Fractional amplitude of low-frequency fluctuations in sensory-motor networks and limbic system as a potential predictor of treatment response in patients with schizophrenia

,


Introduction
Schizophrenia (SCZ) is a debilitating mental disorder characterized by distorted cognition, emotional numbing, and cognitive deficits (Picchioni and Murray, 2007).Afflicting about 1 % of the global population with diverse clinical features (Saha et al., 2007), SCZ poses significant challenges to effective treatment.In a review of 50 studies, only 13.5 % with SCZ met clinical and social recovery criteria (Jaaskelainen et al., 2013).Besides recovery challenges, SCZ reduces life expectancy.Elevated comorbidity rates contribute to premature deaths, mainly due to chronic physical conditions like heart disease, stroke, diabetes, and cancer (Laursen et al., 2014).Consequently, the global burden of SCZ is escalating, leading to severe repercussions for affected families and society at large (Charlson et al., 2018).
Resting-state functional magnetic resonance imaging (rs-fMRI) has garnered considerable recognition as a promising tool for investigating various neurological and psychiatric disorders (Nakamura et al., 2020).The assessment of functional connectivity (FC) during the resting-state offers an intriguing method for studying functional networks or intrinsic brain connections without external stimuli influencing the brain activity (Beckmann et al., 2005;De Luca et al., 2006).A study by Zhang and colleagues revealed that individual with SCZ demonstrated notably heightened network homogeneity (NH) values in the left medial prefrontal cortex (MPFC), coupled with significant reductions in NH values in the bilateral posterior cingulate cortex/precuneus (PCC/PCu) in comparison to healthy controls (HCs) (Zhang et al., 2020).Furthermore, there exists corroborative evidence suggesting that elevated globalbrain functional connectivity (GFC) in the left superior frontal gyrus (SFG) may potentially serve as a distinctive trait of SCZ (Ding et al., 2019).Simultaneously, Yang and colleagues provided evidence that individuals with adolescent-onset SCZ (AOS) exhibited heightened resting-state FC values in the shell and tail of the nucleus accumbens compared to HCs, indicating the potential use of neuroimaging markers for early detection of AOS (Yang et al., 2022).Moreover, neuroimaging studies have revealed structural and functional irregularities in various brain areas among SCZ patients, including the cerebellum, temporal lobe, posterior cingulate gyrus, prefrontal cortex, hippocampus, and thalamus (Bois et al., 2015;Chung and Cannon, 2015;Thermenos et al., 2013).Nevertheless, some investigations have yielded conflicting outcomes related to abnormal FC in specific brain regions (Bluhm et al., 2007;Ongur et al., 2010), underscoring the crucial role of rs-fMRI in SCZ research.
Biswal et al. were the pioneered the revelation of synchronized spontaneous low-frequency fluctuations (LFFs) in resting-state fMRI between the primary motor cortices of both brain hemispheres (Biswal et al., 1995).Building upon this foundational work, Zang et al. introduced the concept of Amplitude of Low-Frequency Fluctuations (ALFF), involving the computation of the square root of the power spectrum within a specific low-frequency range.This computation aimed to gauge the local intensity of spontaneous fluctuations in the blood-oxygen-level dependent (BOLD) signal (Zang et al., 2007).However, due to the vulnerability to physiological noise of ALFF, Zou et al. innovated the fALFF method.This approach effectively mitigates non-specific signal components during resting-state fMRI, thereby enhancing the identification of inherent brain activity in targeted regions.This enhancement boosts both sensitivity and specificity (Zou et al., 2008).The fALFF technique has been widely applied to investigate functional aberrations in individuals with psychiatric disorders (Chai et al., 2020;Gimenez et al., 2017;Mazza et al., 2023;Yang et al., 2018;Zhang et al., 2022).Research indicates that assessing fALFF in baseline resting-state fMRI data can facilitate the use of spontaneous activity within the orbitofrontal cortex as a predictive biomarker for individual responses to antipsychotic medications (Lencz et al., 2022).Additionally, a previous study revealed a connection between genetic factors influencing cognitive processes and brain functions.Notably, fALFF has been found to mediate the connections between candidate single-nucleotide polymorphisms associated with SCZ, gray matter distribution, and cognitive functions, particularly in the domain of working memory (Gao et al., 2022).Collectively, these findings underscore the potential of fALFF as a valuable tool for probing the underlying neuropathological mechanisms of SCZ and as a promise biomarker for predicting treatment response of patients with SCZ.
Cognitive function encompasses a spectrum of intellectual skills such as perception, reasoning, and memory.Conversely, cognitive impairment denotes a reduction or dysfunction in these cognitive capacities (Kelleher, 2012).Comprehensive analyses have revealed that individuals diagnosed with SCZ exhibit more pronounced deficits in diverse cognitive areas when compared to both HCs and patients with mood disorders (Gebreegziabhere et al., 2022).Research findings also underscore a strong connection between the extent of cognitive impairment observed in SCZ and atypical neural activity within several brain regions (He et al., 2019;Xie et al., 2022Xie et al., , 2021;;Yu et al., 2022).The previous studies identified correlations between functional indicators and cognitive function, providing evidence that abnormal brain function may be a pathological mechanism of SCZ (Silberstein et al., 2018;Wen et al., 2021).Nonetheless, the current body of research is still in the process of achieving a thorough understanding of these relationships.
Furthermore, the exploration of neuroimaging as a prognostic biomarker has been extended using machine learning algorithms like support vector regression (SVR) (Zhang and O'Donnell, 2020).Previous studies applying the SVR algorithm with neuroimaging indicators have successfully predicted short-term responses in conditions such as obsessive-compulsive disorder (Yan et al., 2022) and SCZ (Jing et al., 2023).Thus, our aim is to employ SVR analysis based on fALFF to forecast treatment response in individuals with SCZ.
This study aims to enhance our understanding of the neuropathological basis of clinical characteristics by investigating abnormal activity through fALFF in individuals with SCZ and HCs.Additionally, we aim to monitor patients' progress three months after treatment, examining differences between the post-treatment phase and baseline.Our objective is also to assess the potential of these patterns as prognostic biomarkers and explore their responsiveness to pharmacotherapy effects.Building on previous findings linking cognitive impairment to abnormal brain activity in schizophrenia, we hypothesize that deviations in fALFF may correlate with cognitive assessments.

Participants
This study involved 107 participants, comprising 56 individuals diagnosed with SCZ and 51 HCs.HCs were recruited from physical examination centers and local communities, while individuals with SCZ were selected from the Department of Psychiatry at the Third People's Hospital of Foshan.All participants were chronic inpatients.The first MRI data collection and scale assessments were completed within one day, with the second assessment conducted three months after the treatment.During this period, treatment was provided by the doctors based on the patients' condition, and the specific treatment plan and duration of illness are detailed in Table S1.
The inclusion criteria for SCZ participants were as follows: (1) Fulfillment of the Diagnostic and Statistical Manual of Mental Disorders-5 (DSM-5) diagnostic criteria for SCZ; (2) Age ranging between 18 and 55 years; (3) Right-handedness; (4) An education level of at least six years.
The criteria for including HCs were as follows: (1) Absence of personal or familial history of mental illnesses; (2) Age ranging from 18 to 55 years; (3) Right-handedness; (4) A minimum of six years of education.
The participants' psychological and cognitive states were assessed using the Positive and Negative Syndrome Scale

Procedure
At baseline, all participants underwent a 3.0 T brain MRI scan and C. Zhang et al. provided clinical data.After the initial assessment, SCZ patients received a 3-month antipsychotic treatment course and subsequently participated in a follow-up MRI scan.Their clinical symptoms were assessed using PANSS, HAMD, HAMA, and ITAQ at both the beginning and the end of the study.

Measures
The PANSS comprises 30 items, distributed into three subscales: a Positive Scale (P) containing 7 items, a Negative Scale (N) with 7 items, and a General Psychopathology Scale (G) featuring 16 items.Each item is evaluated on a 7-point scale, ranging from 1 to 7. The total score is calculated by summing the subscale scores.Higher scores denote greater SCZ symptom severity (Kay et al., 1987).The HAMD is widely used to assess depressive severity and treatment effectiveness.Scores are collected before and after treatment through conversation and observation, where higher scores indicate greater symptom severity.Similarly, the HAMA assesses anxiety symptoms, with each of the 14 items scored on a scale from 0 to 4. Elevated scores signify increased anxiety symptoms (Bagby et al., 2004).The SDSS consists of 10 items, each assigned 0-2 points.The total score is the sum of all items, where higher scores indicate more pronounced social function impairments in the patient (Huang et al., 2022).The SCSQ functions as a tool to evaluate an individual's coping abilities.It covers 20 questions across two dimensions: positive and negative coping (Fan et al., 2022).The ITAQ comprises an 11-item questionnaire created to gauge patients' comprehension of their mental disorders and treatment requirements.The overall score, assessed on a 3-point Likert scale, reflects the patient's understanding of the illness and its treatments, with higher scores indicating a more extensive comprehension (Kemp and Lambert, 1995).The RBANS assesses cognitive capacities in five areas: immediate memory, language, attention, delayed memory, and visual-spatial construction.Improved cognitive performance is indicated by higher scores (Olaithe et al., 2019).The WCST is a cognitive assessment that targets executive function, particularly emphasizing cognitive flexibility, problem-solving, and adaptability to changing rules.Employing 64 cards displaying diverse shapes, colors, and numbers, individuals must categorize the cards based on evolving classification rules.This enables researchers to scrutinize their capacity to adjust cognitive strategies according to feedback (Nieuwenstein et al., 2001).The SCWT serves as a neuropsychological evaluation, gauging cognitive processing skills like selective attention and cognitive adaptability.It presents participants with conflicting stimulus attributes and assesses their capability to handle interference and switch between tasks (Detailed operating procedures are available in the Supplementary Materials-Methods) (Brugnolo et al., 2016).The MCCB functions as a standardized assessment instrument to appraise cognitive capacities in individuals facing mental health challenges, comprising 10 subtests spanning diverse cognitive domains.The evaluation approach encompasses the administration of cognitive exercises to document performance scores, facilitating the assessment of cognitive shortcomings and intervention impacts.Its widespread adoption in clinical trials underscores its significance (Plichta et al., 2023).

Imaging data acquisition and preprocessing
Resting-state fMRI data was acquired utilizing a GE 3.0 T scanner (GE 3.0 T Signa Pioneer) with the following parameters: repetition time/ echo time at 2000/30 ms, 36 slices, a 64 × 64 matrix, a 90 • flip angle, a 24 cm field of view, 4 mm slice thickness, no gap, and a total of 250 volumes (500 s).Participants were instructed to remain motionless, close their eyes, and stay awake.Earplugs and foam pads were provided to mitigate scanner noise and limit head movement.
Preprocessing, conducted using the Data Processing Assistant for Resting-State fMRI (DPARSF) software within MATLAB.This included slice timing correction, head motion correction, 3 × 3 × 3 mm 3 standard, and maximum displacement and angular motion limits of 2 mm and 2 • , respectively (Chao-Gan and Yu-Feng, 2010).The processed scans then underwent spatial smoothing with an 8 mm full-width at halfmaximum Gaussian kernel.Procedures such as linear trend removal and 0.01-0.08Hz band-pass filtering were also performed (Song et al., 2011).

fALFF calculation
The computation of fALFF was performed using the DPARSF software package, following the method for obtaining fALFF maps as detailed in reference (Zou et al., 2008).Initially, the time series of each voxel underwent conversion into the frequency domain utilizing a fast Fourier Transform, yielding the power spectrum.Then, the square root was computed for each frequency of this power spectrum, generating the mean square root within the frequency range of 0.01-0.08Hz for every voxel.Ultimately, normalization of voxels took place by dividing the fALFF value of each voxel by the overall average fALFF value within the brain mask.

Statistical analysis
For data analysis in this study, SPSS version 25.0 was employed.A Chi-square test assessed sex distinctions between the two groups.Continuous variables, such as age, years of education, and clinical scales, were subjected to scrutiny through two-sample t-tests.A predetermined significance threshold of p < 0.05 was adopted.
The image data were analyzed by using the DPARSF software package.Normalized fALFF maps were subjected to two-sample t-tests at a significance level of corrected p < 0.05 by using the Gaussian Random Field (GRF) theory method (voxel significance: p < 0.001, cluster significance: p < 0.05) for correction of multiple comparisons.To mitigate potential impacts of variables, covariates such as age, sex, education level, and mean framewise displacement (FD) were taken into account.Evaluating the contrast between pre-and post-treatment fALFF was executed through paired t-tests at a corrected p < 0.05 level with the GRF theory (voxel significance: p < 0.001, cluster significance: p < 0.05).Moreover, we extracted the fALFF values from clusters that exhibited abnormal fALFF at baseline to visually depict changes in them after treatment.
Pearson or Spearman correlation analyses were conducted to examine the relationship between abnormal fALFF of patients with SCZ at baseline and symptoms and behavioral assessments, with a significance level set at p < 0.05 (corrected p < 0.003, Bonferroni correction for simultaneously conducting 11 tests).Moreover, correlation analyses were performed between abnormal fALFF at baseline and cognitive assessments, with a significance level set at p < 0.05 (corrected p < 0.0008, Bonferroni correction for simultaneously conducting 53 tests).

SVR analysis
In this research, we employed the SVR algorithm, implemented through the LIBSVM toolbox, to predict the therapeutic outcomes for SCZ patients.These predictions were based on the voxel-wise fALFF values of brain regions in SCZ patients at baseline, which exhibited abnormal fALFF compared to HCs.The observed treatment response quantified as the reduction rate (RR) in PANSS scores following a threemonth treatment period.The computation of RR in PANSS score followed the formula below: RR = (pretreatment score -posttreatment score)/pretreatment score.
A leave-one-out cross-validation was applied alongside linear kernel functions in the SVR analysis.The voxel-wise abnormal fALFF values of SCZ patients at baseline were considered as potential features.A spearman correlation between fALFF values and observed treatment response was conducted to determine features.The fALFF values that showed a significant correlation with the observed treatment response were selected as features for inclusion in the cross-validation process (p < 0.05).Following feature selection, a leave-one-out cross-validation was performed to predict treatment response.Model performance was assessed using mean square error (MSE) and Pearson correlation between predicted and actual treatment responses.
After that, permutation tests were executed to assess whether the correlation coefficient (r) exceeded chance level.This involved 5000 permutations, wherein, for each permutation, the labels (observed RR in PANSS scores) were randomly reassigned to the features.The randomized data were then used to predict RR in PANSS scores, mirroring the approach with the actual data.This process yielded 5000 correlations between predicted and actual treatment responses.To determine the pvalue for the authentic correlation, the count of permutations resulting in a higher correlation than the authentic correlation was divided by the total number of permutations.

Demographic and clinical characteristics
This research involved the enrollment of a total of 56 individuals diagnosed with SCZ and 51 HCs, all of whom were included in the final analysis.Among the 56 SCZ patients, 37 successfully completed the 3month follow-up, with dropouts primarily attributed to the inconvenience caused by the COVID-19 pandemic.No noteworthy disparities are found in age, sex, years of education, and illness duration (month) between the SCZ and control cohorts.There are significant differences between the two groups in PANSS, HAMD, HAMA, SDSS, and RBANS scores.Some sub-items in the SCSQ, WCST, and SCWT scores also exhibit significant differences, while others show no significant distinctions.Detailed data can be found in Tables 1 and S2.
Additionally, there were no significant differences in demographics and psychological status between overall patients (n = 56) and those who completed the follow-up (n = 37) at baseline, indicating that they were demographically and clinically similar at baseline.Please refer to Table S3 for detailed information.

The treatment outcome
Table 2 presents the clinical characteristics of the 37 SCZ patients who completed the follow-up.By the endpoint, these patients demonstrated significant clinical improvement compared to their initial evaluations.

fALFF analysis in pre-treatment patients with SCZ and HCs
At baseline, SCZ patients exhibited lower fALFF values in various brain regions compared to HCs, such as the right postcentral/precentral gyrus, and left postcentral gyrus.Conversely, higher fALFF values were observed in the left hippocampus and right putamen at baseline.Table 3 and Fig. 1 provide more in-depth information.

fALFF analysis in pre-treatment and post-treatment patients with SCZ
After treatment, no significant difference was detected in fALFF after treatment.Furthermore, after treatment, no significant differences in mean fALFF values of the clusters which showed aberrant fALFF at baseline were observed (Fig. 2).Simultaneously, we observed no correlation between baseline fALFF and duration of illness in baseline patients, as detailed in Table S4.

Correlation analysis result
Pearson or Spearman correlation analyses showed that the fALFF values in the left postcentral gyrus (cluster3) were positively correlated with the scores of A e (total) (r = 0.4960, p = 0.0002, Bonferroni corrected) (Fig. 3) (Correlation results for other sections can be found in Table S5).Ae (total) is a component of the SCWT scale, representing the    total score for Task A incorrect responses in patients.

SVR analysis results
The results indicate that using voxel-wise abnormal fALFF at baseline can effectively forecast the reduction rate of PANSS total scores (r = 0.30, MSE = 0.04, p of r permutation = 0.004, Fig. 4), thereby predicting the therapeutic outcomes for SCZ.However, the predictive ability for the other three subscales is not significant (P scale: r = 0.22, MSE = 0.21; N scale: r = − 0.16, MSE = 0.42; G scale: r = − 0.02, MSE = 0.39).

Discussion
In this study, we observed lower fALFF values in the right postcentral/precentral gyrus and left postcentral gyrus, as well as higher fALFF values in the left hippocampus and right putamen, at baseline for SCZ patients compared to HCs.Following 3 months of treatment, significant clinical improvements were demonstrated in positive, negative, and general symptoms when compared to baseline assessments.However, when comparing fALFF values in regions where abnormal fALFF values were noted at baseline for SCZ patients who completed the follow-up, no significant differences were found after 3 months of treatment relative to the baseline data.Moreover, the SVR analysis demonstrated the potential of fALFF as a predictor for treatment response.
Our study reveals that, in compared to HCs, patients with SCZ exhibit lower fALFF values in sensory-motor networks, specifically the right postcentral/precentral gyrus and left postcentral gyrus.Nevertheless, the precise causes of these anomalies and their correlation with the underlying mechanisms of SCZ remain unclear.The right precentral gyrus, situated in the frontal lobe of the right cerebral hemisphere anterior to the central sulcus, predominantly governs motor control and coordination (Hardwick et al., 2013).Within the SCZ patient group, meta-regression analysis indicates a link between deactivations in the right precentral gyrus and poorer performance (Jani and Kasparek, 2018).This observation aligns with findings by Goswami et al., who identified reduced FC in the right precentral gyrus among SCZ patients (Goswami et al., 2020).Our research aligns with this perspective.Likewise, the postcentral gyrus, situated in the parietal lobe, serves as the primary somatosensory cortex, processing tactile and body position sensations.Ferro et al.'s investigation exhibited a more pronounced gray matter volume reduction in the right postcentral gyrus among male firstepisode SCZ patients in comparison to healthy males (Ferro et al., 2015).Additionally, previous studies have indicated diminished ALFF in bilateral postcentral regions among SCZ patients (Chen et al., 2022;Wang et al., 2019).Gao et al. showcased   features (Gao et al., 2022).
Nestled within the temporal lobe, the left hippocampus assumes a central role in cognitive functions like memory formation, learning, and spatial navigation (Zhong et al., 2020).A two-year follow-up revealed that during the initial stages of non-affective psychosis, individuals with SCZ exhibited heightened fALFF in both the anterior and posterior hippocampus.This suggests that hippocampal fALFF might serve as an indicator of vulnerability or the acute state of the illness in psychosis, rather than a persistent attribute of the disorder (McHugo et al., 2022).Simultaneously, research has demonstrated increased directional connectivity in the left hippocampus among SCZ patients, as compared HCs and other brain regions.Moreover, effective connectivity within the hippocampus has proven capable of distinguishing SCZ patients from HCs (Uscatescu et al., 2021).Intriguingly, fALFF in the left hippocampus has been associated with the reported severity of auditory and visual hallucinations (Hare et al., 2017).The proposition that hippocampal hyperactivity stands as a central pathophysiological mechanism in psychosis has gained traction (Lieberman et al., 2018).Treatments designed to reduce hippocampal excitability or activity could potentially mitigate the risk of developing or relapsing into psychosis (Gomes et al., 2016;Koh et al., 2018;Lodge and Perez, 2014).Our findings, indicating higher fALFF values in the left hippocampus in patients with SCZ compared to HCs, align with previous research and lend further support to our results.
The right putamen, a subcortical structure nestled within the basal ganglia, constitutes a cluster of nuclei deeply embedded within the cerebral hemispheres.This component plays pivotal roles in governing motor control, learning, and reinforcement processes, particularly in terms of managing movement on the left side of the body (Grahn et al., 2008;Postuma and Dagher, 2006).Our earlier investigations highlighted that, in comparison to HCs, individuals with SCZ initially exhibited heightened fALFF levels in both putamen.Following one week of olanzapine treatment, these SCZ patients demonstrated a decrease in fALFF levels specifically in the right putamen relative to their baseline measurements (Wu et al., 2019).The putamen, situated within the subcortical region of the striatum, houses elevated concentrations of dopamine, a pivotal neurotransmitter in the brain.In the context of SCZ, dopamine exerts a significant influence over intricate behaviors and cognitive functions (Schultz, 2007).The dopamine hypothesis of SCZ proposes that an excess of dopamine within the striatum, which includes the putamen, is linked to the emergence of positive symptoms such as  hallucinations and delusions (Toda and Abi-Dargham, 2007).Reflecting on these studies, our current research infers that structural and functional alterations in the right putamen might be connected with clinical outcomes in patients diagnosed with SCZ.
Regrettably, our study failed to detect any notable alterations in the fALFF values within the brain regions of SCZ patients following a threemonth treatment period.This lack of significant findings may be attributed to the possibility that the treatment duration was insufficiently lengthy to induce substantial changes in brain function.Alternatively, it could be due to the constraints imposed by a relatively modest sample size.
Our study does possess certain limitations.Firstly, the relatively small sample size in this study and the use of small samples for prediction methods deviate from the best practice guidelines in neuroimaging literature (Poldrack et al., 2020), potentially having an impact on our results.Therefore, our study should be considered as an exploratory one.Second, the treatment duration for SCZ patients is relatively brief, resulting in minimal alterations in the fALFF values within brain regions after treatment.Thirdly, the medication administered to patients is not limited to specific types, and the therapeutic drugs may also change throughout the treatment process according to the evolution of the condition.Thus, different medications may have varying impacts on the outcomes.Last, treating patients with second-generation antipsychotic drugs such as risperidone or aripiprazole, as conducted in studies like Todd Lencz's, can contribute to more accurate research results when receiving standard-category antipsychotic drug therapy (Lencz et al., 2022).However, due to some missing treatment data and the complexity of treatments (involving various therapies and non-pharmacological interventions), we cannot rule out interference caused by the prior treatment.

Conclusion
In summary, this study conducted an exploratory research on changes in fALFF values among healthy controls and individuals with schizophrenia.Our findings suggest that decreased fALFF values in the sensory-motor networks, alongside increased fALFF values in the limbic system, may constitute distinct neurobiological features of SCZ patients.These discoveries could potentially serve as imaging indicators for predicting the prognosis of individuals with SCZ.
Standard Deviation; PANSS = Positive and Negative Syndrome Scale; P = Positive Scale; N = Negative Scale; G = General Psychopathology Scale; HAMD = Hamilton Depression Rating Scale; HAMA = Hamilton Anxiety Rating Scale; ITAQ = Insight and Treatment Attitudes Questionnaire. a The p-values were obtained by paired t-tests.b The data were obtained from 22 patients with schizophrenia.

Fig. 1 .
Fig. 1.Brain regions with significant differences in the fALFF values at baseline between patients with SCZ and HCs.fALFF, fractional amplitude of low-frequency fluctuations.

Fig. 4 .
Fig. 4. In SVR analysis using voxel-wise abnormal fALFF at baseline could effectively predict treatment response in patients with SCZ.

Table 1
Demography and psychological status.
PANSS=Positive and Negative Syndrome Scale; P=Positive Scale; N=Negative Scale; G = General Psychopathology Scale; HAMD=Hamilton Depression Scale; HAMA = Hamilton Anxiety Scale; SDSS=Social Disability Screening Schedule; SCSQ = Simplified Coping Style Questionnaire; ITAQ = Insight and Treatment Attitudes Questionnaire. a The p-values were obtained by two sample t-tests.b The p-value for sex distribution was obtained by a chi-square test.

Table 2
Characteristics of patients who finished the follow-up.

Table 3
Regions with abnormal fALFF values in patients with SCZ at baseline.