Studying emotion in invertebrates: what has been done, what can be measured and what they can provide

Invertebrate research has contributed immensely to our understanding of the brain. In terms of basic neuroscience, pioneering studies with invertebrates helped reveal the existence and structure of neurons (Ehrenberg, 1836; Nansen, 1886; Ramón y Cajal, 1909), and how information is transferred between them (Hodgkin and Huxley, 1939). Invertebrate research continued progress in the neurosciences with Kandel and colleagues' work showing the biochemical and neuroanatomical bases of learning and memory (Kandel, 2006).

Invertebrates are still actively used in many areas of neuroscience, particularly when it comes to the dissection of neural circuitry. This is because it has been possible to link the activity of single identified neurons to behaviour. Hammer (1993) demonstrated in the honeybee that the activation of a single identified neuron could replace sucrose during reward learning. Subsequently, a range of invertebrate studies have pinpointed neural substrates responsible for both reward and punishment, indicating a network of interacting neuromodulators that is organized and functions similarly to the reward system of vertebrates (Perry and Barron, 2013). Using sophisticated genetic tools in Drosophila, researchers have analysed distinct elements in food valuation and mapped out neurons involved in certain features of food reward (e.g. sweetness versus caloric content) at a level of detail that is, as of yet, impossible in vertebrate animals (Das et al., 2016). This type of mapping has been applied to many other aspects of behaviour (Iliadi, 2009), including sleep (Donlea et al., 2014), courtship (Dickson, 2008) and social foraging (de Bono et al., 2002).

Invertebrates are often not considered to have the neural requirements for such sophisticated abilities as emotion (see Glossary) and are assumed to accomplish all they do through sensorimotor responses alone (Allen-Hermanson, 2008). However, accumulating evidence suggests a re-evaluation of these views (Perry et al., 2013, 2017). Invertebrate research has now shown a variety of cognitive phenomena that were previously thought to be restricted to vertebrates and, at one point of course, to the domain of humans uniquely, such as concept learning (Giurfa et al., 2001), numerical cognition (Pahl et al., 2013), categorization of stimuli (Benard et al., 2006), cognitive/behavioural flexibility (Loukola et al., 2017; Mather, 2007) and cultural transmission (Alem et al., 2016).

Darwin (1872) may have first suggested that invertebrates express emotions, but only recently has anyone empirically explored this. Before discussing these works, we will first define what we mean by emotions and consider the importance of emotion research in non-human animals, including invertebrates.

Emotions are transient central states comprising subjective, cognitive, behavioural and physiological phenomena that are triggered by appraisal of certain types of environmental stimuli (Anderson and Adolphs, 2014; LeDoux, 2012; Mendl et al., 2010; Nettle and Bateson, 2012; Scherer, 2001). For example, fear is an internal state that includes increased attentional bias towards potential sources of danger and physiological preparation for fight or flight responses, which is triggered by the appraisal that there is something dangerous in the environment (Nettle and Bateson, 2012). Our conceptual understanding of an emotion is heavily based on its subjective component: i.e. the experience or ‘feeling’ (see Glossary) of the emotion usually in terms of pleasure (valance, see Glossary) and intensity (arousal, see Glossary). The subjective part of emotion has been the main focus of emotion theories for quite some time (Nettle and Bateson, 2012; Box 1). Internal states that humans associate with a ‘feeling’ (e.g. joy, anger, surprise) are what we most strongly consider to be emotions. But these verbally reported subjective feelings are accompanied by cognitive (e.g. perceptual biases), behavioural (e.g. escape behaviour) and physiological (e.g. heart rate) changes (Box 1). By adapting non-verbal techniques used in human psychology, researchers have been able to examine the cognitive, behavioural and physiological components of emotions in a variety of animal species (Boissy et al., 2007; Désiré et al., 2002). We know of no direct evidence of subjective feelings in any animals, but we do not exclude the possibility. We, like others, assume that many animals other than humans, including invertebrates, experience some basic form of subjective experience (Barron and Klein, 2016; Mendl et al., 2010; Nettle and Bateson, 2012; Panksepp, 2005, 2011). What this entails and the ways in which it is like our own subjective experience are currently beyond our scientific capability to investigate. Experimental directions towards answering these questions will include and benefit from, in our opinion, the study of emotion in invertebrates. There is ample evidence now that invertebrates have some form of emotion and therefore future efforts studying all facets of emotion in invertebrates will help provide a more complete picture of how emotion differs across phyla, how emotion has evolved and the neurobiological underpinnings of emotion.

Whether and how invertebrates, or any animal other than humans, experience emotions is of immense societal and scientific concern. From a societal perspective, understanding whether animals have a subjective component of emotion at any level, will be crucial to and will undoubtedly guide how we interact with them (Mendl and Paul, 2004; Dawkins, 2008, 2015). Again, until we develop methods to directly assess the subjective component, we can learn a great deal about emotions from studying the cognitive, behavioural and physiological components (Dawkins, 2015; Mendl et al., 2010; Nettle and Bateson, 2012; Panksepp, 2011). Concurrently, studying elements of animal welfare, health and survival, which are important independently of whether animals are conscious or not, will garner vital information for how we interact with them and will ultimately affect animals' well-being (Dawkins, 2015). From a purely scientific perspective, it will be necessary to determine whether certain animals have emotions in order to understand how our own emotions have evolved and, even more importantly, this will allow neurobiological approaches to elucidate how the brain produces emotions (Panksepp, 1998, 2005, 2011). In addition, the information obtained from a neurobiological study of emotions in animals will be able to help develop potential treatments for emotion-related cognitive dysfunctions that affect all brains. The subcortical regions of the human brain that have been strongly implicated in emotion are, at present and for the foreseeable near future, almost impossible, ethically, to study in any real detail. What non-human animals offer, especially those of the miniature invertebrate kind, are much more limited and accessible neural architectures – not to replace but to complement human and other vertebrates – to study in much greater detail and with more powerful techniques, largely unrestricted by ethical constraints and allowing for a simultaneous whole-brain system approach.

Various methods have been used to successfully study emotional expression in a variety of mammalian species, and only very recently have some of these methods been adapted to a small number of invertebrate species. Here, we attempt to comprehensively review the studies performed so far that explicitly assess emotions in invertebrates from a multi-componential perspective (Table 1). We have not covered any works examining pain (the unpleasant feeling of tissue damage) in invertebrates because others have already discussed these at great length (e.g. Cooper, 2011; Eisemann et al., 1984; Sherwin, 2001), but we do briefly describe the current debate within this field and its relation to animal welfare (Box 2).

We begin by considering cognitive approaches, before analysing behavioural approaches and finish by examining physiological methods, discussing the strength and weaknesses of each in turn. Finally, we consider how current and future research on invertebrate emotion might impact our understanding of the neural basis and evolution of emotions, in both invertebrates and vertebrates, and potentially provide opportunities for new avenues for emotion dysfunction research.

Emotions are triggered by appraisals of stimuli within our environment and help us evaluate the importance of immediate situations (Scherer, 2001; Paul et al., 2005; Anderson and Adolphs, 2014). When we experience a change in emotion, there are concomitant changes in how we view the world, i.e. our cognitive processes and our perception of our environment. These links between cognitive processes and emotional states in humans have been demonstrated in numerous tasks involving attention, perception, memory, expectations and risk assessment (Christianson, 1992; Mathews and MacLeod, 1994; Mathews, 1995; Nygren et al., 1996; Cahill and McGaugh, 1998; Mogg and Bradley, 1998; Loewenstein et al., 2001). However, exploration of cognitive components of emotion in non-human animals is only just beginning. Most of the methods used with humans rely on language-based tasks; however, others do not, meaning that some of these non-verbal tasks lend themselves to being adapted to studies in non-human animals. The first measurement of cognitive components of emotion in a non-human animal came as recently as 2004 (Fig. 1). Harding and colleagues (2004) used a paradigm known as the ‘judgement bias test’ to assess negative emotions in rats. The judgment bias test is based on the fact that humans who report negative emotions tend to make negative judgements about ambiguous stimuli, whereas people reporting positive emotions make positive judgements about the same ambiguous stimuli (Fig. 1A) (Eysenck et al., 1991; Wright and Bower, 1992; MacLeod and Byrne, 1996; Gotlib and Joormann, 2010). Rats in the Harding study (Harding et al., 2004) were trained to press a lever in response to a tone predicting a rewarding food pellet and to avoid pressing the same lever in response to a different tone predicting an aversive white noise burst (Fig. 1B). Once rats learned this discrimination, for several days they experienced either an unpredictable housing treatment designed to induce a negative/unpleasant state [e.g. at random times on any one day the cage might be tilted, the bedding wet or contain a strange conspecific (see Glossary)] or a predictably good housing treatment (control). Subsequently, the rats were exposed to ambiguous tones. As hypothesized, rats in unpredictable housing took longer to respond, and were less likely to respond to ambiguous tones, than rats in control conditions, showing behaviour indicating reduced anticipation of an appetitive/desirable outcome, similar to findings in depressed or anxious humans. The technique has since been applied to a number of different species, predominantly mammals, using different types of tasks (Go/NoGo, Active Choice, Spontaneous Behaviour), stimuli (visual, olfactory, auditory and position), and a variety of positive and negative manipulations (Mendl et al., 2009; Baciadonna and McElligott, 2015; Roelofs et al., 2016). To date, judgement bias paradigms have been employed in three studies in invertebrates. Bateson and colleagues (2011) and Schlüns and colleagues (2017) assessed negative bias in honeybees, whereas Perry and colleagues (2016) assessed positive bias in bumblebees.

Bateson et al. (2011) and Schlüns et al. (2017) used an olfactory learning protocol to investigate the presence of negative judgement bias (Mendl et al., 2011). Bees will naturally extend their proboscis (see Glossary) when sugar solution is touched to their antennae. Honeybees were conditioned to extend their proboscis to a mixture of two odours (1:9 1-hexanol and 2-octanone) by pairing one odour mixture with sugar water (‘reward’) and to withhold their proboscis to another odour mixture (9:1 1-hexanol and 2-octanone) paired with a bitter quinine solution (‘punishment’). After training, half of the bees were subjected to vigorous shaking on a lab table-top vortex machine for 60 s to simulate a predator (e.g. honey badger) attack on the colony. After the shaking manipulation, bees were tested with ambiguous odour mixtures intermediate to the two mixtures used for training (3:7, 1:1 and 7:3 1-hexanol and 2-octanone). In both studies, honeybees subjected to shaking were less likely to respond to the ambiguous odour mixture closest in ratio to the odour mixture associated with quinine during training, suggesting that shaking induces a negative cognitive bias (see Glossary) to ambiguous odour cues. However, it has been argued that shaking may cause bees to become better discriminators – shaking increased haemolymph (see Glossary) concentrations of octopamine, which can modulate sensory function (Adamo, 2016; Giurfa, 2013).

Perry and colleagues (2016) broadened the scope of invertebrate emotion research to consider positive/pleasant states (Plowright, 2017). Bumblebees were first trained on a free-flying task, where they found a cylinder positioned under either a green card at the left on the back wall or a blue card at the right on the back wall of the arena (Fig. 1C). Bees found a 30% sucrose solution reward under one of the colour-location combinations and water (unrewarding) under the other. Bees learned to approach the rewarding configuration faster and avoid entering the cylinder of the unrewarding configuration. After training, bees were tested with ambiguous configurations: intermediate colour (blue-green) and location. Prior to each test, to induce a positive emotion state, half of the trained bees received a small droplet of high-concentration sucrose solution. This unexpected reward amounted to less than 5% of the total crop load of a bee, similar perhaps to a small piece of chocolate for a human. Compared with controls, bees given the unexpected pre-test reward flew faster to intermediate cues, suggesting a positive judgement bias. Control experiments showed that after consumption of the small unexpected reward, bees did not increase their flight speed and were not more likely to explore novel stimuli, suggesting that the small reward did not simply increase the bees' general activity or exploration, but was indeed due to changes in their decision-making processes under ambiguity, thus resembling optimism in humans. Further work supporting this conclusion, in which the optimism response was manipulated pharmacologically through the dopamine pathway, is discussed later in this Review. Additional experiments, described below, showed that the positive effect of the pre-test reward generalized to an entirely different context.

The study of emotional reactions in human psychology and the field of human affective neuroscience (see Glossary) has relied predominantly on verbal reports and subjective rating scales (Russell, 2003; Paul et al., 2005; Oatley and Johnson-Laird, 2014). However, many emotional states are recognized in human infants, and some emotions are accepted in several primate species, and even rats, through facial expressions in response to stimuli (Berridge et al., 2009; Sotocina et al., 2011; Waller and Micheletta, 2013; Steiner et al., 2001; Berridge and Kringelbach, 2008). Invertebrates lack the facial musculature for any real type of comparisons to be made in this regard; however, a substantial amount of work in mammals has utilized other bodily expressions and motor behaviour in response to stimuli to assess both valence (pleasantness) and intensity (arousal) of emotions (Désiré et al., 2004; van der Harst and Spruijt, 2007; Reefmann et al., 2009; Boissy et al., 2011; Briefer et al., 2015).

Humans and other animals can learn to predict aversive events if a neutral conditioned stimulus is paired with an emotion-producing unconditioned stimulus, such as a loud noise or electric shock. The first study showing any form of emotion in an invertebrate came from the Kandel lab, where sea slugs (Aplysia californica) were trained in an aversive classical conditioning paradigm (Walters et al., 1981). Aplysia received shrimp extract for 90 s occurring 1 min before the onset of an electric shock to the head for 30 s (unpaired controls received the extract after the electric shock). Two days after training, animals were tested by delivering shrimp extract to their head, and their behaviour was recorded. The sea slugs that received paired training expressed significant facilitation of four different defensive responses: head withdrawal, siphon withdrawal, inking and escape locomotion. In addition, the feeding behaviour of the sea slugs was markedly decreased in the presence of the shrimp extract. The observed behavioural responses to conditioned stimuli resemble the actions of conditioned fear in mammals (LeDoux, 2012).

Fear conditioning relies on innate fear-inducing stimuli. Fear can result in the expression of a variety of adaptive and defensive innate behaviours, which are aimed at escaping or avoiding a threatening or potentially dangerous stimulus. Reflex reactions serve to protect vulnerable parts of the body (e.g. the eyelid blink to protect the eye) or to facilitate escape from a sudden oncoming stimulus. The startle response, a generally unconscious defensive reaction to threatening stimuli, can even be expressed as entire body movements, such as jumping or perhaps aggression or freezing in position, when the threat is inescapable.

Gibson and colleagues (2015) attempted to examine fear in fruit flies using an innately aversive stimulus: an overhead shadow (Fig. 2A). Drosophila were confined to an enclosed arena and repeatedly exposed to a rotating opaque paddle. Repetitive stimulus exposure resulted in increases in locomotor velocity, hopping and freezing, and dispersed the flies from a food source. To distinguish emotional behaviour from other forms of environmentally induced states, perhaps driven by learning, it is necessary to quantify a range of responses similar to those studied in humans. The behavioural responses of flies to stimuli were both graded (i.e. the more passes of the stimulus, the greater the behavioural responses) and persistent (i.e. the behaviour lasted longer than the presence of the stimulus). The repeated stimulus also dispersed flies from a food source, suggesting negative valence and context generalization. These behavioural results are consistent with the idea of an internal emotional state similar to what we consider to be fear in humans and other vertebrates.

Clear and present sources of threat/danger induce fear and result in escape behaviour. Anxiety, a related negative emotion, is experienced in response to imagined or potential, but unclear, threats. There have been a variety of model paradigms developed to examine anxiety – both normal and pathological – in humans and other vertebrates, mostly rats and mice, and even in zebrafish (Graeff et al., 1998; Liebsch et al., 1998; Belzung and Griebel, 2001; Egan et al., 2009; Stewart et al., 2012). Most of these tests measure how the innate behaviour of an animal changes when exposed to unfamiliar aversive places or threats, such as predators. One such classical and widely used method is the elevated plus maze, a simple paradigm where an animal, usually a rat or mouse, is placed in a four-armed apparatus in the form of a ‘plus sign’. Two of the arms have walls, whereas the other two arms are open. The ratio of the time spent on the open arms to the time spent on the closed arms is taken to indicate the level of anxiety expressed by the animal (i.e. the more time spent in the open arms, compared with the walled arms, the less anxious the animal is). The plus maze relies on the proclivity of the animal towards dark, enclosed spaces and an unconditioned fear of heights/open spaces. This method has recently been adapted and used with crayfish, where the design relied on the preference of crayfish for dark places (Fossat et al., 2014). Crayfish were placed in a dark/light plus maze, submerged in water, where two arms were shaded and two arms were exposed to light. Crayfish that were exposed to a series of electric shocks before testing in the maze, compared with control crayfish, hardly explored and rapidly abandoned the light arms. In addition, the number of entries, mean duration per visit, latency to first visit into a light arm and ratio of retreats from light arms compared with dark arms were significantly different in stressed crayfish. These behavioural results fulfil criteria normally designated for anxiety in mammals, including being innate, being unconditioned, occurring in the absence of a stressor, and expressed in a novel context. A follow-up study from the same lab also showed that, after a fight, the aggressive acts that a winning conspecific displayed towards the losing crayfish while both were in an aquarium (interpreted as harassment behaviour) induced anxious behaviour in the losing crayfish later in the dark/light plus maze (Bacqué-Cazenave et al., 2017), resembling the anxiety effects of psychological harassment in humans.

Repeating, uncontrollable stressful events can induce learned helplessness, where an animal exposed to an uncontrollable stressful situation becomes unable or unwilling to avoid subsequent encounters with stressful stimuli, considered an animal model of depression in humans (Eisenstein and Carlson, 1997; Willner, 1986). Yang et al. (2013) examined this phenomenon in Drosophila. Two flies were placed in separate dark walking chambers and exposed to the same sequence of heat pulses (Fig. 2B). Heat was applied to both when one, the ‘in-control’ fly, stopped moving. When the in-control fly started moving again, the heat ceased. Flies with no control over the heat pulses quit trying to escape as normal, began to walk slower and took more frequent rests, compared with ‘in-control’ flies, behaviours said to indicate a depressed state in flies. These results have also been replicated with electric shocks (Batsching et al., 2016). To support these interpretations, however, it would be valuable to know whether the flies without control were attempting other behaviours in the chamber, such as pushing or jumping, albeit unsuccessfully. Moreover, does the low activity of flies and their reluctance to escape transfer to new similar situations? In addition, how this state affects the flies physiologically (health and longevity), socially and neurobiologically [e.g. serotonin is known to play a crucial role in learned helplessness (Willner, 1986)] would be helpful in interpreting these findings.

Most studies on emotion involving vertebrates, including humans, have traditionally focused on negative emotions. It is argued that the reasons that positive emotions have been neglected in research are because they are few in number – reflected even in the imbalance of English-language words for negative over positive emotions – and are harder to differentiate (Fredrickson, 1998; Fredrickson and Branigan, 2011). The asymmetry might also stem from our understanding that natural selection has shaped emotions more for survival than for prosperity: there are many more ‘threats’ than ‘treats’ in our environment (Fredrickson, 1998). Furthermore, most of the psychological and clinical work with emotions in humans has been focused on solving problems, and therefore dealing with those emotions that pose more problems within our daily lives.

Although only a very limited number of studies have examined emotions in invertebrates, of those that have used a behavioural approach, two have addressed the possibility of positive emotions. Cassill and colleagues (2016) report a behaviour in fire ants (Solenopsis invicta) that, they argue, is similar to bodily expressions indicating pleasure in humans and other animals (Fig. 2C). Experimenters watched many hours of video-recordings from inside a fire ant nest observing ants' behaviours during various situations. They found that ants performed what has been called ‘wagging’, where they position and move their abdomen up and down at around 45°, when they interacted with brood and consume sugar water. This behaviour was not a defensive posture as it was within the hive, their stinger was never extruded and no venom was ever observed on the tip of their abdomen, nor found to be dispersed during wagging. As no sounds were produced during wagging, natural nest conditions are completely dark, and nestmates did not react negatively or positively to wagging, it does not seem to be a form of communication. Interestingly, wagging occurred significantly more during two specific behaviours: tending to brood and consuming sugar water. Cassill and colleagues (2016) suggest that this in-nest behaviour might be analogous to facial expressions and bodily postures of ‘hedonic pleasure’ in humans and other mammals during pleasurable events (Briefer et al., 2015; Proctor and Carder, 2014; Quaranta et al., 2007). However, to substantiate these claims, more experiments would be needed. For example, is this a simple mechanical response to manipulating any item with their mouthparts? Whether interactions with brood or consumption of sucrose cause cognitive biases or changes in neurotransmitter levels indicative of a positive state would also help support a claim for pleasure. At the moment, this work exemplifies how observations of natural behaviour can potentially provide valuable information about invertebrate emotions.

Having already shown that pre-test sucrose resulted in a positive cognitive bias (described above), Perry and colleagues (2016) tested the idea that emotions generalize across contexts (Anderson and Adolphs, 2014). They adapted a protocol used with infants to bumblebees. Infants receiving sugar water before a heel lance (a procedure to collect capillary blood for laboratory tests) tend to cry for a shorter duration and less loudly (Fernandez et al., 2003). In nature, bees are sometimes ambushed at flowers by ‘sit-and-wait’ predators, such as crab spiders, but often escape (Fig. 2D). This predator attack was mimicked using a trapping mechanism in which a constant pressure was applied for 3 s before the bee was released. Half the bees received a small droplet of high-concentration sucrose prior to the attack. Bees that received the droplet of sucrose before the test took much less time to commence foraging, indicating the small pre-test reward was causing a positive emotional state change in the bee across behavioural contexts.

Voices are an important modality for emotional expression (Schirmer and Adolphs, 2017). The voice is produced when air flow from the lung passes the larynx, where air is converted into sound by the vibration of the vocal fold. Before being expelled into the environment, the voice is filtered by the pharynx, oral and nasal cavities (Fant, 1960; Titze, 1994). The structure of vocalisation in mammals depends on the physiology and anatomy of the respiratory, phonatory and filter systems (Fant, 1960; Titze, 1994; Juslin and Scherer, 2005), each of which can be altered by emotional states (Scherer, 2003). Emotions can affect the tension and action of the muscles responsible for sound production, via the somatic and autonomic nervous systems (Scherer, 1986). Vocalisations therefore can, and have been, used as non-invasive markers of animal emotions (Scherer, 2003; Briefer, 2012; Panksepp and Burgdorf, 2000). In many of these efforts, sound features have been quantified using a variety of acoustic parameters clearly linked with the intensity/arousal and valence of emotional experience (Briefer, 2012). Many invertebrates, especially insects, have developed specific features that allow acoustic communication (e.g. for attracting mates or luring prey, defence against predators and territory defence) (Leonhardt et al., 2016). Darwin (1872) was the first to suggest that insects might potentially communicate emotions such as ‘anger’, ‘terror’, ‘jealousy’ and ‘love’ through their stridulation (see Glossary). However, again, this possibility has yet to be tested in invertebrates. Similar to vocalization in mammals, the acoustic signals used for communication in invertebrates could provide useful avenues for exploring the adaptive functions of emotions: e.g. does stress (or enrichment) induce changes in acoustic signalling in ways that might affect social dynamics, health or survival?

Physiological measures of emotion in humans have mostly relied on indices of activation within the sympathetic and parasympathetic nervous systems. Along these lines, studies adapted from what is known in humans to non-human animals have relied on similar methods (Appelhans and Luecken, 2006; von Borell et al., 2007; Kreibig, 2010; Lench et al., 2011). Some of the most common measures include skin conductance levels, skin temperature, heart rate, heart rate variability, blood pressure, neuroendocrine activity, EEG and neuroimaging (Paul et al., 2005). Studies have not been able to reliably relate distinct physiological measures to discrete emotions in humans. Situations that give rise to different emotions can result in similar physiological responses: e.g. increased heart rate due to anxiety and joy (Paul et al., 2005). Instead, most of these measures have pointed to relationships between changes in physiological responses and the dimensional properties of emotion, i.e. valence (positive/negative) and arousal (intensity). However, combinations of physiological measures in humans have provided some support for differentiating basic emotional states (Cacioppo et al., 2000; Rainville et al., 2006). Rainville and colleagues (2006) proposed a heuristic tree to differentiate basic emotions based on the cardiac and respiratory pattern. For example, ‘anger’ is characterized by an increase in heart rate and no visible change in high frequencies of heart-rate variability, whereas ‘fear’ is characterized by an increase of the heart rate and a noticeable decrease of high frequencies of heart-rate variability concomitant with a change in the respiratory activity. Most of these types of measurements are quite difficult to apply to invertebrates, given their often-miniature size and hard carapace and, in the case of insects, an open circulatory system, where heart rate is not increased. It has been shown in some crustaceans and molluscs, however, that sudden changes in the surrounding environment can induce modifications of some physiological variables: e.g. heart and ventilator rate (Schapker et al., 2002; Belzung and Philippot, 2007). Similar to the experiments by Walters et al. (1981), described above, Kita and colleagues (Kita et al., 2011) explored fear conditioning in pond snails. Snails learned to associate the presence of sucrose, which normally elicits feeding behaviour, with an aversive stimulus, KCl, that caused a full body withdrawal. Subsequent exposure to sucrose failed to induce feeding behaviour, showing learning, and caused the hearts of the snails to skip a beat, suggesting physiological responses similar to fear in mammals (Randall and Hasson, 1981).

The similarity of neurochemicals within invertebrate and vertebrate brains has provided another valuable physiological element with which to assess emotions across distant species (Perry and Barron, 2013). In humans, the role of biogenic amines (neurotransmitters) in the regulation of emotion is illustrated in the fact that many of the drugs that affect emotional states target and change the biogenic amine systems (Handley and McBlane, 1993; Dailly et al., 2004; López-Muñoz et al., 2011). A plethora of studies suggest that the three major biogenic amines – serotonin, dopamine and noradrenaline – are essential in the control of emotions (for excellent reviews/syntheses of much of this work, please see Fellous, 1999; Lövheim, 2012). Although cells that produce these neurotransmitters are located within the midbrain, basal ganglia and brainstem regions of vertebrates (raphe nuclei for serotonin, ventral tegmental area and substantia nigra for dopamine; locus ceruleus for noradrenaline), their connections branch to practically all regions of the brain (Lövheim, 2012). Growing evidence suggests that the biogenic amine systems might work concurrently as a final pathway for the simultaneous delivery of emotion-eliciting information to dispersed areas of the brain. The relationship between each amine and any specific emotion appears complex, but a recent model has suggested that the systemic levels of these three amines could potentially predict all of the basic emotions (Lövheim, 2012). Studies of drugs of abuse in humans and animals show that animals respond in similar ways to these drugs and that these observed behaviours rely on similar brain systems, many within the biogenic amine pathways (Panksepp, 2005).

Invertebrate nervous systems contain corresponding biogenic amines that function similarly to neurotransmitters, neuromodulators and hormones. Of the studies assessing emotion in invertebrates, four have applied physiological approaches, and all of these have relied on measuring or manipulating systemic biogenic amine levels. Bateson and colleagues (2011) assessed how systemic biogenic amine levels changed in response to a presumed negative-emotional event. Haemolymph was collected from honeybees after simulating a predator attack (shaking bees on a vortex for 60 s). Analysis of the haemolymph using high-performance liquid chromatography (HPLC) showed that systemic levels of the biogenic amines dopamine, octopamine (chemically similar to noradrenaline) and serotonin all decreased in response to bees being shaken vigorously. In humans, it seems that a depletion of biogenic amines (serotonin, noradrenaline, and dopamine) is responsible for features of depression (monamine hypothesis of depression) (Anderson and McAllister-Williams, 2015). Increasing the levels of these biogenic amines, either through serotonin or noradrenaline reuptake inhibitors, or noradrenaline- or dopamine-enhancing drugs, has anti-depressant effects (Anderson and McAllister-Williams, 2015). Furthermore, and perhaps similar to the observed cognitive bias observed in honeybees, deficient serotonin activity in humans leads to a bias towards negative over positive stimuli, so that negative stimuli have a greater impact on behaviour and cognition (Murphy et al., 2002). The results of Bateson et al. (2011) suggest that the general effects of biogenic amine levels influence emotional states in invertebrates in a manner similar to that found within humans.

Fossat and colleagues (2014) manipulated the systemic levels of serotonin to determine how changes in serotonin would affect anxiety behaviour in crayfish (Fig. 3A). Injection of serotonin into the haemolymph of unstressed crayfish caused avoidance behaviour similar to that exhibited by crayfish stressed with an electric shock. Injection of the anxiety-reducing (anxiolytic) drug chlordiazepoxide prevented anxious behaviours (avoidance of light areas) induced by electric shock or serotonin injection. Similar methods were used in a follow-up study by Bacqué-Cazenave and colleagues (2017), who showed that anxious behaviour induced by social harassment by crayfish conspecifics could be abolished with injection of the anxiolytic drug chlordiazepoxide or the serotonin-receptor antagonist methysergide (Fig. 3A). A separate group also showed that acute exposure to the serotonin reuptake inhibitor fluoxetine (crabs were immersed in a container of seawater containing the drug) reduces anxious behaviour of crabs in a simple light/dark paradigm (Hamilton et al., 2016). Mohammad and colleagues (2016) used interventions of heat-shock stress and a benzodiazepine, diazepam, to verify that the behaviour of flies of staying close to a wall during spontaneous locomotion both relates to anxiety and is dependent on serotonin signalling, comparable to what has been found in a rodent model of anxiety (Fig. 3B). In humans, serotonin has long been implicated in a variety of emotional processes. The specific relationship between serotonin levels and anxiety, as well as with depression, is somewhat complex; however, in general, it seems that fluctuations in serotonin affect diverse neural systems in the control of negative emotions and aversive processing, likely through different receptors, and potentially in combination with the dopamine system (Cools et al., 2008). The results of the above-mentioned invertebrate studies again suggest that the serotonin system might have been conserved, or at least co-opted, through evolution to help regulate emotions such as anxiety in a variety of animal species.

Perry and colleagues (2016) examined the potential role of serotonin, octopamine and dopamine in the optimistic behaviours observed in bumblebees, using topical applications of receptor antagonists for each of these amines (Fig. 3C,D). They showed that dopamine played a role in the positive emotional state of bumblebees. The optimistic behaviour seen in the judgment bias test in response to a pre-test sugar reward was abolished when bees were topically treated with the dopamine-receptor antagonist fluphenazine. Similarly, topical treatment with fluphenazine eliminated the effect that pre-test sugar had on the response of bees to a simulated predator attack. In humans, it seems that the expression and regulation of positive emotions is dependent, at least partially, on the dopamine system. Similar cognitive effects have also been seen in humans, e.g. increasing dopamine levels in humans reduces negative expectations regarding future events in response to negative information (Sharot et al., 2012). Indeed, psychostimulants that increase dopamine levels cause euphoria and have a positive effect in humans, and are rewarding in other mammals (Burgdorf and Panksepp, 2006).

Hughes and colleagues (2016) tested whether propranolol (β-adrenergic receptor blocker) could block stress-related memories in the pond snail. Snails were exposed to various types and combinations of stressors. They found that memories formed during more stressful combinations, e.g. a mimicked encounter with a predator (crayfish effluent+potassium chloride, which induces defensive full-body withdrawal) could be abolished using propranolol after reactivation, but memories formed during less stressful events, e.g. presence of a distant predator (crayfish effluent alone), were not affected. These results suggest phenotypically similar memories were molecularly different, and may help explain the mixed results in human research on disrupting reconsolidation with propranolol in post-traumatic stress disorder. Future work would benefit from addressing these molecular differences. Consistent with similar results in rats, this work also suggests that modulation of memory by emotional events are highly conserved. All of these works indicate that, in invertebrates, similar to what is known in vertebrates, the biogenic amines play a central role in emotions.

The link between certain biogenic amines and emotions in humans and other animals, and recently including invertebrates, is strong. Why would we find biogenic amine involvement in such similar biological phenomena? We feel that it is unlikely that this indicates a true homology of emotion systems across phyla (Barron et al., 2010). We speculate that it is more likely that ancient roles for biogenic amines as neuromodulators of more basic functions, such as in motor response circuits, were co-opted (perhaps even combining some of these more basic functions) for emotions through evolutionary pressure.

Many regions of the mammalian brain have been implicated in the regulation of emotion. However, it is important to note that emotional states have typically been much easier to illicit in humans through stimulation of subcortical circuits, especially those within the midbrain area and basal ganglia, which have also been found to mediate emotional behaviours in other vertebrates (Burgdorf and Panksepp, 2006; Panksepp, 2011). Strausfeld and Hirth (2013) have highlighted developmental, anatomical and genetic evidence arguing that the central complex of the insect brain is analogous to the vertebrate basal ganglia. In addition, Barron and Klein (2016) have made a strong argument that the insect brain is functionally analogous to that of the vertebrate midbrain and potentially capable of subjective experience, defined by the authors as a very basic level of conscious awareness. It does seem evident that the insect brain contains the neural structures able to support emotional states. Given these neurochemical and functional similarities, combined with what has already been shown in crayfish (Fossat et al., 2014), fruit flies (Mohammad et al., 2016) and pond snails (Hughes et al., 2016), it seems promising that pharmacological and genetic manipulations of some invertebrates may be able to provide models of human diseases and/or disorders linked to emotions.

It is also important to realize that emotions are brain states with a wide and vast neural architecture involved in their expression and regulation, and that analogous neuroanatomical features with vertebrates may not be necessary to support similar functions, as there are multiple ways of producing the same output. In studying emotions in invertebrates, as in all animals, we must consider the evolution and ecology of each species, and realize that any of the components of emotion may be expressed differently. Furthermore, given the limited neural architecture available in many invertebrates, it may be that the invertebrate brain evolved a simpler way of producing and regulating emotions. Efforts to understand how exactly this occurs are necessary and will no doubt inform us on the evolution of emotions and shed light on the nature of our own emotions.

Traditionally, emotion research has been limited to large-brained animals, owing to an undeserved bias against small-brained invertebrates, with an assumption that possession of such a limited number of neurons might not lend itself to supporting the complex cognitive functions known to operate in vertebrates. This bias has likely hindered substantial progress in the field of cognitive science. However, a plethora of studies in the past few decades has unveiled a variety of impressive cognitive abilities in these miniature-brained organisms, and various types of neural network models have highlighted how seemingly complex cognitive tasks can sometimes be resolved by relatively limited circuits (e.g. Ardin et al., 2016; Peng et al., 2017). The handful of recent studies reviewed here have ignored this ‘size bias’ and have together provided corroborative evidence that several invertebrate species display emotions on cognitive, behavioural and physiological levels similar to those of vertebrates.

One of the biggest divisions of opinion in the field of emotion research is whether to call what is observed in invertebrates, or other non-human animals, ‘emotions’, as we cannot yet directly measure its subjective component. Because we are ignorant of its neurological basis, we cannot exclude the possibility that invertebrates have some basic form of subjective experience. Furthermore, it is not necessary to fully understand subjective experience in order to study emotions. Therefore, with the help of invertebrate research, studying all facets of emotion will progress us towards answering the most-intriguing and evasive questions about emotion. How did emotions evolve? What are the adaptive functions of negative and positive emotions? In what ways do cognition and emotion interact? How is emotion produced within the brain?

Demonstrating specific emotions in a particular species of animal alone will not provide much useful information at this point. Future research directions should emphasize the mechanisms of emotions, because understanding the mechanisms that produce emotions will ultimately help to answer the questions above and will help expand our efforts to tasks such as building artificial intelligence with emotions and developing therapeutics for emotion-related disorders. A powerful and efficient tool that will support these endeavours will be the use of computational modelling. Computational models can use the data from separate neurobiological and behavioural experiments to simulate and provide a functional account of specific cognitive phenomena (Abbott, 2008; Brodland, 2015). Of course, the smaller the number of neurons in a system, the easier it will be to model. Therefore, many invertebrates, with their limited neural architecture and impressive cognitive abilities, lend themselves well to computational modelling. Three areas of technological advancement will help tremendously in these collaborative efforts. Improved computational behavioural analysis allowing for automatic tracking and discovery of detailed behaviour (Egnor and Branson, 2016) will enable neural computational models to take advantage of massive behavioural data sets with extraordinary detail and help to determine the neural mechanisms of emotion-linked behaviour. Advances in the fabrication and application of electrophysiological recording of many neurons simultaneously in freely moving small animals will provide the temporal and spatial resolution necessary to determine the contributions of specific regions and circuits to emotional processing. Ultimately, to be able to verify models we must be able to establish causal relationships. Establishing causal techniques such as optogenetics in more invertebrate species with rich behavioural repertoires will be vital in determining the causal neural links responsible for emotion and propel the field forward. Miniaturizing and adapting psychological methods for the study of emotions in invertebrates, combining cognitive, behavioural and physiological approaches, and applying computational behavioural analyses and neural network modelling techniques primed for small brains will be necessary to fully understand the evolutionary origins of and the neural mechanisms behind emotions.

We thank HaDi MaBouDi and Andrew Barron for enlightening discussions, and both the editor and two anonymous reviewers for their helpful comments. During the writing process of this review, Jaak Panksepp passed away. We have all lost a dear colleague. We thank him for all of the magnificent work he did in the field of affective neuroscience towards understanding the neurobiological nature of emotion. His research and thoughts undoubtedly contributed to the inspiration of so many young scientists, including us.