Patient characteristics
We recruited 25 patients with dementia (Alzheimer type and mixed type) and 18 matched controls without cognitive impairment in this prospective controlled cohort study. From 2 dementia patients we were not able to collect enough stool and blood samples to do the intended analyses; therefore, they were excluded from the final analysis. (Figure 1) Dementia patients had a lower body mass index (BMI) and erythrocyte count as well as lower serum albumin and total protein levels compared to controls. Accordingly, nutritional status according to MNA-SF was significantly worse in dementia patients. Within the dementia group, erythrocyte count (r=0.669, p=0.002) and albumin (r=0.707, p<0.002) showed a significant positive correlation with MMSE and clock drawing test showed a weak positive correlation with albumin (r=0.485, p=0.019). No significant differences were found regarding age, gender, and other routine biochemistry parameters. BMI did not correlate with MMSE or clock drawing test results. Collinearity analysis showed variance inflation factors (VIF) below 2 for MMSE, clock-drawing test, BMI, albumin and MNA-SF.
Prescription drug intake was significantly different between dementia patients and controls. Dementia patients took three times more prescription drugs compared to controls. Antidepressants, laxatives, opioids, anti-dementia drugs, sedatives, vitamin D and metamizole were prescribed nearly exclusively in the dementia group, whereas proton pump inhibitors (PPI), antihypertensive drugs, statins, nonsteroidal anti-inflammatory drugs (NSAIDS), paracetamol, antidiabetics, thyroid hormones, calcium and magnesium supplements, anticoagulation and phytotherapeutics were equally prescribed for dementia patients and age matched controls. Laxatives, sedatives, metamizole and paracetamol were usually prescribed as needed, whereas the other drugs were prescribed as fixed dose medication. Patient characteristics are shown in table 1.
Table 1: Patient characteristics. Data are given as median and 95% confidence interval unless stated otherwise
|
Dementia patients (n=23)
|
Controls (n=18)
|
p-value
|
Age (years)
|
88 (73;85)
|
75 (74;76)
|
n.s.
|
Gender (f/m) (n)
|
15/8
|
11/7
|
n.s.
|
BMI (kg/m2)
|
24.8 (22.6; 25.9)
|
28.1 (25.2; 31.0)
|
p=0.028
|
MMSE
|
16 (13;21)
|
29 (30;30)
|
p<0.0001
|
Clock drawing test
|
3 (0;5)
|
7 (7;9)
|
p<0.0001
|
Number of prescription drugs
|
9 (6;11)
|
3 (1;4)
|
p<0.0001
|
MNA-SF
|
10 (9;12)
|
14 (14;14)
|
p<0.0001
|
Leukocytes (109/L)
|
6.6 (6.2;8.3)
|
6.1 (5.4;7.5)
|
n.s
|
Erythrocytes (1012/L)
|
4.5 (4.0;4.7)
|
4.7 (4.4; 5.1)
|
p=0.028
|
Thrombocytes (109/L)
|
220 (216;248)
|
216 (205;222)
|
n.s
|
Hemoglobin g/dL
|
13.2 (12.7;14.4)
|
14.1 (13.3;14.7)
|
n.s
|
Creatinine (mg/dL)
|
0.9 (0.8;1.0)
|
1.0 (0.9;1.1)
|
n.s
|
Bilirubin (mg/dL)
|
0.6 (0.5;0.9)
|
0.6 (0.5;0.6)
|
n.s
|
Albumin (g/dL)
|
3.9 (3.7; 4.1)
|
4.2 (4.1;4.4)
|
p=0.006
|
total protein (g/dL)
|
7.0 (6.8;7.3)
|
7.5 (7.3;7.6)
|
p=0.014
|
CRP (mg/l)
|
5 (3;11)
|
2 (1.2; 3.4)
|
n.s.
|
BMI: body mass index, MMSE: Mini mental state examination, MNA-SF: Mini Nutritional Assessment Short Form, CRP: C reactive protein
Gut microbiome composition
Alpha diversity using Chao 1 index (Figure 2A), Simpson reciprocal index (Supplementary figure S1A) or Faith phylogenetic diversity (Supplementary figure S1C) was not significantly different in dementia patients compared to age matched controls. Redundancy Analysis (RDA) showed clear clustering of dementia patients compared to controls (Variance 34.3, F=1.31 p=0.003). (Figure 2B) Alpha diversity also did not change significantly with increasing degree of dementia. (Figure 2C and supplementary figure S1B and D) RDA showed clear clustering of different stages of dementia (Variance 94.7, F=1.2 p=0.001). (Figure 2 D) Linear discriminant analysis of effect size (LEfSe) identified one family, 5 genera and 7 features to differ between patients with dementia and controls. For example, the features Clostridium clostridioforme, Anaerostipes hadrus and Bacteroides dorei were associated with dementia; Lachnospiraceae bacterium MC-35, another Lachnospiraceae sp., and the genus Lachnospiraceae NK4A136 group were associated with health. (Figure 3A) Analysis of Composition of Microbiomes (ANCOM) confirmed that from the taxa identified by LEfSe to discriminate between dementia and control, one uncultured Lachnospiraceae feature as well as the genus Lachnospiraceae NK4A136 group were significantly less abundant in stool of dementia patients. Additionally, the feature Eubacterium rectale was also less abundant in stool of dementia patients (Figure 3B). When looking at different stages of dementia, LEfSe identified one class, 3 orders, 3 families, 18 genera and 20 features, being associated with severity of cognitive impairment. Most notably, three Lachnospiraceae species with the corresponding genus Lachnospiraceae NK4A136 group and the genus Lachnospira were associated with health; Faecalibacterium prausnitzii was associated with mild dementia; moderate dementia was associated with Lactobacillus amylovorus and the corresponding higher taxonomic levels (the genus Lactobacillus, the family Lactobacillaceae and the order Lactobacillales). Severe dementia was associated with several potential pathogens (e.g. Clostridium clostridiforme, Streptococcus salivarius) (Figure 4A) From these discriminating taxa, ANCOM analysis identified the feature C. clostridioforme and the genus Eisenbergiella to increase with severity of cognitive impairment and the family Lactobacillaceae to be highest in patients with moderate cognitive impairment. (Figure 4B)
Association of drug intake and nutrition on microbiome composition
While some drugs were nearly exclusively prescribed in dementia patients (for details see supplementary table S1), other drugs were equally prescribed between dementia and control subjects. To understand how drug use may influence microbiome composition irrespective of the disease, we studied the effect of drugs that were equally prescribed in dementia patients and healthy controls on diversity measures and taxonomic composition: namely PPI, antihypertensive drugs, statins, thyroid hormones and NSAIDS. Paracetamol, antidiabetics and calcium and magnesium supplements were taken by less than 15% of the cohort and therefore these drugs were not included into the analysis. None of the drugs influenced alpha diversity (Chao1, Simpson, Faith phylogenetic diversity). Statins, but none of the other drugs had a significant impact on beta diversity (RDA, Variance 35.58, F 1.36, p=0.003). LEfSe identified several features, genera, families, orders and classes being associated with use or non-use of each drug. ANCOM identified several of these taxa as well as some taxa that were not discriminative on LEfSe to be differentially abundant taxa between drug user and non-user. For details see supplementary tables S2-6. Interestingly, PPI use was associated with increased abundance of oral bacteria (e.g. Streptococcus salivarius) whereas statin and antihypertensive drug use was associated with increased abundance of bacteria known to produce butyrate (e.g. Faecalibacterium sp.).
Since malnutrition was present in 74% of dementia patients but in none of the control persons, microbiome composition in malnourished versus non-malnourished patients was very similar to the results obtained when comparing dementia versus controls. LEfSe identified the feature Ruminococcaceae UCG-014 sp with the corresponding genus Ruminococcaceae UCG014 and the genus Lachnospiraceae NK4A136 group to be associated with normal nutritional state. These taxa were also found to be associated with healthy controls. The genus Eubacterium hallii group was associated with dementia. (supplementary table S7) ANCOM confirmed the feature Ruminococcaceae UCG-014 sp. and the genus Lachnospiraceae NK4A136 group to be differentially abundant between malnutrition and normal nutrition.
Gut barrier dysfunction, inflammation and bacterial translocation
We assessed intestinal permeability by serum diaminooxidase (DAO) and fecal zonulin; inflammation by C-reactive protein, serum lipopolysaccharide binding protein (LBP), soluble CD 14 (sCD14) and fecal calprotectin as well as bacterial translocation by endotoxin, peptidoglycanes and bacterial DNA in serum. Patients with dementia had higher DAO levels and sCD14 levels, indicative for an association with increased gut permeability and increased endotoxin load. (Table 2)
PPI use was associated with significantly increased faecal calprotectin levels (PPI use: 92.5 ng/ml (50.2; 120.5); PPI non-use: 28.1 ng/ml (20.8; 47.9); p=0.008). Antihypertensive use was associated with significantly increased CRP levels (antihypertensive use: 6 mg/dl (3; 11); antihypertensive non-use 1.3 mg/dl (1;4); p=0.016), suggesting complex relations between disease, drug use and inflammation.
Table 2: Biomarker for gut barrier dysfunction, inflammation and bacterial translocation, Data are shown as median and 95% confidence interval
|
Dementia patients (n=23)
|
Controls (n=18)
|
p-value
|
serum diaminooxidase (U/ml)
|
20.8 (9.7;29)
|
11.2 (8.4; 13.8)
|
0.025
|
fecal zonulin (ng/ml)
|
33.8 (31.2; 57)
|
55.1 (40.8; 76.7)
|
n.s
|
C-reactive protein (mg/L)
|
5 (4; 11)
|
2 (1.2;3.4)
|
n.s
|
serum lipopolysaccharide binding protein (µg/ml)
|
17.9 (16.1; 18.6)
|
20.0 (14.6; 21.3)
|
n.s
|
soluble CD 14 (µg/ml)
|
2.4 (1.9; 3.1)
|
1.8 (1.7; 2.1)
|
0.022
|
fecal calprotectin (ng/ml)
|
31.5 (26.6; 85.8)
|
49.0 (18.2; 66.3)
|
n.s
|
endotoxin (EU/ml)
|
0.26 (0.0; 0.33)
|
0.25 (0.09; 0.53)
|
n.s
|
peptidoglycan* (ng/mL)
|
0.96 (0.26; 1.66)
|
0.42 (0.30;1.05)
|
n.s.
|
bacterial DNA (µM)
|
0.06 (0.00;1.46)
|
0.7 (0.0; 1.29)
|
n.s
|
*peptidoglycan was only measurable in 12% of the samples, therefore median and confidence interval only for the positive samples are shown. CD: cluster of differentiation, EU: endotoxin units
Multivariate and network analysis of potential factors influencing microbiome composition in dementia
To understand the main drivers of dysbiosis in dementia we further performed univariate and multivariate RDA to assess the association of clinical variables and biomarkers with microbiome composition. RDA showed that BMI, albumin, total protein, sCD14, statins, NSAIDs, number of drugs, MNA-SF, MMSE, clock-drawing test, sex, number of drugs were explanatory variables for microbiome composition in controls compared to dementia and between different stages of cognitive dysfunction (p<0.1) (supplementary table S8). To the final multivariate RDA model explanatory variables with p<0.1 in the univariate analysis were included and variables with VIF>2 in multicollinearity analysis were excluded. (Table 3) In the multivariate model BMI and statin use were the remaining significant explanatory variables for differences in microbiome composition between dementia and control groups and between the groups of dementia severity in the dementia group only. (Table 3) Network analysis also illustrates the overlap between factors influencing microbiome composition: Genera associated with dementia (red) overlap with genera associated with no statin intake (yellow) and BMI (green), whereas genera associated with health (blue) overlap with genera associated with statin intake (purple). (Figure 5A) When performing network analysis in the subgroup of dementia patients only, the overlaps are less clear, but again genera associated with severe dementia (red) overlap with genera associated with no statin intake (yellow) and genera associated with mild dementia (blue) overlap with genera associated with statin intake (purple). The association with BMI is less pronounced in the dementia subgroup. (Figure 5B)
Table 3: Multivariate RDA to identify the most important explanatory variables for microbiome composition changes.
Variable
|
Control versus Dementia
|
Severity of dementia
|
BMI
|
Variance = 33.18
F = 1.29
P = 0.006
|
Variance = 33.18
F = 1.29
P = 0.008
|
Total protein
|
Variance = 29.19
F = 1.14
P = 0.067
|
Variance = 29.19
F = 1.14
P = 0.070
|
soluble CD14
|
Variance = 29.06
F = 1.13
P = 0.072
|
Variance = 29.06
F = 1.13
P = 0.079
|
Statins
|
Variance = 32.06
F = 1.25
P = 0.009
|
Variance = 32.06
F = 1.25
P = 0.014
|
Clock-drawing test
|
Variance = 25.89
F = 1.01
P = 0.376
|
Variance = 25.89
F = 1.01
P = 0.374
|
Age
|
Variance = 25.88
F = 1.01
P = 0.427
|
Variance = 25.88
F = 1.01
P = 0.409
|
Sex
|
Variance = 27.63
F = 1.08
P = 0.137
|
Variance = 27.63
F = 1.08
P = 0.154
|
NSAIDS
|
Variance = 28.17
F = 1.10
P = 0.098
|
Variance = 28.17
F = 1.10
P = 0.107
|
BMI body mass index, NSAIDS non-steroidal anti-inflammatory drugs