Study population characteristics
A total of 108 participants (40 SCD, 39 MCI and 29 AD) were included in the analysis. Basic demographic, biomarker and neuropsychological assessments of the study sample are described in Table 1. In brief, AD participants were significantly older (67.7 years old) than SCD (59.6 years old) and MCI (61.5 years old) (p = 0.001) whereas sex distribution and education were similar among the three diagnostic groups. CSF Aβ42 levels were lower (p < 0.001) while t-tau and p-tau levels were higher in AD individuals (p < 0.001 for both), after age adjustment compared to the other two groups. The majority of the AD group was positive for amyloid pathology (86.2%), while 20.5% of the MCI and 2.5% of the SCD groups consisted of amyloid positive subjects. AD participants performed worse in memory (p < 0.001) and processing speed (p = 0.021) tests compared to SCD and MCI subjects. There were no significant differences between the groups in terms of depressive symptoms and perceived stress assessments.
Table 1
Demographic, biomarker and neuropsychological characteristics of the study population.
| N | SCD (N = 40) | MCI (N = 39) | AD (N = 29) | p |
Age | 108 | 59.6 (6.0) | 61.5 (7.0) | 67.7 (8.3) | < 0.001 |
Men/Women, % | 108 | 40.0/60.0 | 46.2/53.8 | 40.7/59.3 | 0.621 |
Education years | 108 | 14.2 (3.5) | 13.6 (3.3) | 13.5 (3.0) | 0.657 |
Ab42, pg/mL | 108 | 861 (163) | 727 (193) | 430 (106) | < 0.001 |
Ab42, % positive | 108 | 2.5 | 20.5 | 86.2 | < 0.001 |
t-tau, pg/mL | 108 | 262 (111) | 309 (1567) | 633 (277) | < 0.001 |
p-tau, pg/mL | 108 | 39.4 (14.1) | 43.4 (17.9) | 74.6 (29.8) | < 0.001 |
Memorya | 97 | -0.04 (0.84) | -1.3 (1.2) | -2.8 (0.9) | < 0.001 |
Processing speedb | 91 | -0.32 (1.1) | -0.94 (1.2) | -1.6 (1.0) | 0.021 |
Geriatric Depression Scale (GDS)c | 101 | 5.3 (3.7) | 6.8 (4.7) | 4.9 (3.9) | 0.211 |
Perceived Stress Scale (PSS)d | 96 | 26.3 (6.9) | 27.2 (10.1) | 24.5 (9.5) | 0.687 |
Data are shown as unadjusted mean (SD), unless otherwise stated. P values for age and education were calculated by Kruskal-Wallis test. One-way ANCOVA was applied for analysis of CSF biomarkers and neuropsychological assessments, with age adjustment. Chi square was used for categorical data. P < 0.05 was considered statistically significant.
a Available data for 40 SCD, 33 MCI, 24 AD
b Available data for 39 SCD, 30 MCI, 22 AD
c Available data for 36 SCD, 38 MCI, 27 AD
d Available data for 32 SCD, 37 MCI, 27 AD
Biomarkers of stress, synaptic damage, neuroinflammation and vascular dysfunction among diagnostic groups
Salivary cortisol, a measure representing the free circulating levels of the hormone, is routinely used as a biomarker of the HPA axis response to psychosocial stress 40. In the current study, we measured CAR, which is the increase of cortisol 30 min post-awakening and cortisol slope, reflecting the change in cortisol levels from awakening to bedtime. There were no differences in CAR and cortisol slope between AD, MCI and SCD patients (Table 2). Next, we compared the levels of several CSF biomarkers of synaptic dysfunction, neuroinflammation and cerebrovascular function across the three diagnostic groups (Table 2). The synaptic markers NG and SNAP-25 were elevated in AD compared to both SCD and MCI participants (p<0.001 for both markers), in AD versus SCD (p<0.001 for both) and in AD versus MCI subjects (p<0.0001 for NG and p<0.001 for SNAP-25). Of the 38 inflammatory and vascular biomarkers initially analyzed, half of them passed the quality control criteria and were considered for the final analysis. Of these, IP-10, TARC, CRP, ICAM-1 and VCAM-1 exhibited higher concentrations in MCI patients compared to the other two groups (p=0.030 for IP-10, p=0.029 for TARC, p=0.002 for CRP, p=0.031 for ICAM-1 and p=0.007 for VCAM-1). In pairwise comparisons, IP-10 was higher in MCI compared to AD (p=0.019) and SCD (p=0.040), TARC was higher in MCI versus AD (p=0.012), CRP was increased in MCI compared to AD (p<0.001) and SCD (p=0.012) and ICAM-1 and VCAM-1 were higher in MCI versus AD (p=0.011 for ICAM-1 and p=0.002 for VCAM-1). There were no significant differences for the rest of the biomarkers. All comparisons were performed with age and sex adjustments.
Table 2. Concentrations of salivary cortisol and CSF synaptic, neuroinflammation and vascular dysfunction markers among diagnostic groups.
|
N
|
SCI
(N=40)
|
MCI
(N=39)
|
AD
(N=29)
|
p
|
Salivary cortisol
|
CAR
|
106
|
0.78 (1.5)
|
0.44 (1.0)
|
0.11 (1.3)
|
0.194
|
Cortisol slope
|
107
|
-5.1 (4.1)
|
-5.1 (11.8)
|
-9.4 (5.3)
|
0.105
|
Synaptic loss
|
NG
|
107
|
172 (54.1)
|
176 (76.6)
|
278 (97.8)a,b
|
<0.001
|
SYT-1
|
108
|
33.3 (8.7)
|
34.5 (10.7)
|
38.9 (10.4)
|
0.427
|
SNAP-25
|
108
|
10.8 (1.8)
|
11.2 (2.1)
|
14.1 (2.8)a,c
|
<0.001
|
Neuroinflammation
|
YKL-40
|
108
|
120 (45.0)
|
145 (65.6)
|
164 (57.0)
|
0.427
|
IL-5
|
107
|
0.86(0.22)
|
0.87(0.25)
|
0.91(0.27)
|
0.618
|
IL-6
|
106
|
1.9 (1.3)
|
1.6 (1.1)
|
1.4 (0.45)
|
0.321
|
IL-8
|
106
|
43.5 (9.2)
|
41.6 (7.7)
|
43.5 (9.1)
|
0.388
|
IL-12/IL-23p40
|
104
|
5.6 (2.9)
|
5.5 (2.3)
|
5.1 (1.7)
|
0.551
|
IL-15
|
106
|
3.2 (0.79)
|
3.4 (1.1)
|
3.6 (1.1)
|
0.620
|
IL-16
|
106
|
12.8 (3.4)
|
13.6 (6.2)
|
13.3 (3.8)
|
0.803
|
MCP-1
|
106
|
379 (89.5)
|
350 (74.5)
|
394. (89.3)
|
0.194
|
MIP-1b
|
106
|
15.0 (4.7)
|
16.1 (6.6)
|
15.8 (4.7)
|
0.806
|
TARC
|
106
|
3.9 (1.9)
|
5.2 (4.4)
|
3.8 (1.7)d
|
0.029
|
IP-10
|
106
|
628 (358)e
|
786 (488)
|
633 (310)d
|
0.030
|
Vascular dysfunction
|
Flt-1
|
106
|
14.3 (5.9)
|
16.3 (6.9)
|
18.4 (7.6)
|
0.446
|
PlGF
|
106
|
8.5 (5.5)
|
9.9 (6.8)
|
12.3 (8.9)
|
0.767
|
VEGF
|
105
|
2.7 (0.65)
|
3.0 (1.1)
|
3.3 (1.5)
|
0.806
|
VEGF-D
|
106
|
14.9 (7.7)
|
15.6 (6.7)
|
14.9 (6.1)
|
0.463
|
CRP
|
106
|
3027 (3560)d
|
7974 (12896)
|
2934 (3812)c
|
0.002
|
ICAM-1
|
106
|
1790 (422)
|
2117 (752)
|
1943 (562)d
|
0.031
|
SAA
|
106
|
904 (1317)
|
1292 (1453)
|
893 (838)
|
0.107
|
VCAM-1
|
106
|
6004 (1391)
|
7026 (2501)
|
6250 (1804)f
|
0.007
|
Biomarker levels are shown as unadjusted mean (SD). P values are calculated from one-way ANCOVA adjusted for age and sex. P<0.05 was considered statistically significant. Abbreviations: AD, Alzheimer’s disease; CAR, cortisol awakening response; CRP c-reactive protein, Flt-1, fms-like tyrosine kinase 1; ICAM-1, intercellular adhesion molecule-1; IL-5, interleukin 5; IP-10, interferon g-inducible protein; MCI, mild cognitive impairment; MCP-1, monocyte chemoattractant protein-; MIP-1b, macrophage inflammatory protein 1b; NG, neurogranin; PlGF, placental growth factor; SAA, serum amyloid A; SCD, subjective cognitive decline; SNAP-25, synaptosomal associated protein 25; SYT-1, synaptotagmin 1; TARC, thymus and activation regulated chemokine; VCAM-1, vascular cell adhesion molecule-1; VEGF, vascular endothelial growth factor; YKL-40, Chitinase 3-like 1.
a p <0.001 versus SCD
b p <0.0001 versus MCI
c p <0.001 versus MCI
d p <0.05 versus MCI
e p <0.05 versus MCI
f p <0.01 versus MCI
Individual biomarker cross-correlations
Next, we investigated the bivariate correlations of neuroinflammatory, cerebrovascular and cortisol markers with the traditional AD and synaptic biomarkers in the total population. An overview of the results is presented as a correlation heatmap (Fig. 1). In brief, positive correlations were observed between p-tau, t-tau, synaptic (SNAP-25, NG, SYT-1) and several neuroinflammation (YKL-40, IL-5, IL-8, IL-15, IL-16) and vascular dysfunction markers (Flt-1, ICAM-1, VCAM-1). Ab42 correlated weakly but significantly with VCAM-1, IL-6 and CRP. CAR correlated negatively with IP-10, PlGF and tended to correlate negatively with IL-12/IL-23p40 (p=0.07). Finally, an inverse correlation of cortisol slope with YKL-40 was observed.
Component characteristics and distribution between diagnostic and amyloid pathology status groups
The quantity and variety of the bivariate correlations made the interpretation difficult. Thus, we performed PCA in the total cohort (N=101) to reduce the number of variables and obtain a few hypothetical components that could maximally explain the variance of the data. Only the more exploratory biomarkers (salivary cortisol, neuroinflammation and vascular dysfunction) were added in the analyses, as the traditional AD and synaptic markers would have masked the results. This led to the inclusion of 21 biomarkers, with a variable to subject ratio of ~1:5. Applying this method resulted into 6 principal components (PC), explaining cumulatively 68.9% of the variation. The first three components (PC1, PC2, PC3) described 29.1%, 10.2% and 9.9% of the variation respectively, while the remaining components’ (PC4, PC5 and PC6) contribution was 7.7%, 6.7% and 6.3% respectively. The factor loading matrix is presented in Supplementary Table 1. PC1 correlated the most with Flt-1, IL-5, VCAM-1, IL-15, YKL-40, ICAM-1, VEGF-D and IL-16, PC2 with IP-10, IL-12/IL-23p40 and TARC, PC3 with CRP, SAA and MIP-1b, PC4 with MCP-1, PlGF, IL-8 and IL-6, PC5 with cortisol slope, IL-6 and CAR and PC6 with VEGF and IL-16.
Plotting the coordinates of the observations of the two first components could not visually discriminate either the diagnostic groups (Supplementary Fig. 2A) or the amyloid positive vs negative individuals (Supplementary Fig. 2B). Next, we explored each component's distribution among diagnostic and amyloid status groups. PC5 was found to be significantly lower in AD compared to SCD and MCI participants (p=0.012) (Fig. 2E) and tended to be decreased in the amyloid positive group (p=0.089) (Supplementary Fig. 3E). PC2 was also decreased in subjects positive for amyloid pathology (p=0.011) (Supplementary Fig. 3B). No other differences could be observed for the rest of the components.
Associations of the principal components with CSF biomarkers of Alzheimer’s disease pathology and synaptic damage
We next investigated possible associations of each component with CSF markers of AD pathology and synaptic damage in the total cohort, by linear regression models (Supplementary Table 2). All models were adjusted for age, sex and diagnosis. PC1 associated with t-tau (b=0.58, p<0.0001), p-tau (b=0.64, p<0.0001) and all markers of synaptic dysfunction (b=0.68, p<0.0001 for SNAP-25, b=0.66, p<0.0001 for NG and b=0.802, p<0.0001 for SYT-1) (Supplementary Table 2 and Fig. 3A-E). In addition, PC2 and PC3 associated positively with Ab42 levels (b=0.17, p=0.023 for PC2 and b=0.22, p<0.001 for PC3) (Supplementary Table 2 and Fig. 3F-G). A trend was observed for a positive association of PC1 with Ab42 levels (b=0.16, p=0.056).
Next, we conducted the same analysis after stratifying participants based on their amyloid status. The associations of PC1 with tau and synaptic markers remained significant irrespective of amyloid status (Supplementary Table 3). Interestingly, PC1 and PC3 were associated with increased Ab42 levels in the group negative to amyloid b pathology (b=0.35, p=0.014 for PC1 and b=0.25, p=0.044 for PC3) but not in amyloid positive participants. Finally, no association could be observed for PC2 and Ab42 levels in any of the two groups of amyloid b pathology.
Associations of principal components with cognition, depression and perceived stress
In a next step we sought to investigate associations between components and memory, processing speed, depression and perceived stress. To this end, we performed separate linear regression analyses in the total population with each assessment as the outcome measure and each of the principal components as the regressor, adjusting for age, sex, education and diagnosis. PC4 was associated with a worsened processing speed (b=-0.21, p=0.047). No other associations were observed for the rest of the components (Table 3).
Table 3. Associations of principal components with neuropsychological assessments.
|
Memory
b (p)
|
Processing speed b (p)
|
PSS
b (p)
|
GDS
b (p)
|
PC1
|
0.11
(0.164)
|
0.07
(0.521)
|
0.03
(0.791)
|
0.03
(0.793)
|
PC2
|
-0.12
(0.108)
|
0.00
(0.989)
|
-0.04
(0.706)
|
-0.02
(0.844)
|
PC3
|
0.00
(0.958)
|
-0.07
(0.492)
|
0.07
(0.531)
|
0.12
(0.227)
|
PC4
|
-0.11
(0.164)
|
-0.21
(0.047)
|
0.07
(0.545)
|
0.03
(0.752)
|
PC5
|
0.10
(0.157)
|
-0.14
(0.156)
|
0.01
(0.957)
|
0.01
(0.934)
|
PC6
|
0.01
(0.883)
|
0.03
(0.778)
|
0.02
(0.835)
|
0.06
(0.527)
|
Results are from separate linear regression models with memory, processing speed, GDS or PSS as outcome measure and each component as regressor. Data are shown as standardized b coefficients (p values), after age, sex, education and diagnosis adjustment. P<0.05 was considered statistically significant. Abbreviations: GDS, geriatric depression scale; PC1, principal component 1; PSS, perceived stress scale.
When stratifying for amyloid b pathology status, a trend for PC4 and lower processing speed was observed in the amyloid positive group only (b=-0.40, p=0.064) (Supplementary Table 4). Further, PC2 was related with worsened memory only in amyloid positive participants (b=-0.47, p=0.012). An association of PC3 with depressive symptoms was also observed in the amyloid positive group (b=0.45, p=0.022). No other relationships could be seen for the rest of the analyses.