Forty-seven children (eight to twelve years old) who were poor readers were randomly assigned to either an intensive 100-hour program of systematic and explicit remedial reading instruction focused primarily on developing word-level decoding skills (n=35), or they were assigned to a control group which received normal classroom instruction (n=12). There was also a control group of good readers (n=25) of the same age. The remedial instruction was distributed over about six months of schooling, with instruction occurring in groups of three children with one teacher. (Although the remedial instruction came in one of four alternative forms (see Experimental Procedures), there were no reliable differences among the children assigned to the different forms in either initial behavioral measures or DTI measures, nor in the impacts of the instruction (see Supplemental Results and Discussion, available online). Hence the data reported here are collapsed across the children in the four forms of remedial reading instruction.) The remediated and unremediated poor readers scored equivalently at the pre-instruction scan on multiple measures of reading ability, whereas the group of good readers scored significantly better than both groups of poor readers on every reading ability measure (see Table S1
, available online). The behavioral results indicated that the poor readers who received the remedial instruction showed significant improvement on most of the age-standardized Woodcock Reading Mastery Test - Revised (WRMT-R, Woodcock et al. 1998) reading ability measures when retested following the instruction period, but that the control poor readers did not show improvement on these measures, indicated by a reliable overall group by time effect (F1, 45
= 4.36, p < .05), with means shown in . Individual ANOVAs for each measure indicated that the interaction between group and time was reliable only for the subtest measuring non-word reading ability (Word Attack scores, F1, 45
= 5.22, p < .05), but not for the subtests measuring real word reading ability (Word Identification) or passage comprehension ability (Passage Comprehension). This pattern of outcomes suggests that the instruction specifically improved phonological decoding skills more than the standard reading curricula did. This conclusion was also supported by an analysis of changes in raw scores on all ability measures collected from the poor readers before and after the treatment phase (See Supplemental Results and Discussion and Table S2
Changes in age-standardized Woodcock Reading Mastery Test - Revised scores between the pre-remediation and post-remediation scans
The DTI results indicated that poor readers who received the remedial instruction showed a reliable increase in FA between the pre-remediation and post-remediation scans, with a peak difference in the left anterior centrum semiovale, as shown in . Corresponding contrasts conducted for the two control groups that received no remedial instruction found no areas showing either an increase or decrease in FA between the two scans, indicating that the change in FA among the remediated poor readers was not due to maturational changes over the six-month interval between the two scans. This same region also showed significantly reduced FA at the pre-remediation scan among all poor readers relative to the group of good readers (). The reliable increase in FA between the two scans among the poor readers, but no change in FA between the scans among the good readers, nor among the unremediated poor reader controls, resulted in a significant group (3) by time (2) interaction with a peak F-value in the same region of the left anterior centrum semiovale (), strongly suggesting that intensive remedial reading instruction led to changes in some microstructural property of white matter in a region of left frontal white matter, a region that differed between good and poor readers prior to the treatment. Additional analyses presented in the Supplemental Results confirmed that these findings were not due to the particular voxel-based analysis methods that were used; essentially identical results were obtained using unsmoothed data and non-parametric statistical inference methods (See Supplemental Results and Discussion and Figures S1
Figure 1 (A) Region where the poor reader group showed an increase in FA between the pre- remediation and post-remediation scans (peak t(34) = 5.12, at Montreal Neurological Institute (MNI) coordinates −12 28 36, spatial extent = 450 voxels, p < (more ...)
Figure 2 (A) Location of the cluster of voxels with the maximum F-value (peak F2, 69 = 9.66, spatial extent = 49 voxels, p < .0005 uncorrected, at MNI coordinates −12 26 40) for a test of the group by time interaction. (B) Mean FA for this cluster (more ...)
Because increased FA in highly organized white matter can occur due to either a relative decrease in radial diffusivity or a relative increase in axial diffusivity (or both), a further analysis examined the remediation effect in each of these components separately in the region shown in . It was the radial diffusivity that had changed in the remediated poor readers subsequent to the instruction. There was a reliable group by time interaction for radial diffusivity in this same region (F2, 69 = 5.92, p < .005); this measure reliably decreased among the remediated poor readers (t(34) = 3.98, p < .0005), but showed no change in either the good readers or the poor reader controls, as shown in . This pattern of radial diffusivity effects mirrors the findings for FA (a reliable increase in FA among poor readers who received remedial instruction but no reliable change in FA among the two unremediated groups (). By contrast, the other component of FA, axial diffusivity, showed no significant changes between phases for any group at this location, nor was there a reliable interaction (). The pattern of diffusivity effects indicates that the difference in FA between poor and good readers before remediation is due to initially higher radial diffusivity in the poor readers, and that the change in FA results from a change in some microstructural feature (e.g. myelination, packing density, or axon diameter) that affects radial diffusivity. The pattern of results also argues against the pre-remediation differences in FA between good and poor readers being due to the existence of more crossing fibers or smaller diameter axons in the poor readers in the area, and argues against the proposition that the changes in FA resulting from remediation were due to changes in either of these microstructural features, both of which would be expected to affect axial diffusivity. This leaves increased myelination as a plausible mechanism of the microstructural change.
The findings of increased reading ability and increased FA strongly suggest that the remedial instruction brought about a change in both variables, but say little about the relation between the two variables. To investigate this relation in more detail and to assess which aspects of reading ability were associated with increased FA, an exploratory stepwise hierarchical multiple regression analysis examined how well the change in raw reading scores of an individual poor reader could account for that individual’s change in FA in the region. This analysis (which also took the change in age between scans into account) indicated that a model including the change in raw scores on two subtests from the Test of Word Reading Efficiency (TOWRE, Torgesen et al. 1998) provided the best fit to the change in FA data among poor readers (R2= 0.10, F2, 43 = 2.36, p = .11). The change in Phonemic Decoding Efficiency (PDE, a measure of non-word reading fluency similar to the WRMT-R WA subtest) was positively associated with change in FA (partial r = .23, p = .06). In contrast, the change in the Sight-Word Efficiency (SWE, a measure of real word reading fluency similar to the WRMT-R Word ID subtest) showed a negative partial correlation with change in FA (pr = −.21). No other variables met the criteria for entry into the model. An identical analysis conducted with radial diffusivity in the region as the dependent measure also showed that these same two measures provided the best fit to the data (R2= 0.13, F2, 43 = 3.41, p < .05) with change in PDE significantly negatively associated with change in radial diffusivity (pr = .23, p < .05) and change in SWE positively associated with the change (pr = .29). In contrast, for axial diffusivity, an identical stepwise regression analysis indicated that no change in any of the raw ability measures explained enough variance for entry into the model (p’s > .15). The outcome of these analyses indicate that there is a coupling between the behavioral change in reading and the anatomical change measured by FA and radial diffusivity, and indicates that increased phonological decoding ability provides the best predictor of increased FA and decreased radial diffusivity.
These results and conclusions are further supported by additional analyses (described in the Supplemental Results and Discussion) of the relationships between individual differences in various reading abilities and various diffusion tensor measures in the entire sample of good and poor readers, (using reading and diffusion measures obtained prior
to the remediation), in the cluster that eventually showed an increase in FA among the remediated poor readers. Multiple regression analyses indicated that individual differences in phonological decoding ability (as measured by WRMT-R WA scores) were strongly positively related to FA (see Figure S3
), strongly negatively related to radial diffusivity, and only weakly negatively related to axial diffusivity at the time of the pre-remediation scan (see Figure S4
). These findings suggest that radial diffusivity drives the positive relationship between FA and individual differences in reading ability measured at the initial scan. In addition, both FA and radial diffusivity were more strongly related to Word Attack scores than to Word ID scores, suggesting that connections passing through the cluster area may be more important for phonological processing than for direct access to meaning via a direct orthographic route (see Supplemental Results and Discussion).
To determine the orientation of the tracts showing the remediation-related change and to identify the cortical areas that they likely connect, fiber tractography was carried out on group-averaged diffusion tensor data, using as a seed region the cluster of voxels showing a reliable group by time interaction. These group-averaged tracts were remarkably similar in their gross morphology between the good and poor readers and also across the two scanning sessions, as shown in , indicating the reliability of the data and the tracking methods. The principal direction of diffusion in the region showing a group difference in FA at the pre-remediation scan remained the same at the follow-up scan, and the fibers identified as passing through the regions were remarkably consistent between the two scans for both groups of subjects, suggesting that microstructural changes in the white matter within the region, rather than changes in the orientation of fibers, are responsible for the remediation effect and for the relationship of reading ability to the diffusion measures. The principal diffusion direction was anterior-posterior in all groups, and fibers passing through this region extended anteriorly and medially to terminate in a medial region of the superior frontal gyrus () and extended posteriorly and superiorly to terminate in the left paracentral lobule ().
Figure 3 (A) Consistency of the group-averaged tractography for PR, PC, and GC groups at each of two scans, using a seed region based on the cluster in . Color scale indicates the consistency of the tracking across the groups and phases, with red indicating (more ...)
To check for consistency with previous DTI studies of white matter abnormalities among poorer readers in a left temporo-parietal region (Beaulieu et al. 2005
; Deutsch et al. 2005
; Klingberg et al. 2000
), we tested for group differences and a remediation effect in this region that had shown a relation to reading ability in these previous studies. Although there were no statistically reliable effects in the voxel-wise analyses, the FA was reliably lower among poor readers at the initial scan when the average FA across the entire region of interest was examined and the specific analysis was closely matched to those previous studies. A review of diffusion studies of this region (Ben-Shachar et al., 2007
) suggests that the reduced FA among poorer readers is probably due to increased fiber crossings, and if this is indeed case, then intensive reading remediation would not be expected to change the coherence or the orientation of the fibers. Consistent with this expectation, there was no remediation effect in the region (See Supplementary Results and Discussion).