Distinct Origins of Two-order Hierarchical Cognitive Abilities in Human Adults

Human cognitive abilities are considerably diverse from basic perceptions to complex social behaviors. All human cognitive functions are principally categorized into a two-order hierarchy. Almost all of the first-order cognitive abilities investigated in behavioral genetics have been found to be dominantly heritable. However, the origins of the human second-order cognitive abilities in metacognition and mentalizing so far remain unclear. We here systematically compared the origins of the first-order and second-order cognitive abilities involved in the metacognition and mentalizing tasks using the classical twin paradigm on human adults. Our results demonstrated a double dissociation of the genetic and environmental contributions to the first-order and second-order cognitive abilities. All the first-order cognitive abilities involved in the metacognition and mentalizing tasks were dominantly heritable. In contrast, the shared environmental effects, rather than the genetic effects, had dominant contributions to the second-order cognitive abilities of metacognition and mentalizing in human adults. Hence, our findings suggest that human adults’ monitoring sensitivities in metacognition and mentalizing are profoundly sculpted by their social or cultural experiences, but less preconditioned by their biological nature.


Introduction 32
Everyone has different talents. Some are good at motoric activities (e.g., playing soccer), while 33 others are better in cognitive tasks (e.g., math). There is a long history in questing the origins of 34 individual differences in human abilities, specifically in cognition (Galton, 1870; Plomin and 35 Daniels, 2011). All human cognitive functions are principally categorized into a two-order 36 hierarchy (Flavell, 1979;Vygotsky, 1978). The first-order cognitive functions refer to mental 37 computations on external information from the physical world, while the second-order cognitive 38 functions refer to mental computations on internal information from the mental world that is 39 generated during the first-order cognitive processes. That is, the second-order cognitive 40 processes monitor and manipulate the first-order cognitive processes (Nelson, 1990) (Figure 1A).

41
Critically, the capabilities of the two-order hierarchical cognitive functions might be dissociable.

42
For instance, the capabilities to monitor one's own decisions (i.e., metacognition) are enormously

52
Hence, a greater resemblance of a cognitive ability between the MZ than DZ twins reflects the 53 existence of a genetic factor contribution, while a common resemblance between the MZ and DZ 54 twins instead reflects the existence of shared family environmental contribution (Neale and However, general intelligence cannot similarly explain the second-order cognitive abilities, 66 even though both significantly contribute to the cognitive performance in everyday life (Greven et 67 al., 2009), but see also ref. (Sternberg, 1984). The second-order cognitive process in monitoring 68 one's own current first-order cognitive processes is referred to as metacognition (e.g., I believe 69 how well I made the decision), while the corresponding process in monitoring the others' cognitive 70 processes are termed as mentalizing (e.g., I believe how well you believe you made the decision).

71
Differing from metacognition that can directly introspect the internal information inside one's own 72 brain, mentalizing is unable to access the others' internal information. Instead, it may rely on 73 social cue associations, simulations of metacognitive experiences, or the theory of theory (i.e., 74 theory of mind, ToM), to estimate the others' mental states (Flavell, 2003). Hence, although the 75 two second-order cognitive processes commonly involve in meta-representations of mental states,

91
To compare the genetic and shared family environmental contributions respectively to the first-92 order and second-order cognitive abilities, in the present study we recruited fifty-seven pairs of 93 adult MZ twins and forty-eight pairs of adult DZ twins to participate both the metacognition and 94 mentalizing tasks (Table 1). On each case, two pairs of twins (either MZ or DZ) came together 95 and concurrently conducted both tasks (see Materials and Methods). In the metacognition task, 96 the participant reported the confidence rating immediately after making a decision about the 97 direction of random dot kinematogram (RDK) from four alternative options (Figure 1B, upper).

98
The task difficulty was determined by the fraction of coherently moving dots (i.e., coherence), 99 which was calibrated by a staircase procedure with the performance accuracy towards 50% for 100 each participant prior to the experiment. In the mentalizing task, the participant concurrently 101 observed the partner's task performance on the RDK task and attributed the trial-by-trial 102 confidence to the partner. Critically, to avoid evoking the participant's own decision uncertainty on 103 the RDK task, only those originally coherently moving dots remained moving, while the other 104 originally randomly moving dots were stationary (Figure 1B, bottom). By virtue of this alternation, 105 the RDK stimulus presented to the participant was noiseless, thereby, the participant could 106 perceive the task difficulty, but would not have her/his own decision uncertainty as experienced in 107 the metacognition task. Sequentially, the participant also perceived a progress bar representing 108 the response time that the partner used to make the choice. The participant did not receive the 109 information about the option choice and confidence rating reported by the partner. Each 110 participant conducted the metacognition task and the mentalizing task twice. When the participant 111 conducted the mentalizing task, the partner, either the sibling of the same twin (within-twin) or 112 one of another pair of twin (between-twin), was concurrently conducting the metacognition task, 113 and vice versa. The two pairs of twins were physically separated from one another, while the 114 concurrent metacognition and mentalizing tasks were synchronized by network connection 115 following the TCP/IP protocol via an Ethernet cable ( Figure 1C).

116
There were no systematical differences between the MZ and DZ twins in either first-order or 117 second-order behavioral measures related to the metacognition task (Figure 1 -figure   118 supplement 1). As the participants performed the metacognition task twice, we could evaluate 119 the reliability of these behavioral measures using the test-retest correlation. Except for the 120 stimulus coherence that was the same across the two runs for each participant (

130
We then calculated the intraclass correlation coefficients (ICCs) to measure the resemblance from the same genes or the same family environments. Importantly, the ICCs of all the first-order 136 behavioral measures within the MZ twins were consistently larger than those within the DZ twins 137 (not significant in the RT variance), indicating considerable genetic factor contributions. In striking 138 contrast, those of the second-order behavioral measures had no significant differences between 139 the MZ and DZ twins, or some in the DZ twins were even numerically larger than the MZ twins 140 (e.g., residual AUC), thereby indicating negligible genetic factor contributions.

141
To quantify the genetic and environmental contributions, we used maximum likelihood to 142 compare all the potential structural equation models (SEMs) that decompose the covariance of

151
As shown in Table 2, all the behavioral measures on the first-order cognitive abilities 152 (coherence, median RT, RT variance, and mean confidence) were best accounted for by the AE  abilities, but CE for the second-order cognitive abilities (Figure 2). The dominance of the genetic 163 factor or the shared family environmental factor in each behavioral phenotype in the full ACE 164 model was the same as in the best-reduced model ( Table 2). Hence, the dominant origin in each 165 behavioral phenotype was not biased by the selected models (Hill et al., 2008).

166
In the mentalizing task, the participant estimated the partner's decision confidence by

186
Critically, most of the within-twin ICCs were significantly larger than the corresponding cross-187 twin ICCs (Figure 3 -figure supplement 2), indicating that the resemblance of the mentalizing 188 ability was also specific for twins who shared the genes and family environment. Further, the 189 participants seemed able to tell the partner from the same twin pair (within-twin) from the partner 190 from another twin pair (between-twin), in that the respective ICCs were significantly differential 191 (Figure 3 -figure supplement 2), even though they were physically separated from one another.

192
We thus separately compared the SEM models accounting for the within-twin and between-193 twin behavioral measures in the mentalizing tasks. For the within-twin behavioral measures, the 194 SEM analyses also revealed that the individual differences of the RT-confidence association (RT 195 weights), the gamma correlation, and the raw AUC among the MZ and DZ twins were dominantly 196 accounted for by the AE/DE model ( Table 3). The genetic contributions ranged from 0.36 to 0.52.

197
In contrast, the individual differences of the residual AUC that was calculated after the RT-198 associated component was regressed out were best accounted for by the CE model (Figure 3).

199
The shared family environmental contribution was about 0.19 (Table 3). For the between-twin 200 behavioral measures, Similarly, the individual differences of the RT-confidence association and 201 the gamma correlation were best accounted by the AE/DE model, but those of the raw AUC were 202 best accounted by the CE model (Figure 3). The bootstrapping procedure further confirmed 203 these results (Figure 3). The results in the full ACE model were as the same as in the best-204 reduced model (Table 3), with an exception of the AUC in between-twin mentalizing task.

205
Thereby, the first-order cognitive ability of inference on the basis of the RT-confidence 206 association in mentalizing was genetically influenced, but the second-order cognitive ability of 207 mentalizing sensitivity beyond the RT-confidence association, particularly in the within-twin 208 mentalizing task, was instead primarily influenced by the shared family environmental factor.

209
However, unlike the assumptions that the common environmental factor and the non-

230
A pitfall in the present study was that the sample size of the MZ and DZ twins was relatively 231 small. This may raise concern about the robustness about the results. To assess the robustness

288
On the other hand, in terms of the mentalizing abilities (i.e., cognitive empathy, ToM), the 289 inference on the basis of the association with the external cue (i.e., RT) was heritable, but the 290 intrinsic mentalizing sensitivity was not. The two abilities were also dissociated. The former was a 291 first-order cognitive ability, but played a major role in mindreading (Kliemann and Adolphs, 2018; 292 Lurz, 2011), in that the latter was not further able to predict the other's performance after the cue-293 associated component was regressed out. However, these cue-independent residuals were not 294 simply noises, but were, at least partially, generated from the neural activities in the social brain

343
Two hundred and twenty-six right-handed healthy participants (sixty-five pairs of MZ twins and 344 fifty-five pairs of DZ twins with the same sex, 24.2 ± 2.6 years old, one hundred and eighteen 345 females) participated the present study (see Table 1 for their demographics and socioeconomic

373
The RDM coherence was constant in all trials and was set to match the task difficulty between

391
On the basis of our experience, performance on the metacognition task could become more 392 stable after participants received sufficient practice. Therefore, each participant was trained for 393 forty minutes prior to performing the two tasks. The RDK stimulus coherence (task difficulty) was 394 adaptively adjusted trial-by-trial with a staircase procedure, such that the RDK stimulus 395 coherence was upgraded one level after two continuous correct trials, downgraded one level after 396 two continuous erroneous trials, and otherwise remained unchanged (Levitt, 1971). The RDK 397 stimulus coherence started from 50%, and thereafter reduced and later remained at a stable level 398 throughout the procedure. The performance accuracy rate became close to 0.5 at the end of

408
between-twin), was concurrently conducting the metacognition task, and vice versa. The participants were physically separated from one another by board panels (Figure 1C)

468
The differences in covariance identified the SEM parameters by comparing covariance between 469 MZ and DZ twins along the observed variable as follows where CV !" is the covariance between the MZ twins, CV !" is the covariance between the DZ 473 twins, V ! is the additive variance, and r is a coefficient for the non-additive variance (V !" ), which 474 is set as 1/4 or 0, p is a binary coefficient (0 or 1) for the common environmental variance (V ! ),

475
and V ! is the error variance including unshared environmental variance. If the MZ resemblance is 476 larger than twice the DZ resemblance, then V !" (or dominant genetic effect, D) is considered, but 477 V ! is then ignored. That is r is 1/4 and p is 0, and this model is referred to as the ADE model.

478
Otherwise, r is 0 and p is 1, and this model is referred to as the ACE model.

479
To identify which model that should best account for each behavioral measure, we used two

703
The fraction ratios of the selected best models for the bootstrapping data and the dominant ratios