JaramillopersName, ZadorpersName (2014date). “Mice and rats achieve similar levels of performance in an adaptive decision-making task.” Frontiers in Systems Neurosciencetitle·j.
Object of study — whether miceterm can perform an adaptive decision-making task as well as ratsterm, the established rodent model for it
Research question — how learning speed, performance level and speed of adaptation compare between the two speciesterm·var on one flexible categorizationterm task
Contribution — shows miceterm reach the same performance and adapt as fast as ratsterm though they take longer to learn, and argues the residual learning gap may be an artefact of using a task built for ratsterm
phenomenon · claim-type · cross-domain-transfer · mechanism · claim-relation · replication-status · limitation — grounded in the full text| phenomenon | adaptive decision-making in rodents — “The task required animals to categorize acoustic stimuli as high- or low-frequency sounds, and to adapt quickly after changes in the category boundary.” | |
| claim-type | null-result — “No major differences in subjective categorization boundaries or the speed of adaptation between the two species were found.” | |
| method — evidence | — “Here we compare learning and performance of mice and rats, the preferred behavioral rodent model, during an acoustic flexible categorization two-alternative choice task.” | |
| object of study — evidence | — “Two opposing constraints exist when choosing a model organism for studying the neural basis of adaptive decision-making: (1) experimental access and (2) behavioral complexity. Available molecular and genetic approaches for studying neural circuits in the mouse fulfill the first requirement. In contrast, it is still under debate if mice can perform cognitive tasks of sufficient complexity.” | |
| research question — evidence | — “However, mice offer further advantages over rats, stemming mainly from the availability of hundreds of transgenic lines that enable precise targeting of gene expression to defined neuronal populations (Madisen et al., 2010 ; Taniguchi et al., 2011 ; Madisen et al., 2012 ; Gerfen et al., 2013 ). It is therefore crucial to evaluate whether mice can achieve the high levels of performance on cognitive tasks that has previously been demonstrated in rats.” | |
| contribution — evidence | — “We found that both species achieved similarly high performance levels. On average, rats learned the task faster than mice, although some mice were as fast as the average rat. No major differences in subjective categorization boundaries or the speed of adaptation between the two species were found.” | |
| transfer | from primate systems neuroscience: the two-alternative forced choice psychophysical paradigm, brought down to rodents — “Smaller mammals, rodents in particular, provide several advantages over primates for physiological experiments: faster breeding times, lower costs, smaller housing needs, and currently, easier access to molecular and genetic tools.” | transfers[0] |
| transfer | from psychophysics: the fitted logistic and its two parameters, boundary and slope, as the description of a subject's categorisation — “We then fitted a logistic function to the psychometric performance and evaluated two parameters: the subjective category boundary and the slope of the curve.” | transfers[1] |
| mechanism | behavioural adaptation is modelled as an internal category boundary that shifts between blocks, which is why the change generalises to untrained frequencies rather than being limited to the reversing tone — “the observation that animals generalize to other frequencies is well captured by a model in which an internal category boundary shifts between blocks of trials” | mechanisms[0] |
| mechanism | two candidate physiological routes to the same behavioural switch: synaptic-weight change in the sensory-motor pathway (learning/memory), or on-line integration of sensory and context information without lasting synaptic change (selective attention) — “Similarly, there are at least two possible mechanisms for this adaptation in behavior: (1) changes in synaptic strength in the sensory-motor pathway that result in different behavioral responses for the same sound; or (2) integration of sensory and context information without long-term synaptic changes in the sensory-motor pathway.” | mechanisms[1] |
| mechanism | measured rat superiority may be produced by the tasks themselves, since comparisons use paradigms originally engineered around rat ethology — “This potential confound may explain why rats have been found to be better in a variety of tasks” | mechanisms[2] |
| mechanism | animals could be tracking the outcome of the reversing 7 kHz tone, or using the extreme tones as a cue to which contingency is active, and the paper allows both — “Animals could use various strategies for solving the task described in this study. First, animals may evaluate the outcome of responses to the reversing frequency (7 kHz) to update their sound-action associations. Alternatively, the appearance of an extreme stimulus (3 or 16.3 kHz) could be used as a cue indicating the most likely contingency. These strategies are non-exclusive, and animals may be taking advantage of both.” | mechanisms[3] |
| mechanism | the adaptation could be synaptic change in the sensory-motor pathway or context integration without lasting change - learning or attention — “Similarly, there are at least two possible mechanisms for this adaptation in behavior: (1) changes in synaptic strength in the sensory-motor pathway that result in different behavioral responses for the same sound; or (2) integration of sensory and context information without long-term synaptic changes in the sensory-motor pathway. The former correspond to mechanisms hypothesized for learning and memory (Martin et al., 2000 ), the latter to mechanisms of selective attention (Jaramillo and Pearlmutter, 2007 ), both defining features of cognition.” | mechanisms[4] |
| mechanism | whatever the strategy, the generalisation across untrained frequencies is captured by an internal boundary that shifts between blocks — “Animals could use various strategies for solving the task described in this study. First, animals may evaluate the outcome of responses to the reversing frequency (7 kHz) to update their sound-action associations. Alternatively, the appearance of an extreme stimulus (3 or 16.3 kHz) could be used as a cue indicating the most likely contingency. These strategies are non-exclusive, and animals may be taking advantage of both.” | mechanisms[5] |
| builds-on | Jaramillo et al. (2014), the flexible sound-categorization task and shifting-boundary model — “Here we compare the speed of learning and the performance levels achieved by mice and rats in a decision-making task that tested the animals' perceptual abilities as well as their adaptability (Jaramillo et al., 2014 ).” | claim_relations[0] |
| confirms | vibrotactile comparisons finding mice and rats similar on psychometric curves, reaction times and learning rates — “In a two-alternative forced choice paradigm for vibrotactile discrimination, the performance of mice and rats was found to be very similar when quantifying psychometric curves, reaction times, learning rates, and trial omissions” | claim_relations[1] |
| qualifies | the body of comparisons in which rats outperform mice, by identifying the rat-derived task as a shared confound — “Most comparisons of behavioral performance between mice and rats, including our study, have used tasks originally devised for rats. This potential confound may explain why rats have been found to be better in a variety of tasks” | claim_relations[2] |
| critiques | the assumption that a paradigm can be carried over from rat to mouse unchanged — “Successful behavioral paradigms, however, rely on targeting designs to the idiosyncrasies of the mouse from the outset, rather than simply assuming that mice are little rats.” | claim_relations[3] |
| supports | the reinterpretation of water-maze species differences as swimming differences rather than memory differences — “Behavior of the two species was similar in the navigation tasks when evaluated in the dry-land maze. Differences found in water-maze tasks seem to be accounted for mostly by differences in swimming rather than memory capabilities.” | claim_relations[4] |
| extends | rodent psychophysics paradigms modelled on primate decision-making tasks — “Smaller mammals, rodents in particular, provide several advantages over primates for physiological experiments: faster breeding times, lower costs, smaller housing needs, and currently, easier access to molecular and genetic tools.” | claim_relations[5] |
| sense | the same task across species is an EQUIVALENCE CLAIM resting on declared adjustments: reward volume scaled to each species' normal consumption, tubing gauge changed, sound levels set above each species' own threshold — “The volume of water reward was scaled according to the normal consumption for each species” | senses[0] |
| sense | similar performance here means failure to reject a difference at n=10 per species, not a demonstrated equivalence - the paper's own phrasing is that no major differences were found — “We found that both species achieved similarly high performance levels. On average, rats learned the task faster than mice, although some mice were as fast as the average rat. No major differences in subjective categorization boundaries or the speed of adaptation between the two species were found.” | senses[1] |
| sense | subjective category boundary is the 50% point of a fitted logistic, distinct from the experimenter's boundary, and it is what shifts between blocks — “We then fitted a logistic function to the psychometric performance and evaluated two parameters: the subjective category boundary and the slope of the curve.” | senses[2] |
| sense | speed of switching is measured as the trial at which an exponential fit crosses chance, not as trials to criterion — “We estimated the speed of switching to a new sound-action association by averaging performance for each trial from several sessions, and calculating when an exponential fit crossed the 50% chance performance level” | senses[3] |
| sense | valid trials exclude early withdrawals, so every performance figure is conditioned on the animal having waited out the sound — “If animals withdrew before the end of the stimulus, the trial was aborted and denoted as invalid. Only valid trials were used in calculations of the fraction of correct trials.” | senses[4] |
| control | the water-delivery valves calibrated at least once a week to fix the reward volume — held — rules out reward size drifting over weeks of training, which would confound motivation with the task variable being manipulated — “The amount of water delivered was controlled by valves calibrated at least once a week.” | controls[0] |
| differs in | the assumption that a paradigm can be carried over from rat to mouse unchanged — THE AXIS IS WHETHER THE PARADIGM WAS DESIGNED FOR THE SPECIES. 'Successful behavioral paradigms, however, rely on targeting designs to the idiosyncrasies of the mouse from the outset, rather than simply assuming that mice are little rats.' Reports of poorer mouse performance are attributed to a design decision rather than to the animal - which makes the critique testable and the original results reinterpretable rather than wrong. — “Successful behavioral paradigms, however, rely on targeting designs to the idiosyncrasies of the mouse from the outset, rather than simply assuming that mice are little rats.” — “rather than simply assuming that mice are little rats” | claim_relations[3] |
| note | subjective category boundary —operationalised-as→ psychometric — the boundary is a fitted parameter of the curve, so it exists only relative to the logistic model chosen — “We then fitted a logistic function to the psychometric performance and evaluated two parameters: the subjective category boundary and the slope of the curve.” | concept_edges[0] |
| audit | flexible categorization —operationalised-as→ contingency — Audit 2026-09-28: DOUBTFUL - read against its full span and kept, but the span describes the categorization contingency switching between blocks; 'flexible categorization' is not in it, so operationalised-as is loose. — “The categorization contingency changed from one block to the next without any additional cue indicating the change (besides the sound-action-reward relations). The initial contingency for each session was alternated from one day to the next.” | concept_edges[1] |
| note | flexible categorization —operationalised-as→ contingency — CRITERION: the switch is uncued, so the animal can only detect it from sound-action-reward outcomes — “The categorization contingency changed from one block to the next without any additional cue indicating the change (besides the sound-action-reward relations). The initial contingency for each session was alternated from one day to the next.” | concept_edges[1] |
| note | mice —contrasted-with→ rats — the comparison the paper exists to make - and its own discussion says the task favours the rat — “We found that rats learned the basic discrimination task (reaching stage 6-SB, Switch in Blocks) faster than mice ( p = 0.0014, rank-sum test). On average, rats reached the criterion of 80% correct for all stimuli by session 9.8 ± 0.9 (mean ± SD ), while mice took 18.1 ± 4.2 sessions.” | concept_edges[2] |
| note | selective attention —contrasted-with→ learning and memory — two mechanisms behaviour cannot separate; the paper states that physiology would be needed — “Similarly, there are at least two possible mechanisms for this adaptation in behavior: (1) changes in synaptic strength in the sensory-motor pathway that result in different behavioral responses for the same sound; or (2) integration of sensory and context information without long-term synaptic changes in the sensory-motor pathway. The former correspond to mechanisms hypothesized for learning and memory (Martin et al., 2000 ), the latter to mechanisms of selective attention (Jaramillo and Pearlmutter, 2007 ), both defining features of cognition.” | concept_edges[5] |
| audit | valid trials —distinct-from→ early withdrawals — Audit 2026-09-28: DOUBTFUL - read against its full span and kept, but the span writes 'withdrew before the end of the stimulus ... invalid', not 'early withdrawals'. — “If animals withdrew before the end of the stimulus, the trial was aborted and denoted as invalid. Only valid trials were used in calculations of the fraction of correct trials.” | concept_edges[6] |
| note | valid trials —distinct-from→ early withdrawals — CRITERION: withdrawals are removed from the denominator, so performance is conditional on waiting — “If animals withdrew before the end of the stimulus, the trial was aborted and denoted as invalid. Only valid trials were used in calculations of the fraction of correct trials.” | concept_edges[6] |
| audit | EVIDENCE CORRECTED (RE-GROUNDED) · bias-correction —causes→ chance level — Audit 2026-09-28: DOUBTFUL - read against its span and kept, but bias correction produced BELOW-chance performance; the endpoint 'chance level' is wrong. — “In contrast, mice were either at chance level or had the bias-correction mode activated, resulting in performances below chance.” | concept_edges[7] |
| note | EVIDENCE CORRECTED (RE-GROUNDED) · bias-correction —causes→ chance level — SIGN: pushes measured performance BELOW chance, which is why some mouse curves start under the line — “In contrast, mice were either at chance level or had the bias-correction mode activated, resulting in performances below chance.” | concept_edges[7] |
| retracted | EVIDENCE CORRECTED (RE-GROUNDED) · bias-correction —causes→ chance level — RE-GROUNDED. The old span named neither. This sentence names both and carries the consequence - bias correction RESULTING IN performance below chance. — “In contrast, mice were either at chance level or had the bias-correction mode activated, resulting in performances below chance.” — “Note that for an average of 500 trials per session, performance” | concept_edges[7] |
| weight | EVIDENCE CORRECTED (RE-GROUNDED) · bias-correction —causes→ chance level — 0.7 — “In contrast, mice were either at chance level or had the bias-correction mode activated, resulting in performances below chance.” | concept_edges[7] |
| outcome (stated by the tagger, no quotation) | control — held | controls[0] |
| for which use | limitation — the whole mouse-versus-rat comparison: MOST COMPARISONS ... INCLUDING OUR STUDY, HAVE USED TASKS ORIGINALLY DEVISED FOR RATS - the instrument favours one of the two species compared — “Most comparisons of behavioral performance between mice and rats, including our study, have used tasks originally devised for rats.” | limitations[0] |
| for which use | limitation — any mechanistic reading of the adaptation: BEHAVIORAL MEASUREMENTS ALONE MAY NOT DISSOCIATE BETWEEN THESE POSSIBILITIES (synaptic change versus attention) — “Although behavioral measurements alone may not dissociate between these possibilities, techniques for physiological analysis of neural circuits in the mouse have the potential to reveal the mechanisms that mediate this rapid adaptation in the interpretation of sounds.” | limitations[1] |
| for which use | limitation — the reported training times only: OTHER OBJECTIVES, FOR EXAMPLE MINIMIZING TRAINING TIMES FOR EACH ANIMAL, MAY REQUIRE ADJUSTING PARAMETERS FOR EACH INDIVIDUAL - so these are not the fastest achievable — “Other objectives, for example minimizing training times for each animal, may require adjusting parameters for each individual.” | limitations[2] |
| for which use | limitation — the same scope as limitations[0] - the task was devised for rats — “Most comparisons of behavioral performance between mice and rats, including our study, have used tasks originally devised for rats. This potential confound may explain why rats have been found to be better in a variety of tasks” | limitations[3] |
| for which use | limitation — the same scope as limitations[1] - behaviour cannot separate learning from attention — “Although behavioral measurements alone may not dissociate between these possibilities, techniques for physiological analysis of neural circuits in the mouse have the potential to reveal the mechanisms that mediate this rapid adaptation in the interpretation of sounds.” | limitations[4] |
| for which use | limitation — the performance distributions and any per-species rate: ONE MOUSE NEVER LEARNED THE DISCRIMINATION TASK ... ONE RAT NEVER REACHED THE FINAL STAGE. THESE ANIMALS WERE EXCLUDED — “One mouse never learned the discrimination task, although this animal showed no impairment in initiating trials or collecting reward. One rat never reached the final stage (7-OK). These animals were excluded from subsequent analysis.” | limitations[5] |
| for which use | limitation — the same scope as limitations[2] - parameters set for comparability, not for best performance — “In this study, we set the duration of behavioral sessions to 1 h for all stages of training, kept the amount of reward per trial constant, and used the same criteria to advanced each individual from one stage to the next. Consistency in training parameters helped us compare directly the two species, the main goal of the study. Other objectives, for example minimizing training times for each animal, may require adjusting parameters for each individual.” | limitations[6] |
| for which use | limitation — every auditory result in the mouse: C57BL/6J, A COMMON BACKGROUND FOR TRANSGENICS, suffers EARLY LOSS OF HEARING SENSITIVITY, which had to be checked before the data could be read — “An important consideration when working with mice is that some strains suffer from early loss of hearing sensitivity (Zheng et al., 1999 ). C57Bl/6J, a common background for transgenics used in the study of neural circuits, show elevated auditory brainstem response thresholds at 700 days of age (60 dB higher than normal), but not at 200 days of age.” | limitations[7] |
| replication | untested | |
| limitation | The comparison uses a task originally designed for rats, a confound that may bias species comparisons in the rat's favour. — “Most comparisons of behavioral performance between mice and rats, including our study, have used tasks originally devised for rats.” | limitations[0] |
| limitation | Behavioural data alone cannot distinguish the candidate mechanisms (synaptic change in the sensory-motor pathway versus contextual integration without such change) underlying the adaptation. — “Although behavioral measurements alone may not dissociate between these possibilities, techniques for physiological analysis of neural circuits in the mouse have the potential to reveal the mechanisms that mediate this rapid adaptation in the interpretation of sounds.” | limitations[1] |
| limitation | Training parameters were held constant to permit a species comparison, so the reported training times are not optimized and would differ under individually tuned protocols. — “Other objectives, for example minimizing training times for each animal, may require adjusting parameters for each individual.” | limitations[2] |
| limitation | the paper names a confound that undercuts its own headline result: the task was devised for rats, which may be why rats look faster — “Most comparisons of behavioral performance between mice and rats, including our study, have used tasks originally devised for rats. This potential confound may explain why rats have been found to be better in a variety of tasks” | limitations[3] |
| limitation | behaviour alone cannot separate the learning account from the attention account of the adaptation — “Although behavioral measurements alone may not dissociate between these possibilities, techniques for physiological analysis of neural circuits in the mouse have the potential to reveal the mechanisms that mediate this rapid adaptation in the interpretation of sounds.” | limitations[4] |
| limitation | one mouse never learned the discrimination and one rat never reached the final stage; both were dropped — “One mouse never learned the discrimination task, although this animal showed no impairment in initiating trials or collecting reward. One rat never reached the final stage (7-OK). These animals were excluded from subsequent analysis.” | limitations[5] |
| limitation | training was set for comparability rather than for each animal's best performance, so the levels reported are not ceilings — “In this study, we set the duration of behavioral sessions to 1 h for all stages of training, kept the amount of reward per trial constant, and used the same criteria to advanced each individual from one stage to the next. Consistency in training parameters helped us compare directly the two species, the main goal of the study. Other objectives, for example minimizing training times for each animal, may require adjusting parameters for each individual.” | limitations[6] |
| limitation | the C57Bl/6J strain loses hearing with age, which had to be checked separately before auditory results could be trusted — “An important consideration when working with mice is that some strains suffer from early loss of hearing sensitivity (Zheng et al., 1999 ). C57Bl/6J, a common background for transgenics used in the study of neural circuits, show elevated auditory brainstem response thresholds at 700 days of age (60 dB higher than normal), but not at 200 days of age.” | limitations[7] |
finding · quantity · variable · definition · key-term — grounded in the full text| finding | Rats reached the basic sound-discrimination criterion in significantly fewer sessions than mice. — “We found that rats learned the basic discrimination task (reaching stage 6-SB, Switch in Blocks) faster than mice ( p = 0.0014, rank-sum test).” | content_tags.findings[0] |
| finding | Despite the difference in learning speed, the two species converged on comparably high asymptotic performance. — “We found that both species achieved similarly high performance levels.” | content_tags.findings[1] |
| finding | Of all training stages, the sound-discrimination stage showed the largest mouse-rat difference, about 7 sessions. — “From all stages, stage 5-RC (Required Correct), in which animals learned to discriminate sounds, showed the largest difference between mice and rats (7 sessions on average, p = 0.0003, rank-sum test).” | content_tags.findings[2] |
| finding | Once trained, mice and rats re-adapted to a reversed category boundary at the same speed, needing roughly ten trials. — “We found that both rats and mice needed about 10 trials to switch between contingencies (Figure 5E , p = 0.72, rank-sum test).” | content_tags.findings[3] |
| finding | The size of the shift in subjective category boundary between contingencies did not differ between species, although mice had slightly steeper psychometric slopes. — “The change in subjective category boundary as animals switched from one contingency to the other was not different between mice and rats ( p = 0.89 rank-sum test), but mice showed slightly steeper slopes than rats ( p = 0.019 on the low-boundary block, p < 0.001 on the high-boundary block).” | content_tags.findings[4] |
| finding | Age-related hearing loss in the C57Bl/6J strain did not prevent mice from performing the auditory task over several hundred days. — “We found that C57Bl/6J mice can perform the auditory flexible categorization task for several hundred days without major impairments in performance.” | content_tags.findings[5] |
| finding | rats reached the switching stage in 9.8 sessions on average against 18.1 for mice — “We found that rats learned the basic discrimination task (reaching stage 6-SB, Switch in Blocks) faster than mice ( p = 0.0014, rank-sum test). On average, rats reached the criterion of 80% correct for all stimuli by session 9.8 ± 0.9 (mean ± SD ), while mice took 18.1 ± 4.2 sessions.” | content_tags.findings[6] |
| finding | the gap was concentrated in the discrimination stage, about seven sessions of it — “From all stages, stage 5-RC (Required Correct), in which animals learned to discriminate sounds, showed the largest difference between mice and rats (7 sessions on average, p = 0.0003, rank-sum test). This stage was responsible for most of the difference in the overall length of training.” | content_tags.findings[7] |
| finding | the distributions overlapped and some mice matched the average rat — “Overall, rats reached the final stage (7-OK) faster than mice (Figure 2G ) ( p = 0.0014, rank-sum test), although the distributions largely overlapped and some mice were as fast as the average rat.” | content_tags.findings[8] |
| finding | mice showed a slightly higher average performance on the last stages — “Although both species achieved a similar fraction of correct trials per session (Figure 3C ), mice showed a slightly higher average performance on the last stages ( p < 0.05 rank-sum test, for several consecutive sessions after reaching stage 6-SB, Switch in Blocks).” | content_tags.findings[9] |
| finding | nine of ten animals in each species exceeded 70% on the reversing tone in both contingencies — “Nine out of ten animals from each species achieved performance levels above 70% for this reversing stimulus on both contingencies (Figure 2 ). Performance for the non-reversing stimuli (3 and 16.3 kHz) was always above this level.” | content_tags.findings[10] |
| finding | C57Bl/6J mice performed the task for several hundred days without major impairment — “We found that C57Bl/6J mice can perform the auditory flexible categorization task for several hundred days without major impairments in performance.” | content_tags.findings[11] |
| finding | CONTROL NULL: both species took about ten trials to switch contingency, with no species difference — “We found that both rats and mice needed about 10 trials to switch between contingencies (Figure 5E , p = 0.72, rank-sum test).” | content_tags.findings[12] |
| finding | CONTROL NULL: the shift in subjective category boundary between contingencies did not differ between species — “The change in subjective category boundary as animals switched from one contingency to the other was not different between mice and rats ( p = 0.89 rank-sum test), but mice showed slightly steeper slopes than rats ( p = 0.019 on the low-boundary block, p < 0.001 on the high-boundary block).” | content_tags.findings[13] |
| finding | CONTROL NULL: trials completed per session did not differ reliably, though mice varied more — “We found no statistically reliable difference in the average number of trials per session between mice and rats for the reward amounts used in our study ( p = 0.12, rank-sum test), although the variability across animals was larger in the cohort of mice” | content_tags.findings[14] |
| finding | CONTROL NULL: early withdrawals were about 20% in both species — “Early withdrawals were about 20% on average for both mice and rats ( p = 0.56, rank-sum test), although variability was higher for mice. Analysis of subsequent sessions showed that the average for rats remained in the range 15–20%, while the average for mice fluctuated in the range 20–30%.” | content_tags.findings[15] |
| finding | CONTROL NULL: trials needed to learn discrimination did not predict trials needed to learn switching, in either species — “There was no apparent correlation between the trials required to complete stages 5-RC (Required Correct) and 6-SB (Switch in Blocks) for either mice or rats (|ρ| < 0.12, p = 0.57, Spearman correlation).” | content_tags.findings[16] |
| finding | CONTROL: a side-bias or alternation strategy would have scored under 60% in essentially every session — “Post-hoc analysis of the trials presented in each session verified that a strategy of always choosing one side, would yield performances below 60% on 98% of the sessions. Similarly, a strategy of alternating sides would yield performances below 60% on 99.9% of the sessions.” | content_tags.findings[17] |
| finding | both species ran fewer trials on the first day back after free water at the weekend — “Animals received free water on weekends resulting in a clear difference between the number of trials on the first day of the week and subsequent days for both species ( p = 0.0077, for rats and for mice, Wilcoxon signed-rank test).” | content_tags.findings[18] |
| quantity | 9.8 ± 0.9 — mean number of sessions rats needed to reach the 80%-correct criterion (mice needed 18.1 ± 4.2) — “On average, rats reached the criterion of 80% correct for all stimuli by session 9.8 ± 0.9” | content_tags.quantities[0] |
| quantity | 2320 — average number of trials by which rats beat mice on the discrimination-learning stage — “Stage 5-RC (Required Correct), showed the largest difference between species (rats required 2320 fewer trials on average, p = 0.0013, rank-sum test).” | content_tags.quantities[1] |
| quantity | about 10 trials — trials needed by both species to cross chance performance again after a contingency switch — “We found that both rats and mice needed about 10 trials to switch between contingencies (Figure 5E , p = 0.72, rank-sum test).” | content_tags.quantities[2] |
| quantity | about 20% — share of trials with early withdrawal from the centre port, statistically indistinguishable between species — “Early withdrawals were about 20% on average for both mice and rats ( p = 0.56, rank-sum test), although variability was higher for mice.” | content_tags.quantities[3] |
| quantity | ten and ten — cohort sizes: ten Long Evans rats and ten C57Bl/6J mice — “Ten adult male Long Evans rats (Taconic Farms) and ten adult male C57Bl/6J mice (The Jackson Laboratory, stock #000664) were used in this study” | content_tags.quantities[4] |
| quantity | 56% — performance level below which a 500-trial session counts as chance — “Note that for an average of 500 trials per session, performance below 56% would be considered at chance ( p < 0.01, binomial test)” | content_tags.quantities[5] |
| quantity | 9.8 vs 18.1 sessions — sessions to reach the switching stage, rats against mice — “We found that rats learned the basic discrimination task (reaching stage 6-SB, Switch in Blocks) faster than mice ( p = 0.0014, rank-sum test). On average, rats reached the criterion of 80% correct for all stimuli by session 9.8 ± 0.9 (mean ± SD ), while mice took 18.1 ± 4.2 sessions.” | content_tags.quantities[6] |
| quantity | about 10 trials — trials to recover chance performance after an uncued contingency switch, in both species — “We found that both rats and mice needed about 10 trials to switch between contingencies (Figure 5E , p = 0.72, rank-sum test).” | content_tags.quantities[7] |
| quantity | 24 ul vs 2-3 ul — reward per trial, scaled to each species' normal consumption - the adjustment that makes the task comparable — “The volume of water reward was scaled according to the normal consumption for each species” | content_tags.quantities[8] |
| value source note | quantity — THE VALUE DRAWS ON THE PAPER BEYOND THIS QUOTE: part of it is printed elsewhere in the paper, not in this quote ('weeks Weight 285 ± 29 g 26.1 ± 1.3 g Reward/trial 24 μl 2–3 μl Open in a new tab 2.2.'). The reward volumes are given in the paper's table of animal details. | content_tags.quantities[8] |
| quantity | 0-20 dB-SPL vs 10-40 dB-SPL — hearing thresholds of rats against mice over the frequencies used — “Although rats and mice have different absolute thresholds for the frequencies used here: 0–20 dB-SPL for rats (Heffner et al., 1994 ) and 10–40 dB-SPL for mice (Koay et al., 2002 ), all sounds were presented above threshold for either species.” | content_tags.quantities[9] |
| quantity | 300 — valid trials per block before the categorisation boundary reversed — “A single session (1 h) consisted of several blocks of 300 valid trials.” | content_tags.quantities[10] |
| quantity | 56% — the chance ceiling for a 500-trial session by binomial test — “Note that for an average of 500 trials per session, performance below 56% would be considered at chance ( p < 0.01, binomial test), as indicated by the dotted lines in Figure 3” | content_tags.quantities[11] |
| quantity | 50-70 dB-SPL — roving sound level, used so loudness cannot solve the task — “The intensity of the sounds was variable in the range 50–70 dB-SPL to discourage the use of loudness to solve the task; the perceived loudness of a sound depends on frequency even when the same sound-pressure level reaches the ear.” | content_tags.quantities[12] |
| variable | species (independent) — “Here we compare learning and performance of mice and rats, the preferred behavioral rodent model, during an acoustic flexible categorization two-alternative choice task.” | content_tags.variables[0] |
| variable | categorization contingency (independent) — “The categorization contingency changed from one block to the next without any additional cue indicating the change” | content_tags.variables[1] |
| variable | fraction of correct trials (dependent) — “Only valid trials were used in calculations of the fraction of correct trials.” | content_tags.variables[3] |
| variable | intensity of the sounds (control) — “The intensity of the sounds was variable in the range 50–70 dB-SPL to discourage the use of loudness to solve the task” | content_tags.variables[6] |
| variable | training stages — “We found that rats learned the basic discrimination task (reaching stage 6-SB, Switch in Blocks) faster than mice ( p = 0.0014, rank-sum test). On average, rats reached the criterion of 80% correct for all stimuli by session 9.8 ± 0.9 (mean ± SD ), while mice took 18.1 ± 4.2 sessions.” | content_tags.variables[7] |
| variable | subjective category boundary — “The change in subjective category boundary as animals switched from one contingency to the other was not different between mice and rats ( p = 0.89 rank-sum test), but mice showed slightly steeper slopes than rats ( p = 0.019 on the low-boundary block, p < 0.001 on the high-boundary block).” | content_tags.variables[8] |
| variable | slope — “Although both species achieved a similar fraction of correct trials per session (Figure 3C ), mice showed a slightly higher average performance on the last stages ( p < 0.05 rank-sum test, for several consecutive sessions after reaching stage 6-SB, Switch in Blocks). This observation will be discussed in more detail in the next section when comparing psychometric slopes between the two” | content_tags.variables[9] |
| variable | early withdrawals — “Early withdrawals were about 20% on average for both mice and rats ( p = 0.56, rank-sum test), although variability was higher for mice. Analysis of subsequent sessions showed that the average for rats remained in the range 15–20%, while the average for mice fluctuated in the range 20–30%.” | content_tags.variables[10] |
| variable | duration of behavioral sessions (control) — “In this study, we set the duration of behavioral sessions to 1 h for all stages of training, kept the amount of reward per trial constant, and used the same criteria to advanced each individual from one stage to the next.” | content_tags.variables[11] |
| variable | amount of reward per trial (control) — “In this study, we set the duration of behavioral sessions to 1 h for all stages of training, kept the amount of reward per trial constant, and used the same criteria to advanced each individual from one stage to the next.” | content_tags.variables[12] |
| variable | number of sessions (dependent) — “We quantified how many sessions were needed on each stage for animals to successfully learn the task” | content_tags.variables[13] |
| variable | number of trials (dependent) — “We quantified how many sessions were needed on each stage for animals to successfully learn the task” | content_tags.variables[14] |
| definition | number of trials to switch — “The number of trials to switch was calculated as the point at which the exponential fit crossed the 50% chance level performance.” | content_tags.definitions[2] |
| definition | flexible categorization — “Flexible categorization task After a series of training stages described in the next section, animals were required to perform a frequency discrimination task in which the boundary that separated high- from low-frequency sounds varied within a session.” | content_tags.definitions[3] |
| definition | valid trials — “If animals withdrew before the end of the stimulus, the trial was aborted and denoted as invalid. Only valid trials were used in calculations of the fraction of correct trials.” | content_tags.definitions[4] |
| definition | subjective category boundary — “We then fitted a logistic function to the psychometric performance and evaluated two parameters: the subjective category boundary and the slope of the curve.” | content_tags.definitions[5] |
| definition | chance level — “In contrast, mice were either at chance level or had the bias-correction mode activated, resulting in performances below chance. Note that for an average of 500 trials per session, performance below 56% would be considered at chance ( p < 0.01, binomial test), as indicated by the dotted lines in Figure 3” | content_tags.definitions[6] |
| definition | ethological validity — “Behavioral paradigms for mice should therefore be designed taking into account the ethological validity of the task requirements. Small changes in the training apparatus and protocols can make a difference. For instance, previous measurements in our lab indicated that rats may not need the first stage of training (1-DS) presented in this study, but mice benefit greatly from this conditioning stage.” | content_tags.definitions[7] |
| definition | reversing stimuli — “Previous studies from our group showed that performance for reversing stimuli (in this case, 7 kHz) is lower than for non-reversing stimuli (3 and 16.3 kHz)” | content_tags.definitions[8] |
| definition | non-reversing stimuli — “Previous studies from our group showed that performance for reversing stimuli (in this case, 7 kHz) is lower than for non-reversing stimuli (3 and 16.3 kHz)” | content_tags.definitions[9] |
| key-terms | mice, rats, model organism, flexible categorization, psychometric, subjective category boundary, contingency, valid trials, training stages, transgenic lines, ethological validity, two-alternative |
| concept | transgenic lines — not defined in this paper's own words; the mouse's main advantage, cell-type-specific access — “But the main experimental advantage of the mouse over the rat is the larger number of available transgenic lines that enable cell-type specificity in neurophysiological experiments (Huang and Zeng, 2013 ).” | concepts[11] |
| Wilcoxon rank-sum test p | as written: rank-sum test — “We found that rats learned the basic discrimination task (reaching stage 6-SB, Switch in Blocks) faster than mice ( p = 0.0014, rank-sum test).” | content_tags.statistics[0] |
| Wilcoxon signed-rank test p | — “Animals received free water on weekends resulting in a clear difference between the number of trials on the first day of the week and subsequent days for both species ( p = 0.0077, for rats and for mice, Wilcoxon signed-rank test).” | content_tags.statistics[1] |
| Spearman correlation ρ | — “There was no apparent correlation between the trials required to complete stages 5-RC (Required Correct) and 6-SB (Switch in Blocks) for either mice or rats (|ρ| < 0.12, p = 0.57, Spearman correlation).” | content_tags.statistics[2] |
| binomial test p | — “Note that for an average of 500 trials per session, performance below 56% would be considered at chance ( p < 0.01, binomial test)” | content_tags.statistics[3] |
| maximum likelihood fitting of a logistic psychometric function α, β, γ, λ | as written: a constrained maximum likelihood method was used to fit a logistic function with 4 parameters — “Briefly, a constrained maximum likelihood method was used to fit a logistic function with 4 parameters: α (the 50% threshold, or boundary), 1/β (the slope of the curve), γ (the lower asymptote), and λ (the higher asymptote).” | content_tags.statistics[4] |
| nonlinear curve fitting τ | as written: we fit an exponential function — “we fit an exponential function to the performance in response to the reversing stimulus after a contingency switch” | content_tags.statistics[5] |
| confidence interval | as written: 95% confidence intervals — “Error bars are 95% confidence intervals.” | content_tags.statistics[6] |
| standard deviation SD | as written: standard deviations — “Error bars correspond to standard deviations.” | content_tags.statistics[7] |
| rank-sum test | — “We found that rats learned the basic discrimination task (reaching stage 6-SB, Switch in Blocks) faster than mice ( p = 0.0014, rank-sum test). On average, rats reached the criterion of 80% correct for all stimuli by session 9.8 ± 0.9 (mean ± SD ), while mice took 18.1 ± 4.2 sessions.” | content_tags.statistics[8] |
| Wilcoxon signed-rank test | — “Animals received free water on weekends resulting in a clear difference between the number of trials on the first day of the week and subsequent days for both species ( p = 0.0077, for rats and for mice, Wilcoxon signed-rank test).” | content_tags.statistics[9] |
| Spearman correlation | — “There was no apparent correlation between the trials required to complete stages 5-RC (Required Correct) and 6-SB (Switch in Blocks) for either mice or rats (|ρ| < 0.12, p = 0.57, Spearman correlation).” | content_tags.statistics[10] |
| binomial test | — “Note that for an average of 500 trials per session, performance below 56% would be considered at chance ( p < 0.01, binomial test), as indicated by the dotted lines in Figure 3” | content_tags.statistics[11] |
| logistic function | — “We then fitted a logistic function to the psychometric performance and evaluated two parameters: the subjective category boundary and the slope of the curve.” | content_tags.statistics[12] |
| claim type | method — a second reading DIFFERING from the recorded value; recorded: null-result — “Our results demonstrate that mice are an appropriate model for the study of the neural mechanisms underlying adaptive decision-making, and suggest they might be suitable for other cognitive tasks as well.” | alternatives.claim_type |
| method | ten animals of each species run through seven identical training stages on one acoustic two-alternative choice task with mid-session contingency reversals — a second reading DIFFERING from the recorded value; recorded: behavioural-experiment — “Training parameters (session duration, reward amount, etc.) were set to make conditions as similar as possible for all individuals and across the two species, instead of trying to minimize the training time or maximize performance for each individual.” | alternatives.method |
| phenomenon | flexible categorization: switching between two interpretations of the same sound when the reward contingency changes without warning — a second reading DIFFERING from the recorded value; recorded: adaptive decision-making in rodents — “After a series of training stages described in the next section, animals were required to perform a frequency discrimination task in which the boundary that separated high- from low-frequency sounds varied within a session.” | alternatives.phenomenon |
Full text — tagged
The reference list is not part of the text we hold for this paper. Our copy was extracted from the publisher's page and the bibliography did not survive that extraction, so the article below ends before its references. Nothing has been removed by us.
Miceterm and ratsterm achieve similar levels of performance in an adaptive decision-making task Santiago JaramillopersName Santiago JaramillopersName 1 Cold Spring Harbor Laboratory, Cold Spring Harbor, NY, USA 2 Institute of Neuroscience and Department of Biology, University of Oregon, Eugene, OR, USA 1, 2, * , Anthony M ZadorpersName Anthony M ZadorpersName 1 Cold Spring Harbor Laboratory, Cold Spring Harbor, NY, USA 1 Cold Spring Harbor Laboratory, Cold Spring Harbor, NY, USA 2 Institute of Neuroscience and Department of Biology, University of Oregon, Eugene, OR, USA Edited by: Jonathan B. Fritz, University of Maryland, USA Reviewed by: Gidon Felsen, University of Colorado School of Medicine, USA; Carlos Brody, Princeton University, USA; Henry Heffner, University of Toledo, USA ✉ *Correspondence: Santiago JaramillopersName, Institute of Neuroscience and Department of Biology, University of Oregon, 1254 University of Oregon, Eugene, OR 97403, USA e-mail: sjara@uoregon.edu This article was submitted to the journal Frontiers in Systems Neuroscience.
Received 2014 Jul 12; Accepted 2014 Sep 1; Collection date 2014.
Copyright © 2014 JaramillopersName and ZadorpersName.
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
PMCID: PMC4167002 PMID: 25278849 Abstract Two opposing constraints exist when choosing a model organismterm for studying the neural basis of adaptive decision-making: (1) experimental access and (2) behavioral complexity. Available molecular and genetic approaches for studying neural circuits in the mouse fulfill the first requirement. In contrast, it is still under debate if miceterm can perform cognitive tasks of sufficient complexity. Here we compare learning and performance of miceterm and ratsterm, the preferred behavioral rodent model, during an acoustic flexible categorizationterm two-alternativeterm choice task. The task required animals to switch between two categorization definitions several times within a behavioral session. We found that both speciesterm·var achieved similarly high performance levels. On average, ratsterm learned the task faster than miceterm, although some miceterm were as fast as the average ratterm. No major differences in subjective categorization boundaries or the speed of adaptation between the two speciesterm·var were found. Our results demonstrate that miceterm are an appropriate model for the study of the neural mechanisms underlying adaptive decision-making, and suggest they might be suitable for other cognitive tasks as well.
Keywords: auditory, categorization, psychometricterm, context, flexibility, task switching, cognition 1. Introduction A central challenge of modern neuroscience is to understand the neural mechanisms underlying behavior. Historically, macaque monkeys have served as the preeminent non-human experimental preparation for studying complex cognitive behaviors such as attention and decision making (Desimone and Duncan, 1995 ; Parker and Newsome, 1998 ; Maunsell and Treue, 2006 ; Gold and Shadlen, 2007 ). Data derived from this preparation form the foundation for most modern theories of neural computation. Progress in primate systems neuroscience has been propelled largely by sophisticated conceptual tools for data analysis and computational modeling. However, because of the great value of individual animals, primate research is in many ways technologically conservative. These limitations have slowed the pace of progress in systems neuroscience.
Ratsterm have also been an important experimental model for studying cognition (Tolman, 1948 ; Morris, 1984 ; Dudchenko, 2004 ; Izquierdo and Belcher, 2012 ). Over the last decade, we and others have developed sensory psychophysics and decision-making paradigms for ratsterm explicitly modeled after those used in primates (Uchida and Mainen, 2003 ; Kepecs et al., 2008 ; Otazu et al., 2009 ; Erlich et al., 2011 ; JaramillopersName and ZadorpersName, 2011 ; Raposo et al., 2012 ; Znamenskiy and ZadorpersName, 2013 ; JaramillopersName et al., 2014 ). Because of their low cost—at least two orders of magnitude lower than monkeys—it is possible to conduct physiological and behavioral assays in rodents in parallel, much more efficiently and rapidly than primates. Moreover, it is straightforward to exploit many of the recent powerful optogenetic tools (Zhang et al., 2010 ) in ratsterm. However, miceterm offer further advantages over ratsterm, stemming mainly from the availability of hundreds of transgenic linesterm that enable precise targeting of gene expression to defined neuronal populations (Madisen et al., 2010 ; Taniguchi et al., 2011 ; Madisen et al., 2012 ; Gerfen et al., 2013 ). It is therefore crucial to evaluate whether miceterm can achieve the high levels of performance on cognitive tasks that has previously been demonstrated in ratsterm.
Here we compare the speed of learning and the performance levels achieved by miceterm and ratsterm in a decision-making task that tested the animals' perceptual abilities as well as their adaptability (JaramillopersName et al., 2014 ). The task required animals to categorize acoustic stimuli as high- or low-frequency sounds, and to adapt quickly after changes in the category boundary. We found that miceterm achieved high performance levels in this task similar to that of ratsterm. The distribution of required training sessions for the two speciesterm·var largely overlapped, although on average miceterm required longer training. Our results also indicate that once animals learn the task, miceterm and ratsterm adapt equally fast after a change in categorization contingencyterm·var.
We conclude that this flexible categorizationterm paradigm provides an appropriate model for studying the neural mechanisms underlying adaptive decision-making in the mouse. Moreover, this report serves as a guide for researchers wanting to implement similar complex behaviors in rodents.
2. Materials and methods 2.1. Animal subjects Animal procedures were approved by the Cold Spring Harbor Laboratory Animal Care and Use Committee and carried out in accordance with National Institutes of Health standards. Ten adult male Long Evans ratsterm (Taconic Farms) and ten adult male C57Bl/6J miceterm (The Jackson Laboratory, stock #000664) were used in this study (Table 1 ). Data from an additional set of 5 adult C57Bl/6J miceterm from a different study are also presented ( Figure 7 ). Animals had free access to food, but water was restricted. Free water was provided on days with no experimental sessions. The weight of animals was monitored at least three times a week, and supplemental water was given if an individual had a weight below 80% of its baseline.
Table 1.
Animals in the study .
Ratsterm Miceterm Strain Long evans C57Bl/6J Age 10 weeks 10 weeks Weight 285 ± 29 g 26.1 ± 1.3 g Reward/trial 24 μl 2–3 μl 2.2. Behavioral apparatus Experiments were conducted inside single-walled sound-booths with inner dimensions 28 × 26 × 32 inches (W × D × H), custom designed by Industrial Acoustics Company (Bronx, New York). Each booth contained a three-port chamber (Figure 1A ) similar to those used by Uchida and Mainen ( 2003 ); Otazu et al. ( 2009 ); JaramillopersName and ZadorpersName ( 2011 ). For this study, the center port was used for trial initiation and fixation, and the side ports were used for behavioral choices and reward delivery. Nose pokes were detected by sensing the interruption of an infrared beam across each port. Water rewards were delivered via stainless steel tubing (14 gauge for ratsterm, 18 gauge for miceterm) located in the center of each side port. The amount of water delivered was controlled by valves calibrated at least once a week. Calibration was achieved by estimating the time of valve opening necessary to deliver the desired volume per trial, by running 200 deliveries and calculating water volume by weight. The fluid pressure was not adjusted, but the flow rate changed less than 10% from the beginning to the end of a session. The volume of water reward was scaled according to the normal consumption for each speciesterm·var (Table 1 ). The behavioral system was controlled by in-house software developed in Matlab (Mathworks, Natick, MA), based on the BControl framework ( http://brodylab.princeton.edu/bcontrol ).
Figure 1.
Flexible sound-categorization task . (A) Subjects initiated each trial by poking their nose into the center port of a three-port chamber. A narrow-band sound was presented for 100 ms indicating the location of reward: left for low-frequency sounds, right for high-frequency sounds. (B) Animals were trained using 3 sound frequencies, one of which (7 kHz) changed its meaning several times in a session. After the initial training stagesterm, a behavioral session consisted of alternating blocks of 300 valid trialsterm: in one block type, the category boundary was set to a frequency above 7 kHz; in the other block type the boundary was set to a frequency below 7 kHz.
Sounds were delivered through generic electromagnetic dynamic speakers (HP 5187-2105, Harman Kardon) located on each side of the chamber, and calibrated using a free-field microphone (Type 4939, Brüel and Kjær) to produce 70 dB-SPL in the range of 1–40 kHz at the position of the subject. Measurements were performed with the booth closed and the microphone positioned at the location and orientation of the subject's left or right each when calibrating the left or right speaker respectively. The microphone was connected to a preamplifier (Type 2670, Brüel and Kjær), and signals were digitized with a National Instruments acquisition card (NI 9201) at 500,000 samples per second for analysis with custom software developed in Matlab (Mathworks). A Brüel and Kjær calibrator (Type 4231) that produces 1 kHz at 94 dB-SPL was used to verify the correct output level of the microphone/preamplifier. Waveforms were created in software at a sampling rate of 200,000 samples per second and delivered to speakers through a Lynx L22 sound card (Lynx Studio Technology). We applied rise and fall linear envelopes of 2 ms to all sounds.
2.3. Flexible categorizationterm task After a series of training stagesterm described in the next section, animals were required to perform a frequency discrimination task in which the boundary that separated high- from low-frequency sounds varied within a session. Animals initiated each trial by poking their nose into the center port of a three-port chamber (Figure 1A ). After a silent delay of random duration (250–350 ms, uniformly distributed), a narrow-band sound was presented for 100 ms. The silent delay is useful during electrophysiological experiments as a baseline (fixation) period before stimulus presentation. Animals were required to stay in the center port until the end of the sound, and choose one of the two side ports for reward according to the frequency of the sound (low-frequency: left port; high-frequency: right port). If animals withdrew before the end of the stimulus, the trial was aborted and denoted as invalid. Only valid trialsterm were used in calculations of the fraction of correct trialsterm·var. Left and right trials were fully randomized. Post-hoc analysis of the trials presented in each session verified that a strategy of always choosing one side, would yield performances below 60% on 98% of the sessions. Similarly, a strategy of alternating sides would yield performances below 60% on 99.9% of the sessions.
The acoustic stimuli were chords composed of 16 simultaneous pure tones logarithmically spaced in the range f /1.2 to f × 1.2, for a given center frequency f . The intensity of the soundsterm·var was variable in the range 50–70 dB-SPL to discourage the use of loudness to solve the task; the perceived loudness of a sound depends on frequency even when the same sound-pressure level reaches the ear. Although ratsterm and miceterm have different absolute thresholds for the frequencies used here: 0–20 dB-SPL for ratsterm (Heffner et al., 1994 ) and 10–40 dB-SPL for miceterm (Koay et al., 2002 ), all sounds were presented above threshold for either speciesterm·var. Animals were trained to discriminate between sounds centered at 3 and 7 kHz in one contingencyterm, and 7 and 16.3 kHz in another contingencyterm (Figure 1B ). These frequencies were chosen to obtain equal (logarithmic) separation between the two frequencies on each contingencyterm. A single session (1 h) consisted of several blocks of 300 valid trialsterm. The categorization contingencyterm·var changed from one block to the next without any additional cue indicating the change (besides the sound-action-reward relations). The initial contingencyterm for each session was alternated from one day to the next.
2.4. Training stagesterm Animals passed through 7 stages of training, advancing from one stage to the next once they achieved a performance criterion specific to each transition (Table 2 ). Session at all stages were 1 h long. The first two stages familiarized animals with the behavioral setup for operant conditioning: In the initial stage, 1-DS (Direct Sides), water reward was delivered from one of the two reward ports when the animal poked either on the center port or on the appropriate port for that trial (randomized from trial to trial). A sound associated with the reward port was presented (3 kHz for left, 16.3 kHz for right). Animals advanced to the next stage after the first session with more than 200 rewards. Stage 2-DC (Direct Center) was similar to 1-DS, but reward was only delivered after poking on the center port. Animals advanced to the next stage after the first session with more than 200 rewards.
Table 2.
Stage Name Goal Water delivery Criterion to advance 1-DS Direct sides Get animals to poke and collect water After center or side poke One session with 200 rewards 2-DC Direct center Trial initiation in center poke After center poke One session with 200 rewards 3-RS Require side Require response shortly after trial initiation After trial initiation and associated side poke One session with 200 trials 4-ID Increase delay Require animals to stay in center port longer After trial initiation and associated side poke One session with 200 trials, or 3 consecutive sessions with more than 100 trials 5-RC Require correct Full two-alternativeterm choice sound discrimination Only after correct side poke Two consecutive sessions with 80% correct for each stimulus.
6-SB Switch in blocks Full flexible categorizationterm task, switching every 300 trials Only after correct side poke (switches for one sound) One session with 70% correct for the reversing stimulus 7-OK Ready Ready for psychometricterm measurements Only after correct side poke (category boundary switches) The next two stages familiarized animals with the task structure and timing. In stage 3-RS (Require Side), animals were required to initiate a trial by poking in the center port, and to respond to the stimulus by poking in the reward ports. There was no delay between the center poke and the stimulus presentation. Water was only delivered after the animal poked in the appropriate reward port for that trial, but poking the other ports did not finish the trial. If animals had not collected reward after a period of 4 s, the trial finished and the animal had to initiate a new trial. Animals advanced to the next stage after performing more than 200 trials in one session. In stage 4-ID (Increase Delay), the delay between the center poke and the onset of the stimulus was increased. Initially, animals were allowed to withdraw from the center port at any time after the stimulus onset. The delay was first increased in increments of 10 ms every 10 valid trialsterm. After 50 trials it was increased to 100 ± 20 ms, and at 100 trials to 200 ± 50 ms. After 150 trials animals were required to stay in the center port until the offset of the stimulus. An early withdrawalterm resulted in the termination of the trial. Animals advanced to the next stage after performing more than 200 trials in one session, or 3 consecutive sessions each with more than 100 trials. At this point, the delay to the stimulus was increased to its final value of 300 ± 50 ms.
Stage 5-RC (Required Correct) consisted of the full frequency discrimination task (3 vs. 16.3 kHz). In this stage, reward was delivered only if the animal chose the correct reward port after leaving the center port. Error trials resulted in a timeout of 4 s during which all ports were inactive. If animals showed a bias toward one reward port, we activated a correction procedure for the following session. The method consisted of repeating the previous trial if incorrect. This correction procedure method was not activated beyond this stage. Animals advanced to the next stage after two consecutive sessions with 80% correct trials for each stimulus type.
In the next stage, 6-SB (Switch in Blocks), each session started in one of the two possible contingencies, either 3 vs. 7 kHz, or 7 vs. 16.3 kHz. Contingencies alternated every 300 valid trialsterm. The initial contingencyterm was also alternated from one day to the next. Previous studies from our group showed that performance for reversing stimuliterm·def (in this case, 7 kHz) is lower than for non-reversing stimuliterm·def (3 and 16.3 kHz) (JaramillopersName et al., 2014 ). We therefore set a lower criterion for advancing to the next stage compared to the previous transition. Animals advanced to the next stage after one session with 70% correct trials for the reversing stimuliterm·def under each of the two contingencies.
The last stage (7-OK) was achieved once animals were able to successfully switch between contingencies within a single session, and was used to estimate psychometricterm performance along a range of stimulus frequencies. In this stage, 10% of the trials contained sounds with center frequency logarithmically spaced between 3 and 16.3 kHz. The correct reward port for each stimulus was defined according to boundaries located at the geometric mean between the 3 frequencies used for training: 4.5 kHz for one contingencyterm, 10.6 kHz for the other. The fraction of stimuli of each frequency was balanced to obtain the same number of left- and right-reward trials.
2.5. Analysis of behavioral performance Data were analyzed using in-house software developed in Python ( www.python.org ). Performance traces for a single session ( Figures 5A,B ) were calculated by averaging over a moving window (40 trials long) the fraction of correct trialsterm·var for each stimulus. To calculate the number of trialsterm·var needed to switch between categorization contingencies ( Figure 5 ) we fit an exponential functionterm·stat to the performance in response to the reversing stimulus after a contingencyterm switch. We first estimated the fraction of correct choices for each trial after the switch, by averaging across 20 sessions of stage 7-OK. No time-averaging was applied (the moving window was 1 trial long). We pooled together switches from low-to-high and high-to-low boundary contingencies, and fit the function: f ( k ) = A ( 1 − e − k / τ ) + I (1) where k is the trial index after a switch, and 1 − I is the performance at the end of the previous block, estimated from the last 100 trials before a switch. The parameters A and τ are related to the asymptotic performance and the speed of change in performance respectively. The number of trials to switchterm·def was calculated as the point at which the exponential fit crossed the 50% chance levelterm performance.
Psychometricterm curves were fit using the Python module Psignifit 3.0 (Fründ et al., 2011 ). Briefly, a constrained maximum likelihood method was used to fit a logistic functionterm·stat with 4 parameters: α (the 50% threshold, or boundary), 1/β (the slopeterm of the curve), γ (the lower asymptote), and λ (the higher asymptote).
Ψ ( x ) = γ + ( 1 − γ − λ ) 1 1 + exp ( − g ( x ) ) g ( x ) = x − α β (2) 3. Results To compare performance, speed of learning, and adaptability between miceterm and ratsterm, we used a sound categorization task in which the category definitions changed within a single behavioral session (Figure 1 ). Animals had to report if a target sound (100 ms long) was of high or low frequency by poking in either the left or right reward port after trial initiation. The boundary that defined what sounds belonged to the low- or high-frequency categories varied between two possible values every 300 trials.
We first evaluated how fast 10 individuals of each speciesterm·var advanced through the different stages of training (Table 2 ), and then compared the final performance across the two cohorts. Animals were trained in 1 h sessions every week-day and the amount of reward was scaled according to the normal consumption for each speciesterm·var (Table 1 ). Training parameters (session duration, reward amount, etc.) were set to make conditions as similar as possible for all individuals and across the two speciesterm·var, instead of trying to minimize the training time or maximize performance for each individual.
3.1. Ratsterm learned the task faster than miceterm We defined a fixed criterion to advance from one training stageterm to the next (Table 2 ), and animals were required to perform at least one full session on each stage. We quantified how many sessions were needed on each stage for animals to successfully learn the task (Figures 2A,C ). We found that ratsterm learned the basic discrimination task (reaching stage 6-SB, Switch in Blocks) faster than miceterm ( p = 0.0014, rank-sum testterm·stat). On average, ratsterm reached the criterion of 80% correct for all stimuli by session 9.8 ± 0.9 (mean ± SD ), while miceterm took 18.1 ± 4.2 sessions. From all stages, stage 5-RC (Required Correct), in which animals learned to discriminate sounds, showed the largest difference between miceterm and ratsterm (7 sessions on average, p = 0.0003, rank-sum testterm·stat). This stage was responsible for most of the difference in the overall length of training. One mouse never learned the discrimination task, although this animal showed no impairment in initiating trials or collecting reward. One ratterm never reached the final stage (7-OK). These animals were excluded from subsequent analysis.
Figure 2.
Ratsterm learned the task faster than miceterm . (A) Training stageterm as a function of number of sessionsterm·var for each ratterm. One ratterm never reached the last stage. (B) Training stageterm as a function of the number of valid trialsterm (those in which the animal waited long enough for the stimulus). (C) Same as ( A) for each mouse. One mouse never reached stage 6-SB. (D) Same as (B) for each mouse. (E) Number of sessionsterm·var needed to learn to switch between contingencies (stage 6-SB), after the basic discrimination was learned (stage 5-RC). (F) Number of trialsterm·var needed to learn to switch, after the basic discrimination was learned. (G) Overall number of sessionsterm·var to reach the last stage of training. (H) Overall number of valid trialsterm to reach the last stage of training. Black vertical lines indicate the average for each speciesterm·var.
In stage 6-SB (Switch in Blocks), animals learned to switch between two categorization contingencies within a single behavioral session. The average number of sessionsterm·var spent learning to switch (after learning to discriminate sounds) was less for ratsterm than for miceterm (Figure 2E ), although this difference was not statistically significant ( p = 0.07, rank-sum testterm·stat). Overall, ratsterm reached the final stage (7-OK) faster than miceterm (Figure 2G ) ( p = 0.0014, rank-sum testterm·stat), although the distributions largely overlapped and some miceterm were as fast as the average ratterm.
We also quantified the number of trialsterm·var needed on each stage (Figures 2B,D ) and found results consistent with our analysis of the number of sessionsterm·var. Stage 5-RC (Required Correct), showed the largest difference between speciesterm·var (ratsterm required 2320 fewer trials on average, p = 0.0013, rank-sum testterm·stat). Moreover, ratsterm achieved the final stage in fewer trials than miceterm (Figure 2H , p = 0.0054, rank-sum testterm·stat), although the number of trialsterm·var spent learning to switch was not significantly different (Figure 2F , p = 0.057, rank-sum testterm·stat). There was no apparent correlation between the trials required to complete stages 5-RC (Required Correct) and 6-SB (Switch in Blocks) for either miceterm or ratsterm (|ρ| < 0.12, p = 0.57, Spearman correlationterm·stat).
3.2. Ratsterm and miceterm achieved similar performance levels We quantified changes in performance levels as animals learned the task. Figure 3 shows the overall performance for each training session for each animal across several stages. The plots illustrate how, as described in the previous section, ratsterm achieved high performance levels faster than miceterm on average. Although both speciesterm·var achieved a similar fraction of correct trialsterm·var per session (Figure 3C ), miceterm showed a slightly higher average performance on the last stages ( p < 0.05 rank-sum testterm·stat, for several consecutive sessions after reaching stage 6-SB, Switch in Blocks). This observation will be discussed in more detail in the next section when comparing psychometricterm slopesterm between the two speciesterm·var. Each point in the plot represents the average for each entire session, including transition trials between blocks, and trials with intermediate frequencies on psychometricterm sessions. Performance for the easiest frequencies (3 and 16.3 kHz) was, as expected, higher than these values: above 80% for all animals, and above 90% for most.
Figure 3.
Ratsterm and miceterm achieved similar levels of performance . (A) Percentage of correct trials (ignoring early withdrawalsterm) as a function of number of sessionsterm·var. Dotted lines indicate chance levelterm range (assuming 500 trials, p = 0.01). The gray bar indicates when animals were required to switch between contingencies (not all animals reached that stage at the same time). The plot includes only sessions from stage 5-RC or after (the first stage that required correct responses on the first attempt). Some ratsterm start this stage with performance above chance. The ratterm that did not reach the final stage is not included. (B) Same as (A) for miceterm. The mouse that did not reach the final stage is not included. Some miceterm had bias-correctionterm activated and show performance below chance. (C) Average performance across animals of each speciesterm·var. Miceterm achieved performance level as high as ratsterm, but after more training sessions. Error bars correspond to standard deviationsterm·stat. This figure includes all trials (easy and difficult) in each session. Final performance computed for the easiest frequencies alone is higher than that plotted here: above 80% for all animals, and above 90% for most.
As expected, there was a decay in performance at the transition between the frequency discrimination stage (5-RC, Required Correct) and the contingencyterm switching stage (6-SB, Switch in Blocks), illustrated with a gray horizontal bar in Figures 3A,B . This is visible in the average curve for ratsterm around session 9 (in Figure 3C ). Miceterm showed a similar change in performance at this transition, although this feature in smoothed out in the average plot because the transition session was highly variable across miceterm. Figure 3 only includes sessions after animals were required to make the correct choice immediately after stimulus presentation (stage 5-RC and after). Some ratsterm had already learn to associate sounds with the appropriate reward port and show performance above chance levelterm at the beginning of the plot. In contrast, miceterm were either at chance levelterm or had the bias-correctionterm mode activated, resulting in performances below chance. Note that for an average of 500 trials per session, performance below 56% would be considered at chance ( p < 0.01, binomial testterm·stat), as indicated by the dotted lines in Figure 3 .
Physiological studies of the mechanisms underlying complex behaviors often require hundreds of trial in a single experimental session. Our measurements show that animals perform several hundred valid trialsterm per session (Figure 4 ), enabling a quantification of changes in the response properties of neurons as animals switch between contingencies. Animals received free water on weekends resulting in a clear difference between the number of trialsterm·var on the first day of the week and subsequent days for both speciesterm·var ( p = 0.0077, for ratsterm and for miceterm, Wilcoxon signed-rank testterm·stat). This effect was less pronounced after the first day: from first to second day ratsterm increased the number of trialsterm·var by 72 ± 37; from second to fourth day only by 25 ± 31 (Figure 4A ). Similarly for miceterm, the first increase was 194 ± 116, while the second was 101 ± 68 (Figure 4B ).
Figure 4.
Ratsterm and miceterm performed a similar number of trialsterm·var per session . (A) Average valid trialsterm per session for each ratterm on different days after having free water. Animals performed fewer trials on the first day. Each dot is the average across 4 days from different weeks. (B) Same as (A) for miceterm. (C) Comparison between miceterm and ratsterm (data from day 4 after free water). Average number of trialsterm·var is similar between speciesterm·var, but variability is higher across miceterm than across ratsterm. (D) Percentage of trials in which each animal incorrectly withdrew from the center port before the offset of the sound. Data from the first 4 sessions of stage 7-OK for each animal. (E) Percentage of trials in which each animal chose no reward port after the target was presented. Data from the first 4 sessions of stage 7-OK for each animal.
We found no statistically reliable difference in the average number of trialsterm·var per session between miceterm and ratsterm for the reward amounts used in our study ( p = 0.12, rank-sum testterm·stat), although the variability across animals was larger in the cohort of miceterm (Figure 4C ). Previous measurements in our laboratory have shown that ratsterm can perform more trials than those presented here if behavioral sessions are longer than 1 h.
Animals were required to stay in the center port during an initial silent delay (250–250 ms) plus the whole duration of the target sound (100 ms). Figure 4D shows the fraction of trials in which animals left the center port before the offset of the sound, for the first 4 session of the final stage, 7-OK. Early withdrawalsterm were about 20% on average for both miceterm and ratsterm ( p = 0.56, rank-sum testterm·stat), although variability was higher for miceterm. Analysis of subsequent sessions showed that the average for ratsterm remained in the range 15–20%, while the average for miceterm fluctuated in the range 20–30%. The fraction of trials in which animals listened to the whole duration of the target sound but did not make a choice was less than 1% for all animals except one mouse (Figure 4E ). The average was slightly lower for ratsterm ( p = 0.007, rank-sum testterm·stat). Averages remained under 1% on subsequent sessions.
3.3. Ratsterm and miceterm switched equally fast between categorization contingencies In the final task, the category boundary switched between two possible values from one block of 300 trials to the next. As a result, animals were required to associate a different reward port to the middle frequency sound (7 kHz) on each block. Nine out of ten animals from each speciesterm·var achieved performance levels above 70% for this reversing stimulus on both contingencies (Figure 2 ). Performance for the non-reversing stimuliterm·def (3 and 16.3 kHz) was always above this level. Figures 5A,B show an example session from one ratterm and one mouse respectively, as they switch between contingencies (performance for each sound-action association is shown in a different color). These examples show how animals rapidly modified their response to the 7 kHz after a contingencyterm change. We estimated the speed of switching to a new sound-action association by averaging performance for each trial from several sessions, and calculating when an exponential fit crossed the 50% chance performance level (Figures 5C,D ). We found that both ratsterm and miceterm needed about 10 trials to switch between contingencies (Figure 5E , p = 0.72, rank-sum testterm·stat).
Figure 5.
Ratsterm and miceterm switched equally fast between contingencies . (A) Example of the performance of a ratterm (one session) as the contingencyterm changes. (B) Example session for a mouse. (C) Performance of one ratterm after the contingencyterm switch. In gray is the average for each trial across several sessions. In black is an exponential fit. The dotted line indicates when performance crosses chance levelterm. (D) same as (C) for one mouse. (E) Comparison of speed of switching for ratsterm and miceterm. Each dot corresponds to the number of trialsterm·var it took each animal to cross chance levelterm after a switch.
We also estimated the subjective categorization boundary from each individual under each contingencyterm (Figure 6 ). We first quantified performance in response to several sound frequencies in addition to the three stimuli used for training, interleaved throughout the session. We then fitted a logistic functionterm·stat to the psychometricterm performance and evaluated two parameters: the subjective category boundaryterm and the slopeterm of the curve. Both sets of animals showed a clear change in subjective category boundaryterm between the two contingencies (Figures 6C,E , p = 0.0077 for both ratsterm and miceterm, Wilcoxon sign-rank test), but no consistent change in slopeterm ( p = 0.07 for ratsterm, p = 0.37 for miceterm, Wilcoxon sign-rank test). The change in subjective category boundaryterm as animals switched from one contingencyterm to the other was not different between miceterm and ratsterm ( p = 0.89 rank-sum testterm·stat), but miceterm showed slightly steeper slopesterm than ratsterm ( p = 0.019 on the low-boundary block, p < 0.001 on the high-boundary block).
Figure 6.
Ratsterm and miceterm had similar subjective categorization boundaries . (A) Example psychometricterm curves estimated for one ratterm on each contingencyterm. Data pooled from several sessions. Error bars are 95% confidence intervalsterm·stat. (B) Same a (A) for one mouse. (C,D) Subjective categorization boundary and slopeterm of the psychometricterm curve for all ratsterm. The black line corresponds to the example animal in (A) . (E,F) Same as (C,D) for miceterm. Slopesterm were higher for miceterm, but subjective boundaries were the same for both speciesterm·var.
3.4. C57BL/6J miceterm can perform the task after several hundred days An important consideration when working with miceterm is that some strains suffer from early loss of hearing sensitivity (Zheng et al., 1999 ). C57Bl/6J, a common background for transgenics used in the study of neural circuits, show elevated auditory brainstem response thresholds at 700 days of age (60 dB higher than normal), but not at 200 days of age. For the study of auditory behaviors, it is therefore necessary to test if hearing loss affects performance in the time-scale of the experiments. We found that C57Bl/6J miceterm can perform the auditory flexible categorizationterm task for several hundred days without major impairments in performance. Figure 7 shows psychometricterm curves for each contingencyterm for a 300-day old mouse, as well as categorization boundaries and psychometricterm slopesterm of 4 more animals of similar age. These results are comparable to those from our initial cohort of miceterm (Figures 6E,F ).
Figure 7.
Miceterm continue to perform the task for many months . (A) Example psychometricterm curve for a mouse older than 300 days, showing that any age-induced deafness in the C57Bl/6J strain has not produced any visible effect at this age. (B,C) Subjective categorization boundary and slopeterm of the psychometricterm curve for 5 miceterm older than 300 days. The black line corresponds to the example animal in (A) . Values are similar to those from younger miceterm.
4. Discussion Our study quantified the performance of miceterm and ratsterm in an adaptive decision-making task. The flexible sound categorization task used here required animals to switch between two possible interpretations of a stimulus in order to successfully collect reward. We compared three main behavioral features between the two speciesterm·var: (1) how long it took animals to learn the task, (2) how well they adapted between categorization contingencies, and (3) how fast they adapted between contingencies. We found that under our training conditions: (1) ratsterm were faster than miceterm at learning the task, although the distributions overlapped; (2) both miceterm and ratsterm successfully changed their subjective categorization boundaries after changes in contingencies; and (3) both speciesterm·var adapted equally fast between contingencies. Although both cohorts performed the task well, measurements after the last stage of training showed slightly higher overall performance levels and steeper psychometricterm slopesterm for miceterm than for ratsterm.
In this study, we set the duration of behavioral sessionsterm·var to 1 h for all stages of training, kept the amount of reward per trialterm·var constant, and used the same criteria to advanced each individual from one stage to the next. Consistency in training parameters helped us compare directly the two speciesterm·var, the main goal of the study. Other objectives, for example minimizing training times for each animal, may require adjusting parameters for each individual.
4.1. Rodents in the study of the neural basis of adaptive behavior There are opposing constraints in the choice of a model organismterm for the study of the neural basis of complex behaviors: researchers need animals that allow for a high degree of experimentation, and yet be capable of performing complex cognitive tasks. Studies of the physiological basis of decision-making have traditionally been carried out on non-human primates (Parker and Newsome, 1998 ; Gold and Shadlen, 2007 ), motivated by the similarity of their brain anatomy to that of humans. Smaller mammals, rodents in particular, provide several advantages over primates for physiological experiments: faster breeding times, lower costs, smaller housing needs, and currently, easier access to molecular and genetic tools. And although rodents may not have the cognitive abilities of primates, they can perform a variety of adaptive decision-making tasks (Dudchenko, 2004 ; JaramillopersName and ZadorpersName, 2011 ; Bissonette and Powell, 2012 ; Izquierdo and Belcher, 2012 ; JaramillopersName et al., 2014 ).
Rodents have been used for many decades in studies of learning and sensory perception. However, several of the behavioral paradigms employed required subject handling between trials (affecting among other things, the number of trialsterm·var achievable per session), or measurements that relied on the precision of a human observer (when estimating for example freezing times). Several strategies have been implemented to minimize these issues. In particular, automatic systems have been developed allowing animals to initiate trials without intervention of the experimenter and enabling automatic scoring (Wahlsten et al., 2003 ; Schaefer and Claridge-Chang, 2012 ). These systems permit the collection of hundreds of trials per session, a feature often required in physiological experiments. In addition, the use of quantitative behavioral assays allow for detailed characterization of performance and, in turn, a deeper investigation of potential mechanisms underlying decision-making. These advances enable performing in rodents similar psychophysical measurements to those classically studied in humans. The paradigm described in this manuscript is an example of such quantitative characterization of behavior.
Animals could use various strategies for solving the task described in this study. First, animals may evaluate the outcome of responses to the reversing frequency (7 kHz) to update their sound-action associations. Alternatively, the appearance of an extreme stimulus (3 or 16.3 kHz) could be used as a cue indicating the most likely contingencyterm. These strategies are non-exclusive, and animals may be taking advantage of both. Independent of the strategy used, however, animals consistently change the action associated with a subset of stimuli (in particular, 7 kHz), and the observation that animals generalize to other frequencies is well captured by a model in which an internal category boundary shifts between blocks of trials (JaramillopersName et al., 2014 ). Similarly, there are at least two possible mechanisms for this adaptation in behavior: (1) changes in synaptic strength in the sensory-motor pathway that result in different behavioral responses for the same sound; or (2) integration of sensory and context information without long-term synaptic changes in the sensory-motor pathway. The former correspond to mechanisms hypothesized for learning and memoryterm (Martin et al., 2000 ), the latter to mechanisms of selective attentionterm (JaramillopersName and Pearlmutter, 2007 ), both defining features of cognition. Although behavioral measurements alone may not dissociate between these possibilities, techniques for physiological analysis of neural circuits in the mouse have the potential to reveal the mechanisms that mediate this rapid adaptation in the interpretation of sounds.
4.2. Miceterm vs. ratsterm The ratterm has been the preferred rodent model in the study of the neurophysiological basis of behavior (Buzsàki et al., 1989 ; Dudchenko, 2004 ; Izquierdo and Belcher, 2012 ). The mouse, in contrast, has been favored when genetic manipulations are required (Crawley, 2008 ). When choosing between ratterm and mouse for investigating the neural basis of complex behaviors one needs to take into account several factors. Because of their size, ratsterm can accommodate larger chronically implanted devices than miceterm. It is also easier to target specific regions in animals with larger brains. Miceterm, being smaller, are less expensive to house and less time consuming (e.g., during tissue processing). Their size also makes miceterm better suited for experiments that require covering a larger extent of brain tissue, for example when delivering light for optogenetic experiments or trying to reach deeper cortical layers with two-photon imaging. But the main experimental advantage of the mouse over the ratterm is the larger number of available transgenic linesterm that enable cell-type specificity in neurophysiological experiments (Huang and Zeng, 2013 ). Although, there are now techniques for manipulating the genome of both speciesterm·var (Filipiak and Saunders, 2006 ; Witten et al., 2011 ), the mouse has had a long head start as illustrated by the thousands of transgenic linesterm available from The Jackson Laboratory repository.
Despite these advantages of mouse over ratterm, it is still debated whether the mouse can perform the type of adaptive decision-making tasks that have been successful in studies with ratsterm (Carandini and Churchland, 2013 ). Most comparisons of behavioral performance between miceterm and ratsterm, including our study, have used tasks originally devised for ratsterm. This potential confound may explain why ratsterm have been found to be better in a variety of tasks (Whishaw and Tomie, 1996 ). Behavioral paradigms for miceterm should therefore be designed taking into account the ethological validityterm of the task requirements. Small changes in the training apparatus and protocols can make a difference. For instance, previous measurements in our lab indicated that ratsterm may not need the first stage of training (1-DS) presented in this study, but miceterm benefit greatly from this conditioning stage.
Miceterm and ratsterm have been directly compared during memory-guided navigation tasks (Whishaw and Tomie, 1996 ; Frick et al., 2000 ; Cressant et al., 2007 ) and tasks that require sensory-driven decisions (Prusky et al., 2000 ; Mayrhofer et al., 2013 ). Behavior of the two speciesterm·var was similar in the navigation tasks when evaluated in the dry-land maze. Differences found in water-maze tasks seem to be accounted for mostly by differences in swimming rather than memory capabilities. Nevertheless, the two speciesterm·var seem to use different strategies for solving some of these tasks, and whenever differences in performance levels were found, results favored the ratterm. Similar conclusions have been reached for sensory-driven decision tasks. In a two-alternativeterm forced choice paradigm for vibrotactile discrimination, the performance of miceterm and ratsterm was found to be very similar when quantifying psychometricterm curves, reaction times, learning rates, and trial omissions (Mayrhofer et al., 2013 ). In a visual discrimination task, both speciesterm·var learned the task well, although ratsterm showed higher visual acuity and (at least in the examples reported) learned the task faster (Prusky et al., 2000 ).
In addition to these direct comparisons, researchers have evaluated behavioral performance of miceterm during tasks originally devised for ratsterm. These studies have shown that miceterm are capable of performing tasks requiring delayed matching-to-position (Goto et al., 2010 ), visuospatial attention (Humby et al., 1999 ), attentional shifts (Bissonette et al., 2008 ), or visual reversals (Izquierdo et al., 2006 ). Importantly, researchers are now using tasks that make possible detailed psychometricterm measurements from miceterm (Prusky et al., 2000 ; O'Connor et al., 2010 ; Busse et al., 2011 ; Tai et al., 2012 ; Glickfeld et al., 2013 ).
The mouse is uniquely placed at the interface between experimental access and behavioral complexity, making it an ideal model for the study of adaptive decision-making. Successful behavioral paradigms, however, rely on targeting designs to the idiosyncrasies of the mouse from the outset, rather than simply assuming that miceterm are little ratsterm.
Author contributions Santiago JaramillopersName and Anthony M. ZadorpersName designed the study. Santiago JaramillopersName collected and analyzed the data. Santiago JaramillopersName and Anthony M. ZadorpersName wrote the paper.
Funding This work was supported by the National Institutes of Health (Grants R01DC012565 and R01NS088649) and the Swartz Foundation.
Conflict of interest statement The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Acknowledgments We thank B. J. Burbach and Rigina Gallagher for assistance with preliminary experiments.