A comparative study of theory of mind in taxon-like clusters of psychometric schizotypes and individuals at genetic risk for schizophrenia

ABSTRACT Introduction: Clinical and family studies suggest that alterations of theory of mind (ToM) represent a marker of genetic liability to schizophrenia. Findings regarding ToM in schizotypy are less consistent. The study aimed to explore whether this might be due to an insufficient account of the heterogeneity of schizotypy in prior research and/or the fact that in psychometric schizotypy ToM alterations could manifest as subtle peculiarities rather than overt errors of mentalising.Methods: Individuals without a family history of psychosis (n = 150) were assigned to low, positive, negative, and high mixed schizotypy classes based on a cluster analysis of 1322 subjects who completed the Schizotypal Personality Questionnaire. The classes were compared on their performance of faux pas tasks with 77 adult first-degree relatives of schizophrenia patients, who represent individuals at genetic risk for schizophrenia. Besides overt errors, subtle alterations in ToM were analysed using expert judgment.Results: The relatives tended to make overt errors and demonstrated specific features of intentional reasoning. None of the schizotypal classes showed similar trends.Conclusions: The results complement the literature on the subjective-objective disjunction in psychometric schizotypes and did not provide evidence that ToM anomalies are a marker of genetic liability to schizophrenia in this cohort.


Introduction
Deficits in theory of mind (ToM), an ability to infer intentions, dispositions, and beliefs of others, might underpin some of the characteristic symptoms of schizophrenia (Frith & Corcoran, 1996;Green, 2020). There is abundant evidence of ToM abnormalities in patients, with the effect size of Cohen's d = 1 (Bora & Pantelis, 2013;Fretland et al., 2015;van Neerven et al., 2021), as well as some evidence in favour of the theoretical-based expectations that positive symptoms are primarily associated with an impaired ability to recognise intentions (Dorn et al., 2021;Fretland et al., 2015;Peyroux et al., 2018), while negative symptoms are linked to disturbances of emotion processing (Dorn et al., 2021;Peyroux et al., 2018;Wójciak et al., 2021). In addition, both the reduced capacity to detect mental states (hypo-mentalising) and over-attribution of knowledge/intentions to others (hyper-mentalising) have been demonstrated in schizophrenia, although the idea that the former is linked to negative symptoms and the latter with positive ones has not received consistent support in empirical studies (Abu-Akel, 1999;Bliksted et al., 2019;Dorn et al., 2021;Fretland et al., 2015;Peyroux et al., 2018). The presence of similar but milder ToM deficits (d ∼ 0.4-0.5) in unaffected first-degree relatives of patients suggests that ToM anomalies have a developmental origin and may be a marker of genetic liability to schizophrenia (Bora & Pantelis, 2013;Lavoie et al., 2013).
In support of this view ToM alterations have also been demonstrated in psychometric schizotypy (Bora, 2020), i.e. in people with personality organisation that is thought to convey vulnerability to the development of schizophrenia (Lenzenweger, 2015). However, in the case of schizotypy the effect size for ToM deficits is small (d = 0.24), with some papers reporting null results (Bora, 2020); and the findings regarding differences between positive (characterised predominantly by unusual perceptual experiences/ beliefs and suspiciousness) and negative (asociality, constricted affect, and anhedonia) schizotypy are less consistent than those for schizophrenia (Bedwell et al., 2014;Canli et al., 2015;Fernyhough et al., 2008;Morrison et al., 2013;Pflum et al., 2013;Wastler & Lenzenweger, 2021). This might be due in part to insufficient consideration of the heterogeneity of schizotypy. The nature of schizotypy remains to be a subject of debate (Lenzenweger, 2015). According to the qualitative (categorical) model, dating back to Rado and Meehl, schizotypal individuals are a distinct subgroup of the population (about 10%) with specific genetic background resulting in a neurobiological defect (schizotaxia), which manifests itself in schizotypal traits at the preclinical level and, under certain circumstances, leads to schizophrenia spectrum disorders. In the quantitative, temperament-based model, dating from Eysenck and continued with Claridge and others, schizotypy is part of a normal personality and can have varying degrees of severity in the general population. Of importance in the latter model, schizotypy, which began as a personality trait of psychoticism resembling positive symptoms, has evolved into a multidimensional construct. Moreover, it has been argued that schizophrenia might be considered as the overlap of two genetically distinct dimensionspositive and negative schizotypy (Grant, 2015).
Previous ToM studies have used a correlational dimensional approach or an extremegroup design with arbitrary criteria for assignment to low and high schizotypy groups (see Bora, 2020, for review). However, cluster and latent class analytic investigations of psychometric schizotypy commonly detect taxon-like classes of individuals with both qualitatively and quantitatively different combinations of the schizotypal dimensions. Specifically, the literature suggests the existence of low, positive, negative, and high mixed schizotypy groups (Barrantes-Vidal et al., 2010;Fonseca-Pedrero et al., 2017;Polner et al., 2021). These groups may share with schizophrenia different parts of their genetic background (Morton et al., 2017), which lead to different endophenotypes.
Another reason might be that since schizotypal individuals are healthy people with, at most, modest cognitive anomalies (Chun et al., 2013), changes in their ToM could manifest as subtle peculiarities rather than overt errors of mentalising.
Here, we addressed these possibilities by comparing taxon-like schizotypy classes identified in the general population with unaffected first-degree relatives of schizophrenia patients on errors and subtle peculiarities of performance on faux pas tasks (Stone et al., 1998). Unaffected relatives represent genetically high-risk individuals as they share with the probands the genetic background including the genetic liability to schizophrenia as evidenced by their elevated schizophrenia polygenic risk scores (van Os et al., 2020). At the same time, like psychometric schizotypes, they do not share confounders of psychosis, such as distress related to psychotic symptoms, severe cognitive deficits, or adverse effects of antipsychotic medication, and may therefore have more resources to compensate for difficulties in mentalising. On this basis, one can suggest that ToM alterations in schizotypy should be more similar to those in relatives than in patients. The comparison of individuals at psychometric and genetic risk is a rarely used but promising approach to confirm that a phenotype under investigation might be viewed as a genuine manifestation of schizophrenia liability in the general population. Of note, we treated relatives as a single group because of the unreliability of self-reports of schizotypal signs in this cohort (e.g. Appels et al., 2004; see also Table 1 for social desirability assessments).
Faux pas tasks are an advanced reasoning ToM test that makes it possible to distinguish between cognitive and affective aspects of ToM. It evaluates the ability to realise that the protagonist of a story said something that should not have been said.
For a correct answer, a subject must understand that the protagonist does not know that this should not have been said to the interlocutor, and that the interlocutor is insulted by these words, i.e. the subject needs to understand the intention of the protagonist (cognitive ToM) and the emotion of the interlocutor (affective ToM). To assess subtle peculiarities in mentalising, we supplemented the standard analysis of the faux Note. PSthe group of psychometric schizotypy; GRthe group of genetic risk; Edueducation; SPQthe Schizotypal Personality Questionnaire; CPFthe cognitive-perceptual factor; IPFthe interpersonal factor; DisFthe disorganised factor; ToM totalthe total score of the Faux Pas Recognition Test. Significant differences: a PS groups vs GR; b high mixed PS vs GR and low PS; c between all groups, except for positive PS vs negative PS; d between all groups, except for low PS vs GR; e between all groups, except for positive PS vs GR; f between all groups, except for low PS vs GR and positive PS vs negative PS. pas errors (Stone et al., 1998) with an in-depth analysis of task performance using expert judgment.
Therefore, the aim of the present study was to compare clusters of healthy individuals with different profiles of schizotypal traits with unaffected first-degree relatives of schizophrenia patients on overt errors and subtle peculiarities of ToM. Based on prior findings and speculations (Bora, 2020;Green et al., 2008;Scherzer et al., 2012;Wastler & Lenzenweger, 2021;Zalla et al., 2009) we expected to observe such peculiarities as (1) an excessive attribution of intentions, including intentions not disrupting faux pas, which might signal hyper-mentalising; (2) a shift towards negative valence in explanations of protagonists' behaviour, which might signal negative affectivity and/or hostility bias; and (3) a trend to explain protagonists' behaviour in terms of their personality traits and emotional states instead of mental states related to the situation at hand. We hypothesised that compared to low schizotypy individuals, relatives and the high mixed schizotypy group would have multiple and prominent ToM alterations, which would be consistent with both the categorical and quantitative models of schizotypy, while positive and negative schizotypy clusters might show milder isolated changes in the cognitive and affective ToM, as well as trends to hyperand hypo-mentalising, respectively. Although somewhat inconsistent, previous studies have shown that neurocognition may to some extent influence ToM in different cohorts (Bottiroli et al., 2016;Cella et al., 2015;Kocsis-Bogár et al., 2017). Specifically, cognitive ToM in the faux pas tasks was found to correlate with working memory updating (Bottiroli et al., 2016). Given this, we considered the impact working memory has on ToM in our research.

Participants
The sample was derived from the database of the Mental Health Research Center which contains data on individuals from the general population of Russia as well as on patients with psychosis and their first-degree relatives, as described earlier (Golimbet et al., 2017). All participants gave written informed consent to participate in the study. The study was approved by the Ethics Committee of the MHRC (No. 662/21.07.2020). The procedure involved administering a set of questionnaires and a cognitive battery that included faux pas tasks. In addition, individuals from the general population completed a short questionnaire on socio-demographic characteristics and personal and family history of psychiatric disorders. Relatives of patients were offered to participate when they contacted a psychiatrist to provide information about a proband. The diagnoses of probands and relatives were established according to the ICD-10 criteria. Inclusion criteria for individuals from the general population were the availability of the Schizotypal Personality Questionnaire (SPQ-74) and faux pas tasks data. For relatives, it was only mandatory to perform faux pas tasks. Candidates were not included if they (1) had no secondary education (compulsory in Russia), (2) were younger than 17 or older than 65 years old, (3) had personal psychiatric history, or (4) had substance abuse or another disorder impairing cognitive performance. Candidates from the general population, in addition, were not included if they reported a family history of psychosis.
A large cohort (n = 1,322) of individuals without a family history of psychosis completed the SPQ-74. Of them, 150 individuals also completed faux pas tasks and were included in the psychometric schizotypy group (PS) of the present study. These individuals were assigned to schizotypy classes via a cluster analysis of the entire cohort. The group of genetic risk (GR) consisted of 77 unaffected adult first-degree relatives from 41 families: 60 parents, 14 siblings, and three offspring. Table 1 reports the characteristics of the PS and GR samples.

Assessments
The SPQ-74 is a 74-item true-false questionnaire evaluating nine DSM symptoms of schizotypal personality disorder (Raine, 1991). Its nine scales form three higher-order factors: the cognitive-perceptual/positive factor (Ideas of Reference, Odd Beliefs or Magical Thinking, Unusual Perceptual Experiences, and Suspiciousness), the interpersonal/negative factor (Excessive Social Anxiety, No Close Friends, Constricted Affect, and Suspiciousness), and the disorganised factor (Odd or Eccentric Behaviour and Odd Speech).
The original Faux Pas Recognition Test includes 10 faux pas and 10 control stories (Stone et al., 1998). We used three faux pas tasks to ensure the tolerance of the study, as the tasks were part of a longer examination, administered not only to PS and GR but also to probands. These were the two most informative ToM stories [nos. 2 and 14 (Negrão et al., 2016)] and a self-developed story with a culturally understandable situation, which we had previously employed in schizophrenia patients and healthy individuals (Alfimova et al., 2015). In the combined sample of the present study, the selfdeveloped task was similar to story no. 14 in the ability to assess faux pas comprehension (a Spearmen correlation between the total scores [see below]: rho = .24, p < .001), though the internal consistency of the entire battery of three tasks was low (McDonald's ω = .35; 95% CI .18 -.53). After each story the subject was asked: 1) if someone said something awkward, 2) who said it, 3) why it was awkward, 4) why it was said, 5) whether the protagonist knew/remembered the circumstances and that her words could hurt the interlocutor, and 6) what the interlocutor felt.
For each participant, eight ToM indicators were calculated. The first two were scored by the experimenter (MA): (1) the ToM total score (the maximum score = 18) computed according to the instructions of Stone et al. (1998); and (2) the ability to recognise a faux pas, namely if a subject answered questions 1-3 correctly, he/she received one point for the detection of the faux pas. These variables were further dichotomised (0 vs 1) as follows: (1) error-free performance versus any error (i.e. the total score < 18) and (2) recognising all three faux pas versus not recognising at least one faux pas. To calculate the other indicators, we used the expert judgment of two clinical psychologists (MA and VP) experienced in studying social cognition in different cohorts, including families of patients with schizophrenia. One of the experts (VP) was blind to the status (PS, GR) of a participant. Based on Stone et al. (1998) and Zalla et al. (2009), answers to the question "Why do you think he/she said it?" were classified into intentional and non-intentional actions as follows: (1) "intention error" if the answer meant that the protagonist said the awkward thing deliberately, knowing it would hurt the interlocutor; (2) "another intention"the protagonist said the awkward thing with some purpose, not knowing it would hurt; (3) "faux pas reason"the protagonist did not know or forgot that this should not have been said; (4) "personality" if a subject explained the protagonist's behaviour in terms of her individual traits, for example, stupidity; and (5) "emotional state" if a reference to the protagonist's emotional states or feelings (for example, jealousy) took place. In addition, all reasons and intentions were coded as having positive, neutral, or negative valence. Responses about emotions of interlocutors were classified as obviously inappropriate, to some extent appropriate, and obviously appropriate to the situation. Some participants generated two alternative hypotheses regarding intentions or emotions. All responses were analysed. The interrater agreement for intentions (error/another/no), reasons (faux pas/trait/state), and their valence was high: respectively, Cohen's k = .86, 95%; k = .75, 90%; and k = .72, 82%. The agreement on emotions was low: k = .27, 83%, so only those emotions that were assessed as obviously inappropriate by both experts were considered "emotion error". In the cases of discrepancy, a consensus was reached through discussion.
To assess working memory updating we used a count-down series from 200 in 2s and 5s, with the performance measure being the number of correct answers during the first minute.

Statistical analysis
A k-mean clustering was performed using the nine SPQ-74 scales and the Generalised EM and k-Means Cluster Analysis procedure of the Data Mining module of Statistica 13, the sample being divided into training (n = 690) and testing (n = 632) subsamples for validation.
The following analyses were conducted mainly with JASP 014.1.0 (JASP Team, 2020). First, we compared groups on demographic and SPQ characteristics using ANOVA, Pearson χ2, or Mann-Whitney/ Kruskal-Wallis with post hoc Dunn's tests depending on the variable's distribution. Education was coded as secondary (1) vs any tertiary (2). We then ran a set of logistic regression analyses to assess the influence of sex, age, and education on every ToM parameter in the combined sample. The analysis revealed some specific significant correlations and trends. For consistency, we controlled for these three potential confounders in all subsequent analyses of the ToM measures. Next, since our GR sample included related individuals (parents and siblings from the same families), we tested if this would affect the results. To do so, we transformed the ToM total score to approximate the normal distribution using the Cox-Box method and performed a covariance analysis (ANCOVA) for mixed models, in which family was a random effect, and sex, age, and education were covariates. The main analyses comprised an unadjusted comparison of PS and GR on ToM parameters with Pearson or likelihood ratio χ2 test, which was followed by logistic regressions adjusted for sex, age, and education. The influence of the group membership was considered against a null model that included socio-demographic parameters; 95% bias-corrected bootstrap confidence intervals (95% CI bsa ) were calculated for regression coefficients using 1000 bootstrap samples. The significance level was set at α = .05, two-tailed. Bonferroni correction was applied where appropriate.

As shown in
Of note, among GR, 49 individuals completed the SPQ. They had lower SPQ total and factor scores than the entire PS sample; the significant difference in the disorganised factor survived after adjustment for age, sex, and education (logistic regression, OR = 0.77; 95% CI bca 0.62-0.96; p = .021). Differences in SPQ total and factor scores across the PS clusters were highly significant (Kruskal-Wallis test, all p < .001) and in the expected directions.

ToM in GR vs the entire PS group
In three cases (GR) a subject could not understand the story's content. Otherwise, the level of performance was high in both GR and PS (Table 2). Of note, only three responses from GR and from PS to the question of whether someone said something awkward were erroneous as to who made the faux pas and what exactly was not worth saying. All other "no" answers were accompanied by the explanation that nothing special happened since the protagonist did not know that those words would insult the listener. The influence of family on the ToM total score was not significant, F = 0.79, p = .76. The results of the comparison of GR and PS on the ToM indicators are presented in Table 2. The unadjusted analysis showed that GR made more intention and emotion errors and gave more explanations of the protagonist's behaviour in terms of their personality and in negative valence than PS. However, none of the differences withstood a Bonferroni correction. Subsequent logistic regressions, controlling for age, sex, and education, showed the significant impact of the group on the intention error rate. GR made more errors than PS. When the analysis was additionally adjusted for working memory, the effect of the group on the intention error rate continued to be significant, χ 2 = 7.88, df = 654, p = .005/p corr = .040, R 2 Nagelkerke = .048. The working memory measure itself predicted the intention error rate at a trend level, OR adj = 0.94, 95% CI bca 0.87-0.99, p = .078. Table 3 and Figure 2. The unadjusted analysis revealed a nominally significant difference in the intention error rate, χ 2 lr = 13.54, df = 4, p = .009/p corr = .072, and a trend for the "emotion error" parameter, χ 2 lr = 8.34, df = 4, p = .080. Logistic regressions confirmed the significant effect of the group on intention errors. The effect remained significant when the analysis was additionally adjusted for the working memory measure, χ 2 = 17.63, df = 651, p = .001/p corr = .008, R 2 Nagelkerke = .11. The effect of the group was driven by an increase in the error rate in GR compared to the other groups. Post hoc Fisher exact tests showed that GR differed from the positive (p = .036) and negative (p = .021) PS. The tendency to the difference in the number of emotion errors, which was nominally significant between GR and low PS (p = .022), is also noteworthy. There was no difference between the PS clusters. Nor were there expected trends across the five groups (low PS < negative/ Note. For each variable, the total number of analysed answers is given in parenthesis; OR, 95% CI, and p podds ratio, 95% confidence interval, and p-value for Pearson's χ 2 in contingency tables; corrected α = 0.05/8 = .006. OR adj , 95% CI bca, and p lrodds ratio, 95% bias corrected confidence interval calculated using 1000 bootstrap samples, and p-value for the Wald statistic in logistic regression adjusted for sex, age, and education.

ToM indicators for GR and the PS clusters are reported in
positive PS < high mixed PS ∼ REL). Based on the figure and the hypothesis of the study, we conducted an additional analysis of trends for the low PS, high mixed PS, and GR regarding the "personality" and "negative valence" parameters. For both variables, χ 2 was not significant for either the slope or the non-linearity (see Suppl. Material).

Discussion
Our study showed that, compared to the general population, GR more often attributed intentions to offend interlocutors than PS, although the percentage of such errors was Note. PSthe psychometric schizotypy groups, GRthe group of genetic risk. OR adj , 95% CI bcaodds ratio, 95% bias corrected confidence interval calculated using 1000 bootstrap samples in logistic regression adjusted for sex, age, and education. R 2 N -Nagelkerke R 2 . still small. The ability of relatives to recognise the emotions of interlocutors was also somewhat diminished. In addition, their explanations of the behaviour of protagonists tended to be negatively valenced and in terms of personality rather than the situation at hand, similar to what has been found in adults with high-functioning autism (Zalla et al., 2009). These ToM features might be signs of paranoid thinking, which echoes the data on the relationship between ToM deficits and paranoid delusions in schizophrenia (Frith & Corcoran, 1996). Overall, our results imply that while both cognitive and affective components of ToM might be affected in individuals with genetic liability to schizophrenia, the cognitive one is more severely impacted. Of note, only a few previous studies of relatives distinguished between cognitive and affective ToM and between hypoand hyper-mentalising, and none assessed other characteristics of their reasoning about characters' behaviour. Ho et al. (2015) found ToM impairments but no dissociation between affective and cognitive components of ToM, while Montag et al. (2012) revealed subtle impairments in cognitive but not emotional ToM and elevated error counts for hypo-mentalising. Thus, our findings confirm and expand previous data on the presence of mild ToM impairments detected by faux pas tasks in relatives of schizophrenia patients, suggesting that these impairments might reflect genetic liability to the disease (Bora & Pantelis, 2013;Ho et al., 2015;Lavoie et al., 2013;Montag et al., 2012;Raju et al., 2019;Tikka et al., 2020). We hypothesised that high mixed schizotypes from the general population might represent a group at the highest genetic risk for schizophrenia and as such would show diverse ToM anomalies similar to those seen in unaffected relatives of patients, while groups with isolated positive and negative schizotypal traits might have subtle impairments in attributing intentions and emotions, respectively. None of these hypotheses were confirmed. The high mixed schizotypes showed neither significant differences from the low schizotypy group, nor more similarity to relatives than the other clusters. Both positive and negative schizotypes demonstrated high cognitive ToM, and their affective ToM did not significantly differ from that of low schizotypal individuals. Nor did we find hyper-mentalising in the positive schizotypy cluster in contrast to the hypothesis and a recent study (Wastler & Lenzenweger, 2021).
Our negative results are in line with some other studies reporting null findings regarding the impairment of ToM in psychometric schizotypy (e.g. McCleery et al., 2012;Pickup, 2006), including those using faux pas tasks (Canli et al., 2015;Lam et al., 2016; but see Morrison et al., 2013). Moreover, they complement a range of data on the subjective-objective disjunction, which is a discrepancy between selfreported schizotypal signs and objectively measured levels of dysfunctions, in psychometric schizotypy . Specifically, they show that the schizotypy groups report signs of suspiciousness answering the SPQ questions but demonstrate high performance on tasks involving intention recognition. Finally, our findings suggest that individuals from the general population with schizotypal traits should be used for modelling genetic liability to schizophrenia with caution, as high schizotypes have shown no similarity to the genetically high-risk group. This is consistent with schizotypy studies using polygenic risk scores (PRS) for schizophrenia and some other genetic designs. In particular, Morton et al. (2017) revealed that the presence of a family history of psychosis did not predict belonging to classes with high schizotypy. Johnson et al. (2003) showed that cognitive domains were influenced by schizotypy symptoms only in unaffected co-twins of schizophrenia patients but not in co-twins of healthy individuals. Nenadić et al. (2020) found no correlation of schizophrenia PRS with schizotypy in a cohort of healthy individuals from Germany. Such a correlation was specific only for a particular subgroup of negative schizotypal signs in men in the large population-based cohort Generation Scotland (Docherty et al., 2020) and was longitudinally unstable and in the opposite direction of what was expected in a cohort of Greek conscripts (Hatzimanolis et al., 2018). This suggests that only a small portion of individuals scoring high on schizotypy scales may have genetic liability to schizophrenia, while the other part can represent "phenocopies" or people with an unspecific vulnerability. This assumption is consistent with prior longitudinal studies that have shown low prognostic sensitivity and specificity of schizotypy regarding the transition to psychosis (Debbané et al., 2015).
Apart from this main result, the following observation deserves noting. The authors of the faux pas tasks emphasise the need to use not only the total ToM score but also specific scores to make it clear where a subject has trouble (Stone et al., 1998). Our data also underscores the necessity of a fine-grained analysis of responses to these tasks. Specifically, anyone who answers "no" to the first question gets 0 points for the whole story. Meanwhile, our analysis showed that, in healthy people, such a response in most cases might reflect the "never mind" response style in the presence of the fully preserved ability to comprehend a faux pas, intentions, and emotions. This was especially true of positive schizotypes, who most often denied a faux pas, showing at the same time an almost perfect recognition of the absence of offensive intentions. The underestimation of a response style can therefore lead to the overestimation of the ToM deficiency in psychometric schizotypy.
However, some limitations of the study could impact our findings. First, at the theoretical level, we hypothesised that high schizotypal individuals being genetically predisposed to schizophrenia would have abnormalities in a variety of personality and cognitive domains similar to those seen in relatives of schizophrenia patients. This implicitly suggested that all these abnormalities arose from the set of genes predisposed to schizophrenia as a result of pleiotropy or interactions between the domains. However, this assumption seems to be consistent only with the categorical model of schizotypy (Lenzenweger, 2015). Regarding the temperament-based dimensional model, it has been hypothesised that schizotypy reflects a part of genetic predisposition to schizophrenia, while the development of psychosis requires additional deficits caused by the other gene set related to health and resilience (Grant, 2015). These different sets of genes might co-segregate in schizophrenia families but not in psychometric schizotypy. However, this hypothesis hardly explains the absence of ToM anomalies in schizotypal individuals, since the misinterpretation of intentions is a core of suspiciousness, an important component of schizotypy. Second, regarding the methodology, we did not include probands with schizophrenia from the families studied because some of them had severe neurocognitive deficits. However, faux pas tasks were previously used in samples of schizophrenia patients by us and others and were shown to be able to detect ToM impairments in this population (Alfimova et al., 2015;Ho et al., 2015;Negrão et al., 2016;Tikka et al., 2020). Third, while the power to detect ToM differences between GR and the entire PS sample at the α-level of .05 exceeded 80%, the cluster sizes might have been insufficient to detect some effects. However, none of the ToM variables showed a trend in the expected direction across the clusters. It seems therefore unlikely that our results regarding the absence of ToM anomalies in PS were due to an insufficient power. Fourth, the faux pas test is one of the most used ToM measures within neuropsychiatry, with a number of strengths (reviewed in Eddy, 2019). Specifically, it allows a fine-grained analysis of cognitive and affective mental states, hyperand hypo-mentalising. Nevertheless, the ability to detect a faux pas is just one of multiple aspects of ToM, which also includes the understanding of humour, deception, and non-literal communicative intent. Thus, results on a single task could be insufficient for a deep understanding of the differences in ToM between PS clusters and GR. In addition, like some other advanced ToM tests, it requires cognitive functioning above a certain threshold and is associated with education (Eddy, 2019), for which reason the sample had to be limited to relatively high-functioning people, which reduced the generalisability of the results. Fifth, it has been suggested that the subjective-objective disjunction in schizotypy may reflect the heightened emotional demands associated with cognitive functioning in the real world (Carrigan et al., 2017). It is possible that, in high schizotypes, ToM impairments, like alterations in other cognitive domains, emerge in response to stressful life events, hence the failure of the labbased ToM tasks to reliably identify ToM deficits in this cohort. Future research would benefit from controlling for levels of distress in psychometric and genetic risk groups during ToM assessment.

Conclusions
The study confirmed that unaffected relatives of schizophrenia patients demonstrated an increased error rate in ToM tasks compared to individuals from the general population, with the most pronounced deficit in intention recognition. We also found that relatives tended to refer to personality traits instead of mental states associated with the situation at hand to explain characters' behaviour. In addition, their explanations tended to be negatively valenced. None of the taxon-like schizotypal clusters from the general population showed similar features. The latter finding adds to the literature on the subjectiveobjective disjunction in psychometric schizotypy  and does not provide evidence that ToM alterations are a marker of genetic liability to schizophrenia in schizotypal individuals. It implies that when modelling schizophrenia-related processes, researchers should have in mind that only a small portion, if any, of psychometrically selected high schizotypal individuals might be genetically predisposed to schizophrenia-spectrum disorders. Nevertheless, as a group, high schizotypes are characterised by widespread maladaptation (Barrantes-Vidal et al., 2010;Polner et al., 2021). For this reason, this part of the population needs support irrespective of genetic risk for schizophrenia spectrum disorders. Given the discrepancy between their subjective and objective states, it might be hypothesised that psychotherapeutic interventions specifically targeting their subjective concerns and cognitive schemas would be appropriate for this group.

Disclosure statement
No potential conflict of interest was reported by the author(s).

Funding
This work was supported by the Russian Foundation for Basic Research under Grant 20-013-00230.

Data availability
Original data are available upon request to the corresponding author.