Introduction
Among several explanations of developmental reading difficulties (RD), the presence of a core deficit in the ability to mentally represent and manipulate the sounds of the language (phonological awareness) appears to account for a significant amount of variance in word-level reading scores, especially during the earlier stages of reading acquisition. The ultimate goal of reading, however, is automatic recognition of words and comprehension of their meaning (Gough and Tumner, 1986; Share and Stanovich, 1995; Berninger et al., 2006), which involves a variety of other skills such as vocabulary and working memory (Kaplan et al., 1998; O'Shaughnessy and Swanson, 1998; Willcutt et al., 2001; Storch and Whitehurst, 2002; Ouellette and Sénéchal, 2008). Although in each individual with RD several alternative or complementary factors may account for such difficulties, the observation that RD is often found in children with intact or even superior intelligence and oral (non-phonological) language skills indicates that RD is associated with functional deficits specific to the brain circuitry that supports reading.
With the advent of functional neuroimaging methods, it has become possible to detect differences in the cerebral activation patterns associated with processes such as phonological decoding and word recognition between RD and non-impaired readers (NI). In particular, evidence from positron emission tomography (PET) and functional magnetic resonance imaging (fMRI) studies suggested that performance deficits in children with RD is paralleled by hypoactivation in several distinct regions critical for reading ability, including the superior (BA 22) and middle (BA 21) temporal gyri, the inferior parietal lobule [supramarginal and angular gyri (BA 40/39)] and the ventral portion of occipitotemporal cortex (mainly in the left hemisphere; BA 37; Temple et al., 2001; Shaywitz et al., 2002; Cao et al., 2006; Hoeft et al., 2007; for a recent review and meta-analysis see Maisog et al., 2008). An associated finding of either increased or decreased activation in prefrontal (mainly left inferior frontal, IFG) regions has also been reported in these studies.
In addition to PET and fMRI, magnetoencephalography, also known as magnetic source imaging (MSI), is an alternative neuroimaging technique that offers information both on spatial localization and on the timing (latency) of neurophysiological processes. The capacity of MSI to generate spatial maps of language-specific cortex has been validated through direct comparisons with invasive cortical mapping procedures in large-scale clinical populations (e.g., Papanicolaou et al., 2004; Kamada et al., 2007).
Over the course of several studies we have used MSI to differentiate between children with RD and NI readers during execution of both explicit phonological decoding (Simos et al., 2000b, 2002a, 2007), and continuous visual-word recognition tasks (silent real word reading; Simos et al., 2000a). In the latter study the magnitude of neurophysiological activity in left hemisphere posterior temporal (superior and middle temporal gyri) and inferior parietal regions (supramarginal gyrus) was significantly reduced among children with severe RD as compared to children with average to above average reading achievement. Children with RD also showed increased activity in homologous right hemisphere regions. Another reported feature of aberrant profiles in children with RD on activation tasks involving pseudoword reading involves the relative timing of regional activation, such that the early occipitotemporal (visual association cortex) activity was immediately followed by early engagement of prefrontal activity and only later by, activation of temporoparietal cortex.
Earlier MEG studies of RD suffered from a number of potential limitations. One potential limitation was that spatiotemporal activation profiles were obtained by iterative application of the single equivalent dipole model (Sarvas, 1987). While this method has been routinely applied to estimate source locations for sensory- and language-related magnetic fields, and has been subjected to external validation against invasive diagnostic techniques (Szymanski et al., 2001; Papanicolaou et al., 2004; Kamada et al., 2007), a single-dipole model may not fully represent the observed field distributions, a particular problem for modeling brain activation in complex cognitive tasks involving multiple brain areas. A means of overcoming this potential problem is the application of distributed source models and, in particular, the technique known as minimum-norm estimate (MNE; Hämäläinen and Ilmoniemi, 1994; Moran and Tepley, 2000) chosen in the present study. Furthermore, previous MEG studies (as well as some PET and fMRI studies) were limited by their reliance on participant samples that lacked adequate diversity in terms of socioeconomic status and IQ. Finally, another potential source of experimental confounds in early MEG studies of word recognition was that brain activation profiles were obtained with tasks that placed heavy demands on both word recognition and memory processes, despite evidence that task difficulty may significantly impact the degree of activation in key cortical areas (Cao et al., 2006; Pugh et al., 2008).
The objectives of the current investigation were to (a) replicate and extend previous MEG results on the key anatomical and temporal features of the brain activation profiles associated with silent word reading; and (b) assess the relative importance of time-dependent engagement of posterior temporal, inferior parietal, and frontal regions for word recognition in children who are either typical or struggling readers. In order to address some of the limitations of previous studies outlined above: (i) a larger, more representative sample of the general school population was selected; (ii) brain activation profiles were reconstructed in the context of much easier version of the continuous word recognition task used in our previous study (Simos et al., 2000a); and (iii) task-related neurophysiological activity was estimated by application of the MNE technique, mentioned above.
Discussion
In the present study, distributed source modeling of reading-related MEG data derived from a large, representative sample of typical and struggling readers partially replicated and extended previous reports (Simos et al., 2000a; Shaywitz et al., 2002; Booth et al., 2007; Pugh et al., 2008) regarding the nature of the functional disruption of the brain mechanism for word reading in RD. Sample selection ensured that the distribution of key demographic variables (age, gender, ethnic background), in addition to handedness and general cognitive ability (measured by PIQ), did not differ significantly between groups. Spatiotemporal brain activation profiles were obtained in the context of a continuous recognition task, the level of difficulty of which was titrated to reduce performance differences between groups compared to previous studies (between 70 and 80% correct identification of printed words from the training set). This was achieved by using shorter, higher frequency words and a smaller target set compared to our earlier study (Simos et al., 2000a). In this manner demands for both decoding and word recognition as well as memory encoding and retrieval were kept relatively low for both groups. The task was, nevertheless, successful in activating all key reading-related brain areas with clear left hemisphere dominance in their degree of activation as has been our experience with previous language studies (Simos et al., 2001; Papanicolaou et al., 2004).
The main findings of the present study can be summarized as follows. Firstly, children with RD showed reduced degree of neurophysiological activity in posterior temporal regions (MTG and STG) in both hemispheres. This activity was associated with relatively late stages of stimulus processing (200–650 ms after stimulus onset). Secondly, children with RD showed significant overactivation of the fusiform gyrus and middle frontal areas (between 150 and 450 ms), bilaterally, independent of age. Thirdly, individual differences in the degree and latency of activity in posterior temporal, ventral occipitotemporal (VOC), and mesial temporal regions mapped onto print-related achievement measures.
Although reading group differences in MTG and STG were found in both hemispheres, two additional lines of evidence suggest that disrupted functional brain organization for reading was restricted to the left hemisphere in RD children. Thus, while the degree of activity in left posterior temporal regions was a significant predictor of reading and spelling skill among NI readers, such an association was not noted among RD children. In the latter group, enhanced activity in right occipitotemporal and mesial temporal regions was associated with lower achievement scores. Importantly, reading-skill related differences in the degree of activity in these regions were evident across the age span covered by the present study (7–15 years). Moreover, for NI children, a clear temporal separation was noted between peaks of neurophysiological activity in left occipitotemporal (lateral and ventral visual association cortices), posterior temporal (middle and superior temporal gyri), and inferior/middle frontal regions. In contrast, the regular temporal progression of activity peaks between left hemisphere occipitotemporal, posterior temporal, and inferior/middle frontal regions was disrupted in the RD group.
Temporoparietal cortex and reading ability
There is ample evidence that neurophysiological processes that take place in the posterior portion of MTG (BA 21) reflect lexical/semantic processing (Damasio and Damasio, 1983; Pugh et al., 1996; Gaillard et al., 2001; Booth et al., 2002; Fiebach et al., 2002; Halgren et al., 2002; Simos et al., 2002b; Turkeltaub et al., 2002; McCandliss et al., 2003; Tyler et al., 2005; Wehner et al., 2007a). These operations may include storage or gaining access to lexical/semantic information for printed stimuli (Pugh et al., 1996; Fiebach et al., 2002). Damage to this region has been linked to deficits in a variety of tasks tapping into lexical/semantic processing, such as semantic priming (Tyler et al., 2005). Moreover, the onset latency of magnetic activity in this region significantly correlated with pronunciation latency of orthographically exceptional words while this was clearly not the case for regular words and pseudowords (Simos et al., 2002b).
With respect to cortex lying on the lateral and superior aspects of the STG (BA 22), extending posteriorly into the supramarginal gyrus, the bulk of empirical evidence implicates neurophysiological processes taking place in this region in phonological processing of spoken and written language (Beauvois and Derouesne, 1979; Caplan et al., 1995; Papanicolaou et al., 2003; Specht et al., 2003; Majerus et al., 2005). These processes include, but are not limited to, storage of sublexical phonological representations, automatic retrieval and short-term maintenance of these representations in consciousness (Hickok and Poeppel, 2000; Wise et al., 2001; Simos et al., 2002b; Mustovic et al., 2003). It should be noted, however, that both imaging and lesion data do not clearly indicate exclusive involvement of the left STG in phonological processing. There is also evidence that this region plays a crucial role for lexical/semantic processing of spoken language. Patients exhibiting the clinical profile of fluent aphasia, which typically results from damage to this region (Benson, 1985; Vignolo, 1988), almost invariably display notable deficits in a variety of experimental tasks tapping into lexical/semantic processing (Hagoort, 1993; Rodd et al., 2005; Janse, 2006). There are also reports that at least in adult experienced readers, the posterior portion of BA 22 is involved in lexical-semantic processing during silent reading tasks (Fiez et al., 1999; Cannestra et al., 2000; Haist et al., 2001; Fiebach et al., 2002; Okada and Hickok, 2006; Simos et al., 2009).
Evidence of increased activation in RD
In contrast to the bilateral reduction in MTG and STG activity, a significant increase in the strength of activation in middle frontal regions was noted for the RD group concurring with similar findings from PET and fMRI (Shaywitz et al., 1998; Brunswick et al., 1999; Grunling et al., 2004), as well as our more recent MEG investigations (Simos et al., 2007). It has been suggested that elevated levels of neurophysiological activity in prefrontal regions indicates engagement of cognitive operations supporting alternative strategies employed by individuals with RD to compensate for a deficiency in mapping the sound structure of printed words through covert articulation (Shaywitz et al., 1998; Shaywitz et al., 2004). However, the putative compensatory increase in prefrontal activity in RD is not a universal finding (e.g., Rumsey et al., 1997; Eden et al., 2004). In a recent meta-analysis of nine fMRI and PET studies Maisog et al. (2008) found no evidence to suggest a universal pattern of bilateral prefrontal hyperactivation in RD, but this meta-analysis involved only adults. In a meta-analysis with no age restrictions, Richlan et al. (2009) reported underactivation of the IFG gyrus and hyperactivity of the primary motor cortex and anterior insula. Several lines of evidence indicate that prefrontal hyperactivation may not reflect a key feature of the brain circuit for reading in RD, but is related to task and age-related processes (Hoeft et al., 2007).
Prefrontal hyperactivation in children with RD in the present study may also reflect additional demands for both encoding and retrieval operations posed by our continuous word recognition task. A comparable finding of increased memory-related prefrontal hemodynamic response among RD children has recently been reported (Vasic et al., 2008). There is ample evidence from lesion and functional imaging studies regarding the critical role played by prefrontal (Knowlton and Squire, 1995; Skinner and Fernandes, 2007; MacPherson et al., 2008) and mesial temporal cortices (Brown and Aggleton, 2001; Turriziani et al., 2004; Wixted and Squire, 2004) for recognition of previously encountered stimuli. The notion that task demands were related to increased prefrontal activity in RD is consistent with our finding that this effect was significantly larger among younger students, given that for this group memory/attention demands are expected to be higher than among older participants, rendering our continuous recognition task even more taxing among young RD students.
Increased strength of activation for children with RD was also found in VOC cortices (anatomically corresponding to the fusiform gyrus). This region is suitable for hosting neurophysiological processes responsible for graphemic processing and also possibly for the integration of orthographic, phonological and morphological information derived from print (McCandliss et al., 2003; Flowers et al., 2004; Pammer et al., 2004; Vigneau et al., 2005). Accumulating evidence suggests that VOC cortices are involved in storing or gaining access to familiar orthographic representations, which may take the shape of “word forms” (Kronbichler et al., 2004; Simos et al., 2009), and play an important role in both word recognition and spelling (Philipose et al., 2007; Mani et al., 2008). In agreement with previous reports, group differences in activity in this region were found during early processing stages at the peaks of the current density waveforms (between approximately 150 and 350 ms; e.g., Tarkiainen et al., 1999; Usui et al., 2009). One possibility is that fusiform activation among struggling readers may reflect increased effort to engage a visual/orthographic strategy for encoding and recognition of the printed word stimuli. This possibility is consistent with behavioral evidence in children and adults with reading disability (Lennox and Siegel, 1996; Milne et al., 2003; McNeil and Johnston, 2004, 2008; Best and Howard, 2005; Zoccolotti et al., 2005; Miller and Kupfermann, 2009). An alternative account of increased VOC activation relies on the putative role of this area in graphemic processing using linguistic constraints (Vigneau et al., 2005; Devlin et al., 2006) and in language functions in general (Schäffler et al., 1994).
Timing of regional activation
Another key finding of the represent study was the lack of temporal differentiation between activity peaks in VOC and the left MTG and also between left temporoparietal and IFG ROIs in the RD group. Similar results have been reported previously in MEG studies with younger struggling readers (Simos et al., 2005, 2007, in press). In contrast, non-impaired readers showed a systematic progression of activity peaks between VOC, middle temporal, temporoparietal, and IFG regions. In the absence of empirical data establishing direct links between the temporal progression of activity between any two cortical regions and anatomical connections between these regions, interpretations of these findings can only be speculative at present. The timing data, however, present an intriguing parallel with findings of reduced white matter integrity in poor readers (Deutsch et al., 2005; Niogi and McCandliss, 2006) and indirect evidence from hemodynamic studies for reduced functional “connectivity” between VOC and temporoparietal regions (Pugh et al., 2000; Cao et al., 2008) and between temporoparietal and frontal regions (Hampson et al., 2006).
Comparison with previous MEG studies of RD
A notable difference between the current findings and those of our earlier MEG study (Simos et al., 2000a) concerns the hemispheric profile of reading-group activation differences during the visual-word recognition task. Significant right hemisphere predominance of temporoparietal activity had been observed earlier for RD children, whereas in the present study both groups of children demonstrated L > R activation of posterior temporal cortices. Moreover, whereas RD children had previously been distinguished by a noticeable reduction in activation of the left temporoparietal cortex and accompanying hyperactivation of the right homotopic region, bilateral group effects were found here, namely reduced activation of the left and right STG and MTG, relative to NI children. Although, bilateral hemodynamic hypoactivation in posterior temporal and inferior parietal areas has sometimes been reported (Rumsey et al., 1997; McCrory et al., 2000; Meyler et al., 2007), these discrepancies merit some consideration.
One possible explanation may concern differences in sample selection: The present investigation employed a more representative sample of children selected from schools as opposed to community recruitment, and a slightly more lenient reading achievement criterion to form the study groups. This account is not supported by additional exploratory analyses comparing two age- and PIQ-comparable groups of children derived from the current data set, one with severe RD (N = 20, with word attack, letter–word identification, and spelling scores below 75 points) and the second with high-average reading achievement scores (over 95 points on each of the three subtests; N = 22). The results were essentially identical to those from the entire sample without notable increases in the effect size of interactions involving hemisphere.
Another potential explanation may relate to differences in task demands. The task used in the present investigation was significantly easier to perform than the version used in Simos et al. (2000a) as evidenced by comparable performance between reading achievement groups. Evidence from neuroimaging studies in individuals with RD have shown the degree of brain of activity to be affected by reading task difficulty (Cao et al., 2006), including lateralization of STG and MTG activation (Pugh et al., 2008). Word reading tasks, especially if they do not pose specific demands for phonological processing are likely to be more susceptible to the effects of overall task difficulty, and result in reduced hemispheric asymmetries in perisylvian activation (e.g., McCrory et al., 2000). A similar effect appears to hold for auditory processing as well (e.g., Breier et al., 2003; Wehner et al., 2007b).
Furthermore, the use of a higher-density gradiometer array in conjunction with distributed source modeling may account for the additional group-related findings in prefrontal and mesial temporal cortices. By affording increased sensitivity to smaller sources (mainly in terms of the spatial extent of their respective magnetic surface maps), the current MEG system was more likely to reveal multiple sources in each hemisphere which were invisible or difficult to model using the single-dipole method. Similarly, the potential for detecting smaller, more extended sources of activation is likely to have been facilitated by the use of distributed source modeling.