Focal CA3 hippocampal subfield atrophy following LGI1 VGKC-complex antibody limbic encephalitis

Autoantibodies linked with voltage-gated potassium channel (VGKC)-complex antibody-mediated limbic encephalitis (LE) bind to hippocampal subfields. Miller et al. report that chronic LGI1 antibody VGKC-complex LE is associated with focal CA3 atrophy on 7.0-Tesla neuroimaging and autobiographical episodic amnesia. The results point to antibody-mediated pathogenicity and CA3-mediated impairment of episodic memory.

The T1-weighted sequence was used as the basis for the VBM analyses and to derive total intracranial volumes to compensate for inter-individual variability in head size (Malone et al., 2015). The coronal and sagittal images in Figure 1 were obtained by loading the data from the three-dimensional T2-weighted fast spin-echo sequence and three-dimensional whole-brain T1weighted Phase Sensitive Inversion Recovery sequence into ITK-SNAP 3.2 (Yushkevich et al., 2006) (http://www.itksnap.org). As noted in the main text, ITK-SNAP 3.2 was also used to conduct hippocampal subfield volumetric morphometry.
The optimised 3D-FSE sequence was obtained in addition to the whole-brain T1weighted image because VBM automated segmentation of whole-brain images can fail to detect restricted hippocampal atrophy (Keller and Roberts, 2008;Wagner et al., 2015) and is not designed to report grey matter volume loss described at the level of individual hippocampal subfields. Furthermore, automated hippocampal subfield segmentation (1) needs to be validated when applied to clinical populations (de Flores et al., 2015) and/or on images at ≥3.0-Tesla signal strength to address particular issues in ultra-high field MRI such field inhomogeneity; (2) may collapse across subfield boundaries such as CA2 and CA3 that can be separated with manual quantitative morphometry; and, (3) may incorporate geometric as well as anatomical rules (Wisse et al., 2014a). Improvements, however, continue apace with these automated schemes and will enable future work to compare manual segmentation with automated hippocampal subfield segmentation (de Flores et al., 2015).

Whole-brain voxel-by-voxel morphometry
T1-weighted images from all participants in the LGI1 VGKC-complex-Ab LE group and age-matched control group were bias-corrected and the brain segmented into grey and white matter and CSF probability maps using the unified segmentation approach (Ashburner and Friston, 2005). Inter-subject iterative registration of the grey and white matter segments was performed using the Dartel toolbox (Ashburner, 2007). SPM12 failed to register one patient's scan and so the scan was removed from further VBM analyses. The resulting Diffeomorphic Anatomical Registration Through Exponentiated Lie algebra template and deformations were used to normalize grey and white matter probability maps to the stereotactic space defined by the Montreal Neurological Institute template. The normalization procedure involved modulating the grey matter probability maps by the Jacobian determinants of the deformation field and smoothing with an isotropic Gaussian kernel (8-mm, full-width at half-maximum). Voxels with grey matter values <0.2 (absolute threshold masking) were excluded to avoid edge effects between the tissue types. Total intracranial volumes were included into the model as a covariate of no interest.
The VBM analysis included, for example, voxels in areas adjacent to the hippocampus and other sites enriched in LGI1, such as those reported in a study examining the anatomical localisation of gene transcripts of the LGI1 family (Herranz-Perez et al., 2010). Our VBM analysis was also able to detect, if present, grey matter loss in subcortical regions such as the basal ganglia, insula cortex, hypothalamus or parahippocampal gyrus, which are other less common regions associated with high signal change on T2-weighted scans acquired in encephalitis patients positive for LGI1 antibodies.

Neuropsychology
Neuropsychological assessment was conducted in the LGI1 VGKC-complex-AB LE patient group on the following subtests: General intelligence: Wechsler Adult Intelligence Scale (WASI) -Similarities and Matrix Reasoning (Weschler, 1997); Verbal memory: Logical Memory I and II, Logical Memory I and II themes, and Word Lists I and II from Wechsler Memory Scale-III (WMS-III; (Wechsler, 1997) and Doors and People -People and People Recall Tests (Baddeley et al., 1994); Visual memory: Rey-Osterrieth complex figure immediate-recall (Osterrieth, 1944) and Doors and People -Shapes Test and Visual forgetting scores (Baddeley et al., 1994); Recognition memory: Words Lists II recognition (WMS-III; (Wechsler, 1997)), Recognition Memory Test for Words and Faces (Warrington, 1984) and Doors and People -Names and Doors Tests (Baddeley et al., 1994); Attention: All subtests of the Test of Everyday Attention (Robertson et al., 1994); Language: Graded Naming Test, Letter Fluency from the Verbal Fluency Test and Category Fluency from the Verbal Fluency Test (Delis-Kaplan Executive Function System, D-KEFS; (Delis et al., 2001)) and the Camel and Cactus Test (Bozeat et al., 2000); Executive function: Category Switching from the Verbal Fluency Test, Number-Letter Switching from the Trail Making Test and Colour-Word Interference Test (D-KEFS; (Delis et al., 2001)), and Digit Span (WMS-III) (Wechsler, 1997)); Visuoconstruction: Rey complex figure copy (Osterrieth, 1944); and, Visuomotor: Visual Scanning, Number Sequencing, Letter Sequencing, and Motor Speed (all from the Trail Making Test; D-KEFS; (Delis et al., 2001)). Scores on the standardised neuropsychological tests were first transformed into age-corrected standard values, where available, and then transformed into z-scores and averaged to derive composite scores corresponding to indices for the respective cognitive domains.
No significant deficits were evident outside of delayed verbal recall -comprised of

Logical Memory II, Logical Memory II themes and Word Lists II (WMS-III) and People Recall
Test -on the range of neuropsychological tests (see Table 2), and are broadly in line with the neuropsychological profile previously reported in six of the current 18 patients (McCormick et al., 2016(McCormick et al., , 2017. Furthermore, the severe deficits in episodic recollection evident on the autobiographical interview is not merely due to a loss of executive function or attention because the corresponding indices did not reveal deficits, which is in line with results that suggest the reconstitution of attentional-executive function following treatment (Frisch et al., 2013). The results also demonstrate the merit of testing episodic memory with both standardized tests and with measures that provide extended, objective quantitative assessments of autobiographical episodic memory, because nascent evidence indicates there is little neurocognitive or behavioural overlap (McDermott et al., 2009;Palombo et al., 2015). The results are also compatible with the proposal that the retrieval of autobiographical event knowledge is qualitatively different from other forms of episodic retrieval (Roediger and McDermott, 2013).

Postmorbid autobiographical episodic and semantic memory
Anterograde (postmorbid) first-person autobiographical episodic memory was assessed because it is widely regarded to be dependent on an intact hippocampus in both human (Squire and Alvarez, 1995;Nadel and Moscovitch, 1997;Bontempi et al., 1999;Frankland and Bontempi, 2005) and experimental (Lux et al., 2016) animal lesion studies.
Anterograde autobiographical memory was investigated under the standard retrieval conditions of autobiographical interview (Levine et al., 2002). The autobiographical interview has been used previously to study autobiographical episodic memory in both health and disease (Addis et al., 2007;Rosenbaum et al., 2008), and is designed to separate internal details (episodic recollection -the re-experiencing of all aspects of an event) from external details (non-episodic; e.g., semantic memory) related to autobiographical information (Levine et al., 2002). The basic method applied when scoring the autobiographical interview aligns with standardized tests that examine the retrieval of narrative-based details. In addition, in the autobiographical interview, the details are categorized according to whether they reference a unique episodic aspect of the event in order to distinguish these from generic or semantic content of the memory (Levine et al., 2002).
Data on the autobiographical interview were obtained from 16 of the 18 LGI1 VGKCcomplex-Ab LE patients and 16 participants in the age-matched 7.0-Tesla MRI control group.

Procedure
Instructions were first read aloud to explain that the task involved the recall of details about a specific recent event (within the last year) for each memory that was personally experienced at a particular time and place, and which occurred over minutes and hours but no longer than one day. The instructions also directed participants to generate as much detail as possible about the selected event in response to a series of retrieval cues. Participants were encouraged to free associate, if necessary, in order to facilitate the selection of an appropriate specific event. It was explained that the participant should feature in the autobiographical event and that the participant was free to choose any memory that was compatible with the instructions. Participants were then shown the written instructions. In line with the standardised administration, structured specific probes were provided after the general probes to target and facilitate the recollection of further contextual details from five discrete categories (event, time, place, perceptual, emotion/thought details). Participants were given a time limit of 5 min per event.

Scoring
All event descriptions were recorded on a digital recorder for subsequent transcription and scoring. The events were verified, where possible, by asking relatives and/or friends of the participants. Events were scored in accordance with the standardised procedure of the autobiographical interview, which involved the segmentation of each event into informational details (bits) that were classified as internal or external. Internal details reference the main episode being described, were situated within a specific spatiotemporal context, convey episodic re-experiencing, and thereby correspond to a quantitative measure of episodic memory. Internal details were assigned to one of the five response categories (event, time, place, perceptual, emotion/thought details). By contrast, external details refer to details that were not related to the main event, metacognitive statements or editorialising, repetitions, or semantic facts. Internal and external details were cumulatively summated to derive internal (episodic) and external (semantic) composite scores ( Figure 2B).
Two trained raters independently scored all of the events acquired from LGI1 VGKCcomplex-Ab LE patient group and age-matched control participants. Intra-class correlation coefficients were calculated to assess inter-rater reliability.

Hippocampal subfield segmentation: LGI1-antibody LE subgroup analysis
Re-analysis of the patient group hippocampal subfield segmentation data after removing the non-LGI1 patient (i.e., with n=17 LGI1-antibody positive patients) indicated that the results held both in terms of the main effects, interaction terms, and with the planned comparisons as compared to the results conducted with n=18. In particular, a three-way mixed-model ANOVA was conducted on the subfield volumes of the 17 LGI1-antibody patients, with two withinsubjects variables (subfield and side) and one between-subjects variable (group). Mauchly's test demonstrated that the assumption of sphericity had been violated (χ 2 (9) =81.16, p<0.0001), therefore, degrees of freedom were corrected using Greenhouse-Geisser estimates (ε=0.48).

Also in line with the results based on all 18
LGI1 VGKC-complex-Ab patients, the planned group comparisons revealed a significant reduction in total CA3 volume-again, collapsed across left and right CA3 due to the absence of a significant group interaction with side and subfield-of the LGI1-antibody LE patients (mean=375 (SD=84) relative to age-matched controls (mean=531 (SD=128)) (F (1,32) =17.58, p=0.0001; Cohen's d=1.44), whereas the group differences in subiculum, CA1, CA2, and dentate gyrus volumes were not statistically significant at the alpha criterion corrected for multiple comparisons (Cohen's d all <0.8).
Evidence of a restricted subfield lesion profile in both the LGI1-antibody LE subgroup and LGI1 VGKC-complex-Ab LE group is different from the generalised hippocampal atrophy seen in conditions such as Alzheimer's disease, where all subfields except CA2 are implicated (Wisse et al., 2014b), and may help to explain the differences in memory pathology. More broadly, evidence of a loss of CA3 integrity suggests that computational mechanisms ascribed to CA3 are also likely to be impaired in the patients (Rolls, 2013;Deuker et al., 2014), because disrupted binding of separate properties of an object, such as orientation and location, has been reported in patients with VGKC-complex antibodies (Pertzov et al., 2013).

Additional analyses of CA3 atrophy
The nature of the CA3 atrophy was additionally assessed in two ways. First, individual patient CA3 volumes were compared against the control mean CA3 volume. This demonstrated a mean CA3 proportion of 0.72 (SEM=0.05) on the left and 0.72 (SEM=0.02) on the right. A relative reduction compared to the control mean CA3 volume was seen in 17/18 participants on the left and 18/18 on the right. As in the case of the planned group comparison reported earlier, an independent-samples t-test indicated that there was no significant between left and right CA3 atrophy (t (17) =0.17, p=0.865). Second, we assessed whether the overall contribution of CA3 to the total hippocampal volume differed between left and right hippocampi. The proportions were calculated by dividing the left and right CA3 volumes by the total left and right hippocampal volumes, respectively. This demonstrated that for the patients the proportion of the total hippocampus volume constituted by CA3 was 0.143 (SEM=0.005) on the left and 0.144 (SEM=0.005) on the right, with no significant difference between these values (t (17) =-0.20, p=0.844).

CA3 atrophy and VGKC-complex antibody titre on presentation
High positive VGKC-complex-Ab titres (>400 pmol/l) are associated autoimmune limbic encephalitis that is responsive to immunotherapy (Paterson et al., 2014), and with cognitive impairment and seizures (Klein et al., 2013). By contrast, the relevance of lower antibody titres (≤400 pmol/l) for diagnosis or management appears equivocal (Paterson et al., 2014), and may not differ when compared with high positive titres in terms of memory outcome (Malter et al., 2014). Nonetheless, lower antibody titres may be relevant for clinical evaluation when they cooccur with cell surface antigens leucine-rich glioma inactivated 1, contactin associated protein 2 or both, which was found in 61% of patients when titres were ≤400 pmol/l (Klein et al., 2013).
As noted in the main text, the link between the antibody titre and CA3 pathology was tested by means of a linear regression with Huber correction and did not reveal evidence of a significant association (β1=-0.002, R 2 =0.07, t=1.07, p=0.30). VGKC-c 377 -1.24 Hippocampal CA3 subfield z-scores represent the combined (left and right hemisphere) total CA3 volume z-transformed relative to the age-matched control group. LGI1: leucine-rich glioma inactivated-1 antibody positive. VGKC-complex-Ab conc. on presentation: voltage-gated postassium channel-complex antibody concentration on presentation. VGKC-c: voltage-gated postassium channel-complex antibody positive.

Autobiographical Interview
Autobiographical interview data were acquired from 16 patients in the amnesic group and 16 participants in the age-matched control group. Inter-rater reliability for the full set of data scored by two raters -as assessed by conducting a two-way mixed model design for absolute agreement -indicated a high-level of consistency in scoring, as determined by a high intraclass correlation coefficient, 0.92.
The selective impairment of internal (episodic) details is consistent with evidence from our studies describing the effects of hippocampal damage -characterised at 3.0-Tesla -in six of the 18 patients reported here (McCormick et al., 2016(McCormick et al., , 2017 and with other reports of hippocampal and extra-hippocampal damage (Rosenbaum et al., 2008;Race et al., 2011), and is associated with a similar level of internal detail generated to that associated with left temporal lobe epilepsy (mean=27.6 (SD=24.4)) (Addis et al., 2007). Damage to the MTL that extends beyond the hippocampus or neocortical involvement impairs non-episodic retrieval, whereas more focal lesions preferentially impair episodic detail retrieval -a finding that appears to be consistent across 147 cases of hippocampal amnesia (Spiers et al., 2001). Evidence of no significant impairment on the Graded Naming Test (n=17, ave. z-score=0.70, s.e.m.=0.25, tscore (16) =3.07, p=0.003) indicates that the episodic autobiographical deficit was not related to reduced verbal output/fluency.