Brain Structural Correlates of Metacognition in First-Episode Psychosis

Abstract

Metacognition is impaired in schizophrenia and is an important predictor of functional outcome, but the underlying neuropathology is not clear. Studies have implicated frontal regions and there is also some evidence that the hippocampus might play a pivotal role, but findings are inconsistent. We set out to more comprehensively investigate the neural underpinnings of insight in first-episode psychosis (FEP) using 2 metacognitive measures (the Beck Cognitive Insight Scale [BCIS]) and a perceptual metacognitive accuracy task alongside structural magnetic resonance imaging (MRI). We measured cortical thickness in insula and frontal regions, hippocampal (including subfield) volumes, hippocampal microstructure (using neurite orientation dispersion and density imaging [NODDI]), and fractional anisotropy in fornix. Relative to controls, FEP showed poorer metacognitive accuracy, thinner cortex in frontal regions and lower fornix integrity. In healthy controls (but not FEP), metacognitive accuracy correlated with cortical thickness in frontal cortex and insula. Conversely, in FEP (but not controls), metacognitive accuracy correlated with hippocampal volume and microstructural indices. Subicular hippocampal subregions were particularly implicated. No structural correlates of BCIS were found. These findings suggest that the neural bases of metacognition might differ in FEP: hippocampal (rather than frontal) integrity seems to be critical. Further, the use of objectively measured metacognitive indices seems to be a more powerful method for understanding the neurocircuitry of metacognition in FEP, which has the potential to inform therapeutic strategies and improve outcome in these patients.

Introduction

Metacognition (“thinking about thinking”) is the ability to be aware of one’s own thinking and observe and detect errors in one’s cognitive processes.¹ Metacognitive abilities have received considerable attention in schizophrenia research because metacognitive deficits could contribute to core symptoms: metacognitive failings might underlie increased capacity for false beliefs and delusion formation,² and individual differences in metacognitive ability could moderate the linkage between biological disturbance and functional disability.³ Impaired metacognition has been consistently identified in both first-episode psychosis (FEP) and patients with chronic schizophrenia and impacts functional capacity,⁴ affecting work function,⁵ quality of life,⁶ and social function.⁷ Evidence also suggests that metacognition might mediate relationships between neurocognitive abilities and functional capacity⁸ and predict functional outcome⁹; various therapeutic strategies have emerged specifically targeting metacognitive deficits.^3,10 Determining the neural correlates of metacognition, and how these relate to neuropathological mechanisms in schizophrenia, would help optimize these strategies.

Modular frameworks of metacognition^11,12 differentiate “discrete” metacognition (the appraisal of specific, singular cognitive events or constructs) from “synthetic” metacognition (the ability to integrate context and connections between such events into more complex representations). One element of synthetic metacognition is “cognitive insight.” Cognitive insight is self-awareness of cognitive distortions and misinterpretations¹³ and willingness to modify these with corrective information.^14,15 This contrasts with clinical insight (self-awareness of one’s own mental disorder and effects of medication¹⁶). In schizophrenia, cognitive and clinical insight appear to form independent constructs.¹⁷ To measure cognitive insight, the self-report Beck Cognitive Insight Scale (BCIS) assesses self-certainty (overconfidence) and self-reflection (awareness of dysfunctional reasoning styles and resistance to feedback). Low self-reflectiveness and/or high self-certainty in one’s own (mis)interpretations are proposed to underlie poor insight.¹³ BCIS and Positive and Negative Syndrome Scale (PANSS) symptom scores do not consistently correlate, but self-reflectiveness in FEP can predict severity of symptoms 4 years subsequent¹⁸ and patients with lower self-certainty respond better to cognitive remediation therapy,¹⁹ underlining the importance of cognitive insight in determining functional outcome and its potential as a therapeutic target.¹⁰ Other aspects of synthetic metacognition (not assessed by the BCIS) include emotional awareness (both of self and in others) and mastery (ability to utilize awareness to facilitate problem-solving); these aspects have also been linked to symptoms, functioning, and neurocognition in schizophrenia.¹²

Regarding neural correlates, studies in early psychosis have linked synthetic cognition to gamma activity,²⁰ prefrontal resting-state functional connectivity,²¹ and prefrontal gray matter density.²² Other studies have focused specifically on cognitive insight. These have implicated frontal regions, with thicker right ventrolateral prefrontal cortex (vlPFC) linked to higher self-reflectiveness and lower self-certainty in FEP²³ and self-confidence positively modulating covariance in cortical thickness in frontal networks.²⁴ The hippocampus might also be of particular importance. Buchy et al²⁵ found that self-certainty correlated negatively with bilateral hippocampal volumes in FEP. Orfei et al²⁶ localized this effect to left presubiculum volume. Importantly, no relationships between hippocampal subfield volumes and BCIS were seen in healthy controls, indicating that the role of hippocampus in cognitive insight might be specific to patients. Hippocampal disconnectivity is also implicated. Fractional anisotropy (FA, indexing fiber integrity, derived from diffusion tensor imaging [DTI]) in right fornix (connecting the hippocampus to frontal cortical regions) correlates with self-certainty in FEP.²⁷ Follow-up work, however, from the same groups has been unable to replicate these findings.^15,23 A more in-depth exploration into the role of hippocampal volume and connectivity in the neurobiology of cognitive insight is, therefore, required.

We set out to comprehensively investigate the neural underpinnings of cognitive insight in FEP using the BCIS and structural brain imaging. Fornix integrity (indexed by FA), cortical thickness in vlPFC and hippocampal (including subfield) volumes were measured and correlations with BCIS scores were investigated. Based on previous findings, we predicted positive associations between total BCIS score and cortical thickness in vlPFC and for self-confidence to correlate negatively with hippocampal volumes and fornix integrity. Further, hippocampal microstructure (as well as fornix FA) was assessed using neurite orientation dispersion and density imaging (NODDI), an advanced diffusion MRI technique.²⁸ NODDI provides estimates of both neurite density (NDI) and neurite orientation dispersion (ODI), 2 of the key variables contributing to FA. In gray matter, NDI reflects the density of cortical myelinated axons and ODI relating to the heterogeneity of neurite (dendrites and axons) fiber orientations.²⁹ To date, very few studies have made use of this emerging technique in schizophrenia: Nazeri et al report reduced NDI in hippocampus (which correlated with spatial working memory performance),³⁰ while a study in FEP found reduced NDI to underlie reduced FA in the main white matter tracts.³¹ Thus, the use of NODDI is a valuable and novel aspect of the present study. Here, we examine whether differences in hippocampal microstructure (NDI and ODI) might explain variance in metacognitive ability.

We also included a measure of discrete metacognition, employing a perceptual decision-making task³² to provide an objective measure of metacognitive accuracy (the correlation between metacognitive judgments and actual task performance) in patients and controls. As an objective measure, metacognitive accuracy might provide a cleaner route to elucidating the neurocircuitry of metacognition in schizophrenia, but few studies have explored this possibility. In healthy controls, metacognitive accuracy positively correlates with gray matter volume in medial PFC²⁵; converging evidence suggests this region plays a critical role in accurate metacognitive judgments of performance made retrospectively.³³ This region, alongside vlPFC and left insula, seems to be the key structure underlying self-judgments.³⁴ However, a previous volume-based morphometry study in the same sample used here found lower PFC volumes in the FEP group but no correlation with impaired metacognitive accuracy.³⁵ Thus, PFC structural differences appear not to be the source of metacognitive deficits in schizophrenia and this accords with other evidence of abnormal structure–function relationships.³⁶ Cortical thinning of PFC is well established in patients^36,37 and, while this is likely the biggest contributor to PFC volumetric deficits,³⁸ it has not been interrogated in relation to metacognitive impairment. Here, we investigated PFC cortical thickness differences between FEP and controls in medial PFC, vlPFC, and insula, and assessed correlations with metacognitive accuracy within each group. We predicted reduced frontal cortical thickness in FEP and frontal correlations with metacognitive accuracy in controls but not FEP. Hippocampal indices (volumes, microstructure, and fornix integrity) were also investigated in relation to metacognitive accuracy: on the basis of the BCIS findings,^25–27 we predicted that these hippocampal indices would correlate with metacognitive performance in FEP but not controls.

Methods

Participants and Procedure

Ethical approval was granted by the London-Camden and Islington National Health Service (NHS) Research and Ethics Committee (Ref:11/LO/1877, project ID:72141) and a local NHS research governance committee. Patients were recruited through Early Intervention in Psychosis (EIP) services in Sussex NHS Partnership Trust: inclusion criteria were a primary diagnosis of FEP by a UK psychiatrist, aged 18+, no history of organic neurological impairment, and no primary diagnosis of substance misuse. Duration of untreated psychosis (DUP) was not recorded because DUP appears not to affect metacognitive ability in FEP³⁹ or response to metacognitive training.⁴⁰

Control participants were recruited from the community through local media outlets: inclusion criteria were aged 18+ and no history of psychiatric/neurologic illness or substance misuse. Immediately prior to scanning, all participants completed the perceptual metacognitive accuracy task. The FEP group also completed the BCIS and was assessed using the short form of the PANSS.⁴¹ MRI protocol was the same for both groups. Written informed consent was obtained on the day of the study.

Fatigue prevented 4 patients from completing the metacognition task; 4 were also excluded from analysis due to poor accuracy. One patient was excluded due to poor quality T1 and one due to atypical neurology. The final sample consisted of 31 patients aged between 19 and 39 and 18 age-, gender-, and education-matched healthy controls aged between 18 and 38; see table 1. There were no significant differences between FEP and control populations on age, gender composition, or years of education (P > .05).

Table 1.

Participant characteristics by group (FEP, Control) and between-group P values from χ2 test (gender) and t-tests (age and education)

Characteristics	Controls (n = 18)	FEP (n = 31)	P
Age, M (SD)	24.06 (4.87)	26.16 (5.70)	.196a
Female (n, %)	5 (27.8%)	7 (23.7 %)	.683b
Education (years; SD)	13.66 (1.68)	13.58 (1.70)	.865a
Medication (olanzapine equivalent mg/day; SD)	N/A	13.24	N/A
PANSS (3-item) positive symptoms (mean; SD)	N/A	5.24	N/A
PANSS (2-item) cognitive disorganization (mean; SD)	N/A	3.32	N/A
PANSS (3-item) negative symptoms (mean; SD)	N/A	5.76	N/A
Metacognitive accuracy (Mratio; SD)	0.44 (0.34)	0.23 (0.37)	.036c
BCIS Self-Reflectiveness, M (SD)	N/A	14.44 (5.08)	N/A
BCIS Self-Certainty, M (SD)	N/A	6.23 (3.33)	N/A
BCIS Composite Index, M (SD)	N/A	8.26 (6.92)	N/A

Note: FEP, first-episode psychosis; M (SD), mean (standard deviation); PANSS, Positive and Negative Syndrome Scale; BCIS, Beck Cognitive Insight Scale; N/A, not applicable.

^aIndependent sample t-test.

^bPearson chi-square.

^cANCOVA (test of between-groups effect).

MRI

All imaging data were acquired with a Siemens Avanto 1.5T scanner (Siemens Erlangen, Germany) using a 32-channel head coil and gradient strength 44 mT/m. Volumetric analyses: A high-resolution T1-weighted structural image was acquired and processed using Freesurfer 6.0 to derive measures of cortical thickness in insula, middle frontal and vlPFC, and hippocampal subfield volumes (see Supplementary Material). NODDI: Three diffusion-weighted shells were acquired using b = 300, 800, and 2400 s/mm² with 9, 30, and 60 unique noncolinear diffusion directions, respectively. In addition, 11 nondiffusion-weighted volumes (ie, b ≈ 0) were acquired. A NODDI microstructural model was computed and fitted to the data using the NODDI toolbox for MATLAB (http://www.nitrc.org/projects/noddi_toolbox)²³ that provided measures of NDI, ODI, and FA in native space. After normalization, binarized masks were applied to calculate hippocampal NDI and ODI and fornix FA (see Supplementary Material).

Perceptual Metacognitive Accuracy Task

The perceptual metacognitive accuracy task was adapted from that of Fleming et al.³² The task consisted of a forced-choice visual perception task in which 2 sequential displays were presented. Each display contained 6 Gabor patch stimuli. One of the Gabors was manipulated to “pop-out” by increasing the contrast in the patch itself compared to neighbor patches (target).

Participants made a button-press response to indicate which display they believed the target Gabor had appeared in (first or second) and were then asked to report their confidence in the decision on a scale of 1–6. Meta-d’ (a measure of the correspondence between trial-by-trial accuracy and trial-by-trial confidence) was divided by d’(perceptual sensitivity) to yield a “bias-free” measure of metacognitive accuracy (metacognitive efficiency, “MRatio”).⁴² This measure is the current “gold standard” for quantifying metacognition because it is invariant to biases in decision accuracy and confidence.⁴³ An MRatio value of 1 equates to optimal metacognitive awareness where confidence perfectly tracks accuracy in response to the task, whilst lower values indicate suboptimal metacognition (see Supplementary Material).

Beck Cognitive Insight Scale

Cognitive insight was assessed using the BCIS,¹³ a self-report inventory consisting of 2 subscales: Self-Reflectiveness (SR, ability to understand and modify one’s own mental states, 9 items) and Self-Certainty (SC, overconfidence in the validity of one’s own beliefs, 6 items). Each item is rated on a 4-point scale. A Composite Index (CI) score (SR minus SC) reflects overall cognitive insight.

Statistical Analysis

IBM SPSS Statistics 21.0 was used to conduct statistical analyses. An ANCOVA (age as covariate) was used for the between-group comparison of metacognitive accuracy (MRatio). Age was included as covariate because previous work using the same task showed that age has significant effects on metacognitive performance.⁴⁴ Likewise, between-group comparisons for the MRI structural measures were conducted using ANCOVA (age as covariate). To investigate the relationships between BCIS scores, Mratio, and MRI measures, partial correlations were performed. For BCIS, scores were correlated against vlPFC cortical thickness (by subregion), right/left hippocampal volumes, Fornix FA, and hippocampal NDI/ODI. For Mratio, scores were correlated against these and (additionally) cortical thickness in middle frontal and insula. Follow-up analyses investigated correlations between BCIS/Mratio and hippocampal subfield volumes. To test for differences in correlations between groups, a frequentist approach was taken using a percentile bootstrap method.⁴⁵ After 2000 iterations, the distribution of bootstrap correlation differences was assessed to determine significance. This analysis was performed in MATLAB using code adapted from https://github.com/GRousselet/blog/tree/master/comp2dcorr and is more robust than other approaches.⁴⁶

Results

Behavior—Metacognitive performance: BCIS and Mratio

Means and SDs for BCIS and Mratio are reported in table 1. Metacognitive accuracy (Mratio) was significantly worse (F_(1,46) = 4.680, P = .036, η_p² = 0.092) in FEP (M = 0.23, SD = 0.37) compared to controls (M = 0.44, SD = 0.34), with age as covariate (table 1). In FEP, there were no correlations between BCIS and Mratio and no correlations between these measures and PANSS (supplementary table S1; all P values >.05, age as covariate).

MRI—Structural Differences Between FEP and Control Groups

Between-group differences were assessed using an ANCOVA with age as a covariate (table 2).

Table 2.

Between-group differences in cortical thickness and hippocampal volumes

Middle frontal thickness	Controls	FEP	F	P	η 2
	Mean (SD)	Mean (SD)
Right middle frontal thickness	2.41 (0.08)	2.35 (0.13)	1.661	.204	0.035
Left middle frontal thickness	2.44 (0.10)	2.35 (0.14)	4.666*	.036	0.092
Right vlPFC thickness
Pars opercularis	2.62 (0.10)	2.52 (0.15)	4.907*	.031	0.086
Pars orbitalis	2.73 (0.19)	2.61 (0.15)	5.220*	.026	0.091
Pars triangularis	2.49 (0.12)	2.37 (0.15)	7.118*	.010	0.120
Left vlPFC thickness
Pars opercularis	2.58 (0.09)	2.51 (0.16)	1.494	.227	0.028
Pars orbitalis	2.76 (0.18)	2.60 (0.24)	5.116*	.028	0.090
Pars triangularis	2.49 (0.15)	2.37 (0.14)	6.636*	.013	0.113
Insula thickness
Right hemisphere	3.07 (0.14)	2.98 (0.16)	3.079	.085	0.056
Left hemisphere	3.06 (0.10)	2.96 (0.14)	3.897	.054	0.070
Right hippocampal volume (%ICV)	0.194 (0.0132)	0.2005 (0.0152)	2.007	.163	0.042
Left hippocampal volume (%ICV)	0.192 (0.0102)	0.1952 (0.0102)	0.892	.460	0.012
Fractional anisotropy in fornix
Right fornix	0.2967 (0.0221)	0.2709 (0.0224)	14.912**	.000	0.216
Left fornix	0.2841 (0.0229)	0.2612 (0.0190)	14.238**	.000	0.209
NODDI
Right hippocampal microstructure
ODI	0.4160 (0.0184)	0.4264 (0.0223)	2.450	.123	0.043
NDI	0.4541 (0.0226)	0.4596 (0.0306)	0.252	.618	0.005
Left hippocampal microstructure
ODI	0.4288 (0.0226)	0.4369 (0.0185)	1.708	.197	0.031
NDI	0.4459 (0.0167)	0.4482 (0.0305)	0.024	.876	0.000

Note: FEP, first-episode psychosis; HC, healthy controls; vlPFC, ventrolateral prefrontal cortex; NODDI: neurite orientation dispersion and density imaging; ODI, Orientation Dispersion Index NDI, neurite density. *P < .05, **P < .01.

Cortical Thickness

In FEP, significantly thinner cortex was present in left middle frontal and in vlPFC: bilateral pars triangularis, bilateral pars orbitalis, and right hemisphere pars opercularis.

Hippocampal Volumes

No significant between-group differences were observed in the intracranial volume (ICV)-adjusted hippocampal volumes or within any of the hippocampal subfields (all P values >.05) except for right CA3 volume (P = .046) where greater volume was seen in FEP (supplementary table S2).

Fornix FA

The FEP group had significantly lower FA in both left and right fornix (P values <.001), suggesting compromised fornix integrity in FEP relative to controls.

Neurite Orientation Dispersion and Density Imaging

There were no significant differences between groups in hippocampal ODI or NDI (all P values >.05), suggesting that hippocampal neurite orientation heterogeneity and NDI did not differ in FEP.

Correlations Between Metacognitive and MRI Measures

Partial correlation analyses (controlling for age) were performed to investigate relationships between BCIS, Mratio, and MRI measures.

BCIS (FEP group)

There were no relationships observed with vlPFC cortical thickness, hippocampal volumes, FA in fornix, or any of the NODDI measures (supplementary table S3 and S4).

Metacognitive Accuracy (Mratio)

On cortical thickness, Mratio scores in the control group positively correlated with right pars orbitalis thickness and left hemisphere insula thickness. No relationships were observed with cortical thickness in FEP (table 3). Between groups, a significant difference in correlation was seen for right pars orbitalis thickness (P = .008, 1-tailed). A 1-tailed test was used in accordance with our a priori hypothesis that frontal-metacognition relationships would be evident in controls only. No significant between-group differences in correlation were seen for left insula (P = .254, 2-tailed).

Table 3.

Correlations between metacognitive accuracy (Mratio) and cortical thickness/hippocampal volumes, by group

	Controls		FEP
	r	P	r	P
Middle frontal thickness
Right middle frontal thickness	.206	.428	.211	.300
Left middle frontal thickness	.386	.126	.262	.211
Right vlPFC thickness
Pars opercularis	−.219	.398	−.011	.955
Pars orbitalis	.513*	.035	−.305	.108
Pars triangularis	.265	.305	.173	.368
Left vlPFC thickness
Pars opercularis	.159	.543	.191	.320
Pars orbitalis	.467	.059	.281	.140
Pars triangularis	−.037	.889	.232	.226
Insula thickness
Right hemisphere	.405	.107	.136	.483
Left hemisphere	.533*	.027	.009	.963
Right hippocampal volume	.053	.839	.474**	.008
Left hippocampal volume	−.051	.846	.416*	.022

Note: FEP, first-episode psychosis; vlPFC, ventrolateral prefrontal cortex.

*P < .05, **P < .01.

On hippocampal volumes, Mratio scores positively correlated with total ICV-adjusted hippocampal volume both in right and left hemispheres in FEP, but no associations were seen in controls (table 3; figure 1). Between groups, a significant difference in correlation was seen for right hippocampus (P = .0235, 1-tailed) and left hippocampus (P = .0475, 1-tailed). A 1-tailed test was used in accordance with our a priori hypothesis that hippocampal–metacognition relationships would be evident in FEP only.

Figure 1.

Scatterplots of metacognitive accuracy score (Mratio) against ICV-adjusted hippocampal volumes by group/hemisphere.

No associations were seen with fornix FA for either group. For the NODDI measures in hippocampus, Mratio in FEP positively correlated with NDI (right: P = .008, left: P = .010), while no associations were seen in controls (supplementary table S4). However, between groups, the difference in correlation was not significant for either right hippocampus (P = .087, 1-tailed) or left hippocampus (P = .075, 1-tailed).

Correlations Between Metacognitive Measures and Hippocampal Subfield Volumes

To further explore the associations between hippocampal volumes and metacognitive accuracy, a follow-up, exploratory partial correlation analysis was conducted (controlling for age) between hippocampal subfield volumes and Mratio. In controls, there were no significant relationships between any of the subfield volumes and Mratio scores. By contrast, significant relationships were seen in FEP, including subiculum, presubiculum, and CA4 bilaterally, and left CA1 (table 4). In FEP, neither BCIS (supplementary table S5) nor PANSS (supplementary table S6) correlated with any hippocampal subfield volume.

Table 4.

Correlations between metacognitive accuracy (Mratio) and hippocampal subfield volumes by group

	Controls		FEP
	r	P	r	P
Left hippocampal subfields
CA1	.203	.434	.367*	.046
CA3	.302	.239	.250	.182
CA4	−.039	.881	.418*	.021
Subiculum	−.163	.533	.428*	.018
Presubiculum	−.310	.226	.419*	.021
Parasubiculum	−.379	.134	.083	.661
Molecular layer	.164	.529	.455*	.012
GC-ML-DG	−.110	.675	.382*	.037
Hippocampal tail	.075	.775	.044	.816
Right hippocampal subfields
CA1	.123	.638	.291	.119
CA3	−.014	.956	.224	.233
CA4	.042	.872	.496**	.005
Subiculum	.127	.628	.544**	.002
Presubiculum	−.058	.826	.541**	.002
Parasubiculum	−.226	.384	.416*	.022
Molecular layer	.099	.706	.448*	.013
GC-ML-DG	.001	.997	.472**	.008
Hippocampal tail	.080	.759	.141	.456

Note: CA, cornu ammonis; FEP, first-episode of psychosis; GC-ML-DG, granule cells in the molecular layer of the dentate gyrus.

*P < .05, **P < .01.

Discussion

In this study, we set out to better determine the neural correlates of cognitive insight and metacognitive accuracy in FEP. Poorer metacognitive accuracy was evident in FEP relative to controls, and distinct relationships with brain structural indices were identified.

Cortical thickness was extracted within predefined regions of interest (ROIs) implicated by previous studies. Associations between vlPFC thickness and both BCIS-SR and BCIS-SC scores have been shown previously in FEP,²³ whilst a relationship between BCIS-SR and right vlPFC volume has been found in schizophrenia.¹⁵ Although we found that FEP had significantly thinner cortex within vlPFC (which accords with previous findings^47,48), this did not correlate with BCIS scores. Another recent study also failed to find any association even when assessed longitudinally,¹⁴ suggesting that the relationship might not be robust. It is noteworthy that inconsistencies also exist in findings linking BCIS to frontal (executive) functioning. Some studies have implicated only self-confidence,^49,50 others self-reflectiveness.²⁵

We also assessed in depth the potential role of hippocampus, motivated by previous structural neuroimaging studies linking overconfidence to hippocampal volume and fornix integrity deficits in FEP,^25,27 with left presubiculum particularly implicated.²⁶ Here, BCIS scores did not correlate with hippocampal volumes (including individual subfields), microstructure (from NODDI), or fornix FA. Likewise, follow-up studies by those same groups did not replicate their findings.^14,15,23 This lack of consistency across studies suggests that the BCIS might not be an optimal method for determining the neural structures underlying metacognition in schizophrenia.

A more powerful method might be to adopt an objective measure, such as the perceptual metacognitive accuracy task used here. In contrast to the BCIS findings, correlations between MRI measures and perceptual metacognitive accuracy (Mratio) revealed interesting effects and suggests major differences in the structural bases of metacognition in FEP relative to controls. In controls, correlations between cortical thickness and Mratio were evident in right vlPFC (pars orbitalis) and left insula. These regions have been consistently implicated in self-related processing.³⁴ In contrast, these correlations were absent in FEP; results instead implicated the hippocampus as a key structure underlying metacognitive accuracy in this group. Although no between-group differences were observed in hippocampal volumes, both left and right hippocampal volume correlated with metacognitive accuracy in FEP only. No correlations were seen between hippocampal volumes and symptom severity, which accords with other studies⁵¹ and points to a relationship specific to metacognitive ability rather than overall symptomology. Parallels can be drawn with the previous (albeit inconsistent) BCIS findings linking smaller hippocampal volumes to poorer insight^25,27 with such relationships apparent only in schizophrenia patients but not controls.²⁶ We found that correlations with metacognitive accuracy were particularly striking in CA4, subiculum, and presubiculum, although there were no volume differences in these subfields between FEP and controls. Orfei et al¹⁵ explained their findings linking BCIS-SC to presubiculum volumes in relation to the interconnectivity between subicular complex and the anterior thalamic nucleus (implicated in memory-guided cognitive control); overrealistic self-confidence in schizophrenia, contributing to poor insight, could be due to an inability to consolidate and integrate new episodic memories about the self.

Subiculum hyperactivation is observable after psychosis onset⁵²; it has been suggested that hypermetabolism in this region triggers subsequent hippocampal volume reduction in schizophrenia⁵³ and, thus, might be central to schizophrenia’s neuropathological development. It also might be a critical structure underlying error monitoring deficits in schizophrenia. It has been proposed that dopaminergic dysfunction in schizophrenia is characterized by abnormal connectivity between subiculum and ventral tegmental area, which results in a mismatch between cortical input to CA1 (representing perceptual information) and CA3 input to CA1 (representing predicted information).^54,55 This provides another potential explanation as to why subiculum integrity could be particularly important for metacognitive judgments of perceptual accuracy. The current findings provide strong evidence of a link between subicular complex and metacognition specific to FEP because strong volume-accuracy correlations were observed in subiculum and presubiculum, bilaterally.

The inclusion of NODDI is a valuable and novel aspect of the current work. NODDI is an emerging neuroimaging technique that can quantify gray matter microstructure. A previous study found reduced hippocampal NDI in a schizophrenia population.³⁰ Here, no hippocampal differences in ODI or NDI were seen, although FEP did have lower fornix FA than controls, replicating previous findings.⁵⁶ However, clear associations between hippocampal microstructure (NDI) and perceptual metacognitive ability were observed in FEP only, providing further evidence of the importance of hippocampus in the neurocircuitry of metacognition in FEP. Postmortem studies of schizophrenia patients have reported reduced dendritic density in the subiculum,⁵⁷ consistent with the volumetric findings discussed above.

Overall, two important points can be drawn from these findings. First, the BCIS, as a self-report measure of metacognition might not be an optimal tool for exploring the neuroanatomical substrates of metacognition. Previous findings using the BCIS have been inconsistent and no structural correlates were identified here in any of the neuroimaging measures used. Instead, results suggest that objective indices of metacognitive judgments might be preferable for studying metacognitive neurocircuitry in patient groups. Second, our findings further implicate the hippocampus as a key structure in this circuitry. Previous work with the BCIS has highlighted this possibility but, here, we show volumetric correlations with metacognitive ability in hippocampus and specific subfields, in FEP only, and in the absence of between-group differences. Similar patterns in the microstructural indices lend further support. By contrast, cortical thickness in vlPFC and insula was implicated in the healthy control group, consistent with previous work. Thus, the neuroanatomy of metacognition seems to be strikingly different in FEP, with evidence of hippocampal involvement unique to this group.

It should be noted that the BCIS focuses on but one aspect of synthetic metacognition (cognitive insight) and, as such, the current findings are likely not generalizable to synthetic metacognition as a whole. The BCIS was designed as a brief, self-report tool. Further work employing fuller measures of synthetic metacognition based on semistructured interviews (eg, the Metacognition Assessment Scale⁵⁸) is required in order to explore brain structural correlates of synthetic metacognition.

Strengths of the current study include the use of an age-, gender-, and education-matched control group, inclusion of a rigorous objective measure of metacognition (upon which data is lacking in schizophrenia), and the use of various powerful neuroimaging techniques to comprehensively assess the role of relevant neural structures. In particular, the NODDI data provided novel insight into hippocampal microstructure in FEP: there is no previous published work on this. However, some important limitations should be noted. The study was cross-sectional: further work is needed to explore how the relationships identified here relate to outcome and whether they are responsive to interventions that aim to improve metacognition. It is noteworthy that some of the significant P values were close to the significance threshold. As such, results would benefit from replication in larger samples because the control group, in particular, was small in number. Inclusion of cognitive measures to assess potentially important links with memory performance would also be beneficial.⁴ Future work should also consider the effects of other variables not recorded in the present study. These include effects of ethnicity (which has been linked to prevalence⁵⁹) and effects of nondrug treatments: although the cognitive behavioral therapies routinely offered by front-line services do not directly target metacognitive ability, they might have some indirect benefits. These lines of enquiry merit further study.

Considerable evidence now exists suggesting that hippocampal function and integrity might be central to the pathophysiology of schizophrenia, and this work adds further to that evidence base. Encouragingly, emerging evidence suggests that antipsychotic treatment (particularly second-generation treatments) might stimulate hippocampal growth in FEP⁶⁰; cognitive remediation therapy has also been shown to enhance hippocampal volume in schizophrenia.⁶¹ Future studies should investigate the responsivity of the CA4 and subicular subfields implicated here and whether this can result in metacognitive improvements. Conversely, the present findings are also of relevance to emerging metacognition-oriented therapies (such as metacognition-oriented social skills training⁶²), which show considerable potential for improving clinical and functional outcomes. By describing the brain structural underpinnings of metacognitive deficit in schizophrenia, these findings point to perceptual metacognitive accuracy as a reliable indicator of metacognitive deficit that correlates with FEP-specific brain structural indices. Investigations into whether metacognitive-oriented therapies can affect these indices would provide valuable data that could optimize intervention strategies.

Funding

This work was supported by the Dr. Mortimer and Theresa Sackler Foundation (grant number 1128926).