Introduction

Cognition is defined as the processes by which animals collect, retain, and use information from their environment (Shettleworth, 2010). These processes include perception, learning, memory, and decision-making. These processes can thus play a major role in driving behavior, as most animals must navigate their environment to learn of vital resources, store the locations, and recall where in the environment those resources are for later use (Jander, 1975). Therefore, spatial learning and memory are ecologically relevant cognitive traits with implications for fitness (Domjan & Galef, 1983; Dukas, 2018; Morand-Ferron et al., 2016).

Spatial cognition research has historically focused on vertebrate taxa that exhibit food-hoarding behavior. Although research in this area has been illuminating (see Shettleworth, 2010, Ch. 8 for extensive discussion of spatial cognition research), it is vital that we better understand the cognitive capacities of non-hoarders. It is also vital that we better understand the cognitive capacities of invertebrates, with insects being the obvious choice due to their ease of care and diversity. Doing so will help us address the degree of variation in ecologically relevant cognitive processes with implications for fitness across animal taxa.

Despite early pioneering work by Tinbergen (1972), insects historically were thought to display only the simplest forms of learning (i.e., habituation). More recent experimental studies have revealed sophisticated cognition across both eusocial and non-social insects (Perry et al., 2017). Insects can utilize tools (e.g., Loukola et al., 2017), engage in spatial cognition (e.g., Collett et al., 2013), and learn socially (e.g., Dyer, 2002; Lancet & Dukas, 2012), all phenomena that were once thought to be the exclusive domain of much larger-brained animals. Moreover, early evidence suggests insect learning and memory has adaptive value (Raine & Chittka, 2008; but see Evans et al., 2017). For example, antlions (Family: Myrmeleontidae) provided with the opportunity for associative learning shortened their vulnerable larval stage and pupated sooner, providing important fitness benefits (Hollis et al., 2011). Learning ability has also been shown to evolve rapidly in laboratory-reared fruit flies (Family: Drosophilidae; Mery & Kawecki, 2003).

Here we focus on quantifying cognitive capacities and processes in a non-hoarding invertebrate species, the Texas field cricket, Gryllus texensis. True crickets (Family: Gryllidae) are an ideal system for studying aspects of cognition owing to their ease of care, suitability for breeding programs, simplicity regarding testing, and extensive knowledge of their neuroanatomy, physiology, and behavior (Horch et al., 2017). Gryllidae also exhibit a worldwide distribution with some species spanning broad geographic ranges, and the family contains approximately 2400 closely related species, some of which occupy similar niches (Capinera et al., 2004; Marshall, 2017; Resh & Carde, 2009). Over the last couple of decades, several studies have investigated spatial and other forms of learning in field cricket species. For instance, Gryllus bimaculatus exhibit olfactory learning and both short- and long-term olfactory memory (Matsumoto et al., 2003, 2018; Matsumoto & Mizunami, 2000, 2005, 2006). Field crickets can also learn and remember place information in both food and non-food motivated tasks. Mediterranean field crickets (G. bimaculatus) can learn to associate the location of a hidden cool spot when released in an arena with a floor heated to an aversive temperature (Wessnitzer et al., 2008). Further, our previous research reveals that female Texas field crickets (G. texensis) can learn the location of a reward in a radial arm maze paradigm and that females exhibit consistent individual differences in place learning (Doria et al., 2019). Together, these results position field crickets as a prime study system for understanding cognitive evolution associated with spatial learning.

Although Doria et al. (2019) demonstrated the ability of female field crickets to learn the location of a food reward, the study design did not allow determination of which salient cues the crickets used to learn. Organisms can learn the location of a reward through egocentric navigation, a highly localized cue (i.e., a beacon), proximal spatial cues, or distal spatial cues (i.e., landmarks) (see Shettleworth, 2010, Ch. 8, for a discussion of spatial cues). Given the important role that cues play in learning associations, our current study investigates the strategies and cues crickets use to learn places. Maze rotations enabled us to investigate whether crickets use intra-maze cues (e.g., geometrical shapes and colors in the maze) or extra-maze cues (e.g., cues located in the room around the outside of the maze) when place learning. By rotating the maze 180°, we intentionally introduced a visual cue conflict. This cue conflict enabled us to investigate the visuospatial learning of crickets, as combinations of cues may result in enhancement, overshadowing, or blocking and thus may impact learning (Boddez et al., 2014; Gibson & Shettleworth, 2005; Pavlov, 1927). We predict that if crickets place learn using extra-maze cues (i.e., cues outside the immediate maze), they will use the arm in the same location as the previously baited arm after the 180° maze rotation. In contrast, if crickets place learn using intra-maze cues (i.e., cues located within the maze), then they will use the arm that has the same intra-maze local cue, now located at 180° opposite the original location. If crickets use both intra- and extra-maze cues, they may be unable to find the original location of the food reward.

The value of our study design is that relatively few studies have investigated how cue conflict and overshadowing inform place learning in insects. To further investigate field crickets’ place learning and memory, we used a large sample size (N = 140) of G. texensis male and female crickets. Our design also enabled us to determine whether the sexes differ in their place learning and cue use.

Methods

Study subjects, housing, and care

All crickets were descendants from adult Texas field crickets (Gryllus texensis) collected from Smithville, TX, USA (30°0′31″N, 97°9′34″W) in 2014. Crickets were housed in a greenhouse on a 14:10 h light:dark cycle at 28 ± 2 °C. They were reared in plastic containers (l × w × h = 12.5 × 23.5 × 23.5 cm) in groups of 30. Groups were fed standard cricket food (Teklad Rodent diet no. 8604, Envigo, Madison, WI, USA) and water ad libitum, with cardboard egg cartons for shelter. We removed newly penultimate stage individuals from their rearing bin three times weekly by visually inspecting crickets to determine moulting stages and isolating individuals before sexual maturity. We weighed these individuals using an electronic balance (OHAUS Pioneer Analytical Scale), then transferred them into individual circular housing containers (l × w × h = 11.5 × 11.5 × 7.5 cm), with a water vial, ad libitum standard cricket food, a crumpled sheet of paper towel for shelter, and a screened lid for ventilation.

Individuals were 7 ± 3 days post imaginal moult at the start of the experiment. Forty-eight hours prior to the experiment, we deprived individuals from ad libitum food to ensure crickets were hunger motivated for the remaining experimental procedures. Instead, in the two days preceding the experiment, we provided each individual with two pieces of Royal Gala apple (l × w × h = 4 × 2 × 2 mm) coated in standard cricket food (hereafter “reward”) in their housing container. Our approach also ensured that individuals were familiar with the food reward used in the upcoming learning trials.

Place learning maze

We conducted our experiment from March 2019 to August 2019, generally following the protocol established by Doria et al. (2019) in which crickets were assayed for place learning using a single arm baited radial arm maze (RAM). Each cricket was randomly assigned to one of ten RAMs built out of white plastic (CANUS Inc., 2684 Fenton Rd, Ottawa, ON K1T 3T7, Canada). Each RAM consisted of six arms (l × w × h = 15 × 5 × 6 cm), each radiating from a circular central platform (d × h = 15 × 6 cm) in a star shape. Each arm ended in a differently colored dead end (green, yellow, blue, orange, cream, and red), each with a unique contrasting black shape to provide intra-maze cues (see Fig. 1). Each end housed a metal dish flush with the floor of the RAM. A baited arm was randomly assigned to each cricket and this dish was baited with a piece of the reward for that individual's training trials. During each phase, every time a cricket was removed from the RAM, the RAM was cleaned with a paper towel soaked in 95% ethanol to remove any odor cues.

Fig. 1
figure 1

One of ten identical radial arm mazes (RAM) used in the present study

Full size image

Until the rotation phase (described below under Rotation probe), RAM position and location were constant such that the orientation of the baited arm was always in precisely the same location with regards to extra-maze cues. Extra-maze cues were provided in the form of colored tablecloths with different patterns and similar colors to the intra-maze cues, and the testing room furniture that remained in precisely the same place throughout the testing procedure (see Fig. S1, Online Supplementary Material (OSM)).

Discovery phase

To habituate individuals to the RAM, a discovery phase was conducted three days prior to the initial learning phase. Individual crickets were placed in their assigned novel RAM for the first time under a transparent cup. Following a 5-min acclimation period, the cup was carefully raised via an attached string. During this discovery phase, 12 pieces (l × w × h = 2 × 1 × 1 mm) of the apple coated in cricket food were haphazardly scattered throughout the arms and central platform to encourage exploration of the RAM. Individuals were given 62–64 h to explore the maze, after which crickets were returned to their individual housing with water (no food). Conducting familiarization trials is common practice in cognitive tests to reduce the potential for novelty to confound performance (Morand-Ferron et al., 2016).

Learning and initial probe

All individuals were trained on four consecutive initial learning trials (Trials 1–4). Individuals experienced a single trial each day approximately 24 h apart during the morning hours (between 9:00 a.m.–12:00 p.m.) to control for the effect of time (times when Gryllus crickets are relatively active; French & Cade, 1987). One randomly determined metal dish was baited for each individual and this rewarded arm remained constant for all four of the individual’s training trials. Individuals were given 5 min to acclimate under the cup. The cup was pulled up from outside the maze by a string. The 30-min trial began once the individual left the acclimation zone (the area under the cup). We video recorded each training trial from above for later scoring (see Scoring videos below). Individuals were returned to their housing container following each learning trial and provided with a single piece of the food reward.

Following the four consecutive days of training trials, we completed an initial probe trial (Trial 5). The initial probe was conducted in the same manner as the training trials, except that the previously baited food dish no longer contained a food reward. Our goal for this initial probe was to determine whether our measures of cognition changed now that there were no food rewards or food cues present in the dish that originally contained the food reward. Following the initial probe, individuals were returned to their housing container and provided with two pieces of the reward.

Approximately 70 h following the completion of Trial 5, crickets were returned to the same RAM previously used for Trials 1–5 for an additional three trials, each separated by approximately 24 h. The first two of these trials were relearning trials (Trials 6–7), which were identical to the initial learning trials (Trials 1–4) with the food reward placed into the initial baited arm and the arm remaining in the same spatial location. Following the absence of a reward in Trial 5, we conducted these relearning trials to ensure that the crickets relearned that the baited arm contained a food reward.

Rotation probe

To determine whether crickets rely solely on intra-maze or extra-maze visuospatial cues, we conducted a rotation phase. On the last trial (rotation probe, Trial 8), each maze was rotated 180° clockwise from its original position, thus changing the relationship between the baited arm and extra-maze cues, while maintaining the same intra-maze cues. Rotation did not impact the placement of the RAM within the testing room. Besides the 180° rotation, the rotation probe (Trial 8) was otherwise identical to the initial probe that was conducted in Trial 5.

Scoring videos

Videos were scored using Boris (v7.9.8; Friard & Gamba, 2016) by two observers (F.J.B. and D.T.). For all trials, we measured the (1) latency to emerge from under the acclimation cup, (2) time spent in thigmotaxis (moving while in contact with the wall), and (3) time spent trying to climb walls. To quantify performance on learning and relearning trials (Trials 1–4, 6 and 7) and the probes (Trials 5 and 8), we quantified (1) latency to locate the reward dish after leaving the acclimation zone, (2) the total number of complete arm entries (“visits”) made prior to locating the reward dish (i.e., total number of errors), and (3) the number of unique arm entries made prior to locating the reward dish (i.e., the number of unique errors). For the initial probe (Trial 5), we also quantified the amount of time spent in the rewarded arm relative to all other arms. For the rotation probe (Trial 8), we quantified the amount of time spent in the spatially correct arm based on the extra-maze cues, and the amount of time spent in the spatially correct arm based on the intra-maze cues, relative to the amount of time spent in all arms. For all training trials, video scoring stopped after the individual located the reward dish, or when 30 min had elapsed since the individual had left the acclimation zone, whichever happened first. For all probes (Trial 5 and 8), the video scoring stopped when 30 min had elapsed from the time the individual left the acclimation zone. Trials in which individuals did not find the reward dish were coded as missing data (42/964 = 4.4%).

Statistical analysis

Learning and initial probe

Statistical analyses were performed using the R software (version 3.3.1; R Core Team, 2018). We first used linear mixed models (LMMs) to assess performance in 140 crickets over all six learning trials (Trials 1–4, 6 and 7). Models were fitted using the lme4 package (Bates et al., 2015). LMMs were computed for (1) latency to complete the task, (2) total number of arms visited to complete the task, and (3) number of unique arms visited to complete the task, which represented our three measures of performance. Mixed models were run separately for each of the three response variables. Each model included the trial number as a predictive variable to estimate learning. In addition to this predictor of interest, we included time spent in thigmotaxis and time spent climbing as fixed effects. Sex and mass were included in early versions of these models but because neither term reached statistical significance and the addition of these terms created convergence issues, we removed both for the final models (results of models including sex are included in the OSM, Table S1). A quadratic term for trial was initially tested for each response variable but was never significant and so was removed in the final models. The random structure of each model included cricket ID in addition to the rewarded arm ID (1–6) nested within the RAM ID (1–10) as crickets were tested in different mazes and with different baited arms. Conformity of the models to assumptions of independence, homoscedasticity, and normality was assessed through visual inspection of residuals. All continuous variables were standardized by mean centering and dividing by two standard deviations (Gelman, 2008). Standardization facilitated model convergence and interpretation of estimates (Schielzeth, 2010).

We then compared Trials 4 and 5 on the same three performance measures using LMMs with similar structure as described above to determine if individuals were able to perform the task without the presence of a reward. In this model, we also tested for an interaction between sex and trial number.

Following these analyses, we were specifically interested in the initial probe (Trial 5). Using one-sample t-tests, we compared (1) the proportion of time spent in the “correct” arm relative to all other arms, and (2) the proportion of visits to the “correct” arm to what would be expected if individuals visited arms at random (i.e., 1/6 = 0.167).

Rotation probe

Using one-sample t-tests, we compared (1) the proportion of time spent in the correct arm based on extra-maze cues, (2) the proportion of visits to the correct arm based on intra-maze cues, and (3) the proportion of visits to the correct arm, based on extra-maze cues to what would be expected if individuals visited arms at random.

To further investigate the rotation results, we re-ran the one-sample t-tests restricted to crickets that displayed negative regression slopes during the learning trials across all performance measures.

Results

Sex differences

We found no differences between males and females in any aspect of their learning, in the initial probe, or in the rotation probe (results of models including sex are included in the OSM, Table S1). Because sex was not statistically significant, we removed it from the final models that we present below.

Learning and initial probe

Crickets improved their accuracy significantly over successive learning trials, showing decreases in both the number of total arms visited (Table 1; Fig. 2B) and number of unique arms visited (Table 1; Fig. 2C) prior to finding the reward during Trials 1–4 and Trials 6 and 7. Thigmotaxis and wall climbing were significant covariates for both performance measures (Table 1). Conversely, the latency to find the reward did not significantly decrease over the learning trials (Table 1; Fig. 2A). Thigmotaxis was a significant positive covariate, but wall climbing was not (Table 1).

Table 1 Linear mixed model results for the three models testing the effect of learning trials (Trials 1–4 and 6 and 7) on different measures of performance (latency to complete the task, number of arms visited to complete the task, and number of unique arms visited to complete the task). Thigmotaxis and climbing were used as covariates, and each model used cricket ID and baited arm ID nested within radial arm maze ID as the random structure. N = 669 trials for all models. * Estimate was unobtainable owing to singularity issues (i.e., it had no effect on the model)
Full size table
Fig. 2
figure 2

Learning curves (Trials 1–4 and 6 and 7) for three performance measures over training trials. Points represent the median. Violins represent the distribution of data

Full size image

Compared to Trial 4, there was no statistically significant difference during the initial probe (Trial 5) when the reward was absent for the total number of arms visited, number of unique arms visited prior to finding the reward, or the latency to find the reward (Table 2). Both thigmotaxis and climbing were significant covariates in all three models (Table 2).

Table 2 Linear mixed model results for the three models comparing trials 4 (learning) and 5 (probe). Thigmotaxis and climbing were used as covariates, and each model used cricket ID and baited arm ID nested within radial arm maze ID as the random structure (see Table S1 (OSM) for models including sex). N = 224 trials for all models. * Estimate was unobtainable owing to singularity issues (i.e., it had no effect at all on the model)
Full size table

There was a non-significant trend for a difference between the proportion of time spent in the correct arm and chance (1/6 = 0.17) during the initial probe trial (t123 = 1.84, p = 0.07; mean = 0.20, 95% CI: 0.16–0.24; Fig. 3), but the proportion of visits to the correct arm (i.e., the previously rewarded arm) revealed a significant deviation from chance (t123 = 2.67, p = 0.01; mean = 0.21, 95% CI: 0.18–0.24; Fig. 3).

Fig. 3
figure 3

Proportion of arm entries and proportion of time spent in the previously baited arm during the probe trial (Trial 5) compared to what would be expected at random (dashed line). Displayed are means (triangle) ± 95% CI. *p < 0.05

Full size image

Rotation probe

During the rotation probe trial, neither the proportion of time spent in, nor the proportion of visits to the spatially correct location based on extra-maze cues differed significantly from chance (t108 = -1.61, p = 0.11; mean = 0.14, 95% CI: 0.18–0.24 and t108 = -0.07, p = 0.94; mean = 0.17, 95% CI: 0.14–0.19, respectively; Fig. 4). Similarly, neither proportion of time spent in, nor the proportion of visits to the correct arm based on intra-maze cues differed significantly from chance (t108 = 0.82, p = 0.42; mean = 0.18, 95% CI: 0.15–0.22 and t108 = 1.22, p = 0.22; mean = 0.18, 95% CI: 0.16–0.21, respectively; Fig. 4). Restricting these analyses to only those individuals that showed negative regression slopes across performance measures did not qualitatively change these results (see OSM results, Figs. S2 and S3).

Fig. 4
figure 4

Proportion of entries and proportion of time spent in the correct arm based on intra-maze cues (circles) and correct arm based on extra-maze cues (triangles) during the rotation probe (Trial 8) compared to what would be expected at random (dashed line). Displayed are means (triangle or circle) ± 95% CI

Full size image

Discussion

We found that male and female cricket performance improved over successive learning trials as evidenced by a decrease in the total and unique number of arms visited, and that crickets performed similarly when tested on a probe where the reward was not present. We also found a significant negative effect of thigmotaxis on performance, and in some cases, wall climbing. Unlike Doria et al. (2019), we did not find a significant decrease in latency to find the baited arm over trials. We rotated the RAMs to put intra-maze and extra-maze cues in conflict, and then probed cricket performance for a second time (Trial 8) without a reward present. Intriguingly, cricket performance was near chance during the rotation probe, regardless of whether we considered the correct arm based on intra-maze cues or the correct arm based on extra-maze cues. We found no significant effect of sex on any performance measures.

Over learning trials, individual male and female crickets improved their performance by reducing the total and unique number of arms visited before finding the reward. However, unlike Doria et al. (2019), this improvement was not significant for latency to contact the reward. Given the frequency with which latency to find or access a reward is used in cognitive assays, it is important to consider the disparate findings between Doria et al. (2019) and our study. The relative utility of latency measures compared to accuracy measures of performance has been questioned previously within the context of potential confounds and speed-accuracy trade-offs (e.g., Chittka et al., 2009; Morand-Ferron et al., 2016). Rowe and Healy (2014) also suggest that individuals that are faster at finding the reward may associate the cue with the reward more rapidly, while individuals that are slower, but make the same (or fewer) errors, may be learning more about the actual cues themselves and putting the cues in the context of prior experience. Perhaps most telling, the cross-context repeatability of latency measures appears to be quite variable when considering a range of cognitive tasks, further strengthening the argument that latency measures are highly susceptible to shifts in protocol when purportedly measuring the same cognitive domain (Cauchoix et al., 2018). This may help explain the incongruence between our present results and the previous results. Despite the theoretical issues with latency, it is feasible that latency differences between Doria et al. (2019) and the present study are attributable to variation in rearing, breeding, housing, and social conditions, as well as subtle differences in the maze construction. However, we feel this explanation is unlikely, given that differences between the studies are minute. If these little differences resulted in variation in performance, it would call into question most replicated studies of spatial cognition that don’t find identical results.

In previously tested female G. texensis, thigmotaxis, or wall-hugging behavior, accounted for 20–30% of variation in cognitive performance, and thigmotaxis did not decrease with maze familiarity (Doria et al., 2019). The effect of thigmotaxis on RAM performance in these female crickets parallels results for rats tested in a Morris water maze paradigm (Harris et al., 2009). Sex differences in cognition in rodents has been linked to thigmotaxis, with female rodents generally exhibiting greater thigmotaxic behavior and therefore, worse performance on place learning tasks, compared to males (e.g., Beiko et al., 2004). Thigmotaxic behavior is often considered an anxiety behavior in mammals (e.g., Harris et al., 2009; Kallai et al., 2007), but thigmotaxis is also pervasive in diverse insect species (e.g., Besson & Martin, 2005; Domingue et al., 2021; Doria et al., 2019; Kastberger, 1982; Laurent Salazar et al., 2018; Mosquera & Lorenzo, 2020; Nestel et al., 1995). Like rats, crickets are susceptible to depredation and have evolved antipredator behaviors like avoiding open spaces and light (e.g., Bailey & Haythornthwaite, 1998; Niemelä et al., 2012), and thigmotaxis may represent one such antipredator behavior. However, in the present study, there was no significant difference in thigmotaxis between male and female field crickets. Doria et al. (2019) did not find that thigmotaxis decreased over trials or with familiarity with the maze, questioning its interpretation as an expression of anxiety. Additional research is needed to understand whether there is any adaptive value of thigmotaxis in wild crickets, whether this behavior represents antipredator behavior, whether it represents a reaction to subtle differences in light levels within the maze, and whether thigmotaxis is negatively impacting cognitive performance in other domains or test contexts.

The consistency with which male and female crickets performed in the initial probe (Trial 5) compared to the final learning trial (Trial 4), and the above-chance performance during the initial probe suggest that males and females are attending to some relevant visual cues provided when place learning, and not simply relying on olfactory or visual cues provided by the reward itself. Despite the absence of the olfactory cues from the reward, it is important to consider other cue sources. Given that mazes were wiped clean of olfactory cues between every trial, we can reject the use of olfactory cues left in the maze by a cricket. Crickets could have used within-trial self-generated olfactory cues. However, we found that most crickets that learned the rewarded location during the initial probe failed to find the rewarded location (based on both intra- and extra-maze cues) during the rotation. Another source of cues potentially available to crickets is magnetic compass cues. To our knowledge, there is currently no research on magnetic cue-use for orientation in crickets, but magnetic cues could have been available to crickets. However, use of magnetic cues would have resulted in crickets going to the extra-maze spatial location during the rotation since these cues would have been stable during all learning, the initial probe, and the rotation.

This leaves visual cues as the likely explanation for place learning. Excluding our intended intra- and extra-maze cues, a further source of visual cues that are potentially available to crickets is subtle differences in lighting within the maze. We attempted to ensure that lighting was uniform, but some shadows and light spots could have existed. Regardless, we statistically controlled for lighting, as each maze had consistent lighting across trials and was therefore included in the RAM ID random term. Additionally, use of intra-maze lighting would have been a subset of intra-maze cues and would have resulted in crickets finding the intra-maze spatial location during the rotation, if they were attending to this and solely using intra-maze cues.

Our results are consistent with our understanding of visual cue use in cricket orientation behavior (Kral, 2019) and consistent with other studies on cricket visuospatial place learning (e.g., Doria et al., 2019; Hale & Bailey, 2004; Wessnitzer et al., 2008). However, the initial probe does not allow us to reject the possibility that crickets could have learned an association with a single cue, for instance the intra-maze visual cue located right above the reward, and does not allow comparing the relative saliency of intra- versus extra-maze cues to the crickets. The RAM rotation allowed further exploring the nature of the cues used by crickets in this task.

Crickets did not perform above chance levels when the RAM was rotated relative to its surroundings, which suggests crickets are not relying on a one-cue intra-maze association. The decreased performance of crickets during the rotation probe compared to the initial probe suggests that crickets learned intra-maze and extra-maze cues in a combinatory manner, such that when previously reliable cues were put in conflict, crickets were unable to accurately attend to either of the previously reliable cue sets. Like our crickets, Japanese eels (Anguilla japonica) perform better when intra- and extra-maze cues are available in concert, but unlike the crickets, eels successfully found a place reward when only one cue is present on test trials (Watanabe, 2020). These latter results in crickets are at odds with results for humans, rodents, birds, fruit flies, and bees, which typically show overshadowing of one cue type over another when one of the cue sets is unreliable. Distal and extra-maze cues often overshadow proximal and intra-maze cues in humans, rodents, and flightless Drosophila (e.g., Chamizo, 2003; Diez-Chamizo et al., 1985; Foucaud et al., 2010; Hébert et al., 2017; Spetch, 1995), while proximal visual cues appear to overshadow distal visual cues in some flying animals, like birds and bees (e.g., Bennett, 1993; Cheng et al., 1987; Spetch, 1995). In any event, results from other animals suggest that one type of cue or the other should have been more salient to the crickets. Instead, our results appear more consistent with the reciprocal overshadowing and decreased performance that is seen in rodents presented with conflicting multi-modal cues (e.g., March et al., 1992; Sánchez-Moreno et al., 1999). Since rats tested in a similar apparatus with slightly different protocols exhibited different patterns of overshadowing (Diez-Chamizo et al., 1985; March et al., 1992; but see Introduction in Hébert et al., 2017), it appears that slight differences in the type and shape of the maze, cues presented, modalities involved, or even reinforcement type (e.g., food vs. escape) may all impact the degree to which cues conflict.

In one study, crickets performed better with naturalistic cues compared to artificial cues in an escape-reinforced maze (Wessnitzer et al., 2008). In the present study, the intra-maze cues were high contrast abstract shapes and vibrant colors, while the extra-maze cues were patterned fabrics with colors. Additionally, some colors (e.g., red and orange) may have contrasted less compared to other colors (e.g., blue and green), as cricket color vision allows crickets to see in ultraviolet, blue, and green areas of the spectrum (Horch et al., 2017). It is possible that the saliency and unnatural nature of the cues required crickets to attend to both intra- and extra-maze cues to accurately perform; in the presence of conflict, one cue set was not sufficient. Future studies would benefit from ensuring that all cues are of comparable visibility and naturalistic to address whether these changes increase cue recognition when two cue sets are in conflict. Dukas (2002) even suggests that cryptic cues increase attention and, therefore, visual perception, indicating that cryptic naturalistic cues may serve better than high-contrast cues.

Overall, both male and female crickets can use a combination of visual cues to learn the location of a food reward in the absence of odor cues, but in the present RAM paradigm both intra-maze and extra-maze cues were necessary for increased performance. When cues conflicted, crickets were not accurate at finding the previously baited location, suggesting that: (1) they do not rely on a single-cue association to navigate the maze; (2) they do not rely solely on a motor strategy based on route-following and self-motion cues; (3) they do not rely on their own scent cues to avoid previously visited arms, as none of these strategies should have been disrupted by rotation of the RAM (Paul et al., 2009). As central place foragers (Gangwere, 1961) with well-developed ommatidia and associated visual capacity (Briscoe & Chittka, 2001; Horch et al., 2017), attending to an array of relevant visual cues should be advantageous in locating food and the shelter of the burrow, possibly explaining the crickets’ poor performance when the two sets of cues contrasted. Future research should assess the robustness of this finding using a different maze type and/or more natural cues instead of artificial ones, and should focus on the ability of crickets to use just one set of cues (whether intra- or extra-maze).

Domjan and Garlef (1983) argues that the comparative approach is crucial for bridging psychological and biological research. In recent decades, we have made strides in comparative cognition, but invertebrate cognition research has lagged vertebrate research. We believe the diverse ecology and distribution of Gryllidae crickets make this taxon a promising insect system to rectify this. Our study sets the stage for future laboratory- and field-based studies on the nature of visuo-spatial memory, the degree of individual variation in spatial learning and memory, the comparative relevance of spatial learning in cricket species with different ecologies, and the fitness effects of this variation in field crickets.