Background
Neuroinflammation is now widely accepted as a pathological hallmark of Alzheimer’s disease (AD) and other dementias [
1,
2]. However, in contrast to the classical fluid biomarker hallmarks of amyloid and tau proteins, a standard clinical application of inflammatory markers in the clinical diagnosis of AD is lacking, likely owing to contradictory and heterogeneous findings of numerous studies [
3‐
6]. It is noteworthy that several frequently investigated inflammatory markers are found in low abundance in both brain tissue and cerebrospinal fluid (CSF). These have therefore been excluded from analysis in some more recent studies [
7,
8]. Hence, part of the discrepancy may be due to marker levels being close to detection thresholds in laboratory assays, yielding nonreproducible findings. Another important aspect of such biomarker research is the attainment of sufficient sensitivity and specificity for diagnostic application [
9]. Although high sensitivity and specificity have consistently been reported for the use of CSF amyloid or tau markers, these key diagnostic parameters have barely been studied for inflammatory markers in AD research.
To address these issues, we first analyzed a total of 46 inflammation-associated protein markers to validate assay performance in CSF. The panel was based on a commercial 40-plex previously used by Chen et al
. for a study with a similar concept but involving an investigation of brain samples containing several typical pro- and anti-inflammatory cytokines and chemokines as well as mediators likewise associated with vascular injury [
8]. Additionally, the following markers from the complement system’s classical and alternative pathways, genome-wide association studies, and our own previous research that have previously been related to AD were selected: C1q, C5a, C3aDesArg, soluble triggering receptor expressed on myeloid cells 2 (sTREM2), inhibition of soluble interleukin-1 (IL-1) receptor accessory protein (sIL-1RAcP), and myeloid-related proteins 8/14 [
10‐
14]. Afterward, a panel of 15 markers with robust detectability in CSF was used to study samples derived from patients in a neurological and psychiatric outpatient unit, where decision-making relies on the ability of biomarkers to distinguish AD from other disorders. Investigated cohorts included patients with a diagnosis of AD or mild cognitive impairment (MCI) and nondemented comparator subjects (nondemented neurological patients without cognitive dysfunction, central nervous system [CNS] involvement, or inflammation-associated disorders), supplemented by smaller groups of subjects with other neurodegenerative diseases. We tested for associations between inflammatory markers and pathological and neuropsychological features as well as major potentially confounding factors. Finally, the diagnostic performance was evaluated by comparing the discriminative power (sensitivity and specificity) of significant inflammatory markers with that of standard AD amyloid and tau biomarkers to discriminate across the three groups in pairwise comparisons. This article is supported by Additional file
1, which provides detailed results of the initial test phase for each of the 46 proteins and a statistical supplement reporting those test results not included in the body of the article.
Discussion
Briefly, this study had three main objectives: (1) to ensure robust detectability of all included inflammatory biomarkers in the CSF, (2) to investigate potential associations between those inflammatory markers and important pathological features of AD, and (3) to test whether any inflammatory protein markers would achieve sufficient discriminative power to aid in clinical diagnostic procedures. Studies on CSF inflammation-associated markers in AD have provided quite heterogeneous results, and no powerful marker has yet gained salience for use in clinical practice [
5]. Differences in cohort composition, sample handling, and choice of assays have a major impact on the results obtained in such studies [
21]. Aside from this, low CSF concentrations of inflammatory mediators constitute another important limitation. Several recent studies have highlighted that many proteins included in typical multiplex panels are below detection limits when investigating AD CSF or brain tissue samples [
7,
8]. This is consistent with the initial testing phase in the present study, where roughly one-third of tested proteins, especially very frequently reported pro- and anti-inflammatory cytokines, were found to be below detection limits. The number of detectable cytokines and chemokines was limited to a few analytes such as IL-8, MCP-1, or IP-10. In contrast, robustly detectable markers were proteins from different immune signaling pathways or associated with multiple biological processes, such as soluble receptors, complement factors, or vascular signaling. These findings indicate that data on inflammatory markers with low abundance should be treated with caution. Furthermore, future research should either be focused on analysis of pathways of robustly detectable markers or use more sensitive assay systems to overcome such analytical limitations.
Another technical finding of this study was related to biobank storage time. Storage time was correlated with CSF levels of VEGF, sVEGFR, MIF, and C3aDesArg. In general, it would be expected that samples would lose quality over time and that measured levels would decrease as analytes deteriorate. However, this was observed only for sVEGFR. Statistically, for the other three markers, there was a perplexing positive correlation between storage time and levels. Whether this observation was a data artefact or due to actual biochemical or biophysical processes, such as separation from binding proteins or changes to epitopes recognized by antibodies, could not be clarified with the available dataset. It is noteworthy that storage time was not a critical covariate for the majority of clinical findings in this study.
Among the panel of detectable markers, two proteins stood out when we compared nondemented, MCI, and AD samples: CRP and sTREM2. CRP is one of the most common peripheral blood biomarkers for cardiovascular and inflammatory disorders. Similarly to several other inflammatory markers, elevated peripheral blood levels of CRP have frequently been associated with increased risk of dementias and cognitive decline in multiple studies, although less robustly with AD itself [
22,
23]. It is noteworthy that CRP levels showed a lognormal distribution in CSF, similar to those found in peripheral blood, which hampers interpretation of CRP levels in general [
24]. However, median levels of CRP decreased among groups, from nondemented to MCI to AD samples. This is consistent with findings reported by Schuitemaker et al
., who described lower CSF CRP levels in patients with AD than in those with MCI [
25]. Our study yielded no significant findings for CRP in PD samples, although others have described higher CSF CRP values in patients with PD with dementia than in patients with PD and control patients [
26]. Furthermore, CRP levels were robustly associated with CSF amyloid levels and
ApoE genotype, but not with tau or the clinical cohorts. These results are well in accordance with studies of CRP function in AD. CRP gene variants have been associated with plaque development in AD [
27]. The protein has been linked to amyloid pathology in APP/PS1 (amyloid precursor protein/presenilin 1) mice, and Strang et al
. have described dissociation of pentameric into monomeric CRP induced by amyloid plaques [
28,
29]. Monomeric CRP has been discussed as a linker between vascular damage and inflammation on the one hand and plaque formation, neuronal damage, and dementia risk on the other [
30]. Regarding the association of CRP with the
ApoE genotype, several reports have described statistical interactions between peripheral CRP levels and
ApoE genotype. For example, Hubacek et al
. described lower plasma CRP levels in
ApoE4 carriers than in
ApoE3 carriers [
31]. It should be noted that CRP is produced primarily in the liver, and local production of CRP in the CNS does not appear to contribute significantly to CSF levels [
32]. Hence, on the one hand, it could be hypothesized that CSF levels of CRP in AD are a consequence of decreased blood levels associated with the
ApoE4 genotype, which is simultaneously the strongest genetic risk factor for AD. On the other hand, the negative association found with phonetic fluency in patients with AD is consistent with the fact mentioned above that elevated peripheral CRP levels provide a risk factor for dementia [
22,
23]. Thus, CRP functions as well as CRP levels are likely linked by multiple mechanisms to AD pathology.
The other significant marker, sTREM2, has received tremendous attention since the discovery of
TREM2 gene variants as risk factors for AD. Its functional role in AD and other neurodegenerative diseases, although not without controversy, has been thoroughly investigated and reviewed [
12,
33,
34]. In the present study, sTREM2 was found to be elevated in MCI and AD cohorts and to be strongly associated with age and tau pathology, but with neither
ApoE genotype nor amyloid levels. These findings are highly consistent with previous reports [
35‐
39]. In part, this redundancy is probably due to the use of the same assay (the protocol described by Suárez-Calvet et al
.). Differences between results were of a minor nature, such as those related to the influence of sex on sTREM2 CSF levels.
Given the oppositional regulation of CRP and sTREM2 levels in the CSF, we tested how a ratio between levels of both proteins would impact significance and power. Although this ratio differed significantly between the nondemented and MCI as well as AD cohorts, discriminative power was only slightly improved, resulting in no added value of the ratio compared with its components. An explanation might be that in contrast to the Aβ
42/Aβ
40 ratio, which is made of mechanistically linked components, CRP and sTREM2 do not share a common pathway, aside from being inflammation-associated in general. So far, only in African American women has a
TREM2 variant been linked to higher peripheral CRP levels [
40]. Hence, CRP and sTREM2 are probably involved in different processes and stages of AD pathology, which could be the reason why a ratio does not improve cohort discrimination.
A range of further markers such as VEGF, sVEGFR-1, IP-10, sVCAM-1, MIF, and C1q, were robustly associated with pathological tau levels, independent of patient age. Basically, tau-associated proteins could be clustered into three groups when using age as a covariate: (1) proteins for which age was not a significant covariate at all; (2) markers for which there was an influence of age, whereas the association with tau was still significant; and (3) markers that were apparently entirely age-dependent and no longer significant for tau after adjusting for age. Hence, the influence of aging on CSF marker levels differs from molecule to molecule. Because CSF tau is considered a marker of pathological processes in later stages of AD, such as neuronal death, it is possible that the levels of these markers increase in the CSF owing to inflammatory signaling in response to tissue damage. This would be consistent with the associations found between inflammatory markers and cognitive outcomes, primarily phonetic fluency in AD, which is associated with frontal and temporal lobe integrity. Thus, CSF levels of these markers correlate to some extent with the pathological processes in the brain only in late stages of the disease and with cognitively demanding tests. Importantly, this does not exclude involvement of the immune system and inflammation within the brain in earlier stages of AD. It does, however, indicate that traceability of earlier inflammatory processes in the CSF might be limited owing to lack of detectability and effect strength of respective markers.
Comparison with other neurodegenerative diseases was limited by sample size in this study. Still, there were several striking findings. First, the clustered effects observed for CRP and sTREM2 in the main cohorts were robust in the supplemented cohorts. Second, sIL-1RAcP was elevated in both PD and dementia with DLB, which are both considered synucleinopathies [
41]. Last, sVEGFR and IP-10 differed between the main cohorts and FTD as well as ALS, which are also speculated to belong to one disease spectrum, in particular since the discovery of the
C9orf72 mutations [
42]. These results could indicate potential targets for research on inflammatory biomarkers in the respective disorders, but they require validation in larger cohorts.
When discussing biomarker findings, it is of great importance to consider not only statistical significance but also the effect strength, sensitivity, and specificity of the biomarker candidates [
9,
43]. This was the third and most important aim of this study. Sensitivity and specificity were weighted equally and summarized by the term discriminative power. Despite a large number of significant associations between clinical cohorts or AD pathology markers on the one hand and inflammatory markers on the other, none of the tested scenarios reached a discriminative power high enough for applicability in clinical diagnostic procedures. Overall, inflammatory markers were clearly associated with various pathological features of AD but did not show changes in CSF levels to the extent of established AD amyloid and tau markers. In consequence, the inflammatory processes involved in AD pathogenesis are not reflected in CSF in the same way as amyloidosis or tauopathy. Implementation of assays with higher sensitivity or investigation of signaling mediators from alternative pathways could lead to discovery of candidates with higher potential for use in clinical diagnostic procedures.