Heat stimulation elicited apparent paw withdrawal behavior (3.49 ± 0.08 s, 3.42 ± 0.08 s, and 3.49 ± 0.09 s for the baselines of NS, morphine, and morphine plus naloxone sessions, respectively; see Fig. 1). The two-way ANOVA revealed significant session × treatment interactive effect (F (2, 17) = 31.87, P < 0.0001). Post-hoc analysis showed that pre-treatment with morphine produced significantly longer withdrawal latency (5.15 ± 0.24 s) than that of any other session (Bonferroni post tests, P < 0.001). The effect of morphine was prevented by naloxone pretreatment prior to morphine administration. Rats treated with saline caused no change in pain thresholds compared to the baseline level.

A total of 218 units were recorded during the morphine session (62 SI, 53 VPL, 42 ACC and 61 MD). All of these neurons were spontaneously active and the mean firing rates of SI, VPL, ACC and MD neurons were 4.15 ± 0.46, 2.53 ± 0.25, 2.74 ± 0.27, and 2.16 ± 0.25 spikes/s (mean ± S.E.), respectively. The majority of these neurons exhibited excitatory responses to noxious heat stimulation (Table 1). The pain-related neural activities within each area are shown in Fig. 2. The general response characteristics of SI, VPL, ACC and MD neurons are consistent with what we described before 19 .

The intraperitoneal injection of morphine induced a global depression of the pain-related responses in SI, VPL, ACC and MD (Figs. 2 and 3), as compared with the NS session. Fig. 2 shows an example of morphine's effect with and without pretreatment of systemic naloxone (4 mg/kg) on neuronal responses within each region. As can be seen in Fig. 2A, noxious stimulation alone evoked significant activation of thalamic and cortical neurons. Following morphine administration, the noxious heat induced responses were remarkably suppressed. These effects were always reversed by naloxone pretreatment (Fig. 2B). Injection of normal saline did not cause a change in the magnitude of pain-related response within all recorded areas (Fig. 2C). The mean pain-related responses of SI, VPL, ACC and MD neurons before and after administration of morphine are illustrated in Fig. 3. Comparison indicated that morphine significantly attenuated the neural response magnitude in all of the four areas.

Cluster analysis revealed that the recorded cortical and thalamic areas contain units with several different temporal coding patterns for the noxious radiant heat (Fig. 4A). Except for 7 neurons with inhibitory activities and 68 neurons without obvious responses (C4 and C5, respectively), the neurons with excitatory responses were classified as three categories according to the responding duration: long- (20-30 sec, C1), medium- (10-15 sec, C2) and short-response neurons (<10 sec, C3). Twenty-seven percent (58/218) units displayed long-duration responses, whereas 16% (35/218) had medium responses, and 23% (50/218) exhibited short responses. All of the three types of excitatory responses were shortened by systemic injection of morphine (Fig. 4B). The majority of the activation lasted no longer than 5 sec after morphine administration.

In addition, morphine suppressed the activation by reducing the fraction of responding neurons. The percentage of responsive neurons in each recording area before and after injection of morphine or NS is illustrated in Table 2. There is a significant reduction in proportion of responsive neurons in the recorded cortical and thalamic areas after morphine injection. The change in percentage of responding neurons in SI, VPL, ACC and MD over time is shown in Fig. 5.

All experiments were performed on twelve adult male Sprague-Dawley rats (300-350 g) in accordance with the Institutional Animal Care and Use Committee of Peking University. Animals were housed individually under a reversed light/dark cycle (lights off from 7:00 A.M. to 7:00 P.M.) for 7 days before surgery. Food and water were available ad libitum.

Prior to chronic implants of microelectrode array, initial anesthesia was administered by an intraperitoneal injection of ketamine anesthesia (100 mg/kg, i.p.). Supplementary doses (one-third of the original) of ketamine were given as needed to maintain proper anesthetic depth during surgery. Rats were mounted on a stereotaxic apparatus (David Kopf Instruments, Tujunga, CA) and four arrays of eight stainless steel Teflon-insulated microwires (50-μm diameter, Biographics, Winston Salem, NC) were slowly lowered into primary somatosensory cortex (SI, AP = -1.0 mm, ML = 2.0 mm, and DV = 2.0 mm down; measurements relative to bregma); anterior cingulate cortex (ACC, AP = 3.2 mm, ML = 0.8 mm, and DV = 2.8 mm down); medial dorsal thalamus (MD, AP = -2.3 mm, ML = 0.8 mm, and DV = 5.5 mm down); and ventral posterolateral thalamus (VPL, AP = -3.0 mm, ML = 3.0 mm, and DV = 6.0 mm down) according to the atlas of Paxinos and Watson. Six stainless steel screws were screwed into the skull to serve as anchors for cementing the microwires in place after implantation. Animals received penicillin (16,000 U, i.m.) before surgery to prevent infection. Animals were allowed to recover for 1 week before recording sessions commenced.

To record the neuronal activities, rats were placed in a plastic chamber (44 × 44 × 44 cm³) in a quiet room kept at 22 ± 1°C and allowed unrestricted movement during the entire recording session. Neural spike recording started after the rats become familiar with the experimental environment. Extracellular signal were collected by the chronically implanted microwire assemblies which are connected to a preamplifier via a head stage plug and two light-weight cables. The outputs of the preamplifier were filtered (0.5 and 5 kHz, 6 dB cut-off) and sent to a multichannel spike-sorting device (Biographics, Inc) for on-line signal processing. Individual waveforms were discriminated by setting multiple time-voltage windows using a PC-based software Magnet (Biographics, Inc.). The time stamps of these waveforms were then stored on a personal computer for off-line analyses. Graphical capture of waveforms, interspike interval histograms and autocorrelograms were used to validate the on-line sorting of single units. Spike train activity was analyzed with the PC-based programs Stranger (Biographics, Inc.) and NeuroExplorer (Plexon, Dallas, TX).

Noxious radiant heat from a 12.5-W projector bulb was used as painful stimulation, which was randomly applied to the plantar surface of the rats' hindpaws contralateral to the electrode implantation via a 4-mm diameter opening and through a glass floor (1 mm thick). During the experiment, a plastic frame was placed at proper position under the thin glass board, to protect it against the weight of rats. The nocifensive responses were measured by paw withdrawal elicited by the radiant heat. The focused light was manually turned off when the rat lifted the paw. Time stamps (resolution, 1 ms) of the light onset and paw lift were recorded and synchronized with the neural activities. The intensity of heat stimulus was adjusted by changing the voltage of electricity to obtain a latency within the range of 2-4 s and stimulation of identical intensity was used throughout this experiment. To avoid tissue damage, a cut-off limit of 10 s was used. To minimize sensitization and habituation, the interval between two consecutive noxious stimuli was longer than 3 min. Turning on the light without focusing on the paw was used as sham stimulation. The order of stimulus presentation was randomized. Both real and sham stimuli were delivered only when the animal was quiet and showed no voluntary motor activity. The well-isolated single unit activities were recorded simultaneously throughout the duration of all stimulation.

The effect of morphine on the pain-evoked neural responses in SI, VPL, ACC and MD were tested in 8 rats. At baseline, neuronal activities evoked by noxious heat stimulation were recorded from the abovementioned four areas. Then morphine (5 mg/kg) or equivalent volume of NS was injected intraperitoneally and pain-related activities were recorded 10 min later. The sequence of the injection of morphine or NS was random and different injections were given by a separation of ≥2 days.

To ensure specificity of the morphine effect, four animals were pre-treated with 4 mg/kg naloxone 10 min prior to administration of morphine. A repeated neuronal recording was obtained 10 min following morphine injection.

Behavioral data were expressed as mean ± S.E. Statistical differences between treatments and sessions were analyzed by two-way ANOVA followed by Bonferroni post tests, with significance being defined as P < 0.05.

Electrophysiological data were processed offline with NeuroExplorer (Plexon, Inc., Dallas, TX) program for basic analysis and with MatLab (The MathWorks, Natick, MA) and SPSS (Chicago, IL) for advanced statistics. Bin counts per trial (0.1-s bin size) were calculated using NeuroExplorer program and the results were exported to Matlab in spreadsheet form. Neuronal responses to noxious stimulation were evaluated using a sliding window averaging technique, in which a 1-s time window was slid through the entire period of a trial at 1-bin step. The bin counts of each window were compared with those of a baseline window (-5~-1s) by Student's t-test. Only a change of firing rates from the baseline exceeds the limit of P < 0.01 for at least three consecutive windows were taken as a response. The 'P values' were converted into the information theory concept surprise by performing logarithmic transformation, i.e., -ln P, to highlight the significance of the responses distributed over a time period. These values were used to plot the mean ensemble significance of neuronal responses over time. To compare neural responses between different sessions (for example, pre-morphine vs. post-morphine session), firing rates or surprise values were also compared using the sliding window method (1-s time window, 0.1-s step, P < 0.01 for three consecutive windows). The firing rates for all neurons were normalized and arranged into a spreadsheet. A clustering analysis (K-means, SPSS) was performed to classify neuronal responses depending on the similarities in patterns of excitation or inhibition around stimulation events. Chi-square tests were used to detect the percentage differences between different sessions.

Linear discriminant analysis (LDA) is a well-known method for classification and feature extraction. The aim of LDA is to find the linear combination of features which best separate several classes of objects or events. In this paper, LDA was used to investigate whether morphine affected the capability of neural ensembles to discriminate different types of sensory stimulation. The evoked firing rates were chosen for discriminant analysis. The analysis was performed using the MatLab platform and the SPSS package program. Principal component analyses (PCA) were first performed throughout the recording session for neurons within each recorded area. The firing rates of multiple principle components (PCs) around noxious, sham, or no-stimulation (randomly selected points where no events occurred within 30 seconds around) events were then exported into SPSS. The discriminant function coefficients were estimated, and the correct percentage to differentiate sensory inputs was compared before and after morphine injection.

Partial directed coherence (PDC) analysis was used to determine the amount and direction of information flow among SI, VPL, ACC and MD. The process of PDC analysis has been described in detail elsewhere 45 46 47 . To be brief, PDC is a frequency-domain approach of the key concept of Granger-causality to uncover the link between two neuronal groups 48 . Similar to LDA, we firstly performed principal component analyses for neurons in each brain area. Then the first principal component (PC1) of a given brain area was exported into MatLab to calculate PDC (in 1-50 Hz range). These values for each trail were then averaged. The results were normalized to Z scores relative to baseline data.

After completion of the experiment, animals were deeply anesthetized with pentobarbital and selected tip positions of the electrodes were marked electrolytically (10-20 μA, 10-20 s, anode current). The animals were then sacrificed and perfused with 4% paraformaldehyde-5% potassium ferricyanide solution. The brains were extracted and post-fixed with the same solution used for the perfusion for 48 h. Coronal sections of 40 μm thick were cut through the SI, ACC, and thalamus. In these sections, recording sites were then determined as blue dots. Data of those sites deflecting from the target areas were out of analysis.