Peripheral sTREM2-Related Inflammatory Activity Alterations in Early-Stage Alzheimer’s Disease

Alzheimer’s disease (AD) is a devastating neurodegenerative disorder that impacts 5.8 million Americans and is the sixth leading cause of death in the United States (1). Multiple studies suggest that AD-related pathology, which includes amyloid β (Aβ) plaques and τ neurofibrillary tangles, accumulates years before the onset of dementia symptoms, which delays a clinical diagnosis of AD until after neurodegeneration manifests as cognitive decline and the inability to manage daily living activities (2, 3).

Clinical stages of AD progression are primarily based on clinical signs and symptoms, while a more recent classification method is based on cerebrospinal fluid biomarkers. In the classic symptomatic classification, asymptomatic individuals are classified as “cognitively normal” (CN), often in the absence of testing for cerebrospinal fluid biomarkers and brain pathology. CN individuals are without symptoms of dementia, such as memory loss. The next stage is mild cognitive impairment (MCI), in which cognition has declined to an extent greater than expected for one’s age but without having AD or other diagnoses. This form of MCI thought to progress into AD dementia is referred to as amnestic MCI (4). Finally, a person can progress to AD dementia, of which there are also progressively mild, moderate, and severe forms (5). The ATN continuum is a second method of defining stages of AD dementia and can be based on measures of cerebrospinal fluid biomarker proteins or neuroimaging. ATN refers to “A” for Aβ, “T” for phosphorylated-tau181 (p-Tau181), and “N” for neurodegeneration. Aβ is normally 40 kDa (Aβ40), while a pathogenic form is 42 kDa (Aβ42). The ATN categories can be used to identify different stages of Aβ or τ abnormalities. Individuals without cerebrospinal fluid Aβ or τ abnormalities can be categorized as A−T−N−. The early first stage is A+T−N− if an individual has only altered cerebrospinal fluid Aβ42/Aβ40. The next stage is A+T+N− if an individual has both cerebrospinal fluid Aβ and p-Tau181 pathology. The last, more severe stage is A+T+N+ if an individual has abnormal cerebrospinal fluid Aβ, p-Tau181, and total tau (t-Tau), a biomarker of neurodegeneration, which is more typical in AD than CN individuals (6).

AD is a disease of the brain, but there is evidence that peripheral inflammation is associated with AD (7, 8). Several inflammatory disorders are risk factors for AD, including obesity, metabolic syndrome, traumatic brain injury, and chronic periodontitis development (9). It has been suggested that recurrent inflammatory events over one’s life, including infections, ischemia, and free radical exposure, parallel to the accumulation of Aβ, can increase the risk of AD development because of the repeated cycle of activated immune cells both in the peripheral nervous system and the CNS (10). The mechanism underlying the link between peripheral inflammation and AD is not fully understood.

To better understand the role of peripheral inflammation and AD, studies have measured levels of circulating inflammatory factors. Importantly, there is evidence that immune system dysregulation can also contribute to risk of AD, and it may be the lack of a proper immune system reaction that makes people susceptible to repeated infections and consequently AD. For example, a retrospective review of hospital data found that participants admitted to the hospital with autoimmune disorders were more likely to be diagnosed with AD, and 18 of the 25 autoimmune disorders that were part of the study showed significant correlations with dementia (11). Similarly, age-related decline in immune system function, referred to as immunosenescence, may contribute to risk of AD (12). Meta-analyses of inflammatory factors in people with AD compared with controls often show mixed results, which may be a result of pathological heterogeneity across study groups and confounding factors, including age and apolipoprotein E (APOE) e4 status (13–15).

One immune receptor associated with AD is triggering receptor expressed on myeloid cells 2 (TREM2). Normally, TREM2 functions as a pattern recognition receptor in the innate immune response. It is expressed on microglia in the brain and other myeloid cells in peripheral blood. Several loss-of-function mutations in TREM2 have been shown to increase AD risk (16, 17). The R47H TREM2 mutation results in decreased function of the receptor (16). Another mutation associated with increased AD risk is H157Y. It leads to less full-length cell-surface TREM2 and more of the cleaved version, soluble TREM2 (sTREM2), which is detectible in cerebrospinal fluid and plasma (18, 19). Increased cerebrospinal fluid sTREM2 in MCI and AD has been described in multiple studies, while results from plasma are less clear, and neither cerebrospinal fluid sTREM2 nor plasma sTREM2 is a feasible biomarker given the modest elevation in MCI or AD (20, 21). However, the relationship between elevated sTREM2 and inflammatory activity is unclear and is a critical piece of missing information needed to understand the role of sTREM2 in the broader immune response.

Because there is an association between TREM2 and AD and evidence of peripheral inflammatory factor alterations in AD, but little is known about the relationship between peripheral sTREM2 and peripheral inflammatory activity in AD, the aim of this study was to explore the relationship between plasma sTREM2 and inflammatory activity in AD. The hypothesis was that alterations in plasma sTREM2 are related to altered peripheral inflammatory activity in AD, in a stage-specific manner as defined by MCI or AD pathological stage (ATN category).

Participants defined as CN controls (n = 88), amnestic MCI (n = 37), and AD dementia (AD; n = 75) donated biospecimens under the Lou Ruvo Center for Brain Health Aging and Neurodegenerative Disease Biobank (LRCBH-Biobank) and the Cleveland Alzheimer’s Disease Research Center (CADRC) protocols approved by the Cleveland Clinic Institutional Review Board. Participants were all >55 y of age. Venipunctures were performed after overnight fasting for the collection of whole blood. Plasma was isolated from lavender-top EDTA tubes and stored at −80°C. Cerebrospinal fluid was aliquoted into amber tubes, immediately frozen, and stored at −80°C as previously described (22).

All study participants underwent cognitive testing on enrollment in Cleveland Clinic LRCBH-Biobank or CADRC. The participants were reviewed by neurologists in a formal consensus panel to assign diagnostic groups. Only the groups CN (n = 88), MCI without AD or other diagnoses (n = 37), and AD dementia (n = 75) were included in this study (Table I). A subset of this group had both a blood draw and a lumbar puncture for collection of cerebrospinal fluid for which AD-related biomarkers Aβ40, Aβ42, t-Tau, and p-Tau181 were measured for research studies (CN: n = 33, MCI: n = 36, AD: n = 65) (Table I).

Complete blood counts were performed on n = 49 CN, n = 33 MCI, and n = 60 AD participants via microscopy by trained Cleveland Clinic technicians blinded to sample group. Lumbar punctures were performed for the collection of cerebrospinal fluid from a subset of these participants (CN: n = 13, MCI: n = 22, AD: n = 35) to assess blood–brain barrier integrity (Table I). Blood and cerebrospinal fluid samples were drawn on the same date for each participant to allow for direct comparison of clinical markers and study markers.

Genotyping of APOE was performed from blood samples using the 7500 Real Time PCR System and TaqMan SNP Genotyping Assays (rs429358, rs7412) (Thermo Fisher Scientific) as previously described (23).

Plasma sTREM2 levels were measured using a Luminex 200 3.1 xPONENT System (EMD Millipore, Chicago, IL) and a custom detection method designed to capture the soluble portion of TREM2 protein as previously described (24). The custom-designed plasma sTREM2 assay uses Luminex xMap technology, a reliable and robust bead-based ELISA method (25, 26), which has been published previously by our group (24, 27). Briefly, a capture Ab bound to MagPlex beads binds sTREM2 (R&D: #MAB1828 human TREM2 Ab monoclonal mouse IgG_2B Clone #263602; Immunogen: His19-Ser174). A biotinylated Ab with a SAPE conjugate was used for detection (R&D: #BAF1828; human TREM2 biotinylated Ab; Ag affinity-purified polyclonal goat IgG; Immunogen: His19-Ser174).

A panel of 38 plasma inflammatory factors (cytokines, chemokines, and growth factors) were measured with a human cytokine/chemokine panel using Luminex 200 xMap technology and the MILLIPLEX MAP multiplex kit (Luminex xMAP technology [EMD Millipore, Chicago, IL], kit HCYTMAG60PMX41BK) following the manufacturer’s instructions for analyte detection in human plasma. The Luminex xMAP assays are a bead-based ELISA and have been well validated in the past with documentation of reliability and repeatability (25, 26, 28). This panel was selected because it included both previously studied cytokines and cytokines without previous study with sTREM2 in AD. The inflammatory markers in the panel were epidermal growth factor (EGF), fibroblast growth factor 2 (FGF-2), eotaxin, TGF-α, G-CSF, Flt-3L, GM-CSF, Fractalkine (also known as CX3CL1), IFN-α2, IFN-γ, growth-regulated oncogene (GRO), IL-10, monocyte chemotactic protein-3 (MCP-3, also known as CCL7), IL-12 40 kDa (IL-12p40), macrophage-derived chemokine (MDC), IL-12 70 kDa (IL-12P70), IL-13, IL-15, soluble CD40 ligand (sCD40L), IL-17A, IL-1 receptor agonist (IL-1RA), IL-1α, IL-9, IL-1β, IL-2, IL-3, IL-4, IL-5, IL-6, IL-7, IL-8, IFN-γ–inducible protein 10 kDa (IP-10), MCP-1 (also known as CCL2), macrophage inflammatory protein (MIP) 1α (MIP-1α, also known as CCL3), MIP-1β (also known as CCL4), TNF-α, TNF-β, and vascular endothelial growth factor (VEGF).

Plasma sTREM2 and inflammatory markers were measured in plasma on plates that also contained buffer-alone background wells and serial dilutions of analytes to generate a standard curve. Plates measuring inflammatory markers also contained quality-control proteins. Each plate was evaluated to ensure that the standards and quality control ran in the expected ranges. Samples from different disease groups were run across each plate to control for batch effects. Mean fluorescence intensities (MFIs) of analytes were calculated and averaged between duplicate samples. Our analyses compared MFI rather than calculated concentration because many participants had low levels of cytokines, which were undetectable by the standard curve interpolation. Use of MFI values allowed inclusion of all participants, including the samples at the low range of detection, to avoid biasing the results toward participants with higher values (29–31).

Cerebrospinal fluid Aβ40, Aβ42, t-Tau, and p-Tau181 were measured according to the manufacturer’s specifications (Luminex xMAP technology [EMD Millipore]: HNABTMAG-68K), modified by a 1:10 dilution of cerebrospinal fluid. Each Aβ and tau kit comes with an Aβ and tau standard, as well as Aβ and tau quality controls. The kit provides the expected concentrations of each working standard, as well as each of the quality controls. The standards, controls, and cerebrospinal fluid samples are all run in duplicate. If the coefficient of variation for any of the replicate wells is >25% or if both replicate wells have a bead count of <35 beads for a given analyte, the results are repeated or excluded for that analyte.

Cerebrospinal fluid measurements of Aβ40, Aβ42, t-Tau, and p-Tau181 were performed as described earlier. Participants were categorized into ATN groups based on the following criteria: A+ indicates participants were in the lower 10th percentile of all Aβ42/Aβ40 ratios within the CN group in the cohort, T+ indicates participants were in the upper 90th percentile of all p-Tau181 protein concentrations within the CN group in the cohort, and N+ indicates participants were in the upper 90th percentile of t-Tau protein concentrations within the CN group in the cohort. Cutoff values generated by percentile analysis were as follows: A+ ≤ Aβ42/Aβ40 0.139; T+ p-Tau181 ≥ 65.5 pg/ml; and N+ t-Tau ≥ 695.9 pg/ml. These cutoff values were tested in contingency tables with numbers of CN and AD participants. The cutoffs were selected because they had the highest sensitivity and specificity in our cohort, compared with cutoffs set at 20th/80th percentiles, 15th/85th percentiles, and 5th/95th percentiles (for Aβ42/Aβ40, p-Tau181, and t-Tau, respectively), as well as concentrations previously published (32–34). Using the 10th/90th percentiles and cutoffs listed earlier, our cutoffs had the following values: Aβ42/Aβ40, sensitivity = 0.963, specificity = 0.901 (p < 0.001 by Fisher exact test); p-Tau181, sensitivity = 0.722, specificity = 0.887 (p < 0.001 by Fisher exact test); and t-Tau, sensitivity = 0.426, specificity = 0.811 (p = 0.0266 by Fisher exact test).

We characterized participants into sTREM2 tertiles to compare groups with the highest versus lowest levels of plasma sTREM2. Levels of sTREM2 from participants were sorted from lowest to highest, regardless of disease status and ATN group, and the results were divided into three groups. Tertile 1 (n = 67) refers to the lowest sTREM2 levels, and tertile 3 (n = 67) refers to the highest sTREM2 levels. Tertile 2 (n = 66) was not included in the tertile analyses. Tertiles were used rather than quartiles or quintiles to maximize the sample size while still analyzing two distinct groups of sTREM2 levels.

CN, MCI, and AD demographics, plasma inflammatory marker, and plasma sTREM2 were compared between groups using Mann–Whitney, without and with outlier removal using the ROUT method (Q = 1%). ATN groups were compared by Mann–Whitney tests along the ATN continuum (A−T−N− versus A+T−N−, A+T−N− versus A+T+N−, and A+T+N− versus A+T+N), without and with outlier removal using the ROUT method (Q = 1%). The ROUT method using Q = 1% indicates there is a 1% chance of incorrectly identifying outliers (false discovery rate). This method is appropriate for our relatively small sample size dataset and has been published previously in both cytokine and AD research (35, 36). Sensitivity and specificity of ATN group cutoffs were analyzed by Fisher exact tests, and contingency tables of sTREM2 tertiles in disease groups were analyzed by χ² tests. These data were analyzed, and violin plots and bar graphs were generated using GraphPad Prism version 9.0.2 (GraphPad Software, San Diego, CA). Using R (R version 3.6.1), the “lm” function was used to create a single linear regression model for each cytokine. The standardized β coefficient was then obtained by applying the “standardize_parameters” function from the effectsize package to these models (37). The “cor” function was used to determine Spearman correlations. Correlograms were made using the resulting Spearman correlation matrix with the “corrplot” function in the corrplot package (38) and arranged according to hierarchical clustering using Ward’s method. The p values were added using results from the “rcorr” function from the “Hmisc” function as a parameter in the “corrplot” function (39). To compare the graphs of the correlation matrix, we applied the Steiger test to the correlation matrices with the “cortest.normal” function from the psych package (40). A power analysis using analyte data from a small pilot study indicated a minimum sample size of n = 25 per group was necessary to achieve a significance of 0.05 and a power of 0.8 for these analyses.

A coexpression network was generated using data without outlier removal for each group based on the Spearman rank correlation coefficient ρ. For inflammatory factors, Spearman r > 0.8 and p < 0.05 was used as cutoff for coexpressions; for sTREM2 versus other inflammatory factors, Spearman r > 0 and p < 0.05 was used as a cutoff. The coexpression networks were visualized by Gephi 0.9.2 (41).

Participants consented under the LRCBH-Biobank and the CADRC protocols approved by the Cleveland Clinic Institutional Review Board.

Characteristics of the study populations were compared between CN controls (n = 88), individuals with MCI (n = 37), and participants with AD (n = 75). There were no statistically significant differences in age between groups by Kruskal–Wallis with Dunn’s test for multiple comparisons. There were also no differences in distributions of male/female sex between groups by Fisher exact test. There were significant differences in APOE4⁺ allele status between groups (Table I).

Peripheral blood cell counts were assessed in a subset of participants with available measures from Cleveland Clinic clinical chemistry. Participants with AD had significantly higher WBC counts compared with MCI (p = 0.0239) and significantly lower basophil percentages compared with CN (p = 0.0309). Participants with MCI had significantly lower erythrocyte sedimentation rates compared with CN (p = 0.0188). Participants did not vary by percentages of neutrophils, lymphocytes, monocytes, levels of C-reactive protein, or blood–brain barrier integrity (ratio of cerebrospinal fluid albumin/serum albumin) (Table I).

A subgroup of the cohort had cerebrospinal fluid measurements of Aβ42/40, p-Tau181, and t-Tau (CN: n = 33, MCI: n = 36, AD: n = 65). Significant differences were observed among all three disease groups in levels of AD-related biomarkers, Aβ42/40 and p-Tau181. In addition, both CN and MCI groups had significantly lower levels of t-Tau compared with the AD group (Table I).

Levels of sTREM2 were significantly higher in MCI compared with AD with (Fig. 1A) and without outlier removal (Table II). Immunoregulatory/pleiotropic cytokines, fractalkine and IL-9, had significantly higher levels in MCI, compared with AD, after outlier removal (Fig. 1B). Anti-inflammatory factors, FGF-2, IL-1RA, IL-5, and IL-13, had significantly higher levels in MCI compared with AD after outlier removal (Fig. 1C). Proinflammatory factors, EGF, GM-CSF, IFN-a2, IL-1a, IL-1B, IL-2, IL-17A, IP-10, and VEGF, had significantly higher levels in MCI compared with AD after outlier removal (Fig. 1D). Two inflammatory factors, Flt-3L and IL-2, were higher in controls compared with AD, and two, MIP-1B and IL-1RA, were higher in controls compared with MCI (Fig. 1D). Results from Mann–Whitney analyses with or without outlier removal and the number of outliers removed are shown in Table II.

To determine whether individuals with elevated sTREM2 also exhibit elevated levels of other inflammatory factors, we stratified the cohort by sTREM2 tertiles. This was done by ordering sTREM2 MFI levels from smallest to largest and evenly dividing them into three groups, regardless of disease status. Tertile 1 refers to the lowest sTREM2 levels, and tertile 3 refers to the highest sTREM2 levels. Levels of cytokines between tertiles 1 and 3 with and without outlier removal exhibited significant differences between tertiles for 32 of the 38 inflammatory factors measured (on outlier removal) (Fig. 2, Table III). The cytokines that were significantly higher in tertile 3 were immunoregulatory/pleiotropic cytokines fractalkine, IL-6, IL-7, IL-9, and TGF-α (Fig. 2B, Table III); anti-inflammatory cytokines FGF-2, G-CSF, IL-4, IL-5, IL-10, and IL-13 (Fig. 2C, Table III); and proinflammatory cytokines EGF, GM-CSF, GRO, IFN-α2, IFN-γ, IL-1α, IL-1β, IL-2, IL-3, IL-8, IL-12p40, IL-12p70, IL-15, IL-17A, IP-10, MIP-1α, MCP-3, sCD40L, TNF-α, TNF-β, and VEGF (Fig. 2D, Table III). Results from Mann–Whitney analyses with or without outlier removal and the number of outliers removed are shown in Table III.

Correlation matrices were generated to characterize relationships between inflammatory factors in sTREM2 tertile 1 (low) (Fig. 3A) compared with tertile 3 (high) (Fig. 3B). The plots shown are without outlier removal. Outlier removal was not possible for correlation matrix analyses because of limitations in the software package because missing values resulted in the elimination of participants. The correlation patterns were very different between tertiles 1 and 3, shown visually by a color gradient of positive (blue) to negative (red) correlations, with white showing no correlation (Spearman r = 0). Overall, tertile 1 had fewer significant correlations than tertile 3, including between sTREM2 and other inflammatory factors. Correlation matrices were compared by Steiger test and found to be significantly different (p < 0.0001) (Fig. 3). In a separate analysis that allowed for the removal of outliers, after outlier removal sTREM2 was correlated with each cytokine within tertiles. Tertile 1 (low sTREM2) did not significantly correlate with any of the 38 cytokines after outlier removal, while tertile 3 (high sTREM2) had significantly positive correlations with 24 cytokines: Fractalkine, IL-6, IL-7, IL-9, FGF-2, G-CSF, IL-4, IL-5, EGF, GM-CSF, IFN-α2, IFN-γ, IL-1α, IL-1β, IL-2, IL-3, IL-8, IL-12p40, IL-15, IL-17A, MIP-1α, sCD40L, TNF-β, and VEGF (Spearman correlation with and without outlier removal shown in Supplemental Table I).

There were significantly higher proportions of CN in tertile 3 versus MCI and MCI in tertile 3 versus AD, but not CN versus AD by Fisher exact test, indicating CN and MCI represented a majority of the high sTREM2 values (Fig. 4). When we compared the numbers of individuals in each disease group by sTREM2 tertiles, approximately half of the CN group was in tertile 3 and the other half in tertile 1 (Fig. 4A). In contrast, a greater proportion of MCI was in tertile 3 compared with tertile 1, while MCI numbers were lower than CN in both tertiles (Fig. 4A). MCI numbers were lower in tertile 1, but not in tertile 3, compared with AD (Fig. 4B). Fewer AD were in tertile 3 compared with tertile 1, while more MCI were in tertile 3 compared with tertile 1, and MCI had fewer than AD in tertile 1 (Fig. 4B). There was not a significant distribution difference when comparing AD and CN by tertile (Fig. 4C).

Cutoff values of AD-related biomarkers were generated by percentile analysis (A+ ≤ Aβ42/Aβ40 0.139; T+ p-Tau181 ≥ 65.5 pg/ml; N+ t-Tau ≥ 695.9 pg/ml; see Materials and Methods). The within CN, MCI, and AD distribution of designated ATN categories according to these cutoff values are exhibited in a pie chart (Fig. 5). Characteristics of the ATN continuum categories, A−T−N− (n = 31), A+T−N− (n = 17), A+T+N− (n = 37), and A+T+N+ (n = 39), indicate no significant differences in age, sex, or APOE4 status (Table IV). There were 71 participants who could not be defined by ATN group because of the lack of cerebrospinal fluid material available. There were no statistically significant differences in age (by Mann–Whitney tests) or sex (by Fisher exact test) between groups (Table IV). People in other ATN groups (n = 10), such as A−T+N−, A−T−N+, and A−T+N+, were not included in further analyses (Fig. 5). On stratification by ATN category, increased neutrophil counts in A+T−N− compared with A−T−N− and higher WBC counts in A+T−N− compared with both A−T−N− and A+T+N− were observed (Table IV).

Levels of sTREM2 and inflammatory factors were compared between ATN categories. To focus on designations related to AD stage (6), we analyzed only participants defining the AD-related ATN continuum: A−T−N−, A+T−N−, A+T+N, or A+T+N+ (124 participants of 200 in the total cohort). Groups were compared by Mann–Whitney tests after outlier removal (A−T−N− versus A+T−N−, A+T−N− versus A+T+N−, and A+T+N− versus A+T+N+). Levels of sTREM2 did not significantly differ between ATN groups after outlier removal (Fig. 6A) or without outlier removal (Table V). Levels of immunoregulatory/pleiotropic cytokines fractalkine and IL-7 were significantly higher in the A+T−N− group compared with the A−T−N− and A+T+N− groups after outlier removal (Fig. 6B). The A+T+N+ group had significantly lower levels of anti-inflammatory cytokines FGF-2 and IL-5 compared with the A+T+N− group after outlier removal (Fig. 6C). Similarly, levels of proinflammatory cytokines GM-CSF, GRO, IFN-γ, IL-1β, IL-2, IL-3, IL-8, IL-17A, and MCP-3 were significantly lower in A+T+N+ compared with A+T+N− after outlier removal (Fig. 6D). In addition, higher levels of MCP-3 were observed in A+T−N− compared with A−T−N− after outlier removal (Fig. 6D). Results from Mann–Whitney analyses with or without outlier removal and the number of outliers removed are in Table V.

Correlation matrices were generated to characterize relationships between inflammatory factors by disease status (Fig. 7A) or ATN status (Fig. 7B) on all 39 inflammatory factors (sTREM2 and 38 cytokines). Outlier removal was not possible for correlation matrix analyses because of limitations in the software package because missing values resulted in the elimination of participants. A Steiger test to compare correlation landscapes indicates that the CN group was significantly different from both the MCI and AD groups (p < 0.0001) (Fig. 7A), while there was no difference between ATN groups (Fig. 7B).

CN showed significant positive correlations between sTREM2 and 34 of the 38 inflammatory factors (the exceptions were eotaxin, MDC, MCP-1, and MIP-1β) (Fig. 7A). After outlier removal, Fractalkine, IL-7, TGF-α, FGF-2, G-CSF, IL-10, EGF, GM-CSF, IFN-α2, IFN-γ, IL-2, IL-17A, IP-10, MIP-1α, sCD40L, TNF-α, and VEGF remained significant (Supplemental Table II). This was in contrast with the MCI group, in which sTREM2 significantly correlated with only IL-10, EGF, IL-1α, and IL-1β before outlier removal (Fig. 7A), and none remained significant after outlier removal (Supplemental Table II). In the AD group, sTREM2 significantly correlated with 12 of the 38 inflammatory factors, IL-6, IL-7, IL-10, IL-13, EGF, IFN-γ, IL-1β, IL-3, MCP-3, VEGF, TNF-β, and MIP-1α (Fig. 7A), 5 of which remained significant after outlier removal, IL-10, EGF, IL-1β, IL-3, and VEGF.

The A−T−N− group had significant positive correlations between sTREM2 and multiple inflammatory factors, including Fractalkine, IL-6, IL-7, IL-9, TGF-α, FGF-2, IL-4, IL-5, IL-10, IL-13, EGF, GM-CSF, GRO, IFN-α2, IFN-γ, IL-1α, IL-1β, IL-2, IL-8, IL-12p40, IL-12p70, IL-17A, MCP-3, TNF-β, and VEGF (Fig. 7B). Only two of these inflammatory factors remained significant (IL-5, EGF), and two others gained significance (MDC, IP-10) after outlier removal (Supplemental Table III). In the A+T−N− group there were no significant correlations between sTREM2 and inflammatory factors (Fig. 7B) unless outliers, IL-4 and IL12p70, were removed (Supplemental Table III). The A+T+N− group exhibited significant correlations between sTREM2 and IL-6, IL-9, IL-4, IL-10, IL-13, EGF, IL-1α, IL-1β, MCP-3, and TNF-β (Fig. 7B), while after outlier removal TGF-α gained significance, and IL-10, EGF, and IL-1β remained significant (Supplemental Table III). A+T+N+ showed positive correlations between sTREM2 and IL-6, IL-7, TGF-α, FGF-2, G-CSF, IL-1RA, EGF, GM-CSF, GRO, IFN-α2, IFN-γ, IL-1β, IL-2, IL-3, IL-12p70, IL-17A, TNF-β, and VEGF (Fig. 7B), while after outlier removal only IL-7, FGF-2, GM-CSF, GRO, IFN-α2, IL-1β, and IL-3 remained significant (Supplemental Table III).

To estimate the complexity of the relationship between inflammatory factors, we performed a network analysis (without outlier removal) (Fig. 8). The CN group exhibited a large sTREM2 node at the center of the module representing the most connections to other cytokines with an overall modularity of 0.442 and four communities (each community shown as a different color) (Fig. 8A). In contrast, the MCI group showed a very sparse network where only four cytokines connected to sTREM2, and the modularity (0.548) and number of communities were increased (five communities). The AD group exhibited the lowest modularity with more cytokines connected to sTREM2 and four communities. The MCI and AD groups also showed networks of inflammatory factors that were not connected to sTREM2, while the CN group did not (Fig. 8A). The A−T−N− network analysis showed a large sTREM2 node and a modularity of 0.408 with four communities (Fig. 8B). In contrast, the A+T−N− group exhibited no sTREM2 connections, and the modularity was increased (0.557) and the number of communities increased to six (Fig. 8B). A+T+N− and A+T+N+ both had lower modularity than A−T−N− and A+T−N−, while A+T+N− showed networks of inflammatory factors that were not connected to sTREM2 (Fig. 8B).

The purpose of this study was to explore the relationship between plasma sTREM2 and inflammatory activity in AD. We sought to accomplish this by comparing plasma sTREM2 and plasma inflammatory factors across clinical or ATN categories.

Multiple cytokines exhibited a decrease in AD compared with MCI, which is in contrast with previous studies that have observed no difference or an increase in AD (14, 21), suggesting that heterogeneity in pathological stage across cohorts could contribute to the discrepancy across studies. For example, our cohort was assembled from biobanked samples collected in a dementia clinic and is not representative of the general population, including the CN group. To address the question of whether our CN and MCI subcohorts had underlying early-stage pathology, we assessed a subcohort with cerebrospinal fluid ATN measures. Indeed, the CN group included individuals with underlying neurological pathologies as indicated by their ATN status. Therefore, to address the question of whether heterogeneity in AD-related ATN pathology is associated with a decrease in cytokine levels, we stratified the cohort by ATN status. Three cytokines, Fractalkine, IL-7, and MCP-3, were significantly elevated in the A+T−N− stage compared with A−T−N−, suggesting that the A+ stage is a critical time point in the immune response in AD. Fractalkine can act as a chemoattractant for leukocytes and has been described as elevated in MCI (42). MCP-3 is also a chemoattractant for leukocytes and has been described as either elevated in AD or without a significant difference in AD (14). IL-7 regulates the development and homeostasis of immune cells, including B and T cells, and has been somewhat inconsistently associated with AD (8, 43, 44). Together, this suggests that cell recruitment and development may play an important role in early-stage AD. Multiple other cytokines exhibited decreased levels in the later A+T+N− or A+T+N+ stages of AD, suggesting that there is potential peripheral immunosenescence in the later A+T+N− or A+T+N+ stages of AD. This idea is supported by previous reports of age-related decline in immune function and AD (45, 46).

Comparisons between sTREM2 tertiles regardless of disease or ATN status indicates that when plasma sTREM2 is elevated, many other inflammatory factors in the circulation are also elevated. Tertile 3 (high sTREM2) had significantly higher levels of 31 cytokines, while tertile 1 did not have higher levels of any cytokine. In addition, inflammatory activity matrices were significantly different between tertiles, and MCI is disproportionate in tertile 3, while AD is in tertile 1, further implicating TREM2-related inflammatory activity as playing a key role in early-stage AD, while this role changes later in the disease. An early-stage elevation of cerebrospinal fluid sTREM2 has been associated with protection against cognitive decline (47, 48). In contrast, cerebrospinal fluid sTREM2 is associated with cognitive decline in the later stages of AD (47, 48). In this study, we observed elevation of multiple inflammatory factors in the peripheral circulation that corresponds with elevated plasma sTREM2 in the MCI stage. Together, this implicates alterations in TREM2 function in the early stages of cognitive decline as a critical player in a broader, potentially systemic immune response in AD.

An association between sTREM2 and inflammatory factors was observed that was different in the CN and A−T−N− groups compared with the other groups defined by AD symptoms (MCI and AD) or AD-related cerebrospinal fluid biomarker categories (A+T−N−, A+T+N−, and A+T+N+). Overall, patterns of inflammatory activity and the relationship with sTREM2 were drastically altered in early-stage AD (MCI and A+T−N− or A+T+N−) and remain altered in the later stages of AD (AD and A+T+N+), although less so, suggesting that dysfunctional peripheral TREM2-related inflammatory activity plays a critical role early in disease progression, especially at the MCI and A+T−N− stage. Although increased expression of TREM2 in peripheral cells from MCI patients has been described previously (49, 50), peripheral sTREM2 has been described as either only modestly elevated (20, 21) or as not significantly elevated in AD (51). In contrast with previous findings, we observed a significant difference in plasma sTREM2 between MCI and AD, where levels were decreased in AD and a nonsignificant elevation was observed in MCI compared with CN. An association between elevated plasma sTREM2 and cognitive impairment after acute ischemic stroke and all-cause dementia has been described (52–54), suggesting that underlying heterogeneity may contribute to observed differences across study cohorts. Higher cerebrospinal fluid sTREM2 has been described as attenuating APOE4-related risk for cognitive decline and neurodegeneration, suggesting that CNS early elevation of TREM2 plays a critical role in AD progression (55). In addition, studies in mouse models suggest that TREM2 function impedes τ seeding in neuritic plaques (56) and potentially restrains the enhancement of τ accumulation and neurodegeneration initiated by β-amyloid pathology in the brain (57). In our study, after stratification by cerebrospinal fluid ATN categories, plasma sTREM2 did not differ by ATN, implicating other pathologies rather than amyloid and τ as contributing to higher sTREM2 levels in MCI. However, in support of a possible role of dysfunctional sTREM2 in early ATN, the correlation and network analyses show a complete lack of correlation with other inflammatory factors in the A+T−N− stage, suggesting that early-stage AD is a critical time point.

Interestingly, there were instances where there was not a relationship between sTREM2 and inflammatory factors. The four inflammatory factors that did not correlate significantly with sTREM2 in the CN group (eotaxin, MDC, MIP-1β, and MCP-1) did not significantly correlate in the MCI and AD group either and were not different between sTREM2 tertiles, suggesting that those inflammatory factors are not associated with sTREM2 in this cohort even though previous evidence suggests they are altered in AD (58, 59). Interestingly, MDC did correlate with sTREM2 in the A−T−N− group, suggesting that in individuals negative for AD-related biomarkers there is a relationship between MDC and sTREM2.

The results after outlier removal (summarized in (Fig. 9, Supplemental Tables I–III) indicate that three cytokines stand out as potentially important in AD because they had both significantly different levels between groups (both disease and ATN groups) and a significant relationship with sTREM2 in at least one group, implicating a role in a systemic immune response in AD. These inflammatory factors were FGF-2, GM-CSF, and IL-1β. Fractalkine also stood out because it was increased in the early A+T−N− stage and decreased in the late stages of AD while also correlated with sTREM2 in CN individuals, implicating an early alteration in the normal relationship between sTREM2 and Fractalkine (Fig. 9).

FGF-2 levels were decreased in AD and in the A+T+N+ individuals. It was positively correlated with sTREM2 only in CN and A+T+N+. When sTREM2 is elevated, FGF-2 is also elevated (Fig. 9). FGF-2, also known as basic FGF, is a growth factor and signaling protein that binds to and exerts effects via specific FGF receptors. It is involved in various cellular functions regulating proliferation, differentiation, survival, and migration (60, 61). FGF-2 stimulates pericyte proliferation and production of proteinases from endothelial cells, which can locally degrade the extracellular matrix, allowing cells to migrate for the formation of the new vessels (61–63). It induces formation of tubular structures in endothelial cell cultures in vitro and promotes angiogenic responses and vascular regeneration in vivo (60, 63, 64). Brain endothelial cells form the blood–brain barrier and tightly regulate the transport between the brain and the periphery (65), and it has been suggested that aberrant angiogenesis and senescence of the cerebrovascular system initiates neurovascular uncoupling, vessel regression, brain hypoperfusion, and neurovascular inflammation in AD (65). This previous evidence and our findings suggest that a decline in sTREM2-related FGF-2 levels may be indicative of blood–brain barrier changes in late-stage AD. To our knowledge, a link between peripheral sTREM2 and FGF-2 is a novel finding that is supported only by our previous report that indicates that, in a human microglial cell line treated with Aβ42, TREM2 overexpression inhibits FGF-2 (66).

GM-CSF levels were decreased in AD and in the A+T+N− individuals. They were also positively correlated with sTREM2 in CN individuals, but not in MCI or AD, although they were correlated in A+T+N+ individuals. When sTREM2 is elevated, GM-CSF is also elevated in a pattern of early increase and late decrease in levels (Fig. 9). Interestingly, in the network analysis, GM-CSF is normally part of a community of other inflammatory factors associated with sTREM2 levels that include IFN-γ, IL-17A, VEGF, IL-7, Fractalkine, and IFNa-2, while in the early stages of AD this association with sTREM2 is lost even though some of the relationships with other factors remain. GM-CSF is a cytokine that regulates the numbers and function of cells of macrophage-monocyte lineage (67). It is a product of cells activated during inflammatory or pathological conditions and is secreted by macrophages, T cells, mast cells, NK cells, endothelial cells, and fibroblasts (67). In mouse models of AD, GM-CSF inhibition reduces brain amyloidosis, increases plasma Aβ, and rescues cognitive impairment (68, 69). One study suggests that GM-CSF improves cognition in cancer patients (70). Together, this suggests that in AD there may be a sTREM2-related decrease in GM-CSF that is indicative of an alteration in the normal function of peripheral cells of macrophage-monocyte lineage. To our knowledge, a relationship between sTREM2 and GM-CSF is novel information. This implicates an alteration in the systemic inflammatory response in AD involving sTREM2-related GM-CSF that warrants further investigation.

IL-1β was lower in AD compared with MCI and in A+T+N+ compared with A+T+N−. In addition, IL-1β correlated with sTREM2 in AD and in both A+T+N− and A+T+N+ groups (Fig. 9). In the network analysis, IL-1β was normally part of the same community of inflammatory factors as FGF-2 and TREM2. This community relationship remained in MCI and A+T−N−, while not including sTREM2 in the A+T−N− group, and was lost in AD and A+T+N− or A+T+N+ groups. In contrast with our findings of lower levels of IL-1β in AD than in MCI, previous reports have shown either an increase or no difference in peripheral levels from patients with AD (8, 14, 71), suggesting that heterogeneity in pathological stage across cohorts could contribute to the discrepancy across studies. In support of this idea, there was a slight, although nonsignificant increase in MCI and in the early ATN groups in our cohort with the decrease only in late-stage AD evident in the A+T+N+ group implicating a very late-stage decrease in IL-1β in AD. IL-1β overexpression is associated with reduced amyloid plaques and levels of Aβ42 and Aβ40 in a mouse model of AD (72). Interestingly, IL-1β downregulates expression of TREM2 mRNA in cultures of human peripheral blood monocytes and synovial fluid macrophages from patients with rheumatoid arthritis (73). Together, this suggests that the positive correlation between IL-1β and sTREM2, as well as decreased levels of IL-1β in later stages of AD observed in our study, is not likely related to IL-1β–induced downregulation of peripheral TREM2. However, further study is necessary to fully understand the nature of the relationship between TREM2 and IL-1β in AD.

Fractalkine was significantly higher in MCI than AD and significantly higher in A+T−N− compared with both A−T−N− and A+T+N−, suggesting that this cytokine may increase early in the A+ stage and later decrease as the disease progresses (Fig. 9). Fractalkine was significantly correlated with sTREM2 only in the CN (Fig. 9), suggesting that normally when sTREM2 is elevated, fractalkine is also elevated. Interestingly, in the network analysis, Fractalkine is normally (in CN and A−T−N−) part of the same community as GM-CSF, but not in AD or A+T−N− or A+T+N−, while this relationship is regained in the A+T+N+ stage. This may suggest that this community of inflammatory factors is critically disrupted in early AD progression. This idea is supported by previous evidence that describes fractalkine as an immunoregulatory cytokine that reduces inflammatory signaling in activated microglia and reduces τ pathology when overexpressed (74–79). In addition, both fractalkine and TREM2 are cleaved by the same enzyme, a disintegrin and metalloproteinase domain-containing protein 10, which may link the downstream function of the two proteins (18, 19, 80), suggesting important links between Fractalkine and sTREM2.

A limitation of this exploratory study is small sample size, especially for the ATN groups. Although we did gather evidence for multiple cytokines, our findings will need to be tested in a larger cohort to validate the correlations observed. Even though a preliminary power analysis suggested a sample size of 25 per group would achieve the desired power, the data should be approached with caution because there is a potential for false-negative results for some of the analytes with detectable levels in the very low range. Another limitation is that some important inflammatory factors may have been missed because the 38 inflammatory factors available on the inflammatory panel used is not comprehensive of the multiple immune factors present in the circulation. Outliers were removed using a standard method of ROUT with Q = 1%, indicating a false discovery rate of outliers was, at most, 1%. This method has been published previously by other groups (35, 36). However, it does have the risk of eliminating true data points that are outside of the expected range that may contain biological significance. Follow-up studies of characteristics of participants with very high and low levels of cytokines could add further knowledge to the underlying biological mechanisms. In addition, another limitation is that we used Aβ42/Aβ40 for “A”, p-Tau181 for “T,” and t-Tau for “N” as in other reports (81). Others have argued that only A and T groups should be used or that imaging or neurofilament light should be used for N (82, 83). As with any test that uses a cutoff value for positivity, there is likely a gray area of overlap between the groups that could not be completely separated. Furthermore, given the identified alterations in basophil levels and erythrocyte sedimentation rate in AD, the results should be approached with caution because immune factor levels could have been altered because of factors such as undiagnosed inflammatory illnesses or impending infections. Future studies will benefit from increased sample size, more diverse inflammatory panels, and single-cell–based analyses to further understand the underlying source of these results. Importantly, longitudinal studies will help tease apart these inflammatory effects on disease progression.

In this exploratory study, we identified patterns of sTREM2-related inflammatory activity that differed by clinical diagnosis and ATN category. Plasma sTREM2 was linked to inflammatory activity in the peripheral circulation, where strong connections and patterns were observed in the groups without AD symptoms or cerebrospinal fluid ATN biomarkers (CN and A−T−N−). In addition, they were profoundly altered in the MCI and A+T−N− stages, suggesting that the pathological amyloid stage, before the pathological τ stage, may be the critical stage for peripheral immune system intervention in AD. Notable inflammatory factors that had both a significant relationship with sTREM2 and significantly different levels across groups were GM-CSF, FGF-2, IL-1β, and Fractalkine. This study lays the groundwork for research and therapeutic strategies that seek to understand or target TREM2-related inflammatory activity in AD. Together, the findings suggest that depending on AD stage, therapeutically targeting TREM2 may broadly influence inflammatory activity.