Characterization of the extracellular free water signal in schizophrenia using multi-site diffusion MRI harmonization

Studies applying Free Water Imaging have consistently reported significant global increases in extracellular free water (FW) in populations of individuals with early psychosis. However, these published studies focused on homogenous clinical participant groups (e.g., only first episode or chronic), thereby limiting our understanding of the time course of free water elevations across illness stages. Moreover, the relationship between FW and duration of illness has yet to be directly tested. Leveraging our multi-site diffusion magnetic resonance imaging(dMRI) harmonization approach, we analyzed dMRI scans collected by 12 international sites from 441 healthy controls and 434 individuals diagnosed with schizophrenia-spectrum disorders at different illness stages and ages (15–58 years). We characterized the pattern of age-related FW changes by assessing whole brain white matter in individuals with schizophrenia and healthy controls. In individuals with schizophrenia, average whole brain FW was higher than in controls across all ages, with the greatest FW values observed from 15 to 23 years (effect size range = [0.70–0.87]). Following this peak, FW exhibited a monotonic decrease until reaching a minima at the age of 39 years. After 39 years, an attenuated monotonic increase in FW was observed, but with markedly smaller effect sizes when compared to younger patients (effect size range = [0.32–0.43]). Importantly, FW was found to be negatively associated with duration of illness in schizophrenia (p = 0.006), independent of the effects of other clinical and demographic data. In summary, our study finds in a large, age-diverse sample that participants with schizophrenia with a shorter duration of illness showed higher FW values compared to participants with more prolonged illness. Our findings provide further evidence that elevations in the FW are present in individuals with schizophrenia, with the greatest differences in the FW being observed in those at the early stages of the disorder, which might suggest acute extracellular processes.


INTRODUCTION
Diffusion magnetic resonance imaging (dMRI) has proven to be a powerful tool for the in vivo interrogation of tissue microstructure in the brain.In particular, the application of dMRI approaches to the study of psychosis has contributed significantly toward a greater understanding of the biological bases of the illness [1][2][3].These studies, which historically employed the diffusion tensor imaging (DTI) model and its associated metrics, specifically fractional anisotropy (FA), have consistently reported global alterations in white matter FA at all stages of the illness [1][2][3][4][5].Many of these reports have interpreted the observed changes in FA to be indicative of a central role for white matter in the development of psychosis [1][2][3][4][5].
However, it is important to acknowledge that the dMRI signal is influenced by a variety of underlying biological constructs, ranging from inherent physiological differences such as myelination, axonal packing, and crossing fibers, to more pathological features like edema, demyelination, or axonal degeneration [6,7].
While FA remains the most used dMRI measure in the field of clinical neuroimaging, the development of increasingly sophisticated dMRI models offers an avenue for the more refined investigation of the underlying biological elements that may contribute to changes in the dMRI signal.One such example is Free Water Imaging, an approach that deconstructs the dMRI signal into two compartments: an isotropic compartment measuring the relative amount of unrestricted water diffusion in the extracellular space (Free Water, FW) and a second compartment measuring the extent and orientation of water diffusion in or near cellular processes (quantified by Fractional Anisotropy of the Tissue, FAt) [8].
In the first application of Free Water Imaging in schizophrenia, Pasternak et al. identified significantly increased FW in both white and gray matter of recent-onset psychosis subjects [9].More importantly, this initial report also showed that these FW increases exhibited high spatial overlap with reductions in FA [9].Since then, four additional studies have identified increased FW in independent early psychosis cohorts [10][11][12][13], thereby calling into question whether the biological changes taking place in the early stages of illness are driven by structural changes in white matter, as previously assumed by earlier DTI studies [1,[3][4][5].All five published studies have observed FW increases in cerebral white matter [9][10][11][12][13], with two studies also observing FW elevations in gray matter [9,11].Moreover, two of these studies, Guo et al. and Berge et al. [12,13], included longitudinal imaging follow-ups at 1 and 2 years, respectively, and showed that FW remains elevated after 1 year of illness [12], but begins to exhibit a gradual decline after 2 years of illness [13].Conversely to early psychosis populations, studies of individuals in the chronic stages of illness do not show large or widespread increases in FW [14][15][16][17][18]. Instead, populations of people with chronic schizophrenia tend to show an attenuated FW pathology, suggesting that FW elevations may be more associated with the events surrounding illness onset.
While previous work provides strong foundational evidence that elevated FW is present in the early course schizophrenia compared to more prolonged illness, it has yet to be directly tested.This is because most studies examined participant cohorts at specific illness stages (i.e., first-episode, chronic, etc.), thus limiting insights regarding FW changes across age and duration of illness.To overcome this challenge, our study aggregates dMRI data from 12 international datasets, resulting in a large agediverse population of individuals across a range of illness durations.However, scanner-induced biases in the dMRI signal limit the ability to directly pool multi-site dMRI data.Therefore, we applied a dMRI data harmonization technique developed by our group to remove site-, sequence-, and scanner-specific effects from the dMRI data while preserving inter-subject biological variability [19].We have successfully validated and applied our harmonization framework in multiple studies [5,[20][21][22][23][24][25][26], including a recent study using FA, which revealed patterns of age-related white matter pathology in individuals with schizophrenia [5].Here, we extend this work by applying Free Water Imaging to determine the contribution of extracellular FW changes to these previously reported FA changes across age and duration of illness in schizophrenia [5].Based on previous studies, we hypothesized that FW values would relate to the duration of illness independent of age effects, such that participants with shorter illness durations will display higher FW values than those with prolonged illness.

MATERIALS AND METHODS Participants
This study includes extant dMRI data collected from 12 international sites.Each location collected site-specific schizophrenia participant populations and healthy controls with similar distributions of age and sex.Imaging and clinical data from this harmonized cohort have been previously published [5,[20][21][22].A detailed description of the data collection, quality control procedures, and exclusion criteria performed to attain the final sample for the present study can be found in the Supplementary.Our final sample included a total of 875 participants; 434 individuals were diagnosed with a schizophrenia-spectrum disorder: 148 females and 286 males, age range between 15 to 58 years with mean ± std: 31.87 ± 11.25 years and duration of illness is 10.07 ± 10.22 years.This study also included data from 441 healthy controls: 191 females, 250 males, age range between 15 to 58 years with a mean ± std: 31.18 ± 12.68 years.All dMRI data were harmonized for joint analysis.See Table 1 for the description of sitespecific samples and acquisition protocols and Supplementary Table 1 for the other available clinical and demographic data.For the purposes of this study, we utilized the following standardized variables across 12 sites: symptom severity (SS), Chlorpromazine equivalent doses of antipsychotics (CPZ), IQ, and education (in years).The standardization was done as part of another study [20] and is briefly described in the Supplementary.
Duration of illness was calculated as the difference between age of onset and age at scan.We note that three of the datasets included in this sample have been utilized in other published FW studies (see Supplementary).

Image processing
A standardized minimal pre-processing procedure was carried out for each participant across the 12 sites as part of our earlier study [5].This involved axis alignment and centering using the Psychiatry Neuroimaging Laboratory pipeline: https://github.com/pnlbwh/pnlutil.Next, FSL's eddycurrent correction was applied to correct dMRI data for eddy-currents and motion artifacts [27].Brain masking was performed with the brain extraction tool(BET) [28].
All preprocessed dMRI data were harmonized using our rotation invariant spherical harmonics (RISH) approach [5], thereby removing nonlinear scanner and sequence differences present across the 12 sites.We note that this method has been selected as the best-performing approach in a harmonization community challenge [29].Our method is directly applied to the dMRI signal.Further, it can robustly harmonize dMRI data from multi-site studies with different b-values, spatial resolution, and the number of gradient directions [19].The details of the harmonization pipeline can be found in the Supplementary.
Using the harmonized dMRI data, FA was estimated and registered to the standard FA template in MNI space using the Illinois Institute of Technology Atlas version 4.1 [30].FW was also computed from the harmonized dMRI data using the method described in [8] and then transformed to the standard template using the transformations calculated for FA.Finally, the average FW was computed for the whole brain white matter (as defined by the Illinois Institute of Technology Atlas) in MNI space.

Statistical analysis
Overlap in the distribution of age and duration of illness.The association between illness duration and schizophrenia participant age can be difficult to disentangle because they are assumed to be highly correlated.However, it is an important distinction to make, particularly in the case of late-onset illness (i.e., participants who are chronologically older yet have a shorter duration of illness) or early-onset illness (i.e., participants who are chronologically younger yet have a longer duration of illness).Since our sample includes both early-and late-onset participants, we made the assumption that age and duration of illness are jointly observed independent variables.To quantify the similarity between the age and duration of illness distributions, we applied two approaches: (1) Pearson's correlation coefficient; (2) Bhattacharyya coefficient [31,32].Pearson's correlation can provide a complete description of the association when the variables are normally distributed.On the other hand, Bhattacharyya coefficient [31,32] is a distribution-free overlapping index that can robustly measure the similarity between the distributions of two independent variables; therefore, it is a preferred index for assessing non-normally distributed independent variables.The Bhattacharyya coefficient yields a 0 when there is no overlap between the two distributions and yields a 1 for complete overlap.
Age-related FW trajectories in schizophrenia and in healthy controls.The majority of previous studies utilizing the Free Water model in populations of participants with schizophrenia predominantly observed widespread and/or anatomically diffuse FW pathologies [9,10,[12][13][14][15][16].As such, we focused our analysis on characterizing the trajectory of averaged whole brain FW values in individuals with schizophrenia and healthy controls as opposed to evaluating individual tracts.After running goodness-of-fit tests on multiple parametric (Poisson, gamma, linear, cubic) and non-parametric models (smoothing splines), the model with the highest adjusted r 2 was selected as the best-fitted model, resulting in a quadratic model.Therefore, age-related FW trajectories were modeled with quadratic regression as follows [33]: models the whole brain white matter FW value in schizophrenia, v 2 models the additional contribution of controls, v 3 models the linear relationship between age (centered around the age α) and FW in individuals with schizophrenia, v 4 models the additional linear relationship in controls, v 5 models the quadratic relationship between age and FW in individuals with schizophrenia, and v 6 models the additional contribution of the quadratic term in controls.dx is the categorical variable for diagnosis: controls (0), individuals with schizophrenia (1), κ represents sex (any interaction with age 2 , age, and dx was included in the model), and ϵ is the error term.We note that κ representing sex, IQ, and education in years was also tested separately with the limited data available.The regression model was independently fitted with age centered at 15 and 58 years in yearly increments (α).The v 2 term was used to determine if the differences between individuals with schizophrenia and controls are significant.v 4 and v 6 were used to assess if there were significant group differences in the association with age and age 2 , respectively.Finally, we computed the effect size (Cohen's d) of the between-group FW differences at each age to assess the magnitude of differences.To estimate the age of minima/ maxima and confidence interval (CI) at 97.5 and 2.5% quantiles, quadratic curve fitting was repeated using a bootstrapping procedure with 5000 iterations.
Age-related standardized (z-score) FW trajectories in schizophrenia and healthy controls.To investigate further age-related differences in FW values between individuals with schizophrenia and healthy controls, we used a sliding window approach to estimate the standardized FW values at Table 1.Demographic and diffusion MRI acquisition information for all sites, and the harmonized diffusion MRI data parameters (i.e., harmonized diffusion MRI data b-value = 1000 s/mm 2 and spatial resolution = 1.5 mm 3 ) that were selected based on the common b value across studies and optimal spatial resolution.

Name of the study/location
where i = [15, 16, 17, …, 58] years.N i = number of participants within the window of age, and ws is the window size.A window size of 5 years was applied to each age between 15 and 58 years, such that only participants within an age window contributed to standardized FW estimate.We note that the window size of 5 was selected heuristically.Supplementary Fig. 1 presents participant counts (N i ) for each age with a window size of 5 years, as well as window sizes of 3 and 7 for comparison.Next, using the mean and standard deviation trajectories computed from healthy controls, we standardized the FW values of each participant, including the individuals with schizophrenia, by normalizing the raw FW value: k refers to the index of any participant with age i.Finally, like the trajectory analysis on the raw FW values as described in section "Age-related FW trajectories in schizophrenia and in healthy controls", we modeled the zscore trajectories and computed Cohen's d effect sizes and significance levels to quantify between-group differences in standardized FW estimates at a given age.
Illness duration effects on whole brain white matter FW in schizophrenia.As described in the previous section, standardized FW values (z-scores) model the FW alterations in individuals with schizophrenia relative to the normalized FW trajectories of controls.Standardized FW values can also allow us to explore relationships between FW and schizophrenia-specific parameters (i.e., duration of illness) which are not present in controls.
Standardized FW estimates were modeled along duration of illness (doi) in the schizophrenia group using the linear mixed effects (lme) model.While lme allows us to ascertain illness duration effects on whole brain FW levels in schizophrenia, it also can account for the site effects in the data robustly (e.g., the potential doi variability across multiple sites): where sex, age, education, IQ, SS, and CPZ were included as fixed effects.
To reduce skewness in doi and transform the data as close as to the "normal" distribution, we applied log transformation.

Demographics
Our study examined harmonized dMRI data from a large group of individuals with schizophrenia and age-matched healthy controls (Table 1 and Supplementary Table 2).Supplementary Fig. 2 depicts the age distribution of the participants with schizophrenia and healthy controls, as well as the sex distribution as a function of age and duration of illness.Supplementary Fig. 3 depicts the distribution of duration of illness as a function of age.We observed a high positive correlation between age and duration of illness (Pearson's r = 0.8033).Meanwhile, the Bhattacharyya coefficient, which was used to quantify the overlap between age and duration of illness distributions, yielded an overlap of 0.61, indicative of a moderate overlap [31,32].As a result, we modeled FW changes over age and duration of illness separately in the subsequent steps of the analysis to disentangle their effects on whole brain white matter FW.

Age-related FW trajectories in schizophrenia and in healthy controls
The age 2 was found to be significant (p value of v 5 = 0.0013); thus, the quadratic regression model was considered to provide the most parsimonious explanation of the relationship between FW and age. Figure 1 (Left) shows trajectories of whole brain FW in Fig. 1 Age-related whole brain white matter FW trajectories.Quadratic regression was used to infer age effects on "Raw FW values (left figure)" and "Standardized FW values (Z-scores) (right figure)" from whole brain white matter.Means (solid lines), 95% confidence intervals (dashed lines) and standard deviations (shaded region) are represented by blue for individuals with schizophrenia and by orange for healthy controls.Z-scores were computed using sliding window analysis.See "Age-related FW trajectories in schizophrenia and in healthy controls" and "Age-related standardized (z-score) FW trajectories in schizophrenia and healthy controls" sections for more details of the analysis steps.
individuals with schizophrenia (blue line) and controls line) for the entire cohort plotted over the 43 year-period (between 15 and 58 years) using the quadratic model (refer to Supplementary Fig. 4 for the non-parametric model, which yielded similar outcomes).Overall, there was a significant FW difference between schizophrenia and control groups for several parameters in our model: v 2 p < =0.00013 (v 2 represents the differences between individuals with schizophrenia and controls); v 4 p ≤ 0.0018 and v 6 p ≤ 0.0023 (v 4 represents group differences in the linear association of age; v 6 represents group differences in the quadratic association of age).The model parameters for age association in controls (v 3 ) and (κ) for only sex, as well as (κ) for sex, IQ, and education, were not significant (p > 0.05).Due to the limited data availability of IQ and education, we only report the analysis results of (κ) for only sex.In individuals with schizophrenia, whole brain FW was higher compared to the controls across the entire age range.From 15 to 23 years old, FW elevations were most pronounced, but also followed a monotonically decaying trajectory until reaching a minima at the age of 39 ± 5 years.This was followed by a monotonic increase with a slower rate of change in FW in the later ages (39-58 years).On the other hand, across the entire age span (between 15 and 58 years), the whole brain FW in the controls showed a lower and relatively stable pattern compared to the individuals with schizophrenia.
Differences in standardized FW age-related trajectories between individuals with schizophrenia and healthy controls Figure 1 (Right) presents trajectories of standardized FW values in schizophrenia and controls (refer to Supplementary Figs. 5 and 6, respectively, for the non-parametric model and scatter plot of FW).
Standardized FW values followed similar trajectories as the raw FW values.The largest FW estimates in schizophrenia (z-score >1) were observed between ages 15 and 23 years, with statistical significance for between-group differences (v 2 ; p ≤ 0.00012).We also observed large effect sizes for between-group standardized FW differences in younger individuals with schizophrenia (15-23 years; effect size range = [0.700.87]; Fig. 2).However, with increasing age, effect sizes decreased monotonically until 39 years of age.After 39 years old, the effect size trajectory followed an increasing trajectory, yet with slower rates in the older individuals (39-58 years old effect size range = [0.320.43]) when compared to younger individuals (23-39 years old effect size range = [0.430.70]; Fig. 2).

Illness duration effects on whole brain whiter matter FW in schizophrenia
We examined the relationship between standardized FW values (zscores) and duration of illness, in individuals with schizophrenia.
The relationship between standardized FW estimates and duration of illness was significant (p = 0.005) and was negative and linear in nature.Age, sex, IQ, education, CPZ, and SS were not significant (p > 0.05).Hence, regardless of age and other variables, participants with schizophrenia with a shorter duration of illness showed higher FW values compared to participants with prolonged illness.

DISCUSSION
The present analysis leverages harmonized dMRI data to characterize whole-brain FW values across age and duration of illness in individuals with schizophrenia.We find that, when compared to age-matched controls, greater FW is observed in individuals with schizophrenia between 15 and 23 years of age.
Greater FW is also found in those individuals with schizophrenia who have a shorter duration of illness when compared to those with longer illness.These data suggest that the earlier stages of illness in schizophrenia may be more characterized by pathologies affecting the extracellular space.Our work directly aligns with previous reports and a recent meta-analysis [34] that show FW elevations in early-course schizophrenia populations [9,10,12,13] and attenuated FW signal in individuals with a longer duration of illness [14][15][16][17][18].This study expands upon those previous studies using our well-matched sample of healthy controls and participants with schizophrenia, thus enabling us to characterize the relationships between FW, age, and duration of illness, separately, while considering important clinical variables like SS and CPZ equivalents.Moreover, given the limited number of Free Water imaging studies investigating extracellular FW pathologies over time, our findings demonstrate that FW elevations may persist for longer than previously observed [12,13].
Earlier studies investigating white matter with traditional dMRI approaches have reported progressive reductions in FA across the course of illness in people with schizophrenia [3,33].In fact, our recent work utilizing dMRI harmonization to characterize agerelated trajectories of whole brain FA in 1092 participants with schizophrenia and controls closely aligns with these earlier reports [5].Reductions in FA were observed across all stages of illness, potentially indicating the presence of a continuous pathology that worsens with age and increased illness chronicity, with the greatest reductions observed in the oldest participants [5].In the present study, however, we see that the relationship between FW and age exhibits a very different pattern from those observed in FA.Specifically, the differences in FW between participants with schizophrenia and controls show two distinct periods of time where the divergence from controls is significantly greater.The larger of the two is found in younger participants, the majority of whom have a shorter duration of illness.This finding may have clinical utility for two key reasons.
First, the early stages of the illness, specifically those surrounding the onset of overt positive symptoms, represent a critical With increasing age, effect sizes followed a monotonically decreasing pattern until 39 ± 5 years of age, then followed a monotonic increase in effect size with a slower rate of change in the older individuals (>39 years) when compared to younger individuals (<39 years).
window for diagnosis and therapeutic interventions.Several studies have shown that successful treatment during time can minimize illness chronicity and help to reduce the incidence of psychiatric relapse in people with psychosis [35,36].However, there are presently no imaging markers that are found to be predominantly affected in the early stages of illness.Most imaging metrics exhibit a similar pattern to FA, such that deviations from healthy tend to become greater with more protracted illness and older age.As such, the behavior of FW in schizophrenia spectrum disorders appears to be distinctive.In fact, parallel work performed in this sample showed that FW was the most effective metric to differentiate participants with schizophrenia from controls with a subject-level classification machine learning approach [21].Thus, we propose that the FW metric, together with key imaging measurements used to study psychosis, could be developed further for future use in a clinical setting.
Second, given that greater elevations in FW appear to take place concurrently with, or immediately after, the onset of psychotic symptoms, a more complete understanding of the biological elements contributing to this extracellular pathology may provide novel insights into the pathophysiology of schizophrenia-spectrum disorders.Early reports argue that the elevations in FW found early in the course of illness are more likely the result of immune.While this is difficult to validate in humans, several studies provide evidence to support this claim.Di Biase et al. found a significant positive correlation between FW and plasma levels of pro-inflammatory cytokines IL-6 and TNF-a in individuals with schizophrenia [23].Lesh et al. reported a significant inverse correlation between glutathione, an antioxidant with anti-inflammatory properties, and FW in the dorsolateral prefrontal cortex in early-stage participants with schizophrenia [11].Finally, a preclinical study showed significant FW elevations in rodents who were prenatally exposed to the viral mimic polyriboinosinic-polyribocytidylic acid (Poly I:C) when compared to rodents prenatally exposed to saline [37].Interestingly, the elevations in FW occurred during the post-pubertal period of rodent development (postnatal day 90) [37], a period that coincides with the onset of structural and behavioral differences in the poly I:C MIA model [38].While none of the aforementioned studies demonstrated direct evidence for a relationship between central nervous system inflammation and FW, when taken together, we believe there is convergent evidence to suggest that the observed elevations of FW early in the course of the illness may have an immunological basis.Future studies investigating the relationships between FW and other validated biological indicators of immune activation, such as levels of proand anti-inflammatory cytokines in cerebrospinal fluid [39][40][41], are warranted to solidify these claims.
The present study supports earlier findings that differences in FW elevations between participants with schizophrenia and controls diminish with increased age [14][15][16][17][18].We expand on these findings by showing a gradually decreasing linear relationship between FW and a longer duration of illness independent of relevant clinical and demographic variables.It is important to note, however, that the differences in FW between the two groups significantly increase again between the ages of 39 and 58 years of age.We propose that this increase is more likely attributable to tissue atrophy as both biological processes, atrophy and immune activation, have been shown to lead to increases in isotropic diffusion in the brain [6].While white and gray matter tissue loss has been a consistent neuropathological finding in schizophrenia [42][43][44], it has yet to be elucidated the degree to which this loss is the result of the illness, chronicity, stress, aging processes, cumulative medication effects, or a combination of several disease-related factors, such as cardiometabolic symptoms that arise in subjects posited to be the result of long-term exposure to antipsychotics [45].Several meta-analyses have shown that gray matter tissue loss takes place during the early stages of illness [46][47][48].However, a similar consensus has not been formed for white matter.Widespread white matter tissue loss has been more commonly observed in chronic populations, with early course studies observing more focal differences [2,3] Given that many of the previous imaging studies utilized tools available at the time, mainly DTI, it remains difficult to interpret the biological implications of the previously reported FA reductions.However, given the available evidence, we suggest that the modestly increased FW in the older (>40 years of age) participants is more likely the result of tissue atrophy in the white matter due to several disease-related processes [20].
Finally, it is important to highlight that elevations in FW are not unique to individuals with schizophrenia spectrum disorders.Significant alterations in FW have also been reported in other psychiatric and neurodegenerative disorders.For instance, our group has previously shown that FW increases have also been found in individuals with chronic bipolar disorder [49], adolescentonset bipolar disorder with psychotic features [50], major depression [51], and obsessive-compulsive disorder [52].As such, we do not make claims that changes in FW are unique to schizophrenia.However, at present, there are more studies that have applied Free Water Imaging to schizophrenia-spectrum disorders than any other disorder.More importantly, the trajectory of FW changes across age or duration of illness has yet to be characterized for these illnesses.Further efforts are needed to evaluate the extent to which FW changes with increased age or illness duration in other psychiatric disorders to understand better the potential differences and similarities that may exist.
This study has several limitations that are necessary to note.The primary limitation is that the sample consists of cross-sectional dMRI data, which we utilize to characterize age-related changes in FW values.While we recognize the inherent limitations that arise when aggregating cross-sectional data, we believe that robust harmonization techniques, such as the one used in this study, are an essential contribution to the field.Leveraging the power of harmonization approaches enables the generation of clinical populations with large enough sample sizes in addition to allowing for the testing of the viability of new hypotheses.While longitudinal designs should be preferred for the more direct examination of maturational and aging trajectories, performing such studies from youth through late adulthood is currently more feasible and cost-effective with harmonized cross-sectional studies [53,54].Nevertheless, future longitudinal studies are necessary to verify our claims and understand the degree of interand intra-individual differences in the FW signal over time.Next, we would like to underline that three of the populations in our harmonized sample had been previously studied separately using FW, as outlined in the Supplementary.However, leveraging the harmonized large-scale datasets from 12 different sources, including these three datasets, was essential to examine changes in FW across the illness course effectively.This represents a unique and innovative extension of the previously published studies.Similarly, all dMRI data included in this sample are single-shell acquisitions (i.e., one single b-value).With single-shell data, Free Water imaging relies on spatial regularization and may be biased by T2 effects, as well as by perfusion and pseudo-diffusion effects from the blood [55].The availability of multi-shell acquisition protocols (i.e., more than one b-value) will represent a critical next step for furthering our understanding of the FW trajectories in people affected by illnesses with psychosis.Another important methodological next step would be to evaluate FW in gray matter.While several published studies have shown evidence of significant FW increases in gray matter areas [11][12][13], we recommend that these findings be interpreted cautiously.It is our perspective that the available tools to estimate extracellular FW are not sufficient to perform informed analyses in highly complex tissues like gray matter.Advances in both image acquisition and processing methodologies are being developed that will allow for the interrogation of microstructural gray matter features alongside investigations of microstructural white matter.We acknowledge that duration illness is a difficult measure to assess clinically.One limitation of our study is that we do not have access to the extensive information about the way duration of illness or age of onset was evaluated for each site.Our solution was to address this statistically; however, future studies where the duration of illness is systematically evaluated across participants are needed to validate our present findings.It is also important to note that due to an insufficient amount of clinical or medical data within the discrete age windows, we could not successfully implement our statistical models to evaluate the impact of several risk factors (e.g., diet, smoking, body-mass index, or cardiometabolic factors) on whole brain FW in white matter.Yet, in a recent study investigating a population of community-dwelling healthy elderly participants, neither BMI, hypertension, nor hypercholesterolemia was not found to be a significant predictor of FW values in white matter hyperintensities [56].While this does not rule out the impact of cardiometabolic factors in the present sample, it does present some preliminary evidence that they may not be the primary drivers of FW change in the older participants.Next, we note that over 87% of the datasets excluded individuals with active substance use/abuse (except nicotine or caffeine).While this does not preclude the likelihood of substance use taking place, we believe that its contribution to FW changes is likely to be minimal.However, no study has yet to directly investigate the impact of either active substance use/abuse or a history of substance use/abuse on the FW signal, making it an open question and a limitation of our study Finally, despite the lack of impact of current medication dose on our findings, it is still an important potential confound to address.At present, none of the previous Free Water imaging studies in psychosis reported a significant relationship between white matter FW and current medication dosage.However, Guo et al. showed a trend level increase in gray matter FW in the schizophrenia population who were treated compared to those who were antipsychotic naïve [13].More importantly, no study has yet investigated the relationship between FW and cumulative medication exposure in individuals with schizophrenia.Given the controversy regarding the potential impact of antipsychotic medications on the brain, particularly white matter (see [57]), it remains a critical issue that has yet to be fully explored.
In conclusion, our study provides strong evidence that FW is most significantly elevated in younger people with schizophreniaspectrum disorders who have a shorter duration of illness.It will be important to understand further the potential clinical utility of this measure and continue to explore the biological components contributing to the observed extracellular pathologies.In this work, we demonstrate that enabling pooled large-scale analysis of schizophrenia-spectrum neuroimaging datasets has noteworthy scientific benefits, including the opportunity to characterize brain changes along the lifespan of individuals at different disease stages.The application of a consistent pre-processing pipeline with a robust harmonization approach across sites provides a unique opportunity to analyze the data as if it were to come from a single scanner.Future studies should aim to include harmonized data from large publicly available imaging datasets that focus on younger and/or typically developing populations, such as the Adolescent Brain Cognitive Development study, in addition to clinical high-risk for psychosis populations [58][59][60][61][62][63][64][65].Large-scale analysis of these combined populations will provide information about events occurring prior to illness onset and may improve our understanding of potential risk factors that increase the probability of developing psychotic symptoms, such as traumatic experiences, substance use/abuse, genetics, prenatal or early developmental exposure to infections, and/or stress.In summary, our study provides strong evidence that elevations in FW are present in individuals with schizophrenia but that the greatest differences are observed in those individuals with a shorter duration of illness.

Fig. 2
Fig. 2 Effect size at each age of coefficient v2 from the quadratic model.Colored circles represent the significant differences between individuals with schizophrenia and controls (represented with the coefficient v2 in the quadratic model).We observed large effects for between-group standardized FW differences in younger individuals with schizophrenia (15-23 years; effect size range = [0.70-0.87]).With increasing age, effect sizes followed a monotonically decreasing pattern until 39 ± 5 years of age, then followed a monotonic increase in effect size with a slower rate of change in the older individuals (>39 years) when compared to younger individuals (<39 years).