The thermoneutral zone in women takes an “arctic” shift compared to men

Significance

When evaluating energy metabolism, measurements are typically conducted at thermoneutral temperatures to avoid confounding by cold-induced thermogenesis. However, factors that contribute to interindividual variation in the human thermoneutral zone are not well known. To improve our understanding of human thermoregulation, we measured the energetic response of lean women and men to various ambient temperatures. We found that for both women and men, the lower limit of thermoneutrality is largely a weighted balance between basal heat production, which relates to body surface area, and insulation from body fat. Our findings suggest that human thermoregulation largely conforms to principles of physics such that characteristics, including body size, morphology, and composition, determine an individual’s thermoneutral zone.

Abstract

Conventionally, women are perceived to feel colder than men, but controlled comparisons are sparse. We measured the response of healthy, lean, young women and men to a range of ambient temperatures typical of the daily environment (17 to 31 °C). The Scholander model of thermoregulation defines the lower critical temperature as threshold of the thermoneutral zone, below which additional heat production is required to defend core body temperature. This parameter can be used to characterize the thermoregulatory phenotypes of endotherms on a spectrum from “arctic” to “tropical.” We found that women had a cooler lower critical temperature (mean ± SD: 21.9 ± 1.3 °C vs. 22.9 ± 1.2 °C, P = 0.047), resembling an “arctic” shift compared to men. The more arctic profile of women was predominantly driven by higher insulation associated with more body fat compared to men, countering the lower basal metabolic rate associated with their smaller body size, which typically favors a “tropical” shift. We did not detect sex-based differences in secondary measures of thermoregulation including brown adipose tissue glucose uptake, muscle electrical activity, skin temperatures, cold-induced thermogenesis, or self-reported thermal comfort. In conclusion, the principal contributors to individual differences in human thermoregulation are physical attributes, including body size and composition, which may be partly mediated by sex.

Popular belief, theoretical arguments, and survey data suggest that women are more likely than men to feel cold in ambient temperatures typical to common areas, such as offices (1, 2). Proposed causes for this potential sex difference include a lower basal heat production and a proportionally greater heat loss from higher body surface area (BSA) to volume ratio, which both relate to the smaller body size of women (2, 3). Others point to a warmer core body temperature in women as evidence that an elevated thermostatic setpoint creates greater heat demand (4). However, women generally have a higher body fat percentage than men of similar body mass index (BMI) (5), suggesting that women are more thermally insulated (6, 7). To date, few studies have compared the thermal physiology of women and men under controlled conditions (3, 7–9).

Endotherms, including humans, defend core body temperature independent of ambient temperatures (T_a) by balancing heat loss and production. In thermoneutral environments, heat production is minimal and constant at the basal metabolic rate (BMR), and heat loss is mainly regulated with vasoconstriction and vasodilation. The lower critical temperature (T_lc) is the T_a at which heat conservation mechanisms are maximized, below which additional energy expenditure (EE), termed cold-induced thermogenesis (CIT), is required to maintain body temperature. As T_a declines below T_lc, CIT increases linearly according to Fourier’s law of heat flow along the line of “heat requirement.” The slope of the heat requirement is the endotherm’s heat conductance, and its inverse is the thermal insulation (10, 11). This model has been used to describe sources of thermoregulatory variability between and within species (12, 13).

Recently, we used this approach to compare a sample of young lean men and men with obesity (14). We found that the men with obesity were shifted to what Scholander described as a more “arctic” thermal profile (11). This profile included a colder T_lc and a more gradual rise in CIT as T_a dropped. Surprisingly, we found that the men with obesity produced less CIT than the lean men at a similarly cool T_a before overt, uncomfortable shivering. We posited that the same protocol and analytic approach can be employed to compare the thermoregulation of men and women.

We hypothesized that lean women would have a more “tropical” thermal profile than lean men—they would generate less heat due to their smaller body size resulting in a warmer T_lc. Other thermogenic and thermoregulatory processes studied here include brown adipose tissue (BAT), shivering, skin and core temperatures, and self-reported thermal comfort. We postulated that differences in the percentage of body fat, which provides insulation, would explain some of the thermoregulatory variability among individuals.

Results

Demographics.

Twenty-eight young, healthy, lean volunteers (16 women and 12 men) were studied as inpatients. Men and women did not differ in age, BMI, or BSA-to-volume ratio (SI Appendix, Table S1). Men were taller and had greater body weight, lean body mass, and BSA but lower body fat percentage.

Basal Metabolic Rate Is Directly Proportional to Body Size.

Energy expenditure was measured in fasted subjects exposed to T_a‘s of 17.0 to 31.0 °C (Fig. 1A). The resting EE was plotted against T_a and fitted for each participant using a branched regression commonly referred to as a Scholander model (11) to determine the BMR, T_lc, and thermal conductance (SI Appendix, Fig. S1 and Table S2).

Fig. 1.

Analysis of energy expenditure vs. ambient temperature. (A) Resting energy expenditure (EE) as a function of ambient temperature (T_a) for 16 lean women (red) and 12 lean men (blue), including daily measurements (filled circles). Solid lines and shaded regions are group averages and 95% CIs calculated from individual Scholander models. Dashed lines are group average lines from the lower critical (T_lc) to core temperatures. (B) Data in (A) with resting EE divided by body surface area (BSA). (C) BMR directly correlates to BSA. (D and E) Insulation is greater in women than men (D) and positively correlates to body fat percentage (E). (F and G) T_lc is colder in women than men (F) and is inversely related to insulation (G). (H) Averages and shaded 95% confidence bands calculated from individually measured (red) and predicted (gray) Scholander models for lean women. Predicted values were obtained by entering the BSA and body fat percentage of lean women into equations for BMR [BMR (W) = 43.8*BSA (m²) − 2.9] and conductance [Conductance (W/m²/^oC) = −0.015*Body Fat (%) + 3.33] generated from previously published data of lean men and men with obesity (14). (C–G) Filled circles are averages for each participant and solid black lines are all-participant regressions with 95% confidence bands (x-y plots) or group means with unpaired t test results (groupwise univariate plots).

The average BMR was significantly lower in women than men (SI Appendix, Table S3). BMR was directly correlated to BSA (Fig. 1C), body weight (SI Appendix, Fig. S2C), and lean mass (SI Appendix, Fig. S2D), confirming well-established relationships between heat production and measures of body size (15). When an additional interaction term for sex was included in these analyses, it was found to be nonsignificant (SI Appendix, Table S4). According to Fourier’s law of heat flow, heat production directly relates to surface area and, thus, is expressed per unit BSA (10, 16), as in SI Appendix, Fig. S2B (for each participant) and Fig. 1B (group averages for men and women).

Insulation Is Higher in Women and Related to Body Fat Content.

Women had significantly higher insulation than men, both in absolute units (SI Appendix, Table S3) and after converting to SI units (°C/W/m²) by dividing by BSA (Fig. 1D and SI Appendix, Table S3). Linear regression analysis of all participants shows that insulation per unit surface area was positively correlated with fat mass (r = 0.45, P = 0.015) and body fat percentage (Fig. 1E), mirroring the results of our previous study of men with and without obesity (14, 17) and suggesting that body fat plays a role in human thermal insulation.

Cold-Induced Thermogenesis Is Initiated at a Colder Lower-Critical Temperature in Women.

The T_lc is the lower limit T_a of thermoneutrality. In our previous analysis of lean men and men with obesity, we found a colder T_lc was associated with both a higher BMR (SI Appendix, Fig. S3A) and greater insulation (SI Appendix, Fig. S3B) (14). However, the body fat percentage of the men in that sample also scaled with their BSA (SI Appendix, Fig. S3C), making it difficult to determine whether insulation from body fat and basal heat production from BSA contribute separately to the T_lc. The lean women, on the other hand, had lower BSA than the lean men, while their body fat was ~11% higher (SI Appendix, Table S1 and Fig. S3D). If T_lc was largely determined by BMR, one would expect the lower BMR of the smaller lean women to lead to a warmer T_lc, or a “tropical shift” compared to lean men. In contrast, we detected a significantly cooler T_lc in lean women than in lean men (mean ± SD: 21.9 ± 1.3 vs. 22.9 ± 1.2 °C, P = 0.047, Fig. 1F). The T_lc of the lean women and men was negatively correlated to their insulation (Fig. 1G). When included in the same multivariate regression, BMR (β ± SE: −0.15 ± 0.014, P < 0.001) and insulation (β ± SE: −66.88 ± 3.99, P <0.001) but not sex (P = 0.52) significantly associated with T_lc (adjusted R² = 0.92).

Since insulation correlates with body fat percentage and BMR correlates with BSA, we tested whether the cooler T_lc observed in women could be predicted from these measures of body composition. Using previously published data from the lean men and men with obesity (14, 17), we derived linear regression equations for BMR vs. BSA (SI Appendix, Fig. S3E) and thermal conductance (inverse of insulation) vs. body fat percentage (SI Appendix, Fig. S3F). The women’s BSA and measured body fat percentage and tympanic core body temperature (T_core) were entered into these equations, and the intersection of the predicted BMR and the heat requirement lines did not statistically differ from the measured T_lc (mean ± SD: 22.3 ± 0.6 °C predicted vs. 21.9 ± 1.3 °C measured, P = 0.35 for paired t test, Fig. 1H). The cooler T_lc observed in our primary analysis of the Scholander model parameters indicated an “arctic” shift in this sample of lean women vs. lean men and suggests that an individual’s T_lc results from a weighted balance of their basal heat production and thermal insulation.

We measured several additional relevant thermoregulatory parameters as exploratory outcomes.

Core Body Temperature Was Warmer in Women at Colder Ambient Temperatures.

T_core decreased slightly at colder T_a’s for both sexes but changed less for women than men (Fig. 2A and SI Appendix, Table S5), resulting in a significantly warmer T_core for women than men in the cold (T_a < 22 °C, SI Appendix, Table S6). The reduction in T_core from warmer to colder T_a’s was also directly related to body fat percentage (Fig. 2B) and insulation (SI Appendix, Fig. S4A), further supporting that body fat is an effective insulator and suggesting that T_core is more protected in the cold for individuals of higher body fat content.

Fig. 2.

Cold-induced changes to core temperature, skin temperature, and heart rate as proxies of insulation and vasoconstriction. (A) Core temperature (T_core) varies directly with ambient temperature (T_a), with less change for women (red, n = 16) than men (blue, n = 12) (B) Drop in T_core from warm to cold T_a directly correlates to body fat percentage. (C) Mean skin temperature (T_sk) varies directly with T_a and is slightly warmer for women. (D) Proximal-to-distal T_sk gradient (from upper arm to hand) varies inversely with T_a, widening less for women in the cold. (E) Resting heart rate vs. T_a did not differ for men and women. (F) Slowing of heart rate from warm to cold T_a is inversely related to body fat percentage. (A and C–E) Solid trendline with shaded 95% confidence bands and P-values are for the marginal, fixed effect of mixed-effect regressions with interaction terms for sex and random intercept by participant. Filled circles are daily measurements. (B and F) Filled circles are averages for each participant, solid black lines are all-participant regressions with 95% confidence bands.

Skin Temperature Was Warmer in Women, and Vasoconstriction Was Greater in Individuals with Low Body Fat.

Peripheral vasoconstriction provides insulation in the cold (18) and causes changes to both regional skin temperature (T_sk) and cardiovascular parameters. We measured T_sk from the chest, upper arm, upper and lower leg, hand, and fingertip. In both sexes, T_sk varied directly with T_a, with larger changes for more peripheral sites (SI Appendix, Fig. S4 B–G) consistent with peripheral vasoconstriction. Women had a slightly warmer weighted mean T_sk than men throughout the T_a range (Fig. 2C and SI Appendix, Table S5), which is in contrast to a previous cross-sectional finding of women having cooler T_sk than men (4). The gradient from proximal to distal T_sk is a surrogate of peripheral vasoconstriction (19). The upper arm-to-hand T_sk gradient was not significantly different between women and men in warm temperatures (SI Appendix, Table S6). In cooler T_a’s, the gradient widened for both sexes, but the increase was less pronounced for women (Fig. 2D and SI Appendix, Table S5) and its magnitude was significantly smaller for women than men in the cold (SI Appendix, Table S6).

Vasoconstriction causes an increase in blood pressure and a reduction in heart rate via the baroreflex (20). Resting systolic (SBP) and diastolic (DBP) blood pressure were both negatively correlated to T_a (SI Appendix, Table S5) and SBP was higher for men in the cold (SI Appendix, Table S6). Heart rate slowed in colder T_a’s for both women and men (Fig. 2E and SI Appendix, Table S5). The reduction in heart rate from warm to cold T_a’s was directly related to individual body fat percentage (Fig. 2F), suggesting that the participants with less insulation from body fat vasoconstricted more during mild cold exposure.

No Sex Difference Was Detected in Shivering or Reported Thermal Comfort.

In addition to objectively measured physiological responses, we also assessed whether the perception of cold differed for women and men, asking them to rate thermal comfort after 4 h of exposure. Thermal comfort closely correlated with T_a for all participants with no significant interaction with sex (Fig. 3A and SI Appendix, Table S5). The thermal comfort scores of women and men in the cold were not associated with body fat percentage (Fig. 3B).

Fig. 3.

Thermal comfort and shivering. (A and B) Self-reported thermal comfort (0 = coldest, 100 = warmest) vs. ambient temperature (T_a) did not statistically differ for men and women (A) and does not correlate to body fat percentage in the cold (B). (C) There was not detected sex difference in the pattern of the four-muscle average electromyogram (EMG) over T_a. (D) EMG activity in the cold correlated negatively with body fat percentage. (A and C) Solid lines are trendlines with shading indicating 95% confidence intervals (CI) for the fixed effect terms of mixed-effect regressions with an interaction term for sex and random intercept by participant. EMG data was Box-Cox transformed for analysis and the inverse transform was applied to fixed-effect trendlines and CIs to display on a linear scale. Filled circles are daily measurements. (B and D) Filled circles are averages for each participant, solid black lines are all-participant regressions with 95% confidence bands. (C and D) Data from 12 women are included due to EMG equipment failures for four women.

We also measured muscle activity with surface electromyography (EMG) at the pectoralis, trapezius, biceps brachii, and rectus femoris. The resting four-muscle average EMG increased at colder T_a’s for women and men, but no significant sex difference was detected in the pattern of average EMG with respect to T_a (Fig. 3C and SI Appendix, Table S5). Over the entire group, the increase in EMG muscle activity in the cold (<22 °C) inversely correlated with body fat percentage (Fig. 3D). Thus, while the thermal comfort and shivering responses of women and men did not significantly differ, individuals with lower body fat percentage had higher muscle electrical activity in the cold.

No Differences Were Detected in the Coldest Tolerable Temperature, Cold-Induced Thermogenesis, or Brown Adipose Tissue Glucose Uptake between Women and Men.

On the final day in the room calorimeter, participants were studied at their coldest tolerable T_a before overt shivering preceding imaging of ¹⁸F-fluorodeoxyglucose (¹⁸F-FDG) by positron emission tomography/computed tomography (PET/CT) to quantitate BAT volume and ¹⁸F-FDG uptake (SI Appendix, Fig. S5 A and B) (14). The coldest tolerable T_a did not differ for women and men (SI Appendix, Table S7), nor was it related to body composition. After 5 h of exposure, women had less activated BAT volume than men (Fig. 4A). However, BAT ¹⁸F-FDG uptake was not statistically different among the sexes (Fig. 4B). Neither BAT volume nor ¹⁸F-FDG uptake was related to insulation (SI Appendix, Fig. S5C) or the coldest tolerable T_a (SI Appendix, Fig. S5D).

Fig. 4.

Analysis of brown adipose tissue and cold-induced thermogenesis. (A–C) Brown adipose tissue (BAT) volume assessed via ¹⁸F-FDG PET/CT is greater for men than women (A), but BAT ¹⁸F-FDG uptake did not statistically differ by sex (B). (C) Cold-induced thermogenesis (CIT) capacity—the heat above basal metabolism generated at the coldest ambient temperature (T_a) before overt shivering—did not differ for women and men. (D–F) CIT capacity inversely correlates to the coldest tolerable T_a (D) and directly correlates to electromyogram (EMG) activity averaged from four muscles (E) but does not correlate to BAT ¹⁸F-FDG uptake over all participants (F). Filled circles are averages for each participant and solid black lines are all-participant regressions with 95% confidence bands (x-y plots) or group means with unpaired t test results (groupwise univariate plots). (A, B, and F) One woman declined PET/CT, and (E) EMG was recorded in only eight women at the coldest tolerable T_a.

The magnitude of CIT at the coldest tolerable T_a (CIT capacity) was considerable for both women and men with no significant group difference between sexes (Fig. 4C and SI Appendix, Table S7). CIT capacity was inversely related to the coldest tolerable T_a (Fig. 4D). At the coldest tolerable T_a, EMG activity was not statistically different for women and men (SI Appendix, Table S7) but directly correlated with CIT capacity over all subjects (Fig. 4E). However, CIT capacity was unrelated to BAT volume (SI Appendix, Fig. S5E) and BAT ¹⁸F-FDG uptake (Fig. 4F), suggesting that higher CIT capacities were the result of greater muscle activity for individuals exposed to colder temperatures.

Discussion

In this controlled study of physiology, we focused on the thermoregulatory responses in healthy, lean young women and men over a range of ambient temperatures that allowed us to compare the lower bound of the thermoneutral zone, basal metabolic rate, insulation, and magnitude of cold-induced thermogenesis. Surprisingly, we did not find the women in our sample to have a more “tropical” thermal profile than the men. Instead, the women initiated CIT at a cooler, not warmer, T_a than the men, resembling an “arctic” shift. Across the lean men and women, the onset and amount of additional heat production were related to body size, percentage of body fat, degree of vasoconstriction, and intensity of muscle electrical activity. Thus, our observations suggest that the thermoregulatory responses in healthy young lean participants were associated with factors such as body size, morphology, and composition in both women and men.

To our knowledge, comprehensive metabolic and physiological measurements at different T_a exposures for both women and men have been limited. The Scholander model, rooted in physical principles, was chosen as our primary analytical approach to define three key thermoregulatory parameters—lower critical temperature (T_lc), basal metabolic rate (BMR), and insulation.

Lower-Critical Temperature as a Weighted Balance of Basal Heat Production and Thermal Insulation.

Compared to the lean men, the lean women in this sample had lower basal heat production, as expected from their smaller body size (15). According to some models (13), a lower BMR leads to a more tropical thermal profile. However, the lean women also had a higher body fat percentage than the lean men and, across all participants, body fat percentage correlated to whole-body insulation, supporting prior findings that adipose tissue acts as an effective insulator (7, 9, 16). Since the T_lc is the intersection of the BMR and the line of heat requirement, whose slope is the inverse of the insulation, both BMR and insulation influence the location of the T_lc. Thus, despite a lower BMR, we observed a more artic rather than tropical T_lc for the lean women compared to lean men due to their greater insulation, reflecting the balance of the two parameters. However, these findings should be replicated in larger, more diverse study samples to enhance generalizability.

Evidence of Substantial Cold-Induced Thermogenesis and Additional Insulation from Vasoconstriction.

At their coldest T_a before overt shivering, lean women and men elevated their metabolic rate by 13.1% and 17.4 % above their BMR, respectively, which is greater than the EE increase from common pharmacological agents (3 to 10%) (21–23) or from standing compared to sitting (~10%) (24). BAT glucose uptake and muscle shivering also increased in the cold, suggesting that they both contributed to the CIT increase. However, those with higher proportions of body fat had a smaller reduction to heart rate and lower EMG activity in cooler T_a’s. These changes suggest that those with greater body fat insulation required less vasoconstriction and heat production from CIT.

No Differences in Thermal Perception Were Detected between Women and Men.

Contrary to conventional belief, we did not detect a significant difference in the self-reported thermal comfort of women and men across the span of T_a’s we tested. Clothing insulation contributes to thermal comfort (2) and, in free-living settings, the amount and type of clothing is influenced by environmental temperature as well as social norms (25). Since clothing was standardized, our study could not address the impact of insulation from the preferred attire on thermal perception. Limited observational data suggest free-living clothing insulation does not differ by sex (25), but more detailed analyses, accounting for body composition, age, and cultural dependence of clothing, are warranted. We did not objectively measure thermal perception by direct assessment of thermoreceptor density or neural response (26). However, our observations in self-reported thermal comfort and EMG measured shivering agree with those of a previous study (9).

Significance to the Study of Human Metabolism.

While energetics has been studied at the genetic, epigenetic, cellular, and colonial organism levels (27–29), contributors to interindividual variation in whole-body human thermal and metabolic regulation are still not fully understood. Historically, studies of the human cold response were conducted to better predict physiologic changes that accompany environmental exposures (7). More recent efforts have focused on determining whether CIT can increase EE enough to impact energy balance (30). However, identifying the attributes that affect human thermoneutrality and the metabolic response to cold is also important for understanding the nonthermoregulatory components of EE. Changes to whole-body EE cannot be attributed to a particular intervention (e.g., with drug or diet) without first minimizing interindividual differences in CIT, which can be accomplished by studying all participants in thermoneutral temperatures. However, there is considerable variability in the onset and magnitude of human CIT (31). The results of the current study suggest that this variability is most prominently related to individual differences in basal heat production and thermal insulation. The close relationship of BMR to surface area and insulation to body fat percentage underscores the concept that the relationship between resting EE and T_a in humans, including the T_lc, is related to body size and composition, which should be accounted for in studies of human metabolism.

Limitations.

We conducted a careful comparison of the thermogenic response to varied T_a in lean, young men and women in the follicular phase of their menstrual cycle who volunteered and were rigorously screened in accordance with the protocol. The sample size was based on previous studies of the energetic response to cold (14, 32) and was powered to detect difference in CIT and T_lc. Additional data are needed to enhance confidence in the exploratory, secondary outcome measures of thermoregulation, particularly those involving mixed-effect regressions which are typically applied to larger datasets (33). Similarly, the small, homogenous sample may not represent the broader population.

The Scholander model was originally fit to group data with limited analytic details (11). Here, we applied the model separately to each participant. With 7 to 13 data points per participant, we cannot discount the potential impact of artifacts on the Scholander model fits. Three of the 28 participants each had several outlying data points which resulted in models with non-normal residuals (SI Appendix, Table S2), but we did not have reason to eliminate the outliers. Thus, we performed several sensitivity analyses that yielded results similar to those of our original analysis: 1) we recomputed the average Scholander parameters by sex after removing the three individual fits with non-normal residuals and 2) we pooled data from all participants by sex and computed a single Scholander fit for each sex (SI Appendix, Table S8).

We found that parameters of the human thermal profile, including BMR, insulation, and T_lc, differed by sex and associated with individual body size and composition. Our interpretations were largely based on how the physical properties of these attributes (e.g., the resistance to heat flow provided by body fat), rather than sex per se, might contribute to thermoregulation. However, while body size and composition are the proximate determinants of the thermal physiology, they are also partly mediated by sex, as women are generally observed to have higher body fat percentage (5) and lower BSA (34) than men of a similar BMI range. Thus, even when sex was not statistically significant as an independent variable or interaction in our analyses, we cannot definitively discount sex as a biological mechanism or mediator of the observed differences.

According to convention, some ratio variables were used in our analyses, e.g., body fat percentage and EE divided by BSA, but we cannot completely rule out nonzero intercept artifacts (35). Thus, we replicated all analyses of ratio variables with numerators and denominators as covariates (SI Appendix, Table S9).

Our findings would be enhanced by additional study of women in the luteal phase, when shifts in hormone levels are associated with elevated core temperature (36) and increased resting metabolic rate in some (37, 38), but not all (39), cases. We did not exclude women taking contraceptives and nine of the women continued their contraceptive regimen during the protocol, while the others had natural menstrual cycles. The small group sizes and varied formulations and doses make it difficult to draw conclusions about the effect of contraceptives on thermoregulation from our data, but previous reports suggest women taking oral contraceptives have warmer core temperatures than naturally cycling women in the follicular phase (40, 41). While our results may differ from studies exclusively measuring women with natural menstrual cycles (36), findings from a cohort with mixed contraceptive use may be more informative to free-living conditions for populations with a high prevalence of contraceptives, as in the United States (42). Finally, our conclusions would be further informed by expanding to women with obesity, postmenopausal women, and cohorts of broader age and ethnicity.

Conclusions.

We found that young, healthy women initiated cold-induced thermogenesis at a cooler lower critical temperature than young, healthy men. Their individual thermal physiology indicates a balance between basal heat production, which scales with body size, and resistance to heat loss, which is related to body fat. Exploratory analyses suggest body fat is an effective insulator and those with greater proportions of it required less additional insulation from vasoconstriction and cold-induced thermogenesis from BAT and muscle activity. In summary, body size, morphology, and composition, attributes partly mediated by sex, likely play a role in the mechanisms governing human thermoregulation.

Materials and Methods

Protocol Design.

This was a single-blinded study with a randomized temperature schedule and blinded data analysis. To minimize biological variability in the absence of informative prior research, we studied healthy, young, sedentary, Caucasian males and females between the ages of 18 to 35 y, with a BMI of 18.5 to 25.0 kg/m² (lean cohorts) and 30.0 to 40.0 kg/m² (obese male cohort). Subjects were recruited through https://ClinicalTrials.gov and attended an initial screening visit at the NIH Hatfield Clinical Research Center to determine eligibility following a 12 h fast with ad libitum water intake. After providing written informed consent, they underwent standard screening tests including blood chemistry, lipid panel, complete blood count, thyroid stimulating hormone, ferritin, electrocardiogram (ECG), and urinalysis. Women who were pregnant, had irregular menstrual cycles, or were taking hormonal controls that prevented the identification of menstrual phases were excluded from participating. Nine of the 16 women who participated were taking contraceptives prior to screening and continued their regimen during the study. The study protocol was approved by the Institutional Review Board of the National Institute of Diabetes and Digestive and Kidney Diseases (For further details see: Clinical Trial number NCT01568671). Partial data were published previously to compare BAT in lean men to men with obesity (43) and lean women (44) and to compare CIT in men with vs. without obesity (14).

Energy expenditure was measured while >8-h fasted subjects were exposed to a specified T_a (±0.2 °C) for 5 h daily in a whole-room indirect calorimeter (metabolic chamber) from 0800 to 1300. The calorimeter T_a’s were incremented by ~1.5 °C over the thermal comfort range for healthy individuals with light clothing, 17 to 31 °C (45). The T_a for day 1 was 24 °C. The T_a of day 2 remained 24 °C for men to determine reproducibility at standard room temperature and their day 1 was treated as an acclimation day and data were not used in the final analysis. Women did not repeat the day 1 temperature due to a shortened protocol (7 to 10 d) compared to that for men (13 d) to accommodate follicular phase length. The remaining T_a’s were randomly ordered. Calorimeter studies were terminated if overt, uncomfortable shivering was self-reported, and no colder T_a’s were attempted on the remaining days. The T_a of the final day was the coldest tolerable temperature with minimal shivering, or 17 °C if no shivering was previously reported. Subjects were blinded to calorimeter T_a order.

In the calorimeter, subjects wore standardized clothing provided by the study team, consisting of the same model sleeveless shirt, shorts, and socks for all participants (combined clothing insulation of 0.056 °C/W/m² or 0.36 clo) and an additional sports bra for the women (0.003 °C/W/m² or 0.02 clo). Participants supplied their own underpants. Subjects were asked to remain awake and minimize activity while seated in a mesh chair. During rest periods (1030 to 1100 and 1130 to 1200), subjects refrained from movement. Physiological measurements from the rest periods were averaged and used for analyses, unless otherwise specified (14). Blood pressure was measured before and after each calorimeter study. On the last day’s calorimeter study for men and women in the follicular phase, subjects received a 370 mBq dose of ¹⁸F- FDG at 1200, remained in the calorimeter until 1300, and were transported to the NIH Clinical Center Department of Nuclear Medicine via wheelchair and underwent PET/CT scanning with a Siemens Biograph mCT (Siemens Healthcare) by 1330.

When outside the calorimeter, participants were housed in a temperature-controlled (23 to 25 °C) inpatient room and fed a weight maintenance, caffeine-free diet with prespecified meal times (1300, 1800, and 2200) and a macronutrient distribution of 55% carbohydrate, 15% protein, and 30% fat and a caloric content of 1.4 times estimated resting EE (46), adjusted if body weight changes were detected during the study. Each day, participants completed 30 min of walking on a treadmill or through the hallways of the clinical center at a constant, self-selected pace outside the calorimeter and were not allowed food from midnight until the conclusion of the calorimeter study at 1300.

Room Calorimeters.

Air drawn from seven equidistant sample points on the ceiling of the 30,000 L aluminum-lined interior at a constant, mass-flowmeter-measured rate of ~60 L/min (Teledyne-Hastings, Inc.) is passively replaced by a mixture of conditioned outside air and medical-grade air entering from a small opening several inches from the floor. An interior air-handling unit mixes the calorimeter air and provides a stable internal temperature (±0.2 °C) and relative humidity (30 to 50%) which is continuously monitored (Optica, GE Sensing). A small quantity (1 L/min) of effluent and supply airstreams is dried (PD-Series Nafion Perma Pure) and measured for CO₂ (ABB AO2000) and O₂ (Siemens Oxymat 6E) concentration. Two microwave-based sensors (Museum Technology Source, Inc.) on the ceiling detect the presence or absence of motion each second. Data from all sensors are recorded each minute (CalRQ, MEI Research, Edina, MN) and used to compute O₂ consumption (VO₂) and CO₂ production (VCO₂) (47) using customized software (MATLAB, The MathWorks). EE is computed using the Weir Equation (48). Gas analyzers are calibrated weekly and between analyzer differences are corrected for prior to each study. Monthly propane combustion or gas infusion tests demonstrated no significant differences in the accuracy of the three calorimeters (<1.5% for VO₂ and VCO₂) when temperature is maintained within 0.2 °C. Greater detail can be found elsewhere (32, 47).

Scholander Model.

Branched regression was performed for each subject with resting EE as the dependent variable and T_a as the independent variable (SI Appendix, Fig. S1). The regression calculates three parameters that define two lines and their intersection point. One line (at the cooler T_a’s) is forced to intersect the T_a axis at the measured T_core. The other line (at the warmer T_a’s) is restricted to a zero slope. As defined by Scholander and Kleiber, the T_a at the intersection point is the T_lc, the line at cooler T_a’s is the line of heat requirement and its slope is the whole-body heat conductance, and the EE of the zero-slope line is the BMR (11, 49). Each possible partition of the subject’s data into cooler and warmer T_a’s were fit with two intersecting lines that complied with the described restrictions. The fit that yielded the lowest mean squared error was considered the final model for that subject. The normality of the model residuals was visualized with quantile–quantile plots and tested via the Shapiro–Wilk method (SI Appendix, Table S2).

Physiological Sensors.

Prior to entering the room calorimeter, subjects were instrumented with a Holter monitor (Evo Recorder, Spacelabs Healthcare) to record ECG, wireless surface EMG electrodes (Trigno, DelSys Inc.), and wireless thermistor probes (iButton, Maxim Inc.) to record T_sk. ECG data from one woman were excluded due to an arrythmia. Raw EMG signals (2,000 Hz) were recorded from four muscle groups shown to significantly contribute to whole-body shivering (50): right biceps brachii, pectoralis major, rectus femoris muscles, and left upper fibers of the trapezius. EMG contamination was removed with a Butterworth filter (45 to 500 Hz) and RMS was computed with overlapping 50 ms periods (51) in Matlab. EMG device failures resulted in a lack of data for four women and incomplete data for four other women. T_sk sensors were attached using surgical tape (Medipore, 3M) on the left deltoid (upper arm), dorsal hand, pectoralis major (chest), anterior thigh (upper leg), and shin (lower leg) and logged temperature each minute. Fingertip was measured with an infrared thermal camera (T400, FLIR Systems). Weighted mean T_sk was computed as described previously (52). T_core was measured with a handheld infrared tympanic thermometer (PRO4000, Braun) by participants after staff training. Tympanic measurement was chosen to minimize subject burden and because it closely approximates nasopharyngeal temperature (53) and is well correlated with rectal temperature (54). Since tympanic T_core varied with T_a, to minimize the potential influence of excessive cold or heat the average of T_core measurements at T_a’s from 26 °C to 30 °C was considered the thermoneutral T_core for the Scholander analysis. Data from the calorimeter were time aligned with all physiological sensors during postprocessing. Data from the 20 consecutive minutes of lowest microwave-detected activity for each rest period (1030 to 1100, 1130 to 1200) were averaged for primary analysis.

Brown Adipose Tissue.

Using the PET/CT Viewer plugin for ImageJ (55), trained technicians created regions of interest on each axial slice of the coregistered PET/CT images from the C3 to the L3 vertebrae, carefully avoiding regions that were not metabolically active adipose tissue to minimize false positive detections. Voxels in the regions of interest with a CT density of −300 to −10 Hounsfield Units and PET SUV above 1.2 g/mL/lean body mass (56) were used to compute BAT volume, activity, mean SUV, and max SUV. Greater detail is reported elsewhere (43, 44, 57). One woman declined the PET/CT measurement.

Self-Reported Thermal Comfort.

Volunteers rated thermal comfort level using an electronic slide bar visual analog scales from 0 to 100, administered on an iPAD (Apple) at 1200.

Body Composition.

Weight (Scale-Tronix 5702 digital balance) and height (Seca 242 stadiometer) were taken before and after each chamber study. Body composition, including body fat mass, lean mass, and fat percentage was measured by dual-energy X-ray absorptiometry (iDXA scanner with Encore 11.10 software; GE Healthcare). BSA was calculated with the Dubois formula (58). Body volume was computed with Siri’s equation (59).

Statistical Analysis.

Statistical analyses were performed using R software version 4.3.1 (https://www.r-project.org) in RStudio version 2023.12.0 (Posit Software, PBC). Code and data are included in the Supporting Information and available at the Open Science Framework (60). Figures were created in Prism version 10.0.2 (GraphPad).

The number of subjects for the primary analysis of EE vs. T_a was informed by data from a prior study where a 6.0 ± 4.4% (mean ± SD) increase in EE was measured in 25 participants (males and females, BMI 23.2 ± 2.2 kg/m²) wearing hospital scrubs (~0.7 clo) during 12-h in 19 vs. 24 °C (32). We estimated needing 16 lean women to compare to the 12 lean men to detect a 5% group difference in CIT using a two-tailed, independent t test (α = 0.05, 1-β = 0.8). This participant ratio also enabled detection of a ~6% change in T_lc from the 22.9 ± 1.2 °C (mean ± SD) value found in lean men (14) using the same statistical assumptions. Group differences between men and women were assessed with t tests for single-point variables. Linear regression and Pearson’s correlation were used to determine the presence or absence of associations between single-point variables (e.g., BMR vs. BSA).

Exploratory outcomes measured at each T_a, including T_core, T_sk, blood pressure, heart rate, EMG, and self-reported thermal comfort were assessed using linear mixed effect models with an interaction between T_a and sex and a random intercept by participant (lmerTest v3.1.3 and lme4 v1.1.34 packages in R). Restricted maximum likelihood was used to estimate mixed effect model parameters, and conditional residuals were assessed for homogeneity of variance and normality with visual inspection of residual and quantile–quantile plots, respectively. EMG data were highly skewed and were transformed via the Box-Cox method when used as a dependent variable, as indicated in figure and table legends. An additional analysis was performed on dependent variables of models where sex was found to significantly contribute (P < 0.05). These variables were dichotomized by averaging measurements taken near the extremes of the T_a range, T_a < 22 °C for cold and T_a > 28 °C for warm. Mixed effect models of response variables vs. dichotomized T_a (warm/cold) with an interaction for sex and random intercept by participant were used to test for within-sex differences by temperature (paired companions, warm vs. cold) and between sex difference at warm and cold T_a’s (unpaired comparisons) using post hoc tests adjusted for multiple comparisons with the Bonferroni method. Linear regression and Pearson’s correlation were also used to determine the presence or absence of associations between cold T_a averages or the difference between cold and warm T_a averages and body composition parameters (e.g., EMG activity in the cold vs. body fat percentage).

Supplementary Material

Appendix 01 (PDF)

Code S01 (R)

Dataset S02 (CSV)

Dataset S03 (XLSX)

Data, Materials, and Software Availability

Original Data and source code have been deposited in Open Science Framework (https://osf.io/jbrv3/) (60).