The Neuronal Correlates of Digits Backward Are Revealed by Voxel-Based Morphometry and Resting-State Functional Connectivity Analyses

Digits backward (DB) is a widely used neuropsychological measure that is believed to be a simple and effective index of the capacity of the verbal working memory. However, its neural correlates remain elusive. The aim of this study is to investigate the neural correlates of DB in 299 healthy young adults by combining voxel-based morphometry (VBM) and resting-state functional connectivity (rsFC) analyses. The VBM analysis showed positive correlations between the DB scores and the gray matter volumes in the right anterior superior temporal gyrus (STG), the right posterior STG, the left inferior frontal gyrus and the left Rolandic operculum, which are four critical areas in the auditory phonological loop of the verbal working memory. Voxel-based correlation analysis was then performed between the positive rsFCs of these four clusters and the DB scores. We found that the DB scores were positively correlated with the rsFCs within the salience network (SN), that is, between the right anterior STG, the dorsal anterior cingulate cortex and the right fronto-insular cortex. We also found that the DB scores were negatively correlated with the rsFC within an anti-correlation network of the SN, between the right posterior STG and the left posterior insula. Our findings suggest that DB performance is related to the structural and functional organizations of the brain areas that are involved in the auditory phonological loop and the SN.


Introduction
Working memory (WM) refers to a limited system that provides for the temporary storage and manipulation of the information necessary for complex cognitive tasks and that provides an interface between perception, long-term memory and action [1,2]. The definition of WM has evolved from the concept of short-term memory but is defined in three different ways: as short-term memory applied to cognitive tasks; as a multi-component system that holds and manipulates the information in the short-term memory; and as the use of attention to manage the short-term memory [3]. A widely accepted model of WM has proposed that it consists of four subsystems, including the central executive system, the phonological loop, the visuospatial sketchpad, and the episodic buffer [1,2,[4][5][6]. The phonological loop is specialized for processing verbal materials and is assumed to be a crucial component of the WM system for language acquisition [7]. Moreover, the phonological loop includes two subsystems: a phonological store, which has a limited information capacity and temporal trace (information can be held for a few seconds before it fades); and a subvocal rehearsal system, which continually repeats information to revive the memory trace in WM [2,6]. The visuospatial sketchpad is a parallel to the phonological loop but exists and serves for the processing of visual and spatial information. The central executive system is an attentional control system that is responsible for strategy selection and for the regulation and coordination of the various processes involved in the phonological loop and the visuospatial sketchpad [2,6]. The episodic buffer is assumed to be a limited-capacity system that depends heavily on executive processing but that differs from the central executive system in being principally concerned with the storage of information rather than with attentional control. The episodic buffer is capable of binding information from different modalities into a single multi-faceted code [7].
Most of our knowledge of the neural correlates of WM is derived from lesion studies and functional imaging studies, which have revealed that the phonological store depends largely upon the left inferior parietal cortex [8,9]; the rehearsal processes are based on the left inferior frontal gyrus (IFG) (typically described as Broca's area), the premotor area and the supplementary motor area [10][11][12][13][14]; and the central executive system relies heavily on the frontal lobe, particularly the dorsolateral prefrontal cortex (DLPFC) and the dorsal anterior cingulate cortex (ACC) [15]. Although functional imaging can be used to identify the brain regions engaged in WM, it cannot be used to identify the neural correlates of WM capacity because functional imaging measures active processing, whereas capacity is a constraint on processing and not a process itself [16].
WM capacity is typically assessed using behavioral measures such as the digit-span test, in which participants are asked to perform the immediate recall of digit sequences of increasing length. Digits forward (DF) has been characterized as a simple span test and is thought to measure the storage and maintenance components of the WM by deemphasizing the manipulation of the material. However, Digits backward (DB) requires a transformation to reorder the input verbal digits into a reversed sequence and is believed to involve the phonological loop (phonological store and rehearsal process) and the central executive system in the putative WM model. DB is thought to be a sensitive measure of WM and has been widely used in neuropsychological research and clinical evaluations [17][18][19]. However, the neural correlates of DB capacity have not been well established.
Several studies have been performed to investigate the relationship between the gray matter volume (GMV) and the performance of the digit-span test in different populations using voxel-based morphometry (VBM) analysis. In 109 healthy elderly people, the performance of the digit-span test was positively correlated with the gray matter ratio, i.e., the GMV divided by the intracranial volume [20]. In a study of 34 normal and 40 dyslexic adults, researchers identified a region in the left posterior superior temporal sulcus where the gray matter density was positively correlated with the performance of DF and DB [16]. A study of 58 patients with neurodegenerative diseases found that the DB scores were correlated with the GMVs of the DLPFC and the inferior parietal lobule [21]. However, there is a lack of studies with large sample sizes that investigate the structural correlates of DB capacity in healthy young adults. Furthermore, none of the previous studies have substantiated the question using rsFC analysis, which investigates the correlations of the time series between a region of interest (ROI) and other voxels of the brain.
In the present study, using a VBM analysis across the whole brain, we firstly identified positive correlations between the GMV and the DB score in the right anterior superior temporal gyrus (STG), the right posterior STG, the left IFG and the left Rolandic operculum. Many previous studies on brain disorders have revealed that brain areas with changes in the GMV are commonly accompanied by altered rsFCs between these regions and other related brain areas [22][23][24]. For example, the hippocampus had both a reduced GMV [25] and rsFCs in patients with Alzheimer disease [26]. These findings suggest that structural (i.e., GMV) change in a brain area may be associated with rsFC alteration in this area and that the combination of VBM and rsFC analyses can improve our understanding of the pathology of brain diseases. We further hypothesize that the combination of these two methods may improve our understanding of the neural correlates of DB capacity. Thus, we also investigated the correlations between the DB scores and the rsFCs of the four clusters found in the VBM analysis.

Participants
A total of 324 right-handed healthy young adults participated in this study. Twenty-five of these subjects were excluded from further analysis due to unqualified images (2 subjects) or lack of behavior assessment (23 subjects). The remaining 299 healthy young adults (151 females, 148 males; mean age 22.7 years, range 18-29 years) were finally included. All of the subjects were Chinese speakers and had no history of neurologic or psychiatric disorders.
The participants were recruited from universities or society and were paid for their participation. All of the participants gave written informed consent that was approved by the Medical Research Ethics Committee of Tianjin Medical University.

Behavior assessments
The working-memory capacity was measured by the DB subtest of the digit span test in the Chinese Revised Wechsler Adult Intelligence Scale (WAIS-RC) [27]. The subjects performed the DB task outside of the scanner. In the DB test, a variable number of random digits were read aloud by the examiner at a rate of one digit per second. Then, the subject was asked to repeat the digits in the reverse order immediately. The DB test comprised nine levels of difficulty, with the digit sequences ranging from two to ten digits. Each level consisted of two trials with digit sequences of the same length. The difficulty is progressively increased by increasing the number of random digits and the test is terminated if both trials of a test item are repeated incorrectly. The DB score (2 to 10 points) was the maximum number of digits that the subject successfully repeated in the reverse order.
The handedness of the participants was evaluated by a Chinese questionnaire modified according to the Edinburgh Handedness Inventory [28]. The handedness preference for each of 10 items, namely writing, holding chopsticks, throwing, using a toothbrush, using scissors, striking a match, threading a needle, holding a hammer, holding a racket and washing the face, was assessed for each subject. If the same hand preference was expressed for all of the 10 items, the subject was regarded as strongly right-or lefthanded. If a person preferred using his right or left hand for the first 6 items but used the other hand for each of the last 4 items, this person was regarded as right-or left-handed. If a subject mixed using his or her right and left hands in the first 6 items, this person was regarded as having a mixed handedness. In this situation, the writing hand determined its tendency. Taken together, the handedness was quantitatively strong right, right, mixed right, mixed left, left, strong left strong right, right, mixed right, mixed left, left, strong left scored as 1-6 for the strong right-, right-, mixed right-, mixed left-, left-and strong left-handedness, respectively.

MRI acquisition
All of the MR images were acquired on a Signa HDx 3.0 Tesla MR scanner(GE Medical Systems). Tight but comfortable foam padding was used to minimize head motion, and ear plugs were used to reduce scanner noise. A high-resolution T1-weighted brain volume (BRAVO) 3D MRI sequence with 176 contiguous sagittal slices was performed with the scan parameters of repetition time (

VBM analysis
The VBM analysis was performed using the Statistical Parametric Mapping (SPM8) software (http://www.fil.ion.ucl.ac. uk/spm/software/spm8). The structural MR images were seg-mented into gray matter (GM), white matter and cerebrospinal fluid using the standard unified segmentation model in SPM8. Next, the GM population template was generated from the entire image dataset using the Diffeomorphic Anatomical Registration Through Exponentiated Lie algebra (DARTEL) technique [29]. After an initial affine registration of the GM DARTEL template to the tissue probability map in Montreal Neurological Institute (MNI) space (http://www.mni.mcgill.ca/), non-linear warping of the GM images to the DARTEL GM template was performed in MNI space with 1.5-mm cubic resolution (as recommended for the DARTEL procedure). The GMV of each voxel was obtained through modulation by multiplying the GM concentration map by the non-linear determinants derived from the spatial normalization step. Finally, to compensate for the residual anatomical differences, the GMV images were smoothed with an FWHM kernel of 8 mm 3 . In effect, the analysis of the modulated data tests for regional differences in the absolute volume of the brain and eliminates the confounding effects of different individual brain sizes. After spatial pre-processing, the smoothed, modulated and normalized GMV maps were used for the statistical analysis.
A voxel-based partial correlation analysis was conducted to test the relationships between the GMVs and the DB scores, and age, gender and years of education were entered as covariates of no interest. The purpose of partial correlation is to find the unique variance between two variables while eliminating the variance from the controlling variables. Converging evidence from a number of studies across several cognitive task domains, including WM, suggests that age, gender and years of education are related to cognitive functions, such as WM [30][31][32][33]. Moreover, the GMV and the rsFC are correlated with age [20,[34][35][36]. Thus, we calculated partial correlation coefficients between the GMV or rsFC and the DB scores to remove the effects of age, gender and years of education.
The correction for multiple comparisons was performed using a Monte Carlo simulation. A corrected threshold of P,0.05 was derived from a combined threshold of P,0.001 for each voxel and a cluster size .147 voxels, which was determined by the AlphaSim program in the AFNI software (parameters: single voxel P = 0.001, 5000 simulations, FWHM = 8 mm, cluster connection radius = 2.5 mm, with gray matter mask, http://afni.nimh.nih. gov/).

Resting-state functional connectivity analysis
The preprocessing and statistical analysis of the functional images were conducted using SPM8 (http://www.fil.ion.ucl.ac. uk/spm/software/spm8). The first 10 volumes of each functional time series were discarded for the magnetization equilibrium. The remaining 170 images were corrected for the time delay between the different slices and realigned to the first volume. The head motion parameters were computed by estimating the translation in each direction and the angular rotation on each axis for each volume, which provided a record of the head position. The realigned images were spatially normalized to the MNI EPI template and re-sampled to a 2-mm cubic voxel. The normalized images were smoothed with a Gaussian kernel of 6 mm full-width at half maximum. We required the head motions of all of the subjects to be lower than the thresholds of 2 mm of translation in each of the x, y and z directions and 2u of rotation in each of the x-, y-and z-axes; only 282 subjects satisfied this requirement and were included in the rsFC analysis. Several sources of spurious variances, including the estimated motion parameters, linear drift, global average BOLD signals, and average BOLD signals in the ventricular and white matter regions, were removed from the data through linear regression [37,38]. Finally, temporal band-pass filtering (0.01,0.08 Hz) was performed on the time series of each voxel to reduce the effects of low-frequency drift and highfrequency noises [37,39].
The ROIs were defined as the brain areas in which GMVs showed significant correlations with the DB scores. Thus, four ROIs were extracted and resampled into 2-mm cubic voxels. Next, ROI-based FC analysis was performed to calculate the FC maps of each ROI using the DPARSF software (http://www.restfmri.net). For each subject, the correlation coefficients between the mean time series of each ROI and that of each voxel of the whole brain were computed and converted to z-values, using Fisher's r-to-z transformation to improve the normality. Subsequently, individuals' z-values were entered into a random-effect one-sample t-test in a voxel-wise manner to identify the brain regions that showed significant positive correlations with the seed region. The significant FC maps were corrected for multiple comparisons using the Family Wise Error (FWE, P,0.05) method. Next, a mask that included the brain areas with significant positive rsFCs with the seed regions was generated for the following correlation analyses; the brain areas with negative rsFCs were excluded because their functional significance is still under debate [40,41].
Similar to the VBM analysis, the voxel-based partial correlation analysis was performed to test the relationships between the rsFCs of each of the four ROIs and the DB scores, and age, gender, and years of education were entered as covariates of no interest. The correction for multiple comparisons was performed using a Monte Carlo simulation. A corrected threshold of P,0.05 was derived from a combined threshold of P,0.01 for each voxel and a cluster size .110 voxels which was determined by the AlphaSim program in the AFNI software (parameters: single voxel P = 0.01, 5000 simulations, FWHM = 6 mm, cluster connection radius = 3.3 mm, with gray matter mask, http://afni.nimh.nih.gov/).

Behavioral data
The demographic and behavioral data are reported in Table 1. A total of 299 healthy young adults (151 females, 148 males; mean age 22.7 years, range 18-29 years) were finally included in the VBM analysis. Among these adults, 282 subjects participated in the rsFC analysis. The mean DB score was 7 with a standard deviation of 1.5 and a range from 3 to 10. There were no significant correlations between the DB scores and the age or years of education (P.0.05). There was no significant difference in the DB scores between the male and female subjects (P.0.05).

VBM analysis
The results of the correlations between the DB scores and the regional GMVs, adjusting for age, gender and years of education, are reported in Table 2 and Figure 1. No negative correlations were found between the regional GMVs and the DB scores. However, the VBM analysis revealed that the DB scores were positively correlated with the GMVs in four clusters (P,0.05, corrected). One cluster (ROI 1) was located in the right anterior STG (peak MNI coordinates of x = 55.5, y = 15, z = 24.5). Another cluster (ROI 2) was located in the right posterior STG (peak MNI coordinates of x = 58.5, y = 222.5, z = 13.5), a counterpart of the putative Wernicke's area in the left hemisphere. A third cluster (ROI 3) was located in the left IFG, corresponding to the putative Broca's area (peak MNI coordinates of x = 243.5, y = 25.5, z = 15). The last cluster (ROI 4) was located in the left Rolandic operculum (peak MNI coordinates of x = 245, y = 225.5, z = 19.5), which is an area that is involved in speech.

Resting-state functional connectivity analysis
The four seed regions (ROIs) where the GMVs were positively correlated with the DB scores were selected for the rsFC analysis. The brain areas that had positive rsFCs with each ROI (FWE corrected, P,0.05) are shown in Figure 2. A partial correlation analysis was performed between the rsFCs of each ROI and the DB scores, and the results are shown in Table 3 and Figure 3. The DB scores were positively correlated with the rsFCs of ROI 1 (the right anterior STG) with the dACC (peak MNI coordinates x = 14, y = 28, z = 20) and the right fronto-insular cortex (FIC) (peak MNI coordinates x = 16, y = 20, z = 216). However, no significant negative correlations were observed for ROI 1. In contrast, the DB scores were negatively correlated with the rsFC between ROI 2 (the right posterior STG) and the left insula (peak MNI coordinates x = 238, y = 14, z = 12) (Figure 4). No significant positive correlations were observed for ROI 2. No significant correlations were observed between the DB scores and the rsFCs of ROI 3 and ROI 4.

Discussion
The capacity of the auditory-verbal-digital WM is dependent on the close cooperation of several subsystems, including the auditoryphonological perception and storage system, the subvocal rehearsal system, and the central executive system [2,6,7]. We will discuss our findings from the VBM and rsFC analyses within this framework of the verbal WM.

The VBM analysis
The phonological perception and storage system. The perception of auditory input is critically important for the subsequent storage, rehearsal and executive control processes and thus relates to the verbal WM capacity as assessed by the DB test. The anterior STG belongs to the ''what'' pathway of the auditory system and serves for processing auditory object identity [42,43]. The right anterior STG has been reported to respond to increased spectral variation of auditory input [44,45], to enhance the response to auditory selective attention tasks [46], and to process auditory information about prototypicality [47]. We found that the verbal-digital WM capacity was positively correlated with the GMVs of the right anterior STG, suggesting that the accurate perception of auditory input is important for the capacity of the verbal WM. This assertion is supported by functional imaging findings indicating that better language performance is associated with more involvement of the right hemisphere [48,49] and that higher verbal intelligence was associated with the activation of the  right anterior STG [50]. A VBM study found that Chinese speakers had greater gray and white matter density in the right anterior STG compared with those who did not speak Chinese [51], which is also consistent with our finding in the Chinese population. Moreover, in patients with aphasic stroke, the language performance and outcome are found to be associated with the activation [52] and functional connectivity [53] of the right anterior STG. The posterior STG is involved in phonological processing [54][55][56][57]. The left posterior STG is regarded as a ''receptive'' area for the analysis and identification of auditory verbal stimuli [58]. In a study of 210 stroke patients, the authors found that the structural integrity of the left posterior STG can predict the auditory shortterm memory capacity [59]. It has also been shown that the gray matter density in the left posterior STG is correlated with the auditory short-term memory capacity in both normal and developmentally dyslexic groups [16]. In the present study, we found that the GMV of the right posterior STG can predict the performance of the verbal WM, which may be explained as follows. One explanation is that the right posterior STG serves certain special functions that are important for the verbal WM. This theory is supported by the findings that the right posterior STG is the main locus for the perception of sounds [60], is the preferred hemisphere for the processing of spectrotemporal auditory information [61,47], and serves to enhance the response to auditory selective attention tasks [46]. Although the left posterior STG is the main location of activation during phonological processing, the right posterior STG is additionally recruited during phonological processing in subjects with higher performance on verbal tests. Thus, an alternative explanation is that the recruitment and structural organization of the right posterior STG reflect a capacity for language processing. This idea is consistent with the findings that the right hemispherical STG is more strongly activated in subjects with better language performance [48,49] and higher verbal intelligence [50].
The subvocal rehearsal system. The left IFG (Broca's area) and the Rolandic operculum are the main components of the subvocal rehearsal system in the verbal WM model. The continuous subvocal rehearsal is critically important for the revival and maintenance of the memory trace and thus contributes to the performance of the verbal WM [2,6]. The involvement of Broca's area in the subvocal rehearsal system of the Figure 2. The brain areas that show positive rsFCs with each ROI (FWE corrected, P,0.05). The representation of each ROI is shown in Table 1. Abbreviations: FWE, family-wise error; ROI, region of interest; rsFC, resting-state functional connectivity. doi:10.1371/journal.pone.0031877.g002 Table 3. Brain areas whose rsFCs with the seed ROIs are correlated with the DB scores.  phonological loop was first proposed by Paulesu and colleagues (1992) [10]. Subsequently, functional imaging studies on the verbal WM revealed the neural substrates of the rehearsal of phonological information in Broca's area and related motor areas [11][12][13]62]. A functional MRI study on the DF and DB tasks revealed that the DB task additionally recruited Broca's area (BA44), in contrast to the DF task [63]. These findings are consistent with our finding that the GMV of Broca's area (BA44/ 45) was positively correlated with the individual verbal WM capacity, suggesting that the structural organization of the rehearsal component is one of the restricting factors of the verbal WM capacity. The left Rolandic operculum plays a role in speech production and phonological rehearsal [64], which is supported by findings that lesions in this area are associated with articulatory disorders [65] and the finding that structural and functional abnormalities of this area are associated with stuttering [66,67].

The rsFC analysis
The insular rsFC. The insula is a cytoarchitectonically and functionally heterogeneous region that has been divided into 2 or 3 subregions [68][69][70]. Resting-state fMRI has shown that each insular subregion has a distinct rsFC pattern, is involved in different functional networks, and serves a different function [71,72]. The ventral anterior insula is primarily connected with the pregenual ACC and is involved in emotional processing; the dorsal anterior insula is connected with the dorsal ACC and is implicated in cognitive control; and the posterior insula is connected with the primary and secondary sensorimotor cortices and participates in sensorimotor processing [72]. In another parcellation scheme, the insula is divided into anterior and posterior parts, which form two mutually exclusive and anticorrelated functional networks, i.e., increases in the BOLD signal for one network correspond to decreases in the other one [71]. The anterior insula network is also called the salience network (SN). As illustrated in Figure 2, ROI 1 (the right anterior STG) is involved in the SN, whereas ROI 2 (the right posterior STG) and ROI 4 (left Rolandic operculum) belong to the posterior insula network.
The salience network. The SN is mainly comprised of the FIC and dorsal ACC and serves to identify the subjective salience across several domains, whether it is cognitive, homeostatic, or emotional [67,[73][74][75][76][77][78][79][80]. Specifically, the FIC serves to mark salient stimuli transiently from the vast and continuous stream of visual, auditory, tactile and other sensory inputs and to initiate attentional control signals, which are then sustained by the dorsal ACC and the ventrolateral and dorsolateral prefrontal cortices [74]. Once such a stimulus is detected, the right FIC initiates appropriate transient control signals to engage the brain areas of the centralexecutive network (CEN) mediating attention, WM and other higher-order cognitive processes while disengaging the defaultmode network (DMN) via the von Economo neurons (VENs) [81] with large axons that facilitate the rapid relay of FIC and dorsal ACC signals to other cortical regions [82]. Consistent with this view, event-related fMRI shows by latency analysis that the onset of signal activity in the right FIC occurs earlier compared with the activation in the CEN nodes and the deactivation in the DMN nodes [83]. More importantly, the right FIC plays a critical role in switching between the CEN and the DMN [74,83]. In addition to its role in the perception of the salience of stimuli, the dorsal ACC has been reported to be involved in a variety of cognitive functions, such as attention control [84][85][86], conflict monitoring [87][88][89], error monitoring and detection [90][91][92][93], and response selection [94,95]. The involvement of the dorsal ACC in the WM has been reported in functional imaging studies [96][97][98]. The dorsal ACC shows greater activation during cognitive tasks in subjects with a higher capacity of WM [97,98].
Recently, many pieces of evidence suggest that several disorders, such as frontotemporal dementia, depression, and posttraumatic stress disorder are associated with the dysfunction of the SN [99][100][101][102]. In the rsFC analysis, we found that the DB scores were positively correlated with the rsFCs within the SN (between the right anterior STG and the dorsal ACC and right FIC), which  suggests that better WM performance is related to increased rsFCs within the SN. We also found a negative correlation between the DB scores and the rsFC within the posterior insula network (between the right posterior STG and the left posterior insula), which indicates that better WM performance is associated with reduced rsFC within the posterior insula network, which is anticorrelated with the SN. Taken together, our results are consistent with the notion that different insula subregions are involved in distinct functional networks and serve different functions.
The negative rsFC. Negative rsFCs (anti-correlations) have been found between the task-positive network and the defaultmode network [37,38,103,104] and between the neural systems underlying different components of the verbal WM [105]. Greicius et al. (2003) suggested that intrinsic anti-correlated activity might relate to the differential task-related responses in these regions [38]. Fox et al. (2005) proposed that anti-correlations may serve a differentiating role in segregating neuronal processes that serve opposite goals or competing representations [37]. However, one should be cautious when interpreting anti-correlations derived from the resting-state fMRI studies because the debate remains unsettled regarding whether anti-correlations are artifacts of the global signal regression [40,41] or whether they reflect dynamic, truly anti-correlated functional networks [106]. Thus, in the present study, we did not analyze the correlations between the DB score and the negative rsFCs of the four ROIs derived from the VBM analysis.
A comparison between our findings and the brain activation during the DB task Converging evidence from near-infrared optical tomography, PET and fMRI studies show that the DB task activates the brain regions associated with WM (the bilateral DLPFC and the inferior parietal lobule (IPL), the left Broca's area, and the cerebellum), attention (the ACC), and visuospatial processing (the occipital cortex) [63,[107][108][109]. The left Broca's area and the right dorsal ACC were correlated with the DB capacity in the present study and were also activated in DB tasks in previous studies [63]. However, other related brain areas were either correlated with DB capacity or activated during DB tasks, but not both. These findings supported the concept that functional imaging during the DB task measures active processing but that off-line DB capacity is a constraint on this processing and not a process itself [16]. That is, the brain areas that show DB task-induced activation and those that correlate with DB capacity are related but are not necessarily the same areas.

Conclusions
In the present study, we found that the capacity of the auditoryverbal WM was associated with the structural organization of the brain areas related to the phonological perception and storage and the subvocal rehearsal system. We also found that this capacity is positively correlated with the rsFC between the right anterior STG (an area related to the phonological perception) and the dorsal ACC and the right FIC (two core nodes of the SN) and anticorrelated with the resting-state functional connectivity between the right posterior STG with left posterior insula (two areas included in posterior insula pattern network). These findings suggest that digital backward performance is related to the structural and functional organizations of the brain areas involving in the auditory phonological loop, the SN and its anticorrelated networks.

Author Contributions
Conceived and designed the experiments: CY TJ YZ. Performed the experiments: RL WQ. Analyzed the data: RL WQ. Contributed reagents/ materials/analysis tools: RL WQ. Wrote the paper: RL CY. Critically read the manuscript: CY TJ YZ.