Reading of a narrative text (henceforth ‘text reading’) plays a key role in our daily lives. Text reading requires a dynamic integration of vision, visual attention, and linguistic processes of the materials. When reading words, the ventral visual stream is activated which involves the visual word form area (VWFA) as a central hub (Cohen et al., 2000, 2002; Cohen and Dehaene, 2004). According to the dual-route theory of visual processing (e.g., Goodale and Milner, 1992), the division of labor between a dorsal “where” stream and a ventral “what” stream is one of the most fundamental principles of information processing in the brain (Ungerleider and Haxby, 1994). Similarly, the processes involved in text reading may also follow this dual-route principle. While previous functional magnetic resonance imaging (fMRI) studies have generally implicated a key role of the ventral visual system for word reading (for a review, see Dehaene and Cohen, 2011), recent fMRI studies in naturalistic text reading have proposed that the dorsal visual regions [e.g., intraparietal sulcus (IPS)] play a role in visual/spatial attention or eye movements during reading of an entire sentence (Hillen et al., 2013; Choi et al., 2014). However, little is known about how the dorsal visual system interacts with the ventral visual region during text reading. The present study was specifically aimed at addressing this question.

Brain connectivity analyses, including both functional connectivity (FC) and effective connectivity (EC), have become important for understanding the inter-regional associations in both task-driven fMRI and resting-state fMRI (rs-fMRI; see Friston, 2011 for a review). Such analyses in reading research have revealed the dynamics of cooperation and causal influence among several key language regions including the left perisylvian cortex, as revealed by task-driven fMRI studies of word reading (e.g., Booth et al., 2008; Richardson et al., 2011; Duncan et al., 2013; Kwok et al., 2015) and rs-fMRI studies (e.g., Hampson et al., 2002, 2006; Koyama et al., 2010; Li et al., 2013; Schurz et al., 2014a). Interestingly, recent rs-fMRI studies have consistently shown that the VWFA was functionally connected to dorsal attention regions (Vogel et al., 2011; Wang X.S. et al., 2015). In a rs-fMRI study with children, Zhou et al. (2015) have investigated the FC seed maps of two central nodes for the dorsal and ventral visual streams respectively (i.e., VWFA and IPS), and found that FCs from the VWFA and the left IPS were both connected to the left middle frontal gyrus (MFG). The left MFG was proposed as a convergent region for the dorsal and ventral visual systems during text reading. They also demonstrated that the FC strength between IPS-MFG was positively correlated with the scores of reading fluency but not the scores of lexical decision.

While task-driven fMRI provides an opportunity to study the evoked neural mechanism of cognition (e.g., reading), it is likely influenced by task-induced factors (Friston, 2005; Koyama et al., 2010). In contrast, the rs-fMRI, which measures spontaneously activation during rest, can reveal the brain’s intrinsic functional organization that relates to cognition. Therefore, the combination of task-driven fMRI and rs-fMRI provides a good approach to investigate the fundamental and reliable neural mechanism of cognition (Kannurpatti et al., 2012; Rosazza et al., 2014). However, this approach has been rarely used in the field of reading. It is unknown based on our previous studies whether the above connectivity findings in children could be applied to adults, and whether the findings in rs-fMRI could be generalized to reading-driven fMRI. In order to address these questions, we conducted the FC analyses in a reading-driven fMRI and a rs-fMRI experiments with adults. Although, a number of regions in the perisylvian cortex may also be involved in text reading (e.g., see Price, 2012 for a comprehensive review), the present study mainly focused on the FCs for the dorsal and ventral visual streams respectively (see also Zhou et al., 2015). It has been suggested that there is a common mechanism between task-driven fMRI and rs-fMRI (Mennes et al., 2010), but weak correspondence may exist for regions whose patterns of evoked functional interactions are more adaptive and context-dependent (Mennes et al., 2013; see also Sporns, 2011). Thus, it is important to carry out similar FC analyses in both task-driven fMRI and rs-fMRI, so that we can identify whether the brain connections among the dorsal and ventral visual systems are stable or context-dependent, in text reading and other similar processes.

A further important aim of the current study was to ask how the regions in the dorsal and ventral visual systems causally affect each other in text reading. The directional influence (i.e., top–down versus bottom–up) between the dorsal and ventral visual streams is a hotly debated issue in the discussion of the dual-route mechanisms for visual attention tasks. While the automatic “bottom–up” capture of attention is driven by stimulus properties derived from the ventral visual system, the volitional “top–down” modulation is due to selective visual attention and knowledge about the current task situated in the dorsal visual system (Kastner and Ungerleider, 2000; Buschman and Miller, 2007). As a specific visual-related task, text reading should also require such a dynamically interactive system: when word presentation in the ventral visual system initiates a bottom–up effect (Schurz et al., 2014a), high level factors situated in the dorsal system (e.g., visual attention or situation construction; Yarkoni et al., 2008; Vidyasagar and Pammer, 2010) are likely to exert a top–down control. Compared to the tasks using isolated word presentation paradigms (e.g., lexical decision, semantic categorization), the text reading tasks (e.g., reading fluency) are more likely to be dominated by the top–down control process (Radach et al., 2008). In order to test the existence of bidirectional influences (i.e., top–down versus bottom–up) between the dorsal and ventral visual systems in text reading, we employed Granger causality analysis (GCA) for EC analyses, which has been frequently used in previous task-driven fMRI and rs-fMRI studies (Roebroeck et al., 2005; Seth et al., 2015). While non-directional FC reveals the synchronization between brain regions, directional EC reflects the influence from one brain region to another. The use of both kinds of analysis not only verifies specific coherent systems in the brain but also uncovers the dynamic nature of these systems.

Although, the extant literature has pointed to interactions between the dorsal and ventral visual streams in text reading, the detailed picture of such interaction has not emerged. In this study, we hypothesized that (a) there are significant FCs and ECs among regions in the dorsal and ventral visual systems in reading-driven fMRI, (b) the brain connections in rs-fMRI are generally similar to those in reading-driven fMRI, as the resting-state connectivity is assumed to reflect a history of consistent and repeated co-activations of areas (Dosenbach et al., 2007; Fair et al., 2007), and (c) the dorsal visual connection is more correlated with text reading relative to single word processing. To test these hypotheses, we carried out two fMRI experiments for FC and EC analyses: on the basis of existing data from Wang X.J. et al. (2015), we selected regions of interest (ROIs) as two central nodes (i.e., IPS and VWFA) in the dorsal and ventral visual systems and one convergence node (i.e., MFG) for these two systems in the frontal cortex. Wang X.J. et al. (2015) conducted analyses of whole-brain and ROI-based activations in naturalistic text reading of both English and Chinese. In the current analyses in our Experiment 1, we examined only the dataset from Chinese text reading to identify connectivity patterns so as to compare with rs-fMRI data from Experiment 2. In Experiment 2, the same ROIs as in Experiment 1 were used for FC and EC analyses in a set of data from the participants’ rs-fMRI. Finally, in order to distinguish the functions of the dorsal and ventral visual connections and to identify brain-behavior relations, we calculated correlations between the strength of brain connections in rs-fMRI and the behavioral reading scores that were based on fluent reading against lexical decision tasks. As fluent reading engages more cognitive resources such as visual attention allocation (Rayner, 2009) relative to single word processing, we expected different brain patterns for these two tasks, especially in dorsal attention regions.

In Experiment 2, we conducted the FC and EC analyses with a new rs-fMRI dataset to identify similarities and differences in reading-driven fMRI versus rs-fMRI connectivity patterns. In addition, in order to understand how network connectivity patterns correlate with behaviors, we selected a reading fluency task and a lexical decision task because both tasks include visual aspects of reading, but are different in the degree of engagement in volitional attention control (Rayner, 2009).

Narrative text reading in our daily lives requires readers to integrate multiple words in an entire sentence, but traditional neuroimaging studies in reading have tended to present the stimuli as one word at a time in a rapid serial visual presentation paradigm (RSVP, see Price, 2012 for a review). Due to such methodological limitations, previous research has focused on the role of the ventral visual stream in text reading. More recently, researchers have started to work on paradigms that can present entire sentences or passages in fMRI experiments (e.g., Hillen et al., 2013; Regev et al., 2013; Choi et al., 2014; Schuster et al., 2015; Wang X.J. et al., 2015), and as a result the role of the dorsal visual regions in reading (Corbetta and Shulman, 2002; McDowell et al., 2008; Herweg et al., 2014) has now been examined. Although, the studies reviewed above have identified the distributed activations of both the dorsal and ventral visual regions in text reading, knowledge about the interactions between these two systems is lacking. To fill in this knowledge gap, the present study employed a dual-route approach for FC and EC analyses in task-driven fMRI and rs-fMRI. Although, there has been a “dual-route model of reading” concerned with a dorsal phonological route and a ventral orthographic route among left perisylvian regions (such as VWFA, IFG, inferior parietal lobe; see Coltheart et al., 2001; Jobard et al., 2003 for reviews), that model is focused on single-word reading rather than text reading and therefore it does not address research questions raised in the current study. Instead, our current study highlights a distinctive network that includes the VWFA and the dorsal visual attention regions (i.e., MFG and IPS). The findings of our reading-driven fMRI and rs-fMRI analyses converge to reveal the interactions between dorsal and ventral visual systems for text reading. As we expected, there were significant FCs and ECs among the left MFG, left IPS, and VWFA in both reading-driven fMRI and rs-fMRI.

In particular, the current study has highlighted directional influences among the dorsal and ventral visual regions by using EC analyses. The results have indicated a clear top–down modulation from the left MFG to the left IPS and from the left IPS to the VWFA in reading-driven fMRI and rs-fMRI. Most importantly, the current reading-driven fMRI has revealed dual top–down effects from the left MFG to both the dorsal region (i.e., left IPS) and the ventral region (i.e., left VWFA). These results were in agreement with the top–down effects from frontal-parietal regions to ventral visual regions or from frontal regions to parietal regions in vision research using GCA methods (e.g., Pessoa et al., 2003; Bressler et al., 2008; Hwang et al., 2010; Sneve et al., 2013). As previous studies have indicated, the MFG is an important region for executive functions in dual-task performance (Rossi et al., 2009) and in switching attentional control on the basis of changing task demands (Szameitat et al., 2002). When it comes to text reading, the modulation from the MFG to the dorsal and ventral connections may be corresponding to the coordination between sub-processes of text reading. In contrast, there was no reliable bottom–up effect except for a negative EC from the left IPS to the left MFG at rest. We believe that text reading is a more volitional task relative to single character reading, so it is more likely to be dominated by a top–down control.

The results of FC analyses are largely in line with those of EC analyses, which supported the existence of the cooperation among regions in the dorsal-ventral visual network for text reading. Based on Zhou et al. (2015), the present study has used a dual-route approach to investigate the dorsal-ventral FCs in reading-driven fMRI and rs-fMRI. The current results of FC analyses in rs-fMRI for adults are generally similar to our previous rs-fMRI study for children (i.e., Zhou et al., 2015): the FC strength of MFG-IPS was positively correlated with the scores of reading fluency but not lexical decision, suggesting the unique contribution of the dorsal visual system to text reading abilities. Additionally, the FC analyses in reading-driven fMRI have further implicated the importance of this dorsal-ventral visual network for on-line text reading. Interestingly, we observed a negative correlation between the FC strength of VWFA-IPS and the score of lexical decision. This result may indicate that word processing in the VWFA needs to be supplemented by the deployment of attention in the IPS, especially for individuals with poor ability of lexical processing.

In general, our analyses suggest that the brain connectivity patterns in reading-driven fMRI and rs-fMRI are consistent with each other. Previous research has revealed that there are common mechanisms for reading-driven fMRI and rs-fMRI (Mennes et al., 2010). In reading research, Schurz et al. (2014b) explored the FCs in both reading-driven fMRI (tasks: visual word reading and phonological lexical decision) and rs-fMRI in developmental dyslexia, and found that there were consistent patterns in these two modalities. The present study has found that the FCs and the ECs for MFG-IPS and IPS-VWFA coupling were quite reliable in reading-driven fMRI and in rs-fMRI. As the resting-state connectivity is assumed to reflect a history of consistent and repeated co-activations of areas (Dosenbach et al., 2007; Fair et al., 2007), the identified stable connections are possibly due to more fundamental cognitive function such as visual search, a foundational process for text reading. However, we have observed the disconnected FC and EC between the MFG and the VWFA in resting states. Mennes et al. (2013) have proposed that some areas may show weak correspondence between the brain’s intrinsic and extrinsic functional architectures, and these areas are typically those that display more adaptive and context-dependent functional interactions. VWFA may be one such region, as previous studies have also indicated that the activation of VWFA is likely influenced by context-dependent higher-level processing (Dehaene and Cohen, 2011; Price, 2012). Thus, the magnitude of attentional modulation of VWFA activity in text reading may explain the inconsistent results between resting-state versus reading-driven fMRI patterns in the current study. In other words, during attention-demanding tasks of text reading, the VWFA-MFG ventral pathway is more likely to be activated in order to convert visual input to high-level word-to-text integration, whereas in a resting-state, this ventral pathway is less likely to become active. Notably, as we used datasets from different participants in Experiments 1 and 2, the consistency in brain connection patterns between reading-driven fMRI and rs-fMRI was consolidated whereas their difference should be interpreted with caution.

While the FC reveals non-directional relationships between brain regions, the EC indicates a directional influences from one region to another. However, as there has been little research using both FC and EC in the same analyses, the specific relationship between FC and EC effects is somewhat difficult to explain. At the very least, these two effects are not independent of each other. For example, the correlation (i.e., FC) between two areas can be caused by a common input (i.e., EC) of another area from past fluctuations (Friston, 2011). Consistently, we have observed concurrent and similar pattern of FC and EC networks in the present study, suggesting a complex relationship between FC and EC effects. In addition, while we have successfully discovered the brain–behavior correlations for FC at rest (i.e., correlation between MFG-IPS connectivity and reading fluency), there is no evidence for the relationship between EC and the reading performance scores. We suspect that EC is more dynamic (activity-dependent) relative to FC, so the brain–behavior relationship is relatively difficult to detect at rest. Such discrepancies between FC and EC patterns and the corresponding effects remain to be investigated in future research.

Although, we have identified the important role of the dorsal-ventral visual network in text reading, the nature of this network needs further clarification. For instance, is this network also related to general comprehension? Currently, we believe that the found connections are more associated with visual/attention aspects of reading relative to general comprehension, as typical regions for general language comprehension (e.g., the left IFG and MTG; Constable et al., 2004; Buchweitz et al., 2009) were not included in the dorsal-ventral visual system. However, these connections may be potentially modulated by comprehension processes. Future research is needed to directly compare the dorsal-ventral visual system in self-pace reading with comprehension processes (e.g., text reading) and without comprehension processes (e.g., word list reading). Another noticeable point is that the results of present study are based on text reading in the Chinese writing system. The Chinese text is written in a series of characters of roughly uniform height and width. Unlike alphabetic writing system, Chinese orthographic system is well-known for the complex visual form of characters and the lack of inter-word spaces for word segmentation in text. These properties were tightly associated with the function of dorsal visual regions such as MFG and IPS. However, Wang X.J. et al. (2015) have found that the difference in activation between Chinese and English story reading mainly situated in ventral visual regions. An interesting question for further research is whether the connectivity patterns of the dorsal-ventral visual system in Chinese can be generalized to alphabetic writing systems.

Readers recruit both the dorsal and ventral visual pathways when reading a narrative text. Previous fMRI studies on text reading has identified a distributed set of frontal, parietal, and occipital-temporal regions within this dual-route system. The current study examined the interactions among these regions in text reading using FC and EC analyses in task-driven fMRI and rs-fMRI. Our analyses suggest a dynamic picture in which the dorsal and ventral systems involving MFG, IPS, and VWFA cooperate with each other and causally influence one another during natural text reading.

Study conception and design: HS, WZ; acquisition of data: XW, ZX; analysis and interpretation of data: WZ; drafting of manuscript: WZ, HS; critical revision: HS, PL, YB.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.