Introduction

At least 250,000 plant species have been identified as edible. Of these, 7000 have been consumed by humans but in 1995 only twelve crops and five animals made up 75% of the world’s food. Moreover, at this time 60% of global calories were provided by just three cereals, rice (Oryza sativa), maize (Zea mays) and wheat (Triticum aestivum)1. This bottleneck in agricultural diversity led to interest in underutilised, neglected, or so-called orphan crops2 however, there is debate about the use of such terms3 and now more inclusive and positive terms such as “opportunity” and “indigenous” crops are increasingly being used4,5. Greater use of such species could reduce the demand to adjust conditions to suit the current limited portfolio of major crops and instead allow crops better adapted to local conditions to be deployed6. Thus, increased use of locally adapted species could increase resilience of supply chains and so contribute to improving the lives of the ~10% of the global population who are undernourished7. Better adoption of these species could also help agriculture cope with increasingly unpredictable extreme weather and greater climatic uncertainty that are applying mounting pressure on the global food system8,9. Neglected and opportunity crops are defined as those that have been locally cultivated from sites at which the crop originates, or have become naturalised but been neglected by the global agricultural community10,11. Orphan crops are often defined as being rich in nutrient content, are climate resilient, locally available, and economically viable12,13,14. Orphan crops have already been adopted, with examples including quinoa (Chenopodium quinoa)15 and India’s targeting breeding programme of pearl millet (Pennisetum glaucum)16.

Gynandropsis gynandra (2n = 34) of the Cleomaceae family is a relative of Arabidopsis and Brassica crops in its sister family Brassicaceae. It has many common names including Shona Cabbage, Spider Plant, and Cat’s Whiskers. It is grown as an indigenous vegetable crop in parts of Sub-Saharan Africa and Asia and matches the four criteria outlined above17. It has recently been recognised as being of high relevance to achieve food security in Africa due to its nutritional content and modelled ability to withstand climate change in Sub-Saharan Africa18. Gynandropsis gynandra has been identified as a strong candidate for improvement due to its high vitamin C (ascorbic acid) and β-carotene content along with high levels of phosphorus, potassium, calcium, iron, zinc, phenols and flavonoids compared with commercial cultivars of Brassica oleracea (var. capitata cv. Drumhead) and Beta vulgaris (L. cv. Fordhook Giant)19. Its seeds are high in polyunsaturated fatty acids and are cultivated in Asia for oil20. The active ingredient of PIXALIA (Laboratoires Expanscience) is extracted from G. gynandra leaves for acne treatment, and G. gynandra has been identified as being of medicinal interest with leaves high in flavonoids such as quercetin and kaempferol21 that are thought to have antithrombogenic properties22. Gynandropsis gynandra extracts have also been reported to contain moderate antifungal and antibacterial activities23 and high glucosinolate content, which provides resistance to pests21. Lastly, its seeds are well suited for animal feed20.

In addition to its practical applications, G. gynandra is a model organism for the study of C4 photosynthesis. As such it has been used extensively as a system through which genetic determinants of C4 gene expression have been investigated24,25,26,27,28. Furthermore, a recently published genome sequence for G. gynandra confirmed its genome size of 740 Mb, and highlighted its utility in studying genome evolution29.

Gynandropsis gynandra is grown across a wide geographic range that includes all inhabited continents20, and these divergent environments and populations have resulted in a diverse gene pool. For example, resequencing data for fifty-three accessions identified three genetically distinct groups derived from West Africa, Asia and East/Southern Africa, respectively30. These accessions exhibited natural variation in multiple economically important traits, including plant height, flowering time, and vitamin content31. Local cultivation is a major part of G. gynandra’s natural history, as it has been traditionally used by farmers who have independently selected for desirable traits such as stem and leaf colour, plant height, and flowering time32. Geographically diverse lines of G. gynandra can hybridize, but it is also self-compatible17. There is now a rich germplasm collection, with 471 accessions stocked by the World Vegetable Center (https://avrdc.org/). These factors make G. gynandra an excellent candidate for pre-breeding methods, such as identifying Quantitative Trait Loci (QTL) for use in Marker Assisted Selection (MAS).

Crossing highly divergent lines enables segregating mapping populations from which QTL can be identified, providing insight into trait inheritance and targets for marker assisted selection to enhance germplasm33. To initiate this approach in G. gynandra we generated two F2 mapping populations in 2018 and 2019 from parental lines of Malaysian and Malawian origin. These were assessed for agronomic traits (e.g. plant height, leaf area, flowering time), nutritional compounds (carotenoids and tocopherols), and leaf anatomical features relevant to C4 photosynthesis.

Despite the growing recognition of G. gynandra’s value, no published studies to date have reported genetic linkage or QTL mapping in this species. This study therefore represents the first effort to dissect the genetic basis of trait variation in G. gynandra through bi-parental QTL mapping. We hypothesised that natural genetic variation between the founder lines would enable identification of heritable loci controlling key traits, including those underpinning the C4 pathway.

Although photosynthesis has not been a major focus for crop improvement34, it has been noted that selecting solely for yield can overlook potential avenues for crop improvement such as photosynthetic enhancement35. Since the founders of Wag19 showed differences in traits underpinning the efficient C4 photosynthesis pathway36, we assessed this population for variation in vein density, which is crucial for this pathway37. We anticipate that this work provides foundational tools for future marker-assisted breeding and will contribute to the development of improved G. gynandra germplasm.

Results

Phenotypic and heritability assessment for Wag18

Founders of the Wag18 population demonstrated statistically significant differences for all phenotypes assessed (\(p < 0.05\), Supplementary Table S1). With the exception of alpha-tocopherol, all other traits had higher values in Malawi-02 than Malaysia-03 (Fig. 1). All traits displayed transgressive segregation compared with the founders (Fig. 1) and broad-sense heritability (\({H}^{2}\)) was higher than 0.5 for all traits (Supplementary Table S2), demonstrative of the fact they were grown under tightly controlled environmental conditions. Flowering time, plant height and leaf area correlated positively with each other, and the content of all three carotenoids were highly correlated (Fig. 2).

Fig. 1: Transgressive segregation for all traits in the Wag18 F2 mapping population.
figure 1

Blue and red dotted lines indicate means for the founders, Malaysia-03 and Malawi-02, respectively. The dashed black line represents is the mean of the F2 population. A Violaxanthin content; B Lutein content; C Beta-carotene content; D Total carotenoid content; E Alpha-tocopherol content; F Flowering time; G Plant height; H Leaf area.

Full size image
Fig. 2: Correlation analysis for agronomic and nutrition traits in Wag18.
figure 2

Correlation coefficients were derived by Pearson correlation analysis and are presented in each box. Colour intensity indicates the strength of correlation, with deeper red representing a stronger positive relationship. The diagonal is the label for each phenotype. BSCW Bundle Sheath Cell Width. Significance is shown as *** \({\boldsymbol{p}}{\boldsymbol{ < }}{\boldsymbol{0}}{\boldsymbol{.}}{\boldsymbol{001}}\); ** \({\boldsymbol{p}}{\boldsymbol{ < }}{\boldsymbol{0}}{\boldsymbol{.}}{\boldsymbol{01}}\); *\({\boldsymbol{p}}{\boldsymbol{ < }}{\boldsymbol{0}}{\boldsymbol{.}}{\boldsymbol{05}}\).

Full size image

Four unique QTL were identified in Wag18 across all traits

The Wag18 linkage map consisted of 279 markers covering 984.87 cM with an average chromosome length of 82.07 cM (Fig. 3a, b; Supplementary Table S3).

Fig. 3: Wag18 genetic map.
figure 3

A Heatmap displaying recombination frequencies and LOD scores between each marker in the upper and lower triangles respectively; with yellow indicating a lower recombination frequency/higher LOD score, hence suggesting marker-pairs are linked. B Final genetic map.

Full size image

In Wag18, seven putative QTL (pQTL) were initially identified across all traits (significance less than 0.05 based on 1000 permutations, Supplementary Table S4). However, two pQTL for violaxanthin content, both on chromosome 1 (nearest markers at 5.95 Mb and 13.37 Mb) showed opposing additive effects despite similar genotype trends; individuals with the AB genotype exhibited significantly higher violaxanthin content than the AA (Malaysia-03) genotype at both loci (\(p=0.033{;t}=-2.15{;df}=132.55\) and \(p < 0.001{;t}=-3.43{;df}=121.43\) respectively). The signal at pVio2q was driven by a small number of Malawi-02 genotypes (BB; \(n=20\)) suggesting a spurious effect. Neither locus was retained in the final model following multiple QTL mapping, or single-scan validation and was thus excluded from further analysis.

Conversely, overlapping pQTL for plant height and leaf area showed consistent positive additive effects resulting from the Malawi-02 allele and mapped to a shared region on chromosome 1 around 89 cM (Table 1). Given the strong phenotypic correlation between these traits (Fig. 3), these signals were consolidated into a single robust QTL affecting overall plant size, hereafter referred to as Siz1q.

Table 1 Overall summary of QTL identified in the Wag18 population
Full size table

To summarise, extensive variation and high heritabilites were recorded for Wag18, and this enabled the mapping of four unique QTL associated with six agriculturally important traits (Table 1). Siz1q explained 25% and 18% of the phenotypic variance observed for plant height and leaf area respectively; Lut1q explained 10% of the lutein content variance; Atc1q contributed to 33% of the variance seen for alpha-tocopherol content; Flt1q explained 19% of the observed flowering time.

Phenotypic relationships in the Wag19 population

When the Wag19 founders were assessed, Malawi-01 had higher values than Malaysia-01 for carotenoid content, flowering time, plant height and leaf area, and a lower alpha-tocopherol content (Fig. 4), but the difference observed for all carotenoids and plant height was not statistically significant (Supplementary Table S5). For traits measured in Wag19 that were not assessed in Wag18, compared with Malaysia-01, vein density, trichome density, and stem colour was significantly higher in the Malawi-01 parent, while bundle sheath strand width and cell length were lower (Supplementary Table S5). Traits in Wag19 displayed transgressive segregation, except bundle sheath strand width and vein density, for which Malawi-02 demonstrated extremely low and high values respectively (Fig. 4).

Fig. 4: Phenotypic distributions in the Wag19 F2 mapping population.
figure 4

The blue, red and pink dotted lines indicate the means for Malaysia-01, Malawi-02, and the F1s respectively. The dashed black line represents the mean of the F2 mapping population. A Violaxanthin content; B Lutein content; C Beta-carotene content; D Total carotenoids content; E Alpha-tocopherol content; F Vein density; G Bundle Sheath Strand Width (BSSW); H Bundle Sheath Cell Length (BSCL); I Flowering time; J Plant height; K Leaf area; L Stem colour; M Trichome density.

Full size image

Heritabilities were generally lower in the Wag19 population compared to Wag18, however only plant height (0.49) and bundle sheath cell length (0.19) had heritability of less than 0.5 (Supplementary Table S2). Flowering time correlated with leaf area in Wag19, but this was not the case for plant height (Fig. 5). Rather, in Wag19, plant height was also positively correlated with leaf area. Analysis of variance found that flowering time differed between stem colour categories (\(p < 0.001,F=10.07,{df}=178\)), with stems that lacked anthocyanin flowering earlier. As in Wag18, content of the carotenoids were highly correlated, however unlike in Wag18, a significant correlation was identified between alpha-tocopherol and carotenoid content (Fig. 4). Traits related to C4 photosynthesis such as vein density, bundle sheath strand width, and bundle sheath cell length were strongly correlated, with the former having a negative relationship with the latter two traits. Interestingly, bundle sheath strand width was slightly, albeit significantly negatively correlated with carotenoid content. Plant height was positively correlated with bundle sheath size but negatively with vein density. A Chi-squared test revealed no significant relationship between stem colour and stem trichome density.

Fig. 5: Correlation analysis for agronomic, nutrition and C4 traits in the Wag19 population.
figure 5

Numbers indicate correlation coefficients derived by Pearson correlation analysis. Colour intensity indicates the strength of correlation, with deeper red representing a stronger positive relationship. The diagonal is the label for each phenotype. BSSW Bundle Sheath Strand Width, BSCL Bundle Sheath Cell Length (measured as length of 6 cells). Significance is shown as ***\({\boldsymbol{p}}{\boldsymbol{ < }}{\boldsymbol{0}}{\boldsymbol{.}}{\boldsymbol{001}}\); ** \({\boldsymbol{p}}{\boldsymbol{ < }}{\boldsymbol{0}}{\boldsymbol{.}}{\boldsymbol{01}}\); *\({\boldsymbol{p}}{\boldsymbol{ < }}{\boldsymbol{0}}{\boldsymbol{.}}{\boldsymbol{05}}\).

Full size image

Increased marker density in Wag19 captured more cross-over events compared with Wag18

The Wag19 linkage map consisted of 920 markers covering 1645.02 cM with an average chromosome length of 126.54 cM (Fig. 6a, b; Supplementary Table S3. While the smallest four chromosomes, 3, 15, 16, and 17 (Hoang et al.29) remained unmapped, the improved marker availability in Wag19, due to a less stringent selection process compared to Wag18 meant that chromosome 14 was captured in Wag19.

Fig. 6: Wag19 genetic map.
figure 6

A Heatmap displaying recombination frequencies and LOD scores between each marker in the upper and lower triangles respectively; with yellow indicating a lower recombination frequency/higher LOD score, hence suggesting marker-pairs are linked. B Final genetic map.

Full size image

Eleven unique QTL were initially identified in the Wag19 population

In Wag19, a total of fourteen pQTL were identified (significance less than 0.05 based on 1000 permutations, Supplementary Table S6). Two pQTL were found for vein density, one on chromosome 4 at position 122.00 cM (\({LOD}=3.88\)) and one on chromosome 14 at 6.00 cM (\({LOD}=4.30\); Supplementary Fig. S1a). Looking at the nearest markers for these two traits (4.59646348 and 14.00360911) it was clear the increase in vein density was, as expected, driven by the Malawi-01 (B) allele (Supplementary Fig. S1b, c), however 14.00360911 had just a single individual that was homozygous for the B allele. When this line was excluded only the loci on chromosome 4 remained significant in linear regression (\(p < 0.001\)). We therefore retained only this locus as a robust QTL and hereafter refer to it as Vd1q.

Two additional pQTL for plant height and leaf area were found on chromosome 1 at approximately the same location (84 cM). Both were positively influenced by the Malawi-01 (B) allele. As these traits were significantly correlated (\(r=0.33{;p} < 0.001\), Fig. 5), we consider these loci as a single multi-trait QTL affecting plant size, Siz2q.

For the discrete traits, stem trichome density and stem colour, we identified two pQTL each: pStt1q and pStt2q for trichomes, and pStc1q and pStc2q for colour (Supplementary Table S7). Ordinal regression confirmed that both trichome-associated loci had significant effects (\({AIC}=274.65\); Supplementary Table S7). For stem colour, the optimal model included both loci (\({AIC}=160.91\)), but only pStc1q was significant. A reduced model including only pStc1q had very similar AIC (\({AIC}=161.57\)), therefore we retained only these loci as a QTL for stem colour.

Finally, to ensure statistical robustness, all pQTL associated with stem trichome density and colour were cross-validated using non-parametric single-scan models. These yielded similar profiles for trichome density (Supplementary Fig. S2a), while the non-parametric scan for stem colour showed a slightly lower LOD peak (Supplementary Fig. S2b). Final LOD values reported for these traits are based on the non-parametric models.

In summary, eleven unique QTL affecting eight traits were identified in Wag19 (Table 2). One QTL was found to be significant for violaxanthin content (Vio1q), alpha-tocopherol content (Atc2q), vein density (Vd1q), flowering time (Flt2q), plant height (Ht1q), and stem colour (Stc1q); two QTL were significantly associated with leaf area (La1q and La2q) and stem trichomes (Stt1q and Stt2q); and one QTL was found to effect overall plant size, i.e., having the same effect on plant height and leaf area (Siz2q). Overall, 12% of the variance observed for violaxanthin content, 14% for alpha-tocopherol content, 8% for vein density, 24% for flowering time, 25% for plant height, 29% for leaf area, 21% for stem trichomes and 52% for stem colour was explained by these QTL (Table 2). For traits assessed in both populations, four QTL were mapped in Wag18 (Fig. 7a) and seven were mapped in Wag19 (Fig. 7b).

Fig. 7: QTL locations in both populations.
figure 7

Location of QTL in the (A) Wag18 and (B) Wag19 populations. QTL associated with each trait are colour-coded. Multi-trait QTL are represented in grey boxes. Confidence intervals were identified using the R/qtlTools package67.

Full size image
Table 2 Overall summary of QTL identified in the Wag19 population
Full size table

Agriculturally important QTL found in both populations

In Wag18, the QTL Siz1q was located on chromosome 1 with a peak around 89.9 cM and in Wag19 the QTL Siz2q was located on chromosome 1 with a peak around 85.9 cM (Fig. 7). The nearest marker to both these QTLs were 1.13366757 and 1.17199903 (located at 113.37 Mb and 117.20 Mb) in Wag18 and Wag19 respectively and in both cases, the “African” allele (B) coming from Malawi-02 in Wag18 and Malawi-01 in Wag19, was responsible for the increase in plant size (Fig. 8a, b). The confidence limits based on the overlapping regions of the traits they were associated with (Fig. 7) for these QTL was between 87.7 cM and 90.7 cM for Siz1q and between 79.7 cM and 101.0 cM for Siz2q. Based on the nearest markers, the corresponding physical range for these QTL was 11.74–20.24 Mb and 11.15–61.79 Mb. While this region contains many genes, the overlapping QTL region means these effects may be under the control of the same gene or group of genes38, and hereafter we will refer to this as single QTL, namely Siz1q.

Fig. 8: Phenotypes for agronomically important traits plotted against genotypes at markers nearest to associated QTL.
figure 8

A In Wag18, plant height (indicative of plant size) increased at the marker nearest to Siz1q when plants have the “African” allele; B the same trend was observed for Wag19. C Flowering time increased when plants had the B allele in the marker closest to Flt1q in Wag18 and (D) Wag19. E All individuals homozygous for the Malaysia-01 (A) allele had no visible anthocyanin. Numbers in parentheses are the number of individuals with each genotype. Confidence intervals for average phenotypes within each genotype group are shown as blue for AA, purple for AB, and red for BB. Marker names describe physical location; “Chromosome:bp”.

Full size image

Further, the QTL Flt1q (identified in Wag18) was located on chromosome 6 with a peak around 45.0 cM and Flt2q (identified in Wag19) was on chromosome 6 with a peak around 77.0 cM (Fig. 7). Once again, the African allele was responsible for the later flowering time (Fig. 8c, d). The confidence limits for these QTL were 39.2–54.5 cM and 54.1–86.0 cM corresponding to a physical range of 27.36–44.29 Mb and 36.68–49.41 Mb for Flt1q and Flt2q respectively. This overlap suggests the same underlying gene or group of genes may be responsible for this effect, from henceforth we shall refer to this as a single QTL, namely as Flt1q. Additionally, closer inspection of the nearest marker for the QTL for stem colour (Stc1q), found that just 2 out of 48 AA (Malaysia-01 genotype) individuals had a score of 2 (indicative of high anthocyanin) and no individuals with even one copy of the B (Malawi-01) allele were absent anthocyanin (a score of 0; Fig. 8e). Given its proximity to Flt1q (Fig. 7b), and the observation that stems with reduced anthocyanin flowered significantly earlier (\(p < 0.001,F=10.07,{df}=178\)), these loci may be genetically linked or share a common regulatory basis.

Discussion

Our identification of significant QTL in G. gynandra provides insight into the genetic basis of key agronomic and physiological traits, and paves the way for pre-breeding programmes. These findings not only contribute to the widespread adoption of G. gynandra as a food crop but also offer broader relevance for breeding nutrient-rich, climate resilient crops. Given that G. gynandra is currently harvested as a wild plant with minimal domestication39, we anticipate these QTL will contribute to the development of improved germplasm, enhancing its potential for breeding efforts.

Two QTL were identified, Siz1q and Flt1q, that mapped to the same regions in both populations and could be the result of the same underlying gene(s; Lynch and Walsh, 1998). These controlled plant size and flowering time (Fig. 7). Interestingly, their effects were relatively stable in both populations, in the Wag18 population, Siz1q explained 25% and 18% of the variance associated with plant height and leaf area respectively, which was similar to the variance Siz1q explained by in Wag19 for these two traits, 17% and 10%, especially when considering the lower heritabilities associated with the Wag19 population (Supplementary Table S2). Further, Flt1q explained 19% and 24% of the variance associated with flowering time in Wag18 and Wag19 respectively. The recurrence of QTL in both populations suggests that key traits like plant size and flowering time are under conserved genetic control. This consistency across genetic backgrounds and environments is encouraging for breeding programs, as it indicates the presence of stable, heritable targets.

Considering that G. gynandra is a commercially viable crop with nutritional and medical applications17,19,20,21,22, identifying QTL associated with flowering time and plant size, two traits of value to plant breeders, in multiple environments and populations, and given their physical locations are known (Tables 1 and 2), means markers could be generated so desirable alleles could be incorporated into breeding germplasm via marker assisted selection40. To our knowledge, these are the first example of stable targets for marker assisted selection identified in G. gynandra and could be of great benefit in the next stages of this plant’s transcendence from underutilised to becoming a widely cultivated, climate resilient crop.

Further, across both populations, we identified four QTL associated with the content of the carotenoids, lutein and violaxanthin, and the vitamin E, alpha-tocopherol. The QTL for stem colour (Stc1q), also provides a promising target, explaining over half of the observed variance. Its close proximity with the QTL for flowering time (Flt1q), suggests possible pleiotropy or linkage, and the early visibility of anthocyanin expression may offer a convenient phenotypic marker for breeders.

The vitamins we assessed are involved in protection of eyes from diseases, age-related degeneration41,42, mitigating UV-induced skin damage43 and are important for the immune system44. There is a need for a targeted crop improvement programme for G. gynandra17,39 and these identified QTL could be useful in marker assisted selection for vitamin content. Furthermore, a previously reported negative correlation between carotenoids and tocopherols31 was not identified in the Wag18 population (Fig. 2), and was observed only weakly in the Wag19 population (Fig. 5) suggesting these two traits, can in fact be bred for simultaneously. A brief assessment of the F1 lines in the Wag19 population also found strong heterosis for plant height (Fig. 4j). This has not been reported previously for G. gynandra and indicates its suitability for hybrid breeding45, which is of huge benefit for generating advanced breeding programmes16.

Cleomaceae is the closest related family to the Brassicaceae and therefore contains the most closely related C4 species to the model C3 plant Arabidopsis thaliana with which G. gynandra shares high similarity in gene sequence46. This makes it an excellent model organism for the study of C4 photosynthesis and it has been used as a system through which genetic determinants of C4 enzymes have been investigated24,25,26,27,28. C4 photosynthesis has evolved independently over 65 times47 and is more energy efficient under warm, dry conditions compared to the C3 state. Mathematical modelling suggests that morphological changes such as increased vein density and larger bundle sheath cell size were crucial in the early evolution of the C4 syndrome37.

Kranz traits, vein density and bundle sheath size showed extensive variation (Fig. 4f–h) with one QTL, Vd1q being mapped to chromosome 4 (Fig. 7b). Further, initial QTL scans found an additional pQTL for vein density (pVd2q; Supplementary Table S6). Although we deprioritised pVd2q as an outlier appeared to skew the results, rendering this pQTL association statistically insignificant after its removal, it is possible that this represents a genuine association. This possibility is supported by the additive effect observed in the heterozygote, and the large vein densities found in both ‘BB’ lines assessed (the Malawi-01 founder line and the F2 line in question; Fig. 1f and Supplementary Fig. S1c). A population with a greater prevalence of these alleles would be needed to demonstrate this statistically.

Furthermore, it should be noted that in G. gynandra and other C4 plants, carotenoids and tocopherols are crucial for protecting the photosynthetic apparatus from oxidative stress, ensuring efficient photosynthesis under intense light conditions and other abiotic stressors such as drought and high temperature48,49, therefore the aforementioned QTL linked to these traits could also be useful for identifying lines suitable for growing under more extreme conditions.

To our knowledge, this is only the first example of linkage mapping being employed to identify QTL associated with C4 characteristics in any dicotyledon. In maize (Zea mays) a bi-parental population identified QTL for stomatal size and density50, and so together these data provide strong support for the notion that quantitative genetics could be used to study C4 traits51. Given the region containing Vd1q contains 318 genes29, we have refrained from reporting candidate genes due to the ambiguity in searching for GO terms amongst a relatively large number of genes. Instead, we aim here to prove that reverse genetics approaches such as QTL mapping is a viable option for the study of C4 photosynthesis. Furthermore, alongside a wild diversity panel, there is currently a Multi-Parental Advanced Generation Inter-Cross (MAGIC) population of G. gynandra being developed with the aim of mapping traits related to C4 photosynthesis. Such association mapping methods are prone to false positives52 so populations derived from simple bi-parental crosses, such as that assessed here, that are equally related and therefore absent population structure, can add power to and complement such studies that enable higher resolution and the fine mapping of causal genes40,51. We demonstrate here the effectiveness of mapping QTL in G. gynandra and, given its rapid generation time, phenotypic and genotypic diversity31,36, combined with the power of bi-parental populations40, it is an excellent resource for further mapping of additional traits. Furthermore, many F1 hybrids already exist that can be used as a resource for future study of developmental, physiological or agricultural traits in G. gynandra.

Methods

Production of plant material, growth and phenotyping

Two populations of G. gynandra were developed at Wageningen University & Research in 2018 and 2019, hereafter referred to as Wag18 and Wag19. For Wag18, the female and male parents of the F1 were derived from repeated self-fertilization of an early-flowering, short accession from Malaysia (TOT7200; Malaysia-03; MAY-03) and a late-flowering, tall accession from Malawi (TOT8917; Malawi-02; MAL-02; Sogbohossou et al.17). For Wag19, the female and male parents were derived from repeated self-fertilization of an early-flowering, short accession from Malaysia (TOT7199; Malaysia-01; MAY-01) and a late-flowering, short accession from Malawi (TOT8918; Malawi_01; MAL-01)17, these lines also have wide variation for C4 photosynthetic traits36. Wag18 consisted of 213 F2 lines and Wag19 187 F2 lines. As is standard for F2 populations, each individual was genetically unique. Consequently, phenotypic replication at the level of individual genotypes was not possible. To increase confidence in QTL discovery despite this limitation we used two independent F2 populations derived from divergent founder lines.

The populations were grown in the greenhouse of Wageningen University & Research, the Netherlands from March–June 2018 and April–July 2019. For Wag18 seven replicates of each parent were randomised, however, one replicate of Malaysia-03 was not healthy at the time of phenotyping and was therefore discarded. For Wag19 five replicates of each parent were randomised and two F1 lines were included. Growing conditions were tightly controlled with plants grown under irrigated conditions with temperatures maintained at 24 °C in the day and 20 °C during night, and with artificial lights maintaining a minimum light intensity of 300µmol m−2 s−1 and a photoperiod of 16 hr days and 8 hr nights. Sampling for DNA extraction and subsequent processing was carried out as described previously30.

For both Wag18 and Wag19, plant height was measured after 10 weeks and the leaf area of three fully mature leaves per plant was assessed using image analysis in ImageJ53. Flowering time was measured as the number of days from sowing to flowering. In Wag19, stem trichome density and stem colour were assessed as ordinal variables: 2 = dense trichomes or red colour; 1 = moderate trichome density and red/green colour; and 0 being no trichomes and green colour. Red colouration was indicative of anthocyanin content. Extraction and analysis of carotenoids and tocopherols as well as quantification was performed as described by31. In Wag18, seven plants died prior to vitamin-content assessment, along with an eighth plant that died before flowering. As a result, these plants were not assessed for plant height, leaf area and flowering time. Further, due to poor image quality or lack of replication, accurate leaf area could not be assessed for an additional nine plants. In Wag19, in total nineteen plants were not assessed for vitamin content due to sample contamination. Five plants died before being assessed for plant height, leaf area, stem colour, and stem trichome density. A further three died prior to flowering, and three plants could not be accurately assessed for leaf area. Lastly, six replicates of each parental line were assessed for vitamin content.

C4 leaf anatomical traits were assessed in Wag19 as follows. Plants were harvested over a three-day period 4 weeks after germination. The most recent fully expanded leaf was selected for sampling. Tissue was harvested from the central leaflet of each leaf and placed immediately in a plastic cuvette submerged in a 3:1 100% (v/v) ethanol:acetic acid fixative solution before treatment with 70% (v/v) ethanol solution overnight at 37 °C being refreshed one hour into treatment. Samples were cleared using 5% (w/v) NaOH solution for two hours at 37 °C before being washed with, and replaced in, 70% (v/v) ethanol until preparation for imaging. Immediately prior to imaging, samples were immersed in Lugol’s solution (I3K), and washed with water to highlight vein tissue as described by Simpson et al.54. Samples were mounted with water and imaged on an Olympus BX41 light microscope with a mounted Micropuplisher 3.3 RTV camera (Q Imaging). Images were captured with Q-Capture Pro 7 software. Bundle sheath cells were imaged at ×200 magnification and assessed using the line measurement tool in ImageJ53. Bundle sheath strand width was measured as the length of the bundle sheath strand perpendicular to the vein; six measurements were taken per field of view. Bundle sheath cell length was quantified by measuring the length parallel to the vein of 6 bundle sheath cells and dividing this by 6 to get a single value per field of view. For both parameters, six field of views were measured per plant. Four and five F2 lines were not assessed for bundle sheath cell width and length respectively, in addition to one replicate of Malaysia-01 due to poor clarity of microscopy images. The Wag19 lines were also assessed for vein density, using six fields of view per plant at a magnification of ×100 using the Starch4Kranz pipeline54.

SNP calling

To call and select Single Nucleotide Polymorphisms (SNPs) in the two sets of founder lines for genotyping-by-sequencing in each population, raw reads generated for each founder30 were checked for quality and trimmed using Trimmomatic55. The reads were aligned to the reference genome using the BWA MEM algorithm56 and Samtools57. Duplicate reads were filtered and read groups added using GATK (version 4.0) Variant calling and filtering was performed with the HaplotypeCaller and the VariantFiltration tool of GATK58. In order to select SNPs to be used for the characterization of the F2 population, heterozygous SNPs and SNPs with no call were discarded. Non-polymorphic SNPs between both parents were removed using vcffilterjdk and JEXL expressions59. SNP clusters in 150 bp windows were also removed. Furthermore, in the Wag18 founders, we subsequently used a set of re-sequenced genomes of 48 accessions: 24 from East and Southern Africa and 24 from Asia30 to select SNPs that were present in at least 12 accessions from the same region of origin as the parents and, we identified SNPs that were present in genomic regions of interest for our targeted traits irrespective of whether they were unique to the parental lines or shared by other accessions from the same region. These additional filtering steps were not applied to the Wag19 founders. The final set consisted of 1309 SNPs and 9305 SNPs that were used for the genotyping-by-sequencing in the Wag18 and Wag19 population respectively.

Heritably estimates, linkage maps, and QTL mapping

Broad-sensed Heritability (\({H}^{2}\)) was measured according to Kearsey and Pooni60. The observed phenotypic variance (\({\sigma }_{p}^{2}\)) is the sum of the genetic variance (\({\sigma }_{g}^{2}\)) and environmental variance (\({\sigma }_{e}^{2}\)). Since the homozygotic parental lines are genetically identical, the observed variance within each line is equivalent to (\({\sigma }_{e}^{2}\)) which can be extracted from the residual mean squares of the combined parental data. \({\sigma }_{g}^{2}\) can therefore be measured with the following equation, where \({\sigma }_{p}^{2}\) corresponds to the phenotypic variance excluding the parental lines:

$${\sigma }_{g}^{2}={\sigma }_{p}^{2}-{\sigma }_{e}^{2}$$
(1)

This can then be used to calculate broad sensed heritability:

$${H}^{2}=\frac{{\sigma }_{g}^{2}}{{\sigma }_{p}^{2}}$$
(2)

Heritability estimates were carried out by comparing parental phenotypes to that of the final mapping population.

In both Wag18 and Wag19 populations, reads were aligned to the reference genome29 using Bowtie261 and Samtools57. Haplotype-based variant detection was carried out using Freebayes62 and filtering was undertaken with Bcftools63 before converting to R/qtl format using the R/utl package64. To produce the final SNP set, markers with a Minor Allele Frequency (MAF) less than 0.2 were removed along with any markers missing in more than 70% of individuals and ambiguous genotypes (i.e., being heterozygous in the parents or homozygous in the F1). From our initial sets of 1309 and 9305 SNPs, this resulted in final SNP sets of 521 and 2035 for the Wag18 and Wag19 populations respectively.

Linkage Maps were generated using the following pipeline. First, duplicate individuals were identified. In Wag18, three pairs of duplicate individuals were identified that shared more than 95% of their genotypes; each pair was next to each other in the experimental design suggesting a sampling error. Due to not knowing which individual was correct all 6 individuals were removed. No duplicates were identified in Wag19. Secondly, we removed one individual in the Wag18 population that was completely homozygous. We then constructed genetic maps as detailed by Broman and Sen65 using the package R/qtl66. Marker positions were based on their physical location29, with errors being revealed through investigating marker recombination frequency, and successive use of the ‘droponemarker’ function65. Local marker order was adjusted through rippling to minimize cross-over number by using a sliding window of 7 markers. Finally, individual genotyping errors were removed through study of double cross over rate and the ‘calc.errorlod’ function. Our final map for Wag18 consisted of 206 individuals, 279 markers, with 99.9% genotype rate, and an error rate of less than 0.001 based on loglikelihood scores. For Wag19, we attained the same error rate, 187 individuals and 920 markers, with a 99.7% genotype rate.

QTL were detected in both populations using the Multiple QTL Mapping (MQM) method described by Broman and Sen65. First, single scans were performed using the ‘scanone’ function. Second, two-dimensional scans were carried out using the ‘scantwo’ function which considers QTL of large effect meaning modestly sized QTL can be identified in addition to interactions between QTL. Last, the putative QTL identified by ‘scanone’ and ‘scantwo’ scans were used as the starting model and the function ‘stepwiseqtl’ used to identify the final QTL model for each trait. For all scans, Haley-Knott regression for interval mapping was implemented for QTL detection due to its rapidity and that our genotype datasets had very little missing data. Further, if single scans revealed significant QTL not identified through MQM, these were investigated for significance through linear regression. Significance thresholds were calculated based on 1000 permutations per trait per scan. QTL models were fitted from the output of ‘stepwiseqtl’ and the applying the ‘fitqtl’ function, so that Percentage of Variance Explained (PVE) for each QTL model could be calculated as65:

$${PVE}=1-{10}^{\left(\frac{-2\times {LOD}}{n}\right)}$$
(3)

Where LOD is the Logarithm Of the Odds score associated with the QTL or QTL model in question and n is the number of individuals phenotyped. Confidence intervals for each QTL were calculated as the 1.5 LOD drop intervals based on the final QTL model and ‘scantwo’ permutations using the ‘qtlStats’ function from R/qtlTools67. If the same QTL was identified for multiple traits, confidence limits were taken as being the overlapping region. Prior to mapping, phenotype data was transformed based on assessment of histograms in the mapping populations. Phenotype data were transformed when appropriate so that in the Wag18 population, leaf area and alpha-tocopherol-content underwent log-transformation, and in the Wag19 population, plant height was squared transformed, leaf area, and vein density were square-root transformed, flowering time was cube-root transformed, and alpha-tocopherol-content was log-transformed. To enable sufficient permutations to be generated for the non-parametric traits, stem trichome density and stem colour, normal models were selected while QTL mapping as it is accepted these still produce accurate results65, however single scans were compared to the non-parametric model to ensure QTL identified were correct. All QTL were identified under a significance threshold of 5% based on 1000 permutations derived from both the ‘scanone’ and ‘scantwo’ functions.

Statistical analysis

All statistics was carried out in R (version 4.1.3). Variation between parent lines was analysed using a two-sample Student’s t test in R Studio (V: 4.0.0). Data were checked for normality and equal variance in each group using Shapiro–Wilks test and Bartlett’s test respectively. If data were not normal in one or both groups, the Mann–Whitney U (MWU) test was used in place of the Student’s t test and variance was checked using Levene’s test. Results from Levene’s test determined comparisons made with the MWU test. If Levene’s test found equal variance, the MWU test for significance was based on median values. If Levene’s test found unequal variance between parental groups, the MWU test was based on variance. If data were normal in distribution but had unequal variance between groups, a Welch’s t test was used. For mapping populations, normality tests were determined from data distribution, and pairwise Pearson correlation analysis carried out using R/GGally package68. To identify relationships between binary or discrete traits (stem trichome density, and stem colour) and continuous traits, ANOVA was used, while a Chi-squared test was used to test for a relationship between stem trichome density and stem colour. Linear regression was used to determine environmental effects and multiple QTL mapping performed as described. Plots were generated using R/ggplot269, R/corrplot70, R/qtlTools67, and R/qtl66. Logistic regression was carried out using the R/MASS package71, and p-values derived from resulting t-distributions by applying the cumulative distribution function.