Clostridium butyricum CBX 2021 strain was isolated and screened from the intestines of a healthy 6-month-old boar in our laboratory, and is currently stored at the Guangdong Microbial Culture Collection Center (GDMCC No.62503).

The CBX 2021 strain was incubated under anaerobic environment at 37°C for 12 h in reinforced Clostridium medium. For animal administration, the harvested bacterial solution (3 × 10⁸ CFU/mL) obtained at this time point was orally administered to weaned piglets. For cell culture assays, the bacterial solution was centrifuged at 5,000 × g at 4°C for 10 min to collect the bacteria. Then, the bacteria were resuspended in DMEM/F12 cell culture medium without penicillin/streptomycin (1.5 × 10⁸ CFU/mL) for cell experiments.

The experiment included forty-eight 28-day-old healthy piglets (8.16 ± 0.77 kg) split evenly into 2 groups, with 6 replicates per group and 4 piglets per replicate. The control (CON) group piglets were administered a 5 mL sterile saline orally, while the CB group piglets were given a 5 mL CBX 2021 bacterial solution (with an effective bacterial count of 3 × 10⁸ CFU/mL) orally once daily. The experiment lasted for 28 days and utilized a commercially available, standard-formulated feed. The composition and nutritional levels of it are displayed in Supplementary Table S1.

After the experiment concluded, one piglet from each replicate was chosen at random to gather fresh fecal samples for 16S rRNA sequencing analysis. Meanwhile, the venous blood of piglets was collected. After the blood samples stood overnight, the serum was collected by centrifugation at 3,000 × g at 4°C for a duration of 15 min and stored at −80°C for subsequent analysis of immune, growth-related hormones, and intestinal permeability.

All piglets were weighed on an empty stomach on d 0 and 28, and the daily feed intake for each replicate was noted daily. These values are utilized in the computation of the average daily gain (ADG), average daily feed intake (ADFI), and feed-to-gain ratio (F/G).

Monitor the fecal consistency of each piglet during the trial, and document the number of piglets with diarrhea. Assess the piglet feces according to the following scoring standards reported in previous studies and calculate the frequency and index of diarrhea (Hung et al., 2019). Diarrhea frequency % = [ ∑ ( number of piglets with diarrhea × number of days with diarrhea ) / ( total number of piglets × test days ) ] × 100. Diarrhea index = sum of diarrhea scores / total number of piglets × test days .

As per the manufacturer’s guidelines, the ELISA kit (Enzyme-linked Biotechnology Co., Ltd., Shanghai, China) was utilized to measure serum immune indicators: immunoglobulin A (IgA), IgG, and IgM, as well as interleukin-1β (IL-1β), IL-10, and tumor necrosis factor-α (TNF-α); growth-related hormone indicators: growth hormone (GH), gastrin (GAS), ghrelin (GHRL), peptide YY (PYY), and glucagon-like peptide-1 (GLP-1); intestinal permeability indicators: D-lactic acid (DLA) and diamine oxidase (DAO).

As previously mentioned, we extracted and purified total genomic DNA from piglet fecal samples (Wang J. L. et al., 2016). The quantitative PCR products were then examined through the Illumina NovaSeq platform for sequencing at Beijing Biocloud Biotechnology Co., Ltd. (Beijing, China¹) via paired-end sequencing. The raw data can be found in the NCBI Sequence Read Archive (SRA) database and accessed using the login number PRJNA947735.

Sequencing reads were denoised using the DADA2 method and clustered into amplicon sequence variants (ASVs). The SILVA 138 database was used as a reference for classifying and annotating each representative sequence of the ASVs using a naive Bayes classifier, with a 70% confidence threshold. The unweighted unifrac distance algorithm was used for principal coordinate analysis (PCoA) at the bacterial genus level. ANOSIM function was used for statistical analysis. Student’s t-test or Mann–Whitney U test was used to analyze the differences in microbial distribution at phylum and genus level. Additionally, Linear discriminant analysis (LDA) effect size (LEfSe) was employed to identify biomarkers (from family to genus level), with a significance threshold set at an LDA value of ≥3.0.

The serum samples from the piglets were sent to Beijing Biocloud Biotechnology Co., Ltd. (Beijing, China) for metabolite extraction, detection and analysis. The Progenesis QI software was utilized to process the raw data, including peak detection, calibration, and other essential operations for data processing. Upon peak area normalization, the processed data were exported for further analysis. The orthogonal partial least squares discriminant analysis (OPLS-DA) was utilized to compare and identify the overall metabolic variations in the CON and CB groups. Differential metabolites were chosen based on VIP values >1 in the OPLS-DA model and p-value <0.01. The pathway enrichment analysis of the differential metabolites was then conducted using the Kyoto Encyclopedia of Genes and Genomes (KEGG) database. The Spearman correlation algorithm was applied to analyze the relationships among the top 20 bacterial genera, differential metabolites, and serum biochemical indicators, generating correlation matrices to assess their associations.

IPEC-J2 cells were graciously provided by Dr. Yongjiang Wu (Chongqing University of Arts and Sciences, Chongqing, China). The cells were grown in DMEM/F12 medium supplemented with 10% fetal bovine serum (Gibco, Paisley, USA), 1% penicillin/streptomycin (Solarbio, Beijing, China), and 1% insulin-transferrin-selenium (Sigma-Aldrich, MO, USA). And they were incubated in biochemical incubators (5% CO₂, 37°C).

IPEC-J2 cells were inoculated in a 6-well plate (5 × 10⁵ cells/ well), and divided into 2 groups after the confluence reached 70–80%. CB group was exposed to CBX 2021 bacterial suspension (1.5 × 10⁸ CFU/mL, 2 ml/well), and CON group was treated with an equivalent volume of DMEM/F12 medium, with three replicates in each group. After incubation at 37°C for 4, 6, and 8 h, the cells were treated according to the standard procedure. Subsequently, they underwent fixation with 4% neutral formaldehyde for 15 min before being subjected to gram staining. The samples were examined under inverted fluorescence microscope with magnification of ×40 for analysis. Lastly, 25 cells were randomly selected, and the average number of CBX 2021 bacterial adhesion on visible cell surfaces was calculated and documented through photography.

IPEC-J2 cells were inoculated in a 96-well plate (1 × 10⁴ cells/ well) and grew to the above degree. The cells were classified into 3 distinct groups: the blank group (without cells), CON group, and CB group. Each group had 3 replicate wells (100 μl/ well). After the cells were incubated for 2, 4, 6, 8, and 12 h, the old medium was aspirated and DMEM/F12 medium with 10% CCK-8 was added (100 μl/ well) within the specified culture time, followed by 2 h of additional incubation. Then, OD values at 450 nm were determined using plate reader, and the cell proliferation activity was calculated at each culture time point to ascertain the optimal time for coculture of the bacterial strain and cells.

Cell processing and grouping were as described in the above adhesion experiment. After incubation for 6 h, the coculture supernatant was gathered to measure the secretion levels of IL-1β and IL-10 using porcine ELISA kits as per the manufacturer’s guidelines.

Cell processing and grouping were as described in the above adhesion experiment, and each group had 5 replicates. After incubation for 6 h, cell digestion and collection were performed using 0.25% pancreatase-EDTA (Sigma-Aldrich, MO, USA). The cell samples were then sent to Hangzhou Lianchuan Biotechnology Co., Ltd. (Hangzhou, China) for total RNA isolation, quality control, cDNA library construction, and transcriptome sequencing. The raw data can be found in the NCBI SRA database and accessed using the login number PRJNA948058.

The raw sequencing data underwent quality control, which involved removing reads that had low-quality bases and adapter contamination. Fragment per kilobase per million mapped reads were used to determine gene expression levels. DESeq2 was applied for the differential expression analysis, with the false discovery rate (FDR) controlled through the Benjamini-Hochberg procedure. The screening condition of differentially expressed genes (DEGs) was FDR < 0.05 and |log2(Fold change)| > 1. Additionally, the enrichment analyses were conducted using the OmicStudio cloud platform, including gene ontology (GO) term, KEGG pathway, and gene set enrichment analysis (GSEA). GSEA analysis relied on the GO database and involved calculating a normalized enrichment score (NES) for each gene set through 1,000 permutations of the genome. Gene sets with |NES| > 1 and FDR < 0.05 were considered significantly enriched in the GSEA analysis.

The previously mentioned RNA samples underwent reverse transcription to generate cDNA, utilizing a commercially available reverse transcription kit (Yeasen, Shanghai, China) in a two-step reaction. Then, we followed the instructions provided by the manufacturer of the fluorescent quantitative PCR assay kit (Yeasen, Shanghai, China) in order to conduct our experiments. The 10 μl reaction mixture was prepared containing 1 μl of cDNA, 0.2 μl of upstream and downstream primer, 3.6 μl of RNA-free water, and 5 μl of SYBR fluorescent dye. The mixture underwent cDNA amplification and fluorescence signal collection using a real-time fluorescence quantitative PCR instrument (Thermo Scientific, Wilmington, DE, USA). The amplification program was performed as instructed. In this study, the PCR amplification efficiency was always between 95 and 100%, so the 2^-∆∆CT method was used to analyze the relative mRNA expression levels of the target genes mRNA (Livak and Schmittgen, 2001). A single internal reference gene GAPDH was selected for normalization based on previous studies (Wang H. et al., 2019; Wu et al., 2021; Li et al., 2023). Primers utilized in this research were synthesized by Shanghai Sangon Biotech Co., Ltd. (Shanghai, China) (Supplementary Table S2).

After the data were preliminarily organization in Excel 2021, SPSS 26.0 software was used for statistical analysis. The Shapiro–Wilk test is used to evaluate whether the data obeys normality. Student’s t-test was conducted for the data that followed normal distribution and showed homogeneity of variance, while Mann–Whitney U test was used for non-parametric analysis when variance was unequal. The results are expressed as means ± SEM. Statistical significance was defined as p < 0.05 (*), p < 0.01 (**), and p < 0.001 (***). A significant trend was noted when 0.05 ≥ p > 0.10. Data visualization was carried out using GraphPad Prism 9.0 software.

Table 1 presents the growth performance results of piglets. The CB piglets showed a 10.10% increase in body weight after 28 days compared to the CON piglets (p = 0.06). The ADG of CB treated pigs significantly increased by 24.95%, and F/G obviously reduced by 15.82% (p < 0.05). However, no difference in ADFI was witnessed between CON and CB piglets (p > 0.05). Compared to the CON group, CB piglets experienced a significant reduction in diarrhea frequency and diarrhea index by 36.15 and 40.74%, respectively (p < 0.05). The findings indicate that CBX 2021 improves the growth performance of piglets and helps manage post-weaning diarrhea.

Research suggests that there is a strong correlation between peripheral hormone secretion and improved growth performance (Li et al., 2021). Thus, we analyzed the serum levels of growth- and appetite-related hormones in piglets (Table 2). Compared with CON piglets, the of GH and GAS levels in the CB piglets increased by 6.33 and 7.74%, respectively (p < 0.05). The level of GHRL, a key appetite-stimulating hormone, showed a trend of increase by 3.90% (p = 0.06). Elevated levels of GAS and GHRL expression play a significant role in stimulating digestive fluid secretion, which is vital for increasing feed intake and conversion (Mazzoni et al., 2022; Rodríguez-Viera et al., 2022). However, there were no notable variations in the concentrations of the anorectic hormones PYY and GLP-1 between the piglets in the two groups (p > 0.05). The results indicate that CBX 2021 may positively impact the growth and development of piglets by enhancing endocrine hormone levels.

The serum immunoglobulin levels in Table 2 reflect the humoral immune response level. The CB group piglets showed a significant increase of 5.90, 11.02, and 7.31% in IgA, IgG, and IgM levels, respectively, compared with CON piglets (p < 0.05).

Furthermore, we assessed serum levels of inflammatory mediators as indicators of weaning stress-induced inflammation within the body. The study revealed that compared with CON piglets, IL-10 content of CB piglets increased by 13.05%, and IL-1β content decreased by 15.53% (p < 0.05). However, there was no significant difference in TNF-α level between the two groups (p > 0.05).

The impact of CBX 2021 on piglet’ intestinal permeability was studied by measuring serum DLA levels and DAO activity. Elevated levels of DLA and DAO are typically indicative of heightened intestinal permeability (Cai et al., 2019). The findings indicated that the DLA level and DAO activity in the CB piglets were significantly reduced by 15.27 and 9.98%, respectively (p < 0.05) compared with CON piglets.

To sum up, the results indicate that CBX 2021 effectively enhanced immune response, strengthened intestinal barrier, and alleviated inflammatory stress in weaned piglets. This may lead to an overall improvement in the animals’ health status.

A balanced microbial community structure contributes to immune system development and intestinal health (Wang et al., 2018; Wu et al., 2023). Here, we conducted an analysis on the diversity and structure of fecal microbiota by 16S rDNA sequencing to evaluate the gut bacterial community of piglets. The PCoA scatter plot, which illustrates the β diversity of microbiota, showing that the microbial distribution of the two groups of piglets is obviously different (R = 0.670, p = 0.002, Figure 1A). This result indicated that supplementation with CBX 2021 obviously altered the composition of microbial community in piglet feces.

Subsequently, an analysis of the bacterial population structure was conducted, focusing specifically on phylum and genus levels (Figures 1B,C). A combined total of 12 bacterial phyla were recognized in both groups of piglets. Firmicutes and Bacteroidota were the main phyla among them, comprising over 97% of the total abundance (Figure 1B). Notably, the abundance of Desulfobacterota was significantly decreased by CBX 2021 treatment (p < 0.05, Figure 1D). Figure 1C illustrates the relative abundance of the top 20 genera. Faecalibacterium, a butyrate producer, had a higher relative abundance in piglets receiving CBX 2021 intervention than CON piglets (p < 0.05). Additionally, there was an increasing trend in the number of helpful bacteria, particularly Prevotella_9 (p = 0.087), Subdoligranulum (p = 0.063), and unclassified_Bacteroidales (p = 0.055), with their numbers 1.97-, 1.99-, and 4.65-fold higher than those of the CON piglets, respectively (Figure 1D). Furthermore, the addition of CBX 2021 caused a substantial drop in the number of proinflammatory bacteria, including unclassified_Muribaculaceae and Rikenellaceae_RC9_gut_group (p < 0.05, Figure 1D). The findings suggest that exogenous administration of CBX2021 had a positive impact on the gut microbiota as a whole.

LEfSe analysis was used to study the key bacterial species that drive variations within the microbial population. As shown in Figure 1E, a total of 33 different taxa were found in both groups (LDA score ≥ 3.0). In the CB piglets, seven genera, including Faecalibacterium, Ligilactobacillus, Eubacterium_eligens_group, and Lachnospiraceae_NK4A136_group, were significantly enriched and could serve as biomarker genera. Furthermore, CON piglets exhibited 14 characteristic genera, such as unclassified_Muribaculaceae, Rikenellaceae_RC9_gut_group, Desulfovibrio, and Family_XIII_AD3011_group.

LC-Q/TOF-MS technology was employed to examine the serum metabolome changes in weaned piglets treated with CBX 2021.The OPLS-DA score plots clearly showed clear separation between the two groups, suggesting that CBX 2021 significantly influenced the serum metabolite composition in piglets (Figures 2A,B). The high statistical values of R²Y (explanatory ability parameter) and Q² (predictive ability parameter) in the score plots demonstrated that the models had good applicability and predictability. Furthermore, in the permutation tests of both ion modes, the intercepts of the Q²Y fitted regression lines with the vertical axis were below 0, confirming the reliability of the OPLS-DA model and its suitability for identifying differential metabolites between groups (Figures 2C,D).

A total of 70 differential metabolites were screened between the two groups of piglets, with 38 being positive ions and 32 being negative ions (Figures 2E,F and Supplementary Table S3). In the positive ion mode, the contents of 29 metabolites, such as 1,5-naphthalenediamine, 6-methylquinoline, indoleacrylic acid, and isoquinoline N-oxide, were significantly higher in the CB piglets than CON piglets, while the contents of 9 metabolites, such as selamectin, were significantly lower (p < 0.01). In the negative ion mode, the contents of 19 metabolites, such as PE(22:4(7Z,10Z,13Z,16Z)/22:5(4Z,7Z,10Z,13Z,16Z)), PE(20:1(11Z)/PGF1alpha), and PE(15:0/22:1(13Z)), showed a significant increase, while the contents of 13 metabolites, such as DN-isobutylamide and 4-hydroxycyclohexylcarboxylic acid, exhibited a significant decrease (p < 0.01).

Subsequently, an analysis of KEGG pathway enrichment was conducted on all differential metabolites found in the piglets from both groups. As shown in Figure 2G, CBX 2021 highly affected 5 pathways, including indole alkaloid biosynthesis, and tryptophan metabolism (p < 0.05). The significant increase in L-tryptophan and tryptamine mediated by CBX 2021 mainly contributed to the enrichment of these pathways (p < 0.05, Figure 2H). These outcomes show that CBX 2021 modulated tryptophan metabolism and biosynthesis of various alkaloids by altering the levels of metabolites, thus affecting the metabolic function of weaned piglets.

Metabolomics is considered an important tool for elucidating potential interactions between the microbiota and host phenotypes. Therefore, this study investigated the correlations between the top 20 bacterial genera and serum differential metabolites, along with the relationships between these metabolites and blood indicators. Figure 3A shows that in the CB piglets, upregulated genera (Faecalibacterium, unclassified_Bacteroidales, and Subdoligranulum) and downregulated genera (unclassified_Muribaculaceae and Rikenellaceae_RC9_gut_group) were highly correlated with differential metabolites. Specifically, Faecalibacterium showed positive correlations with upregulated metabolites in the CB piglets; these metabolites included 6-methylquinoline, indoleacrylic acid, isoquinoline N-oxide, L-tryptophan, and tryptamine (p < 0.05). Faecalibacterium exhibited negative correlations with downregulated metabolites in the CB piglets; these metabolites included selamectin, 4-hydroxycyclohexylcarboxylic acid, and DN-isobutylamide (p < 0.05). Interestingly, unclassified_Bacteroidales and Subdoligranulum showed similar trends to that of Faecalibacterium, while unclassified_Muribaculaceae and Rikenellaceae_RC9_gut_group showed opposite trends.

The correlation analysis results between differential metabolites and blood indicators are shown in Figure 3B. Serum immunoglobulins showed positive correlations with a majority of the upregulated metabolites in the CB piglets (p < 0.05), while they showed negative correlations with downregulated metabolites, especially 4-hydroxycyclohexylcarboxylic acid (p < 0.05). Regarding inflammatory factors, IL-1β was negatively correlated with tryptamine, and IL-10 was negatively correlated with selamectin, 4-hydroxycyclohexylcarboxylic acid, and DN-isobutylamide (p < 0.05). Notably, growth-related hormones, including GH, GHRL, and GAS, were consistent with the trend of immunoglobulin. Furthermore, in terms of indicators reflecting intestinal permeability, serum DAO activity showed positive correlations with the content of DN-isobutylamide and selamectin (p < 0.05), and it showed negative correlations with the content of PE(22:4(7Z,10Z,13Z,16Z)/22:5(4Z,7Z,10Z,13Z,16Z)), PE(20:1(11Z)/PGF1alpha), and PE(15:0/22:1(13Z)) (p < 0.05).

The aforementioned findings indicate that alteration in gut microbiota composition mediated by CBX 2021 intervention may positively influence hormone levels, immune responses and gut barrier function by regulating metabolic function, ultimately improving growth, development and overall health of piglets.

Adhesion of strains to the surface of intestinal epithelial cells (IEC) is crucial for achieving their maximum probiotic effect (Kainulainen et al., 2015). Therefore, we assessed the ability of the CBX 2021 strain to adhere to IPEC-J2 cells in this study. The results, shown in Figures 4A,B, indicate that after 2 h of co-culture, there was almost no bacterial adhesion. After 4 h of co-culture, the average bacteria amount of adhesion was 50.04 ± 2.74 CFU/cell. After 6 h of co-culture, the average bacterial adhesion decreased to 33.92 ± 2.48 CFU/cell. When co-cultured for 8 h, there was almost no detectable bacterial adhesion. These findings indicate that the adhesion capability of CBX 2021 to IPEC-J2 cells is time-dependent, reaching its peak at approximately 4 h of co-culture.

The integrity of the intestinal barrier is dependent on the proliferation of IEC (Xie et al., 2019). Therefore, we investigated the impact of CBX 2021 adhesion on the proliferation activity of IPEC-J2 cells. The results in Figure 4C indicate that the strain adhesion did not significantly affect cell viability through 2 h of co-culture (p > 0.05). However, between 4 to 12 h of co-culture, CBX 2021 demonstrated a significant proliferative effect on IPEC-J2 cells (p < 0.01). As the co-culture time increased, cell viability gradually increased and reached its peak at 6 h, with a value of 120.14%. Subsequently, with a further increase in time, cell viability slightly declined, reaching 111.15% at 12 h. However, the CB group cells still demonstrated a significant proliferative effect on the cells at this time point (p < 0.001) compared to the CON group cells. The outcomes suggest that CBX 2021 significantly promoted IPEC-J2 cells proliferation, reaching peak activity after a 6-h co-culture. Therefore, a 6-h co-culture time was chosen for subsequent experiments.

To assess the potential anti-inflammatory properties of CBX 2021 in the intestine, we also analyzed the levels of inflammatory mediators in the cell supernatant. The data shows that CBX 2021 effectively decreased IL-1β levels, while enhancing IL-10 secretion in IPEC-J2 cells following 6 h of stimulation, (p < 0.05, Figure 4D). These findings are consistent with the in vivo experimental results, and thus suggesting that CBX 2021 can reduce the inflammatory response of IEC.

To delve deeper into the molecular mechanisms responsible for reducing inflammation in IPEC-J2 cells through CBX 2021 adhesion, RNA-seq analysis was conducted to confirm changes in the relative gene expression. The PCA scatter plot indicates that the two groups of cells formed distinct clusters, suggesting significant differences in gene expression profiles (Figure 5A). A total of 1892 DEGs were distinguished between the CON and CB groups. Among them, 1,387 genes, including TCIM, ZC3H12A, NFKBIE, and GPR20, showed a marked increase in the CB group compared to the CON group, while 505 genes, including KEAP1, PANK1, and CDCA7, were significantly downregulated (Figure 5B). The heatmap was generated using the top 100 genes with the smallest adjust p-value (Supplementary Figure S1). It is obvious that cell samples within the same group have high repeatability in gene expression. Notably, the TCIM and ZC3H12A genes were mapped to GO database and involved in various immune-related pathways. For instance, the TCIM gene was implicated in negative regulation of the Notch signaling pathway. The ZC3H12A gene is involved in activating immune response signals and negatively regulating the production of IL-1β, IL-6, and TNF. The GPR20 gene is involved in the G protein-coupled receptor (GPCR) signaling pathway (Supplementary Table S4).

Subsequently, enrichment analyses were carried out for all DEGs using GO and KEGG databases. The findings showed that numerous DEGs were significantly enriched in GPCR signaling pathway, inflammatory and immune response, which are related to cellular health and stress adaptation (Figure 5C). The KEGG pathway analysis revealed multiple immune and inflammatory pathways, including cytokine-cytokine receptor interaction (52 DEGs), PI3K-Akt signaling pathway (44 DEGs), JAK–STAT signaling pathway (23 DEGs), were altered by the adhesion of CBX 2021 (Figure 5D). By screening for these DEGs, we found that IL10RA and IL22RA1, which are anti-inflammatory cytokine receptors, were significantly upregulated in the CB-treated cells. In contrast, the pro-inflammatory gene IL17D was significantly downregulated in the cells.

To further reveal the details of changes in cellular immunity and inflammatory response, GESA was conducted using the Go database on all gene sets of the CON and CB groups. As shown in Figures 6A–F, adhesion of CBX 2021 significantly enhanced immune response (ES = 0.58, NES = 2.26), GPCR activity (ES = 0.55, NES = 2.17), and GPCR signaling pathway (ES = 0.55, NES = 2.27) in the cells. In addition, CBX 2021 was able to improve the negative regulation for the Notch signaling pathway (ES = 0.66, NES = 1.85), the negative regulation of the production of IL-6 (ES = 0.67, NES = 1.94) and the negative regulation of tumor necrosis factor (ES = 0.60, NES = 1.80). The changes in these gene sets supported the fact that CBX 2021 decreased pro-inflammatory cytokines like IL-1β in IPEC-J2 cells. It is obvious that this result interprets the molecular pathway for CBX adhesion to reduce cellular inflammation at the gene level.

Next, the transcriptomic result was validated by using qRT-PCR method. Twelve genes (PLAT, TCIM, ZC3H12A, GPR20, IL10RA, IL22RA1, TLR2, KEAP1, PANK1, DACT1, ARMC6, and LMNB1) were subjected to qRT-PCR analysis. The findings indicated a high degree of concordance between the qRT-PCR data and transcriptomics results and thus confirming the reliability of the RNA sequencing (Figure 6G).