How (not) to train your spider: successful and unsuccessful methods for studying learning

ABSTRACT

Spiders are capable of learning in many different contexts, including prey capture, social interactions and predator avoidance. In recent years, there has been an upturn in the number of published studies of learning in spiders, with a particular interest in active hunters that do not build prey-capture webs. However, as is often the case when developing protocols for studying learning in a new taxonomic group, many researchers have tested methods that failed to produce interpretable results. These methods largely remain unpublished, although knowing what has already been attempted would be of great benefit to the community of researchers that work on arthropod learning. Here, we briefly review published methods that have been successful in demonstrating learning in actively hunting spiders, as well as report on unpublished, unsuccessful methods that we collected from the research community.

KEYWORDS:

• Anyphaenidae
• experience
• learning
• Lycosidae
• Oxyopidae
• Pisauridae
• protocol
• Salticidae
• spiders
• Thomisidae

Introduction

Published studies of learning were initially biased toward a few model species, notably laboratory rats and pigeons. This bias arose partly from the legacy of early comparative psychology, which focused on careful, well-controlled laboratory studies that characterised learning processes using a handful of tractable species. However, the study of learning expanded well beyond those model species when the view that evolution has shaped learning differently in different taxa began to gain traction (see Shettleworth 2010 for discussion). Arthropods were initially viewed as unlikely candidates for the study of learning because of their small brains and presumed reliance on stereotypic behaviour, but work on insects proved otherwise (e.g. Papaj & Lewis 1993; Dukas 2008). In spiders, the pioneering work of Robert Jackson and his colleagues revealing surprisingly complex problem-solving capabilities in jumping spiders (e.g. Tarsitano & Jackson 1997; Wilcox & Jackson 1998) helped inspire arachnologists to focus on learning and other cognitive processes (see reviews in Cross & Jackson 2006; Jackson & Cross 2011; Jakob et al. 2011).

Developing workable protocols to study learning in a new taxon is often time-consuming and plagued with difficulties. To name just a few considerations, the animal must be well adapted to the testing environment, presented with a task appropriate to its sensory capabilities, motivated to perform the task and unlikely to perform related behaviours instead of the task (see Breland & Breland 1961 for examples of the last).

In this review, we have two goals. First, we will review briefly some of the more commonly used published methods of studying learning in spiders. Because of the recent reviews on spider learning and behavioural plasticity cited above, we will not revisit the larger goals of these studies, but rather will focus on their methods. In particular, we will focus on associative, appetitive and avoidance learning. Because of space limitations, we will omit other types of cognitive processes, such as priming, search-image formation, insight, path integration and detouring, all of which are covered elsewhere (Jakob et al. 2011). Second, we recognise that we lose much useful information about protocols that fail to produce interpretable results because these studies are never published. Null results in learning experiments can be very difficult to interpret. If the task presented to an animal is not appropriate due to its sensory capabilities, its stress under the testing conditions or a myriad of other factors, negative results do not provide the insight we seek about the animal's learning performance. Other protocols may never even get incorporated into a formal experiment, but instead are tried and then set aside. Information about these methods is often only shared between lab groups through informal conversations at conferences or email exchanges between friends. Here, we provide some of that unpublished information as a guide for others who are interested in studying learning in spiders.

Methods

To find published protocols, we conducted a Web of Science search using the phrase ‘spider* and (learn* or experience or cognition)’, and also included selected studies cited within those articles. We focused on visually hunting spiders rather than web-building spiders. The methods used to study learning in these two groups are, by necessity, substantially different. We then contacted authors of a subset of recent articles and asked for information about unpublished protocols that they had tried.

Results and discussion

We organise our findings by the type of task with which the spiders are presented, loosely modelled after our earlier, more comprehensive review (Jakob et al. 2011). Learning is not expected to evolve under all conditions. It has been argued persuasively that learning is expensive in terms of time, energy and neural resources (reviewed in Snell-Rood et al. 2011). Given that learning has costs, if the environment is entirely predictable, an animal should evolve an innate, fixed response to useful features of the environment. If the environment is entirely unpredictable, learning about its features would provide no better results than randomly guessing about them. If the environment has a moderate amount of predictability, then depending on the cost of learning, an animal might benefit by learning about relevant features of the environment (see review in Dunlap & Stephens 2009).

Contexts in which learning is likely to have value to spiders include: choosing, locating and handling prey; returning to shelter and other locations of value; avoiding predators; interacting, even fighting, with competitors; and choosing mates. Some experiments try to mimic the real-world conditions as nearly as possible in order to more accurately assess whether spiders are likely to demonstrate learning in their daily lives. In contrast, other studies use settings and stimuli that are less analogous to the real world, either because learning is used as a tool to get spiders to ‘tell’ us whether they can distinguish different perceptual stimuli or because the researchers wish to know more about the types of learning that are within a species’ capabilities.

Predation

Successful protocols

Many spiders are euryphagic—they consume a wide range of prey items in their natural diet—and have evolved a host of flexible strategies, including learning, for dealing with different prey types (e.g. Nelson & Jackson 2011). Predation is relatively straightforward to study as spiders are generally motivated to attack prey, even in a laboratory setting. One aspect of predation that has received attention is choice of prey. Preference for a particular type of prey has been tested with three experimental designs (summarised in Nelson & Jackson 2012). In simultaneous prey testing, a spider is offered different kinds of prey at the same time and must choose one. In alternate-day testing, a spider is offered only one prey type at a time. A spider expressing a preference will eat only one type but not the other, even when the less preferred prey is the only food available. Finally, in alternative prey testing, a spider feeding on one prey is offered a different prey. To express a preference, the spider must release the first prey and capture the second. Of these tests, the latter two require the spider to exhibit a higher strength of preference than the first.

The tests described above have been used to examine the innate preferences of spiders towards particular prey types (e.g. see review of work on the mosquito-eating jumping spiders Evarcha culicivora Wesolowska and Jackson, 2003 and Paracyrba wanlessi Zabka and Kovac, 1996 in Jackson & Cross 2015), but can also be employed to test how experience influences preference. For example, to test whether experience with prey influenced later prey choice, Punzo (2002a, 2002b) presented lynx spiders Oxyopes salticus Hentz, 1845 with one of three prey species daily for a week, at the rate of one prey item per day, then offered a simultaneous choice between all three prey types. Spiders were more likely to choose prey with which they were familiar. Wolf spiders Hogna carolinensis (Walckenaer, 1805) also chose familiar food (Punzo & Preshkar 2002). In contrast, regardless of previous diet, a mosquito-specialist salticid species (P. wanlessi) preferred mosquitoes (Jackson et al. 2014).

A slightly different approach is to give spiders experience with prey and then test if they then respond to cues from that prey. For example, wolf spiders Tigrosa helluo (Walckenaer, 1837) fed one prey species for a month preferred filter paper with odour cues from that species (Persons & Rypstra 2000). These techniques can also be adapted to test for generalisation of a learned aversion to novel prey (Toft 1997).

Training can also be accomplished in a shorter period of time. Over the course of a single day, jumping spiders Phidippus princeps (Peckham and Peckham, 1883) were presented with milkweed bugs (Oncopeltus fasciatus Dallas) rendered either palatable or unpalatable by their diet (Skow & Jakob 2005). Spiders interacted with a milkweed bug for 10 min in each of eight trials. Importantly, as a control to ensure that satiation was not an issue, spiders were offered a palatable cricket immediately after the final trial. After demonstrating that spiders learned to avoid distasteful bugs but did not avoid palatable bugs, the authors used this same procedure to investigate whether spiders attended to visual contextual cues during learning. In this design, spiders learned to avoid distasteful prey in the presence of a particular set of background cues, and then were tested a final time with either the same background cues or novel cues. If cues had changed, spiders attacked the distasteful prey as they did prior to training.

Similar methods, but with distasteful fireflies rather than milkweed bugs, were also employed by our lab. We positioned distasteful fireflies next to either flashing or unlit LEDs of the same wavelength as firefly light emissions, and found that the flashing lights enhanced the ability of spiders to learn to stop attacking the fireflies. Again, all training took place in 1 day, with seven 10 min trials separated by a 30-min rest (Long et al. 2012).

Artificially colouring prey offers a way to test the role of colour in prey choice. For example, colour biases in foraging by the jumping spider Habronnatus pyrrithrix (Chamberlin, 1924) were demonstrated by providing dyed water to cricket prey; small crickets are translucent and thus take on the colour of the dye (Taylor et al. 2014). The ability of H. pyrrithrix to generalise about colour to new prey was then tested. Spiders were first fed milkweed bugs, which are black and red. The bugs were rendered either palatable or unpalatable through their diets. One group of spiders was fed unpalatable milkweed bugs for 3 weeks and learned to avoid them, while another group was fed palatable bugs. After training, spiders were presented with a choice between control crickets and those that were artificially coloured red. Spiders experienced with unpalatable bugs avoided the artificially red crickets, while those experienced with palatable bugs chose them (Taylor et al. 2015). Interestingly, subtle aspects of experimental design changed this result. In a different version of the study, where field-caught spiders were used, spiders in the group given palatable bugs received no other prey, while those given unpalatable bugs were also given white-eyed flies, and the testing environment differed slightly from the training environment. Here, while spiders learned the task, they did not generalise (Taylor et al. 2015).

A potential hurdle in studying predatory behaviour is that the behaviour of prey can influence the outcome of a test. The visual system of many spider species is particularly attuned to motion (Zurek & Nelson 2012a, 2012b), and spiders attend to both local and global motion cues of prey (e.g. antennae and leg movement vs. the movement of the entire body; Bednarski et al. 2012). Prey that do not move—an anti-predator behaviour of many insects—may thus avoid detection and can bias results. Besides allowing for variation in prey behaviour when planning sample sizes, the problem can be mitigated by using a small arena so that an active spider will find even motionless prey, as in the studies described above. Alternatively, one might standardise prey movement by presenting videos of prey (e.g. Clark & Uetz 1990) or by using a dead prey item as a lure, which can then be moved in a standardised manner (e.g. Tarsitano & Jackson 1997). However, for learning trials, thought must be given to how to reward spiders for attacking these unrewarding representations of prey. Another approach is to use something besides live insects as a food reward. Liedtke & Schneider (2014) took advantage of the fact that many jumping spiders feed on nectar (Jackson et al. 2001b), and offered drops of coloured sugar water as a reward to Marpissa muscosa (Clerck, 1757). They created a negative stimulus by adding citric acid to the solution. Most spiders readily fed on the sugar drops on their first encounter. This elegant approach allowed the authors to control for initial preference for a particular colour, and also to test for reversal learning. In reversal learning, the animal first learns that a particular conditioned stimulus A, but not stimulus B, predicts an unconditioned stimulus. After the task is learned, the predictors are reversed and the animal must learn the new associations. This provides a test of the animal's ability to learn under changing conditions. With the spiders, both stimulus colour and location were used as predictors. Spiders easily learned the initial associations, and then quickly learned the reversals. Nectar use is not limited to salticids. For example, Hibana futilis (Banks, 1898) (Araneae: Anyphaenedae) can learn to associate odour with artificial nectaries (Patt & Pfannenstiel 2008).

Besides prey choice, another important aspect of foraging behaviour is the selection of a good foraging site. Foraging-site selection is most often studied in web-building spiders, but crab spiders also benefit by waiting for prey on high- rather than low-quality flowers. The role of experience in foraging-site selection by Misumena vatia (Clerck, 1757) has been examined with a number of different protocols, reminiscent of the variety of protocols used in the prey preference trials described above. For example, experienced spiders were given a choice between familiar and novel flower types (Buchsbaum & Morse 2012). In another design, spiders with experience with a particular flower type were placed on either similar or different flowers with the expectation that, if foraging behaviour were based on experience, they would stay on familiar flowers longer (e.g. Morse 2000). Together, a suite of studies suggests that experience becomes increasingly important for crab spiders as they age (Buchsbaum & Morse 2012).

Finally, spiders can improve in their ability to handle and subdue prey. A typical protocol is simply to present spiders with a sequence of prey and see whether their attack success or attack strategy improves over time (reviewed in Jakob et al. 2011). Spiders may also modify their attack strategies using trial-and-error during the course of an encounter with a potential prey. An excellent example comes from Jackson and colleagues, who studied the araneophagic jumping spider Portia fimbriata (Doleschall, 1859). Portia is an aggressive mimic, and moves stealthily onto the webs of other spiders where it broadcasts an array of vibratory signals that resemble those of insects struggling in the web. Over the course of an interaction, P. fimbriata narrows its signals to the ones that are successful in attracting its prey (Jackson & Wilcox 1993; Jackson 2002).

Unpublished protocols

It can be difficult for researchers to identify a suitable aversive prey species to use in experiments. For example, as described above, milkweed bugs are especially useful because they can be aversive or palatable depending on their diet. Milkweed bugs can be purchased in bulk from Carolina Biological Supply (Burlington, NC, USA), but these have been reared for many generations on sunflower seeds and must be transitioned back onto milkweed seeds in order to become aversive. Milkweed bugs can also be collected quite easily in the field in North America, but possible differences between the behaviour of lab- and field-reared species needs to be considered. There can also be natural variation in aversive properties of prey. Milkweed bugs vary in the content of defensive cardenolides (Isman et al. 1977), and our lab found that fireflies of the same species collected from one field were aversive to jumping spiders, but those collected a few miles away were not.

Some researchers have tried manipulating palatable prey to make them aversive. We have found that spiders were undeterred by prey dipped in quinine. Crickets fed on cayenne pepper, Bitrex, cinnamon or ginger until their guts were full were also preyed upon readily by jumping spiders (L Taylor, University of Florida, pers. comm. June 2015). The Jakob lab made artificial prey—gelatin balls that were then wrapped in Parafilm along with a small magnet and dragged from below the arena with another magnet. Spiders readily pursued and attacked the moving balls, and appeared to be repelled by quinine-laced balls, but we deemed the protocol too cumbersome to pursue.

Spatial information

Successful protocols

Spiders may benefit from learning spatial information in several contexts. Many species of actively hunting spiders build retreats that they return to in the evening or during inclement weather (e.g. Hoefler & Jakob 2006). Spiders may also benefit from knowing the location of a reliable food source, such as a blooming plant or an insect-attracting carcass.

Typical experiments on spatial information use in spiders involve testing whether spiders can learn to associate a positive or negative stimulus with a particular spatial cue. For example, Hoefler & Jakob (2006) tested whether jumping spiders (Phidippus clarus Keyserling, 1885) could associate beacons (landmarks close to a goal) with their nest sites. Spiders built webs in nest tubes in the field that were placed on coloured sticks. While still in the field, spiders were then tested with novel sticks of either the same or a different colour, and spiders were more likely to approach sticks of the same colour as their ‘home’ stick. While this experiment had the advantage of being conducted in the field under conditions very similar to what animals encounter naturally, a drawback was that spiders were only motivated to return to their nests at the end of the day, so data collection was restricted in time.

In a laboratory study, Peckmezian & Taylor (2015a) compared jumping spiders Servaea incana (Karsch, 1878) that had a beacon near their nest to those that did not have a beacon. Spiders experienced with a beacon spent more time near a test beacon both in the real world and in a simulated, 3D virtual world, demonstrating convincingly that the spiders learned the beacon's visual characteristics.

To our knowledge, evidence of spatial learning by spiders in complex mazes with multiple turns and dead ends has rarely been published, but such a maze was used to test lycosid (Hogna carolinensis) spiderlings. Punzo & Ludwig (2002) heated the maze with a bright light, and the combination of heat and the light motivated the spiders to move through the maze. Each spiderling received 10 trials daily for 10 days. Spiderlings reared with their mother and siblings had fewer blind alley errors in the maze compared with spiders reared in isolation.

Unpublished protocols

While the use of beacons by jumping spiders has now been shown in several studies, the use of landmarks that are further from the goal has, to our knowledge, not been shown. In a small pilot experiment, the jumping spider Portia fimbriata attended to beacons near the goal but did not use landmarks placed further from the goal (P Taylor, Macquarie University, pers. comm. June 2015). In our lab, we attempted a landmark study in the field. Similar to the Hoefler & Jakob (2006) beacon study, we set out nest tubes for Phidippus clarus to colonise. Their habitat is tall, grassy and flower-filled fields. So that spiders could get a clear view of the three landmarks we provided, we trimmed the vegetation to just above the ground, but unfortunately this species was reluctant to cross the open area and colonise the nest tubes. Landmarks and their configuration might be more profitably examined in species that live in open habitats where multiple potential landmarks are routinely in view.

Because spiders are primarily motivated to return to their nest tubes only at the end of the day, our lab also tried to give spiders a choice between two different small habitats in the lab, with the idea that we could test individuals repeatedly in the same day. One habitat consisted of a platform with various artificial plants and small objects that Phidippus spiders readily explore, and the other consisted of an unadorned platform. The spider was released at the bottom of a stick that then branched to these two habitats. Our plan was to investigate if spiders could learn various cues that indicated the direction to the complex habitat. Unfortunately, in pilot trials, spiders exhibited no preference, and in fact seemed quite content to rest on the stick indefinitely and never make a choice. In retrospect, the stick resembled a plant stem that they would normally rest on in nature.

When designing navigation studies, it is helpful to keep in mind potential behavioural differences between sexes and across the season. For example, Hoefler & Jakob (2006) found that female jumping spiders exhibited more site fidelity than did males, and that female site fidelity increased as the season progressed.

Avoiding predators

Successful protocols

Most studies of predator avoidance by spiders focus on invertebrate predators. Among studies of active hunters, wolf spiders make exceptionally good subjects because they are well known to attend to chemical cues. One study design is to expose spiders to a live predator and then test them for subsequent avoidance of the predator's chemical cues. For example, Punzo (1997) exposed Schizocosa avida (Walckenaer, 1837) spiders to a scorpion. Those that escaped predation by autotomising a leg were more likely to avoid chemical cues from a scorpion on a filter paper disc. Conversely, one can first expose spiders to cues from a predator and then test for enhanced survival in a live encounter, although whether a positive association is due to rapid associative learning or some other process, such as heightened attention and arousal, requires careful consideration (Eiben & Persons 2007).

Unpublished protocols

In a study of antipredator behaviour of Phidippus jumping spiders, our group attempted to mimic predation attempts by a vertebrate predator in the laboratory by using a mounted bird that moved towards the spider. Spider responses were unpredictable; often they ran towards or underneath the predator rather than away, so it was difficult to define meaningful outcome variables.

Social contexts

Successful protocols

Spiders learn through interactions with conspecifics. For example, males in many species compete for access to females. Prior experience—whether one has lost or won previous fights—can influence a contestant's fighting behaviour. The costs and benefits of different methods to manipulate experience are discussed by Kasumovic et al. (2010). Evidence for the effects of experience on fight outcomes has been found in several families of actively hunting spiders, including thomisids and salticids (Dodson & Schwaab 2001; Hoefler 2002; Kasumovic et al. 2010).

Juvenile experience can also influence adult mate choice, as demonstrated with a variety of protocols, primarily from the lycosid literature. A typical protocol is to expose juvenile females to a particular adult male phenotype for 30 min every other day during their last instar before maturation, and then test them after maturation with either a familiar or a novel male (Hebets 2003; Hebets & Vink 2007). This protocol has been used in a number of studies, including one on the effect of experience on the propensity for sexual cannibalism in pisaurids (Johnson 2005). Prior experience as an adult with prospective mates also influences mating propensity in subsequent encounters in wolf spiders (Wilder & Rypstra 2008). Some considerations in designing exposure experiments such as these include deciding whether to allow animals to interact directly during the exposure period, and thus receive both tactile and chemical information, or whether they should receive only visual information (Rutledge & Uetz 2014). Videos rather than live conspecifics can also provide an effective training experience: juveniles that saw more video males were more selective as adults than those presented with fewer males (Stoffer & Uetz 2015). This type of protocol has been extended to other social situations. For example, in a study of eavesdropping by male wolf spiders, males exposed to videos of courting males learned that courtship could indicate the presence of females (Clark et al. 2015).

Unpublished protocols

It is possible to use a similar protocol to those described above to investigate whether spiders learn about potential predators through observing conspecifics. M Persons (Susquehanna University, pers. comm. June 2015) provided an interesting unpublished dataset comparing wolf spiders that observed a predator capturing a conspecific with spiders that observed a conspecific evade predation. When confronted with a live predator, spiders that had observed the capture of a conspecific had a significantly shorter survival time, possibly because they became more active.

Experimental designs for asking general questions about learning

Successful protocols

The studies above generally were concerned with how spiders learn in their natural environments. There has also been considerable interest in developing robust laboratory protocols that can be employed for asking general questions about which learning tasks spiders can perform, for designing psychophysics experiments (such as using learned tasks to understand which sensory stimuli spiders can distinguish) and for comparative work across species. To that end, different labs have been exploring protocols using a variety of appetitive and aversive stimuli that can be consistently applied during the training period.

A positive stimulus that is often used in studies of vertebrate learning is food, but as noted above, food can present challenges in spider studies. While spiders are generally eager predators, it is important to ensure they do not become satiated. Generally, in vertebrate studies, animals are given only small bits of food in order to keep them hungry and thus maintain their motivation for the task. It is difficult to consistently provide spiders with tiny prey as a reward as spiders may not be successful at capturing it. Because spiders generally feed by sucking on prey with their chelicerae embedded in it, wrestling larger prey away from a spider before it reaches satiation can be dangerous to the spider.

Another potential barrier to using prey as a positive stimulus is that (as discussed in the section on Predation) prey behaviour can influence whether a spider gets a reward during a training session. Spiders generally prefer moving prey (e.g. Bednarski et al. 2012), but if the protocol requires that the reward be delivered in a particular place, the prey must be restrained in some way. For example, the jumping spider P. princeps Peckham and Peckham, 1883 learned to distinguish between beacons of two colours that indicated the location of prey (Jakob et al. 2007). Crickets were secured in place by gluing their bodies and legs to small pieces of card stock, and then were placed behind a red or blue block in a T-maze. Spiders were given an hour a day to explore the maze and discover the prey. Only colour predicted the location of prey; locations were randomised. There were four training trials followed by a test trial with no prey, then a second block of four trials. An easier method is that of Liedtke & Schneider (2014), who offered sweetened or aversive coloured drops to spiders (described above) in their test of reversal learning. While not all spiders found the drops during the tests, this method allows one to control the amount of reward and eliminates any confounds caused by the behaviour of the prey. This is a very promising avenue to pursue.

Because it can be difficult to design an experiment in which a spider consistently receives a positive reward, a number of researchers have turned to using aversive stimuli. An aversive stimulus adopted from the vertebrate learning literature is electric shock. Skow (2007) developed a shock apparatus for jumping spiders (P. princeps), which was later modified by Bednarski et al. (2012) in order to test the perceptual capabilities of another species (P. audax Hentz, 1845). The shock apparatus consisted of parallel aluminium strips glued to a wooden arena. Alternating strips were connected to the positive or negative terminals of a power source, so that spiders received a shock when they touched neighbouring strips simultaneously. Spiders watched a video while receiving a shock of 33 V, 8 mA for 1 s every 10 s, for 2 min. After a 5-min break, this procedure was repeated for a total of five training periods. Spiders were then given a choice between the training video and a novel video; if they chose the novel video, it was taken as evidence that they could distinguish between the two videos.

Peckmezian & Taylor (2015b) also used shock to develop a passive-avoidance paradigm for jumping spiders (Servaea incana). Their apparatus consisted of a printed circuit rather than a hand-made shock apparatus. They presented mobility assays to test the effect of different levels of shock, and found 30V to be optimal (their thorough assays present a useful model when expanding this protocol to new species). Two circuit boards, one electrified and one not, were placed at opposite ends of an arena. One side of the arena was in dark and the other in light. Spiders initially preferred the dark side but were able to learn to avoid shock by moving away from the dark.

Vibration has also been used as a negative stimulus in experiments on spider perception (Long et al. 2015), using a similar protocol as Bednarski et al. (2012). In Long et al.'s study, it was important that the spiders viewed the training stimulus from the same static position each time, so spiders were tethered using beeswax and rosin to a microbrush that could be clipped in place. Spiders then watched a training video while their tarsi were periodically vibrated (at 196 Hz) by a motorised platform. After training, spiders were given a choice between the training video and a novel video, and they preferred the novel video.

Heat has been used as an aversive stimulus in testing the ability of spiders to distinguish colours. Jumping spiders Hasarius adansoni (Audouin, 1826) were tested in arenas divided into two halves, each with a different colour of paper. After a control test in the arena for 3 min, one half of the arena was heated from the bottom and spiders had 3 min to explore the arena in a training session. In a post-training test in an unheated arena, spiders avoided the heat-associated papers (Nakamura & Yamashita 2000). A similar protocol was used by VanderSal & Hebets (2007) with another jumping spider (Habronattus dossenus Griswold, 1987), but with the addition of a seismic stimulus provided by a mini-shaker. When the seismic stimulus was turned on, spiders were more successful at discriminating the colours in the test trial, perhaps because the vibrations increased their attention to the task.

In a very creative study, Jackson et al. (2001a) tested the ability of Portia fimbriata to use trial-and-error learning to escape from a water-filled dish. The dish contained a central small island and a raised atoll halfway between the island and the edge of the chamber. Spiders were placed into two categories: swimmers and leapers. Swimmers were rewarded for choosing to swim from the island to the atoll by creating waves that gently pushed them towards their destination, and were punished for choosing to leap by creating waves that pushed them back to the island. Leapers were rewarded for leaping and punished for swimming. Once a spider had made it either successfully to the atoll or was unsuccessfully returned to the island, it could either try the same method of escape or switch to the other method for the second attempt. Successful spiders, both swimmers and leapers, repeated the initial behaviour while unsuccessful spiders switched methods. Subsequently, Cross & Jackson (2015) expanded this test to additional species from the subfamily Spartaeinae. They found that seven species that were web-invading aggressive mimics were proficient at solving this problem, in contrast to two non-aggressive-mimic species.

Many of the protocols described in this review can be adapted to characterise the nature of memory. For example, after training a spider to avoid particular prey, one might simply add extra trials at later periods to see when the memory has decayed. Peckmezian & Taylor (2015b) point out that their passive-avoidance protocol using shock may offer a way to standardise studies of the speed of learning and the rate of memory decay across species. Cross & Jackson (2014) studied memory using a different approach (expectancy-violation methods) that did not involve training spiders. Here, spiders were shown a lure consisting of a dead prey mounted on a disc. Before the spider could attack, the prey was briefly hidden. Then, either a prey of the same type or a new type was revealed. Spiders shown a new type of prey were less likely to attack than spiders shown the original type, suggesting that the spiders had a working memory of the first prey they had seen.

Unpublished protocols

Many laboratories have encountered problems using shock to train spiders. As reviewed in Long et al. (2015), in order for spiders to receive a shock on a shock platform, they must complete an electrical circuit by touching two adjacent metal strips, rods or parts of a grid. We have noticed in our apparatuses that spiders can learn to avoid completing a circuit by carefully positioning their body and tarsi. In addition, the amount of shock a spider receives depends on the part of the body that completes the circuit, as the resistance of the chitinous exoskeleton increases with its thickness.

T Peckmezian (Macquarie University, pers. comm. June 2015) built an automated shuttle box that had independently controlled iPod video screens at either end of an electrified shock floor. When the video came on, the jumping spider (Servaea incana) had to learn to switch sides in order to avoid a shock. Thus, the area of the box associated with punishment was constantly changing. Although the solution to this two-way avoidance task may have been too difficult for the spiders to learn, the shuttle-box protocol may be useful for other tasks or other spider species.

E Hebets (University of Nebraska, pers. comm. June 2015) reports that an attempt to train Schizocosa wolf spiders to distinguish between different visual patterns by using forceps that delivered a charge also failed to produce compelling results in a pilot study.

Heat also produces inconsistent results. Our lab attempted to replicate the Nakamura & Yamashita (2000) experiment described above using several species of Phidippus, and found no evidence that these spiders could learn the task. While our spiders avoided the hot side in the training trials, they immediately resumed exploring the entire arena during the test trial. Whether this is due to a difference in learning capabilities between our species and Hasarius adansoni, to less interesting differences in physiology (our spiders seemed to initially prefer the heated side until we raised the temperature even more; J Houser, pers. comm. June 2015), or to a difference in the details of our methods, is unclear.

Marpissa muscosa was also tested unsuccessfully with heat: in one experiment, spiders were in a cooled T-maze with one arm warmed as a reward, but it was difficult to present the conditioned stimulus (patterned paper) with the unconditioned stimulus (the warm temperature) closely in time, as the maze did not warm quickly enough. In a second pilot experiment using two colours as cues for a heated area, the spiders moved so actively that it was impossible to assess a preference (J Liedtke, Universität Hamburg, pers. comm. June 2015). A general lesson is that a potential confound with heat is that the movement rate of ectotherms is likely to be greatly affected by temperature.

Two labs, to our knowledge, have tried some version of the Morris water maze, which is used to test spatial learning. Often used with rodent species that find being in water aversive, the maze consists of a platform hidden just below the surface of an opaque liquid. The animal swims in the water until it locates the platform, where it can rest. The animal learns about the position of the platform using cues positioned around the maze. In our lab, we used a container of water with exits marked by symbols on the side of the platform. Phidippus jumping spiders could float on the water with their eyes above the surface and ‘swim’ towards the exit, but we found no evidence of learning in pilot experiments. On the other hand, we found that Trite planiceps Simon, 1899 jumping spiders in New Zealand sank immediately. Another lab tested jumping spiders with a version of a Morris maze with shock rather than water, providing a safe zone in the middle of a shock platform. Distal and local location cues were provided. Data are still under analysis, but a shock-based maze may be more promising than the water maze (T Peckmezian, Macquarie University, pers. comm. June 2015).

In our laboratory, we also used sprays of water as an aversive stimulus—spiders that moved into one side of an arena received a spray of water. In pilot trials, there was no evidence of learning: spiders ran wildly when sprayed and then stopped to groom themselves. If accidentally sprayed too heavily, they became trapped by the water droplet.

General considerations in experimental design

Taken together, the tales of failed experiments make the point that different spider species, even those that are quite closely related, may behave very differently from one another. As with most studies of animal behaviour, spending time in getting to know the study species is key. For example, some jumping spider species are calm during experiments while others exhibit heightened arousal and may not attend to the stimulus that the experimenter is presenting. Some species are quick to behave in an experiment while others can sit motionless for long periods. This interspecific variation means that we must carefully tailor our methods to investigate a cognitive process in a particular species (F Cross, University of Canterbury, pers. comm. October 2015).

As with any study of animal learning, the experimenter must consider carefully whether it is better to plan a more natural experiment in the field in which some variables, such as the spiders’ previous experiences, are less controlled, or a more controlled experiment in the laboratory that may have little applicability to normal behaviour. In lab studies, even small changes in arena design can mean the difference between success and failure; for example, a white arena with white walls may be disorienting to a jumping spider, which can be predisposed to be attracted to particular colours (Baker et al. 2009). Possible confounding variables must be considered carefully from the perspective of the sensory system of the spider; for example, chemical cues in nail polish might be detectable to wolf spiders and thus might provide confounding cues in an experiment (Rutledge et al. 2010). In addition, many researchers have noted that spider behaviour varies with time of day and weather, even in the laboratory.

Many researchers expressed to us that they were sometimes frustrated by individual differences in spider behaviour. Sometimes these took the form of particular individuals failing to participate in tasks (D Rao, Universidad Veracruzana, pers. comm. June 2015), differences within a population depending on when individuals were collected in the field and how long they had spent in the lab (G Uetz, University of Cincinnati, pers. comm. June 2015; also see Carducci & Jakob 2000), a decline in performance with age (our lab) or a decline in performance after several generations in the lab (D Harland, AgResearch, Christchurch, pers. comm. June 2015). We have also documented differences between the sexes. Thus, it is important to keep close track of all these variables to help illuminate why experiments sometimes fail and sometimes succeed.

Ultimately, many researchers in spider behaviour would like to understand broad patterns in cognitive abilities across the taxon. The best way for the arachnology community to move towards phylogenetically-informed comparisons of learning capabilities across spider populations and species is to freely share experimental designs and, when possible, to publish experiments with negative data. Whether successful or not, each experiment described here helps us develop robust methods that can be adopted by researchers from around the world.

Acknowledgements

Thank you to Cor Vink for his invitation to participate in this Festschrift issue dedicated to the inspirational and indefatigable Robert R Jackson, to whom we owe so much. Thank you to the people who responded swiftly to our requests for tales of success and woe: Fiona Cross, Gary Dodson, Eileen Hebets, Duane Harland, Chad Hoefler, Jeremy Houser, Jannis Liedtke, Douglass Morse, Ximena Nelson, Tina Peckmezian, Matthew Persons, Dinesh Rao, Ann Rypstra, Lisa Taylor, Phillip Taylor, George Uetz and Daniel Zurek. Al Kamil and Dan Papaj kindly answered queries about the learning literature from EMJ. Adam Porter provided helpful comments on a draft.

Guest Editor: Dr Fiona Cross.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This work was supported by NSF IBN 0952822 to EM Jakob.