Aberrant salience is selectively associated with ventral midbrain connectivity to the dorsal but not ventral striatum

Abstract

The dorsal midbrain-striatal dopaminergic stream encodes salience in animals, while it’s unknown in humans. The midbrain-striatal dopaminergic hyperactivity is postulated to attribute aberrantly-heightened salience to ordinary stimuli, forming delusion and hallucination. While animal studies implicate the ventral midbrain-striatal stream, patient studies suggest the dorsal stream, leaving the precise localization of aberrant salience unclear. We investigated these issues by examining midbrain-striatal functional connectivity. Resting-state functional MRI was acquired from 116 patients with schizophrenia (SCZ) and 224 age- and sex-matched healthy controls (HC). Subjective aberrant salience was assessed using the Aberrant Salience Inventory (ASI). Functional connectivity was computed between six striatal seeds (bilateral ventral limbic, dorsal executive, and dorsal sensorimotor striata) and the dopaminergic midbrain voxels (ventral tegmental area and substantia nigra). We tested (1) the presence of a ventral-to-dorsal connectivity gradient along anterior-to-posterior and inferior-to-superior axes; (2) the main effects of diagnosis and ASI and their interaction on connectivity; and (3) their associations with psychotic symptoms. Both anterior-to-posterior and inferior-to-superior connectivity gradients were confirmed. A significant main effect of diagnosis was observed in connectivity between the left anterior midbrain and ventral limbic striatum, which was reduced in SCZ than in HC. A significant diagnosis-by-ASI sensory subscale interaction was found for connectivity between the left anteroinferior midbrain and the dorsal sensorimotor striatum. This ASI subscale was significantly correlated with positive symptoms (all P < 0.05 family-wise error fully corrected). These findings demonstrate the ventral-to-dorsal midbrain-striatal gradients in humans and indicate a more complex localization of aberrant salience than previously assumed.

Introduction

Salience refers to whether something is particularly noticeable or important. Animals, including humans, automatically allocate their attention to salient stimuli in the surrounding world, which helps them survive. Previous animal studies have indicated that dorsal midbrain–striatal dopaminergic neurons encode salience^1–3. To date, however, few studies have investigated the presence of such localization in the human midbrain–striatum^4,5.

Psychosis is a psychiatric condition that presents symptoms such as delusions and hallucinations, and schizophrenia is the most common psychotic disorder. Schizophrenia significantly impairs the lives of patients and constitutes a major global burden, ranking 18th in years lived with disability (YLDs) according to the World Health Organization’s 2021 report⁶. The precise mechanisms of psychosis are still unclear, but the aberrant salience hypothesis is one of the most prominent pathophysiological hypotheses, which proposes that the elevated dopaminergic activity in the striatum of patients causes excessive salience attribution to daily-life stimuli, leading to formation of psychotic symptoms⁷. Animal studies on aberrant salience have been centered on the ventral stream of the dopaminergic midbrain–striatum from the ventral tegmental area (VTA) to the ventral striatum, i.e., the nucleus accumbens^8–10. Conversely, a recent meta-analysis of human positron-emission tomography (PET) studies revealed elevated dopamine levels in the dorsal associative and sensorimotor striata of patients with schizophrenia compared to healthy controls but not in the ventral limbic striatum¹¹. Thus, which of the ventral or dorsal streams of the midbrain–striatum is relevant to aberrant salience in psychosis is unclear.

Two anatomical aspects are critical in answering to above questions: first, unlike the well-documented parallel organization of the corticostriatal pathway¹², the dopaminergic midbrain–striatal pathway is not only parallelly connected with each other but includes overlapping connections over the ventral-to-dorsal axis¹³. Second, the brains of quadrupedal animals are rotated relative to the brains of bipedal humans, which causes uncertainty in the inter-species correspondence of “ventral” and “dorsal” labels between animals and humans^14,15.

To answer our questions, we aimed to combine functional connectivity (FC) analysis between the dopaminergic midbrain and striatum using resting-state functional magnetic resonance imaging (rs-fMRI) data and subjective aberrant salience experience measurements for a large number of patients with schizophrenia (SCZ) and healthy controls (HC). We intensively denoised rs-fMRI data, harmonized scanner effects, and utilized VTA and substantia nigra (SN) masks optimized for East Asian brain anatomy to increase the reliability of the results. We investigated the ventral-to-dorsal axis in human midbrain–striatal FC in two orthogonal axes: one from the inferior to the superior and the other from the anterior to the posterior. We hypothesized that (1) there would be a canonical ventral-to-dorsal FC gradient (i.e., the ventral midbrain preferentially connects to the ventral striatum, whereas the dorsal midbrain preferentially connects to the dorsal striatum) between the dopaminergic midbrain and striatum, and (2) connectivity in the dorsal midbrain–striatal stream would be associated with the subjective aberrant salience experience in SCZ.

Methods

Participants

One hundred and sixteen SCZ and 224 HC, matched for age, sex, and educational history, participated in the study. The premorbid intelligence quotient (IQ) was estimated using the Japanese version of the National Adult Reading Test short form (JART-25)¹⁶. Handedness was measured by the Edinburgh Inventory¹⁷. SCZ were recruited at the Department of Psychiatry, Kyoto University Hospital in Japan, and HC were recruited from the local community in the Kansai region of Japan. All patients met the criteria for schizophrenia (N = 112) or schizoaffective disorder (N = 4) based on the Structural Clinical Interview for the DSM-IV and were receiving antipsychotic medications. Chlorpromazine equivalents were calculated according to the practice guidelines for the treatment of SCZ, as described previously^18,19. Information on medication history and duration of illness (DOI) was unavailable for two patients. For a detailed description of the participants, see the Supplementary Material.

This study was approved by the Ethics Committee of Kyoto University Graduate School and Faculty of Medicine and conducted in accordance with the Declaration of Helsinki. Prior to participation, each individual was provided with a comprehensive explanation of the study, and written informed consent was obtained.

Clinical and psychological assessments

The Positive and Negative Syndrome Scale (PANSS) was used to assess the symptom severity of patients²⁰. The Aberrant Salience Inventory (ASI) is a 29-item questionnaire with yes/no responses, designed to capture the individual’s subjective experiences of aberrant salience. Total scores range from zero to 29, with higher scores indicating greater levels of aberrant salience²¹. We structured our ASI analysis according to a three-subscale model, which was proposed and validated in a recent mega-analysis²². This framework operationalizes three distinct domains of salience that we hypothesized would map onto the limbic, executive, and sensorimotor striatal subdivisions, respectively²³:

1. Motivational salience: measured by the “Enhanced Interpretation and Emotionality” ASI subscale, with items such as “Do certain trivial things ever suddenly seem especially important or significant to you?”.

2. Cognitive salience: measured by the “Unveiling Experiences” subscale (e.g., “Do you ever go through periods where you feel especially religious or mystical?”).

3. Sensory salience: measured by the “The Sharpening of Senses” subscale (e.g., “Do your senses sometimes seem sharpened?”).

Because of a technical issue, Item 13 of the ASI, belonging to the “Unveiling Experiences” subscale, could not be obtained; therefore, the analyses were performed using the remaining 28 items.

MRI acquisition

MRI data were acquired with two 3-T MR scanners: Trio (N = 99) and Tim Trio (N = 241), with 8- and 32-channel phased-array head coils, respectively (both manufactured by Siemens Healthineers, Erlangen, Germany).

Imaging protocols for resting-state fMRI

All participants were instructed to keep their eyes open (fixated), stay awake, and not concentrate on anything. An echo-planar imaging (EPI) sequence was used for both scanners. The scanning parameters were as follows.

Trio: Repetition time (TR)/echo time (TE) = 2000/30 ms; field of view (FOV) = 256 × 192 mm; matrix = 64 × 48; slice thickness = 4.0 mm with no gap; flip angle = 90°; total volumes = 180.

Tim Trio: TR/TE = 2,500/30 ms; FOV = 212 × 212 mm; matrix = 64 × 64; slice thickness = 3.2 mm with a 0.8 mm gap; flip angle = 80°; total volumes = 240.

Imaging protocols for T1-weighted MRI

T1-weighted 3D magnetization prepared rapid gradient echo (3D-MPRAGE) sequences were used for both sessions. The scanning parameters were as follows.

Trio: TR/TE = 2000/3.4 ms; inversion time = 990 ms; FOV = 225 × 240 mm; 240 × 256 matrix; resolution = 0.9375 × 0.9375 × 1.0 mm³; 208 total axial sections without intersection gaps.

Tim Trio: TR/TE = 2000/4.38 ms; inversion time = 990 ms; FOV = 240 × 240 mm; 256 × 256 matrix; resolution = 0.9375 × 0.9375 × 1.0 mm³; 208 total axial sections without intersection gaps.

Imaging protocols for B0 map

Trio: TR = 511 ms; TE = 5.19 ms and 7.65 ms; flip angle = 60°; FOV = 192 × 192 mm matrix, resolution = 3.0 × 3.0 × 3.0 mm³; 46 total axial slices without gaps.

Tim Trio: TR = 488 ms; TE = 4.92 ms and 7.65 ms; flip angle = 60°; FOV = 212 × 212 mm matrix, resolution = 3.3125 × 3.3125 × 3.2 mm³; 40 total axial slices without gaps.

rs-fMRI data preprocessing

To minimize non-neural signals, including artifacts related to head movements, physiological sources (e.g., pulsation and respiration), susceptibility-induced distortions, and scanner-generated noise, rs-fMRI data were preprocessed using an advanced pipeline incorporating FSL (version 6.0.3, http://www.fmrib.ox.ac.uk/fsl), FreeSurfer (version 7.4.1, http://surfer.nmr.mgh.harvard.edu/), and ICA-FIX²⁴. Preprocessing steps were scripted using a custom shell pipeline and proceeded as follows:

1. Motion correction: Head motion was corrected using MCFLIRT, and six motion parameters (translation and rotation) and framewise displacement metrics were extracted.

2. Bias-field correction of EPI images: A single EPI volume (first frame) was extracted and processed using fsl_anat in the T2-weighted mode. The resulting bias field was applied to the entire 4D EPI and the corresponding magnitude image of the B0 map using fslmaths.

3. Structural preprocessing of the T1-weighted anatomical images, including bias-field correction and segmentation, was performed using fsl_anat.

4. Skull-stripping with SynthStrip: The following three image types (bias-field corrected as above) were skull-stripped using SynthStrip in FreeSurfer.

1. T1-weighted image

2. EPI image (first volume)

3. Magnitude image (used for fieldmap correction)

Brain-extracted images and binary masks were retained for downstream fieldmap preparation and normalization.

• 5.Fieldmap preparation: Phase and magnitude images of the B0 map were processed with fsl_prepare_fieldmap (SIEMENS mode, ∆TE = 2.46 ms) to compute voxel-wise fieldmaps for EPI distortion correction.
• 6.First-level preprocessing and independent component analysis (ICA)-based denoising: First-level preprocessing was performed using FEAT in FSL. MELODIC was used for the ICA decomposition, followed by denoising using FIX (classifier: WhII_Standard; threshold = 20).

  The cleaned fMRI and structural images were stored for further processing in CONN (version 22a, Whitfield et al., 2012, Nieto-Castanon et al., 2022, https://web.conn-toolbox.org/) on SPM (version 12, https://www.fil.ion.ucl.ac.uk/spm/software/spm12/)²⁵ as follows²⁶:

• 7.Outlier detection: Potential outlier scans were identified using ART²⁷ as acquisitions with framewise displacement above 0.9 mm or global blood oxygen level–dependent (BOLD) signal changes above 5 standard deviations²⁸, and a reference BOLD image was computed for each subject by averaging all scans, excluding outliers.
• 8.Normalization and segmentation: Functional and anatomical data were normalized to standard Montreal Neurological Institute (MNI) space, segmented into gray matter, white matter, and cerebrospinal fluid (CSF) tissue classes, and resampled to 2 mm isotropic voxels following a direct normalization procedure²⁹ using the SPM unified segmentation and normalization algorithm^30,31 with the default IXI-549 tissue probability map template.
• 9.Smoothing: Considering the small size of the VTA and SN, the functional data were smoothed using spatial convolution with a Gaussian kernel of 2 mm full width at half maximum (FWHM) as previously described^32,33.
• 10.Further denoising by regression: Functional data were further denoised by regressing the potential confounding effects of white matter (five principal component analysis-based noise components of CompCor), CSF (five CompCor noise components), motion parameters and their first-order derivatives (12 factors)³⁴, outlier scans (below 42 factors)²⁸, session effects and their first-order derivatives (two factors), and linear trends (two factors) within each functional run.
• 11.Temporal filtering: A bandpass frequency filtering between 0.01 Hz and 0.1 Hz was applied to the above-preprocessed timeseries³⁵.

Global signal regression was not performed in our preprocessing pipeline. This decision was made to avoid mathematically introducing artificial negative correlations, which could confound the interpretation of our primary outcome, namely the midbrain-striatal functional connectivity gradients³⁶.

Seed-based FC calculation

The mean time series of the preprocessed fMRI data were extracted from six subdivisions of the striatum (limbic, executive, and sensorimotor in left and right) by applying the Oxford-GSK-Imanova Striatal Connectivity Atlas³⁷. Seed-based FC was calculated using the six time-series data as seeds and each voxel time series within the dopaminergic midbrain mask (combining the VTA and SN masks) as targets. We used VTA and SN masks optimized for the East Asian brain³⁸. Functional connectivity strength was Fisher-transformed to Z values, resulting in six FC maps for each participant.

Harmonization

To substantially minimize scanner-related variability in multi-scanner MRI data^39,40, ComBat was utilized to harmonize the data by adjusting for additive and multiplicative scanner effects after computing the target FC^41–43. Data quality was assessed by calculating the temporal signal-to-noise ratio (TSNR) for each voxel, defined as the mean divided by the standard deviation of its time series. The average TSNR across all voxels was computed using a quality assurance script⁴⁴. Age, sex, and TSNR were included as covariates when performing the ComBat.

Statistical analyses

Demographic and clinical characteristics

Between-group differences in continuous variables (age, education, premorbid IQ, ASI scores, and TSNR) were assessed using independent sample t-tests, whereas categorical variables (sex, handedness, and smoking status) were compared using chi-squared tests in SPSS (version 25.0; IBM, Armonk, NY). The statistical threshold was P < 0.05, and Bonferroni correction over family-wise error (FWE) was applied to the ASI subscale and total scores.

For variables only available in SCZ, descriptive statistics were reported.

Testing of ventral-to-dorsal canonical connectivity gradient in the midbrain–striatum across participants

We tested the ventral-to-dorsal connectivity gradient in two ways.

Standard second-level FC analysis

First, we performed one-sample t-tests on the six above-mentioned first-level connectivity maps within the dopaminergic midbrain mask across all participants to test whether the ventral limbic, dorsal executive, and dorsal sensorimotor striata were significantly connected with the ventral, middle, and dorsal midbrain areas, respectively. Statistical significance was defined as FWE-corrected P < 0.05, using the PALM toolbox of FSL with 10,000 permutations, correcting over voxels (threshold-free cluster enhancement [TFCE]), positive and negative contrasts, and six seeds⁴⁵. Age, sex, and TSNR were treated as confounding variables.

Inferior/anterior bias analyses

Second, to quantify the organization of dopaminergic midbrain–striatal connectivity, we defined and computed “inferior and anterior biases” for each striatal subdivision. We divided the midbrain mask into inferior/superior and anterior/posterior portions based on the median Z-coordinate (inferior/superior) and Y-coordinate (anterior/posterior) in MNI space. For each participant, we calculated the proportion of voxels showing positive FC (Z > 0) within the midbrain masks. The inferior bias was defined as the number of positive voxels in the inferior midbrain mask divided by the total number of voxels across the mask.

$$\mathrm{Inferior}\mathrm{bias}=\mathrm{N\_inferior}/(\mathrm{N\_inferior}+\mathrm{N\_superior})$$

Similarly, the anterior bias was then calculated using the anterior and posterior masks:

$$\mathrm{Anterior}\mathrm{bias}=\mathrm{N\_anterior}/(\mathrm{N\_anterior}+\mathrm{N\_posterior})$$

To formally test whether these biases followed the topographic gradients reported in animal studies (ventral-to-ventral and dorsal-to-dorsal preferential connections between the midbrain and striatum)^2,3,13,46, we performed a beta-regression analysis (using the betareg package in R) which was considered to provide more accurate estimates than traditional linear models for such metrics^47,48. The total number of positively connected (Z > 0) voxels with the entire midbrain for each participant was used as weights in the model. Statistical significance was set at P < 0.05, with a Bonferroni correction for multiple comparisons across the three striatal subdivisions in the left and right hemispheres. Age, sex, and TSNR were considered confounding factors.

Effects of diagnosis and ASI and their interaction on midbrain–striatal FC

We investigated the main effects of diagnosis and ASI and their interaction on midbrain–striatal FC using a general linear model (GLM) framework. Interaction and main effect analyses of the ASI were conducted on a subset of participants for whom ASI scores were available (SCZ = 38, HC = 74). The statistical threshold was P < 0.05, FWE-corrected for voxels (TFCE), two contrasts, and six striatal subdivisions. Age, sex, and TSNR were treated as confounding variables.

To unpack any effect of ASI on the psychopathology of SCZ, we first investigated the correlation between the ASI subscales (showing significance in the ASI main effect and ASI by diagnosis interaction analyses) and PANSS scores (positive, negative, and general psychopathology scales and total scores). We also extracted mean connectivity values from the significant clusters in the above ASI-related main effect and interaction analyses. Within the SCZ group, we only examined their correlations with PANSS scores. Spearman’s rank-order correlation was used, with age, sex, and TSNR included as covariates. Multiple comparisons were Bonferroni-corrected with a threshold of P < 0.05 in R.

Furthermore, to quantitatively evaluate the unique variance explained by the significant ASI subscale beyond that explained by general schizophrenia psychopathology, we performed hierarchical multiple regression analyses within the patient group (N = 38)^49,50. To avoid potential multicollinearity arising from simultaneously entering highly correlated clinical scales into the same model, separate two-step models were constructed while controlling for each significant PANSS subscale. In Step 1 of each model, the baseline covariates (age, sex, and TSNR) and the respective PANSS symptom scores were entered. In Step 2, the significant ASI subscale was added to the model. Statistical significance for the incremental change in explained variance ($\Delta R^2$) was defined as t P < 0.05 after Bonferroni correction in R.

Effects of demographic and clinical variables on connectivity

We examined the effects of educational history, premorbid IQ, smoking status, medication, and DOI on the significant findings reported above. Antipsychotic doses were converted to chlorpromazine equivalents according to the guidelines^18,19,51. For educational history, premorbid IQ, and smoking status, we extracted mean values from the significant clusters, added each of the three variables as a covariate in the GLM used in the analyses above, and tested the main effects and interactions. For medication and DOI, correlational analyses between these variables and mean cluster values were performed only for patients with age, sex, and TSNR as covariates. PALM was used in these analyses at P < 0.05, and FWE was corrected for the number of contrasts and significant seeds.

Sensitivity analysis

To evaluate the robustness and anatomical validity of our findings, we conducted sensitivity analyses using different spatial smoothing parameters (0 mm and 4 mm FWHM) and repeated the above FC analyses, inferior/anterior bias analyses, main effects of diagnosis and the ASI scales, and their interaction. Further details are provided in the Supplementary Materials.

Results

Demographic and clinical characteristics of participants

As shown in Table 1, we recruited 116 patients with schizophrenia (SCZ) and 224 healthy controls (HC). They did not statistically differ in age, sex, handedness, education, and rs-fMRI quality indexed by temporal signal-to-noise ratio (TSNR), while premorbid IQ and smoking status showed significant differences. The three ASI subscale scores and total scores were significantly higher in SCZ than in HC.

Table 1.

Demographic and psychological data.

	SCZ (n = 116)	HC (n = 224)	Statistics	P-valuea
Age (years)	38.7 ± 9.6	36.5 ± 11.2	t = −1.91	0.06
Sex (female/male)	58/58	99/125	χ2 = 1.036	0.31
Handedness (Rt/Lt)	113/3	217/7	χ2 = 0.08	0.78
Educational history (years)b	14.6 ± 2.0	15.1 ± 2.3	t = 1.90	0.06
Premorbid IQ (JART)b	100.9 ± 7.9	106.1 ± 5.3	t = −7.268	<0.001
Smoking status (Y/N)	24/92	13/211	χ2 = 17.46	<0.001
DOI (years)b	13.9 ± 8.6	/	/	/
CPeqb (mg/day)	556.4 ± 419.4	/	/	/
PANSSc Positive	13.5 ± 5.0	/	/	/
Negative	15.4 ± 5.7	/	/	/
General	28.5 ± 8.9	/	/	/
Total	57.4 ± 17.4	/	/	/
ASId Interpretation	7.3 ± 3.5	4.0 ± 3.7	t = −4.45	<0.001
Unveiling	2.1 ± 1.9	0.94 ± 1.4	t = −3.60	<0.005
Sense	1.8 ± 1.5	0.9 ± 1.3	t = −3.53	<0.005
Total	13.1 ± 6.6	7.6 ± 9.5	t = −3.15	<0.001
TSNR	216.5 ± 36.8	220. 8 ± 46.7	t = −0.53	0.60

ASI Interpretation = enhanced interpretation and emotionality; Unveiling = unveiling experiences; Sense = sharpening of senses. ASI scores were compared with an independent t-test by Bonferroni correction for factors, adding age, sex as covariates. One item of the ASI was unavailable at the time of data collection.

SCZ patients with schizophrenia, HC healthy controls, Rt right, Lt left, JART Japanese adult reading test, Y/N yest/no, DOI duration of illness, CPeq chlorpromazine equivalent, PANSS Positive and Negative Syndrome Scale, ASI Aberrant Salience Inventory, TSNR temporal signal-to-noise ratio, SD standard deviation, SD standard deviation.

^aTwo-tailed.

^bData not available for 2 HC.

^cData not available for 3 SCZ.

^dData available for 38 SCZ and 74 HC.

Ventral-to-dorsal canonical connectivity gradient in the midbrain–striatum

In the standard second-level FC analysis, significant positive FC was observed from the ventral limbic striatum to the broad area of the dopaminergic midbrain bilaterally (P = 0.027 and 0.0001, Figs. 1A and 1B, respectively), and from the dorsal executive striatum to the scattered area of the left dopaminergic midbrain (P = 0.045, Fig. 1C) (all FWE-corrected P < 0.05 for TFCE, two contrasts, and six striatal subdivisions). No significant results were observed for the other striatal seeds.

Fig. 1. Significant midbrain-striatal FC across all participants.

One sample t-test was conducted separately for the limbic (orange), executive (yellow), and sensorimotor (white) striatal subdivisions in the left and right hemispheres. A The left ventral limbic striatum showed significant positive FC with a broad area in the dopaminergic midbrain. B The right ventral limbic striatum showed significant positive FC with a broad area in the dopaminergic midbrain. C The left dorsal executive striatum showed significant positive FC with a scattered area in the dopaminergic midbrain. All results were FWE-corrected at P < 0.05 for multiple comparisons across voxels (TFCE), two contrasts, and six striatal subregions. Str Striatum, VTA blue, SN green, significant cluster red.

In the inferior/anterior bias analyses, a significant linear gradient was identified bilaterally for inferior bias across striatal subdivisions, with the bias decreasing from inferior to superior direction (left: β = –0.011, P = 2.39e⁻¹⁶; right: β = −0.016, P < 2e⁻¹⁶ after Bonferroni correction; Fig. 2B), indicating that the ventral striatum is connected to the inferior more than superior midbrain. Similarly, anterior bias was observed with the bias decreasing from anterior to posterior direction (left: β = –0.032, P < 2e⁻¹⁶; right: β = −0.016, P < 2e⁻¹⁶ after Bonferroni correction; Fig. 2C), indicating that the ventral striatum is connected to the anterior more than posterior midbrain. These results revealed a ventral-to-dorsal connectivity gradient between the midbrain and the striatum.

Fig. 2. Inferior and anterior biases of midbrain–striatal FC across all participants.

A On a sagittal MNI slice, the midbrain mask (VTA blue, SN green) was split at the median Z (inferior vs. superior) and median Y (anterior vs. posterior). Midbrain voxels showing positive FC (Fisher’s Z > 0) with the striatal seeds were counted. Each dot represents an individual participant. Each box plot shows the median, interquartile range, and interquartile range across the striatal subdivisions. B The inferior bias was calculated as the proportion of positively connected voxels located in the inferior midbrain out of all midbrain voxels (inferior and superior). C The anterior bias was calculated as the proportion of positively connected voxels located in the anterior midbrain out of all midbrain voxels (anterior and posterior). All biases were computed separately for the limbic (red), executive (green), and sensorimotor (blue) striatal subdivisions in the left and right hemispheres. All results were Bonferroni corrected at P < 0.001 for two hemispheres, two inferior/anterior indices. Str, striatum, Lt, left, Rt, right.

Decreased ventral midbrain–ventral limbic striatal FC in SCZ

We found a significant main effect of diagnosis, with SCZ exhibiting significantly reduced FC than in HC from the left ventral limbic striatum to the left anteromedial dopaminergic midbrain (P = 0.011, β = 0.166, FWE-corrected for TFCE, contrasts, and striatal subdivisions; Fig. 3A), controlling for age, sex, and TSNR. No significant main effect of diagnosis was found in the right hemisphere, nor were there any significant correlations between FC and PANSS scores. No significant results were found for the main effects of ASI subscales.

Fig. 3. Main effect of diagnosis and interaction of diagnosis by ASI on the dopaminergic midbrain-striatal FC.

A The FC between the left dopaminergic midbrain and left ventral limbic striatum (orange, top) was decreased in SCZ compared with HC at P < 0.05 FWE-corrected for multiple comparisons across voxels (TFCE), two groups, two contrasts, and six striatal subregions (significant cluster: red, VTA mask: blue, SN mask: left). B (a) A significant interaction between diagnosis (schizophrenia vs. healthy controls) and the ASI Sharpening of Senses subscale was found on the FC between the left dorsal sensorimotor striatum (white, top) and the left anteroinferior midbrain at P < 0.05 FWE-corrected for multiple comparisons across voxels (TFCE), two groups, two contrasts, and six striatal subregions (significant cluster: red, VTA mask: blue, SN mask: left). B (b) The averaged FC values were extracted from the significant cluster. The significant interaction showed the positive correlation between left midbrain-sensorimotor striatal FC in HC but negative correlation in SCZ after Bonferroni correction for two groups, three ASI subscales (r = 0.33, P = 0.03; r = −0.49, P = 0.02, respectively). This crossover interaction indicated pathological decoupling in SCZ, in which the normal positive integration between sensory salience and midbrain-striatal coupling was disrupted and reversed. C The ASI Sharpening of Senses subscale was positively correlated with PANSS positive symptoms, general psychopathology scores controlling multiple comparisons across three domains, as well as with total scores in SCZ (r = 0.31, P = 0.02; r = 0.35, P = 0.01; r = 0.27, P = 0.02, respectively).

Diagnosis-by-sensory aberrant salience interaction on the ventral midbrain to the dorsal but not ventral striatal connectivity

A significant interaction between the diagnosis and the ASI subscale (Sharpening of Senses) was observed from the dorsal sensorimotor striatum to the anteroinferior dopaminergic midbrain (P = 0.049, β = 0.133, FWE-corrected for TFCE, contrasts, and six subdivisions; Fig. 3Ba). Post hoc analyses revealed a positive correlation for HC but a negative correlation for SCZ (HC: r = 0.33, P = 0.03; SCZ: r = −0.49, P = 0.02 after Bonferroni correction; Fig. 3Bb). This subscale was positively correlated with PANSS positive symptoms and general psychopathology scores after controlling for multiple comparisons of the three subscales (Fig. 3C, left and middle) and was also significantly correlated with the PANSS total score (Fig. 3C, right). However, the mean cluster value of significant FC did not correlate with PANSS scores.

In the hierarchical multiple regression analyses, scores on the ASI “Sharpening of Senses” subscale significantly improved model fit and remained robust predictors of FC after adjustment for PANSS positive, general psychopathology, and total scores, even after Bonferroni correction (controlling for PANSS positive score: $\beta =-0.441$, $P=0.021$, $\Delta R^2=0.169$ [partial $R^2=0.207$]; PANSS general psychopathology score: $\beta =-0.471$, P $=0.012$, $\Delta R^2=0.190$ [partial $R^2=0.230$]; and PANSS total score: $\beta =-0.477$, $P=0.009$, $\Delta R^2=0.198$ [partial $R^2=0.242$]).

No effects of medication, DOI, educational history, premorbid IQ, or smoking status

Post hoc analyses of the significant clusters revealed that they remained significant after controlling for educational history (main effect: P = 0.0001, interaction effect: P = 0.0189), premorbid IQ (main effect: P = 0.0001, interaction effect: P = 0.0184), and smoking status (main effect: P = 0.0001, interaction effect: P = 0.018) (all FWE-corrected). No significant correlations were found between the significant cluster values and either medication (main effect: P (positive/negative) = 0.9999/0.376; interaction effect: P (positive/negative) = 0.9999/0.5456) nor the DOI (main effect: P (positive/negative) = 0.9646/0.9999; interaction effect: P (positive/negative) = 0.9999/0.9914) (all FWE-corrected).

Sensitivity analyses

The analyses with smoothing at 0 and 4 mm showed findings similar to the main analyses above, suggesting the robustness of the significant results in this study. The details are provided in the Supplementary Material.

Discussion

In this study, we first delineated the functional topography of the human dopaminergic midbrain–striatal pathway using FC analysis and identified a ventral-to-dorsal canonical connectivity gradient across all participants. Second, we revealed that the FC of this pathway is altered in SCZ compared to HC, whose aberrant sensory salience experience was selectively linked to the connectivity between the “ventral” midbrain and the “dorsal” sensorimotor striatum in SCZ. Third, aberrant sensory salience is associated with the severity of psychotic symptoms. Here, we discuss these findings in detail.

Seed-based FC between the ventral limbic/dorsal executive striatum and dopaminergic midbrain

Our conventional seed-to-voxel analysis revealed significant positive functional connectivity between the dopaminergic midbrain and the ventral limbic and dorsal executive striatal subdivisions across all participants (Fig. 1). This finding is consistent with the canonical role of the VTA and ventromedial SN in reward-related behaviors and cognitive control^{13,38,52–55}. We did not identify significant positive or negative connectivity between the midbrain and dorsal sensorimotor striatum across all participants. This may be due to the functional heterogeneity within striatal circuits, where different striatal subregions have been reported to have distinct connections and behavioral roles^56–58.

Importantly, on visual inspection, the significant connectivity cluster in the midbrain from the left limbic striatum appeared to be more anterior (ventral) to that from the dorsal executive striatum, qualitatively supporting our hypothesis of a ventral-to-dorsal gradient in this pathway.

Ventral-to-dorsal canonical gradient in dopaminergic midbrain–striatal FC

Our beta-regression analyses identified significant linear decreases in the inferior and anterior biases in midbrain–striatal connectivity along both the inferior-to-superior and anterior-to-posterior axes, directly revealing the ventral-to-dorsal connectivity gradient across all participants (Fig. 2). These findings, for the first time, provide quantitative evidence for the nuanced organization previously detailed in animal studies^{1,13,52,53,55,59,60}. Future studies using high-resolution diffusion MRI tractography would clarify the anatomical gradient of this pathway in humans.

Main effect of diagnosis revealed mesolimbic hypoconnectivity in SCZ

The main effect analysis of the diagnosis revealed reduced FC between the left anterior (the ventral midbrain) and the left ventral limbic striatum in SCZ. This finding is highly consistent with a large body of previous rs-fMRI studies in patients with chronic schizophrenia, first-episode psychosis, and individuals at high clinical risk, suggesting a reproducible pattern of dysconnectivity across different illness stages^32,61–65. In our study, all patients underwent antipsychotic treatment, although no significant effect was found between FC and either chlorpromazine-equivalent dosage or illness duration, indicating that the observed ventral FC reduction was unlikely to be driven solely by medication or chronicity.

Ventral midbrain–dorsal sensorimotor striatal connectivity and aberrant sensory salience in SCZ

Based on the parallel organization of the striatal subdivisions and the domains of aberrant salience^22,66–68, we expected a distinct functional anatomical mapping of aberrant salience; that is, the motivational, cognitive, and sensory salience subscales would be associated with the ventral limbic, dorsal executive, and dorsal sensorimotor striata and their corresponding ventral-to-dorsal midbrain counterparts, respectively. However, we found a significant interaction between aberrant sensory salience and diagnosis with connectivity between the left anterior (the “ventral” dopaminergic midbrain, especially the lateral VTA and ventromedial SN) and the left “dorsal” sensorimotor striatum. While unexpected, this finding is anatomically plausible: animal tracer studies have previously shown that the lateral VTA and medial SN have widespread and overlapping projections that included connections to the dorsal sensorimotor striatum in rodents and non-human primates^13,69. Moreover, recent evidence that dopaminergic signals in the caudal sensorimotor striatum can bias sensory processing in rodents supports our interpretation that altered midbrain–sensorimotor connectivity may underlie aberrant salience⁵⁸. Crucially, our finding may help reconcile the apparent conflict in psychosis studies by bridging the animal findings that have historically emphasized the role of the ventral midbrain–striatum in dopamine dysregulation^9,13,69, and recent human PET studies that have revealed excess dopamine synthesis not in the ventral but in the dorsal striatum^66,70.

The significant interaction can be characterized as a positive correlation in HC and a negative correlation in SCZ. In healthy individuals, greater sensory salience was associated with stronger ventral midbrain–dorsal striatal coupling, whereas in SCZ, greater sensory salience was associated with weaker coupling, suggesting a reversal of normal salience-related integration. Our findings suggest that a dysregulated interaction between these two key regions may represent a critical mechanism underlying aberrant salience in SCZ. In HC, increased functional synchronization between the ventral midbrain and dorsal sensorimotor striatum may reflect a well-coordinated dopaminergic system that mediates the adaptive attribution of salience to behaviorally relevant stimuli. Consequently, healthy individuals exhibiting stronger functional coupling may demonstrate greater baseline salience sensitivity within a physiological range, in the absence of overt psychotic symptoms^11,54. In contrast, in SCZ, elevated aberrant sensory salience was associated with a shift toward negative midbrain–striatal coupling rather than an amplification of normal adaptive salience processing⁷¹. This finding may indicate pathological desynchronization of the dopaminergic salience circuit^7,11. Such impaired coordination may result in the unstable or inappropriate assignment of significance to otherwise neutral internal or external events^72,73. At the phenomenological level, this breakdown of the circuit may manifest as the subjective experience that ordinary perceptions feel unusually significant or personally meaningful, thereby facilitating the emergence of overt hallucinations and delusions⁷⁴. Likewise, internally generated mental events may acquire abnormal salience and be misattributed to external stimuli, disrupting the boundaries of self-experience^61,75. In summary, our findings suggest that aberrant salience in patients, particularly within the sensory/perceptual domain, arises not from simple hyperconnectivity or hypoconnectivity of the dopaminergic system relative to that in healthy individuals, but rather from a pathological disruption of its canonical, coordinated dynamics.

Additionally, the aberrant sensory salience measured by the ASI “Sharpening of Senses” subscale was positively correlated with PANSS positive symptoms, as well as general psychopathology symptoms and total scores, consistent with the aberrant salience hypothesis^7,76. In further exploratory analyses using hierarchical multiple regression, this ASI subscale remained a robust, independent predictor of midbrain-striatal FC after adjustment for PANSS positive, general psychopathology, and total scores, explaining an additional 20–24% of the variance. These findings suggest that the observed ventral midbrain-dorsal sensorimotor striatal dysconnectivity is more closely associated with aberrant salience than with general schizophrenia psychopathology. These findings are also in line with our previous study reporting involvement of the striatal aberrant salience in delusion formation⁷³.

It should also be noted that ventral midbrain–dorsal striatal FC was not directly correlated with PANSS positive symptom scores. Therefore, our findings should be interpreted as identifying a neural correlate of aberrant salience rather than a direct neural correlate of positive symptoms. One possible explanation for the lack of a direct association with positive symptoms is that the mechanisms underlying psychotic symptoms are multifactorial, as we propose⁷⁷. Specifically, although midbrain–striatal aberrant salience may contribute to the development of psychotic symptoms, their conviction and persistence may be mediated by distinct neural networks. Consequently, ventral midbrain–striatal FC alone may not be sufficient to explain the full spectrum of positive symptoms, at least in the present sample. Further studies investigating the contribution of each mechanism to psychotic symptoms are warranted.

Medication, DOIs, educational history, and smoking status did not affect the results, indicating that they were not merely the results of these confounding factors. In summary, our results elaborate and advance the 20-year-old aberrant salience hypothesis in terms of FC and the ventral-to-dorsal midbrain–striatal connectivity gradient⁷⁸.

Strengths and limitations

The key strengths of this study are as follows: (1) using a combined dataset paired with the subjective aberrant salience scale; (2) combining two different MRI datasets using ComBat harmonization to increase reproducibility; (3) substantial noise reduction by combining the latest denoising methods, including ICA-denoising; (4) utilizing the East Asian–optimized VTA and SN masks to increase the precision and accuracy of analyzing the small structure of the midbrain; and (5) conducting sensitivity analysis to increase the reproducibility of our findings.

This study has several limitations. First, all patients were treated with medication, and the effects of medication cannot be completely eliminated, even though we statistically controlled for the effect of medication^79–81. Second, our findings, based on BOLD signals, lack neurotransmitter specificity and likely reflect mixed dopaminergic and GABAergic activities, and should be interpreted with caution regarding pure dopaminergic contribution^82,83. Third, Item 13 of the ASI was unavailable for all participants because of a technical issue during data collection. However, this item is not part of the ASI “Sharpening of Senses” subscale; therefore, our significant findings related to this subscale were not affected by its absence. Furthermore, the total score based on the remaining 28 ASI items demonstrated excellent internal consistency (Cronbach’s α = 0.918), comparable to that reported in previous studies²², and the “Unveiling Experiences” subscale, to which Item 13 belongs, demonstrated good reliability (α = 0.747), comparable to that of the other two subscales (α = 0.868 and 0.740 for “Enhanced Interpretation and Emotionality” and “Sharpening of Senses”, respectively), indicating that the psychometric integrity of the scale was preserved. Fourth, ASI data were available only for a subset of the participants (38 patients with SCZ and 74 healthy controls), which may have reduced the statistical power of the interaction and correlation analyses relative to that of the group-difference analyses. Future studies including larger samples with completed subjective psychopathological assessments are needed to replicate these findings.

Conclusion

Our study identified a canonical ventral-to-dorsal midbrain–striatal FC gradient, reduced ventral midbrain–ventral striatal FC in SCZ, and ventral midbrain–dorsal but not ventral striatal FC associated with aberrant sensory salience. These findings help reconcile the discrepancy between animal and human studies and further advance the aberrant salience hypothesis. Extending these findings may facilitate the development of robust biomarkers of aberrant salience.

Author contributions

J.M. and L.Y. designed and conducted the study. L.Y. analyzed the data, with contributions to data analysis from J.M., N.O., D.Q., Y.N., and Y.L. L.Y. wrote the manuscript with feedback from N.O., D.Q., Y.L., T.K., Y.Y., M.K., Y.N., S.K., T.M., and J.M. L.Y., T.M., and J.M. contributed to reviewing and editing the manuscript. T.M. and J.M. supervised the research process. All authors contributed to the critical revision of the manuscript for important intellectual content.

Funding

This study was funded by: Japan Society for the Promotion of Science and Ministry of Education, Culture, Sports, Science and Technology KAKENHI (JP24H02167 and JP24K10708); Japan Agency for Medical Research and Development (JP21uk1024002); research grants from SENSHIN Medical Research Foundation (to J.M. and M.K.); Takeda Science Foundation (to M.K.); SRF Foundation (to M.K.).

Data availability

The unthresholded data are available upon request for reproducibility purposes (e.g., meta-analysis). Raw neuroimaging data, clinical data, and custom in-house code are not publicly available but may be provided upon reasonable request to the corresponding author, subject to scientific review and Material Transfer Agreements.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41537-026-00795-2.