Recruitment and BSTI Administration for Skin Typing
Our study recruited 792 Korean female and 197 male who underwent comprehensive assessments comprising of skin condition measurements, questionnaires (only female) on the living environment and cosmetic use habits, and microbiome sampling via face wash. The resulting Korean-specific facial skin database included 1) BSTI survey data, 2) skin measurement data, and 3) microbiome data of subjects (Supplementary Table 1). By classifying the skin types based on the clinical skin measurement, we conducted an integrated analysis to identify the microbiota with significant distribution in each type. This prospective study aimed to investigate the relationship between skin conditions and microbiome diversity with a robust and reliable approach, focusing on a final 756 female participants. The study outline is illustrated in Fig. 1a.
Before a step in determining skin type by correlating clinical skin assessment with microbiome analysis, we utilized the Baumann Skin Type Indicator (BSTI) self-report questionnaire to classify the skin type of 752 Korean female participants into one of 16 types, based on a dichotomous classification of four contrasting categories (Supplementary Table 1). After calculating scores for each category and determining the type, we observed the distribution of scores for each category (Fig. 1b). The results showed that 64% of participants had dry skin, while 36% had oily skin. In terms of Sensitive vs. Resistant type, 53.5% of participants had resistant skin and 46.7% had sensitive skin. A vast majority of participants, 97.7%, had non-pigmented skin, while only 2.3% identified as having pigmented skin. For the Wrinkled vs. Tight type, 58.4% of participants identified their skin as tight and 41.6% as wrinkled. When we combined the scores for each control category for each participant, we found that DRNT was the most common skin type (155, 20.6%), followed by DRNW (130, 17.3%), DSNT (106, 14.1%), OSNT (100, 13.3%), DSNW (81, 10.8%), ORNT (73, 9.7%), OSNW (56, 7.4%), ORNW (35, 4.7%), DRPW (5, 0.7%), OSPW (4, 0.5%), and DSPT (2, 0.3%). We also found that DRPTs, DSPWs, ORPTs, ORPWs, and OSPTs were all represented by only one participant each (0.1%) (Fig. 1c).
Advanced Clinical Evaluations and Strategic Metric Selection for Optimal Skin Biophysical Assessment
It is possible that the choice of factors or variables used in a skin BSTI is arbitrary. It is also possible that the results will vary depending on which of the various factors that assess skin health are used. For this reason, a ganzheitliche (holistic) approach to assessing and improving skin health is required, taking into account a variety of factors. Thus, to quantitatively measure an individual's skin condition, a comprehensive assessment was conducted with 989 participants (792 female and 197 male), which assessed 39 biophysical parameters across 9 categories, including oiliness, water content, water loss, skin tone, skin elasticity, skin density, nasolabial folds, lateral canthal lines, and pores. In order to derive representative metrics of skin measurement among 39 variable parameters, we were guided by three main principles. Firstly, repeated and site-specific measurements were combined by taking an average to reduce measurement gap error and increase data reliability. Secondly, water content and water loss were combined to create a single metric representing the skin's moisture status, providing a more comprehensive picture of the skin's hydration level. Finally, representative parameters for skin tone, elasticity, density, pores, and wrinkles were selected based on their statistical correlation with each other and existing research findings, ensuring that the chosen parameters were valid and reliable (Fig. 2a and Supplementary Table 2).
For skin tone, the ITA (Individual Typology Angle) metric, which combines lightness (L*), greenness (a*), and yellowness (b*), was selected as a representative metric. Among the skin elasticity parameters, R7 showed a high correlation (Pearson Correlation Coefficient, PCC) with other elasticity parameters, R2 and R5 (R7-R5: 0.91, R7-R2: 0.92, R2-R5: 0.82). Thus, R7 was selected as a quantitative parameter representing elasticity since it is widely used in existing studies. Skin density (R: skin fold thickness) was excluded as a key parameter since R7 sufficiently represented it and showed moderate correlation (PCC) with other elasticity parameters (R7: 0.44, R: 0.36, R2: 0.37)32, 33. The roughness of the skin surface (Ra value) was selected as a wrinkle-related parameter, precisely one nasolabial fold Ra. Additionally, since there was a high correlation between nasolabial folds and lateral canthal lines (0.61), we selected only one nasolabial fold Ra for the representative parameter34. Through a process of correlation and normalization, six indicators were selected as representative of an individual's skin condition: oiliness, hydration, skin tone (ITA), elasticity (R7), average pore size (Pore), and nasolabial folds (Ra). This comprehensive set of metrics provides a more accurate and complete understanding of an individual's skin condition.
We employed a systematic evaluation process to determine the extent to which the six selected metrics accurately reflected the participants' skin condition and how it changed with age. Our first step involved creating a polygonal chart that visually displayed the trends of each metric against age (Fig. 2b). Our analysis of this chart revealed that skin elasticity and tone values were relatively higher in all age groups, with an average score of above 5, which is considered a relative average. Similarly, oiliness was also found to be above five from the teenage years to the 40s. However, we found that pores and wrinkles had an average value of more than 5 in the 40+ age group. In contrast, moisture levels exhibited a mean value ranging between 4 and 6 for all age groups, except for the teenage group, which had an average score of 3.48. Next, we compared the performance of a machine learning model that uses a full set of 39 parameters to predict age with a model that uses only six metrics to predict age. The Catboost algorithm, which has shown strong performance in recent AI competitions, was implicated in this evaluation35. When all 39 skin parameters were used to predict age, each age group had a ROC-AUC (The Receiver Operator Characteristic-Area Under Curve) score between 0.94 and 1, with a micro-average ROC-AUC score of 0.98 and a macro-average ROC-AUC score of 0.97. (Fig. 2c) The performance of the model in predicting each participant's age using only the six metrics resulted in ROC-AUC scores between 0.85 and 0.94 for each age group, with a micro-average ROC-AUC score of 0.91 and a macro-average ROC-AUC score of 0.90 (Fig. 2d).
Consistency and Efficiency of Age Changes in Skin Type Determination Criteria
After conducting the variable selection process, we re-selected the criteria of oil, moisture, tone, elasticity, pore, and wrinkle based on six major metrics (oiliness, hydration, skin tone (ITA), elasticity (R7), average pore size (Pore), and nasolabial folds (Ra)) to focus on the skin type and conduct patterning analysis. To determine the best representative criteria among the three metrics (elasticity, pores, and wrinkles), we conducted a Pearson correlation coefficient (PCC) analysis. The results revealed that elasticity was the most suitable representative indicator for wrinkles and pores (elasticity-pores: -0.54, elasticity-wrinkles: -0.50, pores-wrinkles: 0.31). Since skin tone and elasticity were found to be highly correlated, we combined them into a single criterion to represent color/elasticity. Similarly, we combined the remaining oil and moisture metrics into a single criterion to simplify patterning. When the entire sample was divided into tertiles for tone/elasticity and oil/moisture, there were clear fractionation differences (Fig. 3a). Among the five complexes based on the two criteria (tone/elasticity and oil/moisture), we focused on four types, HH, HL, LH, and LL without the gray zone, which were not in the medium (M) range in either tone/elasticity or oil/moisture (Fig. 3b). We defined each of the four groups as HH (high in both tone/elasticity and oil/moisture), HL (high in color/elasticity but low in oil/water), LL (low in both), and LH (low in color/elasticity but high in oil/water). The integrated coordinates of our skin criteria exhibit characteristic distributions across the four skin types (Fig. 3b).
The analysis of biological age-related changes in skin parameters revealed a noticeable clockwise shift in the distribution of ages in the integrated coordinates of the skin indicators. Specifically, younger individuals in their 10s and 20s were predominantly located on the HH zone, whereas older participants were in the LL zone (Fig. 3c). This trend is further supported by the skin condition of participants across age groups, particularly in terms of skin tone, which has been shown to generally darken with age36. The median skin tone values of participants by age group were found to be 46.2 in 10s, 43.5 in the 20s, 40.6 in the 30s, 37.5 in the 40s, 35.8 in the 50s, 35.1 in the 60s, 33.1 in the 70s, and 32.2 in the 80s (Supplementary Fig. 1a). Similarly, a gradual decline in skin elasticity was observed across age groups. The median elasticity values for each age group were as follows: 65.0 in 10s, 60.2 in the 20s, 51.5 in the 30s, 47.3 in the 40s, 43.8 in the 50s, 41.2 in the 60s, 39.0 in the 70s, and 38.0 in the 80s (Supplementary Fig. 1b). Oil criteria increase through the 20s and 30s and then decrease with age. Median oil value by age group was 34.3 in 10s, 37.2 in 20s, 37.3 in 30s, 30.8 in 40s, 20.8 in 50s, 11.8 in 60s, 8.7 in 70s, and 5.7 in 80s (Supplementary Fig. 1c). Moisture criteria did not appear to be strongly related to biological age. The median moisture value by age group is 2.6 in teens, 3.1 in 20s, 3.2 in 30s, 2.9 in 40s, 3.3 in 50s, 3.4 in 60s, 3.6 in 70s, and 3.1 in 80s (Supplementary Fig. 1d).
In order to conduct an integrated analysis of all subjects based on the type classification of HH/HL/LH/LL, we rearranged the groups in the grey zone (MH, MM, ML, HM, LM) to be classified into the above four types. In other words, participants with above-average values of the Tone-Elasticity (X-axis) and Oil-Moisture (Y-axis) value were classified as H, and those with below-average values were classified as L. By combining the criteria from these two axes, all participants were reclassified into four types: HH/HL/LH/LL. The skin is a complex phenotype that is strongly influenced by the process of biological aging. However, due to individual differences and varying conditions of the skin across different age groups, accurately and objectively identifying the different stages of aging can be challenging. To address these issues, we conducted a comparative analysis of age-specific changes in the top and bottom tertiles of the four key criteria. Interestingly, despite the diverse skin phenotypes of the 705 women recruited for the study, a marked inflection point was identified at ages 35 and 51 based on four key criteria (Fig. 4a). We found a reversal of the upper and lower groups in the Tone criteria at age 35. The proportion of the upper skin tone group decreased from 80% at age 34 to 27.3% at age 35, while the lower skin tone group surpassed the upper group, accounting for 54.5% at age 35. For the Elasticity criteria, the H and L groups became equal for the first time at age 36, both reaching 27.3%. In terms of the Oil criteria, there were two points of intersection between the upper and lower proportions at 14 and 51 years of age. We considered 51 years of age significant due to the trend of increasing oil volume until the age of 30 and then decreasing. At age 51, the proportion of the upper group was 14.3% and the proportion of the lower group was 28.6%. As for the Moisture criteria, we observed fluctuating proportions of the upper and lower groups from time to time. To minimize age bias error and age stratification error, we organized the aging groups based on the inflection points of the four skin criteria (Fig. 4a). We categorized 310 subjects aged 34 or younger into the Young group, 172 subjects aged 35 to 50 into the Aging I group, and 223 subjects aged 51 or older into the Old group. These groups represented 31.6%, 22.4%, and 44% of all women, respectively, as shown in Fig. 4b. We applied our skin type classification to each of these aging stages for further analysis.
Skin Type Classification with the Aging Groups
We divided the Aging group (Young, Aging I, and Old) into the four skin types (HH/HL/LH/LL) and examined the differences between a total of 12 types in a polygonal chart (Fig. 4c). We found significant differences in the mean values of the 6 metrics: skin tone, elasticity, pores, wrinkles, moisture, and oiliness for different skin types. Specifically, the yHH and yHL types had mean values above 8.7 for skin tone and elasticity, while the oLH and oLL types had mean values between 1.7 and 1.9. The mean values for the other eight groups were distributed between 3.3 and 7.6. As for oiliness, the aLH type had the highest mean value of 8.4, while the oHL and oLL types had the lowest mean value of 2.2. As for moisture, the mean values for the young group were yHH/yLH/yHL/yLL, and the same order was maintained in the Aging I and Old groups: aHH/aLH/aHL/aLL, oHH/oLH/oHL/oLL. As for wrinkles, the oLL type had the highest mean value of 7.9, but the oHH type had a mean value of 7.4. Finally, as for pores, the oLH and oLL types showed a similar trend to skin tone and elasticity, with a mean value of over 7, while the yHH and HL types had a mean value of under 2.1. Similar to the age-specific results, we also observed a clear trend of decreased skin tone and elasticity, and increased pores and wrinkles, as the aging groups progressed. The oiliness criterion showed a similar mean value for the Young and Aging I groups, at 6.5 and 6.2, respectively. However, the Old group exhibited a significant difference with a mean value of 3.2, indicating a noticeable oiliness decrease with age. The moisture criterion showed a mean relative value of 5 between the aging groups, ranging from 4.6 to 5.5. This suggests that there is not a significant difference in moisture levels across different age groups. Overall, the yLL, aLH, and oHH types were closest to the mean values of the respective aging groups, suggesting that these Korean skin cutotypes (KSCs) are most representative of the skin conditions in each aging group (Fig. 4c). We observed changes of criteria in tone/elasticity and oil/moisture according to aging groups. As expected, tone and elasticity gradually decreased from the Young group to the Old group, acting as an important factor in distinguishing age groups. Additionally, tone and elasticity move together in aging groups and KSC types. The H- type is more reflected in the elasticity criteria, while the L- type is more influenced by the tone. In contrast, the difference in oil and moisture according to aging groups was not dramatic, but a minor difference in oiliness was observed. The oil values of each age group show a downward trend. Those means that oil/moisture plays an important role in distinguishing individual phenotypic differences and KSCs across the aging groups (Fig. 5).
Exploring the Overall Status and Changes in Skin Microbial Community
We conducted a comparative analysis of skin microbiome with 950 (756 females and 194 males) among 985 subjects (784 females and 201 males) with age and clinical information (Fig. 6a and Supplementary Table 3). Our results showed that alpha-diversity was higher in females than in males, especially in those aged 50 or older. In females (n=756), the alpha-diversity gradually decreased up to their 40s and then increased in their 50s (p<0.01 for all indexes). In contrast, males (n=194) showed a decreasing trend in diversity from their 30s to their 50s and then an increase until their 60s, followed by a decline. The difference between males and females was significant in their 20s and 30s (p<0.05) (Fig. 6b). In terms of microbial composition, there were significant differences (p < 0.001) in Cutibacterium, Streptococcus, Staphylococcus, Rothia, and Neisseria genera between females and males (Fig. 6c and Supplementary Table 4). These findings provide valuable insights into the role of age and gender in shaping the skin microbiome. Changes in microbial community and abundance by sex and age were highly dynamic and tested for statistical significance using beta-diversity analysis with PERMANOVA test (Supplementary Fig. 2).
Significant diversity differences were observed in the top 10 genus compositions of females and males according to age groups (Fig. 6d). Specifically, in females, Streptococcus, Cutibacterium, Staphylococcus, Rothia, Neisseria, Actinomyces, Haemophilus, and Rhodopseudomonas showed significant differences (p<0.05, Kruskal-wallis), while in males, significant differences were found in Cutibacterium, Staphylococcus, Streptococcus, Rothia, Actinomyces, Neisseria, and Haemophilus (p<0.05, Kruskal-wallis) (Supplementary Table 4). After excluding samples with a significantly lower sample size (age < 10s), we conducted a correlation analysis (Spearman and Pearson) between age and the detected skin microbiome (genus) with a detection rate of over 1%. In females, we found significant age correlations with Rothia (Spearman R=0.383 and Pearson R=0.363), Neisseria (SR=0.283 ans PR=0.204), and Cutibacterium (SR= -0.299 and PR=-0.263) (Supplementary Fig. 3a). In males, we found significant correlations with Parvimonas (SR=0.304 and PR=0.194) and Lactobacillales (SR= -0.293 and PR=-0.275) (Supplementary Fig. 3b). The common skin microbiomes identified in both females and males were Filifactor (female SR=0.272 and PR=0.128, male SR=0.22 and PR=0.17) and Fretibacterium (female SR=0.224 and PR=0.043, male SR=0.255 and PR=0.166). From the age of around 30s, five microbial genera showed a significant difference in the skin microbiota between female and male, with varying levels of increase and decrease. Moreover, the time point of changes in the abundance of specific microorganisms was strongly correlated with the inflection point of clinical changes in the skin above. Interestingly, we observed that Staphylococcus decreased with age in females, whereas in males, it increased with age (Supplementary Fig. 3c). The detailed statistical summary of skin microbiome are annotated in Supplementary Table 4.
We also conducted an integrated analysis to identify vital microbials that influence the aging group of female based on skin clinical measurement criteria and to observe changes in the microbiome by the four skin type. Using data 750 female subjects who distinguished to the aging groups of Young, Aging I, and Old by skin clinical measurement criteria, an alpha-diversity comparison analysis were performed. The results showed no statistically significant difference in diversity between Young and Aging I groups, but both groups showed a significant difference from the Old group (p<0.05, Kruskal-wallis) (Supplementary Fig. 4a). Among the top 10 genera, there were statistically significant differences in Cutibacterium, Streptococcus, Rothia, Neisseria, Actinomyces, Haemophilus, Fusobacterium, and Veillonella (p<0.05, Kruskal-wallis) between the aging groups (Supplementary Fig. 4b). Regarding the aging group, a LEfSe (Linear discriminant analysis Effect Size) analysis was conducted at the Genus level, and it was found that the feature microorganisms for each aging group were Lawsonella (Young group), Cutibacterium (Aging I group), and Streptococcus (Old group) (Supplementary Fig. 4c). All statistical values are summarized in Supplementary Table 5.
Comparison of Skin Microbiome among the Aging Groups in Korean Women
The enterotype algorithm is a widely-used method in microbiome research that classifies the relative abundance of gut microbial communities into different types based on clustering analysis37, 38. This approach has been extensively utilized to investigate the association between microbiome data and various phenotypes. However, the relationship between the skin microbiome and clinical outcomes remains an emerging field. Thus, we aimed to apply the enterotype algorithm to identify skin types of Korean women (cutotype), with the ultimate goal of deepening our understanding of the complex interplay between the skin microbiome and skin conditions.
As a first step, to determine the optimal number of clusters for the Korean female skin microbiome, we implemented the Optimal Clustering method based on the Calinski-Harabasz index (CH), the Silhouette Coefficient algorithm, the within-cluster sum of squares, and the Prediction Strength. Our analysis revealed that the appropriate number of clusters was two in which divided into Streptococcus- and Cutibacterium-dominant groups in all (Fig. 7). Based on the top 10 genera (Supplementary Fig. 5a and Supplementary Table 6), the Streptococcus-dominant cluster (DC1) was dominated by Streptococcus, Rothia, Corynebacterium, Neisseria, Actinomyces, and Haemophilus, while the Cutibacterium-dominant cluster (DC2) showed a remarkably high abundance of Cutibacterium. In addition, within the old group, Streptococcus was found to dominate the skin microbiome of women with all four clinical skin types, with an average relative abundance of 82.3%. In contrast, the Aging I and Young groups showed a relatively lower proportion, with 51.2% and 60.7%, respectively. This suggests that the abundance of Streptococcus in the skin microbiome may increase with age (Supplementary Fig. 5b).
In order to further classify our samples into subgroups within two larger groups, we utilized the DivCom algorithm which identify subpopulations of microorganisms and to infer the ecological and functional roles of different taxa within the microbiome (Fig. 7a)25. As a result, we were able to further divide the DC1 into two subgroups, whereas no subgroups were observed within the DC2. Among the 740 individuals satisfying the criteria for the DivCom clustering method, 726 were successfully classified into three subclusters (DC1-sub1, DC1-sub2, and DC2) using this algorithm (Fig. 7b). Specifically, 247 individuals (DC1-sub1) within this cohort showed a high relative abundance of Staphylococcus, Neisseria, Fusobacterium, Gemella, Prevotella, Granulicatella, Porphyromonas, and Leptotrichia, with Streptococcus. In contrast, 268 individuals (DC1-sub2) exhibited a high relative abundance of Rothia, Corynebacterium, Actinomyces, Lactobacillus, and Lautropia, with Streptococcus. In total, 17 genera were identified as comprising 50% of the detected microbiota across all three subclusters. Comparative analysis of the relative composition of these genera across the three subclusters revealed statistically significant differences (Kruskal-Wallis p<0.05) in the relative abundance of 15 core genera. These 15 core genera were subsequently selected for downstream analysis to highlight potential associations with skin condition (Fig. 7c). Next, we explored associations between the three groups obtained by DivCom analysis and the skin clinical measurements and compared the clinical measurements of the two groups used in the cutotype analysis (Streptococcus vs Cutibacterium) and the corresponding three groups (DC1-sub1, DC1-sub2, and DC2). The clinical metrics including age, elasticity, and oil showed differences between DC1 and DC2 groups (Fig. 7d). Further, we also observed similar results when comparing DC1-sub1 and DC2. Interestingly, differences in skin tone between DC1-sub2 and DC2, and in moisture between DC1-sub1 and DC1-sub2 were observed (Fig. 7e). We identified three microbial subclusters of skin based on 15 core genera and observed their associations with age, elasticity, sebum, skin color, and moisture.
Associations of 12 KSC Types within the Aging Group and 15 Core Genera
In the present study, a correlation analysis was performed to explore the association between the clinical outcomes of 15 core genera and the four KSC types within the aging group using a stepwise approach. To begin with KSC type-dependent microbiome significance within the Aging groups, yHL type in the Young group and oHL type in the Old group showed a statistically significant difference in the relative abundance of Cutibacterium and Streptococcus compared to the other three types. Lautropica showed a significant difference in the aLH type of the Aging I group, and Neisseria showed a significant difference in the oHH & oHL and oLH&oLL types of the Old group, with a clear correlation with skin tone and elasticity between the H and L type (Supplementary Fig. 6 and Supplementary Table 7).
The subsequent comparative analysis aimed to investigate the relative correlation of the 15 core genera for each of the 12 KSC types across all age groups. Using this analytical approach, we were able to confirm mosaic changes in microbial composition within the aging groups and the influence of core genera on each KSC type. We filtered genera with log10(-) transformed p-values exceeding ±1 based on abundance significance. Through our analysis, we observed three distinct changes in the mosaic pattern of the 15 core genera. Overall, the composition of the 15 core genera showed similar patterns between the Young and Aging I groups, but significant differences were observed between these groups and the Old group, indicating the mosaic changes (Cutibacterium, Staphylococcus, Lactobacillus, Rothia, Corynebacterium, Neisseria) (Fig. 8a). Additionally, Actinomyces, Porphymonas, and Prevotella also showed high correlation between the Young and Aging I groups, but aHH&aHL types were similar to the Young group while aLH&aLL types were similar to the Old group (Fig. 8b). These genera in the Aging I group are likely associated with skin tone based on the skin clinical measurement data. Furthermore, Streptococcus showed differences in all aging groups with the yHL and yLH type (Fig. 8c). The rest of genera are also investigated the relative correlation across the 12 KSC type (Supplementary Fig. 7).
We analyzed the microbial relative compositions that can represent each KSC type within the aging group using 15 core species, and constructed a predictive model using CatBoost boosting algorithm to distinguish the accurate KSC types within the same aging group20. As the microbial relative compositions differs according to the aging group, we divided the model into 80% training set and 20% test set for each group. We ran an algorithm to differentiate the microbial balance between one type and the other three types in each group. The overall validation AUC value ranges from about 0.8 to 0.98. The accuracy was about 0.96 on average in the Young and Aging I groups, but only about 0.87 on average in the Old group (Supplementary Table 7). The reason for the lower accuracy in the Old group is that there are minor differences that can differentiate the microbial balance in each of the four KSC types, especially in the oLL type (AUC 0.8). In the Young group, Cutibacterium showed a significant difference compared to the other groups and was identified as a representative species that can determine the Young group. Lactobacillus and Corynebacterium species were also relatively distinguishable in the yHH and yLH types. The Aging I group showed representative microbial balance in all four types, and interestingly, Prevotella in aLH type and Streptococcus and Gemella in aLL type were able to distinguish from other types. The aHL type showed a microbiological composition with a balance of Streptococcus and Staphylococcus. However, the Old group did not have enough abundance of genus composition to determine each type with a low AUC score. Simply, we observed the relative distribution of Cutibacterium and Lautropia in the oHH type, Porphyromonas and streptococcus in the oHL type, Nesseria and Granulicatella in the oHL, and Fusobacterium and Staphylococcus in the oLL (Fig. 8d). Based on the main 15 core microbial genera of Korean women, we identified microbial composition alterations specific to the Young and Aging I groups within three aging groups, and skin type-dependent microbiome changes in tone/elasticity and oil/moisture. The changes in core genera by aging group and skin type were demonstrated using the CatBoost boosting algorithm.
Functional Prediction of Differentially Enriched Skin Microbe by KSCs
We performed PICRUSt2 analysis using the KEGG pathways (n=172) related to the KEGG orthology (n=4,837) of the 15 core genera and the feature KEGG pathways of the aging group selected through LEfSe analysis26. We filtered out duplicate items and pathways with an average of less than 0.5% and focused on investigating 60 pathways in detail (Supplementary Table 8). In the Young group, several pathways were highly expressed, including “ko00052: Galactose metabolism”, “ko00500: Starch and sucrose metabolism”, “ko02010: ABC transporters”, “ko00520: Amino sugar and nucleotide sugar metabolism”, and “ko00910: Nitrogen metabolism”. Additionally, in the Young group, the L- type showed a relatively higher pattern in the Tone/elasticity criteria compared to the H- type. Specifically, the H- type in the Oil/Moisture criteria showed a higher pattern in the -L type, which was highlighted in the yLH group with a high expression of pathways. In the Aging I group, 21 pathways showed high enrichment compared to other groups. Interestingly, the functional pathways in Aging I group showed a similarity to the L- type in tone/elasticity criteria of the young group. However, the L- type showed a higher abundance than the H- type in tone/elasticity criteria in Aging I group. The -H type and -L type samples in oil/moisture criteria showed a relatively low abundacne in these KEGG pathways, which mainly belonged to carbohydrate metabolism, amino acid metabolism, and metabolism of cofactors and vitamin categories. Finally, in the Old group, interestingly, pathways related to genetic information processing showed high abundance, such as "ko03430:Mismatch repair", "ko03440:Homologous recombination", "ko03410:Base excision repair", "ko03020:RNA polymerase", "ko03420:Nucleotide excision repair", and "ko03018:RNA degradation". In addition, the KEGG pathway enrichment results specific to the Old group were very different from those of the Young and Aging I groups, with a very predominant pattern. This indicates that the predicted functional pathway patterns of the Young and Aging I groups are more similar to each other than to the Old group (Fig. 9).