Thirty-six rice varieties (Oryza sativa L. ssp. indica) used in this study, provided by Rice Science Center, Kasetsart University, were grown in July 2022 during the rainy season. They were classified based on biochemical analysis (48, 49) and allele-specific SNP markers into waxy (0–12%), low (12–20%), and high (25–33%) AC (Supplementary Tables S1, S5). The following were varieties included in each group.

Dietary fiber was determined by measuring carbohydrates with a degree of polymerization (DP) of more than two, not hydrolysable by endogenous digestive enzymes. The in vitro analysis uses heat-stable α-amylase, protease, and amyloglucosidase following the AOAC methods 991.43 and 985.29 (K-TDFR, Megazyme) to estimate the dietary fiber content of whole grain rice samples. Briefly, 1 g of finely ground whole-grain rice flour or defatted bran powder was boiled for 30 min with 50 mL of 0.05 M MES/TRIS buffer (pH 8.2) and 0.2 mL of thermostable α-amylase (3,000 Units/ml). After cooling, the solution was incubated for 30 min at 60°C with 0.1 mL of protease (50 mg/mL; 350 tyrosine Units/ml) and adjusted to pH 4.5 with 0.561 N HCl. The partially digested flours were further incubated at 60°C for 16 h with 0.2 mL of amyloglucosidase (3,300 Units/ml). The completely digested solution was filtered to separate the insoluble (residue) and soluble (filtrate) fractions. The insoluble residue was washed with 78% ethanol, 95% ethanol and acetone, then dried and weighed to isolate the insoluble dietary fiber (IDF). The weight of the IDF was corrected with crude protein and ash and expressed as a percentage of whole-grain rice flour or rice bran powder. To estimate the SDF, the soluble filtrates were deionized and passed through a column packed with mixed-bed ion exchange resin, concentrated and filtered again through 0.45 μm-membrane filters, and quantified by HPLC with a refractive index detector (Shimadzu RID-10A HPLC system, Shimadzu Corporation, Kyoto, Japan) based on Ohkuma’s method (52) with modification and AOAC methods 2009.01 and 2011.25 (K-INTDF, Megazyme). SDF was expressed as the whole-grain rice flour or bran powder percentage. Total dietary fiber (TDF) was the sum of IDF and SDF.

The iodine-colorimetric method determined the AC using Juliano’s technique (53). Total fat was determined using Soxhlet extraction with petroleum ether, based on AOAC method 945.16. The gravimetric method measured the moisture content found in ISO method 712:2009. Kjeldahl analysis determined crude Protein according to the AOAC method 2001.11. Crude ash was determined by incineration at 525°C for 5 h, according to AOAC method 942.05.

The alkaline degradation test, alternatively called Alkaline Spreading Value (ASV), was conducted (54) with minor modifications to estimate the gelatinisation temperature (GT) and alkaline-resistant properties of whole-grain rice samples. Eight whole grain kernels were placed in a closed 10-cm Petri dish containing 20 mL of 1.7% (w/v) KOH aqueous solution for 24 h at room temperature. The alkaline spreading value (ASV) was rated visually on a 7-point numerical scale (1 = intact; 7 = extensively dispersed), and the average scores of eight kernels were taken as the spreading value.

Whole grain rice samples were soaked in water – non-waxy rice (1:2 ratio), waxy rice with ASV more than 1 (1:1 ratio), and waxy rice with ASV equal to precisely 1 (1:2 ratio) – in aluminum cups, then cooked in a stainless-steel streamer for 30–45 min until no white starch core was observed before the analysis. A texture profile analysis (TPA) of cooked whole grain rice samples was conducted with a texture analyzer (TA-XT plus, Stable Micro System, Godalming, UK) based on the method used by Parween et al. (20) and Guillen et al. (55), which demonstrated the significant correlation with sensory evaluation by trained panelists. A 50-mm cylinder probe was set at 30 mm above the base. The TPA force-deformation curve was obtained using a two-cycle compression with a force versus distance program. Three warm rice kernels were placed on the base platform under the center of the probe and compressed to 90% of the original cooked grain thickness. Pre-test, test, and post-test speeds were 1, 0.5, and 0.5 mm/s, respectively. Nine measurements were performed for each sample (10 per cup × 3 cups).

A radar chart was developed to display the complex interrelationship between AC, Dietary fiber Index (DFI), the SDF to IDF ratio, protein, and textural characteristics across rice varieties with high, low, and very low AC (Waxy) using ‘fmsb’ package in R program.

The analysis was performed using Statgraphics Centurion XVII software (Statpoint Technologies, Warrenton, VA, USA). The data were analyzed in triplicate by one-way analysis of variance, and Duncan’s multiple range test was used to determine statistically significant differences among the samples. These differences were indicated by different letters in the columns when the p-value was lower than 0.05. Linear regression was also analyzed using the Pearson correlation two-tailed test, with a significance level of 0.05 and 0.01. A correlation matrix was developed to identify the correlations between two variables, using a linear relationship Pearson correlation coefficient at a significance level of 0.05 and 0.01.

We performed Structural Equation Modeling (SEM), an integrated factor and multiple regression analyses, to quantify the structural relationships between causes and effects. All selected variables were standardized to have a mean of 0 and a standard deviation of 1 before the analysis. SEM involved the R Software Platform (56) and the Lavaan Package (57). Path coefficients were partitioned into direct and indirect effect components using Wrigh’s method (58) to illustrate the relationship between correlation and path coefficients. Any missing data were corrected using a pairwise deletion. After creating the SEM, the Adjusted Goodness-of-Fit Statistic (AGFI >0.90) and the Standardized Root Mean Square Residual (SRMR <0.05) were utilized to fit the overall model (59). Only SEM with R-square values below 60 were dropped for further interpretation. All scripts for the analyses were listed in the Supplementary Table S2.

The purple rice population contained significant variations in AC and ASV, followed by IDF and SDF. AC ranged from 6.81 to 29.6% and alkaline spreading value (ASV) from 1.0 to 5.4 (Figure 1 and Supplementary Table S3). We classified rice varieties based on AC and allele-specific SNP markers into waxy (AC < 12%), low amylose (AC from 12 to 20%), and high amylose (AC > 20%). For AC, the G/T SNP and 23-bp insertion precisely classified the 36 rice varieties into high (8), low (17), and waxy (11) groups. Within each AC group, rice varieties varied considerably within the specified ranges of AC (Supplementary Table S3). In this study, the SNP markers showed high accuracy even at the borderline between high vs. low AC and low AC vs. waxy groups (Supplementary Tables S3, S5). In contrast, the GC/TT SNP marker and ASV showed less association with the whole-grain purple rice in this study.

The DFI, the SDF/IDF ratio, ranged from 0.19 to 0.60 (Figure 2). In particular, high DFI rice varieties were primarily found among low AC rice varieties. In comparison, high DFI rice >50 mainly contained a lower IDF or higher SDF than lower DFI rice. In contrast, high IDF varieties were found only in the waxy rice group (Supplementary Table S3). In addition, the average protein content was distributed narrowly but in a higher range, while fat content was equally distributed among the whole-grain purple rice varieties. Interestingly, purple rice’s protein contents were exceptionally higher among waxy and low AC groups, ranging from 8.56 to 14.09% (Supplementary Table S3). Such a collection of purple rice represents the extreme genetic diversity suitable for studying the impacts of nutrients on cooked rice quality.

Within each AC group, rice varieties varied in nutrient parameters, particularly DFI and protein content, which could affect the textural characteristics of whole-grain rice, particularly HRD and ADH. AC strongly affected HRD among the high AC group, which shows a narrow variation in DFI. DFI had little effect on HRD but was significantly high in ADH (Figure 3 and Supplementary Tables S3, S4). The B30-E4, with the highest DFI at 0.34 among the high AC group, showed a slightly lower HRD but significantly lower ADH than the group (Figure 3 and Supplementary Tables S3, S4). Besides regular green-leaf rice varieties, we showed that two highly pigmented-leaf Rainbow Rice varieties, RBR#01 and RBR#05, had high AC and HRD with low DFI at 0.22–0.23.

In contrast, DFI showed a considerable variation among the low AC groups and exhibited a good relationship with HRD—all four high DFI rice varieties, viz. RB, B11-11, KHN, and M.210589 showed high DFI with similar low HRD (Figure 3 and Supplementary Tables S3, S4). In contrast, waxy purple rice varieties generally showed lower DFI but higher HRD than the low AC group, with a few exceptions (Figure 3). F9 was exceptionally low in IDF, resulting in high DFI (Supplementary Tables S3, S4) but low HRD (Figure 3). Nonetheless, the two waxy Rainbow Rice varieties, RBR#02 and RBR#04, with the two top protein content at 13.43 and 14%, respectively, were among the lowest DFI at 0.21 and 0.19, respectively, but high contrast in ADH but similar HRD (Figure 3). Besides the influence of AC, DFI and protein also play roles in HRD and ADH.

For the average 10.5% whole-grain protein content, the purple rice germplasm showed superior seed protein than commercial varieties of brown-colored rice. Notably, waxy purple rice contained higher protein than those low and high-AC purple rice groups (Supplementary Tables S3, S4). RBR#02 and RBR#04, Rainbow Rice varieties, were among the top protein contenders among the waxy rice group (Figure 3 and Supplementary Tables S3, S4). In addition, KDSK, KH, and F9 were among the waxy rice varieties with >12% protein content. Nonetheless, only two high-protein rice varieties reported in non-waxy rice groups were HDSKT 2 (>12% protein, low AC) and RBR#05 (11.48% protein, high AC). Understanding the impacts of protein on HRD and ADH is interesting in this purple rice germplasm.

Protein corresponded well with HRD and ADH among waxy and low-AC rice varieties. In contrast, AC strongly affected HRD and ADH more than protein content in the high AC group but not in the low and waxy groups (Figure 3). Variation in ADH was also significant among waxy rice but less among low AC rice varieties. Nonetheless, ADH varied considerably in the low AC rice varieties, albeit less than in the waxy group, but showed an inconsistent relationship with protein contents. For instance, HDSKT 2, the top protein content among the low AC group, was the least ADH. While RB and B11-B11 differed by less than 1% protein content among the low AC group, they showed a substantial contrast in ADH. RBR#05 (high AC) and PK + 6#3G06 (low AC) differed by less than 1% in protein content but were significantly high in ADH. Still, the difference in AC affected ADH to some degree. We found a more consistent relationship between protein content, DFI and ADH among waxy rice varieties. Interestingly, waxy rice F9, with 11.9% protein content, showed the highest DFI and softest cooked rice among whole-grain purple rice varieties (Figure 3 and Supplementary Tables S3, S4). Therefore, we were interested in dissecting the relationship between protein content, dietary fiber, starch and textural characteristics of whole-grain purple rice through binary correlation and multiple regression analyses.

Correlation analysis demonstrated how complex nutritional components, including starch, dietary fiber, and protein, pairwise interacted with the textural characteristics of whole-grain purple rice varieties (Figure 4). AC, AP, and AC/AP strongly influenced HRD, ADH, and GUM but weakly with COH, whereas SDF, IDF, and DFI were strongly associated with the overall textural characteristics. Specifically, SDF and DFI had a significantly negative association with HRD, COH, GUM, and Chew, while IDF was associated strongly with ADH but weakly with SPR. SDF and DFI showed a significant association, but IDF and DFI were independent. Therefore, variation in SDF, not IDF, affected DFI. Conversely, protein content significantly correlated with ADH but not other nutrient factors and textural attributes (Figure 4). We may conclude from correlation analysis that protein did not affect textural characteristics in the purple rice germplasm.

A pairwise correlation analysis, like the Pearson correlation coefficients, only determines the binary relationship between two variables, regardless of differentiating which independent variables have direct or indirect effects on a dependent variable or how strongly independent variables affect a dependent variable. We designed a Path Coefficient Analysis (PCA) to determine the directions and impacts of nutrient factors (independent variables) on the textural characteristics (dependent variables) in cooked whole-grain purple rice.

Cooked Rice Hardness (HRD): The Pearson correlation indicated that the AC and AC/AP variation showed positive influences, but AP, SDF, and DFI were negatively associated with HRD. Nonetheless, PCA revealed that DFI (−0.5, p < 0.01) had the most potent negative direct effect, followed by IDF (−0.27, p > 0.05), whereas protein (0.34, p < 0.05) and AC/AP (0.32, p > 0.05) directly and positively affected HRD (Figure 5A). The fact that Pearson correlation analysis failed to detect the association between Protein and HRD, PCA strengthened the correlation analysis with the power of differentiation between the actual cause and effect. Although IDF contributed a weaker direct effect on HRD, it exerted a more substantial negative indirect effect via AC/AP and could balance protein effects on HRD. Our results inferred that low AC or waxy pigmented rice containing higher protein and DFI tend to be softer than rice with lower DFI.

Adhesiveness (ADH): From the correlation coefficient, ADH was associated with AP, IDF, and protein but negatively associated with AC, AC/AP, and ASV. However, the PCA revealed AC/AP strongly negatively impacted ADH, followed by DFI, while protein positively affected ADH (Figure 5B).

COH, GUM, SPR, and CHEW: Because their R-square values of the SEM were too low, we did not consider them significant for further investigation (Supplementary Tables S10–S17).