Mnemonic introspection in macaques is dependent on dorsolateral prefrontal but not orbitofrontal cortex

Metacognition refers to the ability to be aware of one’s own cognition. The anterior prefrontal cortex has been associated with meta-perceptual but not meta-memory decisions. A recent study has challenged this notion showing that neural activation in macaques’ prefrontal areas 9 and 9/46d is associated with metamemory of recognition of items. Here, we verified the critical role of sub-regions of prefrontal cortex in the domain of spatial recognition memory. We contrasted performance of monkeys with superior dorsolateral prefrontal lesion with orbitofrontal lesioned monkeys and unoperated controls in spatial recognition memory tasks. We show that monkeys with dorsolateral lesions are impaired in meta-accuracy, but not in recognition performance, in comparison to orbitofrontal-lesioned and control monkeys. Together with the observation that the same orbitofrontal-lesioned monkeys were impaired in updating rule-value in a Wisconsin Card Sorting Test analog, we provide causal evidence towards functional specialisation between dorsolateral and ventromedial prefrontal cortices underpinning the introspection ability in relation to memory recognition in primates.


21
orbitofrontal lesioned monkeys and unoperated controls in spatial recognition memory tasks.

22
We show that monkeys with dorsolateral lesions are impaired in meta-accuracy, but not in 23 recognition performance, in comparison to orbitofrontal-lesioned and control monkeys.

24
Together with the observation that the same orbitofrontal-lesioned monkeys were impaired in 25 updating rule-value in a Wisconsin Card Sorting Test analog, we provide causal evidence 26 towards functional specialisation between dorsolateral and ventromedial prefrontal cortices 27 underpinning the introspection ability in relation to memory recognition in primates. streams supporting metamemory, one for temporally remote items in prefrontal area 9 (or 9/46d) 46 versus another for more recent items in area 6, further corroborating the notion that first-order 47 tasks (e.g., memory recognition) and second-order tasks (i.e., metacognitive processes such as 48 meta-recognition) are dissociable 9 . It follows that in theory we could induce deficits in 49 metamemory of recognition by lesioning these dorsolateral prefrontal (dlPFC) regions (lateral 50 area 9) without changing other aspects such as type 1 memory task performance 10 .

51
Furthermore, a recent neuroimaging study revealed that metacognitive control processes

65
Taking the above into account, and in light of the findings that second-order metacognitive 66 processes could be separated from confidence per se 18 , we therefore set out to verify the causal 67 roles of these two subregions (dlPFC vs. OFC) of the aPFC in memory using two variants of a 68 spatial recognition memory paradigm. We hypothesized that the dlPFC is causally required for 69 accurate memory introspection. Specifically, we contrasted the first-order memory and second-   These meta-indices in principle refer to how meaningful a subject's confidence is in 119 distinguishing between correct and incorrect responses. We accordingly ran two mixed-design 120 repeated-measures ANOVAs on percentage correct with "Group" as a between-subjects 121 variable and "Confidence" as a within-subjects variable for the two tasks separately and 122 obtained a significant interaction with the spatial-variant task F (2, 9) = 5.416, P = 0.029, but 123 not with the temporal-variant task F (2, 10) = 0.355, P = 0.710. Percentage correct in high-124 confidence trials is usually higher than low-confidence trials, P < 0.01 for both CON and OFC 125 monkeys in both tasks, but such effects were disrupted in the sdlPFC monkeys in the spatial-  having "Group" as a between-subjects variable on meta-d' (a sensitivity measure quantifying 129 the ability to discriminate between correct and incorrect judgments) for the two tasks separately 130 also revealed that a significant main effect of "Group" in the spatial-variant task on SDT meta-131 d′: F (2, 9) = 5.701, P = 0.025, post hoc test: CON vs. sdlPFC, one-tailed Dunnet P = 0.015,

133
In order to ascertain that these lesion effects were task-specific (Task: spatial-134 variant/temporal-variant), we ran three separate mixed-design repeated-measures ANOVAs 135 considering only the CON and sdlPFC groups with "Group" as a between-subjects variable and 136 "Task" as a within-subjects variable and confirmed a marginally significant "Task × Group"

142
Given that metacognition is quantified by the correspondence between confidence and 143 type 1 task performance, it is theoretically important to establish that the task (first-order)

144
performances were matched between the groups in order to argue for the presence of a true 145 difference in metacognition caused by the sdlPFC lesion. Despite the deficits in metamemory 147 memory deficits in their type I performance. In two mixed-design repeat-measures ANOVAs 148 we entered the percentage correct or RT with one between-subjects factor "Group" and one 149 within-subjects factor "Condition" and found neither a main effect nor interaction effects with

168
with one between-subjects factor "Group" and one within-subjects factor "Condition". We  175 not been able to ascribe the effects specifically to spatial recognition per se. Is this deficit 176 uniquely ascribable to the metamemory in temporo-spatial recognition, or more generally to 177 the metamemory of learning abstract rules, or other higher cognitive processes? Considering 178 performance supporting WCST demands multi-processes such as memory and acquisition of 179 abstract rules, as well as reward-value evaluation, we thus analyzed some extant data obtained 180 from WCST to test for metacognitive deficits specifically in the sdlPFC monkeys. In contrast 181 to the spatial recognition task, no meta-deficits were found with WCST in the sdlPFC group in

270
An alternative but not mutually exclusive explanation is that given that the frontal vs.

388
SDT toolbox is designed for 2AFC tasks, in which S1 and S2 are always constant in the left or 389 right, but our target and foil are randomly presented at the screen and we only recorded the 390 separation of the two probes and did not track the specific position of the target and the foil.

391
Since the target and the foil were presented randomly on the screen, and the SDT algorithm 392 only requires the distribution of those four kinds of trials, we divided the number of those trials 393 equally to S1 trials and S2 trials in a random manner considering the animals would not have 394 any preference to any given side/location of the screen. In addition, we have also replicated the

407
The Φ coefficient evaluates how optimally each trial was assigned for high or low confidence 408 based on performance in the preceding cognitive judgment, reflecting the correlation between 409 the two binary variables. Note that despite differences in their mathematical assumptions, the 410 three meta-cognitive metrics are highly correlated with each other (FIG. S1).

497
We analyzed WCST data from 12 monkey data points (9 CON vs. 3 sdlPFC). Six out of the 498 nine CON monkey data points here were from the pre-lesion data of the six lesioned monkeys

499
(3 sdlPFC and 3 OFC). We included 3,000 trials (acquired from ten 300-trial daily sessions) 500 per monkey data point. We collapsed all trials and classified the trials into four types of trials 501 for the computation for the meta-efficiency and Phi coefficient. Since the type II SDT toolbox 502 was designed for 2AFC tasks, and the WCST task contained three stimuli, we ran three separate 503 sets of computation, each one discarding only either the bottom, left, or right choice, for each 504 of the three meta-indices. For each monkey, we then computed the mean of these three values 505 as his meta-score to enter into the meta-indices calculation. In this analysis, we did not include 506 the OFC monkeys because the OFC monkeys were severely impaired in the WCST type I task, 507 thus making any analyses on meta-ability invalid (their chance level implies they did not know 508 how to make correct judgments, violating the prerequisite for the meta-assessment of their 509 judgement).

511
Preliminary training. All monkeys completed preliminary training and task acquisition before performing the two main tasks and WCST described above. We conducted the spatial-variant 513 task immediately after the temporal-variant task without any additional training. The monkeys 514 performed one session per day, 6-7 d per week. For the lesioned animals, the task was 515 administered post-operatively (on average 22 months post-lesion). For the two DMP tasks, 516 during task acquisition the monkeys were trained until they reached ≥ 90% performance level 517 within a 100-reward session. All trials in this stage consisted of a short delay interval (1 s), and 518 a wide separation between choice positions (21.7°, or 23 cm) to make the trials "easy" to acquire.

519
Upon reaching criterion, the three groups were not different in the number of errors accrued, F 520 < 1 and number of rewards received, F < 1, indicating that the groups of lesioned monkeys 521 learned to perform these spatial recognition problems as well as controls.

590
Metacognitive accuracy in sdlPFC group was lower than CON group for spatial-variant task