# Source paper archive - Source title: Genomic prediction of resistance to Phytophthora megakarya in cacao from a sparse host-pathogen screen - Author(s): Schlathoelter, I.; Gitto, A. J.; Arciniegas, A. L.; Herrera Corzo, M.; Kolawole, O. O.; Goss, E. M.; Livingstone, D.; Firl, A.; Henderson, D. R.; Drost, D. R.; Rodezno, D.; Marelli, J.-P.; Brawner, J. T. - Source landing URL: https://www.biorxiv.org/content/10.64898/2026.11.03.764016v1 - PDF URL: https://www.biorxiv.org/content/10.64898/2026.11.03.764016v1.full.pdf - PDF SHA-256: 6df7fee0c6ee66384fb3c1fc8a7fe9e80a0d4fc3ba7fc85a5c19efe404127e66 - Converter: pymupdf4llm 0.0.27 - Limits: Automated PDF text extraction; reading order, equations, tables and formatting may be incomplete or incorrect. Images are omitted, and no OCR was performed. Markup and links are neutralized for safe downloading. Consult the original PDF for the authoritative version. --- [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) # **Genomic predic�on of resistance to Phytophthora megakarya in cacao from a** **sparse host-pathogen screen** Ina Schlathoelter [1], Andrew J. Gi�o [1], Adriana L. Arciniegas [2], Mariana Herrera-Corzo [1], Oluwaseun Olawale Kolawole [1,3], Erica M. Goss [ 1], Donald Livingstone [ 4], Alana Firl [ 4], Donna R. Henderson [ 5], Derek R. Drost [ 4], Dayana Rodezno [ 4], Jean-Philippe Marelli [ 4], Jeremy T. Brawner [1] 1 Department of Plant Pathology, University of Florida, Gainesville, FL 32611, USA 2 Centro Agronómico Tropical de Inves�gación y Enseñanza, Turrialba, Costa Rica 3 Cocoa Research Ins�tute of Nigeria, Ibadan, Nigeria 4 Mars Plant Science Center, Mars Snacking, Davis, CA 95616, USA 5 Procurement Sustainability, McCormick & Company, Hunt Valley, MD 21031, USA Corresponding Author: Jeremy T. Brawner Submi�ng author and contact: Ina Schlathoelter 1 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) **Abstract** Black pod rot caused by _Phytophthora megakarya_ is a major disease impac�ng cacao ( _Theobroma_ _cacao_ L.) produc�on in West Africa. Resistance is influenced by gene�c varia�on in both host and pathogen popula�ons. While genomic predic�on approaches have been widely applied in plant breeding, most studies incorporate only host genomic informa�on. The value of explicitly modeling pathogen genomic relatedness in quan�ta�ve disease predic�on remains largely unknown. We genotyped 947 clones from the Interna�onal Cacao Collec�on at CATIE using a 10K-SNP panel and characterized 67 _P. megakarya_ isolates by sequencing 3,505 virulence-associated genes and 2,774 SNP loci. Cacao clones were extensively admixed, while isolates formed three geographically structured subpopula�ons. Mixed linear models were fi�ed to lesion diameters on detached pods of 350 genotyped clones inoculated with 58 isolates in 30 bioassays, covering 5% of possible clone-by-isolate combina�ons. Differences among isolates explained a larger share of lesion-size varia�on than differences among clones (0.34 and 0.27 of phenotypic variance, respec�vely). A�er accoun�ng for broad pathogen subpopula�on structure, incorpora�ng within-subpopula�on genomic relatedness among isolates improved neither model fit nor predic�on, whereas incorpora�ng genomic rela�onships among cacao clones improved model fit and enabled predic�on of phenotypically untested clones with a predic�ve ability of 0.41. Clone rankings were stable across pathogen subpopula�ons (genomic correla�on 0.86 ± 0.06), whereas clone-by-isolate interac�ons accounted for li�le phenotypic varia�on and were not resolved with confidence in the sparse design. An exploratory ancestry analysis associated several Upper Amazon ancestry components with smaller predicted lesions. These are the first genomic predic�ons of resistance to _P. megakarya_ in cacao. In this strongly structured host-pathogen system, genomic rela�onships among hosts carried most of the predictable signal beyond broad pathogen subpopula�on effects, and the predic�ons provide a star�ng point for priori�zing screening and selec�on efforts. **Keywords** Preven�ve Breeding, Genomic Predic�on, Disease Screening, _Theobroma cacao_, _Phytophthora_ _megakarya_, Black Pod Rot, Oomycetes, Host-Pathogen Interac�on Genomics 2 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) **Introduc�on** Cacao ( _Theobroma cacao_ L.) beans are the raw material of the global chocolate industry, valued at approximately US$167 billion in 2024 (IMARC Group, 2024). Origina�ng in the upper Amazon region of South America (Zarrillo et al., 2018), cacao subsequently became established as a major crop throughout tropical regions worldwide. Africa accounts for approximately 70% of global cocoa produc�on, with Côte d’Ivoire and Ghana together contribu�ng nearly half of the global supply (Interna�onal Cocoa Organiza�on, 2025). Cocoa produc�on is predominantly carried out by smallholder farmers, with more than two million producers es�mated in Africa (Kissi & Herzig, 2024). Cacao is a key cash crop in the humid tropics, suppor�ng the livelihoods of millions of people worldwide, historically es�mated at 40–50 million (CacaoNet, 2012). Cacao produc�on is threatened by several diseases. Black pod rot, caused by several species of the genus _Phytophthora_, is one of the most destruc�ve bio�c threats to cacao and accounts for half of reported global yield losses (Marelli et al., 2019). Among the causal agents, _Phytophthora_ _megakarya_ is par�cularly aggressive and currently restricted to West and Central Africa (Wessel & Quist-Wessel, 2015). It can result in up to 90% yield loss without control (Bailey et al., 2016; Nyassé et al., 1999). The oomycete pathogen most likely originated in Cameroon (Mfegue et al., 2012; Nyassé et al., 1999) and subsequently spread westward through West Africa, reaching Côte d’Ivoire by the early 2000s (Akrofi et al., 2015). While _P. megakarya_ has largely displaced _P._ _palmivora_ as the dominant black pod pathogen in Cameroon and Nigeria, both species con�nue to occur in Ghana and Côte d’Ivoire (Ali et al., 2016, 2017). Poten�al biological advantages contribu�ng to the compe��ve success of _P. megakarya_ include earlier and more abundant sporula�on, rapid appressorium forma�on facilita�ng infec�on of unwounded pods, and prolonged survival in soil (Ali et al., 2016; Bailey et al., 2016; Ward & Griffin, 1981). The con�nued expansion of _P. megakarya_ across West and Central Africa presents significant challenges for sustainable disease management. An important effort to make cacao produc�on more sustainable and resilient to diseases is the development of resistant varie�es through breeding. As in other perennial crops, gene�c improvement in cacao is slow. With a juvenility phase of 3-5 years, a single selec�on cycle can take more than a decade. Two or more cycles are usually necessary before releasing a new variety (Kulesza et al., 2024). Breeding for durable disease resistance is further complicated by gene�cally structured pathogen popula�ons, in which isolates can differ in aggressiveness but are rarely characterized alongside the host material. Understanding the gene�c determinants underlying host resistance and pathogen virulence is therefore essen�al. Resistance to _Phytophthora_ spp. in cacao is predominantly quan�ta�ve and polygenic, with a substan�al addi�ve gene�c component (Ali et al., 2017; Nyassé et al., 2007). Although cacao genotypes differ considerably in their suscep�bility to _P. megakarya_, complete immunity has not been demonstrated, and resistance is generally expressed as reduced disease severity (Nyassé et al., 2007; Pokou et al., 2019). Large _ex situ_ collec�ons, such as the Interna�onal Cacao Collec�on (IC3) at the Centro Agronómico Tropical de Inves�gación y Enseñanza (CATIE), conserve diverse 3 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) cacao gene�c resources and provide an important pla�orm for iden�fying and characterizing novel sources of resistance (Engels, 1981; Fister et al., 2020; Mata-Quirós et al., 2018). These collec�ons encompass representa�ves of the major gene�c groups of cacao (Motamayor et al., 2008; Todd et al., 2025; Zhang et al., 2012) and capture both dis�nct ancestral lineages and extensive admixture. Comprehensive disease phenotyping of large germplasm collec�ons is logis�cally challenging and expensive, par�cularly when the pathogen is a quaran�ne organism that must be handled under containment away from the collec�on site. In those cases, only a limited subset of host–isolate combina�ons can realis�cally be assayed. Being able to confidently use those sparse, unbalanced datasets would substan�ally accelerate the discovery of poten�al new sources of disease resistance and guide targeted screening efforts. Genomic predic�on comprises a family of methods that rely on the associa�on between phenotypes and genotypes in a training popula�on to predict genotype performance in a related set of individuals from gene�c markers alone (Meuwissen et al., 2001). It does not rely on the direct iden�fica�on of causal polymorphisms and can be highly effec�ve for selec�ng on polygenic traits or traits with significant genotype-by-environment interac�on. In plant breeding, genomic predic�on approaches have demonstrated considerable poten�al for predic�ng complex quan�ta�ve traits, including disease resistance, and may improve breeding efficiency by enabling the selec�on of genotyped individuals without direct phenotypic evalua�on (Enciso-Rodriguez et al., 2018; Lorenz et al., 2012; Poland & Rutkoski, 2016; Rolling et al., 2020; Rutkoski et al., 2014). Although genomic predic�on is widely used in plant and animal breeding programs, its applica�on in cacao remains limited (Baek et al., 2025, 2026; Bekele et al., 2022; McElroy et al., 2018; Romero Navarro et al., 2017). A genomic-predic�on study of black pod resistance in cacao evaluated resistance under low natural _P. palmivora_ disease pressure (Romero Navarro et al., 2017). They evaluated 148 clones for eight years under natural infec�on and es�mated a broad-sense heritability of 0.52 and genomic predic�on accuracy ranging from 0.46 to 0.51 (Romero Navarro et al., 2017). A second study evaluated genomic predic�on of black pod rot resistance in a popula�on of 495 cacao progeny derived from eight interconnected biparental crosses at CATIE, using ar�ficial pod inocula�ons with _P. palmivora_ (Jacquo�e et al., 2026). Genomic predic�on achieved predic�ve abili�es of 0.45–0.46 under fivefold cross-valida�on, which decreased to 0.36 when en�re families were excluded from the training popula�on. The study further demonstrated that predic�ve ability increased with training popula�on size, whereas gains from addi�onal SNP markers were minimal beyond approximately 1,000 markers (Jacquo�e et al., 2026). Furthermore, the explicit incorpora�on of genomic informa�on from both the host and the pathogen remains uncommon in plant disease predic�on (Domínguez-Rondón et al., 2025; Hudson et al., 2024; Hudson & Brawner, 2025). Hudson et al. (2024) integrated the pathogen gene�c rela�onship matrix through reac�on norm models (Jarquín et al., 2014) and dual-genome similarity matrices, which provide a means to par��on phenotypic variance into components a�ributable to host gene�c effects, pathogen gene�c effects, and their interac�on. This strategy was applied to _Fusarium_ ear rot in maize and _Septoria tri�ci_ blotch in wheat. In both systems, 4 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) pathogen genotypes explained part of the varia�on (Domínguez-Rondón et al., 2025; Hudson et al., 2024; Hudson & Brawner, 2025). Because pathogen gene�c varia�on can substan�ally contribute to disease phenotypes, jointly modeling genomic rela�onships in both the host and the pathogen is an appealing extension of conven�onal host-only genomic predic�on. Recent genomic resources for _P. megakarya_, including a reference genome assembly and annota�on of infec�on-related gene families such as secreted RxLR effectors, Crinkler proteins, necrosis-inducing proteins, proteases, detoxifica�on enzymes, and carbohydrate-ac�ve enzymes, provide new opportuni�es to characterize pathogen diversity at high resolu�on (Morales-Cruz et al., 2020). Popula�on-level genomic analyses have revealed pronounced gene�c structure within _P. megakarya_, o�en associated with geographic origin (Mfegue et al., 2012; Morales-Cruz et al., 2020; Nyassé et al., 1999). However, the extent to which genomic differen�a�on among pathogen subpopula�ons translates into varia�on in virulence, and thereby influences quan�ta�ve host resistance, remains uncharacterized. Here, we integrate genome-wide SNP data from the IC3 cacao collec�on with target capture based sequencing of virulence-associated genomic regions in a geographically diverse panel of _P._ _megakarya_ isolates to characterize black pod rot resistance. The goals of this study were to: 1) study the gene�c varia�on of the IC3 cacao collec�on and the _P. megakarya_ popula�on, 2) proac�vely iden�fy poten�al host resistance sources, and 3) test whether incorpora�ng pathogen genomic varia�on improves genomic predic�on of host resistance. As one of the two interna�onal cacao collec�ons, the IC3 at CATIE in Turrialba, Costa Rica, maintains more than 1,100 cacao clones from Central America, South America, the Caribbean, Asia and Africa, although wild genotypes from the upper and lower Amazon are predominant (Mata-Quirós et al., 2018). The pathogen capture panel was designed to interrogate genes implicated in infec�on processes, including effector repertoires and enzyma�c gene families involved in host �ssue coloniza�on. Together, these marker resources enabled gene�c characteriza�on of both the host and pathogen popula�ons. Resistance was phenotyped as lesion diameter on unwounded, detached pods four days a�er zoospore inocula�on, using a sparse design comprising 1,238 cacao pods from 350 clones and 58 _P. megakarya_ isolates. Detached-pod and leaf-disc tests both correlate with black pod rot field resistance at the level of genotype means (Iwaro et al., 2005; Tahi et al., 2006) and neither consistently outperforms the other. We used mixed linear models to evaluate how genomic informa�on affected variance component es�mates and the predic�on of lesion size. The models progressively incorporated genomic rela�onship matrices for cacao clones and _P. megakarya_ isolates, followed by a dual-genome similarity matrix to account for host–pathogen interac�ons. These analyses were conducted under a sparse screening design, in which the availability of cacao pods largely determined which clones could be evaluated in each bioassay. We further characterized the popula�on structure of both host and pathogen collec�ons, assessed the stability of host performance across pathogen subpopula�ons, and explored ancestry-based trends in predicted resistance as a hypothesis genera�ng framework for breeding. 5 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) By explicitly tes�ng genomic covariance structures for both host and pathogen, this study evaluates when dual-genome informa�on improves predic�on in a sparse perennial-crop disease screen. More broadly, it demonstrates how dual-genome modeling approaches can be implemented in perennial crop pathosystems characterized by complex popula�on structure, sparse tes�ng design, and gene�c diversity on both sides of the host-pathogen interac�on. **Methods** **Gene�c Material** _**Theobroma cacao**_ : The germplasm collec�on at the Interna�onal Cacao Collec�on (IC3) at CATIE in Turrialba, Costa Rica, was sampled for genotyping and provided pods for disease screening assays. The collec�on at the “La Montaña” site, located at 602 m above sea level, with a mean annual rainfall of 2,645 mm and a mean temperature of 22.5°C (Mata-Quirós et al., 2017; W. M. Phillips-Mora et al., 2007). The collec�on was re-established at this site between 2001 and 2005, with three trees per accession, and comprises approximately 1,200 accessions represen�ng diverse cacao gene�c backgrounds (Mata-Quirós et al., 2018; W. M. Phillips-Mora et al., 2007). No disease control was carried out other than the removal and disposal of diseased fruits. Naturally pollinated pods approximately 4 months old (unripe) were harvested in the morning, individually wrapped in foam, placed in a shipping box, and sent to the Plant Pathology Department of the University of Florida in Gainesville, Florida, USA, for screening with _Phytophthora_ sp. isolates. Pods were imported under APHIS PPQ permit 556-20-330-01055 and 556-23-242-14283. _**Phytophthora megakarya**_ : The ini�al study panel comprised 67 _Phytophthora megakarya_ isolates origina�ng from Ghana (21), Cameroon (18), and Nigeria (28). Isolates were obtained from collec�ons maintained by the USDA in Beltsville, USA, Gerben Mar�jn ten Hoopen, and the University of Florida (Supplementary Table 1), and were imported or transferred under APHIS PPQ permits P526P-20-02935 and P526P-21-01779. Isolates originated from black pod-infected cacao or soil samples collected in West Africa. Previously characterized isolates were obtained from collec�ons described by Ali et al. (2016), while addi�onal isolates originated from epidemiological studies of _P. megakarya_ in Cameroon (Ndoungué Djeumekop et al., 2021). Nigerian isolates originated from the states of Abia, Cross River, Edo, Ondo, and Osun, whereas Cameroon isolates originated from Bakoa and Kédia (Supplementary Table 1). Species iden�ty was confirmed by PCR using the _P. megakarya_ -specific primers and amplifica�on condi�ons described by Ali et al. (2016). All isolates were revived by inocula�ng cacao pod �ssue, followed by re-isola�on and subsequent storage. Wounded pod-husk inocula�on assays were conducted using an adapta�on of the methods described by Ali et al. (2017) and Puig et al. (2018). Briefly, cacao pod husks were cut into 6 × 6 cm pieces and placed in sterile plas�c containers for incuba�on. Following lesion development, infected �ssue was excised from the advancing lesion margin using a sterile scalpel, 6 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) and three �ssue pieces were transferred to 5% clarified tomato juice (CT) agar (1.5% agar) amended with 5 mg L⁻¹ pimaricin, 250 mg L⁻¹ ampicillin, and 10 mg L⁻¹ rifampicin (PAR), adapted from Ferguson and Jeffers (1999). A�er hyphal growth was observed, mycelium was transferred to water agar (1.5% agar) for purifica�on, followed by transfer to 10% CT agar (1.5% agar, PAR). For long-term storage, colonized agar plugs (5 × 5 mm) were transferred to autoclaved 8-mL glass vials containing reverse-osmosis water, either with or without sterilized hemp seeds, in duplicate, using a method adapted from Su�on et al. (2015). Alterna�vely, isolates were maintained on 10% CT agar (1.5% agar, PAR) in the dark at room temperature for short-term storage. **Genotyping** _**Theobroma cacao**_ : One healthy young leaf was collected from each clone in February 2022 in the IC3 collec�on in Costa Rica, with the first tree in the three-tree row plot sampled if available. Twelve clones were not sampled because no living trees were available. Leaf samples were collected in paper bags and shipped to Mars Plant Science Center (Davis, CA), with silica gel bags used to capture humidity. BioArk Leaf collec�on kits (LGC Biosearch Technologies, Teddington, Middlesex, UK) were used to collect eight 5 mm leaf discs from a single leaf for each sample. The leaf disc samples were shipped to LGC for DNA extrac�on using their sbeadex™ extrac�on protocol followed by genotyping with SeqSNP [© ] (LGC Biosearch Technologies). SeqSNP [©] employs an enrichment approach to target specific regions of the genome for sequencing. The resul�ng sequence data was aligned to the Ma�na V1 cacao reference (Motamayor et al., 2013) using bwa (Li & Durbin, 2009) and freebayes v1.2.0 (Garrison & Marth, 2012) was used to call variants for all cacao clones. The targeted regions consisted of 10,000 preselected posi�ons, most of which had been previously published (Allegre et al., 2012; Jarquín et al., 2014; Livingstone et al., 2015, 2017; Romero Navarro et al., 2017). The panel contained three sets of markers: 1) 8,635 markers that were evenly distributed across the genome, 2) 864 markers within nucleo�de-binding and leucine-rich repeat genes that were iden�fied in the ‘Criollo’ genome v2 (Argout et al., 2017) using the NLRannotator pipeline v.0.7 (Steuernagel et al., 2020), and 3) 501 markers associated with genes or quan�ta�ve trait loci iden�fied in various other studies. The resul�ng VCF-file was processed using a customized script in R 4.3.2 (R Core Team, 2023) and RStudio v.2025.5.0.496 (Posit team, 2025). The clone names were standardized and harmonized, and biallelic SNPs were retained and genotypes were encoded as addi�ve dosages of 0, 1, or 2 copies of the alternate allele using the package SNPassoc 2.1-0 (Moreno et al., 2022). If mul�ple samples of the same clone were present, similarity was verified using the correla�on of marker calls. For samples showing a correla�on below 95% with the same clone and greater than 98% with one or more other clones, the loca�on in the orchard was added to the clone’s name. An assignment test was applied to infer popula�on membership and admixed ancestry (hybrids or ancestral forms) of the clones in the 11 reference groups, using a model-based clustering method implemented in the so�ware program STRUCTURE (Pritchard et al., 2000). The analysis used selected SNPs and reference accessions represen�ng ‘Amelonado’, ‘Bolivian’, ‘Contamana’, ‘Criollo’, ‘Curaray’, ‘Guiana’, ‘Iquitos’, ‘Maranon’, ‘Nacional’, ‘Nanay’, and ‘Purus’. STRUCTURE was 7 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) run under the admixture model with correlated allele frequencies, using a burn-in of 100,000 itera�ons followed by 200,000 MCMC itera�ons. Ten replicate runs were performed with the number of clusters (K-value) set to 11, assuming that each of 11 reference groups may have contributed to the clones. For each clone, the STRUCTURE membership coefficients (Q) were interpreted as ancestry propor�ons. Clones with ≥75% membership in their predominant reference group were classified as members of that group; clones below this threshold were classified as admixed and named according to the ancestry component with the highest Q value. _**Phytophthora megakarya**_ : _P. megakarya_ isolates were grown on 10% CT plates at room temperature for 7 days in the dark. Three agar plugs were transferred to plates, covered with 20 mL of 10% liquid CT, and grown at 25°C in the dark for 7 days. Mycelia were washed, harvested, and concentrated using a vacuum pump prior to storing at -80°C and freeze-drying for 24 hours using a FreeZone 2.5 Liter -50C Benchtop Freeze Dryer (Labconco Corpora�on, Kansas City, MO, USA). LGC Biosearch Technologies Inc. (Hoddesdon, UK) performed DNA extrac�ons and library prepara�ons using their high-throughput workflow with proprietary chemistry. DNA was sheared to a mean fragment length of 500 bp, fragments were end-repaired and A-tailed, followed by incorpora�on of unique dual-indexed Illumina adaptors and PCR enrichment, equimolar pooling, and library prepara�on. Similar to cacao, enrichment was employed prior to sequencing to focus on virulence-related genes selected from the _P. megakarya_ and _P. palmivora_ genomes, with annota�ons published by Morales-Cruz et al. (2020). The target panel comprised gene families cri�cal to the infec�on process, e.g., proteases, CAZymes, detoxifica�on genes, and effectors (Supplementary Table 2). LGC Biosearch Technologies Inc. (Hoddesdon, UK) designed and synthesized the bio�nylated probes used to enrich the targeted regions, providing a 1x probe �ling across the targets. Probes were evaluated for GC content, homopolymers, and uniqueness within the genome and filtered as needed. The final panel consisted of 54,046 probes, each 120 bp in length. These probes were designed to target 3,505 _P. megakarya_ genes, 2,774 SNP _P._ loca�ons previously iden�fied using genotype-by-sequencing (unpublished), and 2,381 _palmivora_ genes (Supplementary Table 2). The DNA libraries were hybridized to the capture-Seq probes and sequenced on a NovaSeq S4 flow cell (2 x 150 bp). The samples were mul�plexed, and demul�plexing was done by LGC Biosearch Technologies Inc. (Hoddesdon, UK). The sequencing data were checked for quality using FastQC v0.12.1 (Andrews, 2010), and Mul�QC 1.24.1 (Ewels et al., 2016) was used to compile a summary report. Trimming was done using fastp 0.22.0 (Chen et al., 2018) with default se�ngs. Filtered high-quality reads were mapped to the _P._ _megakarya_ GH34 genome (Morales-Cruz et al., 2020) using BWA-MEM 0.7.17 (Li, 2013) with the “-M” flag to mark shorter split hits as secondary. The aligned SAM files were processed with SAMtools 1.19 (Danecek et al., 2021) to convert them to BAM files, sort, fix mate pair informa�on, remove PCR duplicates, filter for quality (MAPQ ≥ 20), and obtain overall mapping sta�s�cs. SNP calling was performed per sample and merged using GATK v.4.4 (Van der Auwera & O’Connor, 2020) _HaplotypeCaller_ with ploidy set to 2 and _CombineGVCFs_ . SNPs were extracted and filtered using the recommended filtering criteria by GATK for read and mapping quality, as determined by thresholds for quality by depth (QD <= 2.0), strand bias (FisherStrand (FS) > 60.0), and root mean 8 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) square mapping quality over all the reads (MQ < 40.0), and filtered for the target regions. Summary sta�s�cs were generated using rtg-tools 3.7.1 (Cleary et al., 2015) and bc�ools 1.22 (Danecek et al., 2021). The resul�ng VCF file was analyzed in R 4.3.2 (R Core Team, 2023) using the vcfR package v1.15.0 (Knaus & Grünwald, 2017) for extrac�on, filtering, and conversion, and the ggplot2 package 3.5.0 (Wickham, 2016) for visualiza�on. Because sequencing depth varied among isolates, genotype calls in the lowest and highest 10% of the depth distribu�on for each sample were excluded before downstream analysis; the corresponding depth thresholds ranged from 27–46× for the lower cutoff and 544–1,577× for the upper cutoff. Biallelic SNPs were retained and encoded as addi�ve genotype dosages (0, 1, or 2), leaving 131,162 SNPs across 67 isolates. **Popula�on structure and genomic rela�onship matrices** The final marker matrices from genotyping were processed using ASRgenomics 1.1.4 (Gezan et al., 2022) for quality control (QC) and the construc�on of rela�onship matrices. Rela�onship matrices were calculated using Van Raden’s method (VanRaden, 2008) to convert the n × m marker matrices into n × n genomic rela�onship matrices (GRM) (Amadeu et al., 2023), where n is the number of individuals and m the number of markers. Of the 1,208 cacao samples submi�ed for genotyping, 28 samples were duplicates (r i,j =1). The samples with fewer missing values across the marker set were kept for analysis. Marker filtering was performed using thresholds for missing values within individuals (≥0.5), or within markers (≥0.5), and rare markers with minor allele frequency (≤0.05) were removed. This filtering process removed four individuals and 251 markers from the cacao popula�on, leaving 1,176 individuals x 8,593 markers. The GRM was computed, with missing data imputed using marker-specific average frequencies. Kinship diagnos�cs used to detect inbreeding and anomalies in relatedness es�mates (F i ≥ 1.9 and ≤ 0.5) iden�fied 159 individuals that were removed prior to a second round of kinship filtering, that resulted in the removal of an addi�onal 19 individuals. A final kinship filtering step was applied to remove duplicates (r ij - 0.98), resul�ng in a cacao GRM (GRM c ) based on 947 individuals and 8,572 markers. The names of clones iden�fied as duplicates were updated to match the remaining clone in the analysis. Rela�onship matrices for the 67 _P. megakarya_ isolates were created following marker filtering using thresholds for missing values within individuals (≥0.5), or within markers (≥0.25), and the minimum minor allele frequency (≤0.05). These filtering steps removed 67,227 markers. An addi�onal 104 homozygous SNPs across the popula�on were also removed (MAF = 1), resul�ng in a matrix of 67 individuals x 63,831 markers. Base R and the so�ware packages ‘factoextra’ (Kassambara & Mundt, 2016) and ‘FactoMineR’ (Husson et al., 2006) were used to inves�gate and iden�fy the popula�on structure in both _Theobroma cacao_ and _P. megakarya_ . The extreme popula�on structure of the _P. megakarya_ isolates was evident when the first two components of a principal component analysis (PCA) using filtered markers were examined (Supplementary Figure 1). The number of subpopula�ons for _P. megakarya_ was defined using PCA and hierarchical 9 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) clustering. For each subpopula�on, the marker matrix was extracted, marker filtering was applied, and a GRM was computed. VanRaden genomic rela�onships are derived from allele-frequency-centered marker dosages, and their scale depends on the marker-frequency distribu�on used in the popula�on. These assump�ons are not met in a predominantly clonal organism like P. megakarya. The mean diagonals of the unscaled subpopula�on matrices were accordingly 0.19, 0.41 and 0.39 for subpopula�ons A, B and C, compared to 1.15 for the cacao matrix. Because a variance component es�mated against a rela�onship matrix K is expressed on the scale of K, each matrix was divided by D = mean(diag(K)) - mean(K) (Legarra, 2016) so that its variance component equals the variance of the gene�c values it describes. Each _P. megakarya_ subpopula�on matrix was scaled by its own D before each subpopula�on-specific GRM was combined into a block-diagonal mul�popula�on isolate similarity matrix, so that a single isolate variance component applies on a common scale across the three subpopula�ons. All variance ra�os reported below are therefore comparable among models and among pathogen subpopula�ons. The final GRM for isolates was defined as the block-diagonal matrix GRM I : [2] 𝐺𝑅𝑀 � = � 𝐺𝑅𝑀 � 0 0 0 𝐺𝑅𝑀 � 0 0 0 𝐺𝑅𝑀 � � The final pathogen GRM had a block-diagonal structure to model genomic covariance among isolates within each subpopula�on, whereas covariance between subpopula�ons was set to zero. Broad differences among subpopula�ons were therefore modeled separately as fixed effects. The following packages were used for visualiza�ons in R: ggplot2 3.5.0 (Wickham, 2016), ggrepel 0.9.5 (Slowikowski, 2016), ggsci 3.0.0 (Xiao, 2016), ggpubr 0.6.0 (Kassambara, 2016), rcartocolor 2.1.1 (Nowosad, 2017), scales 1.3.0 (Wickham et al., 2011), ggvenn 0.1.19 (Yan, 2021), and superheat 0.1.0 (Barter & Yu, 2018). **Disease screening** Varia�on in the �ming of pod availability across CATIE’s IC3 germplasm collec�on prevented the use of balanced experimental designs. The germplasm collec�on was sampled as unripe 4-monthold pods were available. Fruit ripening stages were assessed visually based on genotype-specific changes in pod color associated with matura�on (W. Phillips-Mora et al., 2013). A total of 31 shipments from the CATIE germplasm collec�on to the University of Florida defined different bioassays, with each bioassay containing up to 100 pods. As many as five isolates were used to inoculate the clones in a single bioassay. A total of 1,971 pods, represen�ng 563 clones, were inoculated with one or several of the 67 _P. megakarya_ isolates between May 2021 and May 2023. A�er the first year of screening, a selec�on of clones was made from the clones phenotyped to date to represent the collec�on's gene�c diversity and facilitate replica�on of clone-by-isolate combina�ons. POUND-7 served as control clone and was tested in 22 out of 31 bioassays, because pods were readily available from a plan�ng adjacent to the IC3 collec�on and because it has 10 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) demonstrated resistance to black pod rot caused by _P. palmivora_ (Brown et al., 2007; W. Phillips Mora & Galindo, 1989). _P. megakarya_ GH34 served as a control isolate because its published genome annota�on was used as the reference for probe design (Morales-Cruz et al., 2020), and its aggressiveness on cacao pods had previously been characterized (Ali et al., 2016; Puig et al., 2018). The isolate was grown and prepared for every bioassay and used if one of the selected isolates did not produce enough zoospores for inocula�on (11 of 31 bioassays). Ini�al experimenta�on led to the development of a bioassay protocol, which has been deposited in the protocols.io repository (Gi�o et al., 2026). This ar�ficial inocula�on technique is based on the paper disc method developed by Phillips-Mora and Galindo (1989). The bioassay was designed to yield six observa�ons per pod. Pods were prepared for inocula�on by cleaning the surface with dis�lled water. While pods from different clones were harvested at the same �me, differences in shipping �mes cause the �me between harvest and inocula�on to vary from 4 to 13 days (average 8 days). Damaged pods or pods with obvious disease symptoms that developed during shipping were discarded. The plates used to create the zoospore solu�ons were prepared 1 week in advance of inocula�on. Colonized agar mycelium chunks from up to five isolates were placed on two 10% CT plates per isolate and prepared as described in the protocol. For inocula�ons, zoospore solu�ons were prepared at a concentra�on of 50,000 spores per ml, and 1 cm² pieces of �ssue paper were placed on the pod prior to pipe�ng 30 μl of the solu�on into six evenly spaced posi�ons on the pod surface. Twelve inoculated pods were placed in each incuba�on box, and up to six boxes were placed in two inocula�on chambers (Percival Scien�fic, model I36NL) to maintain a temperature of 25°C at a humidity of approximately 100%. Four days a�er inocula�on, ver�cal and horizontal measurements were taken for each of the six inocula�on sites per pod using calipers. Lesions were measured on both the ver�cal and horizontal axes, and the average of both lesion measurements was calculated per inocula�on site. All analyses of lesion size were then performed using the average of the six mean lesion measurements per pod. Lesions less than 15 mm were within the ini�al area of the �ssue paper-based inocula�on. A response only in this area was considered resistant, because the lesion did not spread beyond the inocula�on area. Lesion measurements below 30 mm were considered moderately resistant, whereas measurements above 30 mm indicated suscep�bility. **Sta�s�cal models** Data were analyzed using a set of linear mixed models to produce variance component es�mates and provide lesion-size predic�ons for all clones, isolates, and clone-by-isolate interac�ons observed in the bioassays, as well as tests of significance for model effects. Covariance structures associated with the random effects in the mixed models were used to compare models that used tradi�onal design matrices to account for bioassay effects, as well as marker-derived rela�onship matrices for cacao clones (GRM C ), _P. megakarya_ isolates (GRM I ), and clone-by-isolate interac�ons. Variance ra�os for host, pathogen, and host-by-pathogen effects and type-B correla�ons for host and pathogen were derived from the fi�ed variance components from the general linear mixed model, presented in matrix nota�on following Mrode and Thompson (2005) as: 11 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) [1] 𝑦= 𝑋𝑏 + 𝑍𝑢 + 𝑒 where: **y** refers to phenotypic observa�ons (mean lesion diameter per pod in mm); **b** is the vector of fixed _P. megakarya_ subpopula�on effects; **u** is the vector of random effects associated with the bioassay, cacao clone, _P. megakarya_ isolate, and the cacao clone by _P. megakarya_ isolate interac�on; **X** is the design matrix rela�ng each observa�on to the **b** fixed effects; **Z** is the design matrix rela�ng the **y** observa�ons to the **u** random effects; and **e** is the vector of residuals or random errors associated with each observa�on. Random effects for clone, isolate, clone-byisolate interac�on, and bioassay were assumed to be normally distributed, with σ [2] Clone, σ [2] Isolate, σ [2] Clone:Isolate, and σ [2] Bioassay defining the variance components associated with each random effect, respec�vely. Separate error variances were es�mated for each _P. megakarya_ subpopula�on to allow for heteroskedas�city of the error variances, and the average of these error variances (σ [2] E ) was used to es�mate gene�c parameter es�mates. Six mixed models differing in their covariance structures for host, pathogen, and host-bypathogen effects were fi�ed to the same phenotype dataset (Table 1). Model fit was compared using the REML log-likelihood (logL) and Akaike’s informa�on criterion, AIC = −2logL + 2k, where k is the number of es�mated variance parameters. All models were fi�ed to the same records with the same fixed effect (subpopula�on), so their REML log-likelihoods are comparable. Several pairs of models are not nested, so models were ranked by AIC; a lower AIC indicates a be�er fit. Nested comparisons of single variance components used likelihood ra�o tests, with the null distribu�on taken as a 50:50 mixture of χ²₀ and χ²₁ because the variance is tested at its boundary (Self & Liang, 1987). Models 2–5 progressively replaced iden�ty covariance structures with marker-derived rela�onship matrices for cacao clones, _P. megakarya_ isolates, or clone-by-isolate combina�ons. Model 1 used tradi�onal design matrices for clone, isolate, and interac�on effects, assuming no covariance or relatedness among cacao varie�es or _P. megakarya_ isolates. In Model 2, GRM C and a GRM C -by-isolate design matrix accoun�ng for clone-by-isolate interac�on effects replaced the tradi�onal design matrices that were used for Model 1. In Model 3, GRM I and GRM I -by-clone design matrices replaced the Model 1 design matrices. In Model 4, GRM C and GRM I replaced their respec�ve Model 1 design matrices. For the replacement of the clone-by-isolate interac�on design matrix, the cacao GRM C and the _P. megakarya_ GRM I were combined to form a covariance matrix, which was incorporated into a reac�on norm model (Jarquín et al., 2014), to generate a dual-genome similarity matrix (GGSM CI ) (Hudson et al., 2024). Model 5 incorporated the GGSM CI rather than the iden�ty matrix to account for similari�es among clone-by-isolate interac�ons. The significance of clone, isolate, and interac�on effects were evaluated with likelihood ra�o test that u�lized Model 2 as the complete model for comparisons with reduced models that excluded the random effects under test. A final Model 6 used GRM C with a heterogeneous correla�on structure ( _corh()_ ) for isolate subpopula�ons that allowed for separate genomic variances for each isolate subpopula�on. Addi�onally, subpopula�on-specific isolate effects were included using the _at()_ direc�ve of ASReml. A full unstructured covariance model ( _corgh()_ ) for clone genomic effects was 12 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) ini�ally evaluated but converged with the correla�on between subpopula�ons A and B at its boundary (1.00), so the heterogeneous model with a single correla�on (corh()) is reported. Predic�ons for cacao clones, _P. megakarya_ isolates, and clone-by-isolate effects are derived from Model 2, with subpopula�on-specific clone predic�ons from Model 6 to provide comparisons among cacao clones and _P. megakarya_ isolates and subpopula�ons, respec�vely (Supplementary Tables 5 and 6). Model 6 had the lowest AIC, but its addi�onal parameters described subpopula�on-specific variance heterogeneity rather than a different overall ranking of host clones as the es�mated genomic correla�on among subpopula�ons was high (0.86). We therefore used Model 2, which provides a single genome-wide host effect, for cross-valida�on and overall clone predic�on. **Gene�c parameter es�mates** Gene�c parameter es�mates are provided for each model. For each model, the propor�on of phenotypic variance a�ributable to the cacao clone component ( 𝐻 [�] �� ) was calculated as the ra�o of gene�c variance associated with varia�on among clones to the total phenotypic variance: i [3] 𝐻� �� = [𝑉�] [�] 𝑉 [�] � i 𝜎� ����� i ������ + 𝜎� �������� i � i 𝜎� ����� ≈ i �����:�������� + 𝜎� �� i � i �������� + 𝜎� �����:�������� i with variance components defined above, 𝑉� � as defined is the sum of the clone, isolate, interac�on and residual variances, not the total observed phenotypic variance, since the bioassay effect is s�ll fi�ed but not counted. The bioassay variance was 49–57 mm² across models, about 16% of the total variance (Supplementary Table 7). 𝑉� � is the genotypic variance of cacao clones. Analogous variance ra�os were calculated for isolate ( 𝑉 [�] � ) and clone-by-isolate effects ( 𝑉� �� ) to permit direct comparison of fi�ed variance components on the same phenotypic scale: i [4] ℎ [�] �� = [𝑉�] [�] 𝑉 [�] � [5] ℎ [�] ��� = [𝑉�] [��] 𝑉 [�] � i 𝜎� ����� i ������ + 𝜎� �������� i � i 𝜎� ������� ≈ i ������ + 𝜎� �������� i �����:�������� + 𝜎� �� i � i 𝜎� ����� i �������� + 𝜎� �����:�������� i � i 𝜎� �����:������� ≈ i �����:�������� + 𝜎� �� i � i �������� + 𝜎� �����:�������� i Type-B correla�ons were calculated as variance-ra�o summaries of the rela�ve magnitude of clone-by-isolate interac�on (Burdon, 1977) following equa�ons 6 and 7. i � [6] 𝑟 �� = [𝜎] [�����] i (𝜎 ������ i ������ + 𝜎 �����:�������� i �(𝜎 ������ + 𝜎 �����:�������� ) i � [7] 𝑟 �� = [𝜎] [�������] i (𝜎 �������� i �������� + 𝜎 �����:�������� i �(𝜎 �������� + 𝜎 �����:�������� ) i Both quan��es are bounded func�ons of the interac�on variance and approach unity as the interac�on variance approaches zero, so they are reported as descrip�ons of the f i �ed variance 13 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) par��on typically used to indicate the s tability of lesion size predic�ons across isolates ( 𝑟 �� ) or clones ( 𝑟 �� ). **Cross-valida�on** The ability of Model 2 to predict phenotypically unobserved clones within the IC3 popula�on was evaluated using clone-wise 10-fold cross-valida�on. The 350 phenotyped clones were assigned at random to ten folds. For each valida�on fold, all lesion observa�ons from the corresponding clones were masked and Model 2 was refi�ed to the remaining records, so valida�on clones were predicted solely through genomic rela�onships with phenotyped training clones. Predic�ve ability was the Pearson correla�on between the predicted genomic values of the masked clones and their observed clone means. Before averaging, each pod was adjusted for the subpopula�on, isolate, and bioassay effects es�mated from Model 2. To obtain predic�on accuracy, we divided predic�ve ability by the square root of the mean heritability of clone means: � � � = [𝜎] (𝜎 � [8] 𝐻 � � � (𝜎 � + 𝜎 � ) � 𝑛 � where 𝑛 � is the number of pods of clone j. We repeated the procedure with five random par��ons. To test whether the pathogen GRM improves predic�on, we evaluated Model 4 on the same par��ons and against the same adjusted clone means and compared the two models within each par��on. **Gene�c-group ancestry and predicted resistance** To relate predicted resistance to the gene�c groups in the collec�on, we regressed the Model 2 GEBVs of the 350 phenotyped clones on their STRUCTURE ancestry propor�ons (K = 11) in a single mul�ple regression. Ancestry propor�ons sum to one, and ‘Amelonado’, the most common group, served as the reference. With ‘Amelonado’ omi�ed as the reference component, each coefficient represents the expected difference in GEBV associated with a one-unit increase in the indicated ancestry propor�on rela�ve to ‘Amelonado’. Collinearity was checked using variance infla�on factors. Unphenotyped clones were excluded, because their GEBVs were derived en�rely from phenotyped rela�ves. Because GEBVs are shrunken and mutually correlated through a GRM derived from the same marker panel that informs ancestry, inferen�al sta�s�cs from this regression are not independent of the genomic model. The analysis was therefore treated as exploratory and used to iden�fy ancestry-associated trends rather than es�mate causal gene�cgroup effects. 14 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) **Results** **One-third of the IC3 cacao collec�on shows high ancestry membership to seven gene�c groups** **of cacao** All living clones of the CATIE IC3 collec�on were genotyped with a 10K SNP marker panel. A�er calling SNPs, cleaning for SNP quality, inbreeding, and duplica�ons, 947 clones and 8,572 markers remained. The clones were assigned to eleven gene�c reference groups described in cacao (Motamayor et al., 2008; Zhang et al., 2012) (Figure 1, Supplementary Table 3). 300 clones were assigned as members of one of seven gene�c groups (≥75% contribu�on), while 647 clones were considered admixed (<75%). The gene�c groups ‘Contamana’, ‘Criollo’, and ‘Curaray’ were represented only by admixed clones, while the ‘Bolivian’ gene�c group was present, but did not strongly contribute to the examined clones. The gene�c group ‘Amelonado’ contributed the most to the IC3 collec�on, both in terms of members (13.8%) and hybrids (34.4%), while ‘Curaray’ showed the lowest contribu�on as a gene�c group (0.7%) (Figure 2C). The average propor�on of ‘Amelonado’ across clones in the whole popula�on was 39.6%, followed by ‘Criollo’ and ‘Nacional’ with an average propor�on of 10% each. There was an overlap of 131 clones with the popula�on described in Motamayor et al. (2008), with 81% of these clones assigned to the same gene�c group (Supplementary Table 3). The rela�onships among all clones were visualized using PCA, with the first two components accoun�ng for 25.2% of the total gene�c varia�on (Figure 2A and 2B). Ancestry-related gene�c structure was evident in the genome-wide SNP-based PCA among clones assigned as gene�c group members (Figure 2A). Clones with ≥75% ‘Amelonado’ ancestry occupied a broad and par�ally structured region of PCA space across PC1–PC3, indica�ng substan�al within-group diversity. ‘Guiana’ members formed a compact region at high PC2 scores, while ‘Maranon’ and ‘Nanay’ also occupied compara�vely high PC2 posi�ons but were dis�nguished with addi�onal components (Supplementary Figure 2A and 2B). ‘Iquitos’ and ‘Purus’ occurred near one another along PC1 but separated more clearly on PC3 and PC4. ‘Guiana’ remained dis�nct across the first four principal components, while the numerous admixed clones occupied intermediate regions among the high-membership groups. **Three subpopula�ons were iden�fied in the** _**P. megakarya**_ **isolates causing black pod** The 67 _P. megakarya_ isolates were genotyped using a customized capture-sequencing approach to examine popula�on structure. The main gene families targeted by this approach were effectors (43%), especially secreted RxLRs (36%), proteases (18%), detoxifica�on genes (15%), and CAZymes (13%) (Supplementary Table 2). The rate of duplicated reads ranged from 37.0% to 62.5% with a median of 48.1%. Over 99% of the reads mapped to the _P. megakarya_ GH34 genome (Morales-Cruz et al., 2020). PCA of 62,203 filtered SNPs revealed a pronounced separa�on of isolates into three �ght clusters, with PC1 explaining 78.9% of the variance (Supplementary Figure 1). This primary division was used as the first step in defining broad genomic groups, which were then examined separately using PCA and rela�onship matrices (Supplementary Figures 3 and 4). Within the western group (including isolates from Ghana and Nigeria), addi�onal structure remained evident (PC1 = 48.6%), mo�va�ng a further subdivision before construc�on of within 15 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) group GRMs. Addi�onally, two isolates, “PH009” and “PH021” from Nigeria, were gene�cally dis�nct from both subpopula�ons. The second subpopula�on, comprising all isolates from Cameroon and some from Nigeria (Supplementary Figure 4), also separated into two clusters, with one cluster comprising three isolates from Nigeria and one isolate from Cameroon (PC1 = 68.5%). Another Cameroon isolate, “NRRL_64259,” was gene�cally dis�nct from both clusters. Seven isolates fell outside the three major genomic groups and were excluded from analyses requiring the block-diagonal pathogen GRM because the small number of outlying isolates did not support es�ma�on of addi�onal group-specific covariance structures. Their exclusion reflects a modeling constraint rather than evidence that these isolates are biologically unrepresenta�ve. The subpopula�ons were named from West to East. The first subpopula�on comprises all isolates from Ghana and two from Nigeria (“A”); the second comprises only isolates from Nigeria (“B”); and the third comprises isolates from Cameroon and five isolates from Nigeria (“C”) (Supplementary Figure 5A-C). A�er spli�ng the popula�on into three subpopula�ons, the marker matrix was extracted per subpopula�on. Marker matrix A contained 23 isolates and 11,753 SNPs, out of which 10,871 SNPs (91.6%) were fixed heterozygous. Similarly, marker matrix B contained 14 isolates and 11,614 SNPs, with 10,659 SNPs (91.8%) fixed heterozygous markers. Lastly, marker matrix C contained 21 isolates and 34,519 SNPs, of which 84.6% (29,210 SNPs) were fixed heterozygous. Sites fixed as heterozygous within a subpopula�on contain no among-isolate varia�on within that group and were therefore excluded before construc�on of the withinsubpopula�on GRM. The removal of those SNPs in one subpopula�on did not preclude finding SNPs in the other two subpopula�ons. The final marker matrices contained 882, 995, and 5,309 SNPs for the subpopula�ons A, B, and C, respec�vely. Within subpopula�ons, SNPs covered 4% (A), 6% (B) and 23% (C) of the targeted gene space (Supplementary Table 4). Effectors made up 52.8% of the SNP-containing genes in subpopula�on C, against 36.5% in A and 29.3% in B, whereas proteases were rela�vely more frequent in A and B (28.5% and 30.3%) than in C (17.8%). Most SNPs were specific to one subpopula�on (5,258 in C, 765 in B and 723 in A); only 6 SNPs and 27 genes, five of them secreted RxLR effectors, carried varia�on in all three (Figure 3). **Varia�on in lesion size reflects diverse cacao and pathogen responses** Pods that developed disease symptoms unrelated to the controlled inocula�ons or where physical damage became evident between inocula�on and assessment were excluded (19%). Two isolates (N174 and N339) produced no visible infec�on on >80% of the clones they inoculated and were excluded from the primary quan�ta�ve lesion-size analysis (Supplementary Table S1), resul�ng in a dataset of 1,508 pods from 435 cacao clones inoculated with one or more of the 64 _P. megakarya_ isolates. Exclusion of the seven isolates not represented in the three modeled genomic subpopula�ons resulted in the final analysis dataset of 1,397 pods from 404 clones inoculated with 58 isolates and assessed for lesion size in 30 independent bioassays between 2021 and 2023. Due to the availability of pods across bioassays and by design, 5.1% (1,184) of the 23,432 poten�al clone-by-isolate combina�ons were tested. _P. megakarya_ isolates were used to 16 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) inoculate 4 to 78 different clones (median = 18.5, Figure 4A) apart from the reference isolate GH34, which was used to inoculate 227 pods (16%) from 88 different clones (22%). Across the 1,653 isolate pairs, the median number of clones tested with both members of the pair was zero: 51% of pairs shared no clone and 87% shared two or fewer. Each clone was inoculated with 1 to 17 isolates (median = 3), except for the control clone POUND-7, which was inoculated with 18 different isolates (31%) on 26 independent pods (Figure 4B). The lesion diameter per pod (4 DAI) ranged from 0 mm to 70.8 mm, with an overall average of 36.7 mm (Figure 4C). The histograms by subpopula�on showed that we collected more data for subpopula�on A because it contains the highest number of isolates, and we used isolate GH34 across bioassays. Subpopula�on B was the smallest (14 isolates) and had the fewest total data points. Lesion-size distribu�ons were mildly shi�ed toward larger lesions, par�cularly for isolates from subpopula�on A. The data collected on isolate GH34 showed a diverse response among the 88 inoculated clones: although 83% were suscep�ble (>30 mm), only 17% (15) were moderately resistant, and no clones were resistant. The isolates tested on the reference clone POUND-7 showed a range of virulence levels. POUND-7 was suscep�ble to seven isolates (39%), six of them from subpopula�on A, moderately resistant to six isolates, and resistant to five isolates; eight of the la�er eleven isolates are from subpopula�on C. In lieu of analyzing trends from observa�ons on the whole dataset (Supplementary Figure 6, Supplementary Figure 7A-D), mixed models were used to draw conclusions across our sparse experimental design. **Host genomic rela�onships improve predic�on of lesion size** A series of six increasingly complex mixed models was compared to evaluate the impact of genomic informa�on on predic�ons of lesion size (Table 1). These models progressively incorporated GRMs for cacao clones (Model 2), _P. megakarya_ isolates (Model 3), both together (Model 4), and a dual-genome similarity matrix used to model clone-by-isolate interac�ons (Model 5). Model 1 a�ributed 0.20 of the phenotypic variance to differences among clones and 0.36 to differences among isolates, whereas clone-by-isolate interac�on accounted for 0.02. Both type-B correla�ons were high. Model 2 replaced the tradi�onal clone matrix with GRM c . In Model 2, the cacao genomic component accounted for 27% of phenotypic variance and model fit improved rela�ve to Model 1, highligh�ng the value of accoun�ng for gene�c relatedness among cacao clones. Model 3 incorporated isolate relatedness while using an iden�ty matrix for clones. This led to a slightly increased pathogen variance es�mate (0.38) compared to Model 1 and a similar clone heritability es�mate (0.20). Together with a slightly worse model fit, this suggests that isolate-relatedness contributed li�le to explaining lesion size varia�on. Model 4 combined gene�c relatedness for both clones and isolates. This model improved fit compared to Model 1, but not compared to Model 2, with an increased clone heritability (0.26), a similar isolate variance es�mate (0.36), and a smaller error term (0.38). The clone-by-isolate interac�on term is at the boundary in Model 4. Model 5 added the dual-genome similarity matrix to explicitly model genomic similari�es in clone-by-isolate interac�ons. This model yielded values similar to those of Model 4 across all parameters and a clone-by-isolate es�mate of 0.01. Apart from providing a clone-by-isolate es�mate, Model 5 performed like Model 4 in fit. Across these five models, Model 17 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) 2 showed the best model fit with an AIC of 7503.73. Likelihood ra�o tests comparing Model 2 with reduced models showed significant variance among bioassays (LR = 165.1, P < 0.001), clones (LR = 204.3, P < 0.001) and isolates (LR = 212.4, P < 0.001), whereas the clone × isolate variance was not significant (LR = 2.23, P = 0.07). All tests had one degree of freedom, with P-values from a 50:50 mixture of χ²₀ and χ²₁ (Self & Liang, 1987). Type-B correla�ons remained high across all models, but both are bounded func�ons of the interac�on variance and approach unity as that variance approaches zero, so they are reported here as descrip�ons of the fi�ed variance par��on rather than as evidence of stability. In clone-wise cross-valida�on, Model 2 predicted phenotypically masked clones within the IC3 popula�on with a predic�ve ability of 0.414 ± 0.008 across five random 10-fold par��ons, corresponding to an es�mated predic�on accuracy of 0.52. Replacing independent isolate effects with the block-diagonal pathogen GRM did not improve predic�on; predic�ve ability differed by only 0.002 ± 0.001 in favor of Model 2. Subpopula�on means differed (Wald χ² = 9.75, 2 df, P = 0.008) with predicted lesion diameters of 40.0 ± 2.3 mm for subpopula�on A, 32.6 ± 2.6 mm for B, and 34.1 ± 2.8 mm for C, and residual variances of 85.3, 107.4, and 134.5 mm² for A, B, and C. Model 2 produced genomic es�mated breeding values (GEBVs) for cacao clones and BLUPs for independent isolate effects; these are reported in Supplementary Tables 5 and 6, respec�vely. Clone GEBVs were centered at zero, with nega�ve values indica�ng gene�cally smaller lesions and posi�ve values indica�ng gene�cally larger lesions rela�ve to the fi�ed popula�on mean (35.6 mm). Predicted lesion diameter on the observed scale was obtained by combining the clone GEBV with the relevant fixed-effect mean. The higher the GEBV, the greater the poten�al to be suscep�ble; the lower the GEBV, the greater the poten�al to be resistant. The three isolates with the largest predicted isolate effects were PH029 (+21.6 mm), N77 (+11.5 mm), and GH12 (+10.9 mm). The three least suscep�ble cacao clones to _P. megakarya_ infec�on, as phenotyped, were HY-2714182 (GEBV -18.5), COCA-3370-5 (GEBV -20.4), and LCTEEN-37 (GEBV -20.9). The three least suscep�ble cacao clones that were not phenotyped but were included in the cacao GRM were POUND-16-B (GEBV -17.6), NA-232 (GEBV -18.4), and NA-670 (GEBV -19.4). Model 6 must be viewed separately from the previously described Models 1 to 5, as it introduced subpopula�on-specific genomic variance components by fi�ng a heterogeneous correla�on structure for clones across subpopula�ons of the _P. megakarya_ isolates, along with subpopula�on-specific variances for isolates. This allowed an independent es�ma�on of variance components within each of the three _P. megakarya_ subpopula�ons and finer biological resolu�on. The heterogeneous correla�on structure was of par�cular interest as it provides es�mates of the genomic correla�ons between subpopula�ons and indicates the stability of resistance across cacao clones. The es�mate of 0.86 indicates strong concordance in cacao resistance predic�ons across subpopula�ons of _P. megakarya_, sugges�ng that popula�on-specific resistance mechanisms were not observed. The propor�on of phenotypic variance a�ributable to cacao genomic effects ranged from 0.19 in subpopula�on B to 0.30 in subpopula�on A, and 0.33 in subpopula�on C. The corresponding isolate variance ra�os were 0.40 in A, 0.49 in B, and 18 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) 0.26 in C. These es�mates indicate heterogeneity in both host genomic and isolate variance across pathogen subpopula�ons, although standard errors of the es�mates are large. Model 6 had the lowest AIC of all models (7491.3, against 7503.7 for Model 2), indica�ng an improvement in fit. Model 6 provided comparisons of gene�c control and resistance predic�ons for each subpopula�on. Because isolate GH34 was sampled in Ghana and subpopula�on A was larger, 51% of the phenotype data were collected from this subpopula�on. This compares with the subpopula�ons B and C, which accounted for 33% and 15% of the phenotypic data, respec�vely (Figure 4C). The three most resistant clones for each _P. megakarya_ subpopula�on and for Model 5 showed some overlap (Supplementary Table 5). In each GEBV list, LCTEEN-37 and COCA-3370-5 were among the three most resistant phenotyped clones, and genomic predic�ons iden�fied NA-670 and NA-232 as the three most resistant non-phenotyped clones. Model 5 and GEBV of subpopula�on B of Model 6 also shared HY-2714182 as a top 3 phenotyped clone, and GEBV of subpopula�ons B and C from Model 6 shared NA-710 as a top 3 non-phenotyped clone. The addi�onal top 3 clones with predicted resistance were IMC-27 (GEBV subpopula�on C) and PNG 414 (GEBV subpopula�on A), both with phenotype informa�on, and NA-807 (GEBV subpopula�on A), without phenotype informa�on. To explore whether predicted resistance covaried with cacao ancestry, Model 2 GEBVs of the 350 phenotyped clones were regressed jointly on their STRUCTURE ancestry propor�ons. Ancestry propor�ons accounted for 26% of the variance in GEBVs among the 350 phenotyped clones (adjusted R² = 0.26). Rela�ve to ‘Amelonado’, ancestry from ‘Nanay’ (−13.4 ± 1.9 mm per unit ancestry, _p_ < 0.001), ‘Contamana’ (−11.0 ± 2.4, _p_ < 0.001), ‘Purus’ (−8.7 ± 2.6, _p_ = 0.001) and ‘Maranon’ (−4.1 ± 1.4, _p_ = 0.003) was associated with smaller predicted lesions. Ancestry from ‘Criollo’ (+6.2 ± 2.4, _p_ = 0.010) and ‘Guiana’ (+4.1 ± 1.7, _p_ = 0.013) was associated with larger lesions (Figure 5). Using this composi�onal regression, replacing one-half unit of ‘Amelonado’ ancestry with ‘Nanay’ ancestry corresponds to an expected GEBV decrease of approximately 6.7 mm. Coefficients for ‘Iquitos’ and ‘Nacional’ ancestry were not dis�nguishable from zero in this exploratory regression, whereas ‘Curaray’ and ‘Bolivian’ ancestry were too rare among phenotyped clones for reliable es�ma�on. Variance infla�on factors were moderate, ranging from 1.0 to 1.5. 19 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) **Discussion** **A sparse screen as a first map of resistance to the emerging pathogen** _**P. megakarya**_ When this screen began, quan�ta�ve differences in aggressiveness among diverse _P. megakarya_ isolates, and the extent to which such varia�on affected genomic predic�on of host resistance were poorly characterized. We therefore sampled broadly across the gene�cally diverse IC3 collec�on rather than restric�ng screening to a small set of previously characterized resistance sources. To our knowledge, this is the largest detached-pod screen against _P. megakarya_ in number of isolates (58) and one of the largest in number of clones (350 genotyped clones). This screening provides the first genomic predic�ons of resistance to this pathogen for the en�re IC3 germplasm collec�on at CATIE. Three findings define the principal results of this study. First, genomic rela�onships among cacao clones improved model fit and enabled predic�on of resistance for all clones in the IC3 collec�on. Second, _P. megakarya_ isolates differed substan�ally in aggressiveness, and broad pathogen subpopula�ons differed in mean lesion severity, while the inclusion of genomic rela�onships among isolates did not improve predic�ons of resistance. Third, clone rankings were largely concordant across pathogen subpopula�ons. **Host relatedness improves fit and predic�on** Ranking the models by AIC gave a consistent picture (Table 1). Subpopula�on-specific variances provided the model that fit the data best (Model 6), followed by inclusion of the cacao rela�onship matrix alone (Model 2). Adding the pathogen rela�onship matrix, alone (Model 3) or as an interac�on term incorpora�ng both rela�onship matrices (Models 4 and 5), did not improve fit. Incorpora�ng the cacao GRM increased the propor�on of phenotypic variance a�ributed to the host genomic component from 0.20 to 0.27 and improved model fi�ng, consistent with genomic rela�onships recovering informa�on that was otherwise unavailable for sparsely phenotyped clones. Clone-wise cross-valida�on showed that this informa�on can be transferred to phenotypically masked clones within the related IC3 popula�on (predic�ve ability 0.41; es�mated accuracy 0.52), a level comparable to host predic�ons in the maize– _Fusarium_ _ver�cillioides_ dual-genome study (Hudson et al., 2024). Romero Navarro et al. (2017) reported accuracies of about 0.5 for black pod in cacao using means of eight years of field observa�ons under natural _P. palmivora_ infec�on. Because clones were assigned to folds at random, valida�on clones generally retained genomically related individuals in the training set. The resul�ng accuracy therefore describes interpola�on within IC3, and further valida�on is required before extrapola�ng model predic�ons into other popula�ons. The analysis illustrates the u�lity of rela�onship-matrix scaling when comparing variance components across different popula�ons. Before the rela�onship matrices were rescaled, models with the pathogen matrix a�ributed 59–61% of phenotypic variance to isolates, nearly twice the es�mate from models trea�ng isolates as unrelated. This infla�on resulted from differences in GRM scale rather than evidence for substan�ally greater pathogen gene�c variance. In this predominantly clonal diploid pathogen, a large propor�on of retained loci were fixed 20 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) heterozygous within subpopula�ons, resul�ng in GRMs whose diagonal scale differed strongly from that of cacao. Scaling each matrix to a common base (Legarra, 2016) removed the discrepancy and suggests that pathogen rela�onship matrices be reported along with their scale. **Isolates differ, but within-subpopula�on genomic relatedness does not explain their** **aggressiveness** Differences among isolates explained more lesion-size varia�on than differences among clones (0.34 against 0.27 of phenotypic variance in Model 2), and in this dataset subpopula�on A produced lesions 6–7 mm larger than subpopula�ons B and C. Working with eight isolates, Ali et al. (2016) found _P. megakarya_ uniformly aggressive on pod husks; our larger panel shows that isolates of this species do differ substan�ally. A�er accoun�ng for mean differences among pathogen subpopula�ons, incorpora�ng within-subpopula�on genomic rela�onships among isolates did not improve model fit or predic�on accuracy. The substan�al varia�on among isolate effects was not well represented by genome-wide relatedness at the captured loci. At least three features of the dataset and pathosystem may explain this result. First, _P. megakarya_ reproduces mainly clonally (Nyassé et al., 1999), and 84–92% of captured SNPs were fixed heterozygous within subpopula�ons, leaving 882–5,309 informa�ve SNPs within the virulence associated genes selected for sequencing. Genome-wide relatedness at the retained loci may primarily reflect clonal descent and broad genomic similarity, rather than varia�on in gene _P._ content or expression differences that underlie aggressiveness. Second, about 84% of _megakarya_ genes are duplicated a�er a recent whole-genome duplica�on (Morales-Cruz et al., 2020). Collapsed or highly similar paralogous sequences can generate apparent heterozygosity when reads from different gene copies map to a single reference locus. Consequently, some of the observed fixed heterozygosity may reflect sequence differences between paralogs rather than allelic varia�on. Although genuine gene�c varia�on among individuals may s�ll be detectable at these loci, collapsing paralogous sequences can mask allelic differences or distort genotype calls, poten�ally leading to an underes�ma�on of gene�c diversity. Third, isolate effects were es�mated with error, because 51% of isolate pairs could not be compared on the same clone and the control isolate GH34 alone was used to inoculate 227 pods. Part of the isolate variance may also be non-gene�c, reflec�ng culture history or inoculum quality, so parameter es�mates for the pathogen are reported as the share of phenotypic variance rather than as a gene�c parameter es�mate for the ‘heritability’ of pathogen virulence. Previous dual-genome studies likewise found limited improvement in host predic�on a�er incorpora�ng pathogen genomic rela�onships, although the pathogen biology and popula�on structures differed substan�ally from those of _P. megakarya_ (Domínguez-Rondón et al., 2025; Hudson et al., 2024). Other pathogen popula�ons studied were haploid ascomycetes with weak popula�on structure and regular ( _Zymoseptoria tri�ci_ ) (Morais et al., 2019) or occasional ( _F._ _ver�cillioides_ ) (Fallahi et al., 2021) recombina�on. Our study therefore tested the approach using a diploid, clonally reproducing oomycete with strong geographic structure. 21 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) The differences between subpopula�ons need one caveat. Our isolates were assembled from collec�ons made for different purposes, and isolates sampled at an epidemic front may be more aggressive than those baited from soil or other parts of the tree; without consistent sample metadata, intrinsic aggressiveness cannot be separated from sampling origin. The broad west-to east genomic structure observed here is consistent with previously described geographic differen�a�on of _P. megakarya_ popula�ons across West and Central Africa by Nyassé et al. (1999), including intermediate genotypes near the Nigeria–Cameroon border. The larger number of segrega�ng SNPs observed in subpopula�on C is consistent with greater diversity in popula�ons in Cameroon, although mapping all isolates to the Ghanaian GH34 reference could also inflate varia�on in more divergent lineages. The sequence data may provide a resource for further studies of the diversity of virulence associated in _P. megakarya_ . Subpopula�ons differed in which effector and protease genes carried SNPs (Figure 3), and a paralogue-aware analysis of effector presence–absence varia�on may be used to iden�fy isolates with different virulence gene repertoires. Screening cacao popula�ons with a few diverse isolates would improve the odds of iden�fying isolate or effector-specific resistance mechanisms within cacao collec�ons. **Concordant clone rankings, unresolved interac�on** The genomic correla�on of clone effects across pathogen subpopula�ons (0.86 ± 0.06) indicates substan�al concordance in host rankings across the three modeled pathogen groups. Unlike type B correla�ons, which approach unity whenever the interac�on variance is small, this correla�on is es�mated as a parameter in its own right. This agrees with earlier evidence of broadly consistent resistance rankings across _P. palmivora_ isolates (Surujdeo-Maharaj et al., 2001), between _P. palmivora_ and _P. megakarya_ (Nyadanu et al., 2012) and across countries (Nyassé et al., 2007). Excep�ons exist: SCA 6 restricts _P. palmivora_ but is invaded by _P. megakarya_ (Ali et al., 2016; Herrera-Corzo et al., 2026), so resistance to one species does not guarantee resistance to the other. The fi�ed clone-by-isolate interac�on accounted for only 0.02 of phenotypic variance and was not sta�s�cally dis�nguishable from zero (P = 0.07). Given the sparse connec�vity of the design, however, this result should be interpreted as limited evidence for interac�on rather than evidence that clone-by-isolate specificity is absent. This reflects the experimental design as much as the biology. Only 5% of possible combina�ons were tested, most of them once, and lesion sizes on POUND-7, tested with 18 isolates, varied in ways that a single variance component averages away. While a fully factorial screen is not required to provide es�mates of interac�on variance, tradi�onal designs using a selec�on of diverse cacao clones and isolates would provide more precise es�mates of variance components. Circular alloca�ons of isolates to hosts, as used by Hudson et al. (2024) may also be used to increase the efficiency per observa�on (Huber et al., 1992) and improvements in the accuracy of predic�ons from sparse designs rise with the overlap of genotypes across test condi�ons (Atanda et al., 2022). Combining the sparse tes�ng strategy of this study, which provides phenotype data for many clones and genomic predic�ons for all clones, with more tradi�onal screening of a small set of varie�es would provide both valida�on 22 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) and evalua�on of the trade-off between tes�ng large numbers of clones and obtaining precise gene�c parameter es�mates. **What lesion size on detached pods measures** The phenotype measured here represents resistance of unwounded pods (harvested at approximately four-month since forma�on) to penetra�on and early lesion expansion four days a�er inocula�on. It captured penetra�on and early lesion expansion on the pod, but escape through flowering �me, frui�ng-cycle length, or pod posi�on also contribute to field resistance (Efombagn et al., 2004). Pods were inoculated 4–13 days a�er harvest due to shipping constrants, so varia�on associated with post-harvest interval likely contributed to residual varia�on and reduced precision of clone and isolate effects. Laboratory and field rankings agree at the level of genotype means, but not always significantly (Efombagn et al., 2011), and natural field infec�on can be so dominated by environment that its broad-sense heritability falls to 0.15 (Nyadanu et al., 2012). Controlled assays therefore remain the prac�cal basis for ranking genotypes, but the top-ranked clones iden�fied here require field evalua�on under _P. megakarya_ pressure for defini�ve valida�on. **Gene�c groups and candidate parents** In the exploratory ancestry regression, higher propor�ons of ‘Nanay’, ‘Contamana’, ‘Purus’, and ‘Maranon’ ancestry were associated with smaller predicted lesion sizes. The nega�ve ‘Maranon’ coefficient is consistent with previous reports of black-pod resistance in PA material and with lower natural black-pod incidence among clones carrying ‘Maranon’ ancestry (Iwaro et al., 2005; Romero Navarro et al., 2017). Our predic�ons also place an established ‘Nanay‘ source, POUND 7 (GEBV −12.2 ± 1.7 mm, 26 pods), and several unphenotyped ‘NA’ clones among the most resistant. ‘Guiana’ ancestry showed the opposite trend in the present dataset, with higher ‘Guiana’ ancestry associated with larger predicted lesions. This finding contrasts with wild ‘Guiana’ genotypes described as resistant to both _P. megakarya_ (Paulin et al., 2008) and _P._ _palmivora_ (Thevenin et al., 2012). Those studies used a single isolate per pathogen species, whereas our clones were challenged with up to four different isolates, many from the more aggressive subpopula�on A; the wild French Guiana popula�ons are not the ‘GU’ clones held at CATIE; and leaf and pod resistance overlap only partly (Iwaro et al., 1997). Not every reported source of resistance ranked highly either: PA-150, a good general combiner for resistance in Ghana (Nyadanu et al., 2012), was close to the popula�on mean (+0.3 ± 3.7 mm). Because the response variable consisted of marker-derived, shrunken GEBVs and the phenotyped clones were a non-random, predominantly admixed subset of IC3, these ancestry associa�ons should be interpreted as hypothesis-genera�ng trends rather than es�mates of causal gene�c-group effects. **From predic�ons to breeding sugges�ons** Valida�on of the genomic predic�on model is needed before making strong recommenda�ons on its applica�on to reduce the impact of black pod disease. The experimental design enabled 23 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) tes�ng of a large number of cacao varie�es, leveraging genomic connec�vity across bioassays to provide unbiased predic�ons for any genotyped cacao clone. This approach accommodates temporal variability in pod availability and is well suited for the ini�al characteriza�on of germplasm collec�ons and variable pathogen popula�ons. The IC3 germplasm collec�on is highly diverse and is likely related to ex-situ germplasm popula�ons used around the world, providing some confidence that the IC3-trained predic�on models may be applicable globally. However, predic�on accuracy may decline in gene�cally dis�nct germplasm collec�ons, as model transferability depends on differences in resistance among popula�ons, shared gene�c ancestry, and consistent linkage between gene�c markers and resistance-associated loci. A prac�cal sugges�on for responding to the black pod epidemic in Africa is to ensure clones from popula�ons demonstra�ng resistance, such as ‘Nanay’, ‘Contamana’, ‘Purus’, are made available for evalua�on in the region. Importa�on of clones with predicted resistance for field evalua�on in areas with high infec�on and for use as parents to cross with local selec�ons is also recommended. Genotyping African germplasm collec�ons using the IC3-trained predic�on model to iden�fy 10 to 20 resistant and suscep�ble clones may be used to focus valida�on on clones with contras�ng resistance predic�ons. Subsequent screening with local isolates would provide the phenotype data required for model valida�on in independent popula�ons. Given the stability of cacao clone rankings across pathogen popula�ons, phenotype and genotype data from African popula�ons screened for resistance to local isolates may be incorporated into a general resistance predic�on model that the cacao community may build upon. As isolates evolve, or if repeatable changes in rankings across cacao clones are iden�fied, isolate genotypes may be incorporated into predic�on models to account for interac�ons not detected in this study. Under the average fi�ed pathogen background and the lesion-size thresholds used for this screening assay, 215 clones had predicted lesion diameters below 30 mm, including 119 clones without direct phenotype records. These values may be used to rank clones for valida�on studies that would provide more defini�ve resistance classifica�ons. Their prac�cal value is in priori�zing clones from a large germplasm collec�on for targeted phenotyping and crossing. Releasing a new cacao variety can take 20–30 years (Kulesza et al., 2024), and _P. megakarya_ is s�ll confined to West and Central Africa. Once validated, predic�ons from a quaran�ne screen could also help breeders in regions not yet invaded to decide which clones to test first, and the framework accoun�ng for both host and pathogen variability can be updated as addi�onal screening and field data accumulate. The heterogeneous correla�on structure was of par�cular interest as it provides es�mates of the genomic correla�ons among subpopula�ons and indicates the stability of resistance across cacao clones. The es�mate of 0.86 indicates strong concordance in cacao resistance predic�ons across _P. megakarya_ subpopula�ons, sugges�ng that subpopula�on-specific resistance mechanisms were not iden�fied. Screening with isolates from different popula�ons provided similar predic�ons of resistance, sugges�ng that these results may be combined with addi�onal 24 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) screening data from African popula�ons to priori�ze clones for the development of resistance and to expand the inferences we can draw using genomic predic�on models. **Conclusions** We combined genotyping of the CATIE Interna�onal Cacao Collec�on with capture sequencing of _P. megakarya_ and a sparse detached-pod screen to produce the first genomic predic�ons of resistance to black pod disease. Genomic rela�onships among cacao clones improved model fit and enabled predic�on of phenotypically untested clones, whereas modeling within subpopula�on genomic rela�onships among _P. megakarya_ isolates provided no measurable improvement beyond adjus�ng for fixed pathogen subpopula�on effects. Clone rankings were broadly concordant across the three pathogen subpopula�ons, and exploratory ancestry analyses iden�fied several Upper Amazon ancestry components associated with lower predicted lesion size. Replicated screening of a well-connected subset of clones and isolates, field valida�on of the top-ranked clones, and examina�on of drivers of varia�on in pathogen virulence are the next steps. More broadly, the study illustrates how genomic rela�onships can be used to connect sparse disease-screening datasets in perennial crops and how pathogen genomic covariance structures can be incorporated into resistance and virulence predic�on models. Host genomic rela�onships can extend informa�on from a sparse screening popula�on to related unphenotyped germplasm, while explicit evalua�on and appropriate scaling of pathogen rela�onship matrices are essen�al before a�ribu�ng phenotypic varia�on to pathogen genomic effects. 25 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) **Literature** Akrofi, A. Y., Amoako-A�a, I., Assuah, M., & Asare, E. K. (2015). Black pod disease on cacao ( _Theobroma cacao_, L) in Ghana: Spread of _Phytophthora megakarya_ and role of economic plants in the disease epidemiology. _Crop_ _Protec�on_, _72_, 66–75. h�ps://doi.org/10.1016/j.cropro.2015.01.015 Ali, S. S., Amoako-A�ah, I., Bailey, R. A., Strem, M. D., Schmidt, M., Akrofi, A. Y., Surujdeo-Maharaj, S., Kolawole, O. O., Begoude, B. A. D., Ten Hoopen, G. M., Goss, E., Phillips-Mora, W., Meinhardt, L. W., & Bailey, B. A. (2016). PCR-based iden�fica�on of cacao black pod causal agents and iden�fica�on of biological factors possibly contribu�ng to _Phytophthora megakarya_ ’s field dominance in West Africa. _Plant_ _Pathology_, _65_ (7), 1095–1108. h�ps://doi.org/10.1111/ppa.12496 Ali, S. S., Shao, J., Lary, D. J., Strem, M. D., Meinhardt, L. W., & Bailey, B. A. (2017). _Phytophthora_ _megakarya_ and _P. palmivora_, Causal Agents of Black Pod Rot, Induce Similar Plant Defense Responses Late during Infec�on of Suscep�ble Cacao Pods. _Fron�ers in Plant Science_, _8_ . h�ps://doi.org/10.3389/fpls.2017.00169 Allegre, M., Argout, X., Boccara, M., Fouet, O., Roguet, Y., Berard, A., Thevenin, J. M., Chauveau, A., Rivallan, R., Clement, D., Courtois, B., Gramacho, K., Boland-Auge, A., Tahi, M., Umaharan, P., Brunel, D., & Lanaud, C. (2012). Discovery and mapping of a new expressed sequence tag-single nucleo�de polymorphism and simple sequence repeat panel for large-scale gene�c studies and breeding of _Theobroma_ _cacao_ L. _DNA_ _Research_, _19_ (1), 23–35. h�ps://doi.org/10.1093/dnares/dsr039 Amadeu, R. R., Garcia, A. A. F., Munoz, P. R., & Ferrão, L. F. V. (2023). AGHmatrix: Gene�c rela�onship matrices in R. _Bioinforma�cs_, _39_ (7), btad445. h�ps://doi.org/10.1093/bioinforma�cs/btad445 Andrews, S. (2010). _FastQC: A Quality Control Tool for High Throughput Sequence Data_ (Version 0.12.1) [Computer so�ware]. Babraham Bioinforma�cs. h�ps://www.bioinforma�cs.babraham.ac.uk/projects/fastqc/ Argout, X., Mar�n, G., Droc, G., Fouet, O., Labadie, K., Rivals, E., Aury, J. M., & Lanaud, C. (2017). The cacao Criollo genome v2.0: An improved version of the genome for gene�c and func�onal genomic studies. _BMC Genomics_, _18_ (1), 730. h�ps://doi.org/10.1186/s12864-017-4120-9 Atanda, S. A., Govindan, V., Singh, R., Robbins, K. R., Crossa, J., & Bentley, A. R. (2022). Sparse tes�ng using genomic predic�on improves selec�on for breeding targets in elite spring wheat. _Theore�cal and Applied Gene�cs_, _135_ (6), 1939–1950. h�ps://doi.org/10.1007/s00122-022 04085-0 Baek, I., Cha, M., Lim, S., Irish, B. M., Oh, S., Bha�, J., Upadhyay, R. K., Kim, M. S., Meinhardt, L. W., Park, S., & Ahn, E. (2025). Integrated phenotypic analysis, predic�ve modeling, and 26 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) _Theobroma cacao_ collec�on. _BMC Plant_ iden�fica�on of novel trait-associated loci in a diverse _Biology_, _25_ (1), 1050. h�ps://doi.org/10.1186/s12870-025-07128-y Baek, I., Upadhyay, R. K., Kim, M. S., Meinhardt, L. W., & Ahn, E. (2026). Longitudinal genomic _Fron�ers_ predic�on of cacao disease resilience iden�fies robust witches’ broom disease targets. _in Plant Science_, _17_, 1837102. h�ps://doi.org/10.3389/fpls.2026.1837102 Bailey, B. A., Ali, S. S., Akrofi, A. Y., & Meinhardt, L. W. (2016). _Phytophthora megakarya_, a Causal Agent of Black Pod Rot in Africa. In B. A. Bailey & L. W. Meinhardt (Eds.), _Cacao Diseases_ (pp. 267– 303). Springer Interna�onal Publishing. h�ps://doi.org/10.1007/978-3-319-24789-2_8 Barter, R. L., & Yu, B. (2018). Superheat: An R Package for Crea�ng Beau�ful and Extendable Heatmaps for Visualizing Complex Data. _Journal of Computa�onal and Graphical Sta�s�cs_, _27_ (4), 910–922. h�ps://doi.org/10.1080/10618600.2018.1473780 Bekele, F. L., Bidaisee, G. G., Allegre, M., Argout, X., Fouet, O., Boccara, M., Saravanakumar, D., Bekele, I., & Lanaud, C. (2022). Genome-wide associa�on studies and genomic selec�on assays made in a large sample of cacao ( _Theobroma cacao_ L.) germplasm reveal significant marker-trait associa�ons and good predic�ve value for improving yield poten�al. _PLOS ONE_, _17_ (10), e0260907. h�ps://doi.org/10.1371/journal.pone.0260907 Brown, J. S., Phillips-Mora, W., Power, E. J., Krol, C., Cervantes-Mar�nez, C., Motamayor, J. C., & Schnell, R. J. (2007). Mapping QTLs for Resistance to Frosty Pod and Black Pod Diseases and Hor�cultural Traits in _Theobroma_ _cacao_ L. _Crop_ _Science_, _47_ (5), 1851–1858. h�ps://doi.org/10.2135/cropsci2006.11.0753 Burdon, R. D. (1977). _Gene�c correla�on as a concept for studying genotype-environment_ _interac�on in forest tree breeding_ . _26_ (5–6), 168–175. CacaoNet. (2012). _A Global Strategy for the Conserva�on and Use of Cacao Gene�c Resources, as_ _the Founda�on for a Sustainable Cocoa Economy_ (Nos. 978-92-9043-920–2). Bioversity Interna�onal. h�ps://www.cacaonet.org/fileadmin/templates/CacaoNet/Uploads/publica�ons/CacaoStrategy -sept2012_full.pdf Chen, S., Zhou, Y., Chen, Y., & Gu, J. (2018). fastp: An ultra-fast all-in-one FASTQ preprocessor. _Bioinforma�cs_, _34_ (17), i884–i890. h�ps://doi.org/10.1093/bioinforma�cs/bty560 Cleary, J. G., Braithwaite, R., Gaastra, K., Hilbush, B. S., Inglis, S., Irvine, S. A., Jackson, A., Li�n, R., Rathod, M., Ware, D., Zook, J. M., Trigg, L., & De La Vega, F. M. (2015). _Comparing Variant Call_ _Files for Performance Benchmarking of Next-Genera�on Sequencing Variant Calling Pipelines_ . Bioinforma�cs. h�ps://doi.org/10.1101/023754 Danecek, P., Bonfield, J. K., Liddle, J., Marshall, J., Ohan, V., Pollard, M. O., Whitwham, A., Keane, T., McCarthy, S. A., Davies, R. M., & Li, H. (2021). Twelve years of SAMtools and BCFtools. _GigaScience_, _10_ (2), giab008. h�ps://doi.org/10.1093/gigascience/giab008 27 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) Domínguez-Rondón, A., Tirado, R., Solís, I., Sánchez-Vallet, A., & Isidro Y Sánchez, J. (2025). Integra�ng host–pathogen interac�on genomics to boost wheat defense against septoria tri�ci blotch. _BMC Genomics_, _26_ (1), 1066. h�ps://doi.org/10.1186/s12864-025-12174-9 Efombagn, M. I. B., Bieysse, D., Nyassé, S., & Eskes, A. B. (2011). Selec�on for resistance to Phytophthora pod rot of cocoa ( _Theobroma cacao_ L.) in Cameroon: Repeatability and reliability of screening tests and field observa�ons. _Crop_ _Protec�on_, _30_ (2), 105–110. h�ps://doi.org/10.1016/j.cropro.2010.10.012 Efombagn, M. I. B., Marelli, J. P., Ducamp, M., Cilas, C., Nyassé, S., & Vefonge, D. (2004). Effect of Frui�ng Traits on the Field Resistance of Cocoa ( _Theobroma cacao_ L.) Clones to _Phytophthora_ _megakarya_ . _Journal of Phytopathology_, _152_ (10), 557–562. h�ps://doi.org/10.1111/j.1439 0434.2004.00895.x Enciso-Rodriguez, F., Douches, D., Lopez-Cruz, M., Coombs, J., & De Los Campos, G. (2018). Genomic Selec�on for Late Blight and Common Scab Resistance in Tetraploid Potato ( _Solanum_ _tuberosum_ ). _G3_ _Genes|Genomes|Gene�cs_, _8_ (7), 2471–2481. h�ps://doi.org/10.1534/g3.118.200273 Engels, J. M. M. (1981). _Gene�c resources of cacao: A catalogue of the CATIE collec�on_ (Technical Bulle�n No. 7). Centro Agronómico Tropical de Inves�gación y Enseñanza, Turrialba, Costa Rica. Unidad de Recursos Fitogené�cos. h�ps://repositorio.ca�e.ac.cr/handle/11554/2231 Ewels, P., Magnusson, M., Lundin, S., & Käller, M. (2016). Mul�QC: Summarize analysis results for mul�ple tools and samples in a single report. _Bioinforma�cs_, _32_ (19), 3047–3048. h�ps://doi.org/10.1093/bioinforma�cs/btw354 Fallahi, M., Somma, S., Javan-Nikkhah, M., Saremi, H., Stea, G., Masiello, M., Logrieco, A. F., & More�, A. (2021). Gene�c structure of _Fusarium ver�cillioides_ popula�ons from maize in Iran. _Fungal Gene�cs and Biology_, _156_, 103613. h�ps://doi.org/10.1016/j.fgb.2021.103613 Ferguson, A. J., & Jeffers, S. N. (1999). Detec�ng Mul�ple Species of _Phytophthora_ in Container Mixes from Ornamental Crop Nurseries. _Plant_ _Disease_, _83_ (12), 1129–1136. h�ps://doi.org/10.1094/PDIS.1999.83.12.1129 Fister, A. S., Leandro-Muñoz, M. E., Zhang, D., Marden, J. H., Tiffin, P., dePamphilis, C., Maximova, S., & Guil�nan, M. J. (2020). Widely distributed varia�on in tolerance to _Phytophthora palmivora_ in four gene�c groups of cacao. _Tree_ _Gene�cs_ _&_ _Genomes_, _16_ (1), 1. h�ps://doi.org/10.1007/s11295-019-1396-8 Garrison, E., & Marth, G. (2012). _Haplotype-based variant detec�on from short-read sequencing_ (Version 2). arXiv. h�ps://doi.org/10.48550/ARXIV.1207.3907 Gezan, S., Murray, D., Avelar De Oliveira, A., & Galli, G. (2022). _ASRgenomics: Complementary_ _Genomic_ _Func�ons_ (p. 1.1.6) [Computer so�ware]. h�ps://doi.org/10.32614/CRAN.package.ASRgenomics 28 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) Gi�o, A. J., Schlathoelter, I., Herrera-Corzo, M., Marelli, J.-P., Goss, E. M., & Brawner, J. T. (2026). _Detached Pod-Zoospore Patch (DP-ZP) Phytophthora spp. Cacao Bioassay_ [Protocol]. University of Florida. h�ps://doi.org/h�ps://dx.doi.org/10.17504/protocols.io.n2bvjekypgk5/v1 Herrera-Corzo, M., Schlathoelter, I., Arciniegas-Leal, A., Bailey, B. A., Puig, A. S., Jaimes Suárez, Y. Y., Sundram, S., Nguyen, T. T. N., Bri�o, D. S., Rodezno, D. C., Drost, D. R., Marelli, J.-P., Brawner, J. T., & Goss, E. M. (2026). Evalua�ng cacao pod resistance to black pod disease caused by three _Phytophthora_ species. bioRxiv 2026.09.17.752234. h�ps://doi.org/10.64898/2026.09.17.752234 Huber, D. A., White, T. L., & Hodge, G. R. (1992). The Efficiency of Half-Sib, Half-Diallel and Circular Ma�ng Designs in the Es�ma�on of Gene�c Parameters in Forestry: A Simula�on. _Forest Science_, _38_ (4), 757–776. h�ps://doi.org/10.1093/forestscience/38.4.757 Hudson, O., & Brawner, J. (2025). Using genome-wide associa�ons and host-by-pathogen predic�ons to iden�fy allelic interac�ons that control disease resistance. _The Plant Genome_, _18_ (1), e70006. h�ps://doi.org/10.1002/tpg2.70006 Hudson, O., Resende, M. F. R., Messina, C., Holland, J., & Brawner, J. (2024). Predic�on of resistance, virulence, and host-by-pathogen interac�ons using dual-genome predic�on models. _Theore�cal and Applied Gene�cs_, _137_ (8), 196. h�ps://doi.org/10.1007/s00122-024-04698-7 Husson, F., Josse, J., Le, S., & Mazet, J. (2006). _FactoMineR: Mul�variate Exploratory Data Analysis_ _and_ _Data_ _Mining_ (p. 2.18) [Computer so�ware]. h�ps://doi.org/10.32614/CRAN.package.FactoMineR IMARC Group. (2024, November 4). _Global Chocolate Market to Reach USD 219.1 Billion by 2033,_ _Propelled by Growing Popularity of Chocolate-Based Food Items_ . IMARC Group. h�ps://www.imarcgroup.com/global-chocolate-market Interna�onal Cocoa Organiza�on. (2025). _Produc�on of Cocoa Beans: LI_ (Quarterly Bulle�n of Cocoa Sta�s�cs, Issue 2). Interna�onal Cocoa Organiza�on. h�ps://www.icco.org/produc�on_qbcs-li-no-2/ Iwaro, A. D., Sreenivasan, T. N., & Umaharan, P. (1997). Foliar Resistance to _Phytophthora_ _palmivora_ as an Indicator of Pod Resistance in _Theobroma cacao_ . _Plant Disease_, _81_ (6), 619–624. h�ps://doi.org/10.1094/PDIS.1997.81.6.619 Iwaro, A. D., Thévenin, J.-M., Butler, D. R., & Eskes, A. B. (2005). Usefulness of the Detached Pod Test for Assessment of Cacao Resistance to Phytophthora Pod Rot. _European Journal of Plant_ _Pathology_, _113_ (2), 173–182. h�ps://doi.org/10.1007/s10658-005-2929-6 Jacquo�e, E., Holzwarth, J. A., Mota, A. Z., Leandro-Muñoz, M. E., Rhoné, B., Todd, E. T., Lepelley, M., Bellanger, L., Descombes, P., Arigoni, F., Fillodeau, A., & Beche, E. (2026). Associa�on mapping and genome predic�on reveal a complex gene�c architecture for resistance to black and frosty pod rot in cacao ( _Theobroma cacao_ L.) popula�ons. _The Plant Genome_, _19_ (4), e70322. h�ps://doi.org/10.1002/tpg2.70322 29 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) Jarquín, D., Crossa, J., Lacaze, X., Du Cheyron, P., Daucourt, J., Lorgeou, J., Piraux, F., Guerreiro, L., Pérez, P., Calus, M., Burgueño, J., & De Los Campos, G. (2014). A reac�on norm model for genomic selec�on using high-dimensional genomic and environmental data. _Theore�cal and Applied_ _Gene�cs_, _127_ (3), 595–607. h�ps://doi.org/10.1007/s00122-013-2243-1 Kassambara, A. (2016). _ggpubr: “ggplot2” Based Publica�on Ready Plots_ (p. 1.0.0) [Computer so�ware]. h�ps://doi.org/10.32614/CRAN.package.ggpubr Kassambara, A., & Mundt, F. (2016). _factoextra: Extract and Visualize the Results of Mul�variate_ _Data_ _Analyses_ (p. 2.2.0) [Computer so�ware]. h�ps://doi.org/10.32614/CRAN.package.factoextra Kissi, E. A., & Herzig, C. (2024). The implica�ons of governance factors for economic and social upgrading in Ghana’s cocoa value chain. _Agricultural and Food Economics_, _12_ (1). h�ps://doi.org/10.1186/s40100-024-00295-w Knaus, B. J., & Grünwald, N. J. (2017). VCFR : A package to manipulate and visualize variant call format data in R. _Molecular Ecology Resources_, _17_ (1), 44–53. h�ps://doi.org/10.1111/1755 0998.12549 Kulesza, E., Thomas, P., Prewi�, S. F., Shalit-Kaneh, A., Wafula, E., Knollenberg, B., Winters, N., Esteban, E., Pasha, A., Provart, N., Praul, C., Landherr, L., dePamphilis, C., Maximova, S. N., & Guil�nan, M. J. (2024). The cacao gene atlas: A transcriptome developmental atlas reveals highly _Theobroma cacao_ L. _BMC Plant_ �ssue-specific and dynamically-regulated gene networks in _Biology_, _24_ (1), 601. h�ps://doi.org/10.1186/s12870-024-05171-9 Legarra, A. (2016). Comparing es�mates of gene�c variance across different rela�onship models. _Theore�cal Popula�on Biology_, _107_, 26–30. h�ps://doi.org/10.1016/j.tpb.2015.08.005 Li, H. (2013). _Aligning sequence reads, clone sequences and assembly con�gs with BWA-MEM_ (Version 2). arXiv. h�ps://doi.org/10.48550/ARXIV.1303.3997 Li, H., & Durbin, R. (2009). Fast and accurate short read alignment with Burrows–Wheeler transform. _Bioinforma�cs_, _25_ (14), 1754–1760. h�ps://doi.org/10.1093/bioinforma�cs/btp324 Livingstone, D., Royaert, S., Stack, C., Mockai�s, K., May, G., Farmer, A., Saski, C., Schnell, R., Kuhn, D., & Motamayor, J. C. (2015). Making a chocolate chip: Development and evalua�on of a 6K SNP array for _Theobroma_ _cacao_ . _DNA_ _Research_, _22_ (4), 279–291. h�ps://doi.org/10.1093/dnares/dsv009 Livingstone, D., Stack, C., Mus�ga, G. M., Rodezno, D. C., Suarez, C., Amores, F., Feltus, F. A., Mockai�s, K., Cornejo, O. E., & Motamayor, J. C. (2017). A Larger Chocolate Chip—Development of a 15K _Theobroma cacao_ L. SNP Array to Create High-Density Linkage Maps. _Fron�ers in Plant_ _Science_, _8_, 2008. h�ps://doi.org/10.3389/fpls.2017.02008 30 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) Lorenz, A. J., Smith, K. P., & Jannink, J. -L. (2012). Poten�al and Op�miza�on of Genomic Selec�on for Fusarium Head Blight Resistance in Six-Row Barley. _Crop Science_, _52_ (4), 1609–1621. h�ps://doi.org/10.2135/cropsci2011.09.0503 Marelli, J.-P., Guest, D. I., Bailey, B. A., Evans, H. C., Brown, J. K., Junaid, M., Barreto, R. W., Lisboa, D. O., & Puig, A. S. (2019). Chocolate Under Threat from Old and New Cacao Diseases. _Phytopathology®_, _109_ (8), 1331–1343. h�ps://doi.org/10.1094/PHYTO-12-18-0477-RVW Mata-Quirós, A., Arciniegas-Leal, A., Phillips-Mora, W., Meinhardt, L. W., Mo�lal, L., Mischke, S., & Zhang, D. (2018). Assessing hidden parentage and gene�c integrity of the “United Fruit Clones” of cacao ( _Theobroma cacao_ ) from Costa Rica using SNP markers. _Breeding Science_, _68_ (5), 545– 553. h�ps://doi.org/10.1270/jsbbs.18057 Mata-Quirós, A., Arciniegas-Leal, A., Phillips-Mora, W., Meinhardt, L. W., & Zhang, D. (2017). _Understanding the gene�c structure and parentage of the clonal series of cacao UF, CC, PMCT and_ _ARF preserved in the Interna�onal Cacao Collec�on at CATIE (IC3)_ . Interna�onal Symposium on Cocoa Research. McElroy, M. S., Navarro, A. J. R., Mus�ga, G., Stack, C., Gezan, S., Peña, G., Sarabia, W., Saquicela, D., Sotomayor, I., Douglas, G. M., Migicovsky, Z., Amores, F., Tarqui, O., Myles, S., & Motamayor, J. C. (2018). Predic�on of Cacao ( _Theobroma cacao_ ) Resistance to _Moniliophthora_ spp. Diseases via Genome-Wide Associa�on Analysis and Genomic Selec�on. _Fron�ers in Plant Science_, _9_, 343. h�ps://doi.org/10.3389/fpls.2018.00343 Meuwissen, T. H. E., Hayes, B. J., & Goddard, M. E. (2001). Predic�on of Total Gene�c Value Using Genome-Wide Dense Marker Maps. _Gene�cs_, _157_ (4), 1819–1829. h�ps://doi.org/10.1093/gene�cs/157.4.1819 Mfegue, C. V., Herail, C., Adreit, H., Mbenoun, M., Techou, Z., Ten Hoopen, M., Tharreau, D., & Ducamp, M. (2012). Microsatellite markers for popula�on studies of _Phytophthora megakarya_ (Pythiaceae), a cacao pathogen in Africa. _American_ _Journal_ _of_ _Botany_, _99_ (9). h�ps://doi.org/10.3732/ajb.1200053 Morais, D., Duplaix, C., Sache, I., Laval, V., Suffert, F., & Walker, A.-S. (2019). Overall stability in the gene�c structure of a _Zymoseptoria tri�ci_ popula�on from epidemic to interepidemic stages at a small spa�al scale. _European_ _Journal_ _of_ _Plant_ _Pathology_, _154_ (2), 423–436. h�ps://doi.org/10.1007/s10658-018-01666-y Morales-Cruz, A., Ali, S. S., Minio, A., Figueroa-Balderas, R., García, J. F., Kasuga, T., Puig, A. S., Marelli, J.-P., Bailey, B. A., & Cantu, D. (2020). Independent Whole-Genome Duplica�ons Define the Architecture of the Genomes of the Devasta�ng West African Cacao Black Pod Pathogen _Phytophthora_ _megakarya_ and Its Close Rela�ve _Phytophthora_ _palmivora_ . _G3_ _Genes|Genomes|Gene�cs_, _10_ (7), 2241–2255. h�ps://doi.org/10.1534/g3.120.401014 31 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) Moreno, V., Gonzalez, J. R., & Pelegri, D. (2022). _SNPassoc: SNPs-Based Whole Genome_ _Associa�on Studies_ (Version 2.1-0, p. 2.3.1) [Computer so�ware]. Comprehensive R Archive Network (CRAN). h�ps://doi.org/10.32614/CRAN.package.SNPassoc Motamayor, J. C., Lachenaud, P., Da Silva E Mota, J. W., Loor, R., Kuhn, D. N., Brown, J. S., & Schnell, R. J. (2008). Geographic and Gene�c Popula�on Differen�a�on of the Amazonian Chocolate Tree ( _Theobroma cacao_ L). _PLoS ONE_, _3_ (10), e3311. h�ps://doi.org/10.1371/journal.pone.0003311 Motamayor, J. C., Mockai�s, K., Schmutz, J., Haiminen, N., Iii, D. L., Cornejo, O., Findley, S. D., Zheng, P., Utro, F., Royaert, S., Saski, C., Jenkins, J., Podiche�, R., Zhao, M., Scheffler, B. E., Stack, J. C., Feltus, F. A., Mus�ga, G. M., Amores, F., … Kuhn, D. N. (2013). The genome sequence of the most widely cul�vated cacao type and its use to iden�fy candidate genes regula�ng pod color. _Genome Biology_, _14_ (6), r53. h�ps://doi.org/10.1186/gb-2013-14-6-r53 Mrode, R. A. (with Thompson, R. & C.A.B. Interna�onal). (2005). _Linear models for the predic�on_ _of animal breeding values_ (Second edi�on). CABI. h�ps://doi.org/10.1079/9780851990002.0000 Ndoungué Djeumekop, M. M., Ngo Bieng, M.-A., Ribeyre, F., Bonnot, F., Cilas, C., Neema, C., & Ten Hoopen, G. M. (2021). Spa�al and Temporal Analysis of _Phytophthora megakarya_ Epidemic in Newly Established Cacao Planta�ons. _Plant_ _Disease_, _105_ (5), 1448–1460. h�ps://doi.org/10.1094/PDIS-09-19-2024-RE Nowosad, J. (2017). _rcartocolor:_ _“CARTOColors”_ _Pale�es_ (p. 2.1.2) [Data set]. h�ps://doi.org/10.32614/CRAN.package.rcartocolor Nyadanu, D., Akromah, R., Adomako, B., Kwoseh, C., Lowor, S. T., Dzahini-Obiatey, H., Akrofi, A. Y., & Assuah, M. K. (2012). Inheritance and general combining ability studies of detached pod, leaf disc and natural field resistance to _Phytophthora palmivora_ and _Phytophthora megakarya_ in cacao ( _Theobroma cacao_ L.). _Euphy�ca_, _188_ (2), 253–264. h�ps://doi.org/10.1007/s10681-012 0717-x Nyassé, S., Efombagn, M. I. B., Kébé, B. I., Tahi, M., Despréaux, D., & Cilas, C. (2007). Integrated management of _Phytophthora_ diseases on cocoa ( _Theobroma cacao_ L): Impact of plant breeding on pod rot incidence. _Crop_ _Protec�on_, _26_ (1), 40–45. h�ps://doi.org/10.1016/j.cropro.2006.03.015 Nyassé, S., Grivet, L., Risterucci, A. M., Blaha, G., Berry, D., Lanaud, C., & Despréaux, D. (1999). Diversity of _Phytophthora megakarya_ in Central and West Africa revealed by isozyme and RAPD markers. _Mycological_ _Research_, _103_ (10), 1225–1234. h�ps://doi.org/10.1017/S0953756299008369 Paulin, D., Ducamp, M., & Lachenaud, P. (2008). New sources of resistance to _Phytophthora_ _megakarya_ iden�fied in wild cocoa tree popula�ons of French Guiana. _Crop Protec�on_, _27_ (7), 1143–1147. h�ps://doi.org/10.1016/j.cropro.2008.01.004 32 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) Phillips-Mora, W., Arciniegas-Leal, A., Mata-Quirós, A., & Motamayor-Arias, J. C. (Eds.). (2013). _Catalogue of cacao clones: Selected by CATIE for commercial plan�ngs_ . CATIE. Phillips-Mora, W., & Galindo, J. J. (1989). Método de inoculación y evaluación de la resistencia a _Phytophthora_ _palmivora_ en frutos de cacao _(Theobroma_ _cacao)_ . h�ps://repositorio.ca�e.ac.cr/handle/11554/10760 Phillips-Mora, W. M., Johnson, E. S., & Astorga, C. (2007). Recent efforts to improve the gene�c and physical condi�ons of the interna�onal cacao collec�on at CATIE. _Proceedings of the_ _Interna�onal_ _Cocoa_ _Producer’s_ _Conference_, _1_, 611–623. h�ps://www.ars.usda.gov/research/publica�ons/publica�on/?seqNo115=205705&utm_source =chatgpt.com Pokou, D. N., Fister, A. S., Winters, N., Tahi, M., Klo�oloma, C., Sebas�an, A., Marden, J. H., Maximova, S. N., & Guil�nan, M. J. (2019). Resistant and suscep�ble cacao genotypes exhibit defense gene polymorphism and unique early responses to _Phytophthora megakarya_ inocula�on. _Plant Molecular Biology_, _99_ (4–5), 499–516. h�ps://doi.org/10.1007/s11103-019-00832-y Poland, J., & Rutkoski, J. (2016). Advances and Challenges in Genomic Selec�on for Disease Resistance. _Annual Review of Phytopathology_, _54_ (1), 79–98. h�ps://doi.org/10.1146/annurev phyto-080615-100056 Posit team. (2025). _RStudio: Integrated Development Environment for R_ (Version 2025.05.0+496) [Computer so�ware]. Posit So�ware, PBC. h�ps://posit.co/ Pritchard, J. K., Stephens, M., & Donnelly, P. (2000). Inference of Popula�on Structure Using Mul�locus Genotype Data. _Gene�cs_, _155_ (2), 945–959. h�ps://doi.org/10.1093/gene�cs/155.2.945 Puig, A. S., Ali, S., Strem, M., Sicher, R., Gu�errez, O. A., & Bailey, B. A. (2018). The differen�al influence of temperature on _Phytophthora megakarya_ and _Phytophthora palmivora_ pod lesion expansion, mycelia growth, gene expression, and metabolite profiles. _Physiological and Molecular_ _Plant Pathology_, _102_, 95–112. h�ps://doi.org/10.1016/j.pmpp.2017.12.002 R Core Team. (2023). _R: A language and environment for sta�s�cal compu�ng._ (Version 4.3.2) [Computer so�ware]. R Founda�on for Sta�s�cal Compu�ng. h�ps://doi.org/10.32614/R.manuals Rolling, W. R., Dorrance, A. E., & McHale, L. K. (2020). Tes�ng methods and sta�s�cal models of genomic predic�on for quan�ta�ve disease resistance to _Phytophthora sojae_ in soybean [ _Glycine_ _max_ (L.) Merr] germplasm collec�ons. _Theore�cal and Applied Gene�cs_, _133_ (12), 3441–3454. h�ps://doi.org/10.1007/s00122-020-03679-w Romero Navarro, J. A., Phillips-Mora, W., Arciniegas-Leal, A., Mata-Quirós, A., Haiminen, N., Mus�ga, G., Livingstone Iii, D., Van Bakel, H., Kuhn, D. N., Parida, L., Kasarskis, A., & Motamayor, J. C. (2017). Applica�on of Genome Wide Associa�on and Genomic Predic�on for Improvement 33 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) of Cacao Produc�vity and Resistance to Black and Frosty Pod Diseases. _Fron�ers in Plant Science_, _8_, 1905. h�ps://doi.org/10.3389/fpls.2017.01905 Rutkoski, J. E., Poland, J. A., Singh, R. P., Huerta-Espino, J., Bhavani, S., Barbier, H., Rouse, M. N., Jannink, J., & Sorrells, M. E. (2014). Genomic Selec�on for Quan�ta�ve Adult Plant Stem Rust Resistance in Wheat. _The_ _Plant_ _Genome_, _7_ (3), plantgenome2014.02.0006. h�ps://doi.org/10.3835/plantgenome2014.02.0006 Self, S. G., & Liang, K.-Y. (1987). Asympto�c Proper�es of Maximum Likelihood Es�mators and Likelihood Ra�o Tests under Nonstandard Condi�ons. _Journal of the American Sta�s�cal_ _Associa�on_, _82_ (398), 605–610. h�ps://doi.org/10.1080/01621459.1987.10478472 Slowikowski, K. (2016). _ggrepel: Automa�cally Posi�on Non-Overlapping Text Labels with_ _“ggplot2”_ (p. 0.9.8) [Computer so�ware]. h�ps://doi.org/10.32614/CRAN.package.ggrepel Steuernagel, B., Witek, K., Kra�nger, S. G., Ramirez-Gonzalez, R. H., Schoonbeek, H., Yu, G., Baggs, E., Witek, A. I., Yadav, I., Krasileva, K. V., Jones, J. D. G., Uauy, C., Keller, B., Ridout, C. J., & Wulff, B. B. H. (2020). The NLR-Annotator Tool Enables Annota�on of the Intracellular Immune Receptor Repertoire. _Plant Physiology_, _183_ (2), 468–482. h�ps://doi.org/10.1104/pp.19.01273 Surujdeo-Maharaj, S., Umaharan, P., & Iwaro, A. D. (2001). A study of genotype-isolate interac�on in cacao ( _Theobroma cacao_ L.): Resistance of cacao genotypes to isolates of _Phytophthora_ _palmivora_ . _Euphy�ca_, _118_ (3), 295–303. h�ps://doi.org/10.1023/A:1017516217662 Su�on, W., Reeser, P., & Hansen, E. (Eds.). (2015). PROTOCOL 03-02.1: Long-term storage of _Phytophthora_ cultures in water. In _Laboratory Protocols for Phytophthora Species_ (pp. 1–2). The American Phytopathological Society. h�ps://doi.org/10.1094/9780890544969.03.02.1 Tahi, G. M., Kébé, B. I., Sangare, A., Mondeil, F., Cilas, C., & Eskes, A. B. (2006). Foliar resistance of cacao ( _Theobroma cacao_ ) to _Phytophthora palmivora_ as an indicator of pod resistance in the field: Interac�on of cacao genotype, leaf age and dura�on of incuba�on. _Plant Pathology_, _55_ (6), 776– 782. h�ps://doi.org/10.1111/j.1365-3059.2006.01453.x Thevenin, J.-M., Rossi, V., Ducamp, M., Doare, F., Condina, V., & Lachenaud, P. (2012). Numerous Clones Resistant to _Phytophthora palmivora_ in the “Guiana” Gene�c Group of _Theobroma cacao_ L. _PLoS ONE_, _7_ (7), e40915. h�ps://doi.org/10.1371/journal.pone.0040915 Todd, E. T., Arigoni, F., Holzwarth, J. A., Bellanger, L., Descombes, P., Beche, E., Lass, T., Guil�nan, M. J., Maximova, S. N., Leandro, M., Mérot, V., Lepelley, M., & Fillodeau, A. (2025). Developing a core collec�on for the conserva�on of _Theobroma cacao_ ’s gene�c diversity. _BMC Genomics_, _26_ (1), 896. h�ps://doi.org/10.1186/s12864-025-11882-6 Van der Auwera, G., & O’Connor, B. D. (2020). _Genomics in the cloud: Using Docker, GATK, and_ _WDL in Terra_ (First edi�on). O’Reilly Media. 34 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) VanRaden, P. M. (2008). Efficient Methods to Compute Genomic Predic�ons. _Journal of Dairy_ _Science_, _91_ (11), 4414–4423. h�ps://doi.org/10.3168/jds.2007-0980 Ward, M. R., & Griffin, M. J. (Eds.). (1981). Soil phase of cocoa _Phytophthora_ . In _Epidemiology of_ _phytophthora on cocoa in Nigeria: Final report of the Interna�onal Cocoa Black Pod Research_ _Project_ (pp. 50–61). Commonwealth Mycological Ins�tute. Wessel, M., & Quist-Wessel, P. M. F. (2015). Cocoa produc�on in West Africa, a review and analysis of recent developments. _NJAS: Wageningen Journal of Life Sciences_, _74–75_ (1), 1–7. h�ps://doi.org/10.1016/j.njas.2015.09.001 Wickham, H. (2016). _ggplot2: Elegant Graphics for Data Analysis_ (2nd ed. 2016). Springer Interna�onal Publishing : Imprint: Springer. h�ps://doi.org/10.1007/978-3-319-24277-4 Wickham, H., Pedersen, T. L., & Seidel, D. (2011). _scales: Scale Func�ons for Visualiza�on_ (p. 1.4.0) [Computer so�ware]. h�ps://doi.org/10.32614/CRAN.package.scales Xiao, N. (2016). _ggsci: Scien�fic Journal and Sci-Fi Themed Color Pale�es for “ggplot2”_ (p. 5.2.0) [Data set]. h�ps://doi.org/10.32614/CRAN.package.ggsci Yan, L. (2021). _ggvenn: Draw Venn Diagram by “ggplot2”_ (p. 0.1.19) [Computer so�ware]. h�ps://doi.org/10.32614/CRAN.package.ggvenn Zarrillo, S., Gaikwad, N., Lanaud, C., Powis, T., Viot, C., Lesur, I., Fouet, O., Argout, X., Guichoux, E., Salin, F., Loor Solorzano, R., Bouchez, O., Vignes, H., Severts, P., Hurtado, J., Yepez, A., Grive�, L., Blake, M., & Valdez, F. (2018). The use and domes�ca�on of _Theobroma cacao_ during the mid Holocene in the upper Amazon. _Nature Ecology & Evolu�on_, _2_ (12), 1879–1888. h�ps://doi.org/10.1038/s41559-018-0697-x Zhang, D., Mar�nez, W. J., Johnson, E. S., Somarriba, E., Phillips-Mora, W., Astorga, C., Mischke, S., & Meinhardt, L. W. (2012). Gene�c diversity and spa�al structure in a new dis�nct _Theobroma_ _cacao_ L. popula�on in Bolivia. _Gene�c Resources and Crop Evolu�on_, _59_ (2), 239–252. h�ps://doi.org/10.1007/s10722-011-9680-y 35 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) **Acknowledgements** We would like to thank the field workers at CATIE for harves�ng and providing cacao pods, and for technical assistance in producing and shipping pods; Robert Henfling for assistance in shipping pods; Dr. Brian Bailey and Dr. G. Mar�jn ten Hoopen for the provision of the _Phytophthora_ _megakarya_ isolates; the University of Florida for facili�es and infrastructure; and Mars, Incorporated for funding this project. **Disclosures, declara�ons, and project funding** Authors of this paper (DL, AF, DD, DH, DR, JPM) are or were employed by Mars, Incorporated, and the company has provided funding to CATIE and the University of Florida to complete this project. **Supplementary Materials** **Supplementary Table 1: Geographic origin and informa�on on the 67 Phytophthora megakarya** **isolates used in this study.** Low virulent = more than 80% of clones showed no infec�on with these isolates. **Supplementary Table 2: Summary of the gene families, the number of candidate genes, and the** **number of targeted genes for the capture sequencing approach used for** _**Phytophthora**_ _**megakarya**_ **isolates in this study.** The gene annota�ons were obtained from Morales-Cruz et al. (2020, File_S1). **Supplementary Table 3: Results of the STRUCTURE analysis on the 947 clones inves�gated in** **this study.** **Supplementary Table 4: Summary of the number and gene family of the genes with gene�c** **varia�on (SNPs) within each** _**Phytophthora megakarya**_ **subpopula�on.** **Supplementary Table 5: Genomic es�mated breeding values (GEBVs) for cacao clone resistance** **obtained from Model 2 and 6** . Given are the number of inoculated pods in the bioassay (inoc), the predicted breeding value for clone c (Best Linear Unbiased Predic�on, BLUPc), the standard error of predic�on for the breeding value of clone c (SEPc), and the standardized Wald-type test sta�s�c Zc (calculated as BLUPc/SEPc) for Model 2. The same metrics are given for Model 6, separated in GEBVs from subpopula�on A, B, and C. **Supplementary Table 6: Genomic es�mated breeding values for isolate virulence obtained from** **Model 2.** Given are the predicted breeding value for isolate i (Best Linear Unbiased Predic�on, BLUPi), the standard error of predic�on for the breeding value of isolate i (SEPi), the standardized Wald-type test sta�s�c Zi (calculated as BLUPi/SEPi), the number of inoculated pods in the bioassay (inoc), and the assignment of a subpopula�on (subpop), either A, B, or C. **Supplementary Table 7: Raw variance components (mm² ± SE) of each model.** 36 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) **Supplementary Figure 1: Popula�on structure of 67 Phytophthora megakarya isolates. (A)** Principal component analysis (PCA) plot showing PC1 and PC2 from 62,203 SNPs across 67 _P._ _megakarya_ isolates. Each point represents an individual isolate, colored by their country of origin. **(B)** Correla�on matrix of 67 P. megakarya isolates based on marker data. The color bar to the le� of the plot shows the country of origin of each isolate. **Supplementary Figure 2. Addi�onal principal component projec�ons of the gene�c structure of** **947 clones of the IC3 cacao collec�on.** Principal component analysis (PCA) was based on 8,572 SNP markers. **(A)** Projec�on of principal components 2 and 3, explaining 12.0% and 7.6% of the gene�c varia�on, respec�vely. **(B)** Projec�on of principal components 3 and 4, explaining 7.6% and 5.7% of the gene�c varia�on, respec�vely. Each point represents an individual clone. Gene�c group assignment was based on the ancestry component with the highest STRUCTURE membership coefficient. Clones with ≥75% membership in their predominant gene�c group were classified as members and are shown in color, whereas clones with <75% membership were classified as admixed and are shown in gray. Gene�c-group nomenclature follows Motamayor et al. (2008). **Supplementary Figure 3: Popula�on structure of 41 Phytophthora megakarya isolates from** **Ghana and Nigeria.** **(A)** Principal component analysis (PCA) plot showing PC1 and PC2 from 12,342 SNPs across 41 _P. megakarya_ isolates. Each point represents an individual isolate, colored by their country of origin. **(B)** Correla�on matrix of 41 _P. megakarya_ isolates based on marker data. The color bar to the le� of the plot shows the country of origin of each isolate. **Supplementary Figure 4: (A) Popula�on structure of 26** _**Phytophthora megakarya**_ **isolates from** **Cameroon and Nigeria.** **(A)** Principal component analysis (PCA) plot showing PC1 and PC2 from 44,633 SNPs across 26 _P. megakarya_ isolates. Each point represents an individual isolate colored by their country of origin. **(B)** Correla�on matrix for 26 _P. megakarya_ isolates based on marker data. The color bar to the le� of the plot shows the country of origin of each isolate. **Supplementary Figure 5: Popula�on structure of the three final three** _**P. megakarya**_ **subpopula�ons defined in this study. (A, C, E)** Principal component analysis (PCA) plots showing PC1 and PC2 from subpopula�on A (882 SNPs across 23 isolates), subpopula�on B (995 SNPs across 14 isolates), and subpopula�on C (5,309 SNPs across 21 isolates). Each point represents an individual isolate colored by their country of origin. **(B, D, F)** Heatmap of the gene�c rela�onship matrices (GRM) of the subpopula�on A (B), B (D), and C (F). **Supplementary Figure 6: Virulence of P. megakarya isolates from three gene�cally dis�nct** **subpopula�ons across cacao clones, based on disease screening data.** Summary of all lesion size assessments for each isolate across clones. Each bar plot shows the average pod-lesion diameter across all measurements for an isolate. Standard devia�on is depicted through a black line if several observa�ons were made. Each _P. megakarya_ isolate was used to inoculate a separate set of cacao clones. The isolate GH34 was used as a control (n = 209). 37 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) **Supplementary Figure 7: Disease screening of cacao pods from clones of the IC3 cacao collec�on** **inoculated with different** _**P. megakarya**_ **isolates.** The barplots show the overall mean pod lesion diameter for each clone across P. megakarya isolates **(A)** and by isolates from subpopula�ons A **(B)**, B **(C)**, and C **(D)** . Standard devia�on is depicted through a black line if several observa�ons were made. The horizontal dashed line at 15 mm and 30 mm represent the thresholds for calling a clone resistant (0-15 mm), moderately resistant (15-30 mm), or suscep�ble (>30 mm). 38 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) **Figure 1: STRUCTURE analysis showing the gene�c group propor�on rela�ve to the eleven gene�c groups of** _**Theobroma cacao**_ **for all clones from the IC3 cacao collec�on included in the trial.** Ten gene�c groups were described by Motamayor et al. (2008), and the gene�c group ‘Bolivian’ was described by Zhang et al. (2012). The color bar above the plot indicates whether the clone is considered a full member of a gene�c group (≥75% membership; colored according to the legend) or an admixed clone (25-75%, lighter color corresponding to the gene�c group with the highest contribu�on). 39 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) **Figure 2: Popula�on structure and composi�on of 947 clones of the IC3 cacao collec�on. (A)** Principal component analysis (PCA) based on 8,572 SNP markers. Each point represents an individual clone and is colored according to its predominant gene�c-group ancestry based on STRUCTURE analysis. Clones with ≥75% membership in their predominant gene�c group were classified as members and are shown as circles, whereas clones with <75% membership were classified as admixed and are shown as triangles. **(B)** Percentage of gene�c varia�on explained by the first ten principal components of the SNP-based PCA, **(C)** Representa�on of the major cacao gene�c groups within the IC3 collec�on, based on STRUCTURE ancestry assignments. Contribu�ons from clones with >75% membership and admixed clones are indicated by circles and triangles, respec�vely. Gene�c-group nomenclature follows Motamayor et al. (2008). **Figure 3: Venn diagram of the total number of SNPs (A), genes with SNPs (B), and RxLR genes with SNPs (C) unique and shared** **among** _**P. megakarya**_ **isolates from the subpopula�ons A, B, and C.** 40 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) **Figure 4: Overview of the disease screening carried out in this study.** **(A)** Disease screening using the _P. megakarya_ isolate GH34 on different cacao clones. **(B)** Disease screening of the cacao clone POUND-7 inoculated with different _P. megakarya_ isolates. The bar plots are colored by subpopula�on. **(C)** Histograms summarizing the phenotypic data, with the mean pod lesion diameter per isolate separated by subpopula�on. 41 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) **Figure 5: Associa�on between gene�c-group ancestry and the genomic es�mated breeding values (GEBV) of 350 phenotyped** **cacao clones (Model 2).** Points are regression coefficients from a mul�ple regression of GEBV on STRUCTURE ancestry propor�ons, with ‘Amelonado’ as reference; bars are 95% confidence intervals. A coefficient of −10 means that a clone of en�rely that ancestry is predicted to have lesions 10 mm smaller than an ‘Amelonado’ clone. Nega�ve values indicate smaller predicted lesions, that is, higher resistance. ‘Bolivian’ is omi�ed because too few phenotyped clones carry this ancestry. 42 [bioRxiv preprint doi: https://doi.org/10.64898/2026.11.03.764016; this version posted November 5, 2026. The copyright holder for this preprint](https://doi.org/10.64898/2026.11.03.764016) (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made [available under a CC-BY 4.0 International license.](http://creativecommons.org/licenses/by/4.0/) **Table 1: Comparisons of gene�c parameter es�mates of all models tested in this study.** The es�ma�ons of the percentage of phenotypic variance (ra�os on V P = clone + isolate + interac�on + residual, bioassay is fi�ed but excluded from V_P) with standard errors explained by gene�c effects for cacao clones ( 𝐻 [�] �� ), _P. megakarya_ isolates, their interac�on, and the residual error term ( 𝑒 � ), along with es�mates of gene�c correla�ons with standard errors for clones 𝑟 �� and isolates 𝑟 �� are given. Those were obtained from linear models incorpora�ng the factors clone and isolate as either unrelated and independent design effects or as correlated effects through marker-derived rela�onship matrices (GRM) for cacao clones and _P. megakarya_ isolates. A dual-genome similarity matrix (GGSM) was used to model interac�ons. Model 6 fits subpopula�on-specific genomic variance components via a heterogeneous correla�on structure for clones and subpopula�on-specific isolate effects. The model also provides a genomic correla�on between subpopula�on (cor.). The subpopula�ons are A, B, and C. The cacao ra�o is a genomic broad-sense heritability; the isolate and interac�on ra�os are propor�ons of phenotypic variance. Phenotypic variance is defined excluding the bioassay variance. Each rela�onship matrix was scaled by 𝐷= 𝑚𝑒𝑎𝑛[𝑑𝑖𝑎𝑔(𝐾)] −𝑚𝑒𝑎𝑛(𝐾), which was 1.136 for cacao and 0.192, 0.413, and 0.391 for pathogen subpopula�ons A, B, and C. logL, REML log-likelihood; AIC, Akaike informa�on criterion (−2logL + 2k, where k is the number of es�mated variance parameters: 7 for Models 1–5 and 11 for Model 6). Lower AIC indicates be�er fit. # Model structure 𝑯 [�] 𝟐𝑪 isolate isolate clone: e² 𝒓 𝒈𝑪 𝒓 𝒈𝑰 logL AIC 1 C, I, C:I 0.199 ± 0.362 ± 0.016 ± 0.423 ± 0.925 ± 0.957 ± −3768.083 7550.17 0.028 0.055 0.026 0.046 0.111 0.065 2 GRM C, I, GRM C :I 0.267 ± 0.344 ± 0.017 ± 0.373 ± 0.941 ± 0.953 ± −3744.866 7503.73 0.039 0.053 0.014 0.039 0.047 0.038 3 C, GRM I, C:GRM I 0.195 ± 0.378 ± 0.012 ± 0.416 ± 0.944 ± 0.970 ± −3769.399 7552.80 0.028 0.057 0.025 0.047 0.113 0.061 4 GRM C, GRM I, C:I 0.258 ± 0.361 ± bound 0.381 ± 1 0.039 0.056 0.039 (bound) 1 (bound) −3747.313 7508.63 5 GRM C, GRM I, GGSM CI 0.258 ± 0.360 ± 0.008 ± 0.374 ± 0.971 ± 0.979 ± −3747.170 7508.34 0.039 0.056 0.014 0.040 0.053 0.038 n/a n/a −3734.655 7491.31 6 _corh_ (subpop: GRMc), _at_ (subpop):GRM I 43 A 0.298 ± 0.059/ B 0.191 ± 0.071 C 0.328 ± 0.061 A 0.396 ± 0.092 B 0.489 ± 0.124 C 0.262 ± 0.079 n/a A 0.307 ± 0.052 B 0.320 ± 0.086 C 0.410 ± 0.058