Cooler snakes respond more strongly to infrared stimuli, but we have no idea why

Some species have sensory systems that detect environmental stimuli that are not detectable by most other organisms. Electroreception in mormyrid fish is one example (Bullock, 1982); detection of infrared radiation (IR) by pit vipers is another. In both cases, these sensory systems bestow some obvious advantages to these taxonomic groups. Electroreception is used by mormyrids not only to detect the faint electrical fields generated by the prey they consume in the murky, light-free environments they live in, but also to communicate with each other (Moller, 1995). Pit vipers use IR detection both during behavioral thermoregulation (Krochmal and Bakken, 2003; Krochmal et al., 2004), and to sense prey in low-light environments (Noble and Schmidt, 1937; Bullock and Barrett, 1968; de Cock Buning, 1983b; Kardong and Mackessy, 1991; Kardong, 1992), an ability that apparently makes them more effective predators of nocturnal rodents when compared with similar species of non-pit-bearing vipers in common garden experiments (Kotler et al., 2016).

Pit vipers sense thermal IR using the eponymous facial pit organs located between the eyes and nostrils. Each pit organ houses a membrane that divides an outer chamber from an inner chamber. The membrane is covered with exceptionally sensitive warm receptors (ca. 0.001°C; Bullock and Diecke, 1956), forming a low-resolution, pinhole-camera eye (Goris, 2011; Bakken et al., 2012). The membrane is innervated by three branches of the trigeminal nerve that terminate in palmate warm receptors packed with mitochondria. These receptors sense the temperature distribution image on the pit membrane created by thermal radiation from environmental surfaces (Otto, 1972; Stanford and Hartline, 1980; de Cock Buning, 1984; Kohl et al., 2012). This image is merged with visual information in the optic tectum, and transmitted to the telencephalon, where it informs behavior (Goris and Terashima, 1973; Hartline et al., 1978; Gruberg et al., 1979; Terashima and Goris, 1979; Berson and Hartline, 1988). Facial pits alone can guide predatory (Noble and Schmidt, 1937; Bullock and Barrett, 1968; de Cock Buning, 1983b; Kardong and Mackessy, 1991; Kardong, 1992) and thermoregulatory behavior (Krochmal and Bakken, 2003; Krochmal et al., 2004), and probably serve other purposes as well (e.g. prey evaluation or general navigation; Schraft and Clark, 2017; Schraft et al., 2018).

Pit organ performance should depend on snake body temperature and thermal contrast between a target and its background. The warm receptors in the pit membrane show a phasic-accommodative response (Bullock and Diecke, 1956; Goris and Terashima, 1973). This is functionally important, as pit vipers are active at a wide range of core body temperatures (T_b) as low as 17–18°C (Brattstrom, 1965), and the facial pit membrane remains within −1 to −3°C of T_b (Cadena et al., 2013). The facial pit functions because the sensitive endings do not respond to slow changes of 20°C, and thus to T_b change, but do sense rapid changes of as little as 0.001°C (Bullock and Diecke, 1956).

Thus, the snake detects rapid membrane temperature changes produced by objects moving across a temperature-contrasting background (Bullock and Diecke, 1956; Van Dyke and Grace, 2010). Gracheva et al. (2010) have proposed TRPA1, a highly temperature-sensitive ion channel receptor, to be the main molecular mechanism mediating temperature sensitivity in the pit neurons. TRPA1 cloned from western diamondback rattlesnakes (Crotalus atrox) was found at a high (400×) concentration in the trigeminal ganglion innervating the pit, but not in either the dorsal root ganglion, which innervates the posterior body, or in non-IR sensing snakes (Gracheva et al., 2010). TRPA1 membrane ion current increased with temperature above a threshold of 27–28°C (Gracheva et al., 2010), consistent with the 28–30°C preferred T_b of crotalids (Brattstrom, 1965; Moore, 1978; Brown et al., 1982).

However, the temperature response of TRPA1 does not predict behavioral response to temperature. Goris (2011) and Bakken et al. (2012) have noted that typical active rattlesnakes have a 17–18 to 34–35°C voluntary T_b range (Brattstrom, 1965). Rattlesnakes studied at room temperature (22–24°C), which is below the response threshold for TRPA1, nevertheless readily respond to moving targets (Safer and Grace, 2004; Ebert and Westhoff, 2006; Van Dyke and Grace, 2010), and wild rattlesnakes commonly forage with T_b values at or below 20°C (Secor, 1995; Beck, 1996; Putman and Clark, 2017).

This difference between TRPA1 and behavioral temperature responses may be due to the methods used. Kang (2016) found a threshold of 26.3°C at low ion currents, and 23.3°C at high currents for rattlesnake TRPA1. Both Gracheva et al. (2010) and Kang (2016) cloned C. atrox TRPA1 in Xenopus eggs, but function may depend on membrane or cytoplasm factors. Isolated C. atrox trigeminal neuron bodies responded from 15–18 to 40°C (Pappas et al., 2004). Also, the ion channels studied by Gracheva et al. (2010) and Kang (2016) were from the nerve body. TRPA1 channels in the pit membrane nerve endings have not been examined, even though the nerve endings function when severed from the nerve body (Bullock and Diecke, 1956; de Cock Buning et al., 1981). Nerve ending physiology suggests that the pit organ can function over a T_b range of 18–38°C (Bullock and Diecke, 1956). In addition, processing in the central nervous system (Hartline et al., 1978; Stanford and Hartline, 1980; Berson and Hartline, 1988; Goris, 2011) may be affected by T_b.

As G. A. ‘Bart’ Bartholomew noted (Bartholomew, 1986), ‘But even when a body of data can be logically linked to past information, the new data may not necessarily be what a prudent and reasonable investigator would have predicted from that information’. To avoid this trap, reductionist studies of cellular-level processes should be informed by, and consistent with, the operation of the facial pit in intact animals over a temperature range that reflects activity in nature. In addition to defining the target and background, temperature also affects physiological processes, from the ion channel to information processing in the telencephalon (Prosser and Nelson, 1981; Angilletta, 2009). A behavioral assay that systematically explores all relevant temperatures – body, background and target – is clearly needed to quantify how whole-organism performance is affected by the interacting influences of thermal contrast and body temperature. Here, we implement such an assay by examining the behavioral response of southern Pacific rattlesnakes (Crotalus oreganus helleri) to an infrared stimulus while systematically varying background, target and body temperatures.

where A (m²) is the area of the pendulum bob, d (m) is the distance from facial pit to the bob, σ=5.67×10⁻⁸ (W m⁻² K⁻¹), π=3.14, T_tgt is the temperature of the bob and T_bkd is the background temperature. Temperatures are measured in kelvin (K=°C+273.16).

Based on published studies (de Cock Buning et al., 1981; Safer and Grace, 2004; Van Dyke and Grace, 2010), we expected snakes to respond to both positive (stimulus warmer than background) and negative (stimulus cooler than background) contrasts, but more strongly to positive contrasts. A strong response to positive contrasts is also expected because a major function of the pit organ system in our species is the detection of endotherms (warm targets) moving at night (cool backgrounds).

Based on a detailed study of western diamondback rattlesnakes (C. atrox; Ebert and Westhoff, 2006), we expected a response at all positive temperature contrasts, but a weak response at 2°C contrast (corresponding to a signal strength of 0.173 W m⁻²), increasing with greater contrasts (4 to 20°C, 0.35 to 1.89 W m⁻²). We expected no response to our control presentation, which was a target presented at the same temperature as the background.

The thermal reaction norm (Angilletta, 2009) of behavioral responsiveness in relation to T_b potentially depends on the effect of temperature on central nervous system (CNS) functions related to cognition, and on muscle function related to locomotion and striking. Functional physiology studies show that the typical effect of temperature on both muscle and neurological function is for the response to increase with temperature to a peak, followed by a decline (Prosser and Nelson, 1981; Angilletta, 2009). The thermal response of TRPA1 Q₁₀ follows the same pattern (Kang, 2016). Bullock and Dieke (1956) found the curve of pulse frequency versus temperature was nearly flat from 17 to 29°C. In addition, several reductionist studies on the pit organ system suggest a lower T_b limit on responsiveness. Gracheva et al. (2010) found that rattlesnake TRPA1 studied in Xenopus oocytes was inactive at room temperature, but robustly activated above 28.0±2.5°C, which predicts behavioral response to stimuli when T_b≥28°C. Kang (2016) found an ion current-dependent TRPA1 threshold from 23 to 26°C, predicting behavioral response for T_b≥23–26°C. Pappas et al. (2004) found that isolated trigeminal neuron bodies responded above 18°C, predicting behavioral response for T_b≥18°C. Thus, we predicted that the thermal reaction norm for response to thermal contrast would be minimal response at low T_b, increasing with T_b, reaching a peak at or above typical preferred T_b of ca 28°C, and then declining with a further increase in T_b.

The San Diego State University Institutional Animal Care and Use Committee approved all procedures (APF 16-08-014C). All experiments were performed using 18 sub-adult to adult (snout–vent length, 0.5–1.0 m) southern Pacific rattlesnakes (Crotalus oreganus helleri Meek 1905). Twelve of these were snakes that were being translocated away from populated areas of the Camp Pendleton Marine Corps Base in San Diego County, California, USA, and had been in captivity for only 2–3 months prior to our experiment. The other six individuals were collected from Rancho Jamul Ecological Reserve in San Diego County, and had been in captivity for 1–3 years prior to the experiment. All 18 snakes had fed at least twice in captivity prior to the beginning of the experiment. Animals were housed in paper-lined 60×40×40 cm plastic terraria and provided with a hide box where they spent most of their time. The box was equipped with rails such that a sliding door could be inserted and removed from above. This allowed us to move snakes to and from the experimental apparatus in their ‘home’ box without direct handling. Thus, experimental trials were done using relatively unstressed animals.

We tested snakes with an experimental stimulus consisting of a pendulum bob moving in front of a background plate (Fig. 1), both regulated at contrasting temperatures. The background consisted of an aluminium plate (61×40×0.635 cm) with an array of copper tubing (0.635 cm o.d.) cemented to the back with aluminium-filled epoxy resin (EpoxAcast 655, Smooth-On Inc., Macungie, PA, USA), and arranged to ensure uniform temperature over the background. The resin was built up to 0.5–0.6 cm thickness to maximize heat transfer to the aluminium plate, so that the final assembly was ∼1.2 cm thick. Insulating foam was added to the back of the plate to reduce unnecessary heat transfer. Water circulated from a temperature-controlled water bath (VWR Scientific model 1160, Chicago, IL, USA) regulated the plate temperature, and plate temperature was recorded as the average of four thermocouples, one embedded in each quadrant of the plate. Temperature uniformity was further verified with a calibrated thermal imager (ThermaCam PM575, FLIR, North Billerica, MA, USA).

The stimulus was a pendulum bob consisting of a square aluminium plate (6.35×6.35×0.5 cm) with a square Peltier element (5×5 cm; model 12711-9M31-24CW, Custom Thermoelectric, Bishopville, MD, USA) and heat sink cemented to the back. It was suspended from two 1 m lengths of 2.6 mm hypodermic tubing, each containing one of the two 0.8 mm conductor heater supply wires and a duplex thermocouple with 0.013 mm diameter conductors (type TT, Omega, Norwalk, CT, USA). The Peltier heater was connected to the 0.9 mm wires. The sensing tips of the thermocouples were insulated and inserted in close-fitting holes drilled in the plate. One thermocouple controlled the temperature of the aluminium plate, and the other monitored the plate temperature. The assembly was suspended from an X-shaped frame with a 1.27 cm wide hinge arranged such that the pendulum would swing parallel to the plate. A 3×3 cm square of galvanized steel sheet ca 0.8 mm thick was attached to one of the tubes. This engaged an electromagnet that held the pendulum to one side behind a visual shield. Interrupting the electromagnet current released the pendulum silently. The purpose of the shield was to cause the stimulus to appear suddenly. As a visual demonstration, Movie 1 shows a thermal image of a 36°C pendulum bob swinging across a 30°C background. A single-turn wheel linked to the pendulum reset the mechanism between presentations. The heating current was supplied by a purpose-built power circuit regulated by a PID (three-mode) controller (Series 1600, LOVE division, Dwyer Instruments, Wheeling, IL, USA). Experiments were conducted in complete visual darkness, but as an additional guarantee that thermal radiation was the only stimulus, the background and pendulum were painted uniform black (1916 Ultra-Flat Black, Rust-Oleum, Vernon Hills, IL, USA). Experiments were conducted at times when the snake housing was illuminated.

During experiments, the hide box and snake were placed on a stand in front of the stimulus assembly, such that the snake's facial pits would be 30±1.5 cm from the stimulus. Each hide box had a short duplex thermocouple (type TT, 0.5 mm conductors, Thermo Electric, Saddle Brook, NJ, USA) positioned on the center floor of the box such that it would be in contact with the snake's body. Preliminary trials found that the temperature recorded by this thermocouple was always within 1°C of the snake T_b as recorded by the calibrated thermal imager. Therefore, this thermocouple temperature was used to represent snake T_b. Prior to each trial the hide box thermocouple was connected to a thermocouple extension wire, and all lights were turned off in the procedure room housing the apparatus as well as the anteroom adjacent to the door of the procedure room. Experimenters could operate the entire experimental apparatus remotely from an adjacent room (the control room), leaving the snake isolated in the procedure room. A video camera with built-in near-IR lamp (Handycam DCR85, Sony Electronics Inc., San Diego, CA, USA) placed above the background plate was aimed into the hide box opening to record snake behavior at 30 frames s⁻¹. The sliding door could be lifted by a string extending through a 6.3 mm copper tube into the adjacent control room. The curved copper tube blocked light from the control room by multiple reflections. A real-time image from the camera was transmitted to a monitor in the control room. A digital data logger (model CR23X, Campbell Scientific, Logan, UT, USA) recorded heater voltage, electromagnet voltage, and temperatures of the snake, pendulum, background and procedure room at 1 s intervals. These values were also displayed in real time on a monitor adjacent to the video feed in the control room.

The experimental design called for three nominal T_b values (20, 27 and 34°C), but we were not able to control T_b effectively once the animals were in the procedure room, which was kept at 18°C. Consequently, T_b at the times the stimulus was presented are distributed fairly uniformly over a range from 18 to 36°C (Fig. 2). Background temperatures were 20, 30 and 40±0.2°C. We examined each snake at one nominal body and background temperature combination, and pendulum (target) temperatures of 20, 24, 28, 32, 36 and 40±0.1°C. Each snake was used three times with different background and nominal body temperatures each time. Apparatus constraints (long time needed to cool the pendulum bob) required that we present pendulum temperatures in an ascending series, but we varied the starting temperature so that each was the initial temperature for one experiment. Thus, it was necessary to wait only once for the pendulum bob to cool. Thus, there were three body, three background and six initial stimulus temperatures (3×3×6=54 experimental combinations), each with a series of six stimulus temperatures, giving N=324 total stimulus presentations. We arranged nominal T_b, background temperatures, and initial stimulus temperatures in a modified Latin square design. As we conducted experiments in visual darkness, control for non-thermal stimuli was the stimulus presentation at 0°C contrast with the background. This avoided the problematic behavioral effects of blindfolding and pit-blocking.

After allowing 5 min for the snake to settle from being moved, we raised the sliding door and observed the snake on the video monitor until activity (head movements and tongue flicking) ceased. Snakes reliably faced out of the hide box towards the background area crossed by the stimulus and rarely attempted to move out of their box, which was placed on a raised platform. After snakes had settled we set the stimulus temperature to the first in the sequence, and released the pendulum bob for three to four swings. We then reset the pendulum and increased the pendulum temperature to the next in the sequence. We waited at least 3 min between sequential presentations of the pendulum, and we also did not present the pendulum until the snake had been still (i.e. no head movements or tongue flicks) for the preceding 30 s.

We analysed behavioral responses in the form of tongue flicks and head movements (typically subtle tracking movements of the head following the motion of the pendulum) from the video recordings. The experimental conditions were hidden from the scorer. For each stimulus presentation, we examined a 5 s video clip of three pendulum swings frame-by-frame. This 5 s clip corresponded to the first three swings of the pendulum, such that we only scored behaviors exhibited prior to the reattachment of the pendulum to the magnet, as this mechanism generated quiet sounds that presumably could be an auditory cue detected by snakes. We counted the total number of frames (out of 150) where the tongue was extruded (for tongue flicks) or the head moved from the previous frame (for head movements). We generated three response variables from these data: tongue flicks, head movements and whether either response (any tongue flicks or head movements) was present (1) or absent (0). For visual demonstration, Movie 2 shows a typical no response and Movie 3 shows a typical positive response. For graphical evaluation, we first grouped the data from all snakes recorded with each pendulum–background contrast and computed the fraction of trials (f=0 to 1) where a response was present.

For each pendulum presentation we recorded the following variables: snake identity, snake T_b, background temperature, pendulum temperature, contrast between pendulum and background, trial number (either the first, second or third trial for that snake), stimulus order (first, second, third, fourth, fifth, sixth) and days since the snake last ate. Because our experiment was designed to balance trial number and stimulus order across replicate trials in a cyclical fashion, we did not include these factors as explanatory variables in the model (thereby improving parameter estimation) and instead examined them graphically to determine whether there was any apparent association with response variables. We performed trials only if snakes had not fed for at least 6 days, and as the time since the last meal varied, we also examined days since the snake last ate graphically.

We examined our three response variables (tongue flicks, head movements and binary responsiveness) independently using a generalized linear mixed model (GLMM) analysis framework. Because they are zero-inflated count data, we used negative binomial distributions for tongue flicks and head movements. We used a binomial distribution for models with yes/no response data. We ran separate models for positive (pendulum warmer than background) and negative (pendulum cooler than background) values of thermal contrast because the shape of the response curve may be different for these contrasts (Fig. 3; see also Van Dyke and Grace, 2010; Chen et al., 2017). Thus, our final models included snake body temperature and thermal contrast (either positive or negative) as fixed effects, and included a random intercept for snake ID to account for repeated measurements on the same individuals. All analyses were performed in R version 3.3.2 (https://www.r-project.org/) using the packages glmmTMB for mixed model analysis and visreg for generating plots from GLMMs. The raw data are available in the supplementary material (Dataset 1).

Snakes exhibited tongue flicking and head movements to the moving pendulum.

When there was no temperature contrast between the pendulum and the background, snakes showed only two responses in 36 trials (f=0.055). This is essentially the background rate when no stimulus was present. Therefore, we conclude that snakes were responding only to temperature contrast, and were not responding to any visual, auditory, electromagnetic or vibration stimuli associated with pendulum movement.

Here, we present results focusing on binary responsiveness (yes/no for exhibiting any response). All four response variables (tongue flicking, head movements, binary responsiveness and binned responsiveness) showed highly similar patterns, and binary responsiveness offers the clearest biological and statistical interpretation – regardless of whether the stimulus was detected. See Figs S3 and S4 for analyses using tongue flick and head movement data. We also include Fig. S1 displaying the relationships between behavioral counts and days since last meal, trial number and trial order, showing that none of these factors showed any apparent relationship with snake behavior.

Although we could not control T_b precisely, a plot of our sample space demonstrated that there is no confounding between T_b and contrast (Fig. S2). Because there is an interaction between contrast and background temperature (positive contrasts only with 20°C background, positive and negative contrasts with 30°C background, and negative contrasts only with 40°C background), we used a graphical evaluation. We first grouped the data recorded with each pendulum-background contrast and computed the fraction of trials (f=0 to 1) where a response was present. We then plotted f versus temperature contrast in Fig. 2 with different symbols for each background temperature.

Snakes responded more strongly to positive than negative contrasts of the same absolute value (Table 1, Fig. 3). The probability of a response increased with positive contrast until reaching a putative asymptote around 80% at contrasts of +15°C or greater (≥0.898 W m⁻²). The increase in the probability of response to increasingly negative contrast is much weaker, and reaches only about 40% even at the largest contrast (−20°C).

Surprisingly, the probability of a response decreased as T_b increased from 18 to 36°C. The thermal reaction norm for probability of response versus T_b is therefore exactly opposite to predictions based on reductionist studies of TRPA1 ion channels and neurophysiological studies of trigeminal neurons. It is also opposite to predictions based on the temperature response curves for motor and neural functions.

We were able to elicit clear and reliable behavioral responses from rattlesnakes by presenting a moving target that exhibits temperature contrast with the background. Snakes were tested in the absence of any other stimuli, and exhibited no response to the target when the temperature did not contrast with the background. As expected, we found very strong positive associations between snake response and both positive and negative thermal contrasts. Snakes detected even low contrast targets. For our smallest positive contrast (+2°C, 0.173 W m⁻² signal strength), about 50% of snakes responded, which is consistent with the study of Ebert and Westhoff (2006) investigating behavioral responses of C. atrox to thermally contrasting targets.

However, contrary to our expectations, we did not find that behavioral responsiveness to the target followed the expected thermal reaction norm of increasing above a threshold temperature, reaching a peak near preferred body temperature, and then declining. Instead, we found a strong negative relationship between body temperature and responsiveness, with cooler snakes responding most strongly; this relationship was much weaker with negative contrasts (where the target was cooler than the background). This result is puzzling, because it is not only the opposite of most physiological and behavioral processes in ectotherms, it is also opposite to the temperature response profile of the TRPA1 ion channel, which is inactive at cool temperatures (<23–26°C, depending on ion currents) and increases linearly with warmer temperatures. We know of no obvious explanation for why the pit organ should be more effective or snakes more motivated to move their heads or flick their tongues at lower body temperatures.

Our results are similar to those of Cadena et al. (2013) for South American rattlesnakes (Crotalus durissus). At very low humidity, they found nasal evaporation caused a 2.6°C maximum cooling of the rostrum (and presumably the pit membrane as well) 30 s after a stimulus, and ≤0.5°C fluctuations at resting respiratory frequency (ca. four or five per minute; see fig. 4A of Cadena et al., 2013). By our result, their maximum observed cooling of 2.6°C would be expected to produce an increase in probability of response of 8.6% (from 76.6 to 85.2%, assuming 25°C initial body temperature), and the resting frequency cooling would produce a 1.9% increase in probability of response. Differences in methods prevent quantitative comparison, but qualitatively Cadena et al. (2013) reported that snakes with a higher degree of rostral cooling by nasal evaporation struck more accurately, struck warmer regions of their prey, and relocated and consumed their prey faster. They concluded that, by cooling their pit organs, South American rattlesnakes somehow increased their ability to detect endothermic prey.

Cadena et al. (2013) proposed three mechanisms to explain the relationship between IR sensitivity and body temperature. First, they suggested that the thermal radiation signal would be increased by increasing the temperature difference between the warm stimulus (mouse) and the pit membrane. The second and third hypotheses proposed by Cadena et al. (2013) are similar in that they assume that the membrane temperature receptors must be ‘reset’ by being cooled after being heated by thermal radiation. The second mechanism is that ‘a decrease in the temperature of TRPA1 channels via respiratory cooling would “prepare” the channels for an upcoming stimulus, especially if cooling results in silencing all ion channel activity prior to a thermal stimulus’. The other ‘reset’ mechanism is more of a supporting argument than a mechanism: ‘The dense blood supply to the pit organ membrane acts as a heat exchange mechanism that effectively removes heat from the heat receptors. As a new stimulus is conveyed to the membrane, this heat exchange process allows the receptor to return to its resting state after activation from a thermal stimulus, preventing the formation of “afterimages” … [this indirectly] supports the notion that a cool pit organ could also be important to maintain the high sensitivity of the heat receptors…’ (Amemiya et al., 1999; Goris et al., 2007, 2000).

We believe this mechanism is unlikely, as the difference between stimulus and background temperatures projected onto the pit membrane is affected equally by a change in membrane temperature. Bullock and Diecke (1956) demonstrated experimentally that temperature contrast with the background determines pit membrane response, and this response is unaffected by membrane temperature. Our results and many other studies also show that only the thermal contrast between an object and the background affects responsiveness, regardless of the absolute temperature of each (e.g. de Cock Buning, 1983b; Grace and Van Dyke, 2005; Van Dyke and Grace, 2010; Goris, 2011).

There are problems with any version of a ‘reset’ mechanism. Notably, it presumes that (1) the scene and target are warmer than the snake; else blood flow would warm rather than cool the receptors, and (2) some shutter mechanism exists whereby a new stimulus is periodically flashed onto the pit membrane. However, there is no shutter that we know of, and radiative heat transfer between the membrane and the environment is continuous. The only way that one continuously varying temperature distribution image can be replaced by another more quickly is to decrease the thermal time constant of the membrane. The thermal time constant is the time required for the temperature of a patch of the membrane to complete (1−1/e)=63% of its temperature response to a step (extremely sudden) change in radiative (or other) heat flow from the scene. In heat transfer terms (Bakken, 1976; Bakken and Krochmal, 2007), the time constant equals (mass of the patch of membrane)×(heat capacity)/(overall thermal conductance to and from the patch by radiation, conduction and convection). Thus, increasing thermal conductance, e.g. by increasing blood flow, will make the time constant smaller and hence the pit membrane will respond more rapidly to a change in radiative heat flux.

However, there is a cost to reducing the time constant. The temperature change of the same patch of membrane in response to a change in thermal radiation heat flux is (change in heat flux)/(overall thermal conductance); thus, increasing thermal conductance to ‘erase’ a previous temperature image decreases the already minute temperature contrasts in membrane images.

There is no evidence of selective pressures to increase conductance to ‘erase’ temperature afterimages. Rather, the contrary is true. Exceptional temperature sensitivity is essential to facial pit function. First, thermal radiation from large (several °C) changes in external surface temperatures produce only minute (0.01 to 0.001°C or less) temperature changes on the pit membrane (Bullock and Diecke, 1956; Bakken and Krochmal, 2007). Second, while the poorly resolved image (Bakken and Krochmal, 2007; Bakken et al., 2012) is apparently sharpened in the lateral descending trigeminal tract (Stanford and Hartline, 1980), computational image sharpening is limited by the dynamic range and noise in the unprocessed image, i.e. by temperature contrast sensitivity (Sichert et al., 2006; Gonzalez and Woods, 2017).

In this context, it is not surprising that natural selection resulted in the pit viper sensory membrane being suspended with an air space on both sides, which reduces the thermal conductance to its surroundings. This optimizes temperature contrast sensitivity (e.g. Bullock and Diecke, 1956) at the expense of rapid response. Consequently, the pit viper facial pit is five to 16 times as sensitive as those of boas and pythons, which have no insulating air space (de Cock Buning, 1983a; Zhang et al., 2015). In technology, the sensitivity advantage of the suspended membrane structure of the pit viper has been borrowed, explicitly and successfully, in biomimetic thermal image sensor design (Zhang et al., 2015).

Thus, several lines of evidence indicate that abundant blood supply to the facial pit membrane is a thermal liability rather than a benefit. An alternative function suggested by Goris et al. (2007) is to supply oxygen and glucose to the sensory endings. This appears likely, as Kang (2016) has demonstrated that very high TRPA1 ion currents maximize the Q₁₀ response to small temperature changes. The need for energy to drive these high ion currents may account for the spectacular concentration of mitochondria in pit membrane sensory endings (Bleichmar and De Robertis, 1962; Meszler and Gennaro, 1970; Zischka et al., 2003). The efficient functioning of these mitochondria requires abundant, efficient and precise delivery of oxygen and glucose to where it is needed, reflected in the dense vascularization and local control of blood flow in the pit membrane reported by Goris et al. (2007). This would be similar to the vascular physiology of other neural tissue, which controls perfusion locally to meet metabolic demand (Itoh and Suzuki, 2012).

There is clear behavioral evidence for both decreased responsiveness with warmer pit membranes (Cadena et al., 2013) and body temperatures (present study), and for accommodation to slow changes in T_b (Bullock and Diecke, 1956). However, our results do not isolate the locus of response inhibition by higher body temperature, which might be found anywhere between cytoplasmic cofactors altering TRPA1 ion channel response to ecological considerations in telencephalon processing of facial pit inputs.

Behavior reflects many inputs in addition to simple sensation, and ecological processes may be significant. This leads to two categories for further work.

First, neurophysiological responses should be traced on up to higher centers in the CNS to identify or eliminate direct temperature effects on brain function and thus behavior. A physiological effect on the CNS seems improbable, based on the results of Cadena et al. (2013). They reported that cooling only the rostrum (i.e. nasal and facial pit tissues) by respiratory evaporation improved facial pit performance. Based on thermographic surface temperature (C. durissus, see fig. 1 of Cadena et al., 2013; C. atrox, Fig. 4), this cooling does not extend to the eyes, suggesting no brain cooling and no CNS physiological mechanisms. However, there may be hidden vascular brain cooling anatomy, and our results and those of Cadena et al. (2013) might reflect different mechanisms. Definitive studies are needed.

Alternatively, ecological considerations, such as predator avoidance, may determine the T_b response. It could be argued that cooler snakes are more motivated to respond to novel stimuli. For example, because a cold snake moves more slowly it is more vulnerable to predation, and so may be more motivated to fully identify the source of a thermal signal, thereby resulting in an increased rate of tongue flicking at low T_b. Against this hypothesis, we obtained similar results for tongue flicks and head movements (Figs S3 and S4). Head movements are more noticeable than tongue flicks and could give away the snake's location. Consequently, head movements may be expected to decrease at cooler body temperatures where snakes feel more vulnerable. Furthermore, increased motivation to respond at cooler T_b cannot explain the finding of Cadena et al. (2013) that the cooling of only the nose increases responsiveness.

Fortunately, the question of behavioral motivation changing with T_b is readily tractable, and additional behavioral experiments will test these alternative hypotheses directly. In general, more ecological studies are needed to fully explore the inhibition of behavioral response at higher body temperatures, and more generally, to identify its adaptive value and what circumstances might have led to its evolution.

Regardless of what role ecological factors may play, the increased responsiveness seen when cooling only the facial pit, as reported by Cadena et al. (2013), suggests that a biochemical and neurological basis for this effect exists in the sensory organ itself. However, this basis is unknown, and is thus a promising area for fruitful future research. The only cellular function that clearly increases significantly as temperature decreases is the number of vesicles in the free nerve endings in the pit membrane (Terashima et al., 1995). How this might be related to the function of the facial pit system is unclear, but it could be related to the energy generation needed to maintain the high TPA1 ion currents that maximize the Q₁₀ response to small temperature changes (Kang, 2016). Additionally, in the current study, we did not obtain contrast response data at low contrast levels to determine whether body/membrane temperature affects threshold contrast sensitivity (cf. Ebert and Westhoff, 2006). This question should be addressed in future research in order to provide the necessary whole-organism parameters for efforts to find the mechanism behind decreased responsiveness at high T_b.

For helping with data collection, we thank Martin Cardona, Hao Duong, Sarah Gleason, Amy Orduno and Brianna Rios. We thank Nathaniel Redetzke for help with obtaining animals for this study. Ray Huey and another journal reviewer improved the clarity of presentation and suggested ideas for the Discussion. We thank Indiana State University for loaning some equipment.