Changes in resting-state measures of prostate cancer patients exposed to androgen deprivation therapy

The aim of the present work is to describe the differences in rs-fMRI measures (Amplitude of low frequency fluctuations [ALFF], Regional Homogeneity [ReHo] and Functional Connectivity [FC]) between patients exposed to Androgen deprivation therapy (ADT) and a control group. Forty-nine ADT patients and fifteen PC-non-ADT patients (Controls) were included in the study. A neuropsychological evaluation and a resting-state fMRI was performed to evaluate differences in ALFF and ReHo. Region of interest (ROI) analysis was also performed. ROIs were selected among those whose androgen receptor expression (at RNA-level) was the highest. FC analysis was performed using the same ROIs. Higher ALFF in frontal regions and temporal regions was identified in Controls than in ADT patients. In the ROI analysis, higher activity for Controls than ADT patients was shown in the left inferior frontal gyrus and in the left precentral gyrus. Lower ALFF in the right hippocampus and the lateral geniculate nucleus of the right thalamus was identified for Controls than ADT patients. Higher ReHo was observed in Controls in the left parietal-occipital area. Finally, ADT patients presented an increase of FC in more regions than Controls. These differences may reflect an impairment in brain functioning in ADT users.

www.nature.com/scientificreports/ rs-fMRI data is regional homogeneity (ReHo). ReHo characterizes the local synchronization of spontaneous fMRI BOLD signals and has been described as an index of local functional connectivity 25 . This method assumes that neighboring voxels within a functional cluster have similar temporal hemodynamic characteristics 26 . Finally, functional connectivity (FC) analysis is one of the most commonly used approaches to manage resting-state data. FC shows synchronization within different brain regions [16][17][18][19] corresponding to functionally relevant resting-state networks [27][28][29] .
The association between all these rs-fMRI measures and cognitive-affective functions has been extensively reported 30,31 . More specifically, many studies focusing on the effect of sex steroids in rs-fMRI measures have been published (for a review, see 32 ). However, only one of these addressed the changes of rs-fMRI in prostate cancer with ADT, where, as mentioned above, brain functional changes are expected after ADT onset. In this sense, the authors of this previous work described changes in the FC of the prefrontal cortex and its relationship with cognitive control measures after ADT exposure 33 .
Bearing in mind that the abovementioned previous work addressed only the changes in FC of the medial prefrontal cortex and that no studies focusing on other rs-fMRI measures have been reported in ADT recipients, we note that it would be useful to have a deeper analysis of rs-fMRI measures in ADT recipients. This may help to clarify the functional changes that are clinically evident or self-reported by the patients. Therefore, the aim of the present work is to describe the differences in rs-fMRI measures (ALFF, ReHo and FC) between a group of PC patients treated with ADT and a PC control group (not treated with ADT), both in the whole brain and in specific regions of interest with high expression of androgen receptors.

Methods
Patients. Forty-nine ADT patients (mean age 78.2 years [SD = 7.5]) and fifteen PC-non-ADT (control) patients (mean age 73.5 years [SD = 6.4]) were included in the study. As described in previous work 34 , all participants were right handed (using a Spanish version of the Edinburgh Handedness Inventory; http:// www. neuro psicol. org/ Proto col/ oldfi eld. pdf). The patients were selected from the PC database of the Department of Urology in Hospital Universitario de Canarias (Spain). They needed to have a diagnosis of PC with a clinical indication for ADT (leuproline, triptoreline or gosereline) and they needed to have been exposed to ADT for at least 6 months. Selected ADT patients were exposed to ADT for a mean of 42.6 months (SD = 35.9). Control patients had a diagnosis of PC of more than 6 months, but without indication for ADT. Patients included in the study did not present any other psychiatric, neural or systemic disease that may have modified rs-fMRI measures and cognitive assessment. None of the patients included in the study had a history of exposure to other antiandrogen drugs. The patients included in the study had also participated in another structural study performed by the authors 34 . Among the patients included in that work, one ADT patient was excluded because of excessive head movement during rs-fMRI acquisition.
Demographic features of the ADT and Control groups are shown in Table 1. Functional status, measured with the Eastern Cooperative Oncology Group (ECOG) scale, was slightly better in ADT patients than the Controls (p = 0.042; Table 1).
Written informed consent was explained and signed by the patients and the control subjects. The study was approved by the Ethics Committee of the Hospital Universitario de Canarias, according to the Declaration of Helsinki. www.nature.com/scientificreports/ Neuropsychological assessment. A neuropsychological evaluation to determine the presence of cognitive impairment was performed by a specialist with 10 years of experience in neuropsychology. This approach was performed as it was conducted in a previous study 34 . In brief, several cognitive domains were evaluated: verbal fluency (phonetic and semantic), visuospatial and visuoperception, processing speed, visual memory and verbal memory. The tests used for the evaluation of each cognitive domain are listed in supplementary table A.1. All assessments were performed in the morning (from 8:00 am to 12:00 am). In agreement with the recommendation of the International Cognition and Cancer Task Force (ICCTF), cognitive impairment was defined when the score of at least 2 tests was equal to or below − 1.5 standard deviations (SD), or 1 test with a score equal/below − 2.0 SD. The Chi-Square test was used to compare the presence or absence of cognitive impairment (using the previous described criteria) between the studied groups. All ADT patients and 80.0% of control patients (12) presented a CI according to ICCTF criteria (Table 1)  A whole-brain three-dimensional structural image was acquired for anatomical reference. A 3D fast spoiled gradient-recalled pulse sequence was obtained with the following acquisition parameters: TR = 10.4 ms, TE = 4.2 ms, flip angle = 20, matrix size = 512 × 512 pixels, 0.5 × 0.5 mm in plane resolution, slice thickness = 2 mm.
After checking the images for artefacts, rs-fMRI data were preprocessed using Statistical Parametric Mapping software SPM8 (Wellcome Trust Centre for Neuroimaging; http:// www. fil. ion. ucl. ac. uk/ spm/). The images were spatially realigned, unwarped, and normalized to the Montreal Neurological Institute (MNI) space using standard SPM8 procedures. The first 10 images were discarded to remove signal equilibration effects. After that, the source of spurious variance was removed through linear regression by including the signal from the ventricular system, the white matter and the whole brain, in addition to the six parameters obtained by rigid body head motion correction.
All these approaches have been also used in previous works 35-37 . ALFF analysis. ALFF analysis was performed in a similar way as previously described 37 . The rs-fMRI Data Analysis Toolkit (REST) version 1.8 was used to perform the analysis 38 . The signal was linearly detrended and a temporal band-pass filter was applied (0.01 Hz < f < 0.08 Hz). The filtered time series was transformed to a frequency domain with a fast Fourier transform (FFT) and the power spectrum was then obtained. The square root was calculated at each frequency of the power spectrum and the averaged square root was obtained across 0.01-0.08 Hz at each voxel. Each ALFF map was spatially smoothed with a Gaussian filter of 4 mm FWHM before statistical analysis. A two-sample t test was then performed on the ALFF maps to compare the two studied groups. Age was included as regressor of no interest to reduce the variance unrelated to the variable of interest. As the aim of the study was to study the differences between the two groups in the whole brain, a combined voxel and cluster-size thresholding approach to correct for multiple comparisons was performed. In this reagard, Monte Carlo Simulations using the AlphaSim software (included in REST), with 10,000 iterations and a voxel-wise p = 0.001, were run to determine the minimum cluster size needed to correct for Type I errors. Simulations for ALFF analysis indicated that a minimum cluster size of 16 contiguous voxels was needed to accomplish a corrected alpha of p = 0.05. (that is, 16 contiguous voxels would occur less than 5% of the time by random noise alone assuming a group of highly significantly activated voxels set at an individual voxel-wise threshold of p = 0.001).
ReHo analysis. The ReHo maps were also generated using the REST toolkit. The signal was linearly detrended and a temporal band-pass filter was applied (0.01 Hz < f < 0.08 Hz). Kendall's coefficient of concordance (KCC) was used to measure the similarity of the time series within a functional cluster based on the regional homogeneity hypothesis 26 . The individual ReHo maps were generated in a voxel-wise fashion, with the 27 nearest neighboring voxels defined as a cluster. The ReHo maps were divided by their own KCC value within the mask for standardization 39 . Finally, the results were included in a two-sample t test to compare the two groups. Age was included as a regressor of no interest to reduce the variance unrelated to the variable of interest. Monte Carlo Simulations (AlphaSim, 10,000 iterations, voxel-wise, p = 0.001) were run to determine the minimum cluster size needed to correct for Type I errors 39 . Simulations for ReHo analysis indicated that a minimum cluster size of 7 contiguous voxels was needed to accomplish a corrected alpha of p = 0.05.

Region of interest (ROI) analysis.
In addition to the voxel-wise analysis for ALFF and ReHo, a region of interest (ROI) analysis was also conducted in cortical and subcortical regions that had the highest expression of the androgen receptor (AR). These regions were identified using RNAseq data provided by The Allen Human Brain Atlas platform (http:// human. brain-map. org/). In brief, AR expression data for five male subjects was downloaded. Only the probe of the Agilent 44 k was considered. The mean of the AR RNA expression in each considered region was calculated for the five subjects. The top-5 regions with high AR expression were selected for ROI analysis (supplementary table A.2). The selected regions were extracted from the Automated anatomical labelling atlas 3 (ALLv3) 40 . It should be noted that slight differences might exist in region-delimitation between the ALLv3 atlas and the one performed by the Allen Human Brain Atlas. Two ROIs for each region (one for each brain hemisphere) were included in the analysis. ROIs data was extracted using the MarsBaR 0.44 toolbox (http:// marsb ar. sourc eforge. net/). The two-sample t test was performed to compare ALFF and ReHo data in each www.nature.com/scientificreports/ ROI between the two groups. Statistical significance was considered when the corrected p-value, using False Discovery Rate, was below 0.1 (FDR < 0.1).

FC analysis.
Functional connectivity (FC) between proximal or distant brain regions can be inferred from inter-regional cross-correlations of the BOLD signal at rest 41 . As described above, using the REST toolkit, the signal was linearly detrended and a temporal band-pass filter was applied (0.01 Hz < f < 0.08 Hz). FC analysis was performed using the previously defined ROIs. This analysis estimated the FC between each ROIs and the rest of the brain (cerebellum and grey matter, including subcortical nucleus). Individual z-score maps were obtained and the two-sample t test was then performed. The statistical significance threshold was also set using the Monte Carlo Simulations (AlphaSim, 10,000 iterations, voxel-wise, p = 0.001). Simulations for FC analysis indicated that a minimum cluster size of 6 contiguous voxels was needed to accomplish a corrected alpha of p = 0.05.
Ethical approval. The study was approved by Hospital Universitario de Canarias ethics committee. The study was carried out in accordance with the declaration of Helsinki.  (Fig. 1, Table 2). The opposite contrast (i.e., ADT patients > Controls) did not show any significant difference with the selected threshold.

Results
Region of Interest (ROI) analysis in ALFF maps showed a higher significant activity for Controls than ADT patients in the left IFG (pars opercularis) and in the left PreCG (Fig. 2, supplementary table A.3). The right PreCG also showed higher activity in Controls, but the corrected p-value did not reach statistical significance (FDR = 0.128). On the contrary, a higher ALFF in the right hippocampus and the lateral geniculate nucleus of the right thalamus was identified in ADT patients than Controls (Fig. 2, supplementary table A.3). The relationship between the scores of the neuropsychological tests and the ALFF in the most significant ROIs is shown in supplementary Figs. 1-7 (see supplementary material).

ReHo analysis. The comparison of ReHo maps between Controls and ADT patients showed a higher ReHo
in the control group in the left superior occipital gyrus (Fig. 3, Table 2). No significant differences were identified in the contrast ADT > Controls. Furthermore, no significant differences were identified between both groups of patients in the ROI analysis (supplementary table A.3).

Functional connectivity analysis.
A comparative analysis of FC in regions with high AR expression was performed between Controls and ADT patients. ADT patients showed a higher FC than Controls for many of the selected ROIs, with many other brain regions (Table 3). In other words, a higher correlation in resting-state brain activity between regions with higher AR expression and other brain areas was found in the ADT patients than in the Controls. This difference was more evident in the FC of the ventral-lateral nucleus of both thalami (largest clusters), which showed an FC with the prefrontal area (middle frontal gyrus), temporal regions (fusiform gyrus and superior temporal gyrus) and cerebellum (Table 3). Lateral geniculate nucleus of both thalami also showed higher FC with frontal areas in ADT patients, as well as both precentral and paracentral gyri which presented higher FC with close frontal-parietal areas (Table 3).
On the other hand, Controls showed higher FC than ADT patients for only a small number of regions and the main difference was located at the right subcallosal anterior cingulate cortex (ACC) ( Table 3).

Discussion
The present rs-fMRI work, using different approaches, has compared the resting-state brain activity between PC patients exposed to ADT and PC patients not exposed to ADT (Controls). On the one hand, higher ALFF and ReHo was identified in the Controls than in ADT patients in fronto-temporal and occipital areas, respectively. Furthermore, ROI analysis in regions with higher AR expression showed that some of these regions showed higher ALFF in the Controls than in ADT patients. On the other hand, the FC analysis showed a higher FC of high-AR-expression regions with other brain areas in the ADT group than in the control group. All these findings will be discussed below.
Changes in ALFF and ReHo associated with ADT. As mentioned above, androgen exhibits a considerable influence on human behavior through the modulation of brain structure and function. Consequently, ADT may alter the brain function and this change may alter fMRI findings. In this sense, Cherrier et al. showed that ADT patients when compared to control patients (non-prostate cancer patients), after 9 months exposure to ADT, presented lower fMRI activity in the right parietal-occipital area during tasks involving manipulation and recall of spatial information 42 . Furthermore, Chao et al. described a significant decrease of brain activation during a cognitive control task in the medial prefrontal cortex, right insula and right middle/inferior frontal gyri after 6 months with ADT 33 . These results agree with another work that, using positron emission tomography (PET) techniques, demonstrated an increase in cerebral glucose metabolism during a mental rotation task in occipital and frontal regions after 12 weeks of testosterone treatment in hypogonadal patients 43 . In other words, the brain of males with low androgen levels shows a lower activation during cognitive tasks. Nonetheless, the present work is focused on resting state measures (i.e., without performing any task www.nature.com/scientificreports/ in mind, another longitudinal PET study described a positive correlation between free testosterone levels and regional cerebral blood flow (rCBF) in the hippocampus and frontal regions (right IFG and anterior cingulate gyrus) 44 which may be associated with rs-fMRI changes found in the present work. More interestingly, a previous work, focusing on transgender people and using the same measures as the present study (i.e., ALFF and ReHo measures), demonstrated a shift of ALFF and ReHo related to androgen levels in the frontal cortex (mainly the IFG, PreCG and premotor cortex), medial temporal cortex (mainly in the parahippocampal gyrus) and in the cerebellum 39 which are compatible with the differences between controls and ADT patients described in the present work. In the present work, patients exposed to ADT presented a decrease ALFF during rs-fMRI in various frontal areas. More specifically, the left primary motor cortex, the premotor cortex bilaterally and the IFG bilaterally showed a higher ALFF in Controls than in ADT patients. This higher activation of frontal regions seems to be compatible with the results reported in previous studies 33,39,[42][43][44] . Furthermore, the ROI analysis showed that, apart from the presence of ALFF differences in frontal areas, the right hippocampus and the right thalamus also showed significant differences. However, in these regions, ALFF was higher for ADT patients than Controls. This www.nature.com/scientificreports/ finding may indicate that the effect of testosterone in BOLD signal varies between different brain regions. In this regard, testosterone levels and AR activation in frontal areas (neocortex) would lead to an ALFF increase, while in the hippocampus (allocortex) or subcortical nuclei (thalamus) would lead to an ALFF decrease. Considering the findings in the ReHo analysis, where higher homogeneity in parietal-occipital areas for control patients was shown, ADT seems to alter the functional status of the occipital cortex. This result is also supported by previous works. Indeed, the reduced parieto-occipital activation during manipulation and recall of spatial information tasks in ADT patients reported by 42 showed that testosterone levels influence occipital cortex functioning. Furthermore, further evidence supporting this finding is the positive relationship between testosterone levels and cerebral glucose metabolism during a mental rotation task in occipital areas 44 .
In summary, although the present study is the first study focusing on ADT patients describing differences in ALFF and ReHo measures, the findings here are compatible with those reported in previous neuroimaging works focused on the effect of androgens on brain function. Other cancer patients (e.g. breast or lung cancer) showed also changes in ALFF and/or ReHo related to the use of chemotherapeutic agents 45,46 . Nevertheless, the finding presented here cannot be associated with chemotherapy, because it was not used in any of the selected patient before their inclusion in the study.
ADT patients showed more diffuse FC than controls. FC reflects the relationship between different brain regions and measures how well individual brain regions activate in a concerted manner as an important index of the integrity of brain functions. Androgens have previously been shown to modulate brain connectivity. In this respect, higher testosterone levels in boys were associated with reduced FC between the amygdala and the orbito-frontal cortex 47 . This finding has also been reported elsewhere in adults 48 . The administration of a single dose of testosterone leads to a reduction in frontal-subcortical and frontal-parietal functional connectivity that potentially contributes to impaired emotion processing and regulation 49 . More evidence about the reduction of FC related to testosterone comes from anabolic steroid users. Reduced FC between the superior frontal gyrus and the dorsal attention network as well as between the amygdala and the default mode network has been described in anabolic steroid users 50 . However, the only study that performed an FC analysis in ADT patients, described a decrease in the FC of the medial prefrontal cortex in ADT patients 33 .
The main finding in the FC analysis in the present study was the presence of higher FC in ADT patients than in Controls in most of the selected ROIs. As described in the Methods section, the selected ROIs were those brain regions that have shown the highest AR-RNA expression, thus they are putatively the regions that may be more sensitive to the lack of circulating androgens. The thalamus (ventral-lateral and lateral-geniculate nuclei) and primary motor regions (PreCG and paracentral lobule) were the areas that showed more FC for ADT patients. On the contrary, the subcallosal region of the anterior cingulate cortex showed a significantly higher FC in Controls than ADT patients. As shown in the ALFF analysis, the effect of androgens in FC seems to depend on the ROI considered. Generally speaking, the present results agree with previous publications that described a general attenuation of FC with higher levels of testosterone [47][48][49][50] . The development of the anterior cingulate Table 2. Significant group differences in amplitude of low frequency fluctuation (ALFF) and regional homogeneity (ReHo). Listed regions are those which survived correction for multiple comparison (p-corrected < 0.05). Coordinates are MNI coordinates. www.nature.com/scientificreports/ cortex, involved in emotion and motivation and belonging to the limbic system, is known to be influenced by sex hormones 51 . In regions that belong to the limbic system, androgens seem to induce the increase of FC with other brain regions and this might explain the anxiolytic and antidepressant effects that have been reported for testosterone 52 . Limitations. The main limitation of the present study is its case-control study nature. Therefore, a longitudinal study would be preferable to confirm that the described functional changes are secondary to ADT use and the temporal relationship between these changes and the onset of cognitive decline (measured by standard cognitive assessment). In this regard, if the rs-fMRI changes come early than cognitive decline measured by standard tests, the rs-fMRI information would help the clinician to make changes in the patient`s management of the (e.g. cognitive rehabilitation or changes in the ADT posology). Furthermore, larger cohorts of ADT and non-ADT patients are also recommended to obtain stronger evidence and the effect of other drugs that are currently used in castration-resistant PC (e.g. Enzalutamide or Abiraterone) which target the AR or the testosterone synthesis, respectively, should also be analyzed in prospective studies.
Another limitation of the present work is that the ADT group showed an older age distribution and the effect of the variability in preprocessing and statistical analysis performed here compared to other neuroimaging studies cannot be estimated.
Finally, the clinically-significant ALFF or ReHo changes cutoff and the reversibility of the rs-fMRI changes once the ADT has been stopped should also be studied in longitudinal-prospective studies.

Conclusion
Differences in rs-fMRI measures have been identified between ADT patients and Controls. Firstly, higher ALFF was identified in Controls than in ADT patients in frontal-temporal regions, although ALFF seems to increase in the right hippocampus and right thalamus in ADT patients. ReHo analysis showed a higher homogeneity in parietal-occipital areas in Controls than in ADT patients. Finally, FC was generally higher in ADT patients compared to Controls.   Table 3. Significant group differences in functional connectivity. Listed regions are those which survived correction for multiple comparison (p-corrected < 0.05). Coordinates are MNI coordinates.