TREM2 risk variants are associated with atypical Alzheimer’s disease

Abstract

Alzheimer’s disease (AD) has multiple clinically and pathologically defined subtypes where the underlying causes of such heterogeneity are not well established. Rare TREM2 variants confer significantly increased risk for clinical AD in addition to other neurodegenerative disease clinical phenotypes. Whether TREM2 variants are associated with atypical clinical or pathologically defined subtypes of AD is not known. We studied here the clinical and pathological features associated with TREM2 risk variants in an autopsy-confirmed cohort. TREM2 variant cases were more frequently associated with non-amnestic clinical syndromes. Pathologically, TREM2 variant cases were associated with an atypical distribution of neurofibrillary tangle density with significantly lower hippocampal NFT burden relative to neocortical NFT accumulation. In addition, NFT density but not amyloid burden was associated with an increase of dystrophic microglia. TREM2 variant cases were not associated with an increased prevalence, extent, or severity of co-pathologies. These clinicopathological features suggest that TREM2 variants contribute to clinical and pathologic AD heterogeneity by altering the distribution of neurofibrillary degeneration and tau-dependent microglial dystrophy, resulting in hippocampal sparing and non-amnestic AD phenotypes.

Introduction

Alzheimer’s disease (AD), the most common form of dementia, is typically clinically characterized by episodic memory deficits followed by progressive impairment in executive, visual, language and neuropsychiatric domains [10]. This clinical course strongly correlates with a stereotypical progression of neurofibrillary tangle (NFT) degeneration, which affects memory-related medial temporal lobe structures prior to neocortical involvement [2, 3]. AD can also manifest as non-amnestic syndromes including logopenic variant primary progressive aphasia (lvPPA), behavioral variant of AD (bvAD) with behavioral/dysexecutive deficits, posterior cortical atrophy (PCA), and corticobasal syndrome (CBS) [10, 16, 36]. This phenotypic variability has been associated with an atypical regional distribution of tau [19, 32, 39], but the mechanisms, including potential genetic modifiers, that contribute to this non-stereotypic distribution are poorly understood.

TREM2 encodes a transmembrane protein that is preferentially expressed in microglia and modulates the innate immune response in the brain [17, 46]. Genetic studies have identified that a rare heterozygous loss-of-function variant p.R47H in TREM2 is associated with an increased risk of AD with an odds ratio comparable to the strongest non-Mendelian genetic risk factor APOE ε4 [14, 21]. Additional coding variants p.R62H, p.H157Y and p.D87N in TREM2 have also been identified as being associated with AD risk [14, 20, 49, 52]. The R47H TREM2 variant is also associated with other neurodegenerative diseases such as Parkinson’s disease (PD) [23], frontotemporal dementia (FTD) [5], and sporadic amyotrophic lateral sclerosis [6]. Homozygous loss of function variants ofTREM2 also cause Nasu-Hakola disease, characterized as a sclerosing leukodystrophy with or without multiple bone cysts [11, 47]. These reports indicate that TREM2 variants have pleiotropic effects which may manifest with variable clinical and pathological phenotypes in the context of neurodegeneration.

Although TREM2 variants have been extensively validated to confer AD risk [27], relatively less is known about the impact of TREM2 variants on clinical and pathological features of AD. Likely due to the low frequency of TREM2 variants, most retrospective studies have conducted in small patient series with inconsistent findings with regard to disease duration [24, 42, 51, 57], age at onset [24, 44, 51], and presenting symptoms [29, 44, 51]. Moreover, most series were based on a clinical diagnosis of probable AD without postmortem confirmation. Neuropathologic studies of TREM2 variant cases with ADNC have described increased tau and amyloid-β (Aβ) burden in multiple brain regions [41, 45], a relative loss of amyloid-associated microglia [35, 41], larger amyloid plaque size [22] and increased severity of α-synucleinopathy [24].

There is a need to better understand the clinicopathologic profiles of AD patients with TREM2 variants in order to better understand the factors that contribute to disease heterogeneity which may improve clinical diagnostic accuracy and eventually to inform the development of disease-modifying therapies. Thus, we investigated the clinical and pathological features associated with autopsy confirmed TREM2 variant AD cases. We report here the largest series of autopsy-confirmed TREM2 variant AD cases to date including comprehensive analysis of clinical phenotypes, regional patterns of NFT and their associations with amyloid and microglial pathologies, and co-morbid neurodegenerative disease pathologies.

Materials and methods

Participants

Genetic, demographic, and diagnostic information on autopsy cases is presented in Table 1. 54 cases with 14 TREM2 variants were identified from 1509 autopsy cases from the Center for Neurodegenerative Disease Research (CNDR) brain bank at the University of Pennsylvania [54]. 18 cases had low or no AD neuropathologic change (ADNC) where the primary neuropathological diagnosis included progressive supranuclear palsy (n=1), multiple system atrophy (n=2), frontotemporal lobar degeneration with TDP-43 inclusions (FTLD-TDP) (n=3), tauopathy (n=1), amyotrophic lateral sclerosis (n=8), low ADNC (n=1), unremarkable adult brain (n=1), and primary age-related tauopathy (PART) (n=1). Of the remaining 36 cases with intermediate or high ADNC, 5 cases were associated with TREM2 variants of unknown significance, resulting in 31 cases with 4 TREM2 variants that have been demonstrated as definite (R47H and R62H) or possible (D87N and H127Y) AD risk modifiers (high ADNC, n=30; intermediate ADNC, n=1). In this cohort, 2 cases were found to also harbor a C9orf72 expansion mutation associated with FTLD-TDP. A comparison cohort of 119 cases without rare TREM2 variants but with intermediate or high ADNC and clinical documentation to define typical amnestic versus atypical non-amnestic AD was identified from cases referred to the CNDR brain bank from the Penn Memory Center.

Table 1.

Clinical characteristics and neuropathologic diagnosis of cases with TREM2 variants

Case No.	Genetic variation	Sex	Age at death (yrs)	PMI (hrs)	Brain weight (g)	Age at onset (yrs)	Disease duration (yrs)	ABC score	Neuropathological diagnosis	Clinical diagnosis	Typical, amnestic AD
Intermediate or high ADNC with TREM2 variants (AD TREM2 variants)
1	TREM2 R47H	F	76	23	1053	68	8	A3B3C3	High ADNC LBD	Probable AD	Yes
2	TREM2 R47H APOE3/3	M	60	12	1136	56	4	A3B3C3	High ADNC	lvPPA	No
3	TREM2 R47H APOE2/3	M	62	8	871	53	9	A3B3C3	High ADNC	Probable AD	Yes
4	TREM2 R47H APOE3/4	F	93	7	999	75	18	A3B3C2	High ADNC LATE	Probable AD	Yes
5	TREM2 R47H	F	87	13	970	76	11	A3B3C3	High ADNC LATE	Probable AD	Yes
6	TREM2 R62H C9orf72 expansion	F	72	5	1200	66	6	A3B3C3	High ADNC FTLD-TDP	svPPA	No
7	TREM2 H157Y APOE3/4	F	79	4	909	69	10	A3B3C3	High ADNC LBD	Probable AD	Yes
8	TREM2 R47H APOE3/3	F	70	18	1010	59	11	A3B3C3	High ADNC LBD	Probable AD, language impairment-predominant	No
9	TREM2 R62H	M	77	4	1100	71	6	A3B3C3	High ADNC	bvFTD	No
10	TREM2 R47H APOE3/4	M	82	9	1285	70	12	A3B3C3	High ADNC LATE	CVD/AD	No
11	TREM2 R62H APOE3/3	F	61	18	997	52	9	A3B3C3	High ADNC	Probable AD	Yes
12	TREM2 R47H APOE3/3	F	83	20	1130	76	7	A3B3C3	High ADNC HS	DLB	No
13	TREM2 D87N APOE3/4	M	74	19	1470	68	6	A3B3C3	High ADNC LBD	DLB	No
14	TREM2 R62H	M	69	5	1187	60	9	A3B3C3	High ADNC	DLB/AD	No
15	TREM2 R47H APOE3/4	F	71	4.5	1087	62	9	A3B3C3	High ADNC LBD	PPA, mixed	No
16	TREM2 R47H APOE4/4	F	87	4	957	63	24	A3B3C3	ALS High ADNC	MND	No
17	TREM2 R47H APOE3/3	M	64	16.5	1320	55	9	A3B3C3	High ADNC LBD	Probable AD	Yes
18	TREM2 R47H APOE3/4	M	78	18	1162	67	11	A3B3C3	High ADNC LATE	FTD-NOS	No
19	TREM2 R62H	M	75	11	1221	62	13	A3B3C3	High ADNC MSA	svPPA	No
20	TREM2 R47H	M	61	11	1471	53	8	A3B3C3	High ADNC LBD	Probable AD	Yes
21	TREM2 R47H APOE3/3	F	80	7	1000	59	21	A3B3C3	High ADNC LATE	Behavioral/dysexecutive variant of AD	No
22	TREM2 R62H APOE3/4	F	65	13.5	1060	51	14	A3B3C3	High ADNC LATE	Probable AD	Yes
23	TREM2 R47H APOE3/4	F	79	43.5	1214	68	11	A3B3C3	High ADNC LATE	Probable AD	Yes
24	TREM2 R47H APOE3/3	F	68	9	1000	60	8	A3B3C3	High ADNC LBD	Probable AD	Yes
25	TREM2 R62H APOE2/3	M	86	12	1243	74	12	A3B3C3	High ADNC	Probable AD	Yes
26	TREM2 R47H APOE3/4	M	71	20	1164	63	8	A3B3C3	High ADNC LBD	PCA	No
27	TREM2 R47H APOE3/4	M	77	9	1280	74	3	A3B3C3	High ADNC LATE	Probable AD	Yes
28	TREM2 R47H APOE3/4	F	74	16	1199	60	14	A3B3C3	High ADNC LBD	Probable AD	Yes
29	TREM2 R47H APOE3/4	M	77	14	1307	68	9	A3B3C3	High ADNC LBD	Probable AD	Yes
30	TREM2 R47H C9orf72 expansion	F	65	3.5	910	55	10	A3B3C2	FTLD-TDP High ADNC	bvFTD	No
31	TREM2 R47H APOE3/3	M	59	14	1352	NA	NA	A2B2C2	Intermediate ADNC	Normal	Yes
32	TREM2 F170L APOE2/3	F	88	5	1236	82	6	A3B3C3	High ADNC LBD	Probable AD	Yes
33	TREM2 L121R	M	49	5	1002	34	15	A2B3C3	Intermediate ADNC	FTD-NOS	No
34	TREM2 L221I APOE4/4	F	64	14	1099	53	11	A3B3C3	High ADNC	Probable AD	Yes
35	TREM2 R47C APOE3/3	M	79	20	1147	72	7	A3B3C3	High ADNC	Probable AD	Yes
36	TREM2 S16F APOE3/4	M	76	14.5	1317	69	7	A3B3C2	High ADNC LBD	Probable AD, language impairment-predominant	No
No or low ADNC with TREM2 variants (Non-AD TREM2 variants)
37	TREM2 c.677–6T>C	M	51	17	1405	-	-	A0B0C0	ALS	PLS	-
38	TREM2 D87N	F	43	14	1079	-	-	A0B0C0	ALS, other (No TDP-43)	MND	-
39	TREM2 L133L	M	91	20	1489	-	-	A0B2C0	Tauopathy, unclassifiable	DLB/FTD-NOS	-
40	TREM2 R47H	M	69	5	1200	-	-	A1B1C0	MSA, CVD	MSA-p	-
41	TREM2 S149G	M	77	NA	1040	-	-	A1B1C2	FTLD-TDP	svPPA	-
42	TREM2 A130V	F	56	6	1353	-	-	A0B0C0	MSA	MSA-p	-
43	TREM2 R62H	M	63	9	1430	-	-	A0B0C0	ALS	MND	-
44	TREM2 D87N	M	48	3	1380	-	-	A0B0C0	ALS	MND	-
45	TREM2 R62H	F	49	6	591	-	-	A1B0C0	FTLD-TDP	FTD-NOS	-
46	TREM2 R47H	F	66	6	951	-	-	A1B1C0	Low ADNC	Schizophrenia	-
47	TREM2 R47H	M	43	30.5	1545	-	-	A0B0C0	Unremarkable adult brain	Depression	-
48	TREM2 R62H SOD1 variant	F	63	6.5	1392	-	-	A0B1C0	ALS, other (No TDP-43)	MND	-
49	TREM2 R47H	F	80	16	1400	-	-	A0B2C0	PART	Schizophrenia	-
50	TREM2 R47H	M	63	NA	120	-	-	A0B0C0	PSP	PSP	-
51	TREM2 R62H	F	61	18	1425	-	-	A0B0C0	FTLD-TDP ALS	bvFTD	-
52	TREM2 R62H	F	56	16	1146	-	-	A1B1C2	ALS HS	MND	-
53	TREM2 R136Q	M	76	4	1213	-	-	A1B1C0	ALS Low ADNC	MND	-
54	TREM2 R47H C9orf72 expansion	F	68	8	1099	-	-	A0B1C0	ALS FTLD-TDP	FTD-NOS/ALS	-

F Female, M Male, PMI post-mortem interval, A Amyloid score, B Braak score, C, CERAD (Consortium to Establish a Registry for Alzheimer’s Disease) score, F female, M male, PMI post-mortem interval, A amyloid score, B Braak score, C, CERAD (Consortium to Establish a Registry for Alzheimer’s Disease) score, ADNC Alzheimer’s disease neuropathologic change, LBD Lewy body disease, FTD-NOS frontotemporal dementia, not otherwise specified, bvFTD behavioral variant of frontotemporal dementia, PPA primary progressive aphasia, lvPPA logopenic variant of primary progressive aphasia, svPPA semantic variant of primary progressive aphasia, MND motor neuron disease, CVD cerebrovascular disease, DLB dementia with Lewy bodies, FTLD-TDP frontotemporal lobar degeneration with TDP-43 (transactive response DNA binding protein 43 kDa) inclusions, LATE limbic-predominant age-related TDP-43 encephalopathy, MSA-p multiple system atrophy-parkinsonian type, ALS amyotrophic lateral sclerosis, PLS primary lateral sclerosis, PCA posterior cortical atrophy, HS hippocampal sclerosis, PART primary age-related tauopathy, PSP progressive supranuclear palsy

Clinical assessment

Clinical records were extracted from the CNDR Integrated Neurodegenerative Disease Database [54]. These include sex, age at death, age at onset, disease duration, presenting symptoms and clinical diagnosis at the time of death by the clinician who assessed the patient, and MMSE scores. Based on the clinical diagnosis and presenting symptoms, clinical phenotype was determined according to the consensus and accepted criteria for typical amnestic or atypical non-amnestic syndromes including logopenic variant of primary progressive aphasia (lvPPA), posterior cortical atrophy (PCA), behavioral variant of frontotemporal dementia (bvFTD), corticobasal syndrome (CBS), and behavioral/dysexecutive variant of AD [1, 8, 12, 36, 43].

Neuropathologic assessment

Human brain tissues were obtained at autopsy and fixed using either 10% neutral buffered formalin or 70% ethanol as described [54]. Tissues were then embedded in paraffin blocks and cut into 6 μm thick sections for histological staining with hematoxylin and eosin (H&E) and well-characterized primary antibodies to detect tau (PHF1), Aβ (NAB228), TDP-43 (1D3), and α-synuclein (SYN303) [54]. Up to sixteen regions are routinely examined in the CNDR neuropathology evaluations as described [54]. Each brain region was semi-quantitatively scored for the severity of neuropathological lesions (0, absent; 0.5, rare; 1, mild; 2, moderate; 3, severe). According to consensus guidelines, proteinopathies and vascular pathology were evaluated as follows. NAB228 and PHF1 positive Aβ, neurofibrillary tangles, and neuritic plaques were evaluated to determine the level of Alzheimer’s disease neuropathologic change (ADNC) [31]. α-synuclein positive Lewy body pathology was assigned into amygdala-predominant, brainstem, limbic or neocortical Lewy body disease (LBD) [30]. TDP-43 proteinopathy was classified into ALS-TDP [4], FTLD-TDP with types A-E [25], and limbic-predominant age-related TDP-43 encephalopathy neuropathological change (LATE-NC) with three stages (stage 1, amygdala; stage 2, hippocampus; stage 3, middle frontal cortex) [34]. The presence of large cerebral infarcts and semi-quantitatively scored cerebral amyloid angiopathy and arteriosclerosis lesions were further evaluated for a low, intermediate, or high likelihood that vascular pathology contributed to cognitive impairment [50].

Quantitative image analysis of regional NFT densities

For quantitative analysis of regional NFT densities, five brain regions were selected as representative association cortices (middle frontal cortex, superior temporal cortex, and angular cortex) and hippocampal subfields (CA1 and subiculum) as previously described [32]. For each case, PHF1 stained regions were scanned at 20x magnification using a Leica Aperio AT2 scanner and analyzed using QuPath software. Using an unbiased method to reduce sampling bias, tiles (0.125 mm² per tile) were systematically selected at regularly spaced intervals across each region and then exported to ImageJ software (NIH, Bethesda, MD) for manual NFT counting. NFT density was expressed as the number of PHF1 positive NFT averaged across the selected tiles in each region. All counts were performed in a blinded manner.

Classification of neuropathological subtypes of AD

On the basis of regional NFT counts, cases with Braak stage V or VI were classified into hippocampal-sparing (HpSp), limbic-predominant (LP), or typical neuropathologic subtypes of AD using the threshold-based mathematical algorithm as previously described by Murray et al [32]. Cases were defined as HpSp AD if they met the following three criteria: 1) The ratio of the average hippocampal NFT to the average cortical NFT was less than 1.1 corresponding to less than the 25^th percentile of AD cases. 2) All three of the hippocampal NFT densities [CA1 (median = 12), subiculum (median = 20), CA1-subiculum average (median = 17)] had to be less than the median values. 3) At least three of the cortical NFT measures [middle frontal (median = 5), superior temporal (median = 10), angular (median = 8), and cortical average (median = 8)] had to be greater than or equal to the median values. Cases were defined as limbic-predominant AD if they met the reverse criteria as follows: 1) The ratio of the average hippocampal NFT to the average cortical NFT was greater than 3.6 corresponding to greater than the 75th percentile of AD cases. 2) All three of the hippocampal NFT densities had to be greater than the median values. 3) At least three of the cortical NFT measures had to be less than or equal to the median values. Cases who did not meet criteria for either HpSp or LP AD were considered typical AD.

Statistical analysis

R-4.1.5 was used for statistical analyses. Continuous variables between AD TREM2 variant and AD TREM2 wild-type groups were compared by unpaired t-test with results expressed as mean ± standard deviation, Mann-Whitney U test with results expressed as median ± SD, or linear mixed effects models with TREM2 genotype and interaction term of TREM2 genotype and brain region as main fixed effects and brain region as a fixed effect covariate. Associations between categorical variables were determined by Fisher’s exact test. Associations between binary or ordinal dependent variables and either of continuous or binary independent variables were evaluated by binomial or ordinal logistic regression. Correlations between continuous variables with a single measurement were determined by Pearson’s or Spearman’s correlation coefficient and those with repeated measurements were evaluated by linear mixed effects models. For longitudinal cognitive analysis, a linear mixed effect model was used with fixed effects including TREM2 genotype and interval from MMSE test to death, and the main effect, the interaction of interval from MMSE test to death and study groups. TREM2 wild-type and middle frontal cortex were set as reference factors for TREM2 genotype and brain region in linear mixed effects models if appropriate. Shapiro-Wilk normality test was conducted if appropriate. A p value < 0.05 was considered statistically significant.

Results

AD TREM2 variants are associated with non-amnestic clinical syndromes

Clinical information is provided for each AD case in Table 1. Clinical features were compared between AD TREM2 variant cases (n = 31, mean age at death ± SD = 73.61 ± 8.91 years, and female = 51.61 %) and AD TREM2 wild-type controls (n = 119, mean age at death ± SD = 75.17 ± 10.14 years, female = 52.94 %; age at death, p = 0.405; sex, p = 1) as summarized in Table 2. With regards to clinical presentation, the AD TREM2 variant group had a high proportion of non-amnestic clinical syndromes compared to the AD TREM2 wild-type group (p=0.002, Table 2). Indeed, of the 31 AD patients with TREM2 variants, only 16 cases (51.61 %, Table 2 and Fig. 1a) met criteria for a typical amnestic syndrome with initial episodic memory deficits and progressive dementia. The remaining 15 cases (48.39 %, Table 2 and Fig. 1a), including 2 cases carrying the C9orf72 expansion that were diagnosed with semantic variant of primary progressive aphasia (svPPA) and bvFTD, had non-amnestic syndromes (behavioral/dysexecutive variant of AD, n=1, 3.23%; bvFTD, n=2, 6.45%; lvPPA, n=1, 3.23%; svPPA, n=2, 6.45%; mixed PPA showing both non-fluent and fluent forms of PPA (n=1), n=1, 3.23%; PCA, n=1, 3.23%; dementia with Lewy bodies (DLB), n=2, 6.45%; frontotemporal dementia, not otherwise specified (FTD-NOS), n=1, 3.23%; motor neuron disease (MND), n=1, 3.23%; probable AD, language impairment-predominant, n=1, 3.23%; mixed cerebrovascular disease/AD, n=1, 3.23%; mixed DLB/AD, n=1, 3.23%).

Table 2.

Clinical features of AD patients with TREM2 variants and TREM2 wild-type

	AD TREM2 variants	AD TREM2 wild-type	p value
N	31	119
Female, N (%)	16 (51.61 %)	63 (52.94 %)	†1
Age at death, yrs	73.61 ± 8.91	75.17 ± 10.14	‡0.405
Clinical phenotype			†0.002*
Typical, amnestic AD, n (%)	16 (51.61 %)	96 (80.67 %)
(Clinical diagnosis)
Probable AD	15 (48.39 %)	87 (73.11 %)
Possible AD	0	6 (5.04 %)
Normal	1 (3.23 %)	3 (2.52 %)
Atypical, non-amnestic AD, n (%)	15 (48.39 %)	23 (19.33 %)
(Clinical diagnosis)
Behavioral/dysexecutive variant of AD	1 (3.23 %)	0
bvFTD	2 (6.45 %)	1 (0.84 %)
CBS	0	2 (1.68 %)
lvPPA	1 (3.23 %)	0
svPPA	2 (6.45 %)	1 (0.84 %)
PPA, mixed	1 (3.23 %)	0
PPA, non-specified	0	1 (0.84 %)
PCA	1 (3.23 %)	1 (0.84 %)
DLB	2 (6.45 %)	3 (2.52 %)
FTD-NOS	1 (3.23 %)	11 (9.24 %)
MND	1 (3.23 %)	0
VaD	0	1 (0.84 %)
Probable AD, frontal features-predominant	0	1 (0.84 %)
Probable AD, hallucination and confusion-predominant	0	1 (0.84 %)
Probable AD, language impairment-predominant	1 (3.23 %)	0
CVD/AD	1 (3.23 %)	0
DLB/AD	1 (3.23 %)	0
N	30	117
Age at onset, yrs	63 ± 7.62	63 ± 10.46	#0.684
Disease Duration, yrs	9 ± 4.53	9 ± 3.92	#0.789
Early-onset AD, n (%)	16 (53.33 %)	61 (52.14 %)	†1
N	18	86
Last MMSE score	8.5 ± 6.54	7 ± 7.61	#0.513

Values are mean ± standard deviation

^†Fisher’s exact test

^‡T-test

^#Mann-Whitney U test

^*p<0.05 is statistically significant.

bvFTD behavioral variant of frontotemporal dementia, CBS corticobasal syndrome, PPA primary progressive aphasia, lvPPA logopenic variant of primary progressive aphasia, svPPA semantic variant of primary progressive aphasia, PCA posterior cortical atrophy, DLB dementia with Lewy bodies, FTD-NOS frontotemporal dementia, not otherwise specified, MND motor neuron disease, VaD vascular dementia, CVD cerebrovascular disease

Fig. 1. Clinical and pathological features of AD patients with TREM2 risk variants.

Pie charts depicts the proportion of (a) clinical phenotypes, (b) neuropathological subtype of AD, and (c) comorbid pathologies observed in AD cases with TREM2 variants versus TREM2 wild-type, showing the relatively high proportion of (a) non-amnestic syndromes and (b) hippocampal-sparing AD, but a similar proportion of (c) mixed pathology in AD TREM2 variant cases compared to AD TREM2 wild-type cases. Abbreviations: bvFTD= behavioral variant of frontotemporal dementia, lvPPA= logopenic variant of primary progressive aphasia, svPPA= semantic variant of primary progressive aphasia, PCA= posterior cortical atrophy, DLB= dementia with Lewy bodies, FTD-NOS= frontotemporal dementia, not otherwise specified, MND= motor neuron disease, CVD= cerebrovascular disease, VaD= vascular dementia, LBD= Lewy body disease, VaP = vascular pathology.

In contrast, the cohort of AD TREM2 wild-type cases had 96 of 119 cases (80.67 %, Table 2 and Fig. 1a) with a typical amnestic syndrome and 23 cases with an atypical non-amnestic syndrome (19.33%, Table 2 and Fig. 1a) including bvFTD (n= 1, 0.84%), CBS (n=2, 1.68 %), svPPA (n=1, 0.84%), PPA, non-specified (n=1, 0.84%), PCA (n=1, 0.84%), DLB (n=3, 2.52 %), FTD-NOS (n=11, 9.24 %), vascular disease [49] (n=1, 0.84%), probable AD, frontal features-predominant (n=1, 0.84%), and probable AD, hallucination and confusion-predominant (n=1, 0.84%). Despite the difference in clinical phenotype, no significant differences were detected between AD TREM2 variant cases and wild-type cases in terms of age at onset (TREM2 variants, n = 30, 63 ± 7.62; TREM wild-type, n = 117, 63 ± 10.46; p = 0.684, Table 2), disease duration (TREM2 variants, n = 30, 9 ± 4.53; TREM2 wild-type, n = 117, 9 ± 3.92; p = 0.789, Table 2), and the proportion of early-onset AD cases (TREM2 variants, n = 16 of 30, 53.33 %; TREM wild-type, n = 61 of 117, 52.14 %; p = 1, Table 2).

For the subset of cases where MMSE scores were available within 5 years of death, there was no difference in MMSE between AD with TREM2 variant cases (n = 18, 8.5 ± 6.54) and AD with TREM2 wild-type cases (n = 86, 7 ± 7.61, p = 0.513, Table 2). However, for the subset of cases with more than two MMSE scores (TREM2 variants, n = 12; TREM wild-type, n =87), we further evaluated the effect of TREM2 variants on longitudinal cognitive decline using a linear mixed effects model. There was a significant interaction effect between genotype and the interval from MMSE to death where TREM2 variant cases exhibited a faster decline in MMSE (β = −0.885, SE = 0.414; p = 0.033, Supplemental Fig. 1) compared to TREM2 wild-type cases. Together, these findings suggest TREM2 variants are associated with non-amnestic clinical syndromes as well as accelerated cognitive decline.

TREM2 variants and Hippocampal NFT density

21 of 31 randomly selected TREM2 variant and sex- and age-matched but otherwise randomly selected 23 of 119 randomly selected TREM2 wild-type cases were included for quantitative image analysis. To explore whether TREM2 variants are associated with an altered distribution of NFT pathology, PHF1-stained sections from three association cortices (middle frontal, superior temporal, and angular cortices) and two hippocampal subfields (CA1 and subiculum) were examined for NFT density (Fig.2, Supplemental table 1). There was no difference in NFT density between TREM2 variant versus wild-type groups in three association cortices including middle frontal cortex (AD TREM2 variants, n = 21, mean ± SD = 7.56 ± 3.80; AD TREM2 wild-type, n = 23, mean ± SD = 6.38 ± 2.68; p = 0.246, Fig.2b), superior temporal cortex (AD TREM2 variants, n = 21, mean ± SD = 10.05 ± 4.28; AD TREM2 wild-type, n = 23, mean ± SD = 9.59 ± 3.98; p = 0.717, Fig. 2b), and angular cortex (AD TREM2 variants, n = 21, mean ± SD = 9.13 ± 4.00; AD TREM2 wild-type, n = 23, mean ± SD = 8.28 ± 2.77; p = 0.419, Fig. 2b). NFT density averaged across these three association cortices also did not differ between the two groups (AD TREM2 variants, n=21, median ± SD = 10.19 ± 3.49; AD TREM2 wild-type, n = 23, median ± SD = 7.71 ± 2.66; p = 0.182, Fig. 2b). Similarly, no difference was detected in NFT density in the hippocampal CA1 subfield between AD TREM2 variants (n = 21, median ± SD = 12.43 ± 5.44) and AD TREM2 wild-type (n = 23, median ± SD = 14.93 ± 8.69; p = 0.256, Fig. 2b). However, NFT density in the subiculum (AD TREM2 variants, n = 21, median ± SD = 9.82 ± 13.16; AD TREM2 wild-type, n = 23, median ± SD = 15.86 ± 9.32; p = 0.002, Fig. 2b) and average hippocampal NFT density (AD TREM2 variants, n = 21, median ± SD = 10.66 ± 8.73); AD TREM2 wild-type, n = 23, median ± SD = 16.29 ± 8.62; p=0.023, Fig. 2b) were significantly lower in AD TREM2 variant cases compared to AD TREM2 wild-type control cases. The ratio of average hippocampal to average cortical NFT density was also significantly lower in AD TREM2 variant cases (n = 21, median ± SD=1.26 ± 0.68) compared to AD TREM2 wild-type controls (n = 23, median ± SD =1.84 ± 1.22; p=0.005, Fig. 2b).

Fig. 2. Regional NFT pathology in AD patients with TREM2 risk variants.

(a) PHF immunohistochemistry shows similar NFT burden in the middle frontal cortex, superior temporal cortex, angular cortex, and CA1 between AD TREM2 variant and AD TREM2 wild-type cases. PHF1 immunostaining of the subiculum shows less numerous NFT pathology in AD TREM2 variant cases than in AD TREM2 wild-type cases. Scale bars = 50 μm. (b) Graphs represent the quantification of NFT density in the middle frontal cortex, superior temporal cortex, angular cortex, cortical average, CA1, subiculum, hippocampal average, and the ratio of the average hippocampal NFT to the average cortical NFT in AD with TREM2 variants versus AD with TREM2 wild-type. NFT density was expressed as the number of PHF1-positive NFT pathology averaged across at least 12 sampling image fields (per 0.125 mm²). Mean (NFT density in middle frontal cortex, superior temporal cortex and angular cortex) and median values (cortical NFT density, NFT density in CA1 and subiculum, hippocampal NFT density, and hippocampal to cortical NFT density) are indicated. *p < 0.05 and **p < 0.01, as determined by unpaired t-test (NFT density in middle frontal cortex, superior temporal cortex and angular cortex) and Mann-Whitney U test (cortical NFT density, NFT density in CA1 and subiculum, hippocampal NFT density, and hippocampal to cortical NFT density).

As analyses were done on both ethanol and formalin fixed tissues, we verified that neurofibrillary tangle counts did not differ between these two types of fixatives based on quantification of a subset of cases for which both ethanol and formalin fixed tissues were available (n=10, r = 0.978; p < 0.001 by Pearson’s correlation coefficient, Supplemental Fig. 2). Moreover, to validate that these NFT quantifications were clinically relevant, we found that NFT density correlated well with MMSE scores obtained within five years of death (n=23, r = −0.478; p = 0.021 by Pearson’s correlation coefficient, Supplemental Fig. 3a). However, Aβ burden (n=15, ρ = −0.081; p = 0.775 by Spearman’s correlation coefficient, Supplemental Fig. 3b) and neuritic plaque density (n=15, ρ = 0.068; p = 0.809 by Spearman’s correlation coefficient, Supplemental Fig. 3c) did not correlate with MMSE. These results indicate our method for determining NFT density are technically sound and clinically relevant, thereby supporting the above finding that TREM2 variants in AD are associated with a decrease in the severity of hippocampal NFT burden relative to the neocortex.

TREM2 variants and Hippocampal-sparing AD

Using criteria defined by Murray et al. [32] with the notable difference that NFT counts were based on PHF1 stained sections as opposed to Thioflavin-S stained sections, cases were neuropathologically assigned to HpSp, typical, or limbic-predominant ADNC. Of the AD cases with TREM2 variants, 15 of 21 cases (71.43 %) including 2 cases carrying the C9orf72 expansion were defined as typical ADNC. The remaining 6 cases (28.57 %, Table 3 and Fig. 1b) were HpSp ADNC. Of the 6 cases with HpSp ADNC, 4 had non-amnestic syndromes including lvPPA (n = 1), mixed PPA (n=1), MND (n = 1), and PCA (n = 1). Of the 23 AD cases with TREM2 wild-type genotypes, 21 cases (91.3 %) were typical ADNC, while one had HpSp ADNC (4.35 %) with FTD-NOS and one with limbic-predominant AD with an amnestic syndrome (4.35 %, Table 3 and Fig. 1b). Fisher’s exact test revealed HpSp ADNC to be more common in AD TREM2 variant cases compared to AD TREM2 wild-type cases (p = 0.046, Table 3). These results suggest that TREM2 variants are associated with HpSp ADNC.

Table 3.

Neuropathologic subtypes of AD TREM2 variants and AD TREM2 wild-type

	AD TREM2 variants	AD TREM2 wild-type	p value (Fisher’s exact test)
N	21	23
Typical AD, n (%)	15 (71.43 %)	21 (91.3 %)	0.046* (Typical AD vs Hippocampal-sparing AD)
Amnestic syndrome, n	8	18
(Clinical diagnosis)
Probable AD	8	18
Non-amnestic syndrome, n	7	3
(Clinical diagnosis)
svPPA	1	0
Probable AD, language impairment-predominant	1	0
CVD/AD	1	0
DLB	1	1
FTD-NOS	1	0
Behavioral/dysexecutive variant of AD	1	0
bvFTD	1	1
CBS	0	1
Hippocampal-sparing AD, n (%)	6 (28.57 %)	1 (4.35 %)
Amnestic syndrome, n	2	0
(Clinical diagnosis)
Probable AD	2	0
Non-amnestic syndrome, n	4	1
(Clinical diagnosis)
lvPPA	1	0
PPA, mixed	1	0
MND	1	0
PCA	1	0
FTD-NOS	0	1
Limbic-predominant AD, n (%)	0	1 (4.35 %)
Typical, amnestic AD, n	0	1
(Clinical diagnosis)
Probable AD	0	1

^*p<0.05 is statistically significant.

ADNC Alzheimer’s disease neuropathologic change, svPPA semantic variant of primary progressive aphasia, CVD cerebrovascular disease, DLB dementia with Lewy bodies, FTD-NOS frontotemporal dementia, not otherwise specified, bvFTD behavioral variant of frontotemporal dementia, CBS corticobasal syndrome, PPA primary progressive aphasia, lvPPA logopenic variant of primary progressive aphasia, MND motor neuron disease, PCA posterior cortical atrophy

Using TREM2 variant cases in the current cohort, we previously reported that TREM2 variants with high ADNC did not exhibit altered regional Aβ burden, but did exhibit decreased Aβ plaque-associated microglia and increased neuritic plaque and tau accumulation, the latter determined by measuring the percent area occupied by PHF1 immunoreactivity [41]. This tau burden data was re-analyzed here using linear mixed effect models to corroborate whether there was evidence of altered regional distribution of tau. This model confirmed that overall tau burden was higher in high AD with TREM2 variants compared to wild-type cases (β = 5.197, SE = 2.319; p = 0.032, Supplemental table 3). However, relative to middle frontal cortex, tau burden in CA1 (β = −7.455, SE = 2.075; p = 0.001, Supplemental table 3) and subiculum (β = −4.796, SE = 2.117; p = 0.026, Supplemental table 3) was significantly lower in high AD TREM2 variant compared to wild-type cases. These findings suggest that TREM2 variants are associated with a relative increase in the severity of overall tau accumulation and a relative sparing of hippocampal tau burden characteristic of HpSp ADNC.

TREM2 variants and Aβ pathology

Based on our observations so far, we hypothesized that the distinct regional patterns of tau accumulation in association of TREM2 variants drive non-amnestic AD. To better understand the relationship between Aβ deposition and neurofibrillary tangle formation in TREM2 variant cases, the relationship between NFT density and Aβ burden or neuritic plaque density were examined across middle frontal cortex, CA1, subiculum, and hippocampus. Neither Aβ burden (β = −0.039, SE = 0.337; p = 0.910, Supplemental table 4) nor neuritic plaque density (β = −0.017, SE = 0.034; p = 0.624, Supplemental table 4) were related to NFT density across middle frontal, CA1, subiculum, and hippocampus in AD with TREM2 variant cases. These suggest that NFT accumulation does not correlate well with Aβ or neuritic plaque deposition in the context of high ADNC with TREM2 risk variants.

To determine whether TREM2 variants are associated with atypical regional patterns of Aβ deposition, we analyzed the ratios of hippocampal to middle frontal cortical Aβ burden and neuritic plaque density [41]. There were no differences in the ratios of hippocampal to middle frontal cortical Aβ burden (AD TREM2 variants, n=9, median ± SD = 0.28 ± 0.57; AD TREM2 wild-type, n = 13, median ± SD = 0.27 ± 0.15; p = 0.744, Supplemental Fig. 4a) or neuritic plaque density (AD TREM2 variants, n=14, median ± SD = 1.12 ± 1.77; AD TREM2 wild-type, n = 12, median ± SD = 1.15 ± 1.56; p = 0.705, Supplemental Fig. 4b) between AD TREM2 variant and wild type cases. Thus, the atypical clinical phenotypes associated with TREM2 variants did not appear to be driven by altered regional distributions of Aβ amyloid or neuritic plaques.[41].

TREM2 variants and microglial response to NFT pathology

We previously described that TREM2 variant cases exhibit decreased numbers of microglia per amyloid plaque and an apparent increase in the proportion of microglia with a dystrophic morphology [41]. To better understand how these altered microglial profiles, in particular dystrophic microglia, are involved in AD pathogenesis in the context of TREM2 variants, the relationships between the proportions of microglial subtypes and the accumulations of specific AD neuropathologies were evaluated across middle frontal cortex, CA1, subiculum, and hippocampus (Supplemental table 5). NFT density (β = −1.254, SE = 0.468; p = 0.014) but not Aβ burden (β = 1.082, SE = 0.779; p = 0.179) or neuritic plaque density (β = 0.053, SE = 0.055; p = 0.345) was negatively correlated with the proportion of homeostatic microglia. None of these AD pathologies were significantly correlated with the proportion of activated microglia (NFT density, β = 0.894, SE = 0.463; p = 0.066; Aβ burden, β = −0.974, SE = 0.775; p = 0.221; neuritic plaque density, β = −0.040, SE = 0.054; p = 0.462). Interestingly, NFT density was positively correlated with the proportion of dystrophic microglia (β = 0.324, SE = 0.130; p = 0.021), while Aβ burden (β = −0.099, SE = 0.216; p = 0.651) or neuritic plaque density (β = −0.011, SE = 0.015; p = 0.496) was not correlated with the proportion of dystrophic microglia. These findings suggest that in addition to the decreased number of amyloid-associated microglia we reported previously [35, 41], the overall altered TREM2-mediated microglia response may be linked to NFT pathology.

TREM2 variants and Concomitant pathologies

To assess whether atypical clinical phenotypes associated with TREM2 variants were associated with other co-existent neurodegenerative disease pathologies in AD, we evaluated frequencies of mixed pathology including vascular pathology, Lewy body disease (LBD), and TDP-43 proteinopathy in AD TREM2 variant compared to AD TREM2 wild-type cases (Table 4 and Fig. 1c). Of the 31 AD TREM2 variant cases, 8 cases (25.81%) had pure AD pathology defined as ADNC only or ADNC together with low probability of cerebrovascular pathology, while the majority of cases had mixed pathology (n = 23, 74.19%, Table 4 and Fig. 1c). Specifically, 7 cases exhibited LBD (22.58%, Fig.1c) and 7 cases exhibited TDP-43 proteinopathy (22.58%, Fig. 1c). 6 cases had both LBD and TDP-43 proteinopathy (19.35%, Fig.1c). One case had both TDP-43 proteinopathy and vascular pathology, one case exhibited both TDP-43 proteinopathy and vascular pathology, and one case exhibited LBD, TDP-43 proteinopathy and a coexistent moderate probability of vascular pathology (3.23% each, Fig.1c).

Table 4.

Concomitant pathologies in AD patients with TREM2 variants and TREM2 wild-type

		AD TREM2 variants	AD TREM2 wild-type	p value (Fisher’s exact)
N		31	119
Mixed pathology, n (%)				0.628
Yes		23 (74.19 %)	94 (78.99 %)
	AD + LBD	7	32
	AD + TDP-43 (LATE)	5	18
	AD + TDP-43 (FTLD-TDP)	1	1
	AD + TDP-43 (ALS-TDP)	1	1
	AD + VaP	0	2
	AD + LBD + TDP-43 (LATE)	6	28
	AD + LBD + TDP-43 (FTLD-TDP)	0	1
	AD + LBD + VaP	1	4
	AD + TDP-43 (FTLD-TDP) + VaP	1	0
	AD + LBD + TDP-43 (LATE) + VaP	1	7
No		8 (25.81 %)	25 (21.01 %)
	AD only	4	8
	AD + Low probability of VaP	4	17
Vascular pathology (%)				0.332
Yes		22 (70.97 %)	95 (79.83 %)
	Low probability	19	82
	Moderate probability	2	7
	High probability	1	6
No		9 (29.03 %)	95 (20.17 %)
Lewy body disease, n (%)				0.229
Yes		15 (48.39 %)	72 (60.5 %)
	Brain stem	1	11
	Transitional or limbic	1	22
	Diffuse or neocortical	3	8
	Amygdala	10	31
No		16 (51.61 %)	47 (39.5 %)
TDP-43 proteinopathy, n (%)				1
Yes		15 (48.39 %)	56 (47.06 %)
	LATE-NC	12	53
	Stage1 (Amygdala only)	2	16
	Stage2 (+ Hippocampus)	10	33
	Stage3 (+ Middle frontal gyrus)	0	4
	FTLD-TDP	2	2
	Type A	2	1
	Type B	0	1
	ALS-TDP	1	1
No		16 (51.61 %)	63 (52.94 %)

LBD Lewy body disease, TDP-43 transactive response DNA binding protein 43 kDa, FTLD-TDP frontotemporal lobar degeneration with TDP-43 inclusions, ALS amyotrophic lateral sclerosis, VaP vascular pathology, LATE-NC limbic-predominant age-related TDP-43 encephalopathy

Likewise, in the cohort of 119 cases with AD TREM2 wild-type genotype, 25 cases (21.01%, Table 4 and Fig.1c) had pure AD and the remaining 94 cases (78.99%, Table 4 and Fig.1c) had co-morbid pathology including LBD (n= 32, 26.89%, Fig.1c), TDP-43 proteinopathy (n=20, 16.81%, Fig.1c), and vascular pathology (n=2, 1.68%, Fig.1c). 29 cases were found to exhibit both LBD and TDP-43 proteinopathy (24.37%, Fig.1c), while 4 cases had both LBD and vascular pathology (3.36%, Fig.1c). The remaining 7 cases had LBD, TDP-43 proteinopathy and vascular pathology (5.88%, Fig.1c). Statistically, the prevalence of mixed pathology did not differ between AD with TREM2 variants (n=23 of 31, 74.19 %) and AD with TREM2 wild-type genotypes (n=94 of 119, 78.99 %; p=0.628, Table 4).

Upon analyzing each type of co-morbid neuropathologic change, the frequencies of vascular pathology (TREM2 variants, n=22 of 31, 70.97 %; AD TREM2 wild-type, n=95 of 119, 79.83 %; p=0.332, Table 4), Lewy body disease (AD TREM2 variants, n=15 of 31, 48.39 %; AD TREM2 wild-type, n=72 of 119, 60.5 %; p=0.229, Table 4), and TDP-43 proteinopathy (AD TREM2 variants, n=15 of 31, 48.39 %; AD TREM2 wild-type, n=56, 47.06 %; p=1, Table 4) did not differ between the two groups. Thus, TREM2 variants did not appear to affect the prevalence of concomitant pathologies in AD.

We also evaluated the extent and severity of co-morbid neuropathologies in cases with or without TREM2 risk variants (Supplemental table 6). TREM2 variants were not associated changes in LBD stage (p = 0.120) or α-synuclein scores in amygdala (p = 0.177), hippocampus (p = 0.226), or middle frontal cortex (p = 0.434). Similarly, TREM2 variants were not associated with changes in LATE-NC stage (p = 0.738) or TDP-43 proteinopathy scores in amygdala (p = 0.766), hippocampus (p = 0.136), or middle frontal cortex (p = 0.704). There were no associations of TREM2 variants with cerebrovascular levels using VCING criteria (p = 0.825), the presence of large infarcts (p = 0.905), cerebral angiopathy scores in occipital lobe (p = 0.087) or arteriolosclerosis scores in occipital white matter (p = 0.391). In addition, TREM2 variants were not associated with the number of non-AD pathologies co-existing with AD (p = 0.496). These findings suggest that TREM2 variants did not appear to have an impact on the extent and severity of concomitant pathologies in AD.

Finally, we examined whether accumulations of specific AD neuropathologies were associated with the extent of non-AD neuropathologies (Supplemental table 7) in cases with TREM2 risk variants. Aβ load in middle frontal cortex was not associated with LBD stage (p = 0.571), LATE-NC stage (p = 0.285), or VCING levels (p = 0.360). Neuritic plaque density (LBD stages, p = 0.638; LATE-NC stages, p = 0.431; VCING levels, p = 0.851) and NFT density (LBD stages, p = 0.744; LATE-NC stages, p = 0.199; VCING levels, p = 0.991) in middle frontal cortex were also not associated with the extent of non-AD neuropathologic change. Likewise, Aβ burden (LBD stages, p = 0.418; LATE-NC stages, p = 0.198; VCING levels, p = 0.561), neuritic plaque density (LBD stages, p = 0.873; LATE-NC stages, p = 0.207; VCING levels, p = 0.958), and NFT density (LBD stages, p = 0.544; LATE-NC stages, p = 0.100; VCING levels, p = 0.233) in hippocampus were not associated the amount of comorbid neuropathologies. Overall, there was no evidence that TREM2 variants or specific AD neuropathologies were associated with increased prevalence or severity of non-AD comorbid neuropathologies.

Discussion

We report here the clinical and pathological phenotypes observed in autopsy-proven AD cases with TREM2 disease risk variants. Clinically, TREM2 variants were associated with non-amnestic clinical syndromes. These non-amnestic clinical phenotypes were not associated with amyloid pathology but rather an atypical, HpSp distribution of neurofibrillary degeneration. While TREM2 variants were associated with accelerated cognitive decline, TREM2 variants in AD were not associated with an increased frequency, extent, or severity of co-morbid neurodegenerative disease pathologies. Finally, the overall proportion of dystrophic microglia correlated with NFT density but not amyloid burden or neurtic plaque density. Thus, TREM2 variants appear to be associated with distinct clinicopathologic features including non-amnestic AD, an atypical distribution of NFT pathology, and more rapid cognitive decline, and these appear to be associated with the accumulation of dystrophic microglia independent of amyloid pathology.

Since the identification of rare variants in the coding sequence of TREM2 in association with risk for AD [14, 21], several groups have explored the effects of the variants on clinical features of the disease in AD patients. However, associations have been heterogeneous thus far, in part due to relatively small sample sizes due to the low allele frequency of TREM2 risk variants. Indeed, the R47H variant has been associated with altered disease duration in some [24, 42, 57], but not all studies [51]. One group showed that the variant decreased the age at onset of AD [51], in contrast with other cohorts [24, 44]. Additional clinical features, including neuropsychiatric symptoms, apraxia, and parkinsonian signs, have been previously associated with TREM2 variants in AD [29], while these atypical phenotypes were not noted in other cohorts [44, 51]. The majority of cases in these studies were categorized based on an antemortem diagnosis of AD without postmortem confirmation.

In this largest autopsy-confirmed cohort of TREM2 variants to date, we found that TREM2 variants were associated with non-amnestic clinical syndromes. The clinical phenotypes included behavioral/dysexecutive variant of AD, bvFTD, PPA, PCA, DLB, FTD-NOS, and MND, as well as mixed AD. Therefore, TREM2 variants may be associated with atypical AD clinical phenotypes, raising the prospect that some of previous associations with non-AD neurodegenerative disease clinical phenotypes may actually be due to underlying ADNC. Moreover, TREM2 variants were associated with accelerated global cognitive function. This result is consistent with the finding that atypical variants of AD often exhibit a more rapid cognitive decline [19].

Regional tau burden appears to correlate with various clinical manifestations in AD. For example, in vivo measurements of tau burden are higher in medial temporal lobe in patients with an amnestic presentation and in the clinically affected neocortical regions in those with non-amnestic presentations [7, 33, 37, 40], and they correlate with impairment in cognitive domains in a region-specific manner [37]. Likewise, postmortem studies identified relatively low hippocampal to cortical NFT burden in atypical variants of AD compared to typical amnestic AD [32, 39]. In the present study, our quantitative analysis revealed low NFT density in the hippocampus and a low ratio of hippocampal to cortical NFT density in AD TREM2 variant cases compared to TREM2 wild-type cases. This contrasts with the typical distribution profile of NFT pathology described by the Braak-staging scheme with relatively higher NFT burden in the medial temporal lobe compared to neocortex [2, 3]. Therefore, TREM2 variants appear to alter the distribution of NFT pathology resulting in a higher proportion of non-amnestic clinical presentations. Although our observation of the regional NFT burden corresponding to clinical syndromes is not surprising, it is notable that we, for the first time, detected atypical patterns of NFT accumulation associated with TREM2 variants.

Murray et al. formalized the definition of atypical neurofibrillary neurodegeneration based on the distribution of NFT pathology which defines three ADNC subtypes consisting of typical ADNC, HpSp ADNC, and limbic predominant ADNC [32]. Adopting these criteria, even though we employed PHF1 immunohistochemistry instead of Thioflavin staining, [32], we found that TREM2 variants were associated with HpSp AD. Given that we observed that TREM2 variants were associated with non-amnestic clinical syndromes, the overabundance of HpSp ADNC is consistent with previous studies demonstrating that HpSp AD is associated with non-amnestic clinical phenotypes [32, 56]. Although NFTs were detected by PHF1 immunohistochemistry instead of thioflavin-S fluorescence used by Murray et al., NFT counts measured by the two methods are strongly correlated [28] and analyses using thiofiavin-S and phospho-tau antibody staining have yielded similar results in terms of the proportion of ADNC subtypes [32, 56]. Moreover, the distinction between atypical and typical ADNC subtypes is statistical in nature, and we have applied here the same statistical criteria to an age- and sex-matched but otherwise random selection of AD cases which revealed significant differences between TREM2 variant versus TREM2 wild-type cases.

TREM2 is expressed on microglia and plays an important role in anti-inflammation and phagocytosis of cellular debris, both of which attenuate neurodegeneration [9]. TREM2 is differentially expressed across human brain regions, with higher levels in hippocampus and white matter and lower levels in cortical regions in healthy individuals [38, 53]. However, the expression pattern is inversed in AD brains showing lower hippocampal TREM2 compared to frontal TREM2 levels [38]. In addition, TREM2 expression in hippocampus remains stable across AD disease severity [38, 53]. These findings suggest that TREM2 may exhibit region-specific effects on AD progression.

Prokop et al. described that TREM2 variants were associated with an increase in overall tau burden including tau-positive dystrophic neurites associated with neuritic plaques, but not with Aβ burden [41]. We incorporated these findings into the current study to further evaluate relationship between NFT and Aβ burden in TREM2 risk variant cases. We found that NFT density was not correlated with Aβ burden or neuritic plaque density. These findings suggest that the increased overall tau burden observed in cases with TREM2 risk variants is not due to an increase in amyloid, raising the possibility that other mechanisms such as the loss of amyloid associated microglia may be responsible for the downstream formation of tauopathy.

Interactions between microglia and Aβ deposition play a pivotal role in inflammatory responses involving AD. Indeed, in vivo studies of TREM2 deficiency have focused on the upstream role of Aβ in the pathogenesis of AD, which includes disruption in microglial ability to Aβ phagocytosis and thereby an increase in Aβ accumulation, eventually facilitating tau pathology and contributing to neurodegeneration. [15, 26]. In support of this, we and others have reported a decrease in amyloid-associated microglia in TREM2 risk variant cases [35, 41]. Furthermore, Prokop et al. reported an altered proportion of dystrophic microglia but not homeostatic or activated microglia in TREM2 risk variants compared to wild-type cases [41]. In the current study, we further examined how the altered microglial profiles interact with AD pathology in the setting of TREM2 variants. We observed that overall dystrophic microglia load was related to NFT density and not Aβ or neuritic plaque accumulation, suggesting that TREM2 variants may alter microglial responses to tauopathy. This raises the potential that altered TREM2-mediated microglia response may be linked to Aβ as well as NFT pathology. Indeed, experimental model studies have suggested that TREM2 risk variants attenuate microglial reactivity in response to tauopathy, with the notable caveat that murine models do not typically exhibit dystrophic microglia [13].

Clinically-defined cohorts of AD have demonstrated considerable heterogeneity including the concomitant pathologies including Lewy body disease, TDP-43 proteinopathy, and vascular lesions [18, 34, 50]. These contribute to dementia clinical phenotypes and influence clinical presentations of AD [18, 34, 48, 50], although NFT pathology is a strong determinant of AD clinical profiles [55]. Thus, we hypothesized that different patterns of co-pathologies may be identified between AD patients with TREM2 variants versus TREM2 wild-type. However, there was no difference in the prevalence, extent, and severity of mixed pathology between AD with TREM2 variant cases and AD with TREM2 wild-type cases. In addition, none of AD pathologies associated with specific types of non-AD pathologic change. These findings are inconsistent with the previous report of increased density of α-synuclein burden observed in a study of one kindred with relatively few autopsied brains with or without the TREM2 R47H variant [24]. Overall, our comprehensive assessments of co-pathologies suggest that the atypical distribution of NFT pathology associated with TREM2 variants appears to be the main driver of the atypical, non-amnestic clinical phenotypes in AD. We have also identified two cases with TREM2 risk variants and the C9orf72 repeat expansion mutation. Both cases exhibited a high level of ADNC together with FTLD-TDP indicating that TREM2 risk variants appear to promote ADNC even in the setting of autosomal dominant FTLD-TDP.

A weakness of our study is the relatively small number of cases due to the relative rarity of TREM2 variants, although this cohort represents the largest series of autopsy-confirmed cases to date. Moreover, a subset of the clinical classifications was based on retrospective analysis of clinical reports. Although further replication studies need to confirm our findings in larger autopsy cohorts, our results suggest that TREM2 variants may be associated with non-amnestic clinical syndromes and an atypical distribution of NFT accumulation, which correlates with dystrophic microglia load. Thus, we speculate that clinical and pathological AD heterogeneity is driven at least in part by genetic variation, and that altered TREM2-dependent microglial reactivity appears to modify downstream patterns of neurofibrillary degeneration.