Honey bees rely on associative stimulus strength after training on an olfactory transitive inference task

Introduction

Research on animal cognition has shown that some species can rank events based on individual experience (Acuna et al., 2002; Bond et al., 2003). This ability to order events is essential for survival. For instance, in a foraging context, animals can improve efficiency by ranking food items according to factors such as nutritional value, abundance, and other relevant criteria (Davis, 1992). Similarly, in social contexts, hierarchies and dominance relationships often depend on ranking among individuals (Bond et al., 2003; Paz-y-Miño et al., 2004; Grosenick et al., 2007). The ability to establish such relationships between stimuli (A > B; B > C; therefore, A > C) is known as transitive inference (Bryant and Trabasso, 1971) and is considered as one of the hallmarks of logical deductive reasoning (Vasconcelos, 2008).

Transitive inference tasks allow researchers to study logical reasoning and knowledge manipulation (Potts, 1974; Woocher et al., 1978; Acuna et al., 2002; Vasconcelos, 2008). It is demonstrated empirically by the ability to infer a relationship (B > D) between non-adjacent items from overlapping premises (A > B, B > C, C > D, D > E) of an underlying series (A > B > C > D > E). A preference for B over D in this context may be attributed to deductive reasoning (von Fersen et al., 1991; Vasconcelos, 2008), where subjects construct and manipulate a unified, linear representation of the implicit hierarchy A > B > C > D > E (Delius and Siemann, 1998; Acuna et al., 2002).

Alternatively, associative theories of transitive inference suggest that animals in this experimental design may respond based on reinforced versus non-reinforced experiences (Werner et al., 1992; Wynne et al., 1992; Siemann and Delius, 1993; Siemann and Delius, 1998; Terrace and McGonigle, 2016). According to this view, animals select stimuli based on associative strength—the number of reinforced versus non-reinforced experiences with each stimulus—rather than relying on deductive reasoning. A critical test to distinguish between these two accounts involves presenting non-adjacent stimuli B and D. If B and D were equally reinforced during training (e.g., A+ vs. B– and B+ vs. C–; C+ vs. D– and D+ vs. E–, where + and – signs indicate the presence and absence of reinforcement, respectively), they would have equivalent associative strengths, as both are equally paired with reinforcement and non-reinforcement. Consequently, subjects guided by associative strength would respond equally to B and D. However, if subjects use a mental representation of the hierarchy learned in training, they should prefer B over D, despite the equal associative strengths.

Beyond humans (Bryant and Trabasso, 1971; Delius and Siemann, 1998), various non-human species have demonstrated the capacity for transitive reasoning. For example, fish (Grosenick et al., 2007), pigeons (von Fersen et al., 1990; von Fersen et al., 1991; Siemann and Delius, 1994; Wynne, 1997), corvids (Bond et al., 2003), pinyon jays (Paz-y-Miño et al., 2004), rats (Davis, 1992; Dusek and Eichenbaum, 1997), squirrel monkeys (McGonigle and Chalmers, 1977; McGonigle and Chalmers, 1992), macaques, (Treichler and Van Tilburg, 1996) and chimpanzees (Gillan, 1981; Boysen et al., 1993) consistently prefer B over D in tests after multiple discrimination training (A+ vs. B–, B+ vs. C–, C+ vs. D–, D+ vs. E–). Transitive inference has been associated with the hippocampus (Dusek and Eichenbaum, 1997; Eichenbaum and Fortin, 2009; Devito et al., 2010), which processes and stores critical relationships among items and events, enabling the flexible use of memories in new situations.

In invertebrates, transitive inference has been studied in an operant context in two insect species—honey bees (Benard and Giurfa, 2004) and wasps (Tibbetts et al., 2019)—with contrasting results. Free-flying honey bees were trained to enter a Y-maze to discriminate between five distinct black-and-white patterns arranged in four overlapping premise pairs, where one stimulus was rewarded with sucrose solution and the other was not (Benard and Giurfa, 2004). This study found no hierarchical ranking of stimuli, as tests with non-adjacent stimuli B and D showed no preference, indicating that choices were guided by the associative strength of each stimulus (Benard and Giurfa, 2004). Polistes wasps were trained with five colors arranged in four overlapping premise pairs displayed on opposite walls of a rectangular box, where one color was paired with electric shock and the other was not (Tibbetts et al., 2019). After training, unlike bees, wasps preferred B over D when tested with these non-adjacent stimuli, indicating a hierarchy of colors based on transitive inference (Tibbetts et al., 2019).

Both studies relied on operant training, raising the issue of reinforcement control, as insects in these setups could move freely. Consequently, reinforcement outcomes depended on the insect’s choices and actions. With free movement and choice, the animals may quickly learn to avoid non-reinforced alternatives, resulting in fewer non-reinforced experiences than initially planned by the experimenter. This highlights the need for precise control over the reinforcement history of each stimulus to determine whether the animal’s choices are influenced by associative factors or transitive inferences. In the case of honey bees, addressing the transitive inference problem with full control over reinforcement history is achievable using a Pavlovian conditioning protocol called olfactory conditioning of the proboscis extension response (PER) (Bitterman et al., 1983; Menzel, 1999; Giurfa, 2007; Giurfa and Sandoz, 2012). In this protocol, restrained honey bees learn to associate olfactory stimuli with a sucrose solution reward. When the antennae of a hungry bee are touched with sucrose solution, it reflexively extends its proboscis to consume the sucrose. While odors alone do not trigger this reflex in naive bees, forward pairing of an odor with sucrose creates an association, allowing the odor to elicit a PER in subsequent tests (Bitterman et al., 1983). In this protocol, the odor acts as the conditioned stimulus (CS), while the sucrose solution serves as the reinforcing unconditioned stimulus (US). Since reinforcement delivery is entirely controlled by the experimenter, the bees’ responses do not influence the learning of the odor-sucrose association (Bitterman et al., 1983). In differential conditioning, where bees must learn to discriminate between a rewarded and a non-rewarded odorant, the protocol allows for the delivery of both reinforced trials (CS+ trials) and non-reinforced trials (CS– trials), in which no reward is provided. An even more effective approach for CS– trials involves presenting the unconditioned stimulus (US) before the conditioned stimulus (CS) in a backward pairing. This method induces inhibitory learning of the CS– (Hellstern et al., 1998), leading to improved discrimination (Schleyer et al., 2018).

The PER protocol has been widely used to study various learning and discrimination tasks in the olfactory domain (see review in Giurfa and Sandoz, 2012). However, no attempts have yet been made to investigate transitive inference, despite the feasibility of conditioning premise pairs using the PER paradigm. This approach allows precise control over the number of excitatory (+) and inhibitory (−) experiences the bees have with each stimulus in the series—a level of control that is difficult to achieve in operant conditioning setups.

We trained honey bees using differential olfactory conditioning of the proboscis extension reflex (PER), in which one odor was rewarded with a sucrose solution (CS+ trials), while the other was backward paired with the sucrose solution (–CS trials; the minus sign was inverted to account henceforth for the backward US delivery). Bees were conditioned with a sequence of four premise pairs of odorants arranged into a defined hierarchy (A > B > C > D > E or A < B < C < D < E). Our aim was to determine whether honey bees could form transitive inferences or if they relied on the associative strength of the stimuli experienced under these conditions, and to explore the mechanisms underlying their responses.

Materials and methods

Subjects

Honey bee foragers (Apis mellifera) were captured as they landed on a feeder containing a 30% sucrose solution (w/w) to which they had been previously trained. The experiments were conducted during late spring and summer, when training to such a feeder is feasible. Each captured bee was placed in a small glass vial and immobilized by cooling in a freezer at −6°C for 3 to 4 min. The bees were then harnessed in small tubes, with the head protruding, allowing only the movement of the antennae and mouthparts, including the proboscis. Afterward, each bee was fed 4 μL of the 30% sucrose solution and left undisturbed in a dark box with moist filter paper for 2 h. Ten minutes before each experiment, bees were tested for intact proboscis extension reflex (PER) by lightly touching their antennae with a toothpick dipped in a 30% sucrose solution (w/w). Extension of the proboscis beyond an imaginary line between the open mandibles was counted as PER (the unconditioned response). Bees that did not show the reflex (<5%) were excluded from the experiments.

Unconditioned and conditioned stimuli

The unconditioned stimulus (US) was always a 30% sucrose solution (w/w). The conditioned stimuli (CSs) were the odorants 1-Hexanol, 2-Hexanone, Heptanal, 2-Nonanol and Eugenol (all obtained from Sigma-Aldrich, Deisenhofen, Germany), which are well differentiated in olfactory PER conditioning experiments. The choice of these odorants was based on a generalization matrix that includes four of the five odorants used (1-Hexanol, 2-Hexanone, Heptanal, and 2-Nonanol), showing low cross-generalization (Guerrieri et al., 2005). Additionally, we had preliminary data for the missing comparisons involving Eugenol, indicating low generalization.

Four microliters of each odorant were applied to a fresh strip of filter paper, which was then placed inside a 20 mL plastic syringe. During each trial, a scented airflow was directed toward the bees’ antennae by gently pressing the syringe from a distance of approximately 5 cm. An exhaust system was positioned behind the bees to remove odor-laden air.

In all three experiments, within each group, and for each bee, a specific odorant was assigned to categories A, B, C, D, and E. The odorant sequence was then shifted for each subsequent bee, ensuring that each odorant was evenly distributed across the categories.

Conditioning

Differential conditioning was used in all experiments, whereby animals learn to respond to a reinforced odorant while inhibiting responses to a non-reinforced odorant. In preliminary experiments using the same odorants, we observed a high rate of generalization between reinforced and non-reinforced odorants. To mitigate this effect, and to achieve a full control of reinforcement history, we implemented differential conditioning in which reinforced trials (CS+ trials) involved forward pairing of the CS and the US, while non-reinforced trials (–CS trials) involved backward pairing, where the US was presented before the CS to reduce responses to the –CS (Hellstern et al., 1998). The use of the backward pairing, which induces inhibitory learning of the –CS (i.e., learning that the CS is not followed by the US; Hellstern et al., 1998), was precisely aimed at reducing odor generalization.

Each trial began by placing the subject 15 cm in front of the exhaust system, where it acclimated for 15 s. In CS+ trials, the CS was presented before the US (forward conditioning): the CS began at the 15-s mark, followed by the US 2 s later. Both the CS and the US were presented for 4 s. The US was delivered by lightly touching the antennae with a toothpick dipped in sucrose solution, allowing the bee to feed for 2 s after proboscis extension. This created a 2-s interstimulus interval with a 2-s overlap between the CS and the US. In –CS trials, the US was delivered first (backward conditioning), starting at the 15-s mark, followed by the CS 2 s later. Both stimuli lasted for 4 s, with a 2-s interstimulus interval and overlap. Each trial, regardless of pairing type, concluded at the 30-s mark. After each trial, bees were returned to their resting positions. A total of 12 bees were trained per experimental run, with a 6-min interval between trials.

Testing

At the end of each conditioning phase, an intermediate retention test was conducted to verify whether the bees had learned the discrimination. These tests determined if the bees had learned not only to respond to the CS+ but also to refrain from responding to the –CS, which could not be evaluated though the learning curves, as responses during the –CS conditioning trials were not measurable (see above). During these tests, bees were presented with the two odors from the just-completed phase without any reward. After completing all five conditioning phases, the bees were additionally subjected to five final tests where they were tested on the five trained odors (A, B, C, D, E) in a random sequence without reinforcement. During intermediate and final tests, the duration of odorant delivery, and the interstimulus interval (inter-test interval) were the same as during conditioning trials.

Response measurement

For CS+ trials, we recorded whether or not a bee extended its proboscis after the onset of the CS and before US delivery, which was counted as the conditioned response. Multiple responses during a single CS were counted as one PER. For –CS trials, no response to the CS could be recorded, as bees always responded to the US presented before the CS. In this case, bees continued extending the proboscis during the overlapping –CS but this cannot be counted as a CS response. Thus, acquisition curves reflect only the variation in responses to the CS+ across trials. At the end of each experiment, all animals were retested for the proboscis extension reflex to the US; bees that did not respond (<5%) were excluded from the analyses.

Experiments

Experiment 1

Two groups of bees (Group I and Group II) were trained on a sequence of four odor discriminations involving five different odorants. Each discrimination phase employed differential conditioning, with the conditioned stimulus (CS+) and the non-conditioned stimulus (–CS) each presented six times, totaling 48 trials (Table 1).