OUP user menu

Anatomical correlates of dyslexia: frontal and cerebellar findings

Mark A. Eckert, Christiana M. Leonard, Todd L. Richards, Elizabeth H. Aylward, Jennifer Thomson, Virginia W. Berninger
DOI: http://dx.doi.org/10.1093/brain/awg026 482-494 First published online: 1 February 2003


In this study, we examined the neuroanatomy of dyslexic (14 males, four females) and control (19 males, 13 females) children in grades 4–6 from a family genetics study. The dyslexics had specific deficits in word reading relative to the population mean and verbal IQ, but did not have primary language or motor deficits. Measurements of the posterior temporal lobe, inferior frontal gyrus, cerebellum and whole brain were collected from MRI scans. The dyslexics exhibited significantly smaller right anterior lobes of the cerebellum, pars triangularis bilaterally, and brain volume. Measures of the right cerebellar anterior lobe and the left and right pars triangularis correctly classified 72% of the dyslexic subjects (94% of whom had a rapid automatic naming deficit) and 88% of the controls. The cerebellar anterior lobe and pars triangularis made significant contributions to the classification of subjects after controlling for brain volume. Correlational analyses showed that these neuroanatomical measurements were also significantly correlated with reading, spelling and language measures related to dyslexia. Age was not related to any anatomical variable. Results for the dyslexic children from the family genetics study are discussed with reference to dyslexic adults from a prior study, who were ascertained on the basis of a discrepancy between phonological coding and reading comprehension. The volume of the right anterior lobe of the cerebellum distinguished dyslexic from control participants in both studies. The cerebellum is one of the most consistent locations for structural differences between dyslexic and control participants in imaging studies. This study may be the first to show that anomalies in a cerebellar‐frontal circuit are associated with rapid automatic naming and the double‐deficit subtype of dyslexia.

  • Keywords: dyslexia; MRI; cerebellum; frontal lobe; rapid automatic naming
  • Abbreviations: ANT = anterior lobe; PTR = pars triangularis; RAN = Rapid Naming of Letters; RAS = Alternating/Switching Letters and Numbers test


Structural MRI studies of dyslexic adolescents and adults have demonstrated differences from controls in a variety of brain regions, including the inferior frontal gyrus, cerebellum, insula, caudate, corpus callosum, left temporal lobe and thalamus (Pennington et al., 1999; Eliez et al., 2000; Brown et al., 2001; Leonard et al., 2001; Rae et al., 2002; Robichon et al., 2000a, b). The cerebellar and inferior frontal gyrus findings have been most consistent, but few studies have measured a broad range of structures and no study has tried to replicate findings prospectively in a new sample.

The widespread neuroanatomical differences between adult dyslexic and control brains support a multivariate approach to characterizing the dyslexic brain. Leonard and colleagues (Leonard et al., 2001) used such an analysis in a study examining the morphology of language‐related areas in a group of dyslexic college students who were characterized by a discrepancy between their poor phonological decoding ability and reading comprehension. Compared with controls, dyslexic subjects were more likely to exhibit a duplication of the left Heschl’s gyrus, extreme leftward asymmetry of the planum temporale and parietale, small right cerebellar anterior lobes and leftward cerebral asymmetry (Leonard et al., 2001). The probability of a dyslexia diagnosis increased with the additional presence of each of these anatomical measures (Leonard et al., 2001). Frontal lobe measurements were not collected in the study of Leonard and colleagues.

Although multiple structural differences have been identified in dyslexics compared with controls, it is not clear whether these findings map onto the behavioural deficits exhibited by dyslexics. These include problems in phonological coding (Shaywitz et al., 1998; Wagner et al., 2000) and rapid naming of letters or switching letters and numbers (Wolf, 1986). Wolf and colleagues (Wolf et al., 2002) examined 144 severely reading‐impaired dyslexics from grades 2 and 3 who exhibited a discrepancy between their reading achievement and cognitive ability. Sixty per cent of these children demonstrated a double deficit in phonological and rapid naming skills. The inferior frontal gyrus and cerebellar findings (Brown et al., 2001; Leonard et al., 2001) could explain the deficits in phonological accuracy, rate or both.

This study was designed to replicate the study of Leonard and colleagues (Leonard et al., 2001) in a group of dyslexic children characterized by poor reading skills but average to superior verbal intelligence. A second goal of this study was to determine the probability of a dyslexia diagnosis for each of the anatomical measures, including a measure of the inferior frontal gyrus [pars triangularis (PTR)]. Finally, this study examined the relation between anatomical measures and measures of reading, spelling, verbal intelligence and selected language skills related to reading.



The University of Washington Institutional Review Board approved this project and participant consent was obtained according to the Declaration of Helsinki. Participants and parents provided written informed consent. The 18 dyslexic participants were recruited from probands in a family genetics study. To qualify as a proband, children had to meet these inclusion criteria for dyslexia: a difference of at least one standard deviation between pro‐rated verbal IQ (verbal comprehension factor) and at least one reading or spelling measure. Although this was the only formal requirement for inclusion, the dyslexic probands in the participating families were actually discrepant on multiple reading and writing measures, had deficits on language markers associated with dyslexia (phonological, orthographic and rapid automatic naming tasks), and a history of difficulty in learning to read despite special help in reading (Table 4). For further information on subject ascertainment see the study by Berninger and colleagues (Berninger et al., 2001). All dyslexic participants had pro‐rated verbal IQs (verbal comprehension factor) in the average to superior range (mean 110.8, range 92–129).

Dyslexia was defined as underachievement relative to verbal IQ rather than performance IQ because prior research showed that verbal IQ is a better predictor than performance IQ of reading achievement in referred (Greenblatt et al., 1990) and unreferred (Wechsler, 1991; Vellutino et al., 2001) populations. The pro‐rated verbal IQ, which is based on all the same subtests (except Arithmetic) as the verbal IQ, was used so that we would have a pure measure of the Verbal Comprehension Factor (Wechsler, 1991) not confounded with number reasoning. This approach allowed us to document that word reading was underdeveloped compared with verbal reasoning in the functional reading system (Berninger, 2001b) of these children with reading disability.

Additional clinical measures confirmed that this sample had specific reading disability (dyslexia) rather than reading problems related to primary language disability. The sample was impaired in both accuracy [see Table 4 for WRMT‐R (Woodcock Reading Mastery Test—Revised) Word Identification and Word Attack] and rate of oral reading of words [mean 82.8, SD 10.4, sight word efficiency; mean 82.4, SD 8.2, real word efficiency (TOWRE, Test of Word Reading Efficiency; Torgesen et al., 2001)] and text [mean 3.9, SD 3.2, rate; mean 4.4, SD 2.0 (GORT‐3, Gray Oral Reading Test; Wiederholt and Bryant, 1992)] and in spelling [mean 80.8, SD 7.8 (WRAT3, Wide Range Achievement Tests—Revised; Wilkinson, 1993); mean 78.7, SD 8.5 (WIAT II, Wechsler Individual Achievement Test, 2nd edition; Psychological Corporation, 2001)]. The sample was not impaired (at or near the mean) in handwriting (three measures used in our research programme for over a decade), expressive language [mean 10.3, SD 2.5 (CELF‐3, Clinical Evaluation of Language Fundamentals—Third Edition, Formulated Sentences; Seml et al., 1995], oral‐motor planning (measure of repetition of alternating consonant–vowel syllables) or fine‐motor planning [sequential finger movements (Finger Succession; Berninger, 2001b)]. None had a history of language delay or disability in the preschool years. Yet they were impaired (at or near a standard deviation below the mean) in phonological coding (Table 4), orthographic coding [see Table 4 for expressive coding and 28th percentile (Word Choice; Berninger, 2001b)] and rapid automatic naming (see Table 4), the three markers associated with the language phenotype for dyslexia (Berninger et al., 2001).

The 32 control subjects were recruited with advertisements. Parental interviews helped to screen for control children who learned to read with ease. Standardized tests of reading and the Wechsler Intelligence Scales for Children—III (WISC III) vocabulary subtest, as an estimate of the verbal comprehension factor, were then used to determine that controls read well for their age and ability. Six controls and one dyslexic did not receive a full behavioural testing battery. In addition, poor image quality precluded anatomical data collection for one male dyslexic. Age of the control and dyslexic children was not significantly different (dyslexic males, 135.4 months; dyslexic females, 138.3 months; control males, 134.9 months; control females, 139.8 months). One dyslexic child had a diagnosis of attention deficit disorder.

Psychometric measures

The psychometric battery included measures of phonological, rapid naming, orthographic and verbal ability. Accuracy for single‐word (Word Identification) and pseudoword (Word Attack) reading was assessed with the WRMT‐R (Woodcock, 1987). Spelling ability was determined with the WRAT‐3 (Wilkinson, 1993). Phonological awareness was assessed using a prepublication version of Elision from the Comprehensive Test of Phonological Processing (CTOPP) (Wagner et al., 2000). Orthographic processing was tested using the Process Assessment of the Learner (PAL‐RW) (Berninger, 2001a). Rapid automatic naming was assessed using the Rapid Naming of Letters (RAN) and Alternating/Switching Letters and Numbers test (RAS) (Wolf, 1986). Pro‐rated verbal IQ (all Verbal Scale subtests except Arithmetic) was obtained using the Wechsler Intelligence Scales for Children—III (WISC‐III) (Wechsler, 1991). Dyslexic hand preference was established using the Edinburgh handedness questionnaire (Oldfield, 1971). All but one dyslexic were strongly right‐handed. Writing hand was used as an index of handedness for the controls, and all were right‐handed.

MRI data acquisition

Sagittal images (1.2 mm) were acquired using a GE Signa 1.5 tesla scanner (version 5.8) (General Electric, Milwaukee, WI, USA) using a 3D fast spoiled gradient echo pulse sequence. Imaging parameters were as follows: TR (repetition time) 11.1 ms, TE (echo time) 2.2 ms, flip angle 25°, field of view 24 cm. The entire acquisition time was 4 min 36 s.

The images were assigned a random blind number to hide identifying information from anatomists at the University of Florida. Programs written in PV‐Wave (Visual Numerics, Boulder, CO, USA) processed the electronically transferred images and placed the images into a single file. The Talairach proportional grid method was used to examine comparable brain regions across subjects (Talairach and Tornoux, 1988). The Talairach system standardizes positions by relating them to an atlas brain where the horizontal plane intersects the anterior and posterior commissure. The scans were reformatted into a final set of 1 mm thick sagittal images to adjust for head tip in each plane of section.

Surface area and volume measurements

Cerebral and cerebellar hemispheres

The volume of each hemisphere was estimated by tracing around the circumference of every sagittal 1 mm section starting at the midline. This measure includes cortical and subcortical structures but excludes the brainstem and cerebellum. Intraclass correlation demonstrated that both intrarater and inter‐rater reliability were 0.99 for the brain volume measure. The volume of the anterior lobe of the cerebellum was estimated by measuring from the mid‐sagittal section to the lateral boundary where the posterior cerebral artery disappeared. The primary sulcus served as the posterior boundary for the measurement. The right anterior lobe volume was used as a predictor variable, as opposed to the asymmetry measure (Leonard et al., 2001), because the right anterior lobe contributed to all of the predictive cerebellar asymmetry variance in the study of Leonard and colleagues. Intraclass correlation demonstrated that both intrarater and inter‐rater reliability were 0.95 for the cerebellar measurement.

Planum temporale and planum parietale

The anterior border of the planum temporale (PT) was defined as the depth of the sulcus that is the posterior border of Heschl’s gyrus (Heschl’s sulcus). The posterior boundary was defined as the origin of the posterior ascending ramus or the termination of the sylvian fissure. The planum parietale rises from the termination of the PT into the parietal cortex. The surface areas of the PT and planum parietale were measured sequentially on every section between x = 46 mm and x = 56 mm. These measurement boundaries were chosen to maximize planar asymmetry (Best and Demb, 1999), ensure reliability and replicate previous MRI studies showing cognitive associations with PT measurements (Foundas et al., 1994; Leonard et al., 1996; Eckert et al., 2001). Intraclass correlation demonstrated that intrarater and inter‐rater reliability was 0.97 and 0.89, respectively, for the PT measurement.

Heschl’s gyri. Many individuals exhibit two Heschl’s gyri. A study of normal variation for the presence of Heschl duplications (H2) in 105 scans found that the frequency of a left H2 increased continuously from medial to lateral cortex (Leonard et al., 1998). The incidence was 17% at x = 34 (Talairach coordinate for the mediolateral axis) but increased to 56% by x = 48. The surface areas of H1 and H2 were traced between their limiting sulci on consecutive sagittal images between x = 34 and x = 48 because this region exhibited the greatest individual variability in Heschl’s gyrus morphology. Intraclass correlation demonstrated that intrarater and inter‐rater reliability was 0.97 and 0.98, respectively, for the Heschl’s gyrus measurement.

Pars triangularis (PTR)

The PTR was measured from 39 to 49 Talairach mm by tracing the surface formed by the anterior ascending ramus (AAR) and the anterior horizontal ramus (AHR) of the sylvian fissure. These are two major branches of the sylvian fissure and thus easily identified. The surface was traced from the tip of the AAR, down to the sylvian fissure and then, following the AHR, to the end. Intraclass correlation demonstrated that intrarater and inter‐rater reliability was 0.95 and 0.95, respectively, for the PTR measurement.


Multivariate analysis of variance (MANOVA) was used to compare anatomical measurements between the control and dyslexic subjects. Significant anatomical measures from the MANOVA were then used in binary logistic regression analysis to determine the probability of a dyslexia diagnosis on the basis of these anatomical measures. Multiple regression was used to examine quantitative relations between behavioural measures predictive of dyslexia and the anatomical variables.


Pearson correlation and t tests were performed to determine if age and gender were related to the anatomical measures. Age was not related to any anatomical variable. Males, however, had larger brain volumes than females [t(1,48) = 3.09, P < 0.005].

Anatomical differences

Table 1 presents descriptive statistics for the brain structure measures by group and gender. The table also includes ANOVA comparisons between the male dyslexic and control subjects. Male dyslexics exhibited a significantly smaller right cerebellar anterior lobe (ANT), right PTR and brain volume compared with male controls. There were too few female dyslexics to compare the female subjects.

View this table:
Table 1

Neuroanatomical means and standard deviations by group and gender

Male controls (n = 19)Male dyslexics (n = 14)Female controls (n = 13)Female dyslexics (n = 4)

Bonferroni corrected post hoc comparisons between male control and dyslexic subjects. *P<0.006, Bonferroni correction; P<0.05. PTA = planum temporale asymmetry; SUMPPAR = planum temporale and parietale asymmetry; LH1 = left primary Heschl’s gyrus (cm2); LH2 = left Heschl’s gyrus duplication (cm2); RANT = right cerebellar anterior lobe (cm3); LPTR = left pars triangularis (cm2); RPTR = right pars triangularis (cm2); CV = cerebral volume (cm3); CVASYM = cerebral volume asymmetry.

MANOVA, controlling for gender, was used to test the prediction that dyslexic brain structures differed from controls. Table 2 presents the results for tests of between‐subjects effects [multivariate test: F(9,48) = 3.76, P < 0.005]. Dyslexic subjects had a smaller right ANT, smaller right and left PTR and a smaller brain volume. Figure 1 presents images of control and dyslexic brains with the ANT and PTR outlined. The significant interaction between gender and group for the left H2 was due to female controls with large left H2 surface areas. There were no differences between groups for measures of the PT or Heschl’s gyrus.

Fig. 1 Comparison of the PTR (45 mm lateral) and right cerebellar anterior lobe (5 mm lateral) between control (A and C) and dyslexic (B and D) images.

View this table:
Table 2

MANOVA results comparing anatomical measurements between reading groups

Corrected modelGroupGenderGroup × gender
CVASYM1.467n.s. 3.950 n.s.0.138n.s.0.066n.s.

*P < 0.05; **P < 0.01; ***P < 0.001; n.s. = not significant. PTA = planum temporale asymmetry; SUMPPAR = planum temporale and parietale asymmetry; LH1 = left primary Heschl’s gyrus (cm2); LH2 = left Heschl’s gyrus duplication (cm2); RANT = right cerebellar anterior lobe (cm3); LPTR = left pars triangularis (cm2); RPTR = right pars triangularis (cm2); CV = cerebral volume (cm3); CVASYM = cerebral volume asymmetry.

Anatomical probability of dyslexia

Hierarchical binary logistic regression was used to determine the odds of group membership on the basis of the anatomical variables that were significantly different between the two groups. The PTR and cerebellar measures were entered in the first level of the regression. Eighty‐eight per cent of the control (28 of 32) and 77% of dyslexic (13 of 17) children were classified correctly in step 1 (–2 log likelihood = 34.8; Cox and Snell R2 = 0.441; Nagelkerke R2 = 0.608). Right ANT, left PTR and right PTR each made a unique contribution in predicting group membership (Table 3). Only one additional control participant was correctly classified when cerebral volume and cerebral asymmetry were added to the regression (–2 log likelihood = 33.8; Cox and Snell R2 = 0.452; Nagelkerke R2 = 0.623). The unique contributions of right ANT, left PTR and right PTR to the prediction of group membership were reduced, however, with this addition. The addition of gender did not unmask any anatomical relation to group membership (–2 log likelihood = 29.1; Cox and Snell R2 = 0.502; Nagelkerke R2 = 0.693). Figure 2 shows the probability of group membership based on the contribution of right ANT, left PTR, right PTR, cerebral volume and cerebral asymmetry.

Fig. 2 The anatomical probability of dyslexic or control group membership compared between the actual dyslexic and control groups. This probability is based on the contribution of the right ANT, left PTR, right PTR, cerebral volume and cerebral asymmetry. The symbols are labelled by gender to demonstrate that anatomical measures (brain volume) predicted both gender and reading group membership.

View this table:
Table 3

Hierarchical binary logistic regression results

CoefficientbetaSEWald χ2d.f.SignificanceExp(B)
Step 1RANT–1.580.577.631**0.21
 Total R2(3,48) = 28.5, P < 0.0001 LPTR–*0.12
Step 2RANT–1.300.634.281*0.27
 Step R2(2,48) = 0.94, n.s.LPTR–
 Total R2(5,48) = 29.4, P < 0.0001RPTR–2.701.165.411*0.07
Step 3RANT–1.250.703.1910.29
 Step R2(1,48) = 4.74, P < 0.05LPTR–2.501.383.2710.08
 Total R2(6,48) = 29.1, P < 0.0001RPTR–2.721.354.031*0.07

P < 0.10; *P < 0.05; **P < 0.01; n.s. = not significant. RANT = right cerebellar anterior lobe; LPTR = left pars triangularis; RPTR = right pars triangularis; CV = cerebral volume; CVASYM = cerebral volume asymmetry.

Anatomical relation to behavioural language measures

The anatomical variables that distinguish dyslexic from control participants were predicted to relate to behavioural measures that characterize dyslexia. Table 4 presents means and standard deviations for the behavioural language measures. Table 5 reports intercorrelations among the behavioural language measures for the entire sample. All of the measures were significantly correlated, except verbal IQ with orthographic coding and naming speed.

View this table:
Table 4

Behavioural language performance for dyslexic and control children

Phoneme elisione11.
Ortho codingf69.621.825.3118.8

aWISC III (Wechsler, 1991); bWord Identification; cWord Attack (Woodcock, 1987); dWRAT III (Wilkinson, 1993); eCTOPP (Wagner et al., 2000); fPAL Expressive Orthographic Coding (Berninger, 2001a); g,hRapid Automatic Naming of Letters and Naming of Alternating/Switching Letters and Numbers (Wolf, 1986).

View this table:
Table 5

Behavioural correlations for dyslexic and control participants

VIQaWIbWAcSPELLdPhoneme ElisioneOrtho CodingfRAN g
Phoneme elision0.345*0.820***0.822***0.785***
Ortho coding0.2590.731***0.784***0.790***0.753***

*P < 0.05; **P<0.01; ***P < 0.001. aWISC III (Wechsler, 1991); bWord Identification, cWord Attack (Woodcock, 1987); dWRAT III (Wilkinson, 1993); eCTOPP (Wagner et al., 2000); fPAL Expressive Orthographic Coding (Berninger, 2001a); g,hRapid Automatic Naming of Letters and Naming of Alternating/Switching Letters and Numbers (Wolf, 1986). VIQ = prorated verbal IQ; WI = Word Identification; WA = Word Attack; SPELL = spelling ability.

Table 6 presents correlations for the behavioural variables with the anatomical measures that distinguished dyslexic from control children. The anatomical variables that differentiate dyslexic from control subjects were consistently significantly correlated with real word reading, pseudoword reading and spelling, the three language skills on which the dyslexic children were reliably different from the controls (Table 4). However, of the four reading‐related language processes (phoneme elision, orthographic coding, RAN and RAS) only RAN was consistently significantly correlated with the anatomical measures that distinguished dyslexic from control participants.

View this table:
Table 6

Anatomical and behavioural correlations for dyslexic and control participants

VIQaWIbWAcSPELLdPhoneme elisioneOrtho codingfRANgRASh

*P < 0.05; **P < 0.01. aWISC III (Psychological Corporation); bWord Identification; cWord Attack (Woodcock, 1987); dWRAT III (Wilkinson, 1993); eCTOPP (Wagner et al., 2000); fPAL Expressive Orthographic Coding (Berninger, 2001a); g,hRapid Automatic Naming of Letters and Naming of Alternating/Switching Letters and Numbers (Wolf, 1986). VIQ = prorated verbal IQ; WI = Word Identification; WA = Word Attack; SPELL = spelling ability.

Elision, orthographic coding and RAN were selected as examples of phonological, orthographic and rapid naming skills to determine if the right ANT, left PTR and right PTR accounted for different sources of variance in these language measures. Three linear multiple regressions were performed for each of the three reading‐related language measures. The results of these analyses are presented in Table 7. Figures 3, 4 and 5 demonstrate these results graphically. To summarize, 20, 20 and 41% of the variance in elision, orthographic coding and RAN, respectively, was explained by the three anatomical measures. The right ANT, right PTR and left PTR predicted the same variance in elision performance. The right ANT uniquely predicted orthographic coding. Although the right cerebellar anterior lobe and brain volume shared some variance in RAN, the right ANT, right PTR and left PTR all uniquely predicted RAN after controlling for brain volume.

Fig. 3 The standardized predictive value of right ANT, left PTR and right PTR exhibits a linear relation with elision performance.

Fig. 4 The standardized predictive value of right ANT, left PTR and right PTR exhibits a linear relation with orthographic coding performance.

Fig. 5 The standardized predictive value of right ANT, left PTR and right PTR exhibits a linear relation with Rapid Automatic Naming of Letters performance.

View this table:
Table 7

Regression results demonstrating the relation between anatomical variables and phonological, orthographic and rapid automatic naming measures

Regression F Total variance explainedStandardized beta T Significance
Phonological (phoneme elisiona)F(3,41) = 3.12, *20%
Orthographic (ortho codingb)F(3,39) = 2.98*20%
Rapid naming (RANc)
Step 1F(3,42) = 9.19***41%
Step 2F(4,42) = 6.79***
R2 change = 0.003, n.s.

n.s. = non‐significant. ∼P < 0.10; *P < 0.05; **P < 0.01; ***P < 0.001. aCTOPP (Wagner et al., 2000); bPAL Expressive Orthographic Coding (Berninger, 2001a); cRapid Automatic Naming of Letters (Wolf, 1986)

Double‐deficit dyslexics

A dyslexic was classified as having a double deficit if performance was more than 1 standard deviation below their pro‐rated verbal IQ on both the elision and the RAN or RAS (Berninger et al., 2001). Sixty‐five per cent of dyslexic children in this study had both phonological awareness and naming speed deficits. Ninety‐four per cent of dyslexics had a naming speed deficit. One dyslexic subject had only a phonological awareness deficit. The small number of dyslexic children with only naming speed or phonological awareness deficits prevented statistical comparisons with the double‐deficit group. Visual inspection of the anatomical data did not reveal any trends that might distinguish the single‐deficit from the double‐deficit group. There were two control subjects with below average naming speed. Anecdotally, the left and right PTRs of these two controls were within the dyslexic distribution for these measures. Their right cerebellar anterior lobe and brain volume were within the normal range, however.


This study shows that measures of the right cerebellar anterior lobe and inferior frontal gyrus distinguished dyslexic children from controls with a high probability. The frontal and cerebellar measures each contributed to classifying a subset of dyslexic participants by differentiating them as a group from controls and by predicted reading skill performance across the sample. Although unexamined brain regions may contribute to the risk of developing dyslexia, these findings suggest that deficits in a frontal‐cerebellar system could lead to symptoms of dyslexia.

Measures of the temporal lobe, including the PT, did not differentiate dyslexic from control participants. This is at least the eighth study demonstrating that individuals with dyslexia do not exhibit reversed asymmetry of the PT (Leonard et al., 1993; Rumsey et al., 1997; Best and Demb, 1999; Heiervang et al., 2000; Robichon et al., 2000a). Planar asymmetry predicts a wide range of verbal ability (Eckert et al., 2001) but does not predict individual variability in phonology when phonology is discrepant from verbal ability (Eckert and Leonard, 2000).

Multivariate approach

A single anatomical marker that perfectly differentiates dyslexic from control participants has not been reported and is unlikely to be found, given the complex nature of the reading process. Multiple neuroanatomical measures were necessary to predict a diagnosis of dyslexia in children. Each measure accounted for overlapping subsets of dyslexic subjects. These findings may help explain why the structural neuroimaging literature has been plagued by failed replications (Eckert and Leonard, 2000).

Genetic findings for dyslexia support this perspective. Careful phenotyping in genetic studies has resulted in positive linkage between dyslexia and measures of phonological processing with genetic markers on chromosomes 1, 2, 3, 6, 15 and 18 (Cardon et al., 1994; Grigorenkoet al., 1997, 2000, 2001; Schulte‐Korne et al., 1998; Fagerheim et al., 1999; Fisher et al., 1999, 2002; Gayan et al., 1999; Nothen et al., 1999; Nopola‐Hemmi et al., 2000, 2001; Petryshen et al., 2001). One gene or a combination of genes within these marker regions could have developmental effects on a variety of neural reading systems, leading to a common behavioural outcome but heterogeneous neurobiological findings.

Comparison with dyslexic college students

In contrast to a previous study of college‐educated dyslexic adults (Leonard et al., 2001), the dyslexic children in this study had smaller brain volumes than controls and an absence of duplicated left Heschl’s gyrus or extreme leftward asymmetry of the PT and planum parietale. The dyslexic college students were defined as having phonological decoding discrepant from their passage comprehension, in contrast to the dyslexic children, who were characterized as having phonological decoding deficits that were discrepant from their verbal IQ. In general, the dyslexic college students had superior passage comprehension and rapid naming performance compared with the dyslexic children in this study. Failure to replicate the duplicated Heschl’s gyrus and exaggerated leftward planar asymmetry findings from the study of Leonard and colleagues (Leonard et al., 2001) could be due to these cognitive differences between the samples. Alternatively, the passage comprehension and rapid naming differences could disappear with age as the dyslexic children learn to compensate for their reading problems. In this case, failure to replicate could be attributable to the post hoc nature of the study of Leonard and colleagues (Leonard et al., 2001).

Age is a possible, but unlikely, explanation for some of the differences between studies. The gyral and sulcal measures are unlikely to exhibit dramatic changes with age. In contrast, a longitudinal study of the dyslexic children might demonstrate protracted increases in brain volume during development.

The one anatomical measure that was comparable in the child and adult dyslexics was that of the right cerebellar anterior lobe (Leonard et al., 2001). This effect may have been due to decreased grey matter rather than white matter. Rae and colleagues (Rae et al., 2002) reported that dyslexic adults failed to exhibit rightward whole cerebellum grey matter asymmetry, in part as a result of reduced volume of right cerebellar grey matter. There was no difference between control and dyslexic subjects for whole cerebellum white matter measures.

The results for cerebellum involvement in reading are consistent with an emerging research literature that shows that the cerebellum activates during learning and not just motor function—the circuits of the cerebellum that are activated during the learning phase differ from those activated during the automatic phase following practice and learning (Nicolson et al., 1999; Poldrack and Gabrielli, 2001). The study by Nicolson and colleagues (Nicolson et al., 1999) showed that dyslexic participants demonstrated less PET activation in the right cerebellar anterior lobe when they performed learned finger movement sequences. The reduced activation volume could be related to reduced anterior lobe size. These findings are consistent with Fawcett and Nicolson’s cerebellar deficit hypothesis, which attributes the cognitive and motor problems exhibited by individuals with dyslexia to abnormal cerebellar development (Nicolson and Fawcett, 1990; Nicolson et al., 2001).

Studies of children with cerebellar tumours demonstrate a critical role for the right cerebellar hemisphere in linguistic performance (Riva and Giorgi, 2000; Scott et al., 2001). For example, small samples of children with right cerebellar tumours exhibited poor verbal and literacy performance, in contrast to children with left cerebellar tumours and spatial deficits (Scott et al., 2001). Akshoomoff and colleagues (Akshoomoff et al., 1992) reported poor naming performance in a child with a right cerebellar hemisphere tumour.

The results are also consistent with the proposal that the cerebellum is affected indirectly in dyslexic adults and children (Ivry and Justus, 2001; Zeffiro and Eden, 2001; Bishop, 2003). The lateral prefrontal cortex, superior temporal sulcus and posterior parietal association cortices project to the cerebellum via pontine nuclei in macaque monkeys (Levin, 1936; Schmahmann, 1997; Schmahmann and Pandya, 1997). The cerebellum also projects to these association cortices through the thalamus (Middleton and Strick, 1997). Cerebellar or frontal anomalies could produce downstream effects on the architecture and function of either region. The dyslexics in this study had problems in learning to read but not in the motor skills associated with learning to read: oral–motor for mouth movements during oral reading and graphomotor for hand movements during spelling. They did have problems in the rate of reading single words and text, which is consistent with the role of the cerebellum in precise timing.

The left and right PTR were smaller in dyslexic than in control children. This finding is consistent with previous research identifying bilateral area and/or volume differences in the inferior frontal gyri of dyslexic and control adults (Robichon et al., 2000a; Brown et al., 2001). These findings support the idea that early frontal lobe dysfunction may play a role in the development of dyslexia. The findings may not be specific to dyslexia, however. Gauger and colleagues (Gauger et al., 1997) reported smaller PTR measurements in children with specific language impairment.

Frontal lobe dysfunction is the centrepiece of the motor‐articulatory feedback hypothesis for dyslexia (Heilman et al., 1996). This hypothesis maintains that an inability to associate the position of articulators with speech sounds is due to a lack of awareness regarding the position of articulators during speech. Heilman suggests that left inferior frontal lobe dysfunction could explain dyslexic symptoms related to motor feedback. This argument is supported by evidence showing that (i) patients with left anterior perisylvian lesions are not capable of making grapheme‐to‐phoneme conversions (Adair et al., 1999), (ii) normal participants exhibit activation of the inferior frontal gyrus during grapheme‐to‐phoneme conversion (Newman and Twieg, 2001; Fiebach et al., 2002), and (iii) this region is abnormally activated in dyslexic subjects (Shaywitz et al., 1998; Georgiewa et al., 2002).

The anatomical findings suggest that impairments in a frontal‐cerebellar network may play a role in delayed reading development in dyslexia. Functional imaging studies show activation in the left inferior frontal and right cerebellar hemisphere during fluency tasks (Schlosser et al., 1998), passive listening to clicks (Ackermann et al., 2001), linguistic working memory tasks (Desmond et al., 1997) and rapid production of consonant–vowel stimuli (Wildgruber et al., 2001). Interestingly, a PET study of aphasic patients with left inferior frontal damage found hypoactivation of the right cerebellar hemisphere (Metter et al., 1987). This network may be important for temporal processing of verbal utterances or more generally for error correction operations. Considering the high percentage of dyslexic children in this sample with the double deficit in rapid automatic naming and phonological awareness (Wolf and Bowers, 1999), this frontal‐cerebellar network may be critical to the precise timing mechanism that Wolf and Bowers (1999) hypothesize to underlie the double deficit. Both the orthographic coding and rapid naming tasks, which were predicted by the cerebellar measure, have time constraints: the orthographic task presents words for a brief moment (1 s), during which they must be encoded into short‐term memory, and the rapid naming task records time for naming highly familiar alphanumeric symbols.

Dyslexic subtypes could be produced by processing deficits anywhere in the frontal‐cerebellar phonological system. The unique contribution of each frontal and cerebellar measure to the classification of dyslexic participants and the prediction of phonological and naming performance supports this view. Each region may play a distinct, but related, role in reading that is responsive to instructional intervention.

Future directions

A twin study of reading‐disabled children reported that orthographic coding and phoneme awareness have unique genetic foundations (Gayan and Olson, 2001). This finding suggests that multiple neurobiological pathways for dyslexia could be the consequence of having unique genetic beginnings for orthography and phonology. Anomalous development of a left posterior cerebral‐cerebellar network may explain the orthographic deficits. Area V5/MT and the fusiform gyrus process motion (Salzman et al., 1990) and visual word stimuli (Cohen et al., 2000), respectively, and each has been implicated in dyslexia (Eden et al., 1996; Demb et al., 1997; Brunswick et al., 1999; Shaywitz et al., 2002). Future studies will examine whether these brain regions constitute a separate structural network that accounts for the orthographic deficits in dyslexia or whether a common structural network accounts for both orthographic and phonological deficits. Future work will also examine whether functional activation for phonological and orthographic tasks map onto posterior cerebral‐cerebellar and frontal‐cerebellar structural network(s) for orthographic and phonological processing.


We wish to thank the children and families who participated in this study and the NICHD (National Institute of Child Health and Human Development) for its support (P50 38812–06).


View Abstract