Acute low dose caffeine affects behavior profile and activity, an examination of male rats with high or low anxiety-like behavior

Anxiety disorders affect up to one third of the population. Caffeine, an adenosine receptor antagonist, is thought to have a dose-dependent effect on anxiety. We recently showed that a high dose of caffeine (50 mg/kg) differentially affected anxiety-like behavior in rats with high or low baseline anxiety-like behavior, replicating findings using relatively high doses in human patient samples. It is not known if low doses of caffeine have similar effects. The elevated plus maze (EPM) was used to categorize male Wistar rats (13 weeks of age) into groups of high or low anxiety-like behavior. Behavior was evaluated using the multivariate concentric square field (MCSF) test and the EPM after a low 10 mg/kg dose of caffeine. Multivariate data analysis demonstrated that caffeine decreased the differences between the high and low anxiety group, whereas the separation remained for the high and low control groups. For the caffeine treated rats, univariate statistics showed an increase in parameters regarding activity in the EPM and duration in the slope of the MCSF. Regarding risk-taking, shelter-seeking, and exploratory behavior, caffeine did not affect the groups differently. In conclusion, these results demonstrate increased activity in the caffeine-treated rats, together with a potentially anxiolytic effect and increased impulsivity that did not differ between the baseline anxiety groups. In contrast to high caffeine doses, a low dose does not generally affect rats with high anxiety at baseline differently than rats with low anxiety-like behavior. Further studies are warranted to fully elucidate the effects of caffeine in anxiety.


Introduction
Anxiety is a growing problem in the world, the number of individuals seeking care is increasing each year, and both pharmacotherapy and psychotherapy are important parts of the treatment [1].However, the pharmacological treatment available today has in several cases proven to be inefficient for many individuals, and has unwanted side effects [2].Anxiety disorders consist of panic disorder, general anxiety disorder, different phobias, and social anxiety disorder, and combinations and variations of these.Due to the large variation from individual to individual the response to treatment differs in every case.
Caffeine is a psychostimulant drug that is commonly known to reduce fatigue [3] and it has long been discussed to be involved in anxiety disorders [3][4][5].It has antagonistic actions on adenosine receptors, with the highest affinity on the subunits A1 and A2A, and seem to have both anxiogenic and anxiolytic properties [6].When stimulated, the adenosine A1 receptor (A1R) is regulating synaptic transmission, and adenosine A2A receptor (A2AR) stimulates synaptic plasticity such as long term potentiation [7].Links between polymorphism in the A2AR gene and a higher sensitivity for anxiety is documented [8,9] as well as polymorphism in other genes linked to anxiety [10][11][12].Furthermore, there is an ongoing discussion regarding the A2AR as a therapeutic target in several psychiatric disorders including several aspects of anxiety [6,13].However, the role of caffeine in relation to anxiety is not fully elucidated.A correlation between trait anxiety and chronic caffeine intake (high doses of 186.4 ± 35.8 mg/day and 388.6 ± 139.5 mg/day of caffeine) is reported in women [14] and there was a correlation between increased frequency in caffeine consumption and anxiety in males [15].A recent meta-analysis by Klevebrant and Frick, demonstrated that individuals with panic disorder have a higher risk of experiencing a panic attack when consuming caffeine (>400 mg/day) [16].However, acute lower doses (10 mg/kg, 30 mg/kg and 25 mg/kg) of caffeine have previously been associated with no effect on the anxiety in adult male rats [17], an anxiolytic response in adult males, and in males more than in female rats, respectively [18,19], or stimulated locomotion [20][21][22] in adult male mice, adult wild type and A2A receptor knockout mice, and in adult rats, respectively.In Hughes et al. [19], even a high dose of 50 mg/kg gave an anxiolytic response in the elevated plus maze (EPM).However, both 25 mg/kg and 50 mg/kg of caffeine differed in their anxiolytic effect depending on surrounding parameters, brightness of the behavior test and sex of the rats.In addition, caffeine has also been reported to affect other behaviors such as decision-making and risk-taking.In adolescents and children a study reported a significant correlation between caffeine and risk-taking behavior, tested with the Iowa gambling task and the balloon analogue risk task [23], and when combined with alcohol caffeine seems to have both beneficial and nonbeneficial effects on alcohol consumption and risky behavior in young adults [24].In adult female and male rats caffeine is seen to diminish decreased risk-taking caused by the steroid nandrolone [25], and in a risky decision-making task, caffeine had no effect in adult male rats [26].
The study of anxiety in animals often utilize behavioral tests in the laboratory.One of the most used is the elevated plus maze (EPM), validated to be an animal model for anxiety using benzodiazepines [27,28].The EPM is based on the rat's natural behavior to avoid open bright areas, and keep to sheltered areas [29].The parameter open arm time is usually used as a measure for anxiety-like behavior.Other parameters such as time spent in the center have earlier been considered as a measure of decision-making [30] and impulsivity.The multivariate concentric square field (MCSF) test is developed to study rodent behavior in a multivariate setting where the animal has a free choice to visit different areas with various conditions in the apparatus [31].The test measures risk and risk assessment behavior, shelter seeking, explorative behavior, and general behavior, giving a behavior profile of the rat in order to evaluate several behaviors at once and more stimulate a normal environment for the rat, where multiple choices often is the norm.
Previous studies using the EPM have found individual differences in anxiety-like behavior, allowing categorization of rats into subgroups of high or low anxiety-like behavior [32][33][34][35][36].We recently showed that an acute high dose (50 mg/kg) of caffeine affects both behavior, such as risk-taking, and expression of certain genes differently in rats categorized as having a high or low anxiety-like behavior [37].However, if a low dose of caffeine affects rats differently depending on their baseline anxiety level has not been extensively studied.Thus, the aim of this paper was to determine how a low acute dose of caffeine (10 mg/kg, 30 min prior to behavior test) affected the behavior profile and if the effect differed between rats categorized as having a low or high anxiety-like behavior with focus on anxiety-like behavior, general activity, and risk-taking.

Animals
Male Wistar rats (n = 30), 13 weeks old, weighing 347-394 g at the beginning of the study, from Envigo (Horst, Netherlands), were put under a reversed light regimen (lights off at 07:00, lights on 19:00).The rats were placed in groups of three per cage in type VI cages, bottom area of 1820 cm 2 , with raised cage lid, a wooden house, and with pelleted food and water provided ad libitum.Temperature was controlled to 20 ± 1 • C with 40 ± 10 % humidity.Rats were allowed to acclimatize 12 days prior to being exposed to behavioral tests.During this time the rats were handled at multiple occasions by the test executioner, to reduce stress from handling during behavioral testing, as well as weighed two times a week.This study was approved by the local Uppsala animal ethics committee (Dnr 5.8.18-08450-2020).All animal handling were in accordance with Swedish rules and guidelines regarding animal experiments (Animal Welfare Act SFS1998:56) and the European communities council directive (86/609/EEC).

Behavioral testing and drug treatment
All testing were performed between 8:30-15:30 Monday to Friday.Animals were weighed two days a week, with the addition of weighing prior to injection, and transported to the behavior room using a small cage with a towel over for light protection.The rats were on experimental day 1 screened using the EPM and based on open arm time, as previously done [32], [37], the rats were divided into three groups: high open arm time (HOAT) (low anxiety-like behavior), intermediate open arm time (IMOAT) (intermediate anxiety-like behavior), and low open arm time (LOAT) (high anxiety-like behavior) and randomly assigned to either control or caffeine.The distribution of the rats within each group was based on the open arm time in relation to each other, giving the slight difference in animals per group.The IMOAT group were excluded from further analysis due to the aim of this study to compare the extremes.The open arm time was calculated as a percentage of the total time spent in the maze.After the first screening session in the EPM, the rats were left to rest from the behavioral tests for 21 days, weighing and handling were continued during these days.On day 21 after the screening session in the EPM, the rats were injected with caffeine or vehicle and exposed to the MCSF test, and on day 28 injected with caffeine or vehicle and exposed to the EPM test again.Two hours after injection the animals were euthanized by decapitation.On experimental day 21 and 28 a dose of 10 mg/kg caffeine (Caffeine sodium benzoate, 1:1; Sigma Aldrich, Darmstadt, Germany) [35,38] dissolved in 0.9 % saline were subcutaneously injected 30 min prior to behavioral testing.Sodium benzoate was added to increase caffeine's solubility.Controls received 10 mg/kg sodium benzoate (Sigma Aldrich) dissolved in 0.9 % saline, injected subcutaneously 30 min prior to behavioral testing.All doses were administered with the injection volume of 0.5 ml/kg.

EPM
The EPM was constructed of four arms, two open arms (40 × 10 cm, length x width) and two closed arms (40 × 10 × 40 cm, length x with x height) in a plus shape with the closed arms facing each other and the same for the open arms, with a 10 × 10 cm center connecting them.The EPM was elevated 50 cm from the floor and illuminated with 30 lux.Each rat was allowed to explore the EPM arena for 5 min, then taken back to its cage.The EPM was cleaned with 10 % EtOH and let dry between each rat.To avoid any first-in-line effect, an out-of-test rat was prior to each session placed in the EPM arena.The tracking software Ethovision XT version 16 (Noldus Inc., Wageningen, Netherlands) was used to track parameters using a video camera placed above the arena.Parameters that were automatically recorded were open and closed arm time (s), time in center (s) frequency to each arm and center, distance moved (cm), and velocity (cm/s).Manually scoring of head dips, rearing, grooming, and SAP (stretch attend posture) was performed.

MCSF
The MCSF test is described in detail elsewhere [31,37].Briefly, the MCSF test consists of an arena 100 cm × 100 cm, the center area is 70 × 70 cm with corridors surrounding three of the sides.The south corridor is linked to the dark corner room in one end, the other two corridors (west and north) meet in the hurdle at the next corner, the last corridor is also linked to the slope leading up to an illuminated (600 lux) bridge (Fig. 1).On experimental day 21 each rat was tested individually in the MCSF for 20 min, 30 min post injection of 10 mg/kg caffeine sodium benzoate or vehicle.Automatic tracking was enabled by Ethovision 16 (Noldus, Netherlands) using a camera placed above the test arena to record each trial.The parameters duration for each zone (Duration, s), latency to first time in each zone (Latency to first, s), frequency for each zone (Frequency), distance moved in the arena (cm), velocity in the arena (cm/s), and latency to leave the center (s) were all automatically recorded.Calculated parameters were duration per visit (Duration/-Frequency, s), total activity (sum of all zone frequencies), and Duration, Frequency, and Duration/Frequency for all corridors combined (total corridors).The risk shelter index is the ratio of time spent in risk (bridge, central circle) versus shelter areas (DCR) of the MCSF arena and were also calculated.Manually scoring of head dips, rearing, grooming, and SAP (stretch attend posture) was performed and fecal boli and urination were counted after each animal.Nose pokes in the hurdle were recorded by a photocell and manually collected by the test executioner.In order to summarize the parameters a trend analysis was preformed using the basic parameters and grouping them into categories based on the type of behavior measured in the parameter, as described earlier [37,39].The trend analysis consists of five categories, General activity, with the basic parameters of total activity, frequency of all corridors, frequency center, duration/frequency of all corridors (inversely ranked); Explorative activity, consisting of duration center (inversely ranked), duration hurdle, and duration for all the corridors combined (inversely ranked), rearing; Risk assessment, consisting on the basic parameters frequency for slope and bridge entrance, duration for slope and bridge entrance, duration/frequency slope and bridge entrance; Risk taking, consisting of frequency, duration, and duration/frequency for bridge and central circle; and Shelter seeking, with the basic parameters of duration, frequency and duration/frequency for dark corner room.For each descriptive parameter that is a part of a functional category, the individual rats (independent on treatment group) are ranked against each other.The ranking procedure enables a normalization for each parameter and they can be grouped together.Thereafter, the individual rank values for the descriptive parameters of a functional category are summed, resulting in group-wise comparison based on the animal's relative position to the entire tested population.

Statistical analysis
Due to the multivariate nature of the MCSF test, Meyerson et al. [39], suggests to compliment traditional univariate statistics with a multivariate data analysis in order to receive a good overview and identify patterns and correlations that cannot be seen with traditional statistics.The software SIMCA 17 (Umetrics AB, Umeå, Sweden) was used to perform principal component analysis (PCA) and partial least square discriminant analysis (PLS-DA) in order to produce an overview of the data from both the MCSF and the EPM.The PCA is an unsupervised method comparing all the parameters and all the individuals without taking into account how the parameters and individuals are linked to each other.Two components are presented as the X and Y axis, they are a product of all the traditional parameters joined together in order to simplify the data, where the first component stands for most of the variation in the data set and the second component for the next highest variation both presented as percentages of the total variation.The test presents a score plot for the relationships between the subjects of the study, and a loading plot which presents the parameters of the study and their effect on the relationships created between the parameters.The data presented are two dimensional and can potentially reveal groups and outliers not visible with more basic statistics.Parameters that are on opposite side of origo in the plot but linked with a straight line through origo (180 • angle) are negatively correlated to each other, a 90 • angle through origo indicates no correlation between the two parameters.A narrow angle, the two parameters are close to each other on the same side of origo, indicates a positive correlation between the parameters.In addition to the PCA, the PLS-DA is a supervised method taking into account grouping of the parameters and individuals and it is a regression extension of PCA.The PLS-DA examines predetermined groups (treatment groups) as the Y variable in relation to the X variables, which are the parameters of study.It presents the parameters of study that are most likely to be responsible for the separation of the groups and as well what parameters affect which group.The loading plot illustrates which parameters have the strongest effect on the different groups, the closer the parameter is to the group in the figure the bigger impact it has.The score plot presents the data in similarity to the PCA score plot.In addition to the score and the loading plot for the PLS-DA, the variable importance to the projection (VIP) plot presents the parameters with the highest overall impact on the test subjects.A parameter with a value higher than 1 is regarded as a significant impact on the test subject separation.From the multivariate tests single parameters were selected for univariate statistics where a difference between groups could be hypothesized from the loading plots of the PCA and PLS-DA.
Data from the MCSF and EPM were expressed as median ± interquartile range.The Shapiro-Wilk test for normality was performed on all data followed by either one-way ANOVA (for normally distributed data) or the nonparametric Kruskal-Wallis test (for nonnormally distributed data), followed by Tukey's multiple comparisons test or Dunn's multiple comparisons test, respectively.Comparisons were performed for the following: HOAT caffeine vs. control, LOAT caffeine vs. control, HOAT caffeine vs. LOAT caffeine, and HOAT control vs. LOAT control, to resemble column and row factor of a two-way ANOVA.Effect size eta-squared (η 2 ) were calculated for all significant parameters for the EPM and the MCSF.Tests were considered significant at p < 0.05.The data was managed using Microsoft Excel version 16.74 (Microsoft) and statistical analysis were performed in GraphPad Prism version 9 software (GraphPad Software Inc., USA).S1).The HOAT groups had the highest time spent in the open arms, followed by the IMOAT group, and then the LOAT group.The HOAT groups consisted of ten animals, the IMOAT group of eight animals, and LOAT of twelve animals.These groups were then separated in to caffeine treated and controls, The IMOAT group were excluded from further analysis due to the intention to compare the extremes.

Principal component analysis of parameters from MCSF and EPM demonstrates differences between the groups
To get an overview of the parameters from the MCSF and the second EPM test a multivariate data analysis principal component analysis (PCA) and the partial least square data analysis (PLS-DA) were used.The principal component analysis (Fig. 3), (n = 24, two components, R 2 X = 0.472, Q 2 = 0.193), with component 1 standing for 25.6 % of the total variation and component 2 with 17.1 % of the total variation.Over all, the caffeine treated groups were in the positive quadrants of the first component and the control groups in the negative quadrants (Fig. 3A).The LOAT groups were clearly separated alongside the first component, where the LOAT caffeine group were in the positive quadrants of the first component, and the control LOAT group in the negative quadrants of the first component.The HOAT group was separated in both the component 1 and component 2, where the HOAT caffeine had a positive component 1 value and a negative component 2 value.The HOAT control group had a negative contribution of component 1 and a positive contribution of component 2. The EPM parameters clustered themselves in the loadings plot except the closed arm time which were placed on the opposite side of the plot.According to the loadings plot, parameters with a high impact on the separation of the groups in the PCA were duration of center in the MCSF, the frequency of the central circle, velocity and distance moved in the MCSF and the EPM, as well as the duration in the DCR (Fig. 3B).

The groups show distinct separations in partial least square data analysis
The result from the PLS-DA (n = 24, two components, R 2 X = 0.384, R 2 Y = 0.449, Q 2 = 0.138) demonstrated a clear separation between the caffeine versus the control groups and thus supports the results of the PCA.The caffeine treated groups have a positive contribution of component 1 while the control groups have a negative contribution of component 1 giving a clear distinction between treated and control (Fig. 4A).The majority of the individuals in LOAT caffeine group have a positive contribution of component 1 and a negative contribution of component 2 with the mean of the group situated there as well.The LOAT control group have a negative contribution of both component and 2. The HOAT control have a positive contribution of component and a negative contribution of component 1, with their mean placed there as well, the HOAT caffeine treated group are spread from having a positive contribution of both component 1 and 2, to having a negative contribution to component 2 and a positive contribution of component , with the mean having similar contributions.Parameters that had the most impact on the LOAT caffeine treated group were latency to leave center in the MCSF, the duration on the central circle of the MCSF, the open arm frequency of the EPM, and the photocell count (Fig. 4B).Parameters affecting the LOAT control group were the duration in the dark corner room and the duration/frequency of the dark corner room of the MCSF.For the HOAT groups, the caffeine group were affected by the photocell count and the duration in all the corridors.For the HOAT controls, the duration in hurdle and time per visit in the hurdle had highest impact.The VIP plot (Fig. 4C), explaining which parameters had the highest overall impact on the total population, presents the closed arm time, open arm time, followed by distance moved and velocity of the EPM, and latency to bridge in the MCSF to have the highest impact.The VIP plot indicated that several of the parameters from the EPM had the greatest effect on the separation of the groups, including DUR/FRQ in the DCR, latency to the slope, latency to the bridge, together with total activity in the MCSF.In the PLS-DA the separation for both caffeine treatment and anxiety-like behavior can be distinctly shown, indicating that both aspects affect the results equally.

Caffeine increases general activity for the LOAT group in the MCSF
Univariate statistical analysis of the MCSF parameters demonstrated that caffeine increased the number of entries to the slope in the LOAT rats compared to LOAT controls, Kruskal Wallis test (H(3)= 8.688, p = 0.0337), and Dunn's multiple comparisons test (p = 0.0400) (Fig. 5A).A significant difference for total activity was found, Kruskal Wallis test (H(3)= 7.959, p = 0.0469), however, the post hoc test was not significant (Fig. 5E).Time per visit in the dark corner room provided a significant difference (Kruskal-Wallis test, H(3) = 8.737, p = 0.0330), however the post hoc test could not determine where the difference was (Fig. 5C).Latency to the slope and the bridge showed a trend (Kruskal-Wallis test, H(3) = 7.747, p = 0.0515, and H(3) = 6.919, p = 0.0745) (Figs.5B and 5F) towards significance.Remaining parameters are presented in supplementary table S2.For the functional category general activity a significant difference between LOAT control and LOAT caffeine was found, where the caffeine treated rats had a higher activity than controls (Kruskal-Wallis test, H(3) = 9.483, p = 0.0235, Dunn's multiple comparisons test, p = 0.0395) (Fig. 5D).There were no significant differences in the other functional categories risk-taking, shelter seeking, explorative activity, or risk assessment (Table 1).

10 mg/kg of caffeine increases time spent in open arms and in the center of the EPM
Selected parameters of the EPM were identified as having a high impact in the VIP plot from the PLS-DA.Caffeine increased the time spent in the open arms and the number of entries to the open arms for the rats in the LOAT group compared to control (Kruskal-Wallis test, H (3) = 15.14, p = 0.0015, H(3) = 13.00,p = 0.0046, respectively) Dunn's multiple comparisons test, p = 0.0086, p = 0.0175, respectively) with a trend for a difference between the HOAT caffeine and control in the open arm time (Dunn's multiple comparisons test, p = 0.0775) (Fig. 6A and 6B).The time in center were higher for both the HOAT and LOAT caffeine treated groups compared to their controls (Kruskal-Wallis test, H(3) = 16.38,p = 0.0009, Dunn's multiple comparisons test, p = 0.0296, and p = 0.0178, respectively (Fig. 6C).Caffeine affected the rearing (Kruskal-Wallis test, H(3) = 15.72,p = 0.0013) for both of the groups increasing the number of rearings for the caffeine treated rats, Dunn's multiple comparisons test, HOAT caffeine vs. control: p = 0.0097, LOAT caffeine vs control p = 0.0437, (Fig. 6D), The head dips were increased in the LOAT group (Kruskal-Wallis test, H(3) = 12.01, p = 0.0073), Dunn's multiple comparisons test, LOAT caffeine vs control: p = 0.0384 (Fig. 6E), grooming (Table 1) and stretch attend posture (SAP) (Fig. 6F) showed no significant differences.The number of entries to center were higher for the HOAT caffeine treated rats compared to their controls (Kruskal-Wallis test, H(3) = 14.37, p = 0.0024, Dunn's multiple comparisons test, p = 0.0124) (Fig. 6G).The duration and frequency of the closed arms differed significantly (Kruskal-Wallis test, H(3) = 15.93,p = 0.0012, p = 0.0046, respectively, with Dunn's multiple comparisons test, p = 0.0132, HOAT caffeine vs control, and 0.0175, LOAT caffeine vs control, respectively) (Fig. 6H and 6I).Both groups differed in distance moved and velocity compared to their control, Kruskal-Wallis test for distance moved (H(3) = 16.01,p = 0.0011), Dunn's multiple comparisons test, HOAT caffeine and control (p = 0.0163), LOAT caffeine and control (p = 0.0234) and  for velocity, (H(3) = 16.01,p = 0.0011) (Table 1), Dunn's multiple comparisons test, HOAT caffeine and control (p = 0.0163), LOAT caffeine and control (p = 0.0234).

Effects of 10 mg/kg caffeine on anxiety-like behavior and locomotor activity
The present study describes behavioral effects of a low acute dose of caffeine in male rats categorized as having a high anxiety-like behavior (LOAT) or a low anxiety-like behavior (HOAT).Earlier studies have demonstrated both anxiogenic and anxiolytic effects of caffeine, and factors such as dose variation, treatment length, sex, or underlying anxiety trait might influence these effects.In this study, male rats were used to facilitate comparisons of results from a previous study using a high dose of caffeine [37], and in addition males occasionally have had a higher response to caffeine [19].However, since previous studies show that there may be sex differences, the effect of caffeine on anxiety in both males and females needs to be further studied.The dose in the present study, 10 mg/kg, is considered a low dose caffeine, and the results suggest an anxiolytic effect of this dose, an effect most prominent between the LOAT caffeine and control rats.Supporting our results, acute administration of 10 mg/kg caffeine decreased anxiety-like behavior [40].Doses of 25 and 50 mg/kg caffeine gave an anxiolytic effect in the open field and the EPM in the presence of a bright light stressor [19], and 30 mg/kg gave an anxiolytic response in the EPM [18] indicating that even higher dosages than used here may elicit an anxiolytic response.10 mg/kg caffeine have also been shown to not affect the open arm time in rats [41], the rats were however tested in the EPM an hour after the caffeine administration which could contribute to the discrepancy between these results.In the EPM the rats moved more and faster when treated with caffeine, while in the MCSF no such difference could be seen, although an increase of general activity was observed in the LOAT caffeine treated rats.Lundberg et al. [42] found no correlations between distance moved and velocity in the MCSF and the EPM, suggesting that the behavior tests measure different kind of activity in the animals.Several studies have shown that acute doses of caffeine, 5, 15, 30 mg/kg [43], 10 and 30 mg/kg [44], 12.5 mg/kg [20] and 25 mg/kg [45] increase locomotor activity in male mice and rats.Comparing 10 mg/kg to 100 mg/kg in the forced swim test, rats given 10 mg/kg increased climbing and swimming compared to the 100 mg/kg [21], further supporting that the 10 mg/kg caffeine is an activity stimulatory dose.There could be a possibility that other parameters in the EPM are affected by the increased movement of the rats, as implied earlier [46], thus lower doses of caffeine may create a primary effect on locomotor activity, and through that, a secondary effect in the anxiety-like behavior.However, the variation would not entirely be dependent on this since the animals have a free choice to enter the open areas in the EPM or to explore them.Interestingly, a higher dose increased both distance moved and velocity in the EPM, but not the open arm time [37], further supporting that the increased open arm time observed in this study is associated with an anxiolytic effect.It should be noted that the rats experienced the EPM twice, the second time treated with caffeine, and they could be affected by the one trial tolerance [47].However, there were 28 days between the trials, something that has been shown to minimize the risk of this phenomenon [48].The open arm time for all groups was decreased in the second trial compared to the first trial, suggesting that the rats did remember the test, thus the retest did not show the same clear separation into groups of high or low open arm time of the controls.Previous studies have demonstrated relatively stable inter-individual differences when retesting in the EPM with the interval time of 25 and 35 days [33] and in Schneider et al., [48] 28 days (2011).A week before the second EPM test, the rats were tested in the MCSF which also could have influenced the behavior during the second EPM trial.Furthermore, more rats per group could have given a better distinction between the groups for certain parameters.

Effects of caffeine on risk assessment, impulsivity, and exploration
In both the PCA and the PLS-DA the HOAT and LOAT controls were more separated than the caffeine treated rats following the second component, supporting the categorization made in the screening EPM trial.This could indicate that the effects of 10 mg/kg of caffeine bring the HOAT and LOAT groups closer together, and thus diminishes the different anxiety-like states of the rats.Implying that caffeine treatment had a higher overall impact than the separation based on anxiety-like behavior, which also is demonstrated in several of the univariate parameters.The caffeine treated groups were placed closer together and moved towards parameters involved in risk-taking and exploration, implying that caffeine removes the difference seen for the controls and have an effect on these behaviors.Previous studies have shown an impact of caffeine on risk-taking behavior [25,37].In the MCSF, the slope is a risk assessment area adjacent to the bridge, rated as a risk area, and in addition linked to decision-making.Following the univariate statistics, the LOAT caffeine group had an increased slope frequency compared to LOAT controls, indicating that the LOAT caffeine treated rats takes longer time assessing whether or not they would enter the risk area of the bridge.This is supported by the PLS-DA where the mean for the LOAT caffeine group was placed closer to several of the risk parameters in comparison to the LOAT control group, suggesting that these parameters have a higher impact on the LOAT caffeine group.Although, caffeine's effect on risky decision-making in rats have been tested and no significant difference were found [26] suggesting other behaviors could be the cause of the risk assessment in the EPM.The caffeine treated rats in our study spent more time in the center of the EPM than their controls, possibly translated into increased impulsivity or risk assessment.Time in the center of the EPM is discussed to be a measure on the rat's decisiveness and their impulsivity, in adult female rats there is a significant positive correlation between time in the center of the EPM and the FI 60′s operant task, a test developed for measuring impulsivity [49] and rats bred for a high anxiety-like behavior display greater impulsivity [50].Furthermore, caffeine (single exposure, 5 and 10 mg/kg) increase sign tracking, which is linked to impulsivity [51].An acute dose of caffeine (2 mg/kg, single exposure) decrease impulsive-like behavior, while a chronic dose (2 mg/kg, two times a day for 21 days) increase impulsivity of rats in a tolerance to delay reward task [52].In humans, adolescents with gambling problem consume higher amount of caffeine than the general public, and have a strong correlation to higher impulsiveness, nicotine consumption, increased impulsive decision-making [53].The caffeine treated rats were placed closer to each other in the PLS-DA and had a higher number of rearings in the EPM compared to their controls, something that caffeine had been seen to affect previously [54], in addition the LOAT caffeine treated rats were placed closer to each other in the PLS-DA and had more head dips in the EPM compared to their controls.Both behaviors are explorative and may have a connection to the impulsivity trait from the center and the risk assessment in the slope from the MCSF, thus an impulsive rat may move around more without a determination of exploring its surroundings.

Conclusions
In summary, here we present that a low acute dose of caffeine (10 mg/kg), increased the distance moved and velocity in both the high and low anxiety-like behavior group and increased the open arm time of the EPM for the LOAT group, possibly an anxiolytic effect.However, in general the low acute dose of caffeine did not affect the rats with high or low anxiety-like behavior differently.In conclusion, the acute low dose of caffeine potentially increased impulsivity, and also had an effect on the overall behavior profile of the rat, indicating that a low dose of caffeine affects the general activity and impulsivity behavior.However, in contrast to a high dose, a low acute dose of caffeine does not affect individuals differently depending on their anxiety state.Further studies are needed to better understand the exact role of caffeine and adenosine receptor signaling in anxiety disorders.

Fig. 2 .
Fig. 2. Cumulative duration of open arm time in the screening round of elevated plus maze in %.Rats were given no prior caffeine treatment for this round.3 groups: High open arm time (HOAT) > 14 % in black, Intermediate open arm time (IMOAT) 8.5 -12 % in pink, Low open arm time (LOAT) < 8.5 % in teal.Data is presented as median with interquartile range of each group, Oneway ANOVA test were performed followed by Tukey's multiple comparisons test, p ≤ 0.05 were considered statistically significant.*** p ≤ 0.001, **** p ≤ 0.0001.

Fig. 3 .
Fig. 3. Principal component analysis (PCA) of the test subjects and the parameters.Scatter plots of A. Individual scores for each test subject in relation to component 1 (x axis) and component 2 (y axis).The four groups presented in different colors: high open arm time (HOAT) Caffeine (green), HOAT Control (blue), low open arm time (LOAT) Caffeine (red), LOAT Control (yellow).B. Loadings for each parameter from the behavior tests, Loadings further from origo have a higher impact on the total variation.The parameters are colored after their functional category: Anxiety-like behavior, Parameters from the elevated plus maze, Explorative activity, General activity, Impulsive-like behavior, Risk assessment, Risk taking, Shelter seeking and Other.Abbrevations: Closed arm time (CAT), Central circle (CTRCI), Distance moved (DIST), Duration (D), Dark corner room (DCR), Duration/Frequency (D/F), Frequency (F), Latency to enter (LAT), all corridors (TOTCORR), all activity (TOTACT), velocity (VEL).

Fig. 4 .
Fig. 4. Partial least square discriminant analysis (PLS-DA) of the test subject and the parameters.Scatter plots of A. Individual scores for each test subject in relation to the other subjects and the first and second component, presents the four groups high open arm time (HOAT) Caffeine (green), HOAT Control (blue), low open arm time (LOAT) Caffeine (red), LOAT Control (yellow), and their relation to component 1 and 2. B. Variable loadings from the PLS-DA of the test subjects and the parameters of the two behavior tests.The test parameters are colored by functional category (Anxiety-like behavior, Elevated plus maze, Explorative activity, General activity, Impulsive-like behavior, Risk assessment, Risk taking, Shelter seeking, and Other) together with the mean score of the four treatment groups (HOAT Caffeine, HOAT Control, LOAT Caffeine, LOAT Control).C. Variable Important to the projection (VIP) plot.Describing the parameters with the highest impact on the individuals in the test, explaining X (parameters from the behaviour tests) and the correlation to Y (the treated rats).The parameters are colored after their functional category, A value above one indicates that the parameter has a noticeable impact on the variation for the test subjects.Errow bars state the confidence intervals for the VIP values.Abbrevations: Closed arm time (CAT), Central circle (CTRCI), Distance moved (DIST), Duration (D), Dark corner room (DCR), Duration/Frequency (D/F), Frequency (F), Latency to enter (LAT), all corridors (TOTCORR), all activity (TOTACT), velocity (VEL).

Fig. 5 .
Fig. 5. Basic parameters and the functional category general activity from the Multivariate Concentric Square Field (MCSF) based on the multivariate data, after 10 mg/kg subcutaneous caffeine treatment.HOAT: high open arm time, LOAT: low open arm time.Caffeine in grey and turquoise and control in black and teal. A. number of entries to the slope.B. the latency to enter the slope in s. C. time per visit in the dark corner room (DCR) in s.D. sum of ranks of the functional category general activity, E. total activity measured as number of entries to each zone in the arena.F. the latency to reach the bridge in s.Data presented as median with interquartile range, * p ≤ 0.05.

Fig. 6 .
Fig. 6.Basic parameters from Elevated Plus Maze (EPM) after 10 mg/kg subcutaneous caffeine treatment.Caffeine in grey and turquoise and control in black and teal. A. open arm time in percent of total time spent in the EPM.B. number of entries to the open arms.C. duration in center in percent of total time spent in the EPM.D. number of rearings in EPM.E. number of head dips in EPM.F. number of stretch attend posture (SAP) in the EPM.G. number of entries to center of EPM.H. time in closed arm in percent of total time in the EPM.I. number of entries to closed arm.Abbrevations: HOAT: High open arm time, LOAT: low open arm time.%OAT: procent open arm time, SAP: stretch attend posture, %CAT: procent closed arm time.Data presented as median with interquartile range, * p ≤ 0.05, ** p ≤ 0.01.

Table 1
Parameters distance moved, velocity, and grooming from the EPM and functional categories from the MCSF.