Abstract

Orthographic processing skills (reading and spelling) are evolutionarily recent and mastered late in development, providing an opportunity to investigate how the properties of the neural networks supporting skills of this type compare to those supporting evolutionarily older, well-established “reference” networks. Although there has been extensive research using task-based fMRI to study the neural substrates of reading, there has been very little using resting-state fMRI to examine the properties of orthographic networks. In this investigation using resting-state fMRI, we compare the within-network and across-network coherence properties of reading and spelling networks directly to these properties of reference networks, and we also compare the network properties of the key node of the orthographic networks—the visual word form area—to those of the other nodes of the orthographic and reference networks. Consistent with previous results, we find that orthographic processing networks do not exhibit certain basic network coherence properties displayed by other networks. However, we identify novel distinctive properties of the orthographic processing networks and establish that the visual word form area has unusually high levels of connectivity with a broad range of brain areas. These characteristics form the basis of our proposal that orthographic networks represent a class of “high-level integrative networks” with distinctive properties that allow them to recruit and integrate multiple, lower level processes.

INTRODUCTION

Written language is a relatively recent human invention: Humans have been communicating in writing for only a few thousand years, whereas they have been using certain other cognitive skills since they first walked the earth. Despite appearing only recently in human history, orthographic processing (reading and spelling) is a very important cognitive ability, such that the failure to acquire literacy or its loss due to brain lesion significantly limits one's ability to function successfully in daily life (Hillis & Tippett, 2014). Because of its evolutionary recency, orthographic processing is fully learned, unlike older cognitive functions, such as motor planning, attention, or vision, that are specified in the genetic blueprint (Dehaene, 2009). Written language, therefore, provides an important opportunity to investigate how the human brain instantiates highly practiced, complex, evolutionarily recent skills. To date, research on the neural bases of literacy has been primarily directed at identifying the brain networks that support reading (and, to a lesser extent, spelling) and at determining if the brain areas involved in reading are specialized for this function. More recently, research has begun to ask if the network properties of the brain areas supporting reading are similar to those of networks supporting other cognitive functions. Because resting-state fMRI (RS-fMRI) is especially appropriate for characterizing intrinsic network properties, we use it to (1) directly compare the coherence properties1 of reading and spelling networks (RdN and SpN, respectively) with those of other networks and (2) examine how the network properties of the key node of the orthographic processing network—the visual word form area (VWFA; Cohen et al., 2000)—compare to those of other nodes. Our analyses identify novel properties of the orthographic processing networks and the VWFA, revealing that these networks are well suited to specific challenges of literacy, which require integration across a diverse set of cognitive skills.

RS-fMRI Investigations of Orthographic Processing

A number of studies have provided support for the notion that RS-fMRI exhibits similar connectivity patterns to those observed in task-based fMRI (e.g., Cole, Bassett, Power, Braver, & Petersen, 2014; Smith et al., 2009). Despite the application of RS-fMRI to a wide range of cognitive skills, only a few RS-fMRI studies have examined reading, and none has investigated spelling. With regard to reading, some studies have examined intrinsic brain connections in individuals with dyslexia (e.g., Horowitz-Kraus, DiFrancesco, Kay, Wang, & Holland, 2015; Schurz et al., 2015; Zhou, Xia, Bi, & Shu, 2015), and others have looked at the resting-state (RS) connectivity of brain regions involved in reading tasks in neurotypical individuals (Stevens, Kravitz, Peng, Tessler, & Martin, 2017; Vogel, Petersen, & Schlaggar, 2014; Li et al., 2013; Vogel et al., 2013; Vogel, Miezin, Petersen, & Schlaggar, 2012; Vogel, Petersen, & Schlaggar, 2012; Koyama et al., 2010, 2011).

Most relevant to the current study, Power et al. (2011) and Yeo et al. (2011) carried out whole-brain graph theoretical analyses designed to identify “communities” of brain areas with correlated activations at rest. Neither found any neural communities corresponding to the RdN, although they did identify, among others, communities corresponding to dorsal attention, frontoparietal, and cingulo-opercular control networks. Following up on this, Vogel et al. (2013) considered a large number of brain areas that have been associated with reading, but they too failed to find strongly knit reading-related communities or networks. Instead, Vogel et al. (2013) found that reading-related regions were strongly connected to regions from a number of different communities. On this basis, they concluded that the regions recruited for reading do not form a true network with the properties observed in other networks. With regard to the VWFA specifically, Vogel et al. (2012) found that it was more strongly connected to regions of the dorsal attention network (DAN) than to other reading-related regions. The VWFA had particularly strong connections to the right anterior intraparietal sulcus, a key node of the DAN (e.g., Behrmann, Geng, & Shomstein, 2004; Corbetta & Shulman, 2002; Corbetta, 1998), and the strength of this connection was positively correlated with reading age. On the basis of these findings, Vogel et al. (2012) concluded that the brain has not remodeled itself for reading and that the VWFA is not specialized for word reading. They proposed instead that the VWFA is a general processor that is designed to act on groups of visual stimuli that require attention for successful processing (Vogel et al., 2014). They argued that, although reading may be one task that requires such a processor, it is by no means the only one and that this area is used whenever this attentional function is required—not only for written words but also for other visual stimuli such as letter strings, unfamiliar characters, or line drawings (Vogel et al., 2012).

Understanding How Orthographic Processing Compares to Other Networks

Although the data-driven approaches of Vogel et al. (2013), Power et al. (2011), and Yeo et al. (2011) showed that a RdN does not emerge as a “community” of regions, none of this work directly compared the connectivity characteristics of the entire RdN with those of other, evolutionarily well-established networks. As a result, we do not know how the properties of specific well-studied networks compare with those of reading (or spelling) networks. Vogel et al. (2012) revealed strong connectivity between the VWFA and regions of the DAN, and Stevens et al. (2017) recently reported a strong connectivity between the VWFA and language areas. However, the possibility of additional strong connectivity between the VWFA and a broader range of networks has not yet been investigated. Furthermore, although Vogel et al. (2012) found that the VWFA—the key node of the RdN—was more strongly linked to regions of the DAN than to regions of the RdN itself, we do not know how unusual this is compared with other nodes of the RdN and nodes of other networks. Therefore, to better understand the characteristics of the RdN, it is critical to directly compare it with other well-studied networks and their nodes.

Finally, it is worth noting that spelling is closely related to reading in its processing and representation of orthographic information and also in terms of its evolutionary recency. Together, reading and spelling make up our orthographic processing abilities, and examining the SpN would provide opportunities to understand if the network characteristics observed for reading correspond to more general properties of orthographic processing.

The SpN and the Role of the VWFA in Spelling

Although reading involves perception and comprehension of written words, spelling involves the expression of orthographic knowledge, via handwriting, typing, oral spelling, and so forth. Two meta-analyses of neuroimaging studies of spelling and writing (Planton, Jucla, Roux, & Démonet, 2013; Purcell, Turkeltaub, Eden, & Rapp, 2011) identified areas consistently involved in spelling. The specific roles of these areas have been elucidated by research on acquired spelling deficits subsequent to brain lesion (e.g., Rapp, Purcell, Hillis, Capasso, & Miceli, 2016). Interestingly, the left mid-fusiform location of the VWFA was identified as a key spelling region in both neuroimaging meta-analyses and brain lesion-symptom mapping results. Furthermore, studies have found that a shared left fusiform area is recruited for both reading and spelling when both tasks were examined in the same individuals (Purcell, Jiang, & Eden, 2017; Purcell et al., 2011; Rapp & Dufor, 2011; Rapp & Lipka, 2011). Given the differences between reading and spelling with regard to input stimulus modality, the finding that the same brain area is recruited for both is a challenge for claims (such as those proposed by Vogel et al., 2012) that the VWFA is strictly involved in the visual processing of letters or other visual stimuli. Instead, these findings suggest that more abstract orthographic representations are processed by this region (Rothlein & Rapp, 2014).

The Current Study

As indicated, previous work has not directly compared the properties of RdN and SpN to those of well-established reference networks, nor have the properties of the VWFA been compared with other regions within these networks. To address these outstanding issues, we identified three well-studied reference networks: (1) the default mode network (DMN), which is active when individuals are not performing a specific task (e.g., Raichle et al., 2001); (2) the DAN, which is involved in top–down orienting or shifting of attention (e.g., Fox, Corbetta, Snyder, Vincent, & Raichle, 2006); and (3) the sensorimotor network (SMN), involved in preparing and executing motor acts (e.g., Biswal, Yetkin, Haughton, & Hyde, 1995; Rumeau et al., 1994). All three of these reference networks have also been studied extensively using RS-fMRI (e.g., Keller et al., 2015; van Meer et al., 2010; Fox, Zhang, Snyder, & Raichle, 2009; Laird et al., 2009; Fransson & Marrelec, 2008; Behrmann et al., 2004; Fox et al., 2006; Corbetta et al., 1998). Using RS-fMRI, we carried out three sets of analyses comparing the two orthographic processing networks to the three reference networks. Two of the analyses focused on network characteristics, and the third focused on the characteristics of individual network nodes and, in particular, the VWFA.

First, we analyzed within-network versus across-network coherence for the five networks (RdN, SpN, DMN, DAN, and SMN). We defined within-network coherence as the average connectivity strength of each node within a network with every other node within the same network. Across-network coherence corresponded to the average connectivity of network nodes with nodes from the other networks. Second, we compared the within-network coherence of each of the networks to that of pseudonetworks composed of randomly selected nodes from across the brain. Third, we examined some of the characteristics of individual nodes using graph theoretical measures and node-based linear mixed-effects models. For the node-based analyses, we first evaluated the extent to which nodes are strongly connected to other nodes, both within and across networks, using the graph theoretical measure known as “degree” (Bullmore & Sporns, 2009). To determine not only whether nodes have many strong connections but also the extent to which these connections involve multiple networks, we used the graph theoretical measure “participation coefficient” (Rubinov & Sporns, 2010). Finally, to understand the extent to which nodes are connected with their own network compared with other networks, we compared the average connectivity of each node with nodes from its own network to its connectivity with nodes from each of the other networks using a linear mixed-effects model.

In combination, these analyses allowed us to better understand how orthographic processing networks compare to evolutionarily well-established networks and, in so doing, further our understanding of the brain's response to the specific challenges of literacy. Briefly, we find that, consistent with previous results, orthographic processing networks do not exhibit key network properties exhibited by the reference networks. However, the orthographic processing networks and the left fusiform (VWFA) region do display distinctive properties. We suggest that these properties arise in response to the specific challenges of literacy to integrate multiple distinct cognitive functions, and we propose that orthographic processing networks may represent a class of “high-level integrative networks” that draw on and integrate lower level processes that span multiple domains.

METHODS

Participants

Twenty-one neurotypical adults (8 men, 13 women) who self-reported no history of reading or spelling disability participated in the study. The mean age of the participants was 56.2 years2 (range = 42–78 years), and they had all completed at least 12 years of education (mean years of education = 16.6). All of the participants reported normal or corrected-to-normal vision. Four participants were excluded from the analysis: three of them because they reported falling asleep during the RS scan and the other one due to incorrect acquisition parameters. All the participants signed informed consent in accordance with the requirements of Johns Hopkins' institutional review board and were compensated $50 for their time.

MRI Data Acquisition

MRI data were acquired on a 3.0-T Phillips Intera Scanner at the F. M. Kirby Research Center for Functional Brain Imaging at the Kennedy Krieger Institute (Baltimore, MD). Participants were fitted with an eight-channel SENSE (Invivio) parallel-imaging head coil. Each participant performed two RS scans (14:20 min total), followed by an anatomical (MPRAGE) scan, and additional task-based scans that will not be reported in this article. Structural images were acquired using an MPRAGE T1-weighted sequence, producing images with 1-mm isotropic voxel resolution (repetition time = 2.3 sec, total scan time = 4:50 min). With the exception of one participant, all scans for each participant were acquired within the same session that lasted approximately 1 hr 15 min. RS scans were always acquired before task scans in all of the participants, so that the task would not affect the resting data.

RS Scans

In the RS scans, participants were asked to lie still, with their eyes open. They fixated a black cross on gray background and were instructed to “try and stay awake and not think of anything in particular.” Participants' eyes were observed during scanning to verify that they are awake, and they were also asked at the end of each scan whether they had fallen asleep. Two identical runs were acquired for each individual, with a short break between them. Each of the two scans consisted of 175 volumes, with 41 transverse slices in each volume acquired in ascending order, and 10 dummy volumes. Acquisition parameters were as follows: repetition time = 2.4 sec, echo time = 20 msec, flip angle = 90°, field of view = 220 × 206.25 × 123 mm, voxel size = 1.72 × 1.72 × 3 mm, length of each scan = 7:10 min (for a total of 14:20 for each participant).

Data Analysis

Preprocessing

fMRI data were preprocessed using Statistical Parametric Mapping (SPM12, Wellcome Trust Centre for Neuroimaging, London, United Kingdom). The structural (MPRAGE) images were aligned to the AC–PC plane, segmented into gray and white matter, and then normalized into Montreal Neurological Institute (MNI) space. The functional scans were preprocessed in the following sequence: slice-timing correction, motion correction (six motion parameters and their derivatives—a total of 12 parameters—were obtained for inclusion in subsequent statistical analysis to account for motion artifact, in accordance with Satterthwaite et al., 2019), and resampling to 2 mm3 voxels. The functional data were then coregistered to the structural image and normalized to MNI space using normalization parameters obtained from the structural images to reduce interpolation artifacts. To better account for head motion-induced signal change, the data were also scrubbed, so that a temporal sample was removed if its frame displacement (average rotation and translation parameter difference) was greater than 0.5 or its DVARS (root mean squared intensity difference of volume N to volume N + 1) is greater than 50. Both measures were computed using FSL's fsl_motion_outliers tool (Jenkinson, Beckmann, Behrens, Woolrich, & Smith, 2012), in accordance with Power, Barnes, Snyder, Schlaggar, and Petersen (2012) and Power et al. (2014). Mean frame displacement was 0.17, and mean DVARS was 38.22. Average remaining frames postscrubbing was 162 per scan (92.5%). For the statistical analysis, white matter and CSF principal components were used as nuisance regressors. Global signal regression was not done, in accordance with research suggesting it contains neurological information (e.g., Scholvinck, Maier, Ye, Duyn, & Leopold, 2010). Data were not smoothed, in accordance with recent papers suggesting smoothing may affect network properties (e.g., Alakörkkö, Saarimäki, Glerean, Saramäki, & Korhonen, 2017). RS scans were processed further using the REST Toolbox (Song et al., 2011). Each RS scan was 0.01–0.08 Hz bandpass filtered and detrended to avoid linear trends (degree = 1) that could be caused by thermal noise. The two RS scans for each participant were then mean-centered, tapered (at the tail of the first scan and the head of the second scan), and concatenated, creating one time series of 350 time points for each voxel per participant.

Specification of Network Nodes

We were interested in the RS relations between nodes of the following functional networks: reading, spelling, default mode, dorsal attention, and sensorimotor. To identify the relevant nodes and to avoid specific biases that might arise from a particular task or set of participants, we considered only meta-analyses that met the following criteria: (1) were based on fMRI or PET data; (2) involved healthy adult participants; (3) provided x, y, z coordinates for areas that were identified as activated across studies (coordinates reported in Talairach space were converted to MNI space using the REST Toolbox's conversion tool; Song et al., 2011); and (4) for the reference networks, included only studies and contrasts that controlled for orthographic stimuli, because our primary interest was in RdN and SpN. In other words, for the reference networks, we selected meta-analyses that excluded studies using orthographic stimuli in both the test and control conditions, thereby ensuring that the reference networks were not influenced by orthography. When a meta-analysis reported results for several contrasts, we chose the most general one, because this was most likely to provide information about the entire network of interest rather than any specific sub-function.

The nodes that were included for each network are reported in Appendix A. For DMN, we used the nine regions reported in the meta-analysis by Laird et al. (2009). For the DAN, we used the seven regions reported in the meta-analysis by Wager, Jonides, and Reading (2004). Specifically, we included only the peaks that were significant (at p < .05) in the “switching density” contrast, which is reported to reveal the dorsal network associated with “shifting attention,” excluding those peaks that only showed up in the “executive function” contrast. For the SMN, we used the nine regions reported in the contrast of right upper limb movement versus rest in healthy adults from a meta-analysis by Rehme, Eickhoff, Rottschy, Fink, and Grefkes (2012). This meta-analysis also evaluated the SMN in stroke patients but reported the contrasts separately for stroke patients and healthy adults. For the SpN, we used the meta-analysis of spelling by Purcell et al. (2011). Nodes were selected for the SpN based on the “all” contrast, which includes regions involved in both central spelling processes (such as orthographic long-term memory and working memory) and peripheral spelling processes (such as motor control for written production). This contrast reported 22 peaks. Because one of our goals was to compare connectivity between and across networks, we wanted the number of nodes in each network to be similar. We therefore applied an additional inclusion criterion for this meta-analysis and included only those peaks from the meta-analysis that were found in at least four studies, resulting in eight nodes.

RdN nodes.

In the case of the RdN, we identified five different meta-analyses that matched the criteria mentioned above (Martin, Schurz, Kronbichler, & Richlan, 2015; Taylor, Rastle, & Davis, 2013; Vogel et al., 2012; Jobard, Crivello, & Tzourio-Mazoyer, 2003; Turkeltaub, Eden, Jones, & Zeffiro, 2002). Rather than choosing one of them, we performed an additional higher level meta-analysis, applying a clustering technique. Using the 66 total peaks reported in those five meta-analyses, we computed the Euclidean distance (in mm) between every pair of peaks. We then generated a dendrogram using the Ward.D2 method of the hclust function in R (R Core Team, 2014), defining a height cutoff for the dendrogram that grouped the nodes into 13 clusters. We then picked only those clusters that contained nodes that were represented in at least three of the contributing five meta-analyses. This resulted in nine clusters that were consistent across the meta-analyses used. For each of those nine clusters, we picked as the representative node the one that was closest to the geometric center of the cluster.

Shared nodes.

Because some of the networks we examined are functionally related (e.g., RdN and SpN), it is reasonable to assume that some nodes involved in those networks are shared and in fact perform a similar function in both networks. Because we used different meta-analyses for each cognitive domain, some shared nodes might have been associated with slightly different coordinates, despite actually corresponding to the same functional region. To address this possibility, we combined nodes that were within 1 cm or less of one another in Euclidean distance into one node. We defined the coordinates of the new node as the geometric center of each “combined” node. We excluded only one node pair with nodes within 1 cm of one another because the two nodes were on opposite hemispheres (right and left primary motor cortex in the SMN, which was also the only pair of nodes that were both within 1 cm and belonged to the same network). This process of combining nodes resulted in a total of 34 unique nodes, 6 of which are shared nodes that participate in more than one network (see Figure 1 and Table A1 in Appendix A for a complete list of nodes).

Figure 1. 

Nodes of all five networks examined in this study. A sagittal view (left) and axial view (right) are displayed on a transparent normalized brain. From top: RdN, SpN, DMN, DAN, and SMN. Images visualized using BrainNet Viewer (Xia, Wang, & He, 2013).

Figure 1. 

Nodes of all five networks examined in this study. A sagittal view (left) and axial view (right) are displayed on a transparent normalized brain. From top: RdN, SpN, DMN, DAN, and SMN. Images visualized using BrainNet Viewer (Xia, Wang, & He, 2013).

Extraction of RS Node Data

Spheres with a radius of 3 mm were defined, centered on the MNI coordinates representing each of the 34 different nodes. For each participant, the BOLD signal from all voxels within each sphere was averaged, resulting in one time series for each node for each participant. The correlation between the time series of every pair of nodes was computed for each participant using Pearson's r, and then, a Fisher's Z transformation was applied to normalize the distribution so that statistical calculations could be performed (see Supplementary Material3 for a full matrix of node–node correlations).

Analysis 1: Within-network versus Across-network Coherence

The goal of this analysis was to determine whether the nodes in the networks we examined have higher connectivity with other nodes from the same network than with nodes belonging to different networks. This is similar to the approach used by Vogel et al. (2012), although they only compared the within-network connectivity of the VWFA with the nodes of the RdN to its cross-network connectivity with nodes of the DAN (Vogel et al. 2012). We defined “within-network coherence” as the average Fisher's Z-transformed pairwise correlations between all the nodes of a single network. Similarly, we defined “cross-network coherence” as the average Fisher's Z-transformed pairwise correlations between nodes belonging to two different networks. For example, the SMN–DAN cross-network coherence corresponds to the average of the correlations between all pairs of nodes where one node belongs to the SMN and the other to the DAN.

In Analysis 1, we statistically compared the within-network coherence values for each of the networks with each of their cross-network coherence values. For example, we compared the RdN's within-network coherence with the cross-network coherence of RdN and DMN, RdN and DAN, and so forth. This was tested using a linear mixed-effects model computed with the lmer function in R (R Core Team, 2014), which had one fixed effect, a categorical predictor of network pair (e.g., RdN–RdN or RdN–DMN), and one random effect, a random intercept for each participant. We evaluated five such models, one for each network. Planned contrasts were used to determine, for each network, whether the within-network coherence was different from the cross-network coherence with each of the other networks.

We took two approaches to the issue of the nodes that are shared by two networks and ran the analysis twice. First, we ran the analyses excluding correlations between shared nodes. This way, a node that is shared between two networks was excluded from the calculation and did not contribute to a stronger average correlation between the two networks. However, because a shared node suggests a joint function between two networks, to include this relationship in the analysis, we also performed the comparison including the shared connections with an automatic value of 1 (perfect correlation). Because the results from both analyses were comparable in terms of direction and significance of the results, we only report the results of the first approach.

Analysis 2: Network Coherence—Comparison to Random Networks

The goal of this analysis was to evaluate whether the coherence values reported in Analysis 1 were greater than would be expected for networks composed of randomly selected regions from across the entire brain. To do so, we randomly generated 1500 “pseudonetworks,” each consisting of nine nodes. The nodes were defined as 3-mm-radius spheres, centered at randomly selected coordinates from a normalized gray matter mask of the brain. For each of the participants, we then extracted the RS signal from the nodes of each of these pseudonetworks and computed the average correlation between all pairs of nodes in each network, as in Analysis 1. This resulted in within-network coherence values for the 1500 pseudonetworks for each of the 17 participants, which were then averaged across participants to get 1500 coherence values for the pseudonetworks. The distribution of the averaged coherence values was used to evaluate the relative strength of within-network coherence values obtained from the five networks in Analysis 1.

Analysis 3: Individual Node Properties

Analysis 3 was directed at understanding the properties of the individual nodes of each of the five networks. In particular, we were interested in the properties of the left mid-fusiform, an area commonly known as the VWFA (Cohen & Dehaene, 2004; Cohen et al., 2000). This area has been used as a “seed” in RS seed-based analyses in the past (e.g., Vogel et al. 2012). We compared the properties of the VWFA and of nodes from the orthographic networks more generally with those of nodes from other networks using three measures: First, we computed for each node the degree (also called “degree centrality”), a graph theoretical measure indicating how connected a node is to other nodes (Bullmore & Sporns, 2009). We then calculated each node's participation coefficient, indicating how segregated or integrated it is with other networks/modules (Rubinov & Sporns, 2010). Both measures were computed by (1) calculating the average pairwise correlations of each node with each of the other 33 unique nodes (averaged across participants) and (2) removing connections that were weaker than the average correlation for all 561 unique pairs. The degree value, which corresponds to the number of connections of each node that passed this threshold, was then calculated using the degrees_und function from the Brain Connectivity Toolbox (Rubinov & Sporns, 2010). A higher degree value indicates a greater number of strong (i.e., above-threshold) connections for a given node, whereas a lower value indicates that more of the node's connections are weak. The participation coefficient was calculated using the participation_coef function from the Brain Connectivity Toolbox (Rubinov & Sporns, 2010) and represents the extent to which a node's strong connections are limited to the other nodes of one network or are distributed across multiple networks. A higher participation coefficient indicates that a node is highly integrated with multiple networks (and might function as a “connector hub”; Rubinov & Sporns, 2010), whereas a low participation coefficient indicates a node is more segregated.

Finally, for each node, the Fisher's z-transformed correlation between this node and every other node across the five networks for every participant was entered into a mixed-effects model. There was one model with five levels for each node (four levels corresponding to each of the “across” networks and the source network of the node serving as a “baseline”). As in Analysis 1, each model had one fixed effect (a categorical predictor of the second [comparison] network) and one random effect (a random intercept for each participant). Planned contrasts were again used to compare the node's connectivity with nodes from its own network to its connectivity with nodes of each of the other networks.

RESULTS

Analysis 1: Within-network versus Across-network Coherence

For each of the five networks, the network's average within-network coherence was compared with its average across-network coherence with respect to each of the other networks (Figure 2). The results indicate that, for all three reference networks (DMN, DAN, SMN), their within-network coherence (0.078, 0.111, and 0.106, respectively) was significantly greater than their cross-network coherence with each of the other networks (p < .01 for every comparison; Table 1).4

Figure 2. 

Within-network and across-network coherence for each of the networks examined. Within-network coherence values are depicted in black. Error bars represent standard error.

Figure 2. 

Within-network and across-network coherence for each of the networks examined. Within-network coherence values are depicted in black. Error bars represent standard error.

Table 1. 
Beta Values for the Comparisons of Within-network Coherence to Cross-network Coherence for Each of the Networks Examined
graphic
graphic

Positive values (shaded gray) indicate that the cross-network coherence is greater than the within-network coherence of the base network (left). Negative values (unshaded) indicate that the cross-network coherence is smaller than the within-network coherence of the base network.

p < .1.

*

p < .05.

Unlike the reference networks, the RdN's within-network coherence (0.050) was marginally lower than its across-network coherence with the DAN (0.065, t = 1.713, p = .087) and numerically lower than with the DMN (0.052) and with the SMN (0.060), although not significantly so (t = 0.296, p > .1, and t = 1.270, p > .1, respectively). For the SpN, its within-network coherence (0.037) was significantly lower than its across-network coherence with the DAN (0.075, t = 3.795, p < .001) and the SMN (0.063, t = 2.782, p = .005) and numerically lower than with the DMN (0.038), but not significantly so (t = 0.103, p > .1; Table 1).

It is clear from these results that the RdN and SpN behave differently from the three reference networks (DMN, DAN, SMN). The orthographic processing networks exhibit the opposite pattern of network coherence compared with the three reference networks: They have lower within-network versus across-network coherence.

Analysis 2: Comparison to Random Networks

The RdN and SpN not only have lower within-network versus cross-network coherence, but their within-network coherence itself is lower than that of the DMN, DAN, and SMN (i.e., each of the reference networks has significantly greater within-network coherence than each of the orthographic networks, p < .001 for all comparisons; see black bars in Figure 2). We had hypothesized that, although the orthographic networks' coherence was lower than that of the reference networks, it would still be greater than the within-network coherence of pseudonetworks, composed of nodes that do not necessarily belong to any specific network. To test this, we compared the within-network coherence of each of the five networks to the within-network coherence values in a sample of 1500 pseudonetworks randomly generated from a gray matter mask of the brain and calculated from the data of the same participants.

The average within-network coherence value of the pseudonetworks was 0.063. As depicted in Figure 3, the SMN and DAN fell within the top 3% of the distribution of within-network coherence values for pseudonetworks, with 98% of pseudonetworks having smaller within-network coherence values than the DAN and 97% having smaller coherence values than the SMN, as we expected. The DMN's within-network coherence just missed the top 20%, with 78% of pseudonetworks having smaller within-network coherence values. However, the picture was strikingly different for the orthographic processing networks. These networks, far from being more coherent than the pseudonetworks, were in fact in the bottom quarter of the distribution. The RdN's coherence value was greater than only 25% of pseudonetworks, and the SpN's coherence value was greater than only 7% of them (Figure 3). Once again, it is clear that the RdN and SpN exhibit a different pattern from the three reference networks (DMN, DAN, and SMN). Given that the RdN and SpN are about as different from the random networks as are the DMN, DAN, and SMN, one would not conclude that they simply behave like collections of randomly selected nodes. Instead, this finding supports the notion that the RdN and SpN constitute a different kind of network.

Figure 3. 

The distribution of within network coherence values for 1500 pseudonetworks composed of nine randomly selected nodes. X-axis represents a running percentage of pseudonetworks. Y-axis represents pseudonetwork density. The relative position of within-network coherence values for the five networks of interest are indicated on the graph with dashed lines. coh = coherence.

Figure 3. 

The distribution of within network coherence values for 1500 pseudonetworks composed of nine randomly selected nodes. X-axis represents a running percentage of pseudonetworks. Y-axis represents pseudonetwork density. The relative position of within-network coherence values for the five networks of interest are indicated on the graph with dashed lines. coh = coherence.

Analysis 3: Individual Node Properties

This analysis provides a closer look at the connectivity patterns of individual nodes and, in particular, the left mid-fusiform gyrus (the VWFA). The average correlation value (across participants and across all 561 unique pairs of nodes) was .055. The degree value for each of the 34 unique nodes corresponds to the number of a node's connections (out of 33 connections each) with values greater than the overall average. The analysis reveals that the average degree for the 34 nodes was 14.9, with a median of 14.5. Table 2 reports the degree of each node in each network (ranked from highest to lowest). Of the 34 nodes, the left fusiform gyrus (VWFA) had the highest degree at 27, followed by the precuneus (associated with the DMN) and the left SMA (associated with the RdN, SpN, and DAN), both at 24. At the other end, the ventral anterior cingulate (part of the DMN) had the lowest degree, with only four connections of above average strength.

Table 2. 
Each Network's Nodes and Their Degree (Deg.), Ranked from Highest (Top) to Lowest Degree
RdNSpNDMNDANSMN
NodeDeg.NodeDeg.NodeDeg.NodeDeg.NodeDeg.
L FG 27 L FG 27 Precuneus 24 L SMA 24 L SMA 24 
L SMA 24 L SMA 24 L MTG 15 L aIPS 22 R dlPMC 20 
L preCG 20 L M1 17 PCC 15 L Occ 21 L S1 18 
L IO 18 L pIPS 14 L MFG 14 R pIPS 21 L M1 17 
L SMG 16 R STS 11 R MTG 13 R premotor 17 L vlPMC 15 
L IFG 10 L IFG 10 MpFC 13 L pIPS 14 L preMotor 14 
L aSTS L aSTS R IPL 11 R aIPS 14 L CB (VI) 13 
L STG L SFG L IPL     R CB (V VI) 
L PH G     VAC         
RdNSpNDMNDANSMN
NodeDeg.NodeDeg.NodeDeg.NodeDeg.NodeDeg.
L FG 27 L FG 27 Precuneus 24 L SMA 24 L SMA 24 
L SMA 24 L SMA 24 L MTG 15 L aIPS 22 R dlPMC 20 
L preCG 20 L M1 17 PCC 15 L Occ 21 L S1 18 
L IO 18 L pIPS 14 L MFG 14 R pIPS 21 L M1 17 
L SMG 16 R STS 11 R MTG 13 R premotor 17 L vlPMC 15 
L IFG 10 L IFG 10 MpFC 13 L pIPS 14 L preMotor 14 
L aSTS L aSTS R IPL 11 R aIPS 14 L CB (VI) 13 
L STG L SFG L IPL     R CB (V VI) 
L PH G     VAC         

See Appendix A for full node names and locations.

To better understand the extent to which a node's connectivity was distributed across multiple networks, we calculated the participation coefficient of each node, indicating how segregated or integrated this node is with regard to the networks we examined (Table 3). A high participation coefficient indicates that a node is more integrated with other networks (i.e., a “global hub”), and a low participation coefficient indicates it is more segregated. We found that the left fusiform gyrus (VWFA) again ranked first among all 34 nodes with the highest participation coefficient. Following it was again the left SMA, an unsurprising finding considering it is a node shared by the RdN, SpN, and DAN. Nodes in the three reference networks were connected, on average, to 55% of the nodes from their own network and only to 44% of the nodes in other networks. The nodes in the orthographic networks, in contrast, were more likely to be connected with nodes in the reference networks (50%) than in their own network (30%). The left fusiform gyrus in particular was connected to 62% of the nodes in the orthographic networks, compared with 95% of the nodes in the reference networks. This percentage of multinetwork connectivity was, by far, the highest percentage of all the nodes we examined (the next highest was 75% for the left inferior occipital gyrus, another node of the RdN that is anatomically close to the left fusiform gyrus). The VWFA's high participation coefficient and its high connectivity with many nodes of the reference networks indicate that it functions as a “connector” or “global” hub (i.e., bridging across multiple networks; Rubinov & Sporns, 2010). In contrast, the nodes of the reference networks function more locally, connecting with other nodes from their own network.

Table 3. 
Each Network's Nodes and Their Participation Coefficient (P.C.), Ranked from Highest (Top) to Lowest Participation Coefficient
RdNSpNDMNDANSMN
NodeP.C.NodeP.C.NodeP.C.NodeP.C.NodeP.C.
L FG 0.749 L FG 0.749 Precuneus 0.743 L SMA 0.748 L SMA 0.748 
L SMA 0.748 L SMA 0.748 R MTG 0.718 R premotor 0.743 L S1 0.742 
L preCG 0.746 L aSTS 0.735 L IPL 0.702 L aIPS 0.741 L preMotor 0.742 
L IO 0.744 L pIPS 0.721 R IPL 0.696 R pIPS 0.738 R dlPMC 0.727 
L STG 0.735 L IFG 0.716 MpFC 0.683 R aIPS 0.731 L CB (VI) 0.719 
L aSTS 0.735 L M1 0.700 L MFG 0.672 L pIPS 0.721 L vlPMC 0.714 
L IFG 0.716 R STS 0.631 VAC 0.662 L Occ 0.690 L M1 0.700 
L PH G 0.713 L SFG 0.612 L MTG 0.624     R CB (V VI) 0.681 
L SMG 0.647     PCC 0.616         
RdNSpNDMNDANSMN
NodeP.C.NodeP.C.NodeP.C.NodeP.C.NodeP.C.
L FG 0.749 L FG 0.749 Precuneus 0.743 L SMA 0.748 L SMA 0.748 
L SMA 0.748 L SMA 0.748 R MTG 0.718 R premotor 0.743 L S1 0.742 
L preCG 0.746 L aSTS 0.735 L IPL 0.702 L aIPS 0.741 L preMotor 0.742 
L IO 0.744 L pIPS 0.721 R IPL 0.696 R pIPS 0.738 R dlPMC 0.727 
L STG 0.735 L IFG 0.716 MpFC 0.683 R aIPS 0.731 L CB (VI) 0.719 
L aSTS 0.735 L M1 0.700 L MFG 0.672 L pIPS 0.721 L vlPMC 0.714 
L IFG 0.716 R STS 0.631 VAC 0.662 L Occ 0.690 L M1 0.700 
L PH G 0.713 L SFG 0.612 L MTG 0.624     R CB (V VI) 0.681 
L SMG 0.647     PCC 0.616         

See Appendix A for full node names and locations.

Finally, after establishing in Analysis 1 that the orthographic networks exhibit a different pattern of within-network versus across-network coherence than the reference networks and after determining that the VWFA was especially well connected, functioning as a “connector hub,” we examined whether the VWFA's correlation with nodes from the reference networks was also statistically greater than its correlation with nodes from the orthographic networks. For each node, we compared the node's connectivity with all other nodes from within its source network to its connectivity with nodes from each of the other networks. The results of this analysis, like the previous ones, reveal that the orthographic networks exhibit a pattern of RS connectivity that is quite different from that of the three reference networks. Although this analysis examines the connectivity of individual nodes (and not entire networks), the same properties are observed: The nodes belonging to the three reference networks were almost uniformly more connected to nodes within their respective networks than they were with nodes across network. The vast majority of nodes from reference networks (18/25) were significantly (p < .05) or marginally (p < .1) more correlated within their source network than they were with any other network.

In contrast, only two of the nodes of the orthographic networks (the left inferior frontal gyrus and the left precentral gyrus, both of the RdN) showed even marginally stronger correlations within than across networks (both were marginally more connected with the RdN than with the DMN, t = −1.740, p = .082 and t = −1.710, p = .088 for the left inferior frontal gyrus and left precentral gyrus, respectively). In fact, 8 of the 13 nodes associated with orthographic networks showed significantly weaker correlations with nodes from within the orthographic networks than across with at least one other network. However, none showed significant effects of the same magnitude as the left fusiform gyrus (VWFA). The left fusiform gyrus had significantly stronger correlations with nodes from both the SMN and DAN (t = 2.266, p = .024, and t = 3.106, p = .002, respectively) than with nodes of the orthographic network and also numerically stronger correlations with the DMN than with the orthographic networks (t = 1.521, p = .129; Figure 4 shows the results for the VWFA alongside three representative nodes from the DMN, DAN, and SMN).

Figure 4. 

Within-network and across-network coherence values of individual nodes. The left fusiform gyrus (VWFA, a key node of the RdN and SpN), the posterior cingulate cortex (PCC; of the DMN), the right posterior inferior parietal sulcus (R pIPS; of the DAN), and the left primary motor cortex (L M1; of the SMN). Average node connectivity with nodes from the source network is in black. Error bars represent standard error.

Figure 4. 

Within-network and across-network coherence values of individual nodes. The left fusiform gyrus (VWFA, a key node of the RdN and SpN), the posterior cingulate cortex (PCC; of the DMN), the right posterior inferior parietal sulcus (R pIPS; of the DAN), and the left primary motor cortex (L M1; of the SMN). Average node connectivity with nodes from the source network is in black. Error bars represent standard error.

In summary, these results reveal that the left fusiform gyrus (the VWFA) is very well connected, in that it has many strong connections, which extend across all the networks examined. This pattern of broad, strong connectivity of the VWFA is clearly depicted in Figure 5. and contrasts markedly with a well-studied node of the DMN—the posterior cingulate cortex. Whereas the VWFA's strong connections (Figure 5, top) are distributed across nodes of multiple different networks (depicted by different colored spheres), the posterior cingulate cortex's strong connections (Figure 5, bottom) are primarily limited to other DMN nodes (depicted as red spheres).

Figure 5. 

Above-average connections for the VWFA (top) and the posterior cingulate cortex (bottom), depicted on a transparent brain. Line color and width represent the strength of the correlation. Node color indicates the network to which the node belongs: RdN and SpN, navy; DMN, red; DAN, fuchsia; SMN, cyan. Image visualized using BrainNet Viewer (Xia et al., 2013). The figure depicts the greater cross-network connectivity of the VWFA compared with the posterior cingulate cortex.

Figure 5. 

Above-average connections for the VWFA (top) and the posterior cingulate cortex (bottom), depicted on a transparent brain. Line color and width represent the strength of the correlation. Node color indicates the network to which the node belongs: RdN and SpN, navy; DMN, red; DAN, fuchsia; SMN, cyan. Image visualized using BrainNet Viewer (Xia et al., 2013). The figure depicts the greater cross-network connectivity of the VWFA compared with the posterior cingulate cortex.

DISCUSSION

Are orthographic processing networks different from other networks? To address this question, we examined various properties of RS connectivity for the orthographic processing networks (both reading and spelling) as well as for their key node—the left mid-fusiform gyrus, often referred to as the VWFA (Cohen & Dehaene, 2004; Cohen et al., 2000). Properties of the orthographic processing networks and the left fusiform gyrus were compared with those of well-established “reference” networks, which were identified from meta-analyses, and their individual nodes: the DMN, the SMN, and the DAN. With regard to network properties, we observed that (1) each of the three reference networks exhibited a “classic” pattern of network connectivity, with stronger connectivity between nodes belonging to the same network (within-network coherence) than between nodes belonging to different networks (across-network coherence). In contrast, both of the orthographic networks exhibited the opposite pattern, with weaker within-network than across-network coherence; (2) when compared with 1500 pseudonetworks composed of randomly selected nodes, the within-network coherence of the reference networks was greater than the majority of the pseudonetworks, whereas the orthographic processing networks exhibited relatively low within-network coherence compared with the pseudonetworks. With regard to individual nodes and, in particular, the left fusiform gyrus (VWFA), we report three key findings: (1) In terms of the graph theoretical measure of “degree,” the VWFA had the highest degree of all 34 network nodes examined; (2) relatedly, the VWFA had the highest participation coefficient of the 34 nodes; and (3) the VWFA exhibited stronger connectivity to each of the reference networks than it did to the two orthographic networks, in direct contrast to nodes from the reference networks, almost all of which showed significantly stronger connectivity with nodes from their respective source networks than with nodes from other networks.

The “Network” Status of Orthographic Processing Networks

Our finding that the RdN exhibited lower within-network versus across-network coherence than the three reference networks (DMN, SMN, DAN) is consistent with previous findings (Cole et al., 2014; Vogel et al., 2013; Yeo et al., 2011), showing that the nodes of the RdN do not form a community of nodes whose activity at rest is highly intercorrelated. Our further finding that the SpN patterns, like the RdN, add to these previous results strengthens the conclusion that, at least at rest, orthographic processing networks do not show the internal coherence and community structure exhibited by other networks.

In addition to exhibiting lower within-network versus across-network coherence, the orthographic networks displayed lower overall within-network coherence values than the reference networks. To better evaluate this finding, we examined the within-network coherence values of 1500 randomly generated pseudonetworks of a similar size. As expected, all three of the reference networks showed higher internal coherence than the vast majority (78–98%) of the pseudonetworks. Based on the argument that RS connectivity reflects the patterns of coactivation during task performance (Tavor et al., 2016; Cole et al., 2014), our expectation was that SpN and RdN would at least show higher within-network connectivity than pseudonetworks, albeit not as high as the reference networks. However, we found that the orthographic processing networks exhibit lower within-network connectivity than the majority of pseudonetworks, placing the RdN in the bottom 25% and the SpN in the bottom 7% in terms of within-network coherence relative to the pseudonetworks. Importantly, this difference between the reference and orthographic processing networks is not due to the average neurotopographic distance between the nodes of the networks, as the RdN's average node distance (58 mm) is smaller than that of the DMN (85 mm), DAN (70 mm), and SMN (64 mm), and the SpN's average node distance (74 mm) places it between the DMN and DAN. This finding underscores the conclusion that orthographic processing networks do not have key characteristics of at least certain evolutionarily well-established networks. However, the finding that their connectivity is considerably lower than that of random sets of nodes is intriguing, and we discuss possible implications of this finding in the final section of this General Discussion.

One question that might arise is whether the properties of the orthographic networks we have identified could be due to improper localization of their constituent nodes. Arguably, this could have occurred given that node selection was based on meta-analyses and also given the possibility of greater individual anatomical variability in node location for orthographic network nodes compared with the nodes of the reference networks. This possibility cannot definitively be ruled out, but there are some reasons that mitigate this concern. First, although it is certainly plausible that there is greater individual variability in node location for orthographic networks than others, this remains to be empirically established. Second, although we used a different approach to identifying RdN nodes than used by Vogel et al. (2013), we found similar results regarding the within-network compared with across-network connectivity patterns. These patterns are also consistent with the more general findings of Cole et al. (2014) and Yeo et al. (2011) regarding the weak community structure of the RdN. Furthermore and more directly to the point, in earlier work on this project, we carried out a subset of the analyses using individually identified VWFA nodes (based on reading localizer data) and found the same relationships between the VWFA and the RdN and DAN, supporting the findings we reported here that are based on the meta-analysis-defined left fusiform gyrus node.

Connectivity of the Left Mid-fusiform Gyrus: Implications

There has been a great deal of interest in understanding whether or not the left fusiform gyrus (VWFA) is specialized for reading (e.g., Vogel et al., 2012; Dehaene & Cohen, 2011; Tsapkini & Rapp, 2010; Binder, Medler, Westbury, Liebenthal, & Buchanan, 2006; Cohen & Dehaene, 2004; McCandliss, Cohen, & Dehaene, 2003). Although this study does not directly address this issue, understanding the connectivity properties of this region may be relevant for understanding its functions. In this regard, we found that the within-network versus across-network connectivity of individual nodes mirrored the patterns observed for their “source” networks: Nodes of the reference networks had greater within-network versus across-network coherence values, whereas nodes of the orthographic networks and, in particular, the VWFA showed the opposite pattern, indicating strong connectivity with nodes outside the orthographic processing network. This characteristic of the VWFA was also apparent in the analysis of the degree of each node, where degree indexes the number of strong (above average) connections a region has with other regions (Bullmore & Sporns, 2009). The degree analysis revealed a very highly connected VWFA. Of the 34 nodes across the five networks, the VWFA had the highest degree value. Finally, when we considered the extent to which each node was strongly connected with nodes of other networks, we found that the left fusiform gyrus region had the highest participation coefficient of all the nodes examined, with strong connections across all three reference networks and specifically with strong connections to 78–100% of the reference networks' nodes (95% in total). Overall, both the reference network nodes and the other orthographic network nodes showed a more narrowly distributed connectivity pattern than did the VWFA.

Our finding that the VWFA is highly connected with a large number of regions across a broad range of functional networks presents something of a challenge for the Vogel et al. (2014) argument. Vogel et al. argue that, due to its strong connectivity with the attention network, the left fusiform gyrus should be considered a visuoattentional processor whose specific function is to act on groups of visual stimuli (not only letters) that require attentional processing. Our findings indicate that this region, although certainly strongly connected with the nodes of the attention network, is also strongly connected with the DMN and the SMN. Consistent with this broad connectivity is the Stevens et al. (2017) finding that the VWFA is highly connected to areas of the language network, especially Wernicke's area. Furthermore, it is worth pointing out that, although Vogel et al. (2014) have proposed that the left fusiform gyrus is a general visuoattentional processor, the region does not generally appear in meta-analyses of attentional networks (e.g., Wager et al., 2004; Corbetta et al., 1998), and attentional deficits have not been reported in response to lesions in this area. Overall, these findings instead suggest that, at least in terms of the intrinsic connectivity revealed by RS-fMRI, this area in the mid-fusiform gyrus has hub-like properties. According to Buckner et al. (2009), these properties make hubs intrinsically well suited for integrating information across multiple, disparate domains and processes. As we suggest below, this may make the left fusiform gyrus region especially appropriate for the specific challenges required for orthographic processing.

High-level Integrative Networks: Meeting the Challenges of Orthographic Processing

The patterns of connectivity revealed by RS-fMRI indicate that the set of brain regions, which has been shown to be consistently active during orthographic processing, lacks some of the key network characteristics exhibited by the evolutionarily well-established networks that served as reference networks in this study. Why is orthographic processing different? Here, we briefly consider several possibilities that, we note, are not mutually exclusive.

One possibility is that the phylogenetic youth of orthographic functions did not allow enough time for the development of the RS network properties seen in the reference networks. Another possibility is that these differences are due to the relative recency of orthographic processing within an individual's ontogenetic developmental trajectory. Reading and spelling are usually only taught formally around age 5 or 6 years, and these skills continue to develop throughout childhood and into early adolescence. It is possible that this later acquisition is responsible for the absence of the network coherence seen in skills practiced since birth/early childhood (Schlaggar & McCandliss, 2007). A third possibility is the one proposed by Vogel et al. (2014), which is that orthographic processing is different because it lacks dedicated neural machinery, relying instead on general-purpose processors. Implicit in this proposal is the assumption that the lack of specialized orthographic processes precludes the development of network structure. Although not unreasonable, this assumption strongly links the issues of specialization and network structure. However, these are not necessarily linked as it is logically possible that general-purpose processors, which are frequently and strongly coactivated, could come to develop network characteristics. In fact, the notion that coactivation while executing tasks should be reflected in connectivity at rest has been a premise of much work on RS-fMRI connectivity (e.g., Power et al., 2011; Smith et al., 2009; Fox & Raichle, 2007). In other words, it does not necessarily follow that the presence of network structure implies specialization or vice versa.

We propose another interpretation of the data: Because of the cognitive requirements of reading and spelling, orthographic processing occurs at a higher level on the neural processing hierarchy than the skills supported by the reference networks. Although skills such as shifting attention and motor control are also highly internally complex, we argue that the cognitive challenge of reading and spelling is that they require the seamless integration of multiple functions not only within a single domain but also across multiple domains. These domains include, as Vogel et al. (2012) have noted, the shifting and directing of attention, but also visual processes (Dehaene et al., 2010), language (Stevens et al., 2017), motor functions (James & Engelhardt, 2012; Longcamp et al., 2008), cognitive control (Horowitz-Kraus et al., 2015), and long-term and working memory (Rapp et al., 2016), among others. In other words, orthographic processing networks may be different from the reference networks, not only because of their phylogenetic and ontogenetic recency but also because they are what we will refer to as “high-level integrative networks.” These are networks that recruit multiple lower level, “precompiled,” modularized functions. We further hypothesize that the hallmark, diagnostic feature of such networks may be that their coherence values are lower than not only those of the lower level functions they draw upon but also the coherence observed for random collections of nodes. Whether or not such complex networks become precompiled and modularized given sufficient phylogenetic or ontogenetic experience is a question to be addressed in future research examining a range of other cognitive functions. This proposal also opens the possibility that networks may vary in terms of their network characteristics depending on where they are situated on the integrative hierarchy and that this, in turn, may interact with their phylogenetic and ontogenetic properties. We acknowledge that this proposal is, for the moment, somewhat post hoc, but as we have argued, it is consistent with a number of observations. Nonetheless, it will certainly need to be more directly evaluated in future research.

This proposal may also provide insights into the role of the VWFA in orthographic processing. As we have shown, this area has high intrinsic connectivity, making it naturally suited for bridging different domains. In other words, it is well positioned to address the requirement of orthographic processing for efficient cross-domain integration. Whether or not this area has become specialized for orthographic processing is a separate issue. What we emphasize here is that, given the cognitive demands of orthographic processing and the extensive intrinsic connectivity of the left mid-fusiform region, this area is well positioned to assume a key role in orthographic processing.

Conclusions

The findings from this research clearly indicate that the set of brain areas involved in orthographic processing have unique network properties relative to those of well-established, well-studied networks, such as the DMN, DAN, and SMN. We propose that because reading and spelling require integration and processing of information across multiple domains, orthographic processing is supported by high-level integrative networks. These high-level integrative networks are characterized by their recruitment of lower level, modularized networks that have traditional network properties. We further suggest that the recruitment of the left mid-fusiform area, known as the VWFA for orthographic processing, is understandable given both the cognitive demands of orthographic processing and the intrinsic strong connectivity of this area with a broad range of brain regions. These hypotheses require further scrutiny, including, but not limited to, the study of other skills that range in their evolutionary recency and complexity and other measurements of RS connectivity (such as modularity, centrality, or efficiency; Bullmore & Sporns, 2009). Furthermore, it will be important to examine, with larger groups of participants, the impact of developmental age and skill level on the emergence of high-level integrative network properties not only in reading and spelling but also in other evolutionarily recent cognitive domains.

Appendix A

Table A1. 
Node Names and Coordinates for Each of the Networks Examined
NetworkNameAbbreviationShared NetworksCoordinates (MNI)
RdN L posterior superior temporal gyrus L pSTG   (−54, −36, 16) 
L supramarginal gyrus L SMG   (−49, −57, 28) 
L fusiform gyrus L FG SpN (−47, −59, −16) 
L inferior lateral occipital cortex L IO   (−44, −74, −4) 
L parahippocampal gyrus L PH G   (−32, −36, −12) 
L anterior superior temporal sulcus L aSTS SpN (−58, −12, −3) 
L inferior frontal gyrus L IFG SpN (−48, 17, 17) 
L precentral gyrus (BA 6) L preCG   (−46, 2, 42) 
L supplementary motor area L SMA SpN, DAN (−4, 19, 50) 
SpN L inferior frontal gyrus L IFG RdN (−48, 17, 17) 
L superior frontal gyrus L SFG   (−22, −8, 54) 
L supplementary motor area L SMA DAN, RdN (−4, 19, 50) 
L primary motor area L M1 SMN (−38, −22, 57) 
L posterior intraparietal sulcus L pIPS DAN (−25, −63, 46) 
L anterior superior temporal sulcus L aSTS RdN (−58, −12, −3) 
R superior temporal sulcus R STS   (52, −12, −6) 
L fusiform gyrus L FG RdN (−47, −59, −16) 
DMN Precuneus Precuneus   (−2, −56, 50) 
Posterior cingulate cortex PCC   (−3, −52, 25) 
Ventral anterior cingulate VAC   (3, 35, −17) 
R inferior parietal lobule R IPL   (58, −26, 24) 
Medial prefrontal cortex MpFC   (−1, 56, 10) 
R middle temporal gyrus R MTG   (51, −67, 23) 
L middle temporal gyrus L MTG   (−44, −67, 23) 
L middle frontal gyrus L MFG   (−27, 23, 43) 
L inferior parietal lobule L IPL   (−59, −34, 31) 
DAN R anterior intraparietal sulcus R aIPS   (32, −40, 46) 
L anterior intraparietal sulcus L aIPS   (−36, −47, 50) 
R posterior intraparietal sulcus R pIPS   (26, −64, 50) 
L posterior intraparietal sulcus L pIPS SpN (−25, −63, 46) 
R premotor cortex R premotor   (40, 2, 46) 
L supplementary motor area L SMA SpN, RdN (−4, 19, 50) 
L occipital cortex L Occ   (−36, −82, 0) 
SMN R cerebellum (Lobes V, VI) R CB (V VI)   (18, −50, −20) 
L supplementary motor area L SMA   (−2, 0, 54) 
L primary motor area L M1 SpN (−38, −22, 57) 
R supplementary motor area R SMA   (2, 0, 54) 
L premotor cortex L preMotor   (−40, −9, 57) 
L cerebellum (Lobe VI) L CB (VI)   (−24, −52, −22) 
R dorsolateral premotor cortex R dlPMC   (44, −8, 56) 
L ventrolateral premotor cortex L vlPMC   (−50, −12, 48) 
L primary somatosensory area L S1   (−40, −34, 60) 
NetworkNameAbbreviationShared NetworksCoordinates (MNI)
RdN L posterior superior temporal gyrus L pSTG   (−54, −36, 16) 
L supramarginal gyrus L SMG   (−49, −57, 28) 
L fusiform gyrus L FG SpN (−47, −59, −16) 
L inferior lateral occipital cortex L IO   (−44, −74, −4) 
L parahippocampal gyrus L PH G   (−32, −36, −12) 
L anterior superior temporal sulcus L aSTS SpN (−58, −12, −3) 
L inferior frontal gyrus L IFG SpN (−48, 17, 17) 
L precentral gyrus (BA 6) L preCG   (−46, 2, 42) 
L supplementary motor area L SMA SpN, DAN (−4, 19, 50) 
SpN L inferior frontal gyrus L IFG RdN (−48, 17, 17) 
L superior frontal gyrus L SFG   (−22, −8, 54) 
L supplementary motor area L SMA DAN, RdN (−4, 19, 50) 
L primary motor area L M1 SMN (−38, −22, 57) 
L posterior intraparietal sulcus L pIPS DAN (−25, −63, 46) 
L anterior superior temporal sulcus L aSTS RdN (−58, −12, −3) 
R superior temporal sulcus R STS   (52, −12, −6) 
L fusiform gyrus L FG RdN (−47, −59, −16) 
DMN Precuneus Precuneus   (−2, −56, 50) 
Posterior cingulate cortex PCC   (−3, −52, 25) 
Ventral anterior cingulate VAC   (3, 35, −17) 
R inferior parietal lobule R IPL   (58, −26, 24) 
Medial prefrontal cortex MpFC   (−1, 56, 10) 
R middle temporal gyrus R MTG   (51, −67, 23) 
L middle temporal gyrus L MTG   (−44, −67, 23) 
L middle frontal gyrus L MFG   (−27, 23, 43) 
L inferior parietal lobule L IPL   (−59, −34, 31) 
DAN R anterior intraparietal sulcus R aIPS   (32, −40, 46) 
L anterior intraparietal sulcus L aIPS   (−36, −47, 50) 
R posterior intraparietal sulcus R pIPS   (26, −64, 50) 
L posterior intraparietal sulcus L pIPS SpN (−25, −63, 46) 
R premotor cortex R premotor   (40, 2, 46) 
L supplementary motor area L SMA SpN, RdN (−4, 19, 50) 
L occipital cortex L Occ   (−36, −82, 0) 
SMN R cerebellum (Lobes V, VI) R CB (V VI)   (18, −50, −20) 
L supplementary motor area L SMA   (−2, 0, 54) 
L primary motor area L M1 SpN (−38, −22, 57) 
R supplementary motor area R SMA   (2, 0, 54) 
L premotor cortex L preMotor   (−40, −9, 57) 
L cerebellum (Lobe VI) L CB (VI)   (−24, −52, −22) 
R dorsolateral premotor cortex R dlPMC   (44, −8, 56) 
L ventrolateral premotor cortex L vlPMC   (−50, −12, 48) 
L primary somatosensory area L S1   (−40, −34, 60) 

Coordinates are in MNI space. Node names are based on the Yale Atlas (Lacadie, Fulbright, Arora, Constable, & Papademetris, 2008). L = left; R = right.

Acknowledgments

This research was supported by National Institutes of Health Clinical Research Center (grant DC012283). We gratefully acknowledge Shanna Murray and Jennifer Shea for their contribution to data collection, Yuan Tao for contribution to data processing, and Robert W. Wiley for his valuable advice on statistical analysis.

Reprint requests should be sent to Gali Ellenblum, Department of Cognitive Science, Johns Hopkins University, 3400 N Charles St., Krieger 239, Baltimore, MD 21218, or via e-mail: gali@jhu.edu.

Notes

1. 

This is similar to an undirected nonthresholded association matrix, as described by Bullmore and Sporns (2009).

2. 

The participants in this study served as a control group for a study involving poststroke language impairments, explaining the older average age than in most neuroimaging studies.

3. 

Supplementary material for this paper can be retrieved from https://jh.box.com/s/jqlgatxervn257npiogcld5ju68of55f.

4. 

At an earlier point in the project, Analyses 1–3 were carried out with eight of the participants in the current set and yielded the same conclusions as we report here. The fact that with the addition of nine participants the same conclusions are supported provides evidence of the reliability of these findings.

REFERENCES

Alakörkkö
,
T.
,
Saarimäki
,
H.
,
Glerean
,
E.
,
Saramäki
,
J.
, &
Korhonen
,
O.
(
2017
).
Effects of spatial smoothing on functional brain networks
.
European Journal of Neuroscience
,
46
,
2471
2480
.
Behrmann
,
M.
,
Geng
,
J. J.
, &
Shomstein
,
S.
(
2004
).
Parietal cortex and attention
.
Current Opinion in Neurobiology
,
14
,
212
217
.
Binder
,
J. R.
,
Medler
,
D. A.
,
Westbury
,
C. F.
,
Liebenthal
,
E.
, &
Buchanan
,
L.
(
2006
).
Tuning of the human left fusiform gyrus to sublexical orthographic structure
.
Neuroimage
,
33
,
739
748
.
Biswal
,
B.
,
Yetkin
,
F. Z.
,
Haughton
,
V. M.
, &
Hyde
,
J. S.
(
1995
).
Functional connectivity in the motor cortex of resting human brain using echo-planar MRI
.
Magnetic Resonance in Medicine
,
34
,
537
541
.
Buckner
,
R. L.
,
Sepulcre
,
J.
,
Talukdar
,
T.
,
Krienen
,
F. M.
,
Liu
,
H.
,
Hedden
,
T.
, et al
(
2009
).
Cortical hubs revealed by intrinsic functional connectivity: Mapping, assessment of stability, and relation to Alzheimer's disease
.
Journal of Neuroscience
,
29
,
1860
1873
.
Bullmore
,
E.
, &
Sporns
,
O.
(
2009
).
Complex brain networks: Graph theoretical analysis of structural and functional systems
.
Nature Reviews Neuroscience
,
10
,
186
198
.
Cohen
,
L.
, &
Dehaene
,
S.
(
2004
).
Specialization within the ventral stream: The case for the visual word form area
.
Neuroimage
,
22
,
466
476
.
Cohen
,
L.
,
Dehaene
,
S.
,
Naccache
,
L.
,
Lehéricy
,
S.
,
Dehaene-Lambertz
,
G.
,
Hénaff
,
M. A.
, et al
(
2000
).
The visual word form area: Spatial and temporal characterization of an initial stage of reading in normal subjects and posterior split-brain patients
.
Brain
,
123
,
291
307
.
Cole
,
M. W.
,
Bassett
,
D. S.
,
Power
,
J. D.
,
Braver
,
T. S.
, &
Petersen
,
S. E.
(
2014
).
Intrinsic and task-evoked network architectures of the human brain
.
Neuron
,
83
,
238
251
.
Corbetta
,
M.
(
1998
).
Frontoparietal cortical networks for directing attention and the eye to visual locations: Identical, independent, or overlapping neural systems?
Proceedings of the National Academy of Sciences, U.S.A.
,
95
,
831
838
.
Corbetta
,
M.
,
Akbudak
,
E.
,
Conturo
,
T. E.
,
Snyder
,
A. Z.
,
Ollinger
,
J. M.
,
Drury
,
H. A.
, et al
(
1998
).
A common network of functional areas for attention and eye movement
.
Neuron
,
21
,
761
773
.
Corbetta
,
M.
, &
Shulman
,
G. L.
(
2002
).
Control of goal-directed and stimulus-driven attention in the brain
.
Nature Reviews Neuroscience
,
3
,
201
215
.
Dehaene
,
S.
(
2009
).
Reading in the brain: The new science of how we read
(1st ed.).
New York, NY
:
Viking Penguin
.
Dehaene
,
S.
, &
Cohen
,
L.
(
2011
).
The unique role of the visual word form area in reading
.
Trends in Cognitive Sciences
,
15
,
254
262
.
Dehaene
,
S.
,
Pegado
,
F.
,
Braga
,
L. W.
,
Ventura
,
P.
,
Filho
,
G. N.
,
Jobert
,
A.
, et al
(
2010
).
How learning to read changes the cortical networks for vision and language
.
Science
,
330
,
1359
1364
.
Fox
,
M. D.
,
Corbetta
,
M.
,
Snyder
,
A. Z.
,
Vincent
,
J. L.
, &
Raichle
,
M. E.
(
2006
).
Spontaneous neuronal activity distinguishes human dorsal and ventral attention systems
.
Proceedings of the National Academy of Sciences, U.S.A.
,
103
,
10046
10051
.
Fox
,
M. D.
, &
Raichle
,
M. E.
(
2007
).
Spontaneous fluctuations in brain activity observed with functional magnetic resonance imaging
.
Nature Reviews Neuroscience
,
8
,
700
711
.
Fox
,
M. D.
,
Zhang
,
D.
,
Snyder
,
A. Z.
, &
Raichle
,
M. E.
(
2009
).
The global signal and observed anticorrelated resting state brain networks
.
Journal of Neurophysiology
,
101
,
3270
3283
.
Fransson
,
P.
, &
Marrelec
,
G.
(
2008
).
The precuneus/posterior cingulate cortex plays a pivotal role in the default mode network: Evidence from a partial correlation network analysis
.
Neuroimage
,
42
,
1178
1184
.
Hillis
,
A. E.
, &
Tippett
,
D. C.
(
2014
).
Stroke recovery: Surprising influences and residual consequences
.
Advances in Medicine
,
2014
,
378263
.
Horowitz-Kraus
,
T.
,
DiFrancesco
,
M.
,
Kay
,
B.
,
Wang
,
Y.
, &
Holland
,
S. K.
(
2015
).
Increased resting-state functional connectivity of visual- and cognitive-control brain networks after training in children with reading difficulties
.
Neuroimage. Clinical
,
8
,
619
630
.
James
,
K. H.
, &
Engelhardt
,
L.
(
2012
).
The effects of handwriting experience on functional brain development in pre-literate children
.
Trends in Neuroscience and Education
,
1
,
32
42
.
Jenkinson
,
M.
,
Beckmann
,
C. F.
,
Behrens
,
T. E. J.
,
Woolrich
,
M. W.
, &
Smith
,
S. M.
(
2012
).
FSL
.
Neuroimage
,
62
,
782
790
.
Jobard
,
G.
,
Crivello
,
F.
, &
Tzourio-Mazoyer
,
N.
(
2003
).
Evaluation of the dual route theory of reading: A metanalysis of 35 neuroimaging studies
.
Neuroimage
,
20
,
693
712
.
Keller
,
J. B.
,
Hedden
,
T.
,
Thompson
,
T. W.
,
Anteraper
,
S. A.
,
Gabrieli
,
J. D.
, &
Whitfield-Gabrieli
,
S.
(
2015
).
Resting-state anticorrelations between medial and lateral prefrontal cortex: Association with working memory, aging, and individual differences
.
Cortex
,
64
,
271
280
.
Koyama
,
M. S.
,
Di Martino
,
A.
,
Zuo
,
X. N.
,
Kelly
,
C.
,
Mennes
,
M.
,
Jutagir
,
D. R.
, et al
(
2011
).
Resting-state functional connectivity indexes reading competence in children and adults
.
Journal of Neuroscience
,
31
,
8617
8624
.
Koyama
,
M. S.
,
Kelly
,
C.
,
Shehzad
,
Z.
,
Penesetti
,
D.
,
Castellanos
,
F. X.
, &
Milham
,
M. P.
(
2010
).
Reading networks at rest
.
Cerebral Cortex
,
20
,
2549
2559
.
Lacadie
,
C.
,
Fulbright
,
R.
,
Arora
,
J.
,
Constable
,
R.
, &
Papademetris
,
X.
(
2008
).
Brodmann areas defined in MNI space using a new tracing tool in BioImage Suite
.
Proceedings of the 14th Annual Meeting of the Organization for Human Brain Mapping
.
Melbourne, Australia
. p.
771
.
Laird
,
A. R.
,
Eickhoff
,
S. B.
,
Li
,
K.
,
Robin
,
D. A.
,
Glahn
,
D. C.
, &
Fox
,
P. T.
(
2009
).
Investigating the functional heterogeneity of the default mode network using coordinate-based meta-analytic modeling
.
Journal of Neuroscience
,
29
,
14496
14505
.
Li
,
L.
,
Liu
,
J.
,
Chen
,
F.
,
Feng
,
L.
,
Li
,
H.
,
Tian
,
J.
, et al
(
2013
).
Resting state neural networks for visual Chinese word processing in Chinese adults and children
.
Neuropsychologia
,
51
,
1571
1583
.
Longcamp
,
M.
,
Boucard
,
C.
,
Gilhodes
,
J. C.
,
Anton
,
J. L.
,
Roth
,
M.
,
Nazarian
,
B.
, et al
(
2008
).
Learning through hand-or typewriting influences visual recognition of new graphic shapes: Behavioral and functional imaging evidence
.
Journal of Cognitive Neuroscience
,
20
,
802
815
.
Martin
,
A.
,
Schurz
,
M.
,
Kronbichler
,
M.
, &
Richlan
,
F.
(
2015
).
Reading in the brain of children and adults: A meta-analysis of 40 functional magnetic resonance imaging studies: Reading in the brain of children and adults
.
Human Brain Mapping
,
36
,
1963
1981
.
McCandliss
,
B. D.
,
Cohen
,
L.
, &
Dehaene
,
S.
(
2003
).
The visual word form area: Expertise for reading in the fusiform gyrus
.
Trends in Cognitive Sciences
,
7
,
293
299
.
Planton
,
S.
,
Jucla
,
M.
,
Roux
,
F. E.
, &
Démonet
,
J. F.
(
2013
).
The “handwriting brain”: A meta-analysis of neuroimaging studies of motor versus orthographic processes
.
Cortex
,
49
,
2772
2787
.
Power
,
J. D.
,
Barnes
,
K. A.
,
Snyder
,
A. Z.
,
Schlaggar
,
B. L.
, &
Petersen
,
S. E.
(
2012
).
Spurious but systematic correlations in functional connectivity MRI networks arise from subject motion
.
Neuroimage
,
59
,
2142
2154
.
Power
,
J. D.
,
Cohen
,
A. L.
,
Nelson
,
S. M.
,
Wig
,
G. S.
,
Barnes
,
K. A.
,
Church
,
J. A.
, et al
(
2011
).
Functional network organization of the human brain
.
Neuron
,
72
,
665
678
.
Power
,
J. D.
,
Mitra
,
A.
,
Laumann
,
T. O.
,
Snyder
,
A. Z.
,
Schlaggar
,
B. L.
, &
Petersen
,
S. E.
(
2014
).
Methods to detect, characterize, and remove motion artifact in resting state fMRI
.
Neuroimage
,
84
,
320
341
.
Purcell
,
J. J.
,
Jiang
,
X.
, &
Eden
,
G. F.
(
2017
).
Shared orthographic neuronal representations for spelling and reading
.
Neuroimage
,
147
,
554
567
.
Purcell
,
J. J.
,
Turkeltaub
,
P. E.
,
Eden
,
G. F.
, &
Rapp
,
B.
(
2011
).
Examining the central and peripheral processes of written word production through meta-analysis
.
Frontiers in Psychology
,
2
,
239
.
R Core Team
. (
2014
).
R: A language and environment for statistical computing
.
Vienna, Austria
:
R Foundation for Statistical Computing
. https://www.R-project.org/.
Raichle
,
M. E.
,
MacLeod
,
A. M.
,
Snyder
,
A. Z.
,
Powers
,
W. J.
,
Gusnard
,
D. A.
, &
Shulman
,
G. L.
(
2001
).
A default mode of brain function
.
Proceedings of the National Academy of Sciences, U.S.A.
,
98
,
676
682
.
Rapp
,
B.
, &
Dufor
,
O.
(
2011
).
The neurotopography of written word production: An fMRI investigation of the distribution of sensitivity to length and frequency
.
Journal of Cognitive Neuroscience
,
23
,
4067
4081
.
Rapp
,
B.
, &
Lipka
,
K.
(
2011
).
The literate brain: The relationship between spelling and reading
.
Journal of Cognitive Neuroscience
,
23
,
1180
1197
.
Rapp
,
B.
,
Purcell
,
J.
,
Hillis
,
A. E.
,
Capasso
,
R.
, &
Miceli
,
G.
(
2016
).
Neural bases of orthographic long-term memory and working memory in dysgraphia
.
Brain
,
139
,
588
604
.
Rehme
,
A. K.
,
Eickhoff
,
S. B.
,
Rottschy
,
C.
,
Fink
,
G. R.
, &
Grefkes
,
C.
(
2012
).
Activation likelihood estimation meta-analysis of motor-related neural activity after stroke
.
Neuroimage
,
59
,
2771
2782
.
Rothlein
,
D.
, &
Rapp
,
B.
(
2014
).
The similarity structure of distributed neural responses reveals the multiple representations of letters
.
Neuroimage
,
89
,
331
344
.
Rubinov
,
M.
, &
Sporns
,
O.
(
2010
).
Complex network measures of brain connectivity: Uses and interpretations
.
Neuroimage
,
52
,
1059
1069
.
Rumeau
,
C.
,
Tzourio
,
N.
,
Murayama
,
N.
,
Peretti-Viton
,
P.
,
Joliot
,
M.
,
Mazoyer
,
B.
, et al
(
1994
).
Location of hand function in the sensorimotor cortex: MR and functional correlation
.
American Journal of Neuroradiology
,
15
,
567
572
.
Satterthwaite
,
T. D.
,
Ciric
,
R.
,
Roalf
,
D. R.
,
Davatzikos
,
C.
,
Bassett
,
D. S.
, &
Wolf
,
D. H.
(
2019
).
Motion artifact in studies of functional connectivity: Characteristics and mitigation strategies
.
Human Brain Mapping
,
40
,
2033
2051
.
Schlaggar
,
B. L.
, &
McCandliss
,
B. D.
(
2007
).
Development of neural systems for reading
.
Annual Review of Neuroscience
,
30
,
475
503
.
Scholvinck
,
M. L.
,
Maier
,
A.
,
Ye
,
F. Q.
,
Duyn
,
J. H.
, &
Leopold
,
D. A.
(
2010
).
Neural basis of global resting-state fMRI activity
.
Proceedings of the National Academy of Sciences
,
107
,
10238
10243
.
Schurz
,
M.
,
Wimmer
,
H.
,
Richlan
,
F.
,
Ludersdorfer
,
P.
,
Klackl
,
J.
, &
Kronbichler
,
M.
(
2015
).
Resting-state and task-based functional brain connectivity in developmental dyslexia
.
Cerebral Cortex
,
25
,
3502
3514
.
Smith
,
S. M.
,
Fox
,
P. T.
,
Miller
,
K. L.
,
Glahn
,
D. C.
,
Fox
,
P. M.
,
Mackay
,
C. E.
, et al
(
2009
).
Correspondence of the brain's functional architecture during activation and rest
.
Proceedings of the National Academy of Sciences, U.S.A.
,
106
,
13040
13045
.
Song
,
X. W.
,
Dong
,
Z. Y.
,
Long
,
X. Y.
,
Li
,
S. F.
,
Zuo
,
X. N.
,
Zhu
,
C. Z.
, et al
(
2011
).
REST: A toolkit for resting-state functional magnetic resonance imaging data processing
.
PLoS One
,
6
,
e25031
.
Stevens
,
W. D.
,
Kravitz
,
D. J.
,
Peng
,
C. S.
,
Tessler
,
M. H.
, &
Martin
,
A.
(
2017
).
Privileged functional connectivity between the visual word form area and the language system
.
Journal of Neuroscience
,
37
,
5288
5297
.
Tavor
,
I.
,
Parker Jones
,
O.
,
Mars
,
R. B.
,
Smith
,
S. M.
,
Behrens
,
T. E.
, &
Jbabdi
,
S.
(
2016
).
Task-free MRI predicts individual differences in brain activitiy during task performance
.
Science
,
352
,
216
220
.
Taylor
,
J. S. H.
,
Rastle
,
K.
, &
Davis
,
M. H.
(
2013
).
Can cognitive models explain brain activation during word and pseudoword reading? A meta-analysis of 36 neuroimaging studies
.
Psychological Bulletin
,
139
,
766
791
.
Tsapkini
,
K.
, &
Rapp
,
B.
(
2010
).
The orthography-specific functions of the left fusiform gyrus: Evidence of modality and category specificity
.
Cortex
,
46
,
185
205
.
Turkeltaub
,
P. E.
,
Eden
,
G. F.
,
Jones
,
K. M.
, &
Zeffiro
,
T. A.
(
2002
).
Meta-analysis of the functional neuroanatomy of single-word reading: Method and validation
.
Neuroimage
,
16
,
765
780
.
van Meer
,
M. P.
,
van der Marel
,
K.
,
Wang
,
K.
,
Otte
,
W. M.
,
El Bouazati
,
S.
,
Roeling
,
T. A.
, et al
(
2010
).
Recovery of sensorimotor function after experimental stroke correlates with restoration of resting-state interhemispheric functional connectivity
.
Journal of Neuroscience
,
30
,
3964
3972
.
Vogel
,
A. C.
,
Church
,
J. A.
,
Power
,
J. D.
,
Miezin
,
F. M.
,
Petersen
,
S. E.
, &
Schlaggar
,
B. L.
(
2013
).
Functional network architecture of reading-related regions across development
.
Brain and Language
,
125
,
231
243
.
Vogel
,
A. C.
,
Miezin
,
F. M.
,
Petersen
,
S. E.
, &
Schlaggar
,
B. L.
(
2012
).
The putative visual word form area is functionally connected to the dorsal attention network
.
Cerebral Cortex
,
22
,
537
549
.
Vogel
,
A. C.
,
Petersen
,
S. E.
, &
Schlaggar
,
B. L.
(
2012
).
The left occipitotemporal cortex does not show preferential activity for words
.
Cerebral Cortex
,
22
,
2715
2732
.
Vogel
,
A. C.
,
Petersen
,
S. E.
, &
Schlaggar
,
B. L.
(
2014
).
The VWFA: It's not just for words anymore
.
Frontiers in Human Neuroscience
,
8
,
88
.
Wager
,
T. D.
,
Jonides
,
J.
, &
Reading
,
S.
(
2004
).
Neuroimaging studies of shifting attention: A meta-analysis
.
Neuroimage
,
22
,
1679
1693
.
Xia
,
M.
,
Wang
,
J.
, &
He
,
Y.
(
2013
).
BrainNet Viewer: A network visualization tool for human brain connectomics
.
PLoS One
,
8
,
e68910
.
Yeo
,
B. T.
,
Krienen
,
F. M.
,
Sepulcre
,
J.
,
Sabuncu
,
M. R.
,
Lashkari
,
D.
,
Hollinshead
,
M.
, et al
(
2011
).
The organization of the human cerebral cortex estimated by intrinsic functional connectivity
.
Journal of Neurophysiology
,
106
,
1125
1165
.
Zhou
,
W.
,
Xia
,
Z.
,
Bi
,
Y.
, &
Shu
,
H.
(
2015
).
Altered connectivity of the dorsal and ventral visual regions in dyslexic children: A resting-state fMRI study
.
Frontiers in Human Neuroscience
,
9
,
495
.