Brain energy metabolism as an underlying basis of slow and fast cognitive phenotypes in honeybees

Slow–fast phenotypic differences in terms of behavior, life history and physiology, integrated into a suite of traits described as the ‘pace of life’, have lately attracted considerable attention (Ricklefs and Wikelski, 2002; Réale et al., 2010). Metabolic rate, the rate at which organisms acquire, transform and expend energy, and therefore often described as the fundamental biological rate (Brown et al., 2004), has been cited as the potential pacemaker that drives this slow–fast phenotypic variation (Glazier, 2015; Biro and Stamps, 2010). Although there is substantial empirical support for an association between metabolic rate and slow–fast phenotypic differences, both between and within species (Wiersma et al., 2007; Pettersen et al., 2016; Wong et al., 2021), the mechanistic relationship between metabolic rate and behavioral traits is complex and far from clear (Careau et al., 2008; Salzman et al., 2018; Chung et al., 2018). In this context, it has also been proposed that such slow–fast differences are fundamentally tied to cognitive differences related to a speed–accuracy trade-off in decision-making, in which fast phenotypes spend less time gathering information, making rapid but less accurate decisions, compared with slow phenotypes (Carere and Locurto, 2011; Sih and Del Guidice, 2012; Dougherty and Guillette, 2018).

Both behavioral ecology and neuroscience research have independently emphasized the role of energy metabolism in different cognitive processes and in the modulation of complex behavioral phenotypes (Niven and Laughlin, 2008; Mathot and Dingemanse, 2015), but few studies have integrated this perspective to understand the underpinnings of behavioral variation (Rittschof et al., 2015; Coto and Traniello, 2021). Neural tissue is widely known to be energetically expensive to produce, operate and maintain (Laughlin et al., 1998; Laughlin, 2001; Ames, 2000; Lennie, 2003). This is substantiated by a large body of work demonstrating that cognitive processes such as learning and memory formation are associated with significant energetic costs (Déglise et al., 2003; Mery and Kawecki, 2005; Jaumann et al., 2013; Plaçais and Preat, 2013). There is also substantial inter-individual variation in energy production and usage in the brain that has, in turn, been linked to behavioral variation, notably in social species (Hollis et al., 2015; Rittschof and Schirmeier, 2018).

In social insects, behavioral variation – the underlying basis for division of labor – has been shown to be correlated to differences in brain organization (Molina et al., 2009; Muscedere and Traniello, 2014; Kamhi et al., 2016). More specifically, in terms of cognitive differences, slow–fast differences in behavior were shown to match the predicted differences in metabolic rate and cognitive traits in honeybees (Mugel and Naug, 2020; Tait and Naug, 2020). This provides a background to test whether slow and fast cognitive phenotypes in honeybees exhibit any differences in the energetic capacity of their brains. In this study, we therefore first classified bees into slow and fast cognitive types using a discrimination and reversal learning assay based on the principle that fast phenotypes are defined by fast discrimination learning and slow reversal learning while the opposite is true for slow phenotypes (Sih and Del Giudice, 2012). We then measured the whole-animal metabolic rate of these bees using flow-through respirometry, followed by measuring the maximum oxidative phosphorylation (OXPHOS)-linked respiration rate of their brain using high-resolution respirometry. The broad goal of the study was to test the hypothesis that slow–fast cognitive differences and the related speed–accuracy trade-off are associated with differences in energy metabolism at the level of the brain, which in turn is reflected in differences in brain size and whole-animal metabolic rate.

The bees used in the experiment came from six source colonies of the honeybee Apis mellifera. Brood frames with pupae were collected one day prior to adult emergence and kept in an incubator set at 32°C. Upon emergence, individual adult bees were marked with paint on their thorax and introduced into an experimental hive that consisted of two brood frames, a full honey frame, a laying queen and workers. Marked bees of foraging age were collected at the hive entrance, immobilized on ice, and harnessed within a 4.5 cm long plastic drinking straw with a small wire around the thorax. Each bee was fed to satiation with 30% sucrose solution and then kept starved for 24 h inside an incubator set at 27°C to increase motivation for appetitive learning. Before the start of the learning assay, all bees were tested for their responsiveness to sucrose by touching their antennae with 30% sucrose solution and any bee that did not extend its proboscis to this stimulus was excluded.

The discrimination and reversal learning ability of an individual bee were determined by using the proboscis extension reflex (PER) assay. This consisted of presenting a bee with an odor A (CS+) followed by a sucrose reward (US) and a second odor B (CS-) followed by a saline punishment in a predetermined pseudorandom sequence (ABBABABBABABAABAABAB) in a series of 12 trials with a 5 min inter-trial interval. Discrimination learning ability of a bee was measured first by pairing one odor with a sucrose reward (A+) and another odor with saline solution (B-) and, after a gap of 60 min, reversal learning ability of the same individual was assessed by reversing the odor pairings (A- and B+). The conditioned response of a bee, the extension of the proboscis to the CS alone prior to the presentation of US, to the sugar reward and its non-response to the saline solution were considered as correct choices and the opposite responses were considered as incorrect choices. Bees that completed both the learning assays were kept in an incubator at 27°C for 30 min before measuring their metabolic rate.

Active whole-animal metabolic rate (MR) of a bee was measured using flow through respirometry. Ambient air, scrubbed of H₂O and CO₂, was passed at a constant rate of 750 ml min⁻¹ through a 250 ml sealed glass chamber containing a single bee and CO₂ in the excurrent airstream was measured for 10 min with a FoxBox gas analyzer (Sable Systems), lightly agitating the chamber to stimulate flight as necessary. Bees that did show flight activity were discarded. Each bee was weighed immediately afterward and MR was calculated as the weight corrected mean CO₂ production (in ml h⁻¹ g⁻¹) for the 180 s with the lowest variance in CO₂ production. This was transformed into a weight-corrected power output (in mW g⁻¹) by multiplying it with 21.4 J ml⁻¹ CO₂ and dividing by 3600 J h⁻¹ W⁻¹ (Mugel and Naug, 2020).

Following the measurement of whole-animal metabolic rate, bees were immediately anaesthetized on ice and their heads were removed. Fresh brains were carefully removed under a dissecting microscope in chilled MiR05 respiration buffer (containing 0.5 mmol l⁻¹ EGTA, 3 mmol l⁻¹ MgCl₂, 60 mmol l⁻¹ K-lactobionate, 20 mmol l⁻¹ taurine, 10 mmol l⁻¹ KH₂PO₄, 20 mmol l⁻¹ HEPES, 110 mmol l⁻¹ sucrose and 1 mmol l⁻¹ g l⁻¹ fatty-acid free BSA). Brains were then transferred to pre-weighed centrifuge tubes containing 2 ml chilled MiR05 buffer and weighed on a microbalance with a resolution of 0.01 mg (Mettler Toledo). Four to six brains were measured in any one round of the assay, which ensured that their oxygen consumption could be quantified simultaneously within 3–4 h of dissection, which is well within the timeframe in which fresh brain preparations remain viable (Williamson and Hiesinger, 2010; Neville et al., 2018).

Oxygen consumption rate in the brain was quantified ex vivo using an Oxygraph-2k high-resolution respirometer (Oroboros Instruments GmbH, Innsbruck, Austria). Before adding brains, chambers were rinsed with 70% ethanol (3×) and milliQ H₂O (6×), then filled with MiR05 respiration buffer and air-calibrated to a starting oxygen content of 160 μM in a circulated chamber (with a 750 r.p.m. stir bar) maintained at 37°C. Following measurement of basal oxygen consumption rate of intact brains, digitonin (10 µg ml⁻¹) was added to permeabilize brain cells while leaving mitochondrial membranes intact for assessments of substrate-specific respiratory capacities. Low flux LEAK respiration state was measured first by adding a combination of substrates (0.5 mmol l⁻¹ malate, 5 mmol l⁻¹ pyruvate and 10 mmol l⁻¹ glutamate) in the absence of ADP, which reflect NADH-linked electron flow facilitated by non-specific proton leak across the inner mitochondrial membrane. The OXPHOS-linked respiration rate was measured next by the addition of 2.5 mmol l⁻¹ ADP to the same chamber, which enables a higher rate of NADH-linked electron flow by dissipating the inner membrane proton gradient through the ATP synthase. Finally, a maximal OXPHOS-linked respiration rate was measured by the addition of 20 mmol l⁻¹ succinate to provide additional electron supply through succinate dehydrogenase (CII), thereby fully reconstituting the supply of reducing equivalents from the tricarboxylic acid cycle to the electron transfer system. During these measurements, mass-corrected oxygen consumption rates were recorded continuously by monitoring changes in the negative time derivative of the chamber oxygen concentration signal following standardized instrumental and chemical background calibrations and recorded as the average of at least 2 min of stable (linear) oxygen consumption readings. Brain energetic capacity was defined in terms of mass-specific brain respiration, calculated from the maximum OXPHOS-linked respiration, a direct indicator of mitochondrial bioenergetics (Sauerbeck et al., 2011; Rittschof et al., 2018).

Slow and fast cognitive phenotypes were defined based on a learning index that was calculated for each bee by subtracting its discrimination learning score from its reversal learning score (reversal score–discrimination score). This resulted in values ranging from −1 to +1, where individuals with negative indices defined fast cognitive phenotypes (low reversal and high discrimination scores) and individuals with positive indices defined slow cognitive phenotypes (high reversal and low discrimination scores). The association among the different traits was assessed using the linear mixed model package lme4 in R (https://github.com/lme4/lme4/) with age and colony of origin used as random effects.

All data used in this research are available in Dataset 1.

Brain mass of bees was positively correlated with mass-specific whole-animal metabolic rate (χ²=5.52, N=37, P=0.01, Fig. 1A) and negatively correlated with mass-specific brain mitochondrial respiration (χ²=44.58, N=51, P<0.001, Fig. 1B). There was, however, no significant relationship between brain mitochondrial respiration and whole-animal metabolic rate (χ²=1.34, N=51, P=0.24, Table S1).

The learning index of a bee was negatively correlated to brain mass (χ²=4.94, N=51, P=0.02, Fig. 2A) but positively correlated to brain mitochondrial respiration (χ²=4.43, P=0.01, N=51, Fig. 2B), which means that bees with fast cognitive phenotypes had larger brains with a lower respiratory rate compared with slow bees. Learning index, however, was not significantly correlated to whole-animal metabolic rate (χ²=0.14, N=51, P=0.7). There was no effect of age or colony of origin on any of the relationships (Table S1).

This is the first time, to the best of our knowledge, slow–fast differences in cognition have been shown to be associated with differences in brain respiration rate, thereby establishing a possible common energetic link that ties the cognitive axis with other parts of the pace-of-life axis. Fast bees were those with larger brains, which were also associated with an overall higher whole-animal metabolic rate compared with slow bees. The observed positive relationship between brain mass and whole-animal metabolic rate is consistent with what has been observed across a wide variety of taxa (Isler and van Schaik, 2006) and can be attributed to the high energetic expense of brain tissue (Mink et al., 1981; Ames, 2000). While the larger brains seem to impose an overall higher metabolic expenditure in the fast bees, it is important to note that the larger brains were also associated with lower mass-specific energy consumption. It has been shown that there is a hypometric relationship in terms of how energy consumption of the brain scales with brain mass (Karbowski, 2007), which along with other mechanisms such as reduced size of other metabolically expensive tissues (Aiello and Wheeler, 1995; Goncerzewicz, et al., 2022) can lower the relative cost of larger brains.

The allometric relationship of larger brains having relatively lower rates of energy usage is generally attributed to decreases in the density of neurons or their firing rate (Karbowski, 2007, 2009). This suggests that differences in brain size could be associated with possible trade-offs in cognitive performance. A speed–accuracy trade-off is an intrinsic part of cognitive performance and fast cognitive phenotypes are defined as those which prioritize speed over accuracy, while slow cognitive phenotypes are those which show the opposite pattern (Sih and Del Giudice, 2012). There is ample evidence of such alternative cognitive strategies in honeybees (Chittka et al., 2003; Burns and Dyer, 2008; Tait and Naug, 2020; 2022) and the results from this study suggest that fast bees are those with larger brains with a high overall but low mass-specific energy consumption. The only other study which we found to have tested for a similar relationship observed a somewhat opposite pattern in guppies such that there was a positive correlation between telencephalon size and the time to make decision, although its relationship with accuracy was less clear (Burns and Rodd, 2008). This would mean that fishes with larger brains are likely to have slow cognitive phenotypes compared with those with smaller brains. However, this study did not measure energy usage in the brain, making it difficult to directly compare their results with what was found here. It also suggests that brain size alone may not be enough to explain how it influences cognitive functioning. Although brain size has often been used as a measure of cognitive capacity (Deaner et al., 2007; Benson-Amram et al., 2016; Collado et al., 2021), it has also repeatedly been shown that the relationship between the two is more complex (Schoenemann et al., 2000; Chittka and Niven, 2009; Logan et al., 2018; Triki et al., 2021; Hooper et al., 2022) and energetic considerations are an important part of it (Herculano-Houzel, 2011; Heldstab et al., 2022).

Energy used in the brain is largely required for the propagation of action potentials and for restoring postsynaptic ion fluxes, and the energetic cost of information processing at the neuronal level is associated with noise and the speed of response (Laughlin et al., 1998; Laughlin, 2001; Niven et al., 2007). During signal transmission, speed is related to the bandwidth over which a signal is transmitted while accuracy is related to the signal-to-noise ratio. Noise can be reduced by averaging the outputs from multiple neurons, although it comes with an increased energetic cost. Therefore, the number of neurons that can be activated will trade off with the average discharge rate of each neuron. Signal-to-noise ratios can also be improved by a higher level of inhibition at the neuronal level that results in slower neuronal integration of the downstream process. The fixed and signaling cost of a neuron both increase supra-linearly with its ability to transmit information and therefore efficiency declines with increasing capacity. It has been shown that while energy usage is positively correlated to the transmission rate of neuronal signals, cells that fire at a higher rate also carry less information (Ames, 2000; Koch et al., 2006). The cost of each spike puts a constraint on how many neurons can be concurrently active, thus implying an upper limit to aggregate neural activity and task accuracy (Levy and Baxter, 1996).

Models of speed–accuracy trade-off in decision-making are largely accumulator models that are based on two parameters: information accumulation and a threshold that defines the amount of information at which a decision is made (Gold and Shadlen, 2007; Bogacz et al., 2010; Heitz and Schall, 2012; Standage et al., 2014). This broadly implies that adjustments in either the rate of information accumulation, or the value of the decision threshold, or both, are at the basis of speed accuracy trade-off. The rate of information accumulation can be directly tied to the firing rate of individual neurons and/or the total number of neurons involved in the process, which in turn will be correlated to the overall energy expenditure (Penconek, 2022). A higher firing rate would allow a given value of decision threshold to be reached sooner, which might suggest that a higher level of energy metabolism in the brain is associated with fast cognitive phenotypes. However, an alternative possibility is that brains with lower rates of energy metabolism are associated with lower decision thresholds, which means that they can reach decisions sooner, consistent with what we report here. There is some evidence that information processing speed is lower during energetic compromise, and hungry neurons cannot maintain the firing speed necessary for optimal computations (Lord et al., 2013; Killeen et al., 2016). While faster decisions can be made by either increasing the rate of information accumulation or lowering the decision threshold, a higher accuracy is more constrained by a high value of the decision threshold and therefore by the quality and efficiency of information transfer in which energetic considerations play a strong role (Schreiber et al., 2002; Lennie, 2003). Our results suggest a higher energetic capacity in the brain might be needed to achieve these higher thresholds that translate to the higher accuracy defining the slow cognitive phenotypes.

Our measurement of brain respiration specifically reflects oxidative phosphorylation and mitochondrial activity, which are particularly important in meeting neuronal energetic demands (Kasischke et al., 2004; Ly and Verstreken, 2006; Hall et al., 2012; Picard and McEwen, 2014). It has been proposed that mitochondrial respiratory capacity could play a key role in the phenotypic variation related to pace of life (Jimenez et al., 2014), but the empirical evidence for this is somewhat mixed, with slow–fast differences being correlated to mitochondrial activity in the liver, but not in the brain or the heart (Chung et al., 2018). This suggests that the connection between mitochondrial function and whole-organism metabolic rate, and thus pace of life, is likely to be complicated. The plasticity of mitochondrial function both in the short-term and across life stages adds a further level of complexity to this issue (Chan, 2006; Jendrach et al., 2008). In addition, one may need to account for differences in the organizational structure of the neurocircuitry, which has a strong role in determining the efficiency of information transfer and energy usage in the brain (Chittka and Niven, 2009; Sengupta et al., 2013; Karbowski, 2019; Farnworth and Montgomery, 2024). Energy metabolism is a critical component of cognitive function and many cognitive impairments are associated with metabolic dysfunction (Sullivan et al., 2005; Lin and Beal, 2006; Frisardi et al., 2010), underscoring the importance of understanding how energetics and metabolism are associated with interindividual differences in cognition.

Slow and fast cognitive phenotypes represent differences in information acquisition strategy (Sih and Del Giudice, 2012; Tait and Naug, 2020, 2022). It has been shown that such differences could be tied to physiological differences (Moreira et al., 2004; Øverli et al., 2007) although our understanding of the underlying mechanisms that drive differences between cognitive types is still limited. The foundational idea of cognitive ability being tied to brain size and accompanying energetic demands (Dunbar and Shultz, 2007; Pérez-Barbería et al., 2007) has not been extended to understand the basis of slow–fast cognitive variation at the individual level. More recently, it has been shown how behavioral variation can be traced to energetic processes at the brain level (Chandrasekaran et al., 2015; Hollis et al., 2015; Kamhi et al., 2016; Rittschof et al., 2018), which provides support for the idea that differences in metabolic rate and energetic capacity may be the common underlying basis of variation across multiple phenotypic levels.