Adaptive changes in the intestinal microbiota of giant pandas following reintroduction

Abstract

The biggest challenge during the reintroduction of captive giant pandas into the wild is their ability to adapt to the natural environment, and the role of gut microbiota in this process remains unknown. Here, the gut microbiota was analyzed and categorized into training, exploration (1–3 months post-release) and stable period (4–6 months post-release) by activity intensity of released pandas. We found that the gut microbiota diversity of pandas was significantly higher during the stable period compared to the training period. Streptococcus was significantly enriched in the training period, but Clostridium became significantly enriched after being released. KEGG functional prediction analysis revealed that during the stable phase, carbohydrate and amino acids metabolism was significantly reduced, while pathways associated with cofactors and vitamins, other amino acids, lipids, nucleotide and energy metabolism were markedly enriched. This suggests that, after a three-month acclimation period, the transformation of the gut microbiota provides reintroduced giant pandas with more diverse energy acquisition strategies suited to the wild environment. This finding highlighted that the first 3 months post-release are a critical exploration period for digestive adaptation to the wild environment, which will help guide the implementation of future monitoring efforts post-release.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-16136-9.

Introduction

As a pivotal conservation strategy, reintroduction into the wild has been widely applied globally to protect endangered wildlife populations across diverse species^1–4. Due to prolonged exposure to a single captive environment, these animals often struggle to adapt to the wild, resulting in a low success rate post-release^5–7. The rapid changes in environmental factors such as food and social interactions between captivity and the wild make it difficult for them to quickly adjust their gastrointestinal functions and acquire necessary nutrients for survival post-release^8,9.

The adaptive transformation of gut microbiota facilitates the host’s acclimatization to new environments¹⁰. The gut microbiota, often considered an essential “organ” that provides nutrition, regulates epithelial cell development, and guides innate immunity, can influence the host’s energy balance and immune function through the production of short-chain fatty acids, vitamins, and other metabolites^11–13. The gut microbiota not only directly protect the host from pathogens but also maintain intestinal and systemic health by modulating immune responses, thereby preserving immune homeostasis¹⁴. Conversely, a reduction in gut microbiota diversity and dysbiosis could lead to metabolic and immune-related diseases, including obesity, diabetes, cardiovascular diseases, and inflammatory bowel disease^13,15.

Currently, due to human activities and environmental factors, the wild giant panda (Ailuropoda melanoleuca) population has been fragmented into 33 isolated groups, 24 of which are at high risk of extinction¹⁶. The reintroduction of captive giant pandas into the wild can effectively increase the genetic diversity of wild populations, thereby aiding the rejuvenation of small isolated populations and reducing their extinction risk³. Compared to wild individuals, captive-born giant pandas are limited by factors such as diet composition, activity range, and environmental enrichment¹⁷. This limitation restricts their opportunities to acquire microorganisms from sympatric species, natural food sources, soil, and water, leading to reduced gut microbiota diversity and relatively simple microbial structures^10,18,19. To enhance the physiological adaptability of reintroduced giant pandas and maintain stable and healthy gut microbiota, ensuring a gradual transition of their microbiota from captivity to the wild is essential for improving adaption to the wild^7,10,18,20.

The physiological structure of giant panda’s carnivorous gut, coupled with their specialized bamboo-eating behavior, underscores the instability and resilience of their gut microbiota, but at the same time exhibits significant plasticity in different environments^21–23. Environmental changes inevitably alter the composition of their gut microbiota, and rapid shifts within a short period can reduce the efficiency of nutrient acquisition from food¹⁷. Systematic pre-release training could effectively enhance environmental enrichment gradually, thereby increasing the diversity and richness of their gut microbiota, and purposefully improving the structure and function of the gut microbiota to enhance adaptability to wild environments post-reintroduction^17,20,24. Influenced by factors such as environment, host age, and associated animals, the gut microbiota of subadult pandas during pre-release training period tended to resemble that of their captive mothers, showed a gradual decrease in the abundance of Streptococcus and Escherichia coli, transitioning to a Pseudomonas-dominated gut type within 6 months⁷. After a full year of adaptive training in the wild habitat, their gut microbiota composition transitions successfully to a Clostridium-Pseudomonas type, closely resembling the Pseudomonas-dominated gut type of wild individuals²⁴. Unfortunately, due to outdated tracking and monitoring methods, as well as the low sensitivity or low transmission success rate of tracking collars, most current studies on the gut microbiota of giant pandas are limited to comparisons among captive, released, and wild states. These studies lack focused research on the adaptive transitions of targeted individuals during the critical early stages of reintroduction.

This study aims to utilize high-throughput sequencing techniques to continuously monitor the gut microbiota of released giant pandas from training states to post-reintroduction. By investigating the changes in the composition, structure, and function of the gut microbiota during the reintroduction process, we seek to elucidate the adaptive characteristics of the gut microbiota after reintroduction, trying to determine the transitional period of gut microbiota changes during the early stages of reintroduction and to enrich the research data on the gut microbiota of rewilded giant pandas. Our findings will provide a theoretical basis for the formulation and optimization of rewilding training systems and monitoring protocols for captive giant pandas.

Materials and methods

Sample collection and grouped

In this study, all samples from the habitat pre-release training phase (Training Phase, TP) were collected in Daxiangling Nature Reserve between 2022 and 2024 (Fig. 1A and C). Fecal samples were obtained from 3 adult female giant pandas (Aged 10, n = 1; Aged 8, n = 2). Post- release phase (including Exploration Phase (EP) during the first 1–3 months post-reintroduction, and the Stable Phase (SP) during the 4–7 months post-reintroduction) fecal samples were collected from 1 released pandas (Aged 10, n = 1) in November 2023 to June 2024. All sample providers maintained good health throughout the training and reintroduction process. Thanks to the successful implementation of artificial assisted soft-release training (Invention Patent in China: A domestication method of wild release and artificial intervention of giant panda; patent number: ZL 2017 1 0576548.4), researchers were able to maintain close contact with the giant pandas for extended periods, allowing for the immediate collection of fecal samples following defecation. Collection personnel strictly adhered to sample collection protocols, wearing disposable surgical masks and PE gloves. Once the pandas defecated, the feces were transferred to disposable sterile surgical drapes. The outer layer of the feces was gently removed, and the internal portion was placed into 50 mL sterile cryogenic vials. After labeling, the samples were rapidly frozen in sampling transport boxes containing dry ice to ensure that the gut microbiota remained unaffected by external environmental factors. The samples were subsequently transferred to a laboratory and stored in a −80 ℃ ultra-low temperature freezer until DNA extraction.

Fig. 1.

The information of study area, sample collection and groups. (A) The study area in China. The Green line indicate the borders of Sichuan Province; (B) The study area in Sichuan Province, China. The green areas indicate the Giant Panda National Park. (C) The study area and sample location in the Giant Panda National Park. The green areas indicate the reintroduction training area, the black line indicate the borders of Daxiangling Nature Reserve, red spot represented the specific collection location; (D) The monthly MCP active area of reintroduced giant panda, the different colors indicate the MCP activity area of the released panda in different months.

By collecting the timed activity location data from the GPS collars worn by the individual during training and post-reintroduction, we utilized the Minimum Bounding Geometry tool in ArcGIS 10.3 to obtain the minimum convex polygon (MCP) of the activity locations for the reintroduced individual. The MCP was used as the home range to statistically analyze their activity range on a monthly basis. Based on the state and changes in the activity range and area of the experimental individual, the collected fecal samples were categorized into 3 phases: the Training Phase (TP), the Exploration Phase (EP) during the first 1–3 months post-reintroduction (Fig. 1D), and the Stable Phase (SP) during the 4–7 months post-reintroduction (Fig. 1D). The Training Phase (TP) involved gradual exposure to wild-like conditions, including natural bamboo foraging, reduced human interaction, and GPS collar monitoring. Considering the potential impact of seasonal feeding variations on the fecal microbiota of giant pandas, we conducted a year-long sample collection for the TP group from November 2022 to November 2023 ²⁵. This collection encompassed both the culm-eating season (December to April, n = 6) and the leaf-eating season (May to November, n = 6). Based on this, we subdivided the TP group into the TP culm group and the TP leaf group. The TP culm group represents the TP stage samples collected during the culm-eating season, while the TP leaf group represents those collected during the leaf-eating season. For the SP group, samples were gathered from March to June 2024 (n = 8). Using the same method, the SP group was also divided into the SP culm group and the SP leaf group. The samples for the EP group were collected from November 2023 to January 2024 (n = 7). Since all samples in the EP group were collected during the culm-eating season, no further subdivision was made for this group. To study the impact of individual differences on the gut microbiota, we analyzed the structural differences in gut microbiota by treating the three individuals in the TP group (TH, TX, TQ) as separate subgroups. The specific times and locations of sample collection are detailed in Table S1. The minimum convex polygon (MCP) of reintroduced individual are detailed in Table S2.

Diet

Based on our vegetation survey conducted in the Daxiangling reintroduction training area and the regions subsequently visited by the released individuals, it was identified that the sole available food source for giant pandas within the study area is arrow bamboo (Bashania fangiana)²⁶. Besides this natural diet, the giant pandas were provided with a small quantity of apples solely during their monthly health checks for blood sampling purposes. Considering that more than 99% of the giant panda’s diet consists of bamboo, the gut microbiota sequencing results of the experimental individuals in this study will not be affected by dietary changes²⁵.

DNA extraction and amplification

The collected fecal samples were pretreated using magnetic beads²⁷, and the total microbial DNA was extracted from the feces using the Qiagen QIAamp DNA Stool Mini Kit (Qiagen, Germany). The DNA quality of each sample was determined by 1% agarose gel electrophoresis. For each sample, the entire 16 S rRNA gene was amplified by PCR using the primer pair 27 F (5′-AGAGTTTGATCCTGGCTCAG-3′) and 1492R (5′-TACCTTGTTACGACTT-3′) and the following PCR program: 95 °C for 3 min; 30 cycles of 95 °C for 30 s, 56 °C for 30 s, and 72 °C for 3 min; 72 °C for 10 min; hold at 4 °C (T100 Thermal Cycler PCR thermocycler, BIO-RAD, USA). The PCR amplification of each sample was performed in triplicate to minimize the stochastic effect. After electrophoresis, The PCR products were purified using the AMPure^® PB beads (Pacifc Biosciences, CA, USA) and quantified with Qubit 4.0 (Thermo Fisher Scientific, USA).

DNA library construction and sequencing

Purified products were pooled in equimolar and DNA library was constructed using the SMRTbell prep kit 3.0 (Pacifc Biosciences, CA, USA) according to PacBio’s instructions. Purified SMRTbell libraries were sequenced on the Pacbio Sequel IIe System (Pacifc Biosciences, CA, USA) by Majorbio Bio-Pharm Technology Co. Ltd. (Shanghai, China). High-fidelity (HiFi) reads were obtained from the subreads, generated using circular consensus sequencing via SMRT Link v11.0.

Data processing and statistical analysis

HiFi reads were barcode-identified and length-filtered. For bacterial 16 S rRNA gene, sequences with a length < 1,000 or > 1,800 bp were removed. The optimized-HiFi reads were de-noised using DADA2²⁸ plugin in the Qiime2²⁹ (version 2020.2) pipeline with recommended parameters, which obtains single nucleotide resolution based on error profiles within samples. DADA2 denoised sequences are usually called amplicon sequence variants (ASVs). Chloroplast and mitochondrial sequences were removed from all samples. To minimize the impact of sequencing depth on subsequent Alpha and Beta diversity analyses, we performed rarefaction on the sequencing data from all samples. After rarefaction, the average sequence coverage (Good’s coverage) for each sample remained at 99%. Taxonomic classification of ASVs was performed using the Naive Bayes classifier in Qiime2 (version 2020.2), based on the Silva 16 S rRNA gene database (v 138). Functional prediction of 16 S data was conducted using PICRUSt2 (version 2.2.0)³⁰.

Statistical analysis

Species annotation for the obtained ASVs data information was performed using Qiime2 software (version 2020.2). Species with an abundance > 0.1% in each sample or group at various taxonomic levels (Phylum, Class, Order, Family, Genus, Species) were selected using the maximum value sorting method and visualized with R software (Version 3.6.1, ggplot2 package). Alpha diversity indices (ACE, Chao1, Shannon, Simpson and Faith’s PD) were calculated and visualized using R software (Version 3.6.1, vegan package). We used the Kruskal-Wallis test to analyze Alpha differences among groups, followed by Dunn’s post hoc test for pairwise comparisons, and applied the false discovery rate (FDR) correction to adjust the p-values obtained from Dunn’s post hoc test. We also conducted linear mixed-effects models (LMM) to analyze Alpha differences using the R software (Version 3.6.1, lme4 and lmerTest package), with season as a fixed blocking factor and panda ID as a random intercept to account for repeated measures across individuals. The stability of microbial communities across groups was assessed by calculating the Average Variation Degree (AVD) using R software (Version 3.6.1, vegan package) and visualized accordingly. Beta diversity was evaluated by calculating the Bray-Curtis distance between groups using R software (Version 3.6.1, vegan package), followed by Principal Coordinates Analysis (PCoA) for visualization. Group similarities and significant differences were statistically analyzed using ANOSIM (Analysis of Similarities) and PERMANOVA (Permutational multivariate analysis of variance)³¹. To assess the impact of seasonal foraging and individual differences on the gut microbiota structure of giant pandas, we conducted inter-group similarity analysis using a multivariate PERMANOVA mixed model with 999 permutations, performed in R software (Version 3.6.1, vegan package), where fecal collection season and sex were considered as restricting factors. Further the linear mixed-effects models (LMM) was used to analyze the difference in Beta diversity by the R software (Version 3.6.1, lme4 and lmerTest package), with season as a fixed blocking factor and panda ID as a random intercept to account for repeated measures across individuals. Inter-group differential analysis and visualization were conducted using R software (Version 3.6.1, stats and ggplot2 package). Differential phyla and genera were identified using the (1) Kruskal-Wallis test, followed by Dunn’s post hoc test for pairwise comparisons, and applied the false discovery rate (FDR) correction to adjust the P-values obtained from Dunn’s post hoc test, and (2) the linear mixed-effects models (LMM) in R software (Version 3.6.1, lme4 and lmerTest package), with season as a fixed blocking factor and panda ID as a random intercept. Additionally, Linear Discriminant Analysis (LDA) was employed to identify significantly different phyla and genera, with an LDA threshold set at 2.0. Microbial functional prediction analysis was performed using PICRUSt2 (http://huttenhower.sph.harvard.edu/galaxy), and the results were annotated against level 2 pathways of the Kyoto Encyclopedia of Genes and Genomes (KEGG) database^32,33. The types and abundances of the metabolic pathways were visualized using GraphPad Prism 9. Differential metabolic pathways between groups were identified using (1) the Kruskal-Wallis test, followed by Dunn’s post hoc test for pairwise comparisons, and (FDR) correction and (2) the linear mixed-effects models (LMM) in R software (Version 3.6.1, lme4 and lmerTest package), with season as a fixed blocking factor and panda ID as a random intercept. Finally, correlation analysis between the differential genera and differential metabolic pathways was conducted using the Spearman coefficient and visualized with R software (Version 3.6.1, vegan package). All statistical results in this chapter were considered significant at 0.01 < p or adjusted p-value (FDR) < 0.05 and highly significant at p or adjusted p-value (FDR) < 0.01.

Ethical approval

The research complied with methods and experimental protocols approved by the Institutional Animal Care and Use Committee and conformed to Chengdu Research Base of Giant Panda Breeding, Sichuan Province, China (IACUC No. 202004). This study is performed in accordance with relevant guidelines and regulations. All methods are reported in accordance with ARRIVE guidelines (https://arriveguidelines.org).

Results

Sequencing data

After filtering and denoising the raw data, a total of 687,773 high-quality clean reads were obtained across the TP, EP, and SP groups, with an average of 25,473 sequences per sample (ranging from 20,722 to 30,262). The average sequence length was 1480 bp, with a maximum length of 1493 bp and a minimum length of 1474 bp. A total of 1253 ASVs were identified across the 27 samples, with an average of 46 ASVs per sample. The number of ASVs per sample ranged from a minimum of 30 to a maximum of 66.

Analysis of microbial diversity differences

A total of 381 ASVs were detected across the three groups, with 179 ASVs in the TP group, 124 ASVs in the EP group, and 201 ASVs in the SP group. Among these, 32 ASVs were shared among all three groups. The TP, EP, and SP groups had 126, 48, and 116 unique ASVs, respectively (Figure S1). The TP and EP groups shared 6 ASVs, the TP and SP groups shared 15 ASVs, and the EP and SP groups shared 38 ASVs (Fig. 2A). The microbial richness increased progressively across the TP, EP, and SP groups (ACE, TP: 42.56 ± 9.91, EP: 45.09 ± 8.74, SP: 53.63 ± 10.46; Chao1, TP: 42.42 ± 9.75, EP: 45.00 ± 8.72, SP: 53.63 ± 10.46), with the SP group showing significantly higher richness compared to the TP group (Chao1, Kruskal-Wallis test: TP vs. SP, adjusted p-value (FDR) = 0.035; LMM: TP vs. SP, adjusted p-value (FDR) = 0.047, seasonal variation: p = 0.453, individual variance: 15.65 ± 4.23, Fig. 2B and C). Microbial diversity initially decreased and then increased with the progression of the recolonization process (Shannon, TP: 2.01 ± 0.34, EP: 1.91 ± 0.17, SP: 2.07 ± 0.23, Fig. 2E; Simpson, TP: 0.28 ± 0.05, EP: 0.37 ± 0.05, SP: 0.34 ± 0.05, Fig. 2F). The Simpson index indicated that microbial diversity in the EP group was significantly lower than in the TP group (Simpson, Kruskal-Wallis test: TP vs. EP, adjusted p-value (FDR) = 0.011, LMM: TP vs. EP, adjusted p-value (FDR) = 0.003, seasonal variation: p = 0.267, individual variance: 0.01 ± 0.01, Fig. 2F), and the microbial diversity in the SP group was also lower than in the TP group (Simpson, Kruskal-Wallis test: TP vs. SP, adjusted p-value (FDR) = 0.095, LMM: TP vs. SP, adjusted p-value (FDR) = 0.026, seasonal variation: p = 0.267, individual variance: 0.01 ± 0.01, Fig. 2F).The Faith’s PD index analysis based on phylogenetic information revealed that the diversity results of the three groups exhibited the same patterns and trends as those observed previously (Faith’s PD, TP: 1.89 ± 0.27, EP: 1.70 ± 0.36, SP: 2.10 ± 0.65), with no significant differences detected between the groups (Three groups: adjusted p-value (FDR) = 0.203, TP vs. EP, adjusted p-value (FDR) = 0.999, TP vs. SP, adjusted p-value (FDR) = 0.630, TP vs. EP, adjusted p-value (FDR) = 0.244). The analysis of the Average Variation Degree (AVD) demonstrated that microbial community stability was highest in the TP group (AVD = 0.6304), followed by the SP group (AVD = 0.6993) and the EP group (AVD = 0.7252); however, no significant differences were observed (Fig. 2D).

Fig. 2.

Analysis results of ASVs distribution and microbial diversity among groups. (A) VENN diagram analysis on ASV levels; (B) Analysis of differences in ACE richness index among groups; (C) Analysis of differences in Chao richness index among groups; (D) Stability analysis of the Average Variation Degree (AVD) of three groups of samples; (E) Analysis of Shannon diversity difference among groups; (F) Analysis of Simpson diversity differences among groups. The gray color represents the Training Phase (TP group), yellow represents the Exploration Phase (EP group), and blue represents the Stable Phase (SP group). The scatter points in the boxplot represents individual samples. Red/blue asterisks (*): Significant differences via Kruskal-Wallis/LMM (FDR-adjusted p < 0.05). *Represents significant difference (0.01 < adjusted p-value (FDR) < 0.05), **represents extremely significant difference (0.001 < adjusted p-value (FDR) < 0.01), ***represents highly extreme significant difference (adjusted p-value (FDR) < 0.001).

Analysis of differences in microbial community structure

The PCoA analysis based on Bray-Curtis distances visualized the spatial relationships among samples. The TP group exhibited higher dispersion among samples and showed clear separation in structure compared to other two groups (Figure S2). The microbial communities of the EP group partially overlapped with those the SP group, indicating a higher similarity in composition. The first two components accounted for 52.71% and17.84% of the total variation (Fig. 3A). Linear mixed-effects models revealed significant beta diversity divergence on PC1 (TP vs. EP: adjusted p-value (FDR) < 0.001; TP vs. SP: adjusted p-value (FDR) < 0.001; seasonal variation, p = 0.633, individual variance: 0.02 ± 0.05) and PC2 (TP vs. EP: adjusted p-value (FDR) = 0.027; seasonal variation, p = 0.496, individual variance: 0.01 ± 0.03), with group differences persisting after accounting for seasonal and individual variability. Statistical verification using Adonis and ANOSIM confirmed significant separation the three groups (Adonis, R² = 0.431, p = 0.001; ANOSIM, R = 0.443, p = 0.001). Pairwise comparisons between groups showed the EP and SP groups had higher structural similarity (Adonis, R² = 0.209, p = 0.005; ANOSIM, R = 0.320, p = 0.006, Table 1), while the TP group exhibited more substantial differences in community structure compared to both the EP and SP groups (Adonis, TP vs. EP: R² = 0.355, p = 0.001; TP vs. SP: R² = 0.379, p = 0.001; ANOSIM, TP vs. EP: R = 0.452, p = 0.003; TP vs. SP: R = 0.590, p = 0.002, Table 1). The ANOSIM analysis of inter-group differences based on the weighted Unifrac distance, which incorporates phylogenetic information, showed results consistent with those observed in the Beta diversity analysis based on Bray-Curtis distances (ANOSIM, Three groups: R = 0.242, p = 0.002; TP vs. EP: R = 0.133, p = 0.019; TP vs. SP: R = 0.210, p = 0.026; EP vs. SP: R = −0.035, p = 0.730).

Fig. 3.

The bacterial composition and structure difference among groups. (A) PCoA based on the Bray-curtis distance matrix to find principal coordinates. (Description: X-axis, PCoA axis1 and Y axis, PCoA axis2. The scale of the X-axis and the Y-axis are the projection coordinates of the sample points in the two-dimensional plane, respectively. A dot represents each sample, and different colors represent different groups). The microbial community bar plot at the (B) Phylum level and (C) Genus level among groups. (D) The composition difference on Phylum level. (E-F) The composition difference of differential bacterial genera stratified by relative abundance threshold (E: high-abundance taxa ≥ 2.5%; F: low-abundance taxa < 2.5%). (G) The Linear Discriminant Analysis (LDA) demonstrated distinct microorganism on Genus level enriched in the TP, SP and SP group. When the default LDA value is > 2.0 and the p < 0.05, the result corresponds to a differential species. Red/blue asterisks (*): Significant differences via Kruskal-Wallis/LMM (FDR-adjusted p < 0.05). *Represents significant difference (0.01 < adjusted p-value (FDR) < 0.05), **represents extremely significant difference (0.001 < adjusted p-value (FDR) < 0.01), ***represents highly extreme significant difference (adjusted p-value (FDR) < 0.001).

Table 1.

The results of fecal bacterial construction similarity analyses between groups.

Group	Adonis		ANOSIM
	R 2	P	R	p
TP VS. EP	0.3549	0.001**	0.4524	0.003**
TP VS. SP	0.3788	0.001**	0.5902	0.002**
EP VS. SP	0.209	0.005**	0.3196	0.006**

**Represents extremely significant difference (p < 0.01).

To investigate the effects of individual differences on the gut microbiota, three individuals in the TP group were treated as separate subgroups and analyzed alongside the EP and SP groups using ANOSIM, with results visualized by PCoA. The analysis revealed no significant differences in gut microbial structure among the three individuals (TH vs. TX, R = 0.167, p = 0.800; TX vs. TQ, R = −0.188, p = 0.967; TH vs. TQ, R = −0.047, p = 0.466; Figure S3A, Table S3). The results of the multivariate PERMANOVA, with individual pandas set as restricting factors, indicated that differences among individuals did not have a significant effect on the inter-group variation in gut microbiota structure across the TP, EP, and SP groups (R² = 0.011, p = 0.743). Further ANOSIM analysis was conducted by dividing the samples into culm-feeding and leaf-feeding seasons. Significant differences in gut microbial structure were observed between the TP group and both the EP and SP groups during the culm-feeding season (TP culm vs. EP, R = 0.669, p = 0.001; TP culm vs. SP culm, R = 0.484, p = 0.001), indicating that, within the same season, variations in living environments led to substantial differences in the gut microbiota structure of giant pandas (Figure S3B, Table S3). Although the seasonal variation in gut microbial structure was not as pronounced as the differences between the TP group and both the EP and SP groups during the culm-feeding season, it was still evident in reintroduced pandas (EP vs. SP culm, R = 0.463, p = 0.021, Figure S3B, Table S3). However, no seasonal differences were detected in the gut microbiota structure within the TP group (TP culm vs. TP leaf, R = −0.107, p = 0.810) or the SP group (SP culm vs. SP leaf, R = 0.083, p = 0.647, Figure S3B, Table S3). The results of the multivariate PERMANOVA, with seasonal foraging type of the giant panda as a restricting factor, showed that seasonal differences did not have an impact on the inter-group variation in gut microbiota structure across the TP, EP, and SP groups (R² = 0.041, p = 0.516). These findings suggest that, compared to the effects of seasonal foraging, differences in gut microbial structure induced by living environments, particularly during the adaptation phase of reintroduction, have a more significant impact.

Analysis of differences in microbial composition

Annotation of ASVs identified a total of 7 phyla, 12 classes, 16 orders, 30 families, 38 genera, and 56 species across all 27 fecal samples. At the phylum level, the TP, EP, and SP groups shared 3 bacterial phyla, which accounted for over 99% of the relatively abundance in all samples (Fig. 3B). The most relatively abundant phylum was Firmicutes (TP = 98.18% ± 1.60%, EP = 98.83% ± 2.11%, SP = 96.54% ± 5.18%), followed by Proteobacteria (TP = 1.79% ± 1.60%, EP = 1.14% ± 2.05%, SP = 3.38% ± 5.20%) and Cyanobacteria (TP = 0.02% ± 0.03%, EP = 0.001% ± 0.004%, SP = 0.08% ± 0.10%, Fig. 3B). The TP and EP groups shared an unclassified phylum (unclassified Bacteria, TP = 0.007% ± 0.020%, EP = 0.034% ± 0.070%). The TP and SP groups shared Actinobacteria (TP = 0.008% ± 0.023%, SP = 0.006% ± 0.018%) and Verrucomicrobia (TP = 0.004% ± 0.011%, SP = 0.003% ± 0.007%, Fig. 3B). Additionally, Bacteroidota was only present in the TP group (0.001% ± 0.003%, Fig. 3B). At the genus level, Clostridium was the most relatively abundant genus in all 3 groups (TP = 72.63% ± 23.36%, EP = 96.11%, ± 4.16%, SP = 91.53% ± 8.64%, Fig. 3C). In the TP group, the next most relatively abundant genera were Streptococcus (25.17% ± 22.73%), Escherichia (1.46% ± 1.50%), Terrisporobacter (0.34% ± 0.30%) and Klebsiella (0.10% ± 0.16%, Fig. 3C). In the EP group, the second to fifth most relatively abundant genera were Streptococcus (1.51% ± 2.76%), Escherichia (1.01% ± 1.92%), Terrisporobacter (0.80% ± 1.52%) and Leuconostoc (0.40% ± 0.76%, Fig. 3C). In the SP group, the second to fifth most relatively abundant genera were Escherichia (3.02% ± 4.76%), Terrisporobacter (2.29% ± 2.64%), Leuconostoc (2.14% ± 6.00%), and Streptococcus (0.32% ± 0.24%, Fig. 3C).

Our result showed that the relative abundance of Firmicutes in the EP group was significantly higher than in the TP group (Kruskal-Wallis test: adjusted p-value (FDR) = 0.047, LMM: adjusted p-value (FDR) = 0.049, seasonal variation: p = 0.443, individual variance: 0.01 ± 0.03, Fig. 3D). Cyanobacteria were significantly more relatively abundant in the SP group compared to both the TP group and EP group (TP vs. SP: Kruskal-Wallis test: adjusted p-value (FDR) = 0.019, LMM: adjusted p-value (FDR) = 0.014; EP vs. SP: Kruskal-Wallis test: adjusted p-value (FDR) = 0.003, LMM: adjusted p-value (FDR) = 0.007, seasonal variation: p = 0.117, individual variance: 0.01 ± 0.02, Fig. 3D). At the genus level, Streptococcus was significantly more relatively abundant in the TP group compared to the EP (Kruskal-Wallis test: adjusted p-value (FDR) = 0.003, LMM: adjusted p-value (FDR) = 0.009) and SP groups (Kruskal-Wallis test: adjusted p-value (FDR) = 0.002, LMM: adjusted p-value (FDR) = 0.005) with group differences persisting after accounting for seasonal and individual variability (seasonal variation: p = 0.443, individual variance: 0.01 ± 0.03, Fig. 3E). Additionally, Clostridium was significantly more relatively abundant in the EP group compared to the TP group (Kruskal-Wallis test: adjusted p-value (FDR) = 0.010, LMM: adjusted p-value (FDR) = 0.018, seasonal variation: p = 0.572, individual variance: 0.02 ± 4.88, Fig. 3E). In the SP group, Terrisporobacter (Kruskal-Wallis test: adjusted p-value (FDR) = 0.048, LMM: adjusted p-value (FDR) = 0.031, seasonal variation: p = 0.243, individual variance: 0.02 ± 0.03) and Leuconostoc (Kruskal-Wallis test: adjusted p-value (FDR) = 0.044, LMM: adjusted p-value (FDR) = 0.038, seasonal variation: p = 0.259, individual variance: 0.01 ± 0.46) were significantly more relatively abundant compared to the TP group (Fig. 3F). Unclassified Cyanobacteria were significantly more abundant in the SP group (EP vs. SP, Kruskal-Wallis test: adjusted p-value (FDR) = 0.003, LMM: adjusted p-value (FDR) = 0.011; TP vs. SP, Kruskal-Wallis test: adjusted p-value (FDR) = 0.017, LMM: adjusted p-value (FDR) = 0.015; seasonal variation: p = 0.117, individual variance: 0.02 ± 0.02, Fig. 3F). Further biomarker analysis using Linear Discriminant Analysis (LDA) revealed that specific genera were significantly enriched in different groups. Streptococcus was significantly enriched in the TP group (LDA Score = 5.099, p = 0.0002), Clostridium in the EP group (LDA Score = 5.076, p = 0.026), and both unclassified Cyanobacteria (LDA Score = 4.371, p = 0.008) and Terrisporobacter (LDA Score = 5.099, p = 0.001) in the SP group (Fig. 3G).

Functional metabolic pathways and correlation analysis of microbial communities

By comparing with the KEGG database, a total of 43 KEGG level 2 metabolic pathways were annotated across the samples from all 3 groups. The top 5 pathways in relative abundance accounted for 65% of the total pathways, which were: Global and overview maps (TP = 39.01%, EP = 38.46%, SP = 38.44%), Carbohydrate metabolism (TP = 11.26%, EP = 11.25%, SP = 11.23%), Amino acid metabolism (TP = 6.59%, EP = 6.55%, SP = 6.53%), Membrane transport (TP = 4.89%, EP = 4.61%, SP = 4.62%), and Metabolism of cofactors and vitamins (TP = 3.94%, EP = 4.02%, SP = 4.03%, Fig. 4A).

Fig. 4.

The difference and correlation analysis of metabolism pathways on KEGG level 2 among groups. (A) The heat map of metabolism pathways on KEGG level 2 in TP, EP and SP group. Global and overview maps refer to broad metabolic pathway categories in KEGG, encompassing general cellular processes. (B) The difference in metabolism pathways on KEGG level 2 among groups. (C) The correlation between the different Genera and metabolism pathways. Red/blue asterisks (*): Significant differences via Kruskal-Wallis/LMM (FDR-adjusted p < 0.05). *Represents significant difference (0.01 < adjusted p-value (FDR) or p < 0.05), **represents extremely significant difference (0.001 < adjusted p-value (FDR) or p < 0.01), ***represents highly extreme significant difference (adjusted p-value (FDR) or p < 0.001).

Among 43 KEGG level 2 metabolic pathways, we identified 11 significantly different metabolic pathways between each 2 groups. The TP group and EP group had 5 significantly different pathways, while the TP group and SP group had 11 significantly different pathways. The relative abundance of the Amino acid metabolism pathway (Kruskal-Wallis test: adjusted p-value (FDR) = 0.011, LMM: adjusted p-value (FDR) = 0.049, seasonal variation: p = 0.446, individual variance: 0.01 ± 0.43), Carbohydrate metabolism (Kruskal-Wallis test: adjusted p-value (FDR) = 0.001, LMM: adjusted p-value (FDR) = 0.002, seasonal variation: p = 0.366, individual variance: 0.05 ± 0.21) and Translation (Kruskal-Wallis test: adjusted p-value (FDR) = 0.012, LMM: adjusted p-value (FDR) = 0.048, seasonal variation: p = 0.230, individual variance: 0.77 ± 0.18) were significantly higher in the TP group compared to the SP group (Fig. 4B, Table S4). The relative abundance of Membrane transport (TP vs. EP: Kruskal-Wallis test: adjusted p-value (FDR) = 0.012, LMM: adjusted p-value (FDR) = 0.097; TP vs. SP: Kruskal-Wallis test: adjusted p-value (FDR) = 0.012, LMM: adjusted p-value (FDR) = 0.042, seasonal variation: p = 0.799, individual variance: 0.02 ± 0.19), was significantly higher in the TP group compared to the EP and SP groups (Fig. 4B, Table S4). Conversely, the relative abundance of the Metabolism of other amino acids (Kruskal-Wallis test: adjusted p-value (FDR) = 0.008, LMM: adjusted p-value (FDR) = 0.023, seasonal variation: p = 0.259, individual variance: 0.29 ± 0.25), Signal transduction (Kruskal-Wallis test: adjusted p-value (FDR) = 0.005, LMM: adjusted p-value (FDR) = 0.045, seasonal variation: p = 0.414, individual variance: 0.03 ± 0.20), Metabolism of cofactors and vitamins (Kruskal-Wallis test: adjusted p-value (FDR) = 0.001, LMM: adjusted p-value (FDR) = 0.032, seasonal variation: p = 0.664, individual variance: 0.19 ± 0.44) and Energy metabolism (Kruskal-Wallis test: adjusted p-value (FDR) = 0.009, LMM: adjusted p-value (FDR) = 0.037, seasonal variation: p = 0.365, individual variance: 0.02 ± 0.01) were significantly lower in the TP group compared to the SP group (Fig. 4B, Table S4). The relative abundance of Nucleotide metabolism (TP vs. EP: Kruskal-Wallis test: adjusted p-value (FDR) = 0.005, LMM: adjusted p-value (FDR) = 0.048; TP vs. SP: Kruskal-Wallis test: adjusted p-value (FDR) = 0.016, LMM: adjusted p-value (FDR) = 0.041, seasonal variation: p = 0.799, individual variance: 0.37 ± 0.03), Lipid metabolism (TP vs. EP: Kruskal-Wallis test: adjusted p-value (FDR) = 0.028, LMM: adjusted p-value (FDR) = 0.046; TP vs. SP: Kruskal-Wallis test: adjusted p-value (FDR) = 0.001, LMM: adjusted p-value (FDR) = 0.025, seasonal variation: p = 0.662, individual variance: 0.68 ± 2.30), and Cell motility (TP vs. EP: Kruskal-Wallis test: adjusted p-value (FDR) = 0.009, LMM: adjusted p-value (FDR) = 0.011; TP vs. SP: Kruskal-Wallis test: adjusted p-value (FDR) = 0.037, LMM: adjusted p-value (FDR) = 0.047, seasonal variation: p = 0.683, individual variance: 0.01 ± 0.08) pathways were significantly lower in the TP group compared to the EP and SP groups (Fig. 4B, Table S4).

Further Spearman correlation analysis was performed to explore the potential relationships between differential genera and metabolic pathways. The results showed that the relative abundance of the carbohydrate metabolism pathway was significantly negatively correlated with the relative abundance of Leuconostoc (r = −0.404, p = 0.037, Fig. 4C, Table S5). The relative abundance of Terrisporobacter was significantly negatively correlated with the relative abundance of carbohydrate metabolism pathway (r = −0.569, p = 0.002), and positively correlated with the relative abundance of energy (r = 0.389, p = 0.045) and lipid metabolism pathway (r = 0.422, p = 0.028, Fig. 4C, Table S5). The relative abundance of Clostridium was significantly negatively correlated with the relative abundance of translation, amino acid metabolism and membrane transport pathways, and significantly positively correlated with the relative abundance of six metabolism pathways including nucleotide, cofactors and vitamins, energy, lipid, signal transduction and cell motility metabolism (Fig. 4C, Table S5). Conversely, the relative abundance of Streptococcus was significantly positively correlated with the relative abundance of carbohydrate metabolism, amino acid metabolism, translation, membrane transport pathways, and negatively correlated with the relative abundance of remaining 7 metabolism pathways (Fig. 4C, Table S5).

Discussion

Previous studies on the reintroduction of captive giant pandas have revealed significant differences in the composition, structure, and function of gut microbiota between captive and wild populations. These studies have further demonstrated that reintroduction training can shape the gut microbiota of giant pandas towards that of wild individuals^20,24. Increasing evidence highlighted that digestive system diseases are a leading cause of mortality in giant pandas³⁴. However, earlier research has not thoroughly described the critical initial period of gut microbiota transformation at the beginning of reintroduction activities. This study closely monitored the adaptive changes in the gut microbiota of giant pandas during the early stages of reintroduction training. Our findings indicate that the gut microbiota undergoes an adaptive transition phase in the first 1–3 months after reintroduction. As these pandas establish their core home ranges, their gut microbiota gradually stabilizes. Significant differences were observed in the diversity, structural composition, and functional aspects of the gut microbiota between the initial reintroduction phase and the reintroduction training phase.

Compared to the reintroduction training phase, post-reintroduction have more opportunities to acquire a diverse microbiome through a wider range of activities, seasonal changes, and interactions with conspecifics and sympatric species. These factors contribute to an increase in the microbial richness of their gut microbiota^35,36. Consistent with previous studies on giant pandas(Ailuropoda melanoleuca)^3,20, Striped polecat(ictonyxstriatus)³⁷, deer mice(Peromyscus maniculatus)³⁸, baiji dolphins(Lipotes vexillifer)³⁹, Tasmanian devils(Sarcophilus harrisii)⁴⁰ and non-human primates⁴¹, our study found that the gut microbiota richness of giant pandas significantly increased following reintroduction compared to the captive or reintroduction training phases. Notably, our study observed a significant decline followed by an increase in gut microbiota diversity in the early post-reintroduction period (1–3 months). This pattern, combined with the AVD (average value difference) stability analysis results, indicates that the first three months post-reintroduction are a critical phase for gut microbiota transformation in giant pandas. Microbial diversity can confer colonization resistance through various mechanisms, such as competition for nutrients and space, direct antagonism against toxins and other harmful compounds, and enhancement of host immunity against pathogens⁴². These findings suggest that close monitoring during the first three months post-reintroduction, including increased monitoring frequency and detailed analysis of activity patterns and fecal microbiota, will be beneficial for assessing adaptive development. If necessary, artificial interventions can be implemented to ensure the health of the reintroduced giant pandas.

Changes in the composition of gut microbiota are a crucial indicator of the successful digestive adaptation of giant pandas to wild environments⁷. The genus Clostridium, dominant in the gut of wild giant pandas, possesses exceptional cellulose-degrading capabilities²⁰. Comparative studies of the gut microbiota composition between captive and wild giant pandas reveal that the relative abundance of Clostridium in the gut of captive giant pandas is less than 50%, whereas in wild giant pandas, Clostridium constitutes over 90% of the gut microbiota⁴³. Additionally, the study indicates that an effective increase in the relative abundance of Clostridium is considered a sign of enhanced digestive adaptation in natural settings¹⁷. This serves as a crucial monitoring indicator to assess whether giant pandas have adapted to survive in wild environments^17,24. In our study, the mean relative abundance of Clostridium in the reintroduction training group was 72.63%, which further increased to over 90% post-reintroduction, demonstrating the gut microbiota’s improved adaptation to the wild environment. Conversely, the relative abundance of Streptococcus decreased across all 3 groups as the reintroduction process progressed, ultimately dropping to 0.322% after 3 months post-reintroduction. Streptococcus is a major functional genus in the gut of both captive and reintroduction-trained giant pandas, but it is replaced by other bacteria following reintroduction²⁴. Streptococcus participates in amino acid synthesis pathways (map00300 and map00400) through protein metabolism, aiding giant pandas in adapting to the amino acids found in bamboo⁴⁴. Unlike Clostridium and Escherichia, Streptococcus is highly susceptible to environmental factors such as food type, bamboo parts, climate, and seasonal changes⁴⁵. After reintroduction, giant pandas can access a greater variety of bamboo species in the wild, allowing them to adjust their gut microbiota structure according to the varying proportions of carbohydrates, proteins, and lipids in their diet, thereby achieving optimal nutrient absorption^25,46. During the reintroduction training phase, the gut microbiota of giant pandas showed significant enrichment in carbohydrate and amino acid metabolism pathways, indicating that these were the primary sources of nutrients and energy at this stage. Additionally, similar to the gut microbiota functions observed in captive and wild giant pandas, the gut microbiota of giant pandas 3 months post-reintroduction exhibited significant enrichment in various metabolic pathways, including cofactor and vitamin metabolism, energy metabolism, and lipid metabolism¹⁸. Similar results have been observed in other species such as European red deer (Cervus elaphus) and woolly monkeys (Lagothrix lagothricha), these post-reintroduction studies have shown a significant reduction in carbohydrate metabolism pathways within their gut microbiota^47–49. Conversely, there have been notable enhancements in the functions related to other amino acids, energy, cofactors and vitamins, and lipid metabolism to varying degrees^47–49. These findings collectively suggest that reintroduction into natural habitats induces substantial shifts in the gut microbiota functional pathways, potentially reflecting adaptive physiological changes necessary for survival in the wild. We hypothesize that as the reintroduction process continues, the gut microbiota of giant pandas undergoes adaptive changes in structure and metabolic function in response to the complex and variable wild environment. This adaptation allows them to obtain the necessary nutrients and energy from a diverse range of bamboo species or other unknown food sources.

It is noteworthy that the relative abundance of Terrisporobacter and Leuconostoc in the gut microbiota of giant pandas during the reintroduction training phase was below 0.35%. However, as the reintroduction process progressed, these genera gradually increased and became significantly enriched in the gut microbiota of giant pandas during the stabilization phase, reaching over 2.42% and becoming dominant genera. Terrisporobacter, a member of the phylum Actinobacteria, is widely found in soil, plants, and water in natural environments⁵⁰. This genus has excellent acetate-producing capabilities, and its metabolic products can serve as a primary energy source for intestinal mucosal cells, reduce the incidence of colonic and allergic inflammation, enhance gastrointestinal digestive function, and improve nutrient absorption^50,51. Studies have shown that a significant increase in the relative abundance of Terrisporobacter can help compensate for the lack of chitin and carbohydrate-degrading enzymes, ensuring the survival of corsac foxes (Vulpes corsac) in harsh environments⁵². Leuconostoc, an important component of the order Lactobacillales, possesses strong acid-producing, antioxidant, and pathogen-antagonizing abilities⁵³. Leuconostoc mesenteroides metabolizes linoleic acid via an intracellular cyclophilin A-dependent pathway to produce energy and maintains high butyrate production by enhancing lipid metabolic pathways⁵⁴. Previous research has demonstrated that feeding Caenorhabditis elegans with Leuconostoc-enriched food significantly improves stress resistance and lifespan⁵⁵. Additionally, supplementing feed with Leuconostoc can competitively inhibit the damage caused by 8 pathogenic bacteria, including Streptococcus thermophilus, to the intestines of loaches (Misgurnus anguillicaudatus), enhance erythrocyte phagocytosis, boost host resistance to pathogens, improve overall health, and increase juvenile survival rates⁵⁶. Studies have also reported that adding Leuconostoc to adjust gut microbiota can enhance the survival abilities of South American white shrimp (Penaeus vannamei), turbot (Scophthalmus maximus L), and rainbow trout (Oncorhynchus mykiss), although the antibacterial regulatory mechanisms remain unclear^57–59. We speculated that the relatively low-abundance probiotics such as Terrisporobacter and Leuconostoc play a significant role in the gut adaptation of giant panda post-reintroduction.

This study compares the composition and function of gut microbiota in giant pandas during the reintroduction training phase and post-reintroduction, demonstrating the adaptive changes and patterns of gut microbiota following reintroduction. It highlights the importance of the first 3 months post-reintroduction as a critical transition period. These findings provide effective monitoring targets and intervention strategies to ensure the health of reintroduced individuals.

This study has several limitations: (1) The small sample size (27 samples from 3 individuals) limits the statistical power and generalizability of our findings. Although the sample size aligns with common practices in endangered species research^3,7,18,20, it may reduce the ability to detect subtle effects (e.g., minor differences in microbiota composition between groups) and increase the risk of Type II errors. Additionally, small samples may amplify the influence of individual variability or outliers. Future studies should prioritize expanding the number of sampled individuals and ensuring balanced group sizes to validate our observations. (2) Seasonal effects were mitigated by year-round sampling for the TP group, but the frequency of sample collection throughout the year still needs to be increased. (3) Functional analyses were predictive; while PICRUSt2 used in this study has good functional prediction capabilities, actual functions need further quantitative validation using metatranscriptomics or metabolomics. In future research, we suggest (1) increasing the number of sample donors and sampling frequency while investigating the changes in habitat, climate, season, and activity intensity to further screen crucial factors affecting the field adaptation of reintroduced giant pandas. (2) Collecting synchronized fecal and blood samples from donors for metabolomic and transcriptomic analyses, utilizing multi-omics approaches to reveal the adaptive mechanisms of gut microbiota during environmental transitions. (3) Conducting in-depth studies on the functional metabolic impacts of probiotics and employing interventions such as fecal microbiota transplantation to further enhance gut adaptability and the success rate of reintroduced giant pandas.

Conclusions

This study closely monitored the adaptive changes in the gut microbiota of giant pandas during the early stages of reintroduction training. As released pandas establish their core home ranges, their gut microbiota gradually stabilizes. After being released, the giant panda can enhance its overall metabolic capabilities, including cofactors and vitamins, amino acids, lipids, and energy metabolism, by increasing the relative abundance of Clostridium in its gut. This adaptation supports more diverse energy acquisition strategies suited to the wild environment. Our finding highlights that the first 3 months post-release are a critical exploration period for digestive adaptation of giant panda to the wild environment, which will help guide the implementation of future reintroduction efforts.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

R. M., Y. X., R. H., X. G., H. G., Q. Z. and D. Q. designed the study and interpreted the data. R. M., Y. X., W. B., J. L., Z. L., W. W., P. L., H. H., M. Z. and H. Y. collected samples. R. M. finished the lab work. R. M. and Y. X. carried out analyses. R. M. and Y. X. wrote the paper. All of the authors contributed to the discussion.All authors reviewed the manuscript.

Funding

Funding was provided and supported by National Natural Science Foundation of China (No. U21A20193; 32400405), Sichuan Science and Technology Program (No. 2023NSFSC1156; 2024NSFSC0023), Chengdu Science and Technology Program (No. 2023-YF09-00017-SN) and Chengdu Giant Panda Breeding Research Foundation (2024CPB-A23; 2024CPB-Y05; CAZG2025C13; 2024CPB-B06).

Data availability

Data availabilityFull-length raw sequences of the V1–V9 hypervariable region of the 16 S rRNA gene for all samples, in FASTQ file format, are available under BioProject accession number PRJNA1156932.

Declarations

Competing interests

The authors declare no competing interests.