Unraveling material characteristics of different pork belly cuts from raw to dish: a comparative study of texture and flavor

Abstract

Pork belly is widely favored for its alternating fat-lean layers, yet the raw material characteristics of different cuts—upper (UPB), middle (MPB), lower (LPB), and second-knife meat (SKM)—remain unclear. This study selected twice-cooked pork belly (TPB) as the model dish to systematically compare the texture, collagen, microstructure, and flavor of four cuts from raw to cooked stages. Texture analysis showed that the skin layer was mainly related to collagen, while the fat layer was influenced not only by collagen content and composition but also by the extracellular matrix network, as revealed by Masson staining. Lean meat texture was primarily determined by its proportion, resulting in 57.55 ~ 124.45% higher chewiness in SKM. Flavor and precursor analysis revealed that UPB was characterized by higher proline and C20:4n6 and was associated with higher relative abundance of sweet/citrus aldehydes. MPB generated fatty aroma linked to C18:1n9c. LPB delivered complex meaty/roasted flavors from C18:1n9c, C18:2n6c, and Phe. SKM exhibited a flavor profile dominated by 2,3-butanedione due to its high branched-chain amino acid content. Furthermore, biplot analysis elucidated the relationships between raw material characteristics and final quality attributes, providing a theoretical basis for selecting the pork belly cuts in industrial production.

Introduction

Pork belly is widely favored for its distinctive alternating layers of fat and lean meat, occupying a significant position in global pork consumption¹. However, despite its widespread consumption, the impact of different anatomical sections on the final dish quality has long been overlooked². Typically, pork belly is obtained by removing the shoulder, leg, and loin sections from the half-carcass, though the specific cuts of pork belly vary by country³. In China, pork belly is typically divided into three cuts: upper pork belly (UPB, close to the ribs), middle pork belly (MPB, central section), and lower pork belly (LPB, close to the abdominal region) (Fig. 1). Nevertheless, pork belly is often sold as a bulk product in commercial distribution without strict differentiation. This lack of standardization forces consumers and food processors to rely on experience for selection, lacking clear scientific guidance on the processing suitability of different cuts. Therefore, elucidating the impact of raw material differences in material composition on the final quality of fat-based dishes could enhance the utilization rate of pork belly resources and improve the consistency of pork belly dish quality.

Fig. 1. Schematic diagram of sample preparation.

Diagram of pork belly cuts, where UPB represents upper pork belly, MPB represents middle pork belly, LPB represents lower pork belly, SKM represents second-knife meat.

TPB, a representative Chinese dish, serves as an ideal model system for studying the industrialization of traditional Chinese fat-based dishes due to its characteristics of soft and sticky in fat, dry and fragrant in lean meat⁴. Traditional TPB uses “second-knife meat” (SKM, the second knife from the rear hindquarter) as raw material, processed through boiling, slicing, and stir-frying. Nowadays, with evolving consumer preferences, pork belly with a varied fat-to-lean ratio has increasingly become the preferred ingredient^5,6. Yet, current research on pork belly-based dishes reveals significant gaps. On one hand, existing studies predominantly focus on emerging cooking techniques while overlooking critical differences at the raw material level. Pork belly could be categorized into UPB, MPB, and LPB based on fat distribution, and the structural variations among these cuts may directly impact the final quality of processed dishes. On the other hand, representative fat-based dishes like Bazi meat and Dongpo pork exist with similar raw material selection challenges^2,7. These studies use pork belly as raw materials, but ignore the difference of fat and lean ratio among UPB, MPB, and LPB, which will directly affect the quality and flavor characteristics of the final dishes. The lack of systematic investigation into the relationship between different cuts and dish quality has resulted in inconsistent raw material standards, further complicating quality stabilization in industrial production.

To break through the limitations of traditional Chinese cooking techniques and explore emerging thermal processing technologies that offer controllable steps and digital standardization, our preliminary research found that superheated steam (SHS) technology holds potential to substitute traditional stir-fried TPB⁴. Sensory and flavor profiles showed no significant differences compared to the traditional stir-frying group. Accordingly, four representative pork cuts (UPB, MPB, LPB, SKM) were selected as raw materials, using TPB as the model dish, to investigate the relationship between raw material characteristics of different cuts and the formation of quality attributes. Firstly, stir-fried TPB prepared using digitalized cooking processes collected from professional chefs was selected as the sensory reference standard. A sensory flavor wheel for TPB was constructed to evaluate the sensory quality of TPB prepared by SHS from different cuts. Furthermore, a comprehensive analysis was conducted to identify differences in textural properties, material composition, microstructure, flavor precursors, and characteristic flavor formation. Finally, by integrating PLSR and PCA analysis, the material characteristics of pork belly from different cuts was revealed. The novelty of this study lies in establishing a raw material-to-dish linkage between anatomical cut, layer-specific structural composition, final sensory and flavor performance under the standardized thermal processing. This study provides a theoretical basis and technical support for precisely selecting raw materials according to different cuts and developing pork products with stable, superior quality in industrial production.

Results

QDA for TPB from different cuts

A primary challenge in the sensory evaluation of Chinese traditional cuisine is lacking of a fully consistent reference standard. To address this, the present study established a reference sample (TH) through digital recording and standardized preparation of the traditional stir-frying process. TH sample served as the reference standard for developing a comprehensive sensory flavor wheel for twice-cooked pork (Fig. 2A), which encompassed five key dimensions: appearance, flavor, texture, taste, and aftertaste. Hence, the sensory flavor wheel of TPB was subsequently applied to profile and differentiate the sensory characteristics of TPB prepared from different cuts (Table S1). Results revealed that 17 out of the 32 evaluated sensory attributes exhibited significant differences (P < 0.05), predominantly in flavor, texture, taste, and overall liking. In terms of flavor, significant variations were detected for meaty, roasted, creamy, and burnt-flavor (P < 0.05). LPB gained the highest scores in meaty, roasted, and burnt-flavor. Comparatively, the creamy of SKM scored significantly higher than UPB and MPB (P < 0.05). These discrepancies may be attributed to varying ratios of fat/lean, which could influence the extent of lipid oxidation/degradation and amino acid degradation during cooking, leading to the generation of distinct volatile compounds^8,9.

Fig. 2. The QDA evaluation results of TPB from different cuts.

A The sensory flavor wheel of TPB. B Correlation loadings between the sensory attributes of TPB and overall-liking rates through PLSR analysis. C Variables of VIP scores were greater than 0.5, where “A-” represents aftertaste. Note: UPB, MPB, LPB, and SKM respectively represent upper/middle/lower pork belly and second-knife meat, and all samples are cooking by SHS.

Regarding texture, the springiness of skin layer in LPB was rated significantly lower than that from other cuts (P < 0.05). Conversely, the soft and sticky score of fat layer in LPB was significantly higher (P < 0.05). These textural differences might be associated with variations in collagen content and composition among different cuts¹⁰. Overall, the overall-liking score of LPB were significantly highest (P < 0.05), as revealed by PLSR and VIP analysis (Fig. 2B, C). This preference was strongly correlated with the flexibility of skin layer and the texture of fat layer (soft and sticky, and chewiness), as well as with overall taste and flavor attributes (roasted, fatty-aroma, and burnt-flavor). Therefore, it is necessary to further analyze the differences by integrating instrumental analysis and physicochemical properties to comprehensively evaluate the material characteristics of different raw material cuts.

Screening key quality indicators from different cuts

Critical quality indicators of skin layer

Given the heterogeneous nature of pork belly, the skin, fat, and lean layers were separately evaluated to accurately reflect textural changes from the raw to the final cooked stage. The textural evolutions of skin layer from different cuts were presented in Fig. 3A-D. The skin from LPB exhibited significantly lower hardness in both raw and cooked states compared to other cuts (P < 0.05), consistent with its corresponding results of lowest flexibility score in sensory evaluation (Table S1). The raw skin from SKM showed the highest hardness and chewiness (P < 0.05), which decreased significantly during cooking stage. Notably, no significant textural differences were observed in the cooked skin between the two cooking methods for SKM, indicating that the textural properties of the skin in TPB were primarily governed by thermal processing rather than the specific cooking technique. In contrast, no significant differences in springiness were found among the raw samples from different cuts. Furthermore, neither springiness nor cohesiveness showed significant variations in cooked stage across all groups, suggesting that the principal textural distinctions stem from hardness and chewiness. The skin layer of UPB and MPB showed the highest hardness and chewiness in the cooked stage, potentially related to their proportion of insoluble collagen. The material basis of pork skin is primarily collagen, endowing the content and composition of collagen in raw skin determines its texture and processing properties to some extent^11,12. As collagen state predominantly influenced the perceived flexibility, shear force was measured to reflect this sensory attribute during mastication. The changes in shear force from raw to cooked states across different sections followed a trend similar to that of hardness (Fig. 3E). Overall, the differences in hardness, chewiness, and shear force among skin from different cuts diminished after cooking.

Fig. 3. The textural properties and collagen content changes of skin layer in different cuts during the stages from raw to cooked.

Among them, A, B, C, D, and E represent hardness, chewiness, elasticity, cohesiveness, and shear force respectively, and F represents the changes in soluble and insoluble collagen content in the skin at different stages. Note: TH is the TPB stir-fried with SKM. Data are presented as mean ± SD.

To further investigate the underlying differences in texture, variations of collagen content were shown in Fig. 3F. In raw skin, the total and insoluble collagen contents in LPB were significantly lower than other groups (P < 0.05), explaining its significantly lower hardness, chewiness, and shear force. The skin from SKM, with a higher proportion of insoluble collagen (65% of total), correspondingly exhibited greater hardness and chewiness. It was noteworthy that heat treatment reduced the differences of collagen content in the raw materials, particularly after the pre-cooked stage. This stage promoted extensive conversion of soluble collagen in the skin, leading to significant decreases in hardness and shear force compared to the raw stage. After further stir-frying into TPB, the insoluble collagen content in LPB showed no significant difference compared to the MPB, SKM, or TH, and was even higher than that of UPB (P < 0.05). However, the hardness of LPB was significantly lower than all other cuts. This indicated that the textural properties of skin layer in TPB were influenced not only by collagen content and proportion but also by the cross-linking degree of collagen during thermal processing¹³.

Critical quality indicators of fat layer

The textural properties, shear force, and collagen content of the fat layers from different cuts were summarized in Fig. 4. In the raw state, the fat layer from LPB exhibited significantly lower hardness and chewiness than all other cuts (Fig. 4A, B), yet its shear force was significantly higher than that of SKM (P < 0.05, Fig. 4E). These divergent trends may be related to the characteristics of collagen. Subsequent analysis revealed that the LPB contained the highest content of both insoluble and soluble collagen in fat layer (Fig. 4F). Notably, during sampling of the subcutaneous adipose layer, a distinct “granular” appearance was visually observed in MPB and LPB, with LPB being the most pronounced (Fig. 4). This macroscopic morphology likely reflects the constraining effect of the extracellular matrix (ECM) network on the adipose tissue, suggesting that the textural properties of fat are governed not only by collagen content but also by the distribution and structural organization of the ECM, which warrants further microstructural characterization¹⁴. In contrast, no granularity was observed in the fat layer of UPB, which displayed the highest hardness, chewiness, and shear force. Although its total collagen content was lower than that of MPB and LPB, it had the highest proportion of insoluble collagen (80.17%). Meanwhile, SKM showed significantly higher hardness and chewiness than LPB, but significantly lower shear force, alongside the lowest total collagen content. This reinforced the notion that shear force behavior in adipose tissue might be associated with the structural network of the ECM¹⁰.

Fig. 4. The textural properties and collagen content changes of fat layers in different cuts during the process from raw to cooked.

Among them, A, B, C, D, and E represent hardness, chewiness, elasticity, cohesiveness, and shear force respectively, and F represents the changes in soluble and insoluble collagen content in the fat layers at different stages. Data are presented as mean ± SD.

After pre-cooking and further stir-frying, no significant differences were observed in the hardness, springiness, or cohesiveness of the cooked fat layer among different cuts, regardless of the cooking method (SHS or traditional stir-frying). This indicated that the standard processing protocol for TPB effectively homogenized the hardness differences among the fat layers of different cuts. For SKM, the two cooking methods yielded no significant differences in the hardness, springiness, or cohesiveness of its fat layer at either the pre-cooked or cooked stage. The only exception was that SHS-treated SKM showed significantly lower chewiness than TH samples (P < 0.05). This may be attributed to the more stable and efficient heat transfer of SHS, which could promote a greater conversion of soluble collagen to gelatin, whereas the open stir-frying environment might accelerate moisture loss and tissue contraction⁴. Besides, the chewiness of fat layer in LPB did not differ significantly from that of TH samples, yet its shear force remained significantly higher. This aligned with its marked increase in insoluble collagen content after cooking (Fig. 4F), which was significantly higher than all other groups (P < 0.05) and consistent with its highest sensory score for chewiness (Table S1). However, the shear force value for LPB was still significantly lower than that of UPB, implying that the perception of “soft and sticky” in fat was determined not merely by collagen proportion but potentially by the specific structural state of the ECM network.

Critical quality indicators of lean layer

The changes in the textural properties, shear force, and collagen content of the lean layers in TPB prepared from different cuts were presented in Fig. 5. From the raw to the cooked state, the hardness, chewiness, and cohesiveness of the lean meat showed an increasing trend across all cuts (Fig. 5A, B, D). Notably, the lean meat of TH samples exhibited significantly higher values for these parameters in the cooked stage compared to all other groups (P < 0.05). This was possibly due to the higher proportion of lean meat in SKM, which may undergo more intense thermal contraction during the twice-cooking process, thereby resulting in higher hardness, chewiness, and cohesiveness. Consequently, among the four cuts subjected to the same cooking method, SKM also exhibited higher hardness, chewiness, and cohesiveness.

Fig. 5. The textural properties and collagen content changes of lean layers in different cuts during the process from raw to cooked.

Among them, A, B, C, D, and E represent hardness, chewiness, elasticity, cohesiveness, and shear force respectively, and F represents the changes in soluble and insoluble collagen content in the lean layers at different stages. Data are presented as mean ± SD.

Furthermore, a distinct pattern was observed for shear force (Fig. 5E). The raw lean layer from LPB had a significantly higher shear force than other cuts, aligning with its significantly highest content of both insoluble and total collagen (Fig. 5F). However, the shear force of LPB significantly lower than that of the SKM groups (whether prepared by SHS or traditional stir-frying), no matter in the pre-cooked or cooked stages. At the cooked stage, the variations of shear force among different cuts corresponded with the trends observed in their textural properties, even though the total collagen content in cooked SKM and TH samples was significantly lower than in LPB (P < 0.05). These findings suggested that the composition and content of collagen were the critical factors influencing its textural properties in raw meat. In contrast, the textural differences in the cooked lean layers appeared to be primarily determined by the intrinsic proportion of lean meat within each cut.

Microstructure differentiations of fat and meat layers

As previously indicated, the textural and shear force properties of fat tissue are related to the distribution and structural state of its ECM network, whose primary components are collagen and elastin fibers¹⁵. Herein, fat tissues were examined by classic Masson staining, where collagen fibers appear blue and non-collagenous ECM components (including elastin fibers) appear red¹⁶. In raw fat tissue from different cuts (Fig. S1), MPB and LPB exhibited more blue-stained areas than others, whereas SKM showed the least, consistent with the results of collagen content (Fig. 4F). These inherent compositional differences in the fat layers further influenced the structural changes upon cooking, as shown in Fig. 6A. The fat tissues showed a marked reduction in blue areas, conversely dominating by the red areas, which suggested that varying degrees of collagen fiber was disrupted during cooking, leading to the exposure of non-collagenous ECM components. This indicated that the soft and sticky mouthfeel of fat tissue was associated with collagen composition and ECM organization, explaining the observed discrepancies between textural changes and collagen content variations in the fat layer.

Fig. 6. Masson staining results of fat and lean tissues in TPB.

A Masson staining results of fat tissues in TPB. B The collagen distribution within muscle fibers of lean meat in both raw and cooked states from different cuts, and C the corresponding muscle fiber diameter distributions. Note: The black arrows indicate the rupture of the lobular septa, while the green arrows indicate swelling and rupture of the basement membrane.

As for the TH samples, traditional stir-frying in an open environment accelerated the rupture of the ECM network within the fat tissue, causing the basement membrane of fat cells to exhibit shrinkage, rupture, and swelling. Compared to the same cut but cooked by SHS (SKM), the lobular septa and basement membrane in TH sample underwent more severe disruption (black arrow and green arrow), correlating with its significantly lowest shear force value (Fig. 4E). For the four belly cuts cooked via SHS, the primary structural change was adipocyte rupture. Among them, UPB exhibited more extensive distribution of lobular septa, which could better maintain the structural integrity of the fat tissue, resulting in a significantly higher shear force. Furthermore, observations under 10× magnification (Fig. 6A) revealed that the ECM network in UPB and MPB was predominantly maintained by the non-collagenous ECM components. In contrast, the structure of LPB and SKM was mainly co-supported by the collagen fibers and non-collagenous ECM components, with LPB exhibiting a relatively more stable basement membrane structure, which corresponded to its soft and sticky mouthfeel.

Besides, the distribution of collagen in the lean layer was shown in Fig. 6B, primarily distributed within the intermuscular spaces. In raw lean meat, LPB showed abundant collagen filling within the muscle fiber gaps, aligning with its highest total and insoluble collagen content and shear force (Fig. 5E, F). According to the textural results, the textural differences in the cooked lean layer were mainly dictated by the proportion and contraction of lean meat, thus muscle fiber diameter was the key factor influencing its texture and shear force characteristics. Therefore, the distribution of muscle fiber diameters across lean meat layers from different cuts was compared (Fig. 6C). In the raw materials, the diameters of myofibers in UPB and SKM were larger than those in MPB, consistent with the significantly lowest shear force of MPB. After cooking, the muscle fibers in the lean meat from each cut exhibited noticeable contraction, with the interstitial fluid in the muscle fiber gaps being “squeezed” out, thereby causing the water content within the muscles decreased¹⁷. This was particularly pronounced in SKM, resulting in its lean meat exhibiting higher chewiness and a “dry” texture (Table S1). For samples from the same cut prepared by different methods, the open-stir-frying environment accelerated muscle fiber contraction, resulting in the smallest inter-fiber gaps. Consequently, its textural properties (hardness, chewiness) and shear force were significantly highest.

Differences in flavor characteristics among different cuts

Precursors analysis

Texture and flavor are critical determinants of consumer acceptance in meat and meat products¹⁸. Thus, the differences in composition and content of major flavor precursors (amino acids and fatty acids) among different cuts were investigated (Fig. 7). Amino acids serve as crucial precursors for promoting flavor and taste development in meat products. Variations in amino acid content among different cuts at the raw material stage were shown in Fig. 7A. The amino acid composition primarily clustered into two distinct groups. MPB, LPB, and SKM exhibited similar amino acid compositions, while UPB was excluded. This was primarily due to significantly higher contents of Pro, Gly, Ala, and Arg in UPB compared to other cuts. Further VIP analysis (VIP > 1) identified Arg and Pro as the key amino acids distinguishing UPB from MPB, LPB, and SKM (Fig. 7B). Pro was found to be one of the key flavor precursors, contributing to the roasted aroma of TPB through its participation in the Maillard reaction^19,20. Additionally, VIP analysis identified Phe and Tyr as characteristic amino acids distinguishing the LPB from other cuts. Phe, an aromatic amino acid, serves as a key precursor for aromatic compounds in the Maillard reaction, potentially enhancing the savory and burnt flavor of TPB. In contrast, SKM contained the highest levels of branched-chain amino acids (Val, Leu, and Ile). These would participate in Strecker degradation during thermal processing, potentially promoting increased aldehyde formation following thermal processing of TPB. Therefore, variations in these key amino acids within the raw materials may endow the TPB prepared from different cuts with rich and distinct flavor characteristics, through the cumulative and synergistic effects of pathways such as the Maillard reaction and Strecker degradation during thermal processing.

Fig. 7. Analysis of flavor precursors among different cuts.

A Amino acid heatmap analysis and VIP analysis B of raw materials from different cuts. Fatty acid heatmap analysis C and VIP analysis D of raw materials from different cuts.

Furthermore, fatty acids play a crucial role as important liposoluble flavor precursors, with their content and composition making vital contributions to the formation of the final TPB characteristic flavor²¹. As shown in Fig. 7C, the fatty acid composition of raw meat from different cuts exhibited distinct clustering patterns. Saturated fatty acids (including C10:0, C12:0, C14:0, C16:0, and C18:0) clustered together, with MPB showing markedly higher relative abundances. In contrast, n-6 and n-9 polyunsaturated fatty acids (PUFAs) and monounsaturated fatty acids (MUFAs) formed another cluster. LPB had a significantly higher content of oleic acid (C18:1n9c), while UPB contained the highest levels of linoleic acid (C18:2n6c) and arachidonic acid (C20:4n6). VIP analysis based on partial least squares discriminant analysis (PLS-DA) (Fig. 7D) identified C18:2n6c as the discriminatory fatty acid (highest VIP score) distinguishing UPB from other cuts. Notably, C18:2n6c was the typical n-6 PUFA that could generate aldehydes via oxidation pathways (e.g., β-oxidation) during TPB cooking, serving as a key precursor for meaty aroma. The high content of C18:1n9c in LPB contributed to the formation of 2-alkenals during processing²². These fatty acid components provide a crucial lipid precursor foundation for the unique flavor development of TPB, which could be converted into various volatile flavor compounds through the synergistic actions of lipid oxidation, degradation, and the Maillard reaction during TPB cooking.

Composition of volatile flavor compounds from different cuts

The flavor contribution of TPB from different cuts significantly influences consumer preference. As shown in Table S2, HS-SPME-GC-MS detected a total of 46 volatile compounds, including nine categories: aldehydes, alcohols, esters, ketones, furans, alkanes, pyrazines, and sulfur-containing compounds. Variations in the types and contents of these nine categories were observed among TPB samples from different cuts (Fig. 8A). Fifteen compounds were common to all samples (Fig. 8B), while each of the four SHS-prepared TPB samples contained unique flavor compounds in varying quantities, partially overlapping with those in the TH sample. Heatmap analysis revealed that samples primarily clustered into two groups based on the flavor contents, LPB and MPB formed one cluster, while SKM, UPB, and TH formed another. However, further PLS-DA analysis of all volatiles (Fig. 8C) showed partial overlap between the flavor profiles of SKM and LPB, with LPB being closer to TH, whereas UPB partially overlapped with MPB. The model exhibits strong explanatory power for dependent variables (R²Y = 0.966), with a high predictive ability (Q² = 0.867).

Fig. 8. Changes of volatile flavor compounds in TPB from different cuts.

A Heatmap analysis, B Venn diagram and C PLS-DA analysis of TPB from different cuts. D Sanky plot of flavor contributors with ROAV > 1, E PLSR analysis between flavor contributors with ROAV > 1 and sensory flavor attributes. F Biplot analysis of variants for key volatiles and flavor precursors from different cuts.

Accordingly, to clarify the contribution of these volatiles to the overall flavor profiles of TPB, ROAV were calculated to screen the potential aroma contributors. Eighteen compounds with ROAV > 1 were identified (Table S3). The contribution and corresponding flavor descriptions of volatiles with ROAV > 1 across different samples were visualized by Sankey diagram (Fig. 8D). Notably, the TH sample primarily exhibited fat and creamy flavors, with 2,3-butanedione being the most abundant. This compound was primarily derived from the Maillard reaction and thermal degradation of lipids, particularly PUFAs (such as linoleic acid) undergoing β-oxidation and thermal degradation to produce the key precursor—acetyl-CoA²³. The rapid heating rate and open cooking environment of pan-frying accelerated lipid thermal degradation⁴, imparting creamy flavor to TH sample. In contrast, TPB prepared via SHS from different cuts exhibited fatty, citrus, and sweet aromas. LPB possessed significantly higher contents of γ-butyrolactone and 2-methyl-3-octanone compared to other groups, contributing to a more intense sweety and fruity aroma, which was mainly derived from lipid oxidation/degradation, esterification, and amino acid degradation²⁴.

Considering that volatile compounds with ROAV > 1 were potential contributors to the overall flavor perception of TPB, which can be reflected in sensory flavor results. Thus, PLSR analysis was conducted where these potential aroma contributors as independent variables and the QDA flavor scores as dependent variables (Fig. 8E). The results revealed that overall flavor scores and attributes related to fatty and roasted flavors were distributed on the right side of the loading plot, indicating that fatty and roasted flavors significantly influence the overall flavor score of TPB. Notably, heptanal, decanal, and octanal showed positive correlations with the overall flavor, fatty, and roasted scores. These three flavor compounds were not detected in TH samples (Table S2) and characterized the unique flavors of SHS-prepared TPB. Heptanal, decanal, and octanal primarily contributed to fatty, fruity, and sweet aroma^22,25. Among the four groups, only SKM showed a significantly lower decanal content compared to the others (P < 0.05), while differences in the others were not significant. This suggested that the three compounds form a synergistic effect through “fat aroma superposition, flavor complementarity, and threshold synergy”, collectively shaping the signature flavor of SHS-prepared TPB.

Given that amino acids and fatty acids are pivotal precursors directly determining the flavor profile of TPB, biplot analysis based on PCA analysis was employed to elucidate the relationships between the potential aroma contributors (ROAV > 1) and these precursor substances (Fig. 8F). In the biplot, blue points represented the TPB samples prepared from different cuts, while red points denoted the variables (key volatiles, amino acids, and fatty acid precursors). Smaller angles between variables indicated stronger correlations, and closer distances between samples and variables signified tighter associations. Results showed that UPB primarily resided in the fourth quadrant, relating with sweet amino acids (Ser, Gly, and Ala) and C20:4n6, where decanal was the main characteristic flavor compound, contributing to the sweety and citrusy aroma. Comparatively, SHS-prepared SKM exhibited similar flavor profiles with TH samples since their raw material were coming from the same cut, primarily distributed in the first quadrant. Their flavor profiles were associated with branched-chain amino acids (Val, Leu, and Ile), may contribute to the formation of the characteristic flavor compound (2,3-butanedione) through specific metabolism²⁶. Besides, LPB was primarily distributed in the third quadrant, partially overlapping with the MPB, which was mainly associated with oleic acid. Oleic acid underwent β-oxidation and thermal degradation during heat processing²⁷, may further promoting the formation of γ-butyrolactone, 2-methyl-3-octanone, and E-2-octenal. Distinctively, MPB was positioned close to C18:1n9c and (E)-2-decenal. This association was consistent with the thermal oxidation pathway of oleic acid, which generated (E)-2-decenal and other unsaturated aldehydes²⁸, imparting fatty aroma to MPB. LPB was also linked to PUFAs including C18:2n6c and C20:3n6, may facilitating the production of lipid-derived flavor compounds such as octanal, heptanal, and E-2-nonenal²⁹. Accordingly, the differing fat-lean composition at the raw material stage across various cuts influenced the variations of flavor precursor contents, thereby contributing distinct characteristic flavors during thermal processing.

Processing characteristics of different pork belly cuts

Based on the heterogeneous nature of pork belly, this study systematically compared the texture, shear force, collagen composition, and microstructure of the three layers (skin, adipose, lean) from four cuts (UPB, MPB, LPB, SKM) during processing into TPB. In the skin layer, the raw LPB exhibited significantly lower total/insoluble collagen content compared to other cuts, resulting in the lowest hardness, chewiness and shear force. Although thermal processing promoted collagen gelatinization and reduced textural differences among the pork skins from different cuts, LPB retained the lowest hardness and optimal tenderness (QDA results) in the final product due to its inherent raw material characteristics. In the fat layer, sensory attributes among different cuts were primarily influenced by collagen content and the ECM constituents. The raw LPB showed the highest insoluble collagen content, whose microstructure revealed collagen and non-collagenous ECM components jointly supporting the structure of TPB, conferring its highest scores in soft and sticky. Whereas, the TPB cooked from UPB and MPB were dominated by the network consisting of the non-collagenous ECM components, exhibiting significantly lower chewiness and “soft and sticky” scores. Within the lean layer, although LPB exhibited the highest shear force at the raw stage due to its high collagen content, the textural differences of cooked lean meat were primarily influenced by the proportion of lean meat. The lean layer of TPB from SKM showed significantly highest hardness and chewiness due to its relatively highest lean layer proportion and intense muscle fiber contraction. Comprehensively, the structural characteristics of pork skin and lean meat significantly contribute to the overall texture. However, the composition of the fat layer, the collagen network structure, and its morphological evolution during thermal processing are the critical factors determining the quality characteristics of fatty dishes like TPB.

Furthermore, the cut-specific flavor profiles originated from fundamental differences in precursor composition (amino acids and fatty acids). The UPB was rich in Pro and C20:4n6, associated with the flavor of TPB dominated by aldehydes such as decanal³⁰, imparting sweet and citrus aromas. The relatively high content of (E)-2-decenal in MPB was related to C18:1n9c. LPB, abundant in C18:2n6c and C20:3n6, presented the most complex and intense flavor profile, characterized by meaty, roasted, and burnt aromas. Besides, SKM exhibited a flavor profile dominated by 2,3-butanedione, produced through the specific metabolism of branched-chain amino acids due to its high content of these compounds²³.

Overall, this study uses Rongchang pigs as the raw material and superheated steam as the heating method, establishes the intrinsic link between raw material properties of different pork belly cuts and the final quality of TPB, providing a preliminary scientific basis for rational raw material selection in industrial production. Future studies can extend this research paradigm to investigate different pig breeds and various heating methods (e.g., traditional stir-frying, slow-braising, roasting). Furthermore, to better capture consumer perception of dish quality, sensory evaluation could be expanded to include consumer preference surveys such as check-all-that-apply (CATA) questions, which would provide more market-oriented insights. Ultimately, these efforts aim to clarify the processing adaptability of different pork belly cuts and establish standardized raw material grading criteria for industrial applications.

Methods

Sample collection

11 Rongchang pigs (castrated males, 8 months old, live body weight 90–110 kg) were provided by the Chongqing Academy of Animal Science. The experimental pig farm and slaughterhouse of the Chongqing Academy of Animal Sciences are certified with HACCP system certifications and ISO 9001 quality management system certifications. All pigs were electrically stunned (220 V and 1.3 A for 3 s) and exsanguinated following standard commercial procedures (GB/T 17236-2019). Carcasses were chilled at 4°C for 24 h post-mortem; the ultimate pH (pH_24h) was 5.53 ± 0.06. All experimental procedures involving animals were approved by the Animal Ethics Committee of Chongqing Academy of Animal Science (IACUC approval number: XKY‑20221208). After slaughter, the carcasses were immediately segmented. The second cut from the hind leg was collected as the “second-knife meat” (labeled as SKM). Bellies were separated into three layers according to anatomical position: upper pork belly (close to the ribs, labeled as UPB), middle pork belly (central section, labeled as MPB), and lower pork belly (close to the abdominal region, labeled as LPB). The schematic diagram of sampling from different cuts was shown in Fig. 1. All samples were transported under cold chain conditions at 4°C and delivered to the laboratory within 2 h post-slaughter to ensure freshness. Samples were subjected to postmortem aging at 4°C for 24 h and then trimmed into blocks of approximately 12 cm×6 cm×4 cm, vacuum-packaged, and frozen at −40°C for further analysis.

Samples treated by SHS

According to our previous study⁴, a standardized two-step SHS procedure was established for sample preparation. Briefly, meat blocks (12 cm × 6 cm × 4 cm, weighed 250.00 ± 3.24 g) were treated by the steam oven (VD7000CNB, Toshiba Electric Co. Ltd., Thailand) at SHS mode, initially cooked at 120°C for 25 min (pre-cooked stage). After that, the slices of each cut (weighed 100.00 ± 2.65 g) were separately again heated by SHS at 180°C for 5 min (cooked stage), finally obtaining twice-cooked pork belly (TPB). The endpoint of the pre-cooked stage and cooked stage was defined by the center temperature of the meat reaching 73.59 ± 1.32°C and 105.36 ± 3.88°C, respectively. The key information regarding the specific grouping and sample preparation was shown in Fig. 1.

For the traditional stir-fried group, the entire traditional stir-frying procedure for TPB was digitally recorded and quantified by professional chefs to establish a standardized traditional preparation protocol⁴. The specific sample preparation was also depicted in Fig.1. For the pre-cooking stage, SKM blocks (weighed 250.00 ± 1.87 g) were fully immersed in boiling water at a meat-to-water ratio of 1:10 (w/w), and cooked by the induction cooker (C22-WT2218, Midea, Guangzhou, China) at a constant power of 800 W for 30 min, with the final center temperature of the meat blocks at 72.76 ± 1.67 °C. Then sliced into 3.5 mm thick to obtain the pre-cooked traditional heated sample (PTH). For the cooking (stir-frying) stage, 100.00 ± 1.36 g of slices was added to a pan preheated with commercial canola oil (oil-to-meat ratio of 1.5:10, w/w) at 120 °C, and stir-fried for a total of 3 min with a programmed power setting: 1600 W for 50 s, followed by 800 W for the remaining 130 s (cooking completion point was 142.85 ± 4.58 °C). The final product obtained from this traditional protocol was defined as traditional heated TPB (TH).

Consequently, samples were systematically collected at three key processing stages in accordance with the preparation protocol of TPB: raw, pre-cooked, and cooked. Given the heterogeneous nature of pork belly, samples were separated into three distinct tissue layers at each stage: skin layer, fat layer, and lean layer, for independent analysis. Meanwhile, intact slices of each sample were collected to evaluate the overall flavor attributes.

Quantitative descriptive analysis (QDA) of sample sensory

The sensory evaluation protocol was approved by the Institutional Review Board of the College of Food Science, Southwest University (IRB NO. HF20251201). QDA was conducted following the methodology of Zhang et al.³¹ with some modifications. The sensory flavor wheel for TPB was established and applied through a four-step process. Initially, a terminology generation session was conducted with five expert chefs specializing in Sichuan cuisines (each with over 10 years of professional experience). Using the TH as the reference standard, potential descriptive terms were discussed and defined for four key components: skin layer, fat layer, lean layer, and the intact slices. Definitions and intensities of descriptive terms were explicitly established. Subsequently, these preliminary terms were refined through testing with variant samples by our sensory panel (6 males and 6 females, aged 22–30 years, experienced in meat product evaluation). Terms were subjected to frequency analysis and consensus evaluation to retain only the most salient and consistently perceived descriptors, resulting in the final sensory flavor wheel (Fig. 2A). The panel then underwent extensive training and calibration using TH samples to ensure scoring consistency and reproducibility. Finally, four TPB samples cooked by SHS were evaluated in comparison to the constant TH reference served in each session. The evaluation conditions followed the method of Tang et al.⁴.

Texture profile analysis (TPA) and shear force analysis

Texture analysis was performed according to the method of Da et al.³². with slight modifications. Samples from the skin, fat, and lean layers of different cuts were uniformly cut into 2 cm × 1 cm × 0.35 cm. Prior to testing, all samples were equilibrated at 40 °C for 10 min in a temperature-controlled incubator to simulate serving temperature. TPA was conducted using a TA-36 probe. The pre-test, test, and post-test speeds were 1 mm/s, with a compression ratio of 50%. Shear force was measured using an A-MORS probe, with a compression ratio of 80%.

Determination of collagen content

The contents of soluble and insoluble collagen were determined according to the method described by Wang et al.³³. Briefly, 5 g samples of the skin, adipose, or lean layers from different cuts at the raw, pre-cooked, and cooked stages were weighed. Each sample was swollen in Ringer’s solutions for 1 h, followed by incubation in a water bath at 77 °C for 1 h. After centrifugation, the supernatant (containing soluble collagen) and the precipitate (containing insoluble collagen) were collected separately for nitration treatment. The hydroxyproline content in each fraction was quantified through hydroxyproline assay kit (G1205W, Suzhou Grace Biotechnology Co.,Ltd). Total collagen content was subsequently calculated by multiplying the measured hydroxyproline content by a conversion factor of 7.25.

Masson staining

The fat portions and lean portions were cut into 0.5 × 0.5 × 0.35 cm and fixed with 4% paraformaldehyde for 24 h, followed by dehydration and embedding in paraffin blocks. Subsequently, the embedded samples were sectioned and stained by Masson trichrome staining kit (Servicebio, Wuhan, China) according to the manufacturer’s protocol. This staining method differentially colors collagen fibers blue. The stained sections were visualized and imaged using an optical microscope (Eclipse C1, Nikon, Tokyo, Japan).

Amino acid analysis

Raw materials from the four cuts (UPB, MPB, LPB, and SKM) were respectively sliced and then homogenized for amino acid and fatty acid composition analysis. Briefly, 1 g of the homogenized sample was accurately weighed, mixed with 15 mL of 6 M HCl, and hydrolyzed at 110 °C for 22 h. The procedure was conducted in accordance with the Chinese National Standard GB 5009.124-2016. The concentrations of amino acids (except cysteine) were analyzed by the amino acid analyzer (Hitachi L-8900, Tokyo, Japan) equipped with a cation-exchange column (4.6 mm × 60 mm, 3 μm particle size).

Fatty acid analysis

10 g of each sample was subjected to crude fat extraction followed by methyl esterification according to the method of Zhao et al.³⁴. Methyl octanoate (5 mg/mL) was added as the internal standard solution before methyl esterification. Fatty acid methyl esters were separated and quantified by the gas chromatography-mass spectrometry system (GC-MS, 8890-5977B, Agilent Technologies, USA) fitted with an DB-FFAP capillary column (30 m × 0.32 mm × 0.25 μm). Instrumental parameters were consistent with our previously validated method⁴, and briefly defined as follows: injection port temperature set at 250 °C, high-purity helium as carrier gas at a constant flow rate of 3 mL/min, split ratio 50:1. The oven temperature program was: initial temperature 70 °C held for 4 min, ramped to 200 °C at 10 °C/min and held for 5 min, then ramped to 230 °C at 5 °C/min and held for 8 min. Data acquisition and processing were performed using Agilent MassHunter Unknowns Analysis software (v10.1), with fatty acid content calculated automatically based on the internal standard.

Composition analysis of volatile flavor compounds

The volatile compounds were analyzed using headspace solid-phase microextraction coupled with gas chromatography-mass spectrometry (HS-SPME-GC-MS). Specifically, five slices of TPB were minced and homogenized. Then, 2.0 g of the homogenate was weighed into a 20 mL headspace vial, then adding the internal standard solution (1 μg of 2,4,6-trimethylpyridine) and immediately sealed. The vial was equilibrated in a water bath at 60 °C for 10 min. Volatile compounds were subsequently extracted by exposing a DVB/CAR/PDMS-coated SPME fiber to the vial headspace (60 °C, 30 min). The extraction conditions were pre-optimized in our previous study⁴ to ensure equilibrium. After extraction, the fiber was inserted into the GC injection port and thermally desorbed for 5 min at 250 °C.

The GC-MS instrument conditions, data acquisition, and processing procedures were identical to those described in our previous study⁴. Volatile compounds were tentatively identified by electron ionization (EI) full-scan spectrum matching against the NIST 17.0 L library using Agilent MassHunter Unknowns Analysis software (v10.1), with minimum match score of 80. The retention index (RI) of all flavor compounds was calculated by injecting a C₇-C₄₀ n-alkane standard series under the same chromatographic conditions. The quantification of volatiles was determined by the peak area of each compound with that of the internal standard³⁵. For further evaluation, the relative odor activity value (ROAV) of each identified compound was calculated. Odor thresholds were obtained from the VCF online database (https://www.vcf-online.nl/OFTVCompoundSearch.cfm). When multiple threshold values were available, values determined in the same reference matrix (water) were selected to maintain consistency in ROAV calculation. The volatile compound with the highest odor activity value in each sample was assigned an ROAV of 100, and the ROAVs of the remaining compounds were calculated relative to this compound.

Statistical analysis

Unless otherwise specified, 3 pigs were randomly selected from the 11 Rongchang pigs and used as independent biological replicates (n = 3) for instrumental and chemical analyses. For each biological replicate, all four cuts (UPB, MPB, LPB, SKM) were collected and processed through all three stages (raw, pre-cooked, cooked). Each analytical measurement was performed in triplicate technical replicates on each biological sample. Results are expressed as mean ± standard deviation (SD) of the biological replicates (n = 3). Prior to statistical analysis, Shapiro-Wilk test and Levene’s test were performed to verify the assumptions of normality and homogeneity of variance, respectively. Significant differences among groups were determined by one-way analysis of variance (ANOVA) followed by Tukey’s honestly significant difference (HSD) post hoc test, with a significance level set at P < 0.05. GraphPad Prism (v10.1.2) and Origin (v2024) software were used for generating graphs. Statistical analysis was conducted by SPSS software (v25.0). Fatty acid and volatile flavor compound data were processed by Agilent MassHunter Unknowns Analysis software (v10.1). Sankey plot was constructed using Chiplot (https://www.chiplot.online/). Partial least squares regression (PLSR), variable importance in projection (VIP) and Principal Component Analysis (PCA) were performed using XLSTAT (v2019.2.2). Heatmap analysis was generated using the Bioladder platform (version 2.0; https://www.bioladder.cn/).

Author contributions

Mi Tang: Methodology, Data curation, Investigation, Formal analysis, Writing- Original draft preparation. Yong Yu: Data curation, Investigation, Formal analysis. Weijie Hu: Methodology, Software. Hankun Zhu, Liang Ma: Validation, Data curation. Kangting Sun: Visualization, Software. Xin Feng: Validation. Jinyong Wang, Xiaoqian Wu: Methodology. Yuhao Zhang: Conceptualization, Resources, Writing - Reviewing and Editing, Project administration, Funding acquisition.

Data Availability

The datasets generated and/or analyzed during the current study are not publicly available due to an unsubmitted graduation thesis but are available from the corresponding author on reasonable request.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41538-026-00967-5.