Introduction

Progranulin (PGRN) is a multifunctional glycoprotein secreted in the central nervous system (CNS) and peripheral tissues, playing a crucial role in regulating lysosome function [1]. In the brain, PGRN is mainly expressed by microglia [2,3,4]. In PGRN-deficient mice, there existed excessive microglial activation, astrogliosis, and an increased expression of proinflammatory cytokines [5]. Additionally, cerebrospinal fluid (CSF) PGRN levels rise significantly during microglial activation [6]. Together, these findings suggest that PGRN may be crucial in modulating microglial-mediated neuroinflammation. Microglial-mediated neuroinflammation is closely associated with the development of Alzheimer’s disease (AD) [7, 8], a common neurodegenerative disorder characterized by β-amyloid (Aβ) plaques and tau protein neurofibrillary tangles. A previous study reported that CSF levels of PGRN were increased across AD continuum [9]. However, the finding has not been validated in an independent population, and how CSF PGRN levels vary in non-AD pathophysiology remains unclear.

Triggering receptor expressed on myeloid cells-2 (TREM2) is expressed abundantly by microglia in the CNS and plays an important role in regulating microglial phagocytosis and responses to inflammatory stimuli [10]. The extracellular domain of TREM2 can be cleaved to soluble TREM2 (sTREM2) and released into the CSF [11, 12]. As an indicator of microglial activity [13], sTREM2 mediates various microglial responses to Aβ and tau phosphorylation during neurodegeneration [14, 15]. CSF sTREM2 levels were also elevated in preclinical AD, and has been correlated with CSF PGRN [16, 17]. This connection raises the possibility that CSF PGRN and sTREM2 may share similar patterns during microglial activation. It was suggested that elevated CSF sTREM2 conferred protection against Aβ accumulation [18]. It remains unclear, however, whether PGRN independently acts or plays a mediating role in influencing Aβ pathology. Although previous studies have described the dynamic changes of CSF PGRN levels and CSF sTREM2 levels in preclinical AD, the evidence regarding their impact on AD pathology is limited. This limitation arises because these studies have not accounted for the potential reverse causality in the causality chain.

Herein, using the biomarker-based A/T/N classification in combination with clinical staging, we explored the associations of CSF PGRN with microglial activity (as indicated by CSF sTREM2), Aβ pathology (as defined as low CSF Aβ1–42), and tau proteins across AD spectrum and non-AD pathophysiology. Besides, we also conducted mediation analyses to explore the potential roles of PGRN and sTREM2 in modulating Aβ pathology. Proteomic and bioinformatic analyses were conducted to explore the potential biological mechanisms underlying the synergistic roles of PGRN and sTREM2 in Aβ pathology. Additionally, we employed linear mixed effects models to test the longitudinal relationships among CSF PGRN, sTREM2 and Aβ1–42.

Materials and methods

CABLE

Subjects

Nine hundred and five cognitively unimpaired adults who were northern Han Chinese were derived from the Chinese Alzheimer’s Biomarker and LifestylE (CABLE) study [19]. Since 2017, CABLE is an ongoing cross-sectional study aiming to explore AD biomarkers in Chinese Han population. Participants in this study were recruited from Qingdao Municipal Hospital, Shandong, China. Inclusion criteria included: 1) self-reported Han Chinese ethnicity, and 2) age between 40 and 90 years. Exclusion criteria were as follows: (1) significant neurological disorders, such as infection, trauma, epilepsy, multiple sclerosis, etc.; (2) major psychiatric conditions, including major depressive disorder and generalized anxiety disorder; (3) malignant tumors; and (4) genetic disorders. All participants underwent neuropsychological assessments and provided bio-samples (blood and CSF) collected by trained medical staff using a structured questionnaire and an electronic medical record system. The CABLE study received approval from the institutional review board of Qingdao Municipal Hospital, and written informed consent was obtained from all participants or their guardians in accordance with the Declaration of Helsinki.

Classification methods

According to 2018 NIA-AA “research framework” for AD diagnosis [20], each CABLE participant was categorized into specific group based on their biomarker profile (as described by the A/T/N scheme [21]). The clinical stage was not considered for classification because all were cognitively normal. The A/T/N scheme includes 3 biomarker groups: “A” represents aggregated amyloid pathology (as indicated by CSF Aβ1–42), “T” aggregated tau (as indicated by CSF P-tau181), and “N” neurodegeneration or neuronal injury (as indicated by CSF T-tau). Based on the CABLE cohort, our previous study using Gaussian mixture model analysis identified the cutoff points for Aβ1–42, P-tau181, and T-tau at 128 pg/mL, 44 pg/mL, and 236 pg/mL, respectively [22]. The reliability of these cutoff values was further confirmed by post hoc analysis. Therefore, the cutoffs were thus chosen to roughly define the A/T/N group for CABLE participants: “A+” participants refer to those with CSF Aβ1–42 < 128 pg/ml, “T+” to those with CSF P-tau181 > 44 pg/ml, and “N+” to those with CSF T-tau > 236 pg/ml. Given that T-tau and P-tau were highly correlate [23], we merged T and N groups to decrease the number of groups to facilitate the analyses, which produced a TN group. TN negative (TN-) was defined as having both the aggregated tau (T) and neurodegeneration (N) biomarkers in the normal range (T- and N-, that is P-tau181 ≤ 44 pg/ml and T-tau ≤ 236 pg/ml). Participants were classified as TN positive (TN+) if either aggregated tau (T) or neurodegeneration (N) were abnormal (T+ or N+, that is P-tau181 > 44 pg/ml or T-tau > 236 pg/ml).

CSF PGRN, sTREM2, and AD core biomarkers measurements

In CABLE, CSF was collected via lumbar puncture into 10 mL polypropylene tubes and transported to the laboratory within 2 h. The CSF samples were then centrifuged at 2000 g for 10 min. The thaw/freezing cycle of tested sample was controlled not to surpass 2 times. All CSF biomarkers were measured using enzyme-linked immunosorbent assay (ELISA) methods on the microplate reader (Thermo Scientific™ Multiskan™ MK3) by experienced researchers who were blinded to the clinical information of the tested samples. Samples were run in duplicate and the mean of measurements were used for analyses. The ELISA kit used were as follows: Human Progranulin ELISA kit (Biovendor-Laboratornimedicinaa.s. Czech Republic), Human TREM2 SimpleStep ELISA kit (Abcam, no. Ab224881), and Human AD core biomarkers (Aβ1–42, P-tau181, and T-tau) ELISA kits (Innotest, Fujirebio, Ghent, Belgium). The within-batch precision values were < 5% for all proteins. The inter-batch coefficients of variation (CVs) were < 15% (10% for PGRN, 5.6% for sTREM2, 9% for Aβ1–42, 12.2% for T-tau, and 10.9% for P-tau181).

APOE 4 measurements

DNA was extracted from blood samples collected after overnight fasting using the QIAamp®DNA Blood Mini Kit, and then amplified by polymerase chain reaction (PCR) with forward primers 5′-ACGGCTGTCCAAGGAGCTG-3′ (rs429358) and 5′-CTCCGCGATGCCGATGAC-3′ (rs7412). The extracted DNA was stored at −80 °C. APOE ε4 genotyping was carried out using restriction fragment length polymorphism (RFLP) technology.

ADNI

Subjects

ADNI (http://www.adni-info.org/) is a multisite study (63 sites in the US and Canada) designed to track the progression of AD in the human brain with clinical, imaging, genetic and biospecimen biomarkers. The participants are volunteers aged 55–90 years with normal aging, mild cognitive impairment, and mild AD dementia. Further information can be found in previous reports [24]. The ADNI study received approval from the institutional review boards of all participating institutions, and written informed consent was obtained from all participants or their guardians.

Classification methods

The classification methods followed those proposed by Marc Suárez-Calvet et al. [9, 23] based on the 2018 NIA-AA “research framework” for AD diagnosis [20]. Each ADNI participant was assigned to a specific group based on both their biomarker profile (using the A/T/N scheme) and clinical stage (as determined by the Clinical Dementia Rating [CDR] score) [21, 23]. In ADNI, “A+“ participants were those with CSF Aβ1–42 levels < 976.6 pg/ml, “T+“ participants had CSF P-tau181 levels > 21.8 pg/ml, and “N+“ participants had CSF T-tau levels > 245 pg/ml. The CSF biomarker statuses established by these thresholds were highly consistent with positron emission tomography (PET) classifications in ADNI [25]. Since the T and N groups were strongly correlated [23], we merged them together to facilitate the analyses. TN negative (TN-) was defined as having both the aggregated tau (T) and neurodegeneration (N) biomarkers in the normal range (P-tau181 ≤ 21.8 pg/ml and T-tau ≤ 245 pg/ml). Participants were classified as TN positive (TN+) if either aggregated tau (T) or neurodegeneration (N) were abnormal (P-tau181 > 21.8 pg/ml or T-tau > 245 pg/ml).

CSF PGRN, sTREM2, and AD core biomarkers measurements

The levels of CSF PGRN, sTREM2, Aβ1–42, P-tau181, and T-tau were extracted from the ADNI database. The CSF procedural protocols used in ADNI have been described previously [26]. CSF PGRN and sTREM2 were measured by an ELISA using the Meso Scale Discovery (MSD) platform [27,28,29]. Streptavidin-coated 96-well multi-array plates were blocked in blocking buffer (0.5% bovine serum albumin and 0.05% Tween 20 in PBS, pH 7.4) overnight. For detection of the target analyte, plates were incubated for 1 h at room temperature with a biotinylated goat anti-target analyte capture antibody (R & D Systems) diluted 1:100 in blocking buffer. Plates were washed four times with washing buffer (0.05% Tween 20 in PBS) before addition of the samples or the standards (GenWay Biotech) and the first detection antibody (mouse monoclonal anti-target analyte antibody; 1:2000 diluted in blocking buffer; R & D Systems). Plates were incubated at room temperature for 2 h, followed by three washing steps. For detection, a SULFO-TAG-labeled secondary anti-mouse antibody (1:1000; MSD) was added, and plates were incubated for 1 h in the dark. After three washes, MSD Read buffer was added, and the light emission at 620 nm after electrochemical stimulation was measured using the MSD Sector Imager 2400 reader. All CSF samples were randomly distributed across plates and measured in duplicate. To minimize batch variability, all antibodies and plates were sourced from a single lot. The mean intraplate CVs were 3.1% (sTREM2) and 2.2% (PGRN); all duplicate measures had a CV < 15%. PGRN and sTREM2 levels were corrected by inter-batch variation and corrected values were used for analyses.

CSF Aβ1–42, P-tau181, and T-tau were analyzed using electrochemiluminescence immunoassays (ECLIA) with the Elecsys system on a fully automated Elecsys cobas e 601 instrument. A single lot of reagents was used for each of the three measured biomarkers. These measurements are for explorative research use only because the performance of the assay beyond the upper technical limit has not been formally established. Samples were normalized across plates using CSF standard values, and the precision of each analyte was assessed by calculating the inter-plate coefficient of variation (CV < 15%).

CSF proteomics measurement

The proteomic levels of 7584 analytes in CSF were measured in baseline samples from the ADNI cohort (n = 758) using the SOMAmer-based capture array method on the SomaScan 7 K platform, a widely utilized tool for biomarker discovery [30]. This platform is highly reliable, with median coefficients of variation for intra- and inter- plate analyses around 5% [31]. The SomaScan assay can detect proteins over a dynamic range from femtomolar (fM) to micromolar (μM), covering 10 orders of magnitude. Its sensitivity surpasses that of traditional immunoassays. After quality control, 7008 analytes from 708 baseline ADNI samples were retained.

Statistical analyses

Prior to the analysis, non-normally distributed CSF markers were log-transformed to achieve normal distributions, as verified by the Kolmogorov-Smirnov test. Extreme outliers (defined as values > 4 standard deviations from the mean) were excluded, and the data were then standardized by converting them to z-scores. These two normalization steps address data skewness and facilitate the interpretation and comparison of coefficients in regression analysis. All analyses were adjusted for age (continuous variable), gender (male = 1), APOE ε4 status (“44/34/24” = 1), educational level (continuous variable), and CDR score (categorical) except where specifically noted.

To test the differences in CSF PGRN across A/T/N framework with different clinical stages, we performed one-way analyses of covariance (ANCOVAs), followed by Bonferroni-corrected post hoc pairwise comparisons. Four comparisons were separately conducted for each biomarker group, including A-/TN+ vs. A-/TN-, A+/TN+ vs. A+/TN- (for the associations with tau-related neurodegeneration), A+/TN+ vs. A-/TN+, and A+/TN- vs. A-/TN- (for the associations with amyloid pathology). A similar approach was used to test for CSF sTREM2 changes. Next, multiple linear regressions were conducted to explore the associations between PGRN (normalized independent variable) and sTREM2 (normalized dependent variable) across biomarker profiles in the A/TN framework. Also, the associations of PGRN and sTREM2 (normalized independent variables) with CSF Aβ1–42, P-tau181, and T-tau (normalized dependent variables) were explored using multiple linear regression models. Given that the likelihood of cognitively unimpaired adults displaying AD pathology varies significantly with age and APOE genotype [32], we included interaction terms between the independent variables and age and APOE ε4 status in the models to further explore how these factors influence the regression results. Additionally, we performed age- and APOE ε4 status-stratified subgroup analyses of the linear regression models. Then, we explored the reciprocal modulation roles of sTREM2 and PGRN in influencing Aβ pathology. To achieve this, mediation analyses were performed using linear regression models, following the approach outlined by Baron and Kenny [33]. Specifically, two models were constructed: the mediator model refers to the linear regression model used for evaluating the relationships between the independent variable and the mediator variable; the outcome model refers to the linear regression model including both the independent variable and mediator variable to assess their effects on the outcome variable, respectively. The magnitude of the mediating effect was quantified by deriving a mediation percentage, which is calculated as the proportion of the indirect effect to the total effect. This percentage was then estimated using Sobel’s test [34], with significance assessed through 1000 bootstrapped iterations. We hypothesize that the CSF protein profiles and the corresponding pathways simultaneously linked to CSF PGRN, CSF sTREM2, and Aβ pathology may provide insights into their synergistic roles. Accordingly, the proteomic and subsequent bioinformatic analyses were performed to investigate the potential mechanisms underlying the interactions between PGRN and sTREM2 in Aβ pathology. Specifically, multiple linear regression models were used to identify CSF proteins (independent variables) simultaneously associated with CSF levels of PGRN, sTREM2, and Aβ1–42 (dependent variables) in the ADNI cohort (Bonferroni-corrected p < 7.13 × 10−6, number of proteins tested = 7008). Functional enrichment analyses were conducted using the STRING database (http://string-db.org). Finally, linear mixed-effects models were used to examine the longitudinal relationships among CSF PGRN, CSF sTREM2, and CSF Aβ1–42 in the ADNI cohort. Linear mixed-effects models were employed because they can handle unbalanced and censored data, while also accommodating time as a continuous factor [35]. A linear mixed-effects model was implemented to investigate the reciprocal relationship between CSF PGRN and CSF sTREM2. Additionally, two other linear mixed-effects models were applied to examine the longitudinal relationships of CSF PGRN and CSF sTREM2 with CSF Aβ1–42, respectively.

Statistical analyses and figure preparation were conducted using R version 4.2.0 (with packages including “lm”, “ggplot2”, “mediation”, and “nlme”) and GraphPad Prism 7.00 software. All tests were two-tailed, with a significance level of p < 0.05.

Results

Participants characteristics

A total of 905 cognitively unimpaired individuals (mean age [SD] = 62.0 [10.7], female = 59.8%) from CABLE and 973 participants (mean age [SD] = 73.1 [7.4], female = 43.8%) from ADNI, including 291 cognitively normal, 580 mild cognitive impairment, and 102 mild dementia, were included in this study. Participants were categorized into four groups according to their A/TN profile (A-/TN-, A+/TN-, A+/TN+, and A-/TN+) as well as the clinical stage (CDR = 0, CDR = 0.5, and CDR = 1 only for ADNI). In both cohorts, compared with the reference group (A-/TN-), individuals with both A-/TN+ and AD continuum (including A+/TN- and A+/TN+) were older. Additionally, the APOE ε4 carriers in the AD continuum were more frequent than those in the A-/TN+ and reference groups (A-/TN-) (Table 1).

Table 1 Characteristics of participants.
Full size table

The changing patterns of PGRN parallels that of sTREM2 within A/TN framework

CABLE participants within the TN+ profile exhibit higher CSF levels of both PGRN (Fig. 1A) and sTREM2 (Fig. 1B) compared to their respective A status groups. The observed trajectories for PGRN (Fig. 1C) and sTREM2 (Fig. 1D) were validated in ADNI participants in CDR = 0 and CDR = 0.5 groups. The number of subjects in the CDR = 1 group was insufficient for a reliable comparison between groups. Accordingly, CSF PGRN and sTREM2 were associated with higher levels of CSF P-tau181 and T-tau in both cohorts, after accounting for the effects of age and APOE genotype in statistical models (Figure S1; Table S1–2). In the CABLE cohort, the associations of CSF PGRN and sTREM2 with TN+ status were significant across different age strata and APOE genotype subgroups. In the ADNI cohort, the association between CSF PGRN and P-tau181 was weakened (p = 0.062) in the age ≤ 65 group (Table S3–4).

Fig. 1: CSF PGRN and sTREM2 within the framework combining A/T/N biomarker and clinical profile.
figure 1

Scatter plot depicting the levels of CSF PGRN A, C and sTREM2 B, D for each of the four biomarker profiles, as defined by the A/T/N framework, coupled with clinical staging, as defined by CDR (for ADNI). The T (tau pathology) and N (neurodegeneration) group were merged. In ADNI, the CDR = 1 stage group includes some biomarker profiles containing insufficient number of subjects, which precludes conducting group comparison. They are still shown for sake of completeness. Each A/T/N biomarker profile is represented in a different color: A-/TN- are depicted in purple, A+/TN- in blue, A+/TN+ in grey, and A-/TN+ in orange. Solid bars represent the mean and the standard deviation (SD). P-values were assessed by a one-way ANCOVA adjusted for age, gender, APOE ε4 status and education. Significant P values after Bonferroni corrected post hoc pairwise comparisons are marked. A Aβ pathology biomarker status, T tau pathology biomarker status, N neurodegeneration biomarker status, AD Alzheimer’s disease, HC healthy controls, CDR clinical dementia rating, CSF cerebrospinal fluid.

Full size image

PGRN positively correlates with sTREM2 independent of A/TN profile

We explored the relationships between PGRN and sTREM2 within the A/TN classification framework. PGRN showed positive correlations with sTREM2 in both cohorts, after adjusting for age and APOE genotype in the models. (Fig. 2A, adjusted p < 0.001), and this association remained significant across age and APOE genotype subgroups in the CABLE cohort (Table S3). Moreover, the relationships remained significant in separate A/TN groups (Fig. 2B for CABLE and Fig. 2C for ADNI).

Fig. 2: The relationships between CSF PGRN and CSF sTREM2.
figure 2

PGRN exhibited positive correlations with sTREM2 in both cohorts A. The significance remained in each A/TN group (B for CABLE and C for ADNI). A Aβ pathology biomarker status, T tau pathology biomarker status, N neurodegeneration biomarker status, CSF cerebrospinal fluid.

Full size image

CSF PGRN was linked to amyloid pathology via mediating sTREM2

Next, we explored the associations of CSF PGRN and sTREM2 with amyloid pathology. In both cohorts, PGRN was associated with higher CSF Aβ1–42 levels in the fully-adjusted models (Fig. 3A, p = 3.31 × 10−10 for CABLE, p = 6.59 × 10−8 for ADNI). Similarly, sTREM2 was significantly associated with higher CSF Aβ1–42 levels in both cohorts (Fig. 3B, p = 2.30 × 10−16 for CABLE, p = 1.65 × 10−12 for ADNI). The relationship of CSF PGRN and sTREM2 with Aβ pathology in the CABLE cohort remained robust across different age strata and APOE genotype groups. In the ADNI cohort, the association between CSF PGRN and Aβ1–42 was weakened in the midlife group (sample size = 136) (Table S3–4).

Fig. 3: The associations of CSF PGRN and sTREM2 with amyloid pathology.
figure 3

CSF PGRN was associated with higher levels of CSF Aβ1–42 A. CSF sTREM2 was associated with higher levels of CSF Aβ1–42 B. PGRN could increase CSF Aβ1–42 partially by modulating CSF sTREM2. Mediation analyses revealed that CSF PGRN could increase CSF Aβ1–42 partially by modulating CSF sTREM2, and CSF sTREM2 could also modulate the relationship between CSF PGRN and CSF Aβ1–42 C. The CABLE data are depicted in blue and ADNI data in red. CSF cerebrospinal fluid.

Full size image

sTREM2 has long been regarded as a marker of microglial activation, which is associated with alleviated amyloid deposition [18, 36]. We thus tested whether the associations of PGRN with AD pathologies were mediated by the microglial marker sTREM2. We found that PGRN was linked to CSF Aβ1–42 by potentially modulating sTREM2 (proportion mediated = 29.7%, p < 2 × 10−16 for CABLE and 33.5%, p < 2 × 10−16 for ADNI). The mediating effects also existed for PGRN in modulating the association between sTREM2 and CSF Aβ1–42 (proportion mediated = 14.6%, p < 2 × 10−16 for CABLE and 15.6%, p < 2 × 10−16 for ADNI) (Fig. 3C).

Proteins underpinning the synergistic roles of CSF PGRN and sTREM2 in Aβ pathology

After Bonferroni correction (p < 7.13 × 10−6), we identified 1539 CSF proteins simultaneously associated with CSF levels of PGRN, sTREM2, and Aβ1–42, with same direction of coefficients. Among these, 403 proteins were negatively correlated, while 1136 proteins were positively correlated. Functional enrichment analyses revealed that the proteins that showed negative correlation were predominantly enriched in immune system process, such as immune response and regulation of cell activation. In contrast, those with positive correlations were primarily associated with nervous system development, including neuronal plasticity and generation of neurons (Fig. 4).

Fig. 4: Proteins and underlying pathways underpinning the associations of CSF PGRN and sTREM2 with amyloid pathology.
figure 4

After Bonferroni correction (p < 7.13 × 10−6), we identified 1539 CSF proteins consistently associated with CSF PGRN, sTREM2, and Aβ1–42. Among these, 403 proteins were negatively correlated A, while 1136 proteins were positively correlated B. The bar charts on the right display the top five enriched pathways for these proteins.

Full size image

The longitudinal relationships among CSF PGRN, CSF sTREM2, and amyloid pathology

To further test the potential causal relationships, linear mixed-effect models were applied to examine the longitudinal relationships among CSF PGRN, CSF sTREM2, and amyloid pathology using the ADNI cohort. Consistent with the cross-sectional analyses, there were significantly positive correlations between baseline CSF PGRN levels and changing rates of CSF sTREM2 (p < 2 × 10−16) as well as between baseline CSF sTREM2 and changing rates of CSF PGRN levels (p < 2 × 10−16). Furthermore, both baseline CSF PGRN and CSF sTREM2 levels were positively associated with changing rates of CSF Aβ1–42 levels (p < 0.05), supporting an interplay role of CSF PGRN and sTREM2 in modulating amyloid pathology (Figure S2).

Discussion

In the present study, we raveled that 1) both PGRN and sTREM2 were positively related to tau proteins in CSF, 2) PGRN was associated with sTREM2 in AD continuum and non-AD pathophysiology, 3) a reciprocal relationship existed between PGRN and sTREM2 in modulating Aβ deposition, 4) CSF proteins associated with PGRN, sTREM2, and Aβ1–42, with similar directionality, were mainly enriched in immune processes and neural plasticity.

Consistent with previous studies [16, 17, 23], we found that both PGRN and sTREM2 in CSF were linked to tau pathology or neurodegeneration, irrespective of Aβ pathology. It is reasonable to infer that 1) increased PGRN and sTREM2 could be a consequence of neuronal injury and neurofibrillary tangle degeneration or 2) elevated PGRN or sTREM2 promoted the tau-related neurodegeneration. It was proposed that elevated levels of these proteins may indicate a microglial response to neuronal damage [29]. Specifically, as tau pathology accumulates and cell death increases, microglia were activated to eliminate apoptotic neurons [37]. Meanwhile, the genes Grn and Trem2 were up-regulated in activated microglia, leading to an increased secretion of PGRN and sTREM2 [38, 39]. Therefore, CSF PGRN and sTREM2 could be markers of neuroinflammation in the context of tau-related neurodegeneration.

On the other hand, artificial regulation of PGRN or sTREM2 expression levels has been shown to exert protection in AD [40, 41]. Likewise, both our cross-sectional and longitudinal analyses demonstrated that higher CSF levels of PGRN and sTREM2 were associated with reduced Aβ accumulation, aligning with previous findings [5, 18]. PGRN may protect against Aβ accumulation through regulating neuroinflammation, as PGRN-deficient mice showed excessive microglial activation, and increased expression of proinflammatory cytokines [42]. Additionally, PGRN provides neuroprotection in various models of acute and chronic neuronal injury. Specifically, extracellular PGRN can activate the MEK/ERK and PI3K/Akt pathways, offering protection against Aβ toxicity [43]. Furthermore, microglia-derived PGRN demonstrates an anti-amyloidogenic effect in vivo, primarily through enhanced phagocytic activity [5]. Consistent with this, other research also suggested that PGRN haploinsufficiency, along with amyloid beta deposition in APP mouse model [44] and the complete loss of PGRN caused a lysosomal storage disorder accompanied by severe neurodegeneration [45, 46]. CSF sTREM2 reflects the amount of active TREM2 on activated microglia, making it a surrogate marker for TREM2-mediated microglial responses. Several mechanisms may explain the link between higher CSF sTREM2 level and reduced Aβ accumulation. Elevated CSF sTREM2 promotes a shift in microglial phenotype from a homeostatic to a disease-associated state, leading to enhanced Aβ phagocytosis [47]. In contrast, TREM2-deficient mouse models showed decreased microglial proliferation and chemotaxis, resulting in a lower number of microglia surrounding amyloid plaques and a reduced ability to phagocytose amyloid pathology [47, 48]. Furthermore, sTREM2 was found to be localized on the surface of neurons, acting as a mediator of microglia-neuron communication [49]. Thus, the protective roles of PGRN and sTREM2 in early AD could be via enhancing microglial phagocytosis of Aβ and promoting neuroprotective functions.

Interestingly, our findings suggest a mutually mediating effect of PGRN and sTREM2 on their association with CSF Aβ1–42, hinting that CSF PGRN and sTREM2 may potentially work synergistically to alleviate Aβ pathology. However, the exact underlying mechanism remains unclear. Since PGRN deficiency results in microglial hyperactivation and TREM2 activation [10], we therefore hypothesize that CSF PGRN and CSF sTREM2 may be interdependent, working together to prevent excessive microglial activation, which could otherwise impair Aβ phagocytosis and even lead to inflammatory damage. Additionally, lysosome function may play a crucial role in this process. The lysosomal pathway is integral to microglial Aβ clearance [50], and dysfunction in lysosomal activity has been linked to impaired phagocytosis and excessive inflammation [51]. Given that PGRN is involved in lysosomal maintenance [1, 52, 53], it is possible that their interplay may directly influence lysosome function, further modulating microglial responses to Aβ accumulation. Furthermore, our proteomics analyses provided clues of biological mechanisms by which PGRN and sTREM2 acted synergically to modulate Aβ pathology. Functional analyses revealed that the immune system and neuronal plasticity could be important mediating pathways. These findings align with the hypothesis that PGRN and sTREM2 might promote neuroprotective mechanisms by modulating neuroinflammation and supporting neural growth and repair in the context of Aβ accumulation [5, 17, 47]. Together, these results further suggested that the synergistic interplay between PGRN and sTREM2 could contribute to brain homeostasis in immune modulation and neuroprotection, offering a potential strategy to constrain the detrimental effects of Aβ pathology.

However, while our findings hinted that CSF PGRN and sTREM2 might act synergistically to reduce amyloid pathology, other studies have proposed an alternative perspective. For instance, it has been proposed that interactions between Aβ and activated microglia may facilitate the spread of tau pathology [54]. This perspective implies that microglia-related inflammatory activity might play a “double-edged sword effect” in dealing with neurodegeneration [55], with their functions varying according to activation level and disease stage. Therefore, future research should explore the impact of supplementing exogenous PGRN and sTREM2 on both Aβ and tau pathologies at different disease stages to deepen our understanding of their roles in AD progression.

It is important to note that, despite the observed consistent findings across both the CABLE and ADNI cohorts, there are discrepancies in the cutoff values for ATN biomarkers between the two cohorts. These differences in cutoff values may stem from various factors beyond the discrepancies in the methods used to determine the cutoff values. Firstly, the CABLE cohort employed ELISA, while the ADNI cohort utilized ECLIA, which could account for some of the variations in biomarker measurements. Additionally, demographic differences between the cohorts may also play a significant role. The CABLE cohort was younger (average age: 62 years vs. 73 years in ADNI), lower educational levels (CABLE: 9.9 years vs. ADNI: 16.0 years), and fewer women (39.4 vs. 56.2%). Moreover, genetic and ethnic differences between the two cohorts could also be contributing factors. The role of genetic factors, particularly APOE ɛ4, in AD pathogenesis is well-documented, and ethnic variations may affect the metabolism of CSF biomarkers. These ethnic differences in biomarker metabolism, particularly during the early stages of AD, warrant further investigation to better understand their potential impact on cutoff values and disease progression.

Caution is warranted given that several limitations in the present study. First, while mediation analysis was applied, in the complex pathological context of AD, this observational study lacks direct experimental evidence and cannot establish causal relationships. However, the findings provide valuable insights for future research. Well-designed experiments, including those using animal models, are needed to validate the effects of PGRN and sTREM2 on AD pathologies. Second, though the cutoff values in the CABLE and ADNI cohorts have been independently validated [22, 25], the prevalence of AD pathology varies greatly with age and APOE genotype [32]. Therefore, using a single cutoff value may overestimate or underestimate AD pathology, depending on the age distribution and APOE genotype. Third, although the cutoff values for the CABLE cohort were based on Gaussian mixture model analysis and the reliability of these cutoff values was further confirmed by post hoc analysis [22], the study is limited by the small sample size and the lack of PET validation. Fourth, we classified the participants according to the A/T/N framework using CSF biomarkers. While the significance of CSF biomarkers in AD is well-established and a full A/T/N profile can only be obtained with these markers, it is important to note that CSF T-tau might not always reflect neurodegeneration but could be associated with the normal production of tau [56]. Nevertheless, CSF P-tau and CSF T-tau show a strong correlation, and the TN- group exhibited normal levels of both, supporting the reliability of our classification. Other biofluid and neuroimaging markers could also be incorporated into A/T/N classification, and future studies incorporating these biomarkers are necessary to validate our findings.

In conclusion, our study provided evidence supporting the idea that PGRN and sTREM2 may work in concert to modulate amyloid pathology. Future experiments are warranted to test the interactive relationship the interaction between lysosomes and microglia-related neuroinflammation in modulating amyloid and tau pathology.

Availability of data and materials

ADNI data are available upon reasonable request or can be obtained from adni.loni.usc.edu.