A daily temperature rhythm in the human brain predicts survival after brain injury

Abstract

Patients undergo interventions to achieve a 'normal' brain temperature; a parameter that remains undefined for humans. The profound sensitivity of neuronal function to temperature implies the brain should be isothermal, but observations from patients and non-human primates suggest significant spatiotemporal variation. We aimed to determine the clinical relevance of brain temperature in patients by establishing how much it varies in healthy adults. We retrospectively screened data for all patients recruited to the Collaborative European NeuroTrauma Effectiveness Research in Traumatic Brain Injury (CENTER-TBI) High Resolution Intensive Care Unit Sub-Study. Only patients with direct brain temperature measurements and without targeted temperature management were included. To interpret patient analyses, we prospectively recruited 40 healthy adults (20 males, 20 females, 20-40 years) for brain thermometry using magnetic resonance spectroscopy. Participants were scanned in the morning, afternoon, and late evening of a single day. In patients (n = 114), brain temperature ranged from 32.6 to 42.3°C and mean brain temperature (38.5 ± 0.8°C) exceeded body temperature (37.5 ± 0.5°C, P < 0.0001). Of 100 patients eligible for brain temperature rhythm analysis, 25 displayed a daily rhythm, and the brain temperature range decreased in older patients (P = 0.018). In healthy participants, brain temperature ranged from 36.1 to 40.9°C; mean brain temperature (38.5 ± 0.4°C) exceeded oral temperature (36.0 ± 0.5°C) and was 0.36°C higher in luteal females relative to follicular females and males (P = 0.0006 and P < 0.0001, respectively). Temperature increased with age, most notably in deep brain regions (0.6°C over 20 years, P = 0.0002), and varied spatially by 2.41 ± 0.46°C with highest temperatures in the thalamus. Brain temperature varied by time of day, especially in deep regions (0.86°C, P = 0.0001), and was lowest at night. From the healthy data we built HEATWAVE-a 4D map of human brain temperature. Testing the clinical relevance of HEATWAVE in patients, we found that lack of a daily brain temperature rhythm increased the odds of death in intensive care 21-fold (P = 0.016), whilst absolute temperature maxima or minima did not predict outcome. A warmer mean brain temperature was associated with survival (P = 0.035), however, and ageing by 10 years increased the odds of death 11-fold (P = 0.0002). Human brain temperature is higher and varies more than previously assumed-by age, sex, menstrual cycle, brain region, and time of day. This has major implications for temperature monitoring and management, with daily brain temperature rhythmicity emerging as one of the strongest single predictors of survival after brain injury. We conclude that daily rhythmic brain temperature variation-not absolute brain temperature-is one way in which human brain physiology may be distinguished from pathophysiology.

Keywords: brain injury; brain temperature; brain thermometry; daily; mortality.

Figure 1

Disrupted temperature rhythms in brain-injured patients. (A) Violin plot of patient T_Br and T_Bo according to sex. Mean T_Br significantly greater than mean T_Bo, mixed effects analysis with Tukey’s for multiple comparisons (****P < 0.0001, females, n = 20 for T_Br and n = 17 for T_Bo; males, n = 85 for T_Br and n = 77 for T_Bo). See also Supplementary Fig. 2. (B) Representative raw data from a 62-year-old female patient (left) showing daily variation in T_Br and T_Bo, with T_Br consistently higher than T_Bo and both parameters in same phase. T_Br sampled once per minute; peak at 05:28 and nadir at 16:12 highlighting inversion of phase relationship with external day-night cycle under intensive care conditions (external time in 24-h clock format). Representative raw data from a 42-year-old male patient (right) showing lack of a daily rhythm in both T_Br and T_Bo. See also Supplementary Fig. 3. (C) Rose plots (left) showing timings of temperature maxima and minima in TBI patients (24-h clock format). For all variables, the null hypothesis of a uniform distribution could not be rejected (Rayleigh test of uniformity; T_BrMax, n = 104, P = 0.20; T_BrMin, n = 104, P = 0.16; T_BoMax, n = 101, P = 0.99; T_BoMin, n = 101, P = 0.86). Contrast with healthy subject rectal temperature data from publicly-available database (right). (D) Linear regression of patient temperature ranges with age; reduction in temperature range significant for brain (slope of −0.016 significantly different from zero; 95% CI −0.029 to 0.003). Shaded areas represent 95% CIs for lines of best fit. Max = maximum; Min = minimum; NS = not significant. See also Supplementary Fig. 4.

Figure 2

Chronotype-controlled temperature variations in healthy adults. (A) Prospective study profile and workflow (see the ‘Materials and methods’ section). (B) Representative actogram displaying typical actigraphy over 1 week from one male volunteer. Horizontal panels represent consecutive days. Note absence of light exposure and activity, with increase in skin temperature during sleep time (activity also absent when device was ‘off-wrist’). Social jet lag refers to large delay in sleeping schedule due to social activities on two consecutive days, highlighted (red box). See also Supplementary Fig. 5. (C) Group-averaged mean ± SEM data for distal skin (wrist) temperature, total light exposure and activity by sex during actigraphy week (left). Females n = 20, males n = 20. Associated rose plots with circular means (acrophases ± SD) displayed (right). For each data type, radial uniformity was rejected for both groups (Rayleigh uniformity test P < 0.0001) and there were no significant differences in circular mean between them (Watson’s two-sample test for homogeneity, P > 0.1). See also Supplementary Fig. 6. (D) Linear mixed modelling results for oral temperature. Solid blue lines represent model fits, shaded areas and double-ended error bars represent 95% CIs, dark grey circles display residuals (single temperature data-points) and smoothed dashed yellow lines represent partial residuals. The x-axis for time summarizes the continuous variable of time distance since the participant’s MSF_sc (proportion of a linearized unit circle, where 0 = MSF_sc and 1 = 24 h). Note time-dependent trend but lack of significant diurnal variation in oral temperature, likely reflecting inherent practical challenges of obtaining accurate oral temperature readings in human subjects. See also Supplementary Fig. 7.

Figure 3

Human brain temperature is spatially heterogenous. (A) Representative annotated MR images to show MRS extraction protocol immediately after whole-brain structural acquisition. T₂-weighted axial (top left) and T₁-weighted mid-sagittal (top right) image showing multivoxel MRS overlay for more superficial brain regions including cerebral grey and white matter; note positioning superior to corpus callosum. From this multivoxel acquisition, MRS data were extracted from each of the numbered voxels individually; for the final statistical model, the whole cerebral region was split into four superficial groups of voxels (Sup 1–4, depicted as separate colours in the overlay, from medial to lateral). T₁-weighted axial, sagittal and coronal images (bottom three images from left side, respectively) showing orthogonal positioning of single voxel in right hypothalamus (yellow box). T₁-weighted coronal image (bottom right) showing positioning of single MRS voxel in right thalamus (yellow box). See also Supplementary Fig. 8. (B) Linear mixed modelling results for global T_Br by sex, age, brain region and BMI, and for deep T_Br (including thalamus and hypothalamus) by sex and age. Solid red lines represent model fits, shaded areas and double-ended error bars represent 95% CIs, dark grey circles display residuals (single temperature data-points) and smoothed dashed yellow lines represent partial residuals. For sex, P-value reflects comparisons of each group with luteal females. For brain region, P-value represents comparisons of each region relative to superficial region 1 (parasagittal group of voxels). Sup 1–4 = superficial brain regions 1–4 from medial to lateral; Hypo = hypothalamus; Thal = thalamus. See also Supplementary Fig. 9.

Figure 4

Healthy brain temperature varies by time of day. (A) Snapshot 3D maps of T_Br at each data collection point. Inferno colour scale is used to assign a temperature to each tissue voxel, to 0.1°C resolution. Aggregate temperatures are displayed in each voxel for luteal females (n = 14) and males (n = 20) separately. (B) Linear mixed modelling results for T_Br by time of day; results for global T_Br (left) and deep brain T_Br (thalamus and hypothalamus, right) are shown. Solid red lines represent model fits, shaded areas represent 95% CIs, dark grey circles display residuals (single temperature data-points) and smoothed dashed yellow lines represent partial residuals. The x-axis for time summarizes the continuous variable of time distance since the participant’s MSF_sc (proportion of a linearized unit circle, where 0 = MSF_sc and 1 = 24 h). (C) Temperature range (maximum versus minimum across three tested time points) for oral and hypothalamic sites for each healthy participant (n = 39). Temperature varied more by time of day in the hypothalamus than orally (repeated measures one-way ANOVA with Sidak’s multiple comparisons test ****P < 0.0001; see Supplementary Fig. 10B for other brain regions). (D) Schematic to model 24-h temperature rhythms of the healthy human brain. Extrapolated T_Br rhythms in healthy luteal females (n = 14) and males (n = 20), without controlling for age, BMI, or chronotype. Extrapolated temperature rhythms were created by duplicating the average temperatures measured at three time points and applying a 24 h sinusoidal fit to these six points. Note higher temperatures in all regions in luteal females relative to males and marked variation in deep brain temperatures in males. Arrows point to predicted T_Br minima around 2–3 am (approaching MSF_sc). Sup1–4 = superficial brain regions 1–4 from medial to lateral; Hypo = hypothalamus; Thal = thalamus.

Figure 5

A daily brain temperature rhythm predicts survival after brain injury. (A–C) Generalized linear mixed model results for outcome in n = 98 TBI patients. Probability of death (‘success’ or ‘hit’ = 1) relative to survival (‘failure’ or ‘miss’ = 0) is depicted on the y-axis. Solid purple lines represent model fits for logit (log of the odds) binomial distribution for a given predictor and dark grey circles display residuals (individual patients). For numerical predictors (A and B), shaded areas represent 95% CIs and smoothed dashed yellow lines represent partial residuals. For the categorical predictor of presence/absence of a daily T_Br rhythm (C), residuals are jittered in the x-axis direction for visibility and 95% CIs are presented as double-ended error bars. (D) Odds of death in intensive care transformed from the data in A–C; the results for these three predictors are significant since the 95% CIs (double-ended error bars) do not include 1. Note also that CIs become numerically asymmetric once transformed from log odds to regular odds. Only factors that demonstrated a statistically significant relationship with mortality are shown. Note logarithmic scale on the x-axis and large effect size for presence of a daily rhythm in T_Br in (D). See the ‘Materials and methods’ section for further details on the generalized linear mixed model and Supplementary material for numerical outputs and related code.