Incomplete hippocampal inversion: a neurodevelopmental mechanism for hippocampal shape deformation in schizophrenia

Participants

This study includes 199 patients with a schizophrenia spectrum disorder diagnosis (86 schizophreniform disorder, 77 schizophrenia, 36 schizoaffective disorder; referred to here collectively as “schizophrenia”) and 161 healthy control participants matched to individuals with schizophrenia on age, gender, race, and parental education. The majority of the 86 schizophreniform disorder patients were followed for 2 years and final diagnoses were: 38 schizophreniform disorder (44.2% of the original diagnoses), 41 schizophrenia (47.7%), and 7 schizoaffective disorder (8.1%). Therefore, our schizophrenia spectrum disorder sample is primarily a schizophrenia cohort (118 out of 199, 59.3%). IHI prevalence and severity differed in both groups (Table 1). We previously reported on IHI criteria, grading, and reliability in this cohort (32). Both groups were drawn from an ongoing data repository, the Psychiatric Phenotype/Genotype Project (PGPP, NCT00762866). Patients in this repository are recruited from the inpatient unit and outpatient clinics at the Vanderbilt University Medical Center Psychotic Disorders Program, and healthy control participants are recruited from the local community via advertisement. The Vanderbilt University Institutional Review Board approved the study. Participants provided written informed consent and were compensated for their time. The Structured Clinical Interview for DSM-IV was used for diagnostic assessment (40). Participant exclusion criteria include significant medical or neurological illness, age under 16 or over 65, pregnancy, head injury, or meeting criteria for substance abuse within the past month. Healthy control participants were excluded if they had a current or past psychiatric illness, a first degree relative with a psychotic illness, or psychotropic drug use. All participants with a T1-weighted MRI scan without motion artifact were selected from the PGPP repository for inclusion in this study.

MRI acquisition and processing

Structural MRI acquisition was completed on a 3T Philips Intera Achieve scanner at the Vanderbilt University Institute of Imaging Sciences (Philips Healthcare, Inc.). Participants received a 3D T1-weighted scan (voxel resolution: 1mm³; field of view = 256²; number of slices = 170; TE = 3.7ms; TR = 8.0ms). All images were visually inspected and determined to be free from motion or other artifacts prior to inclusion in the analysis.

Each image was processed using the Freesurfer 6 (41,42) hippocampal subfield module (43) with standard parameters. Segmentations were visually inspected to correct those with tissue labeled outside the hippocampus or incomplete labeling of the hippocampus. Automated segmentation failed because of two major errors: 1) hippocampal segmentation extending beyond the border of the hippocampus into surrounding structures and 2) amygdala segmentation extending into the hippocampal head. To maximize study inclusion, automated segmentation failures were corrected by manually deleting segmented voxels extending outside of the hippocampus or by manually replacing amygdala segmentation voxels with hippocampal head segmentation voxels at the amygdala-hippocampal border. Manual voxel correction was completed using ITK-SNAP version 3.8.0 (44). Volume data from all participants were included in a previous study (45).

Hippocampal surface processing

Shape analysis was completed using the SPHARM-PDM toolkit (21,46) to generate hippocampal surfaces and their initial spherical mapping and correspondence. A detailed description of this methodology is available in Styner et al. (21). We then adapted spherical surface registration proposed in Lyu et al. (47) to further refine the initial shape correspondence in a group-wise manner that estimates an unbiased representative (average) surface of the cohort. High-resolution hippocampal surfaces were generated with 40,962 vertices per hemisphere via a standard icosahedral subdivision (48). The local shape variation of each participant’s hippocampal surface was calculated from the cohort average of all participants by quantifying the perpendicular distance between surfaces at a vertex-to-vertex level. The quantification of perpendicular change between surfaces was assigned a positive (outward displacement from cohort average) or negative (inward displacement from cohort average) value.

Statistical analyses

To test our hypotheses, shape variation of hippocampal surfaces was examined in surface-based analyses performed using SurfStat (49) implemented in MATLAB (MatLab 2019a). The left and right hippocampus were analyzed separately. Random field theory (RFT) was applied using SurfStat to account for multiple comparisons. To identify significant clusters using RFT, an initial height threshold of p ≤ 0.01 was implemented at the vertex level, and a corrected family-wise error (p ≤ 0.05) was subsequently applied. T-values generated from SurfStat analyses were converted to effect sizes (Cohen’s d) using R (R Core Team, 2019) package compute.es (50).

To replicate previous findings demonstrating hippocampal shape differences between schizophrenia patients and healthy control participants, displacement values were regressed onto Group (schizophrenia patient, healthy participant) using a general linear model adjusted for estimated intracranial volume (ICV), Age, and Gender (Displacement ~ Group + ICV + Age + Gender).

To determine whether hippocampal shape is impacted by IHI, displacement values were regressed onto IHI severity scores using a general linear model fitted by adjusting for ICV, Age, and Gender (Displacement ~ IHI + ICV + Age + Gender). To test the contribution of IHI to shape differences between schizophrenia patients and healthy control participants, displacement values were regressed onto Group and IHI score using a general linear model fitted by adjusting for ICV, Age, and Gender (Displacement ~ Group + IHI + ICV + Age + Gender). To test the impact of IHI on both groups separately, we performed an omnibus analysis regressing displacement values onto Group, IHI score, and the interaction between Group and IHI score using a general linear model fitted by adjusting for ICV, Age, and Gender (Displacement ~ Group + IHI + Group*IHI + ICV + Age + Gender; See Supplemental Information).

Between-group hippocampal shape analysis

A comparison of hippocampal shape between healthy control participants and patients with schizophrenia revealed significant differences in both hemispheres. Significant clusters are visualized on hippocampal surfaces in Supplemental Figure 1. In the left hippocampus, inward shape displacements are located at the uncus (cluster 1 in Supplemental Figure 1), anterolateral body (cluster 2), and posterior-most surface of the medial tail (cluster 3). Outward shape displacement is restricted to the ventral body (cluster 4). In the right hippocampus, inward shape displacements are clustered at the lateral tail (cluster 5), uncus (cluster 6), anterolateral body (cluster 7), ventrolateral tail (cluster 8), and ventrolateral body (cluster 9). Outward shape displacement is limited to the dorsal tail (cluster 10). Peak t-values, effect sizes (Cohen’s d), the number of significant vertices, and RFT corrected p-values for all significant clusters are summarized in Table 2. Additional hippocampal surface maps in Supplemental Figure 2 illustrate the mean difference in vertex displacement in our patient cohort (Panel A) and t-values for the between-group analysis (Panel C).

Effect of IHI on hippocampal shape

IHI severity is associated with outward displacements in the dorsal and ventral hippocampal surfaces and inward displacements in the medial and lateral hippocampal surfaces bilaterally. Areas of significance contain both clusters and individual vertices. There are more significant vertices and larger clusters of significance on the left than the right hippocampal surface. Significant clusters are visualized on hippocampal surfaces in Figure 1, and the peak t-values, effect sizes (Cohen’s d), the number of significant vertices, and RFT corrected p-values for all significant clusters are summarized in Table 3. Surface maps demonstrating the t-values of this analysis are found in Supplemental Figure 2 (Panel B). We found a similar main effect of IHI in the healthy control sample (see Supplementary Information and Supplemental Figure 7).

Between-group hippocampal shape analysis including IHI as a main effect

Including IHI as a main effect in our linear model results in the elimination of significant clusters of shape variation in the left and right hippocampus. In the left hippocampus, the inward displacements located at the anterolateral body (cluster 2) and medial tail (cluster 3) no longer reach statistical significance. The significant outward displacement in the ventral body is also no longer significant (cluster 4). In the right hippocampus, the inward displacement visualized at the anterolateral body (cluster 7) is no longer significant. All other clusters in our previous between-group model remained significant after including IHI in the model. The impact of IHI on significant clusters is visualized on hippocampal surfaces in Figure 2. The changes in all clusters are summarized in Table 2. Surface maps demonstrating the t-values of this analysis are found in Supplemental Figure 2 (Panel D). Covarying for alcohol and cannabis use disorder in our models did not change the impact of IHI on hippocampal shape differences (see Supplemental Information and Supplemental Figures 8–9)

Hippocampal shape is abnormal in schizophrenia

Our analysis replicates significant shape deformation findings reported in previous studies of patients with schizophrenia spectrum disorders. Inward displacements of the uncus (10,14,17–19,35,39), anterolateral body (6,7,18,24,8,9,11,12,14–17), lateral tail (7,14,21), medial tail (13,15,16,19,20), and ventrolateral areas (7,14) and outward displacements of the ventral body (19,39) and dorsal tail (11,12,19,23) have been reported previously.

Our analysis benefits from a large cohort, high-fidelity surface reconstructions with 40,962 vertices, and a conservative statistical approach to allow for the precise localization of significant clusters throughout the hippocampus. There is some evidence that hippocampal shape deformities begin in the CA1 subfield and progress over time (12,51) in individuals with schizophrenia. The mean duration of illness in our patient cohort is 7.70 years, indicating that progressive shape changes may have occurred across the hippocampus.

IHI impacts hippocampal shape

IHI has a significant impact on hippocampal shape regardless of diagnosis. This is evident by the number of significant individual vertices and large effect sizes of some clusters. As expected, the severity of IHI is associated with an outward displacement of the dorsal and ventral surfaces and inward displacement of the medial and lateral surfaces of the hippocampus. This pattern matches the rounded anatomy described in IHI grading criteria and visualized in coronal sections of IHI cases (34). These results also confirm qualitative descriptions that IHI primarily affects hippocampal anatomy in the head and body (34) as opposed to the tail (52). The greater impact of IHI on left hippocampal shape is likely due to the higher prevalence and greater severity of IHI in the left hippocampus.

In our previous work, we reported that IHI affects hippocampal volume and asymmetry, and here we demonstrate the extensive impact of IHI across the hippocampal shape. We conclude from our findings that IHI should be included in future studies of hippocampal morphology. Furthermore, shape analyses should be added to studies of hippocampal volume, as shape deformations suggestive of neurodevelopmental anomalies may have higher sensitivity for detecting small changes in brain morphology (24). For example, previous studies of IHI in major depressive disorder (53) and in an aging cohort (54) did not find any whole hippocampal volume differences, but shape analyses might have revealed group differences.

Our recommendation to include IHI in future studies of hippocampal morphology raises two concerns. First, the grading of IHI is a manual, time-intensive process that will be difficult to sustain in the study of large datasets. Automated methods to identify IHI based on hippocampal shape will facilitate reporting on IHI in morphological studies. Second, shape differences attributable to IHI might still be informative of schizophrenia pathophysiology. For example, localized hippocampal surface differences (that are removed after inclusion of IHI in the analysis) might be related to clinical features.

IHI contributes to CA1 subfield shape deformations in schizophrenia

Incorporating IHI as a main effect in our shape analysis model eliminated the significant group difference in the anterolateral hippocampus in the right (p-value increase from 2.01E-05 to 0.87) and left (p-value increase from 1.34E-06 to 0.99) hemisphere. This hippocampal region consists primarily of the CA1 subfield. Results of an omnibus test analysis reiterate that IHI score is the main predictor of shape differences in schizophrenia (see Supplemental Results and Supplemental Figures 3–6). CA1 subfield deformations, particularly in the anterolateral region of the hippocampus, have been extensively reported in patients with schizophrenia (6,8–12,15–17). In the only study to date to investigate the effect of IHI on hippocampal subfield volumes, IHI was correlated with the volume of CA1, but not other hippocampal subfields, in 60 healthy subjects (54). Here, we show for the first time that IHI contributes to CA1 shape deformation in schizophrenia.

Possible mechanisms of IHI in schizophrenia

The mechanism by which IHI affects hippocampal structure and function in schizophrenia is unknown. IHI might indicate abnormal migration and development of interneurons, a process that occurs in the second trimester (55). Such inhibition/excitation imbalance might later result in hippocampal hyperactivity in the early stage of psychosis (56). A recent study reported that the genetically predicted expression of the brain complement component C4A is correlated with hippocampal shape deformations that co-localize with our CA1 findings (57). There is an increased risk of schizophrenia associated with genetically predicted C4A brain expression, and preclinical models have demonstrated that C4A promotes synapse elimination by microglia (58). These findings support a hypothesis of aberrant synaptic pruning in schizophrenia and suggest an interplay between synaptic pruning and changes in hippocampal morphology that could result in IHI.

Alternatively, IHI may serve as an indicator for other abnormalities in neurodevelopment that affect hippocampal morphology. For example, an arrest in development resulting in IHI could be a marker of a systemic obstetric complication (i.e., a trigger for aberrant in utero development) such as hypoxia or infection, which are associated with smaller hippocampal volumes (59) in schizophrenia patients and elevate the risk for schizophrenia 2–5 fold (60). The CA1 subfield is especially susceptible to hypoxia (17,61) in comparison to other hippocampal subfields (62).

A recent meta-analysis of studies documenting IHI reported increased prevalence rates of IHI in patients with epilepsy and malformations of cortical development (MCD) (63). Commonalities between these illnesses and schizophrenia may reveal information about the mechanism of IHI. The association between epilepsy, particularly temporal lobe epilepsy, and schizophrenia is well documented (64). Individuals with epilepsy have an 8.5-fold higher risk of developing schizophrenia (65). Evidence suggests that aberrant excitatory synaptic transmission resulting in an excitation/inhibition imbalance drives both psychosis and seizure activity (64). The most common MCD associated with IHI is periventricular nodular heterotopia (63), a condition characterized by abnormal neuronal migration during development of the fetal brain resulting in seizures (66). Altogether, the pathophysiologic similarities between these illnesses support a mechanism of IHI characterized by abnormal neuronal development and migration resulting in aberrant hippocampal activity.

Reconciling discrepancies between previous studies

A minority of studies do not illustrate any CA1 shape deformations in schizophrenia. We previously reported that automated hippocampal segmentation methods are more likely to fail in cases of IHI (32). Therefore, in studies using automated hippocampal segmentation to generate hippocampal shapes, IHI can be a potential confound. For instance, if individuals with failed segmentations are excluded from analyses, the mean shape estimates will underestimate the true shape difference in populations with greater IHI frequency or severity (e.g., patients with schizophrenia). Here, we reconstructed hippocampal surfaces using manually corrected automated segmentation failures, allowing us to include all cases of IHI.

Another inconsistent finding reported in the literature is the location of CA1 shape deformations in schizophrenia. As highlighted in Gutman et al. (8), the use of Freesurfer, such as in this study, to generate gray matter boundaries may account for CA1 shape differences that are more posterior than the anterior CA1 shape deformities reported by groups using diffeomorphic surface mapping methods (23). However, our CA1 shape deformities replicate the results of a recent meta-analysis of hippocampal shape, including 2,833 patients with schizophrenia and 3,929 healthy volunteers in the ENIGMA consortium (8), and the largest individual shape deformation study of the hippocampus in schizophrenia (14). Both studies localize areas of significant shape deformation to the anterolateral CA1 hippocampus, which is consistent with our findings.

Limitations

Our study has several limitations. First, the majority of schizophrenia patients in this study received antipsychotic medications, which may affect hippocampal structural measures (67). However, in a recent shape analysis study of patients with schizophrenia spectrum disorders, medication dose did not affect the shape of the hippocampus (14). Of note, this analysis demonstrated anterolateral inward deformations, which the authors attribute to the CA1 subfield. Additionally, shape analyses of antipsychotic-naive patients also show CA1 shape inward deformations (15) in addition to other widespread shape differences.

Second, we could not assess the cumulative exposure of substances, including cannabis and alcohol, to hippocampal shape deformations. Shape analyses illustrate the additive effects of alcohol (19) and cannabis (68) on hippocampal pathology in schizophrenia. Although participants with concurrent diagnoses of alcohol or cannabis use disorder were excluded from participation in this study, we collected the lifetime history of substance use disorders in this cohort (Table 1). Covarying for alcohol and cannabis use disorder in our shape analysis model does not change the results of our between-group analyses, suggesting that a history of substance use disorder does not contribute to the shape deformations in schizophrenia we report.

Third, we did not assess whether participants had a history of obstetric complications. Since obstetric complications are associated with schizophrenia (60), the relationship between illness, obstetric complications, and IHI needs to be established in future studies.

Fourth, we did not assess hippocampal shape asymmetries. Previous investigations have reported abnormalities in hippocampal shape asymmetries in patients with schizophrenia (23,25,38). We have previously reported that IHI contributes to greater hemispheric volume asymmetries in schizophrenia (32). Therefore, future studies should confirm that IHI also contributes to hippocampal shape asymmetry.

Lastly, we did not investigate whether there is a relationship between clinical measures and IHI in the context of hippocampal shape. To date, one study has investigated the link between IHI and clinical outcomes, reporting that patients with visual hallucinations possess more IHI-specific morphological patterns (69). Additional work should attempt to link clinical measures with shape changes and IHI, similar to previous shape analysis studies relating clinical outcomes and cognitive features with shape deformations (6,15–17,23).

Conclusions

This is the first study to report the impact of IHI on hippocampal shape. Our results suggest that IHI has a significant effect on hippocampal morphology and contributes to CA1 shape deformations in schizophrenia. This provides evidence that hippocampal shape differences in schizophrenia can be attributed to an anatomical variant arising during the perinatal period, long before the emergence of the clinical phenotype.