Changes in local brain function in mild cognitive impairment due to semantic dementia

Abstract Aims Mild cognitive impairment due to semantic dementia represents the preclinical stage, involving cognitive decline dominated by semantic impairment below the semantic dementia standard. Therefore, studying mild cognitive impairment due to semantic dementia may identify changes in patients before progression to dementia. However, whether changes in local functional activity occur in preclinical stages of semantic dementia remains unknown. Here, we explored local functional changes in patients with mild cognitive impairment due to semantic dementia using resting‐state functional MRI. Methods We administered a battery of neuropsychological tests to twenty‐two patients with mild cognitive impairment due to semantic dementia (MCI‐SD group) and nineteen healthy controls (HC group). We performed structural MRI to compare gray matter volumes, and resting‐state functional MRI with multiple sub‐bands and indicators to evaluate functional activity. Results Neuropsychological tests revealed a significant decline in semantic performance in the MCI‐SD group, but no decline in other cognitive domains. Resting‐state functional MRI revealed local functional changes in multiple brain regions in the MCI‐SD group, distributed in different sub‐bands and indicators. In the normal band, local functional changes were only in the gray matter atrophic area. In the other sub‐bands, more regions with local functional changes outside atrophic areas were found across various indicators. Among these, the degree centrality of the left precuneus in the MCI‐SD group was positively correlated with general semantic tasks (oral sound naming, word‐picture verification). Conclusion Our study revealed local functional changes in mild cognitive impairment due to semantic dementia, some of which were located outside the atrophic gray matter. Driven by functional connectivity changes, the left precuneus might play a role in preclinical semantic dementia. The study proved the value of frequency‐dependent sub‐bands, especially the slow‐2 and slow‐3 sub‐bands.


| INTRODUC TI ON
Semantic dementia (SD) is a subtype of frontotemporal lobe degeneration, 1 which is the leading cause of dementia in people under 65 years of age, especially in middle age. 2 Because the typical damage is in the anterior temporal lobe, SD can also be regarded as a temporal variant of frontotemporal lobe degeneration. The main feature of SD is the loss of semantic memory domains, 1,3 but some patients initially present with personality and behavioral changes, such as apathy and disinhibition; the differences in SD symptoms are caused by lateralization. 4 Typical SD symptoms often correspond to left anterior temporal lobe atrophy, although in some patients, atrophy begins on the right side. 5,6 This diversity of the initial symptoms makes it challenging to recruit and study patients with preclinical SD.
Local pathophysiological changes in neurodegenerative diseases exist before the emergence of severe clinical symptoms. Previous studies have found tau-related pathological damage in the preclinical stage of frontotemporal lobe degeneration 7 and decreased local metabolism in mild cognitive impairment. 7 Mild cognitive impairment is the preclinical stage of dementia; at this stage, cognitive decline is evident but it has not yet reached the criteria of dementia. 8 Some patients diagnosed with mild cognitive impairment exhibit impaired semantic function, and during subsequent follow-up are diagnosed with SD. These patients with mild cognitive impairment due to SD are suitable for the study of preclinical SD.
Neuroimaging plays a vital role in diagnosis. In the current diagnostic criteria, clinical symptoms and neuroimaging examination are both necessary, in which neuroimaging examination needs to have one of the following manifestations: (a) MRI dominant significant anterior temporal lobe atrophy; or (b) single-photon emission computed tomography (SPECT) or PET dominant anterior temporal lobe hypoperfusion or hypometabolism. 3 Therefore, MRI has always been an essential tool for studying SD. 9 Structural MRI studies have revealed the extent of gray matter (GM) atrophy and its impact on the occurrence and development of disease. [10][11][12] Moreover, functional connectivity studies using resting-state functional magnetic resonance imaging (rs-fMRI) have revealed the changes in brain network. 10,13 However, there are few studies on the changes in local brain activities. It is not clear whether there are changes in regional brain function in patients with SD, and whether the areas in which these changes exist are consistent with the known GM atrophy. Local functional indicators of rs-fMRI can evaluate spontaneous neural activity from different aspects. The amplitude of low-frequency fluctuation (ALFF), fractional amplitude of low-frequency fluctuation (fALFF), percent amplitude of fluctuation (PerAF), and wavelet-based amplitude of low-frequency fluctuation (Wavelet-ALFF) 14,15 can describe the amplitude of local spontaneous nerve activity. Regional homogeneity (ReHo) can describe the consistency of spontaneous neural activity in local brain regions and adjacent brain regions. 16,17 Degree centrality (DC) is used to describe the importance of local brain areas in global networks. 18 Studies using rs-fMRI have primarily focused on the fluctuation of blood oxygen level-dependent (BOLD) signal in the lowfrequency band (0.01-0.08 Hz). [19][20][21] However, such a relatively wide frequency band may not be sensitive enough due to noise or other reasons. 22,23 Therefore, researchers have divided the BOLD signal into four sub-bands according to frequency: slow-5 (0.01-0.027 Hz), slow-4 (0.027-0.073 Hz), slow-3 (0.073-0.198 Hz), and slow-2 (0.198-0.25 Hz). These sub-bands showed specific changes in different diseases. 24,25 In studies of some neurological or psychiatric diseases, the combination of different frequency bands has provided more refined results. [26][27][28][29] In this study, we selected patients who were diagnosed with semantic impaired mild cognitive impairment at the first visit and were diagnosed with SD in the subsequent follow-up to obtain a preclinical SD study cohort. To increase the effectiveness of the study, we used different frequency bands combined with a variety of local functional indicators to reflect local brain function. Specifically, we conducted a voxel-based morphometry (VBM) analysis and an exploratory rs-fMRI analysis in patients with mild cognitive impairment due to SD. The rs-fMRI sub-bands included the main valuable frequency bands, that is, the normal band, slow-2 band, slow-3 band, slow-4 band, and slow-5 band. The fMRI indicators included ALFF, fALFF, PerAF, Wavelet-ALFF, ReHo, and DC. The goals of this study were to explore the changes in local brain function in patients with mild cognitive impairment due to SD, to analyze the relationship between these changes and GM atrophy, and to evaluate the value of the combination of sub-bands and multiple indicators. We hope that the results will help us to understand the characteristics of local brain functional changes in preclinical SD and aid in further longitudinal studies and subgroup analyses. frequency-dependent, mild cognitive impairment, resting-state functional MRI, semantic dementia cognitive impairment with impaired language ability at their first visit and were diagnosed with SD in the subsequent (at least one year later) follow-up. All participants took part in the study at the outpatient department of Huashan Hospital and were recruited from 2013 to 2017. Participants underwent neuropsychological tests and brain MRI examinations. The inclusion criteria were as follows: (a) at the first visit, the Chinese version of the MMSE 30 score ranged from 24 to 28, showing a decline in semantic function, with no obvious defect in professional ability, social ability, and living ability; (b) during subsequent follow-up, the patient developed a worsening naming disorder and a word comprehension problem, with at least the following three characteristics: lack of object knowledge, surface dyslexia, less repetition ability, and less pronunciation ability, and hit the diagnostic criteria for SD 3 ; and (c) manifestation of temporal lobe atrophy on MRI supporting the diagnosis of SD. 3 Half of the cases completed the fluorodeoxyglucose (FDG)-PET scan. The exclusion criteria were as follows: (a) patients diagnosed with non-fluent/semantic variants of primary progressive aphasia (PPA) and logopenic PPA, or other types of dementia were excluded; (b) patients with a history of head injury, head surgery, neurological or mental diseases, or severe visual or hearing impairment; and (c) patients who could not undergo MRI due to metal implants or other reasons. 3D structural MRI was used for imaging diagnosis to exclude patients with severe brain diseases such as brain tumors, acute cerebral hemorrhage, cerebral ischemia, or non-degenerative brain injury. 3D structure MRI was also visually inspected to evaluate the atrophy degree of the left and right temporal lobes by experienced physicians. In the MCI-SD group, the number of patients with left atrophy and right atrophy was equal.

| Behavioral data collection
We used the following six general semantic tasks: (a) Oral picture naming: participants were instructed to verbalize the name of an object whose picture was presented on a screen. (b) Oral sound naming (SN): a sound was played through headphones, and the participants were instructed to verbalize the name of the object whose sound

| Resting-State fMRI imaging data
The data were processed using SPM12 and RESTplus (http://restf mri.net/forum/ restplus) toolkits. 33 The first 10 time points were removed to allow the participant to adapt to the scanning noise and avoid the non-equilibrium effects of magnetization. Slice timing adjustment and realignment for the correction of head motion were also conducted. The threshold of excessive head movement was defined as being when the head rotated more than 3° or shifted more than 3 mm in any direction during the entire scanning process. All participants were below this threshold and were included in the whole process. In the next step, through linear transformation, the individual structural images were coregistered to the mean functional image after head-motion correction. The transformed structural images were segmented into GM, WM, and CSF using a The ALFF, fALFF, PerAF, and Wavelet-ALFF were standardized by dividing the value of each voxel by the global average. Finally, the standardized mALFF, mfALFF, mPerAF, and mWavelet-ALFF graphs were spatially smoothed using a Gaussian kernel (FWHM = 4 mm). We calculated the indicators of the five frequency bands defined above.

| Computing ReHo and DC
We obtained the ReHo by calculating the Kendall coordination coefficient of the time process for each of the 27 nearest neighboring voxels. Standardization and spatial smoothing were carried out in the same way as the above indicators.
DC represents the sum of weights that showed node strength with a given voxel in the weighted graphs. For each voxel, the BOLD time course was extracted, and the Pearson correlation coefficients with every other voxel in the brain were calculated. A matrix of Pearson's correlation coefficients between a given voxel and all other voxels was obtained to construct the whole-brain functional connectivity matrix for each voxel. An undirected adjacency matrix was then obtained by setting a threshold to each correlation at an r value more than 0.25. Then, the weighted DC was calculated as the sum of correlations exceeding this threshold. The DC value of each voxel was divided by the global mean of the DC values for standardization. Finally, the resulting matrices were smoothed with a Gaussian kernel (FWHM = 4 mm) to enable group comparisons.

| Statistical analysis
We used SPSS (IBM SPSS Statistics, Version 26.0. IBM Corp) software for statistical analysis of demographic data and neuropsychological scale scores. Data were tested for normality using a Shapiro-Wilk normality test. Age, education years, and MMSE scores conformed to a normal distribution, and a two-sample t-test was used to analyze the differences between groups; the chi-square test was conducted for sex. The semantic battery (including Oral picture naming, Oral sound naming, Picture associative matching, Word associative matching, Word-picture verification, Naming to definition, Boston naming test, Animal fluency test, Symbol digit modalities test, Shape trials test-A, Shape trials test-B, Complex figure test copy, Complex figure test recall) scores did not conform to the normal distribution, and these data were analyzed using nonparametric tests. The Mann-Whitney U test was conducted to compare the difference in semantic task scores between the two groups. Spearman's rank correlation was used to estimate the correlation between fMRI indicators and semantic task scores.
To compare the volume of GM between the two groups, Data Processing and Analysis of Brain Imaging version 4.0 (DPABI, http://rfmri.org/DPABI) was used to carry out a two-sample t-test between the SD and HC groups. To examine the between-group differences of ALFF, fALFF, PerAF, Wavelet-ALFF, ReHo, and DC, two-sample t-tests were conducted between the SD and HC groups using DPABI v4.0. The mean relative displacements of the GM map and mean framewise displacement were used as covariates to reduce the influence of mixed variables in the statistical analysis. 34 Each voxel's gray matter volume (GMV) was used as a covariate to reduce the influence of GM atrophy. Multiple comparison correction was performed based on the Gaussian random field theory (GRF; voxel-wise, p < 0.001; cluster-wise, p < 0.05, two-tailed).

| Neuropsychological performances
There were no statistical differences in age, sex, and education between the SD and HC groups.
The MMSE score of the MCI-SD group was significantly lower than that of the HC group (

| Local functional changes in sub-bands
We first carried out an analysis of the ordinary frequency normal band. We then analyzed the four sub-bands according to frequency. Because the slow-4 and slow-5 bands coincide with the ordinary band to some extent, whereas the slow-2 and slow-3 bands are basically outside it, we divided the four sub-bands into two groups.

| The normal band
In the normal band, decreased fALFF in the MCI-SD group was ob-  Figure 1). All the regions were located either in or adjacent to the areas of GM atrophy. There was no difference in ALFF, PerAF, Wavelet-ALFF, DC, and ReHo between the two groups.

| Slow-4 and slow-5 bands
In the MCI-SD group, the brain regions with local functional changes began to spread in the slow-4 and slow-5 bands, and more indicators showed changes in these two sub-bands. It is noteworthy that some indicators changed in regions distant from the areas of GM atrophy.  Figure 1). ReHo was decreased in the right triangular inferior frontal gyrus, the right postcentral gyrus, and the right superior marginal gyrus in the MCI-SD group compared to that of the HC group (GRF corrected, cluster >29voxels, FWHMx 8.72 mm, FWHMy 9.19 mm, FWHMz 8.73 mm, voxel-wise p < 0.001, cluster-wise p < 0.05, two-tailed. Table 2. Figure 1).

Slow-5 band
In the MCI-SD group, ReHo was decreased in the left superior pari-   Figure 1).

| Slow-2 and slow-3 bands
In the slow-2 and slow-3 bands, the MCI-SD group showed more brain regions with local functional changes on a further increased number of indicators. In these two sub-bands, more brain regions distant from the atrophic areas showed local functional changes, and these were primarily distributed in the parietal and occipital lobes.
It is noteworthy that in the slow-3 band, all the local functional indicators used in this study showed changes in multiple brain regions.
Moreover, in the slow-3 band, the DC of the left precuneus, which is distant from the atrophic areas, showed a correlation with semantic function.

Slow-2 band
In the slow-2 band, there were differences in some brain regions between the two groups in ALFF, fALFF, Wavelet-ALFF, ReHo, and DC. In the MCI-SD group, ALFF was decreased in the left inferior parietal angular gyrus (GRF corrected, cluster >23voxels,  Figure 1). Wavelet-ALFF was decreased in the left inferior parietal lobule (extending to the left TA B L E 3 The regions of difference between the two groups in the slow-2 and slow-3 bands Note: All results were corrected for multiple comparisons (GRF, voxel-wise p < 0.001, cluster-wise p < 0.05, two-tailed).

| DISCUSS ION
We reviewed the mild cognitive impairment stage of patients diagnosed with SD to establish a study cohort of mild cognitive impairment due to SD and study preclinical SD. Analysis of the patients' neuropsychological scales showed that they had cognitive decline limited to semantic function. We used VBM analysis to obtain the areas of GM atrophy and used GMV as a covariate in the subsequent rs-fMRI analysis to eliminate the effect of GM atrophy on local brain function. In the rs-fMRI analysis, we used an exploratory approach that combined frequency-dependent subbands with multiple indicators. Our study showed that there were more local functional changes in the sub-bands than in the normal band, especially in the slow-2 and slow-3 bands. Some brain regions with local functional changes identified in the sub-bands were distant from the areas of GM atrophy. Among these regions, the left precuneus had a close relationship with semantic functions, and we think this relationship was likely driven by changes in network connectivity.

| Neuropsychological performance
The comparison of the neuropsychological tests showed that the performance of the MCI-SD group in the general semantic tasks and semantic memory tasks was worse than that of the HC group.
However, there was no significant difference in attention, executive function, spatial ability, and non-verbal episodic memory between the two groups. These results showed that the decline of cognitive function in patients with mild cognitive impairment due to SD was primarily caused by semantic impairment, while the function in the other cognitive domains was roughly retained.

| Functional changes in the atrophic areas
We identified local functional changes in multiple regions. Some of these regions, mostly located in the temporal lobe, frontal lobe, and cingulate gyrus, were in or adjacent to the areas of GM atrophy. The temporal pole is regarded as a critical area of abnormal connection in SD. 35,36 The orbitofrontal gyrus, which is close to the temporal pole and receives its signal input, is also a typical area of GM atrophy and reduced metabolism in SD. 4,37 Previous studies have confirmed that disconnection of the frontal lobe in the semantic network is one of the vital connectivity changes in SD. 38 There are few studies on the cingulate gyrus in SD to this end.
Cingulate gyrus dysfunction exists in a wide range of neurological and mental disorders. Some diffusion tensor imaging studies have reported that the structure of the cingulate gyrus in patients with frontotemporal dementia is damaged. 39,40 In a previous fMRI study, the connectivities of the limbic system, including the cingulate gyrus, were significantly decreased in patients with frontotemporal dementia. 41 In frontotemporal dementia, there is a significant weakening of the salient network, which connects the frontal lobe and limbic system, and is characterized by communication between the anterior cingulate gyrus, insular, striatum, and amygdala. 42 These studies primarily focused on the behavioral variant frontotemporal lobe degeneration subtype. However, in a recent study of the composition of the whole-brain language network, 43 the cingulate gyrus (including anterior and posterior) was found to play an important role.
It is generally believed that GM atrophy in SD is more significant on the left side 11,12 ; however, in our study, the local functional changes in the atrophic area were more significant on the right side.
After visually inspected the temporal lobe atrophy of structural MRI, in the MCI-SD group, the number of patients with left atrophy and right atrophy was equal. Unlike the predominance of left SD in most SD cohorts in the past, our preclinical SD cohort did not have obvious laterality. Therefore, local functional changes were found on both the left and right sides. Considering that GMV was used as a covariate in the rs-fMRI statistical process, we think this may indicate that after excluding the laterality of GM atrophy, the local functional changes on the right side are more obvious than those on the left side in the GM atrophy area.

| Functional changes distant from the atrophic areas
Local functional changes also occurred in some regions distant from the GM atrophic areas, indicating that they were not caused by GM atrophy. These regions were found in different sub-bands with various indicators.
In the slow-4 and slow-5 bands, there were local functional changes in the right cuneus, right precuneus, and bilateral supplementary motor area (fALFF decreased in the slow-4 band), and the left inferior parietal lobe (ReHo decreased in the slow-5 band).
As a region of visual cortex, 44 the cuneus is also a part of a largescale network. This network integrates articulatory, auditory, and visual areas; its role is to produce, listen, and read lists of words. 45 In previous studies, supplementary motor areas were considered to be related to motor language 46,47 and were involved in orthographic, phonological, and lexical-semantic tasks. 48 The role of supplementary motor areas in language function may be achieved through the connection with the superior temporal gyrus. 7 The parietal lobe is an important part of the temporal/parietal/occipital module, which is an integral part of the network topology changes in SD and is related to semantic functional impairment. 49 A meta-analysis also showed that the left inferior parietal lobe engages in social cognition and language function, and the theory of mind and language-related processing facets are inextricably linked to this region. 50 In the slow-2 and slow-3 bands, the local functional changes in non-atrophic areas were further extended to the left superior parietal gyrus (ALFF decreased in the slow-2 band), right inferior occipital gyrus (fALFF decreased in the slow-2 band), and precuneus (DC decreased in the slow-3 band).
It was previously considered that the visual word form area is located in the occipitotemporal sulcus. 51 However, an fMRI study recently found that the inferior and middle occipital gyri responded more strongly to Chinese characters than to visual images; this region plays a critical role in processing and representing the category information for words. 52 Our study showed that the change was in the right inferior and middle occipital gyrus, which differs from the result of their study, which showed it was bilateral but mainly on the left side. The reason for this difference should be studied further in a subgroup study of left and right SD to determine whether it is related to the laterality of SD. There have been few studies regarding the superior parietal gyrus and semantic function. An fMRI study showed that this region is involved in the sublexical conversion of orthographic input into phonological codes in naming Chinese. 53 Thus, the role of this region in SD should be explored further.
In the MCI-SD group, the ReHo of the slow-3 band increased in the right temporal lobe. The neuropathological hub in SD patients is Abbreviations: ALFF, amplitude of low-frequency fluctuation; DC, degree centrality; fALFF, fractional amplitude of low-frequency fluctuation; PerAF, percent amplitude of fluctuation; ReHo, regional homogeneity; Wavelet-ALFF, wavelet-based amplitude of low-frequency fluctuation.

TA B L E 4 Summary of sub-bands and indicators combination
on the left temporal lobe. Therefore, in the temporal lobe, as a critical area of semantic function, may appear functional compensation on the right side. ReHo reflects the consistency of local neural activity and does not depend on the number of remaining neurons and the activity intensity. Therefore, the increase of ReHo in the right temporal lobe may be a compensatory manifestation.

| The left precuneus and semantic function
Among all the regions in which local functional changes were observed distant from the atrophic areas, the most important finding was the left precuneus. The precuneus is one of the main centers of the brain. It is an important node of the default mode network, but there are few studies on its relationship with SD, which may be caused by its location is hidden and far away from the typical SD gray matter atrophy areas. Our study found that the precuneus showed decreased fALFF in the slow-4 band and decreased DC in the slow-3 band, suggesting that it was affected in mild cognitive impairment due to SD.
The correlation analysis found that the DC of the left precuneus was positively correlated with WPV and SN, and the correlation with SN survived after Bonferroni correction.
Some studies have suggested that the precuneus is involved in extended face recognition 54,55 and participates in facial emotion perception and context integration. 56 Based on a new neural network analysis model, 57 researchers found that the abnormality of precuneus combines with some risk genes has a good recognition ability for MCI and AD. 58,59 A previous study showed that a decrease in confrontation naming task scores in patients with early frontotemporal lobe degeneration was associated with decreased fMRI signals in the precuneus. 60 The study reported that the effect was similar to that observed in patients with Alzheimer's disease and mild cognitive impairment, but the difference was not related to GM atrophy in the precuneus in patients with early frontotemporal lobe degeneration. Therefore, they hypothesized that default mode network inactivation was not related to local atrophic pathological changes, but to decreased connectivity. DC is a local indicator of connectivity, so our results verified this hypothesis. SD is a suitable model for semantic research, and the relationship between the precuneus and semantic function is worthy of further study in patients with SD. Furthermore, our results suggest to focus on DC in the slow-3 band.

| Value of the combination of sub-bands and multiple indicators
We used the slow-2, slow-3, slow-4, and slow-5 sub-bands in addition to the ordinary frequency band (ie, the normal band). The frequency range of each sub-band is narrower than that of the normal band, while the total range is far greater. In the normal band, only limited regions with decreased fALFF values were observed in the atrophic areas; however, when the combination of sub-bands and multiple indicators was added to the analysis, more local functional changes were observed ( Table 4). however, it has also been proposed that it only plays the role of phonetic representation rather than language comprehension. 63 If future studies use rs-fMRI in SD queues to explore this area further, our results suggest that focusing on Wavelet-ALFF in slow-2 and slow-3 bands may be useful. Wavelet-ALFF is different from other fMRI indicators in that it is obtained by wavelet transform instead of the fast Fourier transform. Wavelet-ALFF has strong adaptability to the local characteristics or instability of data and has better sensitivity and repeatability in the higher frequency band than in conventional frequency bands. 15 In the slow-3 band, PerAF decreased in the frontal lobe and limbic system. PerAF is an indicator that reflects the percentage fluctuation of rs-fMRI and has good repeatability. 14 According to our results, PerAF may be more suitable for analysis in slow-3.
Both the slow-2 band and the slow-3 band showed ReHo changes in the temporal lobe. ReHo decreased in the right middle temporal gyrus in the slow-2 band but increased in the left superior temporal gyrus and middle temporal gyrus in the slow-3 band. The results showed lateralization in the uniformity of neural activity in the temporal lobe. ReHo in some brain regions of the WM showed decreases in the slow-2 band. This result was consistent with the fact that slow-2 is more sensitive to WM than other frequency bands. 24 In summary, we think that the combination of sub-bands and multiple indicators is valuable. At present, diseases with brain atrophy or other structural changes, including frontotemporal lobe degeneration, 64,65 AD, 66 Parkinson's disease, 67 and cerebral microvascular disease, 68 mainly rely on structural MRI for diagnosis and differentiation. For the early stages of these diseases, such as the MCI stage, studies using RS-fMRI mainly focused on changes in functional connectivity, [69][70][71] and there are few studies on changes in local functional activities. We speculate that this may be due to the relatively slight pathological changes in the early stage of the disease, which leads to the insensitivity of the local functional indicators to reflect the nerve damage. A recent study on the interaction between gut microbiota with the local brain activity of amnestic MCI patients found different interaction types in different sub-bands. 72 This study supports our results. We believe that the sub-bands combined with multiple indicators can make rs-fMRI more sensitive and can reflect more changes. It may be worth promoting in the study of local functional activities in the early stages of brain diseases.

| CON CLUS ION
Patients with mild cognitive impairment due to SD showed mild cognitive decline caused by semantic impairment and were consid-

| LI M ITATI O N S
This study had some limitations. Left and right SD subgroup analyses could not be performed because the sample size was small.
Therefore, although we found that some left and right symmetrical brain regions showed distinct differences in different frequency bands, the reasons and clinical significance of these differences were not studied further. If the sample size can be expanded, relevant research in future studies can be considered.

ACK N OWLED G M ENTS
We would like to thank anonymous reviewers for reviewing and commenting on the manuscript and all research participants for their patience.

CO M PE TI N G I NTER E S TS
The authors report no competing interests.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data available on request from the authors.