Seed ripening and quinolizidine alkaloid dynamics in Lupinus mutabilis: from gene activity to metabolites

Lupins are high‐protein feed and food crops, but most species synthesize toxic quinolizidine alkaloids (QAs). Sweet lupins have been bred for low alkaloid levels, yet occasional reversion to the bitter, toxic phenotype requires routine monitoring for food safety. Understanding QA biosynthesis, transport, and seed accumulation is crucial for developing lupins as reliable crops and for strategies aimed at reducing toxicity or valorising alkaloids. Eight major QAs were quantified during seed ripening in Lupinus angustifolius and L. mutabilis. Pearson correlation analyses explored relationships among alkaloids and informed a hypothesis on pod‐to‐seed transport, especially in L. mutabilis. Semi‐quantitative profiling of ten putative minor alkaloids complemented the study, and targeted transcriptomics measured expression of QA‐biosynthetic and candidate transporter genes in pods and seeds. Both species showed distinct temporal accumulation patterns of the eight major QAs. Correlation data supported the hypothesis that pods are the primary QA source, with subsequent translocation to seeds in L. mutabilis. Minor‐alkaloid profiling expanded the alkaloid spectrum. Transcriptomics revealed low QA‐biosynthetic gene expression in seeds but detectable transporter gene expression, consistent with pod‐derived QA import. The evidence indicates that QA synthesis occurs mainly in pods, followed by transporter‐mediated movement into seeds. Recognising pods as the QA source provides a knowledge‐based target for reducing seed toxicity or for exploiting alkaloid transport pathways in future lupin crop improvement.

Lupin seeds are a protein source for animal and human nutrition (van de Noort, 2024). The high protein content rivals global crops like soybean while being rich in fiber and low in starch (Hove, 1974). In addition to exhibiting an increased resistance to climatic and biotic stress, lupins form a taproot system that improves soil structure. They are also able to fix atmospheric nitrogen via symbiotic interaction with rhizobia, thus mitigating soil nitrogen depletion (Pueyo et al., 2021). Agriculturally, bitter lupins, exhibiting a high quinolizidine alkaloid (QA) content toxic for mammals and humans alike, are discriminated from sweet lupins (Schrenk et al., 2019). On the one hand, sweet lupins are less harmful as food and feed; on the other hand, field cultures might suffer from increased susceptibility toward arthropod plant pests and herbivores (Talhinhas et al., 2016; Rodés-Bachs and Van der Fels-Klerx, 2023; Frick et al., 2018). Sweet lupins are a result of plant selection and breeding; however, there might be stress-induced titer increases and revertants bearing a high alkaloid content; thus, alkaloid monitoring is recommended in food affairs (Schrenk et al., 2019). The most widely accepted biosynthetic pathway for QAs (see Figure 1) has been proposed by Mancinotti et al. (202) and Schoofs et al. (1983), starting with the amino acid L-lysine (1). This is decarboxylated to form cadaverine (2) using a L-lysine decarboxylase (Schoofs et al., 1983).

Subsequently, cadaverine (2) is oxidized by a copper amine oxidase (CAO) to 5-aminopentanal (3), which cyclizes to Δ 1 -piperideine (4) (Mancinotti et al., 2022). By tautomerism, Δ 1 -piperideine (4) and Δ 2 -piperideine (5) equilibrate. Both molecules may couple in an aldol-like reaction to (2′ R,3′ R)-tetrahydroanabasine (6) stereoselectively. The absence of enantiomers highlights the involvement of an enzyme. The natural compound 6 is the last analytically known compound of the following biosynthesis hypothesis. This hypothesis is illustrated in Figure 1. Although definite proof could not yet be provided, the most accepted hypothesis is the attachment of yet another Δ 1 -piperideine subunit to provide the bisquinolizidine scaffold (11), which is the basis for all tri- and tetracyclic QAs (Mancinotti et al., 2022; Namdar et al., 2024). Neither the diiminium ion (11) nor its deprotonated states such as compound 12 have ever been detected. From there on, the reaction mechanism is almost completely unknown, with different hypotheses known in the literature: While some argue for separated derivatization of lupanine (13) and sparteine (14) with further modification toward hydroxylupanines and vinyl-side chains, others see sparteine (14) as the precursor alkaloid, which is decorated (Schoofs et al., 1983). Much of the older literature used in vivo radioactive labeling that was mostly unsuccessful, so a new approach was first tried by Namdar et al. (Namdar et al., 2024; Schütte and Hindorf, 1965; Golebiewski and Spenser, 1988; Rana and Robins, 1984). Using correlation-based analysis of different bitter lupin species, the phylogenetic analysis of the heritage of alkaloids was partly uncovered (Namdar et al., 2024).

A recent review on the chemodiversity of QA listed 397 individual compounds, of which more than 170 were lupin based (Cely-Veloza et al., 2023). They differ from each other in line, accession, variety, and environment, with some having pharmaceutical or ligand-type applications (Wrobleski and Long, 1977; Zeng et al., 2025; Villalpando-Vargas and Medina-Ceja, 2016). The biosynthesis of these toxins mainly takes place in the green organs of the plant, with a particular focus on the leaves (Frick et al., 2025; Lee et al., 2006). From the aerial epidermis, the alkaloids are transported into the remaining plant parts, making the plant entirely toxic, with some of the alkaloids even being exported to the surrounding soil by the roots (Hama et al., 2022). This export was already investigated and led to the hypothesis that core alkaloids lupanine (13) and sparteine (14), as well as esterified lupanines, are transported via the phloem to their respective destination. Arriving at the seeds, the alkaloids are transmitted through the pods into the seeds, where they accumulate. Quite recently, this transfer was first investigated by Otterbach et al. (2019) in L. angustifolius.

The genes involved in these biosynthesis and transport processes have already been investigated by Bunsupa et al. (2012), leading to the functional confirmation of the genes encoding LDC and CAO. Loci like iucundus and pauper were linked to the sweetness or bitterness of seeds in L. albus (Frick et al., 2017). In L. angustifolius, an acetyltransferase (LaAT) possibly involved in acetylation in early QA biosynthesis from the pauper locus was identified, showing polymorphism causing sweetness in L. angustifolius (Mancinotti et al., 2023). The APETALA2/ethylene response transcription factor RAP2-7 gene on the locus iucundus determines bitterness in L. angustifolius (Kroc et al., 2019). In addition to these results, identifying biosynthetic genes and transport mechanisms remains one of the biggest challenges in lupin research.

This study used the knowledge of different alkaloid profiles over the developmental stages. The hypothesis was that green tissues, especially in the young plants, are biosynthesizing QAs, which are then transported into reproductive organs where they accumulate and are also modified. Both the L. mutabilis and L. angustifolius [cv. Oskar; Otterbach et al. (2019) used as reference in this study] alkaloid time series were investigated. This sheds light on their major and minor QA relations and biosynthesis. In addition to analyzing the alkaloid content, the study analyzes transcriptomic data to correlate changes in alkaloid content with altered expression of genes involved in and related to QA biosynthesis and transport. For this, 33 gene homologs co-expressed with LDC first found by Yang et al. (2017), including a purine permease 1-like gene encoding a putative alkaloid transporter and a Myb-like transcription factor that might be involved in the regulation of QA biosynthesis, were investigated.

Biochemical methods

Chemicals

All chemicals were purchased at Roth, Sigma-Aldrich, or TCI in the respective highest purity (at least 98%).

Cultivation of lupins

The bitter lupin line L. angustifolius Bo9027 was donated by Saatzucht Steinach GmbH & Co KG, Steinach, Germany. L. mutabilis PI 457964 was obtained from the National Plant Germplasm System, United States Department of Agriculture-Agricultural Research Service (GRIN-Global, https://npgsweb.ars-grin.gov/gringlobal/search) in 2019 (Byrne et al., 2018). For details of the cultivation, see Supplementary Material; in short: The lupins were cultivated in an ED73:Vermiculite substrate in a fully controlled growing chamber. Seeds and pods have been harvested and categorized according to Table 1.

Alkaloid isolation

The isolation of the major alkaloids was performed using a method published previously (Brooke et al., 1996). After grinding the dried seeds (50 g) in a GK 600 blender (Rotor Lips AG) for 13 min (7 min intensity 2 of 7, 6 min intensity 5 of 7), all seeds were ground finely (<1 mm fragment size). The extraction of the alkaloids was performed in a Soxhlet-based approach described by Brooke et al. (1996). A 150 mL Soxhlet containing 50 g of ground material in a 33 mm × 180 mm cellulose thimble was run for 72 h at 85 °C in 300 mL methanol. Afterward, the solvent was removed in vacuo, and the extracts were stored at 4 °C.

pH-based extraction

The extracts were dissolved in a succession of 3 × 20 mL aqua demin. All aqueous layers were combined and transferred into a 500 mL separating funnel. An adjustment to pH 1 was reached by using 700 µL 37% (m/m) HCl in water. An extraction took place with 3 × 120 mL of dichloromethane. After the second extraction, the interphase was disrupted by adding 20 mL brine. The organic layer was discarded. The aqueous layer was adjusted to pH 11 (1.2 mL 10 M NaOH). The alkaloids were extracted using 6 × 120 mL dichloromethane. The combined organic layers with pH 11 were dried in vacuo at 800 mbar and stored at −20 °C under a nitrogen atmosphere (approx. 500–1,000 mg).

Preparative, chromatographic isolation

In accordance with the literature, the isolation of QA was performed using varying solutions of methanol, dichloromethane, and ammonia in silica-60 column chromatography [(Brooke et al., 1996); for details, see SI]. The chromatographic separation from extract (see above) of L. angustifolius Bo9027 seeds (50 g) yielded lupanine (13), isolupanine (20), angustifoline (19), and 13α-hydroxylupanine (16) in purities >95% determined by qNMR (see Supplementary Material). From extracts of L. mutabilis PI 457964 seeds, lupanine (13), sparteine (14), 3β-hydroxylupanine (15), tetrahydrorhombifoline (17), and hydroxytetrahydrorhombifoline (18) have been isolated (purity >90% based on qNMR).

GC-based quantification

After isolation, calibration standards were performed using the previously obtained isolates. Their purity was determined using qNMR (see Supplementary Material). All standard samples were individually dissolved in methanol [using as an internal standard 50 µg/mL caffeine as done in literature, for example, Reinhard et al. (2006)] at concentrations ranging from 0 to 1.5 mg/mL. Additional measurements revealed the limit of blank (LOB) and limit of detection (LOD) for each alkaloid (see Table 2). Calibration curves have been recorded following every maintenance of the analytical device. Intra- and inter-day precision values have not been determined. Matrix effects of the samples have not been elucidated.

Alkaloid extraction

The extraction was performed following the protocol of Muzquiz et al. (1994), where 100 mg of freeze-dried material was disrupted for 3 min at 30 Hz in a Retsch swingmill MM 400. The powder was transferred to a 50 mL plastic reaction tube (Falcon), and 10 mL aqueous 0.5 M HCl was added. The mixture was vortexed for 15 s and then sonified in a bath for 20 min. Afterward, the solids were precipitated by centrifugation for 40 min at 14,000 × g. The supernatant was transferred, and the process repeated by adding another 10 mL 0.5 M HCl. The supernatants (20 mL) were adjusted to pH 11 (1.2 mL 10 M NaOH). At this point, the extract mixture was transferred onto an Extrelut NT20 (Merck, Darmstadt, Germany) column. After 20 min incubation, the alkaloids were eluted by addition of 3 × 20 mL dichloromethane. The combined eluate was evaporated, and the extract was dissolved in 1.5 mL methanol (internal standard: 50 µg/mL caffeine). The sample was analyzed by GC-MS according to the method described (see Supplementary Material).

RNA extraction

RNA was extracted from four replicates of seeds and pods of Lupinus mutabilis PI 457964 collected in the year 2025 at developmental stages 1.5 and 2, harvested from the same plants used for QA quantification. Total RNA was extracted from plant tissues using the Qiagen RNeasy Plant Mini Kit (Qiagen, Germany), according to the manufacturer’s instructions. To eliminate genomic DNA, RNA samples were treated with Baseline ZERO DNase (Biosearch Technologies, Germany, 5.7 µL buffer, 1.7 µL enzyme) and incubated at 37 °C for 30 min. After DNase treatment, volumes were adjusted to 100 µl with RNase-free water, followed by the addition of 350 µl lysis (RLT) buffer (Qiagen RNeasy Plant Mini Kit, Qiagen, Germany) and 250 µL 100% ethanol. The solution was transferred to RNeasy columns (Qiagen RNeasy Plant Mini Kit, Qiagen, Germany), centrifuged at 10,000 × g for 15 s, and washed twice with 500 µL RPE buffer (Qiagen RNeasy Plant Mini Kit, Qiagen, Germany) for 15 s and for 2 min at 10,000 × g. RNA was eluted in two sequential 25 µL aliquots with RNase-free water, each incubated 1 min at room temperature and centrifuged at 10,000 × g for 1 min, yielding 50 µL RNA. Samples were adjusted to equal volumes using RNase-free water and mixed with an equal volume of chloroform, followed by thorough inversion and gentle shaking for 10 min. The mixtures were centrifuged at 10,000 × g for 10 min at room temperature, and the aqueous phase was carefully transferred to fresh 1.5 mL microcentrifuge tubes, avoiding the interphase. To the aqueous fraction, 0.15 volumes of 3 M sodium acetate (pH 6.4) and 3 volumes of 100% ethanol were added and mixed by inversion. Nucleic acids were precipitated by centrifugation at 14,500 × g for 20 min at 4 °C. The resulting pellet was washed with 70% ethanol (v/v) and centrifuged at 10,000 × g for 5 min. After the wash, the supernatant was discarded, and the pellet was air-dried prior to resuspension in RNase-free water. RNA concentration and purity were assessed spectrophotometrically prior to downstream applications. Only samples with an RNA integrity number (RIN) ≥ 7.0 as determined in a 2100 Bioanalyzer system (Agilent, United States) were retained for library preparation. High-quality RNA samples were submitted to Novogene (Munich, Germany), where sequencing libraries with mRNA enrichment through polyA capture were prepared and sequenced on the Illumina NovaSeq X platform to generate 150-bp paired-end reads.

Instrumental methods

GC-MS-based analyses of QA profiles

The GC-MS detection method applied in the detection of the QA-alkaloid profiles (2 min 120 °C, 30 min 120 °C–300 °C (6 °C/min), 2 min 300 °C) was done under constant flow (1.2 mL/min) using an MS transfer line temperature of 320 °C with a split of 20 ml/min. The ion source temperature was set to 230 °C, and a detection of 50–550 amu started 2.5 min after injection to protect the detector.

GC-ToF-MS-based identification of minor QAs

For the detection of minor alkaloids in L. angustifolius and L. mutabilis, a GC-ToF-MS-based method was applied (Paczia et al., 2012). The extraction was performed similarly to GC-MS quantification, but in the last step, the solution was dissolved in 10 mM phosphate buffer pH 6.8 to meet the requirements of the derivatization.

NIST-based analysis

Minor QAs were identified using the National Institute of Standards and Technology (NIST) database and the NIST-compliant internal database JuPoD (NIST, 2020). First, a detection of the isolated major alkaloids yielded their RI values. Based on the mother mass and the fragment patterns, peaks with similar appearances were searched in alkaloid profiles of all aging stages in both species of lupins. As these compounds were all represented under 2% of the total alkaloid amount in the alkaloid profile, the isolation of the compounds from the available sources was impossible, and identification was rather difficult. To ensure trustworthy representation of the findings, fingerprint-like fragments with exact masses, as determined by GC-ToF-MS, were identified. Based on this information, the alkaloids could be subdivided into lupanine type, sparteine type, and esterified lupanines. If the fragment patterns and RI values matched above an 800/1,000 similarity score with an alkaloid measured in the databases above on the same column, it was sorted to its fragment-based group. It must be acknowledged that distinguishing isomers is not possible with this method. After finishing the identification process, the presence of each compound was verified by calculating the quotient from the area of the QA in question divided by the area of the corresponding standard. This result was then normalized individually for each QA and displayed in a color-coded matrix.

Kovats index determination (RI value)

The RI values of the major and minor alkaloids were analyzed in comparison to a 50 mg/L C10-40 alkane standard purchased from Sigma-Aldrich (Art. No. 68281). After running the standard on the GC-MS and the GC-ToF-MS, the RI values of the major and minor alkaloids were calculated using the formula with R I a l k a n e C 10 + 2 n H 22 + 4 n, n = 0;...; 10: = 1000 + 200 · n

where RI is the retention time index, Rt′ is the netto retention time (detection start GC-ToF-MS ∼ 1.5 min), Rt is the brutto retention time, x is component x, alkane1 is an early alkane, and alkane2 is a late alkane.

Bioinformatic analysis

For raw sequencing data (ENA accession ID: PRJEB104093), quality assessment of the reads was performed using FastQC v0.12 (Andrews, 2010), and Illumina adapter sequences were removed using Trimmomatic version 0.40 (Bolger et al., 2014; Beier et al., 2026) using the default parameters. Cleaned reads were then mapped to the L. mutabilis PI 457964 (“Cruickshanks”) reference genome using Salmon v1.10.1 (https://git.nfdi4plants.org/usadellab/Lupinus_mutabilis_cruickshanks_Genome) with quasi-mapping-based transcript quantification using default parameters (Patro et al., 2017). The resulting count matrices were used for downstream differential gene expression analysis. Four biological replicates were used for each developmental stage/tissue. RNA sequencing generated an average of approximately 59 million reads per sample. Gene-level raw counts were imported into R (v4.5.2), and differential expression analysis was performed using the DESeq2 package (Love et al., 2014). Within the DESeq2 framework, counts were normalized to account for differences in library size using size-factor estimation. Differentially expressed genes were identified using a false discovery rate (FDR) < 0.05 and an absolute log 2 fold change (|log 2 FC|) ≥ 2 across four comparisons of contrasting tissues (pods vs. seeds) and developmental stages (stage 1.5 vs. stage 2). All sequences of genes involved in alkaloid production in L. angustifolius genome assembly version 2.0 (https://lupinexpress.org/) and L. albus genome version 1.0 (Genbank ID: GCA_009771035.1) were downloaded, and a BLASTp version 2.15.0 search was conducted against an L. mutabilis PI 457964 (“ Cruickshanks”) custom protein database using an e-value threshold of 1 × 10 −5 (Altschul et al., 1990). To find homologs, multiple sequence alignments were performed using Clustal Omega (Sievers et al., 2011). Homologs in L. mutabilis PI 457964 were confirmed based on the presence of conserved domains, targeting signals for subcellular localization, the catalytic sites in the case of enzymes, and overall protein length.

Establishing a GC-MS-based quantification method

The isolation of the major quinolizidine alkaloids (QA) in L. angustifolius Bo9027 and L. mutabilis PI 457964 was successfully performed (see Figure 2). The accuracy was tested using standard graphs of all eight major alkaloids, with the isolates gained from lupin seeds and caffeine as the standard. The quantification was established on a GC-ISQ-MS rather than the literature-established liquid chromatographic systems to combine major and minor alkaloid detection on one system.

QA quantification during the aging process

The quantification of the major alkaloids enabled the representation of L. angustifolius Bo9027 and L. mutabilis PI 457964 alkaloid profiles. During the generative phase, lupins can display all pod and seed aging stages simultaneously within a single plant, especially when considering branched and indeterminate varieties.

In other studies, the entire plant or specific parts were collected after fixed intervals, which might lead to an inhomogeneous pool of pods or seeds from different ripening stages. To overcome this limitation, here, phenotypical aging stages were used to guarantee harvest of biological samples from identical ripening stages and to facilitate a meaningful comparison between the two lupin species (for identification of the stages, see Table 1 and Figure 3). This also eliminates issues of varying ripening process speeds in seeds in different species and harvest generations. For better comparability, the alkaloid profiles were normalized to dry weight (mg/g dry weight (DW)). Seeds from 2024 of L. angustifolius have been used to compare the method with values obtained from the literature. In contrast, L. mutabilis was harvested for two consecutive years (2024, 2025), ensuring biological reproducibility of the results.

L. angustifolius had a less diverse major alkaloid profile than L. mutabilis and only four major alkaloids could be detected. These were lupanine (13), 13α-hydroxylupanine (16), angustifoline (19), and isolupanine (20). In the pods, lupanine (13) titers were reduced drastically in the early stages, while isolupanine (20) and 13α-hydroxylupanine (16) stayed mostly unaffected. A notable exception is stage 2, during which an increase in 13α-hydroxylupanine (16) was detected. Angustifoline (19) amounts also decreased in pods, while maintaining their content in seeds during development. In the seeds, the amount of lupanine remained the same. The experiments showed increased 13α-hydroxylupanine (16) and angustifoline (19) and decreased isolupanine (20) content over the course of the aging steps (see Figures 4E,F). The stagnation of lupanine is in accordance with previous studies (Otterbach et al., 2019). For comparison to Otterbach et al. (2019), the supporting information provides diagrams with alkaloid content per seed rather than dry weight (see Supplementary Figures 35–37). The profiles are similar in both studies for alkaloids lupanine (13) and 13α-hydroxylupanine (16) shared by Otterbach et al. (2019) and in this study.

For L. mutabilis, in addition to the alkaloids in L. angustifolius, sparteine (14), 3β-hydroxylupanine (15), tetrahydrorhombifoline (17), and hydroxytetrahydrorhombifoline (18) were detected, while angustifoline (19) was undetectable. The early stages have very similar alkaloid profiles for both seeds and pods (see Figures 4A–D). Lupanine (13) and its hydroxylated forms 3β-hydroxylupanine (15) and 13α-hydroxylupanine (16) were the main components of the early stages of the ripening process. During the growth of the seed in stages 2–3, the content of lupanine (13) reduced in both pods and seeds, while 13α-hydroxylupanine (16) appeared to be unaffected. With this reduction in lupanine (13) and lupanine derivatives, the amount of sparteine (14) increased and reached its maximum content in stages 2–3. Tetrahydrorhombifoline (17) and hydroxytetrahydrorhombifoline (18) were rather unaffected during the entire ripening process in seeds but showed reduced content in pods. In stage 3, the maximum weight of the seeds is reached, while the pods started decoloring (see Figure 3). During the drying stages 4–5, the pod alkaloid profile started to differ vastly from the seed alkaloid contents. Lupanine (13) and 3β-hydroxylupanine (15) were reduced in the seeds, and sparteine (14) remained constant. The alkaloid content of the pods was diminished in stage 5. The detection in both years (2024, 2025) for L. mutabilis displayed biological variation in the absolute alkaloid content, but the development of the alkaloid profiles over the course of the aging stages showed reproducible trends in both years. Thus, in the later analyses, these data were combined.

Minor QA analysis

The minor alkaloid analysis was performed using the library provided in SI on the basis of NIST20 (NIST, 2020). The data were acquired based on alkaloid profiles measured by GC-ISQ-MS, whilst exact m/z-ratios were obtained by GC-ToF-MS. In total, 24 putative alkaloids could be identified from the NIST dataset according to exact masses and RI values, respectively. Although an examination of all distinguishable peaks was performed, the library available in the SI might not represent every alkaloid present in L. mutabilis and L. angustifolius and does not distinguish between isomers. Ten putative alkaloids (21 – 30) were detectable in L. mutabilis by GC-ISQ-MS throughout all aging samples (see Table 3). These were clustered by comparison with the NIST database, which gave the most probable QA-type (NIST, 2020). Because no response factor for quantification is available for all these entities, further analyses must be considered semi-quantitative. The obtained data can be found in the Supplementary Table S5. In Figure 8, the normalized abundances have been joined based on the profile similarity. Foremost, data obtained from pods and seeds cluster per tissue except in the case of compounds 21, 25, and 26. Thus, it can be concluded that there are physiological differences between pods and seeds. Furthermore, it is clear that putative minor alkaloids obtained from pods are more abundant in the early stages (1–1.5), whereas seeds contain more of these compounds in the medium (2) or late (3–5) stages of ripening. The esterified, lupanine-type QAs (21) as well as the compounds 25 and 26 resemble the same pattern with a strong abundance in the early stages (<2).

Differential gene expression across tissues and developmental stages

After detection of alkaloid titers, transcriptomics was utilized to further investigate their biosynthetic origin. To gain insight into transcriptional regulation during early pod and seed development in L. mutabilis PI 457964, mRNA sequencing was performed. As illustrated in Figures 4, 5A; Table 3, seeds and pods at stages 1.5 and 2 exhibited varying concentrations of major and minor QAs. These two stages were intentionally selected for two reasons: (a) it was assumed that both stages are highly active in terms of biosynthesis, and (b) they exhibit different alkaloid profiles. A principal component analysis was performed, showing clear clustering of the four replicates and separation by stage and tissue, reinforcing their phenotypical selection (see Supplementary Figure 47). Overall, 93% of the variance was explained by two principal components. At stage 2, seeds were more strongly separated from pods than at the earlier stage 1.5. Differential gene expression was analyzed in four comparisons, contrasting tissues (pods and seeds) and developmental stages (stage 1.5 and stage 2). Differentially expressed genes (DEGs) were defined as those with significant differences in expression levels between sample groups, based on FDR <0.05 and |log 2 FC| ≥ 2. The results revealed 3,708 DEGs in pods (see Supplementary Table S1, sheet 3) and 4,027 DEGs in seeds (see Supplementary Table S1, sheet 2) comparing the two stages (see Figure 5B). Tissue-specific differences were more pronounced, with 5,587 DEGs differentiating pods and seeds at stage 1.5 (see Supplementary Table S1, sheet 4), and 6,946 DEGs at stage 2 (see Supplementary Table S1, sheet 5; Figure 5B). A Pearson’s correlation between quantified gene expression (transcripts per million, TPM) and QA concentrations resulted in large numbers of genes with high and significant correlations. With current knowledge, the possible candidates were too numerous to analyze. This necessitates further investigations into known components of the QA biosynthesis pathway. Pearson’s correlation coefficient results are presented only for the DEGs (see Supplementary Table S1, sheets 2–5). In addition, Pearson’s correlation coefficients for notable transcription factors (see Supplementary Table S1, sheets 6–9) and transporters (see Supplementary Table S1, sheets 10–13) associated with stage- and tissue-specific differences were also provided.

Therefore, 33 genes (see Supplementary Table S1, sheet 1) known to be involved in biosynthesis, transport, and regulation of quinolizidine alkaloids were specifically investigated, identified through homology with L. angustifolius and L. albus as outlined in Subsection 3.2.5 (Yang et al., 2017). The genes responsible for the initial enzymatic steps in QA biosynthesis, lysine decarboxylase (LDC) and copper amine oxidase (CAO), did not show statistically significant changes over time within either tissue (from stages 1.5–2). However, the LDC and CAO genes mikado.chr19G51 (homolog 1) and mikado.chr19G1289 (homolog 1) displayed strong tissue-specific enrichment in pods compared to seeds. They were upregulated in pods with log 2 fold change values of ∼8–9 at stage 1.5 and ∼16–18 at stage 2 (see Supplementary Table S1, sheet 1). Similarly, cinnamoyl-CoA reductase 2 genes (CCR2), involved in plant secondary metabolism, present comparably increased expression in pods. Two acetyltransferase (AT) genes, homologous to LaAT, exhibited strong pod-enriched expression in stage 1.5 and reduction of expression in seeds and pods with development from stage 1.5 to stage 2 (see Supplementary Table S1, sheet 1). The 13-hydroxylupanine O-tigloyltransferase (HMT/HLT) gene mikado.chr5G47 was enriched in pods, also showing some increased expression in seeds in stage 2 compared to stage 1.5. In contrast, 4-hydroxy-tetrahydrodipicolinate synthase (DHDPS) genes, encoding a central enzyme in the primary lysine biosynthesis pathway, showed variable regulation that did not align with the expression patterns of specific QA-biosynthetic genes (LDC/CAO) (Kroc et al., 2019).

Putative transporter genes associated with the QA pathway, specifically purine permease 1 [ PUP1 (Hildreth et al., 2011; Plewiński et al., 2019; Martinez-Hernandez et al., 2025)] homologs mikado.chr12G757, mikado.chr12G1284, mikado.chr19G396, and mikado.chr2G1111 were significantly more abundant in pods than in seeds, with fold changes exceeding log 2 FC from 6 to 11 at both stages. PUP1 genes also showed a reduction in expression with the development of the seeds and pods in stage 2. Regulatory candidates followed a similar trend as LCD and CAO; the AP2/ERF transcription factor RAP2-7 and the transcription factor MYB10 6 were consistently enriched in pod tissue with log 2 FC changes from 6 to 8.

The investigation of pod and seed maturation in L. mutabilis was performed using the major QA titers, the qualitative content of the ten minor QAs, and a gene expression study. Previous studies, such as Otterbach et al. (2019), established a pod-to-seed QA transport hypothesis for L. angustifolius, and Namdar et al. (2024) recently revisited QA biosynthesis in wild lupin species. This work extends the understanding by adding species coverage with L. mutabilis, investigating seed and pod tissue throughout seed development and ripening, integrating transcriptomics and data from two growing seasons. In general, the absolute content of QA varies between stages, organs, and experimental biological repetitions. The difference in seeds compared to pods increases during the aging process, verifying the role as storage tissue. The pods reduce their alkaloid content during ripening with distinct differences between the green phenotypical stages 1–3 and stages 4–5, correlating with their decrease in chlorophyll and water content (see Figures 3, 4). The same phenotypical subclassification can be attributed to the seeds. Although alkaloid titer variance is persistent until stage 5, the major accumulation can be attributed to stages 1–3, with green and putative photosynthetically active seeds (see Figure 3). This suggests the hypothesis that either transport phenomena between the organs or the biochemical transformation of the alkaloids are causally related. One effect is also evident in a rough examination of the alkaloid profiles; for example, in L. mutabilis, the earlier stages within the pods show lupanine (13) and 3β-hydroxylupanine (15), while the older stages contain reduced amounts (see Figures 4B,D). For this reason, the data from stages 1–3 were pooled, and the data from stages 4 and 5 were combined. Pearson’s correlation analyses were performed for all pairs of alkaloid contents within the respective data pools. This was done for pods and seeds separately, with seeds represented in the lower left part, while correlations in pods were placed in the upper right part of the graphs (see Figures 6, 7). These datasets include the technical replicates (N = 3) and also the biological replicates from both growing years. The results are represented in Figure 6 (age groups 1–3) and Figure 7 (age groups 4 and 5). The same analysis was done for the samples of L. angustifolius for the available data of 2024 solely (supplied in the Supplementary Figures 45,46).

The correlations of the alkaloids differed between both phases in L. angustifolius and L. mutabilis (see Figures 6, 7; Supplementary Figure 45,46). In stages 1–3, L. angustifolius, lupanine (13) correlates positively with angustifoline (19) in pods and seeds. In both organs, 13α-hydroxylupanine (16) correlates positively with lupanine only in stages 4–5, while no correlation between the alkaloids can be detected during the growing stages. The correlation between 13α-hydroxylupanine (16) and angustifoline (19) changes from positive to negative in the pods, while it changes from negative to positive in seeds (see Supplementary Figures 45,46). In contrast, isolupanine (20) correlates negatively with all other alkaloids in early pod stages, while positive correlations with its epimer (13) and angustifoline (19) can be seen in early seeds. The correlation with 13α-hydroxylupanine (16) changes from negative to positive in seeds over development (see Supplementary Figure 46). The data hint at stronger transport processes taking place into seeds in later stages, while negative correlations in earlier stages might indicate derivatization in seeds and pods alike. This is coherent with Otterbach et al. (2019), as they also describe accumulation of high titers of lupanine throughout the pod-seed complex maturation. A direct comparison to the previously published result is shown in Supplementary Figures S3, S7. As expected, there are differences in the analyzed alkaloid profile of the two investigated species. For example, angustifoline (19) is found exclusively in L. angustifolius, while sparteine (14), 3β-hydroxylupanine (15), tetrahydrorhombifoline (17) and hydroxytetrahydrorhombifoline (18) are found exclusively in L. mutabilis. The diversity of alkaloids presented here reveals a greater decrease in lupanine during the ripening process than is observed in L. angustifolius.

Lupinus mutabilis has the more complex major alkaloid profile. However, the correlation of almost all alkaloids was positive or very minor throughout both ripening stages, with the exception of lupanine (13), which had negative correlations in both lupin species. The alkaloid values of the most important alkaloids in stage 1.5 are identical in both the pods and the seeds, suggesting that the seeds and pods originate from a common precursor organ (see Figures 4A–D). The major alkaloids present early in development are lupanine (13) and 3β-hydroxylupanine (15). Starting the ripening process, the relation between lupanine (13) and sparteine (14) is negative in pods and seeds during their biosynthetic activity in both years (see Figure 6). This might suggest a precursor hypothesis. Isolupanine (20) has the lowest content of all major alkaloids. It correlates positively the strongest with lupanine (13) and negatively with sparteine (14) in stages 1–3, leading to the hypothesis that the same enzyme might be responsible for both. This putative enzyme might not be stereoselective toward these diastereomers. This enzyme would be downregulated later, explaining the late sparteine (14) accumulation in both pods and seeds alike. The hydroxylation of lupanine (13) through a putative monooxygenase might lead to 3β-hydroxylupanine (15) and 13α-hydroxylupanine (16). There must be an additional source, such as transport, because lupanine (13), 3β-hydroxylupanine (15), and 13α-hydroxylupanine (16) show a very strong positive correlation, which is in accordance with the transport hypothesis posed by earlier research (Lee et al., 2006; Otterbach et al., 2019).

An additional explanation for this phenomenon might be that esterified lupanines (21), which are present at the beginning of the seed ripening process according to Figure 8, might hydrolyze to form the hydroxyl group. Tetrahydrorhombifoline (17) and hydroxytetrahydrorhombifoline (18) accumulate in seeds, although with small titers compared to the other alkaloids. It cannot be distinguished whether these compounds are produced selectively or as a side product from promiscuous (oxidative) enzymes. Being an artifact of the extraction can be ruled out, because these compounds did not occur in extractions of L. angustifolius (see Supplementary Figure 45,46). In stages 4–5, the reliance on import rather than local biosynthesis becomes apparent. As the alkaloid content in the pods reaches its minimum, all alkaloids correlate positively, which is in accordance with the published hypothesis of storage and drying (water loss, color changes, no bioactivity) (Otterbach et al., 2019). This also applied to L. mutabilis, for which bioactivity must have been decreased severely in stages 4–5, as it is monitored by the positive correlations of all alkaloid titers.

As is usual with secondary metabolisms, the late-stage derivatization of compounds is diversified to combat different threats (Firn and Jones, 2003). Based on this hypothesis, lupin alkaloid profiles should consist of some major and many minor alkaloids representing less than 2% of the total alkaloid content. Applying a library of compounds with exact masses gained from GC-ToF-MS measurements (the library is supplied in a repository; https://doi.org/10.26165/JUELICH-DATA/XNAIFH), a typification of the presenting minor compounds was performed (see Supplementary Table S8). Using the fragment patterns of GC-ISQ-MS (see Supplementary Figures 48–57), the compounds were clustered by their quadratic Euclidean distances, pairing similar appearances (see Figure 8). Generally, the data from pods and seeds clustered together, indicating their similarity to the same alkaloids generally appearing in the pods first. Esterified lupanines (21) are an exception to this rule. In the literature, it is hypothesized that esterified lupanines (21) are the transport form of QAs (Lee et al., 2006). This correlation suggests that their transport mechanism is direct, without other locations, or that the biosynthesis of esterified QAs occurs simultaneously in both organs in the early stages. This is supported by the fact that both organs are photosynthetically active at this point. Furthermore, a cluster of each lupanine- (24 – 26) and sparteine-type (27 – 29) can be observed in the pods. While lupanine-type (24 – 26) alkaloids appear rather early, sparteine-type alkaloids appear later, supporting the postulated derivatization of sparteines to lupanines by enzymatic derivatization. However, where the derivatization takes place is not known, as both types appear to transfer from pods to seeds. The hydroxylated product (22) appears in pods first and slowly accumulates in the seeds.

As deduced from experiments presented in this work and previous studies, a biosynthetic pathway hypothesis was postulated (see Figure 9).

As both lupanine (13) and isolupanine (20) exist, the ring-closure mechanism proposed in the literature may lead to a predetermination toward either lupanine (13) or its epimer (20). The negative correlation of lupanine (13) and sparteine (14), as well as structural advantages, leads to sparteine (14) as a precursor accumulating after the responsible monooxygenase is downregulated. The product lupanine is then either hydroxylated or the available esterified lupanines are hydrolyzed to enable an accumulation of hydroxylupanines (15, 16). Additionally, transport mechanisms from pods to seeds increase the titer in the latter tissue, as is also suggested by the transcriptome analyses, while an additional import from other green plant organs is likely. From the hydroxylated QA in the 13-position (16), a ring opening is likely leading to tetrahydrorhombifoline (17) and hydroxytetrahydrorhombifoline (18). As the correlation analysis does not highlight a common temporal envelope, shared transcriptional regulation, or dilution by biomass accumulation separately, these factors could also be responsible for parts of the correlation drifts.

A gene expression study was conducted to further validate the results. Comparison of transcriptomes of seeds and pods at stage 1.5 and 2 reveals extensive changes due to tissue and development. This provided indications for the putative transport mechanisms hypothesized in Figure 9. The expression of key candidate genes is summarized in Figure 10 and visualized as transcripts per million values (TPM). Their differential expression profiles are provided in Supplementary Table S1, sheet 1.

The differential gene expression profiles of the identified QA biosynthesis-related genes indicated a difference in biosynthetic activity in seeds compared to pods. The substantial enrichment of LDC and CAO transcripts in pods relative to seeds, which is a phenomenon observed across the investigated developmental stages, identified the pod as the predominant biosynthetic “source” for the QAs even in early stages. This indicated scaffold building in the pods, which could then transfer either sparteine or lupanine as precursor molecules. These might then be derivatized in all green tissue (seeds and pods). This pattern was also congruent with Lupinus angustifolius, in which LDC expression peaks early in pods and was very low in developing seeds, while QA contents accumulated later in seeds (Otterbach et al., 2019; Frick et al., 2025). The genome-wide transcriptomics allowed a broader investigation of L. mutabilis homologs to known genes involved in regulating, catalyzing, and transporting QAs (see Supplementary Table S1, sheet1).

The L. angustifolius gene AT encoding an acetyltransferase was identified as the gene with a causal mutation in the pauper region for sweet seeds without QAs (Mancinotti et al., 2023). The identified homologs also showed stronger expression in pods than in seeds at stage 1.5, supporting the scaffold import hypothesis. The cinnamoyl-CoA reductase 2 (CCR2), a candidate for alkaloid biosynthesis regulation in other lupin species, followed a similar expression pattern to LDC and CAO as described before (Czepiel et al., 2021). In contrast to the scaffold hypothesis, the expression of predicted QA regulators, the AP2/ERF transcription factor RAP2-7, along with the MYB -family candidate MYB106 (Kroc et al., 2019; Yang et al., 2017), being more abundant in pods, collectively indicated that both the core biosynthetic reactions and also subsequent modifications of QAs occur predominantly in pods at least in the aging stages 1.5 and 2. However, QA biosynthesis and regulator genes were expressed in the early stages in seeds to a lower amount, which might enable a smaller fraction of QA biosynthesis occurring in seeds as described in L. albus in earlier studies (Lee et al., 2006). Several downstream enzyme candidates, including the 13-hydroxylupanine- O -tigloyltransferase (HMT/HLT), showed presence in both organs with a surplus in the pods, which must lead to an increase in esterified lupanines (Okada et al., 2005), which are not detected in either seed or pod. Thus, the production of esterified lupanines either led to import and hydrolysis in the seeds as represented in Figure 9 or an export into other plant organs. Other pathway-related genes showed more nuanced patterns. One of the carboxylesterase 1 (CES1L) genes (mikado.chr2G27), involved in secondary metabolism and potentially in QA regulation, displayed its highest fold change in pods at stage 2, while mikado.chr11G658 shows reduced expression compared to the seeds. 4-Hydroxy-tetrahydrodipicolinate synthase (DHDPS), a key enzyme in lysine biosynthesis, showed expression patterns that do not correlate with QA levels, likely reflecting general metabolic demand for lysine rather than QA-specific regulation. Despite a decline in total QA content in pods during the development from stage 1.5 to stage 2, the sustained high expression of biosynthetic genes indicates a state of elevated metabolic turnover driven by rapid export. However, it cannot be distinguished whether all alkaloids or only scaffolds for late-stage derivatization are imported. In accordance with the literature, the seeds in L. mutabilis predominantly acquire QAs through transporter-mediated uptake rather than through de novo biosynthesis (Otterbach et al., 2019). This hypothesis is eventually supported by the expression of PUP1 transporters being enriched in pods over seeds in both stages.

This work elucidates L. mutabilis PI 457964 QA metabolism during seed and pod maturation. With the accumulation of valuable compounds like sparteine and lupanine, the line contains valuable compounds for putative side stream applications. The analysis showed a clear indication of a consecutive biosynthesis starting with sparteine derivatized to lupanine, which branches to multiple minor QAs. The accumulation profiles and gene expression analysis indicated that QA biosynthesis is not limited to single tissues, but rather a compilation of different import and biosynthesis processes leading to QA accumulation. The investigations into the minor alkaloids led to the finding that mostly structurally similar alkaloids accumulate and transfer at similar times. For all alkaloids, the route of transfer is from pod to seed, although the early stages of green seeds must also be biosynthetically active to perform late-stage modifications.