Adult male Wistar rats, weighing 120–150 g, were provided by SPF (Beijing, China) Biotechnology Co., Ltd. All experimental procedures were approved by the National Institutes of Health and the Center for Laboratory Animals of Shanghai Mental Health Center and performed according to the guidelines of the National Research Council for the care and use of laboratory animals. The animals were kept in specific pathogen-free (SPF) conditions in a 12 h/12-h light/dark cycle at 22 °C–24 °C and 50%–60% relative humidity. All rats had free access to food and water. Every effort was made to reduce suffering and the number of rats used. Metformin (CAS:657-24-9, Mecklin, Shanghai, China) was dissolved in artificial cerebrospinal fluid (ACSF; 1.2 mM MgCl${}_2$, 234 mM sucrose, 25 mM NaHCO${}_{3,}$ 12 mM glucose, 1.2 mM NaH${}_2$PO${}_4$, 2.5 mM CaCl${}_2$, and 3.6 mM KCl) as a stock solution of 100 µM. Rats were randomly divided into six groups to assign to different treatments - the sham group, sham–30 µg of metformin group, sham–100 µg of metformin group, SNL group, SNL–30 µg of metformin group, or SNL–100 µg of metformin group.

For intrathecal catheterization, polyethylene intrathecal catheters (PE-10: OD, 0.61 mm; ID, 0.28 mm, Advanced Biotechnologies Inc, South Dennis, MA, USA) were implanted into rats after inhalation of isoflurane anesthesia (R510-22, RWD, Shenzhen, Guangdong, China, 4%–5% induction, 1.5% maintenance). The catheter was inserted near the L5 spinal segmental level into the L5/L6 intervertebral subarachnoid space. Rats recovering from surgery were housed in individual cages for 14 days. Those that exhibited immediate bilateral hind-limb paralysis were chosen for the study following 10 µL of 4% lidocaine (CAS:137-58-6, Sigma, St Louis, MO, USA) injections. For intrathecal metformin administration, 10 µL of the metformin solution was injected slowly and then flushed with 15 µL of ACSF.

The SNL model was established based on Kim and Chung for neuropathic rats [20]. In brief, the rat was deeply anesthetized by inhalation. The L5 and L6 spinal nerves were isolated carefully and ligatured tightly with 4-0 silk thread. The wound was sutured after complete hemostasis was confirmed. The procedure for the sham group was identical to that of the SNL group, but ligation was not performed. Rats for subsequent experimental use were screened based on obvious unilateral allodynia in response to mechanical stimuli and the absence of motor deficits.

Rats were exposed to the testing environment for at least 30 min. Rats were placed in a plexiglass box on a metal grid to assess mechanical allodynia. The behavior test was carried out by using a 2290 CE electrical von Frey hair (IITC Life Science, Woodland Hills, CA, USA), ranging from 0.1 g to 90 g. The experimenter was blinded to the different treatment groups. The paw withdrawal thresholds (PWT) were defined as the force value when the rat’s contralateral or ipsilateral hind limb suddenly withdrew after being subjected to an increased upward force. The threshold was calculated by averaging the lowest force that evoked a withdrawal response over a 1 min interval [21, 22].

For thermal pain hypersensitivity tests, rats were kept in the testing apparatus for 30 min on a 30 °C heated glass floor. Thermal withdrawal latency was measured by a plantar test analgesic meter (II-390G, IITC Life Science, Woodland Hills, CA, USA). Each rat was tested over 3 sessions, each 5 min in length. To stimulate skin sensitization and tissue damage, the cutoff time was confined to 30 s. The average paw withdrawal latency (PWL) of the three tests was used for data analysis.

Before testing, rats were placed in the plexiglass chamber for 30 min to acclimatize to the experimental environment. Rats were gently restrained and injected subcutaneously with 50 µL of 5% formalin (221008, Yulu, Shangrao, Jiangxi, China) diluted in normal saline in the left hind paw. Rats were then immediately placed into the chamber and nociceptive behavior was observed with a mirror embedded in the rear part of the cage for better monitoring of animal behavior. Formalin-induced flinches were considered pain behavior and manually quantified by the experimenter for 1 min at 10 min intervals until 90 min; the experimenter was blind to the treatments [23]. Formalin-induced flinching was biphasic: phase I was mediated by noxious chemical stimulation immediately after formalin injection and lasted from 0–10 min, whereas phase II was mediated by a combination of peripheral inputs and formalin-induced central sensitization and began at 10 min and lasted until 90 min after formalin injection [24, 25, 26].

After the mechanical allodynia behavioral test, an electrophysiological test was performed on spinal cord slices obtained from rats at 14 days post surgery. In brief, the rats were deeply anesthetized via isoflurane inhalation. The lumber enlargements of L3–L5 were quickly separated and immersed for 90 s in cold high-sucrose ACSF with a mixture of 95% O${}_2$ and 5% CO${}_2$ containing (in mM) 1.2 mM MgCl${}_2$, 234 mM sucrose, 25 mM NaHCO${}_{3,}$ 12 mM glucose, 1.2 mM NaH${}_2$PO${}_4$, 2.5 mM CaCl${}_2$, and 3.6 mM KCl. Thereafter, 300 µm-thick slices were transversely sectioned using a vibratome (VT1200 S, Leica, Wetzlar, Germany). The spinal cord sections were incubated in ACSF containing (in mM) 1.25 mM NaH${}_2$PO${}_4$, 125 mM NaCl, 26 mM NaHCO${}_3$, 3 mM KCl, 1mM MgCl${}_2$, 10 mM D-glucose, and 2 mM CaCl${}_2$. The pH was adjusted to 7.3, and the fluid was filled with a mixture of 95% O${}_2$ and 5% CO${}_2$ for 30 min at 32 °C prior to electrophysiology recordings.

To record glutamatergic transmission, a bath application of 0.5 µM tetrodotoxin (TTX; Aladdin, Shanghai, China), 1 µM strychnine (Sigma, St Louis, MO, USA), and 100 µM PTX (Sigma, St Louis, MO, USA) was used to inhibit sodium channels, glycine, and GABA receptors. The recording was conducted using a borosilicate glass recording electrode (BF150-86-10, Sutter, Sacramento, CA, USA), which was filled with a K${}^{+}$-based internal solution (composition in mM: 10 HEPES, 130 K-gluconate, 2 NaCl, 2 MgCl${}_2$, 5 KCl, 1.1 EGTA, 5 QX314, 10 sodium phosphocreatine, 4 Na${}_2$-ATP, 1 MgSO${}_4$, 0.4 Na-GTP; the pH was adjusted to pH 7.2 with KOH). To record the inhibitory transmission, the electrode was filled with a high-Cl solution (Formula: 10 mM HEPES, 1.1 mM EGTA, 2 mM MgCl${}_2$, 140 mM CsCl, 0.3 mM Tris-guanosine triphosphate and 3 mM MgATP; the pH was adjusted to pH7.4 with CsOH), and was clamped at –70 mV during bath application of TTX (0.5 µM), 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX; Sigma, St Louis, MO, USA) (20 µM), and D-AP5 (50 µM) to block kainate, N-methyl-D-aspartate receptor (NMDAR) activity, and amino-3-hydroxy-5-methyl-4-isoxazole propionic acid receptor (AMPAR), respectively.

Isoflurane inhalation was used to induce deep anesthesia. About 100 mL of normal saline was perfused intracardially, followed by 100 mL of 4% paraformaldehyde (PFA) (w/v) (BL539A, Biosharp, Hefei, Anhui, China) in phosphate-buffered saline (PBS, pH 7.4). Spinal lumbar enlargements were immediately removed and post-fixed for 12 h in the same fixative at 4 °C before being dehydrated in sucrose solutions dissolved in PBS, with concentrations ranging from 10% to 30%. Tissue was mounted in an optimal cutting temperature (OCT) compound and sliced into 30 µm frozen sections using a Leica Microsystems (CM1950, Leica, Wetzlar, Germany). The frozen sections were incubated in 10% goat serum (v/v) (Sigma, St Louis, MO, USA) and 0.5% Triton X-100 (v/v) (CAS: 9002-93-1, Sangon, Shanghai, China) in PBS for 1 h at room temperature before being incubated with the rabbit anti-c-Fos (1:1000; 226008, SYSY, Gottingen, Germany) and mouse anti-NeuN antibodies (1:300; ab128886, Abcam, Boston, MA, USA) for an additional 24 hours at 4 °C. The sections were washed three times in PBST for 5 min each and then incubated with the appropriate secondary antibodies (conjugated with Alexa Fluor 488 or 594; 1:1000; A-10631, A-21205, Invitrogen, Carlsbad, CA, USA) for 2 h at room temperature. Finally, the sections were mounted on slides before being dried, cleared, and cover slipped. Immunofluorescence images were captured with a confocal scanning laser microscope (Fluo View FV1000, Olympus, Tokyo, Japan).

Spinal cord tissue was extracted and homogenized with a Radio-immunoprecipitation assay (RIPA) lysis buffer (Beyotime, Shanghai, China) to determine the expression level of pain-related proteins. Protein concentrations were measured with a BCA protein detection kit (Beyotime, Shanghai, China). The proteins were separated on 8–20% Precast-Gels (Beyotime, Shanghai, China), followed by electro-transference to polyvinylidene difluoride membrane (PVDF) (Sigma, St Louis, MO, USA) membranes. The membranes were then blocked with 5% skim milk for 1 h at room temperature. After being blocked, the membranes were incubated with primary antibodies for c-Fos (1:1000, ab190289, Abcam, Cambridge, UK) and glyceraldehyde 3-phosphate dehydrogenase (GAPDH; 1:1000, Cat No. 60004-1-Ig, Proteintech, Chicago, IL, USA) at 4 °C overnight. Subsequently, they were incubated with corresponding HRP-conjugated secondary antibody (1:1000, 31402, ThermoFisher, Waltham, MA, USA) for 1 h at room temperature. Proteins were detected and quantified using the iBright FL1000 Imaging System (Invitrogen, Carlsbad, CA, USA).

The lumbar enlargements of the spinal dorsal were quickly separated for RNA extraction after rats were sacrificed by decapitation. Total RNA was extracted with Trizol reagent (9108, Takara, Shiga, Japan) and converted into cDNA using the PrimeScript RT reagent kit (RR036A, Takara, Shiga, Japan). The specific primer sequences used for cDNA amplification were as follows: NOS1, forward 5${}^{\prime }$-ATGGAAGAGAACACGTTTGGGGTTC-3${}^{\prime }$ and reverse 5${}^{\prime }$-CTCCCACTTTGCGTTTGAAGA-3${}^{\prime }$, NOS2, forward 5${}^{\prime }$-ATGGCTTGCCCCTGGAAGTTTCTCT-3${}^{\prime }$, reverse 5${}^{\prime }$-CTCCACGTTGTTGTTAATGTCCT-3${}^{\prime }$, c-Fos, forward 5${}^{\prime }$-AGGCAGAGCTGGAGTCGGAGAT-3${}^{\prime }$, reverse 5${}^{\prime }$-GCCGAGGACTTGAACTTCACTCG-3${}^{\prime }$, GAPDH, forward 5${}^{\prime }$-CCAAGGTCATCCATGACGAC-3${}^{\prime }$, reverse 5${}^{\prime }$-TCCACAGTCTTCTGAGTGGC-3${}^{\prime }$. qPCR analysis was performed by SYBR-Green Supermix and a LightCycler 480 system (04887352001, Roche Molecular Diagnostics, Alameda, CA, USA). The ratio of mRNA expression relative to GAPDH was determined by the 2${}^{-\mathrm{\Delta }\mathrm{\Delta }\text{CT}}$ method.

First, we searched for potential targets of metformin. The PubChem database (https://pubchem.ncbi.nlm.nih.gov) contains chemical information such as molecular formula, chemical structure, biological activities, and other identifiers. The SwissTargetPrediction database (http://www.swisstargetprediction.ch) was employed to predict the most possible protein targets of small molecules. We uploaded the structure of metformin from PubChem to acquire its putative targets. Second, we searched the potential pathogenic targets on the GeneCards database (https://www.genecards.org), Online Mendelian Inheritance in Man (OMIM) database (https://www.omim.org), DisGeNET database (https://www.disgenet.org), and Therapeutic Target Database (TTD) database (https://db.idrblab.net/ttd) using “pain” as the keyword. A pain potential target databank was created after deleting duplicate targets. The targets of metformin were compared with the pain databank to obtain the intersection targets. These intersection targets were the potential targets of metformin for pain treatment. Third, the overlapped targets were displayed as a Venn figure (Venny 2.1.0, CNB, Madrid, Spain), and the protein-protein interaction (PPI) network of targets shared by metformin and pain was acquired from the STRING database (https://string-db.org/, version 11.5). The PPI network was uploaded to Cytoscape 3.9.1 (https://cytoscape.org/) to search hub targets according to the topological features “Betweenness centrality”, “closeness centrality”, and “degree” [27, 28]. Finally, to clarify the biological function of shared targets between metformin and pain, an enrichment analysis of pathways from the Kyoto Encyclopedia of Genes and Genomes (KEGG) was performed with “clusterProfier” package in the RStudio 4.2.2 (https://rstudio.com) [29].

The data were expressed as means $\pm$ standard error of the mean (SEM). Two-tailed and unpaired Student t-test, and one-way or repeated measures two-way analysis of variance (ANOVA) were used to generate the statistical significance values using GraphPad Prism 6 (version 6.0, GraphPad Software Inc., San Diego, CA, USA). The post-hoc Student-Newman-Keuls test was performed when the effect of the drug dose was statistically significant. Probability values were two-tailed, and the statistical significance criterion p value was 0.05.

To assess the analgesic effects following metformin treatments, we applied an SNL and formalin-induced hypersensitivity model. An intrathecal injection of 30 or 100 µg (dissolved in 10 µL of ACSF) of metformin was administered to two groups of neuropathic pain rats. The PWT to mechanical or thermal stimuli was measured before and after injections at 1, 2, and 4 h. Intrathecal injections of 30 or 100 µg of metformin and ACSF did not change contralateral hind paws in sham rats over 4 h. In contrast, metformin via intrathecal delivery prolonged the mechanical and thermal response in ipsilateral hind paws in a dose- and time-dependent manner (Fig. 1A, mechanical allodynia F${}_{(5,96)}$ = 509.4, p 0.0001; two-way repeated measures analysis of variance (ANOVA) followed by Sidak’s post-tests, Fig. 1B, thermal hyperalgesia: F${}_{(5,96)}$ = 118.1, p 0.0001; two-way repeated measures ANOVA followed by Sidak’s post-tests).

Fig. 1.

Intrathecal administration of metformin attenuated neuropathic- and formalin-induced pain. (A,B) Intrathecal delivery of metformin exhibited mechanically antiallodynic and thermal hyperalgesia effects in SNL-induced neuropathic pain rats. (C) Intrathecal delivery of metformin significantly reduced flinch counts and exhibited analgesic effects in formalin-induced inflammatory pain rats (n = 5–6 per group). *p 0.05, by two-tailed Student t-test or repeated measures two-way ANOVA followed by Sidak’s post-tests. (D,E) Consecutive intraperitoneal metformin suppressed mechanically antiallodynic and thermal hyperalgesia effects, mechanical allodynia: F${}_{(2,60)}$ = 273.1, p 0.0001; two-way repeated measures ANOVA followed by Sidak’s post-tests, thermal hyperalgesia: F${}_{(2,60)}$ = 28.74, p 0.0001; *p 0.05, two-way repeated measures ANOVA followed by Sidak’s post-tests, n = 5. Data were presented as means $\pm$ SEM. SNL, spinal nerve ligation; ANOVA, analysis of variance; SEM, standard error of the mean.

Furthermore, we assessed the antinociceptive effects of metformin in formalin-induced acute inflammatory pain. The rats displayed flinching behaviors following 5% formalin injection into the left hind paws. These responses were separated into phase I and phase II, and counted with the frequencies of flinching behaviors (Fig. 1C, F${}_{(2,120)}$ = 164.9, p 0.0001). The results showed that metformin did not significantly affect the number of flinching behaviors in phase I, but intrathecal injection of 30 or 100 µg of metformin inhibited the formalin-induced flinching response in phase II and peaked at around 30–40 min. Therefore, metformin produced analgesic effects in SNL and formalin-induced hypersensitivity. Consecutive intraperitoneal injections of 100 mg of metformin also suppressed nerve injury-associated hypersensitivity in rats (Fig. 1D, mechanical allodynia F${}_{(2,60)}$ = 273.1, p 0.0001; two-way repeated measures ANOVA followed by Sidak’s post-tests, Fig. 1E, thermal hyperalgesia: F${}_{(2,60)}$ = 28.74, p 0.0001; two-way repeated measures ANOVA followed by Sidak’s post-tests).

Network pharmacology was used to determine the underlying mechanism mediated by metformin treatment to further assess the underlying target and signaling pathways. Through the online Venn diagram tool, we found 25 shared targets for metformin to treat pain (Fig. 2A). The targets were imported into the STRING database to construct and identify the PPI network and then visualized by Cytoscape. Subsequently, we collected the hub targets, including nitric oxide synthase 1 (NOS1), NOS2, NOS3, epidermal growth factor receptor (EGFR), and plasminogen (PLG), by the Centiscape 2.2 (https://apps.cytoscape.org/apps/centiscape) plugin in Cytoscape based on three topological features: degree, betweenness centrality, and closeness centrality (Fig. 2B). Gene Ontology (GO) and KEGG pathway enrichment analyses of the 25 shared targets were performed using RStudio. The top 10 remarkably enriched biological processes (BP), molecular functions (MF), cellular components (CC), and KEGG signaling pathways involved in the shared targets were identified. The BP of the targets included postsynaptic signal transduction, regulation of hemostasis, and biosynthetic processes (Fig. 2C). The MF comprised neurotransmitter receptor activity and binding, mitogen-activated protein (MAP) kinase kinase kinase activity, and calcium-dependent protein binding (Fig. 2D). The CC consisted of synaptic cleft, neuronal cell body and spine, dendritic spine, and membrane raft (Fig. 2E). The underlying target genes of metformin in pain were specifically and predominantly enriched in the Relaxin signaling pathway, the cyclic guanosine monophosphate (cGMP)-protein kinase G (PKG) signaling pathway, neuroactive ligand-receptor interaction, calcium signaling pathways, neuroactive ligand-receptor interaction, and cholinergic synapses (Fig. 2F). These findings suggest that the main bioactive components of metformin had different action targets in synaptic transmission pathways.

Fig. 2.

GO and KEGG enrichment analyses on metformin. (A) Venn diagram of targets for metformin to treat pain. (B) A PPI network of metformin to treat pain. (C) The top 10 significantly enriched terms in biological process (BP). (D) The top 10 significantly enriched terms in molecular function (MF). (E) The top 10 significantly enriched terms in cellular component (CC). (F) The top 10 significantly enriched terms in KEGG. GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; PPI, protein-protein interaction; BP, biological process; MF, molecular function; CC, cellular component.

Network pharmacology highlighted that metformin significantly remolded the painful information mediated by complex circuits associated with excitatory and inhibitory networks. The protooncogene c-Fos is a gene expression-regulated nuclear protein with elevated expression in neurons following neuronal activation. The increase in c-Fos protein was stimulated by neurotransmitters on membrane receptors under physiological stimuli, especially in pain states in response to mechanical stimulus. To further illustrate neuronal activity after metformin treatment in neuropathic rats, we used double immunofluorescence staining to identify c-Fos expression levels (Fig. 3A–C). The results showed that c-Fos expression significantly increased in neuropathic rats, and intrathecal injections of metformin abolished c-Fos expression in the spinal cord (Fig. 3A–D, immunofluorescence staining: F${}_{(2,17)}$ = 25.67, p 0.0001, one-way ANOVA followed by Sidak’s post-tests; Fig. 3E,F, protein: F${}_{(2,10)}$ = 4.387, p = 0.0429, one-way ANOVA followed by Sidak’s post-tests). In addition, a network pharmacology analysis inferred that NOS may be involved in metformin treatments. We observed that metformin significantly restored NOS2 mRNA without affecting NOS1 in neuropathic pain (Fig. 3G,H, NOS1: F${}_{(2,9)}$ = 0.2736, p = 0.7667, NOS2: F${}_{(2,9)}$ = 30.95, p 0.0001, *p 0.05, n = 4, one-way ANOVA followed by Sidak’s post-tests). Therefore, our study demonstrated that metformin produced anticipative effects by inhibiting spinal neuronal activity in neuropathic pain.

Fig. 3.

Dual fluorescence labeling of c-Fos with NeuN among three different groups. (A) c-Fos with NeuN co-labeling in the sham group. (B) c-Fos with NeuN co-labeling in the SNL group. (C) c-Fos with NeuN co-labeling in the metformin-treated group. Images were captured in laminae I–III of spinal cord (n = 6–8 animals). Scale bar: 100 µm ($\times$10), 20 µm is ($\times$20). (D) Statistics on number of c-Fos positive cells. (E,F) c-Fos expression. (G,H) mRNA expression of nitric oxide synthase 1 (NOS1) and NOS2. NOS1: F${}_{(2,9)}$ = 0.2736, p = 0.7667, NOS2: F${}_{(2,9)}$ = 30.95, p 0.0001, *p 0.05, n = 4. GAPDH, Glyceraldehyde 3-phosphate dehydrogenase.

We demonstrated that metformin inhibited nerve injury-induced neuronal activity, but the underlying plastic mechanism remains unclear. Electrophysiological recording is the most dynamic method used to study neural plasticity [30]. Behavioral tests suggested that metformin may significantly lower nociceptive behaviors in the ipsilateral side after SNL (Fig. 1A). Thus, the mechanism for neural plasticity was assessed after nerve injury. Both the frequency and amplitude of mEPSCs in the spinal lamina II significantly increased during neuropathic pain, and this finding was consistent with our prior studies [31]. Frequency reflected the release of presynaptic glutamate, whereas amplitude was associated with postsynaptic receptors. Bath application of 10–100 µM metformin ameliorated the increase in mEPSCs amplitude and frequency in a dosage-dependent manner (Fig. 4A–E, frequency: F${}_{(5,62)}$ = 71.41, p 0.0001; amplitude: F${}_{(5,62)}$ = 24.23, p 0.0001, one-way ANOVA followed by Tukey’s post-tests). Inhibitory transmission was also associated with neuropathic pain. We found that 100 µM metformin treatment did not significantly inhibit the frequency and amplitude of mIPSCs after the application of D-AP5, CNQX, and TTX in both SNL and sham groups (Fig. 4F–J, frequency: F${}_{(3,38)}$ = 19.28, p 0.0001; amplitude, F${}_{(3,38)}$ = 2.017, p = 0.1279, one-way ANOVA followed by Tukey’s post-tests). These findings demonstrated that metformin alleviated spinal glutamatergic transmission in neuropathic rats.

Fig. 4.

Metformin attenuated glutamatergic transmission in the spinal dorsal horn of the ipsilateral hind paws in pain states. (A) Sample traces of mEPSCs. (B,C) Cumulative distribution of frequency and amplitude in mEPSCs. (D,E) Statistical analysis of frequency and amplitude in mEPSCs. (F) Sample traces of mIPSCs. (G,H) Cumulative distribution of frequency and amplitude in mIPSCs. (I,J) Statistical analysis of frequency and amplitude in mIPSCs. Data were presented as the means $\pm$ SEM (n = 5–6 animals). n.s, not significant. *p 0.05, by one-way or repeated-measures two-way ANOVA followed by Sidak’s post-tests. mEPSCs, miniature excitatory synaptic currents.