Early increase of CSF sTREM2 in Alzheimer’s disease is associated with tau related-neurodegeneration but not with amyloid-β pathology

This is a cross-sectional study in which CSF sTREM2 was measured in 1031 participants of the ADNI project. Among them, 4 individuals did not have the AD core biomarkers measurements and were further excluded from the analysis, rendering a study sample of 1027 subjects. The CSF sTREM2 measurements were uploaded to the ADNI database (http://​adni.​loni.​usc.​edu) on 16/03/2018 and the data used in this study was downloaded on 21/03/2018. The ADNI project (http://​www.​loni.​usc.​edu/) is a multicenter longitudinal study led by Principal Investigator Michael W Weiner with the main goal to develop and validate biomarkers for subject selection and as surrogate outcome measures in late-onset AD [30]. The institutional review boards (IRB) of all participating centers approved the procedures of the study and all participants or surrogates provided informed consent. Our local IRB (LMU) also approved the study.

In line with the recently published 2018 NIA-AA “research framework” for the diagnosis of Alzheimer’s disease [28], we assigned each ADNI participant in a group defined by its biomarker profile, as described by the A/T/N scheme [27], coupled with its cognitive status, as defined by the CDR score [31]. The A/T/N scheme comprises 3 biomarker groups: “A” refers to aggregated Aβ, “T” aggregated tau and “N” to neurodegeneration. Each biomarker group is binarizied in negative (-) or positive (+) based on whether their biomarkers are normal or abnormal. In the present study, we assigned “A+” to those individuals that had a CSF Aβ_1-42 < 976.6 pg/ml, “T+” to those individuals with P-tau_181P > 21.8 pg/ml and “N+” to those individuals with T-tau > 245 pg/ml. We merged the aggregated tau (T) and neurodegeneration (N) groups in order to decrease the number of groups to be compared. 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-tau_181P ≤ 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 (T+ or N+, that is P-tau_181P > 21.8 pg/ml or T-tau > 245 pg/ml). Only 5.4% of the individuals of the total differed between the T and N biomarkers groups.

The combination of the biomarker profile (A/T/N scheme) and the clinical status (CDR) rendered 12 different groups that are displayed in Table 1. We studied CSF sTREM2 in ADNI following two approaches. In a first one, we compared CSF sTREM2 between the different A/T/N categories within each clinical stage. In a second one, we attempted to model the course of AD with biomarker and clinical-based groups, similar to what was proposed by the 2011 NIA-AA criteria [32‐34]. Thus, in this second approach, we compared the ‘CDR = 0 A-/TN-’ group (which corresponds to healthy controls) with those biomarker-based groups that fall into the Alzheimer’s continuum category, that is: ‘Preclinical AD A+/TN-’, ‘Preclinical AD A+/TN+’, ‘AD CDR = 0.5’ and ‘AD CDR = 1’. Since our aim was to study the Alzheimer's continuum, we excluded from this analysis those individuals that fall in the category of suspected non-Alzheimer’s pathology (SNAP) [35‐38] and those symptomatic individuals (CDR > 0) that do not have positive biomarkers for both Aβ deposition (CSF Aβ_1-42) and neurodegeneration/tau pathology (T-tau or P-tau_181P).

In the present study, we used the AD CSF core biomarkers measurements performed with the Elecsys® total-tau CSF, the Elecsys® phosphotau(181P) CSF and the Elecsys® β-amyloid(1–42) CSF immunoassays on a cobas e 601 instrument [39, 40]. The data is available in the ‘UPENNBIOMK9.csv’ file in the ADNI database. These immunoassays are for investigational use only. They are currently under development by Roche Diagnostics and not commercially available yet. The analyte measuring ranges (lower technical limit to upper technical limit) of these assays are the following: 80 to 1300 pg/ml for total-Tau CSF, 8 to 120 pg/ml for phosphor-Tau(181P) CSF, and 200 to 1700 pg/ml for Elecsys® β-Amyloid(1-42) CSF immunoassays. The measuring range of the Elecsys® β-Amyloid(1-42) CSF immunoassay beyond the upper technical limit has not been formally established. Therefore use of values above the upper technical limit, which are provided based on an extrapolation of the calibration curve, is restricted to exploratory research purposes and is excluded for clinical decision making or for the derivation of medical decision points.

CSF sTREM2 measurements were done with a MSD platform-based assay, previously reported and validated [2, 25, 26]. A comprehensive description of the assay is shown in Supplementary methods (see Additional file 1: Supplementary methods). The CSF sTREM2 measurements are publicly available in the ADNI database.

HEK293T cells were cultured in DMEM with Glutamax I supplemented with 10 % (v/v) fetal calf serum (FCS) and 1 % penicillin/streptomycin. 24 hours after seeding, cells were transiently transfected with equal amounts of DNA coding for the different TREM2 variants using Lipofectamine 2000 as the transfection reagent. TREM2 variant constructs were generated by site-directed mutagenesis (Stratagene) of a TREM2 wt construct bearing N- and C-terminal HA and FLAG tags, respectively, as previously described [2]. All constructs were verified by DNA sequencing (GATC Biotech). We collected the conditioned media 48 hours after transfection. Cellular debris was removed by centrifugation at 4°C (13300 rpm, 20 min). Supernatants were subsequently frozen at -20°C until analyses were performed. Cell culture reagents were purchased from Thermo Fisher Scientific unless otherwise noted.

We measured the concentrations of HA-labeled TREM2 protein in the HEK293T conditioned media by two different ELISAs. First, sTREM2 concentrations were determined by the same ELISA used to measured sTREM2 in the human CSF samples, which includes a detection antibody against sTREM2 (see Additional file 1: Supplementary methods). Second, it was measured by a MSD platform-based including an antibody against the HA-tag. This second assay follows the same protocol as the first one but with the following modifications. The detection antibody is a monoclonal rat IgG anti-HA peptide sequence (YPYDVPDYA), clone 3F10 (Roche, Cat. No. 11 867 423 001; 100 ng/mL, 50 μL/well); the secondary antibody is a SULFO-TAG-labeled goat polyclonal anti-rat IgG antibody (MSD, Cat. no. R32AH-1; 0.5 μg/mL, 25 μL/well). The samples were diluted 1:50 and 1:100 in assay buffer [0.25% BSA and 0.05% Tween 20 in PBS (pH = 7.4)], supplemented with protease inhibitors (Sigma; Cat. no. P8340) and measured in duplicates for each dilution. We acquired the electrochemiluminescence response values using the MESO QuickPlex SQ 120. We compared the signal of the sTREM2 ELISA with that of the HA-tag assay for each of the TREM2 rare variants. The percentage between these two assays renders a relative affinity of the sTREM2 ELISA to each of the TREM2 rare variants in relation to its respective HA-tag control.

CSF sTREM2 did not follow a normal distribution (Kolmogorov-Smirnov test: P < 0.0001) and were hence log₁₀-transformed. After transformation, CSF sTREM2 followed a normal distribution as assessed by Kolmogorov-Smirnov test (P = 0.200) and visual inspection of the histogram. All the statistical analysis described in this study are performed with the log₁₀-transformed values.

A one-way analysis of covariance (ANCOVA) was conducted to determine statistically significant differences on CSF sTREM2 between TREM2 rare variants carriers and the non-carriers’ individuals adjusting for the effect of age, followed by a Bonferroni corrected post hoc pairwise comparison. Only those groups of TREM2 rare variants carriers that comprise more than 1 subject were included in the analysis.

The following analyses were conducted excluding outliers’ values of CSF sTREM2, defined as values differing 3 standard deviations from the mean. There were 5 outliers: 2 subjects classified as ‘Preclinical AD A+TN-’ (1 a TREM2 rare variant carrier and 1 a non-carrier), 1 classified as ‘CDR = 0.5 A+TN-’ (TREM2 rare variant carrier), 1 classified as ‘AD CDR = 0.5’ (TREM2 rare variant carrier), 1 classified as ‘AD CDR = 1’ (non-carrier). Including or excluding these outliers do not change the findings of this study.

To test the differences in CSF sTREM2 across biomarker profiles in the A/T/N framework, we applied a one-way ANCOVA including age as covariate, followed by Bonferroni corrected post hoc pairwise comparisons. A similar approach was used to test whether CSF sTREM2 changes across the Alzheimer’s continuum. These analyses were performed including or excluding individuals carrying a TREM2 rare variant and yielded similar results.

Finally, we studied the association between CSF sTREM2 and each of the CSF core biomarkers for AD (T-tau, P-tau_181P, Aβ_1-42) with a multiple linear regression adjusted for age. The analysis was conducted separately in the healthy controls, Alzheimer’s continuum and SNAP groups. We performed the analysis both including or excluding outliers (defined as AD CSF core biomarkers 3 standard deviations below or above the group mean) in order to exclude that the associations were driven by extreme values. The analysis with and without outliers rendered similar results. For CSF Aβ_1-42, the analyses were performed using both the truncated values at the upper technical limit and the exploratory measurements available based on the extrapolation of the calibration curve. In the main text, we report the results using the extrapolated measurements, but using the truncated ones yielded similar results.

Statistical analysis was performed in SPSS IBM, version 20.0, and the free statistical software R (http://​www.​r-project.​org/). Figures were built using GraphPad Prism or free statistical software R. All tests were 2-tailed, with a significance level of α = 0.05.

We studied a total of 1027 participants of the ADNI study. The demographical and clinical characteristics of the whole study population are described in Table S1 (see Additional file 1). Among the participants studied, 43 (4.2 %) had a known TREM2 rare variant (see reference [41] for a comprehensive review of the pathogenicity of each variant). The overall mean levels of CSF sTREM2 of these individuals (M = 3913 pg/ml, SD = 3548, n = 43) were significantly lower than the rest of ADNI participants without a TREM2 rare variant (M = 4136 pg/ml, SD = 2171, n = 984; F_1,1024 = 6.77, P = 0.009, η_p² = 0.007; Table 2, Fig. 1) in a one-way ANCOVA adjusted for age. However, CSF sTREM2 varied considerably between TREM2 variants (F_4,1019 = 8.79, P < 0.0001, η_p² = 0.033) and Bonferroni’s post hoc comparisons test indicated that the p.R47H variant [7‐12] had significantly higher CSF sTREM2 (P = 0.003) and the p.L211P variant [42, 43] significantly lower CSF sTREM2 (P = 0.002) than non-carriers. No differences in CSF sTREM2 were found between individuals with a p.R62H [44, 45] and the p.D87N [7] variants and the non-carriers. There was a single subject carrying both a p.D87N and p.R62H variants, and another single subject carrying a p.H157Y variant, which were not included in the statistical analysis. However, it is worth noting that the subject carrying a p.H157Y TREM2 rare variant had relatively high CSF sTREM2 (Table 2, Fig. 1), an observation that agrees with our previous findings that the p.H157Y variant, which is located exactly at the cleavage site, increases shedding of TREM2 [17]. Given that TREM2 rare variants may influence CSF sTREM2 (as described here and in [24]), all the following analyses are excluding participants carrying these rare variants. Nevertheless, including the TREM2 rare variants carriers did not change the results. In order to test whether the differences in CSF sTREM2 among TREM2 rare variants are influenced by differences in the antibody affinity to the mutant sTREM2, we transfected HEK293T cells with an epitope tagged wild type (wt) and mutated TREM2 and measured sTREM2 released in the media with the same ELISA used for the quantification of CSF sTREM2 and with an ELISA using an antibody against the epitope tag (Additional file 1: Figure S1). This revealed that the p.R47H, p.R62H and p.H157Y TREM2 rare variants were detected with a slightly reduced efficiency in our ELISA; therefore, the increased levels of CSF sTREM2 found in subjects bearing the p.R47H rare variants are even slightly underestimated. On the other hand, the p.L211P TREM2 rare variants were detected efficiently, independently of their individual amino acid exchanges. However, the p.D87N TREM2 rare variant was detected with significant less affinity in our ELISA than using the antibody against the epitope tag. Thus, the decreased CSF sTREM2 found in the p.D87N rare variant should be interpreted with caution.

In the sample excluding the TREM2 rare variants carriers and TREM2 outliers’ values (n = 982, see methods section), we first assessed which demographic and genetic variables are associated with CSF sTREM2 (descriptives summarized in Table 3). Consistent with previous results [21, 23‐26], CSF sTREM2 levels were associated with age (β = +0.275, P < 0.0001, η_p² = 0.073), but not with gender (F_1,978 = 0.029, P = 0.866, η_p² = 0.00003) or APOE ε4 status (F_1,978 = 0.099, P = 0.753, η_p² = 0.0001). Consequently, all further analysis included age as a covariate, but not gender or APOE ε4 status.

In order to assess the impact of Aβ deposition or the downstream processes of the amyloid cascade (i.e. tau pathology and neurodegeneration), we applied the recently proposed A/T/N classification framework of AD, which proposes 3 binary biomarker groups [27]: (1) aggregated Aβ (A+/A-), (2) aggregated tau (T+/T-) and (3) neurodegeneration (N+/N-). Given that CSF T-tau and CSF P-tau_181P were highly correlated, we merged the tau (T) and neurodegeneration (N) groups. ‘TN-’ profile was defined as both CSF P-tau_181P and T-tau within the normal range, whereas ‘TN+’ was defined as abnormal levels of CSF P-tau_181P or T-tau. Thus, we compared 4 different biomarker profiles within each clinical stage, namely: (1) A-/TN-, (2) A+/TN-, (3) A+/TN+ and (4) A-/TN+.

Within the CDR = 0 group (i.e. cognitively normal individuals), a one-way ANCOVA showed a significant difference between the four biomarker profiles after adjusting for the effect of age (F_3,286 = 21.3, P < 0.0001, η_p² = 0.183). A Bonferroni post hoc test revealed that the A+/TN- profile had significant lower CSF sTREM2 compared to either other biomarker profile (Fig. 2). Only the A-/TN+ profile had significant higher CSF sTREM2 compared to the normal biomarkers profile (i.e. A-/TN-).

Within the CDR = 0.5 group (i.e. very mild dementia), there was also a significant difference between the four biomarker profiles (F_3,582 = 40.7, P < 0.0001, η_p² = 0.173) and the A+/TN- profile had also the significantly lowest CSF sTREM2 compared to either other biomarker profile (Fig 2). Both the A+/TN+ and the A-/TN+ biomarker profiles had significant higher CSF sTREM2 compared to the A-/TN- profile (Fig. 2). The CDR = 1 group did not yield a sufficient number of subjects per A/T/N profile to allow for a group comparison.

We repeated the former analysis also including the individuals with TREM2 rare variants (n = 43; demographics in Table S1, see Additional file 1) and this did not change our conclusions derived from the main analysis (Fig. 2).

We also repeated the same analysis classifying the participants based only on their Aβ pathology (A; CSF Aβ_1-42) and tau pathology status (T; CSF P-tau_181P), that is A/T classification, or based only on their Aβ pathology (A; CSF Aβ_1-42) and neurodegeneration status (N; CSF T-tau), that is A/N classification. The results are shown in Figure S2 (see Additional file 1) and they are similar to those shown with the A/TN classification of the main text. Thus, we conclude that the pathological processes that are downstream of Aβ pathology, both tau pathology and neurodegeneration, are associated with increased CSF sTREM2.

Next we asked if CSF sTREM2 changes during the course of the disease as previously described in our study on late onset-AD [25]. We modeled the evolution of AD comparing the biomarker-defined groups (Table 1) that reflect the temporary course of late-onset AD, similar to what was proposed by the previous 2011 NIA-AA diagnostic criteria [32‐34]. Thus, we compared the ‘healthy controls’ group (highlighted in blue in Table 1 and corresponding to the the ‘CDR = 0 A-/TN-’ group), with those belonging to the Alzheimer’s continuum (highlighted in red in Table 1), which included: ‘Preclinical AD A+/TN-’, ‘Preclinical AD A+/TN+’, ‘AD CDR = 0.5’ and ‘AD CDR = 1’. A one-way ANCOVA revealed that CSF sTREM2 was significantly different between groups after adjusting for the effect of age (F_4,575 = 11.5, P < 0.0001, η_p² = 0.074). A post hoc analysis using the Bonferroni criterion for significance indicated that the average CSF sTREM2 was significantly higher in the ‘AD CDR = 0.5’ group than in the ‘healthy controls’ and ‘Preclinical AD A+TN-’ groups (P = 0.034 and P < 0.0001, respectively; Fig. 3). Similar results were obtained when the individuals carrying a TREM2 rare variant were included (Fig. 3). Thus, these results replicate our and other groups previous findings of increased CSF sTREM2 in early symptomatic stages of late-onset AD in an independent sample [23‐25, 29].

Finally, we studied the associations of CSF sTREM2 with each of the CSF core biomarkers of AD, that is T-tau, P-tau_181P and Aβ_1-42, in linear regression models adjusted for age. The associations were tested separately in three groups based on their biomarker profile (see Table 1): (1) healthy controls, (2) individuals of the Alzheimer’s continuum and (3) in the SNAP groups. Consistent with previous findings, CSF sTREM2 is associated with T-tau and P-tau_181P in the three groups studied (Fig. 4). In contrast, no significant associations were found between CSF sTREM2 and Aβ_1-42 (Fig. 4). Including the CSF biomarkers outliers (see Additional file 1: Table S2), or including the individuals carrying a TREM2 rare variants (see Additional file 1: Table S3), did not change our findings.