Main

In a screen to discover new blockers of ASIC channels from animal venoms, we identified the venom of black mamba (Dendroaspis polylepis polylepis) as a potent and reversible inhibitor of the ASIC1a channel expressed in Xenopus oocytes (Fig. 1a). After purification, two active fractions were collected (Supplementary Fig. 1a, b). A partial amino-acid sequence was used to clone by degenerated PCR the corresponding complementary DNA (cDNA) from lyophilized venom (Supplementary Fig. 1c). Two isopeptides were identified and named mambalgin-1 and mambalgin-2. They are composed of 57 amino acids with eight cysteine residues, and only differ by one residue at position 4 (Supplementary Fig. 1d).

Figure 1: Mambalgins represent a new class of three-finger toxins targeting ASIC channels.
figure 1

a, Black mamba venom (0.1 mg ml−1) reversibly inhibits rat ASIC1a current expressed in Xenopus oocytes. b, Three-dimensional model of mambalgin-1 (disulphide bridges in yellow). c, Electrostatic properties of mambalgin-1 and human ASIC1a channel (on the basis of the three-dimensional structure of chicken ASIC1a29) with positive and negative isosurfaces in blue and red, respectively. d, e, Inhibition of rat ASIC channels expressed in COS-7 cells (n = 4–15; peptides applied before the pH drop as in a). Hill coefficients of 0.7–1 suggest a 1:1 stoichiometry between mambalgins and channels. f, Effect of mambalgin-1 on the pH-dependent activation and inactivation of ASIC1a current recorded in Xenopus oocytes (protocol shown in inset). g, h, Inhibition of native ASIC currents in dorsal spinal cord neurons (g) and DRG neurons (h) by mambalgin-1 (674 nM) and PcTx1 (20 nM), n = 14–34. Right panels: effect of mambalgin-1 on neurons expressing no PcTx1-sensitive current. (Vh = −60 mV in a, d, e and −50 mV in f–h.) Mean ± s.e.m.

PowerPoint slide

Full size image

Mambalgins belong to the family of three-finger toxins12 (Supplementary Fig. 1d, e). They have no sequence homology with either PcTx1 or APETx2, two toxins that we previously identified to target ASIC channels13,14. A three-dimensional model of mambalgin-1 has been established from a pool of five templates with known structures (Supplementary Fig. 1e), which shows a triple-stranded and short double-stranded antiparallel β-sheets connecting loops II and III, and loop I, respectively, the three loops emerging from the core of the toxin like fingers from a palm (Fig. 1b). The model structure presents a concave face commonly found in neurotoxins and is stabilized by four disulphide bonds with a pattern identical to that observed in the crystal structure template (Cys 1–Cys 3, Cys 2–Cys 4, Cys 5–Cys 6 and Cys 7–Cys 8; Fig. 1b and Supplementary Fig. 1f). Mambalgins show a strong positive electrostatic potential that may contribute to binding to negatively charged ASIC channels (Fig. 1c).

Mambalgins have the unique property of being potent, rapid and reversible inhibitors of recombinant homomeric ASIC1a and heteromeric ASIC1a + ASIC2a or ASIC1a + ASIC2b channels, that is, all the ASIC channel subtypes expressed in the central nervous system15,16,17,18, with a similar potency for both isopeptides and IC50 values (the toxin concentration half-maximally inhibiting the current) of 55 nM, 246 nM and 61 nM, respectively (Fig. 1d). Mambalgins also inhibit human ASIC channels (Supplementary Fig. 2a). The peptides inhibit ASIC1b and ASIC1a + ASIC1b channels that are specific of sensory neurons19 with IC50 values of 192 nM and 72 nM, respectively (Fig. 1e). Mambalgins, which bind to the closed and/or inactivated state of the channels (Supplementary Fig. 2b), modify the affinity for protons (pH0.5act shifted from 6.35 ± 0.04 to 5.58 ± 0.02, and pH0.5inact shifted from 7.10 ± 0.01 to 7.17 ± 0.01, n = 4, P < 0.001 and n = 5, P < 0.05, respectively; Fig. 1f) and act as gating modifier toxins. They potently inhibit native ASIC currents in spinal cord, hippocampal and sensory neurons (Fig. 1g, h and Supplementary Fig. 3). In central spinal cord neurons, mambalgin-1 (674 nM) decreased ASIC current amplitude to 13.0 ± 2.0% (n = 14) of the control (Fig. 1g) and reduced the excitability in response to acidic pH without unspecific effect on basal neuronal excitability (resting potential) or on the threshold and the shape of evoked or spontaneous action potentials and on spontaneous postsynaptic currents (Supplementary Fig. 4). Mambalgins had no effect on ASIC2a, ASIC3, ASIC1a + ASIC3 and ASIC1b + ASIC3 channels, as well as on TRPV1, P2X2, 5-HT3A, Nav1.8, Cav3.2 and Kv1.2 channels (Supplementary Figs 5 and 6).

Most three-finger toxins, such as α-neurotoxins that block nicotinic acetylcholine receptors20, evoke neurotoxic effects in animals. This is not the case of mambalgins, which do not produce motor dysfunction (Supplementary Fig. 7), apathy, flaccid paralysis, convulsions or death upon central injections (intrathecal or intracerebroventricular) in mice, but instead induce analgesic effects against acute and inflammatory pain (Fig. 2) that can be as strong as morphine but resistant to naloxone, with much less tolerance (Fig. 3a) and no respiratory distress (Fig. 3b). In the tail-flick and paw-flick tests, intrathecal injection of mambalgin-1 (or mambalgin-2) increased the latency for the tail and paw withdrawal reflex from 8.8 ± 0.4 s and 8.0 ± 0.8 s to 23.2 ± 1.3 s and 19.8 ± 1.6 s, respectively, 7 min after injection (n = 15–22, P < 0.001) (Fig. 2a, b). The effects were completely lost in ASIC1a knockout mice (Fig. 2a, b), demonstrating the essential implication of ASIC1a-containing channels. The key involvement of ASIC channels present in central neurons in the analgesic effect of mambalgins was confirmed using intracerebroventricular injections of the peptides (Supplementary Fig. 8). Mambalgin-1 also suppressed inflammatory heat hyperalgesia and produced a strong analgesia evaluated in the paw-flick test after intraplantar injection of carrageenan (Fig. 2c), and drastically decreased acute (phase I) and inflammatory (phase II) pain assessed in the formalin test (Fig. 2d), with a potency similar to morphine (Fig. 2c, d). These effects were not significantly decreased by naloxone.

Figure 2: Intrathecal injections of mambalgin-1 exert potent naloxone-resistant and ASIC1a-dependent analgesia in mice.
figure 2

a, b, Effects on acute thermal pain (46 °C) determined using tail-immersion (a, n = 13–22) and paw-immersion (b, n = 5–15) tests showing a large increase in response latencies, which is lost in ASIC1a–/– mice. c, Effects on inflammatory hyperalgesia induced by intraplantar carrageenan (grey bar) (n = 8–16). Right panels in a–c: area under curve (AUC, s × min) calculated from each mouse. d, Effects on first (0–10 min) and second (10–45 min) phase of formalin-induced spontaneous pain behaviour (n = 11–20). Naloxone was injected subcutaneously 10 min before the peptide. Comparisons are versus vehicle unless specified. Mean ± s.e.m. ***P < 0.001; **P < 0.01; *P < 0.05; NS, P > 0.05.

PowerPoint slide

Full size image
Figure 3: The central analgesic effect of mambalgin-1 shows reduced tolerance compared with morphine, no respiratory depression and involves the ASIC2a subunit.
figure 3

a, Repeated intrathecal injections of mambalgin-1 induce less tolerance than morphine at concentrations giving the same analgesic efficacy (n = 10, comparison with vehicle (*) or morphine (#)). b, Mambalgin-1 (i.t., intrathecal) induces no respiratory depression unlike morphine (i.t., intrathecal or i.p., intraperitoneal), 0.01 and 0.4 mg per mouse, respectively; n = 4–7, comparison with vehicle unless specified). c, Paw-flick latency before (control) and after treatment with non-specific siRNA (si-CTR, n = 24), siRNA against ASIC2a and ASIC2b (si-ASIC2a/2b, n = 27), or siRNA against ASIC2a (si-ASIC2a, n = 19) with or without naloxone. Comparison with si-CTR (#) or untreated control (*), unless specified. d, Paw-flick area under curve calculated after intrathecal injection of mambalgin-1 or vehicle in siRNA-treated mice (n = 5–27; protocol shown in inset above; s.c., subcutaneous). Comparison with mice injected with vehicle after treatment by si-CTR, unless specified. Mean ± s.e.m. *** or ###, P < 0.001; ** or ##, P < 0.01; *P < 0.05; NS, P > 0.05.

PowerPoint slide

Full size image

Mambalgins, unlike the spider peptide PcTx1 (refs 5, 14), inhibit not only homomeric ASIC1a channels but also heteromeric ASIC1a + ASIC2 channels, which are abundantly expressed in central neurons15,16,17,21,22. This led us to analyse the participation of ASIC2 in central analgesia evoked by mambalgins. Intrathecal injections of short interfering RNA (siRNA) to silence either ASIC2 (both variants a and b) or ASIC2a (Supplementary Fig. 9) induced an analgesia that was partly (ASIC2) or fully (ASIC2a) resistant to naloxone (Fig. 3c). In the presence of naloxone, central injection of mambalgin-1 in these knockdown mice had a decreased effect (Fig. 3d), consistent with a contribution of ASIC2a in the pain suppressing effect of the peptide. Compensation by homomeric ASIC1a channels15,16, which are also blocked by mambalgin-1, as well as incomplete in vivo knockdown (Supplementary Fig. 9) account for the residual analgesic effect that is observed.

Because mambalgins are able to target different ASIC channel subtypes expressed in nociceptors, we tested their peripheral effect after intraplantar injections. Mambalgin-1 (unlike PcTx1) has a significant analgesic effect on acute heat pain (Fig. 4a) and reverts or prevents inflammatory hyperalgesia (Fig. 4b, c). However, this effect is clearly different from the previously described effect of central (intrathecal) injection of the peptide because it is still present in ASIC1a knockout mice (Fig. 4a–c) contrary to the central effect (Fig. 2a, b). If ASIC1a is not involved, what are then the mechanisms that support the peripheral effect of mambalgins? ASIC1b is specifically expressed in nociceptors19,23 but its role in pain is not known. A functional expression of ASIC1b-containing channels in dorsal root ganglion (DRG) neurons is demonstrated by the effect of mambalgin-1, which blocks both ASIC1a and ASIC1b, and has more potent effects than PcTx1, which only blocks ASIC1a channels (Fig. 1h). Moreover, silencing of the ASIC1b subunit in nociceptors of ASIC1a knockout mice (that is, where only ASIC1b is present; Supplementary Fig. 9) mimicked the analgesic effect of peripheral injection of mambalgin-1 on both acute pain and inflammatory hyperalgesia, and largely decreased the effect of subsequent intraplantar injection of the peptide (Fig. 4d, e), supporting the specific participation of ASIC1b in the peripheral effect of mambalgin-1.

Figure 4: Intraplantar injections of mambalgin-1 evoke peripheral analgesic effects through ASIC1b-containing channels.
figure 4

a, Effects of mambalgin-1 on acute thermal pain (46 °C) in wild-type (WT) and ASIC1a–/– mice (n = 10–23). b, Effects of mambalgin-1 on carrageenan-induced hyperalgesia when injected before (open symbols) or 2 h after (filled symbols) carrageenan (n = 7–20). Right panels: area under curve (after carrageenan in b). Comparison with vehicle unless specified. c, Mambalgin-1 prevents inflammatory hyperalgesia when co-injected with carrageenan (n = 7–20, comparison with control). d, Paw-flick latency before (control) and after treatment with si-CTR (n = 13) or siRNA against ASIC1a and ASIC1b (si-ASIC1a/1b, n = 10) in ASIC1a–/– mice upon normal and inflammatory conditions. Comparison with si-CTR (#) or untreated control (*), unless specified. e, Paw-flick area under curve calculated after injection of mambalgin-1 or vehicle in siRNA-treated ASIC1a–/– mice (n = 7–11). Comparison with mice injected with vehicle after treatment by si-CTR, unless specified. Mean ± s.e.m. *** or ###, P < 0.001; ** or ##, P < 0.01.

PowerPoint slide

Full size image

Our results indicate that mambalgins have analgesic effects by targeting both primary nociceptors and central neurons, but through different ASIC subtypes (Supplementary Fig. 12). After demonstrating the important role of ASIC3 in peripheral pain and sensory perception in the skin3,4,24, we now show that ASIC1b, but not ASIC1a, is important for cutaneous nociception and inflammatory hyperalgesia. In the central nervous system, injections of mambalgins evoke a naloxone-insensitive analgesia through an opioid-independent pain pathway involving ASIC1a + ASIC2a channels. Central injections of PcTx1, instead, evoke a naloxone-sensitive analgesia through its specific action on homomeric ASIC1a5, and probably heteromeric ASIC1a + ASIC2b channels (Fig. 3c and ref. 22). In addition, mambalgins, unlike PcTx1 (ref. 5), maintain a potent analgesia in mice deficient for the preproenkephalin gene (Supplementary Fig. 10). These results taken together indicate that different pathways involving different subtypes of ASIC channels can lead to different types of central analgesia (opioid-sensitive or insensitive) (Supplementary Fig. 12). They also indicate that despite their capacity in vitro to inhibit homomeric ASIC1a and heteromeric ASIC1a + ASIC2b channels, in vivo, mambalgin central analgesic action is mainly targeted to neurons expressing ASIC1a + ASIC2a channels (Supplementary Fig. 11).

It is essential to understand pain better to develop new analgesics25,26. The black mamba peptides discovered here have the potential to address both of these aims. They show a potent role for different ASIC channel subtypes in both the central and peripheral pain pathways, providing promising new targets for therapeutic interventions against pain, and they are themselves powerful, naturally occurring, analgesic peptides of potential therapeutic value.

Methods Summary

Mambalgins were purified from Dendroaspis polylepis polylepis venom (Latoxan) using cation exchange and reverse phase chromatography steps. The molecular mass and peptide sequence were determined by Edman degradatation and/or tandem mass spectrometry sequencing, and used to design primers for cloning the full-length mambalgin-1 cDNA from venom. The three-dimensional structure was modelled from five templates of three-finger snake toxins. Recordings of recombinant ASIC currents were done after expression in Xenopus laevis oocytes14 and COS-7 cells, and patch-clamp recordings of native ASIC currents were obtained from primary cultures of mouse dorsal spinal cord neurons16, hippocampal neurons and rat DRG neurons. Pain behaviour experiments were performed in C57BL/6J mice after intrathecal (or intracerebroventricular) injection of mambalgins (10 µl at 34 µM), PcTx1 (10 µl at 10 µM) or morphine (Cooper, 10 µl at 3.1 mM), in the absence or in the presence of naloxone (Fluka, 2 mg kg−1), and after intraplantar injection of mambalgins (10 µl at 34 µM) and PcTx1 (10 µl at 10 µM). The effects of mambalgins on acute pain were also tested in ASIC1a–/– (ref. 27) and Penk1–/– (ref. 28) mice. Inflammation was evoked by intraplantar 2% carrageenan (Sigma) or 2% formalin. In vivo ASIC1 and ASIC2 gene silencing experiments were performed by repeated intrathecal injections (2 µg per mouse twice a day for 3 days) of siRNAs targeting ASIC1a + b, ASIC2a + b or only ASIC2a, mixed with i-Fect (Neuromics) as previously described4.

Online Methods

Electrophysiology in Xenopus laevis oocytes

Venom fractions were tested on rat ASIC1a expressed in Xenopus oocytes as previously described14, applied 30 s before the acid stimulation.

Purification, peptide sequencing and mass spectrometry

The venom of the black mamba Dendroaspis polylepis polylepis (Latoxan) was purified by gel filtration and cation exchange30. The active fraction was loaded on a reversed-phase column (C18 ODS, Beckman) and eluted with a linear gradient of acetonitrile containing 0.1% TFA. Molecular mass and peptide sequence were determined by matrix-assisted laser desorption/ionization–time of flight (MALDI–TOF)/TOF-MS (Applied Biosystems). Protein identification was performed with mascot (http://www.matrixscience.com) at 50 p.p.m. mass tolerance against NCBI (non-redundant) and Swiss-Prot databases. Data were analysed using the GRAMS386 software. Partial sequence was obtained by amino (N)-terminal Edman degradation and protease digestion (V8 protease and trypsin) followed by tandem mass spectrometry sequencing. Peptide analysis was performed using a nano-high-performance liquid chromatography offline (Dionex, U3000) coupled with a 4800 MALDI-TOF/TOF mass spectrometer.

Cloning of the mambalgin-1 cDNA

Mambalgin-1 cDNA was cloned from the black mamba venom31. Lyophilized venom (Sigma) was reconstituted in lysis/binding buffer and polyadenylated mRNAs were captured on oligo(dT25) magnetic beads (Dynal). After first-strand cDNA synthesis, PCR-amplification was done with degenerated sense (TGITTYCARCAYGGIAARGT) and antisense (YTTIARRTTICGRAAIGGCAT) primers designed from the partial peptide sequence obtained from biochemical purification. A specific sense primer (ACACGAATTCGCTATCATAACACTGGCATG) was designed from the new sequence and used with an unspecific poly-dT30 antisense primer (ACACGAATTCdT30) to amplify the 3′-coding and uncoding sequences of mambalgin-1. Using the very strong conservation of the 3′- and 5′-uncoding sequences among snake toxins32, we have designed a sense (ACACGAATTCTCCAGAGAAGATCGCAAGATG) and an antisense (ACACGAATTC-ATTTAGCCACTCGTAGAGCTA) primer to amplify the complete open reading frame of the toxin precursor.

Template-based three-dimensional modelling of mambalgin-1

We modelled the mambalgin-1 protein using the semi-automatic pipeline of the webserver @TOME version 2.1 (ref. 33). The amino-acid sequence was submitted to the server to perform the fold recognition and detect structural homologue templates from the Protein Data Bank. Active fold-recognition tools were HHSEARCH34, SP335, PsiBlast36 and Fugue37. Five templates were selected among snake venom toxins with four disulphide bonds and aligned with Muscle38. The homology modelling of mambalgin-1 was performed with Modeller 9v8. The overall quality of models was estimated by calculating the one- and three-dimensional compatibility TITO score39, by analysing the Ramachandran by MolProbity and comparing it with scores of the templates40 and by visual inspection.

Electrostatic potential calculation

Electrostatic properties of mambalgin-1 (isosurfaces of +3 kBT/ec (∼ +77 mV) and −3 kBT/ec (∼ −77 mV)) and human ASIC1a channel (isosurfaces of +10 kBT/ec (∼ +256 mV) and −10 kBT/ec (∼ −256 mV)) have been calculated with the Adaptive Poisson-Boltzmann Solver41.

Electrophysiology in COS cells and neurons

COS-7 cells were transfected with pCI-ASICs mixed with pIRES2-EGFP and jet-PEI. Primary cultures of dorsal spinal neurons were obtained from C57Bl6J mice embryos (embryonic day (E)14)16. Primary cultured hippocampal neurons were prepared from C57Bl6J mice (P3–P5) as previously described for rats42. Primary cultured sensory neurons were prepared from dorsal root ganglia of Wistar rats (5–7 weeks) as previously described43.

Data were recorded in the whole-cell configuration, sampled at 3.3 kHz and low-pass filtered at 3 kHz using pClamp8 software (Axon Instruments). The pipette solution was (in mM) KCl 140, NaCl 5, MgCl2 2, EGTA 5, HEPES-KOH 10 (pH 7.4); the bath solution was (in mM): NaCl 140, KCl 5, MgCl2 2, CaCl2 2, HEPES-NaOH 10 (pH 7.4). MES was used instead of HEPES for pH from 6 to 5. The bath solution for neurons was supplemented with 10 mM glucose, and 20 µM CNQX/10 µM kynurenic acid for central neurons. The pipette solution for neurons contained (in mM) KCl 140, ATP-Na2 2.5, MgCl2 2, CaCl2 2, EGTA 5, HEPES 10 (pH 7.3, pCa estimated to 7). Toxins were perfused at pH 7.4 with bovine serum albumin (0.05%) to prevent non-specific adsorption. Concentration–response curves were fitted by the Hill equation: I = Imax + (Imin − Imax) (C/(C + IC50)) where I is the amplitude of relative current, C is the toxin concentration, IC50 is the toxin concentration half-maximally inhibiting the current, and nH is the Hill coefficient.

Plethysmography

Respiratory frequency (breaths per minute) was recorded from 7 to 67 min after intrathecal injection of vehicle, mambalgin-1 or morphine-HCl (according the same protocol than for pain behaviour) or intraperitoneal injection of morphine-HCl (24.8 mM, 50 µl) with a whole-body plethysmograph (Emka Technologies).

Pain behaviour experiments

Experiments were performed on awake 7- to 11-week-old (20–25 g) male C57BL/6J, ASIC1a–/–27 and Penk1–/–28 mice following the guidelines of the International Association for the Study of Pain and were approved by the local ethics committee (agreements NCA/2007/04-01 and NCE/2011-06). Mambalgin-1 (34 µM), mambalgin-2 (20 µM), PcTx1 (10 µM) and morphine-HCl (3.1 mM; Cooper) dissolved in vehicle solution (in mM: NaCl 145, KCl 5, MgCl2 2, CaCl2 2, HEPES 10, pH 7.4, 0.05% BSA for intrathecal injection, and NaCl 154, 0.05% BSA for intraplantar injection) were injected intrathecally (10 µl) between spinal L5 and L6 segments or intraplantarly (10 µl). Naloxone (Fluka, 2 mg kg−1 in NaCl 0.9%, 50 µl) was subcutaneously (dorsal injection) injected 10 min before intrathecal injection. Inflammation was evoked by intraplantar injection in the left hindpaw of 2% carrageenan (Sigma-Aldrich) (20 µl) 2 h before intrathecal or intraplantar injection of peptides, morphine or vehicle.

Knockdown experiments

Locally designed siRNAs targeting ASIC1 (si-ASIC1a/1b, GCCAAGAAGUUCAACAAAUdtdt), ASIC2 (si-ASIC2a/2b, UGAUCAAAGAGAAGCUAUUdtdt) and ASIC2a (si-ASIC2a, AGGCCAACUUCAAACACUAdtdt) have been validated in vitro in COS-7 cells transfected with myc-ASIC1a, ASIC1b, myc-ASIC2a or myc-ASIC2b, and the relevant siRNA or a control siRNA (si-CTR, GCUCACACUACGCAGAGAUdtdt) with TransIT-LT1 and transIT-TKO (Mirus), respectively. Cells were lysed 48 h after transfection and processed for western blot analysis to assess the amount of ASIC1a protein with the anti-Myc A14 antibody (1:500; Santa Cruz Biotechnology) or the anti-ASIC1 antibody (N271/44; 1:300; NeuroMab) and a monoclonal antibody against actin (AC-40; 1:1,000; Sigma) as a loading control. siRNAs were intrathecally injected into mice (2 µg per mouse at a ratio of 1:4 (w/v) with i-Fect (Neuromics)) twice a day for 3 days. After 3 days of treatment, the paw-flick latency was measured and the residual effect of mambalgin-1 (intrathecal or intraplantar, 34 µM) or the carrageenan (intraplantar, 2%)-induced hyperalgesia was tested. For validation of the in vivo effect of the siRNAs, lumbar DRGs or lumbar dorsal spinal cord were removed after the last siRNA injection for total RNA isolation (RNeasy kits, Qiagen) followed by cDNA synthesis (AMV First-Strand cDNA synthesis kit (Invitrogen) and High Capacity RNA-to-cDNA Kit, (Applied Biosystems)). The relative amounts of ASIC transcripts were evaluated by quantitative reverse-transcription PCR in a Light-Cycler 480 (Roche Products). Pre-designed and validated TaqMan assays for ASIC1 (ASIC1a and ASIC1b; Mm01305998_mH), ASIC1a (Mm01305996_m1), ASIC2 (ASIC2a and ASIC2b; Mm00475691_m1), ASIC3 (Mm00805460_m1) and 18S ribosomal RNA (Mm03928990_g1) were from Applied Biosystems. Each cDNA sample was run in triplicate and results were normalized against 18S and converted to fold induction relative to control siRNA treatment.

Data analysis

Data were analysed with Microcal Origin 6.0 and GraphPad Prism 4. Areas under the time course curves (response latency in seconds × time after injection in minutes) were calculated for each mouse (over the first 37 min for tail-flick and the entire time range for paw-flick) and expressed as mean ± s.e.m. After testing the normality of data distribution, the statistical difference between two different experimental groups was analysed by unpaired Student’s t-test, and between more than two different experimental groups by a two-way analysis of variance followed by a Newman–Keuls multiple comparison test when P < 0.05. For data in the same experimental group, a paired Student’s t-test was used. *** or ###, P < 0.001; ** or ##, P < 0.01; *P < 0.05; NS, P > 0.05.