Introduction
Moringa (Moringa oleifera Lam.) is a widely distributed perennial tree native to the sub-Himalayan valleys. Today it is cultivated across most tropical and subtropical regions, including South and Southeast Asia, Austronesia, the Americas and Africa (Olson Reference Olson2019; Bania et al. Reference Bania, Deka, Hazarika, Das, Nath and Sileshi2023). Despite centuries of cultivation, end-use varies markedly by culture: fruit (pods) predominate in India, leaves and young shoots in the Philippines, seed oil in the Seychelles and ornamental use in Mexico (Palada et al. Reference Palada, Patricio, Ebert and Wu2014). This diversity reflects underlying genetic and biochemical variation that remains largely uncharacterized, limiting the species’ transition to a commercial specialty crop with market-specific cultivars.
The nutritional and medicinal value of moringa is well documented (Chaudhary and Chaurasia Reference Chaudhary and Chaurasia2017; Kou et al. Reference Kou, Li, Olayanju, Drake and Chen2018), reflecting a biochemical profile rich in essential amino acids, provitamin A, vitamins C and E, minerals, and phenolic and flavonoid compounds (Klimek-Szczykutowicz et al. Reference Klimek-Szczykutowicz, Gaweł-bęben, Rutka, Blicharska, Tatarczak-michalewska, Kulik-siarek and Szopa2024), underpinning its recognition as an ‘opportunity crop’ for diversifying food systems in Africa (Asare et al. Reference Asare, Gyan, Bugyei, Adjei, Mahama, Addo and Nyarko2012; van Zonneveld et al. Reference van Zonneveld, Kindt, McMullin, Achigan-Dako, N’danikou, Hsieh and Dawson2023). Dried leaf serves both human nutrition and high-protein animal feed (Moyo et al. Reference Moyo, Masika, Hugo and Muchenje2011), and seeds function as a water flocculant (Eichhorn et al. Reference Eichhorn, Weckmüller and Urban2022). Yet, the only domestication-related traits identified to date are glucosinolate content and growth rate (Chodur et al. Reference Chodur, Olson, Wade, Stephenson, Nouman, Garima and Fahey2018), and the extent of domestication across the species’ range remains unclear.
Systematic evaluation of phenotypic diversity, particularly biochemical traits, across M. oleifera genetic resources remains limited: prior studies focused on isolated traits or specific uses rather than comprehensive assessment of variation within and among germplasm collections (Palada et al. Reference Palada, Patricio, Ebert and Wu2014; Drisya Ravi et al. Reference Drisya Ravi, Nair and Siril2021; He et al. Reference He, Yang, Wei, Wu, Zheng, Zhang and Peng2021). This gap creates constraints: (1) producers cannot reliably match germplasm to target markets (leaf powder, seed oil, fresh vegetable and pharmaceutical); (2) quality remains inconsistent across supply chains; and (3) breeding programmes lack diversity data for parental selection and crossing design, reinforcing reliance on unimproved or poorly characterized material (McMullin et al. Reference McMullin, Stadlmayr, Mausch, Revoredo-giha, Burnett, Guarino, Brouwer, Jamnadass, Graudal, Powell and Dawson2021; van Zonneveld et al. Reference van Zonneveld, Castañeda-álvarez, Achigan-Dako, Almekinders, de Haan and Hugo2026).
To advance this species, it is important to define the ideotype (Donald Reference Donald1968). Tree architecture, days to first flower, leaf morphology and reproductive output define crop ideotypes for specific markets: late flowering with minimal branching facilitates mechanical harvest for leaf powder, whereas heavy branching with precocious flowering maximizes fruit set for pod production. Quantifying phenotypic variation enables matching germplasm to market and identifying genotypes suitable for evaluation or as breeding parents, while trait correlations enable indirect selection, where easily measured traits (e.g., height and branching) predict labour-intensive targets (e.g., leaf yield and fruit production).
Breeding has been constrained by limited germplasm characterization, absent multi-season stability data and a lack of strategies balancing genetic gain with diversity maintenance. Most prior evaluations assessed fewer than 50 accessions in single-year trials within narrow regional collections, precluding reliable estimates of genotype-by-timepoint interactions critical for selection (Mulugeta Reference Mulugeta2014; Palada et al. Reference Palada, Patricio, Ebert and Wu2014). Few moringa collections exist worldwide; the major repository, at the World Vegetable Center (WorldVeg; Taiwan), holds 87 accessions, primarily from South and Southeast Asia, with limited African representation (Palada et al. Reference Palada, Patricio, Ebert and Wu2014). To expand this resource’s geographic scope, we constructed a diversity panel of 32 half-sibling families from 7 countries/programmes drawn from the WorldVeg collection alongside local Hawaiian germplasm, spanning cultivars/landraces, wild, feral and breeding lines. Our objectives were to (1) characterize morphological, phenological and agronomic diversity within and among families and countries/programmes across two growing seasons; (2) quantify multi-season trait stability to assess the reliability of single-season evaluation; and (3) identify market-defining trait antagonisms and the specialized germplasm associated with each major market class. This characterization provides a foundation for transitioning M. oleifera from a generalized multipurpose tree to a commercial crop with defined ideotypes tailored to leaf powder, seed oil, fresh vegetable and pharmaceutical markets.
Materials and methods
Germplasm acquisition and experimental design
Moringa oleifera accessions were sourced from the World Vegetable Center (WorldVeg, Taiwan), University of Hawaiʻi at Mānoa and local Hawaiian collections as half-sibling seed families (Table 1). In total, 24 WorldVeg accessions that were available under the FAO Plant Treaty MultiLateral System were obtained and used in this study. The University of Hawaiʻi at Mānoa families originate from a segregating half-sibling population with ‘Periyakulam 1’ (PKM-1) of Tamil Nadu Agricultural University (India) as the maternal parent, subsequently mass-selected over five generations for precocious flowering and novel red/purple colour. The original segregating population of seed comes from an internationally curated collection of the Florida group Educational Concerns for Hunger Organization (ECHO) with unknown paternal parents. The germplasm screened consists of half-sibling families from open-pollinated, wild and feral individuals with provenances from South and Southeast Asia (India, Thailand and Malaysia), Polynesia (Hawaii and Taiwan), East Africa (Tanzania) and the Caribbean (US Virgin Islands). Provenances represent precise Genesys passport data associated with the Accession IDs.
Germplasm sources and information for the 32 half-sibling families (HARC family ID) characterized in this study, sampled from seven Countries/Programmes. The HARC family ID column represents family-level statistical analyses in this study. The provenance column represents more precise Genesys passport data, used for selection and crossing design
Table 1 Long description
A table of the 32 half-sibling families evaluated in the study, one family per row. Columns give the accession name, the Genesys accession ID, the HARC family ID, biological status, country or program, provenance, and the number of half-siblings. Biological status is traditional cultivar or landrace for most families, with wild types among several Thai families, feral types from Hawaii, and breeding lines from University of Hawaii at Manoa. Countries and programs are Thailand, Taiwan, Hawaii, University of Hawaii at Manoa, US Virgin Islands, Malaysia, and Tanzania, with Thailand the most represented. The number of half-siblings per family ranges from 1 to 6.
The experiment employed a randomized complete block design with unequal family replication due to germination constraints. Families (n = 32) were evaluated across three blocks, with each family represented by 3–6 individuals depending on seed availability and germination success (Table 1). Individual trees, belonging to each family, were randomly assigned to positions within blocks, resulting in 156 plants total (Figure 1). Several mortalities occurred during the trial, resulting in 148 surviving individuals and 133 individuals with complete phenotypic data. Moringa oleifera’s mixed mating system (approximately 30% outcrossing; Muluvi et al. Reference Muluvi, Sprent, Odee and Powell2004) means that individuals within families represent mixtures of full-sibling and half-sibling progeny in unknown proportions.
Moringa oleifera RCBD germplasm field at 3-months post-transplanting and 5-months post-germination.
Figure 1 Long description
A landscape photograph of the Moringa oleifera field trial at Kunia, Hawaii, three months post-transplanting and five months post-germination. Rows of young, thin Moringa saplings with green foliage are spaced widely across a grass mixture cover-crop.
Field site and agronomic management
The field trial was established at the Kunia Research Station of the Hawaii Agriculture Research Center (21.384253°N, −158.038122°W), Kunia, Hawaii, USA. The site features Molokai silty clay loam (very-fine, kaolinitic, isohyperthermic Typic Eutrotorrox) in a tropical climate with mean annual rainfall of 813 mm and mean growing season temperature of 22°C. Soil analysis conducted on 20 May 2024 guided pre-plant amendments (Supplementary Table S1): 50 lbs/1000 ft2 gypsum, 50 lbs/1000 ft2 meat and bone meal (9-7-1 NPK) and 5 lbs/1000 ft2 magnesium sulphate, incorporated to 6-inch depth across the 0.38-acre site.
Seeds were scarified by applying light pressure until the seed coat cracked without fully breaking, then sown on 16 June 2024, into a saturated 1:1 mixture of Sunshine Mix #4 and perlite. Most families achieved complete germination within 1 week, with remaining seeds germinating in the second week. Seedlings were maintained in a polycarbonate greenhouse for 2 months before field transplanting on 20 August 2024. Trees were planted with 9-foot inter- and intra-row spacing across 12 offset rows.
Monthly fertilization consisted of granular urea, triple superphosphate and muriate of potash applied at 10 oz/acre each of N, P and K per month (total annual application: 7.5 lbs/acre per nutrient). Irrigation was delivered via Bowsmith Fanjet Red sprinklers at 30 PSI (29.4 gph per emitter) with 84 emitters distributed across the field. During the rainy season (October–March), irrigation occurred twice weekly (approximately 0.1 acre-feet/week); during the dry season (April–September), four times weekly (approximately 0.2 acre-feet/week). After 1 year of growth (4 August 2025), the entire field was coppiced to 3-foot height to stimulate coppice shoot production for trait evaluation.
Phenotyping
All trees were characterized for morphological, phenological and agronomic traits using ontologies specifically designed based on M. oleifera product markets across two growing seasons (Table 2). Year 1 (August 2024–August 2025) included measurements over 12 months from transplanting for tree height, leaf count, branching, fruit count, days to first flower, leaf morphology and fruit characteristics. Following coppicing to 3-foot height in August 2025, trees were re-evaluated after 6 months of regrowth (August 2025–February 2026) for tree height, leaf count, fruit count and branching. Coppice shoot production was counted separately at 1 and 6 months of regrowth to assess consistency of regrowth vigour. This multi-season evaluation enabled assessment of trait stability and coppicing response, a critical characteristic for leaf and fruit perennial production systems.
Trait descriptions and measurement protocols. Although not explicitly listed under Type, most morphological and phenological have an impact on agronomic performance of moringa, such as tree height, days to first flower, branching, coppice shoots and leaf count
Table 2 Long description
A table describing the 12 traits measured, one trait per row, with columns for trait name, type, and description. Eight traits are morphological: branching, count of branches over 2 cm; coppice shoots, count one month post-coppicing; fruit diameter in mm at the midpoint; fruit length in cm from pedicel to tip; leaf count of healthy compound leaves over 95 percent green; leaflet shape as a length-to-width ratio approaching 1.0 for circular and below 0.6 for oblong; rachilla length, length of rachis in cm between the first and second rachilla; and tree height in cm from soil to apical meristem, measured six times across the first year. One trait is phenological: days to first flower from germination to first inflorescence. Three traits are agronomic: fruit count of mature fruits measured three times across the first year, hundred-seed weight in grams of 100 dehulled seeds, and seed count per pod.
Single-measurement traits assessed at specific developmental stages included: days to first flower (days from germination to first inflorescence), branching intensity (count of branches >2 cm in length), coppice shoot production (count of shoots post-coppicing), leaflet shape index (length:width ratio), rachilla length (cm of rachis between first and second rachilla on newest fully formed leaf), fruit dimensions (diameter at midpoint and length from pedicel to tip, averaged from three randomly selected mature fruits), seed count per pod (averaged from three randomly selected mature fruits) and hundred-seed weight (grams of 100 dehulled seeds, estimated from three random fruit samples). Fold differences in the phenotyped traits were calculated using raw phenotypic values at the individual level (half-sibling/ plot level) by dividing the maximum value by the minimum value across the characterized germplasm field.
Statistical analysis
Phenotypic data from two growing seasons were analysed using linear mixed models to estimate family effects while accounting for spatial and temporal variation (Bates et al. Reference Bates, Mächler, Bolker and Walker2015). For traits measured in both evaluations (tree height, leaf count, fruit count, branching and coppice shoot count), models included family and evaluation timepoint as fixed effects within block and family × timepoint interaction as random effects. Traits measured only in year 1 (days to first flower, leaflet shape, rachilla length, pod dimensions and seed traits) were analysed with family as a fixed effect and block as a random effect. Best linear unbiased estimators quantified family performance adjusted for environmental effects.
Genetic correlations between evaluations assessed trait stability by correlating family-level predictions between evaluations. For the four traits recorded in both year 1 and 6-month regrowth, correlations related year-1 and regrowth predictions; for coppice shoot count, the correlation related 1- and 6-month regrowth counts. Higher correlations indicate that family rankings are largely preserved between evaluations, whereas lower correlations indicate greater genotype-by-timepoint interaction; because each estimate derives from a single pair of measurements spanning a coppice transition, they are interpreted as preliminary indicators of rank stability rather than definitive selection parameters. Statistical significance was evaluated at α = 0.05 using Type III Analysis of Variance with Satterthwaite’s approximation.
Variance components were partitioned to quantify the proportion of phenotypic variation attributable to different sources. For this analysis, parallel models with all effects treated as random were fitted to enable variance component estimation on a common scale. For single-year traits, variance was partitioned among family, block and residual effects. For traits measured in both evaluations, variance was partitioned among timepoint, family, family × timepoint interaction, block and residual effects. Detailed statistical models including equations and computational procedures are provided in Supplemental Materials and Methods. Phenotypic correlations among 12 traits were estimated using Pearson’s correlation coefficient calculated from individual-level values across the 133 individuals with complete observations for all 12 traits (year 1 data).
Selection index construction
To characterize how measured traits partition germplasm among major market classes, we constructed two selection indices as a classification tool. Individual genetic merit was predicted by combining family effects with individual deviations from family means. Market-specific selection indices weighted six traits according to the traits defining each market ideotype: the seed market index prioritized early flowering, compact stature, heavy branching and vigorous coppicing; the leaf market index favoured delayed flowering, tall unbranched architecture and vigorous regrowth. Index values were calculated as weighted linear combinations of individual predictions (Supplemental Materials and Methods), with individuals ranked within each market. A 70th percentile threshold identified top-performing germplasm for each market.
Results
Phenotypic diversity in the moringa germplasm collection
There was phenotypic variation across tree architecture, leaf morphology and reproductive characteristics (Supplementary Figures S1–S3). Significant family-level differences (p < 0.05) were detected for nine of 12 traits, indicating strong genetic control for most traits (Supplementary Tables S2–S3). Tree height at 1 year after planting varied ∼11-fold in year 1 (Fig. 2; Supplementary Table S2), with significant family effects (p < 0.001; Supplementary Figure S1A). Following coppicing to 3-foot height in August 2025, 6-month regrowth reached 356–638 cm, demonstrating vigorous regenerative capacity for leaf and juvenile-fruit production. Family rankings for tree height were largely preserved between year 1 and 6-month regrowth (rg = 0.776; Fig. 3). Height was positively correlated with leaf count (r = 0.61), hundred-seed weight (r = 0.46) and branching intensity (r = 0.47; Supplementary Figure S4). Days to first flower varied from ∼120 to over 400 days (Figure 2; Supplementary Table S2), with highly significant family effects (p < 0.001; Supplementary Figure S1B). Days to first flower were negatively correlated with fruit diameter (r = −0.51) and hundred-seed weight (r = −0.41), indicating that early-flowering genotypes permit extended fruit development periods. Branching intensity varied from zero branches, the characteristic ‘telephone pole’ ideotype for leaf production, to 49 branches in year 1 with significant family effects (p < 0.001; Figure 2; Supplementary Figure S1C). Family rankings for branching showed moderate consistency between evaluations (rg = 0.679; Fig. 3), with family-by-timepoint interaction contributing 19.3% of variance (Supplementary Table S4). Branching was strongly correlated with leaf count (r = 0.88), fruit count (r = 0.50) and coppice shoot production (r = 0.38). Coppice shoot counts 1-month post-coppicing varied ∼34-fold (Figure 2; Supplementary Table S2), with significant family effects (p < 0.001; Figure 4). Coppice shoot count was highly consistent between 1- and 6-month regrowth (rg = 0.896; Supplementary Table S4), indicating that early regrowth vigour predicts later regrowth. UH-C1, UH-D2 and UH-C4 showed particularly strong coppicing, producing significantly more shoots than baseline (Supplementary Table S2). Leaf count displayed significant family-level variation (p < 0.001; Figure 2), spanning roughly 10–98 healthy leaves in year 1 and 56–145 in regrowth (Figure 2; Supplementary Table S2). Family rankings for leaf count showed moderate consistency between evaluations (rg = 0.544; Figure 3), with family-by-timepoint interaction contributing 5.7% of variance (Supplementary Table S4). In contrast, leaflet shape (p = 0.119) and inter-rachilla length (p = 0.074) did not show significant family effects despite substantial genotypic variation (Figure 5). Leaflet shape index ranged from 0.52, extremely oblong, to 1.0, perfectly circular (Supplementary Figure S2A). Inter-rachilla length ranged from 6.4 to 16.5 cm (Supplementary Figure S4C). Fruit count showed significant family effects (p < 0.001), ranging from zero (16 families) to 224 fruits in year 1, and from 0 to 71 during the 6-month regrowth period (Figure 2; Supplementary Table S2). Family rankings for fruit count were largely preserved between evaluations (rg = 0.707; Figure 3), with only 1.2% of variance from the family-by-timepoint interaction (Supplementary Table S4). Fruit diameter (p = 0.003) and fruit length (p < 0.001) both showed significant family effects (Supplementary Table S2). Seeds per fruit varied significantly (p = 0.018; Supplementary Table S2). Hundred-seed weight, however, did not show family-level variation (p = 0.519) despite ∼13-fold phenotypic variation (Supplementary Table S2), with family effects explaining only 5.5% of within-evaluation variance (Supplementary Table S4), suggesting seed size variation is primarily within-family rather than between-family.
Family-level performance (BLUEs) for morphological, phenological, and agronomic traits. Best linear unbiased estimators (BLUEs) with 95% confidence intervals for 32 families: (A) tree height (cm), (B) branching intensity (count), (C) fruit count, (D) days to first flower, (E) coppice shoot count, and (F) leaf count. For traits measured in both evaluations (A, B, C, E, F), BLUEs are adjusted for block and evaluation effects; days to first flower (D) is from Year 1 only. Families are ordered by BLUE within each panel; the dashed line indicates the population mean. Horizontal bars show 95% confidence intervals.
Figure 2 Long description
Six panels each display the 32 families ranked from lowest to highest best linear unbiased estimate (BLUE) for one trait, with each family as a point, a horizontal bar for its 95 percent confidence interval, and a vertical dashed line at zero marking the population mean. Panel A, tree height in cm, spans roughly an eleven-fold range, with University of Hawaii families UH-D2 and UH-C1 and several Taiwan and Thai families at the tall end and families AVRDC-24 through AVRDC-26 near zero at the short end. Panel B, branching, ranges from families with essentially no branches to UH-C1 with the heaviest branching. Panel C, fruit count, shows most families at low to moderate values with families such as KRS-3 and AVRDC-4 separating at the high end. Panel D, days to first flower, spans early families near 120 days to late families exceeding 380 days. Panel E, coppice shoots, shows the UH families UH-D2, UH-C1, and UH-C4 producing the most shoots. Panel F, leaf count, ranges widely, with UH-D2 and UH-C1 highest. Confidence intervals are widest for families represented by fewer surviving individuals.
Family rank stability between Year 1 and 6-month post-coppice regrowth. Family-level best linear unbiased predictors (BLUPs) correlated between Year 1 (full growth) and 6-month regrowth for (A) tree height, (B) leaf count, (C) fruit count, and (D) branching. Genetic correlation coefficients (rg) quantify preservation of family rankings between evaluations; point color denotes stability tier (rg ≥ 0.7, green; 0.5 0.7, orange) and lines show linear fit with 95% confidence intervals. Because each estimate derives from a single evaluation pair spanning a coppice transition, values are preliminary indicators of rank stability rather than selection parameters. Coppice regrowth consistency (one- vs. six-month, rg = 0.896) is reported in Table S4.
Figure 3 Long description
Four panels each plot family-level predictions from Year 1 on the horizontal axis against 6-month post-coppice regrowth on the vertical axis, one point per family, with a fitted linear regression line and shaded 95 percent confidence band. Panel A, tree height, shows points clustered along an upward trend with a genetic correlation (rg) of 0.776, colored green for the higher stability tier. Panel C, fruit count, shows a similar positive relationship with rg of 0.707, also green. Panel B, leaf count, shows a looser upward scatter with more dispersion, rg of 0.544, colored orange for the moderate tier. Panel D, branching, is intermediate at rg of 0.679, also orange. Across all four panels the positive slopes indicate that families largely retained their relative rank between the two evaluations, most consistently for tree height and fruit count and least so for leaf count.
Coppice shoot regeneration variability showing high coppice shoot counts (left) to low counts (right).
Figure 4 Long description
Two field photographs comparing coppice regrowth on cut Moringa stumps. The left image shows a stump with a dense cluster of many vigorous green shoots emerging from the top and sides, representing a high coppice-shoot count. The right image shows a similar cut stump with only a few small scattered shoots, representing a low count.
Mature leaf photos showing morphological characteristics of compound leaves. Rachilla length is top and leaflet shape is bottom, moving left to right from short rachilla length (Mo-TH-10) and oblong leaflet shape (Mo-TW-11) to long rachilla length (Mo-TH-58) and round leaflet shape (Mo-US-13), respectively.
Figure 5 Long description
Four Moringa compound leaves arranged on gridded measurement boards, illustrating variation in two leaf traits. Rachilla length increases and leaflet shape shifts across the examples: the leaves progress from short rachilla length (top left) to long rachilla length (top right) and from oblong leaflets (bottom left) to round leaflets (bottom right). Each leaf shows a single, central compound leaf laid flat against a grid for scale.
Variance component analysis quantified the contribution of different sources to total phenotypic variation (Supplementary Table S4). For traits measured across both evaluations, timepoint effects explained the largest proportion of variance for tree height (68.9%) and leaf count (56.5%), reflecting the shift in mean values between year 1 growth and 6-month coppice regrowth. Analysing evaluations separately, family effects explained a large proportion of variance for many traits; 35.2% for tree height, 47.3% for fruit count, 42.1% for coppice regrowth, 38.5% for branching and 28.4% for leaf count (Supplementary Table S4). For year 1-only traits, family effects explained 43.2–45.1% of variance for days to first flower and fruit length (Supplementary Table S4).
Several morphological variants also appeared at low frequency (<5%). Lateral dominance (versus typical apical dominance) occurred in family UH-C1 (Mo-IN-3, Mo-IN-4 and Mo-IN-7), accompanied by basal adventitious shoots, as well as Mo-TH-50 (AVRDC-13, Chanthaburi, Thailand) and Mo-TH-34 (AVRDC-8, Ratchaburi, Thailand). Anthocyanin pigmentation (purple or red coloration of rachis and fruits, Figures 6–8) occurred primarily in UH-C1: purple rachis and fruits (Mo-IN-3, Mo-IN-4, Mo-IN-6 and Mo-IN-8) and red rachis and fruits (Mo-IN-7), with isolated, red-fruited individuals in five Thai families (AVRDC-3, AVRDC-5, AVRDC-7, AVRDC-17 and AVRDC-19). Dwarf growth habit (<50 cm at 1 year) occurred in 5% of genotypes across five families (AVRDC-3, AVRDC-15, AVRDC-18, AVRDC-23 and KRS-2).
Mature leaf images showing the variation of anthocyanin pigmentation along the rachilla, moving left to right: green (Mo-TH-43), light red (Mo-TH-20), and deep red (Mo-IN-6).
Figure 6 Long description
Three Moringa compound leaves photographed on gridded boards to show variation in anthocyanin pigmentation along the rachis. From left to right the central rachis progresses from green in Mo-TH-43, to light red in Mo-TH-20, to deep red in Mo-IN-6, while the leaflets remain green. Each leaf is laid flat against a grid for scale.
Fruit maturation phases showing green (left, Mo-IN-11) and red (right, Mo-IN-6) fruits.
Figure 7 Long description
Two grouped photographs of mature Moringa fruits, long slender pods, laid on gridded measurement boards for scale. The left group shows green-toned pods from Mo-IN-11 and the right group shows darker red-toned pods from Mo-IN-6, illustrating fruit color variation between genotypes. Several pods exceed 30 cm in length against the grid.
Juvenile fruit (∼2-3 weeks post-fertilization) color variation across the characterized germplasm collection.
Figure 8 Long description
Four juvenile Moringa fruits, about two to three weeks post-fertilization, laid vertically side by side on blue graph paper. The slender ribbed pods show a gradient of color across the collection, from green on the left, through orange-red and mixed green-red in the middle, to deep purple-red on the right, illustrating early-fruit anthocyanin variation among genotypes.
Market-specific germplasm identification through selection indices
To identify germplasm optimized for distinct market applications, we constructed two selection indices from predicted individual genetic values (BLUPs) for six traits. The seed market index weighted coppice shoots (+0.20), fruit count (+0.20) and branching (+0.20) positively, while penalizing tree height (−0.20) and days to first flower (−0.20). The leaf market index weighted coppice shoots (+0.15), tree height (+0.20) and days to first flower (+0.15) positively, while penalizing fruit count (−0.15), branching (−0.30) and rachilla length (−0.05). Individual index values ranged from −38.5 to 49.1 for the seed market and −17.9 to 36.3 for the leaf market across 133 individuals with complete data.
Classification of individuals using 70th percentile thresholds (seed index = 9.9, leaf index = 11.5) identified 80 top-ranked individuals: 40 seed market specialists and 40 leaf market specialists, with no individuals exceeding both thresholds (Figure 9). This clear specialization indicates that seed-production trait combinations (high fruit count, early flowering and moderate height) are negatively associated with leaf-production combinations (tall stature, low branching and delayed flowering). The top 10 performers in each market showed minimal overlap (Figure 9), with the highest-ranking seed market individual (Mo-TW-7, index = 49.1) and highest-ranking leaf market individual (Mo-TH-61, index = 36.3) representing distinct germplasm ideotypes.
Market-specific selection indices and trait contributions. (A) Individual seed-market vs. leaf-market index values (n = 133); dashed lines mark the 70th-percentile thresholds distinguishing top-performing seed specialists (blue) and leaf specialists (green) from lower performers (gray), with the top five individuals per market labeled. No individual exceeded both thresholds. (B, C) Weighted trait contributions to the index for the top 10 (B) seed-market and (C) leaf-market individuals, ordered by index rank; bars above and below zero show positive and negative contributions, respectively.
Figure 9 Long description
Panel A plots each of 133 individuals by seed-market index on the horizontal axis and leaf-market index on the vertical axis, with dashed lines at the 70th-percentile thresholds. Points are colored blue for seed-market specialists, green for leaf-market specialists, and gray for lower performers, with the top five individuals per market labeled; the points form a clear downward diagonal and the upper-right region, which would hold individuals exceeding both thresholds, is empty, showing no individual excelled in both markets. Panels B and C are stacked bar charts of the top ten individuals in the seed and leaf markets respectively, ordered by index rank, with colored segments for each trait extending above zero for positive and below zero for negative contributions. In panel B, fruit count and early flowering form the largest positive segments while tree height contributes small negative segments. In panel C, tree height and coppice capacity form the largest positive segments while branching forms the largest negative segment.
Decomposition of index values for top performers revealed the trait contributions driving performance in each market. For top performing seed market individuals, fruit count provided the largest positive contributions (mean = 25.4 index units), followed by early flowering (mean = 16.2 units) and coppice capacity (mean = 8.7 units), while tree height imposed modest penalties (mean = −3.4 units; Figure 9). For top performing leaf market individuals, coppice capacity (mean = 12.1 units) and tree height (mean = 8.9 units) drove performance, with branching imposing the largest penalties (mean = −11.3 units) due to the negative weighting for leaf production ideotypes (Figure 9). These contrasting profiles confirm that the indices identified germplasm aligned with market-specific trait combinations, demonstrating that the collection contains distinct, market-defining germplasm pools.
Discussion
This study evaluated 148 individuals from 32 families and 7 countries across two growing seasons, exceeding prior evaluations, which have typically assessed fewer than 50 accessions in a single season and prioritized regional germplasm collections (Mulugeta Reference Mulugeta2014; Palada et al. Reference Palada, Patricio, Ebert and Wu2014). Consistent with this increase in size and diversity, we observed ∼11-fold variation in tree height and 34-fold variation in fruit production, exceeding previously reported ranges. These results establish quantitative benchmarks for family-level genetic control across the collection. Rank stability between years differed by trait, with tree height and fruit count being the most stable (rg = 0.71–0.78), while branching and leaf count showed greater genotype-by-timepoint interaction (rg = 0.54–0.68). However, we treat these as preliminary indicators of rank stability rather than definitive selection parameters, as many repeated regrowth cycles are needed to confirm that rankings persist across successive harvests. This distinction matters for breeding decisions as traits with high rank stability can likely be selected reliably from single-season data, whereas branching and leaf count may require evaluation across more than one cycle before rankings are trusted. Coppice shoot production itself was far more consistent between 1- and 6-month regrowth (rg = 0.896), indicating that early regrowth vigour reliably predicts later performance, a trait central to moringa’s leaf and juvenile-fruit production systems but rarely quantified in previous evaluations. Timepoint, rather than family, explained most of the raw variance in tree height and leaf count, reflecting the large shift in absolute values that accompanies coppicing rather than a loss of genetic signal, since family effects for these traits were comparable to other traits once each evaluation was analysed separately (Supplementary Table S4).
Hundred-seed weight was a notable exception to this pattern of genetic control as despite ∼13-fold phenotypic variation (2.6–35 g), family effects were not significant and explained only 5.5% of within-evaluation variance, well below the 19–47% seen for other reproductive traits. This weak genetic signal may reflect within-family heterogeneity arising from moringa’s mixed mating system (Muluvi et al. Reference Muluvi, Sprent, Odee and Powell2004), strong maternal control as documented in other Brassicales (Li et al. Reference Li, Peng, Shi, Wang, Liu and Wang2015) or resource-allocation trade-offs (Smith et al. Reference Smith, Rao and Merchant2018), and argues for caution before using seed size alone as a selection target without multi-environment confirmation. Family-level estimates likely account for moringa’s mixed mating system, in which half-sibling families from open-pollinated mothers contain unknown proportions of selfed and outcrossed progeny (∼30% outcrossing; Muluvi et al. Reference Muluvi, Sprent, Odee and Powell2004). This heterogeneity likely contributes to the substantial residual variance observed for most traits and precludes precise narrow-sense heritability estimation from our half-sibling design without additional molecular data. Even so, the significant family effects detected for 9 of 12 traits indicate exploitable heritable variation: within-family heterogeneity rules out direct release of uniform seed-propagated cultivars, but leaves standing genetic variation for selection and for recovering superior individuals through clonal propagation or targeted crosses.
The selection indices identified 80 individuals specialized for either seed (n = 40) or leaf (n = 40) production, with none excelling at both. This complete separation, together with the underlying trait correlations (days to flower vs. seed weight; tree height vs. leaf production), indicates that seed and leaf ideotypes are built from largely non-overlapping trait combinations rather than opposite ends of a single continuum. Several architectural and phenological mechanisms plausibly underlie this pattern: heavy branching increases fruit-bearing surface area but reduces individual leaf size and impedes mechanical harvest, whereas the unbranched ‘telephone pole’ habit maximizes vertical leaf production at the cost of fruiting sites, and early flowering extends the fruit-development window for seed markets but shortens the vegetative period that favours leaf biomass. Because these trade-offs appear structural rather than incidental, simultaneous improvement for both markets within a single population is unlikely; the specialized germplasm identified here, led by Mo-TW-7 for seed and Mo-TH-61 for leaf traits, instead offers well-characterized starting material for each market. For breeding programmes, this implies that parental selection should be made within a target market from the outset, rather than screening a single population for broadly superior individuals.
Building on the genetic parameters and market-specific germplasm identified here, several extensions would most directly advance moringa breeding. Full-sibling families or clonally replicated trials would resolve the within-family confounding of our half-sibling design and enable precise heritability estimation. Multi-environment trials that repeat the coppice cycle across sites and seasons would test whether the rank-stability estimates reported here, currently based on a single year-1-to-regrowth comparison, hold across locations and successive harvests. Crosses between top seed- and leaf-market performers could test whether rare recombinants or transgressive segregants (Rieseberg et al. Reference Rieseberg, Archer and Wayne1999) can partially break the trade-offs documented above, opening a path towards dual-purpose types. Finally, integrating biochemical characterization (protein, mineral, secondary-metabolite profiles and mature seed/oil composition) with the morphological data presented here would extend this framework to the supplement and oil markets, which were outside the scope of the present study. Because all accessions are public domain, this characterized germplasm, including the market-specific ideotypes identified (unbranched, tall, late-flowering for leaf production; branched, moderate-height and early-flowering for seed production), is immediately available to programmes around the world for prioritizing accessions for further evaluation and distribution.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S1479262126100835.
Acknowledgements
We would like to acknowledge the Cellular Molecular Biology Research Unit (CMBRU) non-competitive funds awarded to Dr Nathan Fumia via the USDA-ARS Daniel K. Inouye U.S. Pacific Basin Agricultural Research Center located in Hilo, Hawaii and two programmes offered through Hawaii Agriculture Research Center (HARC) for their assistance in characterizing the collection: (1) Seeds 4 Tomorrow K-12 education programme, specifically Kiara Baba and Chase Botill; and (2) and Army Career Skills Programme, specifically Jason Grams. We would like to acknowledge Lynda Kuo from WorldVeg for assisting in the selection and preparation of WorldVeg M. oleifera accessions. We would also like to acknowledge Dr Anna McCormick of the University of Hawaiʻi at Mānoa and Dr Corey Broeckling of Colorado State University for comments on early versions of this manuscript.
Author contributions
N.F.: Conceptualization, Data curation, Formal analysis, Funding acquisition, Methodology, Writing – original draft, and Writing – review & editing. M.D.: Data curation, Formal analysis, and Writing – review & editing. M.D.K.: Conceptualization, Methodology, and Writing – original draft, review & editing. M.A.G.: Conceptualization, Methodology, and Writing – review & editing. M.v.Z.: Conceptualization, Writing – original draft, and Writing – review & editing. T.R.: Conceptualization and Writing – review & editing.
Data Availability Statement
Data and scripts are available at https://github.com/Nfumia.
Competing interests
The authors declare that there is no conflict of interest regarding the publication of this article.
Open access
