Tracing the Evolutionary Dynamics and the Transcriptional Partitioning of Terpene Synthase Gene Family in Coffea arabica and Its Diploid Progenitors Authors: Makarim Elfadil M Osman, Mohanad A Ibrahim, Rieham Sallah H Osman, Budour A Alghamdi, Amina I Dirar, Wafa A Alrasheed, Abdelsattar M Omar, Emadeldin Hassan E Konozy Published: 2026 ABSTRACT AbstractTerpenes represent a diverse class of plant secondary metabolites that play important roles in development, ecological interactions, and aroma, and are synthesized by the terpene synthase (TPS) gene family. To investigate the evolutionary history and regulatory diversification of this family in coffee, we performed a genome-wide analysis of TPS genes in Coffea arabica and its diploid progenitors, C. canephora and C. eugenioides. Using domain-based searches targeting the PF01397 module, we identified 69 TPS genes in C. arabica, 40 in C. canephora, and 55 in C. eugenioides. Phylogenetic classification, duplication analysis, and synteny mapping revealed that the tandem duplication is the major driver of TPS expansion, with C. arabica retaining progenitor-derived copies alongside limited lineage-specific duplications. Selection analyses indicated that most TPS duplicates from C. arabica are under strong purifying constraint, while only a single orthogroup (OG0000013) showed limited evidence of episodic positive selection with a minor fraction of inferred sites and a lack of strong BEB support. Indicating that adaptive divergence was highly localized against a predominantly conserved evolutionary background. To link evolutionary divergence with transcriptional regulation, RNA-seq data from multiple tissues and developmental stages were analyzed. A substantial proportion of TPS genes were transcriptionally active and exhibited tissue- and stage-specific regulation. While many duplicates showed conserved or asymmetric expression, the OG0000013 duplicates combined localized adaptive sequence divergence with significant upregulation during late seed development. In conclusion, the retention and asymmetric transcription of progenitor-derived homeologs indicate that regulatory partitioning, rather than extensive gene-family innovation, has the dominant evolutionary outcome following allopolyploidization in C. arabica.Keywords: Coffea sp., terpene synthase, duplication, selection, orthologs, expression 1. Introduction Terpenes are one of the largest and most structurally diverse classes of plant secondary metabolites, encompassing over 55,000 known compounds with functions that range from plant growth and development to defense and ecological interactions. 1 In the genus Coffea, terpenes are of particular significance because they not only contribute to plant defense mechanisms against herbivores and pathogens but also play critical roles in defining the aroma and flavor profiles of coffee beans, which are traits of immense economic importance.2,3 The biosynthesis of terpenes is catalyzed by the terpene synthase (TPS) gene family, a lineage of enzymes that catalyze the cyclization and rearrangement of prenyl-diphosphates into a vast array of mono-, sesqui-, and diterpenes.1,4 Phylogenetically, plant TPS genes are classified into multiple subfamilies (TPS-a to TPS-h) based on sequence similarity, conserved catalytic motifs, and functional characteristics. In angiosperms, most TPS genes are distributed among the TPS-a, TPS-b, TPS-c, TPS-e/f, and TPS-g clades, reflecting lineage-specific expansion and functional diversification associated with specialized metabolism and ecological adaptation. 5 The TPS gene family is evolutionarily dynamic, often expanding through gene duplication events followed by sub-functionalization or neofunctionalization. Such duplications provide a reservoir of genetic diversity that can be selectively maintained or lost depending on ecological pressures.5,6 In plants, tandem duplication has been identified as a predominant driver of TPS family expansion, resulting in clusters of functionally related genes that may undergo lineage-specific diversification. 6 Comparative genomics has shown that while many TPS genes remain under purifying selection, some undergo episodes of positive selection that promote adaptive innovation. 7 Coffea arabica, the most widely cultivated coffee species, is an allotetraploid formed through hybridization between Coffea canephora and Coffea eugenioides approximately 0.5–1 million years ago. 8 As a polyploid, C. arabica provides a unique framework for investigating gene family evolution, as it harbors two subgenomes with distinct evolutionary legacies. 9 Polyploidization often results in both the retention of ancestral duplicates and the generation of novel lineage-specific paralogs, which may contribute to species-specific traits such as stress adaptation or metabolite production. 10 Understanding how TPS genes have expanded, diverged, and been differentially regulated in C. arabica compared to its progenitors can therefore illuminate the evolutionary forces shaping coffee’s secondary metabolism. 2 Recent studies have highlighted the importance of terpene synthases in the development and ripening of fruits. For instance, expression analyses in tomato and grape have shown distinct temporal regulation of TPS genes during fruit maturation, with defensive terpenes being predominant in early developmental stages and aroma-related terpenes enriched during ripening.11-13 In coffee, RNA-seq studies similarly demonstrate dynamic transcriptional changes during bean development, with TPS genes contributing to the biosynthesis of flavor-determining volatiles. 14 Salojärvi and his colleagues (2024) had provided a high-quality chromosome-scale genome assembly of Coffea arabica and examined selected metabolic gene families (i.e., terpene synthases (TPS), N-methyltransferases (NMT), and fatty acid desaturase 2 (FAD2)). Their analyses were primarily framed within genome structure, subgenome evolution, and population genomics. 9 However, in this study, we aimed to investigate TPS genes, focusing on gene copy number, subgenome origin, and broad evolutionary retention patterns within the context of polyploid formation. Our goal is to deliver a targeted, in-depth functional and evolutionary analysis of the TPS family, integrating orthogroup-based phylogenetics, codon-level selection analyses for duplicates (PAML and Datamonkey), and statistically supported multi-tissue transcriptomic profiling across seeds, fruit developmental stages, and other vegetative organs. Particular emphasis was placed on evaluating whether TPS evolution in C. arabica reflects balanced homeolog retention, asymmetric transcriptional partitioning, or lineage-specific divergence following hybrid genome formation. 2. Methodology 2.1. Identification of TPS Gene Family The Arabidopsis thaliana canonical terpene synthase (TPS) sequence model (ID: A0A178U9Y5) containing domains (PF01397) and (PF03936), was retrieved from the InterPro database (https://www.ebi.ac.uk/interpro/), and used to identify members of the terpene synthase family in three coffee species. Initially, the proteome and CDS FASTA files of Coffea arabica (ET-39 Hifi; GCA_036785885.1), 15 C. canephora (AUK_PRJEB4211_V1; GCA_900059795.1), 16 and C. eugenioides (Ceug_1.0; GCA_003713205.1) were downloaded and queried against the model sequence using TBtools II 17 local BLASTP 2.12.0+ (E-value = 1×10-10). This stringent threshold was selected to prioritize high-confidence identification of TPS candidates while minimizing the inclusion of distantly related terpene-associated proteins and partial low-similarity matches commonly encountered in large plant genomes. Candidate proteins were subsequently evaluated and annotated using InterProScan 5, 18 which integrates multiple conserved-domain recognition models, including Pfam-based signatures. Only proteins containing both canonical TPS domains (PF01397 and PF03936) were retained for downstream analyses, as the study specifically focused on structurally complete canonical TPS proteins. Putative truncated or atypical TPS-like proteins lacking one domain were excluded because such sequences may represent incomplete gene models, pseudogenes, or functionally divergent terpene-associated proteins that could compromise orthology and evolutionary inference. Sequence IDs were used to extract gene coordinate information (i.e., chromosome location, start, and end positions) from the NCBI database (https://www.ncbi.nlm.nih.gov/). These coordinates, along with the scale length information from NCBI, were then used to construct chromosomal maps for each species. 2.2. Characteristics of the TPS Putative Sequences The identified putative canonical TPS sequences were analyzed for signal peptides and cleavage sites using SignalP v5.0 (https://www.cbs.dtu.dk/services/SignalP/). 19 Additionally, TMHMM v2.0 (https://www.cbs.dtu.dk/services/TMHMM/) was employed to detect N-terminal transmembrane helices. 20 To classify signal peptides and predict unconventional secretion pathways, Outcyte v1.0 (https://www.outcyte.com/analyse/) was used to distinguish between intracellular (IC) and unconventionally secreted proteins (UPS). 21 Subcellular localization of the putative lectins was inferred using Wolf PSORT (https://wolfpsort.hgc.jp/). 22 Finally, the molecular weight and the isoelectric points (pI) were calculated using Expasy Compute pI/MW (https://web.expasy.org/compute_pi/). 23 All tools were used under default parameters. 2.3. Classification of TPS Genes To classify TPS genes into established TPS subfamilies, a phylogenetic framework based on full-length protein sequences was constructed using the FASTA format of predicted TPS amino acid sequences from Coffea arabica and its diploid progenitors, C. canephora and C. eugenioides. Reference orthologs representing major angiosperm TPS subfamilies were obtained from A. thaliana (TAIR10): AT3G14540.1 (TPS_a), AT4G16730.1 (TPS_b), AT4G02780.1 (TPS_c), AT1G61120.1 (TPS_e/f), and AT1G61680.1 (TPS_g). These reference sequences were incorporated to anchor subfamily classification. 5 Multiple sequence alignment was performed using MUSCLE with default parameters. A maximum likelihood (ML) phylogenetic tree was reconstructed using IQ-TREE under the JTT amino acid substitution model, as it is a commonly used empirical model for protein phylogenetics. Branch support was assessed with 1,000 bootstrap replicates to evaluate clade stability. TPS subfamily assignments for Coffea sequences were determined based on their clustering with the corresponding A. thaliana reference proteins in well-supported clades (bootstrap support ≥70%). 2.4. Duplication of the TPS Within Each Progenitor Species Gene duplication events were analyzed using protein sequences and corresponding GFF files, which contained gene locations, identifiers, and positional data to investigate the expansion of the terpene synthase gene family in C. canephora and C. eugenioides. Homologous pairs were detected via BLASTp (E-value ≤ 1 × 10-10, top 5 hits retained). The MCScanX package, 24 implemented in TBTools-II, was then used to classify duplication events, leveraging its MCScanX wrapper and duplicate_gene_classifier module. From this analysis, tandem duplicates were directly extracted. Further classification was performed to distinguish between proximal and dispersed (transposon-mediated) duplications. Proximal duplicates were defined as homologous pairs located on the same chromosome within a short genomic distance. In contrast, dispersed duplicates were identified based on sequence similarity but distant genomic positions, either on different chromosomes or separated by ≥1 Mbp on the same chromosome. MCScanX-derived duplication categories in the diploid progenitor genomes should be interpreted as operational classifications of duplication context rather than definitive estimates of historical duplication frequency. 2.5. Orthogroup Inference and Gene Tree Reconciliation To study the duplication of TPS from C. arabica, orthogroups were generated and inferred using OrthoFinder (Galaxy server: https://usegalaxy.eu/) 25 with DIAMOND as the sequence similarity search tool. Multiple sequence alignments were generated using MUSCLE, and gene trees were constructed with IQ-TREE under default parameters. The resulting gene trees for each orthogroup were used to identify tandemly duplicated C. arabica TPS genes. Gene tree reconciliation was performed using Notung (v2.9.1) 26 to infer gene duplication and loss events. Rooted orthogroups gene trees were reconciled against the Coffea sp. phylogeny implemented as the reference species tree, using the default duplication-loss parsimony costs model implemented in Notung to ensure standardized inference of duplication and loss events across orthogroups. Rearrangement options were enabled to improve poorly supported branches (support threshold = 80%). 2.6. Collinearity Analysis To investigate genomic collinearity of TPS genes, dual synteny analysis was performed among Coffea arabica, C. canephora, and C. eugenioides. Gene coordinates and annotations were obtained from the respective genome assemblies. Pairwise syntenic relationships were identified using TBTools-II – MCScanX with default parameters. 24 TPS gene locations were mapped onto the syntenic blocks. 2.7. Estimation of the Substitution Rate and the Approximate Divergence Time Synonymous substitution rates (Ks) for terpene synthase genes and their paralogs were calculated using protein and coding sequence (CDS) FASTA files, along with paired gene lists, processed with the Ka/Ks calculator in TBtools II. Selection pressure was inferred from Ka/Ks ratios: purifying selection (Ka/Ks < 1), positive selection (Ka/Ks > 1), or neutral evolution (Ka/Ks = 1). Approximate divergence times (T) were estimated using the equation T = Ks/2r, where Ks is the synonymous substitution rate, and r is the species-specific synonymous substitution rate per site per year (r = 65 x 10-9, which has been widely applied in plant comparative genomics). The factor of 2 accounts for independent mutation accumulation in both lineages post-divergence. For the allotetraploid C. arabica, divergence times reflect the hybridization event between its diploid progenitors (C. canephora and C. eugenioides), estimated at ∼0.5–1.1 million years ago (MYA). 8 Gene pairs with Ks values corresponding to this period were classified as originating from this ancestral divergence. 2.8. Selection Analysis Codon-based selection analysis was conducted using both PAMLX 27 and the Datamonkey web server 28 on duplicates identified from C. arabica. For each gene pair, codon-based alignments were generated using PAL2NAL. 29 Two-gene phylogenetic trees were generated in Newick format. In PAMLX, three categories of models were tested to detect positive selection: the M0 one-ratio model, site models (M1a vs. M2a and M7 vs. M8), branch models to evaluate differences in selective pressure (ω) between duplicate lineages, and branch-site models (test 2) to detect episodic selection acting on specific codons along designated foreground branches. Likelihood ratio tests (LRTs) were applied to compare nested models and identify signatures of positive selection, where ΔLRT = 2 × (lnL1 − lnL0), and significance was evaluated using the chi-square distribution based on the difference in free parameters between compared models. Codons under positive selection were identified using Empirical Bayes (BEB) with a posterior probability threshold ≥ 0.95. Adaptive evolutionary testing was performed using the Datamonkey Adaptive Evolution Server under default settings. The following models were used: MEME (Mixed Effects Model of Evolution) to detect episodic selection at individual sites, FEL (Fixed Effects Likelihood) and FUBAR (Fast, Unconstrained Bayesian AppRoximation) for pervasive selection, RELAX to test for changes in selection intensity, and aBSREL (adaptive Branch-Site REL) to identify branches under episodic diversifying selection. Because selection analyses were restricted to three predefined orthogroups containing C. arabica-specific duplications rather than genome-wide screening, formal multiple-testing correction was not applied. Consequently, all positive-selection signals were interpreted conservatively and evaluated in conjunction with likelihood support, branch length, codon-level support, and complementary Datamonkey analyses. 2.9. Expression Analysis A search of the European Nucleotide Archive (ENA: https://www.ebi.ac.uk/ena/browser/home) and NCBI Sequence Read Archive (SRA) was conducted to identify publicly available Coffea arabica transcriptomic datasets representing multiple tissues and developmental stages. Studies were screened based on strict inclusion criteria: (a) data generated using high-throughput RNA sequencing with paired-end libraries, (b) experiments conducted with a minimum of three biological replicates per tissue/stage, and (c) availability of complete and well-annotated metadata and published studies. From the available resources, four datasets that encompass major vegetative tissues of C. arabica across key developmental transitions were selected for downstream analyses. These included seed developmental stages (green, yellow, and red beans) from PRJEB24137, 14 root transcriptomes from PRJEB52366, 30 leaf transcriptomes from PRJNA851465, 31 and fruit pericarp transcriptomes representing purple, red, and yellow developmental stages from PRJNA743796. 32 To minimize technical variability among independently generated RNA-seq datasets, all transcriptomic samples were processed using unified preprocessing, alignment, and quantification workflow. Moreover, differential expression analyses were conducted within biologically matched datasets rather than through direct global integration across studies, thereby reducing the influence of potential batch effects associated with independent experimental designs. Raw paired-end FASTQ files (R1 and R2) were subjected to initial quality assessment to identify incomplete, corrupted, or improperly paired libraries; all retained samples passed quality control checks. Quality filtering and preprocessing were performed using fastp (v 0.23.4), which included adapter removal, base quality correction where necessary, and filtering of low-quality reads. Post-filtering quality metrics were evaluated to ensure uniform sequencing quality across all datasets. No dataset-specific trimming biases were introduced, allowing direct comparability among tissues and developmental stages. High-quality reads were aligned to the C. arabica reference genome (ET-39 HiFi assembly; NCBI RefSeq accession GCA_036785885.1) using STAR (v2.7.11b) with genome indices built from the corresponding reference FASTA file. Gene-level read quantification was performed using featureCounts (Subread v2.0.6) with a standardized C. arabica GTF/GFF annotation file. Differential expression analysis was conducted in R using the DESeq2 package. Statistical outputs included log2 fold changes (Log2FC), Wald test P-values, and Benjamini–Hochberg adjusted P-values. Genes with an adjusted P-value (FDR) ≤ 0.05 were considered significantly differentially expressed. 2.10. Reporting Guideline and Methodological Transparency No dedicated EQUATOR Network guideline is currently available for comparative genome-wide identification of gene families’ studies. The manuscript was prepared according to transparent bioinformatics reporting practices. The STROBE v4 framework 33 was adapted as the closest general reporting reference to enhance methodological transparency, and the completed adapted checklist is provided as Supplemental File S1. 3. Results 3.1. Overview of the TPS Gene Family in Coffea sp Genomic assemblies of three Coffea species, Coffea arabica (ET-39 HiFi), C. canephora (Ceug_1.0), and C. eugenioides (AUK_PRJEB4211_v1), were analyzed to identify TPS genes using the canonical protein that contains the monoterpene synthase N-terminal domain (PF01397) and terpene synthase metal-binding (PF03936) as a reference. In total, 69, 40, and 55 TPS genes were detected in C. arabica, C. canephora, and C. eugenioides, respectively (Table 1, Supplemental File S2). Their chromosomal distribution is heterogeneous across the haploid set of 11 chromosomes in C. canephora and C. eugenioides, and the haploid allotetraploid set of 22 chromosomes in C. arabica. TPS clusters were observed on chromosomes 1, 2, 4, 5, 6, 7, 8, and 10 (Figure 1A). In C. arabica, chromosomes 2c, 8e, and 8c harbor the largest proportion of TPS genes (46.4%), whereas in C. eugenioides, chromosome 8 alone contains 22 TPS genes, accounting for 40% of the total (Figure 1A). None of the identified TPS genes contained a signal peptide, and except for one or two genes per species, all lacked a transmembrane helix domain. Subcellular localization analysis indicated that TPS proteins are predominantly intracellular, with either chloroplast or nucleocytoplasmic targeting, and in some cases, unconventional secretion pathways were predicted (Figure 1B). All C. arabica genes maintained the canonical domain architecture with no other domains fused to their (PF01397) and (PF03936). Table 1.Main Characteristics of TPS From Three Coffea spPredicted characteristicsTerpene synthase gene C. arabica C. canephora C. eugenioides Predicted genes (No) 694055 Protomer length 279 – 845364 – 822284 – 1180 Chromosomal location (No/Chr) 2(1e), 9(2c), 8(2e), 3(4c), 6(4e), 6(5c), 4(5e), 2(7c), 1(7e), 9(8c), 14(8e), 3(10c), 2(10e)1(1c), 3(2c), 9(4c), 1(5c), 5(6c), 5(7c), 6(10c), 3(Scaffold_292), 1(Scaffold_443), 1(Scaffold_774), 2(Scaffold_1014), 1(Scaffold_1535), 1(Scaffold_4202), 1(Scaffold_1092)2(1e), 10(2e), 7(4e), 7(5e), 1(7e), 22(8e), 2(10e), 1(NW_020862301.1), 1(NW_020864231.1), 1(NW_020864415.1), 1(NW_020864479.1) Other fused domains ⸺PF19086PF19086, PF07859 Signal peptide No (100%)No (100%)No (100%) Transmembrane Single site (1.4%)2 sites (5.0%)Single site (1.8%) Protein targeting * IC(49, 71.01%), UPS(20, 28.9%)IC(28, 70.0%), UPS(12, 30.0%)IC(38, 69.1%), UPS(17, 30.9%) Subcellular localization Chloroplast, Cytoplasm, NucleusChloroplast, Cytoplasm, NucleusChloroplast, Cytoplasm, Nucleus Protomer M.Wt. 32.5 – 97.2 KDa42.3 – 92.8 KDa32.4 – 1360.8 KDa Isoelectric point (pI) 4.91 – 6.174.92 – 6.644.93 – 8.47*Prediction of targeted proteins: IC: intracellula; UPS: Unconventional Protein Secretion. Main Characteristics of TPS From Three Coffea sp *Prediction of targeted proteins: IC: intracellula; UPS: Unconventional Protein Secretion. Distribution of TPS genes and their subcellular targeting. (A) The chromosomal map of TPS from three Coffea sp. Only chromosomes that harbor TPS genes are displayed. (B) Heatmap illustration of the possible subcellular compartment targeting of TPS 3.2. Classification of TPS Maximum-likelihood phylogenetic reconstruction using full-length TPS proteins and Arabidopsis thaliana references (TPS-a: AT3G14540.1; TPS-b: AT4G16730.1; TPS-c: AT4G02780.1; TPS-e/f: AT1G61120.1; TPS-g: AT1G61680.1) resolved five well-supported clades corresponding to canonical angiosperm TPS subfamilies (Figure 2). All Coffea sp. sequences clustered unambiguously with their respective reference lineages, with strong bootstrap support (predominantly ≥95), confirming robust subgroup assignment and absence of orphan or lineage-ambiguous clades. Quantitatively, TPS-a represents the predominant subfamily across all three species, accounting for 46.4% (32) in C. arabica, 42.5% (17) in C. canephora, and 58.2% (32) in C. eugenioides. This marked expansion, particularly in C. eugenioides, indicates lineage-specific amplification of sesquiterpene synthases and suggests adaptive diversification in specialized metabolism. TPS-b constitutes the second largest group, with relatively balanced representation (18–30%), consistent with conserved monoterpene synthase functions across Coffea lineages. TPS-c, TPS-e/f, and TPS-g are comparatively reduced, together forming a minor proportion of the repertoire. TPS-e/f and TPS-g show especially limited copy numbers (≤7.5% per species), consistent with their conserved roles in primary metabolism (e.g., ent-kaurene and diterpene biosynthesis) and stronger functional constraint. The low variability and shorter branch lengths observed in these clades further support purifying selection acting on ancestral, functionally conserved TPS lineages. Maximum-likelihood phylogenetic tree of TPS proteins from Coffea arabica, C. canephora, and C. eugenioides inferred from full-length amino acid sequences with 1,000 bootstrap replicates. Arabidopsis thaliana TPS orthologs (TPS_a: AT3G14540.1; TPS_b: AT4G16730.1; TPS_c: AT4G02780.1; TPS_e/f: AT1G61120.1; TPS_g: AT1G61680.1) were included to define canonical angiosperm TPS subfamilies 3.3. Duplication and Divergence Analysis in C. canephora and C. eugenioides TPS Genes To investigate the origin and evolutionary trajectory of the terpene synthase gene family in Coffea canephora and C. eugenioides, we analyzed gene duplication events using MCScanX (Figure 3). In C. canephora, a total of 40 duplicated gene pairs were identified, with the majority classified as tandem duplications (26 pairs, 65%), for example, CDP05817.1–CDP05818.1 and CDP11840.1–CDP11842.1, indicating that tandem expansion is the dominant duplication mechanism. Other duplication types were also detected, including proximal (4 pairs, 10%; e.g., CDP15198.1–CDO97389.1) and dispersed/transposed duplications (10 pairs, 25%; e.g., CDP03636.1–CDP07894.1). Similarly, in C. eugenioides, 55 duplicated gene pairs were detected. Tandem duplication was again the most prevalent mode (37 pairs, 67%; e.g., XP_027172055.1–XP_027172056.1), followed by proximal (11 pairs, 20%) and dispersed duplications (7 pairs, 13%). Notably, there was no evidence of whole-genome duplication (WGD) contributing to the expansion of the TPS gene family in either species. Distribution and duplication patterns of TPS genes in C. canephora and C. eugenioides. (A) Circos plots present the types of duplication of TPS in each species. (B) Proportions of duplicated gene pairs categorized by expansion type, selection pressure (Ka/Ks), and approximate divergence time inferred from Ks values Most pairs exhibited Ka/Ks ratios indicative of purifying selection, reflecting functional constraint following duplication. In C. canephora, 29 (72.5%) had Ka/Ks ratios < 0.5, and another 6 (15%) fell between 0.5 and 1.0, suggesting that over 87.5% of duplicates are evolving under strong to a more relaxed selective pressure. Only 5 pairs (13.2%) lacked calculable values due to low divergence or alignment issues, and none showed Ka/Ks > 1.0. Similarly, in C. eugenioides, out of 55 duplicated pairs, 33 (60%) had Ka/Ks < 0.5. Whereas 13 (23.6%) were in the intermediate range (0.5 < Ka/Ks < 1), and like C. canephora, no pairs were under positive selection. To contextualize the approximate evolutionary timing of these duplications, divergence times were estimated using the molecular clock formula T = Ks/(2r), with a standard plant substitution rate (r) of 6.5 × 10-9 substitutions/site/year. Taking into account that the substation rates may vary among lineages, gene families, and polyploid genomes, these estimates should be considered approximate rather than absolute evolutionary dates. In C. canephora, 14 pairs (35%) displayed Ks values corresponding to divergence between ∼2.5 and 60 million years ago (MYA), placing them plausibly within the timeframe of Rubiaceae radiation or early Coffea sp. diversification. Another 7 pairs (17.5%) had Ks estimates corresponding to divergence less than 2.5 MYA, overlapping with the estimated divergence between C. canephora and C. eugenioides (1.1–2.5 MYA), suggesting that these duplications may have occurred near or after this speciation event. However, a significant proportion (42.5%) of C. canephora gene pairs showed Ks values consistent with divergence time estimated to predate ∼60 MYA, although temporal resolution at high Ks values may be affected by synonymous-site saturation. This is likely reflecting either ancient duplications retained since early angiosperm evolution or saturation effects that obscure accurate timing. Similar trends were observed in C. eugenioides, where 29 of the 55 duplicated pairs (52.7%) fell in the 2.5–60 MYA range, 5 pairs (9.1%) below 2.5 MYA, and 15 pairs (27.3%) showed Ks that approximately predate ∼60 MYA (Supplemental File S3). 3.4. Evolutionary Dynamics and Duplication Patterns of TPS Genes in C. arabica The terpene synthase-related gene family across three Coffea sp. revealed extensive orthology and duplication relationships, especially within C. arabica (Figure 4). From a total of 164 protein-coding genes analyzed, 148 (90.2%) were assigned to 28 orthogroups, while only 16 (9.8%) remained unassigned. Notably, C. arabica had the highest gene count (69), with 59 (85.5%) placed in orthogroups and 10 (14.5%) unassigned. C. arabica did not contribute to any species-specific orthogroups, consistent with the overall result that no species-specific orthogroups were identified across the dataset. Of the 28 total orthogroups, 17 (60.7%) contained representatives from all three species, indicating a strong level of evolutionary conservation within this gene family. In contrast, only three orthogroups were strict single-copy groups, highlighting limited single-copy retention and suggesting ongoing duplication and divergence dynamics in Coffea sp. genomes. Across all orthogroups, the mean orthogroup size was 5.3 genes, with a median of 4.0. Gene 50 (G50 = 6, for both assigned and all genes), meaning half of all genes in orthogroups are found in groups of size ≥ 6. Orthogroup 50 (O50 values were 7 (assigned) and 8 (all genes), indicating that half of all assigned genes are contained within the seven largest orthogroups. Together, these metrics reflect a moderately balanced distribution of orthogroup sizes, showing some gene family expansion, but without dominance by any single large orthogroup (Figure 4A and B). Comparative orthogroup analysis and evolutionary dynamics of TPS genes in three Coffea species. (A) Distribution of TPS genes across species-specific and shared orthogroups. (B) Summary of duplication events inferred by OrthoFinder, showing duplications mapped to the Coffea species nodes (C. arabica, C. canephora, C. eugenioides), and to the ancestral node (N0), as well as the most recent common ancestor node (N1). Black bars represent all detected duplications, while white bars indicate duplications supported by ≥50% of gene trees. (C) Distribution of duplications across orthogroups. “Duplications (all)” includes all inferred duplication events, while “Duplications (50% support)” are events consistently recovered in at least half of the reconciled trees. “Terminal” events occurred at the tips of the tree (species-specific), while “non-terminal” events occurred deeper in the lineage (shared/ancestral). (D) Reconciled gene trees of the OG0000003, OG0000010, and OG0000013 orthogroups. (E) Duplication of events identified from all the orthogroups. (F) Dual synteny plot illustrating conserved and lineage-specific TPS loci among C. eugenioides, C. arabica, and C. canephora Focusing on Coffea arabica, reconciliation of the orthogroup trees (OG0000000 to OG0000015) revealed that only a limited subset of orthogroups contains true species-specific duplications attributable to recent paralog formation in the C. arabica genome. Among all orthogroups assessed, three groups (OG0000003, OG0000010, and OG0000013) show clear terminal C. arabica duplication events, as supported by both the topology of the gene trees and the duplication nodes inferred through Notung reconciliation (Figure 4; Supplemental File S4). Each of these duplications forms tight sister pairs of C. arabica genes, characterized by near-zero branch lengths signatures consistent with recent, post-speciation duplication rather than inheritance from the ancestral C. canephora and C. eugenioides hybridization. In OG0000003, two C. arabica genes (XP_027093045.1 and XP_027093111.1) form a terminal sister clade with negligible branch divergence, confirming a recent duplication specific to C. arabica, and in OG0000010, another C. arabica pair (XP_027107080.1 and XP_071931897.1) also appears as a well-supported terminal duplication. Similarly, OG0000013 contains a duplicated C. arabica pair (XP_027102632.2 and XP_071930373.1), forming a distinct monophyletic unit, which is again supported by reconciliation as a species-specific duplication. By contrast, the diploid progenitors also carry multiple authentic terminal duplications, for example, C. eugenioides (XP_027182177.1 – XP_027156962.1 in OG0000001), and C. canephora (CDP18843.1 – CDP18846.1 in OG0000002 and CDO99694.1 – CDO99695.1 in OG0000013), showing that small-scale, lineage-specific duplication is a feature of all three lineages rather than being uniquely elevated in C. arabica. The Notung reconciliations also identify several gene losses and missing-homeolog events, marked as *LOST. For example, in orthogroup OG0000000, the reconciliation indicates that expected homeologous copies were not retained in some lineages, with nodes annotated as Coffea arabica*LOST and Coffea eugenioides*LOST. These inferred losses suggest that, despite the allopolyploid origin of C. arabica, not all duplicated homeologs were maintained following hybridization. Nevertheless, C. arabica still contains the highest number of TPS genes (69) among the three species because it preserves representatives from both ancestral subgenomes. In contrast, although C. canephora and C. eugenioides show lineage-specific duplications, their total TPS gene counts remain lower, reflecting both fewer ancestral copies and the loss of some expected homeologs revealed by the reconciliation analysis. It is also worth mentioning that in several reconciled TPS gene trees, Notung inferred “transfer” events where the gene-tree topology could not be fully explained by speciation and duplication alone. These signals do not represent true horizontal gene transfer; rather, they arise from conflict between the gene tree and species tree caused by ancestral duplications, differential gene loss, and the mixed subgenome origin of C. arabica, which together produce placements that the algorithm resolves as transfer-like events. Across the remaining orthogroups, C. arabica generally occurs as single copies or as members of mixed (nonterminal) clades with C. canephora or C. eugenioides, indicating no evidence of C. arabica-specific expansion. This contrasts with the broad duplication calls produced by MCScanX, which classified most C. arabica TPS genes as duplicates. Such inflation likely reflects MCScanX’s sensitivity to genomic neighborhood rather than true phylogenetic duplication. In contrast, the reconciliation approach provides a clearer resolution of paralogy by distinguishing lineage-specific duplications from ancestral copy inheritance, and therefore, more accurately captures the modest but distinct expansion pattern in C. arabica confined to OG0000003, OG0000010, and OG0000013. The TPS gene pairs in C. arabica are also supported by high sequence identity. In OG0000010, (XP_027093111.1–XP_027093045.1) form 100%-identical tandem duplicates on chromosome 10c, indicating a very recent duplication event with no detectable divergence (Ka = Ks = 0). Meanwhile, in OG0000003, the duplicated pair (XP_071931897.1–XP_027107080.1) on chromosome 2c also shows 100% identity, but is classified as dispersed/transposed, suggesting relocation following duplication. The XP_027102632.2–XP_071930373.1 duplicate pair (identity 97.79%) from OG0000013 represents a slightly older tandem duplication that occurred around 0.79 MYA with measurable but low divergence (Ka ≈ 0.0037; Ks ≈ 0.0052), consistent with purifying selection acting to maintain protein function (Supplemental File S2). 3.5. Lineage-Specific and Orthogroup-Dependent Selection Patterns in C. arabica To assess whether C. arabica–specific TPS gene duplicates experienced shifts in evolutionary pressure following duplication, selection analyses were restricted to orthogroups exhibiting species-specific duplications in C. arabica. This restriction is methodologically justified because branch and branch-site codon models implemented in PAML/PAMLX require an explicitly defined foreground lineage in which functional divergence is hypothesized to occur. Orthogroups lacking C. arabica–specific duplications do not contain an appropriate foreground branch; consequently, tests for altered ω (dN/dS) ratios in such cases would be biologically unmotivated and statistically uninformative. Codon-based maximum likelihood analyses were conducted to evaluate selective pressures acting on duplicated TPS genes retained in the allotetraploid genome of Coffea arabica (Table 2). In orthogroups OG0000003 and OG0000010, C. arabica paralogs formed very short terminal branches (≈10-6 substitutions per site) and clustered tightly with orthologs from C. canephora and C. eugenioides, consistent with recent retention following polyploidization. For both orthogroups, site-model comparisons (M1a vs M2a; M7 vs M8) were not significant, and branch and branch-site models showed no evidence of lineage-specific acceleration, with ω values remaining well below or close to 1. These results indicate strong purifying selection and functional conservation of TPS duplicates with minimal post-duplication divergence. Table 2.Summary of Codon-Based Maximum Likelihood Analyses of TPS Gene Duplicates in C. arabica​OrthogroupsOG0000003OG0000010OG0000013 † ​ C. arabica duplicates XP_027093045.1/XP_027093111.1XP_071931897.1/XP_027107080.1XP_071930373.1/XP_027102632.2​ C. arabica branch length ∼1×10-6∼1×10-60.0035–0.0058 Site model Site ΔlnL (M1a–M2a) 0.109.08–10.42 Significance NSNS** Site ΔlnL (M7–M8) 0.4609.33–9.36 Significance NSNS** ω (M8) 1.22130.54 Branch model Branch ω (bg/fg)● 0.38/0.380.58/0.581.0–1.1/1.5–1.6 Branch ΔlnL 0012.66–20.59 Significance NSNS** Branch site model Branch-site ΔlnL 002.0 e-06 Positively selected codons NDNDMinor inferred site classes under M8, no BEB-supported codons Significance NSNS*​ Evolutionary context ConservationConservationAdaptive divergence†For OG0000013, branch and branch-site models were run twice by alternately assigning each C. arabica paralog as the foreground (#1); values are concordant across both tests. Significance at P0.05 = *, P0.01 = **, and P0.0001 = ***. ● bg = background gene, fg = foreground gene. NS = not significant, ND = None detected. Summary of Codon-Based Maximum Likelihood Analyses of TPS Gene Duplicates in C. arabica †For OG0000013, branch and branch-site models were run twice by alternately assigning each C. arabica paralog as the foreground (#1); values are concordant across both tests. Significance at P0.05 = *, P0.01 = **, and P0.0001 = ***. ● bg = background gene, fg = foreground gene. NS = not significant, ND = None detected. In contrast, orthogroup OG0000013 exhibited a distinct evolutionary trajectory. The C. arabica paralogs XP_071930373.1 and XP_027102632.2 formed a well-supported sister pair with measurable terminal branch lengths (0.0035–0.0058), comparable to those observed for C. canephora duplicates, indicating an older duplication event and sufficient evolutionary time for sequence divergence. The branch and branch-site models were applied twice by alternately designating each C. arabica paralog as the foreground branch (#1). In both analyses, site-model comparisons were highly significant (ΔlnL = 9.08–10.42; p < 0.01), and the M8 model inferred a small class of codons evolving under strong positive selection (ω ≈ 30.5). Branch-model results further demonstrated significantly elevated nonsynonymous to synonymous substitution ratios along the foreground C. arabica branches (ωfg = 1.47–1.64) relative to the background, indicating accelerated protein evolution. Notably, comparison of the two foreground-specific tests revealed asymmetric selective pressures between the paralogs: when XP_071930373.1 was specified as the foreground, the estimated ω was higher (ωfg = 1.64) and the likelihood ratio statistic was larger (ΔlnL = 20.59; p < 0.001) than when XP_027102632.2 was treated as the foreground (ωfg = 1.47; ΔlnL = 12.66; p < 0.01), indicating a more pronounced accumulation of nonsynonymous substitutions along the XP_071930373.1 lineage. Branch-site models were also significant under both foreground assignments, with a substantial proportion of sites assigned to positively selected classes (2a/2b), despite the absence of codons exceeding Bayes empirical Bayes posterior probability thresholds. Nonetheless, evidence for adaptive divergence remained highly restricted relative to the broader TPS repertoire. Furthermore, Datamonkey analyses (Adaptive Evolution Server; default settings) using MEME, FEL, and FUBAR to test site-level selection, RELAX to assess shifts in selection intensity, and aBSREL to detect branch-specific episodic selection corroborated the PAMLX results by revealing no detectable selection in orthogroups with near-zero branch lengths for C. arabica duplicated (OG0000003 and OG0000010), while identifying a conserved purifying site in OG0000013 (Codon 306; α=β =47.417, LTR = 4.383; P0.05 = 0.0363). This is consistent with episodic adaptive evolution acting against a background of strong functional constraint. 3.6. Expression Dynamics of TPS Genes in Coffea arabica Out of the 69 identified canonical TPS genes from C. arabica genome, a substantial proportion was transcriptionally active across the analyzed data. The number of expressed TPS genes varied by tissue and developmental stage, ranging from 22 (31.9%) genes in leaves and yellow fruits to 33 (47.8%) genes in yellow-stage seeds, indicating pronounced tissue- and stage-specific regulation of terpene biosynthesis. Roots and seeds exhibited the highest TPS gene representation, with 29 (42%) and 33 (47.8%) expressed genes, respectively, whereas leaves and fruits consistently showed lower proportions. Among fruit developmental stages, red fruits displayed slightly higher TPS expression (26 genes; 37.7%) compared to purple (25 genes; 36.2%) and yellow fruits (22 genes; 31.9%), suggesting dynamic modulation of TPS activity during fruit maturation. Differential expression analysis further revealed clear regulatory divergence of TPS genes between tissues and developmental stages (Figure 5A). The largest number of differentially expressed TPS genes (DEGs) was observed between leaves and roots (17 DEGs), with a predominance of upregulated genes in leaves, highlighting a strong tissue-specific reprogramming of terpene metabolism. Comparisons among fruit stages revealed fewer but consistent expression shifts, particularly between yellow and purple fruits (14 differentially expressed genes, DEGs) and yellow and red fruits (7 DEGs), indicating a gradual transcriptional remodeling of TPS genes during fruit color transitions. Seed development exhibited pronounced stage-specific regulation. Comparisons between red and green seeds yielded the highest asymmetry, with eight TPS genes upregulated and no downregulated genes, suggesting a strong activation of terpene biosynthesis during seed maturation. In contrast, yellow versus red seeds showed a higher number of downregulated TPS genes, consistent with transient repression during intermediate developmental stages. Heatmap visualization of Log2 fold changes (Log2FC) across all comparisons corroborated these trends, revealing clusters of TPS genes with strong tissue-preferential expression and stage-dependent regulation. Several TPS genes displayed marked upregulation in leaves relative to roots, while others showed fruit- or seed-specific induction, particularly during red and green seed stages. The magnitude and direction of expression changes indicate functional specialization within the TPS family rather than uniform regulation. The predominance of TPS_a subclass genes among significantly upregulated transcripts in purple fruit (i.e., XP_027122319.2, XP_071905367.1, XP_071907493.1, XP_071931899.1) suggests stage-specific activation of sesquiterpene-associated pathways during intermediate fruit maturation. Expression dynamics of TPS genes across tissues and developmental stages in Coffea arabica. (A) Heatmap showing Log2 fold change (Log2FC) values for differentially expressed TPS genes across pairwise comparisons among tissues (roots, leaves) and developmental stages of fruits (purple, red, yellow) and seeds (green, yellow, red). Blue strike marks indicate C. arabica TPS duplicates. (B) Adjacent bar plots summarize the number of expressed TPS genes and the number of significantly differentially expressed TPS genes (*FDR ≤ 0.05, **FDR ≤ 0.01, ***FDR ≤ 0.001) Quantitative expression profiling further demonstrated that a subset of TPS genes is highly upregulated in specific tissues and stages (Figure 5B). In leaves, multiple TPS genes exhibited significantly higher normalized expression levels compared to roots (i.e., XP_071904867.1; 7.5-fold ± 1.07; FDR 1.74 e-11), whereas fruit tissues showed stage-dependent expression peaks, with certain TPS genes preferentially expressed in purple or red fruits. Seed tissues exhibited distinct expression patterns, with several TPS genes showing significant induction in red and yellow seeds compared to green seeds, underscoring their potential role in late seed development and maturation-associated metabolic processes. The expression profiles of C. arabica TPS duplicates remain tightly coupled to their evolutionary trajectories when evaluated in conjunction with predicted subcellular localization. These duplicate pairs belong to distinct phylogenetic subclasses, allowing interpretation of their regulatory divergence within a lineage-specific functional context. Interestingly, these paralogs exhibited a complete absence of expression across comparisons of leaves, roots, and fruit stages. For the conserved duplicate pairs XP_027093045.1/XP_027093111.1 (OG0000003), classified within the TPS_b subclass typically associated with plastidial monoterpene biosynthesis, only XP_027093045.1 showed detectable expression during seed development, with modest and statistically non-significant changes across developmental transitions (yellow vs red: −1.11-fold, FDR = 0.33; yellow vs green: −1.17-fold, FDR = 0.26; red vs green: −0.95-fold, FDR = 0.69). In contrast, its duplicate XP_027093111.1 remained transcriptionally silent across all seed stages, indicating asymmetric expression retention despite strong purifying selection at the sequence level. This pattern suggests post-duplication regulatory suppression of one paralog without evidence of compensatory upregulation in the other. A similar, though less extreme, pattern was observed in OG0000010. The paralog XP_027107080.1 displayed detectable but non-significant expression changes during seed development (yellow vs red: −1.34-fold, FDR = 0.43; yellow vs green: 1.73-fold, FDR = 0.07; red vs green: 1.42-fold, FDR = 0.18), whereas XP_071931897.1 showed lower expression amplitude and similarly no statistically supported differential expression. Both paralogs in this orthogroup belong to the TPS_a subclass, which is linked to cytosolic sesquiterpene production and defense-related functions, suggesting retention without pronounced transcriptional specialization across the sampled tissues. On the other hand, the adaptively diverged duplicate pair in OG0000013 showed clear and statistically supported transcriptional partitioning. XP_071930373.1, the paralog inferred to have experienced stronger episodic positive selection in PMALX, was significantly upregulated in red seeds relative to yellow and green stages (2.58-fold, FDR = 0.020; 2.79-fold, FDR = 0.013, respectively), whereas XP_027102632.2 showed weaker and mostly non-significant changes (yellow vs green: −2.05-fold, FDR = 0.159; red vs green: 2.66-fold, FDR = 0.013). Importantly, this pair is also classified within the TPS_b subclass, indicating that even within the same phylogenetic lineage, duplicated copies can undergo distinct regulatory and selective trajectories following polyploidization. This expression asymmetry, together with predicted differences in subcellular localization that indicated chloroplast targeting for XP_027102632.2 (WolF PSORT score = 11) versus broader cytoplasmic (score = 4), chloroplastic (score = 7), and nuclear (score = 3) localization for XP_071930373.1, supports a model in which adaptive sequence evolution is accompanied by regulatory and compartmental specialization following duplication. 4. Discussion We identified 69 TPS genes in C. arabica, compared with 40 in C. canephora and 55 in C. eugenioides. These values fall within the range reported in other angiosperms, such as A. thaliana (32), rice (34), and tomato (>50),4,11 and reflect the combined effect of polyploidy and tandem expansion. The lower-than-expected count in C. arabica relative to its progenitors suggests post-polyploid fractionation, as also observed in cotton and wheat. 9 Chromosomal mapping revealed strong clustering, particularly on chromosome 8 in C. eugenioides (40% of TPS genes), with similar tandem-rich regions retained in C. arabica. Such ‘terpene synthase hotspots’ are a common feature in plants; for example, grapevine TPS genes are concentrated on chromosomes 12, 13, and 18,13 while tomato has major TPS clusters on chromosomes 8 and 10,11 and rice shows clustering on chromosomes 4 and 11.5 Predicted features, including the absence of signal peptides and transmembrane domains with frequent plastid or cytosolic targeting, are consistent with known TPS properties.1,34 Notably, the prediction of unconventionally secreted TPS proteins in coffee may indicate roles in volatile emission, as extracellular localization of TPS activity has been linked to ecological interactions and aroma production in other species.34,35 In both C. canephora and C. eugenioides, tandem duplication was the dominant driver of TPS family expansion, accounting for ∼70% of duplicate pairs, while proximal and dispersed duplications contributed less. This pattern mirrors the general observation in land plants, where tandem duplication has repeatedly been shown to shape TPS proliferation.5,6 For example, grapevine TPS genes form large tandem clusters, 36 and tomato TPS families also expanded primarily through tandem arrays. 11 The predominance of tandem duplications in coffee thus reflects a conserved mechanism underlying the diversification of terpenoid metabolism. Most duplicates in both diploid species were constrained by purifying selection (Ka/Ks < 1), suggesting functional conservation after duplication. Similar evolutionary dynamics have been described in A. thaliana and rice, where most TPS duplicates remain conserved, with few under positive selection linked to ecological specialization.4,37,38 Approximate divergence time estimates placed many tandem duplications within 2.5–60 MYA, aligning with the diversification of Rubiaceae, while a smaller subset of younger duplications may have occurred near or after the split between C. canephora and C. eugenioides (∼1–2 MYA). 8 This indicates that both ancient lineage-retained duplications and more recent species-specific expansions shaped the diploid coffee TPS repertoires. Although Ks-based divergence estimates provide a useful comparative framework for evaluating relative duplication histories, these estimates remain sensitive to lineage-specific substitution-rate variation and synonymous-site saturation, particularly for older paralogs with elevated Ks values. 39 Therefore, the inferred timings were interpreted as approximate evolutionary estimates rather than precise dates of duplication. The expansion of TPS genes in C. arabica reflects the combined legacy of its diploid progenitors and post-polyploidy divergence. An important feature emerging from the present analyses is the apparent absence of strong subgenome dominance within the TPS repertoire of C. arabica. Orthogroup reconciliation and syntenic conservation demonstrated substantial retention of TPS copies from both C. canephora- and C. eugenioides-derived subgenomes, with only limited lineage-specific expansion and selective gene loss. Orthogroup analysis revealed that most gene families were conserved across species, yet C. arabica contained a small number of lineage-specific expansions, often in the form of tightly clustered tandem duplicates. Such expansions are characteristic of allopolyploids, where duplicated subgenomes provide opportunities for gene diversification. 40 Similar trends have been reported in polyploid cotton, where metabolic gene families, including terpenoid-related genes, underwent extensive tandem proliferation following polyploidization. 41 Gene tree reconciliation across orthogroups OG0000000–OG0000015 indicates that TPS expansion in C. arabica is limited and confined to three orthogroups (OG0000003, OG0000010, and OG0000013), each containing clear terminal, species-specific duplication events. In these cases, duplicated gene pairs form tight sister clades with extremely short branch lengths, consistent with recent paralog formation after speciation rather than large-scale post-polyploid amplification. Importantly, no evidence was detected for ancestral duplication inherited uniquely from either diploid progenitor within these orthogroups beyond expected homeolog retention. Instead, the broader evolutionary pattern of the TPS family in C. arabica reflects substantial retention of progenitor-derived homeologs, accompanied by selective gene loss and only modest, recent small-scale duplication. This indicates that TPS repertoire evolution in C. arabica is primarily shaped by homeolog maintenance and limited lineage-specific duplication, rather than extensive genome restructuring. Although homoeologous exchanges have been shown to remodel metabolic gene families in polyploids such as Brassica napus. 42 Our reconciliation and synteny analyses do not provide evidence for large-scale exchange-driven restructuring within the TPS family of coffee. The majority of C. arabica duplicates remain under purifying selection, consistent with functional retention, while only a small subset exhibits signatures of divergence. Such patterns align with broader models of duplicate-gene fate in polyploid plants,40,43 in which most retained copies remain functionally constrained, and only a minority undergo adaptive divergence. Our analyses also indicate that most TPS duplicates in C. canephora, C. eugenioides, and C. arabica are constrained by purifying selection, reflecting functional stability. This is a common pattern across plant gene families, for example, genome-wide studies of MYB and AP2/ERF transcription factors in Brassica napus and Coffea revealed that the majority of duplicates have Ka/Ks < 1, while only a few show signs of positive selection linked to localized adaptive shift.10,44 In C. arabica, most orthogroups remained strongly constrained; however, OG0000013 showed evidence of localized episodic positive selection affecting a limited subset of codon sites and branches, coupled with undetected significantly supported BEB, suggesting localized sequence divergence following polyploidization rather than widespread adaptive remodeling. Comparable cases of episodic selection acting on individual members of otherwise conserved gene families have been documented in Arabidopsis CONSTANS-like genes, where adaptive changes were restricted to specific paralogs rather than the entire family. 45 At the structural level, Coffea sp. TPS proteins are largely characterized by the canonical PF01397 and PF03936 domains, with only rare instances of additional domain configurations. Similar patterns of structural conservation are punctuated by occasional domain rearrangements. They have been reported in plant metabolic gene families such as chalcone synthases and cytochrome P450s, where limited structural variation can facilitate functional divergence without disrupting core enzymatic frameworks.46,47 The expression landscape of TPS genes in C. arabica indicates that possible functional diversification within this family might be driven predominantly by regulatory partitioning rather than widespread coding-sequence divergence. The uneven distribution of expressed TPS genes across tissues, particularly their higher representation in roots and seeds, together with strong transcriptional reprogramming between leaves and roots, suggests that terpene biosynthesis is spatially structured and organ-specific rather than constitutive. Gradual remodeling across fruit developmental stages further supports a model of developmental tuning of TPS deployment. While metabolite-level validation was not performed in this study, the observed stage-specific expression shifts are consistent with reports of stage-dependent terpene synthase regulation in fleshy fruits like grapes and berries.11,13 The enrichment of TPS_a subclass members among upregulated transcripts in purple fruits also points to subclass-level regulatory stratification, indicating that different enzymatic lineages are selectively mobilized during discrete developmental windows. When integrated with evolutionary analyses, these expression patterns suggest a hierarchical model of divergence in the coffee TPS family. Most retained duplicates remain under strong purifying selection and show limited transcriptional asymmetry, consistent with duplicate retention under functional constraint.10,44 However, the OG0000013 pair represents a notable exception, where episodic positive selection coincides with significant upregulation during red seed maturation and divergence in predicted subcellular localization. The convergence of selective, regulatory, and compartmental differentiation in this orthogroup supports the view that functional innovation in polyploid genomes is typically restricted to a minority of paralogs.45,48 However, it is important to note that the present interpretations of TPS diversification are primarily based on comparative evolutionary and transcriptional evidence. Although expression asymmetry, selective divergence, and predicted subcellular differentiation together suggest potential functional differentiation among retained paralogs, RNA-seq profiles alone cannot establish altered enzymatic activity or metabolite output. Therefore, the inferred diversification patterns were interpreted as evidence of regulatory and evolutionary divergence rather than direct confirmation of biochemical functional innovation. 5. Conclusion This study provides a comprehensive evolutionary and transcriptional analysis of the TPS gene family in C. arabica and its diploid progenitors. The TPS repertoire reflects a dual pattern of conservation and limited localized divergence, with almost all duplicates constrained by purifying selection, while only limited evidence of episodic positive selection was detected in a single orthogrgoup. Chromosomal clustering, particularly on chromosome 8, underscores tandem duplication as a principal driver of expansion, whereas polyploidy in C. arabica facilitated the retention of homologous copies with differential evolutionary trajectories. Expression analyses revealed pronounced tissue- and stage-specific regulation, indicating that diversification within the TPS family is mediated primarily through regulatory divergence across organs and developmental phases. Notably, only a minority of paralogs, such as members of OG0000013, showed convergence of adaptive sequence evolution and regulatory specialization, whereas most duplicates remained evolutionarily and transcriptionally constrained. Although functional roles were inferred from expression and comparative analyses, experimental validation of enzyme activity and metabolite production remains necessary. Future studies integrating biochemical assays, metabolomic profiling, and broader comparative genomics across wild Coffea sp. will be essential to directly connect TPS evolutionary dynamics with terpene diversity, ecological interactions, and coffee quality traits. Supplemental Material Supplemental Material - Tracing the Evolutionary Dynamics and the Transcriptional Partitioning of Terpene Synthase Gene Family in Coffea arabica and Its Diploid Progenitors Supplemental Material for Tracing the Evolutionary Dynamics and the Transcriptional Partitioning of Terpene Synthase Gene Family in Coffea arabica and Its Diploid Progenitors by Makarim Elfadil M. Osman, Mohanad A. Ibrahim, Rieham Sallah H. Osman, Budour A. Alghamdi, Amina I. Dirar, Wafa A. Alrasheed, Abdelsattar M. Omar and Emadeldin Hassan E. Konozy in Bioinformatics and Biology Insights. Supplemental Material - Tracing the Evolutionary Dynamics and the Transcriptional Partitioning of Terpene Synthase Gene Family in Coffea arabica and Its Diploid Progenitors Supplemental Material for Tracing the Evolutionary Dynamics and the Transcriptional Partitioning of Terpene Synthase Gene Family in Coffea arabica and Its Diploid Progenitors by Makarim Elfadil M. Osman, Mohanad A. Ibrahim, Rieham Sallah H. Osman, Budour A. Alghamdi, Amina I. Dirar, Wafa A. Alrasheed, Abdelsattar M. Omar and Emadeldin Hassan E. Konozy in Bioinformatics and Biology Insights. Supplemental Material - Tracing the Evolutionary Dynamics and the Transcriptional Partitioning of Terpene Synthase Gene Family in Coffea arabica and Its Diploid Progenitors Supplemental Material for Tracing the Evolutionary Dynamics and the Transcriptional Partitioning of Terpene Synthase Gene Family in Coffea arabica and Its Diploid Progenitors by Makarim Elfadil M. Osman, Mohanad A. Ibrahim, Rieham Sallah H. Osman, Budour A. Alghamdi, Amina I. Dirar, Wafa A. Alrasheed, Abdelsattar M. Omar and Emadeldin Hassan E. Konozy in Bioinformatics and Biology Insights. Supplemental Material - Tracing the Evolutionary Dynamics and the Transcriptional Partitioning of Terpene Synthase Gene Family in Coffea arabica and Its Diploid Progenitors Supplemental Material for Tracing the Evolutionary Dynamics and the Transcriptional Partitioning of Terpene Synthase Gene Family in Coffea arabica and Its Diploid Progenitors by Makarim Elfadil M. Osman, Mohanad A. Ibrahim, Rieham Sallah H. Osman, Budour A. Alghamdi, Amina I. Dirar, Wafa A. Alrasheed, Abdelsattar M. Omar and Emadeldin Hassan E. Konozy in Bioinformatics and Biology Insights. Acknowledgments The authors gratefully acknowledge Dr. Salem S. Alghamdi, Professor of Genetics and Plant Breeding at King Saud University, for his valuable guidance and support during the development of this coffee research study. Also, E. H. E. M. Konozy (the corresponding author) acknowledges the Institute of International Education’s Scholar Rescue Fund (IIE-SRF) for the awarded fellowship. Author Contributions: MEMO: Conceptualization; Methodology; Data curation; Formal analysis; Validation; Visualization; Writing-original draft; Writing-review and editing. MAI: Conceptualization; Formal analysis; Validation. RSHO: Methodology; Formal analysis. BAA: Project administration; Resources; Writing - Review & Editing. AID: Validation; Writing-original draft; Writing-review and editing. WAA: Resources; Writing - Review & Editing. AMO: Validation. EHEK: Conceptualization; Validation; Supervision; Writing-original draft; Writing-review and editing. All the authors commented on previous versions of the manuscript. All the authors read and approved the final manuscript. Funding: The authors received no financial support for the research, authorship, and/or publication of this article. The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article. Supplemental Material: Supplemental material for this article is available online. ORCID iDs Makarim Elfadil M. Osman https://orcid.org/0000-0001-6619-6999 Amina I. Dirar https://orcid.org/0000-0001-9954-2579 Ethical Considerations No specific permits were needed, and all repository data servers were freely accessible. Data Availability Statement All the data generated and analyzed during this study are included in the main article, and supplemental data are provided (additional file documents 1-3). The genomic data of Coffea arabica, C. canephora, and C. eugenioides are freely accessible, and all related UTR links are supplied within the article under relevant mentions.* References Associated Data Supplementary Materials Supplemental Material - Tracing the Evolutionary Dynamics and the Transcriptional Partitioning of Terpene Synthase Gene Family in Coffea arabica and Its Diploid Progenitors Supplemental Material for Tracing the Evolutionary Dynamics and the Transcriptional Partitioning of Terpene Synthase Gene Family in Coffea arabica and Its Diploid Progenitors by Makarim Elfadil M. Osman, Mohanad A. Ibrahim, Rieham Sallah H. Osman, Budour A. Alghamdi, Amina I. Dirar, Wafa A. Alrasheed, Abdelsattar M. Omar and Emadeldin Hassan E. Konozy in Bioinformatics and Biology Insights. Supplemental Material - Tracing the Evolutionary Dynamics and the Transcriptional Partitioning of Terpene Synthase Gene Family in Coffea arabica and Its Diploid Progenitors Supplemental Material for Tracing the Evolutionary Dynamics and the Transcriptional Partitioning of Terpene Synthase Gene Family in Coffea arabica and Its Diploid Progenitors by Makarim Elfadil M. Osman, Mohanad A. Ibrahim, Rieham Sallah H. Osman, Budour A. Alghamdi, Amina I. Dirar, Wafa A. Alrasheed, Abdelsattar M. Omar and Emadeldin Hassan E. Konozy in Bioinformatics and Biology Insights. Supplemental Material - Tracing the Evolutionary Dynamics and the Transcriptional Partitioning of Terpene Synthase Gene Family in Coffea arabica and Its Diploid Progenitors Supplemental Material for Tracing the Evolutionary Dynamics and the Transcriptional Partitioning of Terpene Synthase Gene Family in Coffea arabica and Its Diploid Progenitors by Makarim Elfadil M. Osman, Mohanad A. Ibrahim, Rieham Sallah H. Osman, Budour A. Alghamdi, Amina I. Dirar, Wafa A. Alrasheed, Abdelsattar M. Omar and Emadeldin Hassan E. Konozy in Bioinformatics and Biology Insights. Supplemental Material - Tracing the Evolutionary Dynamics and the Transcriptional Partitioning of Terpene Synthase Gene Family in Coffea arabica and Its Diploid Progenitors Supplemental Material for Tracing the Evolutionary Dynamics and the Transcriptional Partitioning of Terpene Synthase Gene Family in Coffea arabica and Its Diploid Progenitors by Makarim Elfadil M. Osman, Mohanad A. Ibrahim, Rieham Sallah H. Osman, Budour A. Alghamdi, Amina I. Dirar, Wafa A. Alrasheed, Abdelsattar M. Omar and Emadeldin Hassan E. Konozy in Bioinformatics and Biology Insights. Data Availability Statement All the data generated and analyzed during this study are included in the main article, and supplemental data are provided (additional file documents 1-3). The genomic data of Coffea arabica, C. canephora, and C. eugenioides are freely accessible, and all related UTR links are supplied within the article under relevant mentions.*