Inpatients of the Second Affiliated Hospital of Xinxiang Medical University (Xinxiang, Henan, China) were recruited from Match 2020 to July 2021. Inclusion criteria were (i) SP diagnosed according to the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition, (ii) 18–45 years old, (iii) duration of disease <10 years, (iv) no <9 years of education, and (v) right-handed. Exclusion criteria were (i) intellectual disability, a history of neurological diseases, or other serious physical diseases (such as liver/kidney/heart failure), (ii) received electroconvulsive therapy within 2 years, (iii) previous use of typical antipsychotic drugs, (iv) family history of mental disorders, (v) pregnancy, and (vi) a history of drug/substance abuse or addiction (except tobacco). Diagnoses were made by two associate chief physicians in psychiatry. All patients had residual symptoms but were in a relatively stable phase (antipsychotic medication dose stable for more than one week) that enabled completion of the assessment.

Age- and sex-matched HCs were recruited from the community during the same period. All were free of medication and evaluated by an experienced psychiatrist for exclusion of mental illness using the International Neuropsychiatric Interview. Candidate HCs were excluded for a personal or family history of neuropsychiatric illness, drug/substance abuse, addiction (except tobacco), or pregnancy.

The Chinese version of the PANSS (26) was administered as a semi-structured interview to estimate the severity of schizophrenia symptoms. The Chinese PANSS has demonstrated strong internal consistency (Cronbach's alpha = 0.87). The PANSS administrators were well-trained and demonstrated good inter-rater consistency.

Each participant completed a CFT and LFT in succession. Each test required 310 s and tests were administered in a four block-design (see Figure 1). Both tests were present on a computer screen using E-prime 2.0. Before the formal test, a practice session was conducted to make sure participants understood the task. To ensure a stable fNIRS waveform, a 10-s delay was inserted prior to task onset, followed by a 30-s pre-task period during which participants counted “1, 2, 3, 4, 5” repeatedly until the task began. For each 30-s task period, participants were requested to generate as many words as possible in response to a cue. The CFT cues were “four-legged animal,” “fruit,” “domestic appliance,” and “vegetable,” while cues for the LFT were words beginning with “山,” “大,” “白,” and “天.” There was a 30-s rest period between tasks and the participants were also instructed to count “1, 2, 3, 4, 5” repeatedly during this period. In the 70-s post-task period, the participants were asked to count as in the pre-task period. Participants were instructed to avoid movement, speak quietly, and minimize blinking during the test. Valid words were recorded by a researcher.

Cortical activation was measured during the CFT and LFT using a 52-channel NIRS system with 695 and 830 nm detectors (ETG-4100, Hitachi Medical Corporation, Tokyo, Japan). Relative changes in oxy-Hb concentration ([oxy-Hb]) were measured according to the modified Beer-Lambert law. The 52-channel system consisted of 17 light emitters and 16 light detectors fixed with 3 ×11 thermoplastic shells (Figure 2A). This array can measure the cortical activation in the frontal and superior temporal cortices based on the international 10–20 system used in electroencephalography (Figure 2B). A channel was defined as the area between light emitters and light detectors (inter-optode distance of 30 mm) and signals were sampled at 10 Hz. The corresponding spatial information for each channel in Montreal Neurological Institute space was evaluated using NIRS_SPM (version 4.0) (27).

Previous studies have concluded that [oxy-Hb] is more strongly related to blood oxygenation level-dependent (BOLD) signals as measured by functional MRI (fMRI) than [deoxy-Hb] (28–30). Hence, the current study recorded [oxy-Hb] as a measure of cortical activation during the CFT and LFT. All NIRS data were analyzed using NIRS-SPM (https://www.nitrc.org/projects/nirs_spm/). The hemodynamic response function (HRF) and a Wavelet–minimum description length (MDL)-based detrending method were used to remove systemic noise and physiological variation (31). The false discovery rate (FDR) (32) was applied to correct for multiple comparisons (level set at P < 0.05). The visualized brain network was presented using BrainNet Viewer (33).

SPSS version 25.0 (IBM Corporation, USA) was used for all statistical analyses. Categorical variables were compared between groups by chi-square test and normally distributed continuous variables by Student's t-test. Verbal fluency task performance was compared between groups by the Mann-Whitney U-test. One-sample t-test was used to compare within-group task-related [oxy-Hb] vs. [oxy-Hb] = 0 (null hypothesis). Group differences in [oxy-Hb] were compared by covariance analysis with years of education as a covariate. The associations between [oxy-Hb] and PANSS scores were assessed by Spearman correlation tests. A P < 0.05 was considered statistically significant for all tests.

Forty-seven patients with schizophrenia (SP group) and 29 age-, sex-, and body mass index-matched HCs (HC group) completed the study. The demographic and clinical characteristics of all participants are presented in Table 1. There were no significant group differences in basic demographic factors, while years of education was significantly lower in the SP group.

The SP group generated significantly fewer words than HCs during both the CFT [(26.62 ± 6.53) vs. (41.17 ± 7.42)] and LFT [(17.00 ± 6.84) vs. (28.93 ± 8.65)] (Figure 3). In contrast, the difference in task scores (CFT performance minus LFT performance) was similar in both groups [(9.62 ± 7.04) vs. (12.24 ± 5.97)]. Notably, CFT and LFT performance scores still differed between groups when education was included as a covariate (CFT: F = 35.42, P <0.001; LFT: F = 14.98, P <0.001; VFT performance differences: F = 3.12, P = 0.082). However, we found no significant associations between VFT task performances and [oxy-Hb] changes after FDR correction (FDR P <0.05).

The SP group exhibited significant activation of 33 channels (2–4, 6, 8, 13, 14, 17–19, 23–25, 27–29, 34–41, and 44–52; t = 2.379–5.496; FDR P <0.05; 15 channels over the left hemisphere and 17 channels over the right hemisphere, excluding medial channel 37) during the Chinese CFT. In contrast, only six channels showed significant activation during the LFT (28, 29, 39, 40, 50, and 51; t = 3.066–4.600; FDR P <0.05), all of which were over the left hemisphere (Figure 4, top row).

Healthy controls exhibited significant activation of 44 channels during the CFT (all except 2, 20, 32, 33, 36, 37, 43, and 47; t = 2.112–7.101; FDR P <0.05; 24 channels over the left hemisphere and 19 channels over the right hemisphere, ch16 excluded). Healthy controls also showed significant activation of seven channels during the LFT (7, 8, 18, 28, 29, 39, and 50; t = 3.408–5.972; FDR P <0.05), all of which were over the left hemisphere (Figure 4, bottom row).

When years of education was included as a covariate, the SP group demonstrated significant hypoactivation relative to HCs at 7 channels during the CFT (1, 11, 18, 21, 22, 30, and 45; F = 4.047 −8.105; P = 0.006–0.048; Figure 5, upper row of images). However, there was no significant different with FDR set as 0.05. Similarly, during the LFT, when controlling for years of education, the SP group demonstrated significant hypoactivation relative to HCs at four channels (1, 7, 11, and 17) (F = 4.948–15.889; P = 0.000–0.030; Figure 5, lower row). After FDR correction, the significant different was only presented at channel 11 (FDR P <0.05).

Before the FDR correction, there were significantly positive correlations between activation strength and PANSS-Positive factor score during the CFT at channels 12 and 22 (over the right postcentral gyrus, PoCG; rho = 0.336 and 0.292, P <0.05). There were also significant positive correlations between activation strength during the CFT and PANSS-General psychopathology factor score at channel 23 (the right precentral gyrus, PreCG; rho = 0.413, P <0.01), channel 33 (right PoCG; rho = 0.373, P <0.01) as well as at channels 32 and 44 (the right superior temporal gyrus, STG; rho = 0.308 and 0.344, respectively, both P <0.05). However, these correlation coefficients were no longer significant after FDR correction (FDR P <0.05).

During the LFT, before the FDR correction, there were significantly positive correlations between activation strength and PANSS-Positive factor score at channel 1 (right PoCG) and channel 52 (left middle temporal gyrus, MTG) (rho = 0.327 and 0.340, respectively, P <0.05). In addition, there were significant negative correlations between activation strength and PANSS-Negative factor score at ten channels: channels 7, 18, and 24 (bilateral middle frontal gyrus, MFG), channels 29, 40, 41, 50, and 51 (all over the left inferior frontal gyrus, IFG), channel 11 (right supramarginal gyrus, SMG), and channel 31 (right PoCG) (rho = −0.293 to −0.364, all P <0.05). Again, however, these correlation coefficients were no longer significant after FDR correction (FDR P <0.05).

Patients with schizophrenia produced fewer words than controls during both the CFT and LFT, in accord with previous studies reporting pervasive language processing deficits in schizophrenia (34, 35). Moreover, both groups generated more words during the CFT than the LFT, indicating that the inherently superior semantic fluency observed in healthy individuals (36) was preserved in this patient cohort. Indeed, the difference in task score (CFT performance minus LFT performance) was not significantly different between the two groups, indicating that lower phonemic fluency or semantic fluency was not disproportionately greater in patients. This result is inconsistent with several studies reporting either a greater semantic deficit (11, 37–39) or phoneme deficit (9, 14) in patients with schizophrenia. However, most of these studies used English or Japanese VFTs, while our results are in line with a study of first-episode Chinese-speaking patients by Chou et al. (22). Thus, Chinese patients may show unique VFT performance deficits. A previous study (23) using fNIRS to examine cortical activation patterns during Chinese letter and semantic tasks also reported greater word production in healthy controls than patients with schizophrenia, but the authors did not compare performance between tasks, while other studies of Chinese schizophrenics used only one type of VFT (6, 12). Furthermore, years of education was not controlled in statistical analysis despite a potential impact on VFT performance. A study of Spanish patients with severe psychiatric disorders (40) found that CFT performance was less strongly influenced by years of education than LFT. Collectively, these results suggest that the CFT is superior to the LFT for evaluation of psychiatric disorders in Chinese speakers.

In both groups, cortical activation ([oxy-Hb]) was higher and more extensive during the CFT than the LFT, in accordance with an fMRI study conducted in Chinese-speaking health controls (41). This may be explained by some level of phoneme analysis even when retrieving semantic information. Consistent with this notion, the aforementioned study (41) found activation of regions associated with phonemic fluency during the CFT but no significant activation of regions associated with semantic fluency during the LFT.

During the CFT, both patients with schizophrenia and controls exhibited significant activation of the bilateral frontal region and right superior temporal cortex, but left hemisphere activation was relatively more extensive and right frontotemporal activation more limited in healthy controls (i.e., controls showed greater left lateralization). In contrast, this group difference was not observed during the LFT, with both groups showing activation within overlapping regions of the left hemisphere, primarily in left ventrolateral prefrontal cortex. Similarly, a study of healthy individuals (42) also reported greater left hemispheric asymmetry of activation during a letter-cued task compared to a category-cued task. Thus, left lateralization was maintained in patients during the LFT, indicating that the LFT has less capacity to reveal cortical activation markers for schizophrenia.

These findings are in accord with a large number of neuroimaging studies reporting reduced left hemispheric laterality for language in schizophrenia patients (17, 43–46) compared to healthy populations (42, 47). However, Angrilli et al. (45) reported that reduced lateralization in Western patients with schizophrenia was specific for the phonological component of language. Therefore, lateralization of semantic and phonological processing between healthy controls and patients with schizophrenia may also differ between readers of Western and Eastern language script, possible due to the difference in semantic content.

Make mention of CFT, the schizophrenia patients exhibited no significant different reduced activation compared to controls after FDR correction, which was consistent with a previous fNIRS study (48) focusing on first episode schizophrenia and also used both VFTs as activation task but at odds with previous psychometric studies indicating that patients with schizophrenia have more severe semantic fluency deficits (7, 11, 39, 49–51). Similarly, with regard to LFT, the schizophrenia group showed significantly reduced activation in the right SMG (channel 11) compared to healthy controls. Chou et al. (48) suggested that LFT may be a more sensitive indicator of frontal dysfunction in schizophrenia than the CFT. However, when take the VFTs performance into consideration, the situation is vague. In accord with previous fNIRS studies (14, 15, 48, 50), we also found more word production during CFT than LFT in both groups. However, these studies indeed reported greater [oxy-Hb] activities during LFT than CFT. Remarkably, direct comparisons with previous neuroimaging studies of schizophrenia patients during VFTs may be misleading due to differences in task design, methodology, and study population. For instance, Quan et al. (12) found significantly lower [oxy-Hb] increases at 41 channels among patients during a Chinese-LFT compared to controls even after FDR correction. However, the study cohort was composed of chronic schizophrenia patients. In general, future studies lead to understanding the neural mechanisms of verbal fluency in Chinese schizophrenia patients were needed.

In the present study, we found no significant correlations between [oxy-Hb] changes and PANSS scores during either the CFT or LFT, possibly because the patients recruited were in a relatively stable disease stage. A previous fMRI study (17) reported that positive symptoms were associated with decreased left lateralization, suggesting that a relative increase in right hemispheric activity may predict psychosis. Similarly, an event-related potential study (45) and a previous NIRS study (52) both using phonological tasks found that decreased left lateralization was associated with positive symptoms. Moreover, a meta-analysis (53) concluded that reduced language lateralization could be a strong trait marker for auditory hallucinations (a positive symptom) among patients with schizophrenia. In contrast, Chou et al. (54) reported significant negative relationships between PANSS negative scores and activity in bilateral IFG and temporal regions during an LFT among patients with schizophrenia. They proposed that reduced gray matter volume in the frontal and temporal regions may contribute to the negative symptoms of schizophrenia. Consist with our results, Marumo et al. (15) also found no significant relationship between total PANSS score and [oxy-Hb] during either the CFT or LFT. However, the task design and statistical analysis methods differed from ours. Alternatively, Hori et al. (55) proposed that the association between functional laterality and clinical symptoms is strongly dependent on the activation task, so differences in task design may also contribute. Thus, potential correlations between [oxy-Hb] changes and clinical symptom severity are still uncertain. Larger scale studies recruiting participants at different clinical stages are needed to address this uncertainty.

This study has several limitations. First, the sample size was small, so other associations between cortical activity and schizophrenia symptoms may have been missed. Also, patients with schizophrenia show highly heterogeneous symptom expression, but the small sample precluded subgroup analysis. Second, all patients were currently taking antipsychotic drugs, which may have introduced further heterogeneity. However, we included only patients taking atypical antipsychotics based on evidence that these agents are more effective at maintaining cognitive function (56). In addition, a previous fNIRS study (57) reported that patients treated with atypical antipsychotics showed greater [oxy-Hb] during a VFT that patients treated with typical antipsychotics. Thus, the reduced cortical activation observed in schizophrenics may be unrelated to medication use. Moreover, we recruited schizophrenia patients at an early stage of the illness, and both task-dependent activation patterns and associations between these patterns and clinical parameters may differ from chronic patients (58). Last, all patients included in this study were relatively stable inpatients of a psychiatric hospital. Larger-scale studies including Chinese at different disease stages, under different treatment regimens, and from multiple centers are warranted to assess the general applicability of these fNIRS findings for patient evaluation.