Age affects pigeon memory capacity, but not representation of serial order, during a 1 locomotor sequential-learning task

28 Aging affects individuals of every species, with sometimes detrimental effects on 29 memory and cognition. The simultaneous-chaining task, a sequential-learning task, requires 30 subjects to select items in a predetermined sequence, putting demands on memory and 31 cognitive processing capacity. It is thus a useful tool to investigate age-related differences in 32 these domains. Pigeons of three age groups (young, adult and aged) completed a locomotor 33 adaptation of the task, learning a list of four items. Training began presenting only the first 34 item; additional items were added, one at a time, once previous items were reliably selected 35 in their correct order. Although memory capacity declined noticeably with age, not all aged 36 pigeons showed impairments compared to younger pigeons, suggesting that inter-individual 37 variability emerged with age. During a subsequent free-recall memory test, when all trained 38 items were presented alongside novel distractor items, most pigeons did not reproduce the 39 trained sequence in the absence of reinforcement. During a further forced-choice test, when 40 pigeons were given a choice between only two of the trained items, all three age groups 41 showed evidence of an understanding of the ordinal relationship between items by choosing 42 the earlier item, indicating that complex cognitive processing, unlike memory capacity, 43 remained unaffected by age. 44

As we age, so does our brain, and with it many cognitive functions decline. Age-related degeneration of cognitive and memory capacities is well-documented for humans, and is 49 evident even for healthy individuals 1,2 . The impact of aging on nonhuman animal cognition 50 is less well explored 3 , but this knowledge is important for understanding common effects of 51 non-pathological aging on cognition. The aging process affects every living being regardless 52 of species, but the way in which it manifests in cognitive changes or impairments can 53 sometimes differ greatly among individuals. Establishing how, and to what degree, aging can 54 affect vital cognitive capacities is a first step in assessing the range of normal and abnormal 55 impairments. 56 The simultaneous-chaining task 4,5 is cognitively demanding as it requires subjects to 57 reproduce a list of items in a specific sequence, with the only feedback provided regarding 58 the correctness of a choice being the continuation of the trial. This task puts demands on an 59 individual's memory capacity, in terms of both reference memory to learn the sequence, and 60 working memory to update the last choice made in order to determine the next required 61 response. An individual's memory capacity can be measured through the successive chaining 62 of subsequent sequence items 6 . Each time a subject learns to respond correctly to a sequence 63 of n items, another item is added to the chain (n+1). Using this method, pigeons have 64 successfully been trained to reproduce lists of four to five items 4,5 . 65 In addition to the task's adoption for evaluating memory capacity, it has also been 66 selected as an assessment tool for cognitive capacity, and as such has been used to study a 67 diverse range of species, from humans and apes to pigeons and chickens 4 . As suggested by 68 the task name, successful acquisition of the sequence can be achieved through simple 69 associative chaining, with a response to the first item serving as a cue to respond to the 70 second item, and so on. However, a more cognitively complex solution would be to form a 71 mental representation of the item order, whereby each item in the sequence is acquired not 72 groats, red milo, yellow popping corn, and white millet, at 90% of their free-feeding weight 124 and had ad lib access to water and grit inside their cage. The adult and aged pigeons had 125 previous testing experience, but all were naïve to the specific procedures of this experiment. 126

Ethical Note 127
This study was approved by the University of Manitoba's Local Animal Use Committee 128 (protocol number F18-042) in accordance with the Canadian Council on Animal Care and the 129 ARRIVE guidelines. 130

Apparatus 131
Training and testing sessions were conducted in an enclosed rectangular arena made 132 of Styrofoam ® . The walls (60 cm tall) were lined with white Con-Tact ® paper to permit 133 cleaning. The base (200 x 100 cm length x width) was covered with black rubber flooring. 134 White shower curtains enclosed the arena to block access to external visual cues. Six feeders 135 were placed in a circular array inside the arena and fixed to the floor with Velcro (see Figure  136 1). Each feeder consisted of a wooden base (10 x 10 x 10 cm length x width x height) and a 137 wooden ramp covered with sandpaper (10 x 20 cm base with a 26.6° incline) leading up to 138 the base. A plastic cup (6 cm diameter) was attached to the inside of the base. To make the 139 inside of the feeder only visible from the top of the ramp, the three sides of the base facing 140 away from the ramp were lined with white paper (30 cm height), and a white foam cone was 141 attached to the top of the paper barrier. The white cones could be replaced by equally-sized 142 cones of different colours to make each feeder visually distinct (herein these coloured cones 143 are referred to as features). A centrally-mounted Logitech HD Webcam C270, connected to a 144 Dell desktop computer (running 64-bit Windows 10 Enterprise) in the adjacent room, was 145 suspended from the ceiling for recording of trials. Two Conair ® white noise generators were 146 placed external to the arena, at opposing corners to mask sounds. 147

General Procedure 152
A pigeon was transported from the colony room to the procedure room in an opaque 153 white transport container, which also served as a holding container between trials. At the 154 beginning of each trial, the lights in the procedure room and arena were extinguished, and 155 the pigeon was removed from the holding container and placed into the arena at the starting 156 position (see Figure 1). The experimenter exited the procedure room and remotely 157 illuminated the procedure room lights, which delimitated the start of a trial. Once the pigeon 158 completed the choice requirement (see below), or ten minutes elapsed, whichever occurred 159 first, the lights were extinguished, and the pigeon was removed from the darkened arena and 160 returned to the holding container where it remained while the arena was prepared for the 161 next trial. 162

Feeder Training Phase 163
To train the pigeons to retrieve maple peas from the feeders, a single feeder with a 164 white cone (herein referred to as a white feeder) was located in one of the six possible array 165 locations (with location counterbalanced across trials). Five maple peas were placed at each 166 of the following locations: in the plastic cup inside the feeder, on the ramp, and on the floor 167 of the arena at the base of the ramp. With training, the number of maple peas was gradually 168 reduced from fifteen to two, and only placed inside the feeder. Pigeons were given a 169 maximum of 60 minutes to consume the maple peas. A trial ended 30 seconds after the last 170 maple pea was consumed, after which the arena was re-set and a new trial started. As each 171 daily session lasted a maximum of 60 minutes, a pigeon could receive between one and twelve 172 trials per daily session. Pigeons experienced trials with a single feeder until all maple peas 173 were consumed within 5 minutes of trial start, for a minimum of 12 trials in total. Next, maple pea inside each feeder. Pigeons progressed to the Memory Training phase once every 176 feeder was visited within the first 5 minutes of a trial for 5 consecutive trials. 177

Memory Training Phase 178
A daily training session consisted of 10 trials, with the six feeders positioned in the 179 circular array formation. Across four distinct stages, the pigeons were trained to visit four of 180 the six feeders in a specific sequence, as indicated by distinctly coloured features (red, green, 181 blue, yellow, orange and purple; herein referred to as a sequence item). The order in which 182 the feeders had to be visited was consistent for each pigeon, but pseudo-counterbalanced 183 among the pigeons such that one of the six colours was assigned to be the first item in the 184 sequence for two pigeons. The location of each sequence item within the array was 185 randomised across trials so that only the colours on the feeders, but not the feeder locations 186 within the arena or relative to each other, indicated the correct sequence. 187 Originally, the stages described below were administered with all six coloured 188 features placed on the feeders. However, although all subjects quickly passed the first 189 training stage, during which only the first item in the sequence was reinforced, acquisition of 190 any following memory items appeared hindered by this procedure. In particular, following a 191 successful visit to the first item, the subjects stopped exploring the arena and were reluctant 192 to visit any additional feeders, as visiting them had previously not been reinforced. For the 193 majority of subjects, the transition to training stage 2 resulted in reduced attention to the 194 previously learned first item. Following ten sessions of continued decline in performance by 195 all subjects, we revised the training procedures to encourage exploration of the area after the 196 first item had been visited. Thus, the entire training phase was restarted from stage 1 for all 197 subjects as follows.
presented with five white feeders and the first sequence item. Each successive training stage 200 differed from the previous in that the next new sequence item replaced one of the white 201 feeders in the array. Whereas the white feeders were always non-reinforced, during TS1, four 202 maple peas were placed into the first sequence item; the pigeon was allowed to inspect any 203 feeder. During training stage 2 (TS2; Figure 2B), three maple peas each were placed into the 204 first and second sequence items. If the pigeon approached any feeder before visiting the first 205 sequence item, the lights were extinguished immediately to indicate an incorrect response 206 and the trial ended. If the pigeon successfully visited the first sequence item, it could inspect 207 any other feeder. During training stage 3 (TS3; Figure 2C), two maple peas each were placed 208 into the three sequence items. Only after the pigeon had approached the first and second 209 sequence items, it was allowed to inspect any feeder. During training stage 4 (TS4; Figure  210 2D), two maple peas each were placed into the four presented sequence items. Only after 211 visiting the first, second and third items in sequence, the pigeon was allowed to freely inspect 212 any feeder. During each stage, a trial ended with the extinction of the lights after the pigeon 213 consumed all available peas and exited the final feeder, or if the pigeon visited the feeders in 214 any other order than described, or after ten minutes of inactivity. Training progressed to the 215 next stage once a pigeon consistently visited all presented sequence items in the correct order 216 without visiting any other feeders in at least seven out of ten trials of a session, for two 217 consecutive sessions. If a pigeon failed to advance from one training stage to the next within 218 60 sessions, it was considered to have reached the maximum number of sequence items it 219 could memorise, and proceeded to the Testing phase. presenting all sequence items and novel items together. Each session started with at least two 243 reinforced trials, which were identical to training trials during the final stage a pigeon had 244 reached. Following two correct baseline trials, a non-reinforced test trial was administered, 245 during which all six (sequence and novel) items in a predetermined, randomised order. Test 246 trials ended after the pigeon made four visits, including repeat visits to the same feeder. 247 Memory test sessions were continued until a pigeon had completed ten test trials. Due to the 248 criterion of two correct baseline trials preceding any test trial, each session could contain 249 between zero and three test trials. were conducted as t-tests. A simple-effect analysis was conducted to investigate the presence 267 of age-group differences within training phases. Lastly, a Kruskal-Wallis non-parametric one-268 way ANOVA was performed to assess potential age-group differences in the highest daily 269 performance score achieved during the first half of training in TS4. 270 Analyses of test performance were carried out separately for the two groups of 271 pigeons for which sequence item 3 or 4 had been the last trained item. For the Choice Test, 272 choices between two previously trained items were assessed by conducting Friedman non-273 parametric repeated-measures ANOVAs on the percentage of trials during which a subject 274 chose the item that had appeared earlier in the sequence, for each combination of trained 275 sequence items. Choices between a trained and a novel item were assessed by conducting 276 Friedman non-parametric repeated-measures ANOVAs on the percentage of trials during 277 which a subject chose a previously trained item over a novel item. To meaningfully analyse before visiting any other item, to choose at least two sequence items in their correct order 281 (i.e., 1 followed by 2, 2 followed by 3, or 3 followed by 4, regardless of which item was chosen 282 before or after this isolated combination of two items), or to choose the final trained item last. 283 To obtain a measure of progress for TS4, we evaluated the highest daily performance 317 score (correctly completed trials out of the ten daily trials) obtained halfway through the 318 training stage, after 30 sessions of TS4 had been completed. This allowed us to include the 319 data from the two young subjects that had to cease training before completing all 60 sessions 320 of TS4. The analysis revealed that there were no statistically significant differences between 321 the three age groups in this early performance level (Kruskal-Wallis test: χ²2 = 1.56, P = 322 0.459). 323 In summary, increasing the number of sequence items with each training stage led to 324 an increase in the number of sessions required to acquire the sequence. Age differences were 325 only notable for TS3, as young subjects were able to successfully complete this stage within 326 fewer sessions than adult or aged subjects. 327  Table 1 shows the percentage of trials during which a subject chose the item that had 343 appeared earlier in the sequence for each combination of trained sequence items. For the 344 subjects experiencing up to TS3, although performance was numerically worse for choices 345 between items 2 and 3, there was no statistically significant difference in performance 346 between choices (Friedman test: χ²2 = 2.6, P = 0.273). However, for the four subjects that had 347

360
In addition to the above, we examined pigeons' overall encoding of the identity of 361 trained sequence items by examining their choices when presented with a trained sequence 362 item and a novel item. Table 2  Conover) confirmed that the choice of item 1 was significantly higher than choices of items 3 370 or 4 (comparisons of "known option was item 1" to either other option: both P ≤ 0.029; 371 comparison of "known option was item 1" to "known option was item 2": P = 0.071; all other 372 comparisons: P ≥ 0.48). This result indicates that, when trained with four items, the 373 preference for a known item over a novel item decreased the later the item appeared in the 374 trained sequence. 375 376 377 Table 2. Percentage of Choice Test trials during which a subject chose the known item when the choice was between a known (sequence items 1, 2, 3, or 4, as applicable) and an unknown item (sequence items 5, 6, or 4, as applicable), or between two unknown items. Individuals are listed in order of the number of items they were trained on; those that only completed up to TS3 were not presented with sequence item 4 as a known option. As seen in Table 3, the overall tendency to visit items in the correct sequence was low 381 first" to any other option: P ≤ 0.001), and lowest for choosing sequence item 3 after item 2 391 and for avoiding the last item until the ultimate choice (comparison of "2 chosen after 1" to 392 "3 chosen after 2" and to "3 chosen last": both P < 0.001; comparison of "3 chosen after 2" to 393 "3 chosen last": P = 1.0). Similarly, performance was significantly affected by the position of choosing later chunks (comparisons of "1 chosen first" to any other option: Durbin-Conover 397 tests: all P ≤ 0.016). Pigeons also completed the chunk "2 chosen after 1" significantly more 398 often than "3 chosen after 2" and "4 chosen last" (Durbin-Conover tests: both P ≤ 0.036; any 399 other comparisons: P ≥ 0.16). Taken together, all pigeons were decreasingly likely to 400 complete a chunk the later it appeared within the sequence. This decrease was visible as early 401 as the second sequence item, as predominantly only the first item was visited correctly. 402 403  Table 3. Percentage of Memory Test trials (out of ten trials) during which a subject chose sequence item 1 before any other items ("1 first"), chose sequence item 2 directly after item 1 ("1 → 2"), chose sequence item 3 directly after item 2 ("2 → 3"), chose sequence item 4 directly after item 3 ("3 → 4"; only for those pigeons that had experienced training stage 4), chose the final trained items as its last choice ("3/4 last"), and when a subject completed the entire sequence of either three or four trained items in its correct order ("1 → 2 → 3 (→ 4)"). Individuals are listed in order of the number of items they were trained on; those that only completed up to TS3 were not expected to visit sequence item 4. Increasing age can have detrimental effects on memory and cognition 1,2 . Here, we 408 investigated potential age effects on pigeons' memory and the encoding of serial order during 409 sequence learning. Memory capacity declined noticeably with age, although not all aged 410 sequence learning, however, appeared to be independent of age, as all pigeons performed in 413 a way that indicated a mental encoding of sequence order. 414 The training data provided insight into an individual's ability to remember sequences 415 of increasing length. Lists of one and two items were acquired with relative ease by the 416 pigeons regardless of age. However, as early as the three-item sequence, pronounced 417 differences in acquisition rates emerged between the three age groups. Young pigeons 418 learned the list within significantly fewer sessions than adult and aged pigeons, and half of 419 the adult and aged individuals were unable to reproduce the sequence sufficiently within the 420 session limit. No subject reached the learning criterion for the four-item sequence. Although 421 the criterion was deliberately set relatively high to avoid ceiling effects, it was expected that 422 at least the younger pigeons would be capable of reaching this standard, as the one-year-old 423 pigeons in Terrace's 5,6 experiment did achieve a comparable level of performance not only 424 for four-item lists but also for five-item lists. Considering that the current study employed an 425 open-field paradigm, whereas previous studies used computerised tasks, it is difficult to 426 pinpoint the cause of this discrepancy, which could range from visual aspects and differences 427 in the mental representation of 2D and 3D stimuli (cf. 8-10 ) to motoric aspects when pigeons 428 have to move only their heads compared to their entire body from one stimulus to the next. 429 To acquire a measure of progress with the four-item list, we compared the highest 430 reached daily score (correctly completed trials out of the ten daily trials) after the pigeons 431 had received 30 of the 60 TS4 sessions. Although the highest scores were achieved by the 432 youngest age group, older subjects did not fall far behind, and all subjects that had reached 433 this training stage reached the threshold set for success in other related studies (i.e., 30% 434 correct in a single session 4 ). The lack of clear age differences in TS4 despite observable As is the case for humans 11 , there appear to be well-aging individual pigeons that preserve a 437 high level of memory capacity (comparable to the performance of the least successful young 438 subjects) and less well-aging individuals that show noticeable declines in memory. During a 439 highly complex memory task such as the simultaneous-chaining task presented here, the 440 decline is already noticeable at a relatively early age. 441 We incorporated two tests, a Choice Test and a Memory Test, to assess pigeons' 442 encoding of sequential order and infer potential differences regarding the cognitive 443 mechanisms underlying sequence learning across the age groups. There was no indication in 444 the obtained data to suggest performance differed between young, adult and aged pigeons in 445 the two administered tests. 446 The Choice Test allowed for a comparison of the observed behaviour to previous 447 studies. When presented with a forced choice between two list items, pigeons chose the first 448 list item over items presented later in the sequence, and chose earlier items over the final list 449 item, but they showed no preference for earlier sequence items when the choice was between 450 two items presented in the middle of the list. Terrace 5 interpreted these findings as evidence 451 that pigeons chained responses to neighbouring items within a sequence, learning that a 452 response to item n was only correct when preceded by a response to item n-1, but with no 453 further reasoning about sequential order among the items. He further asserted that the 454 special positions of the first and the last sequence items were encoded separately, resulting 455 in correctly ordered responses when the choices included one of these items, but disrupted 456 performance when the choice did not include the first or the last item. The results from our 457 Choice Test only replicate Terrace's findings in part -our pigeons also showed their weakest 458 choice performance during trials that included internal items of the sequence (2 vs 3 for those 459 subjects that had trained with four items). However, performance was just as impaired when with three items), an observation that is directly contrary to Terrace's argument of the final 462 item assuming a "special role". Indeed, reduced performance occurred primarily when the 463 choice was between sequentially neighbouring items, regardless of whether the item pair 464 occurred in the middle or at the end of the sequence. Item pairs that did not consist of 465 immediately neighbouring items within the sequence reliably led to good performance (2 vs 466 4), which was at the same level of accuracy as any choices including item 1. The special role 467 of the first item also mentioned by Terrace persisted in the current study as well, however, 468 this was most likely due to overlearning. Taken together, these results suggest that, instead 469 of associatively chaining responses to neighbouring items, during the current experiment the 470 pigeons encoded the order of items to some degree, although it might have been more in 471 terms of items that appear "earlier" or "later" within the sequence than in terms of individual 472 ordinal positions. The salience of the last item, which was so prominent in Terrace's studies, 473 was not confirmed here. However, Terrace's pigeons were trained on five-item lists, which 474 might have posed an additional level of cognitive difficulty to his subjects, resulting in good 475 memory only of the first and last item but with little representation of the order of internal 476 items. Instead, as noted by Scarf and Colombo 4 , Terrace's subjects relied on the display of all 477 five list items together to be able to reproduce the order. With shorter lists of four or three 478 items, as used in this experiment, the cognitive load may have been reduced sufficiently to 479 allow an encoding of the relative position of items as appearing earlier or later within the list. 480 The Memory Test allowed a further assessment of whether the observed behaviour 481 supported the hypothesis proposed by Terrace 5 regarding chaining of responses without an 482 inherent concept of order, or the more cognitively complex process proposed by Scarf and 483 Colombo 4 . The most obvious result of this test was that item pairs further down the list were 484 increasingly less likely to be chosen in their correct order. It is unlikely that this failure to all items were equally preferred over novel items in the pairwise-choice test. Instead, we 488 observed that subjects abandoned the sequence as soon as their visit to the first item resulted 489 in an absence of reinforcement. This observation is difficult to combine with Terrace's 490 account of chained responses, by which responding to one item automatically cues the 491 response to the next. As the pigeon had to complete two baseline trials correctly before 492 entering a Memory Test trial, it is unlikely that the sudden stop was due to a lack of motivation 493 or an inability to remember the sequence that the pigeon had completed successfully in the 494 immediately preceding trial. Although unexpected, the observed behaviour indicates that 495 pigeons had formed a concept of the goal of completing the sequence (to maximise reward) 496 and acted in a planned way to achieve it. The pigeons did not perceive each item as an 497 individual stimulus-outcome event, but indeed as part of a connected sequence. When the 498 expected reward was not encountered in sequence item 1 during the non-reinforced test 499 trials, this disruption was likely extrapolated to the entire sequence and the subjects changed 500 their behaviour from the formerly goal-directed completion of the sequence to an 501 unstructured search. 502 The combination of results of both tests makes it evident that the arguments put 503 forward by Terrace 5 are not compatible with the behaviour shown by our pigeons. Instead 504 of acquiring simple response-response associations, the pigeons showed evidence of forming 505 a representation of the order of list items, although this representation might be less 506 sophisticated than proposed by Scarf and Colombo 4 , and instead limited to a concept of items 507 that appear earlier and later in the sequence. As Terrace pointed out correctly, the first item 508 takes a special role, likely due to excessive overtraining. Furthermore, although the absolute 509 length of the sequence that could effectively be memorised was reduced for some older 510 Regardless of age, pigeons were able to perform in a way that was consistent with the task 512 goal of maximising reward. 513 Thus, the current study showed that age can have a noticeable effect on the cognition 514 of pigeons. As previously shown for humans 12,13 , aging does not impact all cognitive abilities 515 uniformly, nor are all individuals affected to the same degree. It is possible to identify 516 cognitive capacities that are highly susceptible to age for many species, such as memory, as 517 shown in this study. Our study further supports that aging is an individual process, 518 manifesting itself differently in "healthy-aging" or "poorly-aging" individuals. Although aging 519 research is currently still limited to a few species, it is evident that age plays a crucial role in 520 many aspects of animal cognition and behaviour. Identifying suitable paradigms, like the one 521 presented in this study, to investigate such age effects in a wide range of species is the first 522 step to closing the gap.   Table 1. Percentage of trials (out of three trials) during which a subject chose the 585 item that had appeared earlier in the sequence. Individuals are listed in order of the number 586 of items they received during training; those that only completed up to TS3 were not 587 presented with choices including sequence item 4. 588 Table 2. Percentage of Choice Test trials during which a subject chose the known item 589 when the choice was between a known (sequence items 1, 2, 3, or 4, as applicable) and an 590 unknown item (sequence items 5, 6, or 4, as applicable), or between two unknown items. 591 Individuals are listed in order of the number of items they were trained on; those that only 592 completed up to TS3 were not presented with sequence item 4 as a known option. 593 Table 3. Percentage of Memory Test trials (out of ten trials) during which a subject 594 chose sequence item 1 before any other items ("1 first"), chose sequence item 2 directly after 595 item 1 ("1 2"), chose sequence item 3 directly after item 2 ("2 3"), chose sequence item 4 596 directly after item 3 ("3 4"; only for those pigeons that had experienced training stage 4),