Two breeding stock southern catfish (a male and a female) that feed on their cohousing small fish (several carps) were collected from a pond and then used for offspring reproduction. Fertilized eggs and a feed used in this study were provided by a local commercial southern catfish farm. The experimental protocols were approved by the Animal Ethics Committee of the Huazhong Agricultural University, China under permit number HZAUMO-2016-026.

The fertilized eggs were transported to an incubator in College of Fisheries, Huazhong Agricultural University (Wuhan, China), and hatched at a density of about 18 eggs/L aerated tap water in a flowing water system during a natural light period. After ~2 dpf, sac-fry larvae were transported to two independent flow-through tanks (~1000 fish/tank) for subsequent culture experiment in which fish were fed with minced water earthworm to apparent satiation at 4~5 dpf and then fed with untreated water earthworm from 7 dpf to the end of the experiment (four times daily at 07:00, 13:00, 19:00, and 24:00 before 35 dpf and three times daily at 07:00, 16:00, and 22:00 after 35 dpf). In addition to automatic flowing water exchange, the tanks were regularly cleaned and some fish were removed from the tanks to avoid crowding stress during the experiment. About 120 fish were stocked in each tank at 8 dpf, and 70 at 15 dpf, 33 at 35 dpf, 15 at 65 dpf. Water physicochemical parameters were measured one time a week by using Thermo scientific, ORION STRA A221 pH meter and UNIC 2800UV/VIS spectrophotometer. The dissolved oxygen and pH ranged 6.36–7.21 mg/l and 7.36–7.93, respectively. Ammonia nitrogen (NH₃-N) and nitrite (NO₂-N) never exceeded 0.11 mg/l and 0.3 mg/l, respectively. These parameters were well within acceptable levels for fish growth.

The sampling procedures are shown in Figure S1. Fish were euthanized with MS-222 (100~200 mg/L). Body weight and standard length of the fish were measured during the experimental periods, as shown in Table S1. In brief, at 8 dpf, six gut samples were collected, and nine gut samples at subsequent each time point (at 18, 35, 65, and 125 dpf). Unable to distinguish various digestive regions, the entire gastrointestinal tract was collected at 8 and 18 dpf, and posterior gut was sampled at 35, 65, and 125 dpf. The collected samples were placed into a sterile tube and stored at −80°C until further sample processing.

Environmental microbiota samples were obtained at 8, 35, and 125 dpf. At each time point, one water sample and one food sample were collected from each tank. Approximate 250 ml of water in each tank was sampled and filtered through a 0.22 μm pore size filter. Filters cut into small pieces were put into sterile tubes and stored at −80°C for DNA extraction. Similarly, the prepared food for each tank was collected and stored in sterile tubes at −80°C after homogenization until further analysis.

The bacterial DNA of gut and food samples were extracted by using QIAamp DNA Stool Mini Kit (Qiagen, Germany) according to the manufacturer's protocol. The DNA of water samples was isolated by using Mo PowerWater kits (MoBio, USA). An identical DNA concentration for all samples was prepared for amplification of the 16S rRNA gene spanning the V4-V5 region (515F, 5'-GTGCCAGCMGCCGCGGTAA-3' and 907R, 5'-CCGTCAATTCCTTTGAGTTT-3'). We added an adaptor and sample-unique DNA barcode of 12 base sequences at the 5' end of the PCR primers for sample identification. Amplifications were carried out in triplicate PCR mixtures (50 μl): 0.4 μM forward and reverse primers, 100 ng DNA template, 2.5U of GoTaq Flexi Polymerase (Promega, USA), 200 μM deoxynucleoside triphosphate (dNTP) and 2 mM MgCl₂. Thermal cycling conditions for PCR were: 94°C for 5 min to denature the DNA, with a total of 25 cycles of 30 s at 94°C, 30 s at 55°C and 60 s at 72°C, and with a final extension of 5 min at 72°C. PCR products were purified by using Qiagen Gel Extraction Kit according to the manufacturer's instructions (Qiagen, Germany). The purified DNA was quantified using PicoGreen reagent and equal concentrations were pooled into one final pool. The pooled amplicons were sequenced with an Illumina HiSeq 2500 instrument (HiSeq Reagent Kit V.2, 500 cycles). Raw data are available on the NCBI BioProject under accession number PRJNA396633.

All raw sequence data analyses were done in Quantitative Insights Into Microbial Ecology (QIIME; version 1.8.0). Sequences were filtered to achieve high quality and assigned to respective samples. The processed sequences were clustered into Operational Taxonomic Units (OTUs) defined according to a value of similarity cutoff = 97% using the UCLUST algorithm. We then removed singleton OTUs (only one sequence read from the combined dataset) and classified taxonomically representative sequences into each OTU using Ribosomal Database Project (RDP) classifier with the Greengenes Database. In order to avoid the effects of sequencing depth on alpha and beta diversity, we performed rarefaction curve analysis and rarefied all samples to the lowest sequence reads for downstream analysis.

Alpha and beta diversity in the data set were assessed in QIIME. For alpha diversity analysis, we calculated observed OTUs number, Shannon indices, Simpson indices, phylogenetic diversity whole tree (PD). For beta-diversity analysis, we used Bray-Curtis metric to evaluate bacterial community variations among all the samples and visualized in non-parameters multidimensional scaling analysis (NMDS) plots. In addition, we compared bacterial communities among gut, food and water samples collected at the same time points based on UniFrac distance metrics and plotted these samples using principal coordinate analysis (PCoA). Further, we used hierarchical agglomerative clustering with group average linking to identify group clustering. One-way analysis of variations (ANOVA) test was used to detect the significance of means in alpha diversity and dissimilarity among groups in IBM SPSS Statistics 19. Permutational multivariate analysis of variance (PERMANOVA) was performed to test the effects of host age on gut microbiota and to analyze bacterial community differences between the gut and surrounding environment. For pair-wise comparisons of gut bacterial communities, both PERMANOVA and analysis of similarity (ANOSIM) were used to test community differences based on Bray-Curtis metric and weighted UniFrac metric in PAST 3.0.

A total of 2969724 V4–V5 16S rRNA sequence reads were obtained from 42 gut, six water and six food samples (on average: 49495 per sample; range: 26381-164795) after data quality control filtering and removal of primers, chimeras, and singletons. Alpha diversity of gut microbiota across host development is shown in Table 1. There was an increased trend in PD with host age (ANOVA, p = 0.064; regression model, p = 0.003). Moreover, three other alpha diversity indexes observed OTUs number, Shannon index and Simpson index increased linearly with age (ANOVA, p < 0.05; regression model, p < 0.01).

We observed eight phyla with more than 0.01% of relative abundance in the gut microbiota of southern catfish over the ontogeny. Of them, Tenericutes, Fusobacteria, Proteobacteria, and Bacteroidetes were the most abundant. The taxonomic composition of gut microbiota changed significantly across different age stages (Figure 1A and Figure S2). The phylum Tenericutes was the most abundant in individual samples at 8 and 18 dpf, with more than 75% of average abundance. The proportions decreased with host age (Kruskal-Wallis test, p < 0.001), while the Fusobacteria abundance significantly increased (p = 0.001), from 7.2% at 8 dpf to more than 45% at 65 and 125 dpf. Similarly, the abundance of Proteobacteria showed an increasing trend. Before 35 dpf, there were low values of the abundance of Bacteroidetes, which increased markedly at 65 and 125 dpf (p < 0.001). Of the observed phyla, which were mainly dominated by 21 genera with more than 0.1% of relative abundance at different stages (Figure S3). At 8 dpf Mycoplasma (Tenericutes) peaked in the gut and decreased with age (p < 0.001), meanwhile Cetobacterium (Fusobacteria) increased significantly (p = 0.001), toward the dominance at the late developmental stages. Several genera from Proteobacteria were dominated by unclassified Enterobacteriaceae, Plesiomonas, unclassified Aeromonadaceae, and Morganella.

The dominant OTUs in the gut showed temporal differences (Figure 1B). Microbial community comparisons at the OTU level visualized by NMDS based on Bray-Curtis distance metrics (Figure 2A) showed sample cluster according to age grouping (one-way PERMANOVA, p < 0.001). The individuals at 8 and 18 dpf were closer to each other than those at other developmental stages. Similarly, this trend was reflected in individuals at 65 and 125 dpf. The results of the pairwise comparison of PERMANOVA and ANOSIM consistently showed gut microbial assemblages were not different between 8 and 18 dpf, and between 65 and 125 dpf (Table 2). Moreover, gut microflora at 8 and 18 dpf significantly differed from those at 65 and 125 dpf (Figure 2B). Throughout the 5 stages of the ontogeny, the bacterial communities were similar in the early 2 stages which transited to the late 2 stages with microbial community reaching another similar and stable status. In addition, the average within-group dissimilarity based on UniFrac distance showed significant differences in microbial variations within individuals at the same developmental stages, and microbial variations increasing with host age (ANOVA, p < 0.001 for both distances, Figure 2C).

Gut samples and the corresponding environmental samples collected at 8, 35, and 125 dpf were used to compare the relationship between gut and environmental microbiota throughout southern catfish ontogeny. The total number of microbial OTUs in the gut was higher than that in the rearing water and food and increased in the gut over time (Figure S4). Similarly, the number of unique OTUs augmented in the gut, but not in the food and rearing water. Sample types affected alpha diversity (Table S2). Gut samples took on a lower alpha diversity than environmental samples in which all the diversity indexes were generally higher in the water than in the food at each sampling time point with an exception of PD. Further, PCoA showed a separation of the gut and environmental samples in both microbial structures and members at each time point (Figure S5). The results were consistent with the clustering link among groups of all samples throughout the host development (Figure 2B). Moreover, microbial communities in the water environment were more distantly separated from those in the fish gut than microbial communities in the food were (Figures 2A,B).

OTU classifications of microbial communities in the gut and environmental samples were performed at the genus level. Twenty-one most abundant taxa (on average, more than 1%) are shown in Table 3. Cetobacterium OTUs was the only shared abundant taxon by the gut, food and water samples. It was significantly more enriched in the gut (30.73%) than in the food (1.72%) and in the water (3.17%). The abundance of Mycoplasma was the highest in the gut (48.69%), but it was extremely low in the food (0.1%) and in the water (0.06%). Treponema consisted of 62.75% of microbiota in the food. However, its abundance accounted for only 0.01% in both the gut and rearing water. Other 13 taxa (more than 1%) in the water were either undetected or extremely low in abundance in the gut and food.

Inter-individual variations were minimized by a batch of fertilized eggs from a pair of parents in this study. After the complete absorption of the yolk sac, southern catfish can feed on small aquatic animals at the first feeding. Thus, when southern catfish are supplied with the same food source, their gut microbial variations are relatively controllable and accessible. In our long-term feeding experiment, a gradually increased alpha diversity of gut microbiota in southern catfish was observed over time, demonstrating the direct effects of host age on gut microbial diversity. In several relatively short-term fish studies, microbial assemblages in the developing individuals resulted in an increase of the diversity (Ingerslev et al., 2014; Bakke et al., 2015). This allows us to speculate that the increased microbial diversity is independent of host species during the development. Two other studies aimed to explore the associations between developing hosts, showing the microbial diversity of significant decreases in zebrafish (Zac Stephens et al., 2016) and increases in gibel carp (Li et al., 2017). The resulting differences are possibly attributed to food shifts and different living environments at several critical stages throughout the lifespan of the hosts (Penders et al., 2006).

Gut ecological niche becomes more mature and is complex with host development so that the developmental differences can lead to changes in the community diversity (Zac Stephens et al., 2016). Compared to this study, more significant microbiota turnover was observed in Yan's et al. (2016) study on the same catfish but fed on boiled egg yolk in the first 3 days at the larval stage. The early dietary differences could be greatly responsible for microbial alterations. Several dominant OTUs persistently occurred in the gut of southern catfish. For example, Mycoplasma OTUs enriched in the southern catfish gut at the early stage in this study and was absent in the previous study (Yan et al., 2016). The Mycoplasma was detected to be abundant in both wild and lab fish (Holben et al., 2002; Lowrey et al., 2015). Additionally, widely distributed Atlantic salmon (Salmo salar) populations were also characterized by Mycoplasmataceae phylotypes (Mycoplasma OTUs especially) in the life-cycle stages (Llewellyn et al., 2016), suggesting the correlation between gut niche and developmental change. Notably, Cetobacterium OTUs increased at 18 and 35 dpf in this study, reaching the dominance at the late stages (Bledsoe et al., 2016; Zac Stephens et al., 2016). This is in line with the result observed in juvenile southern catfish fed on the same diet from 18 to 33 days post-hatching, but differing from that in the adult fish collected from a natural lake (Yan et al., 2016). The high abundance of Cetobacterium OTUs was consistently found in the gut of several freshwater fish as well as on-growing zebrafish at the late age (Zac Stephens et al., 2016; Zhang et al., 2017). The discrepancies among studies further underline the importance and necessity of host developmental context in studying host microbiota and their responses to external factors (Goffredi et al., 2014). Furthermore, cross-talk between host and diet at different stages should be addressed in further work (Wong et al., 2015).

Microbial communities of the gut in southern catfish underwent sequential changes with their development. At 8 and 18 dpf, within-group dissimilarity in the gut microbial communities was less diverse; it sharply increased before 35 dpf and stayed stable at the late stages. Evidently, the host was subjected to microbial successions during ontogenesis, potentially supporting the assumption that morphological development drives the changes in the microbiota and vice versa (Hooper et al., 2012; Olszak et al., 2012; Jiao et al., 2015). A parallel phenomenon has been observed in zebrafish (Zac Stephens et al., 2016) and other vertebrates (Jiao et al., 2015). Different gut microbial communities in southern catfish fed on an identical food at different developmental stages at least explain that gradual adaptation of host-microbiota which leads to the alterations in the prevailing microbiota in the gut.

The fine-scale temporal sampling in the food and rearing water allowed us to evaluate microbial assembly relationships between the gut and external environment. The compositional relatedness analysis of microbial communities at multiple time points displayed that microbiota in the gut was differentiated from that in both the food and rearing water, suggesting gut microbial structure could not be simply explained by the environmental sources (Bakke et al., 2013; Wong et al., 2015; Yan et al., 2016). Despite different DNA extraction kits used in this study possibly leading to the differences in microbial communities between gut and environmental samples, a recent study showed that extraction method only explained small part of the community variation and did not result in inherent differences in taxonomic composition of microbial samples (Mackenzie et al., 2015). In the present study, some OTUs were shared by the gut, food and rearing water, but the environmental microbiota only accounted for relatively low abundances in the gut (Wong et al., 2015). Dominant microbiota in the catfish gut such as Mycoplasma was observed to be rather rare in the food and in the water (Llewellyn et al., 2016) and vice versa. For example, the highly abundant Treponema in the food and Flavobacterium in the water was poorly accumulated in the gut. On the contrary, Bledsoe et al. (2016) found OTU most abundant in the rearing water was dominant in the gut of channel catfish feeding endogenously at 3-day post-hatching. An introduction of the exogenous microbiota is likely to affect gut microbiota (Wu et al., 2013) through the competitions between microbiota and microbiota, and the interactions between host and microbiota. This process involves the influx of microbiota through fish feeding and excretion (Wu et al., 2013). Thus, the different abundance and enrichment of gut microbiota suggest the effects of early environmental exposure on initiation of host colonization.

The initial establishment of a nested gut microbial ecosystem is modulated by the ambient environment (Penders et al., 2006). The first exposure of fertilized eggs to water environment to some degree results in microbial colonization on their surface and unoccupied gut after hatching (Hansen and Olafsen, 1999) because microbiota in the aquatic environment can transfer freely from habitats to hosts. An early DGGE-based study of coho salmon (Oncorhynchus kisutch) reported that gut microbiota after first feeding was mainly from water and eggs epibiota (Romero and Navarrete, 2006). Surprisingly, recent work based on high-throughput sequencing showed that microbial compositions in late embryos poorly reflected microbial communities in surrounding water environment (Wilkins et al., 2015a,b), yet microbial communities in early embryos resembled those in the water environment (Wilkins et al., 2015a), potentially indicating that host selective pressures depend on different stages of developing embryos. At present, the lack of understanding of microbial associations between the environment and host at the continuous developments of fertilized eggs or embryogenesis is the case in many previous studies (Bakke et al., 2013, 2015; Ingerslev et al., 2014; Bledsoe et al., 2016). In addition to environmental effects, Wilkins et al. (2016) disclosed that salmonid fish egg-bacteria ecosystems were influenced by host genetics. Nevertheless, it is poorly understood that whether microbial responses to the hatching of fertilized eggs (from early embryos to late embryos) are consistent with their responses to individual growth after hatching (from early juveniles to adult) across the development, and whether the responses can be correlated with environmental microbiota at the initial stage of colonization. Samples before 8 dpf in this study were not collected, resulting in a failure in performing earlier microbial comparisons between the fish eggs and water. However, the dissimilarities of community structure between the developmental gut and water were higher than those between the gut and food, suggesting that microbiota in water has probably less effect on gut microbial community than that on food since the host begins with exogenous feeding (Llewellyn et al., 2014; Li et al., 2017).

In summary, we have demonstrated that host age was an important factor influencing the gut microbial community in southern catfish. An increase in gut microbial diversity occurred during the ontogenesis after exogenous feeding. The influences of host age on microbial community disappeared at the late stages under the controlled exogenous conditions such as diets that are the possible source of microbial community variation, indicating a vital role of early host development in shaping microbial occurrence in the gut. By continuous comparisons between host and environment, we found that gut microbial structure did not reflect environmental microbiota. These results suggest gut microbial changes could result from the enhanced adaptation to gut development. Further work can concentrate especially on the role of embryo hosts and environmental microbiota during the embryogenesis, and disentangle the effects of the initial establishment of an embryos-associated community on late individual microbial development and life history of the host. Quantifying surface microbiota from eggs to embryos stages, and to individual development associated with different sources of environmental microbiota will allow us to fully characterize temporal microbial dynamics of fish development, and to reveal how the early exposure of ontogeny affects late microbial colonization and persistence. It will provide an insight into the host selection of gut microbiota and microbial modulations in teleosts.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.