The Colombian cacao germplasm collection, managed by the Corporación Colombiana de Investigación Agropecuaria – AGROSAVIA (Colombia), is maintained in vivo at the research center Palmira (3°30′41″N 76°19′19″W) located at 1,001 m.a.s.l, 23°C mean temperature and mean annual rainfall of 1,017 mm, and at the research center La Suiza (7°22′12″N, 73°11′39″W) located at 298 m.a.s.l, 30°C mean temperature and mean annual rainfall of 1,800 mm. The germplasm banks are composed of accessions for conservation purposes as well as additional accessions from working collections. An accession is defined as a distinct and uniquely identifiable sample of a cultivar, breeding line, or population. Each accession is typically represented by multiple tree (up to five) samples propagated vegetatively, by grafting or rootstock, which are considered clones. The two germplasm banks were selected because La Suiza is located in one of the Colombia’s main cacao-producing regions, while Palmira offers contrasting climatic conditions and the possibility to evaluate distinct traits to La Suiza.

In this study, we sampled adult leaves of each tree of the 456 accessions in Palmira and 390 accessions in La Suiza, for a total of 4,653 samples collected ( Supplementary Table 1 ). These samples were stored in hermetically sealed bags with approximately 100 grams of silica gel for transportation to the molecular laboratory at Tibaitatá (4°41′45″N 74°12′12″W) research center of AGROSAVIA. Plant material was washed with water, rinsed with 75% ethanol, and dried with a paper towel. Six to eight leaf disks from each tree were punched and loaded into the 96-well BioArk sampling kits from LGC Biosearch Technologies (https://www.biosearchtech.com/). The sample kits were shipped to the USDA-ARS, Beltsville Agricultural Research Center.

Genomic DNA was extracted by the LGC Genomics services using the sbeadex mini plant kit (LGC Genomics) following the manufacturer’s instructions. We selected 96 SNP markers from a larger database containing over 1,000 SNPs, previously published on cacao research (Allegre et al., 2012; Gutiérrez et al., 2021). The selection criteria for the SNP panel included: distributed across the 10 cacao linkage groups, call rate, minor allele frequency (MAF), and Shannon’s information index. The SNP and the flanking sequences were submitted to LGC Biosearch Technologies for genotyping using a Kompetitive allele-specific array (KASP) ( Supplementary Table 2 ). This method is based on a competitive allele-specific dual FRET-based assay (Cuppen, 2007). Genotype calling was performed using the SNPviewer software (LGC Biosearch Technologies, Hoddesdon, UK).

The quality control of raw data for the 96 SNPs was performed using the assurance module of the SNP Variation Suite software v8 (SVS8; Golden Helix Inc., Bozeman, Montana). SNPs with a call rate greater than 90% were retained in subsequent analyses. We generated a genotype accumulation using the genotype_curve function of the package poppr in R v4.3.0 (Kamvar et al., 2014) to determine the minimum number of markers necessary to discriminate between individuals in a population.

Descriptive statistics, including the Shannon information index (I), observed heterozygosity (H_O), and expected heterozygosity (H_E), were calculated using the whole collection, after filtering genetically identical individuals, and for the core collections. The fixation indexes (F_IT, F_ST, and F_IS) per marker were also calculated. A principal component analysis (PCA) was conducted to elucidate genetic relationships among accessions using Nei’s genetic distance matrix. Finally, molecular analysis of variance (AMOVA) was calculated to assess the amount of genetic variation within and among populations. All the analyses were carried out using GenAlEx v6.503 software (Peakall and Smouse, 2012).

The probability that two individuals share the same multilocus genotype (MLG - a unique combination of alleles observed at multiple genetic loci) is commonly referred to as the probability of identity (PID). The probability of identity among siblings (PID-sib) estimates the likelihood that two randomly selected siblings from a population share the same MLG. Both probabilities were calculated using GenAlEx 6.503 (Peakall and Smouse, 2006, 2012) with established formulas (Waits et al., 2001).

Samples with the same accession name and grouped within the same plot (a defined field area where a plant or group of plants are grown), but showing non-matching SNP patterns, were considered cases of mislabeling. These were identified through multilocus matching using the software GenAlEx 6.503 (Peakall and Smouse, 2006, 2012), following the methodology described in previous studies (Padi et al., 2015; Zhang and Motilal, 2016; Olasupo et al., 2018). A plot heterogeneity was declared when a plot had more than one type of SNP patterns. Individuals with different accession names but with fully matching SNP patterns were considered duplicates. The identification of duplicates was based on the identification of synonymous groups (sets of samples that have the same MLGs), having a maximum genetic distance threshold of 0.05. using the functions mlg.filter and mlg.id in the poppr package (Kamvar et al., 2014).

After examining duplicates, an assignment test was applied to infer the ancestry (hybrids or ancestral forms) of the cacao accessions. For this purpose, a model-based clustering method was implemented in the Structure software v2.3.4 (Pritchard et al., 2000). Data from individuals belonging to the ten cacao reference groups (Motamayor et al., 2008) were included to analyze their ancestral contribution to AGROSAVIA’s germplasm. The detailed list of the 10 reference groups was provided in Supplementary Table 3 . The sample size of each reference group was brought up to 200 using the simulation procedure implemented in the computer program ONCOR (Kalinowski et al., 2007). The simulated populations were then analyzed together with the AGROSAVIA collection. The analysis used a mixed model with a cluster number (K value) of 10, corresponding to the possible genetic groups of cacao present in the AGROSAVIA accessions. Ten independent runs were performed using 100,000 iterations after a burn-in period of 50,000. The run with the highest value of Ln Pr (X|K) of the ten runs was chosen and presented in a bar diagram. The Q value represented each germplasm group’s ancestral contribution.

The core collection was identified using Core Hunter 3 software (De Beukelaer et al., 2018), applying a sampling intensity of 20% to the dataset after removing duplicate samples. The algorithm optimized genetic diversity by maximizing Modified Rogers’ distance among selected accessions, ensuring the core subset captures the broad genetic variation of the collection. The heterozygosity of the core collection and of the complete collection were compared to confirm that represents the overall genetic diversity.

In total, 4,653 trees from 553 accessions were genotyped. The two collections shared 304 accessions, with 162 unique to Palmira and 87 unique to La Suiza. Markers with a call rate below 90% were removed from the initial dataset with 96 SNPs, leaving a final dataset of 77 SNPs for 2,597 accessions from Palmira and 2,056 from La Suiza ( Supplementary Tables 1 , 4 ).

The genotype accumulation curve showed a tendency to reach a plateau and had a greatly decreased variance with 77 SNPs, indicating that there were enough markers to identify 100% of the 2,630 MLGs ( Figure 1 ). The diversity indices were similar in both collections ( Table 1 ). The I index quantifies the amount of genetic diversity in a population based on the frequencies of alleles observed across the genetic loci analyzed. Presented a mean value of 0.588 (Palmira = 0.593 and La Suiza = 0.583). Among the 77 SNP markers, the H_O ranged from 0.106 to 0.662, with an average of 0.399 for the two germplasm banks. The Palmira collection showed a mean value of 0.394, and La Suiza had a mean value of 0.404. The H_E ranged from 0.199 to 0.500, with an average of 0.404. The Palmira collection presented a mean value of 0.408, and La Suiza had a mean value of 0.4. After removing the duplicate samples (2,472), the collection consisted of 2,181 samples, with the I and He indexes slightly higher, while the Ho index was lower.

For F statistics, most loci showed an excess of heterozygosity ( Table 2 ), and the F_ST index indicated a low level of differentiation between the two germplasm banks. This is confirmed by the fact that F_IS and F_IT have almost the same value. We obtained a mean value of 0.041 for F_IS, 0.042 for F_IT, and 0.001 for F_ST. AMOVA was employed to assess the distribution of the observed genetic variance between the two collections and showed that genetic variance within accessions contributed 100% to genetic diversity ( Table 3 ), while the variance among populations was 0% of the total variance. This demonstrates that the observed genetic variations primarily arise from variation among individuals within each germplasm bank rather than between different germplasm banks.

To visualize the relationships between the accessions of T. cacao, a PCA was done from Nei’s genetic distance matrix. The first two PCA coordinates accounted for 26% of the variability ( Figure 2 ). In the graph, two general clustering trends can be observed.

The combined PID-sib of the 77 SNP panel was 1.203 x 10–¹⁵ and ranged from 5.95 x 10–¹ to 7.3 x 10^-1. In addition, the PID based on the combination of 77 SNP loci was 1.479 x 10^-29, indicating an extremely low probability of two unrelated individuals sharing the MGL by chance ( Table 4 ).

The results of pairwise multilocus matching showed that a total of 172 groups with 577 samples had intra-plot mislabeling (12.4% of the total samples), demonstrating a medium rate within the collections ( Supplementary Table 5 ). In addition, the results revealed a total of 406 synonymous groups involving 2,472 samples (53.1% of the total) that were identified as duplicates in the AGROSAVIA germplasm banks. Palmira had the highest number of duplicates (1,366), while La Suiza had 1,106 duplicates. The number of duplicated individuals within each group ranged from 2 to 264, with 134 groups consisting of two individuals each. Detailed information about the identified duplicates is listed in Supplementary Table 6 .

Among the 4,653 cacao samples analyzed, all ten known reference genetic groups were identified. However, only six were predominant within the AGROSAVIA collection. Most accessions exhibited admixture ancestry, with the following distribution: Amelonado (28.6%), Criollo (22.6%), Iquitos (13%), Nacional (10.2%), Contamana (7.2%), Marañon (7.1%), Nanay (5.7%), Purus (3.6%), Curaray (1.3%), and Guiana (0.7%) ( Figure 3 ). The last four groups were present at notably low levels.

Figure 4 illustrates the cacao trees belonging to the seven most representative genetic groups observed in our study. Each group, selected for its distinct genetic characteristics, is displayed within the field setting to emphasize the diversity of the cacao population across the studied regions. The visual representation of these genetic groups allows for a clearer understanding of their spatial distribution and potential implications for cacao breeding and conservation efforts.

Using the Core Hunter 3 software, core collections were determined from an initial dataset of 950 cacao samples for La Suiza and 1,223 cacao samples for Palmira. The resulting core subset consisted of 190 and 246 samples, respectively, that captured the full genetic variability present in the original population ( Supplementary Table 7 ). All diversity indices were higher compared to the original collections even when duplicates were removed ( Table 1 ). The I index showed mean values of 0.619 for Palmira core collection and 0.606 for La Suiza. The H_O averaged 0.418 and 0.421, while H_E was 0.431 and 0.419, for Palmira and La Suiza core collections, respectively.

Historically, cacao has benefited from an extensive exchange of germplasm material between genebanks across regions worldwide and within countries (Gopaulchan et al., 2019; Rodriguez-Medina et al., 2019). This movement of germplasm has facilitated the domestication of the species, particularly in South America, likely driven by cultural interactions between the Amazon and the Pacific coast over 5,000 years. The development of several landraces shaped by human activity allowed hybridization between cultivars and wild relatives, favoring the adaptation of T. cacao to new environments and shaping the genetic basis of the present-day cacao (Lanaud et al., 2024). While hybridization still occurs naturally and through controlled breeding programs, cacao germplasm is maintained by vegetative propagation under field conditions (Rodriguez-Medina et al., 2019). However, the continued exchange of plant materials, often conducted without standardized identifiers, consistent labeling, or complete metadata, represents ongoing challenges for accurate accession tracking (Singh et al., 2019). Consequently, new accessions often lack accurate information regarding their origin. This issue is not unique to cacao and affects numerous germplasm banks worldwide, highlighting the urgent need to improve identity management systems (Bakkali et al., 2019; Migicovsky et al., 2019; de Oliveira et al., 2020). The significance of our study lies in its contribution to preserving a meticulously categorized collection with minimal genetic redundancy, enhancing accuracy, and reducing maintenance costs (Panis et al., 2020). Additionally, our findings emphasize the importance for breeders to understand the molecular diversity within germplasm collections. Such understanding can assist breeding strategies aimed at developing improved new cultivars with higher yields, better quality attributes, and greater resilience to both biotic and abiotic stresses.

The required number of markers for accurately identifying cacao germplasm and assessing genetic diversity within a genebank depends on the species’ genetic variability, sample size, and the specific molecular marker used (Alhasnawi et al., 2024). In species like T. cacao, which exhibit significant genetic diversity, it is essential to use a substantial number of markers spread across the genome to effectively detect allelic variations. In the present study, we selected a set of 96 SNP markers that have been used in previous studies (Lukman et al., 2014; Cosme et al., 2016; Lindo et al., 2018; Wang et al., 2020; Dillon et al., 2024). After applying quality control filters, 77 SNPs were retained for analysis, from which we achieved a robust representation of the genetic groups identified for cacao and obtained diversity values comparable to studies that used a larger number of SNPs (Osorio-Guarín et al., 2020). Regardless of the marker type or species diversity, it is crucial to determine the optimal number of markers needed to assess the genetic diversity and accession identification. In our study, the genotype accumulation curve used to establish this optimal number indicated that the selected markers were adequate for distinguishing the cacao accessions, revealing 2,630 unique MLGs among the 4,653 samples.

This study genotyped the largest number of cacao samples to date, with 4,653 cacao trees representing the entire cacao collections maintained in AGROSAVIA. The H_O value is comparable or higher than previous SNP-based diversity assessments reported for cacao (Ji et al., 2013; Takrama et al., 2014; Cosme et al., 2016). The genetic diversity values indicated that both collections, from Palmira and La Suiza, are very similar in their diversity. The Palmira germplasm collection is more diverse in H_E, possibly due to the highest number of individuals (2,597) compared to the La Suiza germplasm bank (2,056 individuals) and because the two collections differ in their working cacao accessions. However, they conserved mainly the same accessions because La Suiza germplasm bank originated as a copy of the Palmira germplasm bank.

Other studies on cacao using KASP technology found similar heterozygosity values. Olasupo et al. (2018) found H_O = 0.248 and H_E = 0.413 using 63 SNPs across 1,457 samples from Nigeria’s cacao breeding programs. They also detected a high rate of mislabeling among recently introduced international germplasm in Nigeria. Bhattacharjee et al. (2023), analyzed 376 samples with 20 SNPs using KASP and observed lower H_O and H_E values compared to our study. However, they also reported high mislabeling rates in African countries like Sierra Leone (33.33%) and Togo (45.45%), which is comparable to our findings.

The present results suggested that the Colombian cacao collections have a high level of allelic diversity, as measured by heterozygosity and AMOVA. However, a gap in genetic diversity was detected, particularly in terms of the underrepresentation of known Amazon cacao populations. In addition, approximately 30% of the accessions in the collection belong to the Trinitario type, which is significantly over-represented and exhibits a high level of genetic redundancy due to the presence of duplicates. The analysis also revealed mislabeling in the introduced international genotypes. Several international accessions labeled as Parinari, Scavina, and IMC were not genetically aligned with their expected reference groups, indicating discrepancies in their identity. Similar patterns were observed for collections in both locations, confirming the similarity between Palmira and La Suiza germplasm banks. Therefore, the La Suiza collection can be considered a backup or reserve collection for Palmira. This finding is further supported by PCA analysis, which shows similar clustering patterns for both collections, with differences potentially explained by the specific cultivars developed by each respective plant breeding program.

The AMOVA results indicated that the majority of genetic variation occurs within collections rather than between them, which is expected given the presence of shared accessions. Although maintaining duplicate collections might seem cost-inefficient, preserving backup germplasm at two distinct sites offers significant advantages. The contrasting environmental conditions in Palmira and La Suiza allow for the evaluation of the same materials under varying climatic and disease pressures.

Notably, Palmira and La Suiza differ in their genetic composition. Palmira holds a greater number of accessions with mixed ancestry compared to the La Suiza genebank. In contrast, La Suiza contains fewer accessions with Amelonado and Iquitos ancestry, although both genebanks maintain comparable numbers of Contamana and Marañon ancestry accessions. Palmira’s broader diversity, particularly in mixed ancestries, provides a wider genetic base for selecting traits such as disease resistance and yield. Meanwhile, La Suiza’s narrower diversity may limit genetic variability but could represent a more focused genetic pool ideal for specific breeding objectives. The germplasm conserved at La Suiza has been evaluated for several agronomic traits, including disease resistance and productivity (Osorio-Guarín et al., 2020). These genetic distinctions between the two genebanks offer a rich and complementary spectrum of resources for breeding efforts.

The genetic resources maintained in national genebanks have been effectively utilized to develop and release local cocoa cultivars adapted to Colombia’s diverse agroecological conditions. AGROSAVIA conducted regional agronomic evaluations of eight selected cacao genotypes across four sites in the Montaña Santandereana and Magdalena Medio subregions. Among these, TCS 13 and TCS 19 demonstrated superior physiological and agronomic performance, including high photosynthetic capacity, improved water-use efficiency, resistance to moniliasis, and self-compatibility. These traits contributed to high yields (1.5–1.8 kg tree¹ year¹) under agroforestry systems, leading to their registration as commercial cultivars in the Colombian National Register in 2017 (Agudelo-Castañeda et al., 2017, 2018, 2023). In addition, AGROSAVIA released in 2014, the cultivars TCS 01, characterized by high self-compatibility and robust yields, and TCS 06, which has strong resistance to FPRD (Suárez et al., 2022). Finally, this dual-site conservation and breeding approach provides a solid foundation for advanced breeding strategies, including genomic selection. It also supports the application of the Focused Identification of Germplasm Strategy (FIGS), which leverages environmental and passport data to identify germplasm with desirable traits. The existence of duplicate collections in distinct environments can enhance the effectiveness of FIGS by providing comparative data sets that improve the accuracy of trait prediction and targeted germplasm mining (Sunitha et al., 2024).

The genetic groups identified using the 77 SNPs panel showed concordance with those previously studied using SSR markers by Motamayor et al. (2008), demonstrating the usefulness of this SNP panel for assessing genetic diversity and population structure in cacao. However, the population structure analysis revealed that the genetic groups Nanay, Curaray, Purus, and Guiana are underrepresented in AGROSAVIA’s collections. The limited number of accessions from these groups belonging to the Amazon genetic groups within the collection fails to capture the comprehensive intra-population genetic diversity of T. cacao. To address this gap, it is imperative to introduce new accessions from international germplasm banks belonging to these specific genetic groups. It is important to note that population assignment in this study was analyzed using the framework of 10 genetic cacao populations (Motamayor et al., 2008). As a result, the analysis was unable to detect accessions that may represent new genetic groups. Such undetected diversity has been highlighted in recent expeditions and studies (Osorio-Guarín et al., 2017; Argout et al., 2023). The population structure of this collection will be further analyzed, including recently collected wild cacao (Argout et al., 2023; Zhang and Motilal, 2016) as reference.

The present results highlighted the need for corrective actions to address mislabeling within the AGROSAVIA cacao collection. Using a plot heterogeneity strategy, we assessed the occurrence of mixed genotypes within field plots and identified a high incidence of mislabeling, defined as two distinct accessions sharing identical SNP patterns. In addition, we identified a high rate of duplicate accessions that must be addressed to reduce genetic redundancy within the collection. The problem of multiple individuals having the same genotype (SNP pattern) at all loci can be addressed by increasing the number of loci in a study, thereby decreasing the probability that two individuals have the same MGL. We used an MLG strategy along with the PID, a measure that two individuals in a population share the same multilocus genotype. PID depends on the number of loci analyzed, the level of allelic diversity, and the degree of relatedness among individuals in the population. To minimize the potential errors in estimating PID, we used the PID-sib metric, which provides a conservative upper bound for determining the number of loci necessary to reliably distinguish individuals (Waits et al., 2001). In this study, both values, PID and PID-sib, demonstrated the ability of the SNP markers to discriminate individuals. Notably, our PID value (1.479 x 10^-29) was significantly higher than the PID reported by Olasupo et al. (2018), who obtained a value on the order of 10⁻⁶. This result demonstrates the good resolution of our 77 SNP panel to verify the true-to-type trees.

Our results highlight a critical issue of genetic misidentification and redundancy that requires immediate attention. Such inconsistencies may originate from errors during the initial establishment of the genebank or the introduction of new accessions. Mislabeling is a common problem in clonal propagation, where seedlings may be mistakenly transplanted instead of true-to-type clones, especially accessions acquired from older expeditions (Turnbull et al., 2003; Motilal et al., 2013). Additionally, planting or grafting incorrect accessions in field plots could contribute to the loss of genetic integrity. In many cases, material was collected in the form of pods (seeds) rather than vegetative cuttings, increasing the likelihood of confusion due to regional variations in accession names.

To improve the accuracy and management of field collections, we recommend permanent labels using quick response (QR) along with routine analysis of molecular and phenotypic data. These steps are essential to ensure the long-term reliability and functionality of the gene bank, and their successful implementation will require close collaboration of germplasm curators. The high rate of intra-accession mislabeling and genetic redundancy found in AGROSAVIA germplasm establishes a strong foundation for revisiting and correcting existing records. Other strategy to address this issue is the conformation of core collections. In fact, we found an increase in diversity indices of defined core collections highlighting its effectiveness in enriching genetic variation while reducing redundancy. This improvement is particularly valuable for cacao breeding programs, as it ensures a smaller, more manageable subset that can be evaluated for agronomical traits retaining the genetic breadth of the original collection. These core collections could provide a robust foundation for selecting parent lines with desirable traits and enhance the potential for genetic gains through both conventional and molecular breeding approaches (Gu et al., 2023).