CSF sTREM2 in delirium—relation to Alzheimer’s disease CSF biomarkers Aβ42, t-tau and p-tau

The hip fracture patients, which were recruited from the Oslo Orthogeriatric trial (OOT), were admitted to the Oslo University Hospital Ullevål (OUS, Ullevål) between September 2009 and January 2012 [24, 25]. Delirium was assessed using the Confusion Assessment Method (CAM) [26] by the study physician or a study nurse. Delirium was assessed daily preoperatively and until the fifth postoperative day or in case of delirium until discharge. Pre-fracture dementia status was decided by consensus and based on the International Classification of Diseases − 10 (ICD-10) criteria for dementia by an expert panel as previously described [25].

The hip fracture patients (n = 120) were grouped both according to delirium and dementia status into either of the following groups (see Fig. 1a and Table 1)

Delirium patients included prevalent delirium (those that developed delirium preoperatively; n = 41) and incident delirium (postoperative delirium in those free from delirium before surgery; n = 21). The sub-classification of delirium onset was applied for delirium onset analyses.

In the hip fracture cohort, CSF was collected in connection with the orthopedic surgery at the onset of spinal anesthesia before administrating the anesthetic agents. CSF of patients with medical delirium was obtained during the diagnostic lumbar puncture (LP) at a median of 1 day after CNS symptoms developed. CSF was collected in polypropylene tubes and centrifuged as soon as possible, and supernatant aliquots were stored in polypropylene tubes at − 80 °C [27].

CSF sTREM2 was assayed by a sensitive TREM2 enzyme-linked immunosorbent assay (ELISA) as previously described [22]. Briefly, plates were incubated with an anti-humanTREM2 polyclonal capture antibody overnight at 4 °C (AF1828, R&D Systems, Minneapolis, MN, USA) and TREM2 detected by a mouse anti-human TREM2 monoclonal HRP-conjugated antibody (1 h incubation at room temperature (RT);SEK11084, Sino Biologics, Beijing, China). Samples were assayed in duplicates (2 h incubation at RT) with known cohort (hip fracture or medical), but with the clinical identity unknown to the operator. Samples with extreme values were assayed again, including the same and an increased sample dilution to verify measurements in the repeated assay. Two internal standard (CSF) samples were included in each assay to assess interday variability and used to adjust the medical delirium cohort which was assayed separately from the hip fracture cohort, with a final CV < 10% across assays.

CSF levels of t-tau, p-tau, and Aβ42 were quantified with commercially available ELISAs; Innotest® hTau Ag, Innotest® phoshoTau (181P), and Innotest® β-amyloid 1–42 as previously described [28‐30] (Fujirebio Europe, Gent, Belgium). CSF Aβ peptide levels were determined with CSF Aβ_1–38 (Aβ38), Aβ_1–40 (Aβ40), and Aβ_1–42 (Aβ42) MSD Triplex assay (Meso Scale Discovery, Rockwilly, MA, USA). All these analyses were performed at the Clinical Neurochemistry Laboratory at Sahlgrenska University Hopsital, Mölndal, Sweden. CSF cut-off for pathological level was < 530 pg/ml (Aβ42+/−), ≥ 60 pg/ml (p-tau+/−), and > 350 pg/ml (t-tau +/−) [31].

Analyses were performed by parametric or non-parametric statistics as appropriate depending on the data distribution. Data distribution was assessed by histogram, probability-probability (P-P), and quantile-quantile (Q-Q) plots. As CSF sTREM2 raw data were skewed, continuous data are reported by median (interquartile range (IQR)) and group differences analyzed by Mann-Whitney or Kruskal-Wallis test. p values of group comparisons were obtained by Mann-Whitney test, unless otherwise reported. The correlation analyses are reported by Spearman’s rho correlation coefficient (r_s).

Multiple linear regression was used for analyses with multiple predictor variables. For stepwise multiple linear regression, predictors were included based on significance in univariate analyses and biological grounds, including predictors with highest assumed degree of explained variability first. The data transformation by the natural logarithm (ln) (ln(CSF sTREM2 ng/ml) approximated a normal distribution and was therefore applied for analyses requiring parametric tests (linear regressions). Standardized residuals in linear regressions met the criteria of normal distribution. In regression analyses with delirium, no delirium was coded as 0 and delirium as 1. In linear regressions with delirium onset; no delirium was coded as 0, incident delirium as 1, and prevalent delirium as 2.

All hypotheses were two-sided and the reported p values are therefore two-tailed. The significance level was set at p < 0.05. Statistical analyses were performed by the Statistical Package for Social Sciences (SPSS, versions 24 and 25; IBM, Armonk, NY, USA). Graphical illustrations were created with GraphPad Prism (version 7.04 Graph Pad Software, La Jolla, CA, USA).

The patients studied did or did not develop delirium during hospitalization after a hip fracture requiring acute surgery. Around half of the patients included were demented before the accident, and a large proportion of those demented patients developed delirium after the hip fracture (≈ 80%; study setup in Fig. 1a). All the patients displayed an advanced age irrespective of dementia diagnosis (median age 85 years). The gender distribution was similar between groups (Table 1). Gender did not influence the CSF sTREM2 level and was therefore not included in the further analyses (data not shown).

Initial analyses of the CSF sTREM2 level showed a considerable intragroup variability, both among patients with and without delirium, with no separation between these two groups (p = 0.25, CSF sTREM2: 7.7(7.9) versus 7.4(5.4) ng/ml, n = 65, n = 54; Fig. 1b). Neuropathology of several dementia disorders involves microglial activation, and the underlying pathogenic processes may be different in delirium superimposed on dementia as compared to delirium in the absence of pre-existing dementia. Interestingly when stratifying patients according to pre-fracture dementia, an increased CSF sTREM2 level was only associated with delirium among those patients who were not demented before the hip fracture (p = 0.046, CSF sTREM2: 11.1(11.0) versus 7.4(5.0) ng/ml, n = 15, n = 44). In contrast, among patients with a pre-existing dementia, the CSF sTREM2 level did not differ between those patients who did or did not develop delirium (p = 0.94, n = 50, n = 10; Fig. 1c and Table 1).

During delirium, microglial reactivity may be transient and CSF sTREM2 levels may therefore fluctuate in a manner not directly related to clinical symptoms. To investigate possible changes in CSF sTREM2 with delirium progression, we stratified the delirium patients by onset of delirium relative to time of surgery when CSF was sampled (see Fig. 1a). Patients with pre-operational delirium were classified as having prevalent delirium, while those with post-operational delirium were categorized as having incident delirium. Among patients without pre-existing dementia, the incident delirium group showed an increased CSF sTREM2 level as compared to patients not developing delirium (p = 0.02, CSF sTREM2: 11.6 (5.0) versus 7.4 (5.0) ng/ml, n = 7, n = 44). The patients with prevalent delirium displayed a large variability in CSF sTREM2 and did not differ statistically neither from the unaffected patient group nor from the group having incident delirium (p = 0.46, p = 0.45, n = 8, n = 44, n = 7). When likewise stratifying patients with pre-existing dementia into subgroups of onset of delirium, the CSF sTREM2 level did not differ between the three groups (p = 0.62, Kruskal-Wallis test, Fig. 2 and Table 1).

A peripheral insult, such as a hip fracture, may trigger a central immune response [32]. The CSF sTREM2 level correlated positively to waiting time for surgery after hospital admission (waiting time for surgery (h); r_S = 0.23, p = 0.01, n = 119). The positive correlation in all patients was presumably due to a stronger positive relationship between CSF sTREM2 and waiting time for surgery among patients with pre-existing dementia, which remained when demented patients not developing delirium were excluded (r_S = 0.39, p = 0.002, n = 60; r_S = 0.40, p = 0.005, n = 49, Fig. 3 and Table 2).

Having found CSF sTREM2 to relate positively to surgery waiting time among patients with pre-existing dementia, we were concerned that this masked an effect of delirium on CSF TREM2 in our previous analyses of patients with pre-existing dementia. We adjusted for surgery waiting time, but delirium did still not affect the CSF TREM2 level in this group of demented patients (multiple linear regression bivariate model; waiting time for surgery (h): β1 = 0.007, p = 0.02, delirium: β2 = − 0.12, p = 0.60, n = 59, Table 3). The same analyses of delirium patients with pre-existing dementia sub-grouped relative to delirium onset (incident or prevalent delirium) reiterated that surgery waiting time, but not delirium onset, influenced the CSF TREM2 level (data not shown).

The study group was of an advanced age (median of 85 years), and CSF sTREM2 did not relate to patient age (r_S = 0.09, p = 0.32, n = 120, Additional file 1: Figure S1). Morbidity or comorbidity might have masked an effect of aging on CSF sTREM2 that we observed in a cohort with a wider age distribution [22]. However, CSF sTREM2 did still not relate to age when restricting the analysis to the patient group having neither pre-existing dementia nor delirium (r_S = 0.12, p = 0.43, n = 44; Additional file 1: Figure S1A and Table 2). The four study groups of the hip fracture cohort, patients with or without delirium and with or without pre-existing dementia, were age-matched when assessing median age. However, the patient group having neither pre-existing dementia nor delirium had a greater proportion of relatively young individuals than the other groups. Neuroinflammation is a feature of aging, and we wanted to ensure that the age distribution did not infer with our analyses. We approached this by adjusting for age and analyzing data with multiple linear regression models. CSF sTREM2 remained significantly increased with delirium in age-adjusted analyses of patients without dementia (p < 0.05, ln(CSFsTREM2); univariate model: delirium β1 = 0.40, p = 0.01; with a bivariate model: delirium β1 = 0.35 p = 0.03; age β2 = 0.01, p = 0.22, n = 59). This finding reiterated when reanalyzing the effect of incident delirium among patients without dementia (ln(CSFsTREM2): incident delirium β1 = 0.52, p = 0.01; a bivariate model: incident delirium β1 = 0.47, p = 0.03; age β2 = 0.01, p = 0.26, n = 51, Table 3).

In several reports, CSF sTREM2 level relates to CSF core biomarkers of Alzheimer’s disease Aβ42, t-tau, and p-tau [19, 21, 22]. We hypothesized observing such relations in the present study, with the comorbidities delirium or pre-existing dementia acting as possible confounders. First the relations were analyzed in all patients, then patients were stratified by dementia status and finally by delirium. After stratification by both factors, only two patient groups, those neither having pre-existing dementia nor delirium (n = 44) and those being afflicted by both conditions simultaneously (n = 50) were of a sufficient group size to merit further statistical analyses (Table 1).

Reduced CSF Aβ42 is assumed to reflect parenchymal Aβ sequestration by plaques in AD brain. CSF sTREM2 and CSF Aβ42 related positively in the entire hip fracture cohort (r_S = 0.22, p = 0.02, n = 117). When stratifying study subjects by pre-existing dementia status, CSF sTREM2 and CSF Aβ42 did not correlate in patients without pre-existing dementia, not even after excluding patients with delirium. In contrast, there was a strong positive relation between CSF sTREM2 and CSF Aβ42 of patients with pre-existing dementia which grew stronger when demented patients without delirium were excluded from the analysis (r_S = 0.40, p = 0.002, n = 60; r_S = 0.53, p < 0.001, n = 50, Fig. 4 and Table 2).

How prone an Aβ peptide is to form amyloid fibrils and be sequestered by a pre-existing Aβ deposit in the brain heavily depends on the C-terminal extension. Indeed, in clinical studies, only the level of hydrophobic Aβ peptides extending to position 42 decrease in CSF when Aβ deposits begin to form in brain [33]. Thus, an increased CSF level of a C-terminal truncated Aβ peptide, e.g., CSF Aβ_1–38 likely reflect Aβ-monomer metabolism (production or catabolism). Here, we related analyses of CSF Aβ38, Aβ40, and Aβ42 MSD ELISA levels to CSF sTREM2. Our intent was to distinguish if a putative relation between CSF sTREM2 and CSF Aβ42 was linked to Aβ-metabolism or Aβ-plaque sequestration. The MSD analyses outcome was reminiscent to the Innotest Aβ42 analysis, with CSF sTREM2 and CSF Aβ relating positively to all three Aβ peptides (Aβ38 _MSD, Aβ40 _MSD, and Aβ42 _MSD). Upon stratification, the effect was prominent among patients with pre-existing dementia and even stronger when further excluding those patients who did not develop delirium (Aβ38_MSDr_S = 0.45, p = 0.001; Aβ40_MSDr_S = 0.51, p < 0.001; Aβ42MSD rS = 0.53, p < 0.001, n = 50; Additional file 1: Figure S2 and Table 2).

CSF t-tau and p-tau concentrations are thought to reflect altered tau metabolism that relates to neurodegeneration and tangle formation in AD [34]. CSF sTREM2 related positively to both CSF t-tau and p-tau in the entire hip fracture patient cohort (r_S = 0.32, p = 0.001 for both, n = 117). Stratification by dementia status gave results reminiscent of the CSF Aβ-data, with strong correlations to CSF sTREM2 restricted to patients with pre-existing dementia (t-tau, r_S = 0.46, p < 0.001; p-tau, r_S = 0.49, p < 0.001, n = 60). The relation between CSF-TREM2 and tau markers remained but were weaker when restricting the analyses to demented patients who also developed delirium (t-tau, r_S = 0.34, p = 0.016, n = 50 and p-tau, r_S = 0.37, p = 0.008, n = 50; Fig. 5, Table 2).

Only among patients with pre-existing dementia did CSF sTREM2 relate positively to the three CSF AD core biomarkers (Aβ42, t-tau, and p-tau). Here in a separate examination, CSF Aβ42 and p-tau were included in multiple linear regressions to explore if AD core biomarkers together better explained the variability of CSF sTREM2. The biomarker CSF p-tau was chosen above t-tau since it is more AD-specific [35] and since previous analyses gave slightly higher correlation coefficients to CSF sTREM2. Such analyses enabled us to examine if effects of AD neuropathology, reflected by biomarkers CSF Aβ42 and CSF p-tau, masked an effect of delirium on CSF TREM2.

Again, the hip fracture patient cohort was stratified for pre-existing dementia. Variables included were delirium, age, CSF Aβ42, and CSF p-tau. In the patient group with pre-existing dementia, the AD CSF biomarkers were included as predictors before incorporating age and delirium in the analysis. CSF p-tau explained most variability as a sole predictor of CSF TREM2 level (R = 0.50). Including CSF Aβ42 as a second predictor increased the explained CSF sTREM2 level variability (R = 0.61), while age and delirium did not further increase predictive ability (ln(CSF sTREM2): CSF p-tau β1 = 0.01, p < 0.001; CSF Aβ42 β2 = 0.002, p = 0.002; age β3 = 0.0001, p = 0.91; delirium β4 = − 0.03, p = 0.86; n = 59). Irrespective of the number of predictors included, the effect size (β) remains essentially the same (Table 3).

The patient group without pre-existing dementia was examined in the same manner with multiple linear regressions. Delirium was included as the initial predictor before incorporating age and the AD core CSF biomarkers Aβ42 and p-tau. Delirium was the only significant predictor in this patient group (R = 0.34 as a single predictor). CSF AD core biomarkers did not have a significant impact and their inclusion and subsequent adjustment did not influence the effect size (β) of delirium (ln(CSF sTREM2: delirium β1 = − 0.47, p = 0.02; CSF p-tau β2 = 0.000, p = 0.97; CSF Aβ42 β3 = 0.000, p = 0.16; age β4 = 0.013, p = 0.13, n = 57). This model suggest that patients without pre-existing dementia, but afflicted by a hip fracture and subsequent delirium would result in an increased CSF TREM2 level (≈ 1.5 ng/ml). This applied to conditions adjusted for interfering factors, age, or CSF biomarker levels indicating AD neuropathology (Table 3).

We then examined ratios of amyloid and tau CSF biomarkers, which are more discriminative than a single biomarker to the diagnosis and prognosis of AD [36, 37]. Biomarker ratios (t-tau/Aβ42 or p- tau/Aβ42) separated non-demented patients with or without delirium (Additional file 1: Table S1), consistent with a previous report [10]. There was no correlation between CSF sTREM2 and CSF t-tau/Aβ42 and p-tau/Aβ42 among those with dementia or without dementia (Additional file 1: Table S1 and Figure S3).

AD is the major cause of dementia. To identify hip fracture patients with pre-existing dementia likely due to AD, we applied cut-off values for the CSF Aβ42 and CSF-tau [31]. Thus patients were stratified for having a CSF Aβ42-level above or below AD cut-off (Aβ−/Aβ+) and likewise tau-level (t-tau or p-tau) above or below AD cut-off (tau+/tau-) [38]. Most patients with pre-existing dementia were below CSF Aβ42 cut-off (≈ 90%), while fewer patients were above CSF-tau cut-off (≈ 40%, Table 1).

Among patients neither having delirium nor pre-existing dementia, CSF sTREM2 level did not differ between those who were above or below CSF-biomarker CSF Aβ42 level cut-off for AD (p = 0.22, n = 26 n = 18). Among patients without pre-existing dementia and a CSF Aβ42 level below cut-off, CSF sTREM2 was higher in those with subsequent delirium (n = 11) as compared to unaffected subjects (n = 26; p = 0.048). The analyses gave similar results as when all patients without pre-existing dementia were included in the analyses (p = 0.046, Fig. 1c). Corresponding analyses for tau (t-tau or p-tau positive) yielded essentially the same results (data not shown). These analyses confirm multiple linear regression analyses of CSF Aβ42 and CSF tau markers not influencing the CSF sTREM2 level in patients without pre-existing dementia.

We also included a group of delirium patients having delirium secondary to a medical condition with or without pre-existing dementia (medical delirium, n = 26, Table 1). There was a trend of a lower CSF sTREM2 level in patients with medical delirium as compared to those with delirium after a hip fracture (p = 0.08; CSF sTREM2 ng/ml: 6.1 (7.3) vs 7.7 (7.9), n = 26, n = 65, Figs. 1b and 6).

Restricting the analyses only to patients without pre-existing dementia, CSF sTREM2 was reduced in medical delirium as compared to patients with hip fracture triggered delirium (p = 0.04; n = 17, n = 15, Figs. 1 and 6; Table 1). Restricting the hip fracture cohort delirium patients to those with incident delirium CSF sTREM2 was lower, and close to significant (p = 0.055; n = 7, n = 17), but not when compared to those with prevalent delirium (p = 0.18, n = 8, n = 17; Figs. 2 and 6).

Analyzing only those with pre-existing dementia, CSF sTREM2 in patients with medical delirium did not statistically differ from hip fracture patients with delirium (p = 0.56, n = 9, n = 50; Fig. 1c and 6 and Table 1), nor when comparing to each subgroup.

Like in the hip fracture cohort, we found CSF sTREM2 not related to age neither when analyzing all medical delirium patients (r_S = 0.23, p = 0.26, n = 26) nor when restricting the analysis to patients without dementia (r_S = 0.41, p = 0.11, n = 17, Additional file 1: Figure S4). Patients admitted to the hospital with medical delirium were younger with a greater age range (66(16) years, Table 1) than the patients having a hip fracture accident. When adjusting for age, the level of CSF sTREM2 was not significantly higher in hip fracture triggered delirium relative to medical delirium. However, the group size is almost too small to allow statistical age adjustment with several covariates, and the loss of significance from univariate analyses may be explained by additional noise from the added predictor (ln(CSFsTREM2) univariate model; delirium: β1 = 0.62, p = 0.05, and with a bivariate model and age included: delirium β1 = 0.27, p = 0.48, age β2 = 0.02, p = 0.15, n = 24).