Thirteen genotypes were evaluated for sensory attributes in this research, including five small-seeded genotypes (S1, S9, S11, S25, and Huayu 20) and eight large-seeded genotypes (L1, L8, L9, L10, L11, L15, L17, and Huayu 33). The control genotypes (Huayu 20 and Huayu 33) were normal oleic acid and high yield peanut varieties, while the others were high yield and high oleic acid peanuts lines ( Table 1 ). The experiments were carried out in Jinzhou (JZ) and Fuxin (FX) located in Liaoning Province, as well as in Weifang (WF) located in Shandong Province. The geographical location, elevation and climatic data of the experimental locations were presented in Table 2 . The soil types in these three locations are all sandy loam soils that are suitable for peanut cultivation. The comprehensive soil parameters regarding the physical and chemical characteristics of the experimental fields are listed in Table 3 . All genotypes were sown under film mulch and harvested under the local standard recommended procedures of fields management for the respective locations in 2017. A randomized complete block design with 3 replications was used in this experiment. After harvesting, peanuts from all three locations were sun-dried, ensure that the moisture content of all seeds is close to 8%, then sound and mature seeds were collected for sensory evaluation.

The peanuts from each location were roasted in a T3-L383B Meidi oven (Meidi Ltd. Company, China). The oven temperature was maintained at 190°C for 5 and 7 minutes for roasting small-seeded and large-seeded genotypes, respectively. After roasting, all peanuts were allowed to cool down to room temperature in a natural way, then samples were packaged in food-grade plastic bags and stored in them, the sensory evaluation was carried out on the same day.

An 8-member trained peanut profile panel (4 female and 4 male) at Shandong Peanut Research Institute, Qingdao, Shandong, participated in the spectrum descriptive analysis. All panelists were chosen based on the following specific criteria: (1) people who have experience in the evaluation of peanut products, (2) people are not allergic to peanuts, (3) people with natural dentition, (4) non-smokers, (5) non-drug users, (6) people who interested in participating, and (7) people who able to verbally communicate regarding the product (Nepote et al., 2009). Sensory panelists were trained and their performance assessed according to The Spectrum™ Method (Sensory Spectrum, Inc., Chatham, NJ, USA). All panelists selected for this study possessed over 1000 hours of experience in sensory evaluation and training specific to peanuts and peanut products. Their comprehensive training included terminology development, introduction to scaling, initial practice, discrimination of small product differences, and final calibration, all conducted using peanuts and peanut products as test materials. The panelists were given the attributes definitions ( Table 4 ) on how to evaluate all samples using a five-point scale (e.g., 1=weak, 5=strong) for crunchiness (1 (weak) to 5 (strong)), fineness (1 (weak) to 5 (strong)), sweetness (1 (weak) to 5 (strong)), bitterness (1 (weak) to 5 (strong)), off-flavor (1(weak) to 5 (strong)), roasted peanutty (1(weak) to 5 (strong)) and overall acceptability (1 (dislike) to 5 (like)). Attributes definitions were developed based on relevant literature, previous research findings, and deliberations by the sensory panel (Johnsen et al., 1988; Nepote et al., 2006; Miyagi, 2017) ( Table 4 ). The panelists evaluated samples from the same location at a time in a randomized complete block design with 3 replications. The order of samples was random for each panelist. Three seeds of each sample were tasted for texture evaluations and pasted samples were tasted for the flavor attributes, then recording the scores in the given form. Before testing each sample, panelists should rinse their mouths with warm water to minimize the influence of the previous sample. When testing with multiple panelists simultaneously, communication and conversation were not allowed.

Biochemical quality of peanut kernels inclusive of proanthocyanidin (Zhang, 2023), sucrose contents (Tang et al., 2018), oleic acid (Yu et al., 2010), linoleic acid (Yu et al., 2010), palmitic acid (Yu et al., 2010), oil (Yu et al., 2003), protein (Yu et al., 2003), and vitamin E (Liu et al., 2018) were analyzed by near infrared reflectance spectroscopy (NIRS) Matrix-I (Bruker Optics Inc., German), and the software version was OPUS 7.5. To reduce measurement errors, warm up the machine for at least 30 minutes. All samples were naturally sun-dried and were at room temperature for over 6 hours before scanning to minimize the impact of the material’s temperature on the results. Each sample was scanned 3 times and the average value was calculated.

The data were collected using Excel 2016 and analysis with DPS 18.1 software package. A multi-site randomized block design was used to evaluate the sensory attributes for the roasted peanuts, the biochemical components for the raw peanuts. A combined analysis of variance was first conducted according to the experiment design, with genotypes and locations as fixed effects. Variation in sensory attribute and biochemical component scores was partitioned into parts due to environment, genotype, G×E interaction, and error. Duncan’s multiple range test (DMRT) was utilized for multiple comparison. The GGE biplot analysis with Which-Won-Where, Mean versus Stability, Discrimitiveness versus representativeness and Ranking genotypes was done by using the R studio version 4.3.1 (Development Core Team, Vienna), the GGE Biplot GUI package. The relationship among crunchiness, fineness, sweetness, bitterness, off-flavor, roasted peanutty, and overall acceptability, and inter-relationships between peanut biochemical components of raw kernels, sensory attributes of roasted kernels were revealed by Pearson’s correlation analysis using R 4.3.1, the corrplot, ggplot2, and ggpubr packages.

The results of the analysis of the variables and G × E interactions are presented in Table 5 . All the sensory attributes exhibited significant differences at the 0.01 level due to genotype × location effects, and there were highly significant differences in the variability among the tested genotypes. Locations also had significant effects on all sensory attributes, except for off-flavor and roasted peanutty. In other words, there was no significant difference in the aroma attribute among locations after roasting. There were no significant differences in the variability among blocks in locations for all sensory attributes.

As shown in Table 6 , peanuts from Weifang showed great crunchiness, more fineness, high sweetness and low bitterness compared to those from the other two locations. Regarding the aroma attribute, there was no significant difference in off-flavor and roasted peanutty among the three locations. Peanuts from Weifang obtained the highest score in overall acceptability.

The means for the 13 genotypes tested ( Table 7 ) indicated that the ranges of crunchiness, fineness, sweetness, bitterness, off-flavor, roasted peanutty, and overall acceptability. The ranges of the fineness and roasted peanutty attributes for small-seeded and large-seeded peanuts were overlapping, and the small-seeded types showed a significantly (P ≤0.05) higher mean than the large-seeded types for fineness (4.02 vs. 3.59 intensity units), and a significantly (P ≤ 0.05) lower mean for roasted peanutty (3.78 vs. 3.98 intensity units). Among all genotypes, S9 had the best sensory attributes, with high crunchiness and sweetness, and at the same time, low bitterness and off-flavor. L10 had higher crunchiness, high roasted peanutty, and the highest overall acceptability among the large-seeded peanuts.

From Figure 1 , there is a visual presentation of the differences in sensory attributes among the three different locations for roasted peanuts.

The GGE biplot for which-won-where of sensory attributes ( Figures 2A–G ) showed that the Component 1 and Component 2 accounted for 85.13%, 85.96%, 90.46%, 85.55%, 79.59%, 88.61%, and 80.97% of total variation for crunchiness, fineness, sweetness, bitterness, off-flavor, roasted peanutty, and overall acceptability, respectively ( Table 8 ). The polygon containing all genotypes was formed by lines which connecting the signs of the genotypes that are farthest away from the biplot origin, and then draw the perpendicular lines to each side of the polygon from origin point, which divided the whole biplot into several sectors. Sectors help to visualize the mega-environments. The winning genotypes for each sector are placed at the vertex. The test environments fell into two of the seven sectors, one of the six sectors, one of the five sectors, two of the seven sectors, two of the five sectors, two of the eight sectors, and one of the five sectors outlined on the polygon view for crunchiness, fineness, sweetness, bitterness, off-flavor, roasted peanutty, and overall acceptability, respectively ( Figure 2 ). Thus, the mega-environments were identified for each sensory attribute. There were one mega-environment for fineness, sweetness, and overall acceptability, respectively. There were two mega-environments for crunchiness and bitterness with FX and JZ grouped together in a mega-environment, while WF is in the second mega-environment. There were two mega-environments for off-flavor and roasted peanutty with WF and FX grouped together in a mega-environment, while JZ is in the second mega-environment. From Figure 2A , 3 locations were divided into 2 sectors, S9 was the best performer at FX and JZ, and L10 was the vertex genotype at WF in crunchiness. From Figures 2B, C, G , 3 locations were belonged to 1 sector, S9 was the vertex genotype at this environment in fineness, sweetness, and overall acceptability, respectively.

The Figure 3 showed the “mean vs. stability” pattern of GGE biplot for sensory attributes of the 13 peanut genotypes evaluated. The abscissa and the ordinate of the average environment coordinate (AEC) axis are the two lines passing through the origin of the biplot, and the AEC showed the trend of average sensory attribute scores of peanuts across different locations. The abscissa points toward increased order of genotype performance. A perpendicular line drawn from the origin to the AEC revealed the tendency of genotype-environment interaction (GEI). The projection on the abscissa toward the ordinate of the AEC is a measure of stability, so a genotype with zero projection or very short direction from the ordinate is considered the most stable, while a genotype with the longest projection from the abscissa is unstable. In our study, the ordinate divides the genotypes into two: those that sensory attributes above and below average. Figure 3A showed the genotypes toward the arrow that crunchiness above the average means: S9, S25, L10, L9, HY20, L11, and L17; In contrast, the following fell below the average means: L15, L1, S11, S1, L8, and HY33. Based on crunchiness, fineness, sweetness, bitterness, off-flavor, roasted peanutty, and overall acceptability, S9, S9, S9, L9, L1, L10 and S9 were the highest genotypes, while HY33, L17, HY20, S9, S9, HY20, and HY33 were the lowest genotypes, and HY20, S9, HY20, S25, L9, L17, and S1 had the shortest projection, while L1, S11, L17, HY33, S25, S1, and L15 had the longest.

Discriminativeness vs. representativeness pattern of the GGE biplot for sensory attributes was presented in Figure 4 . They pinpoint how the best environment can be informative and representative. The two concepts focus on environments in terms of their ability to detect the best genotypes (discriminativeness) and to adequately represent the test environments (representativeness) (Yan and Tinker, 2006; Oladosu et al., 2017; Khan et al, 2021). The vector length for each environment revealed the discriminatory ability of the environment, while the angle formed by each vector with the abscissa denotes representativeness. The longer the vector of an environment, the higher its capability to discriminate among genotypes, while the shorter the angle formed with the abscissa, the more it is representative. The shortest vector for Patterns B, D, and F was FX, for Patterns C, E, and G was FX, and for Pattern A was JZ, while the longest vector for Patterns A and B was WF, for Patterns C, D, E and F was JZ, and for Patterns G was FX. The angles the environment formed with the abscissa line were also recorded. For crunchiness, the shortest angle was formed by JZ. For fineness, bitterness, and roasted peanutty the shortest angle was formed by FX. For sweetness, off-flavor, and overall acceptability, the shortest angle was formed by WF. The longest angles formed by the environment were also observed to be FX for sweetness and overall acceptability, and JZ for fineness, off-flavor, and roasted peanutty, while for crunchiness and bitterness, they were formed by WF. The biplot identified the environments that were closest to the AEC. The environments were JZ for crunchiness, and FX for fineness, bitterness, and roasted peanutty, while WF for sweetness, off-flavor, and overall acceptability.

In Figure 5 , genotype ranking based on PC1 and PC2 of sensory attributes showing G×E interactions of 13 peanut genotypes under 3 locations. The biplot allowed us to identify the best and ideal genotype among the 13 tested genotypes. An ideal genotype is always localized into the innermost circle and virtually closer to the head of the arrow at the center of the circular ring (Esan et al., 2023) ( Figure 5 : Patterns A, B, C, D, E, F and G). The only genotype that fell inside the inner circle was S9; the genotypes next to the ideal inner circle were S25 and L10, followed by L9, HY20, L17, and L11 for crunchiness ( Figure 5A ). Similarly, for fineness, S9, HY20 and S1 were next to the ideal genotype ( Figure 5B ). For sweetness, the only genotype was S9, while L8, L1, S1, L10, S25, L15 and L17 followed the only ideal genotype ( Figure 5C ). For roasted peanutty, the ideal inner circle was L17 and L10 ( Figure 5F ). For overall acceptability, the ideal inner circle was S9, followed by L10 and L17 ( Figure 5G ).

The results of the analysis of the variable components and G×E interactions are listed in Table 9 . All the biochemistry components exhibited significant differences at the 0.05 or 0.01 level due to the effects of locations, genotypes, and genotype × location. There were no significant differences in the variability among blocks in locations except for oil.

As shown in Table 10 , the content of proanthocyanidin and vitamin E in peanuts from Weifang had a significantly (P ≤ 0.05) lower mean, while the content of sucrose and protein in peanuts from Weifang had a significantly (P ≤ 0.05) higher mean than that of the peanuts from the other two locations. The content of vitamin E in the peanuts from the Northeast peanut-growing area in China (Jinzhou and Fuxin) was significantly higher than that from the Shandong peanut-growing area.

The means for the 13 genotypes tested ( Table 11 ) showed the ranges of proanthocyanidin, sucrose, oleic, linoleic, palmitic, oil, protein and vitamin E for raw peanuts. There was no significant difference in the biochemical components of peanuts between the large-seeded and small-seeded peanuts, except for proanthocyanidins. The mean content of proanthocyanidin of large peanuts (1.89 mg/g) was significantly higher (P ≤ 0.05) than that of small peanuts (1.62 mg/g). The three small-seeded genotypes S1, S9 and Huayu20 had the low content of proanthocyanidin compared with the other genotypes.

The GGE biplot for which-won-where of biochemical components ( Figures 6A–H ) showed that the Component 1 and Component 2 accounted for 93.33%, 92.57%, 99.95%, 99.96%, 99.75%, 97.47%, 94.12%, and 90.27% of total variation for proanthocyanidin, sucrose, oleic, linoleic, palmitic, oil, protein and vitamin E, respectively ( Table 12 ). The test environments fell into two of the five sectors, one of the five sectors, one of the seven sectors, one of the six sectors, one of the six sectors, one of the six sectors, one of the seven sectors, two of the six sectors, and two of the six sectors outlined on the polygon view for proanthocyanidin, sucrose, oleic, linoleic, palmitic, oil, protein and vitamin E, respectively ( Figure 6 ). Thus, the mega-environments were identified for each biochemical component. There were one mega-environment for sucrose, oleic, linoleic, palmitic, and oil, respectively. There were two mega-environments for proanthocyanidin with WF and JZ grouped together in a mega-environment, while FX is in the second mega-environment. There were two mega-environments for protein and vitamin E with WF and FX grouped together in a mega-environment, while JZ is in the second mega-environment. From Figure 6A , 3 locations were divided into 2 sectors, one environment was WF and JZ, and FX was another one. L11 was the best performer at JZ and WF, and L1 was the vertex genotype at FX for proanthocyanidin. From Figures 6G, H , 3 locations were divided into 2 sectors, one environment was WF and FX, and JZ was another one. L15 was the best performer at FX and WF, and L17 was the vertex genotype at JZ for protein ( Figure 6G ), while S1 was the best performer at FX and WF, and no vertex genotype at JZ for Vitamin E ( Figure 6H ). From 6B-6F, 3 locations were belonged to 1 sector, S9, S25, HY20, HY33, and HY20 was the vertex genotype at this environment in sucrose, oleic acid, linoleic acid, palmitic acid, and oil, respectively.

The Figure 7 showed the “Mean vs. stability” pattern of GGE biplot for biochemical components of the 13 peanut genotypes evaluated. Based on proanthocyanidin, sucrose, oleic, linoleic, palmitic, oil, protein and vitamin E, L1, S9, S25, HY20, HY33, HY20, L15, and S25 were the highest genotypes, while HY20 (S1), HY20, HY20, S25, S25, L17, HY33, and S11 were the lowest genotypes. For proanthocyanidin, sucrose, oil, protein and vitamin E, L1, L9, L15, L1, and L10 had the longest projection, while HY33, L17, S1, L11, and S25 had the shortest. For oleic, linoleic and palmitic acid, we couldn’t identify the length of the projection through the Figures 7C–E .

Discriminativeness vs. representativeness pattern of the GGE biplot for biochemical components was shown in Figure 8 . The shortest vector for Patterns A, B, C, E, and G was JZ and for Patterns D, F, and H was FX, while the longest vector for Patterns A and C was FX and for Patterns B, E, and G was WF, and Patterns D, F and H was JZ. The angles the environment formed with the abscissa line were also recorded. For proanthocyanidin, sucrose, oleic, and palmitic, the shortest angle was formed by JZ. For oil and protein, the shortest angle was formed by WF. For linoleic and vitamin E, the shortest angle was formed by FX. The longest angles formed by the environment were also observed to be FX for proanthocyanidin, and JZ for oil, protein and vitamin E, while for sucrose and three fatty acids, they were formed by WF. The biplot identified the environments that were closest to the AEC. The environments were WF for oil and protein, while JZ for proanthocyanidin, sucrose, oleic, and palmitic, and FX for linoleic and vitamin E.

In Figure 9 , genotype ranking based on PC1 and PC2 of biochemical components showing G×E interactions of 13 peanut genotypes under 3 locations. The only genotype that fell inside the inner circle was HY33, followed by L11, L15 and L1 for proanthocyanidin ( Figure 9A ). The only genotype that fell inside the inner circle was S9, followed by L17 and S11 for sucrose ( Figure 9B ) and the best genotype HY20 for oil ( Figure 9F ). For oleic, the genotypes except HY20 and HY33 were all inside the circle ( Figure 9C ). For linoleic and palmitic, HY20 and HY33 fell inside the inner circle ( Figures 9D, E ). The genotypes next to the ideal inner circle were S11 and L15, followed by HY20, L10, L17, L8, and S9 for protein ( Figure 9G ). For vitamin E, the genotypes S1 and S25 were all inside the inner circle ( Figure 9H ).

Correlation analysis was used to determine the relationship among all the sensory attributes in roasted peanuts ( Figure 10 ). Overall acceptability was significantly positively correlated with crunchiness (r=0.28), fineness (r=0.19), sweetness (r=0.35), roasted peanutty(r=0.23) and significantly negatively correlated with bitterness(r=-0.25). Sweet-ness was significantly positively correlated with fineness (r=0.26), and significantly negatively correlated with bitterness (r=-0.25). Crunchiness was highly positive significant with fineness, sweetness and overall acceptability and highly negative significant with off-flavor.

The results presented in Figure 10 demonstrate the correlations between sensory attributes and biochemical components in this study. Proanthocyanidins exhibited a highly significant negative correlation with crunchiness, fineness, sweetness and overall acceptability, while showing a highly significant positive correlation with bitterness and off-flavor. Sucrose demonstrated a highly significant positive correlation with sweetness and a positive correlation with crunchiness. The fatty acid content was found to be correlated with crunchiness, fineness, sweetness, roasted peanutty and overall acceptability. Oil was negatively correlated with sweetness and roasted peanutty. Protein was negatively correlated with bitterness, and positively correlated with sweetness and roasted peanutty. Vitamin E was negatively correlated with roasted peanutty. In terms of overall acceptability, it was negatively correlated with Vitamin E and proanthocyanidin, and positively correlated with sucrose, oleic acid, and protein.