The experiment utilized soybean meal, cottonseed meal, and rapeseed meal as primary protein sources, soybean oil and fish oil as primary lipid sources. The composition and formulation of the experimental diets are detailed in Table 1 . The feed fishmeal-free (FL) is used as the basal diet. Experimental diets are prepared by adding 3% (medium level) fishmeal (referred to as FM) and 6% (high level) fishmeal (referred to as FH) to the basal diet. All diets ingredients were ground and filtered through a 40-mesh screen, and then pelletized using a twin screw extruder (TSE65, Modern Yanggong Machinery Technology Development Co., Ltd, Beijing, China) (2 mm × 2 mm). The pellets are subsequently dried in a dryer at 60°C and stored in sealed bags at room temperature until feeding.

The experiment was carried out in a fish farm in Xidongting district (Changde, Hunan, China). Grass carps were obtained from a commercial farm (Changde, Hunan, China). The grass carp are acclimated in cages for two weeks. The grass carps were subjected to a 24-hour fasting period before the formal experiment. A total of 270 healthy grass carps (average weight 70.01 ± 5.34 g) were selected and randomly divided into nine floating cages (2.0 × 1.5 × 1.5 m). The experiment was divided into three treatment groups, each with three replicates (n = 30). After the experiment begins, the fish are fed three times a day (at 7:00, 12:00, and 17:00). During the 60-day experimental period, the mortality and feeding conditions of the grass carp are recorded daily. Culture unit: 20-hectare earthen pond, average depth 2 m. Water volume: ~400,000 m³. Water exchange: 10% of pond volume per day via continuous inflow/outflow. Throughout the experiment, water quality parameters are monitored: water temperature ranges from 24.5 to 30.3°C, pH values range from 7.3 to 7.8, and dissolved oxygen levels range from 6.5 to 7.5 mg/L.

At the end of the experiment, grass carp were fasted for 8 h. Then, three fish were randomly selected from each cage. Their body weight was measured, and samples of the posterior intestine contents were collected. These samples are rapidly frozen in liquid nitrogen and then stored in a -80°C freezer. The samples from three fish in the same net cage were pooled into one sample for further analysis of the intestinal microbial community. The remained grass carp are then fasted for another 16h, and the total number and body weight in each cage were recorded. Additionally, three fish from each cage were randomly selected and anesthetized, and then the body length, body weight, visceral, and liver weight were measured. The section of posterior intestine (1 cm×1 cm) and the left dorsal muscle samples (1 cm×1 cm×1 cm) were collected for histomorphology analysis. The right dorsal muscle samples were collected. The samples from the three fish in the same cage are mixed into one sample, rapidly frozen in liquid nitrogen, and stored at -80°C, and stored in a -80°C freezer for the subsequent measurement of VOCs.

The growth performance of the grass carp was determined as follows:

Intestinal and muscle sections are prepared according to the method of T. Tang (Tang et al., 2021). Sections are photographed using a fluorescence microscopy (Olympus, Tokyo, Japan). The complete intestinal villus height and villus width are measured by ImageJ software. The number of intact myofibers in various regions of the transected muscle is measured, and the fiber density was calculated as the number of fibers per square millimeter (fibers/mm (Zhang et al., 2022)), and the fiber diameter (≥ 10 µm) is calculated based on the area = π × r².

Total DNA is extracted from intestinal contents using the sodium dodecyl sulfate method according to the manufacturer’s instructions. Next-generation HTS is employed to analyze the composition of the gut microbiota. The V4 hypervariable region of the bacterial 16S rRNA gene is amplified using specific primers with barcodes, 515F (5’-GTGCCAGCMGCCGCGGTAA-3’) and 806R (5’-GGACTACHVGGGTWTCTAAT-3’), followed by sequencing on the NovaSeq 6000 platform. Samples are annotated using the SILVA database and the Mothur method. After dereplication of valid tags, singletons are removed based on abundance, and then clustering at 97% similarity. Weighted UniFrac distance is used to calculate the sample distance matrix to investigate similarities among different samples. The UPGMA clustering tree is generated using the QIIME software (version 1.9.1). A species abundance clustering heatmap (top 30) is created based on species annotation and abundance information at the genus level using R software (version 3.1.0). Non-Metric Multi-Dimensional Scaling (NMDS) analysis is conducted using the vegan package in R software. LEfSe (LDA Effect Size) analysis is performed using LEfSe software (version 1.0) to identify species with significant differences among groups (Wen et al., 2021). Sequencing and routine data analysis were conducted by Novogen Co., Ltd. (Novogen, Beijing, China).

VOCs in muscle samples are analyzed using an integrated system composed of an Agilent 490 Gas Chromatograph (Agilent Technologies, Palo Alto, CA, USA) and IMS instrument (FlavorSpec^®, Gesellschaft für Analytische Sensorsysteme mbH, Dortmund, Germany). The detection method follows the protocol outlined by Zhang Q et al (Granato et al., 2018; Zhang et al., 2020). N-Ketones C4-C9 (Sigma-Aldrich, St. Louis, MO, USA) are used as external references to calculate the retention indices (RI) of aromatic compounds. Aromatic compounds in muscle samples are identified by comparing their RI values with standards in the GC-IMS library. Quantification of each VOC in muscle samples is performed using established calibration curves, with results expressed as the relative content of each VOC relative to the total aroma content (n=3) (Song et al., 2021). For downstream analysis, individual VOCs were grouped into chemical classes (aldehydes, alcohols, ketones, etc.). The abundance of each class was calculated as the sum of the peak areas of all compounds belonging to that class.

Statistical analyses were performed using the Statistical Package for the Social Sciences (SPSS) 23 software (IBM, Armonk, USA). Data were presented as mean ± standard deviation (SD), the analysis was conducted using one-way ANOVA, with a significance threshold level set at P< 0.05. Laboratory Analytical Viewer (LAV, G.A.S., Dortmund, Germany) software and GC-IMS Library Search were used to analyzed the aroma profiles “Fingerprint” gallery plots and principal component analysis (PCA) of muscle samples are analyzed using laboratory analytical viewer (LAV, G.A.S., Dortmund, Germany) software and GC-IMS Library Search. In R software, Spearman correlation is used to calculate the correlation between the top 30 bacterial genera in gut microbiota abundance and VOCs with statistical differences.

The growth performance results of grass carp are shown in Table 2 . Compared with the FL group, the FBW, WGR, and SGR of the FM and FH groups are significantly increased (P<0.05). This is similar to the results obtained by Peng, K.-S. et al. regarding the impact of fish meal substitution for soybean meal on the growth performance of grass carp (Peng et al., 2020). Fishmeal is a widely used feed ingredient which contains a higher percentage of protein, energy, minerals, and vitamins (Cho and Kim, 2011). Fishmeal applied with other animal and vegetable proteins in the diet had synergistic effects which could balance the amino acid composition, promote fast growth of fish, and reduce feeding costs (Cho and Kim, 2011). Therefore, completely or partially replacing plant protein with fishmeal can positively impact the growth performance of grass carp.

This experiment adopts a stepwise feed formulation: fishmeal increases from 0% to 6% while rapeseed meal decreases from 35% to 25%. Rapeseed meal contains several known antinutritional factors—such as glucosinolates, tannins, and phytates—that suppress fish growth and disturb the gut microbiota. Hence, part of the observed improvements in growth performance and gut microbiota cannot be ruled out as arising from the reduction of rapeseed meal rather than from the increase in fishmeal, and the overall feed effect may concurrently benefit from the lower rapeseed meal level.

Ethoxyquin was included at 0.05% of the diet as a necessary antioxidant to ensure feed stability during storage. Ethoxyquin was added only as an antioxidant. However, published data indicate that even this level can transiently suppress respiratory burst and lysozyme activity in tilapia (Yamashita et al., 2009) and accumulate as dimer residues in salmonid tissues (Ørnsrud et al., 2011). Such mild immunomodulation could have dampened diet-related inflammatory responses, although growth performance remained unaffected (Wang et al., 2015). At current ingredient prices, the 6% formulation increases feed cost by approximately 8%–10% per tonne compared with the 3% formulation. Therefore, based on the effect–cost trade-off, 3% fishmeal represents the economically optimal recommendation; a 6% fishmeal formulation is considered only if the market offers a premium for “flavour” that exceeds 10%. Subsequent studies quantify the actual revenue gain from this flavour improvement through consumer sensory tests and economic modelling to confirm the strategy’s feasibility.

The morphological results of grass carp intestines and muscles are shown in Figures 1A and 1B . The results in Figure 1C indicate that, compared with the FL group, the intestinal villus height of the FM and FH groups is significantly increased (p< 0.05). The intestine is the primary site for the digestion and absorption of nutrients, and the height of intestinal villi is positively correlated with nutrient absorption (Zhu et al., 2012). Therefore, the addition of 3% and 6% fishmeal may enhance the intestinal digestive and absorptive capacity of grass carp, thereby promoting their growth, by increasing the height of intestinal villi to expand the intestinal absorptive surface area. The results in Figures 1E and 1F show that, compared to the FL and FM group, the fiber diameter of the FH group significantly decreased while the fiber density significantly increased (p< 0.05). Changes in muscle fiber size are achieved through the balance between protein synthesis and degradation (Yu et al., 2017). Nutritional status is mirrored in the muscle fibers’ characteristics, including their size, shape, and classification (Rodriguez-Torres et al., 2020). An increase in fiber density and a decrease in fiber diameter can improve meat tenderness (Kopač and Ručigaj, 2024). Therefore, a fishmeal level of 6% may elevate the quality of grass carp by improving the morphology of muscle fibers.

The sequencing result of the intestinal microbiome is shown in Figure 2 . The results of the UPGMA phylogenetic tree analysis indicate that Firmicutes, Actinobacteria, Bacteroidota, and Fusobactariota were the predominant bacteria at the phylum and species level in all groups ( Figures 2A, B ). This is consistent with the findings of Wu, S. and Zhang, X. et al (Wu et al., 2012; Zhang et al., 2014). Furthermore, the composition of bacterial phyla in the FM and FH groups was closer to each other than to the FL group. Intestinal Firmicutes possess genes responsible for fermenting dietary fiber and interact with the intestinal mucosa to help maintain host homeostasis (Sun et al., 2023). Actinobacteria, especially Bifidobacteria, are crucial for gut homeostasis and widely used as probiotic with beneficial effects in various pathological conditions (Binda et al., 2018). Bacteroidetes play an important role in gut health by producing butyrate, which is the end product of colonic fermentation and is considered to have antineoplastic properties (Kim and Milner, 2007). Fusobacteria activate the host’s inflammatory response to prevent pathogens from promoting tumor growth (Kelly et al., 2018). Dysbiosis, characterized by an imbalance in the taxonomic composition, often manifests as an increase in Firmicutes relative to Bacteroidetes in metabolic disorders (Shin et al., 2015). Studies have shown that an increase in Firmicutes and Actinobacteria, and a decrease in Bacteroidetes and Fusobacteria, are similar to the microbial patterns of diseased fish (Tran et al., 2018). Compared with the FM and FH groups, we observed this disease-like microbial pattern in the FL group. The gut microbiota may indirectly affect intestinal inflammatory pathways and the synthesis and catabolism of nutrients (Przewłócka et al., 2020). There are complex interactions among different groups within the gut microbial community, including competition and symbiosis (Przewłócka et al., 2020). Changes in fishmeal levels may disrupt the balance of these interactions, leading to changes in the relative abundance of certain groups. In the absence of fish meal, certain microbial populations may become more prominent. For example, Photobacterium was highly represented in gilthead seabream (Sparus aurata) fed with a fishmeal-free diet (Kelly et al., 2018). Some studies have also found that when fishmeal is used to replace cottonseed protein, the proportion of the relative abundance of the phylum Proteobacteria, which encompasses both pathogenic and beneficial symbionts in the gut of juvenile golden pompano (Trachinotus ovatus) gradually decreased with the increase in fish meal levels (Thomas et al., 2011). Proteobacteria is a metabolically diverse phylum that includes not only opportunistic pathogens but also beneficial commensals such as Vibrio alginolyticus, which can act as a probiotic and reduce mortalities caused by Aeromonas and Vibrio spp (Spring et al., 1995). Similarly, with the increasement of fishmeal levels, probiotic bacteria (Lactococcus) increase while pathogenic bacteria (Enterococcus) decrease in the common carp (Cyprinus carpio) intestine (Shin et al., 2015). Therefore, increasing fishmeal levels can balance the gut microbiota and maintain intestinal health.

The clustering heatmap based on the top 30 bacterial genera with relative abundance is shown in Figure 2C . ANOSIM analysis of dietary fishmeal on intestinal microflora of grass carp: FL vs. FM and FM vs. FH were showed in Figure 2D . The FL group is dominated by Staphylococcus, Paeniclostridium, Lactobacillus, Kocuria, and Corynebacterium. In contrast, Butyricicoccus, Acinetobacter, Rothia, Dermacoccus, Ralstonia, Pseudomonas, Alteromonas, Brevinema, Geobacter, and Pseudoalteromonas are enriched in the FM group. The dominant genera in the FH group include Dielma, Romboutsia, Clostridium sensu stricto 1, Erysipelatoclostridium, Aeromonas, Shewanella, Anaerostignum, and Vibrio. Recent studies have identified various pathogenic microorganisms (Staphylococcus (Surana and Kasper, 2014), Paeniclostridium ( Li et al., 2022), Acinetobacter ( Joly-Guillou, 2005), Corynebacterium ( Bernard and Funke, 2015), and Brevinema ( Paster, 2015)) and beneficial microorganisms (such as Dermacoccus ( Rangseekaew et al., 2022)) in the intestines of different species. In our study, the three pathogenic microorganisms Staphylococcus, Paeniclostridium, and Corynebacterium accumulate in the FL group. Additionally, the butyrate-producing genera Butyricicoccus and Rothia are enriched in the FM group. Pseudomonas (lipase) (Gong et al., 2021), Alteromonas ( Ivanova et al., 2005), and Pseudoalteromonas (amylase and chitinase) (Parrilli et al., 2021) known for their vigorous metabolic capabilities, are more abundant in the FM group. Clostridium_sensu_stricto_1 ( Gong et al., 2021) and Romboutsia ( Gerritsen et al., 2019) are important for amino acid utilization in animal protein diets, displaying a range of metabolic capabilities, including carbohydrate utilization, fermentation of single amino acids, anaerobic respiration, and production of metabolic end products. Clostridium can induce regulatory T cells in the colonic lamina propria of mice, thereby providing protective effects in mouse models of colitis and allergies (Li et al., 2022). Clostridium and Romboutsia can enhance the digestion of cellulose (Wu et al., 2012), and they are enriched in the FH group.

Through non-metric multidimensional scaling (NMDS) analysis, as shown in Figure 2E , we visualized the structure of microbial communities fed with different levels of fish meal. The results indicate that the three treatment groups are well separated in the NMDS plot, with a Stress value of 0.084. This demonstrates that different levels of fish meal significantly affect the composition of microbial communities and that our two-dimensional ordination plot effectively represents the original distances between samples, ensuring high reliability of the results (Zha et al., 2018).

In the LefSe analysis ( Figure 2F ), different levels of fish meal exert significant effects on microbial community structure. The LDA value, used to assess the impact of microbial groups that show significant differences between groups, indicates that an LDA value greater than 3 is generally considered to represent biomarkers with high credibility (Ma et al., 2020). In the FL group, Actinobacteria, Clostridium_sensu_stricto_11,Sva0485,Clostridium_sp_HY_34_8,Tetrasphaera, and Intrasporangiaceae; in the FM group, Petrimonas, AEGEAN_169_marine_group, Dysgonomonadaceae, and Breznakia; and in the FH group, Bacteroides paurosaccharolyticus, Erysipelatostridiaceae, Enterobacterales,Subdoligranulum, and Coprococcus all exhibit significant responses, with LDA values exceeding the threshold. This highlights the clear response of these microbial groups to different levels of fish meal feeding.

PICRUSt2-level-2 functional profiling revealed that the 3% fish-meal group was significantly enriched in pathways related to metabolism, energy metabolism, lipid metabolism, and cell growth and death ( Figure 2G ). These enriched functions provide mechanistic insight into the improved growth performance and enhanced muscle flavour observed with fish-meal supplementation, thereby furnishing more direct evidence for the hypothesis that gut-microbiota modulation mediates flavour development.

In Figure 3A , the PCA model is selected as the separation model. Samples with similar aroma characteristics overlap or are closely positioned in their respective figures (Song et al., 2020). The cumulative variance contribution rate of PC1 (57.9%) and PC2 (25.28%) is 83.18%, indicating that these two PCs essentially cover the sample information and can characterize the odor composition of grass carp muscle at different fish meal levels (Pan et al., 2022). A distinct separation is observed between the samples of the FH group and the other samples along PC1 (57.9%), indicating that the odor changes in muscle samples with a high level of fish meal are more pronounced.

The muscle VOCs of FL, FM, and FH are analyzed using GC-IMS. VOCs are significant determinants of meat flavor (Liu et al., 2009), and they serve as an important basis for consumers to evaluate meat quality (Chen et al., 2022). Intensity variations in different muscles are represented as a 3D topographical model in Figure 3B . The 2D topographical visualization of the VOCs is depicted in Figure 3C . Most signals are observed to be located in the retention time range of 100–400 s and a drift time of 1–1.5 s. The results in Table 3 show that 53 VOCs are detected in the samples, including 10 aldehydes, 12 alcohols, 9 ketones, 2 nitrogen compounds, 3 sulfides, and 3 esters. In addition, there are nine unknown compounds. Aldehydes, alcohols, ketones, acids, and hydrocarbons are the primary VOCs in fish and fish products (Liu et al., 2009). When compared to the FM group as the background, the FL and FH groups display varying degrees of blue and red spots. The FL group has more blue spots, indicating that more VOCs and their content are lower than in the FM group; the FH group has many red spots, indicating that more VOCs and their content are higher than in the FM group ( Figure 3D ; Table 3 ).

The comparison of VOC fingerprints among the three groups is shown in Figure 3E . The VOCs with significant differences are roughly categorized into aldehydes, alcohols, ketones, esters, sulfur compounds, and furans, and a radar chart is created as shown in Figure 3F . Compared with the FL and FM groups, a large number of aldehydes are significantly increased in the FH group, including 2-methylbutanal D, hexanal D, T, 4-heptenal (Z), heptanal, 2-heptenal (E) D, benzaldehyde D, and n-nonanal M (P< 0.05); the content of butanal and 2-methylbutanal M is highest in the FM group (P< 0.05). Aldehydes are highly volatile and have a robust fatty aroma (Lorenzo et al., 2013). They are crucial to the flavor of fish due to their diversity and low odor thresholds (Pan et al., 2022). It has been reported that butanal, hexanal, heptanal, and benzaldehyde are the key aroma compounds in Hypophthalmichthys molitrix, Aristichthys nobilis, Lateolabrax japonicus, and Parabramis pekinensis ( Chen et al., 2022). Most linear aldehydes are produced through autoxidation reactions initiated by enzymes or microorganisms present in fish (Fu et al., 2015; Zhang et al., 2020). Genomic and transcriptomic data have confirmed that the cecal microbial community affects the VOCs of duck meat, especially pentanal, hexanal, and heptanal, by modulating metabolic pathways related to fatty acid synthesis and degradation (Xu et al., 2024). Hexanal has been recognized as a unique aroma-active component in fresh tuna (Pan et al., 2022) and raw red seabream (Chen et al., 2022), and it exerts a major influence on flavor perception in silver carp samples (Liu et al., 2009). Previous research has indicated that (Z)-4-heptenal is present in extracts from turbot, characterized by a fishy or potato-like odor (Sérot et al., 2001). These aldehydes, which significantly contribute to the aroma of fish meat, impart a pleasant nutty, grassy, and fruity fragrance to the fish.

Compared to the FL and FM groups, the content of many alcohols is significantly increased in the FH group, including 1-penten-3-ol, 4-methyl-2-pentanol, 1-pentanol, 2-hexen-1-ol, 5-methyl-2-furanmethanol, octen-3-ol, and 2-ethyl-1-hexanol (P< 0.05). Alcohols are primarily derived from the oxidative decomposition of lipids or the reduction synthesis of carbonyl groups. They have higher thresholds compared to aldehydes and are closely related to the characteristic fatty flavor of meat (Zhang et al., 2020; Pan et al., 2022). Alcohols generally impart fruity, floral, and grassy notes to fish (Liu et al., 2009). Unsaturated alcohols typically have much lower thresholds than saturated alcohols and can significantly influence the flavor of food (Lorenzo et al., 2013). In our study, the content of ethanol is highest in the FL group, and the content of isopentyl alcohol M is highest in the FM group (P< 0.05). The elevated content of ethanol in the FL group suggests that it may not be conducive to preservation and can lead to the generation of unpleasant odors (Pan et al., 2022). High concentrations of 1-hexanol and 1-pentanol have been detected in mature cod roe, and they may play a critical role in the product’s aroma (Purrinos et al., 2011). The compound 1-octen-3-ol, originating from the hydroperoxide degradation product of linoleic acid, has been identified as one of the most important aroma-active compounds in fish samples (Salum et al., 2017; Pan et al., 2022). This may be one of the reasons why fish flavor compounds are affected.

Compared with the FL group, the content of many ketone compounds is significantly increased in the FH group, including 2-butanone M, 2,3-pentanedione, methyl isobutyl ketone M, and 2-heptanone D (P< 0.05); compared with the FH group, the content of 2-propanone and 2-pentanone M is significantly increased in the FL and FM groups (P< 0.05). Ketones may originate from the thermal oxidation of unsaturated fatty acids, the degradation of amino acids, or microbial oxidation (Zhang et al., 2019). Ketones have a unique buttery and fruity flavor (Pan et al., 2022). 2-Heptanone, primarily produced by the oxidation of linoleic acid, has been identified as a compound that can modulate the flavor of meat and meat products (Han et al., 2020). Studies have shown that 2-heptanone helps enhance the aromatic characteristics of silver carp (Liu et al., 2009). In addition, among the esters, the content of ethyl acetate M is highest in the FM group, and the content of thiolane is highest in the FH group (P< 0.05). Esters are well-known for their significant contribution to fruity flavors (Peng et al., 2020) and play a crucial role in the unique aroma of fish (Liu et al., 2009). Ethyl acetate, with its slight fruity aroma, is considered a VOC that can enhance the aroma of fish (Mansur et al., 2003). The content of the sulfur compound dimethyl trisulfide is highest in the FL group (P< 0.05). Sulfur compounds, with their low thresholds, play an important role in the overall flavor of food, and thiols are generated from the Strecker degradation of methionine. Furans are one of the important compounds that produce meaty flavors (Pan et al., 2024). The content of the furan compound 2-pentylfuran is highest in the FH group (P< 0.05), and it makes a positive contribution to the flavor of grass carp. The diet-induced shifts in key volatile compounds are known to enhance the fresh, sweet, and umami notes while reducing ‘fishy’ odours. Consequently, fillets from fish fed the higher-fish-meal diets are expected to command better market prices and broader consumer acceptance, providing a clear economic incentive for producers.

Studies have proven that microbial communities are associated with muscle metabolic characteristics (Feng et al., 2022). Changes in the gut microbiota can regulate muscle fiber diameter (Lei et al., 2022). Short-chain fatty acids are produced during the fermentation process of gut microbiota, which promote the formation of oxidative muscle fibers (Sérot et al., 2001). Studies have shown that fecal microbiota transplantation is beneficial for the muscle growth and development of germ-free piglets (Qi et al., 2021). The microbial communities in the animal gut are diverse and play an important role in meat quality (Chen et al., 2022). VOCs with significant differences are categorized into aldehydes, alcohols, ketones, esters, sulfur compounds, and furans. Spearman correlation analysis is performed with the top 30 bacterial genera of the gut microbiota. The results are shown in Figure 4 . Ketones and esters show no significant correlation with various bacterial genera (P > 0.05); sulfur compounds are extremely significantly correlated with Dielma (P< 0.01); furans are significantly positively correlated with Aeromonas (P< 0.05); alcohols are significantly positively correlated with Clostridium-sense-stricto-1 and Geobacter (P< 0.05). Some gut microbes can catalyze the production of acids, thereby affecting the formation of flavor compounds in fish. For example, supplementing ducks with lactic acid bacteria has been shown to significantly increase the content of many VOCs in duck meat, including 1-octen-3-ol (Qi et al., 2021). Clostridium-sense-stricto-1 can produce butyric acid through the butyrate kinase pathway (Chai et al., 2019). Aldehydes show significant positive correlations with multiple bacterial genera, including a significant positive correlation with Bacteroides (P< 0.05) and extremely significant positive correlations with Erysipelatoclostridium, Aeromonas, Shewanella, Cetobacterium, and Vibrio (P< 0.01). As a halophilic bacterium, Vibrio can produce a variety of flavor compounds, thereby providing different flavors to the final product (Liu et al., 2021). Therefore, we speculate that the gut microbiota of grass carp shows a significant correlation with muscle VOCs.