Treatment-resistant schizophrenia (TRS) is a severe subtype of schizophrenia. While negative and disorganised symptoms have been associated with cognitive dysfunction, the underlying neurobiological mechanisms remain unclear. Plasma neurofilament light chain protein (NfL), a biomarker of neuroaxonal damage, has shown promising associations in TRS. This study investigates the interrelationships between symptomatology, cognition, and NfL in TRS.
MethodsEighty-two clozapine-treated TRS participants were assessed using the Positive and Negative Syndrome Scale, Scale for the Assessment of Positive Symptoms, and Scale for the Assessment of Negative Symptoms. Cognitive functions were evaluated using the Cambridge Neuropsychological Test Automated Battery and Wisconsin Card Sorting Test. Plasma NfL concentrations were measured using a Single Molecule Array assay. General linear models and moderation analyses were employed.
ResultsIn younger participants, NfL levels negatively correlated with excited factor (r = -0.26, 95% confidence interval [CI] = [-0.47, -0.04]), whereas in older participants, NfL levels positively correlated with blunted affect (r [95% CI] = 0.27 [0.05, 0.47]). Disorganised (unstandardised beta [B] = -0.07, 95% CI = [-0.11, -0.02]) and positive (B [95% CI] = -0.04 [-0.07, -0.005]) factors were significant predictors of executive dysfunction, independent of NfL levels. Importantly, NfL moderated the relationship between positive formal thought disorder and executive dysfunction (B [95% CI = 0.02 [0.001, 0.05]).
ConclusionsThese findings suggest distinct age-related neurobiological trajectories in TRS. While symptom-cognition relationships were relatively independent of NfL, its moderation effects point to potential compensatory or maladaptive responses that vary by illness stages.
Schizophrenia is characterised by disturbances in thought, perception, behaviour, and functioning, often accompanied by significant impairments in social and occupational domains. Treatment-resistant schizophrenia (TRS) affects approximately one-third of individuals with schizophrenia and is defined by persistent symptoms despite adequate trials of at least two antipsychotic medications at therapeutic doses with confirmed adherence.1–3 Compared to treatment-responsive schizophrenia, TRS is associated with more severe symptomatology, greater cognitive dysfunction, and functional decline.4
Symptoms of schizophrenia are classified into positive symptoms, such as hallucinations and delusions, and negative symptoms, such as blunted affect and avolition.5While these core symptoms define the disorder, cognitive impairments, particularly in executive function and memory, are prominent and have been linked to symptom severity and functional outcomes.6Disorganised and negative symptoms exhibit the strongest associations with cognitive dysfunction, whereas the relationship between positive symptoms and cognition remains less clear, with some evidence suggesting that persistent psychosis may indirectly impair cognition via neurocircuitry disruptions.7–9
In TRS, the interplay between symptomatology and cognitive dysfunction may be distinct. The Positive and Negative Syndrome Scale (PANSS) remains a fundamental tool for evaluating symptom dimensions and severity.10 Factor analyses have identified additional disorganised, excited, and depressed factors, offering greater granularity in symptom assessment.11–13 While these frameworks enhance symptom characterisation, specific symptom dimensions, particularly persistent psychosis, disorganisation, and negative symptoms, may hold greater relevance for understanding cognitive deficits in TRS.14
Neuroimaging studies have consistently demonstrated widespread structural abnormalities in schizophrenia, particularly in frontotemporal regions critical for cognition. These changes are dynamic, with grey matter reductions occurring early in the illness, followed by progressive white matter deterioration at later stages.15–17 Grey matter loss is most evident in the medial temporal, prefrontal, and cingulate cortices, with longitudinal studies showing that these abnormalities precede illness onset and worsen following psychosis onset, supporting a neuroprogressive trajectory.17 White matter abnormalities, emerging later, further exacerbate cognitive dysfunction by disrupting connectivity between cortical networks.15,18 Disruptions in cortico-cortical and cortico-subcortical pathways have been implicated in cognitive deficits, linking structural and functional dysconnectivity in schizophrenia.16 Cortical thinning in TRS has been associated with reduced cognitive efficiency, possibly reflecting compensatory mechanisms or accelerated neurodegeneration.19 While these neuroanatomical abnormalities correlate with cognitive deficits, their underlying mechanisms remain unclear. Identifying biomarkers of neuronal integrity may provide insights into the neurobiological underpinnings of TRS.
Neurofilament light chain protein (NfL), a cytoskeletal protein present in cerebrospinal fluid and bloodstream, is an established biomarker of neuroaxonal integrity. While NfL levels increase with healthy ageing, elevated levels are observed following axonal damage, making it a valuable marker across neurological and psychiatric disorders.20–25 In schizophrenia, NfL findings remain inconsistent. Some studies reported elevated NfL levels in individuals with severe symptoms or chronic illness, 21,26 while others, including our group’s prior work, found no significant differences between TRS and healthy controls or unaffected relatives.22,27 These discrepancies may reflect variations in illness stage, age, symptom severity, or methodological differences.27,28
Recent work has provided insights into the role of NfL. Wannan et al.28 identified age-dependent NfL patterns in TRS, where older individuals exhibited higher NfL levels, suggesting accelerated brain ageing and progressive neurodegeneration, while younger individuals showed lower NfL levels compared to controls, indicating distinct mechanisms at earlier illness stages. Additionally, Cilia et al.29 observed significant associations between NfL levels and cortical thinning in the bilateral insula, suggesting that NfL may reflect broader neurobiological processes beyond axonal damage. Together, these findings highlight NfL’s potential as a biomarker for distinct neurobiological pathways in TRS, warranting further investigation into its role in symptomatology and cognition. The present work extends this literature by integrating multidimensional clinical phenotyping with cognitive domains to map the functional relevance of neuroaxonal integrity, including potential variation across illness stages.
The present study aimed to investigate the interrelationships between symptomatology, cognition, and NfL in TRS, by: (1) assessing whether clinical symptoms and cognition are associated with NfL levels, and (2) exploring the relationships between clinical symptoms and cognition and determining whether NfL mediates or moderates these relationships. The primary hypothesis was that NfL moderates the relationship between clinical symptoms and cognition, with potential variation according to age or illness stage.
MethodsParticipantsEighty-two participants with TRS receiving clozapine and 59 age- and sex-matched healthy controls were enrolled in the Cooperative Research Centre (CRC) Psychosis Study,30,31 a TRS Biobank (https://researchdata.ands.org.au/treatment-resistant-schizophrenia-biobank/1325206). Participants, aged 21–60 years, were recruited from Melbourne clinical services between 2012 and 2017. The CRC Psychosis Study protocol and this study were approved by the Melbourne Health Human Research Ethics Committee (references: 2012.069 and 2020.142). All participants provided written informed consent before participation.
Diagnoses of schizophrenia were confirmed using the Structured Clinical Interview for the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, Axis I Disorders32 or the Mini International Neuropsychiatric Interview.33 Participant demographics and clinical characteristics have been described previously.27
Clinical symptomsSymptoms were assessed using the 30-item Positive and Negative Syndrome Scale (PANSS)10 and grouped into five factors (Positive, Negative, Disorganised, Excited, Depressed) per Wallwork et al.13
To complement this, participants were evaluated using the 34-item Scale for the Assessment of Positive Symptoms (SAPS)34 and the 25-item Scale for the Assessment of Negative Symptoms (SANS).35 The SAPS assessed Hallucinations, Delusions, Bizarre Behaviour, and Positive Formal Thought Disorder; the SANS assessed Affective Flattening or Blunting, Alogia, Avolition-Apathy, Anhedonia-Asociality, and Inattention.
Cognitive functionsCognitive performance was evaluated using the Cambridge Neuropsychological Test Automated Battery (CANTAB)36,37 and the Wisconsin Card Sort Test (WCST),38 focusing on memory and executive function (summarised in Supplementary Table S1). Subtest scores were standardised into Z-scores using the healthy control group to account for demographic and regional factors and better match the younger TRS cohort. Composite scores were calculated by averaging Z-scores within domains. Intelligence (IQ) was measured using the Wechsler Abbreviated Scale of Intelligence, Second Edition (WASI-II).39
Plasma NfL measurementFasting blood samples were collected from all participants. Plasma NfL concentrations were quantified using the Single Molecule Array (Simoa) NF-Light Advantage Kit (SR-X, Quanterix) as previously described,27 with a mean detection limit of 0.0552 pg/mL. Concentrations were standardised for age and converted to Z-scores based on a published Australian normative sample (n = 1926) using the same assay.22
Statistical analysisData were analysed using R version 4.4.1 (R Foundation for Statistical Computing, Vienna, Austria).40 Continuous variables were log-transformed to approximate normality, and results from log-transformed data were reported, with consistency confirmed by sensitivity analyses.
Outliers were defined as values exceeding ±3.0 standard deviations. Potential reasons for outliers were examined, and outliers were removed to prevent distortion of statistical analyses; sensitivity analyses were conducted with inclusion of outliers. Given some missing data across variables, a complete-case approach was applied, whereby participants with complete data for each statistical model were included. The analytic sample size (n) for each model was reported for transparency.
Descriptive statistics for demographic, clinical, cognitive, and biological variables were computed. Continuous variables were reported as means (standard deviations [SDs]), and categorical variables as frequencies (percentages). Difference between TRS participants and controls were assessed using t-test for continuous variables and Pearson’s Chi-squared test for categorical variables. Spearman’s correlation test and general linear models (GLMs) were used to examine the interrelationships between clinical, cognitive, and biological variables. Bootstrapping with 2000 replicates was used to compute 95% confidence intervals (CIs). Statistical significance was defined as a 95% CI not including zero or a p-value <0.05.
Primary analysesTwo sets of analyses were conducted to address the study aims:
- •
Aim (1): GLMs were conducted to examine the relationships between plasma NfL Z-scores (independent variable, IV) and clinical symptom scores or cognitive Z-scores (dependent variable, DV). Body Mass Index (BMI) and sex were included as covariate and cofactor, respectively.
- •
Aim (2): GLMs assessed the relationships between clinical symptom scores (IV) and cognitive Z-scores (DV), adjusting for age, sex, duration of illness, WASI-II IQ scores, and plasma clozapine levels. For significant relationships, potential moderating or mediating effects of NfL were further explored using GLMs and PROCESS Macro Model 1 (moderation) and 4 (mediation) for R,41 with NfL Z-scores entered as the moderator or mediator, adjusting for age, sex, duration of illness, plasma clozapine levels, and BMI. The Johnson-Neyman technique was applied to identify regions of significant moderation.42
Age-stratified subgroup analyses were conducted for participants aged 36 years or younger and participants aged over 36 years, based on the age-NfL interaction effect reported by Wannan et al.28
The 36-year cut-off was derived from prior research and is not intended to reflect a strict biological threshold. Given the exploratory nature of these analyses, for significant relationships, GLMs using age as a continuous variable were applied as sensitivity analyses to assess the age × NfL interaction.
Correction for multiple comparisonsTo account for multiple comparisons, the false discovery rate (FDR) correction43 was applied within families of related GLMs (e.g., symptom-cognition and NfL-outcome models). Both uncorrected results and FDR-adjusted p-values are reported to retain sensitivity in this exploratory study.
ResultsStudy cohort characteristicsAfter excluding one control participant with extreme plasma NfL levels (NfL = 60.1 pg/mL; Z-score = 3.5), 82 TRS participants and 58 controls were included. No significant differences were found in age (mean difference [MD] [95% CI] = 0.4 [−3.2, 3.7] years) or sex distribution (χ2 = 0.2, p = 0.7) between cohorts. Sensitivity analyses comparing the inclusion and exclusion of this control participant (outlier) showed no material change in the results.
TRS participants had a mean illness duration of 17.6 year and were receiving clozapine treatment, with plasma clozapine levels available for 81 participants (mean [SD] = 429.2 [226.9] μg/L). Although plasma NfL concentrations did not significantly differ between cohorts (MD [95% CI] = −0.2 [−1.4, 0.9] µg/L), age-standardised NfL Z-scores were slightly lower in TRS participants compared to controls (MD [95% CI] = −0.4 [−0.8, −0.01] Z-scores). Age-stratified analyses showed that younger TRS participants (≤ 36 years; n = 27) had lower mean NfL Z-scores (mean [SD] = −0.6 [1.4] Z-scores) compared to controls (MD [95% CI] = −0.8 [−1.3, −0.2] Z-scores), while older TRS participants (> 36 years; n = 55) exhibited mean NfL Z-scores (mean [SD] = 0.1 [1.3] Z-scores) that did not differ significantly from controls (MD [95% CI] = −0.2 [−0.6, 0.3] Z-scores). Group differences are illustrated in Fig. 1.
Significant differences in clinical and cognitive variables between TRS participants and controls are summarised in Table 1. TRS participants demonstrated greater symptom severity across all PANSS factors, with a mean total PANSS scores of 61.3 compared to 30.1 in controls. Controls performed within the normal range for memory (mean [SD] = 0.02 [0.9] Z-scores) and executive function (mean [SD] = 0.01 [0.6] Z-scores), while TRS participants showed impaired memory (mean [SD] = −2.0 [1.2] Z-scores) and borderline executive functioning (mean [SD] = −1.4 [0.9] Z-scores). Notably, 60% (n = 49) of TRS participants had cognitive impairment in either domain, and 21% (n = 17) had impairments in both. Individual test scores are detailed in Supplementary Table S1.
Study cohort characteristics.
| Variable | TRS | Control | Mean Difference | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| n | (%) $ | Mean | (SD) | n | (%) % | Mean | (SD) | MD | [95% CI] | ||
| Total | N | 82 | (100) | 58 | (100) | ||||||
| Sex female | 23 | (28) | 22 | (38) | χ2 = 0.2 ‘ | p = 0.7 | |||||
| Age | (years) | 82 | (100) | 40.3 | (9.3) | 58 | (100) | 39.5 | (10.7) | 0.4 | [−3.2, 3.7] |
| Duration of illness | (years) | 78 | (95) | 17.6 | (8.4) | - | - | - | - | - | - |
| Weight | (kg) | 71 | (87) | 95.8 | (24.9) | 53 | (91) | 74.9 | (11.4) | 20.0 | [14.1, 26.3] |
| BMI | (kg/m2) | 67 | (82) | 31.4 | (7.2) | 52 | (90) | 25.4 | (5.0) | 6.0 | [3.9, 8.1] |
| WASI-II IQ | 76 | (93) | 86.0 | (18.1) | 53 | (91) | 111.3 | (13.8) | −22.8 | [−27.9, −17.2] | |
| Plasma clozapine | (μg/L) | 81 | (99) | 429.2 | (226.9) | - | - | - | - | - | - |
| Plasma NfL | (pg/mL) | 82 | (100) | 6.6 | (4.5) | 58 | (100) | 6.6 | (2.8) | −0.2 | [−1.4, 0.9] |
| (Z-score) | 82 | (100) | −0.1 | (1.4) | 58 | (100) | 0.2 | (1.1) | −0.4 | [−0.8, −0.01] | |
| PANSS | Positive symptoms (/49) | 81 | (99) | 16.0 | (6.3) | 58 | (100) | 7.2 | (0.6) | 8.5 | [7.3, 9.9] |
| Negative symptoms (/49) | 81 | (99) | 17.9 | (6.1) | 58 | (100) | 7.0 | (1.6) | 10.1 | [8.7, 11.5] | |
| General symptoms (/112) | 81 | (99) | 28.3 | (6.7) | 58 | (100) | 15.9 | (1.3) | 11.9 | [10.6, 13.3] | |
| Total (/210) | 81 | (99) | 61.3 | (16.4) | 58 | (100) | 30.1 | (2.6) | 29.3 | [25.7, 32.7] | |
| PANSS 5-factor model § | Positive factor (/28) | 81 | (99) | 10.1 | (6.0) | 58 | (100) | 4.0 | (0.3) | 5.7 | [4.5, 6.9] |
| Negative factor (/35) | 81 | (99) | 14.2 | (5.6) | 58 | (100) | 5.5 | (1.0) | 7.9 | [6.7, 9.1] | |
| Disorganised factor (/21) | 81 | (99) | 7.6 | (3.2) | 58 | (100) | 3.6 | (1.3) | 3.7 | [3.0, 4.4] | |
| Excited factor (/28) | 81 | (99) | 5.7 | (2.8) | 58 | (100) | 4.2 | (0.5) | 1.5 | [1.0, 2.2] | |
| Depressed factor (/21) | 81 | (99) | 5.8 | (3.1) | 58 | (100) | 3.5 | (0.9) | 2.5 | [1.8, 3.2] | |
| SAPS | Hallucinations (/35) | 82 | (100) | 5.1 | (7.4) | 58 | (100) | 0.1 | (0.5) | 5.1 | [3.7, 6.6] |
| Delusions (/65) | 82 | (100) | 8.2 | (8.7) | 58 | (100) | 0.02 | (0.1) | 7.9 | [6.2, 9.6] | |
| Bizarre Behaviour (/25) | 82 | (100) | 2.1 | (2.6) | 58 | (100) | 0.1 | (0.5) | 2.0 | [1.4, 2.5] | |
| Positive Formal Thought Disorder (/45) | 82 | (100) | 4.4 | (6.3) | 58 | (100) | 0.3 | (1.0) | 4.4 | [3.1, 5.8] | |
| Total (/170) | 82 | (100) | 19.8 | (17.8) | 58 | (100) | 0.5 | (1.2) | 19.3 | [15.9, 23.1] | |
| SANS | Affective Flattening or Blunting (/40) | 82 | (100) | 13.8 | (7.0) | 58 | (100) | 0.7 | (1.3) | 12.2 | [10.7, 13.8] |
| Alogia (/25) | 82 | (100) | 5.1 | (4.1) | 58 | (100) | 0.3 | (1.0) | 4.6 | [3.8, 5.5] | |
| Avolition-Apathy (/20) | 82 | (100) | 7.9 | (4.8) | 58 | (100) | 1.5 | (2.6) | 6.2 | [5.1, 7.3] | |
| Anhedonia-Asociality (/20) | 82 | (100) | 10.2 | (6.0) | 58 | (100) | 1.8 | (3.2) | 8.2 | [6.8, 9.6] | |
| Inattention (/15) | 82 | (100) | 4.4 | (3.8) | 58 | (100) | 1.6 | (2.4) | 2.6 | [1.6, 3.6] | |
| Total (/120) | 82 | (100) | 41.5 | (17.9) | 58 | (100) | 5.9 | (6.3) | 33.8 | [29.9, 37.6] | |
| Cognition | Memory (Z-score) | 72 | (88) | −2.0 | (1.2) | 56 | (97) | 0.02 | (0.9) | −1.9 | [−2.2, −1.5] |
| Executive function (Z-score) | 77 | (94) | −1.4 | (0.9) | 56 | (97) | 0.01 | (0.6) | −1.3 | [−1.5, −1.0] | |
Abbreviations: BMI, body mass index; CI, confidence interval; IQ, intelligence quotient; IQR, interquartile range; kg, kilogram; kg/m2, kilogram per square metre; μg/L, microgram per litre; MD, mean difference; N, sample size; NfL, neurofilament light chain protein; NOS, not otherwise specified; PANSS, Positive and Negative Syndrome Scale; pg/mL, picogram per millilitre; SANS, Scale for the Assessment of Negative Symptoms; SAPS, Scale for the Assessment of Positive Symptoms; SD, standard deviation; TRS, treatment-resistant schizophrenia; WASI-II, Wechsler Abbreviated Scale of Intelligence – Second Edition.
Footnotes:.
Percentage of the total control group (N = 58, after removal of one control participant identified as an extreme outlier with abnormal plasma NfL levels).
The five PANSS factors and items defining them were: Positive factor (P1 Delusions, P3 Hallucinatory Behaviour, P5 Grandiosity, G9 Unusual Thought Content), Negative factor (N1 Blunted Affect, N2 Emotional Withdrawal, N3 Poor Rapport, N4 Passive/Apathetic Social Withdrawal, N6 Lack of Spontaneity and Flow of Conversation, G7 Motor Retardation), Disorganised factor (P2 Conceptual Disorganisation, N5 Difficulty in Abstract Thinking, G11 Poor Attention), Excited factor (P4 Excitement, P7 Hostility, G8 Uncooperativeness, G14 Poor Impulse Control), and Depressed factor (G2 Anxiety, G3 Guilt feelings, G6 Depression).
In TRS participants, NfL Z-scores were negatively correlated with PANSS Excited factor (r [95% CI] = −0.26 [−0.47, −0.04]), while positively correlated with SANS Affective Flattening or Blunting scores (r [95% CI] = 0.27 [0.05, 0.47]). After adjusting for sex, illness duration, and BMI, higher NfL Z-scores predicted lower PANSS Excited factor scores (B [95% CI] = −0.65 [−1.15, −0.08]). This prediction remained significant after further adjustment for plasma clozapine levels (B [95% CI] = −0.74 [−1.32, −0.08]). The association with SANS Affective Flattening or Blunting became insignificant after covariate adjustment (B [95% CI] = 0.99 [−0.55, 2.64]).
Age-stratified analyses revealed that in younger TRS participants, higher NfL Z-scores predicted lower PANSS Excited factor scores (B [95% CI] = −1.12 [−2.08, −0.38]), a relationship that was absent in the older cohort. Conversely, in older TRS participants, higher NfL Z-scores predicted increased SANS Affective Flattening or Blunting scores (B [95% CI] = 2.01 [0.27, 3.72]), a relationship not observed in the younger cohort. However, sensitivity analyses examining the age × NfL interaction showed no significant interaction effects for PANSS Excited factor scores (n = 64, F [7, 56] = 1.69, R2 = 0.17, p = 0.13) or SANS Affective Flattening or Blunting scores (n = 64, F [7, 56] = 1.29, R2 = 0.14, p = 0.27), after adjusting for sex, duration of illness, plasma clozapine levels, and BMI, suggesting limited support for age-dependent effects when age was modelled continuously.
No significant associations were found between NfL Z-scores and other symptom dimensions or cognitive outcomes (Supplementary Table S2).
Associations between clinical symptoms and cognitionThe relationships between clinical symptoms and cognition in TRS participants are summarised in Table 2. Using the five-factor PANSS model, the Disorganised (B [95% CI] = −0.07 [−0.12, −0.01]) and Positive (B [95% CI] = −0.04 [−0.07, −0.01]) factors predicted executive dysfunction. Among SAPS and SANS dimensions, SAPS Bizarre Behaviour (B [95% CI] = −0.05 [−0.12, −0.003]), SAPS Positive Formal Thought Disorder subscales (B [95% CI] = −0.02 [−0.05, −0.001]), and SANS Inattention (B [95% CI] = −0.06 [−0.10, −0.02]) were the strongest predictors of executive dysfunction. These associations remained significant after adjusting for age, sex, illness duration, IQ, and clozapine levels.
Summary of general linear models exploring the associations between clinical symptoms and cognitive functions in the TRS group.
| Memory outcome | Executive function outcome | |||||
|---|---|---|---|---|---|---|
| Predictor | n | GLMB [95% CI] | FDR-correctedp-value | n | GLMB [95% CI] | FDR-correctedp-value |
| PANSS | ||||||
| Positive factor | 65 | 0.02 [−0.03, 0.07] | 0.520 | 69 | −0.04 [−0.07, −0.01] $ | 0.049 % |
| Negative factor | 65 | −0.03 [−0.08, 0.03] | 0.520 | 69 | −0.02 [−0.05, 0.01] | 0.422 |
| Disorganised factor | 65 | −0.08 [−0.20, 0.02] | 0.520 | 69 | −0.07 [−0.12, −0.01] $ | 0.081 |
| Excited factor | 65 | −0.04 [−0.11, 0.12] | 0.520 | 69 | 0.02 [−0.05, 0.14] | 0.557 |
| Depressed factor | 65 | 0.03 [−0.08, 0.14] | 0.520 | 69 | −0.02 [−0.08, 0.03] | 0.536 |
| SAPS | ||||||
| Hallucinations | 65 | 0.01 [−0.03, 0.05] | 0.631 | 69 | −0.02 [−0.05, 0.01] | 0.267 |
| Delusions | 65 | 0.02 [−0.02, 0.04] | 0.631 | 69 | −0.01 [−0.04, 0.01] | 0.355 |
| Bizarre Behaviour | 65 | −0.05 [−0.14, 0.06] | 0.631 | 69 | −0.05 [−0.12, −0.003] $ | 0.185 |
| Positive Formal Thought Disorder | 65 | −0.01 [−0.06, 0.03] | 0.631 | 69 | −0.02 [−0.05, −0.001] $ | 0.185 |
| SANS | ||||||
| Affective Flattening or Blunting | 65 | −0.03 [−0.07, 0.01] | 0.453 | 69 | −0.02 [−0.05, 0.01] | 0.237 |
| Alogia | 65 | −0.06 [−0.12, 0.001] | 0.453 | 69 | −0.03 [−0.08, 0.02] | 0.250 |
| Avolition-Apathy | 65 | 0.01 [−0.05, 0.08] | 0.660 | 69 | −0.02 [−0.06, 0.02] | 0.323 |
| Anhedonia-Asociality | 65 | 0.01 [−0.04, 0.07] | 0.660 | 69 | −0.02 [−0.05, 0.01] | 0.237 |
| Inattention | 65 | −0.03 [−0.11, 0.04] | 0.644 | 69 | −0.06 [−0.10, −0.02] $ | 0.055 |
Abbreviations: B, unstandardised regression coefficient (reported in Z-score); CI, confidence interval (reported in Z-score); FDR, false discovery rate; PANSS, Positive and Negative Syndrome Scale; SANS, Scale for the Assessment of Negative Symptoms; SAPS, Scale for the Assessment of Positive Symptoms.
Footnotes:.
GLMs were adjusted for age, sex, duration of illness, WASI-II IQ scores, and plasma clozapine levels.
The coefficients (B) presented are unstandardised beta coefficients representing change in cognitive Z-scores per unit increase in PANSS, SAPS, or SANS measures. Given that cognitive Z-scores typically lie within a relatively constrained range (approximately −2.0 to +2.0 in clinically meaningful terms), modest coefficient estimates may reflect clinically relevant shifts.
Correlations further validated these findings, as SAPS Positive Formal Thought Disorder was strongly correlated with PANSS P2 Conceptual Disorganisation (r [95% CI] = 0.92 [0.87, 0.97]), a core component of the Disorganised factor. Similarly, SANS Inattention was moderately correlated with PANSS G11 Poor Attention item (r [95% CI] = 0.55 [0.40, 0.71]), reflecting consistent construct validity.
Given the absence of correlations between NfL levels and these clinical and cognitive variables (Supplementary Table S2), post hoc analyses adjusting for NfL Z-scores revealed the relative independence of NfL levels in the observed relationships (Model 2, Supplementary Table S3). Executive function continued to show significant associations with PANSS Disorganised factor (B [95% CI] = −0.07 [−0.11, −0.02]), PANSS Positive factor (B [95% CI] = −0.04 [−0.07, −0.005]), SAPS Positive Formal Thought Disorder (B [95% CI] = −0.03 [−0.05, −0.004]), and SANS Inattention (B [95% CI] = −0.06 [−0.10, −0.02]).
Table 3 details associations between clinical symptoms and specific psychometric tests. PANSS Disorganised factor demonstrated the broadest impact, predicting deficits in set-shifting (measured by CANTAB Intra-Extra Dimensional Set Shift), abstract reasoning and conceptualisation (WCST Trials to Complete First Category), self-ordered spatial working memory (CANTAB Spatial Working Memory), and spatial span (CANTAB Spatial Span). In contrast, PANSS Positive factor specifically predicted set-shifting performance. SAPS Positive Formal Thought Disorder was associated with deficits across set-shifting, self-ordered spatial working memory, and spatial span, while SANS Inattention predicted outcomes in set-shifting and abstract reasoning and conceptualisation. These associations remained significant after adjusting for age, sex, illness duration, IQ, clozapine levels, and NfL Z-scores (Model 2, Supplementary Table S4).
Summary of sensitivity test results exploring the associations between clinical symptoms and specific psychometric tests in the TRS group.
| Predictor | Executive function outcome | n | GLMB [95% CI] | FDR-correctedp-value |
|---|---|---|---|---|
| PANSS Positive factor | CANTAB IED Total Errors | 66 | −0.06 [−0.11, −0.01] $ | 0.044 % |
| PANSS Disorganised factor | CANTAB SWM Between Errors | 65 | −0.11 [−0.18, −0.03] $ | 0.024 % |
| CANTAB IED Stages Complete | 66 | −0.18 [−0.28, −0.08] $ | 0.008 % | |
| CANTAB IED Total Errors | 66 | −0.16 [−0.25, −0.08] $ | 0.008 % | |
| CANTAB SSP Span Length | 64 | −0.11 [−0.17, −0.03] $ | 0.035 % | |
| WCST Trials to Complete First Category | 56 | −0.13 [−0.26, −0.02] $ | 0.100 | |
| SAPS Positive Formal Thought Disorder | CANTAB SWM Between Errors | 65 | −0.05 [−0.08, −0.02] $ | 0.044 % |
| CANTAB IED Stages Complete | 66 | −0.08 [−0.12, −0.03] $ | 0.017 % | |
| CANTAB IED Total Errors | 66 | −0.07 [−0.11, −0.04] $ | 0.011 % | |
| CANTAB SSP Span Length | 64 | −0.06 [−0.10, −0.03] $ | 0.016 % | |
| SANS Inattention | CANTAB IED Stages Complete | 66 | −0.16 [−0.24, −0.08] $ | 0.003 % |
| CANTAB IED Total Errors | 66 | −0.14 [−0.20, −0.07] $ | 0.003 % | |
| WCST Trials to Complete First Category | 56 | −0.10 [−0.22, −0.01] $ | 0.082 |
Abbreviations: B, unstandardised regression coefficient (reported in Z-score); CANTAB, Cambridge Neuropsychological Test Automated Battery; CI, confidence interval (reported in Z-score); FDR, false discovery rate; IED, Intra-Extra Dimensional Set Shift; PAL, Paired Associates Learning; PANSS, Positive and Negative Syndrome Scale; SANS, Scale for the Assessment of Negative Symptoms; SAPS, Scale for the Assessment of Positive Symptoms; SSP, Spatial Span; SWM, Spatial Working Memory; WCST, Wisconsin Card Sort Test.
Footnotes:.
GLMs were adjusted for age, sex, duration of illness, WASI-II IQ scores, and plasma clozapine levels.
The coefficients (B) presented are unstandardised beta coefficients representing change in cognitive Z-scores per unit increase in PANSS, SAPS, or SANS measures. Given that cognitive Z-scores typically lie within a relatively constrained range (approximately −2.0 to +2.0 in clinically meaningful terms), modest coefficient estimates may reflect clinically relevant shifts.
Among all examined relationships, NfL levels moderated the association between SAPS Positive Formal Thought Disorder and executive dysfunction after adjustment for age, sex, duration of illness, plasma clozapine levels, and BMI (n = 60, F [8, 51] = 2.64, R2 = 0.29, p = 0.02). While higher SAPS Positive Formal Thought Disorder scores independently predicted poorer executive function (B [95% CI] = −0.05 [−0.09, −0.02]), its interaction with NfL levels inversely predicted better executive function (B [95% CI] = 0.02 [0.002, 0.05]). However, this moderation effect became insignificant at NfL levels exceeding Z = 0.73, as determined by Johnson-Neyman significance region analysis (Fig. 2).
Interaction plot demonstrating the moderating role of plasma NfL in the association between positive formal thought disorder and executive function in the TRS group (n = 60). Abbreviations: NfL, neurofilament light chain protein; SD, standard deviation; TRS, treatment-resistant schizophrenia.
No other symptom-cognition relationships or age-stratified analyses demonstrated significant moderation effects.
DiscussionThe present study investigated the interrelationships between clinical symptoms, cognition, and plasma NfL levels in individuals with well-characterised TRS. Our findings suggest that NfL (1) may show age-related differences in its associations with some, but not all, clinical symptoms; (2) may be relatively independent of symptom-cognition relationships; but (3) may indirectly influence executive functioning. These insights contribute to a growing understanding of the neurobiological mechanisms underlying symptomatology and cognitive deficits in TRS. Given the cross-sectional design, the below interpretations are speculative. Notably, continuous age × NfL interaction analyses were not statistically significant, indicating that these age-related patterns should be interpreted cautiously.
While most NfL literature has focused on elevated levels as a marker of neuroaxonal damage, our findings raise the possibility that lower NfL levels might also reflect pathological processes, potentially indicating a lack of compensatory or adaptive mechanisms, although this interpretation remains speculative and requires validation in longitudinal studies. Previous studies have reported inconsistent results regarding NfL in schizophrenia, with some identifying elevated levels in severe or chronic cases,21,26 while others observed no significant differences compared to healthy controls or unaffected relatives.27 These discrepancies likely reflect variations in age, illness stage, symptom severity, and methodological differences. Notably, recent findings by Wannan et al.28 demonstrated an age-dependent relationship of NfL, with younger TRS individuals exhibiting significantly lower NfL levels compared to controls, suggesting distinct pathophysiological mechanisms at earlier illness stages.
While prior studies examined group differences in NfL, the present work extends prior studies by integrating multidimensional symptom phenotyping (PANSS/SAPS/SANS), cognitive domains, and moderation analyses to examine the functional relevance of NfL.
In younger TRS participants, higher NfL levels were associated with fewer symptoms of excitement and hostility, consistent with Bavato et al.,21 who reported a negative correlation between NfL levels and the number of psychotic episodes. One possible explanation is that higher NfL levels may reflect processes potentially related to neural remodelling or plasticity aimed at stabilising disrupted circuits and mitigating behavioural dysregulation.20,44 Conversely, lower NfL levels could be consistent with reduced engagement of such mechanisms, aligning with reduced neural plasticity observed in chronic schizophrenia.45 Alternatively, because younger TRS individuals showed significantly lower NfL levels compared to controls,28 relatively higher NfL levels may approximate normality, indicating preserved neural integrity and reduced symptom severity. This may also signify other underlying mechanisms that deplete NfL levels and drive these symptoms in earlier phases of illness.
In contrast, among older TRS participants, higher NfL levels were associated with greater blunted affect. Given that NfL naturally increases with age, this association may be consistent with processes related to neurodegeneration and demyelination.28,46 This aligns with neuroimaging evidence demonstrating accelerated grey and white matter loss in schizophrenia, particularly in frontotemporal regions.15,17 White matter abnormalities, especially affecting the uncinate fascicle, have been linked to blunted affect, likely reflecting axonal loss and myelin damage.47–49 These findings suggest that NfL-symptom relationships may be non-linear and could vary across the course of illness. In younger individuals, higher NfL levels may reflect the brain's capacity for repair, while in older individuals, they may signify cumulative damage or accelerated brain ageing. However, such stage-based interpretations cannot be confirmed in the current cross-sectional design. Our results tentatively suggest the importance of considering both age and illness progression when interpreting NfL in TRS.28
Disorganisation and inattention were significant correlates to executive dysfunction in TRS, particularly affecting set-shifting and abstract reasoning. These results align with a previous meta-analysis demonstrating robust links between disorganisation and cognitive impairments across multiple domains.8 No significant associations were observed between the negative factor and cognition, consistent with findings by Bagney et al.50 and Rodriguez-Jimenez et al.51 These results suggest that, when separated from disorganisation, the negative factor may not consistently predict cognitive burden, reflecting prior conceptual overlaps when item such as PANSS N5 Difficulty in Abstract Thinking was part of the negative symptom subscale.50,51 The PANSS five-factor model reduces overlap between negative and cognitive items, which may account for this discrepancy and should be considered a methodological strength.
The positive factor, comprising delusions, hallucinatory behaviour, grandiosity, and unusual thought content, was also associated with executive dysfunction. Persistent psychosis may relate to cognitive impairments via mechanisms such as chronic dopaminergic dysregulation and maladaptive network reorganisation, particularly affecting cortico-striatal circuits.52–54 Structural MRI studies have shown that cortical thinning in prefrontal and temporal regions correlates with executive dysfunction,16,19 while white matter disruptions in cortico-striatal connectivity may further exacerbate executive deficits in TRS.19 Importantly, these symptom-cognition relationships were independent of NfL, suggesting that cognitive impairments in TRS may not be primarily explained by NfL-measured neuroaxonal damage in this sample. Instead, they may more likely result from functional disruptions, such as neurotransmitter imbalances or network dysconnectivity.9
Although NfL did not directly influence symptom-cognition relationships, it moderated the association between positive formal thought disorder and executive dysfunction. Higher NfL levels were associated with better executive functioning in individuals with severe thought disorder, whereas lower NfL levels predicted worse outcomes. However, among individuals without severe thought disorder, the association was reversed, with higher NfL levels linked to poorer executive functioning (Fig. 1). This cross-over pattern may be tentatively interpreted within frameworks such as the differential susceptibility model rather than the diathesis-stress model.55,56 This may suggest that NfL is associated with both beneficial and detrimental patterns, depending on the severity of psychopathology, reflecting compensatory or maladaptive processes at different stages of TRS.
This parallels the findings for excited symptoms and is consistent with the possibility that higher NfL levels may reflect compensatory processes in earlier or less severe stages of TRS.16,20,44 Conversely, lower NfL levels may signify a lack of these adaptive responses, leading to more severe cognitive impairments. This non-linear relationship, as evidenced by the Johnson-Neyman analysis, may indicate a potential threshold beyond which compensatory processes are less effective, potentially relating to processes associated with brain ageing and functional decline. In more advanced stages, NfL levels may plateau or decline, reflecting a depletion of compensatory capacity and linking to worse cognitive outcomes.
This study has several limitations that warrant consideration. While moderation and mediation analyses were employed to explore the interrelationships between symptoms, cognition, and NfL, these are inherently temporal processes. However, the cross-sectional design of this study precludes the ability to establish causality or confirm the directionality of these relationships. Assumptions about temporality were necessary, but longitudinal studies are needed to validate and clarify these dynamic interactions over time.
While executive function and memory were comprehensively assessed, other cognitive domains, such as language, attention and processing speed, and social cognition, were not evaluated. Future studies incorporating broader cognitive batteries could provide a more holistic understanding of cognitive dysfunction in TRS. Subgroup analyses stratified by age, although informative, were constrained by small sample sizes, particularly the younger cohort, potentially limiting the ability to detect subtle age-dependent effects. Larger, multi-site cohorts are necessary to validate these findings and further investigate whether the role of NfL differs across illness stages.
Another limitation is the lack of neuroimaging data, particularly measures of white matter disease. White matter abnormalities, such as in the uncinate fasciculus, have been linked to both blunted affect and cognitive impairment in schizophrenia47–49 As with prior studies from our group, we could not assess whether white matter pathology moderates or mediates the observed associations of NfL. Future research integrating neuroimaging techniques will be essential to address this gap.
Additionally, although plasma clozapine levels were appropriately controlled for, long-term cumulative exposure, concomitant medications, and individual metabolic differences may still influence symptom severity and cognitive performance. Importantly, clozapine should be considered not only as a statistical covariate but also a potential biological modulator of NfL levels. Future studies should aim to address these potential residual confounders as well as the effect of clozapine on NfL levels.
We also acknowledge the issue of multiple comparisons inherent in our statistical analyses. Given the exploratory nature of this study, both uncorrected and FDR-adjusted results were reported. However, we recognise that this approach may still increase the risk of Type I error. To improve confidence in these findings, future confirmatory studies should apply robust multiple comparison corrections in larger, independent cohorts to validate the observed effects.
To strengthen the generalisability of these findings, longitudinal studies are needed to explore the evolving relationship between NfL, symptoms, and cognition over time and to determine whether NfL reflects distinct neurobiological processes at different stages of TRS. Expanding biomarker panels to include measures of inflammation, oxidative stress, or structural imaging markers could further elucidate the complex neuropathology underlying TRS. Finally, future research should also investigate whether NfL’s potential compensatory or maladaptive effects vary by illness stage or symptom severity, particularly in relation to cognitive outcomes.
ConclusionsThis study provides preliminary insights into the relationships between clinical symptoms, cognitive impairments, and plasma NfL levels in TRS. The disorganised and positive factors emerged as key correlates of executive dysfunction, highlighting distinct symptom-cognition relationships in this subgroup. Age-related differences between NfL and clinical symptoms suggest possible differences in underlying processes across the course of illness, with patterns in younger and older individuals indicating potential variation in neurobiological mechanisms. However, NfL was not directly implicated in symptom-cognition relationships, suggesting that cognitive impairments in TRS may not be primarily driven by neuroaxonal damage. These findings emphasise the importance of considering age and illness progression in future studies investigating TRS pathophysiology. Validation in larger, longitudinal cohorts, alongside neuroimaging integration, will be crucial to further delineate the role of NfL and its relationship with clinical symptoms and cognitive dysfunction in TRS.
GlossaryBMI Body mass index.
CANTAB Cambridge Neuropsychological Test Automated Battery.
FDR False discovery rate.
GLM General linear model.
NfL Neurofilament light chain protein.
PANSS Positive and Negative Syndrome Scale.
SANS Scale for the Assessment of Negative Symptoms.
SAPS Scale for the Assessment of Positive Symptoms.
Simoa Single Molecule Array.
TRS Treatment-resistant schizophrenia.
WASI-II Wechsler Abbreviated Scale of Intelligence—Second Edition.
WCST Wisconsin Card Sort Test.
Author contributionsWei-Hsuan Chiu: Conceptualization; Data curation; Formal analysis; Investigation; Methodology; Project administration; Visualization; Writing – original draft; Writing – review and editing. Anita M.Y. Goh: Conceptualization; Methodology; Supervision; Writing – review and editing. Dhamidhu Eratne: Data curation; Investigation; Writing – review and editing. Charles B. Malpas: Methodology; Writing – review and editing. Alexander F. Santillo: Data curation; Investigation; Writing – review and editing. Shorena Janelidze: Writing – review and editing. Oskar Hansson: Data curation; Investigation; Writing – review and editing. Cassandra M.J. Wannan: Data curation; Investigation. Antonia H. Merritt: Data curation. Mahesh Jayaram: Writing – review and editing. Naveen Thomas: Writing – review and editing. Chad A. Bousman: Data curation; Investigation; Writing – review and editing. Ian Everall: Data curation; Investigation. Dennis Velakoulis: Conceptualization; Data curation; Investigation; Methodology; Supervision; Writing – review and editing. Christos Pantelis: Data curation; Investigation; Methodology; Supervision; Writing – review and editing. Samantha M. Loi: Conceptualization; Methodology; Supervision; Writing – review and editing.
Declaration of competing interestThe author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: OH has acquired research support (for the institution) from ADx, AVID Radiopharmaceuticals, Biogen, Eli Lilly, Eisai, Fujirebio, GE Healthcare, Pfizer, and Roche. In the past 2 years, he has received consultancy/speaker fees from AC Immune, Amylyx, Alzpath, BioArctic, Biogen, Cerveau, Eisai, Eli Lilly, Fujirebio, Merck, Novartis, Novo Nordisk, Roche, Sanofi and Siemens. CP has received honoraria for talks at educational meetings and has served on an advisory board for Lundbeck, Australia Pty Ltd., Servier Australia and TEVA Australia. The remaining authors have nothing to declare.
Funding statementThis research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.
Ethical considerationsThe CRC Psychosis Study protocol and this study were approved by the Melbourne Health Human Research Ethics Committee (references: 2012.069 and 2020.142). All participants provided written informed consent before participation. All procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Declaration of Helsinki.
Data availabilityThe corresponding author had full access to all the data in the study and could share upon reasonable request and approval by the CRC Psychosis Study.
CP was supported by a National Health and Medical Research Council (NHMRC) Senior Principal Research Fellowship (ID: 628386 & 1105825) and NHMRC Program Grant (ID: 566529). AFS was supported by the Alzheimer’s Association (SG-23-1061717), Swedish Research Council (2022-00775), ERA PerMed (ERAPERMED2021-184), Knut and Alice Wallenberg foundation (2017-0383), Strategic Research Area MultiPark (Multidisciplinary Research in Parkinson’s disease) at Lund University, Swedish Alzheimer Foundation (AF-980907), Swedish Brain Foundation (FO2021-0293), Parkinson foundation of Sweden (1412/22), Cure Alzheimer’s fund, Konung Gustaf V:soch Drottning Victorias Frimurarestiftelse, Skåne University Hospital Foundation (2020-O000028), Regionalt Forskningsstöd (2022-1259) and the Swedish federal government under the ALF agreement (2022-Projekt0080). WHC was supported by the Nikolaos and Dimitra Pantelis Travelling Scholarship in presenting her work at the 24th International Neuropsychiatry Association Congress. The TRS biobank was established by IE, CP, CB as part of a major Australian Department of Industry Co-operative Research Centre (CRC) grant—https://researchdata.ands.org.au/treatment-resistant-schizophrenia-biobank/1325206. The role of these funding sources was to support research study staff and bio-sample analyses.
The authors acknowledge the financial support of the CRC for Mental Health. The Cooperative Research Centre (CRC) programme is an Australian Government Initiative. The authors wish to acknowledge the CRC Scientific Advisory Committee, in addition to the contributions of study participants, clinicians at recruitment services, staff at the Murdoch Children’s Research Institute, staff at the Australian Imaging, Biomarkers and Lifestyle Flagship Study of Aging, and research staff at the Melbourne Neuropsychiatry Centre, including coordinators Merritt, A., Phassouliotis, C., and research assistants, Burnside, A., Cross, H., Gale, S., and Tahtalian, S. Participants for this study were sourced, in part, through the Australian Schizophrenia Research Bank (ASRB), which is supported by the National Health and Medical Research Council of Australia (Enabling Grant N. 386500), the Pratt Foundation, Ramsay Health Care, the Viertel Charitable Foundation and the Schizophrenia Research Institute. We thank the Chief Investigators and ASRB Manager: Carr, V., Schall, U., Scott, R., Jablensky, A., Mowry, B., Michie, P., Catts, S., Henskens, F., Pantelis, C., Loughland, C. We acknowledge the help of Jason Bridge for ASRB database queries. The authors are grateful for assistance from Brett Trounson and Dr Christopher Fowler and the team at The Florey Oak St Biobank. Finally, the authors would like to thank all the participants and their families.
