Hypertensive nephropathy is a common complication of hypertension. Traditional Chinese medicine has been used in the clinical treatment of hypertensive nephropathy for a long time, but the commonly used prescriptions have not been summarized, and the basic therapeutic approaches have not been discussed. Based on data from 3 years of electronic medical records of traditional Chinese medicine used at the Affiliated Hospital of Shandong University of Traditional Chinese Medicine, a complex network and machine learning algorithm was used to explore the prescribed herbs of traditional Chinese medicine in the treatment of hypertensive nephropathy (HN). In this study, complex network algorithms were used to describe traditional Chinese medicine prescriptions for HN treatment. The Apriori algorithm was used to analyze the compatibility of these treatments with modern medicine. Data on the targets and regulatory genes related to hypertensive nephropathy and the herbs that affect their expression were obtained from public databases, and then, the signaling pathways enriched with these genes were identified on the basis of their participation in biological processes. A clustering algorithm was used to analyze the therapeutic pathways at multiple levels. A total of 1499 prescriptions of traditional Chinese medicines used for the treatment of hypertensive renal damage were identified. Fourteen herbs used to treat hypertensive nephropathy act through different biological pathways: huangqi, danshen, dangshen, fuling, baizhu, danggui, chenpi, banxia, gancao, qumai, cheqianzi, ezhu, qianshi, and niuxi. We found the formulae of these herbs and observed that they could downregulate the expression of inflammatory cytokines such as TNF, IL1B, and IL6 and the NF-
HN caused by hypertension commonly damages the kidneys [
Glomerulosclerosis is the most common pathological change associated with hypertension. Long-term hypertension leads to thickening of the renal artery intima, hyaline changes in afferent glomerular arterioles and interlobar arterioles, increased renal vascular resistance, and induced inflammatory reactions and renal tubulointerstitial fibrosis at the same time [
In HN cases, complementary therapies can be good choices for timely intervention and protection against renal damage in the early stage of HN. As a supplementary means of modern medicine, traditional Chinese medicine (TCM) has received increasing attention, and it is one of the most widely used complementary therapies in the world. TCM has clear advantages in the treatment of HN [
In this study, complex networks were used to collect relevant information about TCM prescriptions and disease symptoms from electronic medical records (EMRs) and biological databases, which were used to analyze the prescription network as generated by doctors using herbs in the real world, discuss the topological characteristics of the network, and find the core nodes and connection characteristics of the network, which can be used to mitigate the difficulty posed by the diverse clinical manifestations of this disease. This approach allows the rapid and accurate identification of the common herb combinations used for HN and other diseases and specific symptoms by making efficient use of real-world data [
Currently, specific prescriptions for the treatment of HN are lacking. Under real-world conditions, the advantages and characteristics of syndrome differentiation for use of TCM treatment can be fully implemented [
We collected the EMRs of 30695 anonymous patients diagnosed with hypertension at the Affiliated Hospital of Shandong University of Traditional Chinese Medicine between July 1, 2014, and May 31, 2017, including the diagnosis, demographic characteristics (such as age and sex), chief complaint recorded, and formulae in the TCM prescription. We extracted data on 2055 patients diagnosed with HN and used common terms to identify the signs and symptoms of the HN patients on the basis of the chief recorded complaint and manually referred to the NLM Medical Subject Heading (MeSH) database. Combined with its medical history, other related diseases that can lead to hypertension and CKD were deleted, and the quality was controlled by 2 doctors with over 10 years’ experience in the treatment of cardiovascular and renal diseases.
We extracted a list of herbs from the formulae of HN patients and set up the herb network with herbs as nodes and the occurrence frequency of the use of two herbs in different formulae as weights. Then, we ran the hierarchical network extraction algorithm to extract herb pairs from the weighted herb network on the basis of the degree coefficient
We used the relative risk (RR) to find highly sensitive herbs between HN patient formulae and non-HN patient formulae. We used the HN patient formulae as the exposure group and the non-HN patient formulae as the nonexposure group, and single herbs used in each group were the outcomes.
We analyzed the compatibility of the herbs by using the HN prescription of association rules, which are often used to find the conditional dependencies between tuples. We used the classical Apriori algorithm [
Observing the symptoms of the disease is an important link between disease diagnosis and treatment with TCM. Under the guidance of TCM theory, different herbs can be aimed at different TCM symptoms. The SymMap database contains the TCM symptoms corresponding to herbal medicine that have been agreed upon by experts. The TCM symptoms for which these herbs were used to treat HN were retrieved one by one in the SymMap database.
We identified compounds and targets of the herbs for the treatment of HN from online databases: Traditional Chinese Medicine Systems Pharmacology Database and Analysis Platform (TCMSP) [
The HN gene expression data were retrieved from the NCBI GEO database and analyzed by the GEO2R online analysis tool. The GSE99325 data set contained 20 HN samples and 4 control samples. The genes with
In addition, we used disease terms in several established databases of human disease-related genes, including Online Mendelian Inheritance in Man (OMIM) [
The human biological functions cannot be realized by individual genes but require ubiquitous interactions between different genes. In this study, we combined the compounds and targets of various herbs with the PPIN we had established to create a biological network. The key compounds in the biological network were calculated by restarting the random walk (RWR) algorithm to measure the closeness between each node and seed node. We used HN-related targets as seed nodes, and the restart probability was 0.75 [
Using the
Through GO and KEGG enrichment analyses, the direction of each herb’s biological function was determined. To systematically evaluate the similarities and differences of each herb’s biological function, we carried out hierarchical clustering according to the number of enriched genes in each KEGG signaling pathway profile and the GO terms for each herb. Moreover, we carried out hierarchical clustering of the collected herbal medicine-related symptoms and targets.
Hierarchical clustering is an unsupervised machine learning method. Initially, a single sample was regarded as a class, and the distance between each class was calculated. The data were merged according to specific rules.
Initially, we regarded a single TCM symptom, target, or biological process as a class, calculated the ESS of each pair of classes after merging, determined the combination with the least increase in total ESS after merging, and then iterated the process until all objects formed a large class with the
We selected 1449 prescriptions of traditional Chinese medicine from the EMR data of 2055 patients with HN, and these prescriptions included 425 herbs used 35734 times, and each prescription contained, on average, 23.84 ± 3.88 herbs. The core traditional Chinese medicines used in the treatment prescriptions for HN were obtained by hierarchical extraction algorithm: Radix Astragali (huangqi), Radix Salviae liguliobae (danshen), Radix Codonopsis pilosulae (dangshen), Poria (fuling), Rhizoma Atractylodis macrocephalae (baizhu), Radix Angelicae sinensis (danggui), Pericarpium Citri Reticulatae (chenpi), Rhizoma Pinelliae (banxia), and Radix Glycyrrhizae (gancao). After comparing the RR values, herba Dianthi (qumai), Semen Plantaginis (cheqianzi), Rhizoma Curcumae (ezhu), Semen Euryales (qianshi), and Radix Achyranthis bidentatae (niuxi) were determined to be specific traditional Chinese medicines used for the treatment of HN.
We obtained 179 differentially expressed genes related to HN from the GSE99325 data set. As shown in Figure
Differentially expressed genes in HN: a blue dot indicates a downregulated gene, and a red dot indicates an upregulated gene.
We found 41 important association rules, as shown in Figure
Main rules in the Apriori algorithm: the size of the dot indicates the effect of the herbs, and the darker the color of the dot, the larger the lift.
Ten rules for maximum lift values.
LHS | RHS | Support | Confidence | Coverage | Lift | Count | |
---|---|---|---|---|---|---|---|
Chenpi | => | Banxia | 0.258 | 0.701 | 0.368 | 1.939 | 387 |
Banxia | => | Chenpi | 0.258 | 0.714 | 0.362 | 1.939 | 387 |
Chenpi | => | Fuling | 0.294 | 0.799 | 0.368 | 1.337 | 441 |
Baizhu | => | Fuling | 0.261 | 0.795 | 0.328 | 1.330 | 391 |
Chuanxiong | => | Danshen | 0.203 | 0.540 | 0.376 | 1.301 | 304 |
Danggui | => | Huangqi | 0.388 | 0.758 | 0.511 | 1.285 | 581 |
Huangqi | => | Danggui | 0.388 | 0.656 | 0.590 | 1.285 | 581 |
Banxia | => | Fuling | 0.273 | 0.755 | 0.362 | 1.262 | 409 |
Banxia | => | Dangshen | 0.201 | 0.555 | 0.362 | 1.256 | 301 |
Dangshen | => | Huangqi | 0.324 | 0.733 | 0.442 | 1.242 | 486 |
Fourteen kinds of traditional Chinese herbs are used in the clinical treatment of HN, according to their compatibility characteristics, targets, signal pathways, and other information, as determined using MCODE, hierarchical clustering, and other methods for comparative analysis to determine the potential biological mechanism of herbal medicines in the treatment of HN.
Using the SymMap database, we found a total of 68 effective TCM symptoms treated by the 14 herbs, and we established a 14 × 68-dimensional symptom profile in which an herbal medicine that was directed toward a symptom was given a value of 1; otherwise, it was given a value of 0. Then, we used the Ward clustering algorithm for hierarchical clustering in which the similarity between herbs was measured by Euclidean distance. Figure
Multilevel comparison of the 14 traditional Chinese herbs. (a) Heat map and hierarchical clustering in target level. (b) Heat map and hierarchical clustering in target level. (c) and (d) Heat map and hierarchical clustering are biological signal pathway results at KEGG signal pathways and GO terms.
After introduction into the STRING database, a total of 58 targets for the 14 herbs were retained, and we established a 14 × 58-dimensional target profile for hierarchical clustering. Figure
We analyzed the pathway enrichment of various herbs and HN-related genes and retained the pathways in which the number of enriched genes for each herb treatment in a significantly enriched HN pathway was greater than the quartile of enriched genes expressed upon herbal treatment, and these retained pathways were considered the core pathways for hierarchical clustering. Figures
Network of enriched terms represented as pie charts, where pie pieces are color-coded based on the identities of the gene in the herbs and HN. The thicker the line is, the more common the targets of the nodes and the closer their interaction.
The overlap between the herbs and HN: includes the shared term level, where blue curves link genes that to which the same enriched ontology term is attributed. The inner circle represents gene lists, where hits are arranged along the arc. Genes that were hits in multiple lists are colored in dark orange, and genes unique to a list are shown in light orange.
The table shows that the 14 herbs could be divided into 3 or 4 groups in terms of TCM symptoms, targets, GO enrichment terms, and KEGG enrichment terms (Table
Clusters of the herbs at different levels.
Levels | TCM symptom | Target | GO term | KEGG signal pathway | |
---|---|---|---|---|---|
Clusters | 1 | Fuling | Gancao, cheqianzi, niuxi, huangqi | Danggui, baizhu, danshen, fuling, qumai, qianshi, banxia | Danshen, chenpi, banxia |
2 | Danggui | qianshi, banxia, danshen, fuling | Ezhu, chenpi | Qumai, ezhu, danggui, fuing | |
3 | Qumai, gancao, chenpi, baizhu, ezhu, danshen, banxia, qianshi, niuxi | Qumai, chenpi, dangui, dangshen, ezhu, baizhu | Dangshen, cheqianzi, niuxi, gancao | Qianshi, baizhu | |
4 | Huanqi, dangshen | — | Huangqi | Niuxi, cheqianzi, huangqi, gancao |
In terms of TCM symptoms and according to TCM theory, the 14 herbs can be aimed at HN or symptoms related to essential kidney deficiency, such as edema, diarrhea, fatigue, sore waist and knees, fatigue, and diarrhea. Symptoms such as turbid urination, oliguria, and obvious decline of renal function can be specifically treated by fuling.
According to the RWR algorithm, compounds with values in the upper quartile of the RWR were retained as the core effective compounds. Ultimately, 241 effective core compounds were retained, of which the ten compounds with the highest
The top 10 compounds with the highest
PubChem CID | Compound | Formula | RWR |
---|---|---|---|
5257127 | 2-Azaniumylacetate | C2H5NO2 | 1.82E − 04 |
5280343 | Quercetin | C15H10O7 | 1.40E − 04 |
5281708 | Daidzein | C15H10O4 | 1.01E − 04 |
5280460 | Scopoletin | C10H8O4 | 7.49E − 05 |
4276 | Myristicin | C11H12O3 | 7.46E − 05 |
5280443 | Apigenin | C15H10O5 | 6.45E − 05 |
177 | Acetaldehyde | C2H4O | 6.03E − 05 |
7043901 | (2S,3S)-2-Ammonio-3-methylpentanoate | C6H13NO2 | 5.87E − 05 |
5280863 | Kaempferol | C15H10O6 | 5.44E − 05 |
5280489 | Beta-carotene | C40H56 | 4.30E − 05 |
Studies have shown that a variety of traditional Chinese medicine prescriptions exhibit clear efficacy in the treatment of HN [
From the perspective of target and pathway enrichment, 14 herbs were involved with 25 core targets, and CHRM2, ADRB2, CXCL8, GCG, TNF, and CXCL10 were the most frequently identified targets. These herbs, particularly danshen, niuxi, cheqianzi, huangqi, and gancao, could act on inflammatory factors, which can be significantly enriched in inflammatory-related pathways such as the HIF-1 signaling pathway, TNF signaling pathway, NOD-like receptor signaling pathway, and NF-
HN is one of the main causes of chronic nephropathy. Arteriosclerosis and hyaline degeneration caused by hypertension are the main pathological changes in HN and are closely related to inflammation and tubulointerstitial fibrosis [
The herbs obtained by data mining in this study can inhibit kidney injury with Ang II. The targets of dangfui, niuxi, chenqianzi, huangqi, gancao, dangshen, chenpi, and banxia are significantly related to cAMP signaling pathway, mTOR signaling pathway, and cGMP-PKG signaling pathway. These herbs can inhibit the activity of CHRM1 receptor and GCG on the cell membrane and reduce the activation of cAMP signaling pathway and mTOR signaling pathway, thus inhibiting NF-
HN is also a long-term cause of nephron damage. A previous study showed that the pathological renal changes caused by HN were progressive, with glomerular damage appearing first and renal tubular atrophy and interstitial fibrosis appearing as late manifestations in an SHR model [
The therapeutic effects of these herbs on HN are mainly the deceleration of the renal damage caused by hypertension, protection of nephrons and podocytes, reduction in blood pressure and the amount of protein in urine, and slowing renal interstitial fibrosis and maintaining nephron function, but the effect of a single herbal medicine is often limited. To achieve the expected therapeutic effects, these herbs need to be combined. These herbs are often used in the clinic. This finding is consistent with the results in this study, showing that these herbs fall under the important compatibility rules identified by the Apriori algorithm. To some extent, the results of the multilevel analysis indicated the possible effective pathway of these traditional Chinese medicine combinations, which is consistent with the mechanism of renal injury caused by HN, but this study also has many limitations. First, the compound information, targets, and disease-related genes were all identified in open databases, and there is still a lack of experimental studies on proteomics and metabonomics. Second, the effects of these herbs on the NF-
Based on the traditional Chinese medicine prescription data of HN treated in the Affiliated Hospital of Shandong University of Traditional Chinese Medicine from 2014 to 2017, traditional Chinese medicine for HN treatment was found by using a complex network algorithm and calculating the relative risk. Fourteen herbs were compared and analyzed at multiple levels with a clustering algorithm. Through the clustering results obtained for different levels of analysis, these herbs were found to play multilevel roles in the biological regulation of HN. This study also showed that the use of a data mining algorithm can be used to summarize the TCM treatment prescription of HN accurately and with high reliability and can provide a direction for clinical practice and future experimental research.
Hypertensive nephropathy
Chronic kidney disease
The International Society of Nephrology Kidney Disease Data Center
Odds ratio
Confidence interval
The African-American Study of Kidney Disease and Hypertension
The Ramipril Efficacy in Nephropathy
End-stage renal disease
The Modification of Diet in Renal Disease Study
Angiotensin-converting-enzyme
Microalbuminuria
Renin-angiotensin-aldosterone system
Hazard ratio
Traditional Chinese medicine
Electronic medical record
The NLM Medical Subject Heading
Relative risk
Traditional Chinese Medicine Systems Pharmacology Database and Analysis Platform
Encyclopedia of Traditional Chinese Medicine
Protein-protein interaction network
Online Mendelian Inheritance in Man
Restarting the random walk
The error sum of squares
Radix astragali
Radix salviae liguliobae
Radix codonopsis pilosulae
Poria
Rhizoma atractylodis macrocephalae
Radix angelicae sinensis
Pericarpium citri reticulatae
Rhizoma pinelliae
Radix glycyrrhizae
Herba dianthi
Rhizoma curcumae
Semen euryales
Radix achyranthis bidentatae
Rhizoma imperatae
Rehmannia glutinosa
Ophiopogon japonicus
Coptidis rhizoma.
The TCM prescription data used to support the findings of this study are available from the corresponding author upon request.
The authors declare that there are no conflicts of interest regarding the publication of this article.
The authors thank the Affiliated Hospital of Shandong University of Traditional Chinese Medicine for its support of this study. This work was supported by National Key Research and Development Program (No. 2017YFC1703506): Traditional Chinese Medicine Big Data Mining Research and Innovative Application, for which Jian Yu was in charge; the Jinan Science and Technology Bureau Clinical Medical Science and Technology Innovation Program (No. 202019149), based on network pharmacology analysis of the mechanism of real-world core prescriptions for hypertensive renal damage, for which Yi-Fei Wang was in charge; National Key Research and Development Program (2017YFC1703502): Research and Implementation of Traditional Chinese Medicine Clinical Research Information Sharing System, for which Jie Yang was in charge; and the Construction Project of National Clinical Research Base of traditional Chinese Medicine for Hypertension (Issued by the State Administration of Traditional Chinese Medicine [2008] (No. 23)).
Apriori: 41 important association rules of the 14 herbs obtained for the Apriori algorithm. Core target of herbs: the core target of 14 herbs. cRWR of effective compounds: