Bacterial polar metabolites modulate β-amyloid toxicity and cholinergic dysfunction in models of Alzheimer’s disease Authors: Ana Carolina Costa Santos, Joberth Lee Corrêa, Rener Mateus Francisco Duarte, Serena Mares Malta, Tamiris Sabrina Rodrigues, Débora de Oliveira Santos, Paulo Rogério de Faria, Fernanda Naves Araújo do Prado Mascarenhas, Renata Graciele Zanon, Nadla Soares Cassemiro, Carlos Alexandre Carollo, Foued Salmen Espindola, Mario Machado Martins, Ana Paula Mendes-Silva, Ana Maria Bonetti, Anderson Rodrigues dos Santos, Carlos Ueira-Vieira Published: 2026 ABSTRACT AbstractAlzheimer’s disease is characterized by progressive neurodegeneration driven by β-amyloid (Aβ) toxicity, oxidative stress, and cholinergic dysfunction. In this study, we investigated whether polar metabolites derived from a cultivable bacterial isolate could modulate Aβ-associated neurodegenerative phenotypes in complementary experimental models. A bioactivity-guided approach identified an aqueous fraction with high antioxidant capacity in DPPH, FRAP, and ORAC assays. In a transgenic Drosophila melanogaster model expressing human Aβ, treatment with this fraction significantly reduced amyloid accumulation and attenuated neurodegenerative histopathological alterations. In human SH-SY5Y neuronal cultures, the metabolites improved cell viability under therapeutic, but not preventive, conditions following exposure to aggregated Aβ. The aqueous fraction also exhibited significant inhibitory activity against acetylcholinesterase and butyrylcholinesterase. Whole-genome sequencing assigned the bioactive isolate to the genus Providencia, with comparative genomic analyses suggesting its placement within a distinct taxonomic lineage. Metabolomic profiling by LC–ESI–MS/MS revealed a diverse set of polar metabolites, including metabolites putatively annotated based on spectral matching, previously associated with neuroprotective and cholinesterase-modulating activities. Collectively, these findings demonstrate that bacterial polar metabolites can modulate key pathological features of Alzheimer’s disease, supporting their relevance for mechanistic studies of Aβ toxicity and cholinergic dysfunction.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1038/s41598-026-52291-3.Keywords: Alzheimer model, Genomic, Compound, Natural product, BacteriaSubject terms: Biochemistry, Drug discovery, Neuroscience Introduction Oxidative stress, characterized by an imbalance between the production of reactive oxygen species (ROS) and antioxidant defenses, plays a critical role in the pathogenesis of numerous human diseases, particularly neurodegenerative disorders such as Alzheimer’s disease, as well as cardiovascular diseases and cancer1–4. In Alzheimer’s disease, oxidative damage contributes to β-amyloid toxicity, neuronal dysfunction, and synaptic impairment. Antioxidants, which neutralize ROS and prevent cellular damage, have thus attracted considerable attention for their therapeutic and preventive potential. Traditionally, antioxidants have been sourced from plants; however, in recent years, microorganisms have emerged as a promising alternative source of bioactive antioxidant compounds due to their metabolite versatility, rapid growth rates, and ease of cultivation under controlled conditions5–11. Recent studies have demonstrated the potential of microbial antioxidants in pharmaceutical applications as well as applications in food preservation, cosmetics, and bioremediation12. More recently, their application has been extended to neurodegenerative disorders, including Alzheimer’s disease, given their capacity to modulate oxidative stress and enzyme activity relevant to neuronal function. Despite these advances, the exploration of novel microbial strains and the elucidation of their antioxidant mechanisms remain areas of active research. The discovery and characterization of microorganisms capable of producing potent antioxidant compounds are essential for expanding the repertoire of bioactive molecules with therapeutic potential. Microbial production of antioxidants includes a wide range of compounds such as carotenoids, phenolic acids, flavonoids, and exopolysaccharides, which exhibit diverse mechanisms of action, including free radical scavenging, metal ion chelation, and inhibition of oxidative enzymes13–15. These mechanisms are particularly relevant in the context of Alzheimer’s disease, where redox imbalance exacerbates protein aggregation and neurodegeneration. These metabolites are synthesized by bacteria, fungi, yeast, and microalgae, many of which are isolated from extreme environments, where oxidative stress tolerance is essential for survival16–23. Furthermore, microbial-derived antioxidants offer advantages over plant-derived counterparts, including sustainable production, scalability, and the potential for genetic and metabolite engineering to enhance yields and tailor bioactivities15,24. Our research group has been focusing on the characterization of the microbiome of the larval food of Brazilian stingless bees, isolation of microorganisms, and their bioprospecting25–27. Based on this bioprospecting approach, this study aimed to screen bacterial isolates obtained from the larval food of Brazilian stingless bees to produce antioxidant compounds and to assess their potential to modulate Alzheimer’s disease–related phenotypes using a transgenic Drosophila melanogaster model and a human neuronal cell culture model. Results Screening for antioxidant strains Seven bacterial isolates from the microbial biobank were screened for the secretion of antioxidant metabolites, using oxidative stress mitigation as a selection criterion for potential neuroprotective activity. All protein fractions exhibited measurable antioxidant activity; however, only the protein fraction derived from isolate Mq-NUT-54B displayed detectable reducing power in the FRAP assay (Fig. 1A, C, E). Antioxidant activity of protein and metabolite fractions from bacterial supernatants. Antioxidant capacity was evaluated using DPPH (A, B), FRAP (C, D), and ORAC (E, F) assays for protein (A, C, E) and metabolite (B, D, F) fractions obtained from bacterial isolates. Results are expressed as mean ± SEM from technical replicates (n = 3). Statistical analysis was performed using one-way ANOVA followed by Holm–Šidák’s multiple-comparisons test. Differences were considered significant at p < 0.05 (*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001). In contrast, all metabolite fractions exhibited robust antioxidant activity in the DPPH, FRAP, and ORAC assays (Fig. 1B, D, F). Based on the consistency and magnitude of antioxidant effects, the metabolite fractions of isolates Mq-NUT-54B and Mq-ISP-1B were selected for subsequent analyses. Other tested fractions showed markedly lower or no antioxidant activity and were not selected for further analyses. Liquid–liquid partitioning The results demonstrate that sequential liquid-liquid partitioning of the bacterial culture supernatant with hexane, dichloromethane, ethyl acetate, and n-butanol effectively separated metabolites by polarity. Among all fractions, the aqueous phase consistently exhibited the highest antioxidant activity in the DPPH, FRAP, and ORAC assays (Fig. 2). Based on these results, only the aqueous fractions were selected for subsequent in vivo and in vitro experiments. This selection was based on a bioactivity-guided strategy using antioxidant capacity as the primary screening criterion. Antioxidant activity of fractions obtained by liquid–liquid partitioning. Fractions obtained using hexane, dichloromethane, ethyl acetate, n-butanol, and aqueous phase were evaluated using DPPH (A), FRAP (B), and ORAC (C) assays. Results are expressed as mean ± SEM from technical replicates (n = 3). Statistical analysis was performed using one-way ANOVA followed by Holm–Šidák’s multiple-comparisons test. Differences were considered significant at p < 0.05. Effect of metabolites for aqueous fraction in a model of Alzheimer’s disease-like model (AD-like model) The AD-like Drosophila melanogaster model was first validated by quantifying amyloid-β (Aβ) levels. As expected, flies expressing human Aβ showed a significant increase in amyloid accumulation compared to control flies (Fig. 3A). Treatment with aqueous fractions derived from Mq-NUT-54B and Mq-ISP-1B significantly reduced Aβ levels in AD-like flies relative to untreated controls (Fig. 3B), indicating that bacterial metabolites exert anti-amyloidogenic effects in vivo. Relative quantification of amyloid-β using a Thioflavin T (ThT) assay in a Drosophila melanogaster AD-like model. A Validation of the AD-like model by comparing amyloid levels between control flies (GMR-Gal4/+) and Aβ-expressing flies (GMR-Gal4/+; UAS-Aβ/+). B Effect of aqueous fractions from Mq-ISP-1B (1B) and Mq-NUT-54B (5B) on amyloid levels. Amyloid levels were quantified using a Thioflavin T (ThT) assay. Data are presented as individual values with mean ± SEM from biological replicates (n = 3 independent pools, each consisting of 10 heads). Statistical analysis was performed using one-way ANOVA followed by Holm–Šidák’s multiple-comparisons test. Differences were considered significant at p < 0.05. Histopathological analysis The AD-like flies exhibited significant histopathological alterations, including increased total vacuole area (p < 0.05) and decreased retinal thickness (p < 0.05), compared to healthy control flies (Fig. 4). Histopathological characterization of neurodegeneration in a Drosophila melanogaster AD-like model. Representative histological Sect. (3 μm) of fly heads showing the optic lobe. A Overview of the analyzed region (medulla, me) in control flies (GMR-Gal4/+), 40× magnification. B Medulla of control flies, 100× magnification. C Medulla of Aβ-expressing flies (GMR-Gal4/+; UAS-Aβ/+), 100× magnification; arrows indicate vacuolar lesions. D Quantification of total vacuole area. E Overview of the analyzed retinal region (re), 40× magnification. F Retina of control flies, 100× magnification. G Retina of Aβ-expressing flies, 100× magnification. H Quantification of retinal thickness. Data are presented as individual values with mean ± SEM from biological replicates (n = 3 independent flies per group). Statistical analysis was performed using an unpaired two-tailed Student’s t-test. Differences were considered significant at p < 0.05. Treatment with the aqueous fraction from Mq-NUT-54B significantly reduced the total vacuole area (p < 0.05) compared to untreated samples. No significant difference in retinal thickness was observed (p > 0.05). In contrast, metabolites from Mq-ISP-1B did not significantly affect vacuole area (p > 0.05) but induced a significant reduction in retinal thickness (p < 0.05) (Fig. 5). Effect of aqueous fractions on histopathological parameters in a Drosophila melanogaster AD-like model. Representative histological Sect. (3 μm) of fly heads (optic lobe) from GMR-Gal4/+; UAS-Aβ/+ flies, 100× magnification. A–C Medulla: A untreated, B treated with Mq-ISP-1B aqueous fraction, and C treated with Mq-NUT-54B aqueous fraction; arrows indicate vacuolar lesions. D Quantification of total vacuole area. E–G Retina: E untreated, F treated with Mq-ISP-1B aqueous fraction, and G treated with Mq-NUT-54B aqueous fraction; arrows indicate the analyzed region. H Quantification of retinal thickness. Flies were analyzed at 1–2 days post-eclosion (d.p.e.). Data are presented as individual values with mean ± SEM from biological replicates (n = 3 independent flies per group). Statistical analysis was performed using one-way ANOVA followed by Holm–Šidák’s multiple-comparisons test. Differences were considered significant at p < 0.05. Human neurotoxicity Based on its superior in vivo efficacy, the aqueous fraction from Mq-NUT-54B was selected for evaluation in human neuronal cells. Treatment of SH-SY5Y cells with this fraction did not compromise neuronal viability at any of the tested concentrations when compared to untreated controls (Fig. 6), supporting their neuronal safety profile. Cytotoxicity assessment of the aqueous fraction in SH-SY5Y cells. Cell viability was evaluated using the AlamarBlue assay. Data are presented as mean ± SEM from technical replicates (n = 3). Statistical analysis was performed using one-way ANOVA followed by Holm–Šidák’s multiple-comparisons test. No significant differences were observed (p > 0.05). Cell culture Alzheimer-like model (preventive and treatment) Two experimental approaches were employed: a preventive model and a treatment model. In the preventive approach, co-incubation of synthetic Aβ monomers with the aqueous fraction did not prevent the loss of neuronal viability caused by Aβ aggregation (Fig. 7A). Effects of the aqueous fraction in SH-SY5Y Alzheimer-like models. A Preventive model. B Therapeutic model. Cell viability was assessed using the AlamarBlue assay. Data are presented as mean ± SEM from technical replicates (n = 3). Statistical analysis was performed using one-way ANOVA followed by Holm–Šidák’s multiple-comparisons test. Differences were considered significant at p < 0.05. In the treatment model, the aqueous fraction from Mq-NUT-54B was added after neuronal exposure to aggregated Aβ. Aggregated Aβ significantly reduced neuronal viability (p < 0.05). Treatment with the aqueous fraction significantly increased cell viability compared to the Aβ-only condition (p < 0.05 and p < 0.01) (Fig. 7B). Acetylcholinesterase and butyrylcholinesterase inhibition assay The aqueous fraction from Mq-NUT-54B significantly inhibited acetylcholinesterase (AChE) (p < 0.01) (Fig. 8A) and butyrylcholinesterase (BuChE) (p < 0.001) (Fig. 8B). Inhibitory activity of the aqueous fraction on cholinesterases. In vitro inhibition of acetylcholinesterase (AChE) (A) and butyrylcholinesterase (BuChE) (B). Data are presented as mean ± SEM from technical replicates (n = 3). Statistical analysis was performed using one-way ANOVA followed by Holm–Šidák’s multiple-comparisons test. Differences were considered significant at p < 0.05. Bacterial identification and genome features based on whole-genome sequencing Whole-genome sequencing of isolate Mq-NUT-54B revealed a draft genome of 4.08 Mb, with a GC content of 41.78% and 3,704 predicted coding sequences. Comparative genomic analyses indicated close relatedness to Providencia rettgeri. Detailed genomic comparisons are provided in the Supplementary Table 1. In addition, antimicrobial resistance screening using ABRicate identified the presence of the tet (57) gene, associated with resistance to tetracycline. No virulence-associated genes were detected in the genome based on in silico analysis. Putative identification of neuroactive metabolites by mass spectrometry LC-DAD-MS analysis of the aqueous fraction revealed multiple chromatographic peaks, indicating a chemically complex metabolite profile (Fig. 9). Subsequent LC-ESI-MS/MS analysis enabled the putative identification of several metabolites with known antioxidant or neuroactive properties, while the complete metabolite list is provided as Supplementary Table 2. LC–ESI–MS/MS chromatographic profile and putative compound identification of the aqueous fraction. Compound annotation was performed by LC–ESI–MS/MS and is considered putative, based on spectral matching and fragmentation patterns. The complete list of annotated metabolites is provided in Supplementary Table 2. Discussion Our research group has been investigating the microbiome present in the larval food of various Brazilian stingless bee species and establishing a collection of cultivable microorganisms. In addition, we are actively engaged in the nanobiotechnological exploration of these isolates as sources of antimicrobial compounds and in the green synthesis of nanoparticles, patented in Brazil (BR10202400128). Building on this bioprospecting framework, this study expands the current bioprospecting approach by evaluating the neuroprotective and anti-Alzheimer potential of bacterial metabolites derived from stingless bee-associated microbiota. This work represents an initial screening effort rather than a mechanistic characterization study. Oxidative stress plays a critical role in the pathophysiology of Alzheimer’s disease (AD) and is considered a key factor in the onset and progression of neurodegeneration. It results from an imbalance between the production of reactive oxygen species (ROS) and/or reactive nitrogen species (RNS) and the efficacy of endogenous antioxidant defense mechanisms4,28,29. In the central nervous system, excessive ROS levels can oxidize lipids, proteins, and nucleic acids, compromising cellular integrity and function. Neurons are particularly susceptible to oxidative damage due to their high metabolic activity and limited regenerative capacity, making the brain especially vulnerable to redox imbalances4,30. Antioxidant capacity was used as a rational criterion for selecting metabolites with potential neuroprotective effects. However, antioxidant activity is an indirect indicator and should not be interpreted as a definitive mechanism of action. In AD, oxidative stress contributes to hallmark pathological features, such as amyloid-β (Aβ) accumulation and tau hyperphosphorylation31. Elevated ROS levels promote Aβ aggregation and tau phosphorylation, thereby exacerbating mitochondrial dysfunction, synaptic loss, and neuroinflammation32,33. Moreover, oxidative stress can impair protein clearance and autophagy, further accelerating neuronal damage. Antioxidant capacity, therefore, represents one component of a broader multifactorial process rather than a single dominant mechanism underlying neuroprotection32. While dietary intake of exogenous antioxidants is primarily associated with plant-derived sources, increasing evidence highlights that other organisms—including fungi, and prokaryotes from the domains Bacteria and Archaea—also constitute valuable alternative sources of antioxidant compounds. These non-plant organisms may offer structurally diverse metabolites with unique bioactivities, expanding the repertoire of antioxidant candidates for therapeutic applications24,34–36. In this study, we screened a collection of microorganisms associated with Brazilian stingless bee species and identified a bacterial isolate exhibiting a high level of antioxidant compound secretion. To our knowledge, this may represent one of the first reports of a bee-associated bacterium with such properties, although this should be interpreted with caution due to the limited number of studies in this ecological niche. This finding underscores the neurobiological relevance of microbial communities derived from Brazilian biodiversity. As a continuation of this biotechnological prospecting, this study screened seven bacterial isolates for their potential to secrete compounds with anti-Alzheimer activity. One of them, isolate Mq-NUT-54B, demonstrated strong potential to secrete antioxidants and anti-Alzheimer compounds. Whole-genome sequencing showed that this isolate belongs to the Providencia genus. This genus has not previously been recognized as a relevant source of neuroactive metabolites. To date, only one study has reported a metabolite from the P. rettgeri strain BTKKS1 from the marine environment as an antioxidant compound37, highlighting the novelty of the present findings. However, current evidence is insufficient to generalize this observation, and further strain-level investigations are required. In addition, in silico genome mining revealed a limited resistome profile, with the identification of a single tetracycline resistance-associated (tet) gene38. This suggests a low burden of antibiotic resistance determinants in the genome, although their functional roles remain to be confirmed. No virulence-associated genes were detected, based on current database annotations39. Together, these results indicate a minimal resistome and absence of identifiable classical virulence factors in Providencia sp. Mq-NUT-54B, supporting its potential for biotechnological applications. However, experimental validation remains necessary to confirm the functional relevance of the detected tet gene. Sequential liquid-liquid partitioning of the bacterial culture supernatant using solvents of increasing polarity—hexane, dichloromethane (DCM), ethyl acetate (EtOAc), and n-butanol (n-BuOH)—proved to be an effective strategy for fractionating extracellular metabolites based on their physicochemical properties40.This approach enriches compounds according to polarity and facilitates bioactivity-guided fractionation41–43. In this work, only the aqueous fraction was selected for further analysis due to its superior activity in three antioxidant assays: DPPH, FRAP, and ORAC. This result indicates that the most active antioxidant compounds secreted by the isolate are likely highly polar. Notably, this fraction also demonstrated significant neuroprotective effects in subsequent in vivo and in vitro Alzheimer’s disease models. However, this selection strategy prioritizes antioxidant performance and does not exclude the possible contribution of less polar fractions. The use of the GMR-GAL4 driver system in Drosophila melanogaster represents a well-established approach for modeling Alzheimer’s disease-like phenotypes44–47. In this model, Aβ expression in the compound eye induces progressive neurodegeneration, characterized by a rough eye phenotype. Here, treatment with the aqueous fraction from Mq-NUT-54B significantly reduced Aβ burden and attenuated histopathological hallmarks of neurodegeneration, including vacuolization within the optic lobe, indicating a protective effect in vivo. However, due to the tissue-specific expression of Aβ in this model, these outcomes should be interpreted as histopathological improvements rather than as organism-level functional recovery. Another widely used model involves the treatment of cultured human neuronal cells with synthetic Aβ peptides44,48,49. Consistent with the in vivo findings, the aqueous fraction from Mq-NUT-54B did not compromise neuronal viability and significantly restored cell survival when administered after Aβ aggregation, suggesting a predominantly therapeutic rather than preventive effect, although further studies are required to confirm this interpretation. Cholinergic dysfunction represents another central feature of AD pathophysiology50,51. In addition to acetylcholinesterase (AChE), butyrylcholinesterase (BuChE) activity increases in later stages of the disease50,52–55. Here, the aqueous fraction exhibited inhibitory activity against both enzymes. While this dual inhibition may contribute to the observed neuroprotective effects, it should be interpreted as part of a multifactorial response rather than a single dominant mechanism. LC-ESI-MS/MS analysis of the aqueous fraction revealed a complex metabolite profile comprising several compounds previously associated with neuroprotective or anti-amyloidogenic effects56–67. However, these identifications are putative, based on spectral matching, and require further validation. Furthermore, the presence of these metabolites does not establish a direct causal relationship with the observed biological effects. Neuroprotective activity may instead result from the combined action of multiple compounds within this complex mixture. Therefore, the observed bioactivity should be interpreted as property of the metabolite mixture rather than attributed to individual compounds. The bacterial isolate Mq-NUT-54B was initially assigned to Providencia rettgeri by MALDI-TOF26. However, comparative genomic analyses indicated that only a limited number of publicly available genomes meet accepted species-delineation thresholds68–71. Although taxonomic classification was not the primary aim of this study, accurate species identification remains important for reproducibility and future applications, and safety considerations should also be considered when interpreting the potential biotechnological applications of this isolate. HPLC–MS/MS analysis provides valuable insights into the metabolite composition of the aqueous fraction, enabling the detection and annotation of compounds and an overview of chemical diversity72. However, metabolite identification based on this approach remains putative, as spectral matching depends on database coverage and may not fully capture novel or low-abundance metabolites73. In addition, the chemical complexity of bacterial supernatants may affect compound detection and resolution. Therefore, complementary analytical techniques, such as nuclear magnetic resonance (NMR) and high-resolution mass spectrometry, are required for definitive structural elucidation74. Despite these limitations, the results highlight a chemically rich fraction with potential biological relevance and antioxidant activity, a property widely associated with microbial and natural product-derived metabolites in oxidative stress-related contexts1–4. Nevertheless, further studies involving compound isolation and purification (e.g., C18 fractionation), followed by structural elucidation using complementary techniques such as nuclear magnetic resonance (NMR), high-resolution mass spectrometry, and X-ray crystallography, are required74. Moreover, the biological assays were conducted using a complex mixture rather than isolated compounds, which limits mechanistic interpretation, a common limitation in bioactivity-guided screening approaches25,37,76. Future investigations should also include additional in vivo models, such as neuronal-specific Drosophila models, behavioral analyses, and validation in mammalian systems, which are widely used for translational validation in neurodegenerative disease research43–47,75. These future directions are essential to move from bioactivity-guided screening toward mechanistic and translational validation. In summary, the aqueous fraction derived from Mq-NUT-54B ameliorated Alzheimer-like phenotypes in both in vitro and in vivo models. These findings support a multifactorial and not yet mechanistically defined mode of action, potentially involving antioxidant activity, modulation of amyloid-related toxicity, and cholinergic enzyme inhibition. While further studies are required to validate these effects and identify the active compounds, our results highlight stingless bee-associated microbiota as a promising and underexplored source of neuroactive metabolites with potential relevance for Alzheimer’s disease. Materials and methods Bacterial isolates and culture conditions Seven bacterial isolates obtained from the larval food of Brazilian stingless bees and deposited in the CoMISBee collection were used in this study (Table 1). Some isolates had been previously identified by MALDI-TOF mass spectrometry25. Bacterial cultures were grown under standard laboratory conditions, and cell-free supernatants were obtained by centrifugation for subsequent analyses. Table 1Bacterial Strains and Their Codes Utilized by CoMISBee.Stingless Bees Larval FoodCoMISBee CodeBacterial Identification Melipona quadrifasciata Mq-ISP-1B Proteus mirabilis Mq-MRS-9B Providencia rettgeri Mq-TSA-12 Not identified Mq-BHI-20 Vagococcus fluvialis Mq-OAT-27 Bacillus velezensis Mq-NUT-54B Providencia rettgeri Melipona scutellaris Ms-BHI-39B Providencia rettgeri Bacterial Strains and Their Codes Utilized by CoMISBee. Screening for antioxidant strains To evaluate the antioxidant potential of bacterial metabolites, cell-free supernatants from the seven isolates were initially subjected to fractionation into protein-enriched and metabolite-enriched fractions. To obtain the protein and metabolite-enriched fractions, 5 mL of the supernatants were lyophilized and resuspended in 5 mL of methanol, followed by incubation at 4 °C for 3 h and centrifugation at 8000 g for 5 min76. The protein precipitate was separated from the metabolite, and the solvent was evaporated in a speed vacuum. Both protein-enriched and metabolite-enriched fractions were initially screened; only metabolite-enriched fractions were selected for downstream assays. Lyophilized supernatants were reconstituted in water at one-fifth (1/5) of their original volume before analysis. All experiments were performed in triplicate. Antioxidants assays The antioxidant capacity of the fractions of supernatants was evaluated using three different assays: DPPH radical scavenging activity, Ferric Reducing Ability of Plasma (FRAP), and Oxygen Radical Absorbance Capacity (ORAC). For the DPPH assay, the supernatant fractions were mixed with 60 mM DPPH (diluted in methanol) and incubated in the dark for 20 min at 30 °C. The absorbance was then measured at 517 nm. The radical scavenging activity was calculated using the formula: DPPH (%) = [(A control - A sample)/(A control - A blank)] x 100, where A sample represents the absorbance of the DPPH solution with the sample or positive control, A control corresponds to the absorbance of the DPPH solution alone, and A blank accounts for the absorbance of the sample in methanol. Ascorbic acid and methanol were used as positive and negative controls, respectively27,77,78. For the FRAP assay, the fractions were combined with the FRAP reagent, which consisted of 10 volumes of 300 mM sodium acetate buffer (pH 3.6), 1 volume of 10 mM 2,4,6-tris(2-pyridyl)-s-triazine (TPTZ), and 1 volume of 20 mM ferric chloride. The mixture was incubated for 6 min at 37 °C, and the absorbance was recorded at 593 nm. Antioxidant capacity was determined using an analytical curve with Trolox as a standard. Sodium acetate buffer and ascorbic acid were employed as negative and positive controls, respectively27,75,77. For the ORAC assay, the fractions were inoculated in a plate with 0.085 mM fluorescein and incubated for 15 min at room temperature. Afterward, 153 mM 2,2’-azobis(2-amidinopropane) dihydrochloride was added, and fluorescence (485 nm excitation/528 nm emission) was measured for 90 min at 37 °C. All reagents were prepared in 75 mmol/L phosphate buffer (pH 7.4). Antioxidant capacity was determined using an analytical curve constructed with Trolox as a standard. Phosphate buffer and ascorbic acid were used as negative and positive controls, respectively27,77. Liquid–liquid partitioning The fractions that showed the highest antioxidant activity in the FRAP, DPPH, and ORAC assays were subjected to liquid-liquid partitioning (Bittar et al., 2024). This process was carried out sequentially using the following solvents: hexane (HEX), dichloromethane (DCM), ethyl acetate (EtOAc), and n-butanol (n-BuOH). After vacuum drying, the metabolite material was resuspended in 5 mL of water and vortexed for 15 s or until complete homogenization. Next, 5 mL of HEX was added to each tube, followed by vortexing for 30 s to promote phase separation, yielding a HEX phase and a residual metabolite phase. The HEX phase was carefully removed with a pipette, and the procedure was repeated. Each tube underwent two washes with each solvent in the specified order. At the end of the process, the phase that did not interact with any of the solvents was designated as the aqueous phase43,77. After solvent extraction, the residual aqueous phase was collected, dried, resuspended in distilled water, and selected for subsequent biological assays due to its consistently superior antioxidant activity. The other fractions (hexane, dichloromethane, ethyl acetate, and n-butanol) showed markedly lower or negligible antioxidant activity under the same conditions, supporting the selection of the aqueous fraction for downstream analyses. Chemical characterization of metabolites by UFLC–DAD–MS The chemical profile of the aqueous fraction obtained from the Mq-NUT-54B isolate was analyzed using an LC-20AD ultra-fast liquid chromatograph (UFLC, Shimadzu) coupled to a diode array detector (DAD) and a micrOTOF-Q III mass spectrometer (Bruker Daltonics) equipped with an electrospray ionization (ESI) source and quadrupole time-of-flight (QTOF) analyzer. For MS analysis, nitrogen was used as the nebulizer gas (4 bar), drying gas (9 L/min), and collision gas. The capillary voltage was set at 4.5 kV. Data were acquired in both positive and negative ion modes over an m/z range of 120–1300. Chromatographic separation was performed using a Kinetex C18 column (2.6 μm, 150 × 2.1 mm, Phenomenex) maintained at 50 °C, with a flow rate of 0.3 mL/min. The mobile phase consisted of deionized water (A) and acetonitrile (B), both containing 0.1% formic acid. The gradient elution profile was as follows: 0–2 min, 3% B; 2–25 min, 3–25% B; 25–40 min, 25–80% B; and 40–43 min, 80% B, followed by 5 min for column washing and reconditioning79,80. The aqueous fraction was prepared at a concentration of 1 mg/mL in methanol: water (7:3, v/v), and 1 µL was injected into the system via autosampler. Metabolites were putatively annotated based on MS/MS fragmentation patterns and UV spectra, in comparison with literature data and available spectral databases, and are therefore classified as Level 2 identification according to the Metabolomics Standards Initiative. No compounds were structurally confirmed using authentic standards; thus, all annotations should be interpreted as tentative. Drosophila stock The Drosophila melanogaster strains utilized in this study included w1118 (BL#3605), UAS-Aβ (UAS-Aβ1–42 BL#64216), and GMR-Gal4 (GAL4-ninaE.GMR BL#1104), all of which were sourced from the Bloomington Drosophila Stock Center. The flies were reared under standard Bloomington culture conditions, maintained at 25 °C, and subjected to a 12-hour light/12-hour dark cycle during the colony expansion phase44. To generate the desired genotype, virgin UAS-Aβ females were mated with GMR-Gal4 males. The mating pairs were placed in an egg-laying medium, and embryos were collected after 8 h of oviposition. Parental control groups were established using the same protocol by crossing GMR-Gal4 with w1118 and UAS-Aβ x w1118. This model allows the evaluation of neurodegeneration at the histopathological level but does not capture organismal outcomes such as behavior or lifespan due to the eye-specific expression of Aβ under the GMR-Gal4 driver47. Treatment of Drosophila melanogaster with bacterial aqueous fraction Embryos were transferred to vials containing 1 g of enriched mashed potato medium (75 g instant mashed potato flakes, 15 g yeast Extract, 9.3 g glucose, 0.7 g nipagin), hydrated with 5 mL of the respective treatment solution or control (medium with water only). The aqueous fraction was administered at a final concentration of 62.5 µg/mL in the food medium. This concentration was selected based on preliminary in vitro antioxidant activity and cytotoxicity screening assays. At 48 and 72 h, an additional 100 µL of treatment solution was applied to the surface of the medium. Upon eclosion, flies aged 0–1 days post-eclosion (d.p.e.) were collected, sorted by phenotype, and maintained in untreated media for 24 h before further analysis44. Relative quantification of Aβ using Thioflavin T (ThT) To verify the amyloidogenic pathway in an Alzheimer’s disease (AD)-like model, relative amyloid levels were assessed using ThT. The quantification method for Aβ with ThT was previously standardized in earlier studies44,46,75. Flies aged 1 to 2 days post-eclosion (d.p.e.) were collected from all experimental and control groups, euthanized with liquid nitrogen, and preserved at −80 °C for further procedures. On the day of the assay, fly heads were harvested and homogenized in 1X PBS buffer while maintaining the process on ice. The homogenate was subjected to centrifugation at 10,000 × g for 2 min at 4 °C, and the supernatant was collected for total protein measurement using the Bradford method. For amyloid quantification, the Thioflavin T assay was performed by incubating 2 µL of the sample in a 96-well black plate containing 198 µL of filtered 10 µM ThT solution, with the plate agitated for 20 min. Fluorescence was measured at an excitation wavelength of 450 nm and an emission wavelength of 482 nm, followed by multiple normalization steps. Initially, background autofluorescence of ThT in the absence of protein was subtracted. Subsequently, fluorescence readings were adjusted based on protein concentration, expressed in micrograms (µg). Finally, these values were normalized to the GMR-Gal4/+ control, which does not express Aβ42, by dividing the sample fluorescence by the average fluorescence of the control group. For the experiment, three pools of ten heads per group were prepared, and the procedure was conducted in quintuplicate. Histopathological analysis For histological analysis, five adult flies of the GMR-Gal4;UAS-Aβ or control with 1–2 days post eclosion were collected from each control/treatment group, anesthetized with ethyl ether and fixed in Carnoy solution (6: 3: 1, 99% ethanol, chloroform and glacial acetic acid) for 24 h and processed in 70% ethyl alcohol (2x), 80% ethyl alcohol (2x), 90% ethyl alcohol (2x), absolute ethyl alcohol (2x), and xylol (2x) for 15 min in each repetition and 60% liquid paraffin (2x) for 30 min. The fly heads were embedded in paraffin, and the blocks were sectioned at 3 μm thickness using a semiautomatic microtome (SLEE CUT5062). The sections were hydrated, stained with hematoxylin and eosin, mounted, and photographed with a light photomicroscope. Medulla and retina (optic lobe) of three or more adult flies were used to calculate the area of vacuolar lesions and retinal thickness using the ImageJ software. Synthetic Aβ preparation The Aβ peptide was synthesized by AminoTech (Brazil) with a purity of 95%. Following the methodology outlined by Ryan et al. (2013)81, the synthetic peptide was initially dissolved in 10% (w/v) ammonium hydroxide at a concentration of 0.5 mg/mL. The solution was then incubated at room temperature for 10 min, subjected to 5 min of sonication, aliquoted, and subsequently lyophilized. The resulting lyophilized peptide was reconstituted in 60 mM NaOH, yielding a stock solution with a final concentration of 886 µM, and stored at −20 °C for further use. Human neuronal cell culture The SH-SY5Y human neuroblastoma cell line, originally donated by the University of Siena, Italy, was cultured under standard conditions at 37 °C in a humidified atmosphere containing 5% CO₂. The cells were maintained in Dulbecco’s Modified Eagle Medium (DMEM, Cultilab®), supplemented with 10% fetal bovine serum (FBS, Cultilab®) and 1% antibiotic solution (Vitrocell®). Neurotoxicity assay Cells were seeded in a 96-well plate at a density of 10,000 cells per well in 100 µL of DMEM and incubated for 24 h to allow for adherence. Following this incubation, the aqueous fraction of the Mq-NUT-54B was introduced at specific concentrations, with each experimental group receiving a distinct treatment. The aqueous fraction was tested at final concentrations of 62.5, 31.25, and 15.63 µg/mL to allow preliminary evaluation of concentration-dependent effects. The control groups included untreated cells as the negative control and cells exposed to 0.2% Triton-X as the positive control. Although no classical neuroprotective reference compound was included in the present study, this work was designed as an initial screening platform for bioactivity prioritization. At the conclusion of the treatment, cell viability was assessed using the AlamarBlue assay (Invitrogen®). For this, 10 µL of resazurin solution (10%) was added to each well, and the plate was incubated at 37 °C for approximately 3 h. Resazurin, a non-fluorescent blue dye, undergoes intracellular reduction to resorufin, which exhibits a strong fluorescent pink signal. The resulting fluorescence was then quantified using colorimetric absorbance readings at 570 and 600 nm. Cell culture Alzheimer’s disease models Preventive model The synthetic Aβ peptide (stock concentration 886 µM) was initially diluted to 1 µM and sterilized using a 0.22 μm Millipore filter. The 1 µM concentration of Aβ was selected based on previous optimization studies demonstrating consistent induction of neuronal toxicity without excessive cell death under our experimental conditions44. To assess the effectiveness of the treatment in preventing Aβ peptide aggregation, the peptide was combined with the aqueous fraction and incubated at 37 °C for 48 h. The aqueous fraction was tested at final concentrations of 62.5 and 31.25 µg/mL. Meanwhile, cells were plated in a 96-well plate at a density of 5,000 cells per well, supplemented with 100 µL of DMEM per well, and incubated for 24 h to facilitate cell adhesion. Following this period, the Aβ peptide and metabolites in aqueous fractions were introduced into the cell culture and incubated for an additional 48 h, resulting in a total experimental duration of 96 h. For accurate data interpretation, the study included three control groups: a positive control consisting of untreated cells, a negative control of dead cells incubated in DMEM with 0.2% Triton-X, and a control group with untreated Aβ peptide. Therapeutic model The amyloid-beta (Aβ) peptide was diluted to a final concentration of 1 µM, passed through a 0.22 μm membrane for sterilization, and then incubated at 37 °C for 48 h to promote peptide aggregation. Meanwhile, cells were plated in a 96-well plate at a density of 5,000 cells per well in 100 µL of DMEM and maintained under standard conditions for 24 h. After that, the pre-aggregated Aβ peptide was introduced into the cell culture, and incubation continued for an additional 48 h to encourage the formation of senile plaques. After this period, cells were treated with the aqueous fraction at final concentrations of 62.5 and 31.25 µg/mL, and incubation was prolonged for another 24 h to evaluate their potential in reversing pre-established senile plaques. Acetylcholinesterase and butyrylcholinesterase inhibition assay The inhibitory activity of the supernatant fractions against acetylcholinesterase (AChE) and butyrylcholinesterase (BuChE) was assessed following the protocol described by Rhee et al. (2001), with modifications. Three buffer systems were employed: (A) 50 mM Tris-HCl, pH 8.0; (B) 50 mM Tris-HCl, pH 8.0, containing 1% (w/v) bovine serum albumin (BSA); and (C) 50 mM Tris-HCl, pH 8.0, supplemented with 0.6% (w/v) NaCl and 0.4% (w/v) MgCl₂. The aqueous fraction from the 54B supernatant and positive controls (galantamine and rivastigmine) were dissolved in buffer A at a final concentration of 5.5 mg/mL. Subsequently, they were incubated with a reaction mixture consisting of 3 mM acetylthiocholine iodide (for AChE) or 3 mM butyrylthiocholine iodide (for BuChE), 50 mM Tris-HCl buffer with 1% BSA (buffer B), and 3.06 mM 5,5′-dithiobis (2-nitrobenzoic acid) (DTNB) prepared in buffer C. The enzymatic reaction was initiated by adding 0.2 U/mL of either AChE or BuChE to the mixture. Absorbance readings were immediately recorded at 405 nm using a microplate reader (Molecular Devices, Menlo Park, CA, USA) for a total duration of 30 min. Enzyme inhibition was calculated using the following equation: % Inhibition =[(Acontrol −Asample)/Acontrol] ×100, where: Acontrol is the absorbance (or activity) of the enzyme without the inhibitor (negative control), and Asample is the absorbance (or activity) of the enzyme in the presence of the inhibitor82. Galantamine and rivastigmine were included as positive controls for assay validation and benchmarking of inhibitory activity. Genome sequencing, assembly, and bioinformatics analysis To identify the isolate Mq-NUT-54B, which displayed the strongest anti-Alzheimer activity, genomic DNA was extracted for sequencing. A culture of the Mq-NUT-54B strain was grown in BHI agar and incubated for 24 h at 37 °C. After that, well-developed colonies were collected using a bacteriological loop and suspended in 1 mL of extraction buffer (100 mM Tris-HCl, 500 mM NaCl, 50 mM EDTA, and 0.4% SDS). The suspension was vortexed for 15 s, then 50 µL of proteinase K was added, and the mixture was incubated at 56 °C for 30 min. After incubation, samples were centrifuged at 14,000 rpm for 10 min at room temperature. The supernatant was transferred to a clean microtube containing 330 µL of isopropanol, gently inverted 30 times, and centrifuged again at 14,000 rpm for 10 min. The resulting pellet was washed twice with 500 µL of 70% ethanol, centrifuged for 30 s between washes, and air-dried for 15 min. The DNA pellet was then resuspended in 50 µL of ultrapure water, and its quality and concentration were verified using a Nanodrop 2000 spectrophotometer (Thermo Fisher Scientific, Waltham, MA) and agarose gel electrophoresis. Genomic libraries were prepared using the Illumina DNA Prep Kit following the manufacturer’s protocol, including quality assessment, quantification, indexing with unique dual adapters (barcodes), and pooling for sequencing. Sequencing was conducted on the Illumina platform using a paired-end (2 × 150 bp) configuration, yielding approximately 1 Gb of data. For the Mq-NUT-54B isolate, a total of 30,012,500 reads were generated. Sequencing reads were quality-checked and trimmed using SOAPnuke v2.1.983 to remove adapters and low-quality bases. Contigs shorter than 150 bp or containing ≥ 0.1% ambiguous nucleotides were discarded. De novo assembly was performed with SPAdes v3.15.4, and taxonomic identification was confirmed through BLASTN alignment against the NCBI database, supplemented by an in-house classification algorithm84. The in-silico DNA-DNA hybridization (dDDH) was calculated using the GGDC https://ggdc.dsmz.de/home.php85. The Average Nucleotide Identity (ANI) was calculated using the ANI calculator from http://fbac.dmicrobe.cn/. NCBI was queried in October 2025 using the filter ‘Providencia rettgeri AND assembly level: chromosome OR complete’. Fifty genomes were retrieved and included in the ANI and dDDH analyses. The genomes used were: GCA_019048105.1, GCF_019048105.1, GCA_001874625.1, GCF_001874625.1, GCA_019047885.1, GCF_019047885.1, GCA_010320145.1, GCF_010320145.1, GCA_013702025.1, GCF_013702025.1, GCA_010318885.1, GCF_010318885.1, GCA_005155965.1, GCF_005155965.1, GCA_023650895.1, GCF_023650895.1, GCA_029011985.1, GCF_029011985.1, GCA_013255915.1, GCF_013255915.1, GCA_014394705.1, GCF_014394705.1, GCA_025916175.1, GCF_025916175.1, GCA_025583505.2, GCF_025583505.2, GCA_034330845.2, GCF_034330845.1, GCA_038442735.1, GCF_038442735.1, GCA_044788375.1, GCF_044788375.1, GCA_002984195.1, GCF_002984195.1, GCA_018771265.1, GCF_018771265.1, GCA_016406205.1, GCF_016406205.1, GCA_023205015.1, GCF_023205015.1, GCA_013423885.1, GCF_013423885.1, GCA_020985345.1, GCF_020985345.1, GCA_040208285.1, GCF_040208285.1, GCA_044162235.1, GCF_044162235.1, GCA_044162305.1, GCF_044162305.1, GCA_019048545.1, GCF_019048545.1, GCA_020683065.1, GCF_020683065.1, GCA_020808945.1, GCF_020808945.1, GCA_018861255.2, GCF_018861255.2, GCA_018861235.2, GCF_018861235.2, GCA_018861215.2, GCF_018861215.2, GCA_019890815.1, GCF_019890815.1, GCA_044162895.1, GCF_044162895.1, GCA_010319405.1, GCF_010319405.1, GCA_041075285.1, GCF_041075285.1, GCA_030062245.1, GCF_030062245.1, GCA_010319105.1, GCF_010319105.1, GCA_003204135.1, GCF_003204135.1, GCA_023184555.1, GCF_023184555.1, GCA_013702265.1, GCF_013702265.1, GCA_014489375.1, GCF_014489375.1, GCA_000314835.2, GCF_000314835.2, GCA_013694385.2, GCA_015571575.1, GCF_015571575.1, GCA_013283975.1, GCA_036898195.1, GCA_042142345.2, GCA_022846595.1, GCA_050990565.1, GCA_013702245.1, and GCF_013702245.1.” The ABRicate on the Galaxy Platform was used to mine antimicrobial resistance genes and virulence factors86. Statistical analysis Statistical analyses were performed using GraphPad Prism (version 10). Data normality was assessed using the D’Agostino–Pearson test, and all datasets met the assumptions required for parametric analysis. Accordingly, comparisons between two groups were performed using an unpaired, two-tailed Student’s t-test. Whereas comparisons involving more than two groups were analyzed using one-way analysis of variance (ANOVA) with Holm–Šidák’s multiple-comparisons test. Data are presented as mean ± standard error of the mean (SEM). For biochemical and cell-based assays, values represent technical replicates (n = 3). For Drosophila experiments, values represent independent biological replicates (n = 3), as specified in figure legends. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 2 Acknowledgements We thank to Rede Genoma de Minas Gerais, and FAPEMIG, and CNPQ for making this research possible. Author contributions ACCS and CUV conceived and designed the study. ACCS performed the experiments, analyzed the data, and wrote the original draft of the manuscript. JLC contributed to data analysis and interpretation. RMFD and FSE contributed to the antioxidant analyses. SMM performed the Drosophila assays. TSR, DOS, and PRF carried out the histopathological analyses. FNAPM and RGZ performed bioassays and contributed to experimental validation. NSC, CAC, and MMM conducted the metabolomic analyses and metabolite identification. ARS and CUV contributed to the bioinformatics analyses. APMS and AMB contributed to manuscript revision and project supervision. CUV acquired funding, supervised the study, and reviewed and edited the final version of the manuscript. Funding This project was funded by the Research Support Foundation of the State of Minas Gerais (FAPEMIG APQ-02766-17, APQ-00269-22) and the National Council for Scientific and Technological Development (CNPq, grant number: 403193/2022-2), as well as FAPEMIG (grant number: CBB-APQ-03613-17) for INCT-TeraNano. C U-V received a CNPq research productivity fellowship (PQ—Research Productivity, process 303667/2021-4). Data availability The datasets generated and/or analysed during the current study are available in the NCBI BioProject repository under BioProject accession number PRJNA860336 and BioSample SAMN56532426. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Ana Carolina Costa Santos, Email: ana.carolina@ufu.br. Carlos Ueira-Vieira, Email: ueira@ufu.br. References Associated Data Supplementary Materials Supplementary Material 1 Supplementary Material 2 Data Availability Statement The datasets generated and/or analysed during the current study are available in the NCBI BioProject repository under BioProject accession number PRJNA860336 and BioSample SAMN56532426.