CNS and systemic inammation underlying neuropsychiatric symptoms and related cognitive decline in older people.

BACKGROUND Neuroinammation may contribute to psychiatric symptoms in older people, in particular in the context of Alzheimer’s disease (AD). Here, our objective was to determine systemic and central nervous system (CNS) inammatory signatures associated with neuropsychiatric symptoms (NPS) in older subjects, and investigate their relationships with AD pathology and cognitive decline. METHODS We quantied a panel of inammatory markers in both cerebrospinal uid (CSF) and circulating blood serum in elderly subjects with normal cognition or with beginning cognitive decline. We further performed a comprehensive clinical assessment including longitudinal cognitive and neuropsychiatric evaluations and measured CSF biomarkers of core AD pathology. Multivariate analysis selected CSF and serum neuroinammatory molecules associated with the presence of overall NPS and specic symptoms. RESULTS The presence of NPS was associated with distinct inammatory markers proles involving soluble intracellular cell adhesion molecule-1 (sICAM-1), C-reactive protein (CRP), Interleukin (IL) -8 and 10 kDa interferon-γ-induced protein (IP-10) in CSF; and Eotaxin-3, IL-6 and CRP in serum. Further analysis identied specic inammatory marker signatures associated with anxiety, depression and disinhibition. Presenting NPS was associated with subsequent cognitive decline and this association was mediated by CSF sICAM-1. CONCLUSIONS These results suggest that NPS in older people are associated with distinct systemic and CNS inammatory processes. Neuroinammation may explain the link between NPS and more rapid clinical disease progression.


Study procedures 2.2.1 Neuropsychological assessments:
Along with the clinical examination, the Neuropsychiatric Inventory questionnaire (NPI-Q) (28) was administered to assess neuropsychiatric symptoms. Twelve categories, ten behavioral and two neurovegetative (Night-time behavior and Appetite/Eating), were scored for their severity ranging from 0 to 3. Total NPI-Q score was obtained by adding the twelve scores. Participants with a total NPI-Q score of 0 and with no history or evidence of NPS were considered controls. We also collected Clinical Dementia Rating (CDR), CDR sum of boxes (CDR-SoB), and Mini-Mental State Examination (MMSE). Cognitive state was classi ed using CDR.

Biochemical sample collection and handling:
Lumbar and venous punctures conducted during the same visit yielding 10-12 ml of CSF and 40 ml of blood respectively were performed after an overnight fast in the memory center, spun down at 4 o C, immediately aliquoted, and snap frozen at -80 o C until assayed (22). Study personnel blinded to clinical data performed biochemical and genetic analyses.

Biochemical measures:
The CSF albumin index (Qalb) of blood-brain barrier (BBB) impairment along with the APOE genotype were determined as previously described (29).

Volumetric measurements:
All participants underwent a magnetic resonance imaging scan on a 3T MRI system (MAGNETOM Prisma t , Siemens Healthcare, Erlangen, Germany) with a 32-channel head coil. Acquisitions followed the ADNI2 MRI protocol (30). Images were then segmented with the MorphoBox prototype algorithm (31); brie y, this registers subject data to an internal template established by consensus segmentation of neuroradiologists, applies bias eld correction with a 4-tissue class (gray matter (GM), white matter (WM), cerebro-spinal uid (CSF), non-brain) Gaussian mixture model, performs skull-stripping, classi es brain tissue into 5 classes (ventricular CSF, sulcal CSF, cortical GM, deep GM, and WM) via variational expectation-maximisation yielding 5 posterior probability maps, and provides regional volume estimates by summing up these probabilities within template regions. Quality was checked using established automated image quality (32) and segmentation quality metrics (31), and no images were rejected. We analyzed 18 brain structures and features including: amygdala, caudate nucleus, cerebellum, cortical grey matter, grey matter, hippocampus, insula, medulla oblongata, mesencephalon, pallidum, pons, putamen, thalamus, 3rd ventricle, 4th ventricle, total ventricular volume, white matter, white matter abnormalities.
This regional volumetric data was normalized by total intracranial volume (de ned as the sum of gray matter, white matter and CSF).

Data preparation and transformation:
Before analysis, outliers (i.e. data points that exceeded the cutoff value of mean ± 3 × SD) were replaced by the cutoff value. Biomarker data was log n -transformed prior to correlation and regression analyses to approach Gaussian distribution. For participants displaying a positive NPI-Q score, we computed an NPI severity variable by log n -transforming NPI-Q score.

Statistical and analytical approaches:
Descriptive statistics for the cohort were performed using t-tests comparing NPI-Q and control group for continuous variables and Chi-square tests for categorical variables. Box plots and t-tests compared biomarker distribution and concentration between groups. Correlations implicating neuropsychological measures were assessed with Spearman's rho. Benjamini-Hochberg correction of P value for multiple testing was then applied using a false-discovery rate of 0.1. Independence of variables used in regression models was tested with variance in ation factor (VIF). No variable entered in these models had VIF above 7, with a majority below 3, thus absence of multicollinearity was assumed. Considering our effect sizes (> 0.7), the size of this cohort results in a statistical power of 0.9 or more. Statistical data analysis was performed with IBM SPSS Statistics software version 25.

Statistical modelling:
To select neuroin ammatory marker combinations in CSF or serum associated with NPS, we used binary regression models with NPI-Q > 0 or NPI-Q = 0 as dependent variable while entering all CSF or serum markers. We explored the following confounders: cognitive status (CDR = 0 or CDR > 0), CSF AD biomarker pro les (pTau/Aβ 1−42 ratio) and BBB impairment (Qalb) by entering them into the model before considering in ammatory marker concentrations. NPI severity score was then correlated to CSF or serum in ammation markers using linear regression with similar corrections. Associations of in ammatory markers with individual symptom category impairment (NPI-Q > 0 for each individual category) was assessed using a binary logistic regression model corrected for cognitive status as above. All models used a forward selection method based on the signi cance of the score statistic. To construct a reference model, we used a separate binary logistic regression model with NPI-Q > 0 or NPI-Q = 0 as dependent variable while using available demographic and clinical measures, including APOEε4 status, sex, age, years of study and cognitive status to predict the occurrence of NPS. A receiver operating characteristic (ROC) curve and area under the curve (AUC) were computed for this model. CSF markers associated with NPS selected by the above models where then added to this model and ROC curves and AUCs were compared using the DeLong method. Interaction between AD pathology and neuroin amamtion was tested using binary and linear regression models with either: CSF AD biomarkers × neuroin ammatory biomarkers versus NPS or NPI-Q score × CSF AD biomarkers versus CSF biomarkers, respectively. Associations between morphometric data and CSF neuroin ammatory markers were assessed using linear regression models with stepwise selection method and individual CSF molecule concentration as dependent variables entering all volumetric measurements. Linear regression was used to test associations between CSF neuroin ammatory markers and cognitive decline computed by change in CDR-SoB over time corrected for initial CDR-SoB assessment. Binary logistic regression with cognitive decline as dependent variable was used to test the associations of NPI-Q score, sICAM-1 concentration and NPI-Q × sICAM-1 concentration with cognitive decline at 18 and 36 months. In order to verify for possible over tting of the above logistic regression models, we used the Hosmer-Lemeshow test for goodness-of-t. Models with a Hosmer-Lemeshow chi-squared value yielding a P-value > 0.05 were rejected and the previous iteration was considered instead.

Results:
3.1 Characteristics of the cohort Serum, CSF, neuropsychiatric inventory and cognitive measures were available in 87 participants.
Demographics, clinical and biological characteristics of participants by group with or without NPS are given in Table 1

Neuroin ammatory signatures associated with the presence of NPS
We found 5 markers in CSF (Fig. 1A) and 3 in serum ( Fig. 1B) displaying signi cantly different concentrations between participants with (NPI-Q > 0) orwithout NPS (NPI-Q = 0). Overall NPI-Q score was positively correlated with the concentrations of sICAM-1, sVCAM-1, sFLT-1, IL-8 and IL-15, and MCP-1 and MCP-4 in CSF in the whole cohort (Fig. 1C). In serum, the concentration of MIP-1α showed a signi cant positive correlation with NPI-Q, while for CRP and VEGF this correlation was negative (Fig. 1D).  Table 2 NPI-Q scores within the whole cohort for total score and individual categories. "Controls" represents cases were NPI-Q score is 0 for the overall or categorical score = 0. All other cases are classi ed as "Symptoms". The percentage of cognitively healthy individuals within the whole cohort suffering from neuropsychiatric symptoms overall and in individual categories is also shown (CDR = 0). CDR, Clinical Dementia Rating; Ab motor behaviour, Aberrant motor behaviour.
Using binary regression models, we identi ed a combination of 4 neuroin ammatory markers in CSF that best predicted the occurrence of NPS: CRP, IP-10, sICAM-1 and IL-8 ( Fig. 2A). In serum, this combination was: Eotaxin-3, IL-6 and CRP (Fig. 2B). Because we observed positive NPI-Q scores in the absence of cognitive impairment (Table 2), we re ned these models by correcting them for baseline cognitive status. This did not change the combination of CSF in ammation markers associated with the occurrence of NPS, but in serum only CRP remained associated with a positive NPI-Q score ( Fig. 2A and 2B). To investigate the effect of AD pathology on these associations, we corrected our model for the presence or absence of AD pathology at baseline. Once again, this did not change the CSF neuroin ammatory marker associations, only adding the CSF signature of AD pathology to the model ( Fig. 2A). In serum, VEGF-D was added to the model together with the CSF biomarkers (Fig. 2B). We also investigated the effects of blood-brain barrier (BBB) impairment on this association by further correcting our model for Qalb. In this context the CSF in ammatory markers associated with the occurrence of NPS remained the same although the association of IL-8 with NPI-Q was no longer signi cant ( Fig. 2A, P value = .059). In serum, MIP-1α, VEGF and Qalb itself were associated with the occurrence of NPS (Fig. 2B).
When added to a reference model built using APOE status, Sex, Age, Years of study and Presence/Absence of cognitive impairment (Supplementary Table 4); sICAM-1, IP-10, IL-8 and CRP together signi cantly contributed to improve prediction of NPS ( Supplementary Fig. 1, AUC = 0.908, P value = 0.0058) Severity of NPS was only associated with CSF TARC levels when accounting for all conditions mentioned above (Fig. 2C). In serum, only IL-6 was always associated with NPI-Q severity, whereas sICAM-1, IFN-γ, MCP-4 where associated with the severity of NPI-Q in the uncorrected model with the addition of IL-16 in the CSF AD biomarkers and BBB permeability models (Fig. 2D).

Neuroin ammatory markers associated with individual neuropsychiatric syndromes
We also evaluated the association of neuroin ammatory marker concentrations in both CSF and serum with the occurrence of each individual syndrome measured by NPI-Q ( Table 3). Associations of in ammatory markers or combinations of markers in both CSF and serum emerged for nine out of twelve symptom categories. In all cases, CSF and serum associations were distinct. In CSF, sICAM-1 was overrepresented, with strong associations with Depression, Anxiety and Disinhibition. In serum, we found associations in 5 out of 12 symptoms and the strongest association observed was between VEGF-D and Depression (Table 3).Other molecules associated with the occurrence of overall NPS or their severity, such as MCP-4,CRP, sVCAM-1, VEGF or VEGF-D also displayed (albeit weaker) associations with individual symptoms. Conversely, we also found associations of IL-12 and MDC with Irritability and Night-time behavior, respectively even though these molecules were not associated to NPS in general.

Interaction between neuroin ammation, NPS and AD pathology
Since the CSF concentrations of CRP, IP-10, sICAM-1, and AD status at baseline were associated with a positive NPI-Q score, we sought to test whether the interaction between neuroin ammation markers and the presence of AD pathology was associated with NPS. Using regression models with interaction variables we found that only the interaction between AD pathology and sICAM-1 was signi cantly associated with the occurrence of NPS (Nagelkerke R 2 = 0.363; coe cient = 3.386; P value = .027). The converse interaction model, i.e. the interaction between NPI-Q score and CSF AD biomarkers showed no signi cant association with CSF neuroin ammatory markers.

Neuroin ammatory markers and brain morphometry
We next explored the associations between the NPS-relevant neuroin ammatory CSF markers (Fig. 2) and brain structure volumes. Each of these individual CSF neuroin ammatory markers was associated with the volume of speci c brain regions (Table 4). In a separate binary regression model, the occurrence of NPS was associated with changes in volume in the hippocampus and ventricles (Nagelkerke R 2 = 0.525; coe cients = -8.459* and 4.233** respectively).  Table 4. Individual model R 2 as well as β-coe cients for each signi cantly associated neuroin ammatory molecule are shown. * P value < .05; ** P value < .01; *** P value < .001.

NPS, neuroin ammation and cognitive decline
In the whole cohort, the NPI-Q score was positively correlated with both CDR and CDR-SoB and negatively correlated with MMSE (Table 5). Binary logistic regression con rmed this association between NPI-Q score and cognitive status at baseline (Nagelkerke R 2 = 0.206; coe cient = 0.278; P value = .004). When only considering participants with a positive NPI-Q score most individual NPS categories were correlated with cognitive measures (Table 5). We further correlated NPI-Q scores at baseline with cognitive decline as measured by changes in CDR-SoB at follow-up clinical assessments at 18 and 36 months (n = 87 and 85 respectively, Fig. 3A) in the whole cohort and in participants with a positive NPI-Q score. We also tested in a linear regression model the associations between the concentrations of molecules identi ed in the CSF signature and cognitive decline. Only sICAM-1 was positively associated with cognitive decline at 18 and 36 months (Fig. 3B). A mediation model testing the interaction between NPI-Q score and sICAM-1 concentration revealed that the association between NPI-Q and CDR-SoB change at 18 months is mediated by sICAM-1 concentration (Fig. 3C).

Discussion:
We have identi ed combinations of in ammation molecules that best predict the occurrence of NPS in both CSF (CRP, IP-10, Il-8 and sICAM-1) and serum (Eotaxin-3, IL-6 and CRP). Distinct marker combinations in CSF (TARC only) and serum (sICAM-1, IL-6 and IFN-γ) were associated with the severity of NPS. Individual NPS symptoms were associated with speci c neuroin ammatory molecules in both CSF and serum. The identi ed CSF neuroin ammation molecules were associated with volume changes in speci c brain regions. Finally, NPS were associated with more rapid cognitive decline at follow-up and this association was mediated by sICAM-1 CSF levels.

Speci c CSF neuroin ammatory molecules are associated with NPS
Amongst the molecules related to NPS in our study, CRP and sICAM-1 have been previously associated with the presence of AD pathology (22,33). However, the association of these neuroin ammatory markers with NPS is to our knowledge novel. In this study, sICAM-1 in particular, is strongly associated with NPS. This association could be driven by the large number of participants with AD pathology in the NPI-Q > 0 group. However, accounting for the presence of cerebral core AD pathology does not alter the speci c combination of molecules associated with NPS in CSF. Therefore, this association is at least partially independent of AD. In this signature, CRP, IP-10 and sICAM-1 could play different roles as high concentrations of CRP and IP-10 are associated with lower NPI-Q scores, suggesting a "neuroprotective" effect; while a higher concentration of sICAM-1 is associated with the presence of NPS, suggesting a deleterious role.
We have shown this CSF signature is independent of cognitive status suggesting neuroin ammation leading to NPS also occurs in the absence, or before the onset, of cognitive decline. Furthermore, this signature is not dependent on BBB function, suggesting that the related in ammatory process originates within the CNS. We have also shown that it is not only the absolute concentration of these molecules that is associated with NPI-Q score, but rather their relative concentrations between one another and their pattern of expression and interplay. For example, IP-10 does not display a different concentration between participants exhibiting NPS and those who do not, and does not correlate with total NPI-Q score either; it does however strongly associate with the occurrence of NPS when considering changes in CRP and sICAM-1 concentrations. This novel nding is in accordance with the concept that cytokines form a complex network enhancing and/or suppressing the production of each other (34).

A distinct serum in ammation signature is associated with NPS
We identi ed a distinct serum in ammatory signature of the occurrence of NPS containing Eotaxin-3, IL-6 and CRP. Unlike the CSF in ammatory marker pro le, this signature is related to cognitive status, BBB function and the presence of cerebral core AD pathology. This con rms systemic in ammation may both contribute to predisposing to or enhancing CNS in ammation which may further result in cerebral dysfunction and neuronal injury, and the manifestation of both cognitive impairment and NPS (25). In has been previously shown that systemic in ammation may also re ect cerebral pathology (26) and neuroin ammation related to NPS (22). It is also plausible that following BBB breakdown, caused by AD or neuroin ammation (29), cross-talk between both CNS originating and circulating in ammation occurs (35) as is the case in aging (36). These data suggest a deleterious feedback loop where circulating neuroin ammatory molecules such as IL-6 and IFN-γ can further enhance expression of CNS in ammatory molecules (CRP and IP-10 respectively, (37,38)) that together can further damage the BBB (39) and activate microglia (40). The result is an escalation of neuroin ammation, further contributing to CNS processes leading to the manifestation of NPS. While we do not elucidate here the origin of these in ammatory processes, since subjects with manifest unstable medical conditions, including in ammation were excluded from the present study, we show these speci c markers are particularly relevant for the occurrence of NPS.

Speci c mechanisms relate to the severity of NPS
Distinct molecules, with the exception of sICAM-1 were associated with the severity of NPS both in CSF and serum in our study. This suggests that besides the identi ed in ammatory processes associated with the appearance of NPS, additional mechanisms may modulate their extent and severity. We therefore suggest that CNS in ammation and BBB breakdown could trigger the appearance of NPS. Following these events, CSF TARC along with circulating, IL-6, IFN-γ, MCP-4 and IL-16, could regulate and enhance existing symptoms or play a regulatory role in the in ammatory response and dictate speci c and localized responses.

Individual NPS symptoms have speci c pathological mechanisms
Several individual NPS symptoms were associated to speci c neuroin ammatory molecule signatures in CSF and serum. Amongst these, the strongest associations were found with the CSF signatures of Anxiety and Disinhibition and the serum signature of Depression. Symptom speci c pro les differed between CSF and serum in all cases. Some of the molecules involved in these signatures (CRP, sICAM-1, IP-10, VEGF) are part of the signature of overall presence of NPS.
Previous evidence suggests neuroin ammation can lead to a variety of NPS (21). We however describe novel and speci c associations of the identi ed molecules, such as sICAM-1 and VEGF-D, with multiple NPI-Q categories suggesting that these symptoms may have common pathogenic mechanisms involving the associated molecule. In particular, higher CSF sICAM-1 levels were positively associated with symptoms of Depression, Anxiety, Apathy and Disinhibition in our cohort and could therefore play a role in the pathogenesis of all these symptom. We also found concentrations of VEGF-D in serum to be associated with the occurrence of Agitation, Depression, and Night-time behavior disorders. Furthermore, serum VEGF-D was associated with the occurrence of NPS only in the presence of cerebral AD. This suggests a role for this molecule in the pathogenesis of these NPS in the speci c context of AD.
Contrary to these molecules, some of the identi ed molecules were associated with only a single symptom category; some of which have previously been reported, such as serum IP-10 with sleep disturbances (41) and serum IL-8 with Anxiety (42). The associations of MCP-4 with Agitation and CSF MDC with Night-time behavior however are novel. Overall, these ndings suggest that while neuroin ammation results in a wide spectrum of neuropsychiatric manifestations, speci c neuroin ammatory mediators could play a more important role in certain single syndromes.

Brain regions associated with individual NPS symptoms
The molecules we identi ed as part of the CSF neuroin ammatory pro les related to NPS have previously been associated with volume changes in speci c brain regions (43)(44)(45), together suggesting neurodegeneration associated with neuroin ammation within these individual regions. We also observed that the volume of speci c brain regions is associated with the concentration of neuroin ammatory markers.
The strong association of sICAM-1 with the volume of the hippocampus and 3 rd ventricle (re ecting atrophy of surrounding areas) suggests these regions may play a role in the pathogenesis of NPS, although we cannot exclude sICAM-1 is linked to NPS via mechanisms independent of atrophy. Interestingly, these regions have been associated with the progression of AD pathology and in ammation in previous work (46,47). Considering our observations and previous reports (13), we infer that decreased volume of the hippocampus and increased 3 rd ventricle volume may indicate regional neurodegeneration and neuroin ammation together involved in the pathogenesis of NPS of the symptoms associated with sICAM-1 (i.e. Anxiety, Depression and Disinhibition).

NPS, neuroin ammation and clinical disease progression
While the occurrence of NPS is associated with the presence of AD pathology, we have shown that the CSF neuroin ammatory signature of NPS is independent of AD. This nding suggests that at CNS level both AD core pathology and neuroin ammatory processes may engage similar pathways leading to NPS.
CSF sICAM-1 was previously found to correlate with both tau and pTau181 levels, con rming it is involved in AD-related tau-pathology and neural injury (22). In our models, the interaction between the presence of AD pathology and sICAM-1 was signi cant, suggesting that sICAM-1 is present in both the AD and neuroin ammation pathways related to NPS.
Higher NPS severity was associated with more rapid cognitive decline. This is in line with previous research describing an association of NPS with more rapid clinical disease progression in AD (4,10,9).
The sICAM-1 CSF concentration is also associated with cognitive decline and our interaction analysis suggests that the association between NPS and cognitive decline is mediated by increased sICAM-1 in the CNS. A possible underlying mechanism is through altered cerebrovascular reactivity effects with which ICAM molecules are associated (48) which in turn, are known to differ according to cognitive status (49).

Strengths and limitations
Subjects included in this study had no psychiatric affection or symptom that could interfere with cognition and candidates with more marked neuropsychiatric symptoms were not considered.
Consequently, there was a low frequency of some of the single NPS symptoms such as delusions and hallucinations in this cohort which does not allow to address possible relationships between in ammation and these symptoms. Whether the identi ed signatures may be used in clinical practice as markers of neuroin ammation-related NPS and targets for intervention needs to be further investigated in independent samples. Strengths of this study are the inclusion of elderly subjects with normal cognition or cognitive decline and the assessment of a large panel of in ammatory markers in paired serum and CSF samples. Furthermore, we addressed the relationships of in ammation related NPS with the core cerebral AD pathology, regional brain atrophy, and cognitive decline over time.

Conclusion
We have identi ed speci c CSF and serum in ammatory signatures associated with NPS that can be considered both contributors to the underlying cerebral pathology and potential biomarkers of NPS. While the CSF signature identi ed here appears to indicate in ammatory processes that originate within the CNS and interact with the core AD pathology, the serum signature may represent systemic dysregulation of in ammatory activity related to BBB function and impacting CNS processes that lead to NPS. The in ammatory signatures of NPS indicate symptom speci c underlying processes opening the perspective of targeted interventions to reduce NPS and their long-term consequences.

Declarations
Ethics declaration and consent of participants: The local ethics committee of canton Vaud (Switzerland) approved this study (No. 171/2013), and all participants or their legal representatives provided written informed consent.
Availability of data: The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.