Neuroimaging of Reading Intervention: A Systematic Review and Activation Likelihood Estimate Meta-Analysis

A growing number of studies examine instructional training and brain activity. The purpose of this paper is to review the literature regarding neuroimaging of reading intervention, with a particular focus on reading difficulties (RD). To locate relevant studies, searches of peer-reviewed literature were conducted using electronic databases to search for studies from the imaging modalities of fMRI and MEG (including MSI) that explored reading intervention. Of the 96 identified studies, 22 met the inclusion criteria for descriptive analysis. A subset of these (8 fMRI experiments with post-intervention data) was subjected to activation likelihood estimate (ALE) meta-analysis to investigate differences in functional activation following reading intervention. Findings from the literature review suggest differences in functional activation of numerous brain regions associated with reading intervention, including bilateral inferior frontal, superior temporal, middle temporal, middle frontal, superior frontal, and postcentral gyri, as well as bilateral occipital cortex, inferior parietal lobules, thalami, and insulae. Findings from the meta-analysis indicate change in functional activation following reading intervention in the left thalamus, right insula/inferior frontal, left inferior frontal, right posterior cingulate, and left middle occipital gyri. Though these findings should be interpreted with caution due to the small number of studies and the disparate methodologies used, this paper is an effort to synthesize across studies and to guide future exploration of neuroimaging and reading intervention.


Introduction
Young children who are at-risk for reading difficulties (RD) lag behind their peers in academic achievement and this achievement gap may grow over time [1,2], resulting in a lifelong condition that negatively impacts academic achievement, employment opportunities, and social interaction. Consequently, remediating reading difficulties is a paramount goal of education, requiring research that disentangles the causes and symptoms of reading difficulties and develops effective treatment. RD can involve difficulties with reading at the word level or language processing level. While some children have specific comprehension difficulties, children with word level difficulties (decoding and/or word recognition) may have comprehension difficulties as well. In the general population, estimates of reading difficulty in children at the word level (dyslexia) range from 6 to 17% [3] and these children typically have a phonological processing deficit (e.g., [4]). Children with word-level difficulties tend to be identified during the early elementary years [5,6], though some do not experience difficulties until late elementary or middle school years [7][8][9][10].

Behavioral Responsiveness to Intervention
Behavioral interventions have been found effective in remediating reading difficulties in some, but not all, individuals with RD [4,[11][12][13][14][15][16]. Interventions that are effective for most children involve explicit instruction and address the reading components of phonological awareness, phonics, fluency, vocabulary, and comprehension [17][18][19][20]. Determining which children will and will not readily respond to intervention could inform resource allocation such that children who are not likely to respond well could receive more intensive or individually targeted instruction sooner. A review by Al Otaiba and Fuchs [12] determined that the majority of children who exhibit low responsiveness to intervention have a phonological awareness deficit, and other deficiencies may include problems with phonological retrieval or encoding, verbal ability, behavior, and developmental delays. These findings are supported by meta-analyses showing that individual responsiveness to intervention can be influenced by rapid automatic naming (RAN) skills, problem behaviors, phonological awareness, alphabetic principle, memory, and IQ [21], and that preintervention differences in real word identification, word attack, and reading comprehension are predictive of gains following intervention [22]. These studies establish the behavioral characteristics of deficient responsiveness quite well, but perhaps these are overt symptoms of an underlying deficit. Biological factors (genes and the brain) influence cognitive factors which in turn influence behavior, all three of which are impacted by an individual's environment [23]. Understanding the behavioral profile of differential intervention response is necessary and useful, but to further characterize RD and potential underlying causes, the neurobiology of reading, RD, and responsiveness to intervention should be explored.

Neurobiology of RD and Learning
Though extensive literature has characterized the behavioral aspects associated with RD, the underlying cognitive causes of RD are not fully elucidated and are currently under investigation. RD in the general population appears to be part of a normally distributed continuum of reading ability [24,25]. Growing evidence indicates that the behavioral symptoms of RD are associated with anomalous underlying neurobiology which may be distributed along a continuum as well or perhaps diverges from that of typical readers in different ways according to the type of reading deficit as some structural evidence suggests [26]. This underlying neurobiology is likely influenced by multiple factorsgenetics [27], training/instruction [28,29], nutrition [30], epigenetics [31,32] and possibly other factors [33]. The interplay of these factors creates variability in RD and adds complexity to diagnosis and treatment. Understanding the learning process at the neurobiological level and how this process may differ for individuals with RD may improve outcomes in the future for those with RD.
Neuroplasticity. The capacity to develop neural pathways and adapt to cognitive demands is the essence of neural plasticity. Changes in electrical impulses, chemical signaling, and histology underlie the behavioral aspects of ''learning.'' Describing the candidate mechanisms of neuroplasticity in detail is beyond the scope of this paper but to illustrate the complexity of processes involved in neuroplasticity we describe some of what is known and refer the reader to several excellent review papers. At the synaptic level, long-term potentiation (LTP) of the postsynaptic neuron in response to repeated presynaptic stimuli is widely believed to be a primary mechanism of long-term memory and learning [34]. LTP can lead to synaptic modifications that result in enhanced signal transmission [35]. Though short-term synaptic modification does not require protein synthesis, long-term synaptic changes seem to require protein synthesis and these long-term changes are considered to be the cellular correlates of learning and memory [36]. Beyond enhancement of synaptic connections through LTP, gray matter structural mechanisms of neuroplasticity include neurogenesis, axon sprouting, dendritic branching (and synaptogenesis), gliogenesis and glial modifications, and angiogenesis [37]. As numerous gene products are believed to be involved in learning and memory and protein synthesis appears to be required for longterm memory storage, regulation of gene expression is likely to be important in learning processes. Epigenetic mechanisms include histone modification [31] and post-transcriptional regulation of gene expression by microRNAs [38]. Clearly, learning and memory are complex events at the cellular and molecular levels. Yet, what does this mean for children with learning disabilities? Perhaps these processes differ for individuals with RD compared to typically achieving peers. An inefficiency in one mechanism of the intricate processes involved in storing and retrieving information could adversely affect learning. Such neurobiological differences could explain why some children absorb new knowledge effortlessly while others not only struggle to grasp new concepts, but may also have great difficulty in retaining and consolidating information.
If a characterizable neurobiological anomaly underlies poor response to instruction, identifying these individuals and developing an appropriate intervention could possibly compensate for the deficiency. To illustrate, a critical molecular process related to long-term memory is the effect that CREB (cAMP response element binding protein) has on genetic transcription. CREB is stimulus-inducible in neurons and involved in long-term memory by influencing protein synthesis [39]. In a mouse study, long term memory was adversely affected by CREB gene disruption, yet interestingly the learning deficit was overcome by increasing the time between the training events [40]. That is, modifying the behavioral training compensated for a molecular deficiency. Perhaps identifying underlying dysfunction in functional activity can lead to a specific intervention. A challenge lies in understanding these differences and addressing them perhaps in a very prescriptive manner. Whatever the deficiencies may be, characterizing those deficiencies through investigative techniques is a step in the process toward targeted remediation.
Functional neuroimaging. Neuroimaging is a tool that can be used to explore brain activity and can make a valuable contribution to understanding neuroplasticity. Brain activity, as determined by imaging modalities such as fMRI and MEG, not only indirectly reflects the underlying tissue structure and physiology (as neurons must be present and functioning to exhibit activity), but also represents which areas of the brain are actively engaged when presented with stimuli of interest. Numerous studies have used imaging techniques to explore brain activity associated with specific cognitive tasks. A growing number of studies are exploring various aspects of reading to determine how the brain accomplishes the complex task of reading. Further investigations are concerned with how the brain differs in functional activity between skilled readers and readers who struggle.
Neuroimaging in typically achieving readers. Studies with unimpaired adult readers have revealed a left hemisphere reading network comprised of three areas: a ventral posterior region, a dorsal posterior region, and an anterior region [41][42][43][44]. The posterior ventral region is located in an inferior occipitotemporal area. This area appears to be involved in visual processing and recognition of words [41,[45][46][47][48] as even letters and pseudoletters elicit a response in this region [49]. The posterior dorsal region is comprised of the posterior superior temporal gyrus (pSTG), supramarginal gyrus (SMG), and angular gyrus (AG) and appears to be involved in phonological processing, transforming orthographic representations to phonological representations, and semantic processing [41,47,50,51]. The anterior region is located in and around the inferior frontal gyrus (IFG). The IFG appears to be involved in phonological processing (e.g., [49]) and may be involved in articulatory recoding such that phonological input is converted to speech-gesture articulation output [41,42] and semantic processing [47]. The studies establishing these regions have largely focused on single word reading. However, studies exploring reading comprehension have found left SMG and AG [52] and bilateral middle temporal gyri (MTG) and STG [53] to be activated by typical readers during comprehension tasks. While imaging studies of reading in children indicate a large amount of overlap with adults in activation, there are some differences. Some evidence indicates that over the course of development, activation in some more dispersed areas attenuates with age while increases with age are more focal, and these changes are largely independent of performance [54,55]. For example, adults show less activation of left SMG and AG, areas involved in phonological processing [56]. This suggests that children are more actively engaged in using phonology to analyze words as they read, whereas adults have developed such automaticity in word reading that increased use of these phonological processing areas is no longer required. In contrast, adults showed increased activity in frontal and parietal regions thought to be involved in attention and top-down cognitive control [54,55]. It is important to consider developmental changes in functional activity when reflecting on whether reading difficulties are more related to a delay in development or to dysfunctional reading networks that encourage development of compensatory mechanisms. Another recent finding in functional developmental changes is that sensitivity to visually presented words (i.e. activation response to detecting a word among progressively decreasing visual noise) increases over the school-age years in the L posterior occipito-temporal sulcus [57]. This may imply a more refined usage of an area used in visual word reading.
Neuroimaging of reading disabilities. Over the past two decades, numerous studies have reported differences in brain function during reading tasks for people with reading problems relative to controls with typical reading achievement and several narrative reviews have highlighted the commonalities among studies [41,43,[58][59][60][61]. The functional differences between RD and typical readers are generally characterized by reduced activity in L hemisphere regions for RD. Reviews of the literature have indicated that RD involves a dysfunction of the aforementioned three-region reading network: general underactivation of L temporo-parietal region (including STG) and ventral occipitotemporal (including lateral extrastriate, fusiform, and inferior temporal gyrus) and overactivation of the L IFG. This overactivation of the L IFG has been presumed to be due to compensatory articulatory effort and evidence that contradicts this portion of the accepted model has recently emerged [62]. Though less consistent than reduced left hemisphere activation, some studies have also reported increased right hemisphere activation that may signify compensatory activity for people with RD during reading tasks [51,[63][64][65][66] and this compensatory activity may develop as early as second grade [67]. Adding further complexity, there is evidence that, at times, typical readers exhibit less brain activation in reading areas than do readers with RD. Rimrodt et al [64] found that adolescents with RD activated more than typical readers in the left middle and superior temporal gyri when reading incongruent sentences, perhaps suggesting more effortful processing of nonmeaningful sentences. Pugh et al [68] found through manipulating stimuli that for non-impaired adolescent readers, factors that make the word easier to process were associated with relatively reduced activation. However, for readers with RD facilitative factors were associated with increased activation in the same areas, ''suggesting that the LH reading circuitry in adolescent RD is poorly trained but not wholly disrupted'' [68]. Less activation may at times be reflective of knowledge consolidation. If cognitive processing is less effortful, then the resulting efficiency may mean less functional activity. Though evidence is limited, some studies have shown that typically achieving novice performers can exhibit increased activation, yet following training or practice, decreased activation is observed [28,69,70].
While the value of the narrative reviews cannot be discounted, meta-analysis provides a statistical approach to synthesizing across studies. In a meta-analysis of adults with reading disabilities compared to controls, reduced activation was reported for L hemisphere ventral occipitotemporal cortex, inferior parietal cortex, STG, IFG, and thalamus [71]. Another recent metaanalysis, which included both children and adults with reading disabilities, identified underactivation in the L hemisphere inferior parietal, superior temporal, middle and inferior temporal, and fusiform regions and also reported underactivation in the L IFG that coincided with overactivation in the primary motor cortex and anterior insula [72]. In a meta-analysis that compared adults with RD and children with RD, showed similar results except that temporoparietal underactivation was seen only for adults with RD, not for children with RD [73].These meta-analyses are consistent with the literature reviews in identifying a dysfunctionally underactivating left hemisphere network in RD. However, the metaanalyses indicate the presence of underactivation of the L IFG in RD rather than the overactivation assumed in the narrative reviews [62]. The meta-analysis would appear to constitute a consensus across studies, yet appropriate caution should be used when interpreting meta-analyses. Due to the necessary requirement of coordinates of activation to perform functional metaanalysis, not all relevant studies may be included in the analysis. Hence, both the meta-analytic approach and the narrative literature review provide insight into understanding differences between typical readers and those with RD.

Neuroimaging of Reading Intervention Literature Review and Meta-Analysis
Arising from the growing evidence of RD brain activity differences are questions surrounding how reading instruction can impact individuals with RD at the neurobiological level. In addition, questions arise regarding the relation of neurobiological differences and responsiveness to intervention. To answer these questions, we conducted a systematic review of the functional imaging literature associated with reading intervention. We sought studies that explored functional activity differences before, during, and after intervention including studies that examined responsiveness to intervention and associated functional imaging. We sought to include recent studies published since earlier summaries and reviews [74][75][76][77]. In this paper we limit detailed examination to fMRI and MEG, including magnetic source imaging (MSI) which uses MEG in conjunction with MRI. However, it should be noted that in addition to fMRI and MEG, reading intervention has been explored with other methods and modalities. Structural imaging studies have included diffusion tensor imaging (DTI) [78][79][80] and voxel based morphometry (VBM) [81]. Functional activity studies have used event related potential (ERP) [82,83], magnetic resonance spectroscopy (MRS) [84,85] and transcranial magnetic stimulation (TMS) treatment [86,87].
In an effort to quantify changes associated with treatment, we conducted ALE meta-analysis on a subset of experiments contained in the literature review. Meta-analyses have the potential to identify results that are consistent across studies in a more objective and quantitative manner than can descriptive reviews. The meta-analytic process may reveal commonalities across studies that did not command attention in the original studies because they seemed theoretically uninteresting [62]. Since its development [88], ALE has become available as GingerALE on BrainMap.org [89] which has greatly facilitated the use of the meta-analytic approach for functional imaging experiments. Modifications have been made [90][91][92] such that GingerALE version 2.1.1 supports random effects meta-analysis (rather than fixed effects). Also, values generated by ALE are weighted by sample size of each contributing study. According to the User Manual for GingerALE 2.1, the GingerALE process can be summarized in three steps [93]. First, an ALE value is computed for each voxel in the brain and the null distribution of each voxel's ALE statistic is determined. The full width at half maximum (FWHM) value is automatically included as it has been empirically determined [91]. Second, a false discovery rate (FDR) algorithm uses the p values from the previous calculation to compute the ALE map threshold. Third, a cluster analysis of the thresholded map is performed based on the selected minimum cluster size.
Meta-analyses of neuroimaging of reading disabilities have been published in recent years [71][72][73]. As ALE meta-analysis is necessarily restricted to including only experiments that provide coordinates in normalized space (Talairach or MNI), currently few studies are available for inclusion. This analysis explores these limited studies as a step toward synthesizing across studies in an objective and quantitative manner.
This paper is a review of the literature associated with neuroimaging of reading intervention in children and adults with RD. First, we review the literature regarding brain-based studies of reading intervention. This descriptive review examines studies conducting imaging before and/or after intervention. Next, we present a coordinate-based ALE meta-analysis using a subset of the studies in the descriptive review. This subset consists of the fMRI studies that provided post-intervention scanning data in sufficient detail required for ALE meta-analysis.

Retrieval of Studies
Inclusion criteria. To obtain studies that examined brain activity associated with reading intervention, six inclusion criteria were stipulated. First, only peer-reviewed, primary research studies were included. Second, only studies with at least some participants designated as having reading difficulties, reading disabilities, dyslexia, or at-risk status for reading difficulties were included. For studies that used imaging to predict future reading scores, the designation of reading difficulties could be determined at posttest. Case studies were excluded. Third, the reading difficulty must have been idiopathic in nature and not the result of head trauma, stroke, or illness. Fourth, the studies were required to describe reading-related instruction that occurred during the experiment. Though the intended focus was upon intervention, studies in which participants received regular (business-as-usual) instruction rather than implementing an intervention were not excluded. Fifth, the studies were required to include neuroimaging in the modalities of either fMRI or MEG and the experimental design must associate the imaging with the reading instruction. Sixth, the functional imaging task must have been a reading task or a task of reading-related skill (e.g., letter sound matched to visual letter).
Searches. Two search strategies were employed to identify relevant studies and these searches were current as of January 2013. First, searches were conducted using two electronic databases, PubMed and Web of Science. The Web of Science search input was (TS = ((reading disability OR dyslexia OR reading difficulty) AND (neuroimaging OR fMRI OR brain activation) AND (reading intervention OR reading instruction OR reading treatment))) AND Document Types = (Article). The articles were examined and those that did not meet criteria were systematically eliminated (Fig 1) [94]. Next, using references from studies that met inclusion criteria, additional studies were considered. Additionally, Google Scholar searches were performed to locate papers that have cited some of the studies that met inclusion criteria.

Data collection
Data were gathered from articles, supplemental material, and from other works referenced in the paper as needed (e.g.,detailed descriptions of participants or interventions published separately from imaging analysis). Information describing participant groups, neuroimaging techniques, and whether or not the study only charted reading growth or implemented a reading intervention (in which case reading intervention approach was included) is presented in Results. Of the twenty-two studies included in the qualitative review, 21 included an intervention and 1 only charted reading growth (with no intentional intervention, but some participants received various interventions outside the study). Of the eight studies included in the ALE meta-analysis, all included an intervention.

ALE meta-analysis
fMRI studies that included either Talairach or MNI coordinates in their results were included in the meta-analysis. Data presented in MNI coordinates were converted to Talairach coordinates using Lancaster conversion provided on brainmap.org [95,96]. Data presented in Talairach coordinates were included as presented in the original paper regardless of conversion method. Activation likelihood meta-analysis was conducted using GingerALE v.2.1.1 [89,91,92,97]. Only one experiment or contrast per study was included in the meta-analysis such that each participant group is not represented multiple times in analysis. When available, we selected experiments that contrasted activation increases for postscan over prescan in the RD group [63,[98][99][100][101]. When RD Post . Pre contrasts were not available we used RD Followup . Pre [102]. When only postscan data were available, we used group contrasts RD . Control [69] and Responders . Nonreponders and Controls [103]. Eight experiments were included, with a total of 173 participants and 90 activation foci. Of the 90 foci, 1 fell outside the mask, 1.1% of the total. It is typical for a small percentage of foci to fall outside the mask [93]. A false discovery rate (FDR) of 0.05 and the recommended minimum cluster volume of 264 mm 3 (generated by GingerALE2.1.1) were used and images were viewed using Mango version 2.6 [104] It should be noted that we included foci of activation increases that the Eden et al [63] study stated fell short of their significance thresholds of p,0.001 and k.80. For the Richards et al study [99], we included only the orthographic experimental group as these are the coordinates that were reported and the group that showed significant change in associated with treatment.

Results
The literature search resulted in 22 studies that met criteria for inclusion in the descriptive literature review (Fig 1). The studies are presented in chronological order with their participant groups, interventions used, and intervention dosages in Table 1. The imaging modalities, functional tasks, and imaging findings are listed in Table 2. Descriptively synthesizing across studies, key findings by anatomical region are presented below. Following this qualitative synopsis, we summarize the findings of the ALE metaanalysis of the eight studies with post-intervention fMRI data.

Descriptive Findings
Numerous brain regions were found to be associated with reading intervention in these studies. These regions included not only frontal, temporo-parietal, and occipital cortex, but also sublobar and subcortical areas. Below we describe the findings involved in each region in depth.
IFG. The IFG was found to be associated with intervention across multiple studies. Following reading intervention, previously underactivating L IFG in RD more closely resembled controls [98,102,105] with one study showing underactivation of at-risk children shifting to overactivation following intervention [100]. Interestingly, Richards & Berninger [106] found that children with RD showed higher functional connectivity than controls for the L IFG as related to R and L supplemental motor areas and L precentral gyrus as well as R superior frontal gyrus. Following intervention, no difference was observed between children with RD and controls. An additional functional connectivity study [107] similarly found that L and R inferior frontal connectivity was the same following treatment for RD as compared to controls, but in somewhat of a contrast, found that nonresponders to intervention showed less functional connectivity than did responders. Hoeft et al [108] found that activity in the pars triangularis of the L IFG was positively correlated with reading gains, whereas L IFG/Insula activity was negatively correlated. To summarize, it seems that L IFG involvement may be that of underactivation prior to intervention relative to controls, followed by normalization after treatment; however, there is not complete consensus among the studies in this review.
The right IFG is also prominent in findings related to intervention. Prior to treatment, at-risk children showed underactivation in R IFG [100]. Activation increases in R IFG were seen following intervention [63,69,98] and at follow-up [102] and considered to be normalized to the level of controls [99]. Prior to treatment, higher R IFG activity predicted higher reading gains for children with RD [108]. Following treatment, responders exhibited greater R IFG activation than did nonresponders and controls [103]. Nonresponders showed increased duration of activity in bilateral frontal areas [66], and exhibited less functional connectivity between left and right inferior frontal regions than did responders and controls [107].  Additional frontal areas. While the IFG was the most consistently involved region in intervention, other frontal regions emerged in some studies with several studies presenting results in superior frontal (SFG) and middle frontal (MFG) gyri. Prior to intervention, children with RD showed underactivation in R SFG [105] and increases in activation were seen following treatment [98] and at follow-up [69]. Activity in the L SFG was positively correlated in predicting response to intervention [108]. At followup to intervention, children with RD showed increased activation in L SFG/cingulate [69]. Prior to intervention, children with RD showed underactivation in L MFG [105] and L MFG activity contributed to whole-brain multivariate patterns that predicted responsiveness to intervention [108]. Pre-treatment activation levels in R MFG negatively correlated with posttest decoding scores [109]. Following intervention, children with RD exhibited increased levels of activation in R MFG [98,100] whereas adults showed increased activation in R and L MFG [63]. More isolated frontal lobe findings include at-risk children underactivating in bilateral frontal orbital cortex at pre-treatment [100] and responders showing increased dorsolateral prefrontal activation [110].
STG and MTG. Some consistencies emerged among studies in temporo-parietal areas, particularly in the STG and MTG. Before treatment, participants with RD showed underactivation relative to controls in posterior STG and temporo-parietal cortex that increased or normalized to the level of controls after intervention [98,100,102,111]. Increased L STG activation was evident at follow-up as well [100,111]. Responders to intervention showed greater activation than nonresponders in L STG prior to treatment [112,113] and following treatment [114]. Also, responders to treatment showed increased duration of activity in posterior L STG and the L hemisphere sequence of activation changed for responders such that temporoparietal areas activated prior to frontal areas, such that after treatment responders much more closely resembled [66]. In young children considered at-risk for RD, underactivation was seen in the R posterior STG prior to treatment. In adults, increases in R posterior STG/AG activation were seen following treatment [63]. While in one MEG study, responders exhibited bilateral temporal-parietal activation [110], another study showed nonresponders having increased R temporo-parietal activation [66]. Because of the lesser spatial resolution of MEG, these areas are more general than in fMRI.
Though the STG seemed to be the temporo-parietal region with the most consistent findings across studies, the MTG also emerged in findings from several studies. At pre-treatment, relative underactivation of the L MTG was seen, with activation increases observed following treatment [101,102,105,115]. As for responsiveness, higher L MTG activity was predictive of better response to intervention [109,112] and responders activated more than controls in L MTG/AG following intervention [114]. In addition to the L MTG, increases in activation were also shown for RD in the R MTG following treatment [98,115]. However, Shaywitz et al [102] found that R MTG activity was higher in RD at pretreatment than at follow-up and Odegard et al [103] found that  No notable activation differences between responders and nonresponders at baseline. Following intervention, responders showed increased duration of activity in the L posterior STG, SMG, and angular gyrus. The nonresponders showed increased duration of activity in R temporoparietal and bilateral frontal areas. Responders showed changes in the sequence of activation to more closely resemble CT by initiating in extrastriate, followed by temporoparietal, and then frontal areas and this temporal profile was not apparent in nonresponders.  after treatment non-responders showed greater activation relative to controls and responders in R MTG. In addition, one study found that responders had increased activity at baseline in R mesial temporal cortex, an area not identified in other studies [112].
Other temporo-parietal areas. Though less consistent across studies than the STG and MTG findings, activity in other temporo-parietal areas, including inferior temporal gyrus (ITG), inferior parietal lobule, SMG, and AG, was associated with reading intervention in a few studies. In the ITG, children with RD underactivated in the L hemisphere relative to controls at baseline, a difference that was no longer present following treatment [105], a finding in congruence with Gebauer, Fink, Kargl et al [101] that showed increased activation in L ITG for poor spellers/readers at post-intervention. In adults, increases in activation were seen following treatment in R and L inferior parietal lobule and left intraparietal sulcus [63]. At pretreatment, at-risk children underactivated in bilateral SMG, and following treatment activated more than controls in R SMG [100]. Pre intervention L SMG and angular gyrus activity positively correlated with post-intervention fluency gains [113]. Following intervention, responders showed increased activation of L SMG [66] and activated more than controls in L MTG/AG (BA 39). Adults showed increased activity following intervention in R STG/AG (BA 22/39) [63].
Occipital and fusiform. Children with RD underactivated in the occipital/fusiform region [105] and superior lateral occipital cortex [100] prior to intervention. Baseline activation in the R fusiform and R fusiform/mid occipital gyri positively correlated with later decoding scores [109]. At baseline, children who were responders to intervention showed greater activation than nonresponders in L ventral occipitotemporal region [112]. Post treatment L fusiform activation positively correlated with future gains in letter knowledge for young children [116]. Following intervention, increases in activation were seen in adults in the L fusiform gyrus [63] and in children in the L lingual gyrus [98]. Onset latency increased in R lateral occipitotemporal cortex [115]. Of interest, the only region in which children who were on track for typical reading achievement activated more than at-risk children who had received treatment was in the L superior lateral occipital cortex [100]. At a one-year follow-up, increased activation was seen in occipitotemporal regions for children with RD [102].
Pre/postcentral. Limited evidence indicates possible differences in activity in precentral and postcentral gyri associated with intervention. Following treatment, children considered at-risk for RD showed increased activity relative to controls in L precentral gyrus [100]. Responders exhibited increased onset latency [115] of premotor cortex. Increases were seen in R inferior postcentral gyrus (BA 43) in adults following intervention [63] and in bilateral postcentral gyrus in children at follow-up [69].
Subcortical. Post-intervention increases in activation were found in adults with RD in the L hippocampal gyrus [63]. Increases in thalamus activity were observed in the left hemisphere following intervention for adults [63], bilaterally for children following intervention [98], and in L hemisphere for children at follow-up [69]. Following treatment, children with RD showed increased activation in L putamen relative to controls [69]. At follow-up in the same study increased activation was seen in L and R insula/putamen.
Insula. Following treatment, increased activation was seen in R insula/IFG [69,98] and at follow-up in bilateral insula/putamen [69]. Prior to intervention, bilateral insular activity contributed to predicting response to intervention with a positive correlation for R insula and a negative correlation for L insula [108].
Additional sublobar/medial areas. Prior to intervention, the R precuneus contributed to prediction of responsiveness [108]. At treatment follow-up, children with RD showed greater activation than controls in R precuneus [69]. Children with RD showed greater activation than controls in R anterior cingulate after treatment [100] and at follow-up [69]. Prior to treatment, L posterior cingulate activation was negatively correlated in wholebrain multivariate patterns predicting responsiveness to intervention [108]. At post-intervention, poor spellers/readers showed increased activation in R posterior cingulate cortex when compared with a no-treatment group [101]. At treatment followup, children with RD showed greater activation than controls in R posterior cingulate [69].

Results of ALE meta-analysis
Of the 22 studies in the qualitative literature review, eight provided sufficient information (i.e., coordinates for cluster maxima of activations at posttest when available or at follow-up when posttest was not available) to be included in ALE metaanalysis (Table 3). Results from the meta-analysis are presented in Table 4 and Fig 2. ALE revealed five clusters when using FDR of q = 0.05 and a cluster threshold of 264 mm 3 , the experimentally determined recommended threshold. These clusters were located in L thalamus, R Insula/IFG, L IFG, R posterior cingulate, and L middle occipital gyrus.

Descriptive Review
Though relatively few studies have examined the neurobiology associated with reading intervention and differing methodologies have been employed, some commonalities among studies have emerged in this review. Our descriptive literature review revealed that across studies, reading intervention may be associated with differential activation in bilateral IFG, STG, MTG, MFG, and SFG, as well as bilateral occipital regions, postcentral gyri, inferior parietal lobule, and insulae, among others. In almost all of these regions, differences involve underactivation for RD prior to intervention with relative increase following intervention. These results seem to be consistent with literature that describes underactivation in these areas for RD relative to typical readers and points toward normalization through remediation.
Not all of the patterns that emerged across these intervention studies were consistent with previous literature addressing imaging of RD. Several studies in this review indicated underactivation of L IFG prior to intervention for RD. Interestingly, while these findings are similar to that revealed in meta-analyses of RD [62] they are in contrast to the accepted model of L hemisphere dysfunction that includes L IFG overactivation in RD as a compensatory mechanism for underactivating temporo-parietal and occipito-temporal regions [41,42] In the studies we reviewed, reading intervention was generally associated with increased L IFG activation to normalizing levels. However, as one study [100] showed overactivation in the L IFG (among other areas) in RD following intervention, there may be support for the conventional model of RD overactivating L IFG subsequent to intervention. One proposed hypothesis that would account for this pattern is that of an inverted U activation curve [68]. In this model, children with RD may have a different starting point on the curve when performing a reading task. They may start at (and remain longer) at the low point of activation. With intervention, they may exhibit increased activation in relevant areas. In contrast, unimpaired readers may start out with higher activation, but over repeated exposures or training will decrease in activation. While still speculative, the inverted U should be considered when conducting imaging studies with RD, even if the study is not directly concerned with intervention as perhaps undocumented or inadvertent treatment could be underlying increased activity.
Responsiveness to intervention. In addition to exploring the differences in activity associated with intervention, we were interested in finding predictors of intervention response and differences among responders and nonresponders. Individuals may vary in their neurobiological receptiveness to training or instruction. Exploring these differences may eventually facilitate the targeting of intervention to individual needs. A few studies have begun the investigation of neural predictors of reading improvement in children with RD. Pre-intervention fMRI activity in the R IFG [108] L MTG, L STG, L ventral occipitotemporal, and MEG activity in the R medial temporal cortex [112] predicts response to intervention. Hoeft et al [108] found that brain measures were more predictive of reading gains for adolescents with dyslexia than were behavioral measures, with greater activation in the R IFG during rhyme-judgment predicting greater reading improvement over the next 2 K years. Another study of adolescents who struggled with reading showed that brain activity prior to intervention was predictive of behavioral response to a year-long intervention that emphasized vocabulary, comprehension, word study, and fluency. Higher MEG activity in L middle, superior temporal, ventral occipitotemporal regions,and R medial temporal cortex predicted better response as demonstrated by word reading efficiency than was predicted by pre-intervention reading accuracy or fluency measures alone [112]. These areas overlap with the previously discussed posterior areas involved in reading. In general, this suggests that adolescents with neurobiological profiles more closely resembling typically achieving readers are more likely to respond well to intervention.
The studies which looked at differential responsiveness and postintervention imaging indicated that children who demonstrate behavioral response as evidenced by standardized test scores, exhibited more activation in left middle temporal and posterior superior temporal areas [66,114] and left inferior parietal region [103] following intervention as compared to children who did not respond to intervention. Nonresponders (a year after intervention) had increased right middle temporal lobe during a letter-sound task [103]. Functional connectivity data indicate that responders and non-impaired readers exhibit connectivity between inferior frontal regions that is absent in nonresponders [107]. Identifying these patterns of responders and nonresponders may provide insight into determining if these children are somehow at an optimal state to grow or unknowingly primed to respond to intervention, so that the lesser responders can be moved into that zone.
Changes: Normalizing or compensatory. Given that change in functional activity appears to be associated with reading intervention leads to consideration of what the quality or type of change is. That is, does reading intervention lead to readers with RD becoming more like typical readers or does intervention lead to compensatory changes that enhance reading ability? While the limited studies available may not present a definitive answer, they do provide some insight. Several of the studies reviewed provide evidence of normalizing changes [63,66,69,98,99,102,106,115], including normalized functional connectivity [106]. However, a few studies showed evidence of possible compensatory mechanisms [63,66,98]. Interestingly, one study presented evidence that responders to intervention exhibit normalization while nonresponders exhibit compensation [66]. Though the evidence is limited, this may be indicative of differing types or severities of RD (i.e., less severe RD responds better both behaviorally and neurologically to intervention). Alternatively, it is possible that the responders are typical readers who have not previously received adequate instruction and when provided with adequate instruction show the functional activation profile of typical readers. Another study [63] showed adults exhibiting both normalization and compensation, illustrating that not only can adults with RD be remediated but also that the capacity for both normalizing and compensating mechanisms remains intact into adulthood.

ALE of Post-Intervention fMRI
The narrative review of studies provides valuable information. However, synthesizing across studies is challenging when so many brain regions are involved and results are descriptive rather than quantitative. Precision may be gained by using specified coordinates of cluster activation maxima in standardized space which can then be included in meta-analysis. Meta-analysis of eight studies experiments with post-intervention data given in specified maxima coordinates statistically overlapped with the findings summarized in the narrative review; however, some of the regions that appeared most prominent in the qualitative review were not the ones revealed in the ALE analysis.
The meta-analysis showed that following intervention, participants with RD exhibited increased activation in L thalamus, R insula/IFG, L IFG, R posterior cingulate, and L middle occipital gyrus. These regions have been found to be active in processes that would presumably enhance reading ability. The L IFG likely serves multiple roles including phonologic processing and perhaps articulatory recoding [41,42,49]. The right IFG may play a role in attention and detection of important task related cues [117]. The finding of increased R insula activity may relate to increased coordinated relay of information and heightened detection of salient events [118], useful processes in developing the skill of reading. The role of the L thalamus appears to be complex, but activity may be indicative of the role of the left thalamus in language and verbal memory [119]. While the posterior cingulate is believed to be involved in the default mode network, a role that does not particularly coalesce with these results, the posterior cingulate also appears to be involved in memory, language, and as a hub for information exchange [120,121].
While all the regions of the ALE results could be contextualized within the qualitative review (i.e., they overlapped with those reported to show increased activation following intervention), the regions that were statistically significant by the ALE analyses did not include some of the primary ones that are generally discussed in the literature (e.g.,L MTG and L STG). There are two possible explanations for this. On the one hand, it may simply be a result of bias due to small sample size. The meta-analysis only examined functional differences following intervention and was limited to studies that reported coordinates of activation. Because this produced a restricted subset of the studies that were in the literature review, the resulting areas that emerged are unavoidably biased and differed somewhat from the regions that predominated in the descriptive review. Alternatively, these findings could reflect the salient features of growth; that is, while the other regions may also play a role, those found in the ALE results may be the most critical regions that distinguish growth. For example, the thalamic findings are especially intriguing given the thalamus' fundamental role across many types of cognitive processes. As such, it may be a strong predictor of those who are able to compensate through other mechanisms or pathways.

Limitations
This study has its limitations. First among these is the limited number of studies that have explored neuroimaging and reading intervention. Even within these studies there are shared participants (i.e., appears to be identical groups or overlap among studies as noted in Table 1), such that this review is fairly limited in the number of unique participants. Also, due to the limited number of studies, we included studies that differed in several respects. For instance, we included one study that was performed with adult participants, though it has been established that adults and children differ in reading activation in both typical readers and those with RD [54,56,73]. In addition, we included studies that greatly differed in treatment type and treatment dosage. In a study with typical adults, Sandak, Mencl, Frost, Rueckl et al [122] showed that differences in treatment affect brain activation. Clearly, if differences in functional activity are related to treatment, type and dosage of treatment would be important factors. Additionally, though our focus was on RD, there is not a single definition for RD or being at-risk for RD and we found that the definition of RD varied by study.
A further limitation of the study is that neither the narrative review nor the ALE meta-analysis considered the statistical thresholds for imaging analysis used by each experiment. Not all studies approach data analysis in the same way, which presents a concern when synthesizing across studies. An experiment with more liberal thresholds of significance or cluster size would have its coordinates entered into the ALE meta-analysis with equal weight to that of an experiment (of comparable size) that used more stringent thresholds. In some instances, journal articles do not provide all details of their methods and analysis. In more recent years, a call for more explicit description of experiment methodology and data analysis has occurred [123]. As the functional imaging field has grown and evolved, reporting of methods and results have generally improved with the understanding of what information is necessary to the reader. This is a natural progression for a field in its infancy, so it is with this  perspective that we acknowledge that shortcomings in earlier works were likely due to limitations of the time. Nonetheless, it is important to state that at least one study appears to have used an uncorrected significance threshold and in several of the studies the reporting of methodology and statistical analysis is somewhat opaque. In one respect, our analyses are limited by the analyses performed in the original studies. In another respect, meta-analytic techniques are useful in identifying commonalities across studies and could add validity to studies that did not use stringent statistics if the results are consistent across studies. Another limitation of this study is that we did not address task difficulty in our analyses. Task difficulty may have an effect on functional activity and this effect may differ depending on reading proficiency. As Pugh et al [68] showed, factors that make words easier to read (i.e., frequency, imageability, consistency) result in higher activation in reading related brain areas for RD while nonimpaired readers exhibit, reduced activation. Interestingly, there is indication that even in a resting state condition, proficient readers show activation in reading networks [124]. This is an area of research that needs further exploration.
Another important consideration is we did not directly examine age-related differences. Gray matter development is not uniform. Rather, while areas involved in basic function such as motor and sensory areas mature early, areas involved in executive function, attention, and motor coordination mature later, with areas involving spatial orientation, speech and language development maturing in-between. As stated previously, functional changes occur over the course of development as well [55]. As such, when considering activity in a certain region of the brain, it is worthwhile to consider the typical course of functional activity in that region.

Conclusion and Future Directions
Despite the limited number of studies and the disparate methods of experiments included due to this low number, this analysis is a start at examining reading intervention and neuroimaging not only in a broad descriptive sense, but also in a quantitative manner. While acknowledging the limitations, the current results suggest that there are differences in brain activity associated with intervention. More studies using neuroimaging to explore reading intervention are needed to gain fuller understanding of functional activity and its relation to treatment. As reporting of functional data increasingly includes coordinates of cluster maxima, future meta-analyses will be able to incorporate more experiments. Also, we did not explore differences in type of intervention or dosages with regard to corresponding imaging data. Recently, an fMRI study of novel pseudoword learning indicated that the type of training used in new word learning may matter more to less skilled readers than to highly skilled readers [125], so there is certainly a need to explore this in future analyses. As neuroimaging studies become more numerous and investigators become more adept at analyzing and reporting the data in sufficient detail, more targeted meta-analyses can be conducted.
This paper is an effort to review functional imaging findings that are associated with reading intervention. Using neuroimaging to explore intervention is not a new concept [126], yet because these studies are complex, expensive, and time consuming, relatively few have been conducted. This is an early examination of an area of research that has substantial room to grow. Accordingly, caution should be used in interpreting results. Few conclusions can be drawn at this early stage, but this is an effort to guide future research and to underscore the need for larger scale studies. At this point, we are approaching this as a basic science question, seeking to better define one piece of a very complex puzzle. It is far too soon to suggest any practical implications, yet future studies may lead to practical applications that eventually impact treatment of RD.

Supporting Information
Checklist S1 PRISMA checklist.