Marine Fungal Metabolite Butyrolactone I Prevents Cognitive Decits and Inammation Evoked by AlCl3 in zebrash

Mounting evidences indicate that oxidative stress and neuroinammation are related to neurodegenerative disorders (NDs). Butyrolactone I (BTL-I), a marine fungal metabolite, was previously reported as an in-vitro neuroprotectant and inammation inhibitor. However, little is known about its in-vivo effects. Zebrash (Danio rerio) could be used as a convenient model in evaluation of toxicology and central nervous system (CNS) diseases.

Notably, current clinical drugs for the prevention and treatment of NDs just manifest limited e cacy. For example, donepezil partially relieve symptoms of AD without reversing or preventing its progression [5].
The poor understanding of NDs pathogenesis restricted new drug development [17,18]. The repeated failure to develop anti-AD drugs targeting orphan targets, such as Aβ and Tau, may be substantially related to malignant ampli cation induced by neuroin ammation and oxidative stress. Thus, inhibiting in ammation and oxidative stress to protect neurons and intervene in the early stage of disease has become important new strategy in developing novel anti-ND agents [19]. Aryl butyrolactones (BTLs), including butyrolactone I (BTL-I, Fig. 1 inset), are characteristic natural products of fungi (e.g., Aspergillus sp. and Penicillium sp.) [20,21]. Our previous studies have shown that BTL-I has strong in vitro antineuroin ammatory effects, inhibiting LPS-induced in ammatory proliferation of microglia, the release of in ammatory mediators (NO and IL-1β) and ROS, as well as the expression of in ammatory target enzyme cyclooxygenase-2 (COX-2), and intracellular migration of the signaling protein NF-κB p65 [22].
Moreover, BTL-I plays a versatile anti-neurodegenerative role through multiple mechanisms, such as neuronal nutrition and inhibiting neuronal injury [23][24][25][26]. However, since previous BTL-I studies were limited to in vitro, its in vivo effects remain to be elucidated.
Due to their high genetic and physiological homology to humans, the zebra sh has long been used as an powerful in vivo model to assess anti-in ammatory drugs [27,28] and neuronal injury [29]. Zebra sh possess innate and acquired immune systems similar to those of mammals [30], and display wellcharacterized learning, memory, addiction and other behaviors that corresponding to clinical phenotypes [31][32][33][34].
AlCl 3 causes AD-like pathology, aggravating neuroin ammation, oxidative stress and AChE activity in the rodent's brain [12,35,36]. The effect of AlCl 3 on zebra sh cognition remains unclear [37]. In the present study, we established a neurotoxic zebra sh model (induced by AlCl 3 ) to assess potential effects of BTL-I on memory and cognitive impairment in vivo. The putative neuroprotective activity of BTL-I in zebra sh was further investigated by evaluating AChE activity and GSH levels in the brain, and by detecting the levels of the in ammatory cytokines IL-1β and TNF-α in both central and peripheral tissues. 16S rDNA high-throughput method was used to determine the structure and changes of zebra sh intestinal ora. The in-silico tool of ADMETlab was applied to evaluate the absorption, distribution, metabolism, excretion, and toxicity (ADMET) and drug-likeness properties of BTL-I.

Animals and model development
Adult wild-type AB zebra sh (approximately 6-8 months old; 50:50% male: female ratio) used in the present study were obtained from a commercial supplier (Shanghai Jiayu Aquarium, Shanghai, China) and acclimatized in a 50 L aquarium in the aquatic facility of the Guangdong Ocean University for at least 2 weeks. The sh were kept on a 14 h: 10 h light: dark cycle (lights on at 7 am) at a temperature of 25 ± 2 ℃ in a recirculating tank system. The zebra sh were maintained according to standard conditions [38], and fed Artemis larvae twice a day, at 9 am and 2 pm.
As shown in Fig. 1, to establish an AlCl 3 model, 75 zebra sh (3.0 ± 0.4 cm in length) were randomly divided into control, AlCl 3 and BTL-I treated groups (n = 15 per group). BTL-I (25, 50 or 100 mg/kg/day) was administered with food for 20 days. The control and AlCl 3 groups were fed equal amounts of normal food. 20 days later, the AlCl 3 and BTL-I groups were anesthetized and injected (i.p.) with AlCl 3 solution (4.2 mg/mL, 5 µL, pH = 5.0 ± 0.2) using a 10-µL gas phase injection needle 0.5 mm in the outer diameter. The control group was injected with the same amount of saline. Memory testing was performed 24 h later.
Brie y, following a 24-h fasting, the sh were anesthetized and then received an i.p. injection. For this, eugenol was dissolved in 100 mL anhydrous ethanol to prepare a 1 mg/mL stock solution that was added to 5 L of water (28 ± 1 ℃) and stirred evenly. Zebra sh were then group-exposed to the anesthetic, and after stopping swimming (immobile > 2-3 min), quickly injected with AlCl 3 . After the injection, the animals recovered in a water-containing beaker and returned to the holding tanks once their normal swimming resumed [39].

T-maze behavioral testing
The aquatic T-maze was used for cognitive testing, as described previously, with modi cations [40,41].
The maze comprised a long vertical arm (50 cm) and two short horizontal arms (20 cm), with an arm width of 10 cm, a depth of 10 cm, and the water depth of 8 cm. The right arm was connected to a rectangular water tank (22 cm×20 cm×15 cm) with a black outer wall; sand and stones were added to the bottom of the tank, and bait was set inside the tank to provide an enriched chamber (EC) (Fig. 1). During the nal 4 days of treatment, 6 sh were randomly selected from each group and the sh were individually trained for 5 min daily to locate the EC zone. If a sh did not enter the EC zone within the 5min training session, it was guided into the EC zone and kept there for > 30 s. Following 4 days of training, one day later the trained sh were placed individually into the starting area of the long arm for behavioral testing, scoring the latency time (s) to enter the EC zone and stay there for > 30 s. If a sh did not enter the EC zone within the 5-min test, the latency time was recorded as 300 s. Behavioral testing was performed between 10 am and 1 pm. A Microsoft LifeCam Studio 1080p HD camera was used to record videos with Apowersoft software (Apowersoft Co. Ltd., Hong Kong, China). The Supersys software was used for off-line video analyses (Shanghai Xinruan Information Technology Co. Ltd., Shanghai, China), assessing the latency of the rst entry into the EC zone (s), the average swimming speed (cm/s), and the number of EC entries. Reagents BTL-I was isolated for this study from the marine fungus Aspergillus terreus C23-3 as described previously [22]. The BCA protein-, GSH-and AChE assay kits, sh IL-1β enzyme-linked immunosorbent assay (ELISA) kit and sh TNF-α ELISA kit were purchased from Nanjing Jiancheng Bioengineering Institute (Nanjing, China). Eugenol and AlCl 3 were purchased from Huaxia Reagent (Chengdu, China) and Xiya Reagent (Shandong, China), respectively.

Molecular biomarker assays
Twenty-four hours after behavioral testing, the sh were euthanized. Considering that the available assay kits could not measure the tiny tissue samples of individual sh, the 15 sh in each group were randomly divided into 3 subgroups, and samples of brain, peripheral tissue and intestinal tract from each subgroup (5 sh) were collected and combined immediately, and freeze-dried at -80 ℃. All the samples (except for the gut samples) were homogenized in phosphate-buffered solution (PBS) for further assays. The supernatant was collected by centrifugation at 252 g at 4 ℃ for 15 min. Zebra sh brain sample supernatants were used to determine GSH levels and AChE activity. Moreover, zebra sh brain sample and peripheral tissue supernatants were also used to determine the levels of IL-1β and TNF-α [42], following the manufacturer instructions. The results were expressed as U of AChE/mg of protein and µmol of GSH/g of protein. Regression equations for the IL-1β and TNF-α standard curves were calculated according to the OD value, and logistic curves (4 parameters each) were used as the tting models.

Statistical analysis for behavior and molecular biomarkers
Statistical analysis was performed by one-way analysis of variance (ANOVA) followed by a post-hoc Dunnett test. The results are expressed as the mean ± SD. P-value was set at < 0.05 for all tests.
Gut ora sequencing and data analysis Genomic DNA was extracted by protease K lysis. The variable region of 16S ribosomal RNA gene V3-V4 was ampli ed by PCR, and the speci c primer sequences were as follows: 357F 5'-ACTCCTACGGRAGGCAGCAG-3' and 806R 5'-GGACTACHVGGGTWTCTAAT-3'. The bidirectional sequencing was performed according to Illumina high-throughput sequencing requirements, and the library was constructed by a two-step PCR ampli cation. The PCR conditions were 94 ℃ for 2 min; 94 ℃ for 30 s, 56 ℃ for 30 s, 72 ℃ for 30 s (the primary PCR ampli cation 30 cycles, the secondary PCR ampli cation 8 cycles), 72 ℃ for 5 min, and a nal extension at 10 ℃. PCR ampli cation products were recovered by 2% agarose gel electrophoresis. Recycling was performed using AxyPrepDNA gel recovery kit from Axygen.
The PCR ampli ed products of the zebra sh gut samples were sequenced for 16S rDNA using the Illumina-Misq high-throughput sequencing platform (TinyGene Bio-Tech (ShangHai) Co., Ltd, China) and the sequence length was 450 bp. The raw data obtained from sequencing were evaluated for quality and optimized. Trimmomatic was used for sequence ltration and FLASH was used for splicing. Ambiguous, homologous and some chimeras produced in PCR process were subsequently screened using Mothur V.1.39.5 to obtain optimized sequences for subsequent operational taxonomic unit (OTU) clustering and species information analysis.
USEARCH was used to cluster OTUs of the above treated sequences at 97% similarity. The representative OTU sequences were compared with the database Silva for species annotation (con dence threshold: 0.6). The relative abundance percentages of each sample were calculated at the levels of phylum, class, order, family, genus and species. Rarefaction curve re ects the sequencing depth of the samples. The rank-abundance curve explains species abundance and species evenness.
The Venn diagram can be used to count the common and unique OTU numbers of multiple samples, which can directly show the overlap and uniqueness in the OTU composition of different samples.
Alpha diversity analysis re ected the richness and diversity of communities in the samples. Mothur (http://www.mothur.org/wiki/Schloss_SOP#Alpha_diversity) was used to calculate the values of Shannon, Simpson, chao and ace indices, and R(3.4.1) language tool was used for graph plotting.
Jaccard, Bray-curtis, unweighted-Unifrac, and weighted-Unifrac were used to calculate the differences between samples and conduct non-metric multidimensional scaling graphs (NMDS). Matastats (http://metastats.cbcb.umd.edu/) was used for the comparison of the features with different abundances between groups on multiple taxonomic levels. The non-parametric factorial Kruskal-Wallis (KW) sum-rank test was applied to determine the signi cant difference between the richness of the groups. LEfSe (LDA effect size) uses linear discriminant analysis (LDA) to estimate the impact of each component (species) abundance on the difference effect of the groups.
In silico prediction of ADMET and drug-likeness properties of BTL-I To evaluate the pharmacokinetic pro le and toxicity of BTL-I, we employed the ADMETlab 2.0 (https://admetmesh.scbdd.com/pub/), which is a free online platform that facilitates researcher to predict the ADMET and drug-likeness properties of a compound [43,44].

BTL-I improved AlCl 3 -induced memory impairment
Overall, compared with the model group, the BTL-I treated groups showed signi cant treatment effect for zebra sh cognitive performance in the T-maze test (F (4, 25) = 40.60, P < 0.001 for the latency of rst entry to the EC zone on day 5 in Fig. 2a, F (4, 25) = 9.029, P < 0.001 for swimming speed in Fig. 2b, F (4, 25) = 31.65, P < 0.001 for the numbers of EC entries in Fig. 2c).
Subsequent post-hoc testing revealed that zebra sh in the model group had an increased latency of rst entry to the EC zone, reduced swimming speed and the number of EC entries (Figs. 2b and 2c). The swimming tracks also clearly showed the reduced preference of the model group sh to the EC zone (Fig. 3) following the AlCl 3 injections. In contrast, pretreatment with medium and high doses of BTL-I prevented these effects of AlCl 3 .
Furthermore, paralleling their cognitive de cits in the T-maze, zebra sh treated with AlCl 3 exhibited higher brain AChE activity, whereas both moderate and high doses of BTL-dose-dependently inhibited AChE activity (F (4, 10) = 4.474, P < 0.05 in Fig. 4e). This is consistent with the fact that excessive AChE activity is closely related to memory de cits [45].
Besides, treatment with AlCl 3 caused oxidant-antioxidant imbalance in the brain, GSH, the key nonenzymatic antioxidant in the body, has important physiological functions, such as scavenging free radicals, detoxifying, promoting iron absorption or maintaining membrane integrity [46][47][48][49][50]. As GSH is a low molecular weight scavenger of O 2 − , H 2 O 2 and so on, its content is an important indicator of the antioxidant capacity of the body [51]. Here, AlCl 3 treatments decreased GSH levels in the sh brain ((F (4, 10) = 18.90, P < 0.001 in Fig. 4f). Compared with the model group, the BTL-treatment groups did not show antioxidant activity since no higher GSH level was observed in them. However, they dosedependently increased GSH levels in zebra sh.

Results of intestinal ora diversity analysis
OTU is a hypothetical computational taxon (strain, species, genus, group, etc.) that has been set arti cially to facilitate the analysis of phylogenetic or population genetics. Because of the conservatism of 16S rDNA, the sequence obtained by sequencing can represent a species. To understand the composition of a species in a population sample, it is necessary to cluster the sequences. By clustering, the sequence is divided into many groups according to similarity, and one group is an OTU [52]. In this study, 15 samples were investigated, and the number of OTUs received by each sample was shown in Fig. 5a.
The Venn diagram is used to count the common and unique OTU numbers of multiple samples, which can intuitively show the similarity and overlap of the OTU number composition of environmental samples. Figure 5b shows the differences in OTUs between the ve groups. Different colors represent different groups, and the intersecting part is the OTU shared by adjacent groups. The OTUs in each group were as follows: Group B1 (model) 235; Group B2 (control) 213, Group B3 (25 mg/kg BTL-I + AlCl 3 ) 233; Group B4 (50 mg/kg BTL-I + AlCl 3 ) 229; and Group B5 (100 mg/kg BTL-I + AlCl 3 ) 216. The OTUs common between the model group and the other groups were as follows: 163 (Groups B1 and B2); 160 (Groups B1 and B3); 181 (Groups B1 and B4); 179 (Groups B1 and B5) (Fig. 5b).
In order to test the rationality of intestinal ora sequencing of samples, we constructed the rarefaction curves and rank abundance curves of intestinal ora according to the OTU numbers at different sequencing depths. The curve tends to be at from 10000 reads, indicating that the sequencing data volume is adequate, and more data volume will only produce a small number of new OTUs (Fig. 6a). Rank-abundance curve can be used to explain abundance and evenness of species. In the horizontal direction, the abundance of species is re ected by the width of the curve (i.e., the higher the abundance of species the larger the range of the curve). The shape (smoothness) of the curve re ects the evenness of species in the sample (i.e., the atter the curve, the more uniform the distribution of species). The results showed that the rank-abundance curve was smooth except for individual samples, indicating that the species distribution of each sample was evenness (Fig. 6b).
Alpha diversity can re ect the abundance and diversity of microbial communities, including Chao index, Ace index, Shannon index, Simpson index, etc. The Chao and Ace indices re ect the species richness, i.e., the number of species in the sample, without considering the abundance of each species. Shannon and Simpson indices re ect both species richness and species evenness in the community. The comparison between all the ve groups showed that there was no signi cant total difference in bacterial diversity. However, Shannon and Simpson indices displayed relatively larger difference between model group and control group compared with the difference between control group and administration groups (Fig. 6c ~  f).
The Beta diversity analysis was used to compare the differences in species diversity of the paired samples. The contents of each species in the samples were analyzed, and then the Beta diversity values among different samples were calculated.
NMDS method was a data analysis method that simpli es the research objects in multi-dimensional space to low-dimensional space for positioning, analysis and classi cation, while retaining the original relationship between objects. The degree of difference between samples was re ected by the distance between points. Four algorithms including Jaccard, Bray-Curtis, unweighted-Unifrac, and weighted-Unifrac were used for NMDS calculation. The NMDS based on Jaccard algorithm only considers whether the speci c OTU existed in the sample, not its abundance. The NMDS based on Bray-Curtis algorithm considers both OTU varieties and abundances in samples. UniFrac analysis uses evolutionary information of sample sequences to compare whether the samples have signi cant microbial community differences in a particular evolutionary lineage. The unweighted-Unifrac method only considers whether the speci c sequence appears in the community, not its abundance. The weighted-Unifrac method takes both existence and abundance into account. The results by Jaccard, unweighted-UniFrac, weighted-UniFrac methods showed that there was no signi cant difference in OTU varieties or evolutionary lineage between the experimental groups, the control group and the model group (Fig. 7a, 7c, & 7d). However, the results by Bray-Curtis method showed that there were signi cant differences in OTU abundance between the model group and the control group whereas no signi cant difference in OTU abundance between the experimental groups (except for 50 mg/kg group) and the control group (Fig. 7b).
Microbial diversity analysis showed that the intestinal ora of zebra sh included the following 12 major phyla: Proteobacteria, Firmicutes, Actinobacteria, Fusobacteria, Planctomycetes, Chlamydiae, Bacteroidetes, Chloro exi, Tenericutes, Verrucomicrobia, Deinococcus-Thermus and Saccharibacteria. Among these, Proteobacteria, Firmicutes and Actinobacteria were the dominant bacteria at the phylum level (Fig. 8a). The abundances of Firmicutes in the gut of the model group were reduced, whereas those of Fusobacteria, Planctomycetes, Chlamydiae and Chloro exi signi cantly increased, compared with those observed in the control group. In two experimental groups (administration of BTL-25 mg/kg and 100 mg/kg), the abundances of Firmicutes signi cantly increased, whereas those of Fusobacteria and Chlamydiae signi cantly decreased compared with the model group, and basically returned to the same level as the control group. However, in the BTL-treatment groups, there were almost no signi cant reversal effect on the increase of Planctomycetes and Chloro exi abundance (Fig. 8c).
On genus level, a total of 30 major known taxa of intestinal ora were identi ed (Fig. 8b). The rst 8 genera with inter-group abundance differences were respectively: Bacillus, Bosea, Cetobacterium, Alpinimonas, Singulisphaera, Phreatobacter, Mycobacterium, Candidatus-Microthrix (Fig. 8d). Compared with the control group, the abundance of Bacillus in the model group signi cantly decreased, while those of the other seven genera mostly increased signi cantly. In two BTL-treatment groups (25 mg/kg and 100 mg/kg), the abundance of Bacillus had been elevated to nearly normal level, while those of Bosea and the other genera had mostly decreased signi cantly.
LEfSe was able to compare the taxa composition of multiple groups on different taxonomical levels, identify the taxa with signi cant inter-group differences in abundance (i.e., biomarkers), and exhibit their lineage relationship. The results in Fig. 9 showed

Prediction of ADMET and drug-likeness properties
To obtain more information about the pharmacokinetic pro le of BTL-I and whether it has the potential to become a drug, we used ADMETlab 2.0 [43,44] to predict its ADMET and drug-likeness properties. The corresponding predicted results are presented in Table 1, and the physical properties of BTL-I are shown in Table S1 (see Supplementary data). The results demonstrated that the BTL-I possesses acceptable ADMET and drug-likeness properties in general. For example, results showed that BTL-I is active in both human intestinal absorption (HIA) and blood-brain barrier (BBB) penetration. It has acceptable safety pro les, generally performing well on most metrics (e.g., hERG blockers, Ames toxicity and carcinogenicity), and it is in harmony with Lipinski rule [53] and others (such as P zer rule [54] and golden triangle [55]), which indicates the drug-likeness properties of a compound. Unfortunately, BTL-I displayed some disadvantageous nature, such as a high risk of inhibiting CYP2C19, CYP2C9 and CYP3A4, and inducing liver injury.

Discussion
Aluminum has been examined for its broad neurotoxic effects and close relationship with AD, which promote tau hyperphosphorylation, aggregation, and neuro brillary tangle formation in AD brain (via activating tau kinases CDK5 and GSK3β), accumulate in microglia and induce proin ammatory cytokines, bind to Aβ and induce its aggregation, stimulate iron-induced membrane lipid peroxidation and oxidative damage, decrease the activity of antioxidant enzymes, interact with AChE on γ-peripheral site to enhance enzymatic activity resulting in reduced neurotransmission [12,58,59]. Furthermore, activated AChE can deteriorate Aβ aggregation, decrease BDNF expression [58], and further promote oxidative stress and neuroin ammation through a 'cholinergic anti-in ammatory pathway' (CAIP) via α7 nicotinic acetylcholine receptors [59,60]. Besides, in many reports on Aluminum induced AD or toxicity models, the alteration of host gut microbiota had been observed [61, 62].
The present study has established the zebra sh model of subchronic in ammation induced by acute i.p.
AlCl 3 administration, resulting in subchronic peripheral and central in ammatory responses and enhanced oxidative stress and AChE activity in the brain. On behavioral level, administration of AlCl 3 strongly impaired spatial and contextual memory of zebra sh in the T-maze test. On gut microbiota level, high-throughput sequencing results showed that the intestinal ora of zebra sh was dramatically disturbed by acute AlCl 3 administration. Collectively, these ndings are generally consistent with previous evidence that AlCl 3 induce memory de cits in both humans and animals, including zebra sh, and change intestinal ora [61-63]. In contrast, BTL-co-administration reversed these induced memory de cits and microbiota imbalance, indicating potential neuroprotective role of this drug.
In the present study, acute central and peripheral in ammation was characterized by release of proin ammatory cytokines IL-1β and TNF-α following AlCl 3 administration. Supplementation with BTLpotently inhibited acute central and peripheral in ammation in the AlCl 3 -treated zebra sh. Mounting evidence implicates BTL-in multi-targeted neuroprotective activity against oxidative stress, neuroin ammation and neuronal apoptosis, as well as in nerve growth without inducing cytotoxicity [22][23][24][25][26]. Here, we found that BTL-also prevents cognitive de cits (induced in zebra sh by AlCl 3 ) and exerts neuroprotective effects in this zebra sh model.
Previously studies have shown that hepatotoxicity and liver injury potentially induce in ammation [64, 65] and lead to increased levels of in ammatory markers such as IL-1β and TNF-α [66, 67] In this study, we noted that peripheral TNF-α levels were elevated after BTL-I administration, presumably due to BTL-I's potential hepatotoxicity and liver injury effects. This is suggested by ADMET analysis (Table 1). But It remains unclear why peripheral IL-1β and TNF-α behave differently after BTL-I administration, which may be attributable to the different mechanisms, and this needs further investigations.
The inhibition of brain AChE elevates ACh levels, and hence positively affects cognitive function in rats [59,68]. Subchronic exposure of zebra sh to AlCl 3 or i.p. injection of AlCl 3 in mice enhance brain AChE activity [37]. In line with this, our results show that i.p. injection of AlCl 3 also elevate AChE activity in zebra sh brain, whereas BTL-evokes neuroprotection and lowers AChE activity (Fig. 4e). Because the thin-layer chromatography bioautography shows that BTL-does not inhibit AChE catalytic activity (data not shown), this compound seems to indirectly decrease zebra sh AChE activity here, likely involving other molecular pathways.
Furthermore, oxidative stress involves the excessive production of ROS and reactive nitrogen species (RNS) [69], and may result in tissue damage. GSH is the most important nonenzymatic antioxidant, whose neuroprotective role in the brain is critical against oxidative damage caused by catecholamine oxidation or lipid peroxidation [70]. In the present study, GSH levels markedly decreased, 24 h after the AlCl 3 administration. With the pretreatment of BTL-I at the dose of 25 mg/kg and 50 mg/kg, the GSH levels of the zebra sh were even lower than the model group. But, when the dose rose to 100 mg/kg, the GSH content increased to the same level as model group though it was still lower than the control group.
BTL-I seemed to display doubtful antioxidant effect via GSH. The lack of signi cant antioxidant effects of BTL-I relative to the model group may be due to the limited sample size and the resulting low statistical power. Although it may also suggest that the antioxidant mechanism of action of BTL-I may exist elsewhere. Further study with larger sample sizes and better designs are warranted to testify and explain this complicated phenomenon for a solid conclusion.
Intestinal ora plays a crucial role in the stability and balance of intestinal microecological environment, and the composition of human intestinal microbial community remains basically stable after the age of 3 years [71]. In recent years, a growing number of studies had shown that speci c intestinal ora play important role in neuroprotection and their disorders were closely related to neurodegenerative diseases including AD [15,16,72]. Some reports indicated that, in the gut of healthy human or animals, there are higher population of Gram-positive (G + ) bacteria including Firmicutes and Actinobacteria and lower Gram-negative (G − ) bacteria like Bacteroidetes on phylum level [15,16]. At family or genus level, Some G + taxa like Bacillus, Eubacterium, Clostridiaceace in Firmicutes and Bi dobacterium in Actinobacteria show higher abundance in healthy individuals and bene t their hosts through different mechanisms including reducing leakage of gut by the protection of bio lm, inhibiting in ammation, anti-oxidation, reducing Aβ deposition and transferation from gut to brain, etc [15,16,73]. On the contrary, some G − taxa like Bacteriodes, Blautia, Escherichia coli, Shigella, Chlamydia, Fusobacterium, etc., are closely and positively correlated with AD mainly involving the activation of systematic in ammation by their enriched LPS in cell wall, the invasion of proin ammatory cytokines, LPS, and even bacteria into blood circulation system and brain, inducing Aβ deposition and tau phosphorylation [15,16,72,73].
In the present study, the control group zebra shes host higher G + bacteria (Firmicutes on phylum level and predominantly Bacillus on genus level) than AlCl 3 injured model group with memory impairment, while the model group zebra shes have much less G + bacteria than the control group but signi cantly more G − bacteria including Cetobacterium (in family of Fusobacteriaceae) and Chlamydiales (on order level). This is highly consistent with the previous studies especially the report on the bene ts of Bacillus subtilis in delaying neurodegeneration and behavior impairment in the AD model Caenorhabditis elegans and the report on the negative effect of G − bacteria including Fusobacterium and Chlamydia [72][73][74].
Generally, the pretreatment of BTL-I maintained the dominance of G + bacteria VS G − ones in the gut of zebra shes when facing the impact of Aluminum. However, the dose levels exerted different in uence. In low and high dose groups, the abundances of Bacillus have kept at high levels close to that of control group and the abundances of G-bacteria were much lower. Additionally, in low dose group, another G + bacteria Micromonospora in phylum of Actinobacteria was recognized as a biomarker, hinting its possible positive role, when compared with a report [16] about Actinobacteria. Likewise, in high dose group, G + bacteria Lutispora in family of Gracilibacteriaceae and order of Clostridiales was also biomarker. Considering the report on the strongly negative correlation of Clostridiaceae [16] with AD biomarkers in cerebrospinal uid, we speculate that Lutispora may also have some bene t in neuroprotection.
It's intriguing that the middle dose group did not possess high abundance of Bacillus. This discrepancy may be attributable to a non-linear relationship of BTL-I and Bacillus, more concentration gradients of BTL-I will be set in the further study to explore this relationship. However, G + bacteria Clostridiaceae was found to be the key biomarker of this group; besides, G − bactereia Chloro exi bacteria was its another key biomarker but bacteria in this phylum have no LPS in cell walls [75]. These may help to explain the behavioral improvement of this group.
Our study suggests that administration of marine fungal metabolite BTL-I prior to AlCl 3 injection may be able to maintain the predominance of bene cial G + bacteria in the gut of zebra sh to resist the acute injury of Aluminum, the related in ammation and AD pathology. The detailed mechanisms of intestinal ora regulation and the treatment effect on AlCl 3 induced chronic AD model need to be further investigated for BTL-I in future.

Conclusions
This study showed that BTL-I dose-dependently ameliorated AlCl 3 -induced cognitive de cits in zebra sh, reversed the elevation of AlCl 3 -induced central and peripheral proin ammatory cytokine levels and the increase of brain AChE activity, contributed to maintain the predominance of bene cial Gram-positive bacteria in the intestinal ora of zebra sh, which was challenged by AlCl 3 . The in-silico analysis indicated that BTL-I exhibits acceptable drug-likeness and ADMET pro les. In summary, BTL-I has the potential as therapeutic agent for preventing CNS de cits caused by in ammation, neurotoxicity, and gut ora imbalance.
TNF-α Tumor necrosis factor-α Declarations Consent for publication Not applicable.

Availability of data and materials
All data generated or analyzed during this study are included in this published article and its supplementary information le. The datasets used and/or analyzed during the current study are available from the primary author on reasonable request.  The results of biochemical indexes of zebra sh (n=3). a GSH content in zebra sh brain. b AChE activity in zebra sh brain. c ~ f IL-1β and TNF-α content in zebra sh brain and peripheral tissue. *P < 0.05, **P < 0.01, ***P < 0.005, ****P < 0.001, vs the control group; #P < 0.05, ##P < 0.01, ###P < 0.005, ####P < 0.001, vs the model group.  Alpha-diversity. a The rarefaction curves. b The rank-abundance curves. c ~ f Alpha indices. Beta diversity. a Jaccard algorithm (stress value = 0.137). Only the presence or absence of OTU in the sample was considered, not the abundance. b Bray-curtis algorithm (stress value = 0.096). Both the presence or absence of OTU in the sample and the abundance were considered. c unweighted-UniFrac algorithm (stress value = 0.064). It only considers whether the sequence was present in the community, not the abundance of the sequence. d weighted-UniFrac algorithm (stress value = 0.069). It accounts for the abundance of sequences on the basis of unweighted-Unifrac and was able to differentiate species abundance.