Network Analysis of the Herb-Drug Interactions of Citrus Herbs Inspired by the "Grapefruit Juice Effect".

This study was performed to investigate the herb-drug interactions (HDIs) of citrus herbs (CHs), which was inspired by the "grapefruit (GF) juice effect". Based on network analysis, a total of 249 components in GF and 159 compounds in CHs exhibited great potential as active ingredients. Moreover, 360 GF-related genes, 422 CH-related genes, and 111 genes associated with drug transport and metabolism were collected, while 25 and 26 overlapping genes were identified. In compound-target networks, the degrees of naringenin, isopimpinellin, apigenin, sinensetin, and isoimperatorin were higher, and the results of protein-protein interaction indicated the hub role of UGT1A1 and CYP3A4. Conventional drugs such as erlotinib, nilotinib, tamoxifen, theophylline, venlafaxine, and verapamil were associated with GF and CHs via multiple drug transporters and drug-metabolizing enzymes. Remarkably, GF and CHs shared 48 potential active compounds, among which naringenin, tangeretin, nobiletin, and apigenin possessed more interactions with targets. Drug metabolism by cytochrome P450 stood out in the mutual mechanism of GF and CHs. Molecular docking was utilized to elevate the protein-ligand binding potential of naringenin, tangeretin, nobiletin, and apigenin with UGT1A1 and CYP3A4. Furthermore, in vitro experiments demonstrated their regulating effect. Overall, this approach provided predictions on the HDIs of CHs, and they were tentatively verified through molecular docking and cell tests. Moreover, there is a demand for clinical and experimental evidence to support the prediction.

Introduction

As traditional herbal medicines are popularly applied, herb–drug interactions (HDIs) have become a rising concern in the clinical use of conventional drugs. Complicated chemical compositions and potential multiple bioactivities are associated with complex HDIs. Based on different interaction pathways, HDIs can be classified into pharmacokinetic interactions and pharmacodynamic interactions, and the former ones were the focus of past studies, which concentrated on drug transport and metabolism. HDIs may end up quite differently: on the one hand, they affect drug levels and/or activities and therefore potentially cause therapeutic failure or adverse reactions; on the other hand, some HDIs lead to beneficial clinical effects including heightened efficacy and lessened side effects.[1]

The consumption of grapefruit (GF), a citrus fruit, has been found to have potential health benefits such as antioxidation and anti-inflammation activities, lipid metabolism improvement, neuroprotection, and body weight regulation.[2] Nevertheless, it is associated with interactions with certain drugs including calcium channel blockers, immunosuppressants, and antihistamines, which is a well-known food–drug interaction named the “GF juice effect”.[3] According to the principles of pharmaphylogeny and phytochemistry, plants that belong to the same family and genus tend to have similar chemical compositions.[4] Especially, the distribution of coumarins and furanocoumarins that are related to the GF juice effect in citrus species closely matches citrus phylogeny.[5] Given this, citrus herbs (CHs), common components in quantities of herbal formulae, probably share various active compounds with GF. CHs are most frequently used for qi-regulating based on the theory of traditional Chinese medicine, and the most common ones include Chenpi (Pericarpium Citri Reticulatae, PCR), Qingpi (Pericarpium Citri Reticulatae Viride, PCRV), Zhike (Fructus Aurantii, FA), Zhishi (Fructus Aurantii Immaturus, FAI), Juhong (Exocarpium Citri Rubrum, ECR), Huajuhong (Exocarpium Citri Grandis, ECG), Xiangyuan (Fructus Citri, FC), and Foshou (Fructus Citri Sarcodactylis, FCS). Owing to their wide pharmacological effects on the cardiovascular, digestive, and respiratory systems, they have been used commonly in clinical practice to treat diseases involving multiple systems.[6] However, they have the potential to induce HDIs alike the GF juice effect due to their similar chemical compositions. It was once reported that Fructus Citrus maxima induced a 1.5-fold increase in the blood level of tacrolimus in a renal transplant patient.[7]

Network pharmacology is a paradigm shift in pharmaceutical discovery, which is hopeful of deciphering the drug mechanism with a holistic perspective. The research paradigm has shifted from the “one drug for one target” mode to a “multiple components for network targets” mode.[8] Apparently, the principles of network pharmacology are applicable to the phenomena of the GF juice effect and HDIs, involving multiple components and targets. To predict complicated HDIs of CHs, network analysis was employed in this research. Molecular docking and in vitro experiments were also conducted to reveal associations between compounds and targets.

Results

Network Study on the “GF Juice Effect” and HDIs of CHs

A total of 405 and 250 components, respectively, in GF and CHs were identified. Through ADME screening, 249 components in GF and 159 ingredients in CHs exhibited great potential as active compounds, which are listed in Table S1. As for targets, a total of 360 genes related to 141 compounds in GF were collected via public databases for target prediction, while 422 genes were linked to 125 compounds in CHs.

Through the retrieval from the NCBI Gene database, 111 genes associated with drug transport and metabolism were obtained. The results of the Venn diagram suggested that 25 and 26 overlapping genes were identified by matching 360 GF-related genes and 422 CHs-related genes, as well as the above genes related to drug transport and metabolism (Figure S1). The majority of GF-related (18/25) and CHs-related (18/26) targets associated with drug transport and metabolism were identical.

The interactions between compounds and targets related to drug transport and metabolism were visualized. As shown in Figure A, the compound–target network of GF consisted of 51 nodes (26 compounds and 25 targets) and 62 interacting edges. Naringenin, methoxsalen, isopimpinellin, methyl salicylate, sinensetin, apigenin, and isoimperatorin had higher degree values, suggesting that they might affect more drug transporters and drug-metabolizing enzymes. Besides, the compound–target network of CHs included 34 compounds’ and 26 targets’ nodes and 75 edges, which suggested that key compounds with bioactivities including naringenin, hydroxyacetone, isopimpinellin, apigenin, sinensetin, and isoimperatorin were more likely to play a part in drug transport and metabolism (Figure B). GF and CHs shared the majority of key compounds, among which naringenin possessed the highest degree value. The more chemical information is summarized in Table . Pharmacokinetic prediction results of the compounds via SwissADME are specified in Table S2. The distribution of some typical overlapping CH compounds was retrieved from literature reports. The maximum content was visualized via a chord diagram (Figure ), while the content (range) is given in Table S3.

Figure 1

GF (A) and CH (B) compound–target networks related to drug transport and metabolism.

Table 1

GF and CH Compounds Associated with Drug Transport and Metabolism

GF compounds associated with drug transport and metabolism				CH compounds associated with drug transport and metabolism
ID	PubChem CID	chemical name	molecular formula	ID	PubChem CID	chemical name	molecular formula	CH sources
GF1	4133	methyl salicylate	C8H8O3	CH1	5280443	apigenin	C15H10O5	FAI, ECR, ECG
GF2	798	indole	C8H7N	CH2	637566	geraniol	C10H18O	ECR, ECG, FC
GF3	998	phenylacetaldehyde	C8H8O	CH3	3314	eugenol	C10H12O2	FAI, ECG
GF4	8794	phenylacetonitrile	C8H7N	CH4	42607889	5,7,3′-trihydroxy-5′-methoxyflavanone	C16H14O6	PCRV, FAI, ECR, ECG, FA, PCR
GF5	6054	2-phenylethanol	C8H10O	CH5	145659	sinensetin	C20H20O7	PCRV, FAI, ECG, PCR
GF6	2758	eucalyptol	C10H18O	CH6	72281	hesperetin	C16H14O6	FA, FC, ECG
GF7	637566	geraniol	C10H18O	CH7	5281426	umbelliferone	C9H6O3	ECR, ECG
GF8	638011	citral	C10H16O	CH8	179651	limonin	C26H30O8	FAI, ECR, FC, FA, PCR
GF9	5280934	linolenic acid	C18H30O2	CH9	439246	naringenin	C15H12O5	PCR, PCRV, FAI, FA, ECR, ECG
GF10	145659	sinensetin	C20H20O7	CH10	676152	5,7-dihydroxy-2-(3-hydroxy-4-methoxyphenyl)chroman-4-one	C16H14O6	PCR, PCRV, FAI
GF11	5280443	apigenin	C15H10O5	CH11	68077	tangeretin	C20H20O7	PCR, PCRV, FAI, ECG, FA
GF12	5280371	bergaptol	C11H6O4	CH12	72344	nobiletin	C21H22O8	PCR, PCRV, FAI, FA, ECG
GF13	72344	nobiletin	C21H22O8	CH13	1149877	(2R)-5,7-dihydroxy-2-(4-methoxyphenyl)-2,3-dihydrochromen-4-one	C16H14O5	FAI, ECG, FC
GF14	439246	naringenin	C15H12O5	ECG1	7362	furfural	C5H4O2	ECG
GF15	68077	tangeretin	C20H20O7	ECG2	938	nicotinic acid	C6H5NO2	ECG
GF16	68079	isopimpinellin	C13H10O5	ECR1	667495	(2R)-5,7-dihydroxy-2-(4-hydroxyphenyl)-2,3-dihydro-4H-chromen-4-one	C15H12O5	ECR
GF17	68081	isoimperatorin	C16H14O4	ECR2	68081	isoimperatorin	C16H14O4	EC
GF18	6989	thymol	C10H14O	FAI1	5280445	luteolin	C15H10O6	FAI
GF19	16724	d-carvone	C10H14O	FAI2	10212	imperatorin	C16H14O4	FAI
GF20	5280460	scopoletin	C10H8O4	FAI3	373261	eriodyctiol (flavanone)	C15H12O6	FAI
GF21	5281426	umbelliferone	C9H6O3	FAI4	68079	isopimpinellin	C13H10O5	FAI
GF22	1017	phthalic acid	C8H6O4	FAI5	631170	3′,4′,5,7-tetramethoxyflavone	C19H18O6	FAI
GF23	4114	methoxsalen	C12H8O4	FC1	638011	citral	C10H16O	FC
GF24	72281	hesperetin	C16H14O6	FCS1	5280460	scopoletin	C10H8O4	FCS
GF25	15512	3-cyclohexene-1-methanol	C7H12O	FCS2	8468	vanillic acid	C8H8O4	FCS
GF26	150893	3,3′,4′,5,6,7,8-heptamethoxyflavone	C22H24O9	FCS3	259994	(R)-propane-1,2-diol	C3H8O2	FCS
				FCS4	5281612	diosmetin	C16H12O6	FCS
				FCS5	8299	hydroxyacetone	C3H6O2	FCS
				FCS6	637497	(R)-butane-1,3-diol	C4H10O2	FCS
				FCS7	220010	meso-2,3-butanediol	C4H10O2	FCS
				PCR1	6782	diisobutyl phthalate	C16H22O4	PCR
				PCR2	1183	vanillin	C8H8O3	PCR
				PCR3	6781	diethyl phthalate	C12H14O4	PCR
				PCR4	150893	3,3′,4′,5,6,7,8-heptamethoxyflavone	C22H24O9	PCR

Figure 2

Content distribution (maximum) of typical overlapping compounds in CHs. Note: Capi, apigenin; Ches, hesperetin; Clim, limonin; Cnar, naringenin; Cneo, neohesperidin; Cnob, nobiletin; Csin, sinensetin; Ctan, tangeretin.

Protein–protein interaction (PPI) networks of GF and CH targets related to drug transport and metabolism are depicted in Figure . The topological parameters of each target are detailed in Table S4. A total of 24 nodes and 142 edges were involved in the GF PPI network, of which the top seven targets included UGT1A1, CYP3A4, UGT1A7, NR1I2, CYP2B6, CYP1A1, and CYP3A5. All of the targets and their related drugs (examples) are listed in Table S5. The CH PPI network included 25 nodes and 170 edges, in which UGT1A1, CYP3A4, CYP1A1, CYP2B6, CYP1A2, CYP2E1, UGT1A7, and GSTP1 were considered as hub targets, as listed in Table S5 together with their related drugs (examples). Pharmacokinetic pathways of some typical related drugs are displayed in Figure S2. The importance of 18 mutual genes differed in the PPI network of GF and CHs, except for the top two targets, namely, UGT1A1 and CYP3A4.

Figure 3

PPI networks of GF (A) and CH (B) targets related to drug transport and metabolism.

Remarkably, multiple chemical components might interact with various drugs via diverse targets. Some key compounds served as examples below. Caffeine, clopidogrel, etoposide, tamoxifen, theophylline, and verapamil were linked to naringenin with the aid of ABCB1, CBR1, CES1, CYP1A2, CYP1B1, CYP19A1, GSTP1, and UGT1A1. Isopimpinellin might affect CYP1A1, CYP1B1, CYP3A4, GSTP1, and NR1I2, and therefore influence the pharmacokinetics of drugs such as erlotinib, nilotinib, paclitaxel, tamoxifen, and verapamil. Apigenin was associated with the regulation of ABCG2, CYP1B1, CYP19A1, and UGT1A1 and might interact with etoposide and tamoxifen consequently. Potential interactions of sinensetin with caffeine and theophylline were probably mediated by ABCG2, CYP1A1, CYP1A2, and CYP1B1. Isoimperatorin might influence the metabolism of caffeine, tamoxifen, theophylline, and venlafaxine, which was possibly related to its effect on CYP1A2, CYP1B1, CYP2B6, and CYP2D6.

The mutual compounds of GF and CHs were identified with the aid of chemical structures. GF and CHs shared 48 potential active compounds (Table ). As shown in Figure A, 43 compounds were related to 262 target genes, which formed 650 interaction edges. In the network, naringenin, tangeretin, nobiletin, and apigenin possessed more connections with targets.

Table 2

Mutual Compounds of GF and CHsa

ID	PubChem CID	chemical name	molecular formula	CH sources
C1	5280443	apigenin	C15H10O5	FAI, ECR, ECG
C2	5280460	scopoletin	C10H8O4	FCS
C3	985	palmitic acid	C16H32O2	ECG, FCS
C4	8175	decanal	C10H20O	PCR, FC
C5	637566	geraniol	C10H18O	ECR, ECG, FC
C6	638011	citral	C10H16O	FC
C7	443158	(−)-linalool	C10H18O	PCR, PCRV, ECR, ECG, FC
C8	3893	lauric acid	C12H24O2	PCR, FCS
C9	1549025	neryl acetate	C12H20O2	PCR
C10	8748	β-terpineol	C10H18O	PCR
C11	643820	nerol	C10H18O	FCS
C12	5325830	(−)-terpinen-4-ol	C10H18O	PCR, FCS
C13	637531	pichtosin	C12H20O2	FCS
C14	11005	myristic acid	C14H28O2	ECG, FCS
C15	13849	pentadecanoic acid	C15H30O2	ECG
C16	145659	sinensetin	C20H20O7	PCRV, FAI, ECG, PCR
C17	68081	isoimperatorin	C16H14O4	ECR
C18	2355	bergapten	C12H8O4	FAI, ECR, ECG
C19	8417	scoparone	C11H10O4	FCS
C20	1742210	caryophyllene oxide	C15H24O	ECG
C21	8892	hexanoic acid	C6H12O2	FCS
C22	72281	hesperetin	C16H14O6	FA, FC, ECG
C23	7793	(−)-citronellol	C10H20O	FC
C24	5281426	umbelliferone	C9H6O3	ECR, ECG
C25	6428300	trans-(+)-pyranoid linalool oxide	C10H18O2	PCR
C26	8158	nonanoic acid	C9H18O2	ECG
C27	68079	isopimpinellin	C13H10O5	FAI
C28	6826	methyl 2-(methylamino)benzoate	C9H11NO2	PCR
C29	8635	methyl anthranilate	C8H9NO2	ECG
C30	439246	naringenin	C15H12O5	PCR, PCRV, FAI, FA, ECR, ECG
C31	2775	5,7-dimethoxycoumarin	C11H10O4	FC, FCS
C32	443178	(+)-trans-carveol	C10H16O	FCS
C33	8186	undecanal	C11H22O	PCR
C34	150893	3,3′,4′,5,6,7,8-heptamethoxyflavone	C22H24O9	PCR
C35	68077	tangeretin	C20H20O7	PCR, PCRV, FAI, ECG, FA
C36	5281534	α-sinensal	C15H22O	PCR
C37	72344	nobiletin	C21H22O8	PCR, PCRV, FAI, FA, ECG
C38	9727	N-methyltyramine	C9H13NO	FAI, FA, FC
C39	5284503	3-hexen-1-ol	C6H12O	ECG
C40	629964	4′,5,7,8-tetramethoxyflavone	C19H18O6	FAI, ECG
C41	854067	(−)-synephrine	C9H13NO2	FA, FC
C42	6450230	marmin	C19H24O5	FA
C43	1550607	auraptene	C19H22O3	FAI
-	643779	neral	C10H16O	PCR, PCRV, ECR, ECG, FCS
-	1549026	geranyl acetate	C12H20O2	FC
-	454	octanal	C8H16O	PCR, FC
-	5283361	2-dodecenal	C12H22O	PCR
-	8174	1-decanol	C10H22O	PCR

Note: ID “-” means that no target genes related to the compound were obtained.

Figure 4

Network and enrichment analysis of mutual targets. (A) Mutual compound–target network of GF and CHs. (B) Gene Ontology (GO) enrichment analysis (biological process, BP) results of mutual targets. (C) GO enrichment analysis (molecular function, MF) results of mutual targets. (D) GO enrichment analysis (cellular component, CC) results of mutual targets. (E) KEGG pathway analysis results of mutual targets.

Gene Ontology (GO) enrichment analysis results of mutual targets are shown in Figure , which included biological processes (top 5%, Figure B), cellular components (top 10%, Figure C), and molecular functions (top 10%, Figure D). Biological processes (including response to drug, oxidation–reduction process, and negative regulation of the apoptotic process) cellular components (including the plasma membrane, integral component of the membrane, and cytosol), and molecular functions (including protein binding, enzyme binding, and protein homodimerization) manifested themselves. The results of the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis indicated that 285 mutual targets were significantly enriched in 149 signaling pathways (P < 0.05). The top 20 signaling pathways are shown in Figure E. Apart from the pathways of multiple diseases, the metabolism of xenobiotics by cytochrome P450 and cytochrome P450 drug metabolism demonstrated significance.

Molecular Docking of Mutual Compounds and Target Proteins

The details of docking results are listed in Table . The results showed that apigenin had the best docking effect with UGT1A1, while nobiletin had the greatest binding potential with CYP3A4. The three-dimensional (3D) diagrams of molecular docking models are displayed in Figure , which showed the binding mode and sites between each compound and protein.

Table 3

Docking Results of Mutual Compounds and Target Proteins

target protein	compound type	compound name	lowest binding energy/kcal·mol–1	inhibiting constant/μmol·L–1	number of hydrogen bonds	number of hydrophobic interactions
UGT1A1	inhibitor	dacomitinib	–6.61	14.32	1	5
	inducer	phenobarbital	–5.1	182.12	5	3
	tested compound	naringenin	–5.41	107.37	3	6
	tested compound	tangeretin	–4.87	268.93	1	3
	tested compound	nobiletin	–3.9	1380	3	3
	tested compound	apigenin	–5.48	96.34	3	2
CYP3A4	inhibitor	ketoconazole	–7.64	2.52	3	4
	inducer	phenobarbital	–6.81	10.21	3	4
	tested compound	naringenin	–6.92	8.53	6	3
	tested compound	tangeretin	–5.89	48.21	3	2
	tested compound	nobiletin	–7.05	6.78	5	1
	tested compound	apigenin	–6.86	9.42	5	2

Figure 5

Binding mode of mutual compounds and target proteins. (A) Dacomitinib and UGT1A1, (B) phenobarbital and UGT1A1, (C) naringenin and UGT1A1, (D) tangeretin and UGT1A1, (E) nobiletin and UGT1A1, (F) apigenin and UGT1A1, (G) ketoconazole and CYP3A4, (H) phenobarbital and CYP3A4, (I) naringenin and CYP3A4, (J) tangeretin and CYP3A4, (K) nobiletin and CYP3A4, and (L) apigenin and CYP3A4.

In Vitro Tests on the Potential Common Mechanism of the “GF Juice Effect” and HDIs of CHs

Concentration–cell viability curves of the four flavonoids via the CCK-8 assay are provided in Figure S3. Accordingly, the corresponding tested concentrations were selected: naringenin, 10 μmol·L–1; tangeretin, 10 μmol·L–1; nobiletin 20 μmol·L–1; and apigenin, 1 μmol·L–1. As shown in Figure , the UGT1A1 activity of HepG2 cells was significantly raised by naringenin and tangeretin, while nobiletin and apigenin demonstrated different degrees of ability to decrease it. Meanwhile, the western blot test indicated that the CYP3A4 expression of the cells tended to be reduced by the four compounds.

Figure 6

UGT1A1 activity (A) and CYP3A4 expression (B) of HepG2 cells after treatment with naringenin, tangeretin, nobiletin, and apigenin. n = 3, * P < 0.05, ** P < 0.01 versus control group.

Discussion

In this research, it was predicted that various drugs might be affected in pharmacokinetics by GF and CHs. In fact, there was a growing body of evidence for the “GF juice effect”. The interactions between GF and the majority of the predicted related drugs in this research were reported in clinical studies, and the clinical evidence mainly focused on pharmacokinetic interactions. In general, there were mainly two opposite situations. One was that GF can lessen plasma exposure to certain drugs such as etoposide, which may diminish curative effects.[3] The other condition was that plasma exposure to certain drugs was increased, which may enhance curative effects or trigger off side effects.[9] Psychotropic drugs (caffeine, midazolam), cardiovascular drugs (nifedipine, verapamil), immunosuppressive agents (cyclosporine), antineoplastic drugs (nilotinib), antiepileptic drugs (carbamazepine), antidepressant drugs (sertraline), analgesic drugs (methadone), and antifungal agents (itraconazole) were involved in such “GF juice effect”.[3] There also existed some experimental evidence for HDIs of CHs, which certainly fell far behind that for the “GF juice effect”. Pomelo peel, a source of CGE, was found to significantly increase the bioavailability of cyclosporine and tacrolimus in rats.[10] FAI decreased the bioavailability of tacrolimus in rats, while FA showed no remarkable effect.[11]

As for the mechanism, it was widely believed that GF influenced drug-metabolizing enzymes, including cytochrome P450 (CYP450) and UDP-glucuronosyltransferase (UGT), and drug transporters such as P-glycoprotein (P-gp) and organic anion transporting polypeptide (OATP) to disturb pharmacokinetics of certain drugs.[12−15] Given the similar material basis, CHs might be associated with regulating these targets. In this research, UGT1A1 and CYP3A4 were predicted to be hub genes of the “GF juice effect” and HDIs of CHs. Moreover, drug metabolism by CYP450 was also regarded as an important pathway. The regulating effects of the key compounds, which included naringenin, tangeretin, nobiletin, and apigenin, on UGT1A1 and CYP3A4 were also preliminarily observed in vitro. In fact, some recent studies focused on the capability of the compounds to regulate the drug-metabolizing enzymes, although the action tendency might be reversed due to different experimental systems and conditions.[15−19] As widely known, CYP450 played a pivotal role in drug metabolism, since the majority of hepatically cleared drugs depended on CYP450 for metabolism.[20] GF inhibited intestinal and hepatic CYP3A4 in an exposure-dependent fashion, and patients taking CYP3A4 substrates are at risk of developing drug-related adverse events if consuming large amounts of GF.[21] The GF-induced pharmacokinetic change of multiple drugs might be better explained by the impairment of CYP450 in the intestinal wall rather than in the liver.[22,23] In rats, FA upregulated the protein expression of CYP1A2, CYP3A4, and CYP2E1 and the mRNA expression of CYP1A2 and CYP3A4, which indicated that it might be a slight inducer of CYP1A2 and CYP3A4.[24] Besides, the extracts of FA and FAI showed mild inhibition on CYP3A.[25] UGTs were a series of enzymes responsible for conjugative phase II reactions of drugs.[26] A cross-sectional study suggested that the intake of citrus fruits including GF might promote UGT1A1 activity among women with a certain genotype.[27] Transporters including P-gp and OATP were often expressed in tissues related to drug disposition concerning intestinal absorption, uptake into hepatocytes, and renal/bile excretion of drugs.[28] GF showed bidirectional effects on P-gp in rats. On the one hand, it inhibited P-gp-mediated drug efflux in cotreatment, but on the other hand, its chronic administration led to increased levels of P-gp expression.[29] A study indicated that FAI and PCR extracts increased P-gp and CYP3A4 expression via upregulation of the pregnane X receptor in vitro.[30] GF inhibited human enteric OATP1A2 in vitro.[31] The interaction of GF with fexofenadine only at therapeutic concentrations might be better explained as the presence of multiple binding sites on OATP2B1.[32]

There had been some studies paying attention to the exposure dose and duration of the “GF juice effect”. A human study confirmed that GF–dextromethorphan pharmacokinetic interaction was dose-dependent and indicated 200 mL of single-strength GF juice as the “lowest observed effect level”.[33] It was observed that the recovery of GF-induced enteric CYP3A impairment was largely completed within 3 days, consistent with enzyme regeneration after mechanism-based inhibition.[22] The dose–effect relationship and recovery time for HDIs of CHs remained to be explored further.

In regard to vital compositions in these reactions, multiple furocoumarins and flavonoids showed great potential to cause these interactions. Certain furocoumarins in GF exhibited as strong inhibitors of CYP3A4.[34] Bergamottin was observed to equally increase the absorption of nifedipine in rats by comparison with GF, suggesting that bergamottin played a vital role in GF–nifedipine interaction.[35] In vitro, 6′,7′-dihydroxybergamottin was verified as a potent mechanism-based inhibitor of midazolam α-hydroxylation by CYP3A.[22] These two major furanocoumarins in GF differed in intestinal CYP3A4 inhibition kinetic and binding properties. With human intestinal microsomes, 6′,7′-dihydroxybergamottin was a substrate-independent reversible and mechanism-based inhibitor of CYP3A4. In contrast, bergamottin was a substrate-dependent reversible inhibitor but a substrate-independent mechanism-based inhibitor.[36] Last but not least, the role of GF–CH mutual compounds could not be ignored in these interactions. In vitro naringin was a potent competitive inhibitor of caffeine 3-demethylation dependent on CYP1A2.[37] Potent inhibition of CYP3A4 and negligible inhibitory effects on P-gp and other CYP450 by limonin was observed in vitro.[38] Hesperetin and naringenin exhibited strong inhibition on UGT1A1, while UGT1A7 was moderately inhibited.[15] Molecular docking analysis identified favorable binding of nobiletin with the transmembrane region site 1 of homology modeled human ABCB1 transporter, while in vitro experiment demonstrated that nobiletin profoundly inhibited ABCB1 transporter activity.[39]

This research had some limitations as below, whereas our findings interpreted that some drugs were ultimately connected with CHs by means of network pharmacology. First, CHs and GF differed in usage and dosage. GF juice was more popular, and several hundred milliliters of it might be consumed as a drink habitually. In contrast, CHs were usually used with other herbs in the form of decoctions, and the daily dosage was merely several grams. Second, it was reported that furanocoumarins played a crucial role in the “GF juice effect”, and they did not seem so important in the prediction yet. It might be affected by the online databases for component collection and target prediction. Third, this research was specifically aimed at pharmacokinetic interactions instead of pharmacodynamic ones. Moreover, it should be noted that multiple active compounds in CHs varied in content according to diverse factors including species, places of production, collecting time, storage, processing, and preparation, which was bound to have a significant impact on HDIs. The content of GF–CH mutual compounds also remained to be further determined and compared in light of the close connection between dosage and effect.

Conclusions

In the current study, based on network pharmacology, the potential HDIs of CHs were predicted compared with GF, in which diverse conventional drugs were associated with multiple components and targets. Furthermore, molecular docking and in vitro experiments demonstrated that the regulating effects of flavonoids including naringenin, tangeretin, nobiletin, and apigenin on UGT1A1 and CYP3A4 might play a crucial part in HDIs of CHs. Besides, it would benefit from more evidence of clinical practice and scientific experiments in the future.

Materials and Methods

Network Study on the “GF Juice Effect” and HDIs of CHs

Chemical Composition Collection and Screening of GF and CHs

The Plant Chemical Component Database (http://www.organchem.csdb.cn/scdb/main/plant_introduce.asp), which belonged to the Chemistry Database provided by Shanghai Institute of Organic Chemistry,[40] and the Traditional Chinese Medicine Systems Pharmacology Database and Analysis Platform (TCMSP, https://tcmspw.com/tcmsp.php) were, respectively, employed to identify the chemical components of GF and CHs including ECG, ECR, FA, FAI, FC, FCS, PCR, and PCRV.[41] The canonical SMILES and molecular formulas of all involved components were obtained from PubChem at https://pubchem.ncbi.nlm.nih.gov/ with the aid of chemical names or structures. PubChem is a public database for chemical structures of small molecules, which is conducive to chemical structure standardization.[42]

To obtain compounds with higher oral bioavailability, all components were screened via SwissADME (http://www.swissadme.ch/) with the aid of canonical SMILES. SwissADME is a tool to evaluate the pharmacokinetics and drug-likeness of small molecules.[43] The majority of oral components exhibited bioactivities when absorbed by the gastrointestinal tract, and drug-likeness was a qualitative concept designed to estimate solubility and permeability.[44] Given this, components were selected as potential active compounds when the results of predicted gastrointestinal absorption were high and the output of five drug-likenesses filters (Lipinski, Ghose, Veber, Egan, and Muegge) contained no less than two “Yes”.

Collection of Potential Targets

The targets corresponding to the potential active compounds of GF and CHs were acquired from TCMSP. Meanwhile, the canonical SMILES of each compound was imported into SwissTargetPrediction (http://www.swisstargetprediction.ch/) and STITCH (http://stitch.embl.de/), and the species was confined as “Homo sapiens”, while the required probability was screened as no less than 0.700 to predict the targets of compounds.[45] With regard to STITCH prediction, the compound with the highest Tanimoto score, usually 1.000 (match via InChIKey), was chosen, and the target would be included when its score was no lower than 0.700.[46] Then the targets related to the compounds of GF and CHs were ultimately converted into official gene symbols after being retrieved from UniProt (https://www.uniprot.org/) with the species selected as H. sapiens, while the duplicate targets and the compounds with vacant targets were removed ultimately.

Genes related to drug transport and metabolism were collected in the NCBI Gene database (https://www.ncbi.nlm.nih.gov/gene/) with keywords “drug transport”, “drug transmembrane transport”, “drug metabolism”, “drug metabolic”, and organism selected as H. sapiens. Venny 2.1.0 (http://bioinfogp.cnb.csic.es/tools/venny/index.html) was used to analyze the intersections of GF and CH targets with targets related to drug transport and metabolism.[47]

Compound–Target Network Construction and Analysis

Based on the data previously collected, compound–target networks associated with drug transport and metabolism were illustrated via Cytoscape 3.7.1, a network visualization and analysis software.[48] Degree values provided by NetworkAnalyzer, a tool of Cytoscape, were adopted for the importance of compound nodes in the network. In addition, taking multiple compounds shared by CHs into consideration, the content of typical overlapping CH compounds was retrieved from literature reports to compare and contrast the herbs. The compound content (range) was collected, and the maximum content was visualized via Circos Table Viewer v0.63–9 (http://mkweb.bcgsc.ca/tableviewer/visualize/).[49] Furthermore, since GF and CHs shared diverse chemical components, the mutual compounds were checked and the compound–target network was also constructed to seek their common composition that might play a significant role.

Protein–Protein Interaction Analysis

STRING ver.11.0 (https://string-db.org/) was utilized to conduct PPI for the selected targets, which was an online database providing associations between proteins based on curated databases, experimentally determined, gene neighborhood, gene fusions, gene co-occurrence, text mining coexpression, and protein homology. In the process, the species was limited to H. sapiens, and the minimum required interaction score was selected as medium confidence (0.400).[50] Disconnected nodes in the PPI network were hidden. The visualization of PPI was optimized through Cytoscape. Afterward, three topological parameters including degree, between centrality, and close centrality of each target were analyzed to describe influential nodes in the PPI network.

Collection of Related Drugs and Their Pathways

To provide information on the potential interactions of GF and CHs with certain drugs, DrugBank (https://www.drugbank.ca/), a database that combined detailed drug data with comprehensive drug target information, was adopted to gain related drugs of target genes.[51] Moreover, pathways of the drug above were collected via the Pharmacogenomics Knowledge Base (https://www.pharmgkb.org/), a resource that collected, curated, and disseminated information concerning human genetic variation on drug responses.[52]

Gene Ontology and Pathway Enrichment Analysis of Mutual Targets

To interpret the potential mechanism of the mutual targets linked to GF and CHs, Database for Annotation, Visualization, and Integrated Discovery (https://david.ncifcrf.gov/) 6.8 was utilized for GO and KEGG pathway enrichment analysis with the species setting of H. sapiens.[53,54] GO analysis included the aspects of the biological process (BP), molecular function (MF), and cellular component (CC). The P-value was calculated in the enrichment analyses, and P < 0.05 indicated that the enrichment degree was statistically significant. The results were visualized as a bubble chart via OmicShare tools (http://omicshare.com/).

Molecular Docking Simulations

Simulations were conducted between the top two target proteins and four typical mutual compounds of CHs and GF. The 3D structure of the target proteins was obtained from the Protein Data Bank (https://www.rcsb.org/) and the AlphaFold Protein Structure Database (https://alphafold.ebi.ac.uk/).[55,56] Target protein preparation including removing the solvent molecules and the original ligands was performed through PyMOL, while the structure of the compounds was collected from PubChem.[42] Then, molecular docking was performed with AutoDock 4.2.6 software. A total of 50 independent docking runs were conducted for each compound and target protein. The best docking model with the lowest binding energy was selected as the optimal model and used to demonstrate the binding mode and sites. The Protein–Ligand Interaction Profiler (https://plip-tool.biotec.tu-dresden.de/plip-web/plip/index) was utilized to analyze the docking results,[57] and visualization was completed with PyMOL. Meanwhile, a definite inhibitor and an inducer of the targets were chosen to be docked for a comparison.

In Vitro Tests

Cell Culture

Human hepatoma-derived HepG2 cells, generously provided by Prof. Jian Ni (Beijing University of Chinese Medicine, Beijing, China), were cultivated in Dulbecco’s modified Eagle’s medium (DMEM) (Gibco) supplemented with 10% fetal bovine serum (Analysis Quiz, China) and 1% penicillin–streptomycin (Gibco) at 37 °C in a 5% CO2 atmosphere. In vitro experiments were performed using HepG2 cells between passages 15 and 20, which were subcultured at approximately 80% confluence.

Cell Viability Assay

HepG2 cells (4000 per well) were cultured into 96-well plates for 24 h, and they were exposed to a series of concentrations of naringenin, nobiletin, apigenin (Shanghai Yuanye Bio-Technology Co., Ltd, China), and tangeretin (Shanghai Standard Technology Co., Ltd., China) for another 24 h. A cell counting kit-8 (CCK-8) solution (Biorigin Inc., China) assay was conducted to screen the safety concentrations of the four compounds on HepG2 cells. The concentration–cell viability curves were drawn, and half of the value of the concentration when cell viability reached 90% was considered as the tested concentration, at which cell viability was expected to be over 85% in another assay.

Enzyme-Linked Immunosorbent Assay of UGT1A1

The concentration of UGT1A1 in HepG2 cells was determined using human UGT1A1 enzyme-linked immunosorbent assay (ELISA) kits (Jiangsu Meibiao Biotechnology Co., Ltd, China) according to the manufacturer’s instructions, which took advantage of specific antigen–antibody reactions. Meanwhile, the protein concentrations of the samples were obtained using a BCA Protein Assay Kit (Beijing Solarbio Science & Technology Co., Ltd., China). The UGT1A1 concentration per protein concentration was ultimately calculated for normalization by dividing the UGT1A1 concentration by the total protein concentration.

Western Blot Analysis of CYP3A4

The proteins of HepG2 cells from different groups were harvested and then lysed with cold RIPA buffer (Beijing Solarbio Technology Co., Ltd, China) supplemented with a protease inhibitor cocktail for 20 min on ice. The protein concentrations of the supernatant were measured with a BCA Protein Assay Kit. Next, equal amounts (10 μg) of the protein were separated via precast 10% sodium dodecyl sulfate (SDS)-polyacrylamide gel and transferred onto poly(vinylidene difluoride) (PVDF) membranes (Millipore Inc.). After blocking with TBST buffer containing 5% skim milk for 1 h at room temperature, the PVDF membranes were incubated overnight at 4 °C with the CYP3A4 antibody (1:6000, Proteintech Group, Inc.), followed by incubation with the appropriate secondary antibody at room temperature for another 1 h. At last, the blots were visualized by SageCapture software (Beijing Sage Creation Science Company, China), the levels of protein expression were normalized to that of GAPDH, and relative protein expression was quantified using ImageJ software (National Institutes of Health).

Statistical Analysis

The statistical analysis of data as mean ± standard deviation was performed by utilizing SPSS software (version 26.0, International Business Machines Corporation). According to the normality- and variance-related data of each group, the one-way analysis of variance (ANOVA) and the Kruskal–Wallis test were applied to indicate a significant difference, which was associated with the P-value < 0.05.