• In a search for sympathetic premotor neurons subserving thermoregulatory functions, medullary raphé-spinal neurons were studied in urethane-anaesthetized, artificially ventilated, paralysed rats. Extracellular unit recordings were made from a region previously shown to drive the sympathetic supplies to tail vessels and brown adipose tissue. Neurons that were antidromically activated by stimulation across the intermediate region of the upper lumbar cord (the origin of the tail sympathetic outflow) were selected for study.
• Non-noxious cooling stimuli were delivered to the animal's shaved trunk by circulating cold instead of warm water through a water jacket. Cooling increased the activity of 21 out of 76 raphé-spinal neurons by 1.0 ± 0.2 spikes s⁻¹°C⁻¹ for falls in skin temperature of 3-5 °C below a threshold of 35.0 ± 0.6 °C. Their responses followed skin temperature in a graded manner, and did so whether or not there was any change in core (rectal) temperature.
• Indirect observations suggested that seven of the neurons that were activated by skin cooling were also activated by falls in core temperature (by 2.1 ± 0.7 spikes s⁻¹°C⁻¹ below a threshold of 36.1 ± 0.7 °C), while the remainder were unaffected by core cooling.
• An additional 7/76 raphé-spinal neurons showed evidence of inhibition (activity reduced by 2.1 ± 0.5 spikes s⁻¹°C⁻¹) when the trunk skin was cooled.
• Cold-activated raphé-spinal neurons were found in the nuclei raphé magnus and pallidus, centred at the level of the caudal part of the facial nucleus. Their spinal axons conducted at velocities between 3.4 and 29 m s⁻¹ (median 6.8).
• Drug-induced rises in arterial pressure partially inhibited the discharge of 6/14 cold-activated raphé-spinal neurons. Weak-to-moderate cardiac modulation (10–70 %) was present in arterial pulse-triggered histograms of the activity of 11/21 cold-activated raphé-spinal neurons, and 6/6 showed evidence of ventilatory modulation (two strongly, four weakly) in pump-triggered histograms.
• Raphé-spinal neurons responded to cooling in the absence of any change in the electroencephalogram pattern (6/6 neurons).
• Most cold-activated raphé-spinal neurons responded to noxious tail pinch (13/21 inhibited, 6/21 excited), as did most thermally unresponsive raphé-spinal cells in the same region (19/41 excited, 9/41 inhibited).
• It is suggested that these cold-activated raphé-spinal neurons may constitute a premotor pathway that drives sympathetically mediated cold defences, such as cutaneous vasoconstriction or thermogenesis. The data are consistent with the hypothesis that a brainstem reflex, with additional descending input signalling body core temperature, may mediate autonomic responses to environmental cooling.

Recent evidence suggests that the actions of rostral medullary raphé neurons on sympathetic pathways may be quite selective. Morrison and colleagues, for example, demonstrated on rats that neurons in the region of nucleus raphé pallidus play an important role in driving the sympathetic supply to interscapular brown adipose tissue (BAT) but not the splanchnic nerve (Morrison et al. 1999; Morrison, 1999). In rabbits, Blessing and colleagues showed that electrical stimulation of the rostral ventral medullary raphé region caused vasoconstriction of the ear, and this survived inactivation of the rostral ventrolateral medulla (Blessing et al. 1999). In rats we found that the vasomotor supply to the tail, but not the renal nerve, was strongly activated by chemically activating neurons in this same ventral raphé region (Rathner & McAllen, 1999). In these cases, the neurons of the rostral ventrolateral medulla, which are known to drive most sympathetic vasomotor outflows, had little effect on these outflows that responded to raphé neurons (Rathner & McAllen, 1999; Morrison, 1999; Blessing & Nalivaiko, 2000).

The rat's tail acts as a radiator, through which heat loss to the environment is regulated by blood flow under sympathetic control (Dawson & Keber, 1979), being similar in this function to the rabbit's ear (Wathen et al. 1971). BAT is the major tissue of non-shivering thermogenesis (Foster & Frydman, 1978). A common feature, then, which is shared by these sympathetic pathways that have been shown to be selectively driven by ventral medullary raphé neurons, is that they are all engaged as part of the body's cold-defence response. The same must presumably apply to the premotor neurons that drive them. This notion is supported by the finding that many cells in the ventral medullary raphé region express the transcription factor Fos, indicating that they had been activated, after cold exposure in rats (Morrison et al. 1999; Bonaz & Taché, 1994).

In a search with the aim of identifying these sympathetic premotor neurons electrophysiologically, we have examined the responses of raphé-spinal neurons to mild cooling. We have tested the hypothesis that a subset would be excited by this stimulus, and that those neurons’ other properties would be compatible with a premotor role controlling sympathetic cold-defence pathways. To help select those neurons, we studied raphé neurons with axonal projections to the spinal level of the tail vasoconstrictor outflow, the caudal extremity of the preganglionic cell column in the lumbar cord (Smith & Gilbey, 1998; Rathner & McAllen, 1998). While this would not have eliminated neurons projecting to non-sympathetic targets in the lower cord, it would have selected against the premotor neurons controlling most other sympathetic outflows.

Preliminary reports of this work have been presented in abstract form (McAllen et al. 2000a,b, 2001).

Body temperature was maintained via a custom-made water jacket, similar to that described by Dickenson (Dickenson, 1977), through which either warm (42-50 °C at source) or cold (4-9 °C at source) water could be pumped at a constant rate (80-120 ml min⁻¹). Three thermocouples were attached with cyanoacrylate adhesive to different sites on the rat's trunk skin under the water jacket. The mean of these three readings was taken as the measure of skin temperature. Passing cold water through the blanket for 1.5-3 min lowered skin temperature by approximately 2-5 °C. Placing the jacket around the experimenter's forearm confirmed that this was a mild, non-noxious stimulus. A fourth thermocouple was inserted 2-3 cm into the rectum and used to measure body core temperature.

To provide added stability, the spine was clamped at T_10/11 and the animal given a pneumothorax. A laminectomy of the T₁₂ vertebra exposed the L₁-L₂ segments of the spinal cord. An electrode made from two enamel-coated insect pins (tip separation ca 2 mm, tip exposure ca 100 μm) was aligned to span the midline and inserted into the spinal cord at this level, then partially withdrawn so the tips rested approximately 0.5 mm below the dorsolateral sulci (the approximate location of the intermediolateral cell columns). Electrical stimuli of either polarity were delivered bipolarly between the two pins. These were 0.1 ms duration square waves, given either at constant voltage (range 4-100 V) or constant current (range 0.04-7 mA). At the end of the experiment, stimulus sites were marked by passing direct current of 20 μA for 30 s in each polarity between the electrodes. The stimulating electrodes were always confirmed, by dissection of spinal roots at the end of the experiment, to have been in either the L₁ or the L₂ spinal segment, and in most cases their positions within the segment were reconstructed by histological analysis (described below).

In some experiments (5/16), the ventilatory cycle was recorded via a pressure transducer attached to the expiratory ventilation line. The frontal electroencephalogram (EEG) was also recorded in 5/16 experiments from two stainless steel self-tapping screws implanted bilaterally through the frontal bone, approximately 1 mm rostral and 2 mm lateral to bregma. Wires were attached, covered in dental impression material (Reprosil, Dentsply International Inc., Milford, DE, USA), and the signal between them amplified (× 1000 or 2000) and filtered (bandpass 1-40 Hz).

At the conclusion of all surgery, before recording commenced, anaesthesia was carefully switched to urethane (1-1.5 g kg⁻¹, i.v.) over approximately 30 min, during which time isofluorane was gradually withdrawn. The level of anaesthesia was checked at all stages, then re-checked for adequacy and stability before neuromuscular blockade was administered by a bolus dose of pancuronium bromide (2 mg kg⁻¹, i.v.). Neuromuscular blockade was periodically allowed to wear off so as to allow for re-testing the depth of anaesthesia by withdrawal reflexes. Supplementary doses of urethane (20-40 mg, i.v.) were then administered if required, and repeat doses of pancuronium were given only when an appropriate depth of anaesthesia had been confirmed. During neuromuscular blockade, anaesthetic depth was monitored additionally by the absence of significant blood pressure changes, either spontaneous or in response to noxious pinching.

Signals from the microelectrode, blood pressure, skin and rectal temperatures, stimulus trigger signals, EEG and airway pressure were recorded along with a voice/event channel on an instrumentation tape recorder for later analysis. At the time of the experiment, action potentials of raphé neurons were discriminated by a custom-built time/amplitude window discriminator. The output pulses of the discriminator were fed, along with the temperature and pressure signals, into a computer-based analysis system (CED ‘1401 plus’ intelligent interface and ‘Spike 2′ data acquisition and analysis software: Cambridge Electronic Design, Cambridge, UK). Selected recording sites were marked by electrophoretic deposition of Pontamine Sky Blue from the microelectrode (5-10 μA cathodal current for 10 min).

The raphé-spinal neuron's spontaneous activity was then followed during and after 1.5-2 min cooling episodes. Neurons whose firing rate showed a clear, stimulus-locked change in response to cooling were selected for further study and repeated tests. Those that did not respond to cooling were discarded after minimal further investigation. All temperature-responsive raphé-spinal neurons and most others were tested for their response to noxious tail pinch, applied for 5-10 s with a haemostat. Some were tested for barosensitivity by their response to intravenous bolus administration of a pressor agent (0.5-1 μg of noradrenaline or phenylephrine): a > 30 % change in neuronal activity (always a fall) time-locked to a pressure rise of ≥ 30 mmHg was considered to be a positive response.

Cycle-triggered peristimulus histograms of neuronal activity were constructed with 20 ms or 100 ms bin widths, using either the systolic peak of the blood pressure or the rising phase of airway pressure as the trigger. Histograms were smoothed by three-point averaging, and their pattern compared with the averaged blood pressure or airway pressure signal, as appropriate. Histograms were considered to be modulated if they showed a repeating pattern of peaks and troughs of the appropriate periodicity over three cycles of the stimulus waveform. Modulation was measured as the cyclic peak-to-peak fluctuation in the smoothed histogram and expressed as a percentage of the mean bin count. Unless otherwise noted, data are presented as mean ±s.e.m.

Figure 1 Raphé-spinal neurons: identification, location and thermal responses

Figure 2 Raphé-spinal neurons: conduction velocities

Table 1 Properties of cold-activated raphé-spinal neurons

Figure 4 Cold-activated raphé-spinal neurons: responses to skin and core (rectal) temperatures

Figure 6 Cold-activated raphé-spinal neurons: thresholds and slope responses to skin and core (rectal) temperatures

Neuronal responses to cooling and re-warming were often not identical. In Fig. 4 and Fig. 5, for example, firing rate significantly lagged behind the change in skin temperature, while the response of the unit shown in Fig. 1A more closely parallelled the stimulus, and yet other neurons’ responses (not shown) led the measured fall in skin temperature (Table 1). To what extent these differences in response profile could have been methodological (i.e. due to thermocouple positioning) was not investigated, but two neurons recorded in the same experiment did not always follow the same pattern.

Figure 5 Cold-activated raphé-spinal neurons: response to skin but not core (rectal) temperature

Eighteen of the 21 cold-activated raphé-spinal neurons were located histologically from marked recording sites. These have been plotted onto three coronal sections of medulla in Fig. 3A. Recording sites were centred 400 ± 540 μm (s.d.; range, ±1000; n = 18) rostral to the caudal pole of the facial nucleus (CP7). Their distribution approximately followed the extent of nucleus raphé magnus (although this structure is not well defined anatomically), particularly its ventral portion. Some recording sites may have been in nucleus raphé pallidus, but many were beyond the confines of that compact cell group (Fig. 3A).

Figure 3 Locations of recording and stimulation sites

The reconstructed spinal stimulation sites used in these experiments are shown in Fig. 3B. All were confirmed to be in the first or second lumbar cord segment. Stimulation between bilaterally placed electrodes did not allow precise location of the stimulus site within the segment, however. Nevertheless, it was apparent that the stimulus dipole was always centred in the intermediate region of the cord, and always passed close to at least one of the two intermediolateral cell columns (Fig. 3B).

The influence of arterial baroreceptors was studied by two approaches. In the first, the effects on activity of drug-induced rises of ≥ 30 mmHg in arterial pressure were tested for 14 cold-activated raphé-spinal neurons. In six cases there was a partial (30-90 %), but clear, reduction in activity (Fig. 7A). In the second approach, peri-event histograms of the neurons’ spike activity were triggered from the systolic peak in arterial pressure. Cardiac modulation of between 10 and 70 % was apparent in the histograms of 11/21 cold-activated raphé-spinal neurons (including six not tested by a rise in blood pressure; e.g. Fig. 7C).

Figure 7 Cold-activated raphé-spinal neurons: baroreceptor responses and cycle-triggered histograms

When ventilation-triggered histograms were constructed from the activity of six cold-activated raphé-spinal neurons, ventilatory modulation was apparent in all cases. In four, the modulation was quite weak, amounting to 6-22 % of mean activity (e.g. Fig. 7E). In two cases it was strong, the peak-to-peak variation reaching 70 % and 90 % of mean activity, respectively (e.g. Fig. 7F). Peak activity generally occurred around the onset of lung inflation. The ventilatory modulation did not appear to be a simple consequence of the ventilatory fluctuation in blood pressure because in 5/6 cases peak activity coincided with the rising phase of mean blood pressure rather than its nadir (Fig. 7E and F), and three of those six cells showed no cardiac modulation in their activity.

Figure 8 Cold-activated raphé-spinal neurons: responses to noxious tail pinch

Figure 9 Cold-activated raphé-spinal neurons: relation between EEG state and response to cooling

Figure 10 shows the averaged responses of a different raphé-spinal neuron to six consecutive cooling episodes. While the neuron showed an eight-fold increase in activity in response to the fall in skin temperature, the EEG showed no detectable change in either its pattern or the amplitude of its low frequency components. No consistent change in EEG pattern was observed to accompany cooling in any rat. Moreover, all six cold-activated neurons recorded with the EEG responded to cooling independently of the EEG pattern.

Figure 10 Cold-activated raphé-spinal neurons: averaged EEG and neuronal responses to skin cooling

This study has identified a population of spinally projecting neurons in the medullary raphé nuclei magnus and pallidus that responded to mild, non-noxious cooling. Several lines of circumstantial evidence suggest that these may be sympathetic premotor neurons controlling cold-defence responses. First, the sites where they were recorded - presumably their cell bodies - were in the appropriate location. This part of the raphé magnus/pallidus region contains sympathetic premotor neurons (Loewy, 1981; Strack et al. 1989), many of which can be labelled trans-synaptically by pseudorabies virus following its injection into either of two thermoregulatory tissues: the tail (Smith et al. 1998) or the interscapular BAT (Sved et al. 2001). In functional studies, moreover, it was shown that injection into this raphé region of chemical agents that stimulate cell bodies rather than fibres, strongly activated the sympathetic supplies to tail vessels (Rathner & McAllen, 1999) and BAT (Morrison, 1999), but had little effect on the splanchnic or renal sympathetic nerves (Morrison, 1999; Rathner & McAllen, 1999). Second, the cold-activated neurons studied here had long spinal axons with the potential to supply the full extent of the spinal sympathetic outflow, including the most caudal region that supplies the tail (Rathner & McAllen, 1998; Smith & Gilbey, 1998). Third, their exquisite sensitivity to mild environmental cooling matches that of the sympathetic outflow to the tail (Owens & McAllen, 2000), suggesting that they belong to a pathway that is tuned to respond to this ‘early warning’ signal and drive compensatory responses to cold. Fourth, these neurons’ patterns of response to other inputs, such as those from baroreceptors, nociceptors and the ventilatory cycle, were quite similar to those shown by peripheral cutaneous vasoconstrictor fibres (Häbler et al. 1993, 1994; McAllen & May, 1994), including those supplying the tail (Häbler et al. 1999; Chang et al. 2000; Owens & McAllen, 2000).

On the other hand, while these cells were activated antidromically by stimuli delivered across the intermediate region of the upper lumbar cord, we do not know where their axons terminated. We cannot exclude a possible role controlling other functions represented in the lower cord, such as modulating sensory, motor or parasympathetic neural pathways. We have no evidence to suggest that such functions would be affected by mild cooling of the trunk skin, but this could be tested in the future. On balance, however, we think it probable that the cold-activated raphé-spinal neurons constitute a descending thermoregulatory pathway, whose most likely final targets would be either cutaneous (especially tail) blood vessels, and/or BAT. A less likely target would be the pilomotor supply. Like the former two outflows, it is a pathway engaged in the body's response to cold; but unlike them, it is inactive in anaesthetized animals (Grosse & Jänig, 1976; Jänig, 1985). No evidence yet links it to raphé neurons.

An important issue to be resolved was whether these raphé-spinal neurons were indeed responding directly to skin cooling, or if their responses could have been secondary to factors such as the animal's state of arousal (anaesthetic depth). This issue arises because thermal stimulation of the skin can alter the EEG pattern of anaesthetized animals (Nakayama & Hardy, 1969), and many neurons in presumed ‘thermoregulatory’ brain regions change their activity with the animal's state of arousal (Grahn et al. 1989; Grahn & Heller, 1989; Kanosue et al. 1998; Leung & Mason, 1999; Martin-Cora et al. 2000). In the brainstem raphé, Grahn & Heller (1989) found that their sample of neurons responded to warming or cooling the skin, but did so only in parallel with changes in the EEG pattern, not within a single EEG state. When anaesthesia was deepened sufficiently to prevent EEG changes, the raphé neurons no longer responded to skin temperature (Grahn & Heller, 1989). These workers were led to doubt the existence of raphé neurons responding directly to skin temperature (Grahn & Heller, 1989). It is noteworthy, however, that temperature-responsive raphé neurons can still be recorded in decerebrate rats (Dickenson, 1977), where presumably the descending effects of forebrain arousal would have been removed. Nevertheless, it is clearly important to eliminate this confounding factor when investigating central thermoresponsive pathways, especially in animals with intact neuraxes.

Changes in the EEG pattern could not account for the thermoresponsiveness of neurons identified in the present study, however. The raphé-spinal neurons in question responded strongly to small drops in skin temperature, independently of the EEG pattern, which generally did not change. Two factors may account for the differences between the present findings and those of Grahn & Heller (1989). The first is that our animals were already quite deeply anaesthetized, and this is presumably why the EEG pattern remained stable (cf. Grahn & Heller, 1989). Secondly, we evidently selected for primarily thermoresponsive cells by working on a highly specific neural population: raphé-spinal cells in a region known to drive thermoregulatory effectors. Even so, only a minority of these selected neurons proved to be thermoresponsive. They could thus easily have been missed in a study where neurons were randomly sampled from an area encompassing this region and slightly more rostral parts of nucleus raphé magnus (Grahn & Heller, 1989).

A primary responsiveness to skin temperature does not exclude the possibility that cold-activated cells might respond additionally to arousal state or anaesthetic depth. Indeed, very limited evidence suggests that they may do this also. Supplementary anaesthetic doses activated two raphé-spinal cold-responsive neurons, while further synchronizing EEG activity in the case where this was measured. Conversely, these neurons’ activity was inhibited by noxious tail pinch - a presumably ‘arousing’ stimulus. This same combination of responses (without the thermal sensitivity) has previously been described for raphé neurons with putative antinociceptive function (Leung & Mason, 1999). Functionally, this type of convergence between homeostatic (thermal) and ‘arousal’ signals is what one might expect to find in a descending pathway that controls cutaneous blood flow (Blessing, 1997).

Additional features that suggest that at least some of the identified population of raphé-spinal cold-activated neurons may drive a cutaneous vasomotor pathway are their partial sensitivity to baroreceptors and their, apparently independent, modulation by the ventilatory cycle. A relatively minor degree of barosensitivity corresponds to the properties of cutaneous vasoconstrictor neurons (Häbler et al. 1994), including those that supply the rat's tail (Johnson & Gilbey, 1998; Häbler et al. 1999; Owens & McAllen, 2000). The situation is less clear when comparisons are made with the sympathetic supply to BAT, because here the information is based on whole-nerve recordings rather than individual fibres. A low level of resting activity is present in the BAT nerve of warm rats, and this does show cardiac modulation (cold-induced activity does not) (Morrison et al. 1999). While this activity has been attributed to a minority of vasomotor fibres present in the nerve (Morrison et al. 1999), the possibility that this activity is present in the fibres supplying BAT cells has not been ruled out.

Both the supplies to BAT and to tail blood vessels may be modulated by central respiratory drive (Häbler et al. 1999; Chang et al. 1999), and tail sympathetic activity can be entrained directly by the artificial ventilation cycle (Chang et al. 1999). The ventilatory modulation of cold-activated raphé-spinal neurons could have been caused by either or both of these factors because, although not recorded in these experiments, central respiratory drive is likely to have been entrained by the ventilator. Either way, however, the existence of such rhythmicity may be taken as a further similarity between these raphé-spinal neurons and their putative sympathetic targets.

At first sight a puzzling finding is that virtually all raphé-spinal cold-activated neurons responded to noxious tail pinch, most being inhibited. Perhaps more surprisingly, other raphé-spinal cells with the same response to cooling were excited by tail pinch. However, the inhibitory response of the majority of these neurons to noxious stimulation is in line with the behaviour of sympathetic fibres supplying the tail, which are also silenced by tail pinch (Owens & McAllen, 2000). In fact, inhibition by local (but not distant) noxious stimuli is a general feature of cutaneous vasoconstrictor fibres in anaesthetized animals (Jänig, 1985; Häbler et al. 1994). While it is not clear how the BAT sympathetic supply responds to tail pinch in anaesthetized rats, the net effect on BAT activity in conscious rats is strong excitation (Romanovsky et al. 1997). It is thus at least possible that the cold-activated raphé-spinal neurons that are inhibited by tail pinch drive the supply to cutaneous vessels in or near the tail, while those that are excited by tail pinch drive the supply to BAT or perhaps that to more rostrally located cutaneous vessels.

In view of the proposed role of nucleus raphé magnus in modulating nociceptive transmission (Besson et al. 1981), it was not a surprise to find that many neurons in our raphé-spinal population (which included both cold-activated, and thermally unresponsive cells) were excited or inhibited by noxious tail pinch. Caution is needed before making a direct link between them and the ‘on’ and ‘off’ cells characterized by Fields and co-workers (Fields et al. 1983; Vanegas et al. 1984), because our experimental conditions were significantly different. We used urethane anaesthesia of sufficient depth to abolish withdrawal reflexes, while ‘on’ and ‘off’ cells have been defined under anaesthetic regimes (barbiturate or volatile) that preserved withdrawal or tail-flick responses to noxious stimulation (Morgan & Fields, 1994). Nevertheless, the similarities give rise to the question: could their responsiveness to noxious stimuli be an indication that cold-activated raphé-spinal neurons also play a role in modulating spinal afferent transmission? In line with this possibility, they were located near the caudal end of the region whose cells exert antinociceptive actions in the spinal dorsal horn (Morgan et al. 1994), and we do not yet know where their axons terminate. The existence of collateral branches to the dorsal horn (cf. Allen & Cechetto, 1994) cannot be excluded. If such connections did exist, then one would predict that remote cooling of the body surface should influence spinal nociceptive transmission. Indirect evidence suggests that this might indeed happen: cooler ambient temperatures have been reported to reduce the perceived intensity of noxious thermal stimuli (Strigo et al. 2000). However, it is presently unclear whether such effects are attributable to descending modulation of spinal transmission or to other factors, such as interactions between afferent inputs within the spinal cord or at higher levels of the neuraxis.

With regard to the possible modulation of thermal afferent transmission by raphé-spinal neurons, the evidence is indirect and divided. Electrical stimulation of nucleus raphé magnus was found by Dawson and colleagues (1981) to have no effect on thermosensory afferent transmission in the trigeminal system of cats or rats. On the other hand, Sato (1993) observed that such stimuli strongly suppressed the responses of both warm- and cold-sensitive dorsal horn cells responding to rat scrotal skin temperature.

Raphé-spinal neurons with presumed autonomic functions have been recorded in several previous studies (e.g. Morrison & Gebber, 1985; Barman & Gebber, 1988, 1997; Gilbey et al. 1995; Pilowsky et al. 1995), although none has directly tested these neurons for thermal responses. In rats, two studies have reported on spinally projecting neurons recorded from raphé regions a little caudal to that explored in the present study. Gilbey and colleagues (1995) recorded from raphé-spinal neurons with a median conduction velocity of 3.8 m s⁻¹, most of which showed respiratory modulation (of various patterns), but none of which was modulated by the cardiac cycle. These cells were considered likely to innervate sympathetic neurons because they could be activated antidromically from the region of the thoracic intermediolateral cell column (Gilbey et al. 1995). In a smaller study on the same region, Pilowsky and colleagues found raphé-spinal neurons with a predominantly inspiratory modulation pattern, but these cells also showed barosensitivity and were modulated by the cardiac cycle (Pilowsky et al. 1995). Their axons conducted at 2-10 m s⁻¹. Unlike the neurons recorded in the present study, however, they were insensitive to noxious tail pinch (Pilowsky et al. 1995). It is thus not clear whether either of these studies would have encountered the neurons described in the present paper.

Without the knowledge of axonal projections, previous studies on thermoresponsive neurons in the medullary raphé have generally assumed that they form part of an ascending pathway to the hypothalamus and thalamus (e.g. Nakayama & Hardy, 1969; Dickenson, 1977; Young & Dawson, 1987). Such pathways evidently do exist (Taylor, 1982; Berner et al. 1999), and we cannot exclude the possible existence of ascending collateral connections in the present neuronal population; but a descending pathway is now proven. Interestingly, a cold-activated raphé-parasympathetic pathway has been described by Yang, Taché and colleagues (Yang et al. 1993, 1994). This involves a projection by TRH-containing raphé neurons to the dorsal motor nucleus of the vagus, where they act to stimulate gastric secretion (Yang et al. 1993, 1994).

The previous study that most closely matches the present findings is that of Dickenson (1977), who recorded from a population of raphé neurons that, like those of the present study, responded to quite small changes in trunk skin temperature; but unlike those of the present study, they were unresponsive to noxious pinching. These cells were considered to be serotonergic, because they were inhibited by systemic lysergic acid diethylamide (Dickenson, 1977). The raphé-spinal cold-activated neurons identified in the present study showed neither the slow, regular discharge patterns characteristic of serotonergic cells (Mason, 1997; Martin-Cora et al. 2000) nor axonal conduction velocities within the range of unmyelinated fibres, to which group virtually all serotonergic axons in the rat's spinal cord have been reported to belong (Basbaum et al. 1988), although it should be noted that our use of narrow stimulus widths may have reduced the chance of activating unmyelinated axons. Arguing against this view, however, the serotonin-selective neurotoxin 5,7-dihydroxytryptamine has been found to deplete raphé-spinal neurons not only with unmyelinated axons, but also those with myelinated axons conducting at up to 6 m s⁻¹ (Wessendorf et al. 1981). Sixty per cent of the present population of cold-activated neurons conducted faster than this. Overall, it thus seems likely that most, if not all, of the present population were non-serotonergic neurons, but direct experiments will be needed to settle the issue.

Since a number of the raphé-spinal neurons recorded in this study responded to skin temperature independently of core temperature, a novel possibility arises. They might receive this signal relatively directly, rather than as part of the integrated output from the preoptic/anterior hypothalamic area, and so form a ‘short loop’ (spino-bulbo-spinal) thermoregulatory reflex. In line with this possibility, Dickenson (1977) and Taylor (1982) reported that raphé neurons responding to skin temperature could also be found in rats after mid-collicular decerebration. In the intact state, however, other neurons within this population evidently also receive information on core temperature, presumably directly or indirectly from the thermosensitive cell groups in the preoptic/anterior hypothalamic area. Direct descending connections that might serve this function have recently been traced from the preoptic area to the medullary raphé using neuroanatomical methods (Murphy et al. 1999). Descending excitatory and inhibitory pathways that relay through the midbrain tegmentum may also be involved (Zhang et al. 1997). This specialized group of spinally projecting neurons in the nuclei raphé magnus and pallidus may thus integrate thermal and other signals to provide appropriate descending drive to sympathetic outflows involved in cold defence, such as the vasomotor supply to the tail.

We are most grateful to David Trevaks for his expert help with technical and computing aspects of this study, to Professor Tony Dickenson for allowing us to copy his water jacket design, to Alex Sigmund for making it and to Dr Kazuyuki Kanosue for his participation in two experiments. We also thank the National Heart Foundation, the National Health and Medical Research Council of Australia (block grant 983001) and the Ronald Geoffrey Arnott Foundation for supporting this work.