Project 2

Brenda Rapp, Ph.D. and her team study the effects of treatment of spelling abilities using a cognitive neuropsychologically based approach. Participants receive treatment focused on spelling selected sets of words, with expected improvement of trained words as well as related words with overlapping spellings (i.e., their neighbors). Behavioral and neuroimaging measures reveal patterns of recovery of spelling as well as general language ability and the brain mechanisms that support it.

Dr. Rapp is Professor and Chair of the Cognitive Science Department at Johns Hopkins University and Director of the CogNeuro Research Lab. After working in Special Education, helping children with language disorders, she obtained a Ph.D. in Cognitive Science, and now studies acquired language impairments in adults in order to better understand how the brain processes language. She is an expert and leading authority in acquired dysgraphia research.

Johns Hopkins University’s Cognitive and Brain Sciences Laboratory Website

See our recent publications!

Current Projects:

Behavioral results: Treatment, maintenance, generalization and selectivity.

Treatment effects. During this experiment, completed training data sets were collected for 23 individuals with acquired dysgraphia, i.e. impaired ability to spell and write. All participants were trained on 40 words, each trained for an average of 16 trials over an average of 27 sessions, using the Test/Study/Test training approach. Accuracy and response time data were collected on every training trial. The training effect for the group was significant (p < 0.001), indicating a 17% increase in the odds of correctly spelling a letter per training session. For response times (RT), the training effect for the group was also significant (p < 0.01), indicating a 2% decrease in RT (seconds per letter) per training session. Overall, individual subject analyses revealed significant effects for 18 of 23 individuals.

Maintenance and generalization effects.  Generalization to untrained word sets (average n=112 words) was evaluated pre, post and at 3 months follow-up and was compared to the participants performance on the trained words at the same time points (pre, post, and at 3 months follow-up) (n=23 participants). Results (Fig. 1) showed significant improvement from pre to post for both trained and generalization words (p < 0.001 and p < 0.01, respectively), though the improvement was significantly greater for trained words compared to generalization words (p < 0.001).

Spacing of treatment delivery in dysgraphia treatment. For this investigation, subsets of items were trained on either Clustered or Distributed Schedules, with the total number of training trials matched for two schedules. In the Clustered Schedule, words were trained for 3-4 trials within a session, every 8 sessions (on average).  In the Distributed Schedule, words were trained only once within a session, every 2 sessions, (on average). For the Distributed Schedule the variability in number of days between training trials (ISI) and between training and testing trials (RI), allowed for the evaluation of the relationship between ISI and RI.  For accuracy, the results show that (beta = -0.178, p < 0.001), participants improved at a significantly faster rate for words in the distributed, relative to the clustered condition (Fig 2). For the analysis of testing interval (n=21), the results show that as retention intervals increased, performance was relatively better with increasingly longer ISI. For example, while performance for 3-day retention was best with a 1-day ISI, performance was better for a 40-day retention interval with a 30-day (vs. 1-day) ISI (Fig. 2).
 

Neural changes associated with recovery of function. BOLD data were collected from the spelling task during scanning, Letter Probe Protocol. The Letter Probe Protocol includes an experimental and baseline task: Spelling Probe and Case Verification tasks. The tasks involve identical trial events of 7 second duration each, varying only in the instructions (task prompt) given to subjects: (1) an auditorily presented task prompt indicating if it is a Spelling or Case Verification trial; (2) fixation cross, (3) auditory stimulus, (4) a single visually displayed upper-case letter, and (5) a blank response screen On SPELLING trials, participants respond yes/no to indicate if the visually presented letter is in the spelling of the word they heard. However, on the CASE trials, the visually displayed letter appears either in upper or lower case and participants are instructed that the auditorily presented word is irrelevant and that they are to respond with a yes/no if the visually presented letter is in upper case.

Treatment-related changes in mean BOLD. In this experiment, BOLD response (n=20) to the spelling task (stimuli includes treatment words) compared to the control task was collapsed across pre and post time points to identify the spelling network recruited by the dysgraphic participants. A second analysis evaluated differences between pre and post treatment BOLD response for the spelling task (treatment words), with changes indicated in Fig. 3 with black circles showing significant upregulation of bilateral left superior occipital and posterior cingulate cortices.

Recovery-related network-connectivity changes. Background connectivity analysis involves the analysis of the time-course of the residual of a general linear model (GLM) analysis that includes task-related regressors and is considered to measure both intrinsic and task-based activity. Using the spelling-task BOLD response, we calculated the pairwise background connectivity correlation matrices for 235 cortical nodes for neurotypical participants (n=23) and dysgraphics (n=16) at pre and post treatment time-points. Using hierarchical clustering analysis, we identified a reference modularity structure consisting of 8 bilateral clusters. We then calculated graph-theoretic modularity values for each participant. Modularity measures the degree to which inter-node functional connectivity is organized according to the reference modular organization for spelling.  We found that for the dysgraphics:  pre-treatment modularity indexes deficit severity and, importantly, predicts future responses to treatment. Further, we found that modularity increased from pre to post-treatment (Fig. 4D) so that the magnitude of change correlated with behavioral change and that changes were concentrated in ventral temporal-occipital cortex (vOTC) (Fig. 4E). In summary, lower deficit severity was associated with higher modularity, which also resulted in larger treatment gains critically involving vOTC connectivity.  These findings provide, for the first time, a clear link between  the brain’s modularity structure and  treatment-based recovery of spelling function.

Changes in representational integrity: Local heterogeneity regression analysis. In Purcell & Rapp (2018) we developed a novel method –Local Heterogeneity Regression (Local-Hreg)– that analyzes task-based BOLD measuring the between-voxel heterogeneity (sparseness) of neural responses. We validated the method with neurotypical data (n=30) applying it to single-word reading BOLD data.  We found that within left vOTC, well-learned (high frequency) written words produce more locally heterogeneous neural responses than less-well learned (low frequency) or unknown words. In a second set of analyses (n=25), we evaluated the prediction that, in acquired dysgraphia, the recovery of word spellings should produce an increase in the local heterogeneity of BOLD response. This was confirmed from a whole-brain local-Hreg analysis of spelling fMRI, acquired both before and after treatment. Results identify a region within left vOTC in which local-Hreg at pre-treatment indexes deficit severity and also predicts future response to treatment. Furthermore, pre-to-post training increases in local-Hreg occurred within this region (Fig. 5) near the neurotypical vOTC cluster and the canonical VWFA (visual word form area).

Voxel-based lesion-deficit mapping. Our voxel-based lesion-deficit analysis with 32 dysgraphic individuals found that orthographic long-term memory system (O-LTM) deficits were selectively associated with lesions to either the left inferior frontal gyrus/junction (IFJ) or the fusiform gyrus (FG), and orthographic working memory system (O-WM) deficits with lesions to the superior/intraparietal sulcus (IPS) region (Fig. 6). Relatedly, we have examined whether O-WM is part of a general WM system or if there are specific neural substrates dedicated to O-WM.  In Martin et al., (in prep) we compared the lesion distributions of individuals with O-WM deficits phonological WM deficits and found non-overlapping regions. Within the parietal lobe, phonological working memory (PWM) deficits were associated with lesions to the supramarginal gyrus, O-WM lesions were more posterior and superior.

Meta-analysis of spelling substrates. Purcell et al. (2018), identified substrates that were significantly likely to be activated across existing studies with neurotypical adults, finding a highly left-lateralized network (Fig. 7). The two most consistent regions were the fusiform gyrus (FG) and the posterior inferior frontal gyrus/junction (IFJ); 7 additional left hemisphere sites included: superior frontal sulcus (SFS), superior /intraparietal sulcus (IPS), supramarginal gyrus (SMG) superior temporal gyrus (STG), supplementary motor area (SMA) and pre-post central gyri. And in the right hemisphere: the insula and STG.

Resting state fMRI (RS-fMRI).  In Ellenblum et al., (in progress) we examined the resting state neural network of spelling in neurotypical adults, comparing the within- and across-network RS coherence properties of spelling (and reading) networks to those of three reference networks (default mode, sensori-motor, and attention).  Additionally, we identified novel, distinctive properties of the orthographic processing networks, and established that their key node -the Visual Word Form Area (VWFA) – has unusually high levels of connectivity with a broad range of brain areas (Fig. 8).

Figure 1. Average accuracy for training and generalization words at pre, post and 3 months follow-up.
Figure 2. The relationship between retention interval and number of days between training trials (3,10,20,30 days) on spelling accuracy; regression values.
Figure 3. Univariate RFX-GLM Dysgraphic treatment effects. Pre/post spelling upregulation DIFFERENCES due to training (yellow-orange scale (circled in black). Dotted lines on RH depict left coronal slice location. Lesion overlap (n=20); pink/yellow scale.
Fig 4. D. Modularity significantly increases with treatment. E. The connectivity changes supporting modularity increases are concentrated in vOTC)
Figure 5. Left vOTC area where local heterogeneity of BOLD spelling response to training words increases pre to post treatment
Figure 6. Brain areas associated with O-LTM deficits (orange) and O-WM (blue/pink)
Figure 7. Meta-analysis indicating high-likelihood activation peaks in spelling
Figure 8. RS conectivity of the VWFA