Animal Culture: Individual Learning Flashcards
What Is Individual Learning?
Learning refers to a relatively permanent change in behavior as a result of experience (Shettleworth,
1998). This definition does have a downside, in that it is not clear how long a time period is encompassed by the words relatively permanent. That being said, this is a working definition, already adopted implicitly by most ethologists. The definition suggests an interesting
relationship between learning and what evolutionary ecologists refer to as “phenotypic plasticity”.
A phenotype is typically defined as the observable characteristics of an organism (P. Walker, 1989), and phenotypic plasticity is broadly defined as the ability of an organism to produce different phenotypes
depending on environmental conditions. For example, many invertebrates, such as the bryozoan Membranipora membranacea, live in colonies (Harvell, 1998). When living in such colonies, individuals
typically lack the spines that are used as an antipredator defense in related species.
These spines are simply not grown when a Membranipora membranacea colony develops in the absence of predators. Yet individuals grow spines relatively quickly when exposed to predatory cues (Harvell, 1991, 1994; Tollrian and Harvell, 1998). The resultant change, from spineless to spined, constitutes a case of phenotypic plasticity. The phenotype of this bryozoan shifts dramatically as a result of environmental changes—in this case, the addition of a predator—and hence is thought of as “plastic”.
all learning is a type of phenotypic plasticity, but not all
phenotypic plasticity involves learning. To see why, consider the “flushing” behavior often seen in foraging birds. While searching for food, some birds may move their tails and wings in a way that flushes insects out from cover—insects that the bird then eats. In the painted redstart (Myioborus pictus), for example, when individuals are under branches, they increase their wing and feather motion and flush insects from the overhanging branches.
One hypothesis to explain this flushing behavior is that the birds learn that when they are under branches and
flap their wings, they will get food. However, this response could be based on a relatively fixed genetic response rather than learned. Piotr Jablonski and his colleagues designed an experiment to distinguish between these two hypotheses. What they found is that, while it is true that birds in nature increase their wing-flapping behavior when they are under branches, the same increase in wing flapping also occurs in the laboratory, even when the birds are not rewarded for such behavior: when Jablonski’s team put naive birds under branches, they started flapping more as well, even when they got no food for doing so (Jablonski et al., 2006).
These results suggest that flushing insects under branches does represent a case of phenotypic plasticity—the ability of an organism to produce different phenotypes depending on environmental conditions (whether the birds are under trees or not)—
but it is not a case of learning.
How Animals Learn
• Learning from a Single-Stimulus Experience
• Pavlovian (Classical) Conditioning
• Instrumental (Operant) Conditioning
LEARNING FROM A SINGLE-STIMULUS EXPERIENCE
The simplest experience that can lead to learning involves a single stimulus—a stimulus that can take almost any form. For example, let’s imagine that we are interested in studying learning in rats and that
numerous times throughout the day we place an arbitrary cue—a bluecolored stick—in a rat’s cage. Rats will often take note of such a disturbance and turn their heads in the direction of the blue stick. If, over time, the rats become more likely to turn their heads in the
direction of the blue stick—that is, if they become more sensitive to the stimulus with time, sensitization has occurred. Conversely, if, over time, the animals become less likely to turn their head, habituation is said to have taken place. Sensitization and habituation are two simple single-stimulus forms of learning.
The process of habituation can be problematic when it comes to designing an ethological experiment, because it can be difficult to examine behavior if animals habituate quickly to stimuli. For example,
in many experiments involving antipredator behavior, predators may be housed such that visual interactions between predator and prey are possible, but the predator can’t actually harm the prey. This ethical compromise spares the life of the potential prey, but it also creates a scenario in which the prey may now habituate to the predator, having learned that the predator cannot in fact move close enough to
present any real danger (Huntingford, 1984). Because of these sorts of issues, ethologists often need to go to great lengths to be certain that habituation has not occurred in their study system (Rowland and Sevenster, 1985).
An animal’s habituation to a stimulus may interfere with later attempts to get the animal to associate that stimulus with some other cue. For example, if rats habituate to the blue stick, it might prove more difficult
for them to subsequently learn that the blue stick signals the arrival of food. In contrast, if sensitization to a single cue has occurred, it may facilitate the association of the sensitized stimulus with other cues.
PAVLOVIAN (CLASSICAL) CONDITIONING
Suppose that rather than giving a rat a single stimulus like the blue stick, from the start we pair this stimulus with a second stimulus, like the odor of a cat—an odor that rats fear. Seconds after the blue stick is in place, we spray the odor of a cat into one corner of the cage. If the rat subsequently learns to pair stimulus 1 (blue stick) with stimulus 2 (cat odor) and responds to the blue stick by climbing under the chip shavings (a safer location), but before the odor is sprayed in, we have designed an experiment in Pavlovian or classical conditioning (Kim and Jung, 2006).
This form of conditioned learning was first developed by Ivan Pavlov in the late 1800s (Pavlov, 1927). Pavlovian conditioning experiments involve two stimuli—the conditioned stimulus and the unconditioned stimulus (Domjan, 2005, 2006). A conditioned stimulus
(CS) is defined as a stimulus that initially fails to elicit a particular response but comes to do so when it becomes associated with a second (unconditioned) stimulus. In our rat example, the blue stick is
the conditioned stimulus, as initially the rat will have no inherent fear of it. The unconditioned stimulus (US) would be the cat odor, which inherently causes a fear response in rats. Once the rat has learned to hide after the blue stick (CS) alone is in place, we can speak of its hiding as being a conditioned response (CR) to the presence of the blue stick.
To examine Pavlovian conditioning in detail, we need to define a few more terms. In the learning literature, any stimulus that is considered positive, pleasant, or rewarding is referred to as an appetitive stimulus. Appetitive stimuli include food, the presence of a potential mate, a safe habitat, and so on. Conversely, any stimulus that is unpleasant—shock, noxious odors, and so forth—is labeled an aversive stimulus. Another important distinction made in the learning literature is between positive and negative relationships.
When the first event (placement of the blue stick) in a conditioning experiment predicts the occurrence of the second event (cat odor), there is a positive relationship between events. Conversely, if the first event predicts that the second event will not occur—imagine that a blue stick is always followed by not feeding an animal at its normal feeding time— there is a negative relationship. Positive relationships—for example, blue stick predicts cat odor—produce excitatory conditioning. Negative relationships—for example, blue stick leads to no food at a time when food is usually presented—produce inhibitory conditioning.
Pavlovian conditioning experiments can become complicated when second-order conditioning is added on. In second-order conditioning, once a conditioned response (CR) has been learned by pairing US and CS1, a new stimulus is presented before the CS1, and if the new stimulus itself eventually elicits the conditioned response, then the new stimulus has become a conditioned stimulus (CS2). In our case, any rat that has learned to pair the blue stick (CS1) with danger
might now see a yellow light (CS2) preceding the appearance of the blue stick. Once the rat has learned to pair the yellow light (US) with the danger associated with cat odor, second-order Pavlovian conditioning
has occurred.
Overshadowing, Blocking, and Latent Inhibition:
Pavlovian conditioning affects not only behavior but also learnability, that is, the ability to learn under certain conditions. We will explore three types of learnability: overshadowing, blocking, and latent inhibition. Consider an experiment with four groups of rats. Suppose that group 1 individuals undergo a standard Pavlovian paradigm with two stimuli: the blue stick (CS1) and a cat odor (US). In group 2, a second conditioned stimulus (CS2), a yellow light, is always presented simultaneously with the blue stick, just before the cat odor is sprayed. Subjects from both groups are then tested in response to the blue stick alone. If the yellow light is overshadowing the blue stick, rats in group 2 will respond less strongly to the blue stick than will rats in group 1. The CS2—the yellow light—has made it more difficult for the rats to pair the blue stick and the cat odor is sprayed.
In group 3 of our rats, individuals are first trained to associate the blue stick with the cat odor, but after this training, the yellow light is presented at the same time as the blue stick, and this compound stimulus is paired with the cat odor. In group 2, the blue stick and the yellow light were always presented together. Rats in group 3 differ from those in group 2 in that they first learned to associate the blue stick and cat odor before any yellow light is added to the protocol.
Blocking (a reaction) occurs when those in group 3 respond less strongly to the yellow light (when it is presented alone) than individuals in group 2 do (Kamin, 1968, 1969). It is as if initially learning to associate the blue stick alone with the cat smell blocked group 3 subjects’ ability to pair the yellow light with the cat odor.
A fourth group of rats is initially exposed to a blue stick, but no cat odor, for a long period of time. We then attempt to pair the blue stick with cat odor at some subsequent point in time. If we find that the rats
in group 4 have more difficulty learning than the rats in group 1 (where standard Pavlovian pairing has occurred), then we would say that rats in groups are displaying latent inhibition.
INSTRUMENTAL (OPERANT) CONDITIONING
Instrumental conditioning, also known as operant or goal-directed learning, occurs when the response that is made by an animal is reinforced (increased) by the presentation of a reward or the termination of an aversive stimulus, or when the response is suppressed (decreased) by the presentation of aversive stimulus or the termination of a reward. One of the most fundamental differences between Pavlovian and instrumental learning is that, in instrumental learning, the animal must undertake some action or response in order for the conditioning process to produce learning.
The classic example of instrumental learning is a rat pressing a lever (the action) to get food to drop into its cage. Rats associate pressing on the lever (response) with some probability of getting food (outcome) and learn this task. The earliest work on instrumental learning was that of Edward Thorndike and involved testing how quickly cats could learn to escape
from “puzzle boxes” that Thorndike had constructed (Thorndike, 1898, 1911). When Thorndike placed a cat in a locked box, the cat initially tried all sorts of things to get out of its confined space. Some of these behaviors, by chance, led to a successful escape from the box.
Thorndike hypothesized that the cat began to pair certain behaviors that it undertook in the box with a positive effect—escape—and it was then more likely to use such behaviors when confined in the puzzle
box. His data suggested they did. Combining the findings from his puzzle box experiment with other results he had obtained, Thorndike postulated the law of effect, which states that if a response in the
presence of a stimulus is followed by a positive event, the association between the stimulus and the response will be strengthened. Conversely, if the response is followed by an aversive event, the association will be weakened.
Work in instrumental learning was revolutionized by B. F. Skinner, who devised what is now known as a Skinner box (Skinner, 1938). His idea was to create a continuous measure of behavior that could somehow be divided into meaningful units. When a rat pushes down on a lever, it is making an operant response because the action changes the rat’s environment by adding food to it. Because “lever pushing” is a relatively unambiguous event that is easily measurable, and because it occurs in an environment over which the rat has control, the Skinner box has facilitated the work of psychologists doing research within the instrumental learning paradigm.
Why Animals Learn
• Within-Species Studies and the Evolution of Learning
• COGNITIVE CONNECTION: Natural Selection and Associative Learning
• Population Comparisons and the Evolution of Learning
• A Model of the Evolution of Learning
• CONSERVATION CONNECTION: Learning, Alarm Chemicals, and Reintroduction Programs
WITHIN-SPECIES STUDIES AND THE EVOLUTION OF LEARNING
Both Edward Thorndike and Ivan Pavlov argued that, aside from the details, the qualitative features of learning are the same in all animals, including human beings—that is, all animals learn in a fundamentally
similar fashion (Bitterman, 1975; Pavlov, 1927; Thorndike, 1911). This view became widely accepted, and was promoted by such psychologists as Skinner and Harry Harlow (Harlow, 1959; Skinner, 1959). If Thorndike and Pavlov were correct that the particular
environment an organism evolved in has no effect on learning, the same sort of learning should be seen in all creatures that learn, regardless of the sort of learning tasks with which they are presented.
Could it really be the case that natural selection on learning does not lead to differences in learning across populations and in different species? It seems unlikely. The ability to learn should be under strong selection pressure, such that individuals that learn appropriate cues that are useful in their particular environment should be strongly favored by natural selection. This is the “ecological learning” or “cognitive ecology” model, and its influence is getting stronger as our knowledge about evolution and learning increases.
Such ecological learning was first brought to the attention of many researchers with the work of Garcia and his colleagues. Garcia and his team ran a series of experiments that made many researchers in both psychology and biology rethink their approach to the study of learning (Garcia et al., 1972; Seligman and
Hager, 1972). What is striking about Garcia’s work on rats is that in many ways the protocol he used was very similar to that already being used in psychology learning experiments. In essence, Garcia tried to
get rats to form an association between a series of cues (Garcia and Koelling, 1966).
The “bright-noisy” water treatment had water
associated with a noise and an incandescent light and the “tasty” water had water paired with a particular taste, a gustatory cue. He then paired the bright-noisy water or the tasty water with one of the following
negative stimuli: radiation, a toxin, immediate shock, or delayed shock. The radiation and the toxin made the rats physically ill, while the shocks were painful. For example, in the bright-noisy water/radiation treatment,
bright-noisy water would be presented and the rats that were being trained to this cue were exposed to radiation after drinking the water.
Garcia and Koelling found rats quickly learned that tasty water was to be avoided after this cue was paired with X-rays or toxins, but they did not learn to avoid bright-noisy water after it was paired with X-rays or toxins. In contrast, when Garcia and Koelling examined the rats that were given shock treatments, they
found that shock was easily paired with bright-noisy water (audiovisual cue), but not tasty water (gustatory cue). Garcia and Koelling explained their results in terms of adaptation— something quite unusual for psychologists of the 1960s.
They argued that, on the one hand, natural selection would favor the ability to pair gustatory cues (tasty water) with internal discomfort (getting ill). After all, many instances of internal discomfort in nature are likely to be caused by what an animal has consumed, and rarely are food cues associated with audiovisual cues (as in the bright-noisy water treatment). On the other hand, peripheral pain, like that caused by a
shock, might be more commonly associated with some audiovisual cue like hearing or seeing a conspecific or a predator, so again natural selection should favor the ability to pair these cues together.
Garcia also found that learning in their rats occurred
without immediate reinforcement (Garcia et al., 1966). Most psychologists believed that delays in reinforcement on the order of seconds can stifle animal learning, yet Garcia found that learning occurred even after delays of seventy-five minutes, when injections of
noxious substances were paired with drinking saccharin-flavored water. From an adaptationist perspective, one would expect a delay between
the time that a rat consumed a substance and any subsequent negative effect of such consumption. As such, natural selection would have favored rats that were able to associate what they ate with becoming ill,
even if the events were separated by significant time intervals.
COGNITIVE CONNECTION - Natural Selection and Associative Learning:
Garcia’s experiments suggest that knowledge of the selection pressures that animals face in nature will help us better understand the nature of learning. If
researchers were to experimentally manipulate selection pressures on the reliability of the cues used during associative learning, then we should be
able to predict patterns of learning. Aimee Dunlap and David Stephens designed such a study by examining a form of associative learning called prepared learning in fruit flies (Drosophila melanogaster; Dunlap and
Stephens, 2014).
In their experiment, female fruit flies chose between different locations for laying their eggs. In part 1 of the study, one of the locations had quinine added to it—a chemical the flies are averse to. The quinine location had two cues associated with it—a color cue (aqua or blue) and an odor cue (amyl acetate or benzaldehyde). In part 2 of the study, when females were ready to
lay eggs, they were exposed to the two locations again, the quinine was removed, but both color and odor cues were still present.
In treatment 1/4, the color cue in part 2 was a reliable indicator of quinine in part 1 (color always indicated the
“correct” location to lay eggs in), but odor was an unreliable cue. In treatment 2/4, the odor cue in part 2 was a reliable indicator of where to lay eggs, but
color was not. The experiment went on for forty generations—in each generation, only eggs laid in the location with the reliable cue were selected. At generation 40, when the researchers tested the response of flies in treatment 1, they found more individuals learned to pair color and quinine. Flies from treatment 2 showed the opposite pattern with more flies learning to pair odor and quinine. Taken together, these results suggest that selection pressures can indeed change the nature of associative learning.
POPULATION COMPARISONS AND THE EVOLUTION OF LEARNING
Learning, Foraging, and Group Living in Doves:
The research question - Do individuals that live in groups have better learning skills than individuals that are more solitary?
Why is this an important question? Theory predicts that the competition inherent in group-living strongly favors learning.
What approach was taken to address the research question? Two populations of Zenaida doves were tested on their ability to learn about a food source. Individuals in one population lived in groups, while individuals in the second, geographically close, population were more territorial and tended not to live in groups.
What was discovered? Doves from the population where individuals lived in groups outperformed individuals from a more solitary-living population on a
learning-related foraging task.
What do the results mean? Because doves were taken from two different wild populations, and not raised from birth in a controlled environment, these
results could be due to different experiences the doves had in the wild and/or due to stronger selection pressures with respect to learning in the groupliving doves.
Learning and Antipredator Behavior in Sticklebacks:
One way to partially circumvent the confounding effects of learning (in real time) versus natural selection acting on the ability to learn is through the use of controlled laboratory experiments, where it is possible to raise individuals from two very different populations in a similar environment, and in so doing minimize differences associated with experience. Huntingford and Wright used this approach to study avoidance learning in two populations of three-spined sticklebacks (Gasterosteus aculeatus; Huntingford and Wright, 1992).
Some sticklebacks live in locales that contain many predators, and some live in populations in lakes with virtually no predators; and differences in predation pressures across locales have been in place for long periods of evolutionary time. Huntingford and Wright raised individuals derived from predator-rich and predator-free streams in the laboratory, and during their development individuals had no interactions with predators.
If natural selection has acted more strongly on antipredator strategies—including learning about danger—in the sticklebacks descended from predatorrich populations, then we should see such differences in learning abilities across sticklebacks descended from the different populations even in the absence of experience with predators. The researchers began by training eight sticklebacks from each of their population groups to associate one side of their home tank with food.
They found no differences in learning across populations in the context of foraging alone - individuals from both populations were equally adept at learning that food would come to one side of their tank. Then, after a stickleback had learned that one side of its tank was associated with food, fish were subjected to a simulated attack from a heron predator on the side of the tank that contained food. Huntingford and Wright then examined whether between-population differences emerged in terms of how long it took the fish to learn to avoid the side
of the tank associated with heron predation (and food).
While all but one fish from both high and low predation populations eventually learned to avoid the dangerous end of their tank, fish from high-predation areas learned this task more quickly than did fish from
predator-free populations. Two lines of evidence support the hypothesis that natural selection has operated on learning and antipredator behavior in these populations of sticklebacks.
First, the laboratory protocol used minimized the probability that individual experiences differed across the populations they examined. Second, Huntingford and Wright did not find between-population differences in all learning contexts. When the task was a simple association of food and place, interpopulational differences with respect to learning were absent; such differences were found only when the learning task was to avoid feeding in areas associated with danger.
A MODEL OF THE EVOLUTION OF LEARNING
Many cost-benefit models have been developed to examine when natural selection might favor the ability to learn. Imagine a behavioral scenario in which the options are to respond to some stimuli with a fixed genetically programmed response, or to respond to stimuli based on prior experience—that is, by learning.
Are the net benefits associated with learning greater or less than the net benefits that might be associated with a fixed genetic response to some stimuli?
Ethologists, behavioral ecologists, and psychologists have argued that natural selection should favor the ability to learn over the genetic transmission of a fixed trait when the environment an animal lives in changes often, but not too often, because most of these models
assume (1) that there is some cost to learning, even if it is only a very small cost; and (2) that the ability to learn has an underlying genetic basis. Though difficult to measure, evidence for both these assumptions has been found in fruit flies and butterflies (Burger et al.,
2008; Burns et al., 2011; Kawecki, 2010).
The cost to learning can take different forms. In one experiment using a line of Drosophila melanogaster fruit flies that had been selected for better learning skills for more than 50 generations, Mery and Kawecki
subjected flies to a learning test in which they were trained to associate an odor with a mechanical shock (Mery and Kawecki, 2002, 2005). The learning protocol involved five trials to pair odor and shock. After that
the flies were deprived of food and water, and the researchers noted how long it was before they died.
Female flies in the learning treatment died about 9 hours sooner than females flies from the same line that had been selected for better learning abilities, but that had not been exposed to the learning trials or
shocked. That difference could have been due to a cost of learning, the exposure to shock, or both. Evidence for a cost to learning was seen when comparing the flies in the learning treatment to a control that did not involve learning, but did have flies experience the shock that was used in the learning trials.
Flies in the learning treatment died 4–5 hours sooner than those who had just suffered the shock alone.
Reduced survival is not the only cost to learning. Emilie Snell-Rood and her colleagues have found a fecundity cost of learning in the cabbage white butterfly (Pierus rapae) (Snell-Rood et al., 2011). This work allows us to distinguish between two different types of costs to
learning, “constitutive costs” and “induced costs.” Constitutive costs are paid by good learners, regardless of whether they learn some task, while inducted costs are paid only when learning has occurred.
Both costs were examined in the cabbage white butterfly system. Cabbage white butterflies lay their eggs (oviposit) on green plants, and show an innate preference for searching for green items, but there
is genetic variation in the extent to which individual butterflies can be trained to search for red items. When Snell-Rood and her colleagues measured reproductive success, they found that females from families with individuals who best learned to forage for red items produced fewer and less well developed eggs, even when given no learning tasks (Snell-Rood, 2011).
This constitutive cost of learning is likely related to
the investment in the larger brains of individuals seen in individuals in these families (Snell-Rood et al., 2009). On top of this constitutive cost was an induced cost to learning: butterflies in the families selected for the ability to learn to forage on red items, showed an greater decrease in reproductive success after being part of learning trials involving red items.
Returning to our discussion of a model for the evolution of learning, when the environment rarely changes—and hence the environment that offspring encounter is similar to that of their parents—information is best passed on by a fixed genetic rule, since such a means of transmission avoids the costs of learning. On the other end of the spectrum, if the environment is constantly changing, there is little worth learning because what is learned is completely irrelevant in the next situation.
When the environment is constantly changing, acting on past experience is worthless, as past experience has no predictive value and so genetic transmission of a fixed response, rather than a costly learned response, is again favored. Somewhere in the middle, in
between an environment that never changes and one that always changes, learning is favored over the genetic transmission of a fixed response and it is worth paying the cost of learning. The environment is stable enough to favor learning, but not so stable as to favor genetic transmission.
Because many models confuse two types of stability, David Stephens reformulated the way that environmental stability is represented in models (Stephens, 1991, 1993). The model that Stephens developed breaks environmental predictability into two types: (1) predictability within the lifetime of an individual, and (2) predictability between the environment of parents and offspring. These two types of predictability can be very different, and conflating them may hinder our understanding of the evolution of learning (Figure 5.21).
For example, consider a case in which early in life the offspring A move to environments that are far removed from those of their parents, and the environment to which they migrate is stable over the course of their
lifetime. Here, between-generation environmental predictability is quite low, while within-lifetime environmental predictability is much higher. This distinction is lost when models lump together within- and between generation environmental predictability.
Stephens found that learning is favored when predictability within the lifetime of an individual is high, but environmental predictability between generations is low. Predictability within generations is low, so neither strategy does particularly well. But since
learning has a cost associated with it, genetic transmission of a fixed response is favored. Fixed genetic transmission is again favored in box 4, because with high predictability at all levels, the cost of learning is never worth the investment.
Only in box 3, with high within-generation predictability but low between-generation predictability, is learning favored. Learning is favored here because once an organism learns what to do, it can repeat the appropriate behaviors during its lifetime. But isn’t this the sort of predictability that usually favors a fixed genetic transmission? Yes, but now the environment changes so much between generations that fixed genetic transmission would be less advantageous than learning.
Learning, Alarm Chemicals, and Reintroduction Programs:
One way in which conservation biologists try to protect threatened or endangered species is through reintroduction programs (Batson et al., 2015; Ewen et al., 2012; Kemp et al., 2015; Abbott and Richardson, 2015). These programs often involve managers raising individuals of a threatened or endangered species in captivity and then releasing them into an area that the
species formerly occupied. Reintroduction programs have had mixed success.
One problem is that reintroduced individuals are often especially susceptible to predation, in part because they experience no threats while being raised in captivity. A similar issue arises in translocation programs, when individuals are moved from one natural habitat to another, and in fisheries, when fish are released into the wild (Olson et al., 2012). Conservation biologists understand this very well, and many programs now try to present individual animals with the opportunity to learn something about one aspect or another relative to the environment into which they will be released or transferred, before the release or introduction occurs.
For example, hellbenders (Cryptobrancus alleganeinsis), large aquatic salamanders whose natural range has declined dramatically, have an innate
fear response when exposed to an alarm chemical—a white mucus—that is produced by other hellbenders. To better understand how to design reintroduction programs for hellbenders, Crane and Mathis used a classic conditioning protocol in which one group of hellbenders was given the opportunity to pair the alarm chemical with the scent of brown trout—a predator of hellbenders.
For a second (control) group, the alarm chemical
was paired with water (Crane and Mathis, 2011, 2013). The hellbenders that were given the chance to pair the trout odor and the alarm chemical showed more fine-tuned antipredator behaviors in response to trout than the hellbenders in the control group (Figure 5.22). These classical conditioning protocols could be used in reintroduction programs for hellbenders. Individuals trained to show fear responses when exposed to real predator cues would be more likely to take action to avoid the danger and therefore survive when encountering predators in nature for the first time.
What Animals Learn • Learning about Predators • Learning about Their Mate • Learning about Familial Relationships • Learning about Aggression
LEARNING ABOUT PREDATORS
Prey often live in areas that contain both predatory and nonpredatory species, and learning which species is which has fitness consequences. Even encounters with the same predator are not always the same, as at any given time some individuals are in hunting mode
while others are not actively hunting prey (Chivers et al., 1996). If prey can learn to distinguish between dangerous and benign encounters with potential predators, they may free up time for other activities:
learning about possible predation pressure may allow animals to handle the trade-offs they constantly face.
If the food that a potential predator eats produces a chemical cue that is recognizable to its prey, then that cue may provide an opportunity for such prey to learn what is dangerous and what isn’t. Douglas Chivers and his colleagues examined chemical cues and the role of learning in the antipredator behavior of damselfly larvae (Enallagma spp; Chivers et al., 1996). Damselfly larvae are found in ponds with minnows, and both species are often attacked and eaten by pike (Esox lucius). Chivers and his colleagues hypothesized that damselfly larvae might learn about the potential dangers associated with pike encounters by using chemical cues.
To test this hypothesis, the researchers fed pike predators either minnows, damselflies, or mealworms; mealworms served as a control as they are not eaten by pike. After four days on one of these three diets, a pike was removed from its tank, and damselflies that had never before had any contact with a pike were exposed to the water from the pike’s tank. When damselfly larvae were exposed to the water containing
chemical cues from a pike that had eaten damselflies or a pike that had eaten minnows, Chivers and his colleagues found that the damselfies had significantly reduced their foraging behavior.
But the damselfly larvae did not reduce their foraging behavior when they were exposed to the water treated with pike that had eaten mealworms. Because damselflies are found in the same ponds as minnows, but the damselflies tested by Chivers had never before experienced a pike, these results strongly suggest that damselflies innately associate the scent of pike plus damselfly or pike plus minnow with danger, but they make no such association between pike, mealworm, and danger. The damselflies here hadn’t learned anything; they simply were predisposed
to respond to the smell of pike and prey (minnows and damselflies) as dangerous.
Chivers’s team followed up this experiment by examining the role of learning in the antipredator behavior of damselflies. Here they took the damselflies that had been exposed to the three treatments above
(water from pike plus damselfly, water from pike plus minnow, and water from pike plus mealworm) and isolated them for two days. Then each damselfly was exposed to water from a pike that had been fed
mealworms.
With respect to learning and antipredator behavior, there were three groups of damselflies; group 1: damselflies that were initially exposed to water from pike plus damselfly, but that were subsequently
exposed to pike plus mealworm water; group 2: damselflies that were initially exposed to water from pike plus minnow, but that were subsequently exposed to pike plus mealworm water; and group 3: damselflies exposed to water from pike plus mealworms twice.
Damselflies from groups 1 and 2 had responded with antipredator behaviors in the first experiments, but damselflies in group 3 had not. Damselflies in groups 1 and 2 responded to the scent of pike plus mealworm by decreasing their foraging activities. Damselflies in group 3 did not decrease their foraging. Recall that in the first experiment, damselflies did not curtail feeding when they encountered the smell of pike plus mealworm for the first time, but they did curtail feeding in this second experiment.
These results suggest that, based on their earlier
experience in the first experiment, damselflies in groups 1 and 2 in the second experiment had learned to associate pike plus the scent of any potential prey with danger, and this association translated into a
reduced foraging rate even when they encountered the scent of pike and mealworm.
LEARNING ABOUT THEIR MATE
Michael Domjan and his colleagues have studied the role of Pavlovian conditioning in mate choice in the Mongolian gerbil (Meriones unguiculatus), a species of desert rodent that relies on chemical communication during the formation of pair bonds (Ågren, 1984;
Thiessen and Yahr, 1977; Domjan et al., 2000). Villarreal and Domjan first allowed pair bonds to form between a male and a female Mongolian gerbil.
They presented one group of males with an olfactory
cue (mint or lemon) and then gave them access to their partners, and exposed males in another (control) group to the odor, but did not provide them with access to females following this presentation. Males that experienced the pairing of an odor and subsequent access to a pairmate learned relatively quickly to approach an area where access to the female was signaled by an odor, while males in the control group
formed no such association (Villarreal and Domjan, 1998).
Villarreal and Domjan next tested whether females learned to associate an odor with the presence of their pairmates, and whether any male/female differences emerged. Females did learn to pair odor and access to their pairmates—conditioned females responded to odor cues by approaching the area associated with this cue, and differences between males and females disappeared over time.
While no sex differences were found in the Mongolian gerbils, Domjan and Karen Hollis have hypothesized that in species where differences between males and females in their learning abilities when selecting mates do exist, this should be linked with differences in male
and female parental investment (Domjan and Hollis, 1988). The more equally parental investment is shared, the more the sexes should be similar in terms of learning the location of partners.
One way to think about Domjan and Hollis’s hypothesis is in terms of how much each sex is willing to invest in future offspring (Trivers, 1974). If both males and
females provide resources for offspring, selection pressure for learning ability about partners should be strong on both sexes. In such a case we expect males and females to have the ability to learn where their
mate—the co-provider of resources for their offspring—is at any given time.
In many species, only females provide resources for offspring. There are many reasons that females are more likely to take this role (see chapters 7 and 8), but for the purposes of Domjan and Hollis’s hypothesis, what this translates into is that males in such systems
should be better at learning about the location of mates than females. In terms of parental care for subsequent offspring, females are a valuable resource for males, as they alone provide food for offspring.
From a parental care perspective, males are much less valuable a resource to females. Males then are under strong selection pressure to find receptive females, while females can almost always find males that
are willing to mate. In such mating systems, selection for learning locations associated with potential mates should be stronger in males than in females. More experimental work needs to be done in this area,
but some support for this hypothesis has been found.
First, in both Mongolian gerbils and gourami fish, parental investment is shared, and differences in learning about mates between the sexes is small.
Second, in contrast to Mongolian gerbils and gourami fish, in Japanese quail, where there is no parental investment on the part of males, males show greater learning abilities than do females, though the extent of
differences in learning between males and females may be specific to certain mating contexts (Gutierrez and Domjan, 1996, 2011; Hollis et al., 1989; Villarreal and Domjan, 1998).
LEARNING ABOUT FAMILIAL RELATIONSHIPS
If individuals can learn how they are related to others around them, as well as how different individuals in their group are related to one another, natural selection might favor altruistic and cooperative behavior being preferentially allocated to close genetic kin (chapter 9).
Here we examine learning and kin recognition in the context of helpersat-the-nest. As we have seen, in some birds and mammals, individuals forgo direct reproduction and instead help their relatives raise their offspring (Brown, 1987; Solomon and French, 1996; Stacey and Koenig, 1990).
Sometimes these offspring remain at their natal nest and help their parents raise a subsequent brood (the helper’s siblings). But this is only one way in which helping may emerge. For example, young, reproductively active long-tailed tits (Aegithalos caudatus) breed independently as soon as they can, but many nests fail because of predation on the young.
When that happens, breeders often become helpers at the nests of their close genetic relatives, and such helpers accrue indirect fitness benefits by helping raise their kin (Hatchwell and Sharp, 2006; Hatchwell et al., 2004, 2014). How do the birds know who are kin? Do they learn who is kin, and who isn’t, and if so, how?
To address these questions, Stuart Sharp and his colleagues ran a series of experiments that focused on the “churr” call made by longtailed tits.
This call develops before young birds fledge and leave the nest, and it remains consistent throughout the lifetime of an individual (Sharp and Hatchwell, 2005). Churrs are given by males and females in the context of short-range communications, such as those regarding nest-building and aggression (Gaston, 1973; Hatchwell et al., 2001; Sharp and Hatchwell, 2005).
Sharp and his team set up a “playback” experiment, in which an individual heard the taped call of either a close genetic relative or a nonrelative, and the long-tailed tits showed a preference for the calls given by their kin.
Following the playback trials, the researchers designed an experiment to assess whether the birds learned the churr calls of their relatives, or whether their preference
was based on genetic predispositions for certain churr calls. To do this, they ran a cross-fostering experiment (see chapter 2), in which chicks either were raised with their biological parents or were switched to another nest and raised by foster parents.
A number of lines of evidence suggest that the churr call is learned: (1) The calls of foster siblings raised together were as similar as the calls of biological siblings raised together; (2) the calls of biological
siblings raised apart were as dissimilar as the calls of unrelated individuals in nature; (3) the songs of foster parents and their foster offspring were similar, whereas the songs of biological parents and their offspring were different when those offspring were raised by foster
parents.
These results all suggest an important role for learning in the development of churr calls that are subsequently used to distinguish kin from nonkin.
LEARNING ABOUT AGGRESSION
Paper wasps (Polistes fuscatus) live in colonies in which individuals are constantly interacting with one another. Reproduction in paper wasp colonies is tightly linked to the position a wasp holds in a dominance hierarchy, so knowing who is who in such a hierarchy has important consequences for reproductive success. Prior work has shown that paper wasps recognize their hive mates, but how do they do so?
Because these wasps have facial marks that might allow for such recognition, researchers have examined whether facial learning occurs in the species. This type of specialized learning, which is especially prominent in humans, has been demonstrated in other mammals, but
until Michael Sheehan and Elizabeth Tibbets’s work on paper wasps, had not been found in invertebrates (Sheehan and Tibbetts, 2011; Sheehan et al., 2014a,b).
To test for facial learning, Sheehan and Tibbets took wasps and exposed them to the facial images of two other (stimulus) wasps. One of the pictures was paired with an electric shock, the other was not. They tested whether the wasps learned to avoid the facial image
associated with the electric shock, and how quickly they learned. Results indicated that wasps were able to pair a specific facial image with the electric shock (recently candidate genes associated with this ability have been identified: Berens et al., 2016).
One clue that the wasps were truly exhibiting facial learning, and not some general ability to learn, was that when the pictures of the stimulus wasps lacked antennae, or the faces on these pictures had been artificially rearranged, wasps were not capable of pairing one image with an electric shock—only intact faces produced learning. What’s more, wasps were not capable of pairing basic geometric patterns with an electric shock, again strongly suggesting facial learning.
Sheehan and Tibbets wanted to understand the selective forces that might have shaped facial learning in these wasps. They hypothesized that if this sort of facial learning had been selected in paper wasps
because it allows them to recognize individuals in their colony, and if such recognition has effects on reproduction, then in other wasp species that live a more solitary lifestyle and lack specific markings on
their face, facial recognition may be absent.
To test this idea, Sheehan and Tibbets ran experiments like those described above, but this time using Polistes metricus, a species in which individuals typically nest alone, and in which individuals have much less facial pattern variability. No facial learning was observed in this species. While a two-species comparison is far from definitive in terms of what it suggests about how
natural selection has operated on facial learning abilities, it is a starting point to follow-up comparisons among many more species.
Molecular Genetics and Endocrinology of Learning
• Molecular Genetics of Learning in Rats
• Endocrinology of Learning in Rats
MOLECULAR GENETICS OF LEARNING IN RATS
Work on learning has attempted to tie together long-term breeding experiments in rats with the molecular underpinnings of various types of learning. Zhang and his colleagues examined the molecular genetics
of avoidance learning in two lines of rats that have been selectively bred for over forty years (F. R. Brush, 2003; F. R. Brush et al., 1999; Zhang et al., 2005).
The two lines of rats—known as the Syracuse High Avoidance (SHA) line and the Syracuse Low Avoidance (SLA) line—are descended from a single, large population of rats founded in 1965. Individuals in each
generation were tested on their tendency to avoid auditory and visual cues associated with a foot shock. Although the details varied slightly over time, in the basic protocol, a rat was placed in a cage with two
compartments and could move freely between these compartments.
Just before a series of foot shocks was delivered to the compartment that the rat was in, a light and a tone were set off. A rat underwent ten training sessions in such a protocol and was then tested in sixty “avoidance” trials to see how often it would move to the other compartment once the sound and tone were presented. In each generation, those rats that were best at avoiding shock (SHA) were bred with one another, and those that were poorest at such an avoidance task (SLA) were bred with one another.
Over the course of more than forty years of selective breeding, SHA animals eventually avoided shocks in forty of sixty trials (on average), while the SLA rats typically displayed such avoidance learning in none
of the sixty trials. Other work had shown that the SHA and SLA rats were equally active in their normal daily routines and that they did not differ in their ability to detect shock or the visual and auditory cues used
during the experiments.
But these two strains of rats did differ in “fearlessness,” such that the SLA rats showed much higher levels of anxiety in a series of experiments that were separate from the avoidance learning trials described above. It seems that SLA rats were both anxious and poor at learning to avoid unpleasant cues (shocks), while the converse held true for SHA rats. Further support for this association comes from studies that demonstrate that rats that were administered drugs that are known to reduce anxiety became better at avoidance learning (Fernandez-Teruel et al., 2002; Pereira et al., 1989; Sansone, 1975).
To better understand the molecular underpinnings of the learning differences between the SHA and SLA lines of rats, Zhang and his colleagues examined gene expression patterns in the hippocampus— an area of the brain known to be important in avoidance learning as well as anxiety (Zhang et al., 2005). They ran rats from both lines through ten training sessions and then through sixty avoidance learning trials. Then they selected SHA rats that showed avoidance learning in 70 percent (or more) of their trials and SLA rats that displayed such avoidance learning in less than 10 percent of their trials.
After the learning trials, the researchers removed the hippocampus and measured gene expression in each rat. Initially sifting through gene expression in 7,500 genes, and correcting for statistical problems
associated with sampling expression patterns in so many genes, Zhang and his team were able to distill their system down to eight candidate genes that were differentially expressed in the SLA and SHA rat lines.
Four of these genes—Veli1, SLC3a1, Ptpro, and Ykt6p—showed greater expression in the hippocampus of SHA rats, while four others— SLC6A4/5HTT, Aldh1a4, Id3a, and Cd74—were expressed in greater quantities in the brains of SLA rats. From these results, Zhang and his
colleagues argue that complex traits like avoidance learning may be controlled by many genes, each of which contributes a small amount to phenotypic expression (similar sorts of findings have been found when comparing other strains of laboratory rats that have been selected for high and low anxiety behavior; Sabariego et al. 2011, 2013; Diaz-Moran et al., 2013).
Exactly how the differences in gene expression that Zhang and his team found translate into different behavioral phenotypes associated with avoidance learning per se is not yet understood.
ENDOCRINOLOGY OF LEARNING IN RATS
Let’s now review a study in which another proximate factor associated with learning in rats has been examined. Glucocorticoids such as corticosterone are hormones that play a large role in the stress
responses and learning of many animals (de Kloet et al., 1999). Glucocorticoids can cross the “blood-brain” boundary and enter the brain, where they can affect emotional state and cognitive abilities: when pregnant female rats are stressed and glucocorticoid levels rise,
the offspring of such females show high levels of anxiety and perform suboptimally in learning tests (Lemaire et al., 2000; Weinstock, 1997).
Glucocorticoids bind to receptors in the hippocampal section of the brain, including the mineralocorticoid receptor (MR) (de Kloet et al., 1998). To better understand the relationship between glucocorticoids,
stress, and learning, Ana Herrero and her colleagues administered a series of behavioral tests to a group of rats and then measured the level of various hormones. Their results are correlational rather than causal, but they shed light on learning and stress (Herrero et al., 2006).
Herrero and her team first exposed a group of rats to tests designed to examine the fear response: one test involved rats moving through a maze, and the second involved measuring how rats respond to large open fields. Rats fear open environments—environments lacking cover —and both of these tests involve placing a rat in an open environment and examining its stress response to that environment. The researchers assessed the rat’s stress by measuring such variables as the amount of time the rat spent frozen (unmoving) and the rate at which the rat defecated.
The higher these values, the greater the stress
and anxiety attributed to the animal. Some rats were exposed to only one fear test, and others were exposed to both fear tests. Rats exposed to both tests were consistent in that they were either relatively
anxious in both or not anxious in either. Once the researchers had established that rats are consistent in their response to open environments, they ran 140 rats through one fear test, and based on the rats’ behavioral responses, they classified them as either “high-anxiety” or “low-anxiety.”
After this fear test, rats were tested on their spatial learning skills. This involved placing the rats in a
water maze and measuring their abilities to find and remember the location of a submerged escape platform on which they could rest. Herrero’s team measured either the rats’ plasma corticosterone levels
—a rough measure of the amount of corticosterone circulating in their blood—and the number of mineralocorticoid receptors in their hippocampus.
Although all animals—both those classified as high- and low-anxiety —eventually learned to swim to the submerged platforms, high-anxiety individuals took significantly longer than low-anxiety animals to learn to
do so, mostly because high-anxiety animals spent more time swimming close to the edge of the water tank (Figure 5.29). When blood corticosteroid levels were measured in rats that had been run through
the water maze, Herrero found that high-anxiety animals had higher corticosterone levels than did low-anxiety animals (see chapter 3).
In addition, high-anxiety animals had fewer mineralocorticoid receptors in their hippocampus: having fewer mineralocorticoid receptors results in
a reduced ability to bind corticosterone, and indirectly leads to an increase in circulating stress hormones. It is not clear what the cause-and-effect relationship is in Herrero’s studies. It may be that rats with few mineralocorticoid receptors and high-circulating corticosterone became anxious in the open environment tests, and then scored poorly in the water maze test.
Or it could be that the open environment tests caused a change in the availability of mineralocorticoid receptors and circulating corticosterone, and animals with increased circulating corticosterone and decreased availability of mineralocorticoid receptors then did poorly on the water maze test. Further experiments are needed to decipher a cause-and effect relationship, but the work of Herrero and her colleagues suggests
a link between stress hormones, stress hormone receptors, anxiety, and learning.
