Conclusion: Instructional intervention that improved phonological performance in dyslexic boys was associated with changes in brain lactate levels as measured by proton echo-planar spectroscopic imaging.

We have previously reported that dyslexic boys demonstrated a greater area of brain lactate elevation compared to a control group during a phonological task in the left anterior quadrant of the brain (1). Dyslexia is a language disorder characterized by poor reading due to a phonological deficit (1,2). We used a fast spectroscopic imaging technique called proton echo-planar spectroscopic imaging (PEPSI) to non-invasively measure brain lactate changes associated with language activation. Functional MR spectroscopy (fMRS) using the PEPSI technique is an alternative approach for detecting regional brain activation and measures tissue-based lactate changes (a direct measure of metabolism) produced by a temporary mismatch of oxygen delivery and consumption in response to neuronal activation (3). Following completion of the prior fMRS study, the dyslexic boys entered into a treatment program designed to improve their phonological abilities. Both the dyslexic boys and the control boys were re-imaged a year after the initial imaging session. We hypothesized that a positive response to therapy would be reflected by changes in the distribution of brain lactate activation as measured by fMRS. The purpose of this study was to test the effect of this instructionally-based treatment for dyslexia on the brain lactate response during reading-related language tasks.

Eight dyslexic and seven non-dyslexic (control) boys were imaged using PEPSI (4) while performing four different cognitive tasks. Only males were chosen for this study in order to minimize the variance between subjects, especially given that Shaywitz et al have shown large functional MR imaging (fMRI) differences between the sexes during language processing(5). The dyslexic boys were imaged before and after a three week (plus followups) phonologically-driven treatment for dyslexia described by Berninger (6). The control group was also imaged at the same two time points as the dyslexic group but did not receive treatment because the control boys were already reading well above grade-level and age expectations. However, reimaging the controls at the same time as the dyslexics were reimaged permitted evaluation of where any change in the dyslexics might be due to maturation rather than instruction. The scan protocol and the language stimuli were identical across repeated scans and were equally spaced for the two groups. The boys were not re-imaged immediately after treatment because we wanted to evaluate whether the behavioral gain observed immediately after treatment would result in a long lasting change in lactate brain activity. The dyslexic and control groups were well-matched in age, IQ, and in head-size (number of total voxels) but not in reading skills on which they demonstrated marked differences at the first time point, as described below. The experimental tasks were designed to activate phonological and lexical access functions of the brain, while a tone task was used to activate auditory non-language functions of the brain. Two control tasks (listening to scanner noise and passive listening of word lists) were used to subtract out low-level acoustic stimulus effects. The phonological and lexical access tasks engage additional linguistic processes beyond that required for passive listening of stimulus words. Task specific activation was assessed by subtracting out the passive listening condition from the phonological and lexical access tasks. This method provides a means to identify activation related to processing phonological information or accessing word meanings, independent of brain activation due to the characteristics of word stimuli,. The component of brain activation related to auditory processing not specific to language was assessed by subtracting out the scanner noise from the tone judgments.

Conventional MR imaging and PEPSI were performed on a clinical 1.5 Tesla Signa MR imaging system (General Electric, Waukesha) equipped with version 5.7 software and a custom-built radiofrequency (RF) head coil developed by Hayes et al.(7). MR images were acquired in the sagittal plane (TR/TE 600/20 msec) and also in the axial plane (TR/TE1/TE2 2000/35/80 msec). The custom designed head coil was necessary to acquire MR spectroscopic data with high enough signal-to-noise ratio to detect the small lactate peak and also to maintain a reasonably short acquisition time to avoid motion and habituation artifacts from these young subjects. This custom rf head coil has been measured at producing an MR signal with 35% better signal-to-noise than the standard GE head rf coil (product hardware delivered with the scanner) (7). The coordinates of the Sylvian fissure and surrounding language-related structures were determined on the sagittal images which were co-registered with the axial images for both MR imaging and spectroscopic imaging. The areas sampled with PEPSI were based on the work of Ojemann et al. (8) that invasively demonstrated language activation in the anatomic region encompassing the Sylvian fissure and adjacent opercula. The single PEPSI slice was oriented to encompass the frontal operculum and the posterior portion of the superior temporal gyrus. Deeper subcortical structures were also included that are associated (through neuronal connectivity) with the cortical areas. Proton spectra were acquired using PEPSI, a spin-echo pulse sequence developed by Posse et al. (4) that allows fast spectroscopic imaging which is 32 time faster than conventional hydrogen spectroscopic imaging for the same spatial resolution. PEPSI is a method that is somewhat demanding on gradient amplifiers and may not function properly on most conventional scanners without echo-planar capability. PEPSI was chosen in our experiment over a single-voxel technique such as PRESS because we needed to map the lactate distribution over the whole brain slice in order to define the regional metabolic activation. Parameters for data acquisition included: TR/TE 4000/272 msec; 2 averages; 32 x 16 spatial matrix (zero-filled to 32x32 for reconstruction); 512 echoes in the echo-planar acquisition; 32 complex points per echo; full echo acquisition; field of view 24 cm; and slice thickness 20 mm. Voxel size was approximately 1 cm3. and the acquisition time for each PEPSI scan was approximately 4 1/2 minutes. Five PEPSI scans were acquired at each session during: 1) baseline (no extra sounds); 2) phonological task; 3) lexical access task; 4) tone discrimination task; and 5) passive listening as described below. Data were processed as described previously (9). The metabolites were integrated using the following procedure: 1) magnetic field inhomogeneity shifts (B0 shifts) were corrected by finding the maximum point of the NAA peak and resetting the ppm scale to 2.0 ppm for each spectrum; 2) the average baseline was determined from 32 points to the right of 0.0 ppm; 3) the maximum intensity point of the peak was determined within a set spectral range ( NAA = 2.0 +/- 0.07, lactate = 1.3 +/- 0.1 ppm); and 4) integration was performed by summing the spectral intensities for the NAA and lactate for the ppm ranges specified in step 3.

The University of Washington Human Subjects Institutional Review Board approval was obtained for this study, and each subject (as well as parent/guardian) gave written, informed consent. All subjects were right handed (90-100% on the Edinburgh Handedness scale(10)). The control boys had a history of learning to read easily and were reading above normal for age (average was one standard deviation above mean for age using the Woodcock Reading Mastery Test-Revised (11)) . The dyslexic boys had a developmental history of extreme difficulty in learning to read despite many forms of extra assistance at school and also had a family history of multi-generational dyslexia, which was confirmed in a concurrent family genetics study (W. Raskind, personal communication) at our center. At the initial scan, the dyslexic boys were reading on average 1.66 standard deviations below the mean for age on the Woodcock test (11). In addition, all the dyslexic boys were shown to have deficits in three skills that predict ease of learning to read and response to intervention: phonological (phoneme segmentation and/or memory for spoken nonwords), rapid automatized naming, and orthographic (speed of coding written words and/or accuracy of representing them in memory)(12) . Based on independent t-tests, the controls ( M=127.3, SD=10.8) and dyslexics (M=124.3, SD=11.1) did not differ in age in months at the time of the initial scan (t(11) = 0.49,p=0.637). Likewise, at the time of the initial scan, the controls (M=15.6, SD=3.2) and dyslexics (M=13.2, SD=1.6) did not differ in age-corrected WISC-III vocabulary scores (t (11)= 1.68, p=0.12), which provide the best estimate of Full Scale IQ. However, at the time of the initial scan, the controls and dyslexics did differ significantly in age-corrected standard scores for reading real words on the Word Identification (WI) subtest of the Woodcock Reading Mastery Test-Revised (WRMT-R) and for reading pseudowords on the Word Attack (WA) subtest of the WRMT-R: t(11)=6.81, p < 0.001 on the WI subtest and t(10) = 6.02, p<0.001 on the WA subtest. The differences for both real word reading (WI, controls, M=115.1, SD=9.2; dyslexics, M=75.5, SD=11.8) and pseudoword reading (WA, controls, M=110.2, SD=6.8; dyslexics, M=79.0, SD=10.7) were large as well as statistically significant.

Treatment consisted of fifteen 2-hour group sessions in a 3-week reading/science workshop. To motivate the boys who had a long struggle in learning to read, they were told that they were Einstein's Ninja Turtles. The instructor told them the story of how Albert Einstein had had learning disabilities in reading that he overcame (13). She also explained that gene mutations can confer special advantages, like the Ninja turtles, as well as disadvantages, like trouble learning to read. She used the fable of the tortoise and the hare to convince them that despite a slow start in learning to read, they could finish the race as skilled readers. A systems approach to intervention was used in the treatment in which instruction was aimed at all levels of language (subword, word, and text, (6)) during the first hour of each instructional session. Each session began with sound games to remediate their deficits in phonological processing. High-interest, polysyllabic words taken from science texts were presented orally. The boys counted the number of syllables in the spoken word and used colored counters to represent each phoneme in the syllables. Only after they analyzed the phonological structure of each word did they see the same words in written form. Next they were taught to decode the words using syllable patterns of written English that derive from the Anglo-Saxon, Latin/Romance, and Greek origins of English words (14) and correspondences between 1- and 2- letter spelling units and phonemes (12). Then, they competed in a "reading bee" to decode transfer words that varied in predictability and complexity but contained the most common syllable patterns and spelling-phoneme correspondences of English. Finally, they took turns reading orally science texts on the human brain (week 1), astronomy (week 2) and endangered species (week 3). Following this phonologically-driven instruction, the last hour of each session was devoted to hands-on science activities that contained the words practiced in the beginning of the session. Guest speakers who were scientists came to give lectures and information about famous scientists as part of this training. Follow-up sessions over several months were held to facilitate and maintain the improvements. Additional procedural details are reported by Berninger(6).

During repeated MR scanning, the same brain stimulation tasks were used. The children were asked to listen to aurally-presented words, non-words, and tone pairs at a rate of one stimulus pair every 4 sec. Language stimuli were composed of four groupings of word pairs, crossed for lexical status (word vs. non-word) and sound similarity (rhyming vs. non-rhyming) resulting in four sets of stimuli: word/word:nonrhyming (e.g. FLY-CHURCH); word/word:rhyming (FLY-EYE), word/nonword:nonrhyming (CROW-TREEL); word/non-word:rhyming (MEAL-TREEL). Nonwords such as TREEL allow assessment of sound processing without any meaning cues. The presenting order of word pair-types were counterbalanced and thus the ordering effects were controlled for. During rhyming (phonological task), subjects listened to the word pairs and judged whether they rhymed or did not rhyme; whether words were real was irrelevant. During the lexical access condition, subjects listened to the same word pairs, and judged whether the word pairs contained two real words, or contained a non-word; whether or not the words rhymed was irrelevant. Thus same stimulus lists were used for lexical access and rhyming, only the task instructions were changed. Subjects indicated their rhyme and lexical decisions by raising cards held in the right and left hands (the hand used to signal a "yes" response was counterbalanced across subjects). During passive listening, subjects listened to the same stimuli, but were instructed to alternately raise the left and right hand without making any judgments on the stimuli. For the tone judgment task, five pure tones (329.6hz, 350hz, 415.3hz, 440hz, & 523hz), were grouped into pairs of identical tones, or different tones. Subjects were asked to raise one hand if the tones were identical and to raise the opposite hand if the tones were different. The subjects were tested for accuracy of their responses for all tasks during a prescan training session and during the actual MR scanning. For the tone subtraction, the baseline scanner noise was used instead of passive listening. The PEPSI acquisition time for each task for approximately 4 1/2 minutes and a recovery period of 5 minutes was allowed between tasks based in part on the time for lactate recovery measured by Frahm et al (3).

To affirmatively evaluate focal brain activation from PEPSI images, Z-score maps were created from the lactate/NAA ratios based on the following equation:

[lactate/NAA (task) - lactate/NAA (passive listening)] / [standard deviation of lactate/NAA (passive listening)], where (task) in the equation refers to the task given during the scan which was either phonological, lexical access, or tone differentiation. The standard deviation of the lactate/NAA was calculated for each subject using all valid spectra of the control task (either passive listening or scanner noise). This Z-score was calculated for all voxels that contained valid spectra for each language condition(9). Definition of the threshold for lactate elevation indicating brain activation was based on Z-scores greater than 2.0 on a voxel by voxel basis. The cross-sectional PEPSI slice of the brain was divided into 4 quadrants (described below) and the number of voxels with elevated lactate within each quadrant was counted for each subject.

The lactate/NAA ratio was used in order to normalize for radiofrequency inhomogeneity, variable cerebrospinal fluid contribution, and also to standardize the lactate signal across subjects. An automated computer-software mask was applied to the spectra to ensure that the MR lactate signal was not contaminated with scalp lipid signal (1). In this software, a true lactate peak was differentiated from lipid contamination based on MR frequency which was determined from an in vitro lactate measurement (the lipid peak MR frequency was not the same as lactate frequency). Additionally, in previous work, we varied the TE to assess J-coupling properties of the peak at 1.3 ppm which identified this peak to be lactate (15).

Immediately following the intervention, the boys improved, on average, 8.9 points on age-corrected standard scores (0.6 standard deviation unit) of a measure of phonological decoding (WA WRMT-R (11)). On a criterion measure of oral reading of text, three were at the grade level just completed, three were above the grade level just completed, and one was still below grade level for oral reading but was at the grade level just completed for silent reading. Eight months later (about a year following the initial MRI and PEPSI neuroimaging scans), all but two had age-appropriate phonological awareness, the group showed a relative gain of 0.9 standard score units in phonological memory since the initial assessment prior to the first imaging session, and the group maintained a relative gain of 8.7 points on age-corrected standard scores (0.6 standard deviation unit) on phonological decoding since first assessed prior to neuroimaging (Table 1). Their phonological decoding (word attack) was now at the border between low average and average (M=89.7, SD=8.5 (6)). Additional details are reported by Berninger(6).

Phonological Decoding^a 81.0+9.8 89.9+9.6 89.7+8.5