Grey matter volume loss in Parkinson’s disease psychosis and its relationship with serotonergic gene expression: A meta-analysis

BACKGROUND
Neuroanatomical alterations underlying psychosis in Parkinson's Disease (PDP)remain unclear. We carried out a meta-analysis of MRI studies investigating the neural correlates of PDP and examined its relation with dopaminergic and serotonergic receptor gene expression.


METHODS
PubMed, Web of Science and Embase were searched for MRI studies (k studies = 10) of PDP compared to PD patients without psychosis (PDnP). Seed-based d Mapping with Permutation of Subject Images and multiple linear regression analyses was used to examine the relationship between pooled estimates of grey matter volume (GMV) loss in PDP and D1/D2 and 5-HT1a/5-HT2a receptor gene expression estimates from Allen Human Brain Atlas.


RESULTS
We observed lower grey matter volume in parietal-temporo-occipital regions (PDP n=211, PDnP, n=298). GMV loss in PDP was associated with local expression of 5-HT1a (b=0.109, p=0.012) and 5-HT2a receptors (b=-0.106, p=0.002) but not dopaminergic receptors.


CONCLUSION
Widespread GMV loss in the parieto-temporo-occipital regions may underlie PDP. Association between grey matter volume and local expression of serotonergic receptor genes may suggest a role for serotonergic receptors in PDP.

Background: Neuroanatomical alterations underlying psychosis in Parkinson's Disease (PDP) remain unclear. We carried out a meta-analysis of MRI studies investigating the neural correlates of PDP and examined its relation with dopaminergic and serotonergic receptor gene expression. Methods: PubMed, Web of Science and Embase were searched for MRI studies (k studies = 10) of PDP compared to PD patients without psychosis (PDnP). Seed-based d Mapping with Permutation of Subject Images and multiple linear regression analyses was used to examine the relationship between pooled estimates of grey matter volume (GMV) loss in PDP and D1/D2 and 5-HT1a/5-HT2a receptor gene expression estimates from Allen Human Brain Atlas.

Introduction
Non-motor symptoms of Parkinson's Disease (PD) are distressing, and associated with poor quality of life in PD patients Stebbins, 1993, 1995). Hallucinations (or perceptual disturbances such as seeing or hearing things that are not there), which are typically visual in modality in people with PD (Archibald et al., 2011;Goetz et al., 2006), and delusions (or fixed false beliefs) are some of the most common and debilitating non-motor symptoms in PD patients (Ffytche et al., 2017). Collectively, they are referred to as PD psychosis (PDP), often co-exist with other groups of non-motor symptoms such as depression, sleep disorders and anxiety (Aarsland and Kramberger, 2015;Kulisevsky et al., 2008), and are associated with greater burden of care and risk of hospitalisation Martinez-Martin et al., 2015). Therefore, there is a particular need to diagnose and intervene early. Widely used diagnostic criteria for PDP such as that recommended by a National Institute of Neurological Disorders and Stroke (NINDS)-National Institute Mental Health (NIMH) working group (Ravina et al., 2007), as well as scales to assess severity such as the schedule for assessment of positive symptoms (SAPS) recommended for use in this population (Fernandez et al., 2008) or its adaptation for PDP such as SAPS-PD (Voss et al., 2013) are useful in this regard. However, a key issue that affects treatment of PDP is limited availability of effective and well-tolerated interventions, which, in turn is linked to sub-optimal understanding of its causes. PDP was initially considered a by-product of dopamine augmentation treatments (Ecker et al., 2009;Jakel and Stacy, 2012). However, there is evidence that drug-naïve PD patients may also experience psychotic symptoms (Friedman, 2016;Pagonabarraga et al., 2016). Although, the precise cause of PDP remains unclear, a number of risk factors for development of psychotic symptoms have been described, including severity and duration of PD, dopaminergic medications, sleep disorders, cognitive decline, presence of widespread Lewy Body pathology, and later onset of PD (Barrett et al., 2018;Chang and Fox, 2016;Factor et al., 2014;Ffytche et al., 2017;Williams and Lees, 2005). The first line of treatment for PDP often involves reassurance or other simple strategies as well as adjustment of dose of PD medications (Ffytche et al., 2017), specifically dopamine agonists associated with presence of hallucinations (Ecker et al., 2009;Jakel and Stacy, 2012;Marras et al., 2007). When these measures do not provide adequate relief, antipsychotic medications, that block dopaminergic receptors, such as quetiapine, or clozapine (Aarsland et al., 1999;Black, 2017) and more recently Pimavanserin (Cummings et al., 2014), which acts as an inverse agonist/ antagonist at 5HT2A receptors (Meltzer et al., 2010), are available for use. However, these medications are often sub-optimal in terms of efficacy, tolerability and/or need for invasive monitoring (Bosboom and Wolters, 2004;Breier et al., 2002;Chen, 2017;Jethwa, 2017). Hence, there is a need for better understanding of the abnormality underlying psychotic symptoms in PD, so that the clear unmet need for more effective treatments for psychosis in PD may be addressed.
While a range of mechanisms have been proposed for the different non-motor symptoms such as depressive and cognitive symptoms associated with PD (Bose and Beal, 2016;Cao et al., 2022;Chen and Zhang, 2022;El-Kattan et al., 2022;Hestad et al., 2022;Nyatega et al., 2022), the precise mechanisms underlying the emergence of psychotic symptoms in PD remain unclear (Ffytche et al., 2017). For the purposes of this review, we wanted to focus on brain systems level alterations, investigating in particular brain structural alterations and their relationship with markers of neurotransmitter pathways implicated in PD psychosis in order to understand their biological relevance. Notwithstanding the relevance of the various processes implicated in PD and its non-motor symptoms (as partially referred to above), we primarily wanted to focus on dopaminergic and serotonergic pathways because, arguably, evidence is most robust currently for their involvement in psychosis in PD. In particular, dopamine neuron loss is one of the key characteristics of PD (Giguère et al., 2018;Surmeier, 2018), dopamine-replacement medications used in the treatment of motor symptoms of PD are associated with initiation of or exacerbation of the symptoms of psychosis in PD (Kuzuhara, 2001;Poewe and Seppi, 2001) and existing antipsychotic medications (which block dopamine receptors) (Gomes and Grace, 2021) are used in the treatment of symptoms of psychosis in PD (Ballard et al., 2015;Fernandez et al., 2003). We also wanted to focus on serotonergic receptors, as serotonergic mechanisms have been implicated in the development of neurodegenerative diseases (Factor et al., 2017;Stahl, 2016b), their involvement has also been found in PD and PD psychosis patients (Ballanger et al., 2010a(Ballanger et al., , 2010bFox et al., 2009) and pimavanserin, one of the current treatments for PDP acts on serotonergic receptors (Ballard et al., 2020;Cummings et al., 2014). In terms of brain structural alterations, results from structural as well as task-based and resting state magnetic resonance imaging (MRI) and positron emission tomography (PET) studies have generally reported grey matter reductions (Bejr-Kasem et al., 2019;Gama et al., 2014;Nishio et al., 2018;Ramírez-Ruiz et al., 2007) encompassing the ventral and dorsal visual pathways and hippocampus (Alzahrani and Venneri, 2015;Lenka et al., 2015;Yao et al., 2016) as well as altered temporo-parieto-occipital activation, metabolism or functional and structural connectivity in large-scale brain networks (Boecker et al., 2007;Hall et al., 2019;Hepp et al., 2017;Lefebvre et al., 2016b;Shine et al., 2015;Stebbins et al., 2004). This is consistent with a recent meta-analytic synthesis reporting fronto-temporo-parieto-occipital brain grey matter reduction in a mixed group of participants with PD and Lewy body dementia (DLB) with visual hallucinations (Pezzoli et al., 2021), although they did not consider the effect of potential confounders such as PD medications or concomitant symptoms, e.g., cognitive decline. Complementing this, a recent mega-analysis (Vignando et al., 2021) has reported extensive reduction of whole-brain cortical thickness and surface areas in visual cortex, left insula and hippocampus in PD patients with visual hallucinations compared to those without, as well as a relationship between cortical thickness loss and higher regional availability serotonergic (5-HT2a and 5-HT1a) and dopaminergic receptors (D2/D3). Collectively, these reviews focused on related different metrics of brain structure (volume, cortical thickness and surface). Meta-analytic integration of evidence regarding brain structural alterations in PD psychosis may help unravel the key neuroanatomical alterations that may underlie Parkinson's psychosis. Therefore, here we have conducted a quantitative synthesis of evidence from structural neuroimaging studies investigating the neural correlates of psychosis in PD using a systematic review and meta-analysis approach to identify the brain structural alterations associated with PDP. Subsequently, we investigated the relationship between these structural alterations and the candidate neurotransmitter pathways implicated in PD psychosis to explore their biological relevance. Specifically, we examined whether the spatial distribution of brain structure alteration was associated with the spatial architecture of brain expression of the genes for serotonergic (namely 5-HT2a and 5-HT1a) and dopaminergic (D1 and D2) receptors, the key candidate pathways implicated in PD psychosis (Cho et al., 2017;Factor et al., 2017;Hacksell et al., 2014;Makoff et al., 2000;Wolf, 2000;Wolters, 1999). One of the limitations of a previous attempt in this regard by Vignando et al. (2021) is that they examined the relationship between the structural neuroimaging correlate and the density distribution data for each receptor of interest separately as opposed to examining their relationship collectively. Further, as mentioned before, Vignando and colleagues (2021) examined the relationship of receptor density distribution with cortical thickness and surface area and did not examine the relationship with regional brain volume. Unlike the approach adopted by Vignando et al. (2021) where they examined the relationship between the structural neuroimaging correlate and the density distribution data for each receptor of interest separately, we included the key candidate receptors of interest in a multiple regression model to allow investigation of their relationship with neuroanatomical correlates after taking into account the effect of receptors belonging to other candidate pathways of interest.

Search strategy and eligibility criteria
We followed (PROSPERO registration number: CRD42020221904) the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Moher et al., 2009) and a detailed description of the search strategy is outlined in Supplementary Material 1. Studies identified from PubMed, Web of Science, Embase and the Neurosynth database were included if they examined brain alterations associated with psychosis symptoms (i.e., hallucinations and/or delusions) after PD diagnosis using different neuroimaging modalities, and a case-control design and provided brain coordinates (or available statistical maps) in standardised reference spaces, e.g., Montreal Neurological Institute (MNI) or Talairach space. There were no exclusions or restrictions for this search. For this meta-analysis, we solely focused on studies that used structural magnetic resonance imaging (MRI) modality.

Data extraction
Data extraction involved: study details (i.e., authors' names, year of publications), study design, scanner characteristics (e.g., manufacturer), sample size, sample characteristics (e.g., age, gender, education levels), PD onset age, disease duration, PD symptoms, clinical measures of psychosis symptoms, cognitive measures, dopamine-replacement medications (expressed in Levodopa equivalent daily dose (LEDD)), and other clinical outcome measures (e.g., depression). Where only median values were available, this was converted into mean (Luo et al., 2018). Coordinates type (e.g., MNI or Talairach space), and associated t values were also extracted. Where coordinates were expressed in Talairach or other standard normalised spaces, they were converted into MNI coordinates. Authors were contacted for studies that did not report coordinate data. Data extraction was conducted independently by two researchers (SP, YL). Discrepancies were addressed through consensus or discussion with senior researchers.

Data synthesis
A coordinate-based meta-analysis was conducted for structural MRI studies reporting voxel-based morphometry (VBM) results using peak coordinates and/or statistical maps by employing a random-effects approach using Seed-based d Mapping with Permutating Subject Images (SDM-PSI) (version 6.21, https://www.sdmproject.com/) (Anton Albajes-Eizagirre et al., 2019a, 2019b. This involved computation of upper and lower bounds of effect size maps for studies based on the available peak coordinates by imputing voxel-wise effect size (Albajes-Eizagirre et al., 2018;Radua et al., 2012Radua et al., , 2010 and combining them with statistical parametric maps provided by the studies (where applicable) converted into effect size maps to carry out a random-effects meta-analysis. Contrasts of interest were 'PD psychosis (PDP) < PD without psychosis (PDnP)' (i.e., grey matter loss) and 'PD psychosis (PDP) > PD without psychosis (PDnP)' (i.e., increased grey matter) with and without controlling for the effect of PD medications (expressed in LEDD) (Tomlinson et al., 2010), and cognitive scores. Due to the various neurocognitive scales used to assess cognitive abilities in PDP and PDnP patients, we computed standardised scores for each study by dividing the mean score by the standard deviation (SD) for each patients' group. Statistical parametric maps provided by the studies (where applicable) were registered to the SDM template, and their t values were converted into effect sizes and included in the analysis. SDM-PSI provides I 2 statistic to assess heterogeneity in the observed peaks as well as between-study heterogeneity (Higgins et al., 2003). Funnel plots for each peak and Egger's test (Egger et al., 1997) were used to assess publication bias. Clusters smaller than 10 voxels were discarded.
We examined the association between grey matter volume loss (measured using Hedges' g effect-size estimates extracted from the centroid of the SDM-PSI meta-analytic map parcellated across 78 brain regions of the Desikan-Killiany atlas (Desikan et al., 2006)) in PDP compared to PDnP patients with dopaminergic (D1/D2) and serotonergic (5-HT2a/5-HT1a) mRNA microarray gene expression using a multiple linear regression analysis. A secondary analysis included LEDD as a covariate when computing Hedges' g effect-size estimate. In light of the co-occurrence of depression, cognitive decline and psychosis in PD (Diederich et al., 2009;Moustafa et al., 2014) and the association between depression and 5-HT1a abnormalities (Savitz et al., 2009;Wang et al., 2016), we also wanted to account for depression and cognitive scores. As SDM-PSI only allows 2 covariates per analysis, we were unable to include all 3 covariates (LEDD, depression and cognitive scores) simultaneously. Therefore, we conducted two additional multiple linear regression analyses, one including LEDD and cognitive scores another including LEDD and depression scores as covariates when computing Hedges' g effect size estimates. Gene expression data for dopaminergic and serotonergic receptors was extracted from the Allen Human Brain Atlas which includes microarray expression data in tissue samples from six healthy adult human brains with more than 20,000 genes quantified across cortical and subcortical regions, including brainstem and cerebellum (Hawrylycz et al., 2012). Probe-to-gene re-annotation data was downloaded using the information from Arnatkevic̆iūtė et al. (2019). Sample-to-region matching, and extraction of the gene expression data followed established approaches (Arnatkevic̆iūtė et al., 2019;Markello et al., 2021). Gene expression data for D3 receptor could not be matched to the probe and therefore was not extracted.

Quality rating assessment
Study quality was assessed with the Newcastle-Ottawa rating Scale (Wells et al., 2000) for case-control studies. This scale includes three methodological domains: "Selection" (i.e., definitions of cases and controls, and their selection), "Comparability" (i.e., comparison of cases and controls on the variables of interest, inclusion of covariates), and "Exposure" (i.e., how exposure such as condition was defined and ascertained). This rating is based on a star system: studies can be awarded a maximum of one star when the definitions for each "Selection" and "Exposure" items are met, and a maximum of two stars when the criterion for "Comparability" of cases and controls on the basis of study design (i.e., one star) and/or analysis (i.e., one star) is met. Quality ratings were conducted by one researcher. In this review, neural substrates of PDP patients were compared to those in PDnP patients, thus the latter acted as control or comparison group. Therefore, the items "Selection of controls" and "Definition of controls" were specifically related to PDnP patients.

Study quality assessment
Full quality ratings are reported in Table 2. Briefly, all studies had full score on the "Exposure" domain (i.e., ascertainment, non-response rate and ascertainment method for PDP and PDnP patients); "Comparability" was good in all studies due to their matched design, whereby patients were matched on age, gender and other clinical or demographic variables, whilst other studies reported these variables as covariates in the analysis. "Selection" domain included selection and definition of both PDP patients (i.e., cases) and PDnP patients (i.e., controls), selection of PDP patients was assigned one star (i.e., the maximum) in all studies. Similarly, selection of PDnP patients was assigned one star in all studies, whilst definition of such group was clear in one study which was assigned one star in this domain.

Meta-analysis: Voxel-based morphometry (VBM)
Meta-analysis of 10 MRI studies (one provided statistical maps (Lee et al., 2017)) showed that PDP patients had reduced grey matter volume in parietal-temporo-occipital areas compared to PDnP patients with the largest clusters located in the right precuneus (extending to left precuneus and bilateral cuneus; voxel number = 1059, Z = − 2.990, p = 0.001), bilateral inferior parietal gyrus (left inferior parietal gyrus, extending to left angular gyrus and supramarginal gyrus; voxel number = 296, Z = − 2.449, p = 0.007; right inferior parietal gyrus, extending to right angular gyrus; voxel number = 277, Z = − 2.225, p = 0.013), left inferior occipital gyrus (voxel number = 262, Z = − 2.627, p = 0.004), and right middle temporal gyrus (extending to right inferior temporal gyrus; voxel number = 225, Z = − 2.664, p = 0.003) (eFig. 1, Supplementary Material 2). We did not observe any significant peaks in the opposite direction (i.e., 'PDP > PDnP'). Neither between-study heterogeneity as a proportion of total variability (all I 2 statistic <25%, except right inferior parietal gyrus, which was ~27.6%) nor publication bias was observed in these peaks (Funnel plots in Supplementary Material 2). However, none of these brain regions survived correction for multiple testing. After including PD medication dose, expressed as LEDD, as a covariate, grey matter reduction in right precuneus remained significant in PDP patients compared to PDnP patients (extending to left precuneus, and bilateral cuneus; voxel number = 1084, Z = − 2.952, p = 0.001). There was substantial overlap between previously identified peaks, with the largest clusters in the left angular gyrus (extending to left inferior parietal gyrus; voxel number = 389, Z = − 2.596, p = 0.004), right inferior temporal gyrus (extending to right middle temporal gyrus and right fusiform; voxel number = 353, Z = − 0.2921, p = 0.001), the left middle occipital gyrus (extending to inferior occipital gyrus; voxel number = 315, Z = − 2.902, p = 0.001), and the right inferior parietal gyrus (extending to right angular gyrus; voxel number = 304, Z = − 2.525, p = 0.005) (eFig. 2, Supplementary Material 2). No publication bias was observed in any of these analyses (Funnel plots in Supplementary Material 2). However, none of these brain regions survived correction for multiple testing. When cognitive scores were entered as covariate, the right precuneus remained the largest cluster showing grey matter volume reduction (extending to the left precuneus, bilateral cuneus, voxel number = 1186, Z = − 3.287, p < 0.001). In addition, this   Supplementary Material 2). The overlap between brain regions across the unadjusted and adjusted analyses is shown in Figs. 2 and 3, and Table 3 reports peaks coordinates from the three analyses. We did not observe any significant peak in the opposite direction (i.e., 'PDP > PDnP') in either LEDD-or cognitive score-adjusted analyses. None of the areas across the three analyses survived familywise error correction.

Whole brain correlations with D1/D2 and 5-HT2a/5-HT1a gene expressions
Due to presence of multicollinearity in the multiple linear regression model, D2 receptor data were dropped from the analysis (i.e., due to a variation inflation factor, VIF = 6.5). Multiple linear regression analysis showed a significant association between Hedges' g effect-size estimates of grey matter volume unadjusted for LEDD and 5-HT2a gene expression (regression coefficient = − 0.107 (95% CI, − 0.174, − 0.039), t = − 3.147, p = 0.002) and 5-HT1a gene expression (regression coefficient = 0.109 (95% CI, 0.024, 0.193), t = 2.565, p = 0.012) but not with D1 gene expression (D1, p = 0.554) across the 78 regions of the Desikan-Killiany atlas. Separate analysis with Hedges' g effect-size estimates for grey matter volume adjusted for LEDD did not change the result (5-HT2a, regression coefficient = − 0.120, 95% CI − 0.190, − 0.050, t = − 3.408, p = 0.001; 5-HT1a, regression coefficient = 0.126, 95% CI 0.038, 0.213, t = 2.848, p = 0.006; D1, p = 0.597). Presence of multicollinearity was also detected in this model and D2 was removed (VIF = 6.67). Fig. 4 reports the association between grey matter volume loss and receptor density from the LEDD-adjusted analysis. In both multiple linear regression models (adjusted and unadjusted for LEDD), the associations between regional cortical volume and serotonergic gene expressions were consistent in terms of direction of relationship. However, these associations were in opposite directions for the two serotonergic receptors, whereby the less the grey matter volume the more the 5-HT2a gene expression density and the less the 5-HT1a gene expression density across the 78 regions of the Desikan-Killiany atlas. That is, the greater    Additional multiple linear regression analyses adjusted for LEDD and cognitive score and adjusted for LEDD and depression scores showed comparable relationships as with the unadjusted one and that adjusted for LEDD alone, as reported above (please see details in Supplementary Material 4).

Discussion
Findings from individual studies reported extensive areas of lower grey matter volume in PDP without providing clear understanding on whether PDP may be due to structural issues in key regions or diffused anomalies. Here, we have addressed this ambiguity by synthesizing results from structural MRI studies using a quantitative approach to identify neuroanatomical correlates implicated in PDP. The metaanalysis results show widespread grey matter volume reduction across parietal-temporal-occipital regions in PDP patients, specifically, the Fig. 2. A) Overlapping peak areas (shown in purple) with grey matter loss in PD psychosis patients unadjusted (red) and adjusted for LEDD (expressed in mg/day) (blue) as covariate (uncorrected). This is indicated by the red and blue colour bars on which represent the T threshold of the voxels within the map. B) Overlapping peak areas (shown in yellow) with grey matter loss in PD psychosis patients unadjusted (red) and adjusted for cognitive scores (green) as covariate (uncorrected). This is indicated by the red and green colours bar on the right-hand side of the figure which represent the T threshold of the voxels within the map. The left side of the brain is shown on the right side of these brain images. Fig. 3. Forest plot representing the Hedges' g effect-size estimates for the observed peaks from the SDM-PSI unadjusted analysis with peak names, voxel number for each peak, and I 2 statistic as a measure of heterogeneity. right precuneus (extending to the left precuneus), bilateral inferior parietal gyrus, and left inferior occipital gyrus, corroborating previous evidence (Goldman et al., 2014;Lenka et al., 2015;Yao et al., 2014). Volume reduction in right precuneus, bilateral inferior parietal gyrus, left median cingulate/paracingulate cortex, and right inferior and middle temporal gyrus remained significant and was independent of PD medications and cognitive scores. When PD medications and cognitive scores standardised across studies were included as a covariate, there were additional areas of lower cortical volume within occipital (e.g., left middle occipital gyrus), parietal (e.g., right lingual and left angular gyrus), and temporal regions (e.g., right fusiform gyrus). These results also extend on previous evidence (Bejr-Kasem et al., 2019;Jia et al., 2019;Lenka et al., 2015;Yao et al., 2014) showing more extensive cortical involvement in PD psychosis with the more prominent regions being in the temporal, parietal and occipital lobes. Across the analyses, we identified large clusters of grey matter loss in regions associated with higher order visual processing (i.e., dorsal and ventral visual pathways) and the Default Mode Network (DMN), and specifically the precuneus, one of the DMN nodes, which is embedded in the parietal lobe and is involved in information processing (including visual information) (Buckner et al., 2008;Cavanna and Trimble, 2006;Freton et al., 2014;Hafkemeijer et al., 2012;Lenka et al., 2015). Grey matter volume loss in these areas may lead to dysfunctional information integration giving rise to visual hallucinations in PD patients, and also potentially due to an overreliance on endogenous attentional mechanisms (Shine et al., 2011;Shine, O'Callaghan et al., 2014) as suggested by the involvement of DMN nodes (Rektorova et al., 2014).
Furthermore, this review expands on previous evidence by investigating the association between regional grey matter volume (as measured by Hedges' g effect-size estimates from the meta-analysis) and key candidate receptors involved in PDP as indexed by mRNA microarray gene expression extracted from the Allen Human Brain Atlas Table 3 Peak coordinates showing greater grey matter loss in PDP patients compared to PDnP patients, alongside number of voxels, Z scores, and associated p value (uncorrected), I 2 (which indicates the magnitude of between-study heterogeneity as a proportion of total variability within each peak), and publication bias expressed as Egger's test p value. n.s.: non-significant (Hawrylycz et al., 2012). Expression of 5-HT2a and 5-HT1a receptors were associated with estimates of grey matter volume, albeit being in opposite direction. In addition, in the analyses adjusted for PD medication (expressed in LEDD) with and without depression score and cognitive score, we observed similar patterns of results. This may suggest that the relationships between grey matter volume in people with PD psychosis and spatial distribution of expression of serotonergic receptors observed here are unlikely to have been confounded by the dose of dopamine-replacement treatments that they were receiving or concurrent depressive symptoms or cognitive functional status. Whether the inverse relationship observed here between pooled grey matter volume in PDP patients and the pooled expression of 5-HT2a receptors in an independent healthy cohort indicates that in PDP patients the greater the grey matter volume loss the greater is the regional 5-HT2a receptor availability, remains to be tested by direct investigation using both imaging modalities in the same group of participants. Nevertheless, if an association between grey matter loss in PDP and increased 5-HT2a receptor availability is demonstrated in PDP, it would be consistent with previous evidence of increased 5-HT2a receptor availability in those with PDP (Ballanger et al., 2010a(Ballanger et al., , 2010b and pharmacological evidence of the efficacy of the atypical antipsychotic Pimavanserin, a highly selective 5-HT2a receptor antagonist, in the treatment of psychosis in PD (Majlath et al., 2017;Meltzer et al., 2010;Mohanty et al., 2019;Stahl, 2016a;Yunusa et al., 2020). On the other hand, the direct relationship between grey matter volume and the pooled expression of 5-HT1a receptors, may indicate that in PDP patients the lower the grey matter volume the lower the regional availability of 5-HT1a receptors. Whether this holds true when investigated using both imaging modalities in the same cohort of patients remains to be seen. Given the association between 5-HT1a abnormalities and depression (Albert and Le François, 2010;Savitz et al., 2009;Wang et al., 2016), this may reflect the common co-occurrence of depressive symptoms in people with PDP. While the degeneration of dopaminergic neurons in the substantia nigra is the hallmark of PD pathology (Birkmayer and Birkmayer, 1987;Birkmayer et al., 1975;Kordower et al., 2013), it is also followed by degeneration of serotonergic neurons in subcortical areas such as the raphe nuclei (Fox et al., 2009;Joutsa et al., 2015;Stahl, 2016b). Further, upregulation of 5-HT2a receptors has been observed in visual and temporal regions in PDP patients which may be a result of Lewy bodies accumulation in subcortical regions which also was associated with presence of hallucinatory behaviour (Birkmayer et al., 1974;Huot, 2018;Huot et al., 2010). Our results are in line with those from Vignando et al.'s (Vignando et al., 2021) analysis showing association between structural measures and 5-HT2a and 5-HT1a binding but not in terms of relationship with regional D1 density. Presumably, the development of psychosis in PD patients could then be due to dysfunction in networks involved in attention control and visual information processing (Lenka et al., 2015;Muller et al., 2014;Shine et al., 2011;Shine, O'Callaghan et al., 2014) that become pathological due to grey matter reduction in their nodes, as shown in the meta-analysis results, which consequently leads to abnormal functionality at a network level. This may be in line with previous research investigating white matter abnormalities in PD psychosis patients (Hall et al., 2019;Hepp et al., 2017), in PET studies (Boecker et al., 2007) and in task-based functional MRI (fMRI) studies (Lefebvre et al., 2016a) which also reported dysfunctions in these areas. This may indicate that atypical functional patterns in temporal, occipital and parietal regions are a by-product of cortical atrophy of these regions Fig. 4. Gene expression density of 5-HT2a (A), 5-HT1a (B) receptors in cortical and subcortical regions, and Hedges' g effect size in cortical and subcortical regions (C) derived from the covariate meta-analysis (adjusted for LEDD) results showing decreased grey matter in PDP patients compared to PDnP patients (PDP < PDnP), parcellated across the 78 brain regions of the Desikan-Killiany atlas (Desikan et al., 2006). On the right-hand side, scatterplots showing the relationship between 5-HT2a (D) and 5-HT1a (E) with Hedges' g effect-size estimates of grey matter volume adjusted for LEDD in PDP patients compared to PDnP patients (the regression lines are adjusted for all the predictors). (Hall et al., 2019;Shine et al., 2015). Our results are to some extent different from those of the mega-analysis (Vignando et al., 2021) and may be due to the diverse methodologies used (mega-analysis vs. meta-analysis), the parameters used to assess atrophy (cortical thickness and surface area vs. voxel-based grey matter), and the availability of raw data (subject-level data vs. peak coordinates and one T map). They reported reduced surface area and extensive reduction in cortical thickness in occipital, temporal, parietal, frontal and limbic regions in PDP patients with visual hallucinations, identifying asymmetry in the left ventral visual pathway and extensive cortical thinning in bilateral cuneus, left dorso-medial superior frontal gyrus. Conversely, we mainly reported parietal-temporo-occipital regions as areas of significant grey matter loss and two frontal regions were found involved in PDP (the latter observed in the main analysis and in the analysis adjusted for cognitive state), our results however did not survive familywise error correction. Interestingly, we identified the right precuneus (extending to the left precuneus) as the largest cluster affected by grey matter loss in PDP patients across the three analyses (i. e., unadjusted, adjusted for PD medication and for cognition). The cuneus and precuneus were also reported as two of the areas with the greatest effect size in the principal component analysis by Vignando et al. (2021) examining the nodes that contributed the most to cortical thickness reduction in PDP patients, as well as in their network analysis. We conducted a whole-brain association between grey matter loss in PDP patients and receptor density examining dopaminergic and serotonergic receptors. We did not observe any relationships between grey matter volume and regional D1 density. This is in contrast with Vignando et al. (2021) who found associations with dopaminergic receptor density with cortical thickness in regions where PDP and PDnP had significant differences, and with surface area for significant regions of difference as well as across the cortex. However, we were only able to extract data on D1/D2 receptors unlike Vignando et al. (2021) who investigated the relationship with D2/D3 joint receptor density distribution. This discrepancy may also be due to methodological differences: they used mean thickness differences as the outcome variable, whilst we employed Hedges' g effect-size estimates extracted from SDM-PSI used as a measure of grey matter loss in PDP patients, lastly they examined in-vivo PET data from an independent healthy cohort whilst we relied on gene expression data extracted from six healthy donors of the Allen Human Brain Atlas (Arnatkevic̆iūtė et al., 2019;Desikan et al., 2006).

Strengths and limitations
The main limitations were the lack of significant results in the familywise error corrected analyses potentially due to the small number studies and the limited access to raw imaging data as we mainly relied on peak coordinates, a less powerful approach (Salimi-Khorshidi et al., 2009), and that we have fewer studies than recommended guidelines (Müller et al., 2018). We were also unable to relate meta-analytic estimates with psychotic symptoms. These studies focused on visual hallucinations in PD. This may also be due to the different rating scales applied to assess the presence of these symptoms, some measures may not differentiate or may not be sensitive to these symptoms. Whilst some studies referred to the NINDS-NIMH criteria (Ravina et al., 2007), most used different assessment tools or interviews. Although visual hallucinations and illusions are the most prevalent psychotic manifestations in PD, there is evidence of patients experiencing auditory and multimodal (e.g., olfactory, tactile (Chou et al., 2005;Solla et al., 2021)) hallucinations and delusions (Goetz et al., 2006;Marsh et al., 2004;Papapetropoulos et al., 2008) as the disease progresses, leading to increased distress and worsening of quality of life for the patients. Future exploration should compare the neuroanatomical correlates of different psychotic symptoms in PD, as this can shed light on whether there are common pathways shared with visual hallucinations, as those observed in this meta-analysis, or symptom-specific dysfunction. We observed peaks of grey matter volume reduction in temporal, parietal and occipital regions with the largest cluster located in the bilateral precuneus. Volume loss in the precuneus warrants further investigation in terms of its role in psychosis symptom generation in PDP as altered connectivity with precuneus has also been reported in the context of neuropsychiatric symptoms such as psychosis in people with Alzheimer's dementia (Balthazar et al., 2014;Lee et al., 2020). Whether psychotic symptoms such as hallucinations and delusions may be caused by atrophy in this specific area or whether this area contributes the most to this symptom is yet to be investigated. Future research may also investigate whether volume changes in regions identified in this meta-analysis may serve as prognostic biomarkers for PD psychosis. It is worth noting that we used a particular analytic approach to meta-analytically synthesize the results of existing studies (A. Albajes-Eizagirre et al., 2019a, 2019b. However, there are other approaches to synthesize neuroimaging evidence such as using activation likelihood estimation, permutation of inference (Winkler et al., 2014) and network modelling techniques (Smith et al., 2011), Kernel density analysis or Bayesian approaches which address slightly different but related questions (Eickhoff et al., 2012;Radua and Mataix-Cols, 2009;Radua et al., 2012Radua et al., , 2014Wager et al., 2004). Future research may also employ these other approaches to get a better idea about the key brain structural alterations associated with PDP. A further limitation is that we used mRNA expression, a result of transcriptional and translation activity, as a measure of dopaminergic (i.e., D1, D2) and serotonergic (i.e., 5-HT2a, 5-HT1a) gene expression extracted from 6 neurotypical donors (Hawrylycz et al., 2012) which may not be presentative of the PD population and may introduce variability due to inter-individual differences. Nevertheless, several studies employing this methodology have shown the consistency of the receptor spatial architecture in the cortex across individuals (Beliveau et al., 2017;Rizzo et al., 2016Rizzo et al., , 2014Selvaggi et al., 2019;Veronese et al., 2016). Similar approaches have been applied by other studies assessing the relationship between estimates of brain functions (from fMRI studies) and regional gene expression data, also under pharmacological intervention (Anderson et al., 2020;Gryglewski et al., 2018;Richiardi et al., 2015;Selvaggi et al., 2019;Vértes et al., ). Future studies may address this limitation by employing multimodal neuroimaging approaches that combine PET imaging using 5-HT1a and 5-HT2a receptor ligands (for example, Ballanger et al., 2010aBallanger et al., , 2010b and structural MRI in the same set of participants with and without PDP to investigate whether relationship between in vivo receptor availability and brain volume alterations as suggested by results presented here hold true.
Another limitation worth considering relates to the fact that results presented here suggest that anomalies in brain regions involved in processing visual stimuli, over reliance on internal processing, and potential deficits in directing attention and integrating endogenous and externally derived information may underlie psychosis in PD. However, the studies reviewed here did not investigate whether these cognitive processes and function of the brain substrates underlying those processes were impaired in PDP patients included in these studies. Future studies may therefore employ complementary neuroimaging approaches as functional MRI to investigate neurophysiological abnormalities in PDP to develop a more comprehensive neurocognitive mechanistic account of psychosis in PD. In addition, the studies included in this metaanalysis were all cross-sectional. To better understand the role of structural and functional brain changes in psychosis in PD, future studies should consider investigating longitudinal cohort of patients with PD monitored over time to relate any changes with the emergence of psychosis in PD. Furthermore, such a design may pave the way for exploring whether grey matter atrophy and functional brain abnormalities may be clinically useful prognostic or diagnostic biomarkers for psychosis in PD.

Conclusions
In conclusion, we found evidence of grey matter volume loss in the parietal-temporal-occipital regions in PD patients with psychosis which persisted even after adjusting for the effects of PD medications and cognitive scores and an association between volume alterations and serotonergic receptor gene expression. Future studies need to incorporate longitudinal cohort designs with multimodal neuroimaging approaches that combine PET and functional MRI to precisely delineate the neurocognitive mechanisms and neuroreceptor alterations that may underlie psychosis in PD. This may help identify pharmacological (e.g. receptor, neurophysiological or cognitive alterations) as well as nonpharmacological (e.g. cognitive or psychological processes) targets for the development of effective and well-tolerated novel interventions for which there is a considerable unmet need.

Funding
SB, LV, DA, DF, KRC and CB are in receipt of funding from Parkinson's UK for a clinical trial in Parkinson's disease psychosis. SP PhD studentship is funded by Parkinson's UK. The funding source had no involvement in this research. SB is supported by grants from the National Institute of Health Research (NIHR) Efficacy and Mechanism Evaluation scheme and Parkinson's UK. SB has participated in advisory boards for or received honoraria as a speaker from Reckitt Benckiser, EmpowerPharm/SanteCannabis and Britannia Pharmaceuticals. All of these honoraria were received as contributions toward research support through King's College London, and not personally. SB also has collaborated with Beckley Canopy Therapeutics/Canopy Growth (investigator-initiated research) wherein they supplied study drug for free for charity (Parkinson's UK) and NIHR (BRC) funded research. The views expressed are those of the authors and not necessarily those of the NHS, the NIHR or the Department of Health. LV has collaborated with Beckley Canopy Therapeutics/ Canopy Growth (investigator-initiated research) wherein they supplied study drug for free for charity (Parkinson's UK) and NIHR (BRC) funded research.

Conflict of interest
The authors have declared that no competing interests exist.

Data availability
Data extracted from published manuscripts.