Identification of a new QTL associated to reduced quinolizidine alkaloid content in white lupin (Lupinus albus, L.) and development of ultra-low alkaloid recombinants by stacking with the pauper allele.

RESEARCH Open Access © The Author(s) 2025. Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit h t t p : / / c r e a t i v e c o m m o n s . o r g / l i c e n s e s / b y - n c - n d / 4 . 0 /. Patyi et al. BMC Plant Biology (2025) 25:945 https://doi.org/10.1186/s12870-025-06951-7 BMC Plant Biology *Correspondence: András Patyi [email protected] Full list of author information is available at the end of the article Abstract Background White lupin (Lupinus albus, L.) is a grain legume cultivated for its high agronomic and nutritional potential, despite the accumulation of bitter and potentially harmful to health quinolizidine alkaloids (QA) in the grain. Modern sweet (i.e. low alkaloid content) varieties exist, which are exploiting different recessive mutations responsible for the desired low QA chemotype. The most widely used QA-reducing determinant, pauper, has been recently identified enabling marker-assisted selection, but unstable QA content across growing seasons and environments remain a challenge in white lupin cultivation. Results Through Bulked Segregant Analysis of an F2 population, segregating for two different low QA conferring loci, we identified a novel QTL spanning a 1 Mbp region on chromosome 5, a novel source of sweetness apart from pauper. We present marker-trait associations for the new locus tagging low QA content in white lupin within this QTL, described in F2 generation and validated in F3. Together with genotyping of sweetness tagging pauper sweet individuals, we identified, in F3, 50 stacked allele recombinants where the low QA chemotype is further reduced. These individuals exhibit an exceptionally low total alkaloid content (22.8 ± 10.4 ppm), even when compared to genotypes known to carry the allele associated to the most drastic reduction in QAs, pauper (171.7 ± 18.5 ppm). Conclusion The discovery of this novel locus and the development of associated PACE markers, potentially applicable for marker-assisted selection together with pauper (especially after farther validation in a larger panel of accessions), can enhance the improvement of white lupin through the development of new varieties with very low Identification of a new QTL associated to reduced quinolizidine alkaloid content in white lupin (Lupinus albus, L.) and development of ultra-low alkaloid recombinants by stacking with the pauper allele AndrásPatyi1,2*, MiriamKamp3, ChristineArncken1, ElisaBiazzi4, MichałKsiążkiewicz2, Monika M.Messmer1, MichaelSchneider1, AldoTava4 and MariateresaLazzaro1 Page 2 of 14Patyi et al. BMC Plant Biology (2025) 25:945 Background White lupin (Lupinus albus, L.) is a grain legume of Mediterranean origin produced for its agronomic value and highly nutritious seeds. From the nutritional perspective, white lupin grain is appreciated for feed and food for its high protein (33–47% depending on genotype and location) and excellent fibre content [1]. Its desirable fatty acid composition ratio of omega-3 to omega-6, non-starch carbohydrate-, oligosaccharideand antioxidant-content makes the consumption of white lupin seed a great contribution to a healthy diet [2]. Conversely, white lupin seed contains quinolizidine alkaloids (QAs), toxic (primarily due to their ability to interfere with the nervous system) and bitter compounds which set a major challenge for food producers and processors. In the closelyrelated blue lupin (Lupinus angustifolius, L.), the major site of production of QAs is the aerial epidermis, later transported to seeds during fruit development [3]. Seeds of wild forms of white lupin contain up to 127,000 ppm (mg/kg dry weight) QAs which is reduced in bitter white lupin cultivars to 2,000–4,000 ppm [4]. Historically, the grains were soaked in running water to remove the toxic alkaloids before consumption. Improved methods exist for debittering, commonly using water as a solvent, however these processes remain water-use intensive, laborious and producing toxic waste [5], constraining the use and economic viability of white lupin in the value chain, for feed and, especially, food. Lupins contain a mixture of different alkaloids, not all of which are sufficiently known for their toxicity. Recently, Schreiber et al. [6] evaluated the toxicity of five common lupin alkaloids (sparteine, lupinine, lupanine, 13-hydroxylupanine, angustifoline) reporting that these QAs individually and as a mixture are neither cyto-, nor genotoxic in mammalian cells. Concerning their neurotoxicity, the effect of sparteine, presumably and stable alkaloid content based on stacked allele recombinants. This can help to increase the cultivation of this useful yet underutilised crop and its use for human nutrition. Keywords White lupin, Bulked Segregant analysis (BSA), Quantitative trait locus (QTL), Quinolizidine alkaloids (QA), PCR allelic competitive extension (PACE), Plant breeding Graphical abstract Page 3 of 14Patyi et al. BMC Plant Biology (2025) 25:945 the most toxic lupin alkaloid, has been studied best. The available risk assessment is therefore based on the knowledge of sparteine for all alkaloids present in lupins. Safety control is based on the total quinolizidine alkaloid content [7]. In Switzerland, for example, there are no legal limits for alkaloids in food or feed. The principle of self-regulation applies, which means that companies must ensure that only safe food is placed on the market [8]. The German Federal Institute for Risk Assessment (BfR) has established in 2017 the following guideline values to protect human and animal health: <200 ppm and < 500 ppm (dry weight) respectively for food and feed, relative to the final product intended for consumption [9]. Additionally, the import and export standards of the industry organisation Pulse Australia (Australia being the world’s largest lupin producer) set in 2001 a maximum alkaloid content of 200 ppm [7]. The European Commission has recently recommended monitoring of the following quinolizidine alkaloids in lupins and lupin-derived food: albine, anagyrin, angustifoline, lupanine, isolupanine, multiflorine, 13α-hydroxylupanine, lupinine and sparteine [10]. Modern sweet (i.e. bitterness lacking) varieties exist with the alkaloid content ranging between 100 and 800 ppm of seed dry weight [11]. Low QA cultivars were a major achievement of the formal breeding of white lupin, started in the early 20th century with the activity of Reinhold von Sengbusch [12]. He selected against common wild type traits, amongst many others for high QA content, describing the drastic decrease (to 200–500 ppm) in QA content due to the recessive allele pauper, present in many modern white lupin cultivars [13]. Apart from pauper, other loci exist (e.g. exiguus, nutricius, mitis, reductus) representing possible recessive mutations for low QA content, described to reduce QA content to ≤ 1000 ppm [11]. These mutations were described based on crossings between different white lupin individuals and observing the segregation pattern in the chemotype, thus not much is known about the genes involved or transcriptomic regulation of low QA content in white lupins, except for pauper. Recently, the mutation causing the pauper-type sweetness in white lupin was identified [14]. Reportedly the presence of a single SNP (Lalb_Chr18_12359687) confers a greatly reduced total QA content and a distinct QA profile. This mutation is located in the coding sequence of a gene encoding an acetyltransferase, and is responsible for changes in the early biosynthetic steps of QAs, resulting in the low QA chemotype. However, phenotypically sweet variants exist, which do not carry the pauper mutation, confirming non-pauper sweetness types, earlier described in the 20th century [11]. One example was the cultivar ‘Dieta’, in which the genetic determinant causing low QA levels is unknown [14]. Despite the magnitude of the effect of the gene associated with pauper, grains harvested from reportedly sweet pauper accessions have been observed to exceed the advised threshold for consumption (FiBL, unpublished data). Relatively high and unstable QA content in the harvested grain remains a threat for white lupin producers and processers, creating a demand for even lower QA varieties with stable sweetness across growing seasons regardless of the environmental context. This could be achieved through gene stacking (pyramiding)– combining independent loci conferring low QA content in white lupin, which could be further aided with the availability of molecular markers tagging these loci. The present study aimed to map an unknown source of sweetness in white lupin by Bulked Segregant Analysis (BSA) in an F2 population resulting from the crossing of the pauper sweet cultivar ‘Frieda’ and the cultivar ‘Dieta’ carrying the unknown sweetness determinant. Subsequently, through PACE marker development and validation in a diversity panel, we aimed to verify the effect of the individual loci conferring the low QA level phenotype and to assess their combined effect on QA concentration and profile in the F2 and the derived F3 population. Methods Plant material Crossing and F2 population Two commercial cultivars, ‘Frieda’ (♂, pauper sweetness) and ‘Dieta’ (♀, unknown sweetness), were crossed in 2021. Seeds were obtained from Delley Samen und Pflanzen AG (Delley-Portalban, Switzerland) for ‘Frieda’ and from Soya UK Ltd. (Southampton, United Kingdom) for ‘Dieta’. The five harvested seeds from the crossing were propagated in 2022 under protective nets to prevent cross-pollination and seeds of the F1 plants were harvested in a bulk. The F2 plants (own breeding material) were grown in 2023 in the FiBL breeding nursery located in Leibstadt, Switzerland (47°35’26.4"N 8°10’56.0"E), organized in a 10-row plot with 14 F1:2 seeds sown per row, grown under protective nets. Young leaf tissue (ca. 100mg) was collected and frozen from the 112 surviving individuals of the F2 population (Supplementary Table S1). F3 population Thirteen F2 plants, all heterozygous at the pauper locus were selected for the F3 population (own breeding material) and their seeds were grown in 2024 on a field in Full-Reuenthal, Switzerland (47°35’59.7”N 8°12’13.6”E), sown and processed in an identical manner as in previous years. In total, 329 individuals represented the F3 population (Supplementary Table S3). In addition, 10 individuals each of ‘Frieda’ and ‘Dieta’ were grown in the same field and harvested as single plants. Page 4 of 14Patyi et al. BMC Plant Biology (2025) 25:945 Validation panel A diverse panel of 34 individuals was created for the molecular marker validation, consisting of 1 to 7 representatives for 1 breeding line, 16 cultivars, 3 landraces and 1 wild type in addition to the 39 accessions of the white lupin pangenome [15] (Supplementary Table S5). Phenotyping The alkaloid content of F2 and F3 was estimated first on the field using Dragendorff paper test (Supplementary Table S1, Supplementary Table S3), a filter paper soaked in iodine/potassium iodide solution (Dragendorff solution), then dried at 40°C [16]. Individual QA content was estimated by transferring plant sap from the petioles onto the Dragendorff paper during onset of flowering (BBCH 63–65). A visible reaction of the Dragendorff paper turning a reddish-brown colour was described to occur for QA contents of ≥ 500 ppm [17]. The colour reaction was scored from 1 to 4, 1 meaning a purely white stain, the lowest alkaloid content. A score of 2 was assigned to stains, which were mostly white with the red encircling, score 3 described red encircled pinkish stains with a light centre and lastly, a score of 4 was given to samples with an intense red stain without discolouration in the centre (Fig.1). Additionally, bitterness in the F2 population was phenotyped by scoring the taste of one flower per plant in a range from 1 (absence of bitterness) to 9 (extremely bitter). For a more accurate phenotypic score in the F2, we combined the Dragendorff phenotypic score multiplied by two to that of the bitterness tasting score. In this combined phenotypic score, the phenotypic value ranged from 3 to 15, 15 being the most bitter. Chemotyping of F3 individual seeds Approximately 1g of seeds (3–5 seeds) of a curated panel of F3 (Supplementary Table S6) and parental individuals were milled for analytical QA quantification. QA extraction and characterization were performed according to modified versions of the protocol described in Wink [18], Resta et al. [19] and Boschin et al. [20]. Briefly, 100mg of seed flour was suspended in 1.2 mL of 0.1N HCl, with sparteine (CAS 90-39-1; Extrasynthese, France) added as internal standard in an appropriate concentration, and stirred at room temperature overnight. The mixture was centrifuged at 8000g for 45min at 4°C, the supernatant was collected, and the solid was washed twice with 0.8 mL of 0.1N HCl. The gathered extracts were alkalinized with 5% NH4OH to pH 10 − 11 and then applied onto an Extrelut NT 3 column (Supelco, Merck, Darmstadt, Germany). The alkaloids were eluted with CH2Cl2 (3 × 5 mL), and the solvent was evaporated under vacuum. The residue was resuspended in an appropriate volume of dichloromethane and analyzed Fig. 1 Examples of the colour-reaction of Dragendorff paper according to the predefined scoring scale from 1 (very sweet) to 4 (very bitter). 1 - purely white stain; 2 - mostly white with the red encircling; 3 - red encircled pinkish stains with a light centre 4 - intense red stain without discolouration in the centre Page 5 of 14Patyi et al. BMC Plant Biology (2025) 25:945 by gas chromatography. Each sample was independently extracted, then analyzed in duplicate. Total and individual QA content was determined quantitatively using the internal standard methods by gas chromatography/Flame Ionization Detector (PerkinElmer Clarus 500 GC-FID), equipped with a capillary column Elite-5 MS (DB-5, 30m × 0.32mm × 0.25μm; Perkin-Elmer; Milan, Italy). The oven temperature ramp was held at 90°C for 2min, increased to 300°C at 7°C/ min and held at 300°C for 10min. Helium was used as the carrier gas, and the flow rate was set at 2 mL/min. The temperature of the injector was set at 300°C, and the injection volume was 1 µL; the temperatures of the FID was set at 320°C. The absence of sparteine in the samples was demonstrated by performing GC/FID analyses without an internal standard (choosing one sample per recombinant). The response factor of GC/FID was calculated using the ratio between the response of the internal standard (sparteine) and the response of the analyte standard lupanine. The regression coefficient between the analyte concentration and detector response was R2 = 0.991. Qualitative identification of QAs was performed by gas chromatography/mass spectrometry; GC/MS analyses were carried out using a Perkin Elmer Clarus 500 GC equipped with a Clarus 500 mass spectrometer using the same capillary column and chromatographic conditions as for the GC/FID analyses (above mentioned). Mass spectra were acquired over a range of 40–400 atomic mass units (amu) at 1 scan/sec with ionizing electron energy of 70eV and ion source at 230°C. The transfer line was set at 300°C, while the carrier gas was helium at 1.0 mL/min. The QAs were identified by determination of their elution times published mass spectra [18– 20], as well as a peak-matching library search [21]. The standard GC condition described above allowed a practical measurable sensitivity of 10 ng/µl per injection, i.e. a detection limit of 1µg of lupanine g−1 of lupin seed flour (1 ppm). Pooling for bulked segregant analysis Two bulks consisting of DNA (isolated with a modified CTAB method) from eight plants of the sweet and bitter phenotypic extremes of the ‘Dieta’ x ‘Frieda’ F2 experimental population were curated for sequencing. Homozygous individuals for the pauper locus were excluded from this subset to discriminate sweet individuals where the low QA content is conferred by pauper. Thus, the F2 bulks consisted of individuals which were heterozygous or wild type at the pauper locus, suggesting that the low QA phenotype is conferred by a determinant other than pauper. The combined phenotypic score [3–15] of the sweet bulk was between 7 and 8.5, while that of the bitter one was 13–15. Additionally, bulks of the parental cultivars (from seedlings grown for this purpose from the same seed source used for the crossing) consisting of eight plants each were sent for sequencing. Sequencing-data analysis Whole genome sequencing was performed at Novogene UK (Cambridge, UK) with a NovaSeq X Plus Series, sequencing depth of 15x. Paired-end 150bp long reads were generated. The raw sequence data was processed and analysed in a modified version of the procedure published by Schneider et al. 2022 [22]. Quality checks were performed with the fastqc tool [23]. Raw reads were aligned to the reference genome using bwa mem [24]. Subsequently, duplicated reads were removed using the markdub-function of sambamba [25] and sorted by their mapped position. Reads that had been mapped to multiple locations (e.g. secondary alignments) were discarded. The variant calling was conducted using bcftools only on SNPs that had been previously reported [15] to improve the confidence despite the low sample number and read depth. Furthermore, reads with a quality threshold below 25 for the whole read, and 30 per SNV base call were omitted, as well as SNPs with read depths below 8. The allelic depth (AD) was extracted and heterozygous SNPs in the parental bulks were omitted from further analysis. The remaining SNPs were assigned to annotated genes based on their position in the gene or in close proximity to it, as described by Schneider et al. 2022 [22]. For the resulting 23,396 gene-anchored haplotypes, the allele frequencies of ‘Dieta’ and ‘Frieda’ were calculated for the sweet and for the bitter bulk. The frequencies were aggregated with the read depth to count values, as required for BayPass [26], which is a Bayesian approach to identify i.e. selective sweeps between the sweet and bitter populations. BayPass was conducted using the effective population size of 200 to reflect the size of the ‘Dieta’ × ‘Frieda’ F2 population. From the p-values of each haplotype and their mapped positions (Supplementary Table S8), Manhattan plots were generated, with a threshold of -log10(p-value) > 4.5. Genotyping PACE primers Primer triplets (Table1) for 3 selected SNPs in the newly identified QTL and for the SNP associated to pauper were designed by 3CR Bioscience (Essex, UK) and oligonucleotides were acquired from IDT (Coralville IA, USA). For all four loci, the primers were designed based on the sequence of the sense strand. PACE assay For the genotyping assays, we used the PACE master mix (2x concentration, standard ROX level) provided by 3CR Bioscience (Essex, UK). Each individual reaction Page 6 of 14Patyi et al. BMC Plant Biology (2025) 25:945 consisted of 5µl of the PACE master mix, 0.14µl of the primer pool and 5µl of ELGA water (VWS Ltd.; High Wycombe, UK) or DNA template, isolated from leaf samples transferred onto Qiagen FTA cards (QIAGEN GmbH; Hilden, Germany), processed according to the manufacturer`s instructions. The cycling conditions were set according to the manufacturer`s instructions of the analogously functioning Kompetitive Allele Specific PCR (KASP) assay (LGC Group; Teddington, UK) and ran on the CFX Opus 384 Real-Time PCR System (Bio-Rad Laboratories, Inc.; Hercules CA, USA). All 112 F2 individuals were genotyped in 2023 using a newly designed PACE marker tagging the most important SNP associated to pauper [14], Lalb_Chr18_12359687 and only retrospectively genotyped in 2024 upon the availability of the PACE marker tagging the Dieta-associated sweetness. Samples from the F3 population were genotyped with both markers in 2024. Data analysis All data except for the sequencing data was analyzed and visualized in R version 4.4.2 [27]. The phenotypic data (Dragendorff score) was analyzed in the F2 and F3 experimental populations using Linear Mixed Models (LMM). The allelic combinations at loci Lalb_Chr18_12359687 and Lalb_ Chr05_6643621 were included as fixed effect, and the row the specific plant belonged to on the field as a random effect. The estimated marginal means from the above-mentioned linear mixed model were calculated to test the statistical difference (p < 0.05) using emmeans_1.10.7 [28] and were used to assess the association between the allelic states and the phenotypic value in F2 and F3. Similarly, the total alkaloid content (sum of all QAs present in a sample) and the ratio between the concentrations of 13α-hydroxylupanine and lupanine per sweet variant were analysed LMM using lme4_1.4–36 [29]. For both response variables we included the allelic combinations as fixed, and the biological replicate (i.e. set of samples belonging to allelic combinations of the two loci of interest) as random factor. The estimated marginal means from the above-mentioned linear mixed model were calculated to test the statistical difference (p < 0.05) using emmeans_1.10.7 [28] and were used for the following downstream analysis. Results The phenotypic segregation in the F2 generation confirms another determinant for low QA content With pauper known to segregate in the ‘Dieta’ x ‘Frieda’ F2 population by the genotyping at the causal mutation SNP Lalb_Chr18_12359687, the simplest hypothesis of 1: 3 (sweet: bitter) phenotypic segregation ratio could be rejected. Instead, the phenotypic segregation pattern was 7: 9 (sweet: bitter), which coupled with observation of the genotypic segregation 1: 2: 1 at the pauper locus suggests the segregation of at least one additional low QA conferring gene aside from pauper (Table2). Assuming segregation of two independent recessive genes, the phenotypic scores are expected to segregate in a ratio of 7: 9 (sweet: bitter) in QA content (following Mendelian inheritance), which hypothesis could not be rejected (Table2), indicating two independent recessive loci conferring low QA content with no segregation distortion in the F2 population used for the BSA. The assumption of genotypic segregation of 1: 2: 1 could not be rejected for either loci, further supporting the independence of these determinants. Identification of a QTL associated to low QA content through BSA After the confirmation of the presence of a determinant independent of pauper for low QA content in the Table 1 PACE primers used in the study. Target SNP locus on the chromosome, the oligonucleotide sequence for the two forward (F1 and F2) tagging the two possible alleles at the respective locus (underlined), including the sequence of the Xand YTails and the reverse (R) primers for each triplet, and melting temperatures Locus Primer Sequence (5’−3’) Tm (°C) Lalb_Chr05_6643621 F1 GAAGGTGACCAAGTTCATGCTAGACTATGTCATGTTTTTAAACATGTCAC64.2 F2 GAAGGTCGGAGTCAACGGATTGAGACTATGTCATGTTTTTAAACATGTCAA65.4 R GGCTTGAATAGGGTCCATGTGTTTAATAT 57.1 Lalb_Chr05_6643864 F1 GAAGGTGACCAAGTTCATGCTAATCTATTATTGTCCTATTAAAACATTACGAG62.8 F2 GAAGGTCGGAGTCAACGGATTAATCTATTATTGTCCTATTAAAACATTACGAC63.3 RTATATTTTTTAGCTTAGGATGTTTTGGATT 52.3 Lalb_Chr05_6645052 F1 GAAGGTGACCAAGTTCATGCTCGATAAGAGAGTGGGAGGACCAA67.7 F2 GAAGGTCGGAGTCAACGGATTGATAAGAGAGTGGGAGGACCAC67.8 R CGTGGCATGGACCCCAGTTATATAA 58.5 Lalb_Chr18_12359687(pauper) F1 GAAGGTGACCAAGTTCATGCTAAATGCTATCAGGATAGGGTCTATGT65.3 F2 GAAGGTCGGAGTCAACGGATTAAATGCTATCAGGATAGGGTCTATGA65.7 RCTTCCATTGTTCTTCCTCTATCTRCACTT 57.3 Page 7 of 14Patyi et al. BMC Plant Biology (2025) 25:945 segregating F2 population, we identified the genomic region responsible for this unidentified determinant of low QA content through BSA. From the raw reads ranging between 97 and 117 million per bulk (read length of 150bp), 47–57 million remained after quality filtering, mapping and removal of redundancies (Supplementary Table S7). There were 1.55million polymorphic SNPs (1.76million before q40 filtering) from the lupin browser that were polymorphic between the parents, reported by bcftools, from which 514,982 remained after filtering out monomorphic, low quality and low depth variants. Following, 23,396 gene-anchored haplotypes with an average of 21 SNPs and an average read depth of 302 per haplotype were constructed. The selective sweeps analysis resulted in a clear peak on chromosome 5 (Fig.2) that allows to differentiate the ‘sweet’ from the ‘bitter’ bulk. Candidate gene selection and marker development Ten genes were associated with haplotypes showing significant (-log10(p-value) > 4.5) differences in allele frequencies between sweet and bitter bulks within the entire QTL region spanning from 5,799,140bp to 6,737,917bp on chromosome 5 (Supplementary Table S9). Although none of them could be directly linked to the biosynthetic pathway, the gene Lalb_Chr05g0222381 encoding a “Putative aminoacyltransferase” (a function similar to the gene responsible for the pauper-conferred low QA content) was chosen for molecular marker development. We investigated all individual SNPs within as well as 1,000bp upand downstream of this gene. There are 54 polymorphic loci between ‘Dieta’ and ‘Frieda’ in this region. To determine whether these are also polymorphic within other accessions as well, we collected the allelic states for both loci of all 39 sequenced accessions from the white lupin genome browser (Supplementary Table S5). Out of these 54 SNPs, only 9 are highly polymorphic within the 39 sequenced accessions [15]. Furthermore, only 3 loci are consistent with the assumed alkaloid status of these accessions, Lalb_Chr05_6643621, Lalb_Chr05_6643864, Lalb_Chr05_6645052. These three were tested on the validation panel of 34 accessions to further confirm their association to alkaloid status (Supplementary Table S5), in comparison to the marker-trait association of the marker tagging pauper. At the locus Lalb_ Chr05_6643864, an old cultivar, Nährquell (described to carry the sweetness mutation nutricius) carried the allele Table 2 Segregation ratios and χ2 statistics in the phenotype (Dragendorff score) and at the pauper, and Dieta-associated locus for the entire F2 population Segregating population Expected ratio Observed ratio df χ2 p-value Phenotypic segregation ‘Dieta’x ‘Frieda’ 1:3 41:69 1 8.8364 0.00295 ** 7:9 41:69 1 1.8753 0.1709 Genotypic segregation ‘Dieta’ x ‘Frieda’ pauper segregation 1:2:1 31:47:34 2 3.054 0.2172 Dieta-associated segregation 1:2:1 22:63:27 2 2.1964 0.3335 Fig. 2 Manhattan plot of -log10(p-value) (y-axis, threshold set to -log10(p-value) = 4.5), for differences in allele frequencies of haplotypes between two bulks of the F2 population representing high and low QA levels. The x-axis indicates the position of the haplotypes in each of the 25 chromosomes. A zoomed-in plot is included for chromosome 5 for detail Page 8 of 14Patyi et al. BMC Plant Biology (2025) 25:945 associated to the Dieta-type sweetness, and the wild-type allele at the other two selected loci in the region thus this locus was not chosen for further analysis. Similarly, at the locus Lalb_Chr05_6645052, when comparing two different phenotypically sweet individuals of the cultivar ‘Start’ (described to carry the sweetness mutation exiguus) lead to inconsistent results, thus we selected only the SNP Lalb_Chr05_6643621 for further genotyping of the F3 segregating population. Interestingly, French cultivars which belong to the panel of resequenced accessions, namely ‘Magnus’, ‘Orus’, ‘Luxe’ and ‘Clovis’ carried the alleles associated to sweetness both for the pauper locus and the newly described one on chromosome 5. Marker-trait associations reveal low QA stacked allele recombinants The marker-trait association between the newly described Lalb_Chr05_6643621 SNP, the pauper causal mutation SNP Lalb_Chr18_12359687 and the phenotypic score (Dragendorff score) was confirmed for the individuals in the F2 and F3 populations (Supplementary Figure S1-S3), in comparison to their parental lines. In F2, we were able to statistically differentiate between individuals carrying the sweet genotype associated to pauper, and discriminate for bitter individuals based on the Dragendorff score (Supplementary Table S2). However, in F3, only pauper sweet individuals, and bitter individuals were clearly statistically grouped together based on the Dragendorff score only (Supplementary Table S2, Supplementary Table S4). to obtain more nuanced observations on the quantitative value of alkaloids, we performed GC/MS and GC/FID analytical QA quantification in a subset of the F3 population representing different combinations of allelic states at the pauper causal mutation SNP on chromosome 18 and the locus Lalb_Chr05_6643621 on chromosome 5, used for the following downstream analysis. This subset included the two parental lines and the most important allelic combinations at both loci (Table3) and the corresponding estimated marginal means values from the post-hoc analysis after model fitting. Comparing the concentrations of the total alkaloid content (Fig.3), we observed a decrease in the estimated marginal means of the total alkaloid content in all sweet chemotypes. The F3 plants carrying the allele pauper as sole contributor to the low QA phenotype (PAU) were grouped together with ‘Frieda’ (FRA), the parent conferring the pauper-type sweetness (Fig.3). Similarly, the F3 progeny plants in which the low QA phenotype was conferred by the parent ‘Dieta’ (DSW), were grouped together with the parental genotype (DIA), as well as the F3 individuals heterozygous for pauper and homozygous sweet at Lalb_Chr05_6643621 (HET), though with higher average QA content compared to the plants that are sweet because of pauper (PAU). The stacked allele recombinants (STA) have a very low alkaloid content (range raw data 11–62.3 ppm), lower compared to either parent or progeny, indicating a clear additive effect of stacking two low QA conferring determinants. Stacked allele recombinants (STA) represented the most drastic reduction to 12% and 8% of the total average alkaloid content of the parental cultivars, ‘Dieta’ and ‘Frieda’, respectively. The total estimated marginal mean for the alkaloid content was 22.8 ± 10.4 ppm dry weight in seeds of the stacked allele recombinants, close to 10 times lower than the suggested threshold for human consumption. QA profiles in the F3 population highlights the stacked allele effect Based on the presence of specific QAs in the different allelic combinations studied, we observed distinct QA profiles associated to different low-QA determinants and their combination. In the wild-type, bitter phenotype (BIT) 14 QAs were identified (Fig.4). Table 3 Subset of F3 individuals analyzed with GC/MS and GC/FID, their allelic state, number of samples (n), estimated marginal means of total Quinolizidine alkaloid (QA) content (in ppm) indicating statistical grouping, and standard error (SE). BIT– bitter wild type progeny; FRA– ‘Frieda’ parent, homozygous sweet at pauper, wild type at Lalb_Chr05_6643621; DIA– ‘Dieta’ parent, wild type at pauper, homozygous sweet at Lalb_Chr05_6643621; PAU– homozygous sweet at pauper, wild type at Lalb_Chr05_6643621; DSW– wild type at pauper, homozygous sweet at Lalb_Chr05_6643621; HET– heterozygous at pauper, homozygous sweet at Lalb_Chr05_6643621; STA– homozygous sweet at pauper, homozygous sweet at Lalb_Chr05_6643621 Code Allele nQA content (ppm) SE (ppm) Lalb_Chr18_12359687 (pauper) Lalb_Chr05_6643621 BIT(bitter progeny) AA wild type CC wild type 4 9715.2 2284.8 DSW(Dieta sweet progeny) AA wild type AA homozygous sweet 10 296.4c15.7 HET TA heterozygous AA homozygous sweet 10 267.1c15.7 DIA(‘Dieta’ parent) AA wild type AA homozygous sweet 10 257.4c15.7 PAU(pauper sweet progeny) TT homozygous sweet CC wild type 7 171.7b18.5 FRA(‘Frieda’ parent) TT homozygous sweet CC wild type 10 176.2b15.7 STA(stacked allele recombinants) TT homozygous sweet AA homozygous sweet 23 22.8a10.4 Page 9 of 14Patyi et al. BMC Plant Biology (2025) 25:945 Fig. 4 Gas chromatogram of alkaloids from L. albus seeds. IS - internal standard (spartein); 1 -ammodendrine; 2 - albine; 3 - tetrahydrorhombifoline; 4 - angustifoline; 5 - α-isolupanine; 6 - lupanine; 7 - N-methylalbine; 8 - multiflorine; 9–17-oxolupanine; 10–13α-hydroxylupanine; 11–13α-hydroxymultiflorine; 12–13α-angeloyloxylupanine; 13–13α-tigloyloxylupanine; 14 − 13α-angeloyloxymultiflorine Fig. 3 Individual QA content and composition, and total QA content (estimated marginal mean value in ppm) of different recombinants in the F3 population compared to the parental genotypes. Values with the same letter are not significantly different at 0.05 confidence level. BIT– bitter wild type progeny; FRA– ‘Frieda’ parent, homozygous sweet at pauper, wild type at Lalb_Chr05_6643621; DIA– ‘Dieta’ parent, wild type at pauper, homozygous sweet at Lalb_Chr05_6643621; PAU– homozygous sweet at pauper, wild type at Lalb_Chr05_6643621; DSW– wild type at pauper, homozygous sweet at Lalb_Chr05_6643621; HET– heterozygous at pauper, homozygous sweet at Lalb_Chr05_6643621; STA– homozygous sweet at pauper, homozygous sweet at Lalb_Chr05_6643621