Combined decoction selectively modifies chemical composition of traditional Asian medicine: a systematic analysis on Palmijihwang-tang using theoretical additive model

Abstract

Background

Combined decoction (CD)—boiling multiple herbs together—is a fundamental preparation method in Traditional Asian Medicine. While biological synergies are well documented, chemical interactions during the decoction process itself remain largely unexplored. Understanding these preparation-dependent chemical modifications is crucial for standardizing and modernizing traditional herbal medicines.

Methods

We developed a systematic analytical framework combining theoretical additive modeling with structural similarity analysis to investigate structure-dependent extraction in CD. Using Palmijihwang-tang as a proof-of-concept model, we used liquid chromatography–quadrupole time-of-flight mass spectrometry (LC-QTOF-MS) to compare chemical profiles of CD with predicted profiles from individual decoctions (ID). We then predicted drug-likeness and bioactivity in silico and validated our findings with cell-based assays.

Results

CD systematically modified compound extraction based on structural features, selectively enhancing certain structural groups while suppressing others. These distinct extraction patterns correlated with the compounds' physicochemical properties, structural complexity, and predicted therapeutic properties—suggesting an empirically optimized process. Cell-based assays confirmed these chemical alterations lead to measurable biological differences between CD and ID preparations.

Conclusions

Our findings suggest that CD selectively modifies the chemical composition of herbal formulations through structure-dependent extraction, providing a preliminary chemical basis for better understanding traditional decoction practices. This highlights the critical importance of considering preparation methodology in herbal medicine standardization. Our analytical framework opens new avenues for investigating these CD effects across a wider range of traditional prescriptions. We believe this approach may ultimately contribute to the improved understanding and standardization of this fundamental method in Traditional Asian Medicine.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12906-025-05153-w.

Introduction

Combined decoction (CD), where multiple herbs are boiled together, represents a fundamental preparation method in Traditional Asian Medicine (TAM). TAM's therapeutic approach relies heavily on strategic herb combinations, employing multiple herbs simultaneously to achieve desired clinical outcomes [1]. Modern pharmacological studies have validated various biological effects of these herb combinations through both synergistic and antagonistic mechanisms [2, 3]. However, most research has focused on the final therapeutic effects rather than the preparation process itself. The decoction method significantly influences a formulation's final chemical composition. For instance, whether herbs are boiled together (CD) or separately (individual decoction, ID) can lead to different chemical profiles. Despite this importance, the underlying chemical mechanisms for these effects remain largely unexplored.

Previous research on herb combinations in traditional Asian Medicine (TAM) has primarily focused on biological synergies, particularly through approaches examining herb-target interactions in protein–protein interaction networks [2, 3]. While these studies have illuminated the biological basis of herbal synergy, they largely overlook a critical aspect: the chemical interactions occurring during the preparation process itself. Most chemical analyses have been limited to studying individual herbs or simple herb pairs [4–6], examining basic co-extraction effects or precipitate formation. Recent studies have begun to reveal that co-decoction is more than a simple physical mixing, but rather a process involving complex chemical interactions that can alter the final chemical profile [7]. The use of modern analytical techniques, such as metabolomics and high-resolution mass spectrometry, has been instrumental in characterizing these changes in traditional formulas [8, 9]. These efforts highlight the importance of understanding how the decoction process itself influences the therapeutic and safety profiles of a formula at the molecular level.

This analytical gap poses significant challenges for understanding and standardizing TAM preparations. Current quality control approaches typically focus on analyzing individual herbs [10], potentially missing crucial preparation-dependent changes in chemical composition [11]. Without a comprehensive understanding of how CD affects compound extraction, it becomes difficult to ensure that modern pharmaceutical adaptations maintain the intended therapeutic properties of traditional formulations.

We hypothesized that CD induces structure-dependent extraction patterns, selectively enhancing certain compound classes while suppressing others based on their chemical properties. To test this hypothesis, we developed a systematic analytical framework combining theoretical additive modeling with molecular networking. This approach allows us to differentiate CD-induced changes from simple mixing effects and examine how structural features determine extraction patterns.

To demonstrate this concept, we investigated Palmijihwang-tang (PM), a representative eight-herb formula widely used in contemporary TAM practice. PM comprises eight herbs: Rehmanniae Radix Preparata (RRP, root of Rehmannia glutinosa Liboschitz ex Steudel), Corni Fructus (CF, fruit of Cornus officinalis Siebold et Zuccarini), Dioscoreae Rhizoma (DR, root of Dioscorea batatas Decaisne), Poria Sclerotium (PS, sclerotium of Poria cocos Wolf), Alismatis Rhizoma (AR, tuber root of Alisma orientale Juzepzuk), Moutan Radicis Cortex (MRC, root bark of Paeonia suffruticosa Andrews), Cinnamomi Cortex (CC, stem bark of Cinnamomum cassia Presl), and Aconiti Lateralis Radix Preparata (ALRP, prepared lateral root of Aconitum carmichaeli Debeaux) [12]. This formula serves as an ideal model system for several reasons. First, its composition exemplifies the complex multi-herb combinations typical in TAM. Second, its established clinical applications in treating conditions such as diabetes [13], hypertension [14], and peripheral arterial disease [15] provide a relevant therapeutic context. Third, its role as a foundation for numerous derivative prescriptions [16–18] highlights its significance in TAM's theoretical framework.

Using PM as a proof-of-concept model, we systematically investigated how CD modifies chemical extraction compared to ID, with particular focus on identifying structure-dependent patterns in these modifications. We further examined how these chemical alterations correlate with therapeutic properties, validating our findings through experimental assessment of cytotoxicity and antioxidant effects.

This investigation provides a chemical basis for understanding traditional decoction practices while establishing an analytical framework applicable to other multi-herb formulations. By elucidating how CD selectively modifies chemical extraction based on structural properties, our findings offer insights for standardizing preparation methods while preserving their therapeutic benefits.

Material and methods

Herbal extracts

Commercial herbal raw materials were purchased from GMP-certified suppliers, and authentication and quality control of each raw herb were performed by the manufacturers (MRC, Entab Herb Co., ETC22802-1, China; PS, Sun-il Co., SQ-22029–1, Korea; CF, Sun-il Co., SQ-2303–1, Korea; RRR, Shin Hung Pharmaceuticals Co., SHG-DD1-001–23-021, China; AR, Sun-il Co., 5C-21066–3, Korea; CC, Entab Herb Co., ETV22671-1, Vietnam; ALRP, Shin Hung Pharmaceuticals Co., SHG-002–001–22–001, China; DR, Sun-il Co., SQ-22032–1, Korea). For internal reproducibility and traceability, voucher specimens were deposited in the KM Science Research Division, Korea Institute of Oriental Medicine, with the following specimen numbers: RRP KIOM-KPM2_5, CF: KIOM-KPM2_3, DR: KIOM-KPM2_4, PS: KIOM-KPM2_2, AR: KIOM-KPM2_6, MRC: KIOM-KPM2_1, CC: KIOM-KPM2_7, and ALRP: KIOM-KPM2_8. Extracts for PM and each herb in PM were prepared by CD and ID, respectively. For CD, which follows traditional preparation methods, all constituent herbs of PM were simultaneously decocted together in water (1:10 w/v) at 85 °C for 3 h, in accordance with the MFDS guideline for ‘standard decoction’ (Notification No 2012–22). For ID, which serves as a control to evaluate herb-herb interactions during decoction, each herb was separately extracted under identical conditions (1:10 w/v, 85 °C, 3 h), and the resulting individual extracts were subsequently combined in proportion to their original formula ratios. Both CD and ID extracts were filtered through filter paper (Whatman, No 1, pore size 11 μm) to remove particulate matter and insoluble residues, concentrated under reduced pressure at 60 °C, and lyophilized to obtain the final extracts. Detailed information on the decoction including exact weights of individual herbs and extraction yields is provided in Supplementary Table S1.

Chemical profiling

LC/MS grade water, methanol, acetonitrile (> 99.9%, Fisher Scientific, Waltham, MA, USA), and formic acid (Sigma-Aldrich, St. Louis, MO, USA) were used. Extracts (10 mg) were analyzed in quintuplicate using an Agilent 6546 Q-TOF MS coupled with a 1290 Infinity II LC System (Agilent Technologies, Santa Clara, CA, USA). Separation occurred on a Zorbax Extend-C18 column (2.1 × 50 mm, 1.8 μm) at 40 °C with an injection volume of 1 μL. Mobile phases were (A) 0.1% formic acid in water and (B) 0.1% formic acid in acetonitrile. The gradient was as follows: 0–2.0 min, 95% A; 2.0–13.0 min, 15% A; 13.0–16.0 min, 95% A; 16.0–20.0 min, 95% A, with 0.1 mL/min flow rate. The Q-TOF MS operated in positive ion mode (50–1700 m/z), with 325 °C gas temperature, 35 psig nebulizer pressure, and 350 °C sheath gas temperature.

Data processing

Raw data were processed using MS-Dial with a mass accuracy tolerance of 0.01 Da and a minimum peak height of 1000. We excluded features whose values were less than five times the mean value of the method blank from further analysis for quality check from LC/MS. Feature-based molecular networks were constructed using Global Natural Products Social Molecular Networking (GNPS) (Supplementary Table S2) [19, 20] with modular structures identified by the Louvain algorithm [21, 22]. When calculating mean precursor abundance for each extract, the middle three values from five replicates were used, after excluding precursors with zero abundance. Compounds were considered predominant in a specific herb if its abundance of the compounds in the herb was more than twice that observed in any other herb.

Theoretical additive model for chemical compositions

To evaluate potential herb-herb interactions during decoction, we compared chemical profiles from CD with theoretical profiles representing ID. For each compound , its theoretical abundance in ID () was calculated as the weighted average of individual herb extracts as Eq. (1):

1

where is the compound 's abundance in herb extract, and is the weight (in grams) of herb in PM. Compounds were excluded if their abundance was less than five times the mean value of method blank samples. To quantify CD-induced changes in abundance for each compound , we calculated its abundance ratio score as Eq. (2):

2

where is the observed abundance of compound in CD, and is the sigmoid function. The sigmoid transformation was applied to normalize the fold-change ratios onto a bounded scale from 0 to 1. This facilitates intuitive visualization and statistical comparison. We established significance thresholds at scores of 0.731 and 0.269, which directly correspond to a two-fold increase () and a two-fold decrease (), respectively. The two-fold change criterion is a widely adopted convention in metabolomics and other -omics studies to distinguish biologically meaningful changes from baseline experimental variability and analytical noise. This threshold allows for the robust identification of compounds whose extraction is substantially and systematically altered by the combined decoction process.

Principal component analysis (PCA)

PCA model was fitted and transformed on z-scored data where its rows and columns indicate ID-prepared extracts of single herbs (8 herbs * 5 replicates = 40 rows) and precursors (120 columns), respectively. Abundance data of CD-prepared and ID-induced PM (5 rows for CD and 1 row for ID, and 120 columns) was transformed by the fitted PCA model. Loadings of individual compounds were calculated to identify key compounds contributing to the observed separation patterns.

Prediction of compound properties

Physicochemical properties (molecular weight, LogP, H-bond donors/acceptors, quantitative estimate of drug-likeness (QED) [23]) were calculated using RDKit. Toxicity probabilities for 12 mechanisms were predicted using Generative Toolkit for Scientific Discovery [24] trained on Tox21 datasets [25]. Spearman correlations were calculated between these properties and abundance ratio scores. For constant feature values across compounds, correlation coefficients were set to 0.

In vitro assays

To confirm our predictions, in vitro assays were performed on human renal tubule epithelial cells (HK2). HK2 were obtained and authenticated from ATCC (ATCC® CRL-2190, Manassas, VA, USA). Treatments with PM prepared by CD or ID were administered at concentrations ranging from 0 to 200 μg/mL (0 μg/mL, 10 μg/mL, 50 μg/mL, 100 μg/mL, and 200 μg/mL). More specific methods are described in Supplementary Material S1.

Statistical analysis

The Kruskal–Wallis test was used to compare continuous variables across modules, including mass-to-charge ratios, retention times, and abundance changes, as the data did not meet the assumption of normal distribution. Fisher's exact test was applied to analyze categorical distributions, such as compound ontology and herbal origins across modules. Correlations between compound properties and abundance changes were assessed using the Spearman correlation coefficient, which is robust to outliers and does not assume a linear relationship between variables. For comparison between two groups in cell-based assays, the Mann–Whitney U test was performed. Statistical significance was set at p < 0.05.

Results

CD generates distinct chemical compositions from ID

LC/Q-TOF–MS analysis of PM and individual herb extracts identified 120 precursors (Fig. 1A, B). PCA model was fitted to the chemical profiles from individually decocted herbs, revealing distinct clustering patterns with ALRP showing higher values in the first principal component (PC1) and CF showing higher values in the second principal component (PC2). When this established PCA model was used to transform the PM samples, the CD-prepared PM showed higher PC1 and PC2 values compared to the theoretically predicted values for its ID counterpart. This indicates that CD enhances the characteristic features of ALRP and CF beyond what would be expected from simple mixing (Fig. 1C). Analysis of loading values identified compounds including benzoylmesaconine, benzoylhypaconine, fuziline, talatisamine (PC1), morroniside, loganin, and sweroside (PC2) as key contributors to this separation (Fig. 1D). These results suggest that these compounds are not only specialized in ALRP and CF but also increased in PM by CD. Quantitative analysis for an abundance of each compound confirmed these analysis results. ALRP-derived compounds (benzoylmesaconine, benzoylhypaconine, fuziline) showed higher abundance in PM compared to theoretical PM (Fig. 1E). Similarly, CF-derived compounds (morronside, loganin) showed increased abundance in PM compared to their counterparts in the theoretical additive model (Fig. 1F). These results demonstrate that CD selectively enhances the extraction of certain compounds, particularly those characteristic of ALRP and CF, while potentially suppressing others.

Fig. 1.

Chromatographic analysis and multivariate statistical evaluation of Palmijihwang-tang (PM) and herbal extracts. A, B Base peak chromatograms obtained from PM by combined decoction (A) and individual herb extracts by individual decoction (B) using LC-QTOF-MS analysis. C, D Principal component analysis showing (C) precursor profiles from individual extracts with their clustering patterns (arrow indicates the gap between observed abundance and theoretical abundance of PM) and (D) key precursors contributing to the largest principal component (PC) values. E Abundance profiles of compounds with the highest PC1 contributions: benzoylmesaconine, benzoylhypaconine, and fuziline. F Abundance profiles of compounds with the highest PC2 contributions: morronside, loganin, and acetovanillone primeveroside. Colors indicate different herb sources: BRP (blue), DR (orange), CF (green), MRC (turquoise), PS (purple), AR (khaki), CC (gray), and ALRP (yellow). RRP: Rehmanniae Radix Preparata; CF: Corni Fructus; DR: Dioscoreae Rhizoma; PS: Poria Sclerotium; AR: Alismatis Rhizoma; MRC: Moutan Radicis Cortex; CC: Cinnamomi Cortex; ALRP: Aconiti Lateralis Radix Preparata

These systematic differences between observed and theoretical profiles suggest that CD selectively influences the extraction of compounds based on their specific structural characteristics. To systematically analyze these structure-dependent changes, we next employed molecular networking to classify compounds into structurally similar groups and examine how structural features relate to extraction patterns.

Molecular network modules reflect compound ontology and herbal origins

Structural similarity analysis using the Louvain algorithm clustered 57 compounds into four distinct modules (Table 1, Fig. 2A). These modules showed significantly characteristic (Fisher's exact test, p < 0.001) patterns in both compound ontology (Fig. 2B), while basic physicochemical properties were similar (Kruskal–Wallis test, p = 0.436 and 0.373) across modules (Fig. 2C). Module 1 predominantly contained glycosides (7 out of 10 compounds), including dianthoside and morroniside. Module 2 contained a diverse array of terpenoids, flavonoids, and phenolic compounds, while Module 3 was particularly rich in terpenoids. Module 4, the largest cluster with 17 compounds, almost exclusively consisted of alkaloids and their derivatives.

Table 1.

Representative precursors included in the molecular network. A complete list of all precursors is provided in Supplementary Table S3. m/z: mass-to-charge ratio; Rt: retention time

Module	Ontology	Pubchem ID	Name	Rt (min)	Formula	m/z	Adduct	△ ppm
1	Glycosides	127258930	7-O-Methyl morroniside	7.225	C18H28O11	443.1533	[M + Na] +	−2.91
1	Glycosides	21602024	Daphylloside	6.717	C19H26O12	469.1307	[M + Na] +	−1.49
1	Glycosides	87691	Loganin	6.472	C17H26O10	413.1427	[M + Na] +	−6.51
1	Glycosides	11228693	Morroniside	5.972	C17H26O11	429.1374	[M + Na] +	−0.96
1	Glycosides	92043450	Hastatoside	6.321	C17H24O11	427.1212	[M + Na] +	−0.47
2	Alkaloids and Derivatives	14163819	Fuziline	5.836	C24H39NO7	454.2813	[M + H] +	−2.86
2	Glycosides	3084296	Citrusin C	7.937	C16H22O7	349.1275	[M + Na] +	7.16
2	Glycosides	24721183	Acetovanillone primeveroside	6.608	C20H28O12	483.1489	[M + Na] +	−3.93
2	Terpenoids and Diterpenoids	24721502	Ginkgolide C	6.32	C20H24O11	441.1376	[M + H] +	3.47
2	Terpenoids and Diterpenoids	14036813	Alisol C Monoacetate	11.375	C32H48O6	529.3509	[M + H] +	−1.74
3	Glycosides	45360282	Myxopyroside	2.015	C18H26O13	473.127	[M + Na] +	0.02
3	Glycosides	3038513	Primeverin	6.266	C20H28O13	499.1416	[M + Na] +	0.8
3	Terpenoids and Diterpenoids	12305221	Bayogenin	11.889	C30H48O5	511.3417	[M + Na] +	−3.32
4	Alkaloids and Derivatives	11452543	Isotalatizidine	5.086	C23H37NO5	408.2767	[M + H] +	−5.54
4	Alkaloids and Derivatives	78358535	Benzoylhypaconine	7.903	C31H43NO9	574.3012	[M + H] +	1.39
4	Alkaloids and Derivatives	24832659	Benzoylmesaconine	7.463	C31H43NO10	590.2982	[M + H] +	3.03
4	Alkaloids and Derivatives	441749	Napelline	5.42	C22H33NO3	360.2535	[M + H] +	−0.47
4	Alkaloids and Derivatives	441742	Karakoline	4.902	C22H35NO4	378.2642	[M + H] +	−0.82
4	Alkaloids and Derivatives	245005	Aconitine	8.819	C34H47NO11	646.3228	[M + H] +	−0.93
4	Alkaloids and Derivatives	71456946	Songorine	5.338	C22H31NO3	358.2387	[M + H] +	3.60
4	Alkaloids and Derivatives	441737	Hypaconitine	8.818	C33H45NO10	616.3151	[M + H] +	−5.6
4	Alkaloids and Derivatives	76189740	Hypaconine	5.802	C24H39NO8	470.2743	[M + H] +	−0.66
4	Alkaloids and Derivatives	10003218	Bullatine B	6.06	C24H39NO6	438.2859	[M + H] +	9.38
4	Glycosides	69634125	Forsythoside E	5.71	C20H30O12	485.164	[M + Na] +	−8.24

Fig. 2.

Modular analysis of Palmijihwang-tang compound networks and their characteristics. A Molecular network visualization representing structural similarities between compounds, where edges connect structurally related compounds and nodes are colored by their respective modules. B-D Distribution analyses of (B) compound ontology across modules (numbers in cells indicate compound counts), (C) mass-to-charge ratio (m/z) and retention time characteristics of each module, and (D) compound origin patterns across modules. Statistical significance of inter-module differences is indicated (p-values shown where significant). E Compound-specific abundance ratio scores () indicating relative abundance changes in PM combined extraction, with statistical significance assessment (p = 0.053). Each dot represents an individual compound. F Identification of compounds showing scores above or below threshold values, with module-specific color coding indicating compound classification. m/z: mass-to-charge ratio; RRP: Rehmanniae Radix Preparata; CF: Corni Fructus; DR: Dioscoreae Rhizoma; PS: Poria Sclerotium; AR: Alismatis Rhizoma; MRC: Moutan Radicis Cortex; CC: Cinnamomi Cortex; ALRP: Aconiti Lateralis Radix Preparata

The distribution of modules, each representing compounds with similar structures, showed significant variation (Fisher's exact test, p < 0.001) across different formulas and their constituent herbs (Fig. 2D). Module 4 compounds were almost exclusively derived from ALRP in PM, representing approximately 85% of all compounds in this module. Module 1 and 3 compounds showed distinct distribution patterns, with Module 1 compounds predominantly originating from CF and Module 3 compounds mainly derived from CC.

These module-specific patterns in compound distribution and structural similarity provide a systematic framework for understanding how CD affects different classes of compounds. Using this modular classification, we next investigated how CD specifically altered the abundance of compounds within each structural class to reveal patterns of enhancement and suppression across different compound types.

CD affects chemical abundance with module-specific patterns

To evaluate CD effects, we compared two chemical compositions: the actual profile of CD-prepared PM and a theoretical profile (null profile) representing ID-prepared PM without any herb-herb interactions. CD in PM significantly altered the abundance of 22 compounds, with 16 showing increased and 6 showing decreased abundance compared to ID. We observed distinct patterns in how compound abundances changed across different modules (p = 0.053). Nearly half of the compounds in Module 4 showed significant increases, while Module 3 contained half of all compounds that showed significant decreases (Fig. 2F). These patterns indicate that a compound's structural features may determine whether its concentration increases or decreases during CD.

To elucidate the relationship between specific structural features and abundance changes, we constructed molecular networks for each module based on structural similarity. Module 1, which primarily contained glycosides such as morroniside, loganin, and melezitose, showed an overall trend of increased abundance. Specifically, iridoid O-glycosides including morroniside showed marked increases (= 0.83 and 0.80), while 7-O-Methyl morroniside showed a decrease (=0.15) (Fig. 3A). This selective enhancement indicates that O-glycosyl methylation affects how compounds are extracted during CD.

Fig. 3.

Module-specific molecular networks and their representative compounds. A-D Detailed visualization of molecular networks for Modules 1–4, respectively. Chemical structures of compounds showing significant abundance changes in Palmijihwang-tang are highlighted (pink boxes: increased abundance; blue boxes: decreased abundance). For each compound, retention time (Rt), mass-to-charge ratio (m/z), ontology classification, and abundance ratio score () are provided. Node colors indicate relative abundance changes (red: increased; blue: decreased), with edge connections representing structural similarities between compounds

Module 2, comprising diverse compounds including phenolics, terpenoids, and glycosides, exhibited ontology-dependent changes. Specifically, diterpenoid alkaloids such as fuziline and phenolic glycosides like citrusin C (=0.88 and 0.75, respectively) showed increased abundance, while flavonoid-7-O-glycosides decreased (=0.33) (Fig. 3B). These differential effects highlight how subtle structural variations within similar compound classes can lead to distinct extraction behaviors during CD.

Module 3, dominated by terpenoids and glycosides, demonstrated a general trend toward decreased abundance during CD. This effect was particularly pronounced for high-molecular-weight glycosides, including primeverin, CID 10839700, CID 45783027, and CID 71694455 (=0.35, 0.07, 0.18, and 0.35, respectively) (Fig. 3C). This pattern suggests that molecular weight may be a key determinant of extraction efficiency within the glycoside structural class.

Module 4, characterized by alkaloids including benzoylhypaconine, karakoline, benzoylmesaconine, and songorine (=0.78, 0.83, 0.94, and 0.91, respectively) showed consistent increases in abundance (Fig. 3D). This systematic enhancement suggests that CD in PM may specifically facilitate the extraction of ALRP-derived alkaloids compared to ID.

These systematic differences in compound changes across modules demonstrate that extraction yields strongly depend on chemical structure. To understand the therapeutic implications of these structural-dependent changes, we next examined how these alterations in chemical composition correlate with key pharmacological properties of the compounds.

Chemical profile changes correlate with compound properties

To understand the patterns underlying these CD-induced changes, we examined their relationship with compound properties (Supplementary Table S4). Abundance changes showed systematic correlations with key molecular features. In PM, changes correlated positively with LogP (Spearman correlation r = 0.31, n = 120) and QED (r = 0.43, n = 120), but negatively with H donors (r = −0.45, n = 120) and H acceptors (r = −0.29, n = 120) (Fig. 4A). These correlations aligned with general ADME thresholds: compounds with enhanced extraction levels typically met ADME criteria (Fig. 4B, C).

Fig. 4.

Correlation and functional analyses of compound properties and combined decoction (CD)-induced changes (). A Correlation heatmap between bioavailability and abundance ratio scores in Palmijihwang-tang (PM) (red: positive correlation; blue: negative correlation). B, C Module-specific correlation analyses between abundance changes and (B) quantitative estimate of drug-likeness (QED) or (C) number of hydrogen donors. Each point represents an individual compound, colored by its molecular network module (Module 1: blue, Module 2: orange, Module 3: green, Module 4: red). Solid lines indicate linear regression fits, with shaded areas representing 95% confidence intervals. Spearman correlation coefficients (r) are shown for each correlation. D Comprehensive correlation matrix showing relationships between toxicity pathways (Heat Shock Element [HSE], Antioxidant Response Element [ARE], Peroxisome Proliferator-Activated Receptor gamma [PPAR-gamma], Aromatase [CYP19A1], Aryl Hydrocarbon Receptor [AhR], Tumor Protein p53, ATPase Family AAA Domain-containing protein 5 [ATAD5], Estrogen Receptor Ligand Binding Domain [ER-LBD], Androgen Receptor Ligand Binding Domain [AR-LBD], Estrogen Receptor [ER], Androgen Receptor [AnR], and Mitochondrial Membrane Potential [MMP]) and abundance changes across different modules and compound classes. E–G Pathway-specific correlation analyses for (E) heat shock element, (F) antioxidant response element, and (G) aromatase (CYP19A1) activity. Other details in this figure are the same as (B) and (C). H Cytotoxicity comparison between CD and individual decoction (ID) (mean ± SD, n = 8). (I,J) Representative fluorescence microscopy images showing dihydroethidium (DHE) staining for reactive oxygen species (ROS) detection. Samples were treated with tert-butyl hydroperoxide (tBOOH, negative control), N-acetylcysteine (NAC, positive control), and extracts obtained from CD or ID at various concentrations. Statistical significance: #p < 0.1, *p < 0.05, **p < 0.01, ***p < 0.001; n.s.: not significant

Machine learning prediction further revealed that abundance changes correlated with compounds' potential interactions with toxicity-related pathways. In PM, compounds predicted to be associated with the heat shock response (HSE), antioxidant response (ARE), and aryl hydrocarbon receptor pathway showed decreased abundance (Fig. 4D). These correlations were particularly strong for Module 1 and 4 compounds (Fig. 4E-G), further highlighting the distinct behavior of O-glycosides and alkaloids under different extraction conditions.

These systematic correlations between structural features and CD-induced changes suggest that CD selectively modifies chemical compositions based on compound properties. This property-based selectivity may reflect centuries of empirical optimization in traditional medicine, where decoction methods were refined to preferentially extract compounds with favorable therapeutic properties while potentially limiting the extraction of less desirable components. Such chemical selectivity could represent one of the underlying mechanisms through which traditional practitioners empirically optimized herbal formulations for both efficacy and safety, even without explicit knowledge of molecular structures.

To validate these in silico predictions experimentally, we compared the cytotoxicity and antioxidative effects between CD- and ID-prepared PM in HK2. Consistent with our computational predictions showing negative correlations with toxicity-related pathways, CD-prepared PM showed significantly lower cytotoxicity than its ID counterpart across various concentrations (Mann–Whitney U test, p < 0.05; Fig. 4H). Moreover, when we measured antioxidative effects using dihydroethidium (DHE) staining, CD-prepared PM exhibited stronger antioxidant effects than its ID counterpart (Mann–Whitney U test, p < 0.1; Fig. 4I and J), aligning with our prediction of enhanced ARE pathway modulation. These experimental results support our computational findings that CD alters not only the chemical composition but also the biological effects of the herbal formula.

Discussion

Herb combinations are fundamental to TAM, with practitioners historically claiming enhanced therapeutic effects from specific herb combinations. The traditional "Jun-Chen-Zuo-Shi" principle exemplifies this approach, classifying herbs into four functional roles: monarch, minister, assistant, and courier herbs [26]. Recent research has begun to validate these traditional concepts through advanced methodologies, particularly by analyzing herb-target interactions in protein–protein interaction networks [2]. While these studies have illuminated the biological mechanisms underlying herbal synergy, they largely overlook a critical aspect of traditional medicine: the preparation process itself. Specifically, the chemical interactions that occur during decoction and their influence on therapeutic efficacy remain poorly understood.

To address these limitations in understanding herb-herb interactions, we developed a novel analytical framework that combines three complementary approaches. First, we employed LC/MS-Q-TOF measurements to accurately quantify chemical abundances in extracts. Second, we created a theoretical additive model to predict theoretical chemical profiles in the absence of herb-herb interactions. Finally, we conducted in silico analyses to understand the pharmacological implications of observed chemical changes. This comprehensive approach provides a unique solution to the fundamental challenge of studying multi-herb formulations, where direct observation of non-interacting states is impossible.

Using this analytical framework, we uncovered significant differences in compound abundance between CD and ID. Most notably, these differences showed distinct patterns among compound classes, with alkaloids demonstrating particularly striking changes. The observed increases in alkaloid content varied systematically with predicted bioavailability and toxicity, suggesting that CD selectively enhances the extraction of therapeutically favorable alkaloids while potentially limiting the extraction of more toxic compounds. This result is consistent with recent reports on other herbal pairs. For example, Ginseng-Fuzi co-decoction was reported to show a "directionally detoxification effect," where the process selectively decreased toxic diester alkaloids while simultaneously increasing the concentration of less toxic, therapeutic monoester alkaloids. The proposed mechanism involves fatty acids abundant in ginseng acting as catalysts to facilitate nucleophilic substitution reactions that promote the hydrolysis of toxic alkaloids [27]. PM similarly contains diverse herbs beyond ALRP, with CF being rich in organic acids and DR containing substantial amounts of saponins. We hypothesize that these components may perform roles analogous to the fatty acids in the Ginseng-Fuzi studies. Specifically, organic acids from CF or saponins from DR could form complexes with ALRP-derived alkaloids, potentially enhancing solubility or catalyzing specific chemical reactions that modify structural features, ultimately increasing their concentrations in the decoction.

Another plausible mechanism involves the formation of self-assembled phytochemical complexes. Previous studies have demonstrated that co-decoction of herb pairs can generate nanospheres with enhanced antimicrobial activity, whereas simple mixtures form nanofibers with lower activity [28]. The structure-dependent abundance patterns observed in our study may be influenced by differential abilities of compounds to integrate into these colloidal structures, which could serve as a key factor in altering compound stability and solubility in the final extract.

Our data suggest that these co-decoction-induced chemical changes are directly linked to the observed alterations in biological activity, namely the reduced cytotoxicity and enhanced antioxidant effects. This underscores the critical role of the structure–activity relationship (SAR) in determining the final therapeutic properties of the formula. Indeed, SAR analysis remains a cornerstone in natural product research for elucidating how the structural features of phytochemicals govern their biological functions [29, 30]. Subtle structural modifications such as methylation or glycosylation patterns in flavonoids and other natural products have been consistently reported to dramatically alter biological activities [30]. Furthermore, the CD-induced degradation of diester structures and subsequent generation of monoester or amino alcohol forms of alkaloids may serve as a key mechanism for simultaneously enhancing both the safety and efficacy of the formulation.

The differential changes in compound abundances during CD likely arise from multiple mechanisms. These include direct chemical interactions between compounds from different herbs during decoction, alterations in extraction medium properties such as pH or polarity due to the presence of multiple herbs [31, 32], and competitive extraction processes based on differential solubility.

The duration of herb co-exposure during decoction appears particularly crucial in determining the final chemical composition [33], suggesting that traditional preparation methods may have been empirically optimized over generations of practice.

Our cell-based assays provided strong experimental validation for these in silico predictions. The reduced cytotoxicity and enhanced antioxidant effects observed in CD-prepared PM, compared to its ID counterpart, demonstrate that decoction-induced chemical alterations directly influence biological activity. This alignment between computational predictions and experimental results not only validates our analytical approach but also highlights the critical importance of preparation methods in determining therapeutic properties.

These findings have significant implications for quality control in herbal medicine manufacturing. Current quality control standards typically focus on analyzing individual herbs within prescriptions [10], largely ignoring the chemical changes that occur during combined preparation. Our results demonstrate that decoction methodology significantly influences final product composition, suggesting that quality control standards for multi-herb medicines should incorporate analyses of combined extracts to ensure an accurate representation of the final product's chemical profile.

Our study extends previous research on herb pair co-extraction in several important ways. Earlier studies on co-extraction primarily focused on precipitate formation or basic chemical differences between preparation methods [6, 8, 34, 35]. In contrast, our analysis reveals that combined decoction (CD) induces systematic changes within the liquid extract itself. We found that these changes in chemical composition directly correlate with the formulation's therapeutic properties. We demonstrate that compound concentrations vary significantly depending on specific herb combinations, these changes correlate with drug-likeness and potential toxicity, and significant compositional changes occur within the liquid extract itself, not just in precipitates.

These findings suggest a need to reevaluate current approaches to herbal medicine standardization. Traditional methods that assess only the presence and quantity of specific marker compounds may miss crucial preparation-dependent changes in chemical composition. Moreover, alternative preparation methods that deviate from traditional decoction processes, such as consuming herbs without extraction, may yield substantially different compound interactions and therapeutic effects [11]. This highlights the importance of considering preparation methodology not just as a practical necessity, but as a crucial determinant of therapeutic outcome.

The systematic correlation between CD-induced changes and compound properties provides particular insight into the empirical wisdom embedded in traditional medicine practices. Our results suggest that traditional decoction methods may have been unconsciously optimized over centuries. This empirical optimization appears to achieve two goals: it preferentially extracts compounds with favorable therapeutic properties, and it simultaneously limits the extraction of potentially harmful components. This optimization appears to operate through multiple mechanisms, including selective enhancement of specific compound classes and modification of extraction efficiency based on physicochemical properties.

These insights have relevance for the modernization of TAM. As traditional formulations are increasingly developed into standardized pharmaceutical products, our findings emphasize the importance of carefully considering preparation methodology. The significant differences we observed between CD and ID in PM suggest that modernized production methods should be carefully evaluated to ensure they preserve the beneficial chemical interactions that occur during traditional preparation processes.

Our study, while focused on PM as a model system, establishes an analytical framework that could be applied to other multi-herb formulations. The combination of theoretical additive modeling with molecular networking provides a systematic approach for understanding how CD modifies chemical compositions. This analytical framework could be valuable for future research. A key investigation would be to determine if the structure-dependent extraction patterns observed in PM represent general principles of CD. For example, patterns like enhanced alkaloid extraction and selective glycoside modification could be tested in other formulations.

Future research should focus on validating these patterns across different traditional prescriptions while elucidating the specific mechanisms underlying CD-induced changes. This could include systematic evaluation of how different decoction parameters (temperature, duration, herb ratios) affect final chemical compositions, potentially leading to optimized preparation protocols that maximize therapeutic benefits while maintaining traditional principles. Such investigations could provide crucial insights for standardizing CD processes across various traditional formulations while preserving their intended therapeutic properties.

Conclusions

This study establishes a systematic framework for understanding how CD modifies the chemical composition of herbal formulations through structure-dependent selective extraction. Using PM as a proof-of-concept model, we demonstrated that CD selectively enhances certain compound classes while suppressing others. Furthermore, these chemical modifications correlated well with the formulation's therapeutic properties. Our experimental results validated this correlation, showing enhanced antioxidant activity and reduced cytotoxicity. Our analytical approach, combining theoretical additive modeling with molecular networking, provides a robust methodology for investigating CD-induced changes in complex herbal preparations. These insights into structure-dependent extraction patterns not only provide a chemical basis for traditional preparation methods but also offer crucial guidance for standardizing CD processes while preserving their intended therapeutic properties. While our findings are based on a single formulation, the analytical framework established here opens new avenues for investigating CD effects across various traditional prescriptions, potentially leading to improved understanding and standardization of this fundamental preparation method in TAM.

Abbreviations

ADME

Absorption, Distribution, Metabolism, and Excretion

AhR

Aryl Hydrocarbon Receptor

ALRP

Aconiti Lateralis Radix Preparata

AnR

Androgen Receptor

AR

Alismatis Rhizoma

AR-LBD

Androgen Receptor Ligand Binding Domain

ARE

Antioxidant Response Element

ATAD5

ATPase Family AAA Domain-containing protein 5

CC

Cinnamomi Cortex

CD

Combined decoction

CF

Corni Fructus

DR

Dioscoreae Rhizoma

ER

Estrogen Receptor

ER-LBD

Estrogen Receptor Ligand Binding Domain

FBMN

Feature-based Molecular Network

HSE

Heat Shock Element

ID

Individual decoction

LC/Q-TOF-MS

Liquid Chromatography coupled with Quadrupole Time-of-Flight Mass Spectrometry

MMP

Mitochondrial Membrane Potential

MRC

Moutan Radicis Cortex

p53

Tumor Protein p53

PM

Palmijihwang-tang

PPAR-gamma

Peroxisome Proliferator-Activated Receptor gamma

PS

Poria Sclerotium

QED

Quantitative Estimate of Drug-likeness

RRP

Rehmanniae Radix Preparata

TAM

Traditional Asian Medicine

Authors’ contributions

J.J. conceptualized the research idea and hypothesis of investigating chemical composition changes during combined decoction. D.J. developed the theoretical additive model methodology. S.S. and S.L. performed the chemical analysis including sample preparation, LC/MS-Q-TOF measurements, and compound identification and quantification. S.M.L. conducted the in vitro experiments and collected experimental data. D.J. performed comprehensive data analysis integrating the chemical profiling and in vitro results, and wrote the original draft of the manuscript. C.E.K. and J.J. supervised the overall research process and provided critical revision of the manuscript. All authors have read and approved the final version of the manuscript for publication.

Funding

This work was supported by a grant from the Korea Institute of Oriental Medicine (grant number: KSN2313021) and the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (RS-2024–00339889).

Data availability

Detailed information about software versions and parameters is provided in Supplementary Table S5. The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

All experiments involving the human cell line (HK2) were conducted in accordance with the institutional biosafety guidelines of Korea Institute of Oriental Medicine. This study did not require separate ethical approval as it did not involve animal or human subjects.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.