Identification of Chlamydia pneumoniae and NLRP3 inflammasome activation in Alzheimer's disease retina

Abstract

Chlamydia pneumoniae is an intracellular bacterium implicated in Alzheimer's disease (AD), but its role in retinal pathology and disease progression is unclear. Here we identify Chlamydia pneumoniae inclusions in the retina, showing higher burden in AD retina and brain, increasing with APOEε4, disease stage, and cognitive deficit. Retinal and cortical proteomics reveal bacterial-infection and related NLRP3-inflammasome pathways. Retinal NLRP3 is elevated in mild cognitive impairment and activated in AD dementia, evidenced by increased caspase-1, cleaved interleukin-1β, and cleaved N-terminal gasdermin-D. Chlamydia pneumoniae associates with amyloid-β₄₂, inflammation, apoptosis, pyroptosis, and AD status. In neuronal cultures and APP_SWE/PS1_ΔE9 model mice, infection induces amyloid-β, inflammasome activation, neuroinflammation, and neurotoxicity, and chronic infection worsens cognition. Fewer pathogen-colocalized microglia are found in AD retinas, implying impaired clearance. Machine learning detects retinal Chlamydia pneumoniae or NLRP3, combined with amyloid-β₄₂, as predictors of AD diagnosis and stage. These findings support a disease-amplifying role for Chlamydia pneumoniae and propose NLRP3-attenuation or antibiotic-based early interventions.

Fig. 1. Identification of Chlamydia pneumoniae inclusions in retinas and brains from MCI and AD patients and correlations with disease status.

A Schematic of Retinal isolation and cross-section preparation. Red puncta (red) indicate putative Chlamydia pneumoniae (Cpn) inclusions. Analyses workflow summarizes cohort sizes. Numbers in parentheses indicate brain subsets for Cpn histology. B Immunofluorescence using anti-Cpn polyclonal antibody (pAb, green, white arrow; 3 repetitions) in 3 AD versus 3 normal cognition (NC) retinas. C Peroxidase-based immunohistochemistry using anti-Cpn monoclonal antibody (mAb, brown, red arrows) in 6 MCI and 9 AD versus 3 NC retinas; hematoxylin (purple) and IgG negative control (3 repetitions). D Retinal Cpn mAb immunofluorescence (red; white arrows, 3-repetition); cytosolic inclusions at higher magnification. E Quantification of retinal Cpn⁺ cell count (21 NC, 14 MCI, 34 AD). F Quantification of Cpn immunoreactivity (%IR area) in same retinal cohort (21 NC, 14 MCI, 34 AD) and paired brain tissues (5 NC, 5 MCI, 6 AD). Subjects with %IR area above the NC mean (red line) were classified as Cpn-positive. G Giemsa staining visualizes retinal Cpn-like inclusions (dark-blue, red arrows; BV, blood vessel; n = 8 donors, 4 repetitions). H FISH confirms retinal Cpn-specific genomic material (green, white arrow; n = 11 donors, 3 repetitions). I qPCR curves (Ct values) for the Cpn argR gene in 2 NC, 1 MCI, and 2 AD retinas. Pearson’s correlation (r_P) between retinal Cpn and (J) brain Cpn, (K) retinal Aβ₄₂, or (L) PHF-tau. M Spearman’s correlation (r_s) between retinal Cpn and brain NFT severity. Retinal Cpn stratified by (N) Braak stage (n = 12 Braak 0–II, 17 Braak III–IV, 31 Braak V–VI), (O) APOEɛ4 genotype (n = 12 carriers, 25 non-carriers), and (P) MMSE score (n = 26 MMSE ≥ 24, 9 MMSE 17–23, 16 MMSE ≤ 16). Q Correlation (r_s) between retinal Cpn burden and CDR score. M-male, F-female, and age (y-years) are shown). All scale bars, 10 μm. Data are shown as individual values with means ± SEMs. Fold changes are shown in red. p values were determined by one-way ANOVA with Tukey’s post-hoc test (E, F, N, P), two-sided unpaired t-test (P), or Mann–Whitney U-test (O). Illustration A was created in Biorender.com. Fuchs, D. (2026) https://BioRender.com/msadzfx. Source data are provided as a Source Data file.

Fig. 2. Bacterial infection-associated proteome pathways in AD retina and cerebral cortex.

Gene ontology (GO) analysis of differentially expressed proteins (DEPs) related to bacterial infection (A) in the cerebral temporal cortex and (B) in the temporal hemi retina from 2 separate cohorts of human donors with AD (n = 10 brains, 6 retinas) versus NC (n = 8 brains, 6 retinas). The analysis was carried out in Metascape and included the Reactome, Kyoto Encyclopedia of Genes and Genomes (KEGG), and WikiPathways (WikiPath) databases. Red arrows indicate the shared pathways between brain and retina. Bar and symbol graphs represent z-scores, and Benjamini-Hochberg adjusted p-values from Metascape analysis, respectively. Range of p-values is presented as color-coded symbols. Volcano plots display the fold changes [log₂(FC)] and significance level [-log₁₀(p)] by two-sided t-test in the (C) cortex and (D) retina of AD versus NC subjects for Chlamydia inclusion interactome. Top 10 DEPs by FC upregulated (orange) and downregulated (purple) interactors are shown. The highlighted proteins, five downregulated and five upregulated, were found in both tissues. E GO network (Metascape) of enriched retinal pathways related to bacterial infection, immune response cell death, and mitochondrial collapse. The size of the nodes represents the number of DEPs, with the inner ring showing the proportion of these DEPs that are downregulated (purple) or upregulated (orange) in AD. The thickness of the green border represents the number of DEPs that interact with Chlamydia inclusion. The thickness of the connection lines between nodes represents the shared DEPs (association score) between pathways. F Heatmaps of upregulated (orange) and downregulated (purple) DEPs [-log₁₀(p) by two-sided t-test and FC] normalized by unit variance scaling and generated in ClustVis in AD versus NC retina for selected pathways. Only proteins connected to gram-negative bacterial infection (Metascape analysis) and Chlamydia infection (Chlamydia interactome and literature) are shown for each pathway. Clustering of DEPs was carried out manually based on their involvement in select pathways for visual clarity. Source data are provided as a Source Data file.

Fig. 3. Effects of Chlamydia pneumoniae infection in SH-SY5Y cells and AD⁺ mice.

A Schematic of Cpn infection in SH-SY5Y cells. B Immunofluorescence of SH-SY5Y cells ± Cpn showing IL1β (green), Cpn (red), NLRP3 or Aβ₄₂ (H31L21, white). C Immunoreactive area quantification in SH-SY5Y cells for NLRP3 (−Cpn = 41, +Cpn = 43), IL1β (−Cpn = 74, +Cpn = 61), and Aβ₄₂ (−Cpn=44, +Cpn=63). D Quantification of LDH release in SH-SY5Y cells culture medium ± Cpn (n = 6 wells/group). E, F Western blot (WB) gels and densitometric analysis of NLRP3 (n = 3/group), IL1β (n = 3/group), NGSDMD (n = 3/group), and 12F4-Aβ₄₂ (n = 6/group). G Schematic of acute and long-term intranasal Cpn infection (1 × 10⁶ inclusion-forming units, IFU) in APP_SWE/PS1_ΔE9 (AD⁺) mice. H Cerebral Cpn load at 7 days post-infection, quantified by live Cpn growth ((IFU; −Cpn=5, +Cpn=9) and by qPCR (Cpn DNA copy number; −Cpn=5, +Cpn=7). I Immunofluorescence of mouse hippocampi ± Cpn infection, showing IBA1 (microglia, green) and GFAP (astrocytes, magenta). J Quantification of IBA1 and GFAP immunoreactivity in mouse hippocampi (−Cpn=5, +Cpn=14). K Cerebral mRNA expression levels of Il6 (−Cpn=6, +Cpn=7), Il1β (−Cpn=8, +Cpn=7, and Nlrp3 (−Cpn=8, +Cpn=7) in AD⁺ mice. L–P Behavioral tests in AD⁺ mice 6-month Cpn post-infection or PBS administration (WT = 6, AD⁺-Cp=8, AD⁺+Cp=9). L Locomotor activity in open field test. M, N Percent alternations and transitions in color-mode or contrast-mode X-maze. O, P Number of errors at 1–4-day acquisition phase, day-7 memory retention, the 8–9-day reversal phase, and day-9 search coverage, in Barnes maze. Q Immunofluorescence of cortical and hippocampal Aβ (6E10, green), IBA1 (red), and GFAP (white). R, S Quantification of cortical and hippocampal 6E10, IBA1, and GFAP immunoreactive area (n = 7/group). Scale bars, 50 μm (I) and 20 µm (Q). Individual data points and group mean ± SEMs are shown. Violin plots display median, upper, and lower quartiles. Fold or percent changes are in red. p values by one-way ANOVA and Tukey’s or Fisher’s LSD post-hoc test (L–P), two-sided unpaired t- test (C, D, F, R, S), or Mann–Whitney U test (H, J, K). Illustrations A and G created in Biorender.com. Fuchs, D. (2026) https://BioRender.com/lj8g4yb and https://BioRender.com/k77gi93. Source data are provided as a Source Data file.

Fig. 4. Retinal NLRP3 inflammasome, pyroptotic, and apoptotic markers and associations with Chlamydia pneumoniae infection in early and advanced AD.

A, B Immunofluorescence images of retinal cross-sections from AD and MCI donors compared with NC controls, showing NLRP3 inflammasome activation markers: A NLRP3 (green), Caspase-1 (Casp1, red), and nuclei (blue); B ASC (green), Cpn inclusions (red), and nuclei (blue). Quantification of retinal percentage immunoreactive (IR) area for C Casp1, D ASC, and E NLRP3 in 8 NC, 8 MCI, and 11 AD donors. F Representative images of retinal cross-sections stained for the pyroptotic marker N-terminal cleaved gasdermin D (NGSDMD, green), Cpn (red), and nuclei (blue). G Quantification of retinal NGSDMD %IR area in 8 NC, 8 MCI, and 11 AD donors. H Representative images of retinal cross-sections stained for early-apoptotic marker cleaved caspase-3 (CCasp3⁺, green), Cpn (red), and nuclei (blue). I Quantification of CCasp3 %IR area in a subset of donors with 7 NC, 7 MCI, and 11 AD. J Gaussian distribution curves and Pearson’s correlation (r_p) analyses presented as scatter plots with adjusted p values, showing relationships among retinal markers: Cpn, NLRP3, Casp1, ASC, CCasp3-apoptosis, and NGSDMD-pyroptosis. K Western blot gel images and quantification of pro- and mature IL1β in retinal homogenates with band intensity normalized to GAPDH. [(4 NC, 1 MCI (green dot) and 5 AD (magenta dots)]. L Schematic summary of correlation strength (r_p) among three categories: (1) inflammasome activators (Cpn, Aβ₄₂, oligomeric tau); (2) active NLRP3-inflammasome markers (NLRP3, Casp1, ASC); and (3) cell-death markers (CCasp3, NGSDMD). M Heatmap showing pairwise Pearson’s correlations (r_p) between retinal Cpn–related markers and retinal atrophy, and Spearman’s correlations (r_s) with brain atrophy, Braak stage, and MMSE score. Stars indicate significance based on unadjusted p values; middle-row numbers show r_p or r_s, and lower-row values indicate sample sizes. All scale bars, 25 μm. Data are shown as individual values with group means ± SEMs. Fold increase or percent changes are marked in red. *p  <  0.05, **p  <  0.01, ***p  <  0.001, ****p  <  0.0001, by one-way ANOVA with Tukey’s post hoc test (C, D, E, G, I), two-sided unpaired t-test (K), or pairwise Pearson’s/Spearman’s correlation analyses (M). Source data are provided as a Source Data file.

Fig. 5. Chlamydia pneumoniae-associated glial activation and phagocytosis in MCI and AD retina.

A Representative immunofluorescence images of retinal cross-sections from AD and MCI donors versus normal cognition (NC) controls stained for macrogliosis using GFAP (green), Cpn (mAb, red), and nuclei (blue). The high-magnification image show Cpn inclusions within an astrocyte. B, C Quantification of retinal GFAP⁺ cell count and percentage immunoreactive (IR) area in 8 NC, 10 MCI, and 10 AD donors. D Pearson’s correlation (r_p) between retinal Cpn and GFAP %IR area. E Quantification of retinal vimentin IR area in 6 NC, 8 MCI, and 7 AD donors. F Pearson’s correlation (r_p) between retinal Cpn %IR area and vimentin IR area in the same cohort. G Immunofluorescence images of retinas from NC, MCI, and AD donors stained for microglial marker IBA1 (green), Cpn (red), and DAPI (blue). H, I Quantification of retinal IBA1⁺ cell counts and %IR area in 9 NC, 9 MCI, and 14 AD donors. J Pearson’s correlation between retinal Cpn and IBA1 %IR area. K Representative images showing three stages of microglial involvement in phagocytosis of Cpn-infected cells: recognition, engulfment, and ingestion. L, M Quantification of Cpn-associated IBA1⁺ cells (%) and retinal Cpn-associated IBA1⁺ cells relative to Cpn load (%/Cpn) in 15 NC, 12 MCI, and 17 AD donors. N Immunofluorescence images showing involvement of TMEM119⁺IBA1⁺-microglia in Cpn phagocytosis (n = 8 donors, 3 repetitions): TMEM119 (green), Cpn (red), IBA1 (blue), and nuclei (grey). O Pearson’s correlation analysis between retinal Cpn load and Cpn-associating microglia. P Heatmap illustrating Pearson’s correlations between retinal gliosis and retinal Aβ₄₂, oligo-tau, inflammasome, and cell death markers. Spearman’s correlations (r_s) of retinal gliosis with Braak stage and MMSE score. Stars denote significance (unadjusted p values), middle-row numbers show r_p or r_s, and lower-row values indicate sample sizes. Scale bars, 25 µm (A, G), 10 µm (K, N). Data are shown as individual values with group means ± SEM. Fold or % changes are shown in red. *p  <  0.05, **p  <  0.01, ***p  <  0.001, ****p  <  0.0001, by one-way ANOVA with Tukey’s post hoc test (B, C, E, H, I, L, M). Source data are provided as a Source Data file.

Fig. 6. Multivariable predictions of brain AD pathology and cognitive dysfunction conferred by retinal Chlamydia pneumoniae, NLRP3, CCasp3, and Aβ₄₂ markers.

Random forest regressor using 80 estimators was trained on the data to predict several brain pathologies. Box plots show the spread of results in quartiles with the same number of samples; the mean is displayed above each box. including A ABC average (n = 24/25 for train/validation), B Braak stage (n = 24/25 for train/validation), and C mini-mental state examination (MMSE) score (n = 20/20 for train/validation). The distributions show the spread of models trained on different folds of the 5-repeated 2-fold cross-validation. Only models performing with a variance coefficient of determination r² > 0.15 (gray dotted line) were retained. D Box plots representing the AUC measure for retinal Cpn for each diagnostic groups NC, MCI, and AD. For each model, AUC was measured (n = 28/28 for train/validation) using features either individually or combined with retinal Aβ₄₂, retinal gliosis (IBA1, GFAP, and Vimentin), or retinal atrophy. E The ROC curves for different retinal biomarkers, including Chlamydia pneumoniae (Cpn), Aβ₄₂, NLRP3, CCasp3, and retinal atrophy, either individual or combined with retinal Aβ₄₂. Each model was obtained by averaging the curves across diagnosis separately in each cross-validation fold. In the ROC curves plot, AUC is listed for each curve and unadjusted. P-values for whether the results were different from baseline dummy models using permutation tests with k = 10,000 iterations (estimated ***p  <  0.001). F The models were compared by using a Wilcoxon signed-rank test, and p values were adjusted for multiple comparisons using Benjamini-Hochberg correction. The heat map shows that among the top 5 performing models, we have 3 that are different from one another. The model trained on retinal Cpn + retinal Aβ₄₂ performed best and was significantly different from the second-best model (retinal NLRP3 + retinal Aβ₄₂) with p < 0.05. The other three models were different from the top 2, but not from one another. Red arrows highlight retinal markers, individually or in combination, which were significantly different among the performance models to predict disease diagnosis. Statistics: *p  <  0.05 and **p  <  0.01, adjusted for multiple comparisons with Benjamini-Hochberg procedure. Source data are provided as a Source Data file.