Dose-Weighted Network Pharmacology: Evaluating Traditional Chinese Medicine Formulations for Lumbar Disc Herniation

Introduction

In the literature indexed by the Web of Science Core Collection, Albeck¹ was the first to use the term “Lumbar Disc Herniation” (LDH) to describe the protrusion of disc contents caused by the rupture of the annulus fibrosus. Over the past thirty years since the publication of this work, epidemiological reports have indicated that LDH affects the daily lives of approximately 9% of the global population, with the incidence increasing in tandem with population aging.²

Given the lower back pain and economic burden associated with LDH, extensive research has focused on treatment strategies. Studies on patient adherence suggest that addressing lower limb pain can significantly improve treatment compliance.³ While surgical interventions may yield superior outcomes, non-surgical treatments remain the preferred choice for most newly diagnosed LDH patients.^2,4 Among non-surgical options, traditional Chinese medicine (TCM) has shown unique strength.⁵

The unique strength of TCM lies in its flexibility to adapt formulations based on specific conditions.⁶ The advent of network pharmacology has facilitated a more intuitive understanding of the molecular mechanisms underlying TCM, addressing prior uncertainties.⁷ By employing complex network models, network pharmacology enables a systematic evaluation of TCM efficacy, and its approach has gained substantial recognition.⁸ However, challenges remain in simplifying TCM formulations within these models. In particular, the omission of dosage considerations leads to overly idealized assumptions, such as different doses producing the same levels of active ingredients, which could skew predictions of therapeutic outcomes. Despite these limitations, the “drug-active ingredient-target-disease” pathways have validated the potential for data transmission, and with appropriate computational models, drug dosage information could be incorporated into network analyses.

Accordingly, this study included Taohong Siwu Decoction (TSD) and a modified version of TSD for LDH, known as Yaotong Jizheng Decoction (YJD), developed by Professor Yang Lixue, a renowned TCM practitioner from Shaanxi Province. These were compared with the commonly used Chinese patent medicine Panlong Qi Tablet (PQT). Utilizing the Vlsekriterijumska Optimizacija I Kompromisno Resenje (VIKOR)⁹ scoring method, we developed a TCM formula efficacy scoring algorithm based on the protein–protein interaction (PPI) network and drug dosage. This study aims to explore the feasibility of incorporating dosage factors into TCM omics analysis. The complete research process is illustrated in Figure 1.

Figure 1 Technical roadmap for weighted network scoring and experimental validation. Abbreviations: DC, Degree Centrality; CC, Closeness Centrality; BC, Betweenness Centrality; GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; VIKOR, Vlsekriterijumska Optimizacija I Kompromisno Resenje; Drug VIKOR, VIKOR parameters in terms of drug; Final Score, The final target parameter after mixing the two parameters (Drug VIKOR & LDH VIKOR); LDH VIKOR, VIKOR parameters in terms of Lumbar disc herniation; Drug Scores, The total number of scores for the targets contained in the drug.

Materials and Methods

Drug Quantification and Target Collection

The 6 drugs in TSD use doses recorded in the Qianlong 7th Year Wuying Palace printed edition. The 9 drugs in YJD use common clinical doses. The 29 drugs in PQT (National Medicine Standard Z61020050) use doses according to the national standards.

After completing data entry, traditional network pharmacology methods are applied, using oral bioavailability (OB) ≥30% and drug-likeness (DL) ≥0.18 as screening criteria. Active ingredients of each drug are searched in the TCMSP database.¹⁰ For drugs not included in the TCMSP database, relevant studies^11–13 are used to supplement the data. The chemical names, molecular formulas, and Canonical SMILES of each active ingredient are collected from the PubChem database¹⁴. Subsequently, the Canonical SMILES are input into the SwissTargetPrediction¹⁵ database to identify the corresponding targets of the active ingredients.

Collection of LDH Targets

LDH-related targets were retrieved from the GeneCards,¹⁶ DisGeNET,¹⁷ OMIM,¹⁸ NCBI¹⁹ and CTD²⁰ databases using the search term “Lumbar Disc Herniation”. The obtained targets were then organized, and duplicates were removed.

Discovery of Core Targets Based on PPI Network

Previous studies suggest that constructing a PPI network using the intersection of drug and disease targets can reveal the differential importance of drug action targets on diseases^21,22. This study advances conventional network pharmacology by applying an extremization approach to construct the PPI network for three drugs and the disease. This approach hypothesizes a disease that perfectly matches all drug action targets, allowing each target to exert its effect, thus highlighting the importance of each drug target. Conversely, it also hypothesizes a drug whose targets perfectly match all pathogenic targets of LDH, enabling “treatment” of each target and highlighting their importance in treating LDH.

Based on these hypotheses, the targets of the three TCM formulas collected in Drug Quantification and Target Collection and the LDH targets collected in Collection of LDH Targets were input into the STRING²³ database for PPI network construction. The resulting networks were imported into Cytoscape software, where network topology parameters, including BC, DC, and CC, were analyzed using the Centiscape tool.

Comprehensive Target Scoring Algorithm VIKOR

Given the complexity of the network, using a single indicator to describe the importance of targets would be inadequate.²⁴ To overcome the limitations of a single indicator, this study employs a target importance ranking method based on multi-criteria decision-making (MCDM) — the VIKOR index.⁹

Matrix Construction

Given the undirected PPI network after calculating topological parameters using Centiscape, where represents the set of all target nodes, , with the total number of targets denoted as , and the set of target evaluation criteria (tec) is represented as , with a total of criteria. The matrix can be formulated as follows:

Here, represents the value of the −th target for the −th criterion. In this study, the evaluation criteria are defined as .

Matrix Normalization

To eliminate dimensional differences between the criteria, the matrix D is normalized as:

In this formula, the numerator and denominator are fixed for each , indicating the normalization of the data for criterion . Once all criteria are normalized, the resulting normalized matrix is denoted as .

Assigning Weights to Criteria

For this study, all evaluation criteria are considered equally important, so the weight assigned to each criterion is .

Ideal and Negative-Ideal Solutions

The ideal solution and the negative solution are defined as:

Here, represents the set of benefit-type criteria, where higher values indicate greater importance of the target, and represents the set of cost-type criteria, where higher values are less favorable. In this study, all three criteria belong to the benefit-type.

Utility Measure and Regret Measure

The utility measure and the regret measure for all targets are computed as:

VIKOR Index Calculation

The importance of each target is quantified using the VIKOR index, calculated as:

Where , , , . Here, is the weight for the strategy of maximizing group utility, and is the weight for individual regret, with .

VIKOR Index Transformation

Since the VIKOR index is a cost-type criterion and , is transformed to facilitate subsequent calculations:

Bilateral Target Weight

Frequency Conversion of Drug Dosage and Active Ingredient Production

Since the current research cannot accurately obtain the specific relationship between drugs and the output of their active ingredients, this study simplifies the issue by assuming that each gram of a drug corresponds to one production run of the active ingredient. Therefore, for “a” different drugs producing the same active ingredient A, we calculate:

Where represents the mass of drug in grams.

Frequency of Active Ingredient Production and Target Interaction

Similarly, for “b” active ingredients acting on the same target B, the cumulative effect on the target is expressed as:

where is the number of times target B is affected, and represents the number of production runs of active ingredient y. In this study, we assume that the more frequently a target is affected, the more frequently the biological processes associated with that target are triggered after binding.

Drug-Specific Weight Settings

By combining the formulas from Frequency Conversion of Drug Dosage and Active Ingredient Production and Frequency of Active Ingredient Production and Target Interaction, we can establish a logical pathway linking drug production of active ingredients to target influence, as depicted in Figure 2.

Figure 2 Schematic Network of Drug-Active Ingredient-Target-Disease. Abbreviations: M, mass; nopr, Number of production runs.

The value represents the number of times target is affected. A higher indicates a stronger impact of the drug on target . After normalization, this is converted into a drug-specific weight:

Disease-Specific Weight

As shown in the logical pathway in Figure 2, the disease-specific influence on the target does not follow the hierarchical structure observed on the drug side. The weight for the disease side is derived from the VIKOR score for LDH targets , which reflects the variability in the target’s role in “treating” LDH, often due to individual differences. In this study, it is assumed that:

Drug Scoring Calculation Method

Bilateral Target Weight Scoring

For each LDH target, the weight is calculated as follows:

For each drug target, the corresponding weight is calculated as:

Normalization of Drug-Specific Scores

Since the introduction of drug-specific weights in Bilateral Target Weight Scoring results in having a range outside , a normalization step is performed to bring in line with :

Intersection Target Acquisition and Final Target Scoring

Using the VENNY 2.1 tool, the intersecting targets of the three TCM formulas with the LDH targets are identified. Each target has both a disease-specific score () and a drug-specific score (). The final target importance score () is determined by taking the weighted average of these two scores. In this study, the weights for disease and drug importance are considered equal:

Drug Scoring

Sum the of the intersection targets for the three groups (, , ). The drug score () is the comprehensive score of the targets contained in the drug and the importance of these targets:

Predictive Calculation of Experimental Results

The aim of this study is to evaluate whether the proposed calculation method can accurately predict the therapeutic effects of the drugs on LDH. Using the algorithm described above, the therapeutic outcomes of the intersecting targets of (, , ) are predicted. These predictions are then validated through animal experiments, which are divided into three treatment groups: TSD, YJD, and PQT (the specific dosage for each group is detailed in the animal experiment section). The steps outlined above are used to forecast the efficacy of the drugs in each experimental group, and the predictions are compared to the actual results after the experiments are completed.

GO and KEGG Enrichment Analysis of Targets

The ClusterProfiler tool was used to perform GO functional annotation and KEGG pathway enrichment analysis for the three sets of intersecting targets (, , ) independently. A significance threshold of p < 0.05 was applied to select relevant gene annotation categories and signaling pathways for GO functional annotation and KEGG pathway enrichment. The GO analysis was conducted across three domains: biological processes (BP), cellular components (CC), and molecular functions (MF). The top 10 results from both GO and KEGG analyses, based on enrichment significance, are presented.

Animal Experiment Validation

Experimental Animals

Twenty-five Sprague-Dawley (SD) rats, aged 6–8 weeks, were purchased from Chengdu Dashuo Laboratory Animal Co., Ltd. All animals were kept under a 12-hour light/dark cycle with controlled temperature (23 ± 1°C) and were provided with free access to food and water. The ethical approval for the animal experiments was granted under registration number SUCMDL20240304006.

Primary Experimental Materials

The Chinese medicinal herbs were obtained from the pharmacy of the Affiliated Hospital of Shaanxi University of Chinese Medicine. PQT tablets were sourced from Shaanxi Panlong Pharmaceutical Group Co., Ltd. (Shaanxi, China; specification: 0.3g×48 tablets; National Medicine Approval Number: Z61020050; product batch: 20231108). The following equipment and reagents were used: Hargreaves Infrared Plantar Test Device (Life Science, USA), Electrophoresis system (Tanon, Shanghai, China), Western blot and immunoprecipitation (IP) cell lysis buffer (Beyotime, Shanghai, China), General protease inhibitor (Biosharp), BCA Protein Assay Kit (Beyotime, Shanghai, China), SDS-PAGE loading buffer (Biosharp), Pre-stained Protein Marker (Abclonal), PVDF membrane (Sigma-Aldrich), Tween-20 (Servicebio), 30% Acrylamide Solution (Biosharp), HRP-conjugated Goat Anti-Mouse and Anti-Rabbit IgG (Abclonal, Affinity), GAPDH, IKK-β, and NF-κB p65 primary antibodies (Abclonal, Affinity, Proteintech), Ultra-sensitive ECL Western substrate (Zen-Bio), Molpure^® RNA extraction kit (YEASEN), PrimeScript RT Reagent Kit and TB Green™ Premix Ex Taq™ II (Takara Bio Inc)., Absolute ethanol (Xilong Scientific, Sichuan), Real-time PCR and thermal cycler instruments (ThermoFisher), ELISA kits for Rat IL-1β, IL-6, IL-18, and TNF-α (Yuanju Biotechnology, Shanghai).

Establishment and Treatment of Rat LDH Model

After a 7-day acclimation period, 20 rats were randomly selected and anesthetized via intraperitoneal injection of sodium pentobarbital (50 mg/kg) and placed in a prone position. Based on a comparative study of three modeling techniques,²⁵ the annulus fibrosus puncture method was chosen to induce LDH in rats. The procedure involved disinfecting the rats’ backs with 75% alcohol, making a longitudinal incision, and sequentially dissecting the skin, subcutaneous tissue, and lumbar fascia. After exposing the facet joints, the mammillary process was excised, and a No. 9 puncture needle was inserted at a 15° downward angle to a depth of approximately 3 mm. Upon feeling slight resistance followed by a release, 10 μL of IL-1β was injected into the disc using a microsyringe²⁶. The 20 successfully modeled rats were randomly assigned to four groups: the Model group, PQT group, YJD group, and TSD group. An additional group of 5 rats, which did not undergo modeling, served as the Control group.

In this experiment, the difference in dosage was considered to lie in the concentration of the decoction rather than the volume of the liquid. Based on the literature,²⁷ the optimal gavage volume for rats is 10 mL/kg, and with an average rat weight of 220 g, the volume of liquid administered for gavage was fixed at 2.2 mL. The concentration of the herbal decoction was adjusted by altering the ratio of raw herbs before decoction.

The doses for the YJD and TSD groups were calculated using body surface area conversion between humans and rats^28,29. The YJD group received 7.40 g/kg·d, and the TSD group received 3.70 g/kg·d. The required amount of raw herbs for each group was determined based on the number of animals and gavage schedule. The herbs were soaked in water for 30 minutes, then boiled for 20 minutes. After filtering, the herbs were boiled again for an additional 20 minutes. The two decoctions were combined and concentrated according to the planned gavage dosage, then refrigerated for later use. In the PQT group, the tablets were dissolved in pure water at a dose of 0.081 g/kg·d. Drug administration began 24 hours post-surgery, with the treatment groups receiving their respective drugs via gavage, while the control and model groups received the same volume of distilled water. All treatments were administered once daily for 4 weeks.

Thermal Hyperalgesia Test

Thermal hyperalgesia tests were conducted before the modeling procedure, before drug administration, and then every 7 days after the start of drug treatment for 4 weeks. Paw withdrawal latency (PWL) was measured using the Hargreaves infrared plantar stimulation device, which evaluated the thermal sensitivity of the plantar surface of the hind paw. Each animal underwent three measurements during each session, with a 5-minute interval between consecutive tests. The average of the three measurements was calculated as the thermal pain threshold. Thermal stimuli lasting 10 to 15 seconds were applied, with a 20-second limit to avoid tissue damage.³⁰

H&E Staining

At the end of the drug treatment period, the rats were euthanized, and disc tissue samples were collected. The samples were placed in EDTA for decalcification, followed by embedding, sectioning, and staining with hematoxylin and eosin (H&E). After rinsing the sections with running water to remove excess stain, they were sealed and dried.

Western Blot Analysis

Rat disc tissue samples were placed into 2 mL EP tubes, with 2 grinding beads (3 mm in diameter) and RIPA lysis buffer added to each tube. The tubes were then processed using a high-speed cryogenic tissue grinder (set to −20°C) for four rounds of grinding, each lasting 60 seconds. The samples were then placed in a 4°C refrigerator for 30 minutes to allow for lysis. Afterward, the samples were centrifuged (at 4°C, 12,000 rpm for 10 minutes), and the supernatant was collected. Protein concentrations were determined using a BCA Protein Assay Kit. The supernatant was mixed with 5× loading buffer, followed by heating at 95°C for 15 minutes to denature the proteins. The protein samples were then subjected to PAGE gel electrophoresis for separation and subsequently transferred onto a PVDF membrane. The membrane was blocked in TBST buffer containing 5% skim milk for 2 hours on a shaker at room temperature. After blocking, the membrane was washed three times with TBST, 5 minutes per wash. The membrane was then incubated overnight at 4°C with the appropriate primary antibody, diluted in blocking buffer. Following incubation, the membrane was thoroughly washed with TBST and then incubated with the secondary antibody at room temperature for 2 hours on a shaker. Protein bands were visualized using ECL chemiluminescence, and the exposed bands were captured using the Tanon Fluorescent Imaging System (version 2.0). The band intensities were analyzed using Gel-Pro Analyzer 4 software, with the results expressed as the integrated optical density (IOD) of the target proteins.

RT-PCR Experiment

Rat disc tissue samples were combined with lysis buffer in EP tubes and homogenized using a high-speed cryogenic tissue grinder. The homogenate was then transferred to an RNA adsorption column and centrifuged to collect the RNA-containing filtrate. After adding a binding buffer and mixing, the solution was transferred to a new RNA adsorption column, centrifuged again, and the filtrate was discarded. The column was washed multiple times with protein removal solution and rinse buffer, and the RNA solution was finally collected. To eliminate genomic DNA, a reaction system was prepared by mixing the Total RNA with a DNA removal reagent and incubating at 42°C for 2 minutes. A reverse transcription system was then set up to synthesize cDNA from Total RNA, which was performed in a thermal cycler. Full gene sequences were obtained from the NCBI database, and gene-specific primers were designed using Primer Premier software, synthesized by Shanghai Shenggong Biological Engineering Technology Services Co., Ltd. The PCR reaction system, along with the primers, underwent 50 cycles in a thermal cycler. The CT (Threshold cycle) values for each sample were obtained using Thermo Scientific PikoReal software, and the relative gene expression levels were calculated.

ELISA Experiment

Standard and sample wells were prepared. For each standard well, 50 μL of standard solution at varying concentrations was added. For the sample wells, 40 μL of sample dilution buffer was added, followed by 10 μL of the test sample, resulting in a 5-fold final dilution of the sample. No sample or enzyme-labeled reagent was added to the blank well, though all other steps were carried out the same. To all wells (except the blank), 100 μL of enzyme-labeled reagent was added. The plate was then sealed and incubated at 37°C for 60 minutes. A 20× concentrated washing buffer was diluted with distilled water at a 1:20 ratio. After incubation, the sealing film was carefully removed, and the liquid was discarded from the wells. Each well was washed with washing buffer, left for 30 seconds, and discarded, repeating this washing step 5 times, then blotting the plate dry. Subsequently, 50 μL of chromogenic reagent A and 50 μL of chromogenic reagent B were added to each well, mixed gently, and incubated at 37°C in the dark for 15 minutes. Afterward, 50 μL of stop solution was added to each well, stopping the reaction (the liquid color changed from blue to yellow). Finally, the absorbance at 450 nm was measured for each well using a microplate reader.

Result

Target Collection

This study included a total of 38 types of traditional Chinese medicines (with some repetitions) for drug dosage quantification. Detailed results are presented in Figure 3 and Supplementary Table 1.

Figure 3 Quantification of Drug Dosage.

After screening through the TCMSP database and additional literature sources, 732 targets were identified for TSD, 1017 targets for YJD, and 1180 targets for PQT. By integrating data from five different databases and removing duplicates, 1010 LDH-related targets were obtained, as shown in Supplementary Table 2.

PPI Network Construction

The target sets for the three TCM formulas and LDH were imported into STRING, with the species set to “Homo sapiens” and the minimum required interaction score set to the highest confidence level (0.9). This generated PPI networks for TSD (570 nodes), YJD (811 nodes), PQT (931 nodes), and LDH (702 nodes), with the corresponding TSV files. These files were then imported into Cytoscape, where the Centiscape 2.2 plugin was used to calculate the network topology parameters: Betweenness Centrality (BC), Degree Centrality (DC), and Closeness Centrality (CC). Detailed results are presented in Supplementary Tables 3–6.

Assignment of VIKOR Values

This section selects partial data as a simple example to illustrate the detailed steps of the proposed method. Based on the calculated data in PPI Network Construction, the TSD matrix can be constructed as:

Normalize the TSD matrix according to the normalization formula:

Calculate the remaining elements of the TSD matrix, and the normalized matrix TSD_R is:

The weights of the network topology parameters are calculated as:

Determine the ideal and negative solutions according to the formula:

Next, calculate the utility measure and regret measure:

where:

Then, the VIKOR index is calculated as:

Finally, transform :

Use the above method to calculate the TSD, YJD, PQT, and LDH matrices one by one. The specific results are shown in Supplementary Table 7.

Intersection Target and Weight Calculation

This section uses the data example from Assignment of VIKOR Values, selecting the well-structured data for demonstration. The data structure is shown in Table 1.

Table 1 Progressive Process

According to the logical chain in Figure 2 and the specific values in Table 1, we have:

After normalization:

After intersecting the TSD, YJD, and PQT target sets with the LDH target set, we obtain 167 targets, 210 targets, and 220 targets. The intersection targets have both and ,

Thus:

Using the above method, calculate the for the three groups of intersection targets (, , ). The specific results are shown in Supplementary Table 8.

Summing the target scores for each group, the predicted experimental results are as follows:

Based on these results, the predicted treatment efficacy ranks as YJD > PQT > TSD, with YJD and PQT showing highly similar effectiveness.

GO and KEGG Enrichment Analysis

The intersecting target sets (, , ) were individually analyzed using ClusterProfiler for GO enrichment across biological process (BP), cellular component (CC), and molecular function (MF). The comparative results are displayed in Figure 4.

Figure 4 Comparative GO and KEGG Enrichment Analysis of the Three Intersecting Target Sets. Abbreviations: YJD, Yaotong Jizheng Decoction; PQT, Panlong Qi Tablet; TSD, Taogong Siwu Decoction.

GO and KEGG analyses indicated a significant overlap in biological processes and signaling pathways among the three groups of intersecting targets, with many terms related to oxidative stress, inflammatory responses, and their associated pathways. All three groups were notably involved in the PI3K-Akt, MAPK, and NF-kappa B signaling pathways, which play key roles in inflammation and cellular stress responses.

Thermal Hyperalgesia Test and H&E Staining

As shown in Figure 5A, no significant differences were observed between the model group and the treatment groups before the 7th day of drug intervention (P > 0.05). From the 7th day onwards, the YJD group demonstrated a statistically significant difference compared to the model group (P < 0.05). By the 14th day, both the YJD and PQT groups exhibited significant differences compared to the model and TSD groups (P < 0.05), while the TSD group remained statistically similar to the model group. After 28 days of treatment, all treatment groups showed significant differences compared to the model group (P < 0.05), and the difference between the YJD group and the control group disappeared (P > 0.05). Detailed results are provided in Supplementary Table 9.

Figure 5 Comparative Analysis of Thermal Pain Threshold (A) and H&E Staining Results (B). Abbreviations: YJD, Yaotong Jizheng Decoction; PQT, Panlong Qi Tablet; TSD, Taogong Siwu Decoction.

In the H&E staining analysis (Figure 5B), the model group showed lower cell density and more pronounced degeneration compared to the control group. Among the treatment groups, YJD and PQT exhibited tissue morphology more similar to the control group, while TSD showed characteristics closer to the model group. Taken together, the thermal hyperalgesia test and H&E staining results confirm successful modeling and suggest a preliminary gradient of treatment efficacy among the groups.

Western Blot Analysis

The expression of IKK-β and NF-κB p65 proteins in rat tissues across different groups is shown in Figure 6A and B.

Figure 6 Protein Expression in Rat Tissues. (A) Western blot analysis of relevant proteins (IKK-β and NF-κB p65) in rat tissues. (B) Expression levels of IKK-β across groups. (C) Expression levels of NF-κB p65 across groups. “#” indicates a statistically significant difference compared to the control group (P < 0.05); “*” indicates a statistically significant difference compared to the model group (P < 0.05).

Compared to the control group, the model group exhibited significantly increased levels of both IKK-β and NF-κB p65 (P < 0.05). Among the treatment groups, only the TSD group showed no significant difference in NF-κB p65 expression compared to the model group (P > 0.05). Notably, the YJD group showed significant differences in the expression of both proteins compared to the TSD group and displayed no statistical difference from the healthy control group (P > 0.05).

RT-PCR Analysis

The primers and nucleotide sequences used in this experiment are listed in Table 2.

Table 2 Primers and Nucleotide Sequences

The mRNA expression levels in rat tissues for each group are displayed in Figure 7.

Figure 7 mRNA Expression in Rat Tissues. “#” indicates a statistically significant difference compared to the control group (P < 0.05); “*” indicates a statistically significant difference compared to the model group (P < 0.05).

Compared to the control group, the model group exhibited significant differences in mRNA expression for IKK, IKB, NLRP3, and caspase-1. In the YJD group, significant differences were observed in IKK, NLRP3, and caspase-1 expression compared to the model group, except for IKB (P < 0.05). The PQT and TSD groups did not show significant differences in any of the four markers compared to the model group. Among the treatment groups, a significant difference in caspase-1 expression was observed between the YJD and TSD groups.

ELISA Analysis

The ELISA detection results for IL-1β, IL-6, IL-18, and TNF-α are presented in Figure 8.

Figure 8 ELISA Results. “#” indicates a statistically significant difference compared to the control group (P < 0.05); “*” indicates a statistically significant difference compared to the model group (P < 0.05).

Compared to the control group, the model group showed significant increases in IL-1β, IL-6, IL-18, and TNF-α levels. In the TSD group, significant differences were observed in IL-1β, IL-18, and TNF-α, while the PQT group showed a significant difference only in TNF-α. However, compared to the model group, the TSD group did not exhibit significant differences in IL-1β, IL-18, or TNF-α levels. Among the treatment groups, a significant difference in IL-18 levels was found between the YJD and TSD groups.

Discussion

Previous studies indicate that LDH typically begins at an average age of 40 to 45 years^31–33 According to the TCM classic “Shanggu Tianzhen Lun”, it is stated, “At six sevens (age 42), the three Yang meridians decline in women, the face becomes haggard, and hair begins to turn white”, and “At five eights (age 40), the kidney Qi declines in men, causing hair loss and teeth to wither”. This indicates that the age range of 40–45 already shows a gradual weakening of physical functions, especially in women. This was strongly supported by the fact that in the study of 353 patients in 2020³⁴ and 149 patients in 2023,³⁵ over 55% of the female patients were in this age range. Combining these points from a TCM perspective, most LDH patients face not only pain caused by Qi and blood stasis but also physical depletion. Therefore, Dr. Yang Lixue, a renowned TCM doctor in Shaanxi Province, has, through many years of clinical practice, used TSD as a base formula to promote blood circulation. He modified the formula by adding herbs that nourish Yin and replenish blood according to the characteristics of the patient group, resulting in a new formulation, YJD, which is more suitable for the LDH population and specifically targets LDH symptoms. The positive control drug selected in this study, PQT, is an anti-inflammatory and analgesic drug that has been extensively studied^36,37 and has long been used in clinical practice.

The experimental results indicate that the formulas YJD, PQT, and TSD show significant similarity in GO and KEGG pathway analyses, with all three formulas modulating LDH through mechanisms related to oxidative stress and inflammation. Oxidative stress arises from excessive production of reactive oxygen species (ROS) or an insufficient antioxidant defense. Excessive ROS activates the NF-κB and MAPK signaling pathways, disrupting the balance between degradation and synthesis of the extracellular matrix in intervertebral disc cells, while also promoting the secretion of pro-inflammatory factors. These alterations lead to the loss of intervertebral disc cells and establish a persistent inflammatory microenvironment, further degrading disc tissue³⁸. In addition, the expression of inflammatory mediators such as IL-1β and TNF-α is strongly associated with disc degeneration and herniation.^2,39 According to the GO and KEGG enrichment analyses, several pathways were highly enriched, including the PI3K-Akt signaling pathway, MAPK signaling pathway, and NF-κB signaling pathway. These pathways are considered key areas for further investigation due to their relevance to LDH pathology.

The results from the three assays show that after activation, IKK-β (as detected by Western Blot, WB) becomes phosphorylated and degrades IKB (RT-PCR), allowing NF-κB p65 (WB) to be released and translocated into the nucleus, where it activates the expression of downstream inflammatory genes, including IL-1β (ELISA), IL-6 (ELISA), and TNF-α (ELISA)⁴⁰. NF-κB p65 also promotes the expression of the NLRP3 gene (RT-PCR), which increases its assembly into inflammasomes and activates caspase-1, leading to the maturation and release of pro-inflammatory cytokines IL-1β (ELISA) and IL-18 (ELISA) into the extracellular space^41,42. Additionally, TNF-α and IL-6 can feedback to further activate NF-κB p65, forming a continuous positive feedback loop.⁴⁰ At the upstream level of this signaling cascade, the detection of IKK and IKK-β revealed that YJD showed results comparable to the control group. In the IKK-β assay, YJD showed statistically significant differences from TSD while being on par with PQT. IKB degradation was minimal in the YJD group, closely resembling the control group. Similarly, the NF-κB p65 results paralleled those of IKK-β, with YJD showing the best performance among the three groups: YJD > TSD and YJD ≈ PQT. In subsequent experiments involving NLRP3, caspase-1, inflammatory factors, and H&E staining, YJD continued to show results similar to the control group and comparable to PQT, with a significant advantage over TSD. Furthermore, the results from thermal hyperalgesia experiments showed that YJD was closest to the control group among the three treatment groups. These findings align with the initial computational predictions, confirming the algorithm’s validity in animal experiments.

The algorithm proposed in this study is based on a PPI network, which is composed of the interactions between proteins. These interactions play critical roles in various biological processes and are fundamental for understanding disease mechanisms and developing new therapies.^43,44 Previous research in network pharmacology often relies on PPI networks to model drug-disease interactions,⁴⁵ using network topology parameters to assess the importance of each node. Building upon these prior studies, our research integrates multiple network topology parameters to create a comprehensive scoring system. In this system, the dosage of each drug is progressively passed to the target proteins, allowing for an importance-weighted score. Despite these advancements, this study still faces limitations, particularly in the idealization of the model, as constructing a precise database of drug active ingredient output is a substantial challenge. To address this, we revised our approach by independently decomposing the weights and converting them into frequencies, which allowed us to create a more refined model. Moreover, we externalized all the weights in the algorithm, enabling researchers to adjust the importance of specific parameters or components based on their quantitative results and understanding. This flexibility helps achieve predictions that better align with clinical realities. Rather than introducing a completely new evaluation system, this study functions as a numerical aggregation of multiple dimensions. The core evaluation framework remains based on the traditional, well-established system, with the innovation being the expanded data range. By distinguishing between two key aspects—identifying which targets are most critical for treating LDH and determining which drug formulas can best address those targets—we were able to aggregate these insights into a more comprehensive predictive model.

When all drug dosages are set at standard clinical levels, YJD and PQT demonstrated a clear advantage over TSD in the calculations. However, the variation in the number of compounds remains an important confounding factor. While YJD and TSD have a similar number of ingredients, YJD’s total dosage is nearly twice that of TSD. In contrast, PQT dosages only about one-fortieth of TSD, yet contains five times as many different compounds. In traditional network pharmacology, an increase in the number of compounds generally expands the range of potential therapeutic targets, leading to greater overlap with disease-related targets. Although impractical in real-world applications, from a purely theoretical network pharmacology standpoint, taking all herbal medicines at once could potentially activate numerous unnecessary targets, but it would bring the treatment closer to activating all the targets associated with the disease. However, activating targets related to conditions such as cancer while treating LDH could have catastrophic consequences in clinical practice. For this reason, the number of compounds in a treatment should be carefully limited to activate only the most critical targets while minimizing unnecessary ones. In this study, YJD targeted significantly more LDH-related sites than TSD, approaching the efficiency of PQT. However, the difference in the number of ingredients between PQT and YJD—29 compared to 9—suggests that YJD may be near the optimal efficiency point for treating LDH with TCM. This principle, known in TCM as “fewer medicines, stronger effects”, refers to achieving maximum therapeutic results with the minimum number of ingredients.

Combining the results of animal experiments with our group’s previous research, it is suggested that while TSD theoretically has some therapeutic potential for treating LDH,⁴⁶ its actual effectiveness is quite limited⁴⁷. Long-term clinical observations also indicate that patients tend to show a stronger preference and greater willingness to cooperate with YJD and PQT treatments, rather than with TSD. Given these observations, we have reason to believe that, despite the inherent challenges of network pharmacology—such as issues with data updates and synergistic effects—placing it at some distance from direct clinical application, the computational results offer valuable insights as an initial screening strategy. These results help clarify certain aspects of clinical efficacy, allowing for theoretical simulations to compare the potential effects of different treatment formulas.

Conclusion

This study introduced dosage as a new factor into network pharmacology research, and the results preliminarily demonstrated the potential of using a comprehensive scoring algorithm to compare different formulas at standard clinical doses. This approach offers valuable insights for the future design and evaluation of treatment formulas. However, due to the inherent limitations of network pharmacology and current detection technologies, dosage control in clinical practice still requires significant reliance on physicians’ expertise and extensive data. In the future, the development of a specialized database could further enhance our algorithm, enabling the expansion of computer-aided drug design (CADD) in the field of targeted treatment.