Seedling development
The different cacao genotypes' germination, emergence, and development were evaluated under in vitro conditions. Over 75.0% of all genotypes' seeds germinated in five to nine days. The germination percentage was higher for IMC67, AC9, and ROS2 with 97.2%, followed by TCS01, TCS19, and ROS1, with 96.9, 88.6, and 77.8%, respectively. ICS95 was the cacao genotype with the lowest germination percentage (75.0%). Meanwhile, seedlings' emergence was higher for TCS01, with 71.0%, followed by ROS2, and AC9, with 65.7 and 62.9%, respectively. Cacao genotypes with the lowest seedling emergence percentages were TCS19, IMC67, ROS1, and ICS95, with 51.6, 51.4, 42.9, and 40.7%, respectively.
Library preparation and sequencing for 16S rRNA and ITS-5.8S regions
By sequencing 16S rRNA and ITS-5.8S regions, we examined the microbial community composition for the different cacao genotypes. We obtained successful amplifications in 84.8% of the samples. The success in library construction varied between sequenced regions, genotypes, and tissue. For 16S rRNA, we constructed libraries for 93.2% of the samples, while for ITS-5.8S, the success of generated libraries was for 76.4% of the samples (Fig. S1).
For the 16S rRNA region, 266 libraries were successfully sequenced, with an average sequence depth of 5257.18 (± 2891.29) reads per sample. After performing the quality control, 263 libraries were retained, with 69.6% of the original reads. For the ITS-5.8S region, sequences for 221 samples were obtained with an average sequence depth of 5718.12 (± 3123.20) reads per sample. Only 0.6% of the reads were lost after performing quality control, allowing the retention of 99.4% of reads. The sequencing depth for all the libraries ranged between 285 and 15385 reads, with an average of 5686.68 (Table S1).
Genotypic and tissue-specific variations in bacterial abundance
Taxonomic classification allowed us to separate the fraction of reads belonging to bacterial 16S rRNA from those belonging to plant organelles (mitochondria and chloroplast). Bacterial reads were obtained from most seedlings' libraries, but few sequences from seed tissue. Most reads were classified as mitochondrial or chloroplast 16s rRNA, suggesting that the abundance of bacteria within seeds was low compared to the number of bacteria in other tissues.
A closer examination of the data sets from seedlings showed that the relative abundance of reads classified as bacteria, chloroplast, and mitochondria was similar within tissues (e.g., more similar within roots of all genotypes), showing that the relative abundance of bacteria is tissue dependent. To compare the taxonomic diversity patterns of bacteria between tissues and their relative abundance, we used the mitochondrial 16S rRNA as an internal marker. This marker allowed us to normalize the abundance of bacteria in each sample. To explore these differences, we calculated the fold change between the number of bacteria and plant mitochondria 16S rRNA reads and evaluated the relative abundance of bacteria in each plant tissue for each cacao genotype (Fig. 1).
The bacterial to mitochondrial proportion of 16S rRNA reads reveals tissue-specific patterns of bacterial abundance. Roots tissue had the highest abundance of bacteria reads compared to mitochondria, in the case of TCS01, bacterial reads in root tissues were 31.7 times higher than mitochondria, and similar tendencies were found in ROS1, AC9, ICS95, ROS2, and TCS19 genotypes with 11.3, 9.3, 6.7, 4.8 and 3.0 times more bacterial reads. Only IMC67 had a lower number of bacterial to mitochondrial reads with 0.87 times. The leaves tissue was the second with the highest bacterial abundance, in this case, TCS01, ROS2, ROS1 AC9, and TCS19 genotypes had a higher number of bacteria with 7.7, 4.7, 3.7, and 1.5 times the number of mitochondria while IC95 and IMC67 had lower abundance with 0.98 and 0.41 times. Cotyledon and stem tissues had a lower abundance of bacteria in all genotypes as shown by the negative values of the fold difference (Fig. 1a), exceptions were observed for three seedlings of the TCS01 genotype.
The fold difference analysis also shows differences among genotypes. The AC9, ROS2, ROS1, TCS01, and TCS19 genotypes had a higher average value of bacterial abundance, with 3.6, 2.5, 3.3, 12, and 1.3 times the number of bacterial to mitochondrial reads. This contrasts with the observed in the most domesticated genotypes IMC67 and ICS95 with average values of 0.37 and 0.96. In all evaluated tissues, the IMC67 genotypes had the lower overall bacterial abundance; for instance, tissues such as the IMC67 stem had almost 32 times more mitochondria than bacteria on average (a fivefold difference, Fig. 1a).
Bacterial and fungal taxonomic composition
Bacterial composition was evaluated after excluding 16S rRNA plant mitochondria and chloroplast sequences. We obtained 2,514,150 reads belonging to 240 libraries. These reads were classified into taxonomic groups using the GreenGenes database. Most of the 16S rRNA reads were assigned to the Proteobacteria (86.3%) and the Firmicutes phyla (11.3%, Fig. 1b). A few other reads were assigned to the phylum Actinobacteria (2.2%).
At a finer resolution, 87 groups were classified at the genus level. From this, 61 had more than 1.0% abundance, and only 33 were in more than one sample. The Pseudomonas and Pantoea genera (Proteobacteria) were the most prevalent in 107 and 102 samples, respectively. Pseudomonas was dominant in the seedling tissues of AC9, ROS1, and TCS01 genotypes, while Pantoea was more prevalent in ROS2 and TCS19. The abundance of these two genera was negatively correlated (r = -0.47, p < 0.05), indicating possible ecological exclusion. In the case of ICS95 seedlings, Agrobacterium and Bacillus were predominant. In contrast, in IMC67, we did not observe the prevalence of one specific bacteria genus. Instead, we detected the presence of multiple genera and families, such as Methylobacterium, Mycobacteriaceae, Symbiobacteriaceae, and Bradyrhizobiaceae (Fig. 1b).
Some differences were observed between seedlings from the same genotype. For instance, seedling 34 of the AC9 genotype presented a greater diversity of bacteria than the other replicates. In ROS1, for seedling 17, the genus Methylobacterium predominated, except in the root where Bacillus and Pseudomonas genera were detected. In seedling 14, only the phylum Bacteroidetes was presented in the stem, and in seedling 27, a dominant group belonging to the class Alphaproteobacteria was observed (Fig. 1b).
For the fungi libraries, we obtained 1,256,757 ITS reads after filtering and trimming, and we used them to evaluate the taxonomic composition. After the QIIME2 analysis, 213 ASVs from which 203 were retained for fungi classification. The results obtained with the UNITE database were too unspecific; therefore, we used BLAST to attempt a better classification. This analysis showed that the great majority of the fungal reads in cacao tissues belonged to the phylum Ascomycota (99.9%), and only a few reads belonged to the phyla Basidiomycota (0.02%) and Mucoromycota (0.01%). Within the Ascomycota, 99.7% of the reads belonged to the subphylum Pezizomycotina and few to the subphylum Saccharomycotina (0.3%) (Fig. 2).
Most genotypes and tissues had a similar fungal community (Fig. 2). The genus Penicillium was the most dominant, with 78.8% of reads (Fig. 2). Few samples had other fungi from the orders Pleosporales and Hypocreales and subclass Chaetothyriomycetidae. Some seedlings presented other dominant taxa than Penicillium. For instance, tissues from individuals 34 of AC9 and 18 of ICS95 had more abundance of fungi of the subclass Chaetothyriomycetidae (Fig. 2).
Species Richness Analysis
The bacterial and fungal species richness indices were calculated for each sample to determine if there were significant differences between genotypes and tissue types (Table S2). The ITS species richness analysis did not show significant differences between genotypes or tissues (p > 0.05). The analysis of bacterial ASVs (with relative abundance > 1%) showed that the seedlings host a small number of species (on average, four ASVs per tissue). For some seedling tissues, we did not retrieve any bacterial reads. An average of six ASVs were found in the ITS libraries. The ICS95 rhizosphere libraries had a higher number of 16S rRNA ASVs (34 ASVs). Concerning the genotypes, the 16S rRNA variant analysis showed that the TCS01 genotype had a significantly higher number of endophytic bacterial species (observed features) than AC9, ROS1, ROS2, and TSC19 (p < 0.005). To account for the phylogenetic distance among ASVs, we calculated the Faith phylogenetic diversity (PD) index [52]. There were significant differences between the diversity values of AC9 and other genotypes (p < 0.05). Similarly, the diversity values of ROS2 were significantly different to TCS19, IMC67, TCS01, and ICS95 (p < 0.05). No significant differences were observed between genotypes ICS95 and IMC67 and between genotypes AC9 and ROS2.
Regarding the tissues, the different diversity indices (Faith-PD and observed features) showed that the number of observed species and phylogenetic diversity was, on average, higher in roots than in the other tissues. Statistical evaluation of Faith-PD distances shows that there are significant differences between root tissue with cotyledon and leaf tissues (p < 0.05) (Table S3).
Interestingly, when the frequency of seedling emergence is mapped against the phylogenetic diversity index, the most diverse genotypes had a lower seedling emergence, except for ROS1, in which seed emergence was low (42.9%). For instance, the genotypes with the highest emergence values, TCS01 (71.0%), ROS2 (65.7%), and AC9 (62.9%), had the least diverse endophytic communities of bacteria, as reflected in Faith distance distributions (Table S3). In contrast, IMC67 (51.4%) and TCS19 (51.6%), with lower emergence values, had more diverse communities. Similarly, ICS95 (40.7%) was the genotype with the most diverse bacterial community and had the lowest emergence values. A significant correlation (Pearson's r) is observed between Faith PD and seed germination and emergence when the values for genotype ROS1 are excluded from the analysis (p < 0.05).
Bacterial and fungal composition of different genotypes and their tissues
A Principal Coordinate Ordination Analysis (PCoA) of the seed-borne bacteria was constructed with a matrix of weighted UniFrac distances between samples (Fig. 3). The PCoA was used to visualize the similarities or differences between the endophytic bacterial communities and to evaluate if the distribution was structured based on the genotype or plant tissues. The first, second, and third axes explained 51.5, 15.7, and 10.3% of the variability, respectively (Fig. 3). When samples were colored by tissue, no clear separation was observed (Fig. 3a), but when the samples were differentiated by genotype, three clustering groups were observed, one composed of TCS01, AC9, and ROS1, a second of ROS2 and TCS19, and a third consisting primarily of IMC67 and ICS95 (Fig. 3b).
We conducted a PERMANOVA analysis using the weighted and unweighted UniFrac distances to establish the significance of the genotype clustering. Read subsampling was set to 500 reads because, at this depth, the species diversity was saturated based on the rarefaction curves. When analyzing the samples based on the weighted UniFrac distance, we found four groups of genotypes. The first contained only AC9, the second composed of ROS1 and TCS01, the third had ROS2 and TCS19, and the last included IMC67 and ICS95. Each of these groups showed significant differences compared to the others (p < 0.005). Interestingly, the UniFrac average distance, having AC9 as a reference, showed a clear pattern where less domesticated genotypes had a shorter distance between them (AC9, ROS1, and ROS2), slightly higher distance with recently liberated ones (TCS01 and TCS19), and the most considerable distance with the commercial genotypes (IMC67 and ICS95, Fig. S2).
When a similar comparison was made using the unweighted UniFrac distance, three groups were observed: AC9 - ROS1, TCS19 - IMC67 - ICS95, and ROS2 - TCS19. The comparison of tissue distances showed a significant difference between stem and roots, indicating that the most differentiated endophytic community is in roots (Fig. S3).
The Principal Coordinate Ordination Analysis (PCoA) of the fungi community was constructed with a matrix of weighted UniFrac distances and no clustering based on tissues or genotypes was observed (data not shown). Statistical analysis of the fungal community diversity between tissues showed no significant differences. Evaluation of genotypes showed few differences, the IMC67 genotype was significantly different from the others based on the unweighted UniFrac (p < 0.05) and ROS1 based on the weighted UniFrac (p < 0.05) (Table S4).
Isolation of endophytic bacteria and evaluation of their plant growth-promoting capabilities
Bacteria endophytes were isolated from seeds and seedlings to determine their ability to promote plant growth. A total of 35 morphotypes were isolated from seedlings; 45.7% of the bacteria were isolated from the TCS01 genotype, 20.0% from the AC9, 11.4% from the TCS19, and 22.9% from the IMC67 genotype. Only two endophytic morphotypes were isolated directly from seeds of the AC9 genotype. We identified 33 of the isolated bacteria based on their 16S rRNA sequence. Most bacteria belonged to the Pseudomonas (42.4%) and Bacillus (30.3%) genera, while others belonged to Stenotrophomonas (12.1%), Pantoea (9.1%), and 6.1% to other genera (Table S5).
The evaluation of the potential for plant growth promotion of the bacterial isolates was assessed to determine the 1-aminocyclopropane-1-carboxylic acid (ACC) deaminase activity and indole acetic acid (IAA) production. Twelve morphotypes were found to have ACC deaminase enzyme activity with average growth curves in ACC deaminase presence above the growth curve of the MgSO4 control treatment (Fig. 4a). A spectrophotometric calibration curve was done to estimate the IAA concentration. This calibration curve showed an r2 value of 0.99, indicating the method's reliability for estimating the production of IAA of the different morphotypes (Table S5). Among the evaluated morphotypes, P. agglomerans P1901, Pseudomonas sp. P2502 and strain P2104 (unidentified) produced more than 10.0 ppm of IAA, outperforming the control strain Azospirillum brasiliensis SP7, which generated 9.48 ppm after 72 h of growth (Fig. 4b). Furthermore, 11 isolates displayed IAA production levels exceeding 7.0 ppm after 72 h. These findings highlight the potential of these isolates to generate substantial amounts of IAA, suggesting their promising role in plant growth promotion.