Synergistic pectin deconstruction is a prerequisite for mutualistic interactions between honeybee gut bacteria

Abstract

The honeybee gut microbiome is crucial for degrading diverse pollen glycans. Yet it is unclear how this process shapes the interactions among bacteria. Here, we demonstrate a conditional mutualistic interaction between strains of two honeybee gut bacteria Bifidobacterium asteroides and Gilliamella apicola. When co-occurring in vitro and in vivo, Bifidobacterium provides complementary demethylation service to promote Gilliamella growth on methylated homogalacturonan, an enriched polysaccharide of pectin. In exchange, Gilliamella shares digestive products with Bifidobacterium, through which a positive interaction is established. This positive interaction vanishes when Bifidobacterium is not required on a non-methylated diet. Results from biochemical and gene expression analyses combined with model simulation further suggest that the ratio change of the two major homogalacturonan breakdown products, galacturonic acid (GalA) and di-GalA, determines the bacterial interaction. This study unravels how glycan metabolism may shape the interactions between honeybee gut bacteria.

Introduction

Honeybees rely on a specialized diet consisting primarily of pollen and nectar. However, the glycan constitutions derived from the pollen cell wall, including cellulose, hemicellulose, and pectin, are indigestible by the honeybee host. Alternatively, honeybees possess a simple yet specialized gut microbiome, which plays a significant role in degrading dietary glycans^1,2. Corresponding to polysaccharide varieties associated with pollen diet, members of the honeybee gut bacterial community demonstrate diverse functions in polysaccharide degradation. Genes responsible for these functions are heterologously distributed across bacterial members of the honeybee gut microbiome. These differentiated gene contents have led to the hypothesis of labor division among bacteria, with varied strains specialized in particular sets of glycan substances, which have been examined using comparative genomics, transcriptome assays, and cellular biochemical experiments^1,3,4,5. On the other hand, given the diversity of the pollen diet consumed by the honeybees, complex polysaccharides can result in varied interactions between different bacterial members, as suggested in other model systems, such as humans: direct competition for nutrients or alternatively, cooperation through cross-feeding behaviors^6,7,8. These interactions subsequently influence the microbial community in terms of its size, diversity, and composition^9,10,11,12. However, the specific interactions between different honeybee gut bacteria in relation to the utilization of pollen-derived glycan structures have not been characterized.

The pollen pectin is composed of interwoven polysaccharides that contain primarily galacturonic acid (GalA). The α−1,4-linked D-GalA makes up homogalacturonan, the pectin backbone structure, representing up to 65% of the total mass¹³. In addition, the highly repetitive homogalacturonan is typically characterized by acetyl (at the O-2 and O-3 positions) and methyl (at the C-6 position) decorations¹⁴, with methylation being the most common, reaching up to 90% of the GalA residues¹³.

Of the core honeybee gut microbiome (in which 5–8 bacterial genera account for >95% of the total community), some species of Gilliamella are specialized to break down the pectic backbone^1,3. These Gilliamella possess all enzymes involved in homogalacturonan degradation, including pectate lyases of family 1 (PL1) that cleave the α−1,4-glycosidic linkages of polygalacturonate through a β-elimination mechanism, which generates oligogalacturonides with unsaturation at the C4-C5 position of the nonreducing end. However, the activity of PL enzymes is typically hindered by methylation of the homogalacturonan backbone¹⁵, with only a few exceptions from two Paenibacillus amylolyticus strains¹⁶. Alternatively, the pectin methylesterase, which belongs to the carbohydrate esterase family CE8, is responsible for removing methyl-ester groups from the C-6 positions of D-galactosiduronic acid units in pectic compounds. This de-esterification subsequently enables pectinases (PLs) to break down the highly methyl esterified homogalacturonan in plant cell walls¹⁷. Strikingly, the de-esterification function has not been reported from any Gilliamella. Instead, two lineages of honeybee-derived Bifidobacterium encode the CE8 gene¹⁸. Bifidobacterium is another main glycan degrader among the honeybee gut bacteria, targeting the hemicellulose^1,19. We hypothesize that the two core bacteria cooperate to degrade methylated pectin, the dominant structural component of the pollen cell wall. The coordinated homogalacturonan breakdown mediated by genes from different bacteria (e.g., PL from Gilliamella and CE8 from Bifidobacterium) could have been involved in shaping the interactions between bacteria.

Here, we show that homogalacturonan methylation significantly hinders the pectate lyase activity of PL1 in Gilliamella. The suppressed function is released in the presence of Bifidobacterium due to its role in de-methylation through the CE8 enzyme. When Gilliamella and Bifidobacterium are co-cultivated in vitro, the presence of methylation on the homogalacturonan backbone is the prerequisite for the development of a cross-feeding relationship between the two bacteria. However, when demethylated homogalacturonan is supplemented, Bifidobacterium no longer gains rewards from Gilliamella through homogalacturonan decomposition and the synergistic interaction breaks down. The change in interaction between the two bacteria is determined by changes in carbon sources released during the breakdown of homogalacturonan, through which the expression of genes involved in utilizing these shared carbon sources is modulated. A simple consumer-resource model further indicates that both the dynamics of homogalacturonan degradation and the utilization of liberated products by the two bacteria affect the observed interaction transitions. Finally, in vivo experiments confirm the interaction between strains of Gilliamella and Bifidobacterium in the honeybee gut. This study unravels the variability of interactions among microbiota members under different dietary conditions and highlights the importance of acknowledging this dynamic nature in fully understanding the ecology of honeybee gut communities.

Results

The coordinated pectate lysis activity of Gilliamella was impeded by homogalacturonan methylation

We examined 189 publicly available Gilliamella genomes associated with Apis mellifera, A. cerana, and bumblebees, for carbohydrate-active enzyme (CAZyme) genes that are responsible for pectate deconstruction and utilization. The results revealed that Gilliamella is a primary degrader of homogalacturonan, the most abundant pectic glycan consisting of a linear α−1,4 linked galacturonic acid (GalA)²⁰. Metabolism of homogalacturonan by Gilliamella is orchestrated by two CAZyme gene clusters (CGCs): CGC-1 is involved in depolymerizing the primary pectin backbone structure homogalacturonan, and CGC-2 is responsible for the utilization of liberated oligogalacturonides (Fig. 1A). The two CGCs are heterologously distributed across Gilliamella species, but relatively conserved within lineages (Supplementary Fig. 1A and B). Particularly, the CGC-1 is constrained to three Gilliamella species derived from A. mellifera (including G. apicola) and A. cerana (Supplementary Fig. 1B). The CGC-1 consists of three pectate lyases (PL): PL1, PL9, and PL22. PL1 is predicted as a pectate lyase, primarily functioning extracellularly to deconstruct the highly polymerized homogalacturonan³. PL9 is predicted as a periplasmic exo-polygalacturonate lyase, releasing unsaturated digalacturonates from the reducing end of homogalacturonan²¹. Both PL1 and PL9 contain an N-terminal Sec/SPI signal peptide (SP) (Fig. 1B), which enables transmembrane transportation of enzymes. PL22 is a cytoplasmic oligogalacturonate lyase, with a strong substrate preference for digalacturonate²². The consistent formation of the CGCs thus suggests a coordinated homogalacturonan degradation process: while the homogalacturonan is depolymerized extracellularly by PL1, the resultant oligosaccharides are transported into bacterial cells, where degradation is performed by PL9 in the periplasm and by PL22 intracellularly (Fig. 1C). Although CGC-1 contains a pectin acetylesterase from the carbohydrate esterase family 12 (CE12), which cleaves acetyl groups from the homogalacturonan backbone²³, it lacks the methylesterase of the carbohydrate esterase family 8 (CE8).

Fig. 1: Gilliamella showed limited ability to degrade pectin in the presence of methyl substituents on the homogalacturonan backbone.

A The genetic structures of the two primary CAZyme gene clusters (CGCs) involved in the degradation of homogalacturonan in Gillamella apicola. The direction and length of arrows indicate the orientation and size of genes. B The modular architecture of the three primary pectin lyases (PLs) encoded by G. apicola. The pink color represents the catalytic domain of the enzyme, whereas the green color indicates the signal peptide (SP). C The schematic illustrates the degradation and utilization of homogalacturonan by Gilliamella. The three PLs are spatially and hierarchically orchestrated in the pectin degradation process. PL1 acts extracellularly targeting for large molecular homogalacturonan to produce unsaturated oligogalacturonides. PL9 in the periplasm subsequently cleavages the oligosaccharides into unsaturated disaccharide galacturonic acids (di-GalA). PL22 further degrades di-GalA into one unsaturated and one saturated GalA. CE12 is a pectin acetylesterase, cleaving acetyl groups from the homogalacturonan backbone. D The growth of Gilliamella on either de-methylated polygalacturonic acid (PGA) or citrus pectin (PC). The shadow indicates the 95% confidence interval (CI). PGA or PC was supplemented into the glucose-deleted BHI medium for a final concentration of 2 mg/mL. Ga Gilliamella, Mo monoculture. E The growth rate of Gilliamella on the PGA or PC substrate. (***p = 0.0006, n = 3 cultures in each group, two-sided Student’s t test). F The specific activity of the purified PL1 enzyme on either the PGA or the PC substrate. The bars indicate the mean, whereas the error bars represent SD. Data were analyzed using the two-sided Student’s t test. (***p = 0.0001, n = 3 independent enzyme treatments in each group). Source data are provided as a Source Data file.

The catalytic efficiency of pectate lyases can be affected by the level of methylation of the homogalacturonan backbone¹⁵. To test whether homogalacturonan demethylation was essential for fulfilling PL functions in Gilliamella, one representative strain carrying the CGC-1 (strain W8127 of G. apicola) was cultured on two homogalacturonan substrates characterized by varied levels of methylation: citrus pectin with a high level of methylation (PC, >55% methyl esterified) or de-methylated polygalacturonic acid (PGA, <5% methyl esterified). The results showed that both the bacterial yield and growth rate of W8127 were substantially decreased on the heavily methylated PC substrate (Fig. 1D and E, Student’s t-test, p < 0.001). We further validated whether the growth reduction was attributed to the impeded activity of PL1 on PC deconstruction. The PL1 enzyme of W8127 was heterologously expressed and purified (Supplementary Fig. 2A and B) to measure enzyme activities on PC and PGA substrates (Supplementary Fig. 2E–H). The enzymatic assay showed that the initial reaction rate and specific cleavage activity of PL1 were both significantly decreased on the PC substrate compared to PGA (Fig. 1F, Student’s t-test, p < 0.001, Supplementary Fig. 2C and D), which explains the impeded growth of W8127 on PC. Strikingly, despite the necessary involvement of methylesterase in pectate lysis, no methylesterase was identified in any of the 189 examined Gilliamella genomes, in congruence with a previous study¹ (Supplementary Fig. 1B). Therefore, complete homogalacturonan degradation cannot be performed by Gilliamella alone.

Pectin methylesterase is only encoded by Bifidobacterium

Among the core bacterial taxa of the bee gut microbiome, the carbohydrate esterase family 8 (CE8) methylesterases have so far only been reported from Bifidobacterium¹⁸. We conducted a thorough profiling of carbohydrate-active enzymes (CAZymes) among all publicly available genomes of the gut bacteria associated with honeybees, bumble bees, and carpenter bees. Our findings show that the encoding of CE8 is indeed confined to the two previously documented Bifidobacterium lineages¹⁸, respectively derived from A. mellifera (AmC4, representing B. asteroids) and A. cerana (AcC1, Fig. 2A). In both lineages, the CE8-coding sequence and a GH28 (α-d-galacturonidase, hydrolyzing the saturated oligomers of homogalacturonan to generate GalA) domain together constitute a multi-domain pectin methylesterase gene, with a signal peptide located at the N terminal (Fig. 2B). Structure modeling of the CE8 domain using Alphafold validated a conserved structure, resembling that of the pectin methylesterase characterized from the plant pathogenic bacterium Dickeya dadantii 3379 and the carrot Daucus carota, bearing all preserved catalytic amino acid sites (Supplementary Fig. 4A and B).

Fig. 2: The distribution, genetic structure, and catalytic function of the pectin methylesterase (carbohydrate esterase family 8, CE8) gene found in Bifidobacterium.

A The phylogeny of Bifidobacterium species derived from Apis mellifera and Apis cerana, and the prevalence of genes involved in the degradation and utilization of homogalacturonan among these species. Bifidobacterium strains associated with bumble bees and carpenter bees were used as outgroups. The tree was inferred using the concatenation of the single copy core genes. The names of the bacterial species are labeled to the right of the branches. The CE8 coding gene is limited to two clusters of Bifidobacterium, which originated from A. cerana and A. mellifera, respectively (AcC1 and AmC4, marked in bold). GH28 encodes a putative galacturonidase, which hydrolyzes the saturated oligomers of homogalacturonan to generate GalA. GH105 encodes a putative unsaturated galacturonidase. B The multi-domain pectin methylesterase (PME) gene possesses a modular architecture consisting of a GH28 domain, CE8, and a signal peptide located at the N-terminal. C The genetic architecture of the CGCs in Bifidobacterium, which are involved in the utilization of oligogalacturonides. D The specific activity of purified CE8 on either the PGA or PC substrates. The bars indicate the mean, whereas the error bars represent SD. Data were analyzed using the two-sided Student’s t test. (***p = 0.0001, n = 3 independent enzyme treatments in each group). Source data are provided as a Source Data file.

Flanking the pectin methylesterase gene are genes related to the utilization of the deconstruction products of homogalacturonan, forming a CAZyme gene cluster (CGC), which contains an ABC sugar transporter (ycjNOP)^24,25 and genes for utilizing saturated galacturonic acid (GalA) residues (uxaCBA, kgdK, eda) (CGC1, Fig. 2C). In addition, a CGC2 coexists with CGC1 in Bifidobacterium, and carries genes related to di-GalA transport (aguGFE), di-GalA digestion (yteR of GH105 family, an unsaturated galacturonidase) and unsaturated GalA utilization (kduI and kduD). These genomic features imply that Bifidobacterium can utilize liberated mono-GalA or di-GalA produced from homogalacturonan degradation, suggesting potential nutritional competition with Gilliamella.

To verify the methylesterase activity of the CE8 found in bee-associated Bifidobacterium, we heterologously expressed and purified the CE8 domain of the pectin methylesterase gene from one strain of B. asteroids (W8118) (Supplementary Fig. 5A and B). The activity of the purified CE8 was assayed on a house-made methylated mono-galacturonate (see Methods for details) and also PC substrates. The enzyme activity was evaluated by measuring demethylated GalA generation using mass spectrometry. Our results validated the methylesterase activity of CE8 on the two substrates of varied polymerization grades (Fig. 2D, Supplementary Fig. 5C–F, Supplementary Fig. 3D). Conversely, no demethylated GalA was detected in reactions catalyzed by the purified Gilliamella PL1, or by the purified GH28 domain co-situated with CE8 (Supplementary Fig. 3D), indicating that the methylesterase activity is constrained to Bifidobacterium CE8.

CE8 aids PL1 in pectin deconstruction and facilitates Gilliamella growth

We then asked whether Bifidobacterium CE8 could cooperate with Gilliamella PL1 on pectin deconstruction by removing the methyl group, which eventually facilitated the growth of Gilliamella on pectin. To this end, we first examined whether CE8 could improve the deconstruction efficacy of PL1 on the PC substrate. The results revealed a much reduced PL1 digestion efficacy on PC than on PGA, when applied alone (Fig. 3A vs. B). CE8 addition had little impact on PGA degradation (Fig. 3A), but apparently improved PC deconstruction (Fig. 3B). The thin layer chromatography showed that the dominant products liberated from PGA or PC deconstruction by PL1 were mono-GalA and di-GalA (Supplementary Fig. 6A). The supplementation of CE8 to the PL1 catalyzing PC reaction resulted in stronger degradation product bands (Supplementary Fig. 6A). The mass spectrograph further identified the released products as mono-GalA, di-GalA, and oligo-GalA, all with a Δ4,5-unsaturated residue at the reducing end (Supplementary Fig. 6B). These results indicated that PL1 cleavage had generated predominantly unsaturated galacturonates from the nonreducing end of PGA, which explained the slow migration of these degradation products on the thin layer chromatography.

Fig. 3: CE8 promoted the performance of PL1 in citrus pectin (PC) degradation, leading to enhanced growth of Gilliamella on the PC substrate.

A The cleavage of PGA by PL1 was not influenced by the presence of CE8. The amount of unsaturated sugar generated by the cleavage of homogalacturonan by PL was measured by the absorbance at 232 nm. PGA: polygalacturonic acid. B The presence of CE8 enhanced the cleavage of PC by PL1. C CE8 treatment elevated the Gilliamella growth on the PC but not the PGA substrates. Ga Gilliamella, Mo monoculture. The shadows indicate the 95% confidence interval (CI). D The oligo-GalA products released to the suspension of the Gilliamella culture on either PGA or PC substrates were analyzed using high-performance anion-exchange chromatography equipped with pulsed amperometric detection. E The titer of the released GalA or di-GalA was determined by calculating the peak area under the HPAEC-PAD detection curve in (D). Multiple condition comparisons were analyzed by two-way ANOVA (p = 0.0001) with Tukey’s test (n = 3 independent culture in each group). The letters above each bar indicate statistical differences among the treatments (p < 0.05 significance level). The bars indicate the mean, whereas the error bars represent SD. Source data are provided as a Source Data file.

We then tested whether CE8 treatment could promote the growth of Gilliamella (strain W8127) on a PC substrate. All substrates were adjusted to 2 mg/mL, as Gilliamella displayed optimal growth on this concentration (both for PGA and GalA, Supplementary Fig. 7A and E). Consistently, CE8-conditioned PC resulted in increased Gilliamella growth, which was comparable to that on the PGA substrate (Fig. 3C). The products released during the logarithmic growth phase of Gilliamella on both the PC and PGA substrates were measured by the high-performance anion exchange chromatography with pulsed amperometric detection (HPAEC-PAD). Results showed that substantially less mono-GalA was liberated from the Gilliamella culture on the PC substrate compared to that on the PGA substrate (Fig. 3D and E). Combined, these results suggested that the defective growth of Gilliamella on the PC substrate was primarily attributed to the impeded PL1 activity due to the methylation of the homogalacturonan backbone.

PL1 provides nutrients to Bifidobacterium and promotes its growth on PGA

Our monoculture experiments showed that Bifidobacterium alone could not grow on either PC or PGA (Supplementary Fig. 7B and D), with no simple GalA saccharides released from either substrate (Supplementary Fig. 8D and E). This is in line with the finding that Bifidobacterium lacks depolymerization enzymes (Supplementary Fig. 3A and C). In contrast, Bifidobacterium displayed significantly improved growth on the PL1-treated (with or without CE8 addition) PGA or PC substrate (Supplementary Fig. 8A–C). Both saturated and unsaturated GalA oligosaccharides are potentially accessible to Bifidobacterium (Supplementary Fig. 3B). However, the growth of Bifidobacterium was restricted when cultivated on saturated GalA or di-GalA substrates (Supplementary Fig. 7F and H). This result indicates that Bifidobacterium favors unsaturated GalA oligosaccharides for its growth, which are dominant in the PGA degradation products released by PL1 (Supplementary Fig. 6B).

A reciprocal interaction between Gilliamella and Bifidobacterium is established based on synergistic pectin utilization

Given that the key genes encoded by Gilliamella and Bifidobacterium form a synergistic relationship in the degradation of the backbone structures of pectin, we further tested whether a mutualistic relationship could also be confirmed between bacteria. Representative strains of Bifidobacterium and Gilliamella were cocultured on either PGA or PC substrate, and the growth was compared to that of each monoculture. Results showed that the growth of both bacteria on PC was promoted in coculture (Fig. 4A–C)²⁶. The dependent growth of Bifidobacterium on PC indicated it had utilized PC degradation products produced by Gilliamella (Fig. 4B). Similarly, the growth of Gilliamella was also improved through PC demethylation enabled by Bifidobacterium (Fig. 4B). Thus, through sharing nutrients that could otherwise not be produced in sufficient amounts in monocultures, a reciprocal interaction was established between the two bacteria (Fig. 4C). Interestingly, Gilliamella showed a selfish utilization of all degradation products when cocultured on the same amount of PGA, despite the presence of Bifidobacterium (Fig. 4D–F). Thus, the positive interactions between Gilliamella and Bifidobacterium disappeared when the demethylation service was no longer required by Gilliamella (Fig. 4F).

Fig. 4: Reciprocal interactions between Gilliamella and Bifidobacterium were observed on the PC substrate, while such interactions were not observed on the PGA substrate.

The bacterial growth of either monocultures or coculture of Gilliamella and Bifidobacterium on the PC (A) or PGA (D) substrate was measured by absorbance. Ga Gilliamella, Bi Bifidobacterium, Mo monoculture, Co coculture of Gilliamella and Bifidobacterium. PGA polygalacturonic acid, PC citrus pectin. B, E The growth of Gilliamella and Bifidobacterium in the monoculture or coculture was measured using qPCR with the species-specific primers. The shadows indicate the 95% confidence interval (CI). C, F bottom The area under the growth curve (AUC) for PC (C) in (B) and PGA (F) in (E). Two-sided Student’s t test (ns, p ≥ 0.05, *p < 0.05, **p < 0.01, ***p < 0.001, n = 6 cultures in each group). C, F top The pairwise interactions between Gilliamella and Bifidobacterium were determined by calculating AUC for both PC and PGA substrates. Positive/negative interactions indicate that the species at the end of an arrow grew significantly better/worse in the presence of the species at the beginning of the arrow. Arrow thickness represents interaction strength. Orange circle represents Gilliamella. Blue circle represents Bifidobacterium. G Schematic illustration of bee inoculation and the sampling for bacterial abundance assay. Gilliamella and Bifidobacterium were mono- or co-inoculated into newly emerged bees. The bees were then reared under defined diets: sucrose syrup (SUC) and sucrose syrup diet supplemented with either PGA or PC. The bacterial loads were quantified on day 3 and day 7. H–K Bacterial loads of Gilliamella and Bifidobacterium under different dietary conditions. Bacterial loads of Bifidobacterium in mono-(H) or co-inoculation (I). Bacterial loads of Gilliamella in mono- (J) or co-inoculation (K). The bars indicate the mean with SD. Multiple comparisons were analyzed by generalized linear mixed model (GLMM) with Tukey’s test. (*p < 0.05, ** p < 0.01, ***p < 0.001, n = 24 bees from 3 cages in each group, df = 68). Source data are provided as a Source Data file.

Interaction between Bifidobacterium and Gilliamella strains is also confirmed in bee gut

We then explored whether bacterial interactions mediated by glycan metabolism could affect bacterial abundance in vivo. B. asteroides W8118 and G. apicola W8127 were mono- or co-inoculated (1:1 ratio) to microbiota-depleted newly emerged honeybees. The bees were reared under a sucrose syrup-only diet (10% w/v, used as the control), or supplemented with the same amount of either PGA or PC (2 mg/mL) (Fig. 4G). The abundances of the two bacteria were quantified using qPCR and species-specific primers on day 3 and day 7. The results showed that neither PC nor PGA supplementation changed the abundance of Bifidobacterium in mono-colonization in the gut, mirroring its deficiency in utilizing these polymerized glycans (Fig. 4H, generalized linear mixed model (GLMM) with Tukey’s test). In contrast, the abundance of Bifidobacterium was significantly increased by the supplementation of PC, but not PGA in the presence of Gilliamella (Fig. 4I GLMM with Tukey’s test). This recapitulated the in vitro cultivation results, indicating that Bifidobacterium could gain extra growth only when its methylesterase function was required. Meanwhile, the abundance of Gilliamella exhibited a slight but significant increase by the supplementation of PGA, but not PC, when mono-inoculated (Fig. 4J, GLMM with Tukey’s test), which was in congruence with our genomic and in vitro results. When co-inoculated with Bifidobacterium, Gilliamella gained extra growth on both PGA and PC, as indicated by elevated significance level in gained growth, when compared with the control (i.e., sucrose only) (Fig. 4J vs. K, GLMM with Tukey’s test). In particular, co-cultivation had clearly benefited Gilliamella on PC. The same trends were observed on days 3 and 7, suggesting a stabilized bacterial interaction. Therefore, both in vivo and in vitro results indicate that when provided with methylated polysaccharides, the presence of Gilliamella has led to a colonization benefit for Bifidobacterium via synergistic collaboration in digestion.

The liberated degradation products varied between PC and PGA substrates

The growth of Bifidobacterium relies on the availability of the public goods in the environment (i.e., the net amount of unsaturated GalA and di-GalA available in the supernatant, which represents the difference between the total production by Gilliamella and what is consumed by both bacteria). To understand what had triggered the alteration in bacterial interactions, we measured the constituents of the liberated depolymerization products using HPAEC-PAD. The composition of the public goods varied significantly between PGA and PC conditions. Coculture of Bifidobacterium and Gilliamella on PGA produced predominantly mono-GalA and a small amount of di-GalA (Fig. 5A and B). In contrast, coculture on PC produced substantially decreased mono-GalA but significantly increased di-GalA (Fig. 5A and B). However, HPAEC-PAD failed in measuring potential cross-feeding substrates (primarily unsaturated GalA and di-GalA) from the gut samples, likely due to interferences of complex unknown gut ingredients on the isolation or detection process. Collectively, given that Bifidobacterium can only gain benefits from coculture with Gilliamella on PC, but not PGA, we speculated that the changes in public goods had favored Bifidobacterium growth.

Fig. 5: The deconstruction of PC and PGA produced varied intermediate products, which was in congruent with expression patterns of corresponding genes in Bifidobacterium.

A The oligo-GalA products (predominantly GalA and di-GalA) released during the exponential growth phase of the coculture of Gilliamella and Bifidobacterium on either the PGA or PC substrates were measured using HPAEC-PAD. B The titer of the released GalA or di-GalA was determined by calculating the peak area under the HPAEC-PAD detection curve in panel A. Multiple condition comparisons were analyzed by the two-way ANOVA (p = 0.0014) with Tukey’s test (n = 3 independent cultures in each group). The bars indicate the mean with SD. C Schematic illustration of experiments for gene expression assay. Bifidobacterium was mono- or co-inoculated with Gilliamella into newly emerged bees individually on day 0. Bees were subsequently reared under sucrose syrup diet supplemented with either PGA or PC. The guts were dissected on day 3 for gene expression assay. D The gene expressions were matched with the corresponding reactions in the oligo-GalA degradation process performed by Bifidobacterium. Solid arrows (indigo) represent the reaction catalyzed by Bifidobacterium. Dashed arrows (gray) represent the degradation by Gilliamella. Green and blue boxed panels are the results from in vitro culture and those from the in vivo colonization, respectively. The gene expression was normalized to the mean of the control group, which exclusively received a diet consisting of sucrose syrup (see “Methods”). The PGA or PC was supplemented into 10% sucrose syrup (w/v) for a final concentration of 2 mg/mL. Ga Gilliamella, Bi Bifidobacterium, Mo monoculture or mono-inoculation, Co coculture or co-inoculation of Gilliamella and Bifidobacterium, PGA polygalacturonic acid (red), PC citrus pectin (yellow). Two-way ANOVA with Tukey’s test for multiple comparisons. (for in vitro n = 6 cultures in each group, for in vivo n = 9 bees from 3 cages in each group). The bars indicate the mean, whereas the error bars represent SD. The letters above each bar indicate statistical differences among the treatments (p < 0.05 significance level). Source data are provided as a Source Data file.

A growth assay for Bifidobacterium on the two substrates mixed at a ratio gradient would have been a strong proof of this hypothesis. However, the unsaturated GalA and di-GalA are not commercially available, and the facility and technologies for the isolation of these degradation products from PL1-digested PGA are not accessible to us. Therefore, we opted to look for concerted evidence from gene expressions related to GalA and di-GalA (both saturated and unsaturated) utilization, using qPCR and gene-specific primers for Bifidobacterium, to characterize bacterial usage of these public goods (Fig. 5C). Mirroring its growth pattern in in vitro cocultivation, the Bifidobacterium genes involved in mono-GalA and di-GalA utilization were only significantly upregulated on the PC substrate, but not PGA (Fig. 5D, green boxes). The same bacterial gene expression pattern was also observed in vivo, in bees co-inoculated with the two bacterial strains (Fig. 5D, blue boxes). These results represent further evidence to support that cross-feeding from Gilliamella to Bifidobacterium only occurred when the methylesterase function of Bifidobacterium was required, which might be triggered by variations in public goods.

Understanding the impact of degradation products on bacterial interactions using a consumer-resource model

Based on both in vitro and in vivo results presented above, we propose that the mode of bacterial interaction is determined by the growth kinetics of the two bacteria (G. apicola W8127 and B. asteroides W8118) in response to variations of the two major degradation products (GalA and di-GalA). To test this, we employed the classical MacArthur consumer-resource model (see Methods) to simulate the bacterial growth and the interaction outcomes under different substrates (PGA or PC). Firstly, we examined concentrations of GalA and di-GalA along the simulated bacterial growth. In simulation, PC is postulated as fully methylated and cannot be degraded by Gilliamella unless it is demethylated into PGA by Bifidobacterium. We then introduced a partitioning coefficient ($\eta$) to accommodate varied ratios of GalA to di-GalA present in degradation products (“Methods”). With an appropriate value of $\eta$, the GalA to di-GalA ratio aligned well with our experimental results, where elevated di-GalA was detected under the PC condition (Fig. 6A). These results suggest that the composition of digestive products, in terms of GalA/di-GalA ratio, are determined by the methylation level of the homogalacturonan.

Fig. 6: Understanding the pairwise interaction between Gilliamella and Bifidobacterium under different growth conditions using a consumer-resource model.

A The ratio between GalA and di-GalA at the exponential growth phase, as a function of the partitioning coefficiency ($\eta$) and the ratio of initially supplied PGA. B The ratio of interaction strength of Gilliamella onto Bifidobacterium under different substrates (${\lambda }_{{GB}}={I}_{{GB}}({PC})/{I}_{{GB}}({PGA})$, “Methods”), in relation to the growth kinetics (Monod constant K, and the growth rate γ) of Bifidobacterium on GalA. The higher of the interaction strength ratio, the larger variation in the promotion effect of Gilliamella to Bifidobacterium between PC and PGA conditions. ${\lambda }_{{GB}}$ is negatively correlated with the growth of Bifidobacterium on GalA (peaking at high ${K}_{{BA}}$ and low ${\gamma }_{{BA}}$, depicted by the green point). C ${\lambda }_{{GB}}$ changed with the growth kinetics of Bifidobacterium on di-GalA. The color dots illustrated three representative conditions showing diverse bacterial growth and pairwise interactions, with the yellow best capturing the experiment results. D The growth curve of Bifidobacterium in monoculture (Mo) or in coculture with Gilliamella (Co) in PC and PGA conditions. The kinetic parameters on di-GalA are defined as the colored dots in (C). The pairwise interactions between Gilliamella and Bifidobacterium are calculated in the same way as Fig. 4C and F (“Methods”) and are given under each growth plot. The orientation of the arrow indicates the direction of interaction. The thickness of the arrow represents the strength of the interaction. Ga Gilliamella, Bi Bifidobacterium. E Schematic diagram showing the mechanisms underlying the interaction shift. The orange arrows show the metabolite flow. The blue and green arrows point to the active site of the key enzymes in Gilliamella and Bifidobacterium, respectively.

Next, we explored how the growth kinetics (i.e., the Monod constant $K$, representing the concentration of a resource that supports the half-maximal growth rate of the bacterium, and growth rate $\gamma$) of the two bacteria influenced bacterial interactions. Simulation results showed that adjusting the Monod constant $K$ and growth rate $\gamma$ changed the strength of bacterial interactions, but did not alter the interaction patterns. Gilliamella maintained a positive impact on Bifidobacterium under both PGA and PC conditions, while Bifidobacterium imposed a positive impact on Gilliamella under PC condition, which changed into a negative impact under PGA condition (Supplementary Fig. 9E–H).

We then refined the kinetics parameters for Bifidobacterium until the bacterial interactions approach neutral under the PGA condition, but remain mutualistic under the PC condition (Fig. 4C and F). The ratio of interaction strength (Gilliamella on Bifidobacterium) under PC and PGA conditions was calculated as: ${\lambda }_{{GB}}={I}_{{GB}}\left({PC}\right)/{I}_{{GB}}\left({PGA}\right)$ (See “Methods” for details). An elevated ratio represents larger variations in interaction strength between the two bacteria under PC and PGA conditions. Simulation results showed that a high Monod constant (${K}_{{BA}}$) and a low growth rate (${\gamma }_{{BA}}$) of Bifidobacterium on GalA increased the interaction strength ratio (Gilliamella on Bifidobacterium) between PC and PGA conditions (${\lambda }_{{GB}}$). When the growth kinetics of Bifidobacterium on GalA was set fixed (depicted by the green dot in Fig. 6B), both an increased growth rate (${\gamma }_{{BD}}$) and a lower threshold setting for the Monod constant (${K}_{{BD}}$) on di-GalA produced a high ${\lambda }_{{GB}}$ (depicted by the yellow dot, Fig. 6C and D). These simulation results suggest that di-GalA is favored by Bifidobacterium (a high growth rate with a moderate growth induction concentration), which underpins the observed interaction variation between PC and PGA conditions (Fig. 6E).

PL1 carrying Gilliamella and CE8 carrying Bifidobacterium coexisted in most honeybees

Lastly, an active bacterial cooperation on pectin degradation requires the cooccurrence of CE8-coding Bifidobacterium strains and PL1-coding Gilliamella strains in the same individual honeybee. We screened honeybee gut metagenomic data retrieved from NCBI (78 of A. mellifera and 19 of A. cerana) using MIDAS and a customized bacterial genomic reference database²⁷, to examine whether the expected bacterial strains (belonging to specific species that have a high prevalence of the corresponding enzymes) could be found together in the same bee gut. The results confirm that strains of CE8-coding Bifidobacterium and PL1-coding Gilliamella almost always live in the same bee gut (Supplementary Fig. 10A–C), suggesting that the synergistic pectin digestion enabled by CE8 and PL1 genes is possible in live bees.

Discussion

Even though several studies have unraveled the division of labor in pollen-sourced glycan metabolism among bacterial members in the bee gut^1,3,4, these studies provide no evidence of how glycan metabolism affects the interactions between bacteria. Here, by leveraging the syntrophic deconstruction of the pectic backbone structure (methylated homogalacturonan) by Bifidobacterium and Gilliamella in the honeybee gut, we reveal dynamic interactions between these core bacteria. Particularly, we demonstrate that reciprocal interactions between Gilliamella and Bifidobacterium are established upon demand for collaborative degradation of the glycan substrate. Once the reciprocal dependence collapses, e.g., a single bacterium can complete the digestion, such a cross-feeding relationship disappears as well. In our case, the methylation of glycan plays a key role in determining the interaction between the two bacterial strains. Given the degree of methylation in homogalacturonan varies considerably across different plant sources²⁸, our results indicate that the constitutional features of pollen are probably involved in shaping honeybee bacterial interactions, which explains why dietary variation has served as a key driver underlying gut profile change across honeybee populations²⁹.

A major challenge faced by the study of host-associated microbiota is to understand the ecological dynamics that shape these communities. The periodicity and variability in the supply of dietary fibers can give rise to dynamic regimes of niche availability, which subsequently affects interactions among diet-responsive microbes^{11,12,30,31,32}. Bacterial cross-feeding, where one bacterium uses another organism’s byproducts as a resource³³, is pervasive in microbial communities, playing a vital role in supporting bacterial coexistence and in promoting microbial diversity^34,35. However, substrate provision is generally costly to the producer, although exceptions have been reported³⁶. In principle, natural selection shall favor a bidirectional relationship that is reciprocally beneficial³⁷. Therefore, we speculate that reciprocal interactions should be more common in microbial communities. Nevertheless, there is a general lack of empirical evidence that characterizes the “returning favor” from the cross-feeder in response to substrate provision. This knowledge gap is, at least in part, due to the fact that the onset interaction is often responsive to environmental flux, a pattern sometimes termed “context-dependent interaction”^38,39. In this case, synergistic interactions could only be detected at the “right” condition and timing. For instance, in a pair of marine bacteria, the reciprocal interaction occurs only when the producer reaches growth arrest under stress, after which the cross-feeder begins to perform deacidification service in exchange for nutrient provision from the producer⁴⁰. Regular investigation is typically terminated once a bacterium ceases to grow, in which case, the reciprocal interaction would have been missed. In the present study, we reveal a conditional reciprocal interaction that is determined by digestion needs. The change of interaction mode between Gilliamella and Bifidobacterium would not have been revealed if we did not apply varied polycarbonate substrates (PGA vs. PC). We anticipate that increasing evidence for reciprocal interactions shall be discovered if investigations are extended to cover temporal series and additional nutrient settings.

It remains to be explored whether the novel interaction reported in this study represents a prevalent phenomenon in the honeybees. We note that here only two representative strains (for Bifidobacterium and Gilliamella, respectively) were tested for interaction dynamics. However, evidence suggests that this interaction can also be expected in other bacterial strains that share similar genomic characteristics. For example, comparative genomics revealed that the genes responsible for the focal functions (i.e., methylesterase of CE8 and pectate lyase of PL1), the foundation of the reciprocal interaction, were also found in closely related bacterial strains of the same species, as well as in close related species of the same genera (Supplementary Fig. 1B, Fig. 2A), all reported with a high prevalence in corresponding honeybee hosts (A. mellifera and A. cerana). Moreover, enzymes encoding for these functions share a high sequence similarity across bacterial species (Supplementary Fig. 1A, Supplementary Fig. 3B). Therefore, it’s plausible that similar interactions will hold valid for any pair of Bifidobacterium and Gilliamella strains that encode for corresponding enzymes. Indeed, the expected combination of Bifidobacterium and Gilliamella strains was detected in nearly all honeybee gut metagenomics data examined in the present study. For bacterial species that lack enzymes for glycan digestion, their roles in interactions with other bacterial members remain to be investigated. Given that some of these species/strains are capable of using liberated glycan degradation products (e.g., Gilliamella AmC3, AmC5, and AcC5 encoding for the utilization of saturated GalA, a minority product liberated from PGA deconstruction; Bifidobacterium AmC3 encoding for the utilization of both saturated and unsaturated GalA), we speculate that they may have developed complementary routes to compensate for dietary benefits gained from other bacteria. Nevertheless, limited interaction details have been unraveled for bacterial genera associated with the honeybees, including most of the core bacteria.

Finally, the bacterial interaction revealed here is under confined diet conditions (sucrose, sucrose supplemented with PGA or PC) and is far from being a thorough investigation of the inter-specific relationships under a complex diet (nectar with pollen). However, complex glycans (e.g., arabinan and xylan) derived from pollen that are accessible to Bifidobacterium¹ will inevitably obscure the specific interactions focused in the present study. The results demonstrated here may represent a partial picture of the real honeybee gut system but are a necessary and important step toward understanding the full mechanism behind the complex and likely dynamic interactive network governing the honeybee gut microbiome. This highlights future research priorities that are necessary to deepen our understanding of the interaction networks in the “simple” gut microbial system of the honeybees.

To summarize, our findings demonstrate that the efficient digestion of the primary component of pectin in the honeybee diet relies on collaborative enzymatic activities of two core members of the gut bacterial community, namely Bifidobacterium and Gilliamella. The assistance in pectin digestion is a prerequisite for Bifidobacterium to access the liberated public goods. This study underscored the notion that a comprehensive understanding of the principles governing bacteria is crucial for understanding ecological factors that influence microbial community structures in the honeybee gut.

Methods

Phylogenetic analyses of Bifidobacterium and Gilliamella

Phylogenetic reconstructions were conducted for Bifidobacterium and Gilliamella using publicly available bacterial genomes. A total of 65 Bifidobacterium asteroides and 189 Gilliamella apicola genomes retrieved from the NCBI database (Supplementary Data 1) were first annotated using Prokka (version 1.14.0) with default parameters⁴¹. Core bacterial genes were identified by Roary (version 3.12.0) using the parameter -blastp 75⁴². Nucleic acid sequences of 232 core genes for Bifidobacterium and 80 for Gilliamella were applied to construct a maximum-likelihood phylogeny, using RAxML with a GTR + G + I model and 1000 bootstrap replicates in CIPRES Science Gateway (version V.3.3 http://www.phylo.org/)^43,44. The species names of the taxonomic clusters of B. asteroides and G. apicola were given referring to existing publications. Those without species names were designated by the abbreviation of the host species (Am representing Apis mellifera, Ac representing Apis cerana) with cluster ID (e.g., C1). The phylogenetic tree was visualized using iTOL (version 6.7.4, https://itol.embl.de/)⁴⁵.

Identification of pectin degradation genes from honeybee gut bacterial genomes

Genomes of bacterial isolates (65 Bifidobacterium asteroides and 189 Gilliamella that used above) were utilized to identify genes involved in pectin degradation. Protein sequences of each bacterial strain were run through the dbCAN3 (https://bcb.unl.edu/dbCAN2/) to identify the CAZyme family utilizing HMMER (CAZyme domain HMM database with E-value < 1e–15 and coverage > 0.35) and DIAMOND (CAZy database with E-value < 1e–102)⁴⁶. Then the CAZyme gene clusters (CGCs) were identified using CGCFinder (Distance ≤ 2, signature genes = CAZyme + Transporters). EggNOG-mapper (version 2.1.9, http://eggnog-mapper.embl.de/) was utilized to annotate bacterial genomes against eggNOG database 5.0, as an additional reference for further annotation validation of pectin degradation genes^47,48. The structure of CGCs involved in pectin degradation was illustrated using genoPlotR⁴⁹.

Analysis of PL and CE8 gene structure and construction of CE8 phylogenetic tree

Amino acid sequences of PL1, PL9, and PL22 of Gilliamella apicola W8127 and that of the methylesterase-containing gene of Bifidobacterium asteroides W8118 were screened for signal peptides using SignalP 6.0⁵⁰. The conserved domains of these genes were searched using BLASTP (https://blast.ncbi.nlm.nih.gov/) and dbCAN3 (https://bcb.unl.edu/dbCAN2/) with HMMER (HMM database of CAZyme subfamilies, E-Value < 1e-15, coverage > 0.35). Closely related enzyme sequences of the CE8 domain from Bifidobacterium were also retrieved from the results of BLASTP and CE8 HMM database used in dbCAN (Supplementary Table 2). All sequences were aligned using the MUSCLE alignment in MEGA X⁵¹. After trimming of the overhanging ends, a maximum-likelihood phylogenetic tree was reconstructed using RAxML (GTR + G + I model) with 1000 bootstrap replicates through CIPRES Science Gateway^43,44. Furthermore, the amino acid sequence of CE8 from Bifidobacterium asteroids was aligned with that of two homologs with characterized structure: 2NTB from Dickeya dadantii 3379 and 1GQ8 from Daucus carota using Snapgene (version 6.0.2) based on the MUSCLE method. The structure of the enzymes was aligned using PyMOL (The PyMOL Molecular Graphics System, version 2.5) with the Super method. The 3D model of CE8 of Bifidobacterium was generated using AlphaFold (version 2.3.2)⁵², according to instructions in the AlphaFold Colab notebook (https://colab.research.google.com/github/deepmind/alphafold/blob/main/notebooks/AlphaFold.ipynb#scrollTo=pc5-mbsX9PZC).

The prevalence and coexistence of pectin methylesterase and lyase in individual honeybee guts

To explore the co-occurrence of the two pectin degradation-related enzymes in individual guts, 97 gut metagenomes (78 associated with Apis mellifera and 19 associated with Apis cerana) were retrieved from the NCBI database (Supplementary Data 2) and the presence of the CE8 carrying Bifidobacterium cluster and PL1 carrying Gilliamella clusters were inferred. To this end, the clean reads data were firstly mapped against the reference genome of A. mellifera (GCA_003254395) and A. cerana (GCA_001442555) using BMTagger (version 0.3.102.4, ftp://ftp.ncbi.nlm.nih.gov/pub/agarwala/bmtagger/) to filter out host-derived reads. The strain-level community profiling was then performed using the Metagenomic Intra-Species Diversity Analysis System (MIDAS) pipeline based on a customized honeybee gut microbial genomic database^27,53. The coexistence of bacterial species with PL1 and CE8 functions were plotted using Excel 2020 and GraphPad Prism (version 9.3.1).

In vitro cultivation

The strains Gilliamella apicola W8127 (GaW8127) and Bifidobacterium asteroides W8118 (BiW8118) used in this study were kindly provided by Prof. Hao Zheng of China Agricultural University. GaW8127 and BiW8118 were retrieved from the −80 °C glycerol stocks by streaking on the Brain Heart Infusion agar (BHIA) (Oxoid, UK) supplemented with 5% (vol/vol) defibrinated sheep’s blood (Solarbio, CN). The plates were incubated in an anaerobic chamber with 2% H₂, 5% CO₂, and 93% N₂ at 35 °C. Strains were confirmed by amplification of the 16S rRNA gene using primers 27F (5′-AGAGTTTGATCCTGGCTCAG-3′) and 1492R (5′-GGTTACCTTGTTACGACTT-3′) and Sanger sequencing.

The bacterial cultivation medium was prepared based on glucose-depleted BHI liquid medium (gdBHI, Hopebio, China, containing peptone 10.0 mg/mL, dehydrated calf brain extract 12.5 mg/mL, dehydrated buffalo heart extract 5.0 mg/mL, sodium chloride 5.0 mg/mL and disodium phosphate 2.5 mg/mL). Both bacteria displayed an extremely low growth on this medium when no extra carbon source was supplemented (Supplementary Fig. 7I). The carbon source including polygalacturonic acid (PGA, Sigma-Aldrich, DE), Citrus Pectin (PC, Sigma-Aldrich), galacturonic acid (GalA, Sigma-Aldrich) or digalacturonic acid (di-GalA, Megazyme, IE) was then supplemented into the gdBHI as indicated to a final concentration ranging from 0.5 to 10 mg/mL. GaW8127 and BiW8118 retrieved from BHI agar plates 48 h after cultivation were scraped off and resuspended in PBS to an OD₆₀₀ of ~1. For the monoculture, 2 μL of each bacterial suspension was inoculated into 200 μL of the defined medium on a 96 well plate (Corning, USA). For the co-cultivation, GaW8127 and BiW8118 stains were mixed at a 1:1 (vol/vol) ratio. Then 2 µL of the mixed bacterial culture was inoculated into 200 μL of the defined media. The original cell density determined at OD₆₀₀ by subtracting the blank (culture medium) well was about 0.05. These plates were incubated in 5% CO₂ at 35 °C without shaking. Cell growth determined by OD₆₀₀ was monitored using a BioTek Synergy HTX plate reader every 3 h until 96 h after inoculation. The growth rate of Gilliamella cultivated on PGA or PC substrates was calculated using the Growthrates package of R.

Bacterial growth on enzyme-conditioned media

The PGA or PC at a concentration of 4 mg/mL were conditioned with either CE8 or PL1 at 42 °C and pH 6 for 48 h to reach near completion of the reactions. The enzymes were then denatured by heating at 85 °C for 5 min. The enzyme-conditioned substrates were centrifuged at 12,000 × g for 10 min and the suspension was filtered by a 0.22 μm filter (Millipore, DE). The resulting substrate solutions were mixed with the gdBHI liquid medium at a 1:1 (vol/vol) ratio to obtain a final substrate concentration of 2 mg/mL for bacterial culture. The mediums supplemented with an equal amount of unconditioned PGA or PC were used as control. Then GaW8127 and BiW8118 were cultivated respectively with different substrates in 96 well plates (Corning) at 35 °C and 5% CO₂. The OD₆₀₀ was measured every 3 h for a total of 96 h.

The qPCR quantification of bacterial growth in coculture

The bacterial culture suspension from each well was collected every 6 h, with a total volume of 200 μL. Subsequently, the total DNA was extracted using the FastPure Bacteria DNA Isolation Mini Kit (Vazyme, CN). GaW8127 and BiW8118 bacterial loads were determined by quantitative PCR using species-specific 16S rRNA primers (Supplementary Table 1). The 16S rRNA genes were cloned into the plasmid (pEASY®-T1 Simple Cloning Kit; TransGen, Beijing, China) and the copy number of each plasmid was calculated based on the DNA quantity measured by Qubit 4 (Invitrogen, USA). The plasmid solution was serially diluted and used for the construction of the standard curve. qPCR was performed with the QuantStudio 1 real-time PCR system (Applied Biosystems, USA) in triplicate in a total volume of 20 μL (0.2 μM forward and reverse primers, 1 μL of DNA template, and 1x ChamQ Universal SYBR qPCR Master Mix, Vazyme). The thermal cycling conditions were set as follows: denaturation stage at 50 °C for 2 min followed by 95 °C for 2 min, 40 cycles of denaturation at 95 °C for 15 s, annealing/extension at 60 °C for 1 min.

Interspecific interaction assay

The interactions between GaW8127 and BiW8118 were evaluated based on the definition described in a previous study with some modifications²⁶. Positive or negative interactions were assigned between two bacterial members when the area under the bacterial load growth curve (AUC) of the coculture was greater or smaller than that of each monoculture, respectively. The AUC of GaW8127 and BiW8118 in monoculture and coculture within the defined medium was calculated by GraphPad Prism (version 9.3.1). The Student’s t test was used to assess significant differences. Interaction strength was further quantified by calculating the fold change in AUC and was illustrated graphically by the thickness of the arrows connecting the two strains. The arrow thickness was proportional to log10 (AUC (coculture)/AUC (monoculture)).

Protein expression and purification

Escherichia coli Fast-T1 (Vazyme, CN) was used for enzyme gene cloning, and E. coli BL21 (DE3) (Thermo Fisher Scientific, USA) was used as the host for protein expression. All E. coli strains were cultivated aerobically in Luria Bertoni Broth media (LB, Solarbio) at 37 °C, with 100 μg/mL ampicillin addition when indicated. The CE8-containing gene and PL1 genes were amplified from the genomic DNA template extracted from the cultures of BiW8118 and GaW8127, respectively. The primers targeting these genes are listed in the Supplementary Table 1. The 2x Phanta Flash Master Mix (Vazyme) was used for PCR. The amplified fragments were cloned into pCE2-TA/Blunt-Zero vectors (Vazyme) and the successful constructs were verified by Sanger sequencing (Sangon Biotech, CN). CE8, GH28, and PL1 fragments were amplified from the vector using a specific primer set (Supplementary Table 1), which were then cloned into the pCold TF vector (Takara, Japan) using a ClonExpress Ultra One Step Cloning Kit (Vazyme) and sequenced to ensure correct insertion (Sangon Biotech). The constructed plasmids carrying CE8, GH28, and PL1 coding sequences were transformed into E. coli BL21 (DE3) strains. The transformants were cultivated in 100 mL Terrific Broth (Solarbio) in 500 mL baffled Erlenmeyer flasks at 37 °C with shaking until an OD₆₀₀ of 0.6–0.8. Then it was cooled to 15 °C using ice water and maintained for 30 min. After that, 1 mM isopropylthio-β-galactoside (IPTG; Invitrogen, USA) was added and the culture was incubated with shaking at 15 °C for 24 h. The cells from each culture were harvested by centrifugation and stored at −20 °C until needed. For enzyme isolation, the cell pellet was resuspended in 6 mL of PBS and incubated with lysozyme (20 mg/mL, Vazyme) at 37 °C for 1 h, then disrupted by ultrasonication. The lysate was filtered through a 0.45 μm polyethersulfone (PES) membrane filter (Millipore) after centrifugation at 48,000 × g at 4 °C for 15 min. A GE column with Ni-NTA resin (Sigma-Aldrich, DE) connected to ÄKTA pure (GE Life Sciences, UK) system was used for enzyme isolation. Samples were run through the column at a flow rate of 0.5 mL/min. Then the column was washed by PBS with 10 mM imidazole. The enzyme was eluted with 500 mM imidazole. Samples were dialyzed into 25 mM Tris (pH 7.2) with 25 mM NaCl and 10% glycerol and were frozen at −20 °C. The purity of target proteins was evaluated using SDS-PAGE. TF tags were not cleaved before enzymatic dynamic characterization because of the low solubility of the target protein.

Enzymatic assays

The activity of CE8 was determined using the titration assay as described previously with some modifications⁵⁴. A 25 mL reaction mixture containing 2 mg/mL of PC (Sigma-Aldrich) and 100 μL of enzyme solution was incubated at 42 °C and pH 6.0 for 30 min. The enzymatic activity was measured by titrating the solution with 0.1 M NaOH to neutral. Then the volume of the consumption of NaOH was calculated. All reactions were terminated by heating at 80 °C for 5 min. A reaction mixture with a thermally inactivated enzyme was used as a control. One international unit (IU) of the CE8 activity was defined as the amount of enzyme, which releases 1 µmol of carboxyl groups per minute. The reaction mixture was incubated at different temperatures from 25 °C to 55 °C with 5 °C intervals for 30 min to determine the optimal reaction temperature. Reactions incubated at different pH (2.0–8.0, with a 1.0 interval) were conducted to determine the optimal pH. The acetate solution and the Tris buffer were used for adjusting the pH. The max of CE8 activity was used for normalization. The CE8 enzymatic kinetics was determined by a titration assay. PC concentrations ranging from 0.01 to 5 mg/mL were used for incubating the CE8 enzymes under pH 6.0 at 42 °C for 10 min. Maximum velocity (Vmax), Michaelis–Menten constant (Km), and catalytic rate constant (Kcat) of BiCE8 were calculated.

The PL1 activity was determined by monitoring the formation of unsaturated products from polygalacturonate spectrophotometrically at 232 nm⁵⁵. A total of 200 μL of substrate solution containing 2 mg/mL PGA (Sigma-Aldrich) or PC (Sigma-Aldrich) was used for PL1 enzymatic assay with the addition of 1 mM CaCl₂. The reaction solution was incubated in a 96-well UV-Transparent plate (Corning) at pH 6.0 and 42 °C for 1 min. Then the absorbance at 232 nm was measured immediately using a BioTek Synergy HTX spectrophotometer. Enzyme activity was calculated according to the increased absorbance. One international unit (IU) of PL1 activity was defined as the amount of enzymatic releasing of 1 µmol of unsaturated galacturonate in 1 min. The 4600 molar extinction coefficient was used for the transition from the absorbance to the unsaturated product concentration⁵⁵. Optimal temperature and pH were detected in different substrate solutions containing 2 mg/mL PC or PGA. A pH gradient from 2.0 to 10.0 with a 1.0 interval and a temperature gradient from 25 to 65 °C with a 5 °C interval was set up. The absorbances were determined after incubating with the enzyme for 1 min. The activity of PL1 at a range of substrate concentrations from 0.125 to 2.5 mg/mL. Reactions were performed at 42 °C with 1 mM CaCl₂ and pH 6.0 and were monitored for 1 min. Enzyme concentrations in the reaction mixtures were 5 μM. Nonlinear least-squares minimization fitting to the Michaelis-Menten model was used to calculate the kinetic parameters. Each assay was done for triplicates and the results were analyzed using GraphPad Prism (version 9.3.1).

The synergistic interactions of CE8 and PL1 were measured by combining them into 200 μL reaction solutions containing 2 mg/mL of either PC or PGA as the substrate. The reactions were carried out at pH 6.0 and 42 °C. The absorbances at 232 nm were monitored for releasing of unsaturated hydrolyzed products per hour. Reactions were terminated by heating at 85 °C for 5 min after 48 h. Then the supernatants were extracted for the liberated product analyses. All reactions were performed for triplicates.

Degradation product analyses

Thin-layer chromatography analysis

Reaction and cultivation suspensions centrifuged at 12,000 × g for 10 min. The supernatants were filtered through a 0.22 μm membrane (Millipore). Then 1 μL of the filtered solutions was spotted on one end of the silica gel 60 aluminum sheets (15 × 10 cm; Merck, DE). Chromatography was performed using the ascending method with the following mobile phase: ethyl acetate: acetic acid: formic acid: water (9: 3: 1: 4) as the solvent system⁵⁶. For visualization, the dried spot plates were sprayed with 0.2% (w/v) orcinol dissolved in 10% sulfuric acid: methanol (1: 9), followed by heating at 105 °C for 5 min⁵⁷.

Methylated galacturonate synthesis

The synthesis of methyl galacturonate followed a published method^58,59, in which D-(+)-galacturonic acid (1 g, 51.55 mmol) was refluxed with methanol (10 mL) and concentrated sulfuric acid (0.5 mL) under nitrogen protection. After a reaction time of 2 h, the progress of the reaction was monitored by TLC using a solvent system consisting of DCM and MeOH at a ratio of 15:1. Upon completion of the raw material reaction, the resulting product was concentrated via rotary evaporation and subsequently purified using silica gel column chromatography. The final product was confirmed by mass spectrometry.

Liquid chromatograph-mass spectrometer analyses

The enzymatic breakdown products of PL1 or CE8 were acidified with formic acid before LC-MS analysis (Ultimate 3000 UPLC/ISQ EC, Thermo Fisher Scientific). A volume of 20 μL of the supernatant solution was injected into the C18 column (EC-C18, 2.7 μm, 4.6 × 100 mm, Agilent Poroshell, USA) with a matching guard column and step-gradient elution using solvent solutions A and B (0% B for 1.5 min, then 0–100% over 7.5 min, constant at 100% until 10 min; A: water + 1% acetonitrile + 0.1% formic acid, B: acetonitrile + 2% water + 0.1% formic acid; flow-rate 1.2 mL/min; temperature 30 °C). Data analyses were performed using Chromeleon 7 software (Thermo Fisher Scientific).

High-performance anion-exchange chromatography with pulsed amperometric detection (HPAEC-PAD) analyses

Analyses of mono- and oligo-galacturonic acids were carried out following Versari et al. ⁶⁰ using an ICS-5000 Plus Ion Chromatography system (Thermo Fisher Scientific). Recommended parameters were used for the PAD setting (Carbohydrates, Quad Potential, Ag/AgCl Ref). The column was a CarboPac PA-100 (Dionex, USA) coupled to a guard column of the same material (250 × 4 and 50 × 4 mm, respectively). All determinations were carried out at 25 °C with a flow rate of 1 mL/ min. A total of 15 μL culture medium was first filtered by a 0.22 μm filter (Millipore) and then injected by an autosampler (Thermo Fisher Scientific). Analyses were done using H₂O as eluent A, sodium hydroxide 0.5 M as eluent B, and sodium acetate 1.5 M as eluent C. Solvents were HPLC grade (Thermo Fisher Scientific). The following elution gradient was used [time (min)/B(%)/C(%)]: t1 = 0/20/0.7; t2 = 5/20/0.7; t3 = 35.5/20/67; t4 = 43/20/67; t5 = 48.5/20/0.7; t6 = 52/20/0.7. The elution curves were visualized using GraphPad Prism (version 9.3.1), and the peak area was calculated in OriginPro (OriginLab).

Bee inoculation and bacterial abundance measurement

To explore the influence of glycans on bacterial abundance in vivo, Gilliamella and Bifidobacterium were mono- or co-inoculated into newly emerged honeybees following the method described by Zheng et al. ⁶¹. In brief, late-stage pupae of Apis mellifera were removed manually from brood frames and placed in sterile plastic cup cages. The pupae were then kept in a 35 °C incubation chamber with 70% humidity until eclosion. Newly emerged bees were then separated into cup cages, each with 30 bees. The bacterial strains were retrieved on BHI agar plates, incubated at 35 °C for 48 h, and resuspended in PBS to achieve an OD₆₀₀ of 1. The inoculum for co-colonization was prepared by adding 10 μL bacterial suspension of each GaW8127 and BiW8118 into 980 μL of 10% (w/v) sucrose solution. The 1 mL bacterial suspension was then fed to bees in each cup cage for 24 h. After that bees were reared with defined diets: sterile sucrose syrup (10%, w/v), sterile sucrose syrup supplemented with either PGA (to 2 mg/mL within 10% sucrose, w/v) or PC (to 2 mg/mL within 10% sucrose, w/v). Each treatment had 3-cup cage replicates. Bees were sampled after being reared in defined diets for 3 and 7 days, respectively. A total of 24 bees were sampled for each treatment (8 bees per cup) at each time point, and the hindguts were dissected and stored at −80 °C until DNA extraction. Total DNA of the hindgut was extracted using the CTAB method⁶². The bacterial abundance of Gilliamella or Bifidobacterium was determined using the qPCR method and species-specific primers as described above. A generalized linear mixed-effects model (GLMM) with family Poisson was used to address the effect of glycan supplementation on bacterial colonization, where cage division was considered as a random effect. Tukey’s test was subsequently used for multiple comparisons. The lme4 and emmeans packages of R were used for the statistics.

In vivo and in vitro analyses of gene expressions

To investigate the impact of polysaccharides on bacterial structure changes in vivo, honeybees were inoculated by GaW8127 and BiW8118. Gnotobiotic bees were obtained following Zheng et al. ⁶¹. In brief, late-stage pupae of Apis mellifera were removed manually from brood frames and placed in sterile plastic bins, then kept in a 35 °C incubation chamber. Bacterial strains were cultured on BHI agar plates for 48 h and resuspended in PBS to achieve an OD₆₀₀ of 1. For mono-colonization, the bacterial suspension was mixed with an equal volume of 20% (w/v) sucrose solution. For co-colonization, GaW8127 and BiW8118 were initially mixed at a 1:1 ratio (v/v) and subsequently combined with a 20% (w/v) sucrose solution at 1:1 (v/v). The newly emerged bees were individually fed with 5 μL of the mixed solution. Afterward, a total of 30 bees were placed into one cup cage and were consistently supplied with sucrose syrups (10%, w/v) supplemented with either PGA or PC (final concentration of 2 mg/mL). Each group contained three cups of bees and bees fed with only sucrose syrup (10%, w/v) were used as the control. After being reared for 3 days, nine bees from each group (3 bees from each of the 3 cups) were dissected and their whole guts were stored at −80 °C until RNA extraction.

The expressions of genes related to pectin degradation and utilization in BiW8118 were determined using RT-qPCR. For in vitro cultivation experiments, a subsample of 200 µL was aliquoted from the 2 mL bacterial culture after 24 h. RNA was extracted using an RNeasy Protect Bacteria Mini Kit (Qiagen, DE), and cDNA was synthesized from 1 μg of the RNA sample using the HiScript® III 1st Strand cDNA Synthesis Kit (Vazyme) with random hexamer primers from the kit. The qPCR was performed using the ChamQ Universal SYBR® qPCR Master Mix (Vazyme) on the QuantStudio 1 real-time PCR system (Applied Biosystems, USA) in a standard 96 well block (20 μL reactions; 40 cycles of denaturation at 95 °C for 10 s, annealing/extension and the plate read at 60 °C for 20 s). Relative gene expressions were normalized to the expression of the RecA gene using the 2^−ΔΔCT method⁶³. Then the relative gene expressions were compared to those in the cultivation on the gdBHI medium, from which fold changes were calculated. For the gene expression assay applied on gut samples, RNA was extracted using the TRIzol reagent (Invitrogen, USA). The expressions of bacterial genes from the glycan administration groups were compared with that of the control group. Two or three technical replicates were performed on the same plate for each qPCR reaction. Primers used in this study were listed in Supplementary Table 1.

Model definition

A Monod consumer-resource model was used to explore how the utilization of the public goods affects the interaction outcomes between Bifidobacterium (Bi) and Gilliamella (Ga). In this model, the growth of Bi and Ga are assumed to depend only on the consumption of two primary resources released from PGA degradation: GalA and di-GalA. The growth of the two bacteria can be described as the following functions:

$$\frac{d{x}_{G}}{{dt}}=\left({\gamma }_{{GA}}\cdot \frac{A}{{K}_{{GA}}+A}+{\gamma }_{{GD}}\cdot \frac{D}{{K}_{{GD}}+D}\right){x}_{G}$$

(1)

$$\frac{d{x}_{B}}{{dt}}=\left({\gamma }_{{BA}}\cdot \frac{A}{{K}_{{BA}}+A}+{\gamma }_{{BD}}\cdot \frac{D}{{K}_{{BD}}+D}\right){x}_{B}$$

(2)

Here, ${x}_{G}$ and ${x}_{B}$ represent the biomass of Gilliamella (Ga) and Bifidobacterium (Bi), and $A$ and $D$ are the concentrations of GalA and di-GalA, respectively. The model is parameterized by the growth rate ${\gamma }_{{ij}}$, which stands for the growth rate of species $i$ (G or B) on substrate $j$ (A or D) (nutrient-specific growth rate when the resource concentration is not limited). The ${K}_{{ij}}$ is the concentration of resource j at which species i reaches its half-maximal growth rate on that resource (Monod constant of species $i$ on substrate $j$).

The substrate PGA (in a concentration of $P$) can be degraded by Ga into GalA and di-GalA, the substrate PC (in a concentration of $C$) can be de-methylated by Bi into PGA. PGA and PC are also autonomously hydrolyzed at a very low rate scaled by ${r}_{P},\, {r}_{C}$ (with a value $\ll \, 1$), and the Monod constant ${K}_{P}\, \gg \, P,\,{K}_{C}\, \gg \, C$.

$$\frac{{dP}}{{dt}}=\frac{-P}{{K}_{{GP}}+P}{x}_{G}-{r}_{P}\frac{P}{{K}_{P}+P}+\frac{C}{{K}_{{BC}}+C}{x}_{B}+{r}_{C}\frac{C}{{K}_{C}+C}$$

(3)

$$\frac{{dC}}{{dt}}=\frac{-C}{{K}_{{BC}}+C}{x}_{B}-{r}_{C}\frac{C}{{K}_{C}+C}$$

(4)

The parameters ${K}_{{GP}}$ and ${K}_{{BC}}$ represent for the Monod constant of Ga and Bi in degrading PGA and PC respectively.

The GalA and di-GalA are produced from the degradation of PGA.

$$\frac{{dA}}{{dt}}=\frac{-A}{{K}_{{GA}}+A}{x}_{G}-\frac{A}{{K}_{{BA}}+A}{x}_{B}+\alpha \cdot \left(\frac{P}{{K}_{{GP}}+P}{x}_{G}+{r}_{P}\frac{P}{{K}_{P}+P}\right)+{r}_{D}\frac{D}{{K}_{D}+D}$$

(5)

$$\frac{{dD}}{{dt}}= \frac{-D}{{K}_{{GD}}+D}{x}_{G}-\frac{D}{{K}_{{BD}}+D}{x}_{B}+1-\left[1-\alpha \right]\cdot \left(\frac{P}{{K}_{{GP}}+P}{x}_{G}+{r}_{P}\frac{P}{{K}_{P}+P}\right)\\ -{r}_{D}\frac{D}{{K}_{D}+D}$$

(6)

The ratio of GalA in the degradation products, denoted by $\alpha$ ($\alpha \in [\mathrm{0,\, 1}]$) is set to be dependent on the substrate (PGA) concentration P:

$$\alpha=\frac{1}{{\left(\frac{2\eta }{{\rm{P}}}\right)}^{2}+1}$$

(7)

Di-GalA can also be autonomously hydrolyzed at a very low rate into GalA, where ${r}_{D}\, \ll \, 1$ and$\,{K}_{D}\, \gg \, D$.

The initial values of the variables are set as shown in the table below.

Description	Variable	Initial value
The biomass of Ga and Bi	${x}_{G},\, {x}_{B}$	0.05a
The concentration of PGA	$P$	2 for PGA and 0.6 for PC
The concentration of PC	${\rm{C}}$	0 for PGA and 1.4 for PC
The concentration of GalA	${\rm{A}}$	0
The concentration of di-GalA	$D$	0

1. ^aThe initial biomass is set to 0 when simulating the growth of monoculture of the other bacterial species.

The interaction between Ga and Bi is calculated as the ratio of the AUC of the growth curve of each species in coculture to that in monoculture.

$${I}_{{BG}}=\int \frac{{x}_{G}\left({x}_{G0}=0.05,\, {x}_{B0}=0.05\right)}{{x}_{G}({x}_{G0}=0.05,\, {x}_{B0}=0)}{dt}$$

(8)

$${I}_{{GB}}=\int \frac{{x}_{B}\left({x}_{G0}=0.05,\, {x}_{B0}=0.05\right)}{{x}_{B}({x}_{G0}=0,\, {x}_{B0}=0.05)}{dt}$$

(9)

To better capture the change of interactions between Ga and Bi, we utilized the ratio of interaction ${I}_{{GB}}$ under the two substrates to optimize parameters (Fig. 6B and C).

$${\lambda }_{{GB}}=\frac{{I}_{{GB}}\left({PC}\right)}{{I}_{{GB}}\left({PGA}\right)}$$

(10)

The best-fit estimates of these parameters are listed in Supplementary Table 3 (represented by yellow dot condition in Fig. 6C and D).

The simulation was performed in Matlab (Matlab, R2022) using custom scripts (https://github.com/JacobZuo/Consumer-resource-model-of-mutualistic-interaction).

Statistical analyses

Statistical analyses were performed using GraphPad Prism (version 9.3.1). For comparisons between two conditions, the Student’s t test was used. For multiple comparisons, the two-way ANOVA followed by the Tukey’s test was used. A generalized linear mixed-effects model (GLMM) with family Poisson was used to address the effect of glycan supplementation on bacterial colonization, where cage division was considered as a random effect. Tukey’s test was subsequently used for multiple comparisons.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.