Phenotyping the genetic diversity of C. canephora germplasm collections, especially in the geographical origins of the species, is an important first step in discovering promising genetic material for developing new coffee varieties. Our exploration of the morphological diversity and agronomic potential of the understudied C. canephora genetic resources from the DRC revealed a substantial variation in morphological traits. The yield traits exhibited the most variation, while the coffee cherry and seed dimensions exhibited the least variation. The variation in morphological traits was not consistent with the genetic background, and no morphological traits were found to classify and predict the genetic background of the genotype accurately. The variation in agronomic and quality-relevant traits was substantial, with the ‘Lula’, ‘Lula’ – Wild hybrid, and ‘Lula’ – subgroup A hybrid material containing promising genotypes for further evaluation.
Regarding the morphological diversity of the genotypes, substantial variation in all traits was found, especially in the yield traits. Akpertey et al. (2019) studied 71 genotypes from germplasm in Ghana and reported similar coefficients of variation for the leaf and coffee cherry dimensions. Similarly, they reported a higher variation in tree architecture and branch traits compared to the coffee leaf and cherry traits. Previous research on Conilon varieties (Congolese subgroup A) reported less variation for anther and corolla tube length compared to our study but similar positive correlations between petal, anther, and corolla tube lengths (Silva et al. 2021). We found a substantial range in SLA and stomatal density across the studied genotypes but this variation was not consistent with the genetic background of the respective genotypes. Dubberstein et al. (2021) also reported a wide morphological diversity in stomatal density for Conilon genotypes. In our study, no significant correlation was found between SLA and coffee cherry production or between stomatal density and coffee cherry production. In comparison, previous research reported a correlation between specific leaf area and coffee plant yield (Gagliardi et al. 2015). In line with Pompelli et al. (2010) on C. arabica and also previously reported in other plant species (Xu and Zhou 2008; Sun et al. 2014), we observed a negative correlation between SLA and stomatal density. The environmental plasticity of these two functional leaf traits likely has a genetic basis (De Kort et al. 2020) and plays an important role in the climate adaptation of food crops (Bertolino et al. 2019). Genotypes with a low SLA and stomatal density can be expected to have a fitness advantage under dry environmental conditions (Poorter et al. 2009; Bertolino et al. 2019). This trait combination was found in the ‘Lula’ – Wild hybrids G0023 and G0067 (Supplementary Table S1), which both, interestingly, also scored a high sensory quality, with G0067 exhibiting a large screen size (Fig. 4).
Accurate prediction of the genetic background of the genotypes based on their morphological traits was not achieved. Cross-validation of the LDA model revealed a low prediction accuracy (43.33%) for the constructed linear discriminants. The classification errors for genetic background in both the LDA and random forest models were high, and the reported morphological traits of importance in the two classification models were not the same. No combination of morphological traits was able to accurately classify an accession according to its genetic background, consistent with previous research on Robusta genotypes. Akpertey et al. (2019) reported high coefficients of variations for most studied traits in Robusta genotypes, with origins in Togo, Ghana, Cameroon, and the Ivory Coast and the grouping of the genotypes based on morphological traits was inconsistent with the genetic background. Similarly, Robusta genotypes from Uganda exhibited morphological variation but this was not predictive for their genetic background (Ngugi and Aluka 2019). Advancements in molecular research have increased the efficiency and cost-effectiveness of genotyping plant material, making it easier to analyze the genetic composition of Robusta coffee's genetic resources. However, classifying Robusta genetic material based on morphological traits still requires further evaluation, as it could provide a practical method for screening genetic resources for desirable traits. Additionally, access to molecular research is often limited in the regions where Robusta's genetic resources are found.
A substantial variation in traits linked to agronomic performance was observed across the genotypes. If the green bean traits and sensory quality would be prioritized, promising material was found within the ‘Lula’ – Wild hybrid and ‘Lula’ material. Genotypes G0067 and G0035 from the ‘Lula’ – Wild and genotype G0077 from the ‘Lula’ class had a total sensory quality score of 83 points or higher, and were characterized by large coffee beans (> 50% retained by screen 17), a high 100 bean weight, and a moderate peaberry rate. The relatively large coffee bean size of the ‘Lula’ – Wild material, combined with their high sensory quality scores, indicates the potential of this hybrid material for further breeding efforts. These genotypes exhibited a tree height above three meters, an agronomic trait that is less desired but can be managed through adapting cultivation practices. As a comparison, screen sizes of Ugandan Robusta accessions averaged around screen 15, but with a high level of morphological and bean size variability (Ngugi and Aluka 2019). Also genotype G0240 from the ‘Lula’ class is noteworthy, as it had the largest screen size distribution in the studied material, with good coffee production and sensory quality.
If agronomic production would be prioritized, promising material was found within the ‘Lula’ – subgroup A class. The PCA indicated that the ‘Lula’ – subgroup A hybrids were associated with higher coffee cherry production. More specifically, genotypes G0086 and G0010 had the highest total cherry production across two harvest years, with G0010 recording a maximum node of 52 coffee cherries. Genotype G0102 recorded the most fructifying nodes, which correlates with cumulative coffee production over time (Cilas et al. 2006). The relatively high number of fructifying nodes in these ‘Lula’ – subgroup hybrids is an interesting characteristic. Congolese subgroup A material has its geographical origin in Gabon, the Republic of the Congo, and western DRC, from which the ‘Petit-Kwilu’ variety and the Conilon material were derived. These coffee varieties exhibit high coffee production but small coffee bean dimensions, whereas the ‘Lula’ material is known for its larger green bean size. Genotype G0087 (subgroup A) originated from the Luki region of western DRC and was labeled as ‘Petit-Kwilu’ material, exhibiting a small coffee bean size. Hybridization between the Congolese subgroup A and ‘Lula’ material could result in coffees with a high productivity and larger bean size, as was suggested by genotypes G0222, G0220, and G0010. On the other hand, genotype G0073 from this hybrid class scored exceptionally high in Total score and performed well in agronomic production but lagged in coffee bean size. These results indicate a promising starting point for future breeding activities between these Congolese subgroup A and ‘Lula’ material. Inter-varietal hybridization between Congolese subgroup A (‘Petit-Kwilu’) and subgroup BE (‘Lula’) for the development of drought-resistant varieties with large beans was initiated at INERA (Montagnon et al. 1998) but it was never completed. Our ‘Lula’ – subgroup A hybrids are possibly the descendants of these initial inter-varietal crossing experiments. The direct parents of these hybrids are not known, so the level of heterosis could not be evaluated. Our results indicate that specific genotypes from ‘Lula’ – subgroup A class exhibited large bean sizes and performed well on sensory quality. It would be interesting to evaluate their performance in dry environments as a considerable range in SLA and stomatal density was reported in this hybrid class (Supplementary Table S1). The discovered genotypes with promising morphological traits could be used as parent plants in breeding programs.
In this context, no hybrid crossings between the Wild and Congolese subgroup A were found in the INERA Coffee Collection (Verleysen et al. 2023). Future research could undertake targeted crossings between these two genetic classes, as the genetic distance of the parental material correlates with phenotypic performance (Ferrão et al. 2024a). Previous research on hybrid crossings between Conilon (Congolese subgroup A) and a Robusta variety (Congolese subgroup E) reported better agronomic and yield performance, with a high expression of heterosis in the hybrid material (Teixeira et al. 2017; Carvalho et al. 2019; Alkimim et al. 2021). Similar hybrid vigor observations were made in the hybrid material of Ghanaian accessions (Akpertey et al. 2022) and Conilon varieties of Brazil (Ferrão et al. 2024a). These breeding experiments mainly focused on the already established genetic resources from the same origin groups. The ongoing introduction of local wild materials from the Yangambi rainforest into the INERA Coffee Collection will result in newly captured genetic diversity available for crossing experiments. These hybrids could potentially exhibit a desirable combination of traits.
The multiplication of accessions at the INERA Coffee Collection through seedlings and open pollination resulted in the hybridization of the initial genetic material into many new and unique genetic fingerprints (Verleysen et al. 2023). This resulted in a substantial variation in sensory quality, and promising sensory profiles within the different genetic classes of the collection were discovered (Bollen et al. 2024). Similar observations were made in terms of agronomic potential. Promising material was found within the ‘Lula’ – Wild hybrid, ‘Lula’, and ‘Lula’ – subgroup A hybrid material. Ngugi and Aluka (2019) reported similar observations in agronomic traits from crossings between cultivars and local wild material in Ugandan germplasm. The Wild and Congolese subgroup A material did not perform well, but their sample sizes were too small to draw solid conclusions. A more comprehensive screening of the INERA Coffee Collection is advised in this context, as only 70 of the 263 unique genetic identities reported in Verleysen et al. (2023) were studied. Furthermore, the initial genetic analysis included only a fraction of the collection’s total genetic resources, leaving many opportunities for further exploration. Additionally, the continuous introduction of accessions from plantations and wild populations expands the available genetic resources for phenotypic screening (Verleysen et al. 2024).