Soybean seeds of 17 soybean cultivars were generously provided by Dr SK Lal, Division of Genetics, ICAR-Indian Agricultural Research Institute, New Delhi, India. The genistein and daidzein standards were procured from Sigma-Aldrich, Merck (Burlington, MA, USA). TRIzol and PrimeScript^TM 1^st strand cDNA synthesis kits were purchased from TaKaRa Bio Inc. (Shiga, Japan). The pENTR^TM/D-TOPO^TM vector was purchased from Thermo Fisher Scientific (Waltham, MA, USA) and the SYBR^R Green JumpStart^TM Taq ReadyMix^TM was bought from Sigma-Aldrich, Merck. Thiobarbituric acid (TBA), 2,4-dinitrophenylhydrazine and other reagents were purchased from Sisco Research Laboratories Pvt. Ltd. (SRL) (Mumbai, India).

Isoflavone mass fraction and composition were determined using HPLC. The sample extract, standard curve and HPLC conditions were prepared as per Kumar et al. (18) with slight modifications. A mass of 125 mg of ground soybean seeds was weighed and extracted using 5 mL ethanol and 1 mL concentrated H₂SO₄ for 2 h in the boiling water bath. The extracts were centrifuged (Heraeus Multifuge X3R; Thermo Fisher Scientific) at 10 000×g for 10 min, and the supernatant was neutralised using 10 M NaOH. After syringe filtration (0.45 µm), the samples were injected into the HPLC. The chromatography was performed in a Waters 2695 chromatograph (Waters Corporation, Milford, MA, USA) with a 2998 photodiode array detector and a Phenomenex Luna C₁₈ (250 mm×4.6 mm, 5 µm; Torrance, CA, USA) analytical column as the stationary phase. A binary gradient of acetonitrile (ACN) and water was used as the mobile phase with a flow rate of 1.5 mL/min. For the first 1 min of HPLC separation, 13 % ACN was used, with the ACN content reaching 30 % at 20 min and again reducing to 13 % at the end of the run, i.e. 25 min. The absorbance was measured at 254 nm using spectrophotometer (EVOLUTION One Plus; Thermo Fisher Scientific) and the relative mass fraction of isoflavones was determined by CSW17 software (Data Apex®, Prague, Czech Republic) and expressed as µg/g of seed dry mass.

Malondialdehyde (MDA), a degradation product of lipid hydroperoxides, is produced by lipid peroxidation. The MDA formed in soybean samples was measured as TBA value based on the adduct development after the reaction of MDA with TBA. The adduct gives a pink colour with the absorbance measured at 532 nm (19). Approximately 400 mg of soybean seeds soaked overnight were homogenised with 4.0 mL of distilled water, followed by centrifugation at 13 000×g for 35 min to obtain the clear extract. Then, 0.2 mL of this extract was mixed with 0.2 mL of 8.1 % sodium dodecyl sulphate, 1.5 mL of 20 % acetic acid (pH=3.5) and 1.5 mL of 0.8 % TBA and kept in a boiling water bath for 60 min. After cooling, 1.0 mL of distilled water and 5.0 mL of n-butanol pyridine (15:1) were added, and the tubes were shaken vigorously. The mixture was then centrifuged at 4000×g for 10 min. The pink-coloured organic layer formed at the top was separated from the aqueous phase and absorbance was measured at 532 nm. The standard curve was prepared using 1,1,3,3-tetraethoxypropane ranging from 10 to 70 nmol and the concentration of lipid peroxides was expressed in nmol of MDA per g fresh mass.

The concentration of carbonyl compounds was determined using the method described by Henick et al. (20). The 2,4-dinitrophenyl hydrazine reacts with the aldehydes and ketones to form their 2,4-dinitrophenylhydrazone derivatives, which are converted to wine red-coloured complexes by KOH under alkaline conditions. The samples were extracted (0.5 mL) in the same manner as for the TBA value, then mixed with 0.5 mL of 2,4-dinitrophenylhydrazine (0.05 %) and heated in a water bath at 60 °C for 30 min. After cooling, 1.0 mL of KOH (4 %) was added, and a characteristic wine-red colour appeared. Absorbance was measured at 480 nm 10 min after the appearance of the wine-red colour and the concentration was determined with the following formula:

where A is absorbance, ε_max=2.72·10⁴ L/(mol·cm), c is concentration (mol/L) and l is path length (cm).

The quantitative real-time gene expression analysis of isomers IFS1 and IFS2 was performed on days 35, 45 and 55 after flowering (DAF) in contrasting soybean genotypes grown in the open field at ICAR-IARI, New Delhi, India. Total RNA was isolated using the TRIzol^® method (21) and the first strand of cDNA was synthesised from it. All the primers for quantitative polymerase chain reaction (qPCR) (~150 bp) were designed using PrimerBlast (22) available at NCBI (http://www.ncbi.nlm.nih.gov/tools/primer-blast/). The housekeeping gene actin 2/7 was used as an internal control for data normalisation (Table S1). The qPCR was carried out using a Bio-Rad CFX96 machine (Bio-Rad, New Delhi, India) in a reaction volume of 20 mL containing SYBR^® Green JumpStart^TM Taq ReadyMix^TM (Sigma-Aldrich, Merck), forward primer (10 mM), reverse primer (10 mM) and 10 ng/mL of cDNA using conditions of 94 °C for 5 min, then 40 cycles at 94 °C for 15 s, 55 °C for 30 s and 72 °C for 45 s. The relative expression level was calculated following the 2^-ΔΔCt method (23). Standard errors and standard deviations were calculated from three technical and biological replicates. Seed samples 35 DAF were considered the calibrator. The resulting PCR products were also analysed by agarose gel electrophoresis to determine the specificity.

The total RNA was isolated from soybean seeds using RNAiso plus (TaKaRa Bio Inc.) following the manufacturer's protocol. The full-length IFS1 and IFS2 coding sequences were amplified from the soybean cultivar DS2706. Both forward and reverse primers were designed using PrimerBlast (22). The amplified PCR products were eluted from the gel using a Quick Guide PCR purification kit (SolGent, Daejeon, Republic of Korea) and kept for ligation in a pENTR/D-TOPO vector, an entry vector for the Gateway® System, for 30 min at 23 °C in a molar ratio r_{insert,vector}=2:1. A volume of 3 µL of ligation mix was transformed into TOP10 chemically competent E. coli cells, and 6 colonies from each clone were inoculated for plasmid isolation. PvuII restriction digestion and PCR amplification confirmed the cloning of the cDNA sequence. The confirmed clones were subjected to commercial Sanger sequencing (Chromus Pvt. Ltd., Bengaluru, India) to discern their nucleotide sequence.

The nucleotide and translated amino acid sequences of isomers IFS1 and IFS2 were compared using NCBI BLAST (https://blast.ncbi.nlm.nih.gov/Blast.cgi accessed on 9 June 2022) set to default parameters (24). The three-dimensional structures of the IFS1 and IFS2 proteins from their amino acid sequences were generated using RoseTTAFold (25), a deep learning-based structure prediction tool (https://robetta.bakerlab.orgaccessedon4July2022). This nonhomology-based algorithm works by detecting patterns in protein sequences, interactions among the amino acid residues and probable protein tertiary structures, simultaneously determining everything. The Ramachandran plots of predicted models were generated on the RoseTTAFold server, where the favoured rotamers and Ramachandran plot distribution with φ and ψ angles were analysed (26). The interactions of IFS1 and IFS2 with other proteins were studied using STRING (https://string-db.org/ accessed on 11 January 2023) (27), a database containing known and predicted protein-protein interactions.

The mass fraction and composition of isoflavones in soybean genotypes vary widely (28), with malonylated derivatives of isoflavone glucosides, i.e. genistin and daidzin being the most abundant (29). The health benefits, bioavailability and LOX inhibition potential of various isoflavones could also vary with structural characteristics. Due to the prevalence of these substituted derivatives, these isoflavone glucosides were converted into their corresponding isoflavone aglycones by acid hydrolysis.

Seventeen different soybean genotypes were analysed for total isoflavone content and composition to investigate their relationship with indices of off-flavour development in soybeans. Enormous differences in the total isoflavone content and the mass fraction of individual forms of these isoflavones were found among cultivars (Table 1 and Fig. S1). The total isoflavone mass fraction was expressed in μg/g of soy flour, ranging from (153.5±7.2) µg/g for PUSA 40 to (1146±44) µg/g for Bragg. Bragg and PUSA 40 were identified as the soybean genotypes with high and low isoflavone mass fraction, respectively, and were selected for gene expression analysis of isoflavone synthase (IFS). Similar studies elsewhere have also found enormous differences in isoflavone content and composition among soybean genotypes grown in different ecoregions (28). PK 1042 and PK 416 were identified as the soybean genotypes with a maximum (1.64) and minimum (0.33) genistein/daidzein mass ratio, respectively. This is significant as genistein is the isoflavone with higher antioxidant activity than daidzein (16, 30). The analysed genotypes showed enormous variation in both isoflavone content and composition, which can prove useful for developing isoflavone-rich soybean varieties through breeding programmes.

The TBA value is a well-established measure of lipid peroxidation (31). It expresses the lipid hydroperoxides formed during oxidation in nmol per g of seed. The TBA value in soybean genotypes ranged from (46.7±4.2) nmol/g for BS1 to (307.3±8.1) nmol/g for UPSL 340 (Fig. 1). Similarly, the carbonyl value measures the amount of aldehydes and ketones associated with the formation of off-flavours. Its values were expressed in nmol per g of seed and ranged from (379±5) nmol/g for EC 109514 to (674±5) nmol/g for UPSL 340 (Fig. 2). The extent of correlation between the isoflavone mass fraction and the parameters used as indices of off-flavour was measured using Pearson's correlation coefficients. The observed value for the correlation coefficient between the total isoflavone content and the carbonyl value was 0.089, and that of the TBA value and total isoflavone content was 0.016 (all statistically insignificant at p<0.1) (Table 2). As mentioned above, genistein has been reported to have the highest antioxidant activity among the various soy isoflavones. Therefore, the correlation of particular forms of isoflavones with these parameters is expected to be different. Hence, the correlation coefficients of the parameters determining off-flavour were also calculated for the genistein and genistein/daidzein mass ratio. The correlation between the genistein content and the carbonyl value was -0.12, and the correlation between the TBA value and the genistein content was -0.15 (not statistically significant at p<0.1). In contrast, the correlation coefficients for the genistein/daidzein mass ratio for the TBA and carbonyl values were -0.41 and -0.42, respectively, indicating a weak but statistically significant correlation (as both values are statistically significant at p<0.1 but not significant at p<0.05). The higher antioxidant potential of genistein than daidzein can be the reason for this observation (32). Genistein is also a more potent inhibitor of soybean LOX, with a half inhibitory concentration (IC₅₀) value of 107 µM compared to 140 µM for daidzein (16, 33). Previously, Dahuja and Madaan (9) had also observed an inverse relationship between the contents of antioxidant enzymes and parameters determining off-flavour such as TBA and carbonyl values. Therefore, the ability of genistein to inhibit LOX and to be a general antioxidant seems to be responsible for a reduction in the indices of off-flavour formation. However, isoflavones themselves can contribute to bitterness by binding to taste receptors (34–36). Since the beany off-flavour is mainly caused by lipid peroxidation, a delicate balance must be achieved in the quantity and type of antioxidants.

Isoflavone synthase (EC 1.14.14.87) is the key enzyme of isoflavone biosynthesis responsible for the synthesis of isoflavone aglycone backbones. There are two different isoforms of IFS in soybean. To determine the relative contribution of these isoforms to isoflavone accumulation in soybean seeds, we carried out gene expression analysis of IFS1 and IFS2 in two contrasting soybean genotypes, Bragg (high isoflavone content) and PUSA 40 (low isoflavone content) at three stages of seed development. Expression profiling suggests that the expression level of IFS1 increased with the progress of the developmental stage in PUSA 40 (Fig. 3). However, a nonlinear and irregular trend in the expression of IFS1 was observed in Bragg; it first increased 2.34-fold from 35 to 45 days after flowering (DAF) and then decreased 2.22-fold until 55 DAF. An increasing trend of IFS2 expression was observed in Bragg, but the expression level decreased with each developmental stage in PUSA 40. In quantitative terms, an overall 9.34-fold increase in IFS2 expression was observed in Bragg on 55 DAF compared to its expression on 35 DAF. In comparison, the gene expression in PUSA 40 was reduced to approx. 6 % of the value on 35 DAF.

The corresponding mass fractions of the total isoflavone in each stage were also determined using HPLC. The total isoflavone mass fraction in Bragg remained higher than that in PUSA 40 in all three stages of seed development. In addition, the total isoflavone mass fraction in Bragg increased from 14.07 to 78.4 µg/g on 55 DAF, but in PUSA 40, this increase was only approx. 2.19-fold from 35 DAF to 55 DAF.

Dhaubhadel et al. (37) observed the accumulation of IFS2 transcripts in the later stages of seed development, but IFS1 expression remained constant throughout. Gutierrez-Gonzalez et al. (15) also observed a more significant increase in IFS2 expression of up to 20-fold 70 days after pollination compared to 30 days after pollination during seed development. IFS1, in contrast, increased only 4-fold in the same period.

The expression of IFS2 matched the pattern of isoflavone accumulation in developing soybean seeds. However, no such relationship could be observed with the expression levels of IFS1, as the two contrasting cultivars, despite having similar expression levels of IFS1, differed significantly in the isoflavone mass fractions in all the stages of seed development, with the difference in the isoflavone being more pronounced on 55 DAF. Thus, IFS2 may directly affect isoflavone accumulation during the grain-filling stage.

The qRT-PCR and the analysis of isoflavone content indicated a putative role of isoflavone synthase in the accumulation of isoflavones in soybean seeds. Considering this, these cloning and in silico analyses of isoflavone synthase isoforms were performed. Specific primers for polymerase chain reaction (PCR) amplification of IFS cDNA genes were designed based on the sequence information available at NCBI (https://blast.ncbi.nlm.nih.gov/Blast.cgi accessed on 9 June 2022). The PCR products were run on an agarose gel, and the approximate size of the amplicon was 1566 bp (Fig. S2) for both IFS1 and IFS2. The obtained PCR products were then eluted from the gel and cloned into the pENTR/D-TOPO vector. The cloning of IFS1 and IFS2 was confirmed by plasmid PCR and restriction digestion (PvuII). The PvuII restriction enzyme has two cut positions (174 and 812) in the TOPO entry vector. We found linear bands of 2204 bp (1566 bp insert+638 bp vector) and a 1942 bp vector backbone in both IFS1 and IFS2. The gene sequences of the cloned products were submitted to the National Center for Biotechnology Information (NCBI) GenBank (Bethesda, MD, USA) under accession numbers KP843618 and KT581120). The 1566 bp long IFS1 and IFS2 coding sequences showed an identity of 99.04 % with difference in 15 nucleotides (Fig. S3). The amino acid sequences of IFS1 and IFS2 were also aligned and showed an identity of 98.27 %, with a difference of 9 out of 521 amino acid residues (Fig. S4). The 3D models of both IFS1 and IFS2 were generated using RoseTTAFold (Fig. 4). The Ramachandran plot for IFS1 and IFS2 showed over 94.8 and 96.5 % of the residues in the most favoured region, respectively (Fig. S5). The IFS isoforms showed minor differences in their tertiary structures. Nevertheless, their predicted interactions with other proteins differed significantly. Interestingly, in the interaction study performed using STRING, IFS2 showed a maximum interaction with chalcone isomerase (CHI) with a score of 0.933 (Fig. 5 and Fig. S6), which was not the case for IFS1. In the phylogenetic tree prepared using the translated IFS2 protein sequence (Fig. 6, highlighted in yellow), the IFS1 sequences were grouped separately, and more than one copy of IFS was observed in many other legume species.

Dastmalchi et al. (38) studied the interactions and subcellular localisation of the most important enzymes in isoflavone biosynthesis. They found the interaction of chalcone reductase (GmCHR14) with GmIFS2 but not with GmIFS1, which is consistent with the results obtained in this study. The GmCHR14 is upstream of isoflavone synthase in the phenylpropanoid pathway and is responsible for isoflavone biosynthesis. In a recent study, a CRISPR-mediated knockdown of IFS1 was performed and nearly all the physiological processes in the mutants were intact, indicating a role of IFS2 (39). However, a curious finding of this study was the greater effect of IFS1 knockdown on genistein rather than daidzein accumulation, which could explain the higher amount of daidzein in Bragg than genistein. In view of this, IFS2 appears to be better suited for the formation of a metabolon (a multienzyme complex), which may lead to efficient and effective substrate channelling and thus to a higher accumulation of isoflavones through this pathway. Thus, overexpression of IFS2 can increase isoflavone mass fraction.