LWT - Food Science and Technology 168 (2022) 113929 Contents lists available at ScienceDirect LWT journal homepage: www.elsevier.com/locate/lwt Evaluation of in vitro digestion methods and starch structure components as determinants for predicting the glycemic index of rice Putlih Adzra Pautong a,b,c,1, Joanne Jerenice Añonuevo a,1, Maria Krishna de Guzman a, Rodolfo Sumayao Jr. b, Christiani Jeyakumar Henry d, Nese Sreenivasulu a,* a International Rice Research Institute, Los Baños, 4030, Philippines b De La Salle University, Department of Chemistry, Manila, 0922, Philippines c Mindanao State University, Gen. Santos City, 9500, Philippines d Clinical Nutrition Research Centre, Singapore Institute for Clinical Sciences, 14 Medical Drive, #07-02, Singapore A R T I C L E I N F O A B S T R A C T Keywords: Mainstreaming the low glycemic index (GI) trait in breeding programs is constrained by low-throughput and Rice high-cost clinical GI phenotyping. This study aimed to evaluate the potential of starch fine structure components Glycemic index and simulated digestion parameters in predicting GI in rice. Amylose (AM1 and AM2; r = − 0.94 and r = − 0.80, In vitro digestion respectively, p < .05) and amylopectin fine structure (MCAP, SCAP, and SCAP1; r = 0.78-0.86, p < .05) measured Starch fractions Debranched starch through size-exclusion chromatography along with resistant starch (r = − 0.81, p < .05) in seven (7) rice ac- cessions showed high correlation with in vivo GI. Meanwhile, starch hydrolysis extent (SH) and the corresponding area under the digestion curve (AUC) obtained through in vitro digestion were found to be of higher correlation with GI, even within shorter digestion periods of 5 min or 30 min (r = 0.96, p < .01). These results highlight the potential use of these parameters as predictors of GI, with improved predictive capacity through a multiple regression model. Higher correlations of simulated digestion AUC with GI may be due to its ability to account for the overall food matrix native macro- and micro-structures, gaining an added advantage over SEC method as a predictive tool in studying rice GI variability. Validation in a larger population is an inevitable next step. 1. Introduction is an important target trait in food since high postprandial glucose levels have been associated with various non-communicable diseases such as Shift in dietary practices and lower energy expenditure brought obesity, type 2 diabetes, and cardiovascular diseases (Blaak et al., 2012). about by various socio-economic developments linked with urbaniza- Monitoring the quality of carbohydrates (which account for 40–80% of tion has led to the emergence of obesity as a serious global concern. daily energy intake) is an essential intervention for glycemic control. Recent statistics reported that 1.9 billion adults are overweight; of them More than half of the global population, mainly in Asia, derives more 422 million have diabetes (World Health Organization, 2020). than 50% of daily calories from rice (GRiSP, 2013). Incidentally, 60% of Consuming energy-rich foods and drinks create chronic hyperglycemia people with diabetes live in the continent (Nanditha et al., 2016), (Yan, 2014). Obesity and hyperglycemia contribute to the increased making rice a vital food component in diet-based solutions to the incidences of non-communicable diseases (NCDs) like diabetes, cardio- growing epidemic of diabetes, at least for rice-consuming populations. vascular diseases (CVDs), hypertension, and some cancers, among Starch-rich cereals which include rice are typically of intermediate to others, which account for more than 70% of annual deaths worldwide high GI (Anacleto et al., 2019; Fitzgerald et al., 2011; Kaur, Ranawana, (World Health Organization, 2021). & Henry, 2016; Miller, Pang, & Bramall, 1992). In rice-based diets, Starch is the major source of carbohydrate in the diet. Interestingly, reducing the glycemic index (GI) of a variety through genetics, post- not all starches are created equal due to various intrinsic and extrinsic harvest processing, cooking, and diet diversification solutions are crit- factors affecting their digestibility (Jukanti, Pautong, Liu, & Sreeniva- ical (Jukanti et al., 2020). GI is a clinical measure of the tendency of food sulu, 2020; Toutounji, Farahnaky, et al., 2019). Slow starch digestibility or drinks containing 50 g of available carbohydrates to influence the * Corresponding author. E-mail address: n.sreenivasulu@irri.org (N. Sreenivasulu). 1 These authors contributed to this work equally as co-first authors. https://doi.org/10.1016/j.lwt.2022.113929 Received 13 April 2022; Received in revised form 24 August 2022; Accepted 3 September 2022 Available online 8 September 2022 0023-6438/© 2022 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by- nc-nd/4.0/). P.A. Pautong et al. L W T 168 (2022) 113929 blood glucose response upon intake relative to the same amount of first 20 min, between 20 and 120 min, and remains undigested after 120 glucose standard (Jenkins et al., 1981). GI classifies food as low (≤55), min, respectively. Furthermore, correlations between these fractions intermediate (>55–69), or high GI (≥70) (Atkinson, Foster-Powell, & and amylopectin fine structure have been examined in rice (Benmoussa, Brand-Miller, 2008). Moldenhauer, & Hamaker, 2007), where among the amylopectin frac- With more than half of the global population being dependent on rice tions, those with degrees of polymerization (DP) of less than 13 showed as a staple carbohydrate source, it is crucial that low GI rice cultivars are a positive correlation with RDS while longer DPs showed negative made available to consumers to ease the burden of NCDs. To breed low correlation. GI varieties, (a) understanding of the genetics of glycemic index in rice Based on these premises, we hypothesize that fine structure com- through genome-phenotype associations and identifying diagnostic ponents of amylose (AM1 and AM2) and amylopectin (medium-chain markers and (b) a rapid high-throughput GI phenotyping exhibiting MCAP, short-chain SCAP, and its subfractions SCAP1, SCAP 2, and SCAP higher correlations with gold standard in vivo clinical method are pre- 3, that differ in degrees of polymerization) can be used to predict the GI requisites. The lack of a low-cost, high-throughput in vitro method to in rice. We further hypothesize that in vitro amylolysis protocols replace the gold standard in vivo GI screening with human volunteers, employing low sample requirements, shorter digestion period, and/or which is both expensive and time-consuming, remains a serious limita- lower enzyme activities can be modified for rice samples such that tion. To date, in vitro technique to estimate GI has been described by amylolysis parameters SH and AUC strongly correlate with in vivo GI. Goñi, Garcia-Alonso, and Saura-Calixto (1997), which makes use of The purpose of the study was to evaluate and compare the performance gastrointestinal enzymes pepsin, pancreatic α-amylase (AA), and amy- of these methods for predicting the GI in rice. A reliable, high- loglucosidase (AMG) in the digestion of common dietary starchy foods. throughput, and more affordable GI phenotyping method may fast- Briefly, glucose released through hydrolysis is quantified as a percentage track the research efforts on understanding the genetics of GI in rice, of starch hydrolyzed at different time points (SH) typically 0–180 min at with the ultimate goal of mainstreaming this human health-promoting 30 min intervals. Taken from a plot of these values is the area under the trait in rice breeding programs. digestion curve (AUC) of the test food, whose ratio (expressed in per- centage) with the AUC of a reference food, is referred to as hydrolysis 2. Materials and methods index (HI). Two equations from the experiments of Goñi et al. (1997) were found to best estimate in vivo GI values of the test foods used in 2.1. Samples their study: GI = 39.21 + 0.803(SH90) (Equation 1, r = 0.909, p ≤ .05) which can be used to estimate GI using hydrolysis value at a single time The milled rice samples of seven indica rice cultivars previously point (t = 90 min), and GI = 39.71 + 0549(Hl) of food but lower linear validated through in vivo GI estimation through human subjects (Ana- correlation (Equation 2, r = 0.894, p ≤ .05) which makes use of HI cleto et al., 2019; Pasion et al., 2021) were used to establish in vitro GI derived from multiple time points less accurate. The said equations have prediction methods. These cultivars include GQ02522 (50.4, low), been used in numerous studies to estimate the GI of various food GQ02497 (51.5, low), IRRI147 (55, low), IRRI162 (57, intermediate), matrices (Chung, Shin, & Lim, 2008; Germaine et al., 2008; Kim & IRRI163 (64, intermediate), IR64 (66, intermediate) and IR65 (90, White, 2012; Lal et al., 2021; Leoro, Clerici, Chang, & Steel, 2010) high). including rice (Deepa, Singh, & Naidu, 2010; Fernandes, Madalena, Meanwhile, the variability of predicted GI values across various Pinheiro, & Vicente, 2020; Kale, Jha, Jha, Sinha, & Lal, 2015; Kunyanee linear regression models and parameters was then demonstrated using & Luangsakul, 2018; Tutusaus, Srikaeo, & Diéguez, 2013). Although 10 market samples, of these three are brown rice and the rest are milled widely used, it should be noted that in vitro GI equations are derived (Table 5). from a wide range of food types with varying starch hydrolysis char- acteristics (Goñi et al., 1997) for which the hydrolytic process was found 2.2. Profiling of starch indices and debranched starch regions to follow the first-order equation C–C∞(1-e-kt). However, some slowly digestible foodstuffs reach a plateau by 90 min of hydrolysis (essentially Starch indices such as total starch (TS), digestible carbohydrates zero-order) and thus cannot be subjected to first-order kinetics analysis; (DC) and resistant starch (RS) were measured using commercially which will otherwise result to unreliable C∞ values (Edwards, Cochetel, available kits (Megazyme K-TSTA and K-RSTAR) with downscaled Setterfield, Perez-Moral, & Warren, 2019). Starch digestion is known to enzyme buffers as described in Alhambra, Dhital, Sreenivasulu, and be affected primarily by starch multi-scale structure spanning hierar- Butardo (2019) with minor modification. Meanwhile, the starch regions chical structural levels such as compositional, short-range, helical, (AM1, AM2, MCAP, and SCAP) were characterized using size exclusion crystalline, lamellar, and morphology structures, which is perhaps made chromatography following the procedures described in Guzman et al. further complex by proteins, lipids, and secondary metabolites that (2017). modify the said structures, as thoroughly reviewed elsewhere (Chi et al., 2021). 2.2.1. Total starch (TS) quantification Several alternative protocols that differ in digestion conditions such Rice flour (10 ± 0.1 mg) weighed in 2-mL microfuge tubes was as activities and incubation times were proposed (Brodkorb et al., 2019; dispersed in 0.5 mL of 80% (w/v) ethanol and incubated at 85 ◦C for 5 Edwards et al., 2019; Fitzgerald et al., 2011; Germaine et al., 2008; min. Another 0.5 mL of 80% ethanol was added followed by centrifu- Toutounji, Butardo, et al., 2019). In contrast to methods that aim to gation at 3000 rpm for 10 min. After carefully decanting the superna- compare GI across various food matrices such as the INFOGEST method tant, the pellet was re-suspended with 1 mL of 80% ethanol. After (Brodkorb et al., 2019), a protocol intended for a specific food such as another centrifugation step, the supernatant was decanted once again, rice could benefit from lower variability in food matrix composition and and the remaining pellet was added with 0.2 mL of 2M KOH. A magnetic structure due to a more narrow range of protein (4.91%–12.08%) spin bar was added to aid mixing, and solubilization of the pellet was (Banerjee, Chandel, Mandal, Meena, & Saluja, 2011) and lipids (0.5%– allowed to proceed for at least 20 min in an ice bath. After the incuba- 0.8%) (Juliano, 1985, pp. 59–174) compared to a wider range when tion, 0.8 mL of 1.2 M sodium acetate buffer (pH 3.8) was added followed comparison is made across various botanic sources. by the addition of 10 μL each of thermostable α-amylase and AMG. The Moreover, the association between GI and starch digestion rate has sample was then incubated at 50 ◦C for 30 min with intermittent vortex been previously studied by Englyst, Vinoy, Englyst, and Lang (2003). mixing every 15 min. After the enzymatic hydrolysis, an aliquot (0.1 mL) They classified starch into fractions that differ in their rate of hydrolysis was added with distilled water to bring the final volume to 1 mL. The in the small intestine: rapidly digestible starch (RDS), slowly digestible mixture was mixed and centrifuged at 13,000 rpm for 10 min. An aliquot starch (SDS), and resistant starch (RS) that are hydrolyzed within the (10 μL) of the supernatant was mixed with 0.3 mL of GOPOD reagent, 2 P.A. Pautong et al. L W T 168 (2022) 113929 incubated at 50 ◦C for 20 min, and absorbance was read at 510 nm using then transferred to an SEC vial for analysis (stop time, 35 min; flow rate, a microplate reader (SPECTROstar Nano, BMG Labtech, Germany). 0.5 mL/min; injection volume, 40 μL; sample temperature, 40 ◦C; and column temperature, 60 ◦C). The degree of polymerization (DP) was 2.2.2. Resistant starch (RS) quantification derived from the molecular weight computed using the Mark-Houwink- Rice flour samples (10 ± 0.1 mg) weighed in a 2-mL microfuge tubes Sakurada Equation, and the four regions were determined using this were dispersed in 0.4 mL of enzyme solution containing 10 mg/mL distribution: AM1>1000DP, AM2 121-1000DP, MCAP 37-120DP, and pancreatin and 3 U/mL AMG in 0.1 mM sodium maleate buffer with 5 SCAP 6-36DP. SCAP was further distributed into three regions, SCAP1 mM CaCl2⋅H2O (pH 6.0). The tubes were then secured on a rack and 25-36DP, SCAP2 13-24DP, and SCAP3 6-12DP, based on Hanashiro, incubated horizontally in a shaking water bath set at 100 strokes/min Abe, and Hizukuri (1996). and 37 ◦C for 16 h. Twelve minutes prior to the end of the incubation period, the tubes were removed from the water bath and individually 2.3. In vitro starch digestion protocols dried using a paper towel. At exactly 16 h, the reaction was stopped by adding 0.4 mL of 99% ethanol followed by vortex mixing. The tubes Three different in vitro GI methods were standardized for sample were centrifuged at 18,231×g for 30 min. The supernatant was then preparation, digestion, and/or glucose quantification in this study carefully decanted in a 15-mL conical tube and was set aside for the (Table 2). quantification of total digestible carbohydrate (DC). The remaining pellet was then re-suspended in 0.2 mL of 50% ethanol and mixed using 2.3.1. Method 1 a vortexer and added with a further 0.6 mL of 50% ethanol. Following A slightly modified version of the procedure by Goñi et al. (1997) another centrifugation, the supernatant was carefully decanted into the was employed to test its applicability in rice using reference samples of conical tube containing the first decantate, pooling both decantates seven cultivars with established in vivo GI (Anacleto et al., 2019; Pasion together. On the other hand, a stir bar was placed in the tube containing et al., 2021). In addition, Method 1 was modified (referred to as Modi- the pellet over an ice bath and 0.2 mL of 2 M KOH was added, allowing fied Method 1) by incorporating the following changes: whole grains complete solubilization for at least 20 min. At the end of the reaction (100 ± 5 mg) were used instead of flour (50 mg, passed through a 425 time, 0.8 mL of 1.2 M sodium acetate buffer (pH 3.8) was added fol- μm mesh), a 1:2 rice to water ratio was used instead of 3 mL water, and lowed by the immediate addition of 10 μL of AMG (3300 U/mL). After all incubations were conducted at 37 ◦C to simplify the process, as mixing, the tubes were incubated in a water bath at 50 ◦C for 30 min opposed to the two incubation temperatures in the original Method 1. (with vortex mixing after the first 15 min). Without the stir bars, the Briefly, rice grains (100 ± 0.3 mg) were added with 0.2 ml distilled tubes were then centrifuged at 18,231×g for 10 min, and glucose con- water and cooked for 23 min over vigorously boiling water in a covered centration in a 10-μL aliquot of the resulting supernatant was quantified pot. The cooked sample was then allowed to cool down for 5 min and as in the preceding section. equilibrated in a 37 ◦C water bath for another 5 min. The grains were then minced in 0.5 ml of HCl–KCl buffer (0.05 M, pH 1.5) using a 2.2.3. Total digestible carbohydrates (DC) stainless steel spatula with 20 downward strokes. Then, 9.7 ml of The pooled supernatants (previous section) were diluted to 10 mL HCl–KCl buffer (0.05 M, pH 1.5) containing 40 U of pepsin (Sigma; 4.12 with 0.1 mM sodium acetate buffer (pH 4.5). After mixing, a 10-μL U/mL or 95.9 μg/mL) was added and allowed to incubate for 1 h with aliquot was collected, added with 2 μL of AMG suspension (300 U/mL in constant magnetic stirring (90 rpm). The pH was then adjusted to 6.9 by 100 mM sodium maleate buffer with 5 mM CaCl2⋅2H2O; pH 6.0) and the adding 2 mL of aqueous NaOH (~0.163 M), followed by the addition of mixture was vortexed and incubated in a 50 ◦C water bath for 20 min. 14.8 mL of Tris-maleate buffer (0,05 M, pH 6.9) containing 2.6 U of Glucose was then quantified using GOPOD reagent (0.3 mL) as in 2.2.1. α-amylase (A3176, Sigma; 0.176 U/mL or 11.7 μg/mL). Starch hydro- lysis was allowed to proceed for 180 min in the water bath (37 ◦C, 90 2.2.4. Profiling of debranched starch rpm). Sample aliquots (0.1 mL) were collected at 0, 30, 60, 90, 120, and Waters Alliance 2695 HPLC with 2414 Refractive Index Detector and 180 min and immediately placed in boiling water for 5 min to deactivate fitted with Waters Ultrahydrogel 250 Å column was first calibrated for the enzymes. The aliquot was then centrifuged at 13,500 rpm for 10 min, molecular weight using pullulan standards (P-82 Shodex, Showa Denko, and 10 μL of the resulting supernatant was added with 30 μL AMG (2.75 K. K. Kawasaki, Japan). Mark-Houwink-Sakurada equation for universal U/mL) in sodium acetate buffer (0.4 M, pH 4.75), vortexed, and incu- calibration (Pullulan: K = 0.0126 mL g− 1 and a = 0.733; Linear starch: K bated at 50 ◦C for 20 min to convert starch oligosaccharides into free = 0.0544 mL g− 1 and a = 0.486) as described in Castro, Dumas, Chiou, glucose. After incubation, 60 μL of Milli-Q water was added, vortexed, Fitzgerald, and Gilbert (2005) was used with 0.05M NH4OAc with and a 10-μL sub-aliquot was transferred into a 0.6-mL microfuge tube. 0.02% sodium azide (pH 4.75) as mobile phase. Glucose was then quantified using a GOPOD reagent as in Section 2.2.1. Gelatinization of rice flour (50 ± 0.1 mg in a glass scintillation vial of A reagent blank was incubated under the same conditions, and absor- known weight) was done by adding 0.4 mL of 95% ethanol and 1 mL of bance at different time points was also measured to correct for in- 0.25 M NaOH, followed by heating at 150 ◦C for 12 min. Within the terferences due to the digestion medium. The calculation of SH from heating period, successive aliquots of 0.8 mL hot water (100 ◦C) were glucose concentrations and predicted GI (pGI) value is described in added at 0, 4, and 8 min from the onset of heating period to prevent Section 2.3.4. drying of sample. After heating, the final weight of the solution was adjusted to 4 g by adding hot water (60–65 ◦C). Debranching of the 2.3.2. Method 2 gelatinized starch was then induced by adding 0.206 mL of sodium ac- The second method employed the protocol reported by Alhambra etate buffer (prepared by mixing 10 mL 0.2M NaOAc at pH 4.0 with et al. (2019) with modifications. Whole milled rice grains (500 ± 10 mg) 0.360 mL glacial acetic acid) to a 0.794-mL aliquot of the sample. The were soaked for 10 min, and brown rice samples were soaked for 20 min resulting mixture was then incubated with 10 μL of isoamylase in 6 mL distilled water using 50-mL tube with a foil cap, under room (P113541, Megazyme) in a 50 ◦C water bath for 2 h with mixing by temperature. Afterwards, the sample was cooked in a beaker with inversion every 15 min. Isoamylase was then inactivated by placing the boiling water (heater set at 250 ◦C) for 20 min. Excess water was tubes in a vigorously boiling water bath. After centrifugation at removed, and rice was allowed to cool for 5 min. The tubes were then 12,5000×g for 10 min, the supernatant is carefully decanted into a 1.5- placed in a 37 ◦C water bath and 0.5 mL α-amylase (Megazyme; 75 U/mL mL microfuge tube containing ~0.1 g ion exchange resin (Bio-Rad AG in simulated salivary fluid) was dispensed. The grains were minced 501-X8 (D)) which was then incubated at 50 ◦C for 30 min (with mixing using a spatula for 20 s to mimic the buccal phase, followed by the by inversion every 10 min). An aliquot (0.15 mL) of the supernatant was addition of 5 mL pepsin (P6778, Sigma; 1 mg/mL in 0.2M HCl). Spin 3 P.A. Pautong et al. L W T 168 (2022) 113929 bars were added, and the speed was set to 200 rpm. After 30 min, 5 mL of calculated using the equation: NaOH (0.2 M) was added to neutralize the pH, followed by the addition of 20 mL sodium acetate buffer (0.1 M, pH 6.0), and 5 mL pancreatin SH =(glucose x DF x V x 162 / 180)/(sample weight x (%TS) / 100) x 100 (P1750, Sigma; 2 mg/mL buffer)-AMG (10115-5G-F, Sigma; 25.4 U/mL where glucose is in mg/ml, DF is the dilution factor used on the sample or 0.41 mg/mL buffer) mixture while the speed of stirring was increased aliquot, V is the volume of the digestion medium at the time point during to 700 rpm. This reaction was allowed to carry out for 180 min, and 0.2 which the aliquot was drawn, 162/180 is the conversion factor from ml of aliquots were collected at time points 0, 5, 10, 20, 30, 45, 60, 90, glucose to starch, and %TS is the total starch content of the sample. 120, and 180 min, and placed in a 0.6 ml microfuge tube. To quench the Finally, AUC was calculated from SH values based on the trapezoidal reaction, the tubes were placed on an ice bath prior to centrifugation at rule. 13,000 rpm (4 ◦C) for 10 min. Samples were diluted with 0.1M sodium acetate buffer (pH 6.0) when needed (dilution starts at aliquots 5 min and beyond). Aliquots of 2.4. Derivation of linear regression predictive GI (pGI) equations 50 μl were added with 5 μL of amyloglucosidase (300 U/mL) and incubated for 20 min at 50 ◦C. Starch hydrolyzed was quantified using The Pearson correlation coefficient (r, P < .05; R package “corrplot” the GOPOD assay at 510 nm (Beckman Coulter DU 800 spectropho- (Wei & Simko, 2021)) was used to assess the correlation between in vivo tometer). The calculation of predicted GI (pGI) value is described in GI and each of the parameters tested. Following the tests for linearity Section 2.3.4. (ANOVA for linear regression), normal distribution of the residuals of regression (Shapiro-Wilk; R package “dplyr”; Wickham, François, 2.3.3. Method 3 Henry, & Müller, 2022), and homoscedasticity (R package “lmtest”; A third starch digestion protocol was employed following the Zeileis & Hothorn, 2002) of each parameter showing a good correlation method of Germaine et al. (2008) with some modifications. Briefly, with in vivo GI, linear regression analysis was then performed to whole milled rice (300 ± 0.3 mg) with 1:2 rice-to-water ratio was generate the corresponding pGI equations and values. Statistical anal- cooked in 50-ml tube over boiling water for 23 min, after which the ysis was performed using RStudio 2022.2.3.492 (RStudio Team, 2020). sample was set aside at room temperature for 5 min. After equilibrating in a 37 ◦C water bath for another 5 min, it is added with 1 mL of 0.1 M 2.5. Evaluation of derived pGI equations on market samples sodium potassium phosphate buffer (pH 6.9) and minced using a stainless-steel spatula (35 downward strokes within 10 s). After placing The applicability of the derived equations for predicted GI (pGI) a magnetic spin bar, 2 mL of α-amylase (A3176, Sigma; 55.5 U/mL or using different amylolysis parameters were demonstrated for ten (10) 3.7 mg/mL of 0.05 M sodium potassium phosphate buffer, pH 6.9, 37 ◦C) market samples. SEC experiments were performed in triplicates while was added, and amylolysis was allowed for 75 s in a water bath with simulated digestion were performed in duplicates. constant stirring (37◦, 60 rpm). The reaction was stopped by adding 3 mL of aqueous HCl (pH ~0.92) to acidify the digestion medium. Then, 3 3. Results and discussion mL of pepsin (Sigma; 19.5 U/mL or 0.45 mg/mL of 0.1 M sodium po- tassium phosphate buffer, pH 1.5, 37 ◦C) was added and allowed to 3.1. Evaluation of starch indices and starch regions as proxy parameters incubate for 30 min (37 ◦C water bath, 60 rpm). Enzymatic reaction was to estimate GI quenched by adjusting the pH to 6.9 using 15 mL of aqueous NaOH (pH ~12.6). A 15 mL enzyme solution containing pancreatin (28.4 μg/mL; The range of starch indices (TS, RS, and DC) and debranched starch P1750, Sigma) and amyloglucosidase (10115-5G-F, Sigma; 13 U/mL or regions separated through SEC (AM1, AM2, MCAP, and SCAP and its 0.208 mg/mL) in 0.1 M sodium potassium phosphate buffer (pH 6.9) subfractions) in seven (7) rice accessions are presented in Table 1. RS, was then added and allowed to incubate for another 180 min (37 ◦C, AM1, and AM2 declined whereas MCAP, SCAP, and SCAP increased with 120 rpm). Sample aliquots (0.70 mL) were withdrawn at 0 (just before increasing GI. On the contrary, no such statistically significant trends pancreatin-AMG addition), 30, 60, 90, 120, 150, and 180 min and were observed for TS and DC (p > .05, Table S1). Although direct pos- placed on ice until the succeeding centrifugation step (13,500 rpm, 10 itive correlations (MCAP, r = .780; SCAP, r = .855; SCAP1, r = 0.856; min, 4 ◦C). A 1-μL aliquot of the supernatant was transferred to a 0.6-mL AP, r = .886) and inverse correlations (RS, r = − 0.809; AM1, r = microfuge tube and was added with 9 μL of AMG solution (33.3 U/mL in − 0.941; and AM2, r = − 0.802) were observed with in vivo GI values 0.4 M sodium acetate buffer, pH 5.0), vortexed, and incubated in a water (Fig. 1), they were found to be less than those achieved using amylolysis bath (50 ◦C) for 20 min. Finally, glucose in the mixture and reagent parameters SH and AUC for in vitro GI methods 2 and 3 (Fig. 4) even blank were measured by adding GOPOD reagent and analyzed as in when individual fractions are pooled together (AM and AP, r = − 0.886; Method 1. AMAUC and AMAUC, r = − 0.880). This observation is also reflected in the corresponding R2 values when linear regression model is applied 2.3.4. Calculation of starch hydrolyzed (SH), AUC, and derivation of pGI (Fig. S1). Further inspection of the SCAP fractions revealed that GI was equations significantly correlated with SCAP1 only (r = 0.856, p = .014) which is Sample and glucose standard absorbance values at 510 nm were slightly higher than that for the totality of SCAP (r = 0.855, p = .014). corrected using a reagent blank, and converted into glucose values (mg/ These parameters passed the assumptions of linearity with GI, and ml of aliquot) using the following equation: normal distribution and homoscedasticity of the corresponding residuals based on ANOVA, Shapiro-Wilk test and Breusch-Pagan test, respec- Glucose=Absorbance of sample/Absorbance of glucose standard x 100 x 10/1000 tively, which served as the bases for linear regression analysis (Table 3). where the fraction 10/1000 was used to convert mcg glucose/0.1 ml Based on these results, only RS, AM1, AM2, SCAP1, AM, and AMAUC aliquot to mg/ml. The percentage of starch hydrolyzed (SH) was then were used as proxy parameters in succeeding analyses (Section 3.3). AP 4 P.A. Pautong et al. L W T 168 (2022) 113929 Table 1 Distribution of various starch parameters across seven (7) rice varieties. Parameters Samples GQ02522 (GI = 50.4) GQ02497 (GI = 51.5) IRRI147 (GI = 55) IRRI162 (GI = 57) IRRI163 (GI = 64) IR64 (GI = 66) IR65 (GI = 90) TS 85.21 ± 1.03 87.15 ± 0.68 85.94 ± 0.61 85.75 ± 1.17 83.34 ± 1.03 84.39 ± 1.23 85.79 ± 1.57 DC 83.14 ± 1.10 86.24 ± 1.87 83.66 ± 2.32 83.04 ± 4.30 85.79 ± 2.00 82.70 ± 2.64 86.27 ± 1.71 RS 2.08 ± 0.27 1.16 ± 0.14 1.66 ± 0.92 1.77 ± 0.46 1.01 ± 0.25 1.65 ± 0.38 0.29 ± 0.05 AM1 11.47 ± 2.10 11.31 ± 1.42 8.63 ± 1.47 7.45 ± 2.36 6.18 ± 1.37 6.18 ± 2.34 2.16 ± 1.30 AM2 14.24 ± 1.92 12.36 ± 2.35 11.02 ± 0.02 8.25 ± 0.72 6.64 ± 2.06 7.74 ± 1.16 5.71 ± 1.25 MCAP 24.81 ± 0.76 25.74 ± 0.55 26.38 ± 0.40 28.38 ± 1.86 28.69 ± 1.19 25.77 ± 0.71 30.34 ± 0.67 SCAP 49.48 ± 0.68 50.59 ± 4.32 53.97 ± 1.10 55.92 ± 1.81 58.49 ± 1.57 60.31 ± 1.34 61.79 ± 1.86 SCAP1 13.41 ± 0.40 13.95 ± 0.64 15.08 ± 0.01 15.35 ± 0.12 15.99 ± 1.26 15.46 ± 0.28 16.80 ± 0.35 SCAP2 28.96 ± 0.23 29.43 ± 2.67 32.86 ± 1.08 32.34 ± 1.17 33.62 ± 1.13 37.53 ± 1.09 36.09 ± 1.12 SCAP3 7.12 ± 0.28 7.21 ± 1.01 6.03 ± 0.00 8.22 ± 0.80 8.87 ± 0.82 7.33 ± 0.58 8.91 ± 0.73 AM 25.71 ± 0.58 23.67 ± 3.77 19.66 ± 1.50 15.70 ± 1.96 12.82 ± 2.76 13.92 ± 1.63 7.87 ± 1.81 AP 74.29 ± 0.58 76.33 ± 3.77 80.34 ± 1.50 84.30 ± 1.96 87.18 ± 2.76 86.08 ± 1.63 92.13 ± 1.81 AMAUC 24.25 ± 0.88 20.91 ± 3.11 18.08 ± 1.83 14.00 ± 2.28 10.05 ± 0.52 12.90 ± 1.59 5.90 ± 1.83 APAUC 75.75 ± 0.88 79.09 ± 3.11 81.92 ± 1.83 86.00 ± 2.28 89.95 ± 0.52 87.10 ± 1.59 94.10 ± 1.83 AUCa 47.1 ± 6.1 38.4 ± 0.6 45.9 ± 5.8 63.5 ± 0.5 55.4 ± 1.4 83.2 ± 7.1 125.0 ± 30.5 AUCb 534 ± 86 551 ± 37 481 ± 57 615 ± 45 709 ± 140 866 ± 72 1703 ± 83 The values are expressed as the mean of six (6) replicates for TS and DC, five (5) for RS, and three (3) for SEC-derived parameters ± standard deviation. aAUC(0–5) from Method 2; bAUC(150–180) from Method 3; TS = Total Starch; DC = Digestible Carbohydrates; RS = Resistant Starch; AM1 = percent amylose (DP > 1000 to 20,000); AM2 = percent long-chain amylopectin (DP > 120 to 1000); MCAP = medium-chain amylopectin (DP > 36 to 120); SCAP = short-chain amylopectin (DP 6–36); SCAP1 (DP 6–12); SCAP2 (DP 13–24); SCAP3 (DP 25–36); AM (AM1+AM2); AP (MCAP + SCAP); AMAUC and APAUC are percent amylose and amylopectin based on visual inspection of SEC peaks. (Chi et al., 2021). While the DP of amylopectin has been shown to affect the digestibility of starch (Martens, Gerrits, Bruininx, & Schols, 2018; Srichuwong, Sunarti, Mishima, Isono, & Hisamatsu, 2005), we report a contrasting result with respect to MCAP which displayed a positive correlation with GI. This can be explained by the “building block and backbone” model of amylopectin structure (Bertoft, 2004; Perez & Ber- toft, 2010) which suggests that DP > 36 (which includes MCAP) mostly interlink through α-(1,6)-linkages to form the backbone to which the building blocks of double helices of DP ≤ 36 (SCAP) are subtended. A huge proportion of the backbones are situated perpendicular to the crystalline clusters and are therefore mostly found in the amorphous lamella which is more accessible to hydrolytic enzymes. In addition, among the SCAP sub-fractions, our results show that only SCAP1 (DP 25–36) is significantly positively correlated with GI, contrary to previously reported negative correlation with starch di- gestibility while positive correlations were observed for shorter SCAP fractions (DP 6–24) only (Lin et al., 2016; Srichuwong et al., 2005). This may imply that double helices formed by longer chains (DP 25–36) believed to have higher resistance (Nakamura, 2018) could exist as amorphous double helices (Kim, Choi, Choi, Park, & Moon, 2020) and may actually potentially cause uneven packing and thus reduce crys- tallinity, as more often attributed to shorter helices (Chi et al., 2021). DP 25–36 has been previously reported to introduce defects into the structure of rice starches (Koroteeva et al., 2007), and thus potentially increase digestibility which is consistent with our result. Moreover, the backbone model suggests that SCAP3 of DP 6–8 anchored to the back- bone rather than the helical clusters or building blocks may be present, which may introduce defects in the crystalline structure. It is thus Fig. 1. Correlations between and among starch parameters and in vivo GI. tempting to speculate that SCAP1 (DP 25–36) also contributes to such structural defects and to a greater extent. With these premises, superior and AP haplotypes for bHLH transcription factor on chromosome 7 identified to AUC were not included since they both give pGI equations with the same value of slope (but of opposite sign) as that of the pGI for AM elevate AM1 fraction over SCAP (Butardo et al., 2017) will be useful as a and AM quick screening technique to enrich the germplasm of low GI potential AUC (Figs. S1h and S1j, respectively). The trends in AM1 and AM2 (and collectively as AM or AM from the gene bank accessions (Brotman et al., 2021). AUC) in this study corroborate previously reported negative correlation between Furthermore, the “glucan trimming” hypothesis (Ball et al., 1996) amylose and GI (Fitzgerald et al., 2011; Goñi et al., 1997; Guzman et al., behind the water-insoluble properties of amylopectin could further 2017; Hu, Zhao, Duan, Linlin, & Wu, 2004) since low amylose di- explain the observed correlations. Briefly, a tightly branched “pre- gestibility is attributed to its linear nature that limits the surface area for amylopectin” produced by the action of starch synthase (SS) and starch hydrolytic enzyme action. In addition, amylose is known to provide the branching enzyme (SBE) isoforms, is trimmed down by the debranching structural integrity to retard swelling and disruption of starch structure enzyme (DBE). The trimming process occurs simultaneously with two during cooking and reassemble into ordered structures upon cooling important structural changes: (a) the remaining branches of the 5 P.A. Pautong et al. L W T 168 (2022) 113929 “preamylopectin” form the characteristic double helices in the building 75.40% (IR65). As shown in the corresponding digestion curves in blocks, and (b) the amylopectin backbone being cleared of several Fig. 2a, only IR64 and IR65 are clearly separated from the rest, whereas chains is being further elongated (Tetlow & Bertoft, 2020). The both intermediate and low GI lines are grouped together in a narrow inter-block chain length (IB-CL), or the segment between successive range of SH values despite the huge differences in their in vivo GI values. building blocks, could affect the crystallinity of the resulting amylo- Strong correlations with in vivo GI (Pearson correlation coefficient, r; pectin supramolecular structure, whereby shorter IB-CL limits parallel Fig. 3a) were found between various SH (SH60-SH120 with r = packing of double helices leading to low onset gelatinization tempera- 0.76-0.86) and all AUC (r = 0.82-0.86, except for AUC(120–180)) (p < ture characteristic of decreased crystallinity (Vamadevan, Bertoft, & .05, Table S2), while a very strong correlation was found for SH30 (r = Seetharaman, 2013). The possibility that backbones of interlinked AM2 0.90, p = .005). The limitation of Method 1 to separate the groups was (DP > 120) accommodate longer IB-CL (more crystalline amylopectin made apparent when equations 1 and 2 are applied using single-point structures) than MCAP (DP 37–120) backbones could potentially explain hydrolysis rate SH90 and hydrolysis index HI (encompassing 0–180 their opposing correlations with GI. Similarly, RS negatively correlated min), respectively (Fig. 3c and Fig. A1). Actual in vivo GI values ranged with MCAP only (p < .05), while the latter negatively correlated with between 50.4 and 90, while predicted GI values using equations 1 and 2 AM1 and AM2, and positively with SCAP1 and SCAP3 (p < .05). Still (pGI1a and pGI1b, respectively) fall between 64.9-71.1 and 66.9–73.0, based on the backbone model (Vamadevan et al., 2013), short IB-CL is respectively. Consequently, the plot between pGI1a and pGI1b and in vivo thought to be associated with higher number of building blocks per GI only achieved R2 values of 0.732 and 0.690. cluster which may explain positive correlation between MCAP and To check whether the more intact grains will improve the separation SCAP, whereas longer IB-CL (which we hypothesize to be correlated of digestion profiles, Method 1 was slightly modified (referred as with AM2) had fewer building blocks and may thus explain the strong Modified Method 1). As hypothesized, a significant improvement in the negative correlation between AM1 and fractions MCAP and SCAP. The separation of digestion curves according to GI categories (Fig. 2b) and propensity of a rice cultivar to form more AM2 than MCAP backbone as correlation with GI (Fig. 3b) was established for all SH and AUC pa- the amylopectin chains are being trimmed and elongated could lead to a rameters in the modified method compared to that of the Method 1 more stable crystalline structure with fewer amylopectin short chain (Figs. 2a and 3a). Among the AUC ranges, highest correlation was double helices and thus lower GI. Consistent with this, moderate RS lines established between GI and AUC (30–60) (r = 0.97, p < .001) for which were determined to have elevated AM2 as compared to other starch the linear relationship is described by the equation pGI1c = 0.04 × AUC regions (Parween et al., 2020). (30–60) + 43.86 (R2 = 0.932, Fig. 3d). Simultaneously, Method 2 was performed to test whether well- resolved digestion profiles consistent with the trend in the GI of the 3.2. Evaluation of in vitro digestion methods for GI prediction samples would be achieved. Final hydrolysis rates were between 49.73% (IRRI147) and 60.08% (IR65), as presented in Fig. 2c. It showed better The soundness of the most widely used Method 1 (Goñi et al., 1997) separation during the first 60 min according to in vivo GI values, as an in vitro method to predict the GI was evaluated using the same set resulting to higher positive correlation with GI (with r = 0.93-0.95 be- of milled rice samples with known in vivo GI. It was compared to a tween 0 and 10 min and r = 0.77-0.89 between 10 and 60 min, p < .01- second in vitro method (Alhambra et al., 2019) previously used for .05, Fig. 4a), while succeeding time points had no significant correlation in-house GI screening. Finally, a slightly modified version of the simu- with GI. AUC(0–5) achieved the highest correlation with GI (r = 0.96, p lated digestion method by Germaine et al. (2008) employing lower = .0007) and this linear relationship is represented by the equation pGI enzyme concentrations than Method 2 was used. The components of 2 = 0.44 × AUC(0–5) + 33.43 (R2 = 0.917, Fig. 4c). each method are summarized in Table 2. Fig. 2d shows the digestion curves of the same samples when sub- As a starting point, Method 1 was employed to predict GI based on jected to Method 3 which produced better separation between GI groups starch hydrolysis rate at 90 min (SH90) and HI, as previously employed compared to Method 1 and 2. In contrast to the previous methods, in other studies (Deepa et al., 2010; Fernandes et al., 2020; Kale et al., Method 3 produced a gradual increase in starch digestion rates between 2015; Kunyanee & Luangsakul, 2018). By the end of the digestion 0 and 180 min, with final values ranging between 16.78% (IRRI147) and period, hydrolyzed starch values ranged between 63.96% (IRRI163) to 57.66% (IR65). Curves for GQ02522, GQ02497, and IRRI147 (GI = 50.4–55.1), and IRRI163 and IR64 (GI = 64–66) were clustered Table 2 accordingly while that of IR65 (GI = 90) was clearly separated from the Summary of the steps and components of the simulated digestion methods used rest. Correlation of the SH values and corresponding AUCs calculated for in vitro GI prediction. from successive time point ranges with in vivo GI (Fig. 4b, Table S5) was Step Method 1 Method 2 Method 3 highest for SH180 (r = 0.980, p < .001) and AUC (150–180 min) (r = Cooking 50 ± 0.1 mg 500 ± 10 mg milled 300 ± 3 mg milled 0.976, p < .001). A good linear relationship between in vivo GI and AUC flour grains grains (150–180) is described by the equation pGI3 = 0.031 × AUC(150–180) +3 mL water +6 mL water (excess +0.6 mL water + 37.60 (R2 = 0.953) to estimate GI based on AUC(150–180) as shown water discarded after in Fig. 4d. However, it is important to note that very high significant cooking) Amylase – 37.5 U/0.5 mL 111 U/3 mL correlations were also established using any of the SH and AUC values, digestion medium digestion medium with the lowest being r = 0.957 for AUC(0–30). Moreover, it should be (75 U/mLa) (37 U/mLa) noted that SH180 values of Method 3 (16.78%–30.60%, excluding IR65) Incubation – 20 s, 37 ◦C 75 s, 37 ◦C, 60 rpm fall within the SH10 values of Method 2, which may be due to the higher Pepsin 0.93 mg (40 U; 5 mg (17,820 U; 1.47 mg (63 U; 7 U/ 3.03 U/mLa) 3240 U/mLa mLa) enzyme concentration and conceivably higher digestion rates in the Incubation 60 min, 40 ◦C, 30 min, 37 ◦C, 200 30 min, 37 ◦C, 120 latter (Table 2). This also explains higher SH/AUC correlations with GI 90 rpm rpm rpm established at earlier time points in Method 2. As observed with the Amylase or 2.6 U amylase/ 10 mg Pancreatin- 0.35 mg Pancreatin- various starch structure parameters (Section 3.1), selected simulated Pancreatin- 30 mL digestion 127 U AMG/35.5 mL 195 U AMG/39 mL digestion variables AUC(0–5) from Method 2 and AUC(150–180) from AMG medium (0.087 digestion medium digestion medium U/mLa) (0.28 mg/mLa, 3.57 (0.009 mg/mLa, 5 U/ Method 3 comply with the assumptions of linear regression (Table 3). U/mLa) mLa) Various in vitro enzymatic digestion conditions such as enzyme Incubation 180 min, 37 ◦C, 180 min, 37 ◦C, 700 180 min, 37 ◦C, 120 concentration and sample form affect the rate of starch digestion 120 rpm rpm rpm (Woolnough, Monro, Brennan, & Bird, 2008) and may explain the a Final concentration in the digestion medium. observed discrepancy across the three methods. In contrast to the first 6 P.A. Pautong et al. L W T 168 (2022) 113929 Fig. 2. Percentage of starch hydrolyzed in various samples during simulated digestion using (a) Method 1, (b) modified Method 1, (c) Method 2, and (d) Method 3. The samples and their corresponding in vivo GI values are represented as follows (in decreasing order): (—■—) IR65, GI = 90; (—□—) IR64, GI = 66; (—◆—) IRRI163, GI = 64; (—⋄—) IRRI162, GI = 57; (—▴—) IRRI147, GI = 55; (—△—) GQ02497, GI = 51.5; (—○—) GQ02522, GI = 50.4. The bars represent standard errors of the mean (SEM) using three (3) to five (5) replicates. Fig. 3. Correlation matrix showing the association between in vivo GI and SH and AUC at various time points and ranges measured in (a) Method 1 and (b) modified Method 1, and the corresponding linear regression plots between in vivo GI and (c) predicted GI values (pGI1a) using the single time point equation by Goñi et al. [9] (pGI1a = 39.21+(0.803 × SH90); converted from white bread to glucose as reference by multiplying by 0.71, or setting GI = 71 for white bread), and (d) AUC(30–60) from the modified Method 1 to derive the new equation pGI1c = 0.04 × AUC(30–60) + 43.86. 7 P.A. Pautong et al. L W T 168 (2022) 113929 Fig. 4. Correlation matrix describing the association between in vivo GI and SH and AUC at various time points and ranges measured in (a) Method 2 and (b) Method 3, and their corresponding linear regression plots between in vivo GI and (c) AUC(0–5) from the Method 2 and (d) AUC(150–180) from Method 3 to derive the equations for predicted GI (pGI2 and pGI3, respectively). Table 3 Summary of pGI equations derived in the study using selected parameters. Linear Regression ANOVA Shapiro-Wilk Breusch-Pagan R2 F-value p valuea W-stat p valueb BP value p valuec pGI(RS) = − 18.42 × RS + 87.30 − .809 9.470 .028 .939 .629 0.885 .347 pGI(AM1) = − 3.96 × AM1 + 92.16 − .941 38.657 .002 .909 .393 1.773 .183 pGI(AM2) = − 3.47 × AM2 + 94.65 − .802 9.017 .030 .914 .428 2.800 .094 pGI(MCAP) = 5.33 × MCAP − 82.85 .780 7.782 .038 .913 .414 2.074 .150 pGI(SCAP) = 2.48 × SCAP − 76.12 .855 13.557 .014 .899 .326 3.496 .062 pGI(SCAP1) = 10.11 × SCAP1 − 91.17 .856 13.688 .014 .894 .296 2.869 .090 pGI(AM) = − 1.92 × AM + 94.71 − .886 18.326 .008 .955 .775 3.185 .074 pGI(AMAUC) = − 1.89 × AMAUC + 90.67 − .880 17.114 .009 .939 .631 3.390 .066 pGI2 = 0.44 × AUC(0–5) + 33.43 .917 38.282 .002 .904 .357 1.960 .162 pGI3 = 0.031 × AUC(150–180) + 37.60 .953 102.361 <.001 .963 .845 1.038 .308 pGI(M) = (0.021 × AUCd) − (1.66 × AM1) + 58.65 .995 403.2 <.001 .880 .227 2.483 .289 a <.05 corresponds to compliance with linearity assumption. b .05 corresponds to compliance with normal distribution of the residuals of linear regression. c .05 corresponds with compliance to homoscedasticity. d AUC(150–180) from Method 3. The summary of linear regression analysis is presented in Tables A5-A.14. two methods which produced high starch hydrolysis rates at the onset of absence of AMG and initial α-amylase digestion. Meanwhile, although the digestion period, Method 3 induced slower digestion as well as lower both Method 2 and 3 made use of whole milled grain samples, the use of final SH across samples. This could be attributed to food matrix integ- higher amounts of enzymes at all three stages in the former most likely rity. For instance, the use of cooked rice flour in Method 1 may have caused the higher SH values, despite the higher AMG concentration in caused the high onset SH across all samples despite the use of lower units Method 3. In addition, this effect may have been compounded by the fact of α-amylase (0.087 U/mL digestion medium) compared to that of that Method 2 employed excess water in the cooking step and possible Method 2 (11.3 U/mL) and Method 3 (0.441 U/mL), and even in the increased mechanical breakdown due to higher stirring speed. The same 8 P.A. Pautong et al. L W T 168 (2022) 113929 effect of cooking in excess water on starch digestibility has been reduced food matrix macro-structure in both Method 1 and 2 may have observed in previous studies (Hsu, Lu, Chang, & Chiang, 2014; Huynh, diminished the inherent variations in starch fine structure and other Shrestha, & Arcot, 2016). It is highly likely that higher SH also will have inherent hierarchical structures across samples, with SH almost reaching been measured if the excess water used for cooking in Method 2 was not a plateau at earlier time points. In contrast, the more intact samples discarded, since amylose and amylopectin leach out of the starch digested at lower pancreatin and AMG concentrations (as in Method 3) granules upon cooking (Ong & Blanshard, 1995). In this regard, we may have introduced lower digestion rates that best reflected SCAP DP hypothesize that certain levels of enzyme concentrations, to some variations. For instance, it can be observed that final SH in Method 3 extent, is only of secondary importance to matrix integrity when it were substantially below %TS content and may likely include mostly comes to affecting starch digestibility. Overall, the use of whole milled SCAP digestion. This can be supported by the results from Benmoussa grain, lower water-to-rice ratio (2:1 vs in excess), and lower enzyme et al. (2007), where they determined that amylopectin fine structure activities in Method 3 may have produced the more gradual and lower distribution affects in vitro digestibility of rice cultivars, although their extent of starch hydrolysis compared to both Method 1 and 2. In fact, results showed a negative correlation between MCAP and RDS. modifying Method 1 by simply employing whole grains (100 mg) and To establish an in vitro GI method that can predict in vivo values with 1:2 water to rice ratio, instead of rice flour and excess water, respec- higher accuracy, SH and AUC measured from in vitro digestion of milled tively, led to the establishment of higher correlations with in vivo GI grains are thus superior parameters to use than starch fractions (Fig. 5). (Fig. 2b), thereby supporting this hypothesis. This could be due to the ability of the method to account for the effects of The discrepancy between the correlations with GI established by various starch parameters (AM1, AM2, SCAP2, RS) and other underlying starch fine structure components and simulated digestion of more intact factors such as the presence of starch-lipid and starch-lipid-protein starting material may highlight the importance of the overall macro- and complexes (Wang et al., 2020), dietary fiber (Qi, Al-Ghazzewi, & micro-structures in the release of glucose from starch over the total Tester, 2018), phenolic compounds (Giuberti, Rocchetti, & Lucini, 2020; content of amylose and amylopectin alone (Chi et al., 2021). For Zhu, 2015) which are known to affect starch digestibility and thus GI. instance, the distribution of DP within amylose and amylopectin chains Although the use of the single time-point measurement (SH) poses an have been found to influence digestibility; higher short-medium advantage over AUC measurements with respect to throughput, the use amylose were found to be associated with lower digestibility (Gong, of the latter is less sensitive to errors than SH and is therefore more Cheng, Gilbert, & Li, 2019; Yu, Tao, & Gilbert, 2018) due to their ten- preferred. The earlier time points of Method 3 (e.g. 0–30 min) can also dency upon retrogradation to form smaller and densely packed gel be used without significantly affecting the predicted GI (R2 = 0.91, networks that are less accessible to amylolytic enzymes (Yu et al., 2018). Fig. A3). The comparison of three different in vitro digestion methods With respect to amylopectin, short-chain double helices (DP 6–12) are and fitting the results into models suggests that both Method 2 and generally positively associated with increased digestibility while DP Method 3 show potential as in vitro screening methods for GI in rice. 25–36 and DP ≥ 37 are known to reduce it (Lin et al., 2016; Srichuwong However, their full utility as an alternative to in vivo measurements is et al., 2005). However, these fractions could behave otherwise; short remains to be further explored involving large number of varieties with helices (DP 6–12) may contribute to the slowly digestible starch (SDS) in vivo GI evidence. fraction by either packing into ordered structures (Lin et al., 2016) or promoting steric hindrance against proper enzyme-substrate complex 3.3. Evaluation of the predictive capacity of linear regression models for formation (Li & Zhu, 2017), while long-chain helices may comprise the GI readily digestible starch (RDS) fraction when not in an ordered structure (Kim et al., 2020; Zhang, Ao, & Hamaker, 2008). Going up the hierar- A summary of the equations derived from the various methods and chical starch structure, short-range ordered structures associated with parameters which showed statistically significant correlation with in SDS and RS (Chi et al., 2019), and starch-complexes with non-starch vitro GI were then used to calculate the pGI of the seven samples of moieties such as hydrocolloids, proteins, and phenolics decrease di- known in vivo GI (Table 4), where pGI(AM1), pGI2 and pGI3 equations gestibility due to localized molecular interactions upon cooking (Chen gave GI values with less than 7% mean percentage error (MPE = 6.6%, et al., 2019). Densely-packed single and double helices by amylose and 5.4%, and 4.0%, respectively). Interestingly, a multiple linear regression amylopectin, respectively, form different crystalline structures A-, B-, C-, model built from AUC(150–180) of Method 3 and AM1 (Table A.15) and and V-type (starch-hydrophobic molecule complexes) categorized to described by the equation pGI(M) = (0.021 × AUC(150–180)) – (1.66 × have high, low, intermediate digestibility and resistant, respectively. AM1) + 58.65 (R2 = 0.995, p < .001) gave an MPE of only 1.1%. This Due to the less dense structures of B- and C-type starch crystals that implies increased predictive capacity when multiple parameters that allow for the incorporation of water molecules, they are less readily strongly correlate with GI are used. This result was not surprising given digested compared to A-type (Shrestha et al., 2012). It was reported that that SEC provides details on the fine structure while simulated digestion cereals of low to normal amylose content had proportionally high con- could account for the macro- and micro-structure of the food matrix. tent of A-type crystals with high short:long amylopectin ratio (DP < 24: Bland-Altman plots (Fig. 5) show the bias (deviation from an ideal DP > 36), B-type crystals with longer amylopectin side chains were mean difference of zero) and agreement limits (within which 95% of found in potato starch, while high-amylose starches in cereal and pea differences between the two methods lie) (Giavarina, 2015) between in had intermediate proportions (Martens et al., 2018). It was also reported vivo GI and pGI when selected parameters obtained for the 7 samples in the same study that both %A-type structure and amylopectin were fitted in their corresponding linear regression model (except for side-chain length distribution were the only parameters among those pGI1a which was calculated using the equation by Goñi et al., 1997). tested (amylose content, granule size, and number of pores) that Another set of in vitro values (pGI4, measured at the Commonwealth explicitly predicted the variations in starch digestion kinetics (in vitro Scientific and Industrial Research Organization (CSIRO) and extracted pig model) across selected botanical types (rice, barley, corn, wheat, from Anacleto et al. (2019) was also evaluated (Fig. A4h). In terms of the potato and pea). Thicker crystalline lamellae relative to the amorphous mean of differences, most of the models gave a zero bias (or mean dif- lamella (where disordered amylose and amylopectin chains are located), ference of 0) except for pGI1a and pGI1b (− 5.2 and − 7.0 units) and pGI4 highly-ordered reassembled aggregates, and granule surface proteins (+3.8 units), indicating that the said in vitro methods will give estimates and lipids are all associated with reduced digestibility. In essence, ori- that are 5.2 and 7.0 units higher, and 3.8 units lower, respectively, than entations at various levels of the starch hierarchical structure that limits in vivo values. In terms of the range of differences between in vitro and in or slows down enzyme action reduce starch digestibility (Chi et al., vivo values, the linear regression model with the most accurate agree- 2021). ment range with in vivo GI was observed for pGI3 (− 6.0 to 6.0 GI units), High rates of digestion due to high enzyme concentration and which means that 95% of in vitro GI measurements will differ by ±6.0 9 P.A. Pautong et al. L W T 168 (2022) 113929 Fig. 5. Bland-Altman plots for (a) pGI(RS), (b) pGI (AM1), (c) pGI1a, (d) pGI2, (e) pGI3, and (f) pGI(M) showing the differences between their estimated values and in vivo GI vs. the mean of both measure- ments. The middle horizontal line represents the “bias” (deviation from an ideal mean difference of zero) while top and bottom horizontal lines represent upper and lower limits within which 95% of the dif- ferences will fall (calculated as bias ± 1.96 std. de- viation, respectively). pGI(M) is a multiple linear regression model derived AUC(150–180) of Method 3 and AM1. Table 4 pGI values of the seven (7) standard rice samples using various regression models. Predicted GI Rice Samples GQ0255 GQ02497 IRRI147 IRRI162 IRRI163 IR64 IR65 MPEa pGI(RS) 49.1 ± 5.1 65.9 ± 15.0 56.7 ± 2.6 54.6 ± 8.4 68.6 ± 4.6 56.9 ± 7.0 81.9 ± 0.9 9.7 pGI(AM1) 46.8 ± 8.3 47.4 ± 5.6 58.0 ± 5.8 62.7 ± 9.3 67.7 ± 5.4 67.7 ± 9.2 83.6 ± 5.1 6.6 pGI(AM2) 45.3 ± 6.7 51.8 ± 8.1 56.4 ± 0.1 66.0 ± 2.5 71.6 ± 7.1 67.8 ± 4.0 74.9 ± 4.3 8.7 pGI(MCAP) 49.5 ± 4.0 54.4 ± 2.9 57.8 ± 2.1 68.5 ± 9.9 70.2 ± 6.4 54.6 ± 3.8 79.0 ± 3.6 10.3 pGI(SCAP) 46.4 ± 1.7 49.1 ± 10.7 57.5 ± 2.7 62.3 ± 4.5 68.7 ± 3.9 73.2 ± 3.3 76.8 ± 4.6 8.5 pGI(SCAP1) 44.4 ± 4.0 49.9 ± 6.5 61.3 ± 0.1 64.1 ± 1.3 70.5 ± 12.8 65.1 ± 2.8 78.7 ± 3.5 9.0 pGI(AM) 45.4 ± 1.1 49.3 ± 7.2 57.0 ± 2.9 64.6 ± 3.8 70.1 ± 5.3 68.0 ± 3.1 79.6 ± 3.5 7.9 pGI(AMAUC) 44.8 ± 1.7 51.1 ± 5.9 56.5 ± 3.5 64.2 ± 4.3 71.7 ± 1.0 66.3 ± 3.0 79.5 ± 3.5 7.3 pGI2 53.9 ± 2.7 50.2 ± 0.3 53.4 ± 2.5 61.1 ± 0.3 57.6 ± 0.6 69.7 ± 3.1 87.9 ± 13.3 5.4 pGI3 54.3 ± 2.7 54.8 ± 1.1 52.6 ± 1.8 56.8 ± 1.4 59.8 ± 4.4 64.7 ± 2.2 90.9 ± 2.6 4.0 pGI(M) 50.6 ± 1.8 51.2 ± 0.7 54.2 ± 1.2 58.9 ± 0.9 62.9 ± 2.9 66.2 ± 1.5 90.0 ± 1.7 1.1 in vivo GI 50.4 51.5 55 57 64 66 90 – a Mean percentage error. units compared to the mean of the two measurements. In contrast, wider “passed”, and otherwise, “failed”. Based on AUROC (Hosmer et al., agreement intervals measured from the bias were observed for the 2013), both pGI(AM1) and pGI2 gave acceptable predictions while pGI3 following (in increasing order): pGI(AM1) (±6.0), pGI(AM2) (±6.2), pGI has an excellent predictive ability. Interestingly, the multiple linear (AM) (±6.3), pGI(AMAUC) (±6.4), pGI(SCAP1) (±6.5), pGI(SCAP) regression model pGI(M) derived from the parameters AM1 and AUC (±6.8), pGI(MCAP) (±7.2), pGI2 (±7.7), pGI4 (±11.7), pGI(RS) (±15.8), (150–180) of Method 3 gave and outstanding prediction of GI. These pGI1a (±22.9), and pGI1b (±23.2). Notably higher predictive capacity results further support the conclusion that these models may be further was observed for the multiple regression equation pGI(M), achieving a explored for the potential to predict GI at acceptable capacities or levels. narrow ±1.9 units of agreement interval. A total of 10 market samples were then used to demonstrate the Receiver operating characteristic (ROC) curves (Fig. 6) were gener- phenotypic variability of estimated GI across different methods ated for pGI(AM1), pGI2, and pGI3 using individual replicates for the (Table 3). Models pGI(AM1), pGI(MCAP), and pGI3 were able to predict seven rice accessions to graphically illustrate their ability to predict in GI values that are close to that reported for low-GI rice (LGR) at 49.9, vivo values. For this purpose, pGI measurements that fell within ±1 to 45.8, and 55.3, respectively (Table 5). Estimates of GI for Iddly rice (IR) ±10 units (at increments of ±1) from the in vivo GI were classified as were also within, or at least close to, the low GI range except for pGI(RS) 10 P.A. Pautong et al. L W T 168 (2022) 113929 Table 5 Predicted GI (pGI) values of market samples across various pGI models. Samples GI Predicted GI pGI(RS) pGI pGI pGI pGI (AM1) (AM2) (MCAP) (SCAP) IR 38a 61.7 ± 37.7 ± 59.8 ± 41.7 ± 54.6 ± 5.1 0.6 0.4 1.9 1.6 CR 83a 77.4 ± 57.2 ± 75.7 ± 55.1 ± 72.0 ± 2.5 3.9 2.0 2.3 4.9 LGR 54a 73.4 ± 49.9 ± 70.1 ± 45.8 ± 67.7 ± 2.0 2.0 0.0 2.3 2.3 LBR 52a 68.6 ± 56.3 ± 71.2 ± 58.7 ± 66.6 ± 7.9 4.8 2.4 6.0 7.5 BR 50b/ 76.3 ± 63.7 ± 70.9 ± 55.9 ± 72.3 ± 58c 0.7 2.7 1.1 4.7 1.3 WR 91a 79.1 ± 79.3 ± 80.6 ± 79.0 ± 78.2 ± 3.7 0.2 2.9 18.2 10.4 WGR 93a 80.9 ± 91.2 ± 92.3 ± 115.2 ± 77.3 ± 2.8 0.2 2.0 9.0 5.7 Fig. 6. Receiver operating characteristic (ROC) curves comparing the in vitro GI BGR 42a/ 80.5 ± 86.8 ± 90.8 ± 103.7 ± 78.7 ± values obtained using linear regression models pGI(AM), pGI2, pGI3, and pGI 74d 2.4 2.1 0.2 12.3 7.1 (M) against in vivo GI. A single measurement using either method was catego- TRR 76d 78.9 ± 68.5 ± 71.3 ± 81.0 ± 63.9 ± rized as “passed” (otherwise, “failed”) when it is within ±1 up to 0.8 19.8 15.4 7.4 20.0 ±10 GI units (at increments of ±1 unit) of the in vivo value of the seven (7) rice samples used. BJR 74 81.0 ± 86.1 ± 87.6 ± 80.7 ± 86.6 ± 1.5 0.4 2.9 3.8 3.6 pGI(M) is a multiple linear regression model derived AUC(150–180) of Method 3 and AM1. The area under the respective ROCs are as follows (in increasing Samples pGI pGI pGI pGI2 pGI3 order): (—) chance diagonal, 0.5; (—△—) pGI(AM1), 0.714; (—◆—) pGI2, (SCAP1) (AM) (AMAUC) 0.781; (—⬤—) pGI3, 0.891; and (—▴—) pGI(M), 0.969. Interpretation: 0.5 < IR 38a 49.9 ± 49.0 ± 47.9 ± 63.0 ± 57.7 ± ROC <0.7 = poor prediction; 0.7 ≤ ROC <0.8 = acceptable prediction; 0.8 ≤ 1.2 0.5 0.7 15.0 0.7 ROC <0.9 = excellent prediction; ROC ≥0.9 = outstanding prediction (Hosmer, CR 83a 68.2 ± 67.2 ± 65.5 ± 90.0 ± 74.8 ± Lemeshow, & Sturdivant, 2013); TPR = True Positive Rate; FPR = False Posi- 2.5 3.0 3.2 6.6 2.1 LGR 54a tive Rate. 58.7 ± 60.6 ± 59.2 ± 56.6 ± 55.3 ± 0.5 1.0 0.8 0.6 0.4 LBR 52a 65.5 ± 64.4 ± 63.5 ± 51.6 ± 64.9 ± and pGI2 which gave higher values. All models were able to assign high 2.8 3.6 3.4 11.0 0.1 GI values for both glutinous white and black rices (GWR and GBR) which BR 50b/ 68.5 ± 67.8 ± 66.8 ± 58.4 ± 72.6 ± is expected for waxy rices (Kaur et al., 2016). All models also classified 58c 2.4 0.7 0.6 0.4 1.2 WR 91a 83.8 ± 80.7 ± 79.1 ± 67.5 ± 74.4 ± the Jasmine rices (WR and BJR) accordingly as high GI cultivars, while 1.4 1.5 1.4 4.2 1.0 Thai Red as intermediate to high. Meanwhile, the Basmati rices (LBR WGR 93a 106.9 ± 93.0 ± 89.8 ± 93.2 ± 97.8 ± and BR) got low to high pGI values. The differences in pGI using eight 3.6 1.2 0.5 3.2 0.0 models was smallest for Extra-Long Basmati Rice (LBR), differing at least BGR 42a/ 105.0 ± 90.0 ± 86.8 ± 48.4 ± 68.7 ± 10.1 units, while the pGI for White Glutinous Rice (WGR) and Black 74d 2.6 1.1 1.3 9.4 0.3 TRR 76d 79.9 ± 70.3 ± 64.8 ± 66.7 ± 79.3 ± Glutinous Rice (BGR) varied for more than 40 units across models, 20.3 18.2 13.8 5.3 0.4 highlighting the variability of the predictive capacities of different BJR 74 100.3 ± 87.9 ± 84.4 ± 71.4 ± 85.5 ± methods, and thus, the need to come up with the most accurate one. 3.6 1.4 1.1 1.6 0.1 The values are expressed as the mean of three (3) replicates for RS, two (2) for 4. Conclusion SEC-derived parameters, pGI2, and pGI3 ± standard deviation. aBased on liter- ature search by Kaur, Lim, Chusak, and Henry (2020); cRanawana, Henry, This work aimed to establish an in vitro method for GI screening in Lightowler, and Wang (2009); dFoster-Powell, Holt, and Brand-Miller (2002); rice by exploring the use of amylolysis, starch indices, and starch frac- eIndrasari, Purwani, Wibowo, and Jumali (2010); IR=Iddly Rice; CR= Premium tions as parameters for GI prediction. This study demonstrated that Calrose Rice; LGR = Kangaroo Australian Low GI Rice; LBR = Extra Long Bas- resistant starch (RS), amylose (AM1, DP > 1000, and AM2, 121 < DP ≤ mati Rice; BR=Basmati Rice; WR=Thai Hom Mali Premium Quality Fragrant 1000), and short-chain amylopectin (SCAP, particularly SCAP of DP Rice (Jasmine white rice); WGR=White Glutinous Rice; BGR=Black Glutinous 25–36) are strongly to very strongly correlated with in vivo GI. Mean- Rice; TRR = Premium Thai Red Rice; BJR=Brown Jasmine Rice. while, higher correlations were observed for SH and AUC measured through modified in vitro amylolysis protocol employing intact grains, and AM2, starch-lipid and starch-lipid-protein complexes, and enzyme- low sample requirements, and lower enzyme activities, even within inhibiting phytochemicals) affecting starch digestibility. However, AM1 shorter digestion periods of 5 and 30 min. However, lower mean per- data can further improve predictive capacity. In addition, we found that centage errors were achieved when AUC(150–180) and AM1 were previously reported pGI equations did not accurately predict in vivo GI combined into a multiple regression model. We conclude that simulated for the rice accessions used in this study, and hence, the need to generate digestion Method 3 can be explored for in predicting the GI in closer to in more accurate pGI equations specific to an in vitro digestion protocol, vivo situation. We infer that amylolysis of cooked whole milled grains at which was demonstrated herein. Future studies involving higher num- conditions that do not diminish the effect of grain structural integrity ber of samples with profiles on protein, lipids, and phenolics content, (whole milled grain, lower rice-to-water ratio during cooking, and lower and more variety of botanical sources could be used to overcome some of concentration of amylolytic enzymes) and thus improved in vitro the limitations of the current study. methods exhibit more accurate predictive capacity than the use of more specific starch indices and starch fractions. This could be due to the Funding ability of digestion models, with certain incubation conditions, to simultaneously account for differences in various factors (e.g. SCAP2 Nese Sreenivasulu acknowledge the funding support of CGIAR Rice Program, Foundation for food and agriculture research (FFAR), 11 P.A. Pautong et al. L W T 168 (2022) 113929 Department of Agriculture, Philippines (OneRicePH), Temperate Rice Brodkorb, A., Egger, L., Alminger, M., Alvito, P., Assunção, R., Ballance, S., et al. (2019). Research Consortium and Indian Council of Agriculture Research (ICAR- INFOGEST static in vitro simulation of gastrointestinal food digestion. Nature Protocols, 14(4), 991–1014. https://doi.org/10.1038/s41596-018-0119-1 IRRI work plan). A portion of this work is a component of a PhD Brotman, Y., Llorente-Wiegand, C., Oyong, G., Badoni, S., Misra, G., Anacleto, R., et al. dissertation partially funded by the Commission on Higher Education (2021). The genetics underlying metabolic signatures in a brown rice diversity panel (CHED), Philippines. and their vital role in human nutrition. The Plant Journal, 106(2), 507–525. https:// doi.org/10.1111/tpj.15182 Butardo, V. M., Anacleto, R., Parween, S., Samson, I., de Guzman, K., Alhambra, C. M., Declaration of competing interest et al. (2017). Systems genetics identifies a novel regulatory domain of amylose synthesis. Plant Physiology, 173(1), 887–906. https://doi.org/10.1104/pp.16.01248 Castro, J. V., Dumas, C., Chiou, H., Fitzgerald, M. A., & Gilbert, R. G. (2005). Mechanistic The authors declare that they have no known competing financial information from analysis of molecular weight distributions of starch. interests or personal relationships that could have appeared to influence Biomacromolecules, 6(4), 2248–2259. https://doi.org/10.1021/bm0500401 the work reported in this paper. Chen, L., Zhang, H., McClements, D. J., Zhang, Z., Zhang, R., Jin, Z., et al. (2019). Effect of dietary fibers on the structure and digestibility of fried potato starch: A comparison of pullulan and pectin. Carbohydrate Polymers, 215, 47–57. https://doi. CRediT authorship contribution statement org/10.1016/j.carbpol.2019.03.046 Chi, C., Li, X., Huang, S., Chen, L., Zhang, Y., Li, L., et al. (2021). Basic principles in starch multi-scale structuration to mitigate digestibility: A review. Trends in Food Putlih Adzra Pautong: Methodology, Validation, Formal analysis, Science & Technology, 109, 154–168. https://doi.org/10.1016/j.tifs.2021.01.024 Investigation, Writing – original draft. Joanne Jerenice Añonuevo: Chi, C., Li, X., Lu, P., Miao, S., Zhang, Y., & Chen, L. (2019). Dry heating and annealing Methodology, Validation, Formal analysis, Investigation, Writing – treatment synergistically modulate starch structure and digestibility. International Journal of Biological Macromolecules, 137, 554–561. https://doi.org/10.1016/j. original draft. Maria Krishna de Guzman: Methodology, Validation. ijbiomac.2019.06.137 Rodolfo Sumayao: Conceptualization, Writing – review & editing. Chung, H.-J., Shin, D.-H., & Lim, S.-T. (2008). In vitro starch digestibility and estimated Christiani Jeyakumar Henry: Writing – review & editing. Nese Sree- glycemic index of chemically modified corn starches. Food Research International, 41 (6), 579–585. https://doi.org/10.1016/j.foodres.2008.04.006 nivasulu: Conceptualization, Writing – review & editing, Visualization, Deepa, G., Singh, V., & Naidu, K. A. (2010). A comparative study on starch digestibility, Supervision, Project administration, Funding acquisition. glycemic index and resistant starch of pigmented (“Njavara” and ’Jyothi’) and a non- pigmented (’IR 64’) rice varieties. Journal of Food Science & Technology, 47(6), 644–649. https://doi.org/10.1007/s13197-010-0106-1 Data availability Edwards, C. H., Cochetel, N., Setterfield, L., Perez-Moral, N., & Warren, F. J. (2019). A single-enzyme system for starch digestibility screening and its relevance to Data will be made available on request. understanding and predicting the glycaemic index of food products. Food & Function, 10(8), 4751–4760. https://doi.org/10.1039/C9FO00603F Englyst, K. N., Vinoy, S., Englyst, H. N., & Lang, V. (2003). Glycaemic index of cereal Acknowledgements products explained by their content of rapidly and slowly available glucose. British Journal of Nutrition, 89(3), 329–340. https://doi.org/10.1079/BJN2002786 Fernandes, J.-M., Madalena, D. A., Pinheiro, A. C., & Vicente, A. A. (2020). Rice in vitro We acknowledge the following individuals for their contributions: digestion: Application of INFOGEST harmonized protocol for glycemic index Anna Natoza from the Consumer-Driven Grain Quality and Nutrition determination and starch morphological study. Journal of Food Science & Technology, (CDGQN) Cluster at IRRI who assisted in SEC experiments and other 57(4), 1393–1404. https://doi.org/10.1007/s13197-019-04174-x Fitzgerald, M. A., Rahman, S., Resurreccion, A. P., Concepcion, J., Daygon, V. D., starch-related assays; Luster May Serrano-Labarga from CDGQN for Dipti, S. S., et al. (2011). Identification of a major genetic determinant of glycaemic HPLC troubleshooting tips and insightful discussions with PATP; and, index in rice. Rice, 4(2), 66–74. https://doi.org/10.1007/s12284-011-9073-z Asst. Professor Anna Jean Garcia from the University of Southern Foster-Powell, K., Holt, S. H. A., & Brand-Miller, J. C. (2002). International table of Mindanao for her comments on statistical analysis. gylcemic index and glycemic load values. American Journal of Clinical Nutrition. https://doi.org/10.1093/ajcn/76.1.5, 2002. Germaine, K. A., Samman, S., Fryirs, C. G., Griffiths, P. J., Johnson, S. K., & Quail, K. J. Appendix A. Supplementary data (2008). Comparison of in vitro starch digestibility methods for predicting the glycaemic index of grain foods. Journal of the Science of Food and Agriculture, 88(4), 652–658. https://doi.org/10.1002/jsfa.3130 Supplementary data to this article can be found online at https://doi. Giavarina, D. (2015). Understanding Bland Altman analysis. Biochemia Medica, 25(2), org/10.1016/j.lwt.2022.113929. 141–151. https://doi.org/10.11613/BM.2015.015 Giuberti, G., Rocchetti, G., & Lucini, L. (2020). Interactions between phenolic compounds, amylolytic enzymes and starch: An updated overview. Current Opinion in References Food Science, 31, 102–113. https://doi.org/10.1016/j.cofs.2020.04.003 Gong, B., Cheng, L., Gilbert, R. G., & Li, C. (2019). Distribution of short to medium Alhambra, C. M., Dhital, S., Sreenivasulu, N., & Butardo, V. M. (2019). Quantifying grain amylose chains are major controllers of in vitro digestion of retrograded rice starch. digestibility of starch fractions in milled rice. In N. Sreenivasulu (Ed.), Methods in Food Hydrocolloids, 96, 634–643. https://doi.org/10.1016/j.foodhyd.2019.06.003 molecular biology (Vol. 1892, pp. 241–252). New York, NY: Humana Press. https:// Goñi, I., Garcia-Alonso, A., & Saura-Calixto, F. (1997). A starch hydrolysis procedure to doi.org/10.1007/978-1-4939-8914-0_13. estimate glycemic index. Nutrition Research. https://doi.org/10.1016/S0271-5317 Anacleto, R., Badoni, S., Parween, S., Butardo, V. M., Misra, G., Cuevas, R. P., et al. (97)00010-9 (2019). Integrating a genome-wide association study with a large-scale GRiSP (Global Rice Science Partnership). (2013). Rice Almanac: Source book for one of transcriptome analysis to predict genetic regions influencing the glycaemic index the most important economic activities on earth. In IRRI (4th ed.). International Rice and texture in rice. Plant Biotechnology Journal, 17(7), 1261–1275. https://doi.org/ Research Institute. Los Baños, Philippines. 10.1111/pbi.13051 Guzman, M. K. de, Parween, S., Butardo, V. M., Alhambra, C. M., Anacleto, R., Seiler, C., Atkinson, F. S., Foster-Powell, K., & Brand-Miller, J. C. (2008). International tables of et al. (2017). Investigating glycemic potential of rice by unraveling compositional glycemic index and glycemic load values. Diabetes Care. https://doi.org/10.2337/ variations in mature grain and starch mobilization patterns during seed germination. dc08-1239, 2008. Scientific Reports, 7(1), 5854. https://doi.org/10.1038/s41598-017-06026-0 Ball, S., Guan, H. P., James, M., Myers, A., Keeling, P., Mouille, G., et al. (1996). From Hanashiro, I., Abe, J., & Hizukuri, S. (1996). A periodic distribution of the chain length of glycogen to amylopectin: A model for the biogenesis of the plant starch granule. Cell, amylopectin as revealed by high-performance anion-exchange chromatography. 86(3), 349–352. https://doi.org/10.1016/s0092-8674(00)80107-5 Carbohydrate Research, 283, 151–159. https://doi.org/10.1016/0008-6215(95) Banerjee, S., Chandel, G., Mandal, N., Meena, B., & Saluja, A. (2011). Assessment of 00408-4 nutritive value in milled rice grain of some Indian rice landraces and their molecular Hosmer, D. W., Jr., Lemeshow, S., & Sturdivant, R. X. (2013). Assessing the fit of the characterization. Bangladesh Journal of Agricultural Research, 36. https://doi.org/ model. In D. W. Hosmer, Jr., S. Lemeshow, & R. X. Sturdivant (Eds.), Applied logistic 10.3329/bjar.v36i3.9265 regression (pp. 153–225). https://doi.org/10.1002/9781118548387.ch5 Benmoussa, M., Moldenhauer, K. A. K., & Hamaker, B. R. (2007). Rice amylopectin fine Hsu, R., Lu, S., Chang, Y.-H., & Chiang, W. (2014). Effects of added water and structure variability affects starch digestion properties. Journal of Agricultural and retrogradation on starch digestibility of cooked rice flours with different amylose Food Chemistry, 55(4), 1475–1479. https://doi.org/10.1021/jf062349x content. Journal of Cereal Science, 61. https://doi.org/10.1016/j.jcs.2014.03.002 Bertoft, E. (2004). On the nature of categories of chains in amylopectin and their Huynh, T. D., Shrestha, A. K., & Arcot, J. (2016). Physicochemical properties and connection to the super helix model. Carbohydrate Polymers, 57(2), 211–224. https:// digestibility of eleven Vietnamese rice starches with varying amylose contents. Food doi.org/10.1016/j.carbpol.2004.04.015 & Function, 7(8), 3599–3608. https://doi.org/10.1039/C6FO00661B Blaak, E. E., Antoine, J. M., Benton, D., Björck, I., Bozzetto, L., Brouns, F., et al. (2012). Hu, P., Zhao, H., Duan, Z., Linlin, Z., & Wu, D. (2004). Starch digestibility and the Impact of postprandial glycaemia on health and prevention of disease. Obesity estimated glycemic score of different types of rice differing in amylose contents. Reviews. https://doi.org/10.1111/j.1467-789X.2012.01011.x Journal of Cereal Science, 40(3), 231–237. https://doi.org/10.1016/j.jcs.2004.06.001 12 P.A. Pautong et al. L W T 168 (2022) 113929 Indrasari, S. D., Purwani, E. Y., Wibowo, P. S. L., & Jumali. (2010). Glycemic indices of mechanisms. Plant Biotechnology Journal, 18(8), 1763–1777. https://doi.org/ some rice varieties. Indonesian Journal of Agriculture, 3(1), 9–16. 10.1111/pbi.13339 Jenkins, D. J. A., Wolever, T. M. S., Taylor, R. H., Barker, H., Fielden, H., Baldwin, J. M., Pasion, E. A., Badoni, S., Misra, G., Anacleto, R., Parween, S., Kohli, A., et al. (2021). et al. (1981). Glycemic index of foods: A physiological basis for carbohydrate OsTPR boosts the superior grains through increase in upper secondary rachis exchange. American Journal of Clinical Nutrition, 34(3), 362–366. https://doi.org/ branches without incurring a grain quality penalty. Plant Biotechnology Journal, 19 10.1093/ajcn/34.3.362 (7), 1396–1411. https://doi.org/10.1111/pbi.13560 Jukanti, A. K., Pautong, P. A., Liu, Q., & Sreenivasulu, N. (2020). Low glycemic index Perez, S., & Bertoft, E. (2010). The molecular structures of starch components and their rice—a desired trait in starchy staples. Trends in Food Science & Technology, 106, contribution to the architecture of starch granules: A comprehensive review. Starch - 132–149. https://doi.org/10.1016/j.tifs.2020.10.006 Stärke, 62, 389–420. https://doi.org/10.1002/star.201000013 Juliano, B. (1985). Polysaccharides, proteins and lipids of rice. Qi, X., Al-Ghazzewi, F. H., & Tester, R. F. (2018). Dietary fiber, gastric emptying, and Kale, S. J., Jha, S. K., Jha, G. K., Sinha, J. P., & Lal, S. B. (2015). Soaking induced changes carbohydrate digestion: A mini-review. Starch - Stärke, 70(9–10), Article 1700346. in chemical composition, glycemic index and starch characteristics of Basmati rice. https://doi.org/10.1002/star.201700346 Rice Science, 22(5), 227–236. https://doi.org/10.1016/j.rsci.2015.09.002 Ranawana, D. V., Henry, C. J. K., Lightowler, H. J., & Wang, D. (2009). Glycaemic index Kaur, B., Lim, J., Chusak, C., & Henry, C. J. (2020). Microwave cooking enhances of some commercially available rice and rice products in Great Britain. International glycaemic potential of rice: An in vitro study. Malaysian Journal of Nutrition, 26(1), Journal of Food Sciences & Nutrition. https://doi.org/10.1080/09637480802516191 117–128. https://doi.org/10.31246/MJN-2019-0106 RStudio Team. (2020). RStudio. Boston, MA: Integrated Development for R. RStudio, Kaur, B., Ranawana, V., & Henry, J. (2016). The glycemic index of rice and rice products: PBC. URL http://www.rstudio.com/. A review, and table of GI values. Critical Reviews in Food Science and Nutrition, 56(2), Shrestha, A. K., Blazek, J., Flanagan, B. M., Dhital, S., Larroque, O., Morell, M. K., et al. 215–236. https://doi.org/10.1080/10408398.2012.717976 (2012). Molecular, mesoscopic and microscopic structure evolution during amylase Kim, H. R., Choi, S. J., Choi, H.-D., Park, C.-S., & Moon, T. W. (2020). Amylosucrase- digestion of maize starch granules. Carbohydrate Polymers, 90(1), 23–33. https://doi. modified waxy potato starches recrystallized with amylose: The role of amylopectin org/10.1016/j.carbpol.2012.04.041 chain length in formation of low-digestible fractions. Food Chemistry, 318, Article Srichuwong, S., Sunarti, T. C., Mishima, T., Isono, N., & Hisamatsu, M. (2005). Starches 126490. https://doi.org/10.1016/j.foodchem.2020.126490 from different botanical sources I: Contribution of amylopectin fine structure to Kim, H. J., & White, P. J. (2012). In vitro digestion rate and estimated glycemic index of thermal properties and enzyme digestibility. Carbohydrate Polymers, 60(4), 529–538. oat flours from typical and high β-glucan oat lines. Journal of Agricultural and Food https://doi.org/10.1016/j.carbpol.2005.03.004 Chemistry, 60(20), 5237–5242. https://doi.org/10.1021/jf300429u Tetlow, I. J., & Bertoft, E. (2020). A review of starch biosynthesis in relation to the Koroteeva, D. A., Kiseleva, V. I., Sriroth, K., Piyachomkwan, K., Bertoft, E., Yuryev, P. V., building block-backbone model. International Journal of Molecular Sciences, 21(Issue et al. (2007). Structural and thermodynamic properties of rice starches with different 19). https://doi.org/10.3390/ijms21197011 genetic background Part 1. Differentiation of amylopectin and amylose defects. Toutounji, M. R., Butardo, V. M., Zou, W., Farahnaky, A., Pallas, L., Oli, P., et al. (2019). International Journal of Biological Macromolecules, 41(4), 391–403. https://doi.org/ A high-throughput in vitro assay for screening rice starch digestibility. Foods, 8(Issue 10.1016/j.ijbiomac.2007.05.010 12). https://doi.org/10.3390/foods8120601 Kunyanee, K., & Luangsakul, N. (2018). The utilization of ultrasound and chilling Toutounji, M. R., Farahnaky, A., Santhakumar, A. B., Oli, P., Butardo, V. M., & treatment to reduce GI in Thai glutinous rice (RD6). International Journal of Blanchard, C. L. (2019). Intrinsic and extrinsic factors affecting rice starch Agricultural Technology, 14, 1365–1378. https://scholar.google.com/scholar?hl digestibility. Trends in Food Science & Technology, 88(10170), 10–22. https://doi. =en&as_sdt=0%2C5&q=Kunyanee%2C+K.%2C+%26+Luangsakul%2C+N.+% org/10.1016/j.tifs.2019.02.012 282018%29.+The+utilization+of+ultrasound+and+chilling+treatment+to+re Tutusaus, J. A. M., Srikaeo, K., & Diéguez, J. G. (2013). Effects of amylose and resistant duce+GI+in+Thai+glutinous+rice+%28RD6%29.+&btnG=. starch on starch digestibility of rice flours and starches. International Food Research Lal, M. K., Kumar, A., Raigond, P., Dutt, S., Changan, S. S., Chourasia, K. N., et al. (2021). Journal, 20, 1329–1335. Impact of starch storage condition on glycemic index and resistant starch of cooked Vamadevan, V., Bertoft, E., & Seetharaman, K. (2013). On the importance of organization potato (Solanum tuberosum) tubers. Starch - Stärke, 73(1–2), Article 1900281. of glucan chains on thermal properties of starch. Carbohydrate Polymers, 92(2), https://doi.org/10.1002/star.201900281 1653–1659. https://doi.org/10.1016/j.carbpol.2012.11.003 Leoro, M. G. V., Clerici, M. T. P. S., Chang, Y. K., & Steel, C. J. (2010). Evaluation of the Wang, S., Chao, C., Cai, J., Niu, B., Copeland, L., & Wang, S. (2020). Starch–lipid and in vitro glycemic index of a fiber-rich extruded breakfast cereal produced with starch–lipid–protein complexes: A comprehensive review. Comprehensive Reviews in organic passion fruit fiber and corn flour. In Food science and technology (Vol. 30, pp. Food Science and Food Safety, 19(3), 1056–1079. https://doi.org/10.1111/1541- 964–968). scielo. 4337.12550 Lin, L., Guo, D., Huang, J., Zhang, X., Zhang, L., & Wei, C. (2016). Molecular structure Wei, T., & Simko, V. (2021). GitHub - taiyun/corrplot: A visual exploratory tool on and enzymatic hydrolysis properties of starches from high-amylose maize inbred correlation matrix (Version 0.92). https://github.com/taiyun/corrplot. lines and their hybrids. Food Hydrocolloids, 58, 246–254. https://doi.org/10.1016/j. Wickham, H., François, R., Henry, L., & Müller, K. (2022). Dplyr: A grammar of data foodhyd.2016.03.001 manipulation. R package version 1.0.8 https://CRAN.R-project.org/package=dplyr. Li, G., & Zhu, F. (2017). Amylopectin molecular structure in relation to physicochemical Woolnough, J. W., Monro, J. A., Brennan, C. S., & Bird, A. R. (2008). Simulating human properties of quinoa starch. Carbohydrate Polymers, 164, 396–402. https://doi.org/ carbohydrate digestion in vitro: A review of methods and the need for 10.1016/j.carbpol.2017.02.014 standardisation. International Journal of Food Science and Technology. https://doi.org/ Martens, B. M. J., Gerrits, W. J. J., Bruininx, E. M. A. M., & Schols, H. A. (2018). 10.1111/j.1365-2621.2008.01862.x Amylopectin structure and crystallinity explains variation in digestion kinetics of World Health Organization. (2020). Malnutrition. starches across botanic sources in an in vitro pig model. Journal of Animal Science and World Health Organization. (2021). Non-communicable diseases. World Health Biotechnology, 9(1), 91. https://doi.org/10.1186/s40104-018-0303-8 Organization. https://www.who.int/news-room/fact-sheets/detail/nonco Miller, J. B., Pang, E., & Bramall, L. (1992). Rice: A high or low glycemic index food? mmunicable-diseases. American Journal of Clinical Nutrition. https://doi.org/10.1093/ajcn/56.6.1034 Yan, L. J. (2014). Pathogenesis of chronic hyperglycemia: From reductive stress to Nakamura, Y. (2018). Rice starch biotechnology: Rice endosperm as a model of cereal oxidative stress. Journal of Diabetes Research, 2014. https://doi.org/10.1155/2014/ endosperms. Starch - Stärke, 70(1–2), Article 1600375. https://doi.org/10.1002/ 137919 star.201600375 Yu, W., Tao, K., & Gilbert, R. G. (2018). Improved methodology for analyzing relations Nanditha, A., Ma, R. C. W., Ramachandran, A., Snehalatha, C., Chan, J. C. N., Chia, K. S., between starch digestion kinetics and molecular structure. Food Chemistry, 264, et al. (2016). Diabetes in Asia and the Pacific: Implications for the global epidemic. 284–292. https://doi.org/10.1016/j.foodchem.2018.05.049 Diabetes Care, 39(3), 472. https://doi.org/10.2337/dc15-1536. LP – 485. Zeileis, A., & Hothorn, T. (2002). Diagnostic checking in regression relationships. R News, Ong, M. H., & Blanshard, J. M. V. (1995). Texture determinants of cooked, parboiled rice. 2(3), 7–10. URL https://CRAN.R-project.org/doc/Rnews/. II: Physicochemical properties and leaching behaviour of rice. Journal of Cereal Zhang, G., Ao, Z., & Hamaker, B. R. (2008). Nutritional property of endosperm starches Science, 21(3), 261–269. https://doi.org/10.1006/jcrs.1995.0029 from maize mutants: A parabolic relationship between slowly digestible starch and Parween, S., Anonuevo, J. J., Butardo, V. M., Jr., Misra, G., Anacleto, R., Llorente, C., amylopectin fine structure. Journal of Agricultural and Food Chemistry, 56(12), et al. (2020). Balancing the double-edged sword effect of increased resistant starch 4686–4694. https://doi.org/10.1021/jf072822m content and its impact on rice texture: Its genetics and molecular physiological Zhu, F. (2015). Interactions between starch and phenolic compound. Trends in Food Science & Technology, 43. https://doi.org/10.1016/j.tifs.2015.02.003 13