Microglial reactivity correlates to presynaptic loss independent of β-amyloid and tau

Abstract

Objects:

Triggering receptor expressed on myeloid cell 2 (TREM2) and progranulin (PGRN) are critical regulators of microglia activation and could be detected in cerebrospinal fluid (CSF). However, whether microglial reactivity is detrimental or neuroprotective for Alzheimer’s disease (AD) is still debatable.

Methods:

We identified 663 participants with baseline β-amyloid (Aβ) PET and CSF biomarkers data, including phosphorylated tau181 (p-Tau₁₈₁), soluble TREM2 (sTREM2), PGRN, and growth-associated protein-43 (GAP-43). Among them, 254 participants had concurrent longitudinal CSF biomarkers. We used multivariate regression analysis to study the associations of CSF microglial biomarkers with Aβ PET, CSF p-Tau₁₈₁, and CSF GAP-43 cross-sectionally and longitudinally. A Chinese aging cohort’s independent CSF samples (n = 65) were analyzed as a validation.

Results:

Higher baseline levels of CSF microglial biomarkers were related to faster rates of CSF sTREM2 increases and CSF PGRN decreases. Elevated CSF p-Tau₁₈₁ was associated with higher levels of CSF microglial biomarkers and faster rates of CSF sTREM2 increases and CSF PGRN decreases. In both cohorts, higher Aβ burden was associated with attenuated CSF p-Tau₁₈₁ effects on CSF microglial biomarkers increases. Independent of Aβ PET and CSF p-Tau₁₈₁ pathologies, higher levels of CSF sTREM2 but not CSF PGRN were related to elevated CSF GAP-43 levels and faster rates of CSF GAP-43 increases.

Interpretation:

These findings suggest that higher Aβ burden may attenuate the p-Tau-associated microglial responses, while TREM2-related microglial reactivity may independently correlate to GAP-43-related presynaptic loss. This study highlights the two-edged role of microglial reactivity in AD and other neurodegenerative diseases.

Introduction

Alzheimer’s disease (AD) is characterized by the accumulation of β-amyloid (Aβ) plaques and neurofibrillary tau tangles concurrent with progressive neuronal and synaptic loss strongly correlated with cognitive decline^1–3. Although the underlying mechanism linked to synaptic loss is not well-established, microglial-mediated phagocytosis and neuroinflammation in the brain, which are now acknowledged as central players in AD pathogenesis, may contribute to synaptic degeneration beyond the synaptotoxic effects of Aβ and tau^1,4,5. Triggering receptor expressed on myeloid cells 2 (TREM2) and progranulin (PGRN) are critical regulators of microglial activation, phagocytosis, and proliferation^6,7. TREM2 deficiency prevented microglia-mediated synaptic removal by directly inhibiting microglial phagocytosis, and TREM2 overexpression conversely exacerbated synaptic impairment^8,9. PGRN deficiency could also facilitate microglia-mediated synaptic elimination through complement activation^6,10. Although the findings in animals are sufficiently investigated, knowledge of the involvement of microglia in the synaptic loss in humans of aging and AD is still limited.

Soluble TREM2 (sTREM2) and PGRN, primarily shedding from microglia, can be detected in cerebrospinal fluid (CSF). Thus the CSF levels of sTREM2 and PGRN are considered markers of microglial TREM2 and PGRN signaling, reflecting the status of microglial reactivity^11–13. This was supported by decreased CSF and plasma levels of sTREM2 and PGRN in patients with loss-of-function variants of TREM2 and PGRN^14–16. In mice carrying the TREM2 p.T66M missense mutation, the generation of brain sTREM2 also apparently decreased, accompanied by reduced activation and phagocytosis of microglia¹⁷. Previously, cross-sectional studies demonstrated that patients with symptomatic AD had elevated CSF sTREM2 and CSF PGRN in relation to CSF levels of Aβ and p-Tau, indicating microglial immune responses to primary AD pathologies^12,13,18. Accumulating evidence from human and animal models showed that higher CSF sTREM2 and microglial activation were associated with reduced Aβ and tau deposition^19–21. Elevated CSF sTREM2 was also related to attenuated brain atrophy, glucose hypometabolism, and cognitive decline^19,21–24. However, one recent study reported the accelerated effects of CSF sTREM2 and microglial activation on future tau deposition²⁵. Supporting this, CSF sTREM2 has also been linked to Aβ-related tau aggregates^24,26. Together, the roles of TREM2-dependent and PGRN-dependent microglial reactivity in the course of AD remain controversial.

Here, we investigated the associations of CSF microglial biomarkers (sTREM2 and PGRN) with primary AD pathologies and subsequent presynaptic loss using extensive cross-sectional and longitudinal data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) cohort and validated the primary analyses in an independent aging cohort from China. The specific aims of this study are to determine (1) how CSF microglial biomarkers correlate with each other and their association with primary AD pathologies and (2) how CSF microglial biomarkers correlate with presynaptic biomarker growth-associated protein-43 (GAP-43) in CSF. This study may provide novel insights into understanding the association of Aβ plaques, CSF p-Tau, and presynaptic loss with microglial reactivity and have critical clinical implications for the therapeutic strategies targeting microglia in neurodegenerative diseases.

Methods

Participants

The data in this study were obtained from the ADNI database (ida.loni.usc.edu). The ADNI was established in 2003 known as a public-private partnership, led by Principal Investigator Michael W. Weiner, MD. The main objective of ADNI is to determine whether serial magnetic resonance imaging (MRI), PET, other biological markers, and clinical and neuropsychological assessments can be combined to track the progression of mild cognitive impairment (MCI) and early AD. The ADNI study was approved by institutional review boards of all participating centers, and written informed consent was obtained from all participants or their authorized representatives.

We identified 663 participants in this study, including 202 cognitively unimpaired (CU), 350 MCI, and 110 dementia. All participants had simultaneous (within one year) baseline ¹⁸F-florbetapir (FBP) Aβ PET and CSF biomarkers data, including p-Tau₁₈₁, sTREM2, PGRN, and GAP-43. Among them, 254 participants (77 CU, 166 MCI, and 11 dementia) had concurrent longitudinal CSF biomarkers data with at least two measurements in a median (range) of 2.1 (1.3 – 6.0) years of follow-up. Following clinical cognitive status at baseline, all participants were divided into CU and cognitive impairment (CI, including MCI and dementia).

Additionally, we collected 65 participants (8 CU, 21 MCI, 36 dementia) with CSF samples from the Greater-Bay-Area Healthy Aging Brain Study (GHABS) in China (ClinicalTrials.gov ID: NCT06183658) as a validation cohort²⁷.

CSF biomarkers measurements

In the ADNI cohort, CSF p-Tau₁₈₁ was quantified using the fully automated Roche Elecsys at the University of Pennsylvania²⁸. CSF microglial biomarkers were measured using an MSD platform-based assay (Haass group) at the Ludwig-Maximilians Universität München^13,18. CSF GAP-43 was measured by an in-house ELISA as previously described²⁹. Linear mixed effect (LME) models were used to calculate slopes of CSF biomarkers for all the participants with longitudinal CSF data, controlling for the following independent variables: time, age, sex, and a random slope and intercept.

In the validation cohort, CSF Aβ₄₂ and CSF p-Tau₁₈₁ were quantified using the commercial Neurology 4-plex E kit (cat: 103670, Quanterix) and Advantage V2.1 kit (cat: 104111, Quanterix) by the Simoa HD-X. CSF sTREM2 was measured using an MSD platform-based assay developed by the Haass group^13,18. CSF GAP-43 was detected using the commercial Human GAP-43 ELISA Kit (cat: abx250779, Abbexa). All CSF biomarker measurements were conducted at the Shenzhen Bay Laboratory.

Amyloid PET and MRI imaging

More details on FBP PET image and structural MRI acquisition can be found online (http://adni-info.org). FBP PET data were acquired in 4×5-min frames from 50 to 70 min post-injection. PET images were motion-corrected, time-averaged, and summed into one static frame. Cortical FBP uptakes in 68 regions of interest (ROIs) defined by the Desikan-Killiany atlas³⁰ in FreeSurfer (V7.1.1) were extracted from each FBP PET scan that coregistered to individual corresponding structural MRI scan (closest in time to FBP PET scan). A composite standardized uptake value ratio (SUVR) was calculated by referring FBP uptake in AD summarized cortical regions (including frontal, cingulate, parietal, and temporal regions) to the mean uptake in the whole cerebellum³¹. The Aβ positivity (Aβ+) was defined as AD summarized cortical regions SUVR ≥ 1.11³¹.

Statistical analysis

Statistical analyses were performed using the statistical program R (v4.1.1, The R Foundation for Statistical Computing). Data in this study are summarized as the number (%) or median (range). We compared the demographics and clinical characteristics between CU and CI participants using either a two-tailed Mann-Whitney U test or Fisher’s exact test with a significance threshold value of p < 0.05. LME models were used to investigate the longitudinal changes in CSF sTREM2 and CSF PGRN over time, and multivariable linear regression models were used to explore their association with each other cross-sectionally and longitudinally. Age, sex, education, and APOE-ε4 status were included as covariates for all the models in this study.

To investigate how Aβ and p-Tau pathologies affect the TREM2-related and PGRN-related microglial reactivity, we used baseline Aβ PET or CSF p-Tau₁₈₁ as dependent variables, and baseline or longitudinal CSF microglial biomarkers as independent variables in multivariate linear regression models. We also tested the interaction effect between Aβ PET and CSF p-Tau₁₈₁ on CSF microglial biomarkers, as shown in the following equations:

$$CSFmicroglialbiomarkers~A\beta PET\times CSFp-{Tau}_{181}+covariates$$ (1)

Subsequently, we performed mediation analyses (R; Lavaan package) to determine further the association of Aβ PET, CSF p-Tau₁₈₁, and CSF microglial biomarkers.

Another major aim of this study is to assess the direct effects of TREM2-related and PGRN-related microglial responses on presynaptic loss. To this end, we explored the association of CSF microglial biomarkers with CSF GAP-43 cross-sectionally and longitudinally. To investigate the independent effect of CSF microglial biomarkers, the main effects of Aβ PET and CSF p-Tau₁₈₁ were further adjusted in our models. We also tested the interaction effect between Aβ PET and CSF microglial biomarkers on CSF GAP-43. The equations for the interaction are shown as follows:

$$CSFGAP43~CSFmicroglialbiomarkers\times A\beta PET+CSFp-{Tau}_{181}+covariates$$ (2)

To ensure whether our findings were driven by clinical impairment or the presence of abnormal amyloid plaques, we repeated all analyses when restricting the models to CU, CI, or Aβ+ participants only.

Results

Table 1 summarizes all participants’ baseline demographics, CSF biomarker levels, and Aβ PET stratified by cognitive status. The CU and CI subgroups significantly differed in sex, duration of education, and prevalence of APOE-ε4 carriers but not in age. Longitudinal data of CSF biomarkers are also displayed in Table 1. The demographics of the validation cohort are shown in Supplementary Table 1.

Table 1.

Demographics and characteristics of participants

Characteristics at baseline	ALL	CU	CI
	n = 663	n = 202	n = 461
Age, year	72.4 (55.2–91.5)	72.6 (56.4–86.0)	72.4 (55.2–91.5)
Female	299 (45%)	106 (52%)	193 (42%)
APOE-ε4 carrier	308 (46%)	57 (28%)	251 (54%)
Education, year	16 (8–20)	16 (8–20)	16 (9–20)
Aβ PET, SUVR	1.15 (0.84–2.00)	1.05 (0.84–2.00)	1.26 (0.84–2.00)
CSF biomarkers
p-Tau181, pg/mL	23.3 (8.0–97.0)	19.1 (8.0–60.1)	25.4 (8.2–97.0)
sTREM2, pg/mL	3421 (504–12012)	3514 (504–12012)	3391 (518–11714)
PGRN, pg/mL	1524 (538–3664)	1519 (538–2806)	1534 (654–3664)
GAP-43, pg/mL	4614 (1088–19971)	4306 (1167–17927)	4724 (1088–19971)
Longitudinal CSF biomarkers	n = 254	n = 77	n = 177
Visits of microglial biomarkers	2 (2–3)	2 (2–3)	2 (2–3)
Duration of microglial biomarkers, years	2.1 (1.4–5.1)	2.1 (1.4–4.4)	2.0 (1.7–5.1)
Visits of GAP-43	2 (2–4)	2 (2–3)	2 (2–4)
Duration of GAP-43, years	2.1 (1.3–6.0)	2.2 (1.3–5.0)	2.1 (1.7–6.0)

Data are presented as median (range) or participant’s number (n) and percentage (%).

Abbreviations: APOE: apolipoprotein E; CU: cognitively unimpaired; CI: cognitive impairment; Aβ: amyloid-β; PET: positron emission computed tomography; SUVR: standard uptake value ratio; CSF: cerebrospinal fluid; p-Tau181: phosphorylated tau 181; sTREM2: soluble TREM2; PGRN: progranulin; GAP-43: growth-associated protein-43.

Longitudinal changes in CSF microglial biomarkers

Overall, CSF sTREM2 showed significant increases (standard β (β_std) = 273 [95% confidence interval (ci), 214, 333], p < 0.001) over time contrary to the slight decreases in CSF PGRN (β_std = −13.3 [95% ci, −26.3, −0.3], p = 0.045). Higher baseline levels of CSF microglial biomarkers were associated with faster rates of increases in CSF sTREM2 (sTREM2: β_std = 0.81 [95% ci, 0.73, 0.90], p < 0.001; PGRN: β_std = 0.24 [95% ci, 0.12, 0.37], p < 0.001) but with faster rates of decreases in CSF PGRN (sTREM2: β_std = −0.25 [95% ci, −0.37, −0.12], p < 0.001; PGRN: β_std = −0.93 [95% ci, −0.98, −0.89], p < 0.001). Furthermore, higher CSF sTREM2 was associated with elevated CSF PGRN at baseline (β_std = 0.34 [95% ci, 0.26, 0.41], p < 0.001), but faster rates of increases in CSF sTREM2 were related to more rapid decreases in CSF PGRN over time (β_std = −0.21 [95% ci, −0.33, −0.08], p < 0.001). Congruent results were obtained when tested in CU, CI, and Aβ+ participants (Supplementary Table 2).

Aβ plaques attenuate CSF p-Tau₁₈₁-associated CSF microglial biomarkers increases

Across all participants, elevated CSF p-Tau₁₈₁ was associated with higher baseline levels of CSF sTREM2 (β_std = 0.42 [95% ci, 0.35, 0.50], p < 0.001) and CSF PGRN (β_std = 0.25 [95% ci, 0.17, 0.33], p < 0.001) as well as faster rates of increases in CSF sTREM2 (β_std = 0.41 [95% ci, 0.29, 0.53], p < 0.001) and decreases in CSF PGRN (β_std = −0.17 [95% ci, −0.30, −0.04], p = 0.012) (Figure. 1). The associations with CSF p-Tau₁₈₁ remained significant even though controlling for the main effect of Aβ PET. In contrast, augmented Aβ PET was associated with lower baseline levels of CSF sTREM2 (β_std = −0.20 [95% ci, −0.28, −0.11], p < 0.001) and CSF PGRN (β_std = −0.16 [95% ci, −0.26, −0.07], p < 0.001) but not with slopes of CSF microglial biomarkers after accounting for CSF p-Tau₁₈₁. When the models were restricted to CU, CI, or Aβ+ participants, the same results were yielded for the associations with CSF p-Tau₁₈₁ (Supplementary Table 3). The associations with Aβ PET at baseline were retained in CU and CI participants but marginally in Aβ+ participants. Besides, there was an association between higher Aβ PET and faster rates of increases in CSF PGRN in CU participants.

Figure 1. Association of CSF microglial biomarkers with CSF p-Tau₁₈₁.

Association of baseline CSF p-Tau₁₈₁ with baseline and longitudinal CSF sTREM2 (A-B) and CSF PGRN (C-D). The dash and solid lines represent each group’s regression lines. The baseline Aβ PET × CSF p-Tau₁₈₁ interaction effect was computed. The presented p values were calculated using generalized linear models across all participants, controlling for age, sex, education, and APOE-ε4 status.

Moreover, we found an interaction effect between Aβ PET and CSF p-Tau₁₈₁ on CSF microglial biomarkers across all participants (Figure. 1), in which higher Aβ PET was related to attenuated associations of higher CSF p-Tau₁₈₁ with greater baseline levels of CSF microglial biomarkers and faster rates of increases in CSF sTREM2. We also noticed a marginal interaction on the slope of CSF PGRN. Only CI participants retained the interaction on baseline CSF sTREM2 when tested in different subgroups (Supplementary Table 4). The mediation analyses further showed that CSF p-Tau₁₈₁ mediated the associations between Aβ PET and CSF microglial biomarkers in the whole cohort (Figure. 2). The direct effects of Aβ PET on CSF microglial biomarkers were opposite to the indirect effect mediated by CSF p-Tau₁₈₁, leading to the total disappeared impact of Aβ PET. Mediation analyses generally yielded similar results in CU, CI, and Aβ+ participants (Supplementary Figure. 1).

Figure 2. Mediation analysis of Aβ PET, CSF p-Tau₁₈₁, and CSF microglial biomarkers.

Baseline CSF p-Tau₁₈₁ mediated the association of baseline Aβ PET with baseline CSF sTREM2 and CSF PGRN (A) as well as slopes of CSF sTREM2 and CSF PGRN (B). The solid and the dashed lines show the significant and non-significant pathways, respectively. Total, direct, and indirect associations were computed by a 5,000-bootstrapping procedure, controlling for age, sex, education, and APOE-ε4 status.

CSF sTREM2 levels correlate to CSF GAP-43 independent of Aβ PET and CSF p-Tau₁₈₁

In the primary analyses, higher baseline levels of CSF microglial biomarkers were associated with the more presynaptic loss measured by greater CSF GAP-43 levels (sTREM2: β_std = 0.38 [95% ci, 0.31, 0.45], p < 0.001; PGRN: β_std = 0.22 [95% ci, 0.15, 0.30], p < 0.001; Figure. 3A–B) and faster rates of increases in CSF GAP-43 (sTREM2: β_std = 0.37 [95% ci, 0.26, 0.49], p < 0.001; PGRN: β_std = 0.19 [95% ci, 0.07, 0.32], p = 0.002; Figure. 3C–D). Augmented rates of increases in CSF sTREM2 and decreases in CSF PGRN were also related to faster rates of increases in CSF GAP-43 (sTREM2: β_std = 0.37 [95% ci, 0.26, 0.48], p < 0.001; PGRN: β_std = −0.16 [95% ci, −0.28, −0.03], p = 0.012; Figure. 3E–F). After controlling for Aβ PET and CSF p-Tau₁₈₁, the associations between CSF sTREM2 and CSF GAP-43 were preserved in the whole cohort and partially in CI and Aβ+ participants (Table 2 and Supplementary Table 5). In contrast, no association was obtained for CSF PGRN when adjusting for Aβ PET and CSF p-Tau₁₈₁. These findings suggest that CSF sTREM2 rather than CSF PGRN correlates to GAP-43-related presynaptic loss, independent of Aβ PET and CSF p-Tau₁₈₁.

Figure 3. Association of CSF GAP-43 with CSF microglial biomarkers.

Association of baseline and longitudinal CSF GAP-43 with baseline CSF sTREM2 (A, C) and CSF PGRN (B, D). Association of longitudinal CSF GAP-43 with longitudinal CSF sTREM2 (E) and CSF PGRN (F). The points (blue, CU; red, CI) and solid lines represent the individuals and regression lines, respectively. The presented p values were computed using generalized linear models across all participants, controlling for age, sex, education, and APOE-ε4 status.

Table 2.

Interaction between CSF microglial biomarkers and Aβ PET on CSF GAP-43

Baseline CSF GAP-43	Main effect		Aβ PET interaction
	βstd (95% ci)	p-value	βstd (95% ci)	p-value
Baseline CSF sTREM2	0.08 (0.02 to 0.14)	0.005	0.05 (−0.01 to 0.10)	0.079
Baseline CSF PGRN	0.05 (−0.01 to 0.10)	0.094	−0.06 (−0.11 to −0.01)	0.027
Slope of CSF GAP-43
Baseline CSF sTREM2	0.10 (0.01 to 0.20)	0.036	0.11 (0.03 to 0.19)	0.007
Baseline CSF PGRN	0.05 (−0.03 to 0.14)	0.229	−0.03 (−0.11 to 0.05)	0.483
Slope of CSF sTREM2	0.11 (0.02 to 0.20)	0.016	0.10 (0.02 to 0.19)	0.011
Slope of CSF PGRN	−0.04 (−0.12 to 0.05)	0.390	0.04 (−0.04 to 0.12)	0.311

Abbreviations: ci: Confidence Interval; Aβ: amyloid-β; PET: positron emission computed tomography; sTREM2: soluble TREM2; PGRN: progranulin; GAP-43: growth-associated protein-43.

Subsequently, we further determined whether Aβ pathology modulates the associations between CSF microglial biomarkers and CSF GAP-43. Across all participants, Aβ PET had a significant interaction with CSF sTREM2 on longitudinal CSF GAP-43 (Table 2). This showed that individuals with high levels of Aβ PET had the associations of greater baseline levels and longitudinal increases in CSF sTREM2 with faster rates of increases in CSF GAP-43 (Baseline: β_std = 0.13 [95% ci, −0.01, 0.27]; slopes: β_std = 0.14 [95% ci, 0.01, 0.27]). Conversely, individuals with low levels of Aβ PET had the associations of greater baseline levels and longitudinal increases in CSF sTREM2 with slower rates of increases in CSF GAP-43 (Baseline: β_std = −0.14 [95% ci, −0.28, 0.004]; slopes: β_std = −0.12 [95% ci, −0.25, 0.01]). The interactions with CSF sTREM2 were sustained in Aβ+ participants and partially in CU participants cross-sectionally and longitudinally (Supplementary Table 6). In addition, Aβ PET showed an interaction with CSF PGRN on CSF GAP-43 at baseline across all participants, in which higher Aβ PET was related to attenuated correlation between elevated levels of CSF PGRN and CSF GAP-43 (PGRN × Aβ PET interaction: β_std = −0.06 [95% ci, −0.11, −0.01], p = 0.027), and remained significant in Aβ+ participants.

Association of CSF sTREM2 with Aβ, p-Tau, and presynaptic loss in the validation cohort

In the independent CSF samples (n = 65) obtained from the GHABS cohort, both higher levels of CSF Aβ₄₂ and CSF p-Tau₁₈₁ were marginally associated with higher CSF sTREM2 levels (Aβ: β_std = 0.22 [95% ci, −0.01, 0.46], p = 0.065; p-Tau: β_std = 0.22 [95% ci, −0.02, 0.45], p = 0.072) when we conducted multivariate analyses with CSF Aβ₄₂ and CSF p-Tau₁₈₁ as predictors in one model adjusted for age and sex. Similar to the ADNI cohort, there was an interaction between CSF Aβ₄₂ and CSF p-Tau₁₈₁ on CSF sTREM2, in which lower CSF Aβ₄₂ was related to attenuated association of CSF p-Tau₁₈₁ with CSF sTREM2 (Figure. 4A). Mediation analysis also showed a marginally direct effect of CSF Aβ₄₂ on CSF sTREM2 opposite to the indirect effect mediated by CSF p-Tau₁₈₁ (Supplementary Figure. 2).

Figure 4. Association of CSF sTREM2 with CSF p-Tau₁₈₁ and CSF GAP-43.

Association of CSF sTREM2 with CSF p-Tau₁₈₁ (A) and CSF GAP-43 (B) at baseline in the validation cohort. The points and solid lines represent each group’s individuals and regression lines. The baseline Aβ PET × CSF p-Tau₁₈₁ interaction effect was computed. The presented p values were calculated using generalized linear models across all participants, controlling for age and sex.

Similar to the ADNI cohort, higher CSF sTREM2 was related to elevated CSF GAP-43 (β_std = 0.37 [95% ci, 0.15, 0.59], p = 0.001; Figure. 4B) adjusted for age and sex. The association remained significant after controlling for CSF Aβ₄₂ and CSF p-Tau₁₈₁ (β_std = 0.45 [95% ci, 0.23, 0.67], p < 0.001). No interaction effect on GAP-43 was obtained, which may be because most of the participants were CI (88%) in the validation cohort. The limitation of sample sizes did not allow meaningful analyses among different subgroups.

Discussion

In this study, we showed that elevated CSF p-Tau₁₈₁ was related to higher CSF microglial biomarkers. The effect size of p-Tau₁₈₁-associated CSF microglial biomarkers increases was attenuated by higher Aβ burden. Furthermore, mediation analyses showed that the direct effects of Aβ PET on CSF microglial biomarkers were opposite to the indirect effect mediated by CSF p-Tau₁₈₁. Independent of Aβ PET and CSF p-Tau₁₈₁, higher CSF sTREM2 but not CSF PGRN was associated with elevated CSF GAP-43 and faster rates of CSF GAP-43 increases. Given that CSF sTREM2 and CSF PGRN may reflect expression levels of TREM2-dependent and PGRN-dependent signaling in microglia^11,13, these findings provide novel insights into TREM2-related and PGRN-related microglial responses to primary AD pathologies and modulation on subsequent presynaptic loss (Figure. 5).

Figure 5. Interlinking schematic among Aβ, p-Tau, microglial reactivity, and presynaptic loss.

The findings in the current study suggests that early p-Tau pathology may trigger TREM2-dependent and PGRN-dependent microglial reactivity, which is attenuated by higher Aβ pathology. Besides, TREM2-dependent microglial reactivity may independently correlate to GAP-43-related presynaptic loss.

Several preclinical studies in genetic knock-out mice reported that TREM2 deficiency locked microglia in a homoeostatic signature, whereas PGRN deficiency led to microglial hyperactivation^7,15,32, indicating that TREM2 and PGRN may participate in triggering and restricting microglial activation respectively. This was favored by the fact that TPSO PET signals assessing microglial activation decreased in TREM2-deficient mice but increased in PGRN-deficient mice¹⁵, and loss of TREM2 function could suppress microglial hyperactivation and phagocytosis in PGRN deficiency mice⁷. We and others²¹ found significant longitudinal increases in CSF sTREM2 over time. Intriguingly, we first observed more rapid decreases in CSF PGRN over time parallel with higher baseline levels in this study. The opposite longitudinal changes correspond with the different effects of TREM2 and PGRN signaling in regulating microglial reactivity^7,15,32. Supporting this, we further noticed an association between faster increases in CSF sTREM2 and decreases in CSF PGRN. Given that both sTREM2 and PGRN expressions increased upon microglial activation^33–38, it is likely that the earliest TREM2-related microglial responses to toxic proteins are protective by simultaneously enhancing the PGRN-dependent signaling to suppress microglial hyperactivation. In the later stage, excessive pathological burdens disrupt the microglial homeostasis, eventually reducing PGRN-dependent signaling and increasing detrimental cellular subtypes of microglia.

For Aβ pathology in the brain, early studies revealed that different Aβ aggregates could induce glial activation, but the smaller soluble Aβ oligomers showed a far stronger stimulation to microglial reactivity and neurotoxicity than other Aβ species^1,39. In return, reactive microglia was found to protect against Aβ pathology in amyloidosis mice, especially in the early Aβ seeding stage in a TREM2-dependent manner^40–43. Supporting this, two longitudinal studies in sporadic and familial AD reported the association of increased CSF sTREM2 with slower Aβ accumulation^20,21. In line with our findings, two observational studies reported that individuals with evidence of Aβ pathology only had decreased CSF sTREM2 levels in preclinical and symptomatic AD^18,44. Similarly, trend-level decreases in CSF PGRN were observed in preclinical AD with Aβ pathology only¹³. It may be interpreted by the formation of Aβ plaques sequestering soluble Aβ oligomers³⁹, thereby limiting their potential to induce microglial reactivity. Another alternative explanation may be that reactive microglia are recruited and form a barrier around plaques; thereby, the sTREM2 and PGRN generated from microglia are restricted within plaques^18,45. Regarding tau-related pathology, we and others^13,18,44 observed positive associations between CSF p-Tau₁₈₁ and CSF microglial biomarkers at baseline. Our findings further showed that elevated CSF p-Tau₁₈₁ predicted longitudinal increases in CSF sTREM2 and decreases in CSF PGRN over time. However, another study in familial AD found no relationship between CSF p-Tau₁₈₁ and longitudinal CSF sTREM2 increases²¹. The conflicting results could be explained by several factors, including the differentiative TREM2-related microglial responses to the pathophysiology between sporadic and familial AD or the larger sample sizes and higher age in the current study. Indeed, age-related increases in CSF sTREM2 and activated microglia were characterized in both human brains and mouse models^46–48, implying more susceptible microglia responding to brain pathology in older adults. Intriguingly, the p-Tau effect on CSF microglial biomarkers increases was suppressed by higher Aβ deposition in this study. Mediation analyses further demonstrated the opposite associations of Aβ PET and CSF p-Tau₁₈₁ with changes in CSF microglial biomarkers. Similar results for CSF sTREM2 were yielded using CSF Aβ₄₂ as a predictor instead of Aβ PET in the validation cohort. These results suggest that Aβ PET and CSF p-Tau pathologies have opposite relationships with the levels of microglial TREM2 and PGRN signaling measured by CSF microglial biomarkers.

Growing evidence supports the role of microglia in synaptic elimination during aging and AD. One post-mortem study⁴⁹ demonstrated that microglial processes contained synaptic elements in the hippocampus of AD patients, and the amyloidosis mice experienced greater synaptic engulfment by microglia compared to wild-type mice. Animal studies showed that TREM2 deficiency was linked to reduced microglial phagocytosis and increased dendritic spine densities and synaptic proteins in the brains of aged and amyloidosis mice^9,50. Supporting this, another study found that loss of function in TREM2 markedly reduced synaptic engulfment by microglia in vivo and in vitro experiments⁸, and TREM2 overexpression in the hippocampus could also induce synaptic loss⁹. These findings in animal models suggest that TREM2 signaling plays an active role in microglia-mediated synaptic removal. Importantly, our results in two independent cohorts provided further evidence from living humans by showing positive associations between CSF sTREM2 and CSF GAP-43 cross-sectionally and longitudinally independent of Aβ PET and CSF p-Tau pathologies, supporting the opinion on microglia-associated presynaptic elimination in a TREM2-dependent manner^8,9. Moreover, it is crucial to notice that cortical Aβ deposition significantly modulated the association between CSF sTREM2 and CSF GAP-43. Specifically, higher levels of Aβ PET were related to augmented relationships between CSF sTREM2 and longitudinal increases in CSF GAP-43, suggesting a destructive role of Aβ pathology on microglial TREM2 signaling-related presynaptic dysfunction. This was in accordance with a previous study in non-human primates showing that administration of oligomeric Aβ could trigger an increased uptake of the cortical synapses by microglia paralleled by neuroinflammation⁵¹.

In this study, we systematically investigated the association of CSF microglial biomarkers with Aβ PET, CSF p-Tau₁₈₁, and presynaptic loss measured by CSF GAP-43 in a large dataset. The crucial findings were validated in an independent cohort. However, some caveats should be addressed when interpreting the current results. First, CSF sTREM2 and CSF PGRN are only indirect microglial TREM2 and PGRN signaling measurements. Thus, the levels of microglial activation cannot be concluded unless confirmed by autopsy or PET imaging. Nevertheless, our findings may reflect, at least partly, the TREM2-related and PGRN-related microglial reactivity, given that both sTREM2 and PGRN in CSF are primarily shedding from microglia in the brain^10,15,25. To the best of our knowledge, this study is the first to explore the association between biomarker-based evidence of microglial reactivity and presynaptic integrity in living humans. Therefore, more validations in independent cohorts with large sample sizes are needed in the future, especially by using different synaptic biomarkers (i.e., postsynaptic protein neurogranin⁵² and presynaptic protein SNAP-25⁵³, whose sample sizes are limited in the ADNI cohort). Finally, the current study is observational in nature, and causal associations should not be drawn from our findings.

In conclusion, this study demonstrated that Aβ pathology may attenuate p-Tau-related increases in microglial TREM2 and PGRN signaling, which are independently linked to GAP-43-related presynaptic loss. These findings extend the understanding of the association among primary AD pathology, microglial reactivity, and presynaptic dysfunction in AD and other neurodegenerative diseases, which may have important clinical implications for developing therapeutic strategies targeting microglial TREM2 to prevent the progression of AD.

Data availability

The data used in the current study were obtained from the ADNI database (available at https://adni.loni.usc.edu) and the GHABS cohort. Derived data is available from the corresponding author on request by any qualified investigator subject to a data use agreement.