In this study, we provide immunohistochemical evidence of Chlamydia pneumoniae inclusions in postmortem retina and matched brain from individuals with normal cognition, MCI (due to AD), and AD dementia, with retinal findings validated by fluorescence in situ hybridization (FISH), Giemsa staining, and genomic DNA quantitative polymerase chain reaction (qPCR). We show that retinal and cerebral Chlamydia pneumoniae burden is elevated in AD dementia, but not MCI, and increases with APOEε4 and clinical-neuropathological severity. Consistently, targeted retinal and cortical proteomics reveal dysregulated bacterial-infection pathways, including a Chlamydia interactome coupled to NLRP3-inflammasome signaling. In human neuronal cells and APPSWE/PS1ΔE9 murine model of AD, infection induces Aβ accumulation, NLRP3 activation, neuroinflammation and cytotoxicity, and chronic infection exacerbates neuropathology and cognitive decline. In human retinas, Chlamydia pneumoniae associates with Aβ42, gliosis, apoptosis, gasdermin-dependent pyroptosis, and tissue atrophy, alongside a relative deficit of pathogen-engaged microglia, consistent with impaired clearance. Finally, machine-learning models indicate that retinal Chlamydia pneumoniae and NLRP3, alone and combined with Aβ42, may predict AD diagnosis, stage, and cognitive status, supporting a role for Chlamydia pneumoniae in disease progression and motivating NLRP3-targeted and/or antibiotic-based early treatment strategies.
To investigate Chlamydia pneumoniae infection and its association with neuroinflammation and neurodegeneration in the AD retina, we analyzed retinal and brain tissues from a cohort of 104 patients: 51 with AD dementia (mean age ± SD: 85.90 ± 10.06 years, 29 females/22 males), 16 with MCI due to AD (89.43 ± 6.90 years, 7 females/9 males), and 37 normal cognition controls (83.96 ± 10.96 years, 21 females/16 males). There were no significant differences in age, sex, or postmortem interval across groups. Demographic, clinical, and neuropathological characteristics of donor cohorts for retinal and brain histological and biochemical analyses are summarized in Fig. 1A, Table 1, and Supplementary Tables 1-5. To investigate protein expression profiles related to Chlamydia pneumoniae infection, we reanalyzed our recently reported mass spectrometry (MS)-based proteomic datasets from human retinal and cortical tissues. Complementary Western blot and genomic DNA qPCR assays were performed on retinal samples. Immunohistochemical (IHC) analyses focused on the superior-temporal (ST) retina and dorsolateral prefrontal cortex (A9), given their strong association with AD pathology. In parallel, we examined whether Chlamydia pneumoniae infection modulates AD-related pathological progression in human neuronal cells and AD⁺ mice (n = 45).
We applied a stepwise strategy to establish the existence and distribution of Chlamydia pneumoniae in the human retina. Initially, we broadly screened for Chlamydia species by performing IHC with an anti-Chlamydia polyclonal antibody (pAb) verified against Chlamydia pneumoniae, Chlamydia trachomatis, and Chlamydia psittaci, and thus capable of cross-reacting with these species. Using this antibody, Chlamydia-positive inclusion bodies were readily visualized by both fluorescence-based (Fig. 1B; n = 6) and peroxidase-based (Supplementary Fig. 1A, B; n = 6) immunolabeling. We then confirmed the specific presence of Chlamydia pneumoniae inclusions in retinal cross-sections using an anti-Chlamydia pneumoniae monoclonal antibody (mAb) that does not cross-react with other Chlamydia species, with peroxidase-based (n = 18) and fluorescence (n = 69) detection (Fig. 1C, D and Supplementary Fig. 1C, D). All quantitative analyses of Chlamydia pneumoniae burden reported in this study, including positive cell counts, immunoreactive (IR) area measurements, and correlation analyses, are based exclusively on this Chlamydia pneumoniae-specific mAb. The retinas of MCI and AD patients, compared with those of normal cognition controls, exhibited predominant Chlamydia pneumoniae-positive signals in the retinal ganglion cell layer (GCL) and inner nuclear layer (INL). Most Chlamydia pneumoniae inclusions were observed as cytosolic puncta aggregates, whereas other inclusions were detected in the peri-nucleus or nucleus and colocalized with DAPI, resembling the patterns observed using the anti-Chlamydia pAb. Chlamydia pneumoniae inclusions were also detected with the mAb in the corresponding A9 cortices of AD patients (Supplementary Fig. 1C, E), identifying typical intracellular inclusions similar to previously reported inclusion patterns in the AD brain.
To quantify Chlamydia pneumoniae burden in the retina, we first performed a cell count-based analysis, which revealed 2.1- and 1.9-fold increases in retinal Chlamydia pneumoniae-positive cells in the AD group compared with normal cognition and MCI groups, respectively (Fig. 1E; p < 0.001-0.0001). In addition, we analyzed the IR area of retinal and paired brain Chlamydia pneumoniae to determine a more integrated measure of bacterial burden, revealing significant 2.9- and 4.1-fold increases in Chlamydia pneumoniae inclusions, respectively, in AD patients compared with normal cognition controls (Fig. 1F; p < 0.0001; retina: n = 69, brain: n = 16). Consistent across both analyses (Fig. 1E, F), no significant difference in Chlamydia pneumoniae load was observed between normal cognition and MCI groups, indicating that expansion of Chlamydia pneumoniae infection likely spreads later in disease progression, during the clinical dementia stages of AD. Indeed, Gaussian distribution curves for both retinal and brain Chlamydia pneumoniae levels showed strong overlap between the normal cognition and MCI groups compared with the AD group (Supplementary Fig. 1F, G). Yet, the proportion of individuals with retinal or brain Chlamydia pneumoniae levels exceeding the normal-cognition mean (red line) was 60-79% in MCI and 100% in AD dementia, compared with only 38-40% in the normal-cognition group (Fig. 1F). Examination of retinal Chlamydia pneumoniae spatial distribution indicated a uniform burden across central (C), mid-peripheral (M), and far-peripheral (F) ST subregions (Supplementary Fig. 2A, B). Retinal Chlamydia pneumoniae levels did not differ between males and females within any diagnostic group (Supplementary Fig. 2C), indicating no sex-specific dimorphism in retinal burden.
The existence of Chlamydia pneumoniae in the human retina was further validated using three complementary histological and molecular approaches: Giemsa staining, FISH, and genomic DNA detection by real-time qPCR (Fig. 1G-I; extended data in Supplementary Fig. 3-4). Although Giemsa staining is not specific to a particular bacterial species, inclusion bodies in retinal cross-sections appeared as dark blue structures, consistent with those observed in Chlamydia pneumoniae-infected mouse lung tissues (Fig. 1G and Supplementary Fig. 3 A, B; n = 8 donors) and with Chlamydia pneumoniae inclusions identified by immunostaining with the mAb in the AD retina and brain (Fig. 1C, D). A FISH analysis using a fluorescently labeled Chlamydia pneumoniae-specific DNA probe further verified the presence of this bacterial genomic material within retinal tissues (Fig. 1H and Supplementary Fig. 4; n = 11 donors), which was absent in the no probe retinal AD and normal cognition tissues. Both Giemsa and FISH analyses demonstrated a higher burden of inclusions in AD retinas compared with those from individuals with normal cognition. Notably, qPCR analysis detected the Chlamydia pneumoniae-specific arginine repressor (argR) gene in retinal tissues from 2 of 2 AD cases, 0 of 1 MCI case, and 1 of 2 normal cognition controls, confirming the presence of Chlamydia pneumoniae in the human retina (Fig. 1I; n = 5 donors). Moreover, Pearson's correlation (r) between retinal and corresponding brain Chlamydia pneumoniae burdens revealed a strong concordance of bacterial load in both central nervous system (CNS) tissues (Fig. 1J; r = 0.62, p = 0.0143); extended correlations between Chlamydia pneumoniae load across retinal subregions and the brain showed the strongest association for the mid-periphery (Supplementary Fig. 5A, B; r = 0.71, p = 0.0046).
Given the strong connection between retinal and brain Chlamydia pneumoniae burden, we next examined how this bacterial load relates to AD relevant pathology and disease severity across both CNS tissues. We found a strong correlation between retinal Chlamydia pneumoniae burden and retinal Aβ species (Fig. 1K and Supplementary Table 8; Aβ: r = 0.63, p < 0.0001, Aβ: r = 0.65, p = 0.0014), with no correlation with intracellular Aβ oligomers (Supplementary Table 8), indicating a specific association with the extracellular plaque-dominant Aβ and vascular-dominant Aβ alloforms. Significant correlations between retinal Chlamydia pneumoniae burden and markers of retinal tauopathy were also detected, including paired-helical filament (PHF)-tau (r = 0.54, p = 0.0085, Fig. 1L), hyperphosphorylated (p)S396-tau (r = 0.38, p = 0.0116), T22 oligomeric tau (r = 0.43, p = 0.0040), and citrullinated tau (CitR: r = 0.48, p = 0.0028; Supplementary Table 8). Retinal Chlamydia pneumoniae burden showed no significant association with retinal AT8-positive p-tau or MC-1-positive mature tau tangles (Supplementary Table 8), nor with p-tau/total tau ratios at phosphorylation sites S404, S396, S199, S231, or S214 quantified by NanoString GeoMx digital spatial profiling in a subset of this cohort (Supplementary Fig. 5C-G; n = 22 donors). Overall, these data indicate that Chlamydia pneumoniae inclusions occur in the MCI and AD retina, predominantly in the GCL and INL, closely interact with amyloidogenic Aβ, and modestly associate with certain retinal tau isoforms but not with others.
Next, we determined whether retinal Chlamydia pneumoniae burden associates with AD-related brain pathology, apolipoprotein E (APOE) ɛ4 allele, disease staging, or the extent of cognitive deficit (Fig. 1M-Q; extended data in Supplementary Fig. 5H-J and Supplementary Table 8). We found that retinal Chlamydia pneumoniae significantly correlated with the severity of brain NFTs (Fig. 1M; r = 0.54, p < 0.0001) and was 2.1-2.4-fold higher in patients with advanced Braak stages (Fig. 1N; Stage III-IV or V-VI versus 0-II: p < 0.05-0.001, n = 60 donors), suggesting Chlamydia pneumoniae's involvement in brain tauopathy progression. Additionally, retinal Chlamydia pneumoniae burden significantly correlated with the following brain pathologies: Aβ plaques (r = 0.40, p = 0.0014), ABC severity score (r = 0.54 p < 0.0001), neuropil threads (NT; r = 0.37, p = 0.0033), cerebral amyloid angiopathy (CAA; r = 0.35, p = 0.0057), gliosis (r = 0.40, p = 0.0016), and brain atrophy (r = 0.48, p = 0.0001; Supplementary Table 8). Notably, both retinal and brain Chlamydia pneumoniae burdens were higher in APOE ɛ4 allele carriers compared with non-carriers, regardless of AD diagnosis (Fig. 1O; p = 0.037, n = 37; and Supplementary Fig. 5H; a trend, p = 0.06, n = 13).
Common bacterial infections, such as Helicobacter pylori, Chlamydia pneumoniae, Borrelia burgdorferi, and spirochetal Treponema have previously been linked to cognitive decline and increased dementia risk in elderly. In this cohort, retinal Chlamydia pneumoniae burden inversely correlated with Mini-Mental State Examination (MMSE) scores (Fig. 1P, n = 47 donors; and Supplementary Fig. 5I; r = -0.53, p < 0.0001), Clinical Dementia Rating (CDR) scores (Fig. 1Q; r = -0.43, p = 0.0010), and Montreal Cognitive Assessment (MOCA) scores (Supplementary Fig. 5J; r = -0.56, p = 0.0334; Supplementary Table 8), reinforcing the association between retinal Chlamydia pneumoniae load and global cognitive impairment. Despite the modest cohort size (n = 15), brain Chlamydia pneumoniae burden strongly correlated with increased AD brain pathology, including ABC score, Braak stage, NFTs, NTs, gliosis, and atrophy (Supplementary Table 9; r = 0.60-0.77, p < 0.05-0.001), showed a moderate association with cerebral amyloid angiopathy (CAA) scores but not Aβ-plaque burden, and was also strongly associated with poorer MMSE performance (r = -0.73, p = 0.0043). Collectively, these data tightly link retinal and brain Chlamydia pneumoniae burden to APOE ε4 status, widespread AD neuropathology, and global cognitive deterioration.
Detection of gram-negative Chlamydia pneumoniae inclusions in the retinas and brains of AD patients prompted us to investigate infection-driven protein dysregulation in these tissues by performing a secondary MS-based proteomic reanalysis in independent human retinal and cortical cohorts (Supplementary Tables 2-5; retina: n = 12, brain: n = 18). Metascape gene ontology (GO) analysis identified multiple dysregulated human proteins implicated in response to bacterial infection, including gram-negative intracellular bacteria, in the brains and retinas of AD patients (Fig. 2A, B), suggesting a significant involvement of bacterial infection in AD pathogenesis. To gain a closer look at Chlamydia infection, we searched for differentially expressed proteins (DEPs) in AD versus normal cognition brains and retinas that were included in the Chlamydia interactome (Fig. 2C, D; extended data on up- and down-regulated DEPs in Supplementary Tables 10 and 11). Of the 787 proteins in the Chlamydia interactome, 607 were identified in human brain, of which 84 were differentially expressed (52 downregulated, 32 upregulated; 13.8%) in AD patients compared with normal-cognition controls (Fig. 2C). Importantly, despite being derived from separate cohorts, similar bacterial infection-associated pathways (Fig. 2A, B) and dysregulated Chlamydia interactome DEPs were identified in the AD retina (Fig. 2D and Supplementary Fig. 6A), with 52 downregulated DEPs and 40 upregulated DEPs (13.0% DEPs) among the 710 identified (Fig. 2D). GO network analysis further revealed enrichment of proteins involved in immune responses to microorganisms and cell death in AD brains and retinas (Fig. 2E, F, Supplementary Fig. 6B-D, and Supplementary Fig. 7A-C). Inflammation-related proteins were primarily associated with cytokine signaling, toll-like receptor (TLR) pathways, interferon responses, nuclear factor kappa B (NF-κB) activation, NLRP3 inflammasome activation, and pyroptosis, pathways typically triggered by gram-negative bacteria in peripheral tissues. These data suggest shared infection- and immune-associated mechanisms in AD brains and retinas.
Chlamydia has been shown to trigger the host's innate immune response, requiring TLR2/MYD88 signaling and NLRP3/ASC/caspase-1 inflammasome activation. Indeed, both MYD88 innate immune signal transduction adaptor (MYD88) and PYD and CARD domain containing protein (PYCARD or ASC) were upregulated in the AD retina (Fig. 2F). Additionally, the DNA pathogen sensor and Chlamydia interactor, leucine-rich repeat-binding FLII interacting protein 1 (LRRFIP1), which positively regulates TLR4 by competing with FLII actin remodeling protein (FLII) for interaction with MYD88, was upregulated in both the AD brain and retina (Fig. 2F and Supplementary Fig. 6A). Importantly, retinal AD proteome was enriched in proteins linked to pyroptosis (Fig. 2E and Supplementary Fig. 6B, D), a form of inflammatory regulated necrosis triggered by intracellular pathogens, including Chlamydia. Notably, three members of the gasdermin (GSDM) family, GSDMD, GSDME (or DFNA5), and GSDMA, which are involved in pyroptosis/necrosis, were upregulated in the AD retina (Fig. 2F). Proteins involved in apoptosis, pyroptosis, and inflammation were generally associated with levels of cerebral and retinal tau isoforms quantified by MS and with retinal Aβ measured by enzyme-linked immunosorbent assay (ELISA) (Supplementary Figs. 7D-H and 8A-J). Notably, retinal Aβ levels strong-to-very strongly correlated with proteins associated with cell degeneration, including Casp3 (r = 0.77, p = 0.0099), Bcl-2-associated athanogene 3 (BAG3, r = 0.76, p = 0.012), and GSDMD (r = 0.89, p = 0.0006), as well as inflammatory regulators such as dermcidin (DCD, r = 0.78, p = 0.0084) and LRRFIP1 (r = 0.81, p = 0.0046) (Supplementary Fig. 8A-E). In contrast, cytoprotective and anti-inflammatory proteins, including thiol methyltransferase 1 A (TMT1A, r = -0.77, p = 0.015) and adaptor protein complex 2, alpha 2 subunit (AP2A2, r = -0.88, p = 0.0008), were inversely correlated with retinal Aβ. Similar to retinal amyloidosis, Chlamydia interactome proteins also exhibited significant associations with retinal (0N4R) tau isoform and brain (1N3R, 2N4R) tau isoforms (Supplementary Figs. 8F-J and 7G, H). In particular, retinal 0N4R tau strong-to-very strongly correlated with GSDMD (r = 0.65, p = 0.022), BAG3 (r = 0.85, p = 0.0005), RAD23 nucleotide excision repair protein B (RAD23B, r = 0.90, p < 0.0001), LRRFIP1 (r = 0.81, p = 0.0015), calpastatin (CAST, r = 0.91, p < 0.0001), and TMT1A (r = -0.92, p < 0.0001; Supplementary Fig. 8F-J). Overall, these findings indicate enrichment of proteins implicated in intracellular gram-negative bacterial infection, including Chlamydia-associated proteins, together with signatures of inflammasome activation and degeneration in AD brain and retina.
To determine whether Chlamydia pneumoniae acts as a driver rather than a bystander in AD, we next tested whether infection of neuronal cells and AD transgenic mice is sufficient to trigger inflammasome activation and exacerbate AD-related pathology. Infection of SH-SY5Y human neuroblastoma cells with Chlamydia pneumoniae (multiplicity of infection [MOI] 5) for 68 hours markedly induced Aβ, NLRP3, and IL1β levels and triggered cell membrane damage, as assessed by lactate dehydrogenase (LDH) release (Fig. 3A-F). Immunocytochemistry confirmed robust Chlamydia pneumoniae infection of SH-SY5Y neurons and revealed that infected cells, compared with uninfected controls, exhibited 2.5-fold higher NLRP3, 3.2-fold higher IL1β, and 3.5-fold higher H31L21 Aβ IR areas (all p < 0.0001; Fig. 3B, C; extended data in Supplementary Fig. 9A; n = 6 wells per condition, n = 41-74 cells per group). Moreover, pronounced 3.9-fold increase in LDH leakage was detected in infected neuronal cells versus uninfected controls (Fig. 3D, p < 0.001, n = 6 wells per group), as assessed by LDH release into the culture medium, suggesting that Chlamydia pneumoniae infection induces neurotoxicity. We further substantiated these findings by Western blot analysis, which demonstrated increases in NLRP3, cleaved IL1β (1.5-fold, p < 0.05), 12F4 Aβ (6.7-fold, p < 0.05), and N-terminal cleaved gasdermin D (NGSDMD) in infected neurons compared with controls (Fig. 3E, F; n = 3-6 wells per group). Together, these findings demonstrate that Chlamydia pneumoniae infection is sufficient to drive NLRP3 inflammasome activation, pyroptotic cell death, and Aβ accumulation in neuronal cells — cellular features of AD pathology. Future studies will be required to delineate the molecular pathways by which Chlamydia pneumoniae amplifies Aβ production, sustains inflammasome signaling, and ultimately promotes neurodegeneration.
These in vitro observations prompted us to examine the impact of acute and long-term Chlamydia pneumoniae infection on in vivo Alzheimer-like pathology and cognition in APP/PS1 (AD) mouse models (Fig. 3G-S). In the acute Chlamydia pneumoniae infection paradigm, we examined 8 phosphate-buffered saline (PBS)-treated AD⁺ controls and 14 infected AD⁺ mice (n = 22). In the long-term paradigm, we studied 6 PBS-treated wild-type (WT), 8 PBS-treated AD⁺, and 9 infected AD⁺ mice (n = 23). Intranasal inoculation with Chlamydia pneumoniae (1 × 10⁶ inclusion-forming units, IFU) resulted in a marked increase in bacterial inclusions in the AD⁺ mouse brain (Fig. 3G, H), as confirmed by higher Chlamydia pneumoniae IFUs in HEp2 cells treated with brain lysates from infected versus uninfected mice and by increased Chlamydia pneumoniae genomic DNA copy numbers (Fig. 3H). Seven-days (acute) Chlamydia pneumoniae postinfection caused a 3.2-fold increase in ionized calcium-binding adaptor molecule 1 (IBA1)⁺ microgliosis and a 2.5-fold increase in glial fibrillary acidic protein (GFAP)⁺ astrogliosis in the hippocampi and cortices of infected versus uninfected AD⁺ mice (Fig. 3I, J; p < 0.05-0.01; extended data in Supplementary Fig. 9B-D), indicating amplified neuroinflammation due to infection. Furthermore, infected AD⁺ mice exhibited elevated mRNA expression of Il6 (p < 0.01), Il1β (p < 0.05), and Nlrp3 (p < 0.01) (Fig. 3K), further supporting activation of NLRP3-inflammasome signaling in response to Chlamydia pneumoniae infection. These findings demonstrate that acute intranasal infection is sufficient for Chlamydia pneumoniae to reach the brain, establish infection, and subsequently trigger inflammasome activation and neuroinflammation.
The long-term behavioral and pathological consequences of Chlamydia pneumoniae infection in AD⁺ mice were assessed 6 months after a single intranasal inoculation (1 × 10⁶ IFU Chlamydia pneumoniae or PBS) administered at 8 months of age (Fig. 3L-S; extended data in Supplementary Fig. 9E-J). Multidomain behavioral testing was conducted over 12 days in 14-month-old mice, with PBS-treated WT animals serving as healthy behavioral controls. In the open field and X-maze tests, Chlamydia pneumoniae infection did not affect locomotor function of AD mice, as indicated by rearing and ambulatory activity or total arm entries (Fig. 3L and Supplementary Fig. 9E, H). However, alternations in both color- and contrast-stimuli modes of the X-maze, which assess visuo-cognitive function and are decreased in AD mice, were further reduced in Chlamydia pneumoniae-infected AD mice (Fig. 3M, N). In the color-mode X-maze, while infection did not affect specific arm entries, it further decreased bidirectional blue (B)↔white (W) transitions in infected AD mice, indicating color vision dysfunction (Fig. 3M and Supplementary Fig. 9F, G). In the contrast-mode X-maze, entries into the arm with the white object were increased in the PBS-control AD mice and further increased in the Chlamydia pneumoniae-infected AD mice (Fig. 3N and Supplementary Fig. 9I). The infected AD mice also exhibited increased black (B)↔W and decrease of black (B)↔clear (C) bidirectional transitions, indicating a worsening of contrast sensitivity vision due to infection (Supplementary Fig. 9J). In the Barnes maze, PBS-control AD mice (vs. WT) made significantly more errors prior to finding the escape box, during the 4-day acquisition phase, the long-term memory retention phase, and the 2-day reversal phase (Fig. 3O). Importantly, Chlamydia pneumoniae infection in AD mice further increased the number of errors made on reversal day 9 (Fig. 3O, P), which measures spatial learning and cognitive flexibility. Search coverage analysis showed that Chlamydia pneumoniae-infected AD mice made more errors locally in the area that is both on the side of the old and new escape box locations (Fig. 3P). These results indicate that long-term Chlamydia pneumoniae infection exacerbates visuocognitive dysfunction in AD⁺ mice without affecting locomotor function.
We subsequently examined AD-related pathology in the cortex and hippocampus of long-term infected versus uninfected AD⁺ mice (Fig. 3Q-S). Our analysis revealed significant increases in 6E10 Aβ plaques (1.6-fold, p < 0.001), IBA1 microglia (1.3-fold, p < 0.01), and GFAP astrocytes (1.3-fold, p < 0.05) in the cortex of Chlamydia pneumoniae infected AD mice compared with PBS-administered AD mice (Fig. 3R). Similar increases were also observed in the hippocampus (Fig. 3S). These findings demonstrate that long-term Chlamydia pneumoniae infection in AD⁺ mice aggravates neuroglial activation and Aβ pathology, supporting the hypothesis that chronic infection exacerbates AD-like neuropathology.
Convergent evidence from MS-based proteomics, Chlamydia pneumoniae-infected cell cultures, and AD mouse models implicating this pathogen in cerebral NLRP3 inflammasome activation, together with prior murine infection studies demonstrating Chlamydia pneumoniae-driven NLRP3 activation, led us to test whether a similar inflammasome axis and associated cell death mechanisms operate in the human AD retina. To this end, we first applied a quantitative IHC analysis to retinal cross-sections from patients with MCI due to AD and AD dementia, compared with matched non-AD individuals with normal cognition (Fig. 4A-I; extended data in Supplementary Fig. 10; n = 25-27 donors). Retinal NLRP3 expression was significantly elevated in MCI and further increased in AD compared with normal-cognition controls (2.1- and 3.6-fold, respectively; p < 0.001-0.0001), with strong colocalization with caspase-1, which itself was upregulated 2.5-fold in AD but not MCI. This inflammasome-activation signature was accompanied by a 3.1-fold increase in Chlamydia pneumoniae-associated ASC speck signals in AD, but not in MCI (Fig. 4A-E; extended data in Supplementary Fig. 10A, B). The early rise in NLRP3 immunoreactivity and later induction of caspase-1 and ASC markers may suggest that NLRP3 is activated by earlier processes such as misfolded Aβ and tau accumulation in the retina.
Next, we investigated the impact of retinal Chlamydia pneumoniae-mediated NLRP3 inflammasome activation on key apoptotic and pyroptotic components. Chlamydia pneumoniae-infected cells in AD retinas frequently co-expressed the pyroptotic effector NGSDMD and the early apoptotic marker cleaved caspase-3 (CCasp3) (Fig. 4F-I; extended data in Supplementary Fig. 10C, D). Quantitative IHC analysis revealed significant 2.2- and 3.0-fold increases in retinal NGSDMD and CCasp3 signals, respectively, in AD versus normal-cognition controls (p < 0.001-0.0001), whereas MCI retinas exhibited a nonsignificant trend toward elevation (Fig. 4G, I). Most Chlamydia pneumoniae-positive cells coexpressed either pyroptotic or apoptotic markers, suggesting that retinal Chlamydia pneumoniae infection in AD engages both cell death pathways. Moreover, the marked increase in NGSDMD in AD retinas provides functional evidence of NLRP3 inflammasome-mediated pyroptotic activation.
The inter-relationships between retinal Chlamydia pneumoniae burden, NLRP3 inflammasome components, and cell death markers were subsequently evaluated (Fig. 4J; extended data in Supplementary Table 12). Multivariate correlation analysis in our cohort revealed that retinal Chlamydia pneumoniae burden was strongly to very strongly associated with retinal NLRP3 inflammasome components, particularly caspase-1 (r = 0.87, p < 0.0001), NLRP3 (r = 0.70, p < 0.0001), and ASC (r = 0.60, p = 0.0012) (Fig. 4J). In addition to their correlations with retinal Chlamydia pneumoniae load (Supplementary Table 8), both retinal Aβ and T22 oligomeric tau were strongly to very strongly associated with NLRP3 and caspase-1 (r = 0.60-0.81, p < 0.01-0.0001; Supplementary Table 12), supporting their role as potential activators of the retinal NLRP3 inflammasome. Retinal oligomeric tau was strongly correlated with both apoptotic and pyroptotic markers, including CCasp3 (r = 0.80, p < 0.0001) and NGSDMD (r = 0.77, p < 0.0001), and retinal Aβ likewise showed strong associations with CCasp3 (r = 0.77, p = 0.0003) and NGSDMD (r = 0.64, p = 0.0247; Supplementary Table 12). Retinal NLRP3 inflammasome components were significantly inter-correlated (r = 0.58-0.83, p = 0.0016- p < 0.0001; Fig. 4J), with NLRP3 and caspase-1 most tightly linked. All three components were strongly to very strongly associated with CCasp3 (r = 0.76-0.81, p < 0.0001), whereas NGSDMD pyroptosis was most closely related to NLRP3 (r = 0.74, p < 0.0001) and only moderately correlated with caspase-1 and ASC (r = 0.57-0.58, p < 0.01; Fig. 4J and Supplementary Table 12). To assess the canonical downstream effector of NLRP3 inflammasome activation, we quantified the pro-inflammatory cytokine IL1β by Western blot in retinal homogenates from AD patients and normal-cognition controls. Consistent with inflammasome engagement, AD retinas exhibited a marked shift from pro-IL1β to its active cleaved form, with significantly reduced pro-IL1β by 49.8% and markedly elevated mature IL1β by 2.1 folds (Fig. 4K; p < 0.01-0.001, n = 10 donors). Together, these observations delineate a Chlamydia pneumoniae-, Aβ-, and oligomeric tau-linked retinal NLRP3 inflammasome axis that converges on IL1β maturation and apoptotic/pyroptotic cell death in AD (summarized in Fig. 4L).
We next examined how retinal Chlamydia pneumoniae-related NLRP3 inflammasome activation components relate to retinal atrophy and a spectrum of cerebral AD pathological indices and clinical outcomes (Fig. 4M; extended data in Supplementary Tables 12 and 13). Retinal NLRP3 and caspase-1 were very strongly and positively correlated with retinal atrophy severity, quantified as a thinning index from the inner limiting membrane to the outer limiting membrane (r = 0.82-0.86, p < 0.001-0.0001), and showed moderate positive correlations with global brain atrophy scores (r = 0.41-0.47, p < 0.05). Retinal ASC was strongly associated with retinal atrophy (r = 0.62, p = 0.0171), but not with brain atrophy. Retinal Chlamydia pneumoniae burden exhibited a strong association with retinal atrophy (r = 0.75, p < 0.0001) and a moderate correlation with brain atrophy (r = 0.48, p < 0.001). Retinal CCasp3 showed moderate to very strong correlations with both brain and retinal atrophy (Fig. 4M; r = 0.45-0.86, p < 0.05-0.001), whereas NGSDMD was strongly correlated with retinal atrophy (r = 0.60, p = 0.0247) but not with brain atrophy. Similar to retinal Chlamydia pneumoniae, all retinal inflammasome components and related cell death markers correlated moderately to strongly with Braak stage (r = 0.55-0.72, p < 0.01-0.0001) and inversely with MMSE cognitive performance (Fig. 4M; r = -0.49 to -0.69, p < 0.05-0.001). Together, these relationships position retinal Chlamydia pneumoniae burden and NLRP3 inflammasome activation as integrated indices of local neurodegeneration and global AD severity. They support a model in which pathogen-linked retinal inflammasome signaling tracks with, and may contribute to, parallel brain atrophy and cognitive decline.
Chlamydia pneumoniae infects and extensively replicates in astrocytes and neurons, whereas microglia are primarily involved in Chlamydia pneumoniae phagocytosis. Although previous studies, including our own, have documented robust retinal gliosis in MCI and AD retinas, the potential interplay between glial cells and Chlamydia pneumoniae in these patients' retinas remains unexplored. Here, we observed prominent spatial interactions between Chlamydia pneumoniae inclusions and retinal microglia, astrocytes, and Müller glia in MCI and AD, with glial cells frequently surrounding or internalizing bacteria-positive inclusions (Fig. 5A-J; extended data in Supplementary Fig. 11A, B; n = 21-32 donors). Quantitative IHC analyses revealed significantly increased GFAP astrocyte and IBA1⁺ microglial cell counts in the AD retina (1.5-fold, p < 0.05-0.001) but not in the MCI retina (Fig. 5B, H). Interestingly, analysis of the IR area of GFAP⁺ and vimentin⁺ macroglia, as well as IBA1⁺ microglia, thereby accounting for cell morphology and process hypertrophy, showed significant expansion of all gliosis markers in MCI retinas (1.5-1.9-folds, p < 0.05-0.01), with further marked expansion observed in AD retinas (2.3-2.8-folds, p < 0.0001) relative to normal-cognition controls (Fig. 5C, E, I; extended data in Supplementary Fig. 11C-E). These findings suggest that glial activation and process hypertrophy emerge early along the retinal AD continuum, followed by overt glial proliferation in established AD dementia.
We next examined whether retinal bacterial load scales with gliosis burden. Retinal Chlamydia pneumoniae burden showed strong associations with GFAP⁺ astrocytosis and IBA1⁺ microgliosis (Fig. 5D, J; r = 0.65-0.70, p < 0.0001) and a moderate association with vimentin⁺ macroglial reactivity (Fig. 5F; r = 0.55, p = 0.0090). Consistent with these retinal interactions, strong correlations were detected between brain Chlamydia pneumoniae burden and brain gliosis (r = 0.77, p = 0.0008), as mentioned above. The strong association between Chlamydia pneumoniae burden and gliosis indicates potential chronic inflammatory cascades linked to bacterial infection. Notably, while retinal astrocytes appeared to be infected by Chlamydia pneumoniae, retinal microglia appeared to phagocytose bacterial inclusions or bacteria-infected cells (Fig. 5A, K). Specifically, retinal microglia appeared to exhibit different stages of responses to Chlamydia pneumoniae-infected cells, with most cells in close proximity and partial contact with Chlamydia pneumoniae-positive cells (recognition phase). Other microglia were directly involved in engulfing or ingesting Chlamydia pneumoniae-infected cells (Fig. 5K). The percentage of Chlamydia pneumoniae-associated microglia (CAM) cell count was increased by 50% in the AD retina, but not in the MCI retina, compared with normal-cognition controls (Fig. 5L; 1.5-fold, p < 0.01, n = 44 donors). However, the ratio of retinal CAM cell count to bacterial load was reduced by 61% in the AD retina versus normal-cognition controls (Fig. 5M; p < 0.0001, n = 44; extended data on microglia recognizing, engulfing or phagocytosing Chlamydia pneumoniae-infected cells in Supplementary Fig. 12A-C), implying defective microglial phagocytosis of Chlamydia pneumoniae in the AD retina. To further verify that Chlamydia pneumoniae-associated IBA1⁺ cells were resident microglia rather than perivascular or infiltrating macrophages, we co-labeled retinal sections with transmembrane protein 119 (TMEM119), a marker enriched in resident microglia in the human brain and retina (Fig. 5N, n = 12; extended images in Supplementary Fig. 12D). All observed IBA1⁺ cells engaged in recognition, engulfment, or ingestion of Chlamydia pneumoniae inclusions co-expressed TMEM119, indicating a microglial identity. We refer to these cells as CAM. The percentage of retinal CAM cells relative to bacterial load was strongly and inversely correlated with retinal Chlamydia pneumoniae burden (Fig. 5O; r = -0.72, p < 0.0001), with an even stronger correlation among individuals with normal cognition (Supplementary Fig. 12E; r = -0.79, p = 0.0004). Within the normal-cognition group, three female individuals exhibited a disproportionately low percentage of CAM relative to retinal bacterial burden, comparable to the pattern observed in AD (Fig. 5M and Supplementary Fig. 12E). Notably, these individuals also showed the highest retinal Chlamydia pneumoniae loads (Fig. 1F), with comparable levels of retinal and brain AD-related pathology, suggesting a selective impairment of microglial engagement with bacteria-infected cells even in some clinically normal individuals. Together, these data indicate that, despite increased recruitment of CAM in the AD retina, their phagocytic engagement relative to bacterial burden is markedly reduced, consistent with impaired microglial clearance of retinal Chlamydia pneumoniae in AD.
A multi-interaction heatmap between retinal gliosis and various AD biomarkers in the retina and brain (Fig. 5P; extended data in Supplementary Tables 12 and 13) revealed very strong associations between retinal GFAP or vimentin macrogliosis and NLRP3 load (r = 0.84-0.91, p < 0.01-0.0001), and between retinal GFAP astrogliosis and CCasp3 apoptosis (r = 0.85, p < 0.0001). As it relates to amyloidosis and tauopathy, retinal Aβ and oligo-tau burdens most closely correlated with retinal IBA1 microgliosis levels (Fig. 5P; r = 0.69-0.85, p < 0.0001). In addition, retinal GFAP strongly correlated with Braak scores (r = 0.78, p < 0.0001). These findings suggest close interactions between retinal glial cells and Chlamydia pneumoniae-infected cells, strongly correlating with NLRP3 inflammasome components and apoptosis/pyroptosis cell death markers, and a potentially impaired ability of microglia to phagocytose and clear Chlamydia pneumoniae infection in the AD retina.
We next tested whether retinal Chlamydia pneumoniae burden and associated markers could serve as predictors of AD diagnosis, cerebral pathology severity, disease stage, and cognitive dysfunction (Fig. 6; extended data in Supplementary Figs. 13-15). In addition to retinal Chlamydia pneumoniae, we included key retinal markers that demonstrated significant correlations with infection and AD pathology, specifically NLRP3, CCasp3-apoptosis, and Aβ. These markers were analyzed either in isolation or in combination with retinal gliosis (IBA1+GFAP+vimentin), atrophy, and Aβ; the latter can potentially be imaged in living patients. We evaluated 80 biomarker-derived estimators for both regression and classification, using a diagnosis-stratified split of 56 donors (80%) for model training and 14 donors (20%) for testing. Multivariable analysis employing random forest machine learning models indicated that retinal Chlamydia pneumoniae alone weakly predicted AD-related brain pathologies, including ABC score and Braak stage (Fig. 6A, B), as well as cognitive function (MMSE, Fig. 6C; MOCA, Supplementary Fig. 13B). Multiple models were compared using 5×2 cross validation to obtain distributions of model performance. The predictive power of retinal Chlamydia pneumoniae was generally enhanced when combined with retinal Aβ or gliosis (Fig. 6A-C and Supplementary Fig. 13A-E). We found that retinal Aβ alone was a good predictor of brain NFTs, ABC score, Braak stage, and the MMSE score (Fig. 6A-C and Supplementary Fig. 13A). Notably, the combined retinal Chlamydia pneumoniae and gliosis index was the best predictor of ABC score (Fig. 6A, r = 0.34) and brain gliosis (Supplementary Fig. 13E, r = 0.26). The best predictor of Braak stage was retinal Aβ combined with CCasp3 (r = 0.41), followed by combinations with NLRP3 (r = 0.38), and Chlamydia pneumoniae (r = 0.28; Fig. 6B). In addition, the combined retinal NLRP3 and Aβ index was the best predictor of MMSE score (Fig. 6C, r = 0.25), and retinal Aβ combined with either Chlamydia pneumoniae or CCasp3 also predicted MMSE (Fig. 6C, r = 0.22-0.23), while no individual marker predicted brain atrophy (Supplementary Fig. 13D).
We further evaluated the performance of these variables using the area under the receiver operating characteristic (ROC) curve (AUC) for disease diagnosis (Fig. 6D-F and Supplementary Fig. 14A-D; separate analyses for each diagnostic group in Supplementary Fig. 15A-D). When combined with retinal amyloidopathy (Aβ), the AUC for retinal Chlamydia pneumoniae increased from 0.80 to 0.94 (Fig. 6D, E), with particular gains in distinguishing MCI from normal-cognition diagnoses (Fig. 6D). These findings indicate that retinal Chlamydia pneumoniae in combination with Aβ may constitute a highly informative marker pair for discriminating disease status (Fig. 6F). Retinal Aβ alone (AUC = 0.87) showed its best predictive performance for classifying normal-cognition subjects (AUC = 0.92; Supplementary Fig. 15D). Consistent with the random forest model results, retinal NLRP3, CCasp3, and atrophy exhibited strong AUC values for disease diagnosis, with notable further improvements when combined with Aβ (0.92, 0.87, and 0.89, respectively; Fig. 6E and Supplementary Fig. 14B-D), including within individual diagnostic groups (Supplementary Fig. 15A-C). Based on the results presented in Fig. 6E, F, we selected the model using Chlamydia pneumoniae and retinal Aβ for evaluation on the test set, consisting of 3/4/7 patients with normal cognition/MCI/AD diagnoses, respectively; the results are reported in Table 2. The model performed poorly for subjects with MCI but performed reasonably well for identifying normal cognition and AD. We note that the normal cognition and MCI groups were underrepresented in the test set.
We further examined the selected model's performance by comparing prediction probabilities with a clinical disease severity (scores 0-3), based on premortem CDR and MMSE cognitive performance. Using the test set, we obtained the prediction probability for each diagnosis and compared these against disease severity (Supplementary Fig. 15E). There are high correlations between the prediction probabilities of both normal cognition (r = -0.78) and AD (r = 0.79) against disease severity. No correlation was observed between MCI prediction and disease severity, as expected. Looking at the prediction probabilities for each diagnosis, we see that the standard deviation across normal cognition, MCI, and AD is 0.41, 0.22, and 0.34, respectively. Highly certain predictions would result in higher standard deviations; predictions for MCI patients are less certain, indicating the possibility that MCI could be detectable with additional training samples.
