Patient and healthy control group assessment

The patient group, consisting of adults with suspected AP spectrum syndromes, was recruited from current and former patients at the Department of Psychiatry and Psychotherapy of the Medical Center, University of Freiburg, Germany. No diagnosis was made more than 10 years before the research MRI/EEG examinations. Eligible patients were adults (age ≥18 years) who tested positive for: (1) well-characterised neuronal antibodies, (2) novel CNS antibodies, or (3) systemic antinuclear (on human embryonic kidney cells) or thyroid antibodies with distinct evidence of inflammatory brain involvement in further diagnostic work-ups (i.e., they had been diagnosed with neuropsychiatric lupus or Hashimoto encephalopathy; Jeltsch-David & Muller, Reference Jeltsch-David and Muller2014; Graus et al., Reference Graus, Titulaer, Balu, Benseler, Bien, Cellucci, Cortese, Dale, Gelfand, Geschwind, Glaser, Honnorat, Höftberger, Iizuka, Irani, Lancaster, Leypoldt, Prüss, Rae-Grant, Reindl, Rosenfeld, Rostásy, Saiz, Venkatesan, Vincent, Wandinger, Waters and Dalmau2016; Legge & Hanly, Reference Legge and Hanly2024). Clinically, the term “psychosis” lacks precise definition in the international diagnostic criteria for AP (cf. Pollak et al., Reference Pollak, Lennox, Müller, Benros, Prüss, Tebartz van Elst, Klein, Steiner, Frodl, Bogerts, Tian, Groc, Hasan, Baune, Endres, Haroon, Yolken, Benedetti, Halaris, Meyer, Stassen, Leboyer, Fuchs, Otto, Brown, Vincent, Najjar and Bechter2020). From a clinical perspective, this approach is justified, as autoantibody-mediated processes often do not align with classification systems for primary mental disorders, such as paranoid schizophrenia. Other authors therefore suggested the concept of APS, which was followed here for patient inclusion (Al-Diwani et al., Reference Al-Diwani, Pollak, Langford and Lennox2017; Hansen et al., Reference Hansen, Lipp, Vogelgsang, Vukovich, Zindler, Luedecke, Gingele, Malchow, Frieling, Kühn, Denk, Gallinat, Skripuletz, Moschny, Fiehler, Riedel, Wiedemann, Wattjes, Zerr, Esselmann, Bleich, Wiltfang and Neyazi2020), so that also a subgroup with predominant affective spectrum syndromes was included. Cases in which neurocognitive manifestations co-occurred with schizophreniform psychosis or were the predominant clinical syndrome, were classified as “schizophreniform psychosis spectrum”. Obsessive–compulsive syndromes in combination with affective syndromes were classified as “affective spectrum syndromes” (as was one severe isolated predominant obsessive–compulsive syndrome with a previously diagnosed comorbid depressive episode). Patients were included regardless of whether they had experienced acute, chronic, and (partial) remitted disease stages. A time criterion for a subacute onset (e.g., of three months as in the Pollak criteria; Pollak et al., Reference Pollak, Lennox, Müller, Benros, Prüss, Tebartz van Elst, Klein, Steiner, Frodl, Bogerts, Tian, Groc, Hasan, Baune, Endres, Haroon, Yolken, Benedetti, Halaris, Meyer, Stassen, Leboyer, Fuchs, Otto, Brown, Vincent, Najjar and Bechter2020) was not required for study participation. Immunotherapies could have been implemented before the MRI/EEG research measurements; all psychopharmacological therapies were recorded. As part of the diagnostic routine work-up, all patients underwent blood analyses, CSF testing, clinical EEG (analysed by the responsible physicians), and conventional MRI (analysed by senior board-certified neuroradiologists). Most patients also received a cerebral FDG-PET, which was analysed by senior board-certified nuclear medicine specialists (Endres et al., Reference Endres, Meixensberger, Dersch, Feige, Stich, Venhoff, Matysik, Maier, Michel, Runge, Nickel, Urbach, Domschke, Prüss and Tebartz van Elst2020b; Endres et al., Reference Endres, von Zedtwitz, Matteit, Bünger, Foverskov-Rasmussen, Runge, Feige, Schlump, Maier, Nickel, Berger, Schiele, Cunningham, Domschke, Prüss and Tebartz van Elst2022a). Before starting the study, we assumed that almost all patients would show clinical EEG alterations based on the data available from NMDA-R encephalitis at the time (Titulaer et al., Reference Titulaer, McCracken, Gabilondo, Armangué, Glaser, Iizuka, Honig, Benseler, Kawachi, Martinez-Hernandez, Aguilar, Gresa-Arribas, Ryan-Florance, Torrents, Saiz, Rosenfeld, Balice-Gordon, Graus and Dalmau2013), but this was not the case. EEGs were additionally automatically analysed by the “avg-q” software developed in house (https://github.com/berndf/avg_q). IRDA/IRTA rates were detected with a frequency between 2 Hz and 7 Hz, and the amplitude threshold for IRDA/IRTA classification was set at >1 µV (cf. Endres et al., Reference Endres, Maier, Feige, Mokhtar, Nickel, Goll, Meyer, Matthies, Ebert, Philipsen, Perlov and Tebartz van Elst2017b; von Zedtwitz et al., Reference von Zedtwitz, Feige, Maier, Schaefer, Nickel, Reisert, Spiegelhalder, Venhoff, Brumberg, Urbach, Dersch, Schiele, Domschke, van Elst, Coenen, Hannibal, Prüss, Maier and Endres2025a, Reference von Zedtwitz, van Elst, Feige, Matteit, Schlump, Lange, Runge, Nickel, Venhoff, Domschke, Prüss, Rau, Reisert, Maier and Endres2025b). IRDA/IRTA rates for pre-hyperventilation condition and IRDA/IRTA differences (IRDA/IRTA rate after versus before hyperventilation) were reported. The serum of all patients was tested for well-characterised neuronal antibodies against intracellular antigens using an immunoblot (Ravo®, Freiburg, Germany). Both, serum and CSF, were analysed for antibodies against well-characterised neuronal cell surface antigens using a fixed cell-based assay (Euroimun®, Lübeck, Germany). In addition, routine CSF parameters were determined, including WBC counts, protein levels, AQs, IgG indices, and OCBs. Most patients were additionally tested for novel CNS antibodies in serum and CSF using a tissue-based assay on unfixed murine brain tissue (Prof. Prüss, Laboratory for Autoimmune Encephalopathies, DZNE Berlin and Department of Experimental Neurology, Charité Berlin, Germany; Kreye et al., Reference Kreye, Reincke, Kornau, Sánchez-Sendin, Corman, Liu, Yuan, Wu, Zhu, Lee, Trimpert, Höltje, Dietert, Stöffler, von Wardenburg, van Hoof, Homeyer, Hoffmann, Ab-delgawad, Gruber, Bertzbach, Vladimirova, Li, Barthel, Skriner, Hocke, Hippenstiel, Witzenrath, Suttorp, Kurth, Franke, Endres, Schmitz, Jeworowski, Richter, Schmidt, Schwarz, Müller, Drosten, Wendisch, Sander, Osterrieder, Wilson and Prüss2020).

The HC group was recruited via public announcements. A lifetime diagnosis of an axis I or II disorder according to the DSM-IV (assessed with a self-report questionnaire and verified using the Structural Clinical Interview SCID-I screening and SCID-II for borderline personality disorder), recent drug use (except for episodic cannabis use) within the last six months, current or past use of psychopharmacological medication, any relevant concomitant physical illness that may affect target parameters, past brain injury (e.g., traumatic brain injury, meningitis, encephalitis, seizures/epilepsy, hydrocephalus, or space-occupying processes), systemic autoimmune diseases with potential brain involvement (e.g., lupus erythematosus) or autoimmune diseases treated with immunotherapy (e.g., steroids, azathioprine) resulted in study exclusion.

Exclusion criteria for both the patient and HC group were pregnancy or lactation, lack of legal capacity or inability to comprehend the nature, significance, and scope of the study, as well as MRI-related contraindications, such as pacemakers, claustrophobia, or intrauterine contraceptive devices.

A structured questionnaire and a test battery summarised in Supplemental Table 1 were used to collect psychometric and neuropsychological data (cf. von Zedtwitz et al., Reference von Zedtwitz, van Elst, Feige, Matteit, Schlump, Lange, Runge, Nickel, Venhoff, Domschke, Prüss, Rau, Reisert, Maier and Endres2025b). Missing psychometric or neuropsychological test results did not automatically lead to exclusion, as long as the eligibility criteria were met and the diagnosis was clinically established. This flexibility was particularly useful for patients who were unable to complete the entire test battery due to their illness.

Table 1.

Characteristics and routine diagnostic findings of the suspected autoimmune psychosis spectrum syndrome group. Only the predominant antibody was mentioned (if several antibodies were positive)

Abbreviations: ANAs, Anti-nuclear antibodies; AP, Autoimmune Psychosis; CASPR2, Contactin-associated protein-like 2; CSF, Cerebrospinal Fluid; CNS, Central Nervous System; EEG, Electroencephalography; FDG-PET, [¹⁸F] Fluorodeoxyglucose positron emission tomography; HC, Healthy Controls; HV, Hyperventilation; IRDA/IRTA, Intermittent Rhythmic Delta/Theta Activity, LGI1, Leucine-rich glioma-inactivated 1; MOG, myelin oligodendrocyte glycoprotein; MRI, Magnetic resonance imaging; N, Number; NMDA-R, N-methyl-D-aspartate receptor; OCBs, oligoclonal bands; TPO, Thyroperoxidase; WBC, white blood cell.

A total of 40 suspected AP spectrum patients and 61 HCs initially agreed to participate. The results of the structural MRI data have been published seperately (Zedtwitz et al., Reference von Zedtwitz, Feige, Maier, Schaefer, Nickel, Reisert, Spiegelhalder, Venhoff, Brumberg, Urbach, Dersch, Schiele, Domschke, van Elst, Coenen, Hannibal, Prüss, Maier and Endres2025b). Reasons for exclusion are summarised in Fig. 1. Following an automatic matching approach to account for significant group differences in age and sex, data from 28 patients and 28 HCs could be analysed.

Figure 1.

Recruitment flowchart. *The morphometric imaging data were described in another publication (von Zedtwitz et al., Reference von Zedtwitz, van Elst, Feige, Matteit, Schlump, Lange, Runge, Nickel, Venhoff, Domschke, Prüss, Rau, Reisert, Maier and Endres2025b). Abbreviations: AP, autoimmune psychosis; HC, heathy controls; MRI, magnetic resonance imaging.

Magnetic resonance spectroscopy measurements

MRI and MRS measurements were performed at the imaging centre of the University Medical Center Freiburg in Germany using a Magnetom Prisma Fit 3 T system (Siemens Healthineers®, Erlangen, Germany) equipped with a 64-channel head and neck coil. High-resolution anatomical images were acquired using a T1-weighted magnetisation prepared rapid gradient echo (MPRAGE) sequence (time of repetition [TR]: 2000 ms; echo time [TE]: 4.11 ms; field of view [FOV] = 256 × 256 × 160 mm³; voxel size = 1 × 1 × 1 mm³). These images were used for region-specific spectroscopy analysis and to accurately position the MRSI volume of interest (VOI) along the anterior to the posterior commissure line. For MRSI, a spiral-encoded Mescher-Garwood localised adiabatic selective refocusing 3D-MRSI sequence (“MEGA-LASER 3D-MRSI”) (Bogner et al., Reference Bogner, Gagoski, Hess, Bhat, Tisdall, van der Kouwe, Strasser, Marjańska, Trattnig, Grant, Rosen and Andronesi2014) was applied with the following parameters: TR = 1600 ms, TE = 68 ms, flip angle = 90°, and 32 averages. The MRSI protocol included real-time motion and scanner instability correction (Bogner et al., Reference Bogner, Gagoski, Hess, Bhat, Tisdall, van der Kouwe, Strasser, Marjańska, Trattnig, Grant, Rosen and Andronesi2014). In addition, the LASER technique was used to reduce B1 inhomogeneity artefacts, minimise chemical shift errors, and improve the signal-to-noise ratio. The VOI was sized 80 × 90 × 80 mm and centered within a FOV of 160 × 160 × 160 mm, which encompassed the entire brain in most cases (see Fig. 2a). However, in a few instances, the cerebellum was not fully captured. The nominal voxel size was 4.096 cm³ with a resolution of 10×10×10 voxels. By interpolation, a voxel size of 1 cm³ could be achieved with 16×16×16 voxels. In edit-on/edit-off mode, the Gaussian editing pulses were applied at 1.9/7.5 parts per million (ppm) with a specified bandwidth of 60 Hz for a duration of 14.8 ms, and the centre frequency of the localisation pulses was 3.0 ppm. A total of 32 acquisition-weighted averages were acquired and a two-step phase cycle was used.

Figure 2.

The voxels studied in magnetic resonance spectroscopic imaging are shown in A). A total of 16×16×16 voxels with a size of 1 cm² were examined. The field of view area is shown in yellow. B) shows an exemplary spectrum. Abbreviations: GABA, gamma-aminobutyric acid; Glx, glutamate+glutamine; ppm, parts per million; NAA, N-acetylaspartate; Cr, creatine.

Magnetic resonance spectroscopy analyses

The MRSI data were processed using a two-step pipeline developed by Bogner et al. (Reference Bogner, Gagoski, Hess, Bhat, Tisdall, van der Kouwe, Strasser, Marjańska, Trattnig, Grant, Rosen and Andronesi2014) and Spurny et al. (Reference Spurny, Heckova, Seiger, Moser, Klöbl, Vanicek, Spies, Bogner and Lanzenberger2019). This approach integrates several tools, including Matlab version 2018a (MathWorks, Natick, MA, USA), Bash (Bourne-Again Shell, GNU Operating System), MINC version 1.5 (Medical Image NetCDF, McConnel Brain Imaging Center, Montreal, QC, Canada), linear combination of model spectra (LCModel) version 6.3 by S. Provencher (Oakville, ON, Canada; Provencher, Reference Provencher2001), and FreeSurfer (version 6.0; Fischl, Reference Fischl2012). This pipeline facilitates automatic identification and quantification of the metabolite ratios from individual voxel spectra (Fig. 2b). In the first step, LCModel was applied to each voxel to estimate neurochemical concentrations based on individual voxel spectra. This involved automatic estimation of metabolite signals with the corresponding Cramér-Rao lower bounds (CRLBs), and spectral quality measures such as signal-to-noise ratio and line width of the spectrum. Voxels were classified as “valid” if they met the criterion of CRLB<30 for Glx, GABA, tCr, and tNAA ratios. In the second step, voxel-wise metabolite estimates were mapped onto FreeSurfer-derived regions of interest (ROIs) (details see Spurny et al., Reference Spurny, Heckova, Seiger, Moser, Klöbl, Vanicek, Spies, Bogner and Lanzenberger2019). This relied on the automatic segmentation of structural brain data in FreeSurfer (version 6.0) using the recon-all command with default parameters on T1-weighted MPRAGE images to create structural ROI masks. These masks enabled regional quantification of metabolite concentrations by calculating the mean values across all valid voxels within each ROI. Regions with less than 80% voxels valid for tCr were excluded from the analysis and for the sub-analyses, regions with <80% valid voxels for the corresponding metabolites were also excluded. Additionally, participants with fewer than 80% valid voxels in more than 50% of the ROIs were excluded. To minimise susceptibility artefacts, the FOV was set slightly larger than in Spurny et al. (Reference Spurny, Heckova, Seiger, Moser, Klöbl, Vanicek, Spies, Bogner and Lanzenberger2019). Ultimately, 16 ROIs could be included in the final analysis (Fig. 3).

Figure 3.

Group comparisons of magnetic resonance spectroscopic imaging-derived ratios to tCr (left) and tNAA (right) are illustrated. The data presented here are adjusted for the Cramér-Rao-Lower bounds, but not corrected for multiple testing. *Significantly higher Glx/tCr ratio in the right caudate nucleus of the suspected AP spectrum syndrome group compared to the healthy control group and a significantly lower tCr/tNAA ratio in the left pallidum were identified; however, no significant results were present after correction for multiple comparisons (padj = 0.713 and padj = 0.656, respectively). Abbreviations: AP, autoimmune psychosis; HC, healthy controls; GABA, gamma-aminobutyric acid; Glx, glutamate + glutamine; tNAA, total Nacetyl aspartate; tCr, total creatine.

Group comparisons of metabolite ratios, including GABA/tNAA, GABA/tCr, Glx/tNAA, Glx/tCr, tNAA/tCr, and tCr/tNAA, were conducted for the selected ROIs. The pipeline ensured robust and region-specific quantification of neurochemical metabolites, facilitating comparisons across groups while adhering to rigorous quality control criteria.

Statistical analyses

Statistical analyses were conducted using R software version 3.6.0 (R Foundation for Statistical Computing, Vienna, Austria). Group comparisons of clinical data for categorical variables (e.g., sex) were performed using the chi-squared (χ²) test, while continuous variables (e.g., age) were analysed using the Welch two sample t-test or the non-parametric Mann–Whitney U test for independent samples. For the spectroscopy data, in a first step, the normality of metabolite ratios using the Shapiro–Wilk test in R (rstatix) was evaluated and the distribution of ratios with histograms was visually inspected. These analyses indicated that the metabolite ratios did not follow a normal distribution. To identify potential covariates (sex, age, body mass index, cigarette packs per year, intelligence quotient [IQ], CRLBs of the according metabolites in the ratio) influencing the metabolite ratios, the Boruta feature selection algorithm (Boruta package) was employed on the data. Among the tested potentially confounding variables, only the CRLBs emerged as a significant covariate. To adjust for the influence of CRLBs on metabolite ratios, a robust linear model (lmRob, a robust package) was applied, to predict and remove the influence of CRLBs for the corresponding metabolite ratios, re-centering the metabolite ratios based on the mean CRLBs of the two metabolites of each ratio across all metabolite ratios and valid ROIs. All subsequent statistical analyses were performed using these adjusted ratios. Given that the MRSI results were mostly non-normally distributed, robust statistics using Wilcoxon rank sum test were used for group comparisons. For correlation analyses, robust correlations were calculated using the pbcor function (WRS2 package; Wilcox, Reference Wilcox2012). For the test of attentional performance (TAP) results, again the Boruta model was applied to assess a potential influence of age and sex on the raw test measures. Age emerged as the only significant covariate affecting the TAP results. To adjust for age-related effects, a linear model (lm(TAP ∼ age)) and the predict function were used to re-center the test measures to the mean age, ensuring comparability across participants. Subsequent analyses were then performed using these age-adjusted TAP scores. Finally, all results were corrected for multiple comparisons using the Benjamini–Hochberg approach (Benjamini and Hochberg, Reference Benjamini and Hochberg1995), with a significance threshold of p < 0.05. A “trend” was considered to exist when findings were significantly different before (but not after) correction for multiple comparisons.