The cellular coding of temperature in the mammalian cortex – Nature

Mice

All experimental procedures were carried out in accordance with the State of Berlin Animal Welfare requirements and European animal welfare law. Male and female mice older than 2 months were maintained on a 12:12 h light–dark cycle with experiments carried out during the light phase of the cycle. Mice were housed in groups at 22 ± 2 °C temperature and 55 ± 10% humidity with ad libitum access to food and water unless stated otherwise. Thy1-GCaMP6s (C57BL/6J-Tg(Thy1-GCaMP6s)GP4.3Dkim/J) mice were used for calcium imaging and for behavioural experiments (The Jackson Laboratory, stock no. 024275)^39,40. VGAT-ChR2 (B6.Cg-Tg(Slc32a1-COP4*H134R/EYFP)8Gfng/J) mice were used for behavioural experiments (The Jackson Laboratory, stock no. 014548)⁴¹. For awake experiments, mice were gradually habituated to head and paw fixation and tilted for optical access to pIC or S1.

Surgery

Mice were deeply anaesthetized (ketamine 120 mg kg⁻¹ and xylazine 10 mg kg⁻¹) and injected subcutaneously with dexamethasone sodium phosphate (2 mg kg⁻¹) to prevent cerebral oedema, as well as metamizol (200 mg kg⁻¹). Anaesthetized mice were fixed with a nose-clamp, eye gel (Vidisic, Bausch + Lomb) was applied to both eyes and body temperature was maintained at 37 °C with a heating pad and rectal probe. After surgery, mice received a subcutaneous injection of warm sterile saline solution and were placed on a warm blanket until they awoke from anaesthesia. To avoid post-operative pain, metamizole was dissolved in the drinking water for 3 days post-surgery.

Implantation of cranial windows

To implant a window over pIC, mice were anaesthetized as above, and then the parietal and temporal bones of the left hemisphere were exposed and the left temporalis muscle was carefully separated from the temporal bone and partially removed. An approximately 3 × 3 mm craniotomy was carried out over pIC as identified by local anatomical landmarks (rhinal vein, middle cerebral artery, zygomatic bone). To implant a window over primary somatosensory forepaw cortex (S1) a 3-mm-diameter craniotomy was carried out over the left hemisphere S1 identified by the anatomical location respect to bregma (1.5–2.5 mm lateral and 0.5–1 mm anterior). To implant windows over pIC and S1 in the same mouse, we carried out two 3-mm craniotomies. The cortical surfaces were then rinsed with Ringer’s solution and two glass coverslips were placed on the surface of the cortex. The lower glass (diameter 3 mm or 3 × 3 mm) and the upper glass (diameter 4 or 5 mm) were glued together with optical adhesive (NOA 61, Norland Products). Dental cement and cyanoacrylate glue were used to attach a metal head post and the upper coverslip to the skull. Following pIC surgery, the skin lying over the temporal muscle was sutured.

Clear skull preparation

To gain simultaneous optical access to S1 and pIC, we used a through-skull preparation (Fig. 1f,g). Under deep isoflurane anaesthesia, the left temporalis muscle was partially removed and the dorsal surface of the skull was cleared of skin and periosteum. Finally, the exposed skull was sealed with cyanoacrylate glue.

Cortical pharmacology

For pharmacological inactivation, ≈500-μm-diameter craniotomies were drilled over cool (32–22 °C)-responsive areas in S1 and pIC in mice anaesthetized with isoflurane (1.5–2% in O₂). Both regions were identified functionally using the widefield calcium imaging response to thermal stimuli. Following the craniotomy, the dura was covered with transparent silicone gel (3-4680, Dow Corning). A 300 nl volume of muscimol or Ringer’s solution was injected at a rate of 100 nl min⁻¹ 300–500 μm below the pial surface using a pulled glass pipette and a hydraulic injection system (MO-10, Narashige). Muscimol (Abcam, ab120094) was dissolved in Ringer’s solution to a concentration of 5 mM. Imaging sessions were carried out >10 min following the end of Ringer’s solution or muscimol injection (Fig. 1f,g).

Sensory stimulation

Thermal

For widefield imaging, behavioural and electrophysiological experiments, thermal stimuli were delivered by an 8 × 8 mm Peltier element regulated by a feedback-controlled stimulator (Yale Medical School or a custom-made device, ESYS GmbH Berlin). Different thermal stimuli (2-s duration with an onset ramp speed of 20 °C s⁻¹) were interleaved randomly and delivered every 30 or 60 s. During paw stimulation, we took care to place the forepaw or hindpaw pad glabrous skin into direct contact with the centre of the Peltier element. During face stimulation, the Peltier surface was positioned on the orofacial region of the snout. For single-cell two-photon experiments, thermal stimuli were delivered using a Peltier-based thermal stimulator (QST-lab). Thermal stimuli of different durations (2, 5, 10 s) or with different onset ramp speeds (approximately 130, 10, 3.3, 2, 1 °C s⁻¹) were interleaved randomly and delivered every 30 or 60 s.

Acoustic

Acoustic stimuli of 8 kHz, ≈65 dB sound pressure level and 1-s duration were delivered every 60 s using a loud speaker (Visaton) positioned 10 cm from the contralateral ear. The frequency of 8 kHz was chosen because it is well represented in the pIC auditory field²³. For behavioural training, a 14 -kHz tone (≈65 dB, 1-s duration) was chosen as it is weakly represented in the pIC auditory field²³.

Vibrotactile

Vibrotactile stimuli (100 Hz) were generated by a piezoelectric actuator (PL127, PI) equipped with a 5-cm glass rod bent at the tip. The contralateral forepaw or hindpaw was tethered to a rigid support with a hole (diameter 5 mm) to allow the bent tip of the glass rod to stimulate the centre of the palm of the paw. In some experiments vibrotactile stimuli were generated using a small motor (Pololu shaftless vibration motor 8 × 3.4 mm) positioned on top of the paw tethered to the Peltier. Vibrotactile stimuli had a 0.5-s duration and were delivered every 30–60 s.

Widefield imaging

Setup

Imaging was carried out with a Leica MZ10F stereomicroscope. Blue excitation light (470 nm) was emitted from a LED (pE-300, CoolLED) and bandpass filtered (470/40 nm). Emission light was bandpass filtered (525/50 nm) before recording with a scientific complementary metal–oxide–semiconductor camera (ORCA-Flash4.O LT, Hamamatsu). Images were acquired at a rate of 20 Hz and 35 ms exposure time. Frame size was 1,024 × 1,024 pixels (2 × 2 binning); for cranial window preparation the field of view was about 2.7 × 2.7 mm and for the cleared skull preparation the field of view was about 6.7 × 6.7 mm. Acquisition trials lasted 10–15 s and had an inter-trial-interval time of 30–60 s. During pIC imaging, animals were tilted to allow optical access. As a result of physical constraints, pIC and S1 widefield imaging was carried out in different animals and sessions, except during the cortical pharmacological inactivation experiments (Fig. 1f,g). In this case, we used a ‘clear skull’ preparation and imaged pIC and S1 simultaneously. The setup was always adjusted to allow the paw to be placed in a similar position to all other experiments. Data acquisition was controlled by custom-written code (Python).

Widefield image analysis

Videos were motion corrected, and areas in the recordings outside the brain were masked. We calculated the relative change in fluorescence as ΔF/F = (F(t) − F₀)/F₀, in which F₀ is the 15th percentile of the activity in the trial. The average activity of the last 500 ms before stimulus was subtracted from ΔF/F. The activity of a given region was quantified by averaging the fluorescence in a region of interest (ROI; diameter 15 pixels) before calculating ΔF/F. When placing ROIs, we avoided visible blood vessels, but otherwise positioned the ROI at the area with the peak stimulus-driven response. The ROI positions were kept the same over days and recording sessions by comparing the position relative to the blood vessels. Auditory stimuli activated several fields in the auditory cortex and one area in pIC (insular auditory field).

Somatotopic maps

Somatotopic maps of sensory input to pIC (Fig. 1) and S1 (Extended Data Fig. 2) are grand averages across mice and trials. Not all stimuli were delivered to all mice. Using the ROI of the thermal forepaw response as a reference location, ΔF/F videos for all trials were aligned by translation. The grand averages were smoothed with Gaussian filter (σ = 20 pixels) and thresholded to show only peak activity (>85% of the maximum activity for thermal and >90% for tactile stimulation). For pIC, because of the simultaneous responses in the auditory insular field and auditory cortex to sound stimuli, responses were analysed independently. We carried out this analysis only on data with grand average peak response amplitude >3% ΔF/F.

Response onset estimation

We defined the onset of the widefield response as the peak of the second derivative of the widefield signal in a time window from the start of stimulus to the peak of the first derivative. If negative values of the first derivative appeared after stimulus start, the time window was from the last of such negative values. We carried out this analysis only on data with a peak response amplitude >3% ΔF/F.

Two-photon imaging

Setup

Fluorescence signals were recorded using a two-photon microscope (ThorLabs Bergamo II, 12 kHz scanner) with a Nikon 16× water-immersion objective (NA 0.8) giving a field of view of 540 × 540 µm. The scope was operated using ThorImageLS software (v4.0.2019.8191, Thorlabs). In most experiments the microscope was rotated by 45° from vertical. Owing to physical constraints, pIC and S1 two-photon imaging was carried out in different animals and sessions. The excitation laser (InSight DeepSee, Spectra-Physics) was tuned to 940 nm, and the power never exceeded 100 mW. Emitted photons were bandpass filtered 525/50 (green) onto a GaAsP photomultiplier tube. Multi-plane (512 × 512 pixels) acquisition was controlled by a fast piezoelectric objective scanner, with planes spaced 45 μm apart in depth. Seven planes were acquired sequentially, and the scanning of the entire stack was repeated at about 5 Hz. Beam turnarounds at the edges of the image were blanked. Acquisition trials lasted 26 s and had an inter-trial interval of 30–60 s. Stimuli generation and hardware synchronization were carried out on a computer with a National Instrument card running custom-written Python code.

Two-photon analysis

Motion correction of data, identification of putative neurons and calculation of ΔF/F was carried out using the Suite2p package (v0.9.3) in Python⁴². Identified neurons were manually verified by visual examination of the traces and the spatial footprints of the neurons. A Savitzky–Golay filter was applied to traces presented in figures for visual purpose alone.

Criteria for responsive neurons

Neurons with a significant increase in activity to 10 °C cooling and warming stimuli were included for further analysis. Cool and warm neurons were defined as those significantly responding to either 10 °C cooling or warming stimuli. Broadly tuned neurons had significant response to both cooling and warming stimuli. To identify an increase in response amplitude, the activity in a time window before stimulus onset was compared with the activity in a time window during stimulus. A neuron was identified as being responsive if the change in activity was significantly larger (estimated by bootstrapping) than two times the noise level of the neuron measured by the interquartile range of activity before the stimuli.

The thermal bias index (Fig. 2d,e) for a thermally responding neuron is the normalized difference between its cool and warm responses, (warm − cool)/(warm + cool), in which ‘warm’ is the average peak response to 32–42 °C and ‘cool’ is the average peak response to 32–22 °C. Grand average (Fig. 3b,c) calculated from all significantly responding neurons shown in Fig. 2 and includes 470 neurons with significant responses to cooling and 401 to warming. Of these, 125 responded to both. The duration index (Fig. 3d) measures the change in the fluorescence level between the initial peak value (f_init) and the end (f_end) of a 2-s thermal stimulus, (f_init − f_end)/f_init. The adaptation index (Fig. 3g) was calculated as the difference in the maximum fluorescence level during fast thermal stimulation (about 130 °C s⁻¹; f_fast) and during slow thermal stimulation (≈1 °C s⁻¹; f_slow), (f_fast − f_slow)/f_fast.

Behavioural experiments

Setup

Head-fixed mice (VGAT-ChR2 and Thy1-GCaMP6s) were implanted with a glass window over pIC and trained on a go/no-go stimulus detection paradigm. Mice were rewarded with a water drop (about 2 µl) if they reported a randomly timed thermal stimulus with at least one lick of a water spout (capacitance sensor) within a window of opportunity (from the start of thermal stimulus to start of offset ramp). Correct rejections were not rewarded and incorrect responses were not punished, although premature licking in the 5 s before the stimulus onset would postpone the next trial by 5 s. Mice were water restricted and their weight was monitored daily. For behavioural training, a 200-μm-diameter, 0.22-NA optic fibre (Thorlabs) was coupled to an LED (470 Plexon LED source) and placed on the coverslip surface orthogonal to the thermal region of the pIC or S1 using the blood vessel pattern. In a subset of mice, cranial windows were implanted above both S1 and pIC, and LEDs were placed above both areas (Extended Data Fig. 7). Control of behavioural training and data collection was carried out using either custom-written Labview software (16.0f5, National Instruments, USA) or the Bpod system (1.8.2, Sanworks, USA).

Thermal perception task

Thermal perception (Fig. 5) training involved several stages: (1) Free access to water rewards from the water spout; (2) automatic water rewards paired to the presentation of 10 °C, 2-s cooling and warming stimuli from AT 32 °C; (3) training to report a 10 °C cooling and warming stimuli by licking for a water reward within a 2 s window from the start to the end of the plateau phase of the stimulus; thermal stimuli trials and catch trials were presented with a ratio of 1:1 with an inter-stimulus-interval of 15–20 s; (4) once a hit rate >70% and false alarm rate <30% had been reached, 4 randomized stimulus amplitudes were used from AT 32 °C (cooling, 22 °C, 30 °C, 31 °C and 31.5 °C; warming 42 °C, 40 °C, 38 °C and 34 °C) with a 4:2 ratio of stimulus trials to catch trials; (5) once a hit rate of >70% and false alarm rate <30% had been achieved for at least 1 amplitude, in the next session we presented the same 4 thermal stimuli with 1 catch trial with LED on or off (on/off trials) with a ratio of 2:1 (randomized). The LED was on for the entire duration of the stimulus and delivered at 20 Hz, 50% duty cycle with a power of 12.5 mW (measured at tip of fibre). Before the start of stage (4) or (5) sessions, mice were exposed to a brief session of the stage (3) protocol to quench initial thirst.

Thermal discrimination task

The thermal discrimination task training (Extended Data Fig. 8) involved several stages: (1) free access to water rewards from the water spout; (2) automatic water rewards paired to the presentation of 14-kHz, ≈65-dB, 0.5-s acoustic stimulus; (3) training to report 14-kHz stimulus by licking for a water reward within a 1.4-s window from the start of the stimulus (go trial); acoustic stimuli (14 kHz, about 65 dB) delivered 0.6 s after the beginning of a 2-s thermal stimulus (either cooling or warming) (no-go trial) and catch trials (no stimuli) were not rewarded; go, no-go and catch trials were presented with a ratio of 1:1:1 with an inter-stimulus-interval of 15–20 s; training was continued until go trial hit rate >70%, no-go false alarm rate around 50% and false alarm catch rate <30%; (4) once a mouse learned to discriminate between acoustic stimulus alone (go) and acoustic stimulus presented together with thermal stimuli (no-go), in the next session we presented the same stimuli interleaved at a ratio of 2:1 into trials with the LED off (off) and trials with the LED on (on). The LED was on for the entire duration of the thermal stimulus and delivered at 20 Hz, 50% duty cycle with a power of 12.5 mW (measured at tip of fibre). Before the start of stage (4) sessions, mice were exposed to a brief session of the stage (3) protocol to quench initial thirst.

Acoustic perception

The behaviour training to report acoustic stimuli (Extended Data Fig. 8) had the same structure as the thermal training above, but used a 14-kHz, ≈65-dB, 2-s acoustic stimulus.

Free licking

To monitor the impact of pIC optogenetic manipulation on licking behaviour, water restricted mice were allowed to freely lick from a lick spout with continuous rewards. Licking rates during the 2 s before the onset of the light stimulus (Off) were compared with those during the 2 s after the onset of the light stimulus (On). Light stimuli were delivered for 5 s at 12.5 mW, 20 Hz and 50% duty cycle (Extended Data Fig. 8).

Datasets and analysis

Mice were randomly allocated to be trained first to report cooling or warming. In a second training round, they were trained to report the opposite stimulus. Data measuring the thermal perceptual ability of mice without optogenetic manipulation (Fig. 5b–d) were generated from the same mice used in optogenetic manipulations (Fig. 5e,f), the day before optogenetic testing. The effect of the optogenetic manipulation (bottom panels of Fig. 5e,f and Extended Data Fig. 7) was quantified by the mean change in the percentage of lick trials.

In vivo electrophysiology

Electrophysiological recordings for measuring the impact of optogenetic inhibition on thermal responses were carried out in VGAT-ChR2 mice implanted with a head post. Mice were habituated for head and paw fixation. On the day of the recording, the mouse was briefly anaesthetized (about 30 min, 1% isoflurane) and a craniotomy of about 2 mm diameter was carried out over the pIC. The dura was removed in the ventral part of the craniotomy and the brain was covered with Kwik-Cast silicon (World Precision Instruments). The animal was placed in its home cage to recover from anaesthesia for at least 2–3 h before head fixation in the recording setup.

We carried out 7 recordings in 4 mice using 32-channel silicon probes (Neuronexus) in either a linear or four-shank ‘Buzsaki32’ configuration. In some experiments, the tips of the probe were covered with DiI (Sigma). The probe was inserted orthogonal to the pIC with a micromanipulator (Luigs&Neumann) at 1–2 µm s⁻¹. Recordings were acquired at 30 kHz with a Neuralynx system (Cheetah). The extracellular recordings were spike sorted using Kilosort (version 2)⁴³. The data were collected to evaluate the effect of ChR2 on neural activity, so we carried out limited manual sorting to exclude obvious noise but did not take any steps to exclude multi-unit activity. Units with a mean activity during stimulus representation exceeding a threshold of 2 s.d. above the background were considered to be thermally responsive.

Two-second-long thermal stimuli of 10 °C cooling and warming were interleaved randomly and delivered every 30 s. The LED fibre (200-μm-diameter, 0.22-NA optic fibre, Thorlabs) was tilted slightly to accommodate the silicone probe. The LED was on for the entire duration of the thermal stimulus and delivered at 20 Hz, 50% duty cycle with a power of 12.5 mW (measured at tip of fibre).

Histology

To label the thermal cortical representation, animals were deeply anaesthetized by intraperitoneal injection of ketamine and xylazine. The centre of the response area was marked with a glass pipette painted with fluorescent dye (DiI, 5 mg ml⁻¹). Mice were then perfused transcardially using cold phosphate-buffered saline (PBS; 0.1 M, pH 7.4) and fixed with paraformaldehyde (PFA; 2% in 0.1 M PBS). The brain was removed and kept for 1–5 h in 2% PFA. After washing with PBS, the hemispheres were separated and cortices were removed and flattened between two glass slides separated by a spacer of 1–2 mm. Glass slides were weighed down for approximately 3–8 h at 4 °C in 2% PFA. After washing with PBS, 70–80-μm sections were cut on a Vibratome (Leica VT1000s). Sections were stained for cytochrome oxidase activity (2 mg cytochrome c, 6 mg diaminobenzidine). After the staining procedure, sections were mounted on glass slides with Mowiol mounting medium. Images were acquired with a Zeiss microscope (AX10) using a 5× objective and processed using Fiji (ImageJ, NIH, USA). Area borders were manually delineated following the contrast of the cytochrome oxidase stain and are comparable to previously reported data⁴⁴. Coronal sections of 50 μm in thickness of PFA-fixed mouse brains were stained for 48 h with the following primary antibodies: mouse anti-Gad67 (catalogue number MAB5406; clone 1G10.2; Millipore; 1:800); chicken anti-GFP (catalogue number ab13970; Abcam; 1:250); mouse anti-NeuN (catalogue number MAB377; clone A60; Millipore; 1:100). The secondary antibodies Cy3 goat anti-mouse (A-21422; Invitrogen; 1:250) and A488 goat anti-chicken (A-11039; Invitrogen; 1:250) were incubated for a few hours at room temperature. Cell counting was carried out manually in Fiji using the Cell Counter plug-in on epifluorescence images acquired using a Zeiss microscope (AX10) using a 10× objective.

Data analysis and statistics

No statistical methods were used to predetermine sample sizes, but our sample sizes were similar to those used in previous publications. Experimenters were not blind to trial order, but trial order was randomized during the experiment. All data analysis was carried out using Python. Uncertainty of means are reported with standard error of the mean. Bootstrapping was used for estimating central 95% confidence intervals using 5,000 resamples. Some histograms in Fig. 3 and Extended Data Fig. 4 do not show outlier data points. In Fig. 1f,g, we used two-sided paired t-test. For Fig. 1f: cool S1, t = −3.78, P = 0.0323; warm S1, t = −0.17, P = 0.8767; cool pIC, t = 0.44, P = 0.6929; warm pIC, t = −0.45, P = 0.6839. For Fig. 1g: cool S1, t = 0.55, P = 0.6182; warm S1, t = −0.68, P = 0.5454; cool pIC, t = −3.61, P = 0.0366; warm pIC, t = −8.52, P = 0.0034.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.