Evidence supportive of a bacterial component in the etiology for Alzheimer's disease and for a temporal-spatial development of a pathogenic microbiome in the brain

Abstract

Background: Over the last few decades, a growing body of evidence has suggested a role for various infectious agents in Alzheimer's disease (AD) pathogenesis. Despite diverse pathogens (virus, bacteria, fungi) being detected in AD subjects' brains, research has focused on individual pathogens and only a few studies investigated the hypothesis of a bacterial brain microbiome. We profiled the bacterial communities present in non-demented controls and AD subjects' brains.

Results: We obtained postmortem samples from the brains of 32 individual subjects, comprising 16 AD and 16 control age-matched subjects with a total of 130 samples from the frontal and temporal lobes and the entorhinal cortex. We used full-length 16S rRNA gene amplification with Pacific Biosciences sequencing technology to identify bacteria. We detected bacteria in the brains of both cohorts with the principal bacteria comprising Cutibacterium acnes (formerly Propionibacterium acnes) and two species each of Acinetobacter and Comamonas genera. We used a hierarchical Bayesian method to detect differences in relative abundance among AD and control groups. Because of large abundance variances, we also employed a new analysis approach based on the Latent Dirichlet Allocation algorithm, used in computational linguistics. This allowed us to identify five sample classes, each revealing a different microbiota. Assuming that samples represented infections that began at different times, we ordered these classes in time, finding that the last class exclusively explained the existence or non-existence of AD.

Conclusions: The AD-related pathogenicity of the brain microbiome seems to be based on a complex polymicrobial dynamic. The time ordering revealed a rise and fall of the abundance of C. acnes with pathogenicity occurring for an off-peak abundance level in association with at least one other bacterium from a set of genera that included Methylobacterium, Bacillus, Caulobacter, Delftia, and Variovorax. C. acnes may also be involved with outcompeting the Comamonas species, which were strongly associated with non-demented brain microbiota, whose early destruction could be the first stage of disease. Our results are also consistent with a leaky blood-brain barrier or lymphatic network that allows bacteria, viruses, fungi, or other pathogens to enter the brain.

Keywords: 16s sequencing; Alzheimer’s disease; Bayesian; Cutibacterium; blood brain barrier; glymphatic network; latent dirichlet allocation; microbiome.

Figure 1

Cartoon of the LDA algorithm.

Figure 2

PCA was performed on clr-transformed composition. (A) Scree plot for the PCA ordination. Each bar corresponds to one axis of the PCA; the height is proportional to the amount of variance explained by that axis. (B) PCA ordination plot. Each colored point represents a sample. Points are colored by diagnosis and shaped by biopsy location (EC: entorhinal cortex, F: frontal lobe and T: temporal lobe).

Figure 3

Heatmap that represents the clr-transformed OTU counts within each sample of the 80 most variable OTUs (higher relative abundance corresponds to darker colors). Unsupervised grouping of samples with similar OTU composition (columns) and OTUs with similar abundance across samples (vertical) into clusters was achieved by hierarchical clustering using the Euclidean distance between clr-transformed compositions. The sample’s subjects, biopsy brain locations and diagnosis are indicated by the vertical colored strips. AD, Alzheimer’s disease; C, controls; EC, entorhinal cortex; F, frontal lobe; T, temporal lobe.

Figure 4

Differences in relative abundance between the Alzheimer’s disease (AD) group and the age-matched control group. The relative abundances were estimated for each OTU from each group through Dirichlet-multinomial modelling. The vertical axis shows the estimated differences in the relative abundance of each OTU between the AD and control groups. Points are the means of the posterior probability distribution of differences (PPD) and the whiskers show the 95% equal tail probability intervals of PPD (see Materials and Methods and Supplement).

Figure 5

Type I graph. Results from summation of five runs. Nodes are samples. Colors are maximum classes. Principal bacterial genera and abundance levels indicated for each color. The inset contains the percentage of samples that come from AD subjects by class, called AD statistics.

Figure 6

Graphs showing samples with abundances levels 12–14 with enlarged nodes for three main microbes.

Figure 7

Several definitions of M+ compared to C. acnes (11–13). (A) Objects with level 14 that occur two or more times in any class, (B) objects of level 14 that occur two or more times in magenta or red, (C) objects of level 14 that occur two or more times and their corresponding objects of level 13 in magenta or red, and (D) C. acnes (11-13).

Figure 8

This is the same as Figure 2 , but the nodes are colored by the class colors.

Figure 9

Color pair relationships. (A) Green-Orange: A_j -(13-14), (B) Orange-Blue and Green-Blue: A_j,-(9-13), (C) Blue- Red: Cu_a-14, (D) Blue-Red: Cu_a-(13-14), (E) Red-Magenta: Cu_a-(11-13), (F) Red- Magenta: M+ -(13-14).

Figure 10

Temporal network of classes.

Figure 11

Color class of samples by subject (F: frontal lobe, T: temporal lobe, E: entorhinal cortex).

Figure 12

This figure shows possible ecosystem structure at the microscopic scale. The arrows roughly indicate the human cellular scale. Scenario 1 suggests ecosystems dominated by one principal bacterium predominate around a particular cell, whereas scenario 3 suggests that ecosystems composed of multiple principal bacteria predominate around a particular cell. A physical sample would comprise all or large fractions of the above arrays.

Figure 13

Idealized depiction of distribution of ecosystems. Each dot is an ecosystem dominated by a particular species: blue for C. acnes, orange for A. junii, green for C. jiangduensis, and magenta for M+. The large circles are class mixtures also labeled by colors. The small black circles depict samples. Green is dominated by C. jiangduensis; blue is dominated by C. acnes; orange by A. junii, and magenta by the M+ set.

Figure 14

Each array represents a large area of the brain. Each element is a single class mixture like the ones from Figure 13 . The size of the element could be from centimeters to several centimeters. The left-hand side produces statistics like the inset below (except orange). The right side produces flatter statistics without 1s, like orange. The inset contains the number of class occurrences by subjects for each class for comparison with simulation of macroscopic distribution scenarios.