Dysfunction of the auditory thalamus indevelopmental dyslexiaBegoña Díaza,1, Florian Hintza, Stefan J. Kiebelb, and Katharina von Kriegsteina

aMax Planck Research Group Neural Mechanisms of Human Communication and bDepartment of Neurology, Max Planck Institute for Human Cognitiveand Brain Sciences, 04103 Leipzig, Germany

Edited by Michael Merzenich, W. M. Keck Center for Integrative Neuroscience, San Francisco, CA, and approved July 16, 2012 (received for reviewDecember 5, 2011)

Developmental dyslexia, a severe and persistent reading andspelling impairment, is characterized by difficulties in processingspeech sounds (i.e., phonemes). Here, we test the hypothesis thatthese phonological difficulties are associated with a dysfunction ofthe auditory sensory thalamus, the medial geniculate body (MGB).By using functional MRI, we found that, in dyslexic adults, theMGB responded abnormally when the task required attending tophonemes compared with other speech features. No other struc-ture in the auditory pathway showed distinct functional neuralpatterns between the two tasks for dyslexic and control partic-ipants. Furthermore, MGB activity correlated with dyslexia di-agnostic scores, indicating that the task modulation of the MGB iscritical for performance in dyslexics. These results suggest thatdeficits in dyslexia are associated with a failure of the neuralmechanism that dynamically tunes MGB according to predictionsfrom cortical areas to optimize speech processing. This view ontask-related MGB dysfunction in dyslexics has the potential toreconcile influential theories of dyslexia within a predictive codingframework of brain function.

functional MRI | speech recognition | auditory processing | magnocellular

Developmental dyslexia, with a prevalence of 5% to 10% inchildren, is the most common learning disability (1). In

children with normal intelligence and educational opportunities,dyslexia manifests as an inability to reach normal reading andspelling skills that persists throughout the lifespan (1, 2). Thesereading and spelling difficulties not only limit education success(2), but also frequently lead to social and emotional problems(3). However, dyslexic persons can learn to compensate for theirdifficulties and successfully graduate from college (4).There are various theories about the underlying cause of dys-

lexia. One of the most influential is the phonological deficit hy-pothesis, which posits that dyslexic persons have difficulties inprocessing and manipulating small units of speech, i.e., speechsounds or phonemes (1, 5, 6). This theory is based on findings thatdyslexic persons score below controls in tasks that require them toprocess and manipulate speech sounds, for instance, to judgewhether two words rhyme (as in /hat/ and /cat/) (7) or to repeat twowords while transposing the initial speech sounds (e.g., /green/and /ball/; correct response, /breen/ and /gall/) (8). Recently, pro-ponents of the phonological theory have emphasized that dyslexicpersons’ phonological deficits are apparent only if the phonologicaltasks impose cognitive demands, such as fast responses, highmemoryload, or noisy listening conditions (9). The relevance of task-induceddifficulties is also the hallmark of the sluggish attention shifting(SAS) hypothesis. It proposes that dyslexics are impaired inshifting attention in tasks that have a fast rate of stimulus pre-sentation (10). An important assumption of the phonologicaldeficit hypothesis is that the deficits are caused at the level of thecortex and are specific to phonological phenomena (1, 5, 6). Incontrast, the magnocellular theory, another influential view, positsthat abnormal subcortical function is the origin of poor phono-logical processing in dyslexic persons (11, 12). This theory is largelybased on findings that the left medial geniculate body (MGB) is

altered in postmortem brains of dyslexics (13). The MGB is thefirst-order auditory thalamus, the last station in the sensorypathways before peripheral auditory information reaches thecortex (14). Consequently, the magnocellular theory assumes thatsensory deficits are at the basis of dyslexia and predicts that dys-lexics’ deficits are not specific to linguistic processing, but thatthey extend to sensory processing in general (11, 12, 15).Here, we attempt to reconcile these two opposing views by

testing the hypothesis that the phonological deficit in dyslexia is(i) associated with a dysfunction of the MGB, as claimed by themagnocellular theory; but (ii) that the expression of this dys-function can be evoked by phonological tasks, as predicted by thephonological deficit and SAS hypothesis. There are three keyexperimental findings that support this combined hypothesis, and,in addition, shed light on the role of the MGB in speech pro-cessing. First, the MGB engages in speech processing in healthyparticipants, preferentially in processing phonologically relevant,fast spectrotemporal changes in speech compared with moreslowly occurring features, such as voice characteristics (16). Sec-ond, the left MGB has been found to be altered in postmortembrains of dyslexic persons (13), leading to the speculation thatdyslexia can be attributed to MGB function (11, 12). Third, re-cent studies emphasize the fact that dyslexics’ difficulties seem toresult from not using previous or contextual information to fine-tune sensory perception for optimal performance (17, 18).These findings fit well with the recent reconceptualization ofsensory thalamus function: although sensory thalamic structureswere previously considered as simple relay stations, they are nowdescribed as smart gatekeepers because they are tuned by cor-tical areas to the relevant properties of sensory input (reviewedin refs. 19, 20).To test the anatomical–functional hypothesis that the phono-

logical deficit in dyslexia is associated with MGB dysfunction, weused a functional MRI (fMRI) protocol optimized for sub-cortical, auditory structures (21), and an experimental designthat reliably evokes MGB responses specific to processing pho-nologically relevant, fast changes in speech in healthy subjects(16). We hypothesized that (i) dyslexics have abnormal MGBresponses when processing fast spectrotemporal changes inspeech, and (ii) the strength of this dysfunction is related to theseverity of the phonological deficit in dyslexics.Dyslexic adults (n = 14) and matched control participants

(n = 14; Table S1) performed two tasks while blood oxygen-leveldependent (BOLD) activity was measured (Fig. 1A). In the twotasks, participants attended to speech sound changes (i.e., pho-nological task) or to changes in voice characteristics (i.e., speaker

Author contributions: B.D. and K.v.K. designed research; B.D. and F.H. performed re-search; B.D., F.H., S.J.K., and K.v.K. analyzed data; and B.D., S.J.K., and K.v.K. wrotethe paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.1To whom correspondence should be addressed. E-mail: diaz@cbs.mpg.de.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1119828109/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1119828109 PNAS | August 21, 2012 | vol. 109 | no. 34 | 13841–13846

NEU

ROSC

IENCE

task), within sequences of six syllables (Fig. 1B). Participantsreported via button press whether each syllable within a sequencediffered in content from the previous one (i.e., phonological task)or was spoken by a different voice (i.e., speaker task). Critically,the phonological task required processing of fast spectrotemporalchanges, whereas the speaker task required processing of acousticfeatures on a slower time scale (Methods). Both tasks were per-formed with the same stimulus material, i.e., a difference in acti-vation between tasks cannot be attributed to acoustic differencesin the stimuli. The matched controls were expected to show anincrease in the BOLD signal in the MGB when performing thephonological task, compared with the speaker task (16). In con-trast, according to the present hypothesis, such a task-dependentmodulation in the MGB should be smaller or nonexistent in thegroup of dyslexic participants. In a second experiment, partic-ipants passively listened to sentences (Fig. 1C). This experimentserved to functionally localize auditory structures and to in-vestigate MGB response in the two groups independent of task.

ResultsAbnormal MGB Response in Dyslexic Subjects. As hypothesized, inthe left MGB, we found a larger BOLD signal in control thandyslexic participants for the phonological task compared with thespeaker task [group–task interaction, P = 0.0338, family-wiseerror (FWE) corrected for left MGB; Fig. 2 A and B]. In thecontrol group, there was significantly greater BOLD activity inthe phonological than the speaker task in the right MGB[“phonological − speaker task,” P = 0.0336, FWE corrected;Montreal Neurological Institute (MNI) coordinates, 9, −28, −8]and a trend toward significance in the left MGB (P = 0.067,FWE corrected; MNI coordinates, −9, −28, −8), whereas no sig-nificant differences in MGB activity between the tasks was present

in dyslexics (Fig. S1 and Table S2). To clarify whether there is noMGB modulation in dyslexic participants or, rather, its modu-lation was in the opposite direction relative to controls, we alsotested the “speaker − phonological task” contrast in dyslexics. Inneither of the two MGBs did dyslexic participants show signifi-cantly greater BOLD activity for the speaker task relative to thephonological task, even at a lenient statistical threshold (P =0.01, uncorrected).The group–task interaction in left MGB was location-specific.

To test this, we investigated regions of interest (ROIs) in theentire auditory subcortical pathway, the auditory cortices (ACs),auditory association cortex, and cortical regions that are oftenreported as dysfunctional in dyslexics (reviewed in ref. 22) (Figs.S2 and S3). We found that the group–task interaction was sig-nificant only in left MGB and not in any other regions, even at alenient statistical threshold (P = 0.01 uncorrected within ROI orP = 0.001 uncorrected at whole-brain level; Table S3).The difference between dyslexic and control participants in

left MGB activation cannot be explained by behavioral differ-ences. Control participants’ accuracy rates were 91.82% (SD,1.68%) and 92.88% (SD, 1.79%) and dyslexic participants’ ac-curacy rates were 88.22% (SD, 1.68%) and 88.04% (SD, 1.79%)for the phonological and speaker tasks, respectively. The per-formance in the phonological and speaker tasks was matched[main effect of task: F(1,26) = 0.24, P = 0.62], and there was nobehavioral group–task interaction [F(1,26) = 0.48, P = 0.49].Dyslexic participants showed a trend toward being worse thancontrol subjects at both tasks [main effect of group: F(1,26) =3.37, P = 0.078].

Cortical Compensatory Mechanisms in Dyslexic Participants. To testwhether dyslexics achieved high levels of performance in thephonological task by compensatory mechanisms, we analyzed

Fig. 1. Experimental design. (A) In experiment 1, stimuli were presented in blocks (blue and orange boxes) of six auditory syllables (syllable onsets arerepresented by black lines within colored boxes). Participants were asked to perform the phonological or the speaker task via a visual instruction screen (blackrectangles). After each block, one brain volume was acquired (gray boxes). (B) Example of a block in experiment 1. A task instruction screen was presented atthe beginning of the block. According to the instruction participants performed the phonological task or speaker task. Depending on the task, a voice change(blue arrow) or a syllable change (orange arrow) was the target. Word balloons represent auditory stimulus examples. The same stimuli were used for the twotasks. (C) In experiment 2, sentence blocks (yellow boxes) and silence blocks (white boxes) were presented. Sentence blocks contained four auditory sentences.Participants were asked to press a button at the end of each sentence via a visual task instruction screen (black rectangles). After each block, one brain volumewas acquired (gray boxes).

13842 | www.pnas.org/cgi/doi/10.1073/pnas.1119828109 Díaz et al.

whether there is an indication for such mechanism in right in-ferior frontal gyrus (rIFG), a cortical area that is thought tocompensate for dyslexic participants’ difficulties (23). For thesimple main effects of the phonological task (“dyslexics/phono-logical task − controls/phonological task”) and speaker task(“dyslexics/speaker task − controls/speaker task”), as well as forthe main effect group [“(dyslexics/phonological task + dyslexics/speaker task) − (controls/phonological task + controls/speakertask)”], significantly greater BOLD signal was found in the rIFG(Table S4). An exploratory analysis in several ROIs (Methods)revealed greater BOLD signal for the more specific interactioncontrast “(dyslexics/phonological task − dyslexics/speaker task) −(controls/phonological task − controls/speaker task)” in the rightcochlear nucleus (CN; P = 0.0347, FWE corrected for ROI) andthe right inferior colliculus (IC; P = 0.0291, FWE corrected forROI). However, the significance levels in these subcorticalstructures did not survive correction for testing multiple ROIs,for which we did not have a prior hypothesis. There were nofurther effects in any other subcortical or cortical ROIs, nor anyeffects at the whole-brain level. There were also no corticalregions that were less active in dyslexics than controls in thesimple main effects of task or the group main effect, even ata lenient statistical threshold (P = 0.01 uncorrected within ROIor P = 0.001 uncorrected at whole-brain level).

MGB Dysfunction in Dyslexia Is Task-Dependent. MGB activity indyslexic participants was not generally reduced during speechperception relative to controls. There was no effect of group inleft or right MGB in experiment 1 for either task (“controls/phonological task − dyslexics/phonological task” and “con-trols/speaker task − dyslexics/speaker task”; Table S3). Fur-thermore, in experiment 2, MGB responses to passive listening

to speech (“sentences – silence”) were similar in the two groupsand there was no significant difference between groups in the leftor right MGB (Table S5).

MGB Activity Is Correlated with Dyslexia Diagnostic Scores.Does thetask-dependent modulation of the MGB relate to the phono-logical deficits in dyslexia? We addressed this question by cor-relating the difference of BOLD responses between the tasks(i.e., “phonological task − speaker task”) in each MGB withsubjects’ scores in diagnostic tests. Only in dyslexics did theamount of modulation of the left MGB correlate positively withreading comprehension (P = 0.024, FWE corrected; MNIcoordinates, −15, −31, −11) and negatively with the time neededto name letters and numbers aloud, a measure of phonologicalaccess (P = 0.007, FWE corrected; MNI coordinates, −15, −31,−8; Fig. 2C). These significant correlations show that MGBmodulation during speech processing is related to literacy andphonological deficits of dyslexic participants. In controls, signifi-cant correlations with dyslexia diagnostic scores were present incortical regions that have previously been reported to be associ-ated with reading abilities (Table S6).

DiscussionThe results show that task-dependent modulation of the leftMGB is different in controls and dyslexics. Although there isa positive modulation by the phonological task in contrast to thespeaker task in control participants, this is not the case in dyslexicparticipants. In addition, we found, in dyslexics only, that themodulation amplitude in the left MGB and the dyslexia diag-nostic scores are correlated. The difference in modulation be-tween dyslexic and control participants as well as the correlationin dyslexics are location-specific; they are significant only in the left

Fig. 2. Differences between normal readers and dyslexics in MGB activity during speech processing. (A) The between-group comparison reveals greater task-dependent modulation in the left MGB for control vs dyslexic participants (P = 0.0338, FWE corrected for MGB; local activation maxima, −9, −28, −8, MNIcoordinates). The color bar represents t values. (B) Estimated signal changes (as percentages) extracted from the left MGB for the contrast “phonological task− speaker task” for controls and dyslexics. Error bars represent SE. (C) Task-dependent modulation of the left MGB correlates positively with readingcomprehension and negatively with time required for phonological access (i.e., RAN) in dyslexic but not control participants.

Díaz et al. PNAS | August 21, 2012 | vol. 109 | no. 34 | 13843

NEU

ROSC

IENCE

MGB but not in any other structure of the auditory pathway orcortices. These findings suggest that a dysfunction of the neuralmechanism that modulates MGB processing plays a key role inphonological processing skills in dyslexia. In contrast, in controlparticipants the MGB function does not seem to be the limitingfactor for reading skills, but specific cortical areas seem to be in-volved in determining the behavioral performance.Almost two decades ago, dyslexia was suggested to be caused

by an abnormal thalamic function at the level of the sensorythalami (11, 12). The magnocellular layers of the visual thala-mus, the lateral geniculate nuclei, and the large cells in the leftMGB were found to be altered in postmortem brains of dyslexicpersons compared with normal brains (13, 15), which led to theterm “magnocellular theory.” In monkeys, damage to magno-cellular layers of the lateral geniculate nuclei impairs motion andflicker detection (24). Similar behavioral deficits have been reportedfor dyslexics (25, 26). Consequently, the magnocellular theoryposits that dyslexics’ reading and speech perception difficultiesare caused by an impairment in sensory processing of fast tran-sient stimuli (11, 12). The theory predicts a general processingdeficit in the perception of rapidly varying visual and auditorystimuli in dyslexic participants that are not specific to speech (11,12, 15). In accordance, several studies have shown that dyslexicsare less sensitive than controls in detecting changes in auditorytransient stimuli, e.g., amplitude and frequency modulation (27,28), and that this sensitivity correlates with nonword readingcapabilities (27). A general auditory deficit for fast, transientsounds is also a key feature of the “temporal auditory processing”theory (29, 30), which posits that dyslexia is caused by an auditorysensory deficit in perceiving fast sequences of brief stimuli.However, the causal relation between sensory deficits and dys-

lexia is highly controversial because several studies have found that(i) not all dyslexic persons have sensory deficits and (ii) dyslexics’sensory impairments are not confined to tasks that require rapidsensory processing (31–33). The lack of a consistent sensory deficitassociated with dyslexia has been taken to support the phonologicaldeficit hypothesis, i.e., the view that dyslexia is caused by a deficitthat is specific to phoneme processing (1, 5, 6). An assumption ofthe phonological deficit hypothesis is that the underlying cause is ata cortical rather than at a subcortical level, as claimed by themagnocellular theory (5). This cortical argument is strengthenedfurther by findings that phonological deficits in dyslexia are ap-parent if the tasks or the context impose difficult cognitivedemands, such as high memory load, limited response time, ornoisy listening conditions (9, 34, 35).Here, we propose that key features of both these views, the

magnocellular theory and the phonological deficit hypothesis, canbe usefully reconciled by readjusting the standard view of the roleof the MGB in speech processing. Sensory thalamic structuresreceive massive cortical backward connections, which are believedto play an important role in the processing of sensory information(14). Currently, there are two views, which are not necessarily inopposition, on the functional role of these corticothalamic con-nections. One view is that the sensory thalamus is modulated byefferent cortical connections depending on attentional demands toregulate the amount of sensory information that is forwarded tocortical areas (36, 37). Another view is that sensory thalamic struc-tures serve the function of dynamic gatekeepers, which are dy-namically tuned by specific feedback from cortical areas to optimizedemanding sensory processing (16, 38, 39). In this view, thalamicsensory areas could be considered to serve cognitive functions byoptimizing sensory processes accordingly to cognitive demands.Our findings are in line with the concept of sensory thalamic

structures as dynamic gatekeepers. We assume that cortical ef-ferent connections modulate MGB response properties by pro-viding specific dynamic predictions about the sensory input. Wepropose, as postulated in the magnocellular theory, that one roleof the MGB is to process fast transient changes in the auditory

input. However, in addition, we suggest that this processing ismodulated by cognitive requirements and the predictability ofthe stimulus material. This would mean that the MGB is an in-tegral and important part of the speech processing hierarchy,dealing with fast, predictable transients in speech signals. Thisadaptive role of the MGB is in line with theoretical accounts andcomputational hierarchical models of brain function that pro-pose that the brain uses internal dynamic models of its envi-ronment to predict the trajectory of the sensory input (40, 41). Inparticular, computational accounts of speech recognition suggestthat, for optimal recognition, lower levels of the auditory hier-archy (e.g., the MGB) must process faster dynamics than higherlevels (e.g., ACs) (42, 43). This computational scheme of rec-ognition posits that cortical predictions should optimize auditoryprocessing at early, subcortical stages of the auditory processinghierarchy, especially for fast, complex, and highly predictablestimuli such as speech sounds. Building up accurate predictionsabout fast sensory dynamics would result in higher accuracy andprocessing speed when the perceptual system is confronted withtaxing conditions, such as fast stimulus presentation rates or highmemory load, i.e., precisely the situations in which dyslexicparticipants perform consistently worse than control participants(8, 29, 44). Because several cognitive operations may requireprecisely tuned sensory processing, a failure in dyslexia to tuneauditory processing to relevant spectrotemporal properties ofspeech sounds might explain not only poor phonological pro-cessing (1, 5, 6), but also several other symptoms of dyslexia suchas the failure to improve the discrimination of speech and non-speech sounds with repetition (17, 18), difficulties understandingspeech in noisy environments (18, 34, 45), and attentional limi-tations in processing fast sequences of stimuli (10). This accountwould also be in line with key features of two further theoriesabout dyslexia, i.e., the SAS hypothesis and the anchoring deficittheory (10, 17). The SAS hypothesis is based on findings thatdyslexics have difficulties with shifting attention which is for ex-ample required for detecting a stimulus that is presented shortlyafter another stimulus (10, 46). In congruence with this hy-pothesis, we found that the MGB shows an abnormal response indyslexic participants when attending to fast, compared with slow,spectrotemporal changes in speech. The anchoring-deficit theory(17) is based on the finding that dyslexics, contrary to controlparticipants, do not benefit from stimulus repetition in same-different judgment tasks on tones (see also ref. 18). This theoryposits that dyslexics’ perceptual system is deficient in predictingincoming sensory input. This fits the present findings well,whereby performance in the phonological task could be optimizedbased on exploiting predictable information in speech stimuli. Wespeculate that control participants are able to use fine-tuning ofthe MGB to optimize auditory processing by prediction, whereasthis mechanism is dysfunctional in dyslexic participants.A deficient tuning of sensory processing in dyslexia can also

account for the difficulties that dyslexics have in understandingspeech in noisy environments, for example with perceiving subtlebut distinctive features of speech sounds as in “bad” and “dad” innoisy environments (18, 34, 45). In this case, cortical predictionscould shape MGB responses to the subtle and transient butpredictable spectrotemporal cues that enable the categoricalrecognition of speech sounds. This tuning would enhance therobustness of speech perception, and a dysfunction of thistuning may explain the susceptibility of dyslexics’ speech per-ception to noise.Given the present fMRI findings, it is not clear whether the

MGB dysfunction is caused at the level of the MGB or/and basedon a deficient mechanism for tuning the MGB by cortical areas.Although the structural alteration finding in the left MGB (13)and the present functional findings speak for a key role of MGBin the underlying neural mechanism, we cannot exclude thepossibility that cortical function, or a dysfunctional connectivity

13844 | www.pnas.org/cgi/doi/10.1073/pnas.1119828109 Díaz et al.

between cortical areas and MGB, take a causal role in the MGBdysfunction. Note that we did not find significant differencesbetween groups in the task-dependent modulation [i.e., “(con-trols/phonological task − controls/speaker task) − (dyslexics/pho-nological task − dyslexics/speaker task)”] in cortical ROIs.However, it may be that the modulation of cortical areas is weak ordistributed and cannot be detected by the present analysis.Previous neuroimaging studies on reading and phonological

tasks have reported, in dyslexic participants relative to controls,higher activity in IFG (47, 48) and less activation mostly in theleft parietotemporal, left occipitotemporal, and left frontal areas(49–51), although this is not the case in all studies (e.g., refs. 52,53). Although we did not find less activity in dyslexic participantsin a cortical area, there was higher activity in dyslexics comparedwith controls in the right IFG, supporting the view that this regionis part of a compensatory mechanism in dyslexia (23, 47, 48). Suchcompensation might explain the absence of significant differencesbetween dyslexic and control participants in task performance.The right IFG did not play a specific role for phonological pro-cessing in the present study: the activation difference betweengroups was similar in the two tasks. A specific higher activation forthe phonological task in dyslexic vs. control participants was onlyfound in right-hemispheric subcortical structures (i.e., right IC,right CN), which may point to their compensatory role in thepresence of a dysfunctional modulation of the left MGB in pho-nological tasks. However, as these regions showed only a trend tosignificance, this interpretation remains speculative.Dyslexic participants performed the phonological task at the

same accuracy level as the speaker task and also similarly to thelevel of control participants. This means that the BOLD differ-ences for the task–group interaction cannot be attributed tobehavioral differences. We assume that the phonological task weused during fMRI scanning was not taxing enough to revealsignificant behavioral differences given a potential compensatorymechanism used by the dyslexic participants.The concept of an impaired cognitive mechanism, which

optimizes sensory processes, is compatible with recent sugges-tions that it is not the phonological representations per se thatare deficient in dyslexia, but rather that phonological deficits areapparent only under certain task or attentional demands (9, 10, 17).We speculate that, in control participants, these cognitive demandsdrive the fine tuning of the representations in the MGB andthereby optimize phoneme perception in a task-dependent fashion.We assume that this cortical top-down influence is reflected in themodulation of theMGB in controls. In contrast, the relative lack ofmodulation of the MGB indicates a task-specific dysfunction indyslexic participants. The correlation between MGB modulationand dyslexia diagnostic scores suggests that there is residual normalMGB function in some dyslexic participants, which can be used toprovide a certain degree of task-specific fine tuning.In summary, we suggest that (i) several auditory difficulties in

dyslexia may be related to a task- or context-dependent impair-ment in fine-tuning early auditory processing according to cor-tical predictions, and (ii) key features of theoretical accounts ofdyslexia can be usefully combined by readjusting the standardview of the role of the MGB in speech processing.

MethodsParticipants. Two groups of adult German speakers participated in the study;one group included 14 participants with developmental dyslexia and theother included 14 controls (SI Methods and Table S1). Groups were matchedin age, sex, handedness, educational level, and nonverbal IQ. Dyslexic par-ticipants scored lower in literacy tests compared with control participants(spelling, reading speed, and text comprehension; SI Methods and Table S1).Skills of phonological access were measured by testing rapid automatizednaming (RAN) (44) of letters, numbers, and objects. The time required toname letters and numbers predicts reading ability and is slower in dyslexicscompared with normal readers, whereas the time to name objects is not areliable predictor of reading ability in adults (35). In our sample, dyslexic

participants were slower than controls in the RAN task for letters andnumbers but not for objects (Table S1).

Stimuli and Experimental Design. Experiment 1. The experimental design wasadapted from von Kriegstein et al. (16). We used 150 different vowel-con-sonant-vowel syllables recorded by one male speaker from a database (54).Each syllable was resynthesized by using the software STRAIGHT (55) tosimulate nine speakers by manipulating the acoustic effect of the vocal tractlength and the glottal pulse rate. This manipulation ensures that the speakertask could be performed based only on slow-varying acoustic features of thevoice (SI Methods). Vocal tract length and glottal pulse rate are based on thevocal tract and glottal fold anatomy of the speaker (56). In contrast,naturally recorded voices might differ in these relatively time-constant vocalfeatures, but also in speaking habits expressed in fast-varying spectrotemporalacoustic features, (e.g., a lisping /s/ or rolling /r/) (57). For the experiment,syllables were presented in sequences of six (Fig. 1 A and B). All sequenceslasted 8.4 s (1,100 ms per syllable and 300-ms pause between syllables).Syllables were pseudorandomly presented within the sequence (SI Methods).In the scanner, participants were asked to perform two tasks: they respon-ded via button press whenever a syllable within a sequence changed inthe content (i.e., phonological task) or in the voice (i.e., speaker task) fromthe previous syllable. The two tasks were presented in random order and thesame 80 sequences were used for the two tasks. Before fMRI scanning,participants performed a brief training session of eight trials.Experiment 2.A total of 40five-word sentences from one speaker were used (SIMethods). During fMRI scanning, 40 sequences of four randomly selectedsentences were presented. Sequences lasted 8.4 s (1,800 ms per sentence and400-ms pause between sentences). Participants were asked to report the endof each sentence via button press. In addition, there were 40 silent periodsof 8.4 s duration. Auditory stimulation and silent periods were presented inrandom order (Fig. 1C).

Data Acquisition. MRI was performed on a 3-T MAGNETOM Trio Tim (Sie-mens). To optimally image subcortical sensory structures, we used a sparseimaging protocol with cardiac gating (21). This allows presentation of theauditory stimuli without the noise of the scanner gradients, and, in ad-dition, reduces the artifacts caused by the pulsatile motion of the brain-stem (SI Methods).

Data Analysis. Behavioral data were analyzed with PASW Statistics 18.0 byusing a repeated-measures ANOVA on percentage of correct responses withtask (i.e., phonological and speaker) as a within-subjects factor, and group(i.e., control and dyslexic) as a between-subjects factor. MRI data were an-alyzed with SPM8 software (Wellcome Trust Centre for Neuroimaging) withstandard procedures (SI Methods).Definition of ROIs. Structures in the auditory pathway—i.e., the AC, MGBs, ICs,and CN—were functionally mapped by contrasting “sentences − silence”(experiment 2; SI Methods, Fig. S2, and Table S5). ROIs for auditory areason Heschl gyrus (i.e., Te1.0, Te1.1, and Te1.2) (58) were created basedon the probabilistic maps provided by the SPM Anatomy Toolbox (http://www.fz-juelich.de/inm/inm-1/EN/Forschung/_docs/SPMAnatomyToolbox/SPMAnatomyToolbox_node.html). ROIs for cortices that have been pre-viously implicated in dyslexia [planum temporale (PT), left posterior tem-poral regions, left parietal regions, lIFG and rIFG, left middle temporalgyrus, left inferior temporal gyrus, and right postcentral gyrus] werebased on coordinates taken from the literature (22, 23), or, if available,on probabilistic maps, i.e., PT (59) (SI Methods and Fig. S3).Categorical analysis. In experiment 1, the contrast of interest was the inter-action between task and group: “(controls/phonological task − controls/speaker task) − (dyslexics/phonological task − dyslexics/speaker task).” Fortesting compensatory mechanisms, we also investigated the reverse in-teraction “(dyslexics/phonological task − dyslexics/speaker task) − (controls/phonological task − controls/speaker task),” as well as simple main effectsand main effects of group. In experiment 2, the contrast of interest was“(controls/sentences − controls/silence) − (dyslexics/sentences − dyslexics/si-lence).” Group analyses were performed by using a one-sample t test acrossthe contrast images of all participants (i.e., random-effects analyses). Groupdifferences were computed by comparing single-subject contrast imagesbetween the groups by means of a two-sample t test (“controls − dyslexics”).Correlation analysis. For both groups separately, statistical parametric mapswere generated with the behavioral scores on (i) nonverbal IQ, (ii) literacytests (spelling, reading comprehension, and reading speed), (iii) word/non-word reading measures (time and errors), and (iv) times on RAN tasks fornumber, letters, and objects as covariates (SI Methods).

Díaz et al. PNAS | August 21, 2012 | vol. 109 | no. 34 | 13845

NEU

ROSC

IENCE

Statistical thresholds. For categorical and correlation analyses, activity differenceswere considered significant if they were present at P < 0.05 (FWE corrected)within the ROIs for which a prior hypothesis could be generated from previousliterature. These were the MGB (13, 16), cortical regions that are frequentlyassociated with less activity in dyslexia compared with control subjects (PT, leftposterior temporal regions, left parietal regions, lIFG, left middle temporalgyrus, left inferior temporal gyrus, right postcentral gyrus; Fig. S3) (reviewed inref. 22), and cortical areas that are associated with more activity in dyslexia incontrast to control, i.e., the rIFG (23) (Fig. S3). In addition, ROIs of the auditorypathway (left and right CN, IC, AC, and Te1.0, Te1.1, and Te1.2 in Heschl gyri)were tested to investigate the location specificity of the MGB responses (Figs.S2 and S3). Because, for these regions of the auditory pathway except theMGB,

there were no specific hypotheses, we consider responses within the ROI sig-nificant only if they survived a subsequent Bonferroni correction for testing ofmultiple ROIs without a prior hypothesis. An overview of the results for all ROIsis shown in Tables S2–S6. Note that these tables also list, in a descriptivefashion, the P values for ROIs (without prior hypothesis) that were significantwithin ROI but did not survive the subsequent Bonferroni correction. Theseresults were considered as showing a trend to significance.

ACKNOWLEDGMENTS. We thank David R. R. Smith for helping with stimulusmanipulations and Judy Song and Timothy D. Griffiths for comments on themanuscript. This work was supported by a Max Planck Research Group Grant(to K.v.K.).

1. Shaywitz SE (1998) Dyslexia. N Engl J Med 338:307–312.2. Gabrieli JD (2009) Dyslexia: A new synergy between education and cognitive neuro-

science. Science 325:280–283.3. Carroll JM, Iles JE (2006) An assessment of anxiety levels in dyslexic students in higher

education. Br J Educ Psychol 76:651–662.4. Finucci J, Gottfredson L, Childs B (1985) A follow-up study of dyslexic boys. Ann

Dyslexia 35:117–136.5. Ramus F (2003) Developmental dyslexia: Specific phonological deficit or general

sensorimotor dysfunction? Curr Opin Neurobiol 13:212–218.6. Goswami U (2002) Phonology, reading development, and dyslexia: A cross-linguistic

perspective. Ann Dyslexia 52:139–163.7. Bryant PE, MacLean M, Bradley LL, Crossland J (1990) Rhyme and alliteration, pho-

neme detection, and learning to read. Dev Psychol 26:429–438.8. Landerl K, Wimmer H, Frith U (1997) The impact of orthographic consistency on

dyslexia: a German-English comparison. Cognition 63:315–334.9. Ramus F, Szenkovits G (2008) What phonological deficit? Q J Exp Psychol (Hove) 61:

129–141.10. Hari R, Renvall H (2001) Impaired processing of rapid stimulus sequences in dyslexia.

Trends Cogn Sci 5:525–532.11. Galaburda AM (1993) Neurology of developmental dyslexia. Curr Opin Neurobiol 3:

237–242.12. Stein J, Walsh V (1997) To see but not to read; the magnocellular theory of dyslexia.

Trends Neurosci 20:147–152.13. Galaburda AM, Menard MT, Rosen GD (1994) Evidence for aberrant auditory anat-

omy in developmental dyslexia. Proc Natl Acad Sci USA 91:8010–8013.14. Jones EG (1985) The Thalamus (Plenum, New York).15. Livingstone MS, Rosen GD, Drislane FW, Galaburda AM (1991) Physiological and an-

atomical evidence for a magnocellular defect in developmental dyslexia. Proc NatlAcad Sci USA 88:7943–7947.

16. von Kriegstein K, Patterson RD, Griffiths TD (2008) Task-dependent modulation ofmedial geniculate body is behaviorally relevant for speech recognition. Curr Biol 18:1855–1859.

17. Ahissar M, Lubin Y, Putter-Katz H, Banai K (2006) Dyslexia and the failure to forma perceptual anchor. Nat Neurosci 9:1558–1564.

18. Chandrasekaran B, Hornickel J, Skoe E, Nicol T, Kraus N (2009) Context-dependentencoding in the human auditory brainstem relates to hearing speech in noise: im-plications for developmental dyslexia. Neuron 64:311–319.

19. Cudeiro J, Sillito AM (2006) Looking back: Corticothalamic feedback and early visualprocessing. Trends Neurosci 29:298–306.

20. Suga N, Ma X (2003) Multiparametric corticofugal modulation and plasticity in theauditory system. Nat Rev Neurosci 4:783–794.

21. Thompson SK, et al. (2006) Representation of interaural time delay in the humanauditory midbrain. Nat Neurosci 9:1096–1098.

22. Richlan F, Kronbichler M, Wimmer H (2009) Functional abnormalities in the dyslexicbrain: A quantitative meta-analysis of neuroimaging studies. Hum Brain Mapp 30:3299–3308.

23. Hoeft F, et al. (2011) Neural systems predicting long-term outcome in dyslexia. ProcNatl Acad Sci USA 108:361–366.

24. Schiller PH, Logothetis NK, Charles ER (1990) Role of the color-opponent and broad-band channels in vision. Vis Neurosci 5:321–346.

25. Cornelissen P, Richardson A, Mason A, Fowler S, Stein J (1995) Contrast sensitivity andcoherent motion detection measured at photopic luminance levels in dyslexics andcontrols. Vision Res 35:1483–1494.

26. Martin F, Lovegrove W (1987) Flicker contrast sensitivity in normal and specificallydisabled readers. Perception 16:215–221.

27. Witton C, et al. (1998) Sensitivity to dynamic auditory and visual stimuli predictsnonword reading ability in both dyslexic and normal readers. Curr Biol 8:791–797.

28. Menell P, McAnally KI, Stein JF (1999) Psychophysical sensitivity and physiologicalresponse to amplitude modulation in adult dyslexic listeners. J Speech Lang Hear Res42:797–803.

29. Tallal P (1980) Auditory temporal perception, phonics, and reading disabilities inchildren. Brain Lang 9:182–198.

30. Tallal P, et al. (1996) Language comprehension in language-learning impaired chil-dren improved with acoustically modified speech. Science 271:81–84.

31. Ramus F, et al. (2003) Theories of developmental dyslexia: Insights from a multiplecase study of dyslexic adults. Brain 126:841–865.

32. Sperling AJ, Lu ZL, Manis FR, Seidenberg MS (2005) Deficits in perceptual noise ex-clusion in developmental dyslexia. Nat Neurosci 8:862–863.

33. White S, et al. (2006) The role of sensorimotor impairments in dyslexia: A multiplecase study of dyslexic children. Dev Sci 9:237–255.

34. Ziegler JC, Pech-Georgel C, George F, Lorenzi C (2009) Speech-perception-in-noisedeficits in dyslexia. Dev Sci 12:732–745.

35. Semrud-Clikeman M, Guy K, Griffin JD, Hynd GW (2000) Rapid naming deficits inchildren and adolescents with reading disabilities and attention deficit hyperactivitydisorder. Brain Lang 74:70–83.

36. Crick F (1984) Function of the thalamic reticular complex: The searchlight hypothesis.Proc Natl Acad Sci USA 81:4586–4590.

37. O’Connor DH, Fukui MM, Pinsk MA, Kastner S (2002) Attention modulates responsesin the human lateral geniculate nucleus. Nat Neurosci 5:1203–1209.

38. Krupa DJ, Ghazanfar AA, Nicolelis MAL (1999) Immediate thalamic sensory plasticitydepends on corticothalamic feedback. Proc Natl Acad Sci USA 96:8200–8205.

39. Edeline JM, Weinberger NM (1991) Thalamic short-term plasticity in the auditorysystem: Associative returning of receptive fields in the ventral medial geniculatebody. Behav Neurosci 105:618–639.

40. Friston K (2005) A theory of cortical responses. Philos Trans R Soc Lond B Biol Sci 360:815–836.

41. Kiebel SJ, Daunizeau J, Friston KJ (2008) A hierarchy of time-scales and the brain.PLOS Comput Biol 4:e1000209.

42. Balaguer-Ballester E, Clark NR, Coath M, Krumbholz K, Denham SL (2009) Un-derstanding pitch perception as a hierarchical process with top-down modulation.PLOS Comput Biol 5:e1000301.

43. Kiebel SJ, von Kriegstein K, Daunizeau J, Friston KJ (2009) Recognizing sequences ofsequences. PLOS Comput Biol 5:e1000464.

44. Denckla MB, Rudel RG (1976) Rapid “automatized” naming (R.A.N): Dyslexia differ-entiated from other learning disabilities. Neuropsychologia 14:471–479.

45. Boets B, Wouters J, vanWieringen A, Ghesquière P (2007) Auditory processing, speechperception and phonological ability in pre-school children at high-risk for dyslexia: Alongitudinal study of the auditory temporal processing theory. Neuropsychologia 45:1608–1620.

46. Lallier M, et al. (2009) Auditory and visual stream segregation in children and adults:An assessment of the amodality assumption of the ‘sluggish attentional shifting’theory of dyslexia. Brain Res 1302:132–147.

47. Shaywitz SE, et al. (1998) Functional disruption in the organization of the brain forreading in dyslexia. Proc Natl Acad Sci USA 95:2636–2641.

48. Georgiewa P, et al. (2002) Phonological processing in dyslexic children: A studycombining functional imaging and event related potentials. Neurosci Lett 318:5–8.

49. Shaywitz BA, et al. (2002) Disruption of posterior brain systems for reading in childrenwith developmental dyslexia. Biol Psychiatry 52:101–110.

50. Paulesu E, et al. (2001) Dyslexia: Cultural diversity and biological unity. Science 291:2165–2167.

51. Raschle NM, Zuk J, Gaab N (2012) Functional characteristics of developmental dyslexiain left-hemispheric posterior brain regions predate reading onset. Proc Natl Acad SciUSA 109:2156–2161.

52. Temple E, et al. (2000) Disruption of the neural response to rapid acoustic stimuli indyslexia: evidence from functional MRI. Proc Natl Acad Sci USA 97:13907–13912.

53. Steinbrink C, Ackermann H, Lachmann T, Riecker A (2009) Contribution of the an-terior insula to temporal auditory processing deficits in developmental dyslexia. HumBrain Mapp 30:2401–2411.

54. Ives DT, Smith DR, Patterson RD (2005) Discrimination of speaker size from syllablephrases. J Acoust Soc Am 118:3816–3822.

55. Kawahara H, Irino T (2005) Underlying principles of a high-quality speech manipu-lation system STRAIGHT and its application to speech segregation. Speech Separationby Humans and Machines, ed Divenyi P (Springer, New York), pp 167–180.

56. Smith DRR, Patterson RD (2005) The interaction of glottal-pulse rate and vocal-tractlength in judgements of speaker size, sex, and age. J Acoust Soc Am 118:3177–3186.

57. Remez RE, Fellowes JM, Rubin PE (1997) Talker identification based on phonetic in-formation. J Exp Psychol Hum Percept Perform 23:651–666.

58. Morosan P, et al. (2001) Human primary auditory cortex: Cytoarchitectonic sub-divisions and mapping into a spatial reference system. Neuroimage 13:684–701.

59. Westbury CF, Zatorre RJ, Evans AC (1999) Quantifying variability in the planumtemporale: A probability map. Cereb Cortex 9:392–405.

13846 | www.pnas.org/cgi/doi/10.1073/pnas.1119828109 Díaz et al.

Supporting InformationDíaz et al. 10.1073/pnas.1119828109SI MethodsParticipants. Informed consent was obtained from all participants.The study was approved by the Ethics Committee of the MedicalFaculty, University of Leipzig, Germany. Nonverbal IQ was in thenormal range in both groups (i.e., ≥85). The control participantsreported no language-related difficulties and scored within thenormal range on the literacy tests (i.e., spelling ≥90, readingspeed and reading comprehension ≥40) (1, 2) (Table S1). A totalof 14 control participants were included. None of the partici-pants had a history of auditory, neurological, or psychiatric dis-orders. One additional control participant was excluded becausehis performance in the experimental tasks was two SDs below themean of his group.In the dyslexia group, seven participants had a previous di-

agnosis of developmental dyslexia and the other seven partic-ipants reported having spelling and reading problems sincechildhood. Eleven of the dyslexic participants scored two SDsbelow the mean of the control participants on at least two of theliteracy measures and three scored two SDs below themean of thecontrols on one literacy measure and one SD below on the othertwo literacy measures.In addition, phonological skills were measured with a non-

standardized word and nonword reading tests (3). The time andaccuracy in reading 48 German nonwords and 48 words weremeasured. Dyslexics were slower than controls in reading thenonword list and less accurate in both nonword and wordreading (Table S1), as reported previously (3). The results of thetests of phonological access (Methods) are listed in Table S1.

Stimuli. Experiment 1. We used 150 different vowel-consonant-vowel syllables taken from a database (4). The syllables were re-corded from one male speaker in a quiet room using a ShureSM58-LCEmicrophone and a high-quality PC sound card (SoundBlaster Audigy II; Creative Labs) with 16-bit resolution and 48-kHz sample rate. The syllables were resynthesized by usingSTRAIGHT software (5) to simulate nine speakers for eachsyllable by manipulating the acoustic effect of the vocal tractlength (VTL) and the glottal pulse rate (GPR). Each syllable wasresynthesized with three VTL values (9, 13, and 17 cm) and threeGPR values (80, 170, and 280 Hz) to ensure that the speaker taskcould only be performed based on slow-varying acoustic featuresof the voice.Syllables were randomly presented within the sequence with

the constraints that (i) each syllable/speaker occurred at leasttwice and (ii) changes between two consecutive syllables/speak-ers occurred between two and three times within a sequence.There were 80 syllable sequences in total. Forty included

changes in VTL with GPR kept constant, and, conversely, theother 40 sequences contained changes inGPRbut a constantVTL.Experiment 2. We used 40 five-word sentences that were seman-tically neutral (e.g., “Der Junge trägt einen Koffer,” “The boycarries a suitcase”) and syntactically similar (i.e., subject-verb-object). They were recorded from a 22-y-old male Germanspeaker in a soundproof booth by using a condenser microphone[TLM 50 (Neumann); Mic-Amp F-35 preamplifier Lake People],a high-quality sound card (built-in sound card for Power Mac G5Dual 1.8 GHz; Apple), and the software Sound Studio 3 (FeltTip) with 16-bit resolution and 44.1-kHz sample rate.

Data Acquisition. MRI was performed on a 3-T Magnetom TIMTrio (Siemens). The functional MRI (fMRI) data were acquiredby using a gradient-echo EPI sequence (echo time, 30 ms; flip

angle, 90°; acquisition bandwidth, 116 kHz; slice thickness, 2 mm;interslice gap, 1 mm; 42 slices; axial and ascending acquisitionstarting at the pontomedullary junction). To optimally image sub-cortical sensory structures, we used a sparse sampling imagingprotocol with cardiac gating (6). Sparse sampling allows pre-sentation of the auditory stimuli without the noise of the scannergradients by acquiring a volume only at the end of an auditorysequence (7). Cardiac gating reduces the artifacts caused by thepulsatile motion of the brainstem by starting each volume ac-quisition at the same point of the participants’ cardiac cycle. Forthat reason, the scan repetition time varied slightly (time to re-peat, 2.8 s for volume acquisition + 8.4 s stimulation + time tonext pulse). Participants’ pulse was measured by means of a pulseoximeter. Task instructions (experiment 1: “Silbe,” ”syllable,” or“Sprecher,” “speaker”; experiment 2, “Satzende,” “sentence end”)were visually presented during the last 10 slice acquisitions of eachvolume. Experiment 1 included 170 brain volumes for each par-ticipant (five runs of 34 volumes each). Experiment 2 included 84brain volumes per participant (two runs of 42 volumes each). Thefirst two volumes of each run were not included in the analysis.In addition, high-resolution anatomical images (voxel dimen-

sions of 1 × 1 × 1 mm3) were acquired by using a T1-weighted 3Dmagnetization-prepared rapid gradient echo sequence with se-lective water excitation and linear phase encoding (repetitiontime, 1,300 ms; time to echo, 3.93 ms; field of view, 256 × 240 mm2;slab thickness, 192 mm; sagittal orientation) (8).

Analysis of MRI Data.MRI data were analyzed with SPM8 (www.fil.ion.ucl.ac.uk/spm) and Matlab 7.10 (R2010a; MathWorks). Scanswere realigned and unwarped. For each participant, functionalimages were coregistered to the participant’s anatomical imageand normalized to Montreal Neurological Institute (MNI) stan-dard stereotactic space. For the analysis of subcortical structures,images were spatially smoothed with a Gaussian smoothing kernelof 4 mm full width at half maximum, and, for cortical structures, asmoothing kernel of 8 mm was used. For all analyses, statisticalparametric maps were generated by modeling the evoked hemo-dynamic response for the different conditions as boxcar functionsconvolved with a synthetic hemodynamic response function usingthe general linear model approach (9).Definition of regions of interest. For the localization of structures inthe auditory pathway, the auditory cortices (AC), the medialgeniculate bodies (MGBs), the inferior colliculi (IC), and thecochlear nuclei (CNs) were functionally mapped by contrasting“sentences − silence” (experiment 2; Fig. S2 and Table S5). Toseparate the clusters of the subcortical structures into individualregions of interest (ROIs), the functional clusters were combinedwith a 6-mm sphere centered at the location of the maximumstatistic for each. ROIs for cortices that have been previouslyimplicated in dyslexia were based on coordinates taken from theliterature (10, 11) or, if available, on probabilistic maps, i.e.,planum temporale (PT) (12) (Fig. S3). Coordinates for poten-tially dysfunctional cortical regions in dyslexia were taken froma metaanalysis of neuroimaging findings in dyslexia (11). Amongthe three most recent metaanalyses (11, 13, 14), we selected one(11) because MNI coordinates as well as the range of the acti-vations (i.e., limits of the activation clusters of the metaanalysisin x, y, and z direction) were reported. For the ROIs, we createdspheres with their center corresponding to the center of theactivations and with a size based on the mean distance of thethree coordinate dimensions (x, y, z) reported by Richlan et al.(11): left posterior temporal regions (lPTRs; MNI, center −53,

Díaz et al. www.pnas.org/cgi/content/short/1119828109 1 of 9

−53, 2; 15-mm sphere), left parietal regions (lPRs; MNI center−53, −45, 46; 8-mm sphere), left inferior frontal gyrus (lIFG;MNI center −50, 18, 6; 6-mm sphere), left middle temporal gyrus(lMTG; MNI center −51, −26, −1; 7 mm sphere), left inferiortemporal gyrus (lITG; MNI center −43, −26, −20; 5 mm sphere),and right postcentral gyrus (rPCG; MNI center 55, −17, 30; 4-mmsphere). The coordinate for the right inferior frontal gyrus (rIFG;MNI center 57, 7, 19), a region that is implicated in compensatorymechanisms in dyslexia, was taken from Hoeft (10). The Talair-ach coordinates reported by Hoeft (10) were converted to MNIcoordinates by using the tal2mni.m function of MNI brainand Talairach atlas (http://imaging.mrc-cbu.cam.ac.uk/imaging/MniTalairach). The ROIs were created with the MarsBaR Tool-box (http://marsbar.sourceforge.net).Correlation analysis. For the plots in Fig. 2C, parameter estimateswere extracted for each participant from the voxel with the

maximum statistic for the group [reading comprehension, −15,−31, −11, MNI; rapid automatized naming (RAN) for numbersand letters, −15, −31, −8, MNI]. To obtain the Pearson R and fordisplay purposes (Fig. 2C), parameter estimates were analyzedand plotted by using PASW Statistics 18.0. Note that the P values[significant at P < 0.05, family-wise error (FWE) corrected] wereobtained with SPM and that the definition of the ROIs was basedon a prior hypothesis and independent datasets (i.e., functionalactivations in experiment 2, previously published coordinatesand anatomical masks).

SI ResultsThe time required to name objects in the RAN task did notcorrelate with the MGB modulation (or any other ROI). Thisresult is in line with the finding that naming objects is not a re-liable predictor of reading ability in adults (15).

1. Kersting M, Klaus A (2004) Rechtschreibungstest (Hogrefe, Göttingen, Germany).2. Schneider W, Schlagmüller M, Ennemoser M (2007) Lesegeschwindigkeits- und

verständnistest für die Klassen 6-12 (Hogrefe, Göttingen, Germany).3. Schulte-Körne G (2007) Genetik der lese- und rechtschreibstörung. Monatsschr

Kinderheilkd 155:328–336.4. Ives DT, Smith DR, Patterson RD (2005) Discrimination of speaker size from syllable

phrases. J Acoust Soc Am 118:3816–3822.5. Kawahara H, Irino T (2005) Underlying principles of a high-quality speech manipulation

system STRAIGHT and its application to speech segregation. Speech Separation by Humansand Machines, ed Divenyi P (Springer, New York), pp 167–180.

6. Thompson SK, et al. (2006) Representation of interaural time delay in the humanauditory midbrain. Nat Neurosci 9:1096–1098.

7. Hall DA, et al. (1999) “Sparse” temporal sampling in auditory fMRI. Hum Brain Mapp7:213–223.

8. Mugler, JP, 3rd, Brookeman JR (1990) Three-dimensional magnetization-preparedrapid gradient-echo imaging (3D MP RAGE). Magn Reson Med 15:152–157.

9. Friston KJ, Ashburner J, Kiebel SJ, Nichols TE, Penny WD (2007) Statistical ParametricMapping: The Analysis of Functional Brain Images (Academic/Elsevier, Amsterdam).

10. Hoeft F, et al. (2011) Neural systems predicting long-term outcome in dyslexia. ProcNatl Acad Sci USA 108:361–366.

11. Richlan F, Kronbichler M, Wimmer H (2009) Functional abnormalities in the dyslexicbrain: A quantitative meta-analysis of neuroimaging studies. Hum Brain Mapp 30:3299–3308.

12. Westbury CF, Zatorre RJ, Evans AC (1999) Quantifying variability in the planumtemporale: A probability map. Cereb Cortex 9:392–405.

13. Richlan F, Kronbichler M, Wimmer H (2011) Meta-analyzing brain dysfunctions indyslexic children and adults. Neuroimage 56:1735–1742.

14. Maisog JM, Einbinder ER, Flowers DL, Turkeltaub PE, Eden GF (2008) A meta-analysisof functional neuroimaging studies of dyslexia. Ann N Y Acad Sci 1145:237–259.

15. Semrud-Clikeman M, Guy K, Griffin JD, Hynd GW (2000) Rapid naming deficits inchildren and adolescents with reading disabilities and attention deficit hyperactivitydisorder. Brain Lang 74:70–83.

Fig. S1. Task-dependent modulation of the MGB for groups of control and dyslexic participants. Within the functionally localized MGB (location approxi-mately indicated by gray circles; Fig. S2 shows exact definition), the group of normal readers showed greater blood oxygen level-dependent signal duringattending to speech sounds (phonological task) than to speaker characteristics (speaker task; left MGB, P = 0.067 FWE corrected; right MGB, P = 0.034 FWEcorrected). No such task-dependent modulation of the MGB was present in the group of participants with dyslexia. The statistical parametrical maps for eachgroup of participants are rendered on coronal sections of the mean normalized, structural MRI volume. The color bar represents t values.

Díaz et al. www.pnas.org/cgi/content/short/1119828109 2 of 9

Fig. S2. Definition of ROI based on functional data from experiment 2. The MGBs, ICs, CN, and AC were functionally localized by the contrast “sentences −silence” in experiment 2. Images in the left column display statistical parametrical maps for all participants on sections of the mean normalized, structural MRIvolume. For subcortical structures, the functional clusters were then combined with a 6-mm sphere centered at the statistic peak location with MarsBaR (http://marsbar.sourceforge.net) to separate the clusters into individual ROIs. Images in the right column show the ROIs overlaid on an MNI template brain. The colorbars represent t values.

Fig. S3. ROIs based on probabilistic maps and previous studies on dyslexia. (A) The ROIs for ACs are overlaid on an axial section of the mean normalized,structural MRI volume. The ROIs were created based on probabilistic maps of primary ACs and auditory association cortices: Te1.0, Te1.1, and Te1.2 (1), and PT(2). (B) The ROIs for cortical areas that are described as less active in dyslexic vs. control participants in previous literature of reading and phonological tasks(lPTR, lIFG, lMTG, and lPR) (3) (Table S3) are overlaid on a sagittal section of the mean normalized, structural MRI volume. (C) The ROI for rIFG, a region impliedin compensatory mechanisms in dyslexia (4) (Table S4), is overlaid on a sagittal section of the mean normalized, structural MRI volume.

1. Morosan P, et al. (2001) Human primary auditory cortex: Cytoarchitectonic subdivisions and mapping into a spatial reference system. Neuroimage 13:684–701.2. Westbury CF, Zatorre RJ, Evans AC (1999) Quantifying variability in the planum temporale: A probability map. Cereb Cortex 9:392–405.3. Richlan F, Kronbichler M, Wimmer H (2009) Functional abnormalities in the dyslexic brain: A quantitative meta-analysis of neuroimaging studies. Hum Brain Mapp 30:3299–3308.4. Hoeft F, et al. (2011) Neural systems predicting long-term outcome in dyslexia. Proc Natl Acad Sci USA 108:361–366.

Díaz et al. www.pnas.org/cgi/content/short/1119828109 3 of 9

Table S1. Biographic, cognitive, and language-related data for both groups of participants

Characteristic Normal readers (n = 14) Dyslexic participants (n = 14) Two-sample t tests (df = 26)

Demographic dataMean age ± SD, y 23.5 ± 2.76 24.5 ± 2.40 NSMale sex 14 14 —

No. right-handed 12 12 —

Education level 13 undergraduate students,1 high school diploma

13 undergraduate students,1 high school diploma

—

Screening tests for dyslexia, mean ± SDNonverbal intelligence* 107.07 ± 15.17 101.78 ± 12.84 NSSpelling† 102.21 ± 6.19 82.14 ± 7.44 t = 7.75, P < 0.001Reading speed‡ 58.00 ± 8.44 41.57 ± 6.50 t = 5.76, P < 0.001Reading comprehension‡ 63.28 ± 7.18 46.79 ± 4.22 t = 7.40, P < 0.001

Phonological tests, mean ± SDRAN numbers

Time (s) 16.49 ± 2.35 22.03 ± 6.01 t = 3.21, P < 0.01Errors (%) 0.84 ± 1.28 0.28 ± 0.72 NS

RAN lettersTime (s) 16.09 ± 2.60 21.37 ± 4.20 t = 3.98, P < 0.001Errors (%) 0.70 ± 1.84 0.70 ± 1.26 NS

RAN objectsTime (s) 30.84 ± 5.85 32.57 ± 4.57 NSErrors (%) 0.84 ± 1.28 1.00 ± 1.30 NS

Word readingTime (s) 38.58 ± 7.61 45.38 ± 14.19 NSErrors (%) 0.87 ± 1.33 4.75 ± 6.60 t = 2.14, P < 0.05

Nonword readingTime (s) 78.08 ± 16.89 110.96 ± 38.76 t = 2.91, P < 0.01Errors (%) 3.70 ± 2.83 25.58 ± 20.5 t = 3.95, P < 0.001

*Raven matrices, scores are based on standard scores (mean = 100, SD = 15).†R-T writing test, scores are based on standard scores (mean = 100, SD = 10).‡Reading speed and comprehension tests, scores are based on t-standard scores (mean = 50, SD = 10).

Díaz et al. www.pnas.org/cgi/content/short/1119828109 4 of 9

Table S2. Experiment 1: local fMRI activation maxima for the task comparison for each group ofparticipants

Phonological task − speaker task (P Value)

ROI Controls Dyslexics

Functionally definedMGB

Left −9, −28, −8 (0.0670) NSRight 9, −28, −8 (0.0336) NS

ICLeft NS NSRight NS 3, −40, −5 (0.0563)

CNLeft NS NSRight NS NS

ACLeft NS NSRight NS NS

Based on probabilistic maps (1, 2)Te1.0

Left NS NSRight NS NS

Te1.1Left NS NSRight NS NS

Te1.2Left NS NSRight NS NS

PTLeft NS NSRight NS NS

Based on previous dyslexia studies (3, 4)lPTR NS NSlPR NS NSlIFG −45, 14, 7 (0.0580) NSlMTG −54, −28, −2 (0.0031) −54, −31, −5 (0.0715)lITG NS −39, −22, −20 (0.0211)rPCG NS NSrIFG NS 57, 8, 19 (0.0219)

Numbers represent MNI coordinates (x, y, z) of local activation maxima for the contrasts “phonological task −speaker task” (experiment 1) for each group of participants. All P values reported are FWE corrected within theROI. MGB, IC, CN, and AC regions were functionally defined (experiment 2). Primary ACs (Te1.0, Te1.1, andTe1.2) are anatomically defined regions in Heschl gyri that are included as probabilistic maps in the SPMAnatomy Toolbox (2) (SI Methods and Figs. S2 and S3). ROIs for cortices that have been previously implicatedin dyslexia were based on coordinates taken from the literature (lPTR, lIFG, lMTG, lPR, lITG, rPCG, rIFG) (3, 4) or, ifavailable, on probabilistic maps, i.e., PT (1). For information purposes, we also report regions showing onlya trend to significance after the small volume correction. NS, not significant.

1. Westbury CF, Zatorre RJ, Evans AC (1999) Quantifying variability in the planum temporale: A probability map. Cereb Cortex 9:392–405.2. Morosan P, et al. (2001) Human primary auditory cortex: Cytoarchitectonic subdivisions and mapping into a spatial reference system. Neuroimage 13:684–701.3. Hoeft F, et al. (2011) Neural systems predicting long-term outcome in dyslexia. Proc Natl Acad Sci USA 108:361–366.4. Richlan F, Kronbichler M, Wimmer H (2009) Functional abnormalities in the dyslexic brain: A quantitative meta-analysis of neuroimaging studies. Hum Brain Mapp 30:3299–3308.

Díaz et al. www.pnas.org/cgi/content/short/1119828109 5 of 9

Table S3. Experiment 1: local fMRI activation maxima indicating potentially dysfunctional mechanisms in dyslexicparticipants

Controls − dyslexics (P value)

ROIPhonological task −

speaker taskPhonological task +

speaker tasksPhonological

taskSpeakertask

Functionally definedMGB

Left −9, -28, −8 (0.0338) NS NS NSRight NS NS NS NS

ICLeft NS NS NS NSRight NS NS NS NS

CNLeft NS NS NS NSRight NS NS NS NS

ACLeft NS NS NS NSRight NS NS NS NS

Based on probabilistic maps (1, 2)Te1.0

Left NS NS NS NSRight NS NS NS NS

Te1.1Left NS NS NS NSRight NS NS NS NS

Te1.2Left NS NS NS NSRight NS NS NS NS

PTLeft NS NS NS NSRight NS NS NS NS

Based on previous dyslexia studies (3, 4)lPTR NS NS NS NSlPR NS NS NS NSlIFG NS NS NS NSlMTG NS NS NS NSlITG NS NS NS NSrPCG NS NS NS NSrIFG NS NS NS NS

Numbers represent MNI coordinates (x, y, z) of local activation maxima. The P values reported are FWE corrected within ROIs. Noregion besides the left MGB was activated even if we used a lenient statistical threshold of 0.01 uncorrected within the ROI (see text).Details about the ROIs are provided in SI Methods, Table S2, and Figs. S2 and S3. Primary ACs (Te1.0, Te1.1, and Te1.2) are anatomicallydefined regions in Heschl gyri that are included as probabilistic maps in the SPM Anatomy Toolbox (2). NS, not significant.

1. Westbury CF, Zatorre RJ, Evans AC (1999) Quantifying variability in the planum temporale: A probability map. Cereb Cortex 9:392–405.2. Morosan P, et al. (2001) Human primary auditory cortex: Cytoarchitectonic subdivisions and mapping into a spatial reference system. Neuroimage 13:684–701.3. Hoeft F, et al. (2011) Neural systems predicting long-term outcome in dyslexia. Proc Natl Acad Sci USA 108:361–366.4. Richlan F, Kronbichler M, Wimmer H (2009) Functional abnormalities in the dyslexic brain: A quantitative meta-analysis of neuroimaging studies. Hum Brain Mapp 30:3299–3308.

Díaz et al. www.pnas.org/cgi/content/short/1119828109 6 of 9

Table S4. Experiment 1: local fMRI activation maxima indicating potential compensatory mechanisms in dyslexic participants

Dyslexics − controls (P value)

ROIPhonological task −

speaker taskPhonological task +

speaker tasks Phonological task Speaker task

Functionally definedMGB

Left NS NS NS NSRight 18, −28, −8 (0.0514) NS NS NS

ICLeft NS NS NS NSRight 6, −37, −5 (0.0291) NS NS NS

CNLeft NS NS NS NSRight 6, −37, −41 (0.0347) NS NS NS

ACLeft NS −60, −13, 1 (0.0893) NS NSRight NS NS 45, −28, 10 (0.079) NS

Based on probabilistic maps (1, 2)Te1.0

Left NS NS NS NSRight NS NS NS NS

Te1.1Left NS NS NS NSRight NS NS 45, -28, 10 (0.088) NS

Te1.2Left NS NS NS NSRight NS NS NS NS

PTLeft NS NS NS NSRight NS NS NS NS

Based on previous dyslexia studies (3, 4)lPTR NS NS NS NSlPR NS NS NS NSlIFG NS NS NS NSlMTG NS NS NS NSlITG NS NS NS NSrPCG NS NS NS NSrIFG NS 62, 2, 19 (0.0472*) 63, 2, 19 (0.0164*) 63, 2, 22 (0.0268*)

Numbers represent MNI coordinates (x, y, z) of local activation maxima. All P values reported are FWE corrected within ROIs. Significant results are reportedin bold font. Primary ACs (Te1.0, Te1.1, and Te1.2) are anatomically defined regions in Heschl gyri that are included as probabilistic maps in the SPM AnatomyToolbox (2). NS, not significant.*Significant at at P < 0.05 FWE corrected in regions for which a prior hypothesis was based on literature. As expected, the rIFG showed significantly more task-dependent modulation in dyslexic vs. control participants, indicating a potential compensatory mechanism. Regions without a prior hypothesis (Methods) wereadditionally Bonferroni-corrected for testing of multiple regions without prior hypothesis. We considered regions that do not survive this Bonferroni correctionas showing a trend to significance. Note that the table reports the FWE P values before a potential Bonferroni correction. For information purposes, we alsoreport here regions that do not survive the Bonferroni correction or/and show only a trend to significance in the correction for the ROI (SI Methods and Table S2).

1. Westbury CF, Zatorre RJ, Evans AC (1999) Quantifying variability in the planum temporale: A probability map. Cereb Cortex 9:392–405.2. Morosan P, et al. (2001) Human primary auditory cortex: Cytoarchitectonic subdivisions and mapping into a spatial reference system. Neuroimage 13:684–701.3. Hoeft F, et al. (2011) Neural systems predicting long-term outcome in dyslexia. Proc Natl Acad Sci USA 108:361–366.4. Richlan F, Kronbichler M, Wimmer H (2009) Functional abnormalities in the dyslexic brain: A quantitative meta-analysis of neuroimaging studies. Hum Brain Mapp 30:3299–3308.

Díaz et al. www.pnas.org/cgi/content/short/1119828109 7 of 9

Table S5. Experiment 2: local fMRI activation maxima for the contrast of sentences and silence

Sentences − silence (P value)

ROI All participants Controls Dyslexics Controls − dyslexics Dyslexics − controls

Functionally definedMGB

Left −15, −28, −5 −15, −28, −5 (0.0082) −15, 28, −5 (<0.001) NS NSRight 12, −28, −8 15, −25, −8 (0.0018) 12, −28, −8 (0.0020) NS NS

ICLeft −6, −37, −14 −9, −34, −11 (0.0348) −6, −37, −14 (0.0150) NS NSRight 6, -37, −5 6, −37, −5 (0.0008) 3, −40, −8 (0.0036) NS NS

CNLeft −9, −37, −41 NS −12, −40, −41 (0.0081) NS NSRight 9, −34, −41 9, −34, −41 (0.0444) 9, −34, −41 (0.0014) NS NS

ACLeft −60, −13, 4 −57, −16, 7 (0.0114) −51, −16, 1 (<0.001) NS NSRight 45, -25, 7 39, −37, 7 (0.0038) 57, −25, 7 (<0.001) NS NS

Based on probabilisticmaps (1, 2)Te1.0

Left −51, −16, 4 −54, −16, 7 (0.0190) −51, −13, 1 (<0.001) NS NSRight 42, −22, 7 42, −22, 7 (0.0182) 42, −22, 7 (<0.001) NS NS

Te1.1Left −39, −31, 13 −39, −31, 10 (0.0241) −39, −28, 13 (<0.001) NS NSRight 45, −25, 7 45, −25, 7 (0.0049) 45, −22, 7 (<0.001) NS NS

Te1.2Left −54, −16, 4 −57, −13, 7 (0.0022) −51, −10, 1 (<0.001) NS NSRight 57, −7, -2 NS 57, −7, 2 (<0.001) NS NS

PTLeft −54, −16, 4 NS −51, −16, 1 (<0.001) NS NSRight 45, −25, 7 45, −5, 7 (0.0492) 57, −25, 7 (<0.001) NS NS

Related to dyslexia (3, 4)lPTR NS NS −45, −43, 7 (0.0155) NS NSlPR NS NS NS NS NSlIFG NS NS NS NS NSlMTG −54, −22, 1 (<0.001) NS −54, −22, 1 (<0.001) NS NSlITG NS NS NS NS NSrPCG NS NS NS NS NSrIFG NS NS 60, 2, 19 (0.0313) NS NS

Numbers represent MNI coordinates (x, y, z) of local activation maxima for the contrast “sentences − silence” (experiment 2). All P values reported are FWEcorrected within ROIs. Details about ROIs are provided in SI Methods and Table S2. Primary ACs (Te1.0, Te1.1, and Te1.2) are anatomically defined regions inHeschl gyri that are included as probabilistic maps in the SPM Anatomy Toolbox (2). NS, not significant.

1. Westbury CF, Zatorre RJ, Evans AC (1999) Quantifying variability in the planum temporale: A probability map. Cereb Cortex 9:392–405.2. Morosan P, et al. (2001) Human primary auditory cortex: Cytoarchitectonic subdivisions and mapping into a spatial reference system. Neuroimage 13:684–701.3. Hoeft F, et al. (2011) Neural systems predicting long-term outcome in dyslexia. Proc Natl Acad Sci USA 108:361–366.4. Richlan F, Kronbichler M, Wimmer H (2009) Functional abnormalities in the dyslexic brain: A quantitative meta-analysis of neuroimaging studies. Hum Brain Mapp 30:3299–3308.

Díaz et al. www.pnas.org/cgi/content/short/1119828109 8 of 9

Table S6. Experiment 1: correlations maxima for the contrast of phonological and speaker tasks with diagnosticdyslexia scores

Phonological task − speaker task (P value)

ROI Controls Dyslexics

Functionally definedMGB

Left NS −15, −31, −11 (0.0237)*†; −15, −31, −8 (0.0074)*‡

Right NS NSIC

Left NS NSRight 6, −40, −8 (0.0459)§ NS

CNLeft NS NSRight 6, −37, −41 (0.0460)§ 9, −37, −41 (0.0449){

ACLeft NS NSRight NS NS

Based on probabilistic maps (1, 2)Te1.0

Left NS NSRight NS NS

Te1.1Left NS NSRight NS NS

Te1.2Left NS NSRight NS NS

PTLeft NS NSRight NS NS

ROIs related to dyslexia (3, 4)lPTR 42, −46, 4 (0.037)*§ NSlPR −48, −40, 43 (0.0458)*{ NSlIFG NS NSlMTG NS NSlITG NS NSrPCG 54, −16, 28 (0.0424)*† NSrIFG NS NS

Numbers represent MNI coordinates (x, y, z) of local activation maxima for the correlations between the contrast of phonological −speaker tasks (experiment 1) with the dyslexia diagnostic scores. All P values reported are FWE corrected within ROIs (SI Methods, TableS2, and Figs. S2 and S3). Primary ACs (Te1.0, Te1.1, and Te1.2) are anatomically defined regions in Heschl gyri that are included asprobabilistic maps in the SPM Anatomy Toolbox (2). NS, not significant.*Significant difference at P < 0.05, FWE corrected in regions for which a previous hypothesis was based on literature (MGB, PT, lPTR,lPR, lIFG, lMTG, lITG, rPCG). Regions without a previous hypothesis (CN, IC, AC, and TE1.0, Te1.1, Te1.2) were additionally Bonferroni-corrected for testing of multiple regions without previous hypothesis. We consider regions that do not survive Bonferroni correction asshowing a trend to significance. Note that the table reports FWE P values without the Bonferroni correction for all ROIs.†Reading comprehension score.‡RAN of numbers and letters.§Number of errors when reading a list of nonwords.{Time to read a list of nonwords.

1. Westbury CF, Zatorre RJ, Evans AC (1999) Quantifying variability in the planum temporale: A probability map. Cereb Cortex 9:392–405.2. Morosan P, et al. (2001) Human primary auditory cortex: Cytoarchitectonic subdivisions and mapping into a spatial reference system. Neuroimage 13:684–701.3. Hoeft F, et al. (2011) Neural systems predicting long-term outcome in dyslexia. Proc Natl Acad Sci USA 108:361–366.4. Richlan F, Kronbichler M, Wimmer H (2009) Functional abnormalities in the dyslexic brain: A quantitative meta-analysis of neuroimaging studies. Hum Brain Mapp 30:3299–3308.

Díaz et al. www.pnas.org/cgi/content/short/1119828109 9 of 9