A heterothermic spectrum in hummingbirds

Torpor is an energy-saving strategy documented in over 200 species of birds and mammals (Boyles et al., 2020; Ruf and Geiser, 2015). Torpor is a fascinating ability – torpid animals save energy by lowering their metabolic rate and body temperature. These animals are often described as having species-specific minimum torpid body temperatures (between −2 and 29.6°C; Barnes, 1989; Hainsworth and Wolf, 1970; McKechnie and Lovegrove, 2002; Richter et al., 2015; Ruf and Geiser, 2015). Depending on their minimum torpid body temperature, some animals only use a ‘shallow’ form of torpor (e.g. fasted doves, body temperature 28–36°C; Fig. 1C), while others use ‘deep’ torpor, in which body temperature is low (more than 20°C below normothermic body temperature, e.g. Arctic ground squirrels, hummingbirds, −2 to 18°C; Fig. 1B). Both these states are characterized by lower body temperatures than the 1–2°C drop below resting daytime body temperature that occurs following a circadian rhythm in normothermic sleep (‘nocturnal’ or ‘rest-phase’ hypometabolism; Geiser, 2021; Walker et al., 1983).

Much of the relatively recent work on the metabolic torpor spectrum has focused on mammals (Boyles et al., 2020; Kräuchi and Deboer, 2011; Reher and Dausmann, 2021; Ruf and Geiser, 2015; van Breukelen and Martin, 2015). Energetic, neurological (electroencephalogram, EEG), transcriptomic and ecological evidence exists for a physiological continuum from shallow to deep torpor in mammals, as found in several ground squirrel species, marmots and kangaroo rats (Berger, 1984; Canale et al., 2012; Florant and Heller, 1977; Glotzbach and Heller, 1976; Heller, 1979, 1978; Walker et al., 1977, 1979). Some bird species are known to use shallow torpor at night or when fasted, while others regularly use deep torpor (Ruf and Geiser, 2015). Though avian shallow torpor and deep torpor have separately received research attention (Beckers and Rattenborg, 2015; Brigham et al., 2000; Powers et al., 2003; Roth et al., 2010; Wolf and Hainsworth, 1972), the differences and potential trade-offs between these states in birds are poorly studied relative to mammals. There are some hints in the literature that such a metabolic spectrum exists in birds under specific conditions (in mousebirds; Schleucher, 2004). We use the word ‘spectrum’ rather than ‘continuum’ here because whether this spectrum is continuous or discrete is unknown (some studies argue that it is a continuum between species: Barclay et al., 2001; while others argue that it is discrete: McKechnie and Lovegrove, 2002). Though birds constitute 65% of extant endotherms, the data on avian heterothermy are sparse compared with mammalian data (Geiser and Ruf, 1995; McKechnie and Mzilikazi, 2011), as are data on this avian heterothermy spectrum. Of the 42 bird species reported to use daily torpor, only hummingbirds (Trochilidae), nightjars (Caprimulgidae) and one mousebird (Coliidae) species have minimum body temperatures colder than 20°C; the rest use a relatively shallow form of torpor (Ruf and Geiser, 2015). Torpor is identified by myriad thresholds across taxa, and by body temperature, metabolism and heart rate changes (Boyles et al., 2020; Reher and Dausmann, 2021; Ruf and Geiser, 2015). By necessity, these thresholds are species and sometimes individual specific (Barclay et al., 2001), because they occur in response to changes in ambient temperature, fat stores, moonlight and a variety of other environmental and endogenous factors (Eberts et al., 2021; McKechnie and Lovegrove, 2006; Powers et al., 2003; Ruf and Geiser, 2015; Smit et al., 2011). Exploring the range, variability and flexibility of avian torpor can help elucidate behavioral and physiological mechanisms underlying thermoregulation, energy regulation and torpor use across vertebrates, and move us closer to understanding the evolution of homeothermy versus heterothermy.

Deep torpor likely reflects a trade-off between its benefits – an average of 60% energy savings relative to basal metabolic rate (Hainsworth et al., 1977; Shankar et al., 2020) – and its potential costs in mammals such as susceptibility to predation, high rewarming costs, immune suppression and sleep deprivation (Bouma et al., 2010; Boyles et al., 2020; Carr and Lima, 2013; Deboer and Tobler, 2000; Kräuchi and Deboer, 2011; Palchykova et al., 2006; Prendergast et al., 2002). If these trade-offs are similar in birds, it might be beneficial for birds that use the deepest possible form of torpor to sometimes use a shallower form, to allow moderate energy savings while minimizing some of the potential costs of deep torpor (Boyles et al., 2013). Yet, in contrast to mammals, bird species have been observed to use either shallow or deep torpor, but usually not both (Berger, 1984; Brigham et al., 2000; Hainsworth and Wolf, 1978; Hiebert, 1990; Kruger et al., 1982; McKechnie and Lovegrove, 2002; Ruf and Geiser, 2015). One mousebird species has been described to use both shallow and deep torpor when starved over several days, with their depth of torpor deepening as their energy stores were depleted (McKechnie and Lovegrove, 2000; Prinzinger et al., 1992). However, mousebirds are thought to have diverged early in the avian phylogeny and their unusual combination (among birds) of low-quality plant diet with a relatively small body size makes them physiologically distinct in other ways. It is also possible that they display a form of ‘proto-torpor’ without the standard entry and rewarming patterns of other avian lineages (Bosque et al., 2017; Downs et al., 2000; McKechnie and Lovegrove, 2000). The rarity of shallow torpor in birds that use deep torpor, and vice versa, would imply that shallow and deep torpor are mutually exclusive and relatively inflexible states. The possible existence of an avian torpor continuum has been hinted at in the literature (McKechnie and Lovegrove, 2002; Smit et al., 2011), but evidence supporting or disproving its existence within species is scarce.

Hummingbirds have long been known to use deep torpor to save energy overnight, with minimum body temperatures varying from 3 to 22°C (Carpenter, 1974; Hainsworth and Wolf, 1970; Hainsworth et al., 1977; Wolf et al., 2020). One past study reported a shallower form of torpor in hummingbirds, but its experimental conditions may have prevented deep torpor: those birds were maintained at warm temperatures, were not free living, and were frequently disturbed at night (Morrison, 1962). The ability to use both shallow and deep torpor would indicate that they are facultatively using a range of thermal states. This flexibility in body temperatures is almost never described in birds, but such a capacity could contribute to hummingbirds' ability to thrive under diverse and variable environmental conditions, from deserts to tropical forests and from sea level to the high Andes, despite their small body size and extreme metabolic demands. Previous work suggested that some larger hummingbird species had more variable metabolic rates than smaller hummingbirds (Shankar et al., 2020; Wolf et al., 2020), and our preliminary data from sites in the high Ecuadorian Andes has also suggested that some hummingbird species there might be using a range of shallow and deep torpor (A.S., Ana Morales, Erich R. Eberts, Julisa Ricart, Joshua van Bourg, Gabriela K.Z. Córdova, Gabriela M. R. Urgilés, Boris A. Tinoco, C.H.G, D.R.P., unpublished observations).

Here, we tested whether hummingbirds are capable of shallow torpor by recording night-time surface temperatures in three species sympatric at sites in Arizona (USA) where night-time temperatures are cold enough to allow deep torpor. We know from previous work that all three species use deep torpor (Powers et al., 2003; Shankar et al., 2020). We hypothesized that these hummingbirds might facultatively use shallow torpor, either alone or in addition to deep torpor, to balance the energy savings and physiological costs of using deep torpor alone. Hummingbirds appear to delay torpor until they have reached some minimum threshold of energy stores (Hainsworth et al., 1977; Hiebert, 1992; Powers et al., 2003). We therefore expected birds to use shallow torpor in one of two ways: either exclusively with normothermy (Fig. 1C), or before entering deep torpor, as a strategy to delay the onset and potential costs of deep torpor (Fig. 1D). Given that hummingbirds seem to reach a minimum fat threshold before entering deep torpor (Eberts et al., 2021; Powers et al., 2003), we expected that once a bird entered deep torpor, it would stay in deep torpor for the remainder of that night, and then rewarm to normothermy before flying off, rather than using shallow torpor after deep torpor. We used thermal imaging to study hummingbird torpor under near-natural conditions. This study design allowed us to assess torpor use under natural light and temperature cues, as well as near-natural energy stores. Such physiological control in torpor would in turn imply that a broad and perhaps continuous avian heterothermic spectrum exists, much like in mammals.

We studied males of three hummingbird species at the Southwestern Research Station (SWRS) in the Chiracahua mountains of Arizona (latitude: 31.9, longitude: −109.2): the blue-throated mountain-gem, Lampornis clemenciae (Lesson 1829) (8.4 g, n=14); Rivoli's hummingbird, Eugenes fulgens (Swainson 1827) (7.6 g, n=12); and the black-chinned hummingbird, Archilochus alexandri (Bourcier and Mulsant 1846) (2.9 g, n=7). Two blue-throated mountain-gem individuals had some bill corrugation and were likely late-stage juveniles. Within this hummingbird community, both the black-chinned and Rivoli's hummingbirds are subordinate to blue-throated mountain-gems (i.e. with less exclusive access to floral resources) (Powers et al., 2003; Sandlin, 2000). We collected data between 10 and 19 June 2017, and 20 May and 7 June 2018. All protocols associated with hummingbird care and experimentation were approved by the Stony Brook University Institutional Animal Care and Use Committee (IRBNet number: 282617-6). Field protocols were approved by US Fish and Wildlife in Arizona (USFWS MB75714A-0).

We captured hummingbirds using modified Hall traps at hummingbird feeders (Russell and Russell, 2001) within 1.5 h before sunset, to allow them to store energy naturally through the day, but also acclimate to our experimental setup. The majority of birds were already banded (this is a long-term bird monitoring site), but un-banded birds were marked with a unique set of small dots of non-toxic paint on the forehead (2017: 4 birds unbanded, 8 banded; 2018: 6 birds unbanded, 16 banded). We recorded capture mass, allowed the birds to feed ad libitum, and weighed them again for mass after feeding. They were then placed outdoors (individually) in five-sided acrylic chambers (either 18×17×22 cm or 46×23×46 cm), exposed to natural light and temperatures. All night-time recordings were between 19:00 h and 05:59 h. The front face of the chamber was covered by a clear plastic sheet to prevent the bird escaping. This sheet caused the thermal reading of the bird's surface temperatures to be up to 2°C cooler than readings without the sheet, so once the bird was observed to settle down, the plastic sheet was removed. We placed a wire grill at the base of the chamber to encourage birds to perch with their sagittal plane facing the camera, usually ensuring that recordings included a direct view of the bird's eye. Recordings without this view were excluded from analyses.

Bird eye surface temperature seems to closely reflect internal physiological state (e.g. body condition), from recent work in blue tits (Jerem et al., 2018). Hummingbirds have low feather density around the eye, so skin eye temperature patterns should closely reflect the patterns of core body temperature, minimizing the confounding effects of feather insulation, unlike in larger animals for which skin and core temperatures might vary because of reduced peripheral blood flow in torpor (Arnold et al., 2011; Barclay et al., 1996; Brigham et al., 2000; Dausmann, 2005; Powers et al., 2015). Powers et al. (2017) used thermal imaging to measure heat dissipation areas in hovering hummingbirds during the day in three species. They found that across all three species, eye surface temperature remained relatively constant across a range of ambient temperatures, with an intercept of 32–33°C (see fig. 2 in Powers et al., 2017). Although this is lower than core body temperature, it is consistent with what is observed in individuals from these same species that are clearly normothermic. This supports our measurements of 32°C being a common resting normothermic body surface temperature. A recent study of hummingbird body temperature in torpor across six species showed that many individuals maintained normothermy at body (cloacal) temperatures between 35.5 and 40°C (Wolf et al., 2020). Additionally, there seems to be high concordance between eye surface temperature and metabolic rate as measured in ruby-throated hummingbirds (Erich Eberts, University of Toronto Scarborough, personal communication).

We used a FLIR SC6701 infrared video camera (640×480 pixel resolution, accurate to 1°C at measured temperatures, FLIR Systems, Inc., Wilsonville, OR, USA) to record surface temperatures of hummingbirds. We assumed emissivity was 0.95 across all surfaces of the hummingbird (Cossins and Bowler, 1987; Powers et al., 2015). We monitored birds continuously through the night, and sampled surface temperatures by recording 10 s of 30 Hz video approximately every 10 min, using ResearchIR (FLIR Systems, Inc., Wilsonville, OR, USA). From one frame per recording, a region including the bird and a slight buffer to include ambient temperature was marked as a region of interest and exported to csv files for analysis in R (v.3.5.1; http://www.R-project.org/). From each exported region of interest, we extracted maximum surface temperature (in °C) and mean surface temperature of the bird, as well as minimum temperature of the background (our proxy for ambient temperature; details below). We verified that maximum surface temperature corresponded with maximum surface temperature of the region surrounding the eye (hereafter eye surface temperature, though these were never the temperatures of the actual surface of the eye, which is usually colder than the skin/feathers surrounding the eye), and validated outliers in temperature measurements to ensure that they were reliable measurements. We also exported entire single-frame images from selected recordings and used ImageJ (NIH) to construct 3D images to assess how surface temperature changed over the entire surface of the bird.

We used minimum temperatures from thermal image regions of interest as an estimate of ambient temperature, verifying that these closely matched independent ambient temperature measurements from iButtons (DS1921, Maxim Integrated, San Jose, CA, USA) or thermocouples (Cu-Cn type-T, recorded on a TC-1000; Sable Systems, Las Vegas, CA, USA). The FLIR camera was factory calibrated and verified by imaging a surface of a known temperature. Thermocouples and iButtons were calibrated by using a Percival (i.e. a temperature-controlled cabinet, model I-35LV, Percival Scientific, Inc., Perry, IA, USA) at controlled temperature steps, and checked against a thermometer traceable to the National Institute of Standards and Technology.

We assigned bird surface temperature measurements at each time point to one of four categories: normothermy, shallow torpor, transition to and from deep torpor, or deep torpor. We defined these categories using individually assigned thresholds for each bird based on the values and patterns of eye surface temperature and ambient temperature. We used eye surface temperature of the bird once it had settled, but its eyes were still open, to define resting normothermic temperature per individual. Once the eyes were closed, we considered the bird asleep (Mascetti, 2016). If eye surface temperature dropped more than 2°C below the resting temperature (Walker et al., 1983), we visually classified the birds into one of the other three categories based on (1) rate of temperature change (stable, slow change, rapid change), and (2) magnitude of decrease of eye surface temperature below normothermic temperature, and above ambient temperature.

Birds were considered to be in shallow torpor if they dropped more than 2°C but less than 20°C below their resting temperature (but were still above ambient temperature), and maintained that temperature for more than 10 min (stable temperatures). These thresholds are certainly somewhat arbitrary, but they clearly overlap a range of temperatures that have been previously defined by others to be torpor (Geiser, 2021). Past studies have advocated for individual-specific thresholds to define normothermy and torpor (Barclay et al., 2001). An equivalent of shallow torpor has previously also been referred to as nocturnal hypothermia (e.g. McKechnie and Lovegrove, 2002), but more recent work suggests using the term ‘hypothermia’ only for a pathological condition that an animal cannot rewarm from by itself, and using ‘torpor’ for controlled body temperature reduction with independent rewarming (Geiser et al., 2014). Normothermy has been defined as a state when body temperatures are within ±2°C of resting body temperature (Hetem et al., 2016). Thus, normothermy is, by definition, not associated with the flexibility in body temperature that these birds are clearly capable of; and at those times that we defined as shallow torpor, they are not at their minimum torpid body temperature. From past usages of the term shallow torpor (Berger, 1984; Geiser, 2021; Jonasson and Willis, 2012), shallow torpor is the most appropriate term for what we observed. Measurements were assigned to the transition category if they dropped or increased rapidly between normothermy and deep torpor, or between shallow and deep torpor (i.e. transitions were defined by rapid, large temperature changes; mean±s.e. rate of change of 0.45±0.06°C min⁻¹, up to 3°C min⁻¹; see Fig. S2 for details). Birds were considered to be in deep torpor if eye surface temperature was close to ambient temperature, or if it was maintained below 20°C without dropping any lower (stable, low temperatures), for an extended period (highest reported hummingbird torpid body temperature is 22°C; Bech et al., 1997; Hainsworth and Wolf, 1970). This is by no means steady-state torpor, but does cover a range of temperatures that is not normothermy. Categories varied slightly across individuals because we assigned category thresholds per bird based on its surface temperature patterns relative to resting and ambient temperature, and based on the rate of temperature change.

Normothermy, shallow torpor and deep torpor could be distinguished by the relationship between eye surface (response) and ambient (predictor) temperature. While a normothermic homeotherm can maintain a relatively stable body or surface temperature over a large range of ambient temperatures, in deep torpor the body and surface temperatures become a positive function of ambient temperature. Therefore, we would expect deep torpor to have a steep slope and a very low intercept. If shallow torpor exists, then the slopes for normothermy and shallow torpor should be similar and low, while their intercepts should vary (shallow torpor lower than normothermy). Additionally, we expected species to use these torpor categories differently, and expected mass to negatively influence torpor use (birds with greater energy stores should use torpor less, Powers et al., 2003).

Ambient temperature usually declined steadily over the course of the night (e.g. in Fig. 2). In 2017, ambient temperature averaged 13±4.6°C (mean±s.d., range 3 to 23°C), and in 2018 it was 11±5.6°C (range −1 to 24.6°C). Most nights ranged between 5 and 20°C (mean 12°C), except for one particularly cold night when ambient temperature was between −1 and 14°C (20 May 2018), and one especially warm night (5 June 2018) during which ambient temperature ranged between 15 and 25°C.

The surface temperature of normothermic birds and birds in shallow torpor peaked near the eye and decreased from the eye towards the tail (Figs 2 and 3). Birds in deep torpor were evenly cold. Night-time eye surface temperature varied overall between 5.9 and 38°C (see Fig. S1). Active birds at the beginning of the night had normothermic temperatures ranging between 31 and 38°C. This wide range included birds that were flying around in the chamber before settling and birds at rest. When birds settled down, normothermic temperatures usually stabilized (when the bird was resting with eyes open) at 31°C, so we usually considered minimum normothermic resting surface temperature to be around 31°C. In some cases, birds stabilized at 29°C, at both the start and end of the night, with minimal fluctuation; in these cases we set the resting normothermic threshold to be 29°C. Maximum eye surface temperature ranged from 29–38°C in normothermy to 19.5–29°C in shallow torpor, and 5.9–24.1°C in deep torpor.

The full model for surface temperature (where the slopes and intercepts vary by category and species) allowed us to identify and quantify the various thermal categories, including shallow torpor (Fig. 3; Tables S3 and S4). Mass did not seem to have a large effect on surface temperature given the other factors, but year did.

In this model of the factors affecting surface temperature, the normothermic and shallow torpor categories had similar, very low slopes (0.11), while the normothermy intercept was about 4°C higher than the shallow torpor intercept. A 4°C drop from normothermy has previously been categorized as being a form of torpor (Ruf and Geiser, 2015). In contrast with these thermoregulating states, hummingbirds largely thermoconform in deep torpor (down to the ambient temperatures we measured them at, which were all above their minimum body temperature in deep torpor). In deep torpor, their surface temperature was closely tied to ambient temperature (slope of 0.85) and a low intercept about 20°C lower than the normothermy intercept. The transition category is a non-equilibrial physiological state, with an intermediate intercept 17°C lower than normothermy.

Shallow torpor was used by all species, but at varying frequencies (Fig. 4; Fig. S1). Of the 34 individuals we studied, all were normothermic for part of the night; 25 used shallow torpor for part of the night; 8 used shallow but not deep torpor; 20 transitioned between deep torpor and normothermy; and 20 used deep torpor (Fig. 4A).

All seven black-chinned hummingbirds used deep torpor, for an average of 49% of the night, while only three of these individuals used shallow torpor, for an average of 5% per night. Black-chinned hummingbirds spent 34% of the night on average in normothermy, and 12% in transition. The 14 blue-throated mountain-gems remained largely normothermic (67% of the total night-time), and used shallow torpor an average of 25% of the night, with four individuals remaining normothermic all night, 10 individuals using shallow torpor for at least some time and only four using deep torpor (4% of the night), with 3% of the night spent transitioning on average. The 12 Rivoli's hummingbirds were the most variable in their use of the various metabolic states, with 10 individuals using shallow torpor (33% of total night-time), six individuals using all four categories, four using normothermy and shallow torpor, one individual remaining normothermic all night, and one individual using all categories except shallow torpor. The Rivoli's hummingbird individuals spent an average of 43% of the night in normothermy, 8% in transition and 17% in deep torpor.

The model of thermal category frequency across species showed that there were overall clear differences in the thermal categories across all species. There were also species-specific differences in the use of normothermy, shallow torpor, deep torpor and the transition categories (Fig. 4B; Table S3). The model resulted in estimates that were more evenly distributed than the raw data (the two panels on Fig. 4B), but the differences between species were still clear.

The two blue-throated mountain-gems that we studied on an especially warm night maintained high surface temperatures all night (33–35°C). Another two blue-throated mountain-gem individuals that could have been late-stage juveniles appeared to behave similarly to adults: one used all four categories, and the other used only normothermy and shallow torpor. Contrary to our expectations, several Rivoli's hummingbirds and a few blue-throated mountain-gems used shallow torpor for 1–2 h after coming out of deep torpor, while the black-chinned hummingbirds never used shallow torpor after deep torpor.

We describe and quantify the novel use of shallow torpor in birds that are known to use deep torpor. Similar to mammals, and in contrast with previous studies that either describe birds as using shallow torpor or deep torpor, hummingbirds appear capable of using both. Hummingbirds in shallow torpor appear to thermoregulate to maintain surface temperature below normothermy. In contrast, hummingbirds in deep torpor largely thermoconform to ambient temperature. The intermediate shallow state could serve to balance night-time energy savings with the potential ecological and physiological costs of deep torpor. Reflecting what previous studies have found (Powers et al., 2003), birds with larger energy stores seem to have greater flexibility in avoiding deep torpor. The two larger species in our study used normothermy and shallow torpor for a greater proportion of the night than the smaller species. Our minimally invasive study design allowed us to thermally image hummingbirds under near-natural temperature cycles, without disturbing or touching the birds through the night, and allowed us to discover a new level of flexibility in hummingbirds' management of their night-time energetic needs.

All three species used all thermal categories, but unequally. Rivoli's hummingbirds used shallow torpor the most, followed by blue-throated mountain-gems. The small black-chinned hummingbirds used shallow torpor the least. It therefore appears that black-chinned hummingbirds, the smallest of the three study species, might have the least flexibility in managing their night-time energy budget, while blue-throated mountain-gems, the large territorial species, have the most flexibility. Individuals of the two larger species appeared to have more flexibility in regulating their night-time body temperature, commonly using shallow torpor or a combination of shallow and deep torpor. The more limited use of deep torpor in these two species is consistent with previous findings that these species tend to avoid deep torpor (Powers et al., 2003; Shankar et al., 2020).

Our results support the argument that there must be either physiological or ecological costs of deep torpor (Boyles et al., 2020; Ruf and Geiser, 2015), because hummingbirds that are clearly capable of using deep torpor sometimes use shallow torpor or avoid torpor altogether. Using shallow torpor rather than deep torpor could be especially beneficial in three scenarios. First, deep torpor in mammals (especially hibernation) is usually considered helpful in avoiding predation because torpid animals are less conspicuous to predators; but these animals are usually hidden in hibernacula or dens (Ruf and Geiser, 2015; Turbill et al., 2011). Torpid birds in trees, in contrast, might be more conspicuous, making shallow torpor more efficient than deep torpor in allowing them to respond to potential predators (Carr and Lima, 2013). Hummingbirds in shallow torpor could afford quicker rewarming times (<5 min), and quicker responses to predators or other external stimuli, relative to deep torpor for which rewarming to normothermy takes an average of 20–30 min (Bucher and Chappell, 1992; Shankar et al., 2020). Some mammals in torpor and even birds at least in shallow torpor (e.g. common poorwills; Austin and Bradley, 1969) are capable of locomotive activity (Geiser, 2021). Second, at least in mammals, the physiological costs of torpor include rewarming costs, immune suppression (Bouma et al., 2010; Prendergast et al., 2002), increased oxidative stress (Buzadžić et al., 1997) and potential sleep deprivation (Kräuchi and Deboer, 2011; Palchykova et al., 2002). There are hints that daily heterothermic mammals (Djungarian hamsters, Phodopus sungorus; 26 g) enter a euthermic state after torpor to recover from sleep deprivation (Deboer and Tobler, 2000; Palchykova et al., 2002). Shallow torpor would allow higher levels of metabolic function than deep torpor, perhaps facilitating some of the restorative functions of sleep, immunity and lowered oxidative stress. Though avian sleep has been studied to some extent (Beckers and Rattenborg, 2015), little is known about the physiological basis for torpor versus sleep in birds. Third, for nesting birds that need to keep their nest warm, shallow torpor could help balance the birds' need to maintain energy balance with the need to supply heat to their eggs or chicks. Nesting hummingbirds have been found to generally avoid torpor (with exceptions when energy stores seemed to be low); but the use of shallow torpor by nesting birds has not been evaluated (Calder, 1971; Howell and Dawson, 1954; Smith et al., 1974). If deep torpor had no ecological or physiological consequences, hummingbirds would likely maximize torpor use, or remain in deep torpor for the duration of the night after entering torpor. Instead, some individuals used shallow torpor not just before a deep torpor bout as we predicted but after emerging from a deep torpor bout, indicating that they may be trying to save energy but also balance these energy savings with the potential costs of deep torpor.

Here, we identified four thermal categories in hummingbirds: two thermoregulatory categories – normothermy and shallow torpor; a thermoconforming category – deep torpor; and the transition between deep torpor and the other categories. Though we have assigned them discrete names, it is thus far unclear whether the underlying physiological differences between these states in birds are continuous or discrete. In normothermy and shallow torpor, the animal actively thermoregulates to maintain a constant body temperature across a range of ambient temperatures. Based on the similar surface temperature slopes of normothermy and shallow torpor (Fig. 3), and the rapid transitions between normothermy and shallow torpor that we often observed, these two states seem metabolically continuous in hummingbirds (as they seem to be in doves; Walker et al., 1983). Shallow torpor, as defined here, could potentially be a metabolically inhibited form of normothermic sleep, but it is unclear whether the shallow torpor and deep torpor we report are on a similar metabolic spectrum, and we were unable to definitively distinguish sleep using only body temperature measurements. Multiple lines of evidence, especially in ground squirrels and pocket mice, from EEGs, measurements of brain temperature and metabolic rates indicate that mammals slow their metabolism continuously down from sleep into torpor, and regulate their body temperatures variably above minimum body temperature (Berger, 1984; Boyles and McKechnie, 2010; Boyles et al., 2013; Ruf and Geiser, 2015; Walker et al., 1979). Thus, though sleep and torpor appear to be on a continuous spectrum in mammals, it remains to be seen whether they are on a continuum in birds.

Given that hummingbirds can regulate between shallow and deep torpor, the biological relevance of minimum body temperature measurements must be assessed. Recent work with high-elevation Andean hummingbirds found that minimum torpid body temperature showed a phylogenetic signal, indicating that minimum torpid body temperature, at least at very cold sites, is evolutionarily conserved (Wolf et al., 2020). Shallow torpor can occur either because a bird's minimum possible torpid body temperature is relatively high (i.e. it does not have the capacity for deeper torpor), or when a bird regulates at a high, sub-normothermic, body temperature despite its minimum torpid body temperature being much lower (e.g. 15°C, indicating that it is capable of deep torpor; Fig. 1D). These two shallow torpor scenarios are indistinguishable (as in Fig. 1C) unless the species' ‘true’ minimum body temperature is known. If a bird regulates at a body temperature above its minimum, even though ambient temperature was lower, these measurements might appear to be minimum body temperature measurements although they are not. In Rivoli's hummingbirds, for instance, we found that eye surface temperature went as low as 5.9°C. However, Rivoli's individuals in the laboratory were previously reported to regulate their minimum body temperature at 12°C despite ambient temperature going lower (Wolf and Hainsworth, 1972). Such a large disparity in birds in deep torpor is unlikely to be due to differences between core and skin temperature, and could indicate either intra-specific differences in minimum body temperature or that the birds in the previous study were using a shallower form of torpor. This disparity could also be caused by birds reducing their blood circulation around the eye during torpor, but such regional variation seems unlikely given the small size of hummingbirds and the even distribution of low surface temperatures we observed in torpid birds. Minimum body temperature may therefore be lower than has been reported in some species. Currently, minimum body temperature across all hummingbirds is thought to vary from 3 to 22°C (Bech et al., 1997; Carpenter, 1974; Wolf et al., 2020). But if some of the hummingbirds measured were using shallow rather than their deepest possible torpor, the range of hummingbirds' true minimum body temperatures would be narrower or lower. Additionally, torpor studies in hummingbirds are often conducted in laboratory conditions, which could alter torpor responses (McKechnie and Lovegrove, 2002; Ruf and Geiser, 2015).

We propose three reasons for why this form of shallow torpor in birds has so rarely been detected (Prinzinger et al., 1992). First, small drops in oxygen consumption or body temperature might have been overlooked. Second, most studies of bird torpor are either done under controlled laboratory conditions or involve handling the birds many times at night to record body temperature. Birds in captivity are often overweight and have to be starved to enter torpor (Schleucher and Prinzinger, 2006). Laboratory torpor studies conducted at controlled temperature steps might have pre-empted the use of shallow torpor, because shallow torpor is presumably a fine-scale response to energetic state and environmental conditions, and controlled temperature steps or repeated handling might not elicit the same physiological responses as natural decreases in night-time temperature would (Geiser et al., 2000). Third, birds in laboratory settings are known to show altered torpor use relative to free birds: free-living animals often use torpor more frequently, and drop to a lower body temperature in torpor than laboratory animals (McKechnie and Lovegrove, 2002; McNab, 1989; reviewed in Ruf and Geiser, 2015). Taken together, under relatively predictable natural temperature patterns, hummingbirds might be able to use intermediate torpor states more often, while in the laboratory, low temperatures, the factor most often tested, might cause the bird to either stay awake or drop into deep torpor if energetically necessary.

At the whole-animal level, the next step in understanding avian torpor would be to combine respirometry, thermal measurements and measurements of breathing or heart rate while keeping in mind the possible existence of shallow torpor. These measures have been found to sometimes be uncoupled in torpor (O'Mara et al., 2018; Toien et al., 2011), and therefore studying whether they vary together in shallow torpor would be the first step in identifying the physiological differences between sleep, shallow torpor and deep torpor. A promising future avenue for research would be to investigate which metabolic and genetic pathways shut down at different temperatures in hummingbird torpor. It remains to be seen whether other hummingbird and bird species that use deep torpor are also capable of shallower torpor, or whether such control over their torpid metabolism is unique to these two hummingbird clades. Our data indicate that these hummingbird species in a temperate environment with cold ambient temperatures often use shallow torpor; it is therefore possible that tropical species at sites with high ambient temperatures might be doing the same.

We thank Alexis Stark for help with data collection; Liliana M. Dávalos for discussions, statistical analyses and comments on the manuscript; Benjamin Van Doren for help with statistical analyses; Erich Eberts and Elise Lauterbur for useful discussions; and Jeffrey Levinton and Irby Lovette for helpful comments on the manuscript.