Regional abnormalities in the hippocampus and verbal memory impairment in craniofacial dystonia Authors: Gang Liu, Huiming Liu, Linchang Zhong, Yuhan Luo, Zhengkun Yang, Jiana Zhang, Xiuye He, Zilin Ou, Weixi Zhang, Kangqiang Peng, Jinping Xu, Zhicong Yan, Yue Zhang Published: 2025 ABSTRACT BackgroundPatients with craniofacial dystonia (CFD) often present with verbal memory deficits, but their neuroanatomical basis is not yet clear. This study aims to determine whether verbal memory deficits in CFD are associated with structural atrophy of specific hippocampal subfields, and to isolate dystonia-specific pathological changes through comparison with patients with hemifacial spasm (HFS).MethodsWe recruited 50 patients with CFD, 48 patients with HFS, and 50 healthy controls (HCs). Verbal memory and global cognitive function were assessed using the Rey Auditory Verbal Learning Test (RAVLT) and the Mini-Mental State Examination (MMSE), respectively. Volumes of hippocampal subfields were quantified from high-resolution T1-weighted magnetic resonance imaging (MRI) using FreeSurfer. Group comparisons were performed after controlling for relevant covariates.ResultsWhile global cognition (MMSE) scores did not differ significantly among groups, patients with CFD demonstrated significant verbal memory deficits. Compared with HCs, they performed worse across immediate, short-, and long-delay recall trials, with medium-to-large effect sizes (all P-FDR ≤ 0.002; Cohen’s d = −0.70 to −0.82). Similar deficits of medium effect sizes were observed when compared with patients with HFS (all P-FDR ≤ 0.027; Cohen’s d = −0.49 to −0.52). Crucially, patients with HFS were unimpaired relative to HCs, establishing this memory deficit as specific to the dystonia pathophysiology. The imaging analysis revealed that patients with CFD were associated with significant atrophy in the left granule cell layer of the dentate gyrus (GC-DG) (P-FDR = 0.014; Cohen’s d = −0.58) and CA4 (P-FDR = 0.010; Cohen’s d = −0.60) compared with HCs, and with significant atrophy in the right GC-DG (P-FDR = 0.032; Cohen’s d = −0.52) and CA4 (P-FDR = 0.041; Cohen’s d = −0.50) compared with HCs. However, the magnitude of the atrophy showed no significant correlation with verbal memory scores, disease duration, or motor severity, revealing a critical structure-function dissociation.ConclusionOur findings reveal a structure-function dissociation in CFD. We propose its verbal memory deficits, despite hippocampal atrophy, may stem from broader network dysfunction or microstructural pathology not seen on conventional MRI. This challenges models assuming a direct link between macrostructural atrophy and cognitive symptoms. Introduction Idiopathic adult-onset dystonia (IAOD), a common movement disorder, is characterized by motor features and non-motor manifestations. Among the non-motor manifestations, sensory and neuropsychiatric abnormalities are more commonly reported. However, deficits in cognitive function, including memory, aspects of social cognition, executive function, and changes in information processing speed, are also widely reported in patients with IAOD (Czekóová et al., 2017; Yang et al., 2017; Burke et al., 2020; Monaghan et al., 2021). Within the clinical spectrum of dystonia, increasing evidence has highlighted the link between cognitive impairment and patient quality of life (Girach et al., 2019; Ndukwe et al., 2020). However, the mechanisms underlying cognitive dysfunction in IAOD remain largely unknown. Deficits in cognitive function in patients with dystonia were traditionally thought to be linked to basal ganglia dysfunction, as supported by evidence of impaired cognitive function in neurodegenerative disorders such as Parkinson’s, Huntington’s, or Wilson’s disease, as well as brain injury with damage to the basal ganglia (Tremblay et al., 2015; Galantucci et al., 2017; Hu et al., 2021; Puig-Davi et al., 2021) or related circuit dysfunction, including cortico-basal ganglia networks and cerebello-thalamo-cortical and cerebello-thalamo-basal ganglia circuits (Defazio et al., 2024). Indeed, as recent evidence underscores, the disruptions in the dentato-rubro-olivary pathway, which is essential for motor coordination, may represent a component of circuit dysfunction involving the cortico-basal ganglia networks and cerebello-thalamo-cortical and cerebello-thalamo-basal ganglia circuits, all of which contribute to the pathophysiology of movement disorders (Ogut et al., 2023). The basal ganglia are involved in not only prominent sensorimotor functions but also in cognitive operations and emotional-motivational processes (Rinnerthaler et al., 2006; Gan et al., 2022). Nevertheless, the neuroanatomical substrates differ among different types of IAOD (Neychev et al., 2011). Blepharospasm (BSP) may be related to structural abnormalities in motor cortices such as the supplementary motor area (Xu et al., 2023). In addition, cognitive deficits are reportedly caused by the distracting effects of motor symptoms of dystonia (Redondo-Vergé, 2001). However, a literature review reported that deep brain stimulation did not improve concurrent mild cognitive impairment (MCI) despite improved motor symptoms in patients with dystonia (Owen et al., 2015; Cernera et al., 2019). Moreover, recent studies reported a lack of association between the severity of dystonia and the severity of cognitive deficits in patients with isolated dystonia (Yang et al., 2016; Foley et al., 2017; Monaghan et al., 2021; Rafee et al., 2023; Defazio et al., 2024), providing further evidence that cognitive dysfunction is not a mere consequence of abnormal movements. Interpreting cognitive deficits in dystonia requires rigorously addressing key confounders. Pharmacological agents, particularly anticholinergics, are well-documented to have direct effects on cognitive function (Taylor et al., 1991). More significantly, the high prevalence of depression and anxiety in this population presents a major methodological challenge. Because these mood disorders are themselves associated with memory and executive dysfunctions, this creates a critical ambiguity: are the observed cognitive deficits a core feature of dystonia, or simply a byproduct of co-occurring mood symptoms (Lange et al., 2016)? While some evidence suggests these deficits persist after controlling for such factors, disentangling these intertwined variables demands a robust study design (Romano et al., 2014; Defazio et al., 2024). Therefore, a primary objective of our study was to methodologically isolate the neuroanatomical correlates of verbal memory impairments intrinsic to dystonia. To achieve this, we prospectively mitigated these confounders through stringent exclusion criteria for medications with potential cognitive side effects, and subsequently controlled for subclinical mood symptoms in our statistical models. This design permits a more direct, unconfounded examination of the link between hippocampal integrity and memory function. Studies in recent decades have highlighted the crucial role of the medial temporal lobe (MTL), particularly the hippocampus, in memory (Cutsuridis and Wennekers, 2009; den Heijer et al., 2012; Huhn et al., 2018). Moreover, the hippocampus contains distinct subfields that serve different types of memory functions. Specifically, verbal memory is mainly associated with the anterior subfields of the left hippocampus, while visual memory is associated with many posterior subfields of the bilateral hippocampus (Travis et al., 2014). Therefore, the hippocampus should not be considered as a single homogeneous structure, as doing so may disregard potentially useful information about distinct hippocampal subfields. However, to our knowledge, no studies have explored the structural alterations in distinct hippocampal subfields and their relationship with memory alterations in patients with dystonia. A statistical atlas of the hippocampal formation was recently constructed using ultra-high-resolution ex vivo magnetic resonance imaging (MRI) combined with in vivo data (Iglesias et al., 2015). This atlas is integrated into FreeSurfer 61. This tool enables the accurate measurement of each hippocampal subfield. This study aimed to elucidate the neuroanatomical basis of verbal memory deficits in dystonia by examining hippocampal subfield integrity. Our study compared Rey Auditory Verbal Learning Test (RAVLT) scores and hippocampal subfield volumes across the three groups, and crucially, examined the links between hippocampal atrophy, memory deficits, and key clinical variables such as disease duration and severity. The present study focuses specifically on patients with craniofacial dystonia (CFD), a subtype that offers a unique clinical model for two key reasons. First, memory impairments are consistently reported as cognitive deficits in this population (Alemán et al., 2009; Yang et al., 2016; Maggi et al., 2019). Second, the clinical presentation of CFD permits a robust comparative design. By including a carefully matched control group of patients with hemifacial spasm (HFS), which is a condition with similar facial hyperkinesia but of peripheral origin, we can methodologically disentangle central dystonic pathophysiology from the confounding effects of chronic facial muscle overactivity (Dias et al., 2009). Materials and methods Participants The patients were recruited from an outpatient clinic for movement disorders at the First Affiliated Hospital of Sun Yat-sen University. The diagnoses of HFS and CFD were made by two senior neurologists (G Liu and WX Zhang) using standard criteria (Lefaucheur et al., 2018; Defazio et al., 2019, 2021; Neuro-ophthalmology Group of Ophthalmology Branch of Chinese Medical Association et al., 2025). Specifically, patients with CFD were further classified into subtypes based on the distribution of dystonic movements: (1) Blepharospasm (BSP) is primarily manifested as stereotyped, bilateral, synchronous spasms of the orbicularis oculi muscle (leading to eyelid narrowing/closure) and an abnormally elevated blink frequency (>16 times/min), in addition to sensory symptoms such as ocular dryness, foreign body sensation, and photophobia. The characteristic sensory tricks phenomenon is also observed. Furthermore, the symptoms disappear during sleep. (2) Blepharospasm-oromandibular dystonia (BOD), building upon the clinical manifestations of BSP, involves involuntary spasmodic contractions of multiple muscle groups including the orbicularis oris, masseter, and lingual muscles. It presents as perioral twitching and abnormal mandibular movements, which may impair functions such as mastication, swallowing, and articulation. Additionally, this subtype demonstrates a tendency to spread to adjacent muscle groups. In addition, patients were excluded if they: (i) received botulinum toxin (BoNT) injections within 3 months before the MRI scan; (ii) reported evidence of stroke, traumatic brain injury, Parkinson’s disease, Alzheimer’s disease, epilepsy; (iii) had a family history of movement disorders or history of antipsychotic medication use before dystonia onset; (iv) were receiving benzodiazepines, anticholinergics, or other medications with potential cognitive side effects; (v) had any contraindications to cerebral MRI; (vi) had evidence of major psychiatric disorders. Exclusion of major psychiatric disorders was determined through a multi-component, clinician-administered assessment conducted by two senior neurologists, consistent with the principles of an integrated neuropsychiatric evaluation (Trapp et al., 2022; Pallanti, 2024). The assessment included: (a) a comprehensive neuropsychiatric history, incorporating a thorough review of medical records with an emphasis on the patient’s personal experience of events and careful screening for underlying or comorbid conditions; (b) a systematic clinical interview to screen for core symptom clusters of major mood, anxiety, and psychotic disorders; (c) a formal mental status examination assessing appearance, behavior, affect, thought process, and content; and (d) the administration of the Hamilton Depression Rating Scale (HAMD)-17 (Hamilton, 1960), the Hamilton Anxiety Rating Scale (HAMA)-14 (Maier et al., 1988), and the Mini-Mental State Examination (MMSE) (Folstein et al., 1975) to quantify symptom severity. A participant was excluded if this integrated clinical judgment, supported by direct observation, scale scores, and historical data, indicated a major psychiatric disorder that could significantly confound the study’s outcomes. Healthy controls (HCs) were recruited using the same exclusion criteria. Clinical assessment Before MRI, the participants’ demographic and clinical characteristics, including age, sex, education, disease duration, and BoNT injection duration, were collected during face-to-face interviews. The Burke-Fahn-Marsden Dystonia Rating Scale (BFMDRS) (Burke et al., 1985) and Cohen’s Scale (Cohen et al., 1986) were used to assess the severity of CFD and HFS, respectively. HAMA and HAMD were used to assess anxiety and depression symptoms, respectively. Cognitive assessments Cognitive function was evaluated using the MMSE for general cognition and the Chinese version of the Rey Auditory Verbal Learning Test (RAVLT) for verbal memory (Tsai et al., 2015). All assessments were administered in Mandarin Chinese by a single trained neurologist (Z.L. Ou) in a quiet, distraction-free environment. Standardized instructions were followed according to the respective test manuals, and all delay intervals for the RAVLT were strictly timed. From the RAVLT, we derived several key verbal memory metrics: immediate recall (total score across trials 1-3), short-delay recall, long-delay recall, cued recall, and recognition. Additionally, we calculated the short-delay forgetting rate (SDFR) and long-delay forgetting rate (LDFR), based on the formulas: SDFR = (short-delay recall score - trial 3 score)/trial 3 score, and LDFR = (long-delay recall score - trial 3 score)/trial 3 score. The RAVLT is a well-established measure with good reliability and validity for assessing verbal memory (Pliskin et al., 2021). MRI data acquisition Three-dimensional T1-weighted data were collected using a 3T MRI scanner (Tim Trio; Siemens, Erlangen, Germany) with a magnetization-prepared rapid-acquisition gradient-echo pulse sequence. The main parameters were as follows: repetition time, 2,530 ms; echo time, 4.45 ms; inversion time, 1,100 ms; flip angle, 7°; matrix dimensions, 256 × 256; voxel size, 1 × 1 × 1 mm3; and 192 slices. Data pre-processing All T1 images were processed using the standard segmentation pipeline available in FreeSurfer 6.0. The technical details have been described elsewhere (Fischl and Dale, 2000). We used the command for volumetric segmentation (“recon-all”). The main steps included motion correction, skull stripping, intensity normalization, automated Talairach transformation, gray or white matter tessellation, and topology correction (Fischl et al., 2001; Ségonne et al., 2007). The subcortical structures were segmented using a non-linear warping atlas (Fischl et al., 2002). The total hippocampal volumes were obtained and estimated total intracranial volume (eTIV) was calculated. Subsequently, the hippocampal subfields were segmented using a Bayesian inference approach and a novel atlas algorithm for hippocampal formations built primarily on ultra-high-resolution ex vivo MRI data from autopsy brains (Iglesias et al., 2015). The segmentation accurately delineated various hippocampus subfields in each hemisphere, including the parasubiculum, presubiculum, subiculum, cornu ammonis (CA) 1, CA2-3, CA4, granule cell layer of dentate gyrus (GC-DG), hippocampus-amygdala transition area (HATA), fimbriae, molecular layer hippocampus (HP), hippocampal fissure, and hippocampal tail (Figure 1). Finally, Freeview2 was used to visualize the hippocampal subfields. We selected FreeSurfer 6.0 as it was the stable, validated, and widely adopted standard at the study’s outset. First, it introduced a novel module for hippocampal subfield segmentation based on an ultra-high-resolution ex vivo atlas (Iglesias et al., 2015). Second, its excellent test-retest reliability has been well-documented in large-scale studies (Haddad et al., 2023). Third, given that different software versions can yield systematic volumetric differences affecting statistical outcomes (Sämann et al., 2022; Haddad et al., 2023), we used this single, consistent pipeline for all processing. This approach ensures both the internal consistency of our dataset and its comparability with the substantial body of literature using this well-characterized tool. Hippocampal subfield segmentation in a patient with dystonia. (A) A coronal view of the left hippocampus. (B) A sagittal view showing the longitudinal extent of the left hippocampus. (C) An axial view of the left hippocampus. White arrows indicate the segmented hippocampal formation. Anatomical directions are marked on each panel (L, Left; R, Right; A, Anterior; P, Posterior; S, Superior; I, Inferior). The color legend defines the different hippocampal subfields. CA, cornu ammonis; GC-DG, granule cell layer of dentate gyrus; HATA, hippocampus-amygdala transition area; HP, hippocampus. Scale bar = 1 cm. Volume calculations in the hippocampal subfields The volumes of the bilateral hippocampal subfields were calculated and intergroup comparisons were performed using analysis of covariance (ANCOVA), with age, sex, education, eTIV, HAMA, and HAMD scores included as covariates. The statistical method was specifically chosen to isolate the effect of group by statistically controlling for the influence of these potential confounders. By accounting for variance associated with the covariates, ANCOVA increases statistical power and provides a more precise and less biased estimate of the adjusted group means (Porter and Raudenbush, 1987; Egbewale et al., 2014). Moreover, these comparisons were all corrected for multiple testing using the False Discovery Rate (FDR) method with the Benjamini-Hochberg procedure. Statistical significance was set at P-FDR < 0.05. Correlation analyses To explore potential confounding effects of clinical variables within the CFD group, we conducted partial correlation analyses to examine the relationships between (1) RAVLT memory scores and hippocampal subfield volumes, (2) RAVLT memory scores and clinical variables (duration and disease severity), and (3) hippocampal subfield volumes and clinical variables (duration and disease severity). All partial correlation analyses controlled for age, sex, HAMA, and HAMD scores as covariates. Given the exploratory nature of these analyses, we applied FDR correction for multiple comparisons. Statistical significance was set at P-FDR < 0.05. Statistical analyses Differences in demographic variables among the three groups were assessed based on their distribution. For normally distributed data, one-way analysis of variance (ANOVA) was used, while the non-parametric Kruskal-Wallis H-test was employed for variables such as age and educational level, which did not meet the assumption of normality. Sex comparisons were performed using chi-square (χ2) tests. ANCOVA was used to assess intergroup differences in RAVLT scores. Following a significant main effect in the omnibus ANCOVA, planned post hoc pairwise comparisons were conducted using the FDR method. Statistical significance was set at P-FDR < 0.05. These analyses were performed using IBM SPSS Statistics for Windows, version 26.0 (IBM Corp., Armonk, NY, United States). Results Participant characteristics and behavioral assessment This study enrolled 50 patients with CFD (40 with BSP and 10 with BOD), 48 patients with HFS, and 50 HCs. The demographic information, clinical characteristics, and behavioral test scores of all participants are detailed in Table 1 and the details of ANCOVA results for RAVLT scores among three groups are presented in Table 2. Age, sex, education, or MMSE scores did not differ significantly among the three groups. Patients with CFD displayed significantly lower performance in measures of immediate recall, short-delay recall, and long-delay recall compared with the patients with HFS (all P-FDR ≤ 0.027; Cohen’s d = −0.49 to −0.52), and HCs (all P-FDR ≤ 0.002; Cohen’s d = −0.70 to −0.82). All reported P-FDR values were corrected for multiple comparisons using the FDR method. Moreover, patients with HFS did not show evidence of deterioration in these measures relative to HCs. In addition, cued recall, SDFR, LDFR, and recognition did not differ significantly among the groups. TABLE 1Participant characteristics and behavioral assessment.CharacteristicsCFD (n = 50)HFS (n = 48)HCs (n = 50) P CFD < HFSCFD < HCsHFS < HCsSubjects (BSP/BOD)40/10––––––Median age, y (range)54 (30–68)52 (30–69)52 (29–65)0.698b–––Median education, y (range)10.5 (0–16)12 (0–16)12 (0–16)0.399b–––Sex (female/male)29/2129/1929/210.906a–––eTIV, mm3 (range)1489590.1 (921606.1–1968076.7)1442859.7 (1082096.8–1960182.6)1499080.0 (1088275.0–1819172.5)0.872c–––Median MMSE scores (range)27 (24–30)27 (24–30)28 (24–30)0.418b–––Median disease duration, y (range)2 (0.170–15)2 (0.080–20)–0.195d–––Median Cohen scores (range)–3 (1–4)–-–––Median BFMDRS scores (range)4.5 (3–10.5)––––––BoNT injections (yes/no)31/1927/21–0.07a–––Median BoNT injection duration, y (range)0 (0-15)0 (0-19)–0.958d–––Median HAMA scores (range)4.5 (0–14)3 (0–9)2 (0–10) 0.002b 0.031 0.002 0.105Median HAMD scores (range)4 (0–12)2 (0–10)1 (0–13) 0.001b 0.005 0.001 0.444 RAVLTe Median immediate recall (range)12 (4–26)14.5 (9–25)17 (8–26) <0.001 0.022 <0.001 0.059Median short-delay recall (range)4 (0–9)5 (2–10)5 (0–11) 0.002 0.018 0.002 0.278Median long-delay recall (range)3 (0–8)4 (0–10)5 (0–10) 0.001 0.027 <0.001 0.125Median SDFR (range)−0.293 (−1.000 to 1.000)−0.236 (−0.667 to 0.600)−0.211 (−1.000 to 0.250)0.580–––Median LDFR (range)−0.400 (−1.000 to 1.000)−0.317 (−1.000 to 0.500)−0.236 (−1.000 to 0.125)0.880–––Median cued recall (range)3 (0–10)4 (0–8)5 (0–10)0.287–––Median recognition (range)20 (15–24)20 (15–24)21 (11–24)0.963–––aRepresents χ2 test.bRepresents Kruskal-Wallis H-test between three groups.cRepresents one-way analysis of variance, and subsequent post hoc analyses between any two groups.dRepresents Mann-Whitney U test between two groups.eRepresents ANCOVA between three groups on each RAVLT sub-score with age, sex, education, HAMA, and HAMD scores as covariates and the results were corrected for multiple comparisons using False Discovery Rate. Statistically significant differences (P < 0.05) are highlighted in bold. ANCOVA, analysis of covariance; BFMDRS, Burke-Fahn-Marsden Dystonia Rating Scale; BOD, blepharospasm oromandibular dystonia; BoNT, Botulinum toxin; BSP, blepharospasm; CFD, craniofacial dystonia; eTIV, estimated total intracranial volume; HAMA, the Hamilton Rating Scale for Anxiety; HAMD, the Hamilton Rating Scale for Depression; HCs, healthy controls; HFS, hemifacial spasm; LDFR, long-delay forgetting rate; MMSE, Mini-Mental State Examination; RAVLT, Rey Auditory Verbal Learning Test; SDFR, short-delay forgetting rate; y, years. Participant characteristics and behavioral assessment. aRepresents χ2 test. bRepresents Kruskal-Wallis H-test between three groups. cRepresents one-way analysis of variance, and subsequent post hoc analyses between any two groups. dRepresents Mann-Whitney U test between two groups. eRepresents ANCOVA between three groups on each RAVLT sub-score with age, sex, education, HAMA, and HAMD scores as covariates and the results were corrected for multiple comparisons using False Discovery Rate. Statistically significant differences (P < 0.05) are highlighted in bold. ANCOVA, analysis of covariance; BFMDRS, Burke-Fahn-Marsden Dystonia Rating Scale; BOD, blepharospasm oromandibular dystonia; BoNT, Botulinum toxin; BSP, blepharospasm; CFD, craniofacial dystonia; eTIV, estimated total intracranial volume; HAMA, the Hamilton Rating Scale for Anxiety; HAMD, the Hamilton Rating Scale for Depression; HCs, healthy controls; HFS, hemifacial spasm; LDFR, long-delay forgetting rate; MMSE, Mini-Mental State Examination; RAVLT, Rey Auditory Verbal Learning Test; SDFR, short-delay forgetting rate; y, years. TABLE 2ANCOVA results for RAVLT scores.RAVLT measures F P ηP2 CFD vs. HFS MD (95% CI) P-FDR, Cohen’s dCFD vs. HCs MD (95% CI) P-FDR, Cohen’s dHFS vs. HCs MD (95% CI) P-FDR, Cohen’s dImmediate recall5.33<0.0010.071−2.23 (−4.02, −0.45) 0.022, d = −0.50−3.94 (−5.88, −2.00) <0.001, d = −0.81−1.70 (−3.48, 0.07) 0.059, d = −0.39Short-delay recall3.170.0020.043−1.07 (−1.89, −0.24) 0.018, d = −0.52−1.58 (−2.48, −0.68) 0.002, d = -0.70−0.51 (−1.45, 0.42) 0.278, d = −0.22Long-delay recall4.540.0010.061−1.04 (−1.90, −0.18) 0.027, d = −0.49−1.80 (−2.67, −0.93) <0.001, d = −0.82−0.76 (−1.73, 0.21) 0.125, d = −0.31SDFR0.550.580.008–––LDFR0.130.880.002–––Cued recall1.260.2870.018–––Recognition0.040.9630.001–––The results shown are from a series of ANCOVA used to compare the three groups on each RAVLT sub-score with age, sex, education, HAMA, and HAMD scores as covariates. The results were corrected for multiple comparisons using FDR. ηP2, partial eta-squared. Statistically significant differences (P < 0.05) are highlighted in bold. ANCOVA, analysis of covariance; CFD, craniofacial dystonia; CI, confidence interval; FDR, False Discovery Rate; HAMA, the Hamilton Rating Scale for Anxiety; HAMD, the Hamilton Rating Scale for Depression; HCs, healthy controls; HFS, hemifacial spasm; LDFR, long-delay forgetting rate; MD, mean difference; RAVLT, Rey Auditory Verbal Learning Test; SDFR, short-delay forgetting rate. ANCOVA results for RAVLT scores. The results shown are from a series of ANCOVA used to compare the three groups on each RAVLT sub-score with age, sex, education, HAMA, and HAMD scores as covariates. The results were corrected for multiple comparisons using FDR. ηP2, partial eta-squared. Statistically significant differences (P < 0.05) are highlighted in bold. ANCOVA, analysis of covariance; CFD, craniofacial dystonia; CI, confidence interval; FDR, False Discovery Rate; HAMA, the Hamilton Rating Scale for Anxiety; HAMD, the Hamilton Rating Scale for Depression; HCs, healthy controls; HFS, hemifacial spasm; LDFR, long-delay forgetting rate; MD, mean difference; RAVLT, Rey Auditory Verbal Learning Test; SDFR, short-delay forgetting rate. Hippocampal segmentation After controlling for sex, age, education, eTIV, HAMA, and HAMD scores, the imaging analysis revealed that CFD was associated with significant atrophy in the left GC-DG (P-FDR = 0.014; adjusted mean difference (MD) = −14.92; 95% confidence interval (CI) = −25.12 to −4.72; Cohen’s d = −0.58; reduction 5.13%) and CA4 (P-FDR = 0.010; adjusted MD = −12.75; 95% CI = −21.15 to −4.35; Cohen’s d = −0.60; reduction 5.11%) compared with HCs, with significant atrophy in the left GC-DG (P-FDR = 0.043; adjusted MD = −12.24; Cohen’s d = −0.45; reduction 4.23%), and with significant atrophy in the right GC-DG (P-FDR = 0.032; adjusted MD = −15.21; 95% CI = −26.78 to −3.64; Cohen’s d = −0.50; reduction 5.05%) and CA4 (P-FDR = 0.041; adjusted MD = −12.49; 95% CI = −22.35 to −2.63; Cohen’s d = −0.50; reduction 4.87%) compared with HCs (Tables 3, 4). All reported P-FDR values were corrected for multiple comparisons using the FDR method. Forest plots visualizing the adjusted MD, 95% CI for all key comparisons are presented in Figure 2 (CFD vs. HFS) and Figure 3 (CFD vs. HCs). TABLE 3Volumetric data of hippocampal subfields.Hippocampal subfieldsCFD (mean ± SD)HFS (mean ± SD)HCs (mean ± SD) P CFD < HFSCFD < HCsHFS < HCs Left Parasubiculum61.328 ± 12.44959.154 ± 11.01159.201 ± 10.7210.394–––Presubiculum320.889 ± 35.300316.965 ± 37.330316.385 ± 34.5800.724–––Subiculum442.200 ± 48.928439.333 ± 46.112447.750 ± 46.6100.697–––CA1598.621 ± 64.763603.535 ± 63.242606.312 ± 63.1950.525–––CA2-3184.786 ± 21.143188.644 ± 22.735189.608 ± 26.7040.228–––CA4236.582 ± 20.297246.227 ± 24.791249.329 ± 21.992 0.001 0.056 0.010 0.513GC-DG275.724 ± 25.240287.964 ± 29.179290.641 ± 26.158 0.001 0.043 0.014 0.633HATA56.249 ± 7.33355.414 ± 6.00956.549 ± 6.6720.809–––Fimbria80.850 ± 18.43485.271 ± 21.48186.106 ± 16.5600.343–––Molecular_layer_HP548.375 ± 53.394554.685 ± 53.148558.287 ± 45.5940.290–––Hippocampal-fissure154.936 ± 28.204146.540 ± 28.923151.036 ± 22.9120.459–––Hippocampal_tail569.500 ± 69.409566.490 ± 74.745569.200 ± 77.2260.948––– Right Parasubiculum56.811 ± 12.31552.594 ± 9.58954.061 ± 10.5800.098–––Presubiculum306.591 ± 37.610306.084 ± 41.495304.382 ± 33.5020.778–––Subiculum455.142 ± 50.922457.239 ± 52.207454.169 ± 42.2800.716–––CA1631.283 ± 74.170640.200 ± 75.034648.471 ± 65.3780.306–––CA2-3195.765 ± 24.668201.140 ± 26.051205.422 ± 29.1100.065–––CA4244.250 ± 23.787250.801 ± 25.841256.740 ± 25.859 0.008 0.258 0.041 0.258GC-DG285.670 ± 28.283294.238 ± 29.400300.880 ± 30.017 0.005 0.217 0.032 0.271HATA55.411 ± 6.80656.906 ± 6.41956.744 ± 6.6760.405–––Fimbria76.917 ± 14.99075.652 ± 17.06577.547 ± 16.7590.725–––Molecular_layer_HP568.537 ± 60.176576.942 ± 61.311585.210 ± 51.8600.211–––Hippocampal-fissure160.431 ± 29.158158.803 ± 35.048155.395 ± 23.9760.543–––Hippocampal_tail591.478 ± 80.276597.056 ± 83.348598.874 ± 89.2430.869–––Differences of hippocampal subfields volume among three groups were analyzed using analysis of covariance with age, sex, education, eTIV, HAMA, and HAMD scores as covariates. The results were corrected using False Discovery Rate method. Statistically significant differences (P < 0.05) are highlighted in bold. CA, cornu ammonis; CFD, craniofacial dystonia; eTIV, estimated total intracranial volume; GC-DG, granule cell layer of dentate gyrus; HAMA, Hamilton Rating Scale for Anxiety; HAMD, Hamilton Rating Scale for Depression; HATA, hippocampus-amygdala transition area; HCs, healthy controls; HFS, hemifacial spasm; HP, hippocampus. Volumetric data of hippocampal subfields. Differences of hippocampal subfields volume among three groups were analyzed using analysis of covariance with age, sex, education, eTIV, HAMA, and HAMD scores as covariates. The results were corrected using False Discovery Rate method. Statistically significant differences (P < 0.05) are highlighted in bold. CA, cornu ammonis; CFD, craniofacial dystonia; eTIV, estimated total intracranial volume; GC-DG, granule cell layer of dentate gyrus; HAMA, Hamilton Rating Scale for Anxiety; HAMD, Hamilton Rating Scale for Depression; HATA, hippocampus-amygdala transition area; HCs, healthy controls; HFS, hemifacial spasm; HP, hippocampus. TABLE 4ANCOVA results for volume of the hippocampal subregions.Hippocampal subfields F P ηP2 CFD vs. HFS MD (95%CI) P-FDR, Cohen’s dCFD vs. HCs MD (95%CI) P-FDR, Cohen’s dHFS vs. HCs MD (95%CI) P-FDR, Cohen’s d Left Parasubiculum0.940.3940.013–––Presubiculum0.320.7240.005–––Subiculum0.360.6970.005–––CA10.650.5250.009–––CA2-31.490.2280.021–––CA47.01 0.001 0.092−9.64 (−18.71, −0.58) 0.056, d = −0.43−12.75 (−21.15, −4.35) 0.010, d = −0.60−3.10 (−12.49, 6.29) 0.513, d = −0.13GC-DG6.87 0.001 0.090−12.24 (−23.17, −1.31) 0.043, d = −0.45−14.92 (−25.12, −4.72) 0.014, d = −0.58−2.68 (13.78, 8.43) 0.633, d = −0.10HATA0.510.6010.007–––Fimbria1.080.3430.015–––Molecular_layer_HP1.250.2900.018–––Hippocampal-fissure0.780.4590.011–––Hippocampal_tail0.050.9470.001–––RightParasubiculum2.360.0980.033–––Presubiculum0.250.7780.004–––Subiculum0.340.7160.005–––CA11.190.3060.017–––CA2-32.790.0650.039–––CA44.98 0.008 0.067−6.55 (−16.50, 3.40) 0.258, d = −0.26−12.49 (−22.35, −2.63) 0.041, d = −0.50−5.94 (−16.31, 4.43) 0.258, d = −0.23GC-DG5.50 0.005 0.073−8.57 (−20.13, 3.00) 0.217, d = −0.30−15.21 (−26.78, −3.64) 0.032, d = −0.52−6.64 (−18.56, 5.28) 0.271, d = −0.22HATA0.910.4050.013–––Fimbria0.320.7250.005–––Molecular_layer_HP1.570.2120.022–––Hippocampal-fissure0.610.5430.009–––Hippocampal_tail0.140.8690.002–––The results shown are from a series of ANCOVA used to compare the three groups on volume of the hippocampal subregions with age, sex, education, eTIV, HAMA, and HAMD scores as covariates. The results were corrected for multiple comparisons using FDR. ηP2, partial eta-squared. Statistically significant differences (P < 0.05) are highlighted in bold. ANCOVA, analysis of covariance; CA, cornu ammonis; CFD, craniofacial dystonia; CI, confidence interval; eTIV, estimate total intracranial volume; FDR, False Discovery Rate; GC-DG, granule cell layer of dentate gyrus; HAMA, the Hamilton Rating Scale for Anxiety; HAMD, the Hamilton Rating Scale for Depression; HATA, hippocampus-amygdala transition area; HCs, healthy controls; HFS, hemifacial spasm; HP, hippocampus; MD, mean difference. ANCOVA results for volume of the hippocampal subregions. The results shown are from a series of ANCOVA used to compare the three groups on volume of the hippocampal subregions with age, sex, education, eTIV, HAMA, and HAMD scores as covariates. The results were corrected for multiple comparisons using FDR. ηP2, partial eta-squared. Statistically significant differences (P < 0.05) are highlighted in bold. ANCOVA, analysis of covariance; CA, cornu ammonis; CFD, craniofacial dystonia; CI, confidence interval; eTIV, estimate total intracranial volume; FDR, False Discovery Rate; GC-DG, granule cell layer of dentate gyrus; HAMA, the Hamilton Rating Scale for Anxiety; HAMD, the Hamilton Rating Scale for Depression; HATA, hippocampus-amygdala transition area; HCs, healthy controls; HFS, hemifacial spasm; HP, hippocampus; MD, mean difference. ANCOVA results of verbal memory and hippocampal subfields (CFD vs. HFS). The forest plot displays adjusted mean difference and 95% confidence interval. P-values are FDR-corrected. Thresholds for partial eta-squared (ηp2) effect sizes are shown. Models controlled for age, sex, education, HAMA, and HAMD (eTIV for subfield volumes). ANCOVA, analysis of covariance; CA, cornu ammonis; CFD, craniofacial dystonia; eTIV, estimated total intracranial volume; GC-DG, granule cell layer of the dentate gyrus; HAMA, the Hamilton Rating Scale for Anxiety; HAMD, the Hamilton Rating Scale for Depression; HATA, hippocampus-amygdala transition area; HFS, hemifacial spasm; LDFR, long-delay forgetting rate; SDFR, short-delay forgetting rate. ANCOVA results of verbal memory and hippocampal volumes (CFD vs. HCs). The forest plot displays adjusted mean differences and 95% confidence interval. P-values are FDR-corrected. Thresholds for partial eta-squared (ηp2) effect sizes are shown. Models controlled for age, sex, education, HAMA, and HAMD (eTIV for subfield volumes). ANCOVA, analysis of covariance; CA, cornu ammonis; CFD, craniofacial dystonia; eTIV, estimated total intracranial volume; GC-DG, granule cell layer of the dentate gyrus; HAMA, the Hamilton Rating Scale for Anxiety; HAMD, the Hamilton Rating Scale for Depression; HATA, hippocampus-amygdala transition area; HCs, healthy controls; LDFR, long-delay forgetting rate; SDFR, short-delay forgetting rate. Correlation analyses Within the CFD group, we performed partial correlation analyses to investigate potential associations between cognitive performance, hippocampal subfield volumes, and clinical characteristics, while controlling for age, sex, HAMA, and HAMD scores. These analyses specifically examined the relationships between: (1) RAVLT memory scores and hippocampal subfield volumes; (2) RAVLT memory scores and clinical variables (disease duration and BFMDRS scores); and (3) hippocampal subfield volumes and clinical variables (disease duration and BFMDRS scores). After applying FDR correction across all comparisons, none of these correlations reached statistical significance (all P-FDR > 0.05) (Table 5). TABLE 5Partial correlation analyses of hippocampal volumes, memory scores, and clinical variables in patients with CFD.Variable 1Variable 2 r P-FDR Memory scores vs. Hippocampal subfield volumes Immediate recallLeft GC-DG−0.1470.798Left CA4−0.1240.921Right GC-DG0.0180.971Right CA40.0410.971Short-delay recallLeft GC-DG−0.0050.971Left CA40.0150.971Right GC-DG0.0110.971Right CA4−0.0070.971Long-delay recallLeft GC-DG0.0130.971Left CA40.0180.971Right GC-DG0.0970.971Right CA40.0570.971 Memory scores vs. Clinical variables Immediate recallDisease duration0.1530.798BFMDRS0.0080.971Short-delay recallDisease duration0.1640.798BFMDRS0.0540.971Long-delay recallDisease duration0.2030.798BFMDRS0.1500.798 Hippocampal subfield volumes vs. clinical variables Left GC-DGDisease duration0.0630.971BFMDRS0.3730.146Right GC-DGDisease duration0.0220.971BFMDRS0.3290.171Left CA4Disease duration0.1630.798BFMDRS0.3560.146Right CA4Disease duration0.0330.971BFMDRS0.2910.263The results shown are from a series of partial correlation used to compare the variables with age, sex, HAMA, and HAMD scores as covariates. For every comparison, this table provides the partial correlation coefficient (r) and the precise P-FDR value, corrected for multiple comparisons using the Benjam-Hochberg FDR procedure. BFMDRS, Burke-Fahn-Marsden Dystonia Rating Scale; CFD, craniofacial dystonia; FDR, False Discovery Rate; HAMA, the Hamilton Rating Scale for Anxiety; HAMD, the Hamilton Rating Scale for Depression. Partial correlation analyses of hippocampal volumes, memory scores, and clinical variables in patients with CFD. The results shown are from a series of partial correlation used to compare the variables with age, sex, HAMA, and HAMD scores as covariates. For every comparison, this table provides the partial correlation coefficient (r) and the precise P-FDR value, corrected for multiple comparisons using the Benjam-Hochberg FDR procedure. BFMDRS, Burke-Fahn-Marsden Dystonia Rating Scale; CFD, craniofacial dystonia; FDR, False Discovery Rate; HAMA, the Hamilton Rating Scale for Anxiety; HAMD, the Hamilton Rating Scale for Depression. Discussion The present study provides compelling evidence for a specific verbal memory impairment in patients with CFD, which persists even when compared to a clinical control group with similar hyperkinetic movements. This cognitive deficit appears to be an intrinsic feature of the disorder, as it showed no correlation with motor severity or disease duration. Concurrently, we identified significant bilateral atrophy in the GC-DG and CA4 hippocampal subfields. The pivotal finding of our study, however, is the striking dissociation between these structural and functional alterations: the extent of this bilateral atrophy was not associated with the severity of verbal memory impairment. This lack of a direct relationship is particularly paradoxical, as more extensive, bilateral damage to memory-related structures would logically be expected to correlate with functional decline. These results firmly establish altered cognitive function as an integral part of the CFD clinical spectrum and suggest a complex, non-linear pathophysiology that cannot be explained by macroscopic atrophy alone. Future longitudinal research with larger sample sizes is essential to unravel the mechanisms underlying this structure-function dissociation. In this study, while MMSE scores did not differ significantly between patients with CFD and the HCs, the immediate recall, short-delay recall, and long-delay recall scores were significantly lower in patients with CFD compared with the HCs, suggesting that although the overall cognitive function may appear to be normal; certain specific cognitive domains such as verbal memory are impaired in patients with CFD. Impairment of memory function has been reported using other neuropsychological batteries in primary blepharospasm and cranial-cervical dystonia (Romano et al., 2014). Such subtle cognitive alterations may be associated with the distracting effects of abnormal movements, as suggested by improvements in attention impairment after BoNT treatment in patients with cervical dystonia (Allam et al., 2007). The excessive movement in patients with CFD likely did not affect their scores on the verbal memory tests because these patients performed significantly worse on the RAVLT when compared with patients with HFS, consistent with findings of previous literature demonstrating impairments in cognitive flexibility in patients with BSP compared with patients with hyperkinetic symptoms of non-dystonic origin (Lange et al., 2016). The absence of a significant association between symptom severity in patients with CFD and verbal memory deficits, including immediate, short- and long-delay recall, also supports this theory. These findings suggest that the altered verbal memory function in patients with CFD results from specific pathophysiological processes underlying dystonia. A key finding of the present study is the observed dissociation between hippocampal subfield atrophy and the severity of verbal memory deficits in patients with CFD. This paradox is underscored by our discovery of bilateral atrophy in the GC-DG and CA4 subfields. Given that these subfields, particularly within the left hemisphere, are widely regarded as critical neural substrates for verbal memory (Maguire, 2001; Ezzati et al., 2016; Tsalouchidou et al., 2023), one would logically anticipate that more extensive, bilateral structural damage would manifest in a clear structure-function relationship. Instead, the complete absence of such a correlation directly challenges the conventional assumption of a linear link between macroscopic atrophy and cognitive impairment in dystonia. This observation suggests that conventional etiological models may be insufficient, pointing to the potential value of a more comprehensive framework to better understand the pathophysiology of cognitive deficits in CFD. However, before proposing novel biological frameworks, it is imperative to critically evaluate the methodological factors that could potentially account for this null finding. Several methodological factors warrant consideration as potential explanations for the lack of a significant correlation. First, statistical power limitations are a well-documented challenge in clinical neuroscience research. As seminal work highlights, underpowered studies not only have a reduced chance of detecting a true effect but also decrease the likelihood that a statistically significant result reflects a true effect (Button et al., 2013). While our study identified significant group differences, the sample size may have been insufficient to detect a potentially subtle covariance between subfield atrophy and cognitive scores, risking a Type II error. Second, the technical challenges of hippocampal subfields segmentation must be acknowledged. Automated segmentation relies on standardized protocols and atlases, which, while robust, can be susceptible to inaccuracies, especially in brains with existing atrophy or anatomical variability (Goubran et al., 2020). Such measurement variance could obscure a true, underlying structure-function relationship. Third, the cross-sectional nature of our study limits causal inference. It is plausible that atrophy and cognitive decline follow different temporal trajectories. These methodological considerations demand caution in interpreting the null finding as definitive biological dissociation. Finally, we did not separately analyze the BSP and BOD subgroups. Given the imbalanced sample sizes in our cohort (40 patients with BSP vs. 10 patients with BOD), a formal statistical comparison would have been severely underpowered, and its results could be unreliable or misleading. Future studies with larger, more balanced cohorts are warranted to investigate potential subtype-specific effects. Nevertheless, while these limitations warrant caution, the consistency of our primary findings encourages a deeper exploration of the potential biological mechanisms underlying this dissociation. Assuming the observed dissociation is not solely a methodological artifact but reflects a true biological phenomenon, we propose that the primary driver of verbal memory deficits in CFD may not originate from focal hippocampal pathology, but from dysfunction within large-scale cortico-basal ganglia networks. Dystonia is increasingly understood not as a disorder of the basal ganglia alone, but as a brain-wide network disorder (Gill et al., 2023). Foundational work has established that these circuits, subserving both motor and non-motor functions, are critically implicated in the pathophysiology of dystonia (Alexander et al., 1986; Stamelou et al., 2012; Ray et al., 2020). Crucially, cognitive and psychiatric features are now recognized as core components of the “non-motor syndrome” of dystonia (Stamelou et al., 2012). Verbal memory impairment, as observed here, may be attributable to disruptions in frontally-mediated executive strategies—such as inefficient encoding and impaired strategic retrieval—which are direct consequences of basal ganglia pathophysiology (Monaghan et al., 2021). This perspective is further supported by evidence that chronic dysregulation within cortico-striatal circuits can induce remote structural changes in connected regions like the hippocampus. Such processes may operate through mechanisms of altered trophic support (Brito et al., 2014). In this “network-first” view, the focal hippocampal atrophy we observed may represent a downstream consequence of chronic network dysregulation or a bystander marker of a global disease process, rather than the causal driver of the cognitive symptoms. Secondly, the dissociation between macroscopic structure and function may reflect pathological processes occurring at a scale not captured by conventional volumetric MRI. Although we did not find a significant correlation between atrophy in the bilateral GC-DG and CA4 subfields and delayed recall performance in CFD patients, previous experimental animal studies have shown that hippocampal subregions such as CA1 and CA3 are highly vulnerable to oxidative damage, suggesting that structural and molecular disruptions in these areas—whether from toxic exposure or underlying disease—can impair memory-related processes (Ogut et al., 2019). This is highly relevant, as a growing body of evidence implicates mitochondrial dysfunction and oxidative stress as core pathophysiological concepts in dystonia (Koptielow et al., 2025). Therefore, it is plausible that molecular disruptions (e.g., from oxidative stress) in vulnerable subfields impair memory-related processes long before they culminate in detectable atrophy. This concept of functional impairment preceding volumetric loss is powerfully illustrated in the neurodegeneration literature. In Alzheimer’s disease, cognitive decline correlates more strongly with synaptic loss—a “synaptopathy”—than with frank neuronal loss (Terry et al., 1991; Selkoe, 2002). It is therefore plausible that the verbal memory deficits in our CFD cohort reflect an underlying synaptopathy within the memory network (perhaps driven by the oxidative stress mechanisms suggested by animal models), while the volumetric changes we measured represent a delayed, less sensitive proxy. Conclusion This study reveals a critical apparent paradox in CFD: focal hippocampal atrophy occurs alongside significant verbal memory deficits, yet the two are not directly correlated. This finding challenge simplistic structure-function models. After rigorously considering potential methodological limitations (e.g., statistical power, segmentation accuracy), we hypothesize this dissociation may reflect two possibilities: (1) the cognitive impairment is driven by large-scale basal ganglia network dysfunction, for which hippocampal atrophy is a downstream marker; or (2) the deficits stem from underlying microstructural pathology (e.g., synaptopathy) not captured by conventional MRI. Clinically, our findings establish verbal memory impairment as a core feature of CFD that warrants assessment independently of structural MRI findings. Future longitudinal, multi-modal neuroimaging studies are necessary to test these competing network and microstructural hypotheses.