Late, not early mismatch responses to changes in frequency are reduced or deviant in children with dyslexia: an event-related potential study

Developmental dyslexia (hereafter ‘dyslexia’) is typically diagnosed where a child has difficulty learning to read and spell, despite having no known physical, intellectual, neurological, emotional, educational, or socio-economic problems which might account for these difficulties [1]. Researchers largely agree on those factors that would preclude a diagnosis of dyslexia, yet the causal factors that lead to the disorder are still a matter of debate. There are now several influential theories which state that dyslexia, at least in a subset of individuals, arises from low-level impairments in the perception and processing of auditory information (e.g. [2‐5]). These theories propose a variety of different auditory processing impairments as underlying dyslexia, from deficits in ‘rapid’ auditory temporal processing [5], to frequency discrimination [6], and the detection of rate of change of amplitude at the onset (rise time) and/or during the speech envelope [2, 3]. Nonetheless, they are united in the premise that these impairments lead to difficulties in analysing the incoming speech stream and subsequent problems with phonological processing (i.e. in discriminating, categorizing, and manipulating speech sounds). In turn, these difficulties with phonological processing are thought to lead to problems in learning to read in the case of dyslexia (cf. [7]). A subset of these theories also accounts for the oral language problems of those with specific language impairment (SLI) (e.g. [3, 5]).

There is now a large number of studies showing that both children and adults with dyslexia tend to perform more poorly than matched peers on behavioural measures of auditory processing (for review, see [8]). These measures include, but are not limited to, those of auditory frequency discrimination (i.e. the ability to discriminate differences in pitch) [8]. However, not all studies have replicated findings of deficits in auditory processing in this population (for reviews, see [8, 9]). One factor that may contribute to these mixed results is the nature of the psychophysical tasks used to assess auditory processing. Behavioural evidence can be difficult to interpret, as elevated thresholds on such tasks can arise due to poor (selective or sustained) attention, memory, or motivation, even when stimuli can be accurately discriminated [10‐12]. These confounds are characteristic of children, but particularly of children with dyslexia, who frequently also experience deficits in attention and memory in addition to their reading difficulties [13‐17]. Researchers have therefore turned to the auditory event-related potential (ERP), in the hope that it will provide a more objective measure of auditory processing in dyslexia.

The mismatch negativity (MMN) component of the ERP has rapidly become the method of choice for assessing auditory discrimination in dyslexia. The auditory MMN is typically elicited using an oddball paradigm in which a train of repeated standard auditory stimuli includes occasional deviant stimuli that differ from the standard in one or more acoustic dimensions (e.g. frequency, duration, or rise time; [18]). Responses to standards and deviants are averaged separately and then subtracted from each other. Typically, this method reveals the presence of an enhanced negative response to deviants, occurring approximately 100–250 ms after stimulus onset, with the latency and amplitude of the negativity increasing with the difference of the deviant from the standard tone [19, 20]. Hence, it is often viewed as a discriminative response. Importantly, the MMN can be elicited passively, i.e. without the need for participants to perform a task or attend to the stimuli. Note that this paradigm has also been shown to elicit a component known as the late discriminative negativity (LDN) [21] (also ‘late MMN’ [22]), which is reflected as a prolonged period of negativity occurring around 300–550 ms post-stimulus onset [21]. Unlike the MMN, there is evidence that the LDN is larger for small rather than large deviants and may reflect additional processing of auditory stimuli that occurs when the salient features of the stimulus are difficult to discriminate [23].

Given that the MMN is an objective measure of auditory discrimination, it is surprising that studies assessing the auditory MMN in dyslexia often yield less consistent findings than behavioural studies [8]. For instance, studies assessing responses to frequency deviants have demonstrated an MMN in dyslexic groups that is reduced [24‐26], enhanced [27], or not significantly different in amplitude from that of matched controls [28‐32]. Other studies have found that whether or not a dyslexic group showed differences in the MMN depended on the particular electrode(s) chosen for analysis, stimuli used, or participant characteristics [33‐38]; (see also [39] for review). In a review of the literature, Bishop [40] suggested that inconsistencies in the reported findings may reflect a combination of methodological differences between studies, including differences in statistical power, participant characteristics, stimuli, and presentation rate, as well as factors associated with the measurement of the MMN itself (e.g. the specific time window used to define the MMN—and whether or not it overlaps with the LDN time window). These factors, combined with the poor reliability of the MMN at the individual level (e.g. [41, 42]), have led some researchers to question whether the MMN is truly the gold standard measure of auditory discrimination ability that it was initially hailed to be [43].

In addition to traditional MMN techniques which involve averaging ERPs, further information can be obtained by analysis of the component waveforms that underlie the MMN. The background electroencephalogram (EEG) comprises an ensemble of cortical oscillations across a range of different frequencies (from delta (0–4 Hz) through theta (4–7 Hz), alpha (8–12 Hz), beta (12–30), and gamma (30–100 Hz)). One explanation of the auditory sensory ERP is that following the presentation of a sound, cortical oscillatory activity is synchronized in phase with the incoming auditory signal (‘phase resetting’ [44]). Magnetoencephalography (MEG) and EEG studies have revealed an increase in phase locking of oscillatory activity to deviant auditory stimuli during the MMN time window, but confined to the theta range only [23, 45‐47]. In contrast, the LDN has been linked to an event-related desynchronization of oscillations extending across the delta, theta, and alpha ranges [48]. These responses showed strong developmental effects, with cortical synchronization in the theta band corresponding to the MMN increasing with age in typically developing children ranging from 7 years to adulthood [23]. In addition, synchronization in the theta band over the MMN time window was significantly correlated with behavioural thresholds on a measure of frequency discrimination [23] (c.f. [48]). Of particular relevance to the current study, Bishop et al. [48] found that children with SLI failed to show the expected event-related desynchronization during the LDN time window that was seen in typically developing controls. Moreover, the drop in event-related spectral power associated with the LDN to syllables was correlated with performance on a measure of nonword repetition (a measure of phonological processing and potential gold standard test for language impairment [49]), indicating that it was those children who were poor at nonword repetition that failed to show desynchronization.

The relationship between SLI and dyslexia is not clear cut (e.g. [50]). However, as many as 50% of children with SLI have reading difficulties [51], and both children with SLI (e.g. [52]) and children with dyslexia (e.g. [53]) are poor at nonword repetition. Given that dyslexia is also associated with difficulties with phonological processing (see [50] for review), we would expect to see similar patterns of brain activity in children who have experienced difficulties in learning to read.

In the current study, we compared mismatch responses to changes in auditory frequency in a group of children with dyslexia and a group of typically developing controls who were matched in age. We used a subset of the stimuli and an identical paradigm to that of Bishop and colleagues [23, 48], in which a repeated standard stimulus (a sinusoid of a set frequency) was interrupted by two rarer deviant sinusoids of different frequencies. We compared the responses of the two groups in both the MMN and LDN time windows, using both conventional techniques, and time-frequency analyses of spectral composition and phase locking. We asked: (1) Are children with dyslexia different from typically developing controls in their mismatch responses to frequency deviants during the MMN or LDN time windows when assessed using conventional analysis techniques? (2) Are children with dyslexia different from typically developing children in the spectral composition or in the extent of phase locking of the MMN or LDN? (3) Are any measures of the MMN or LDN related to behavioural measures of frequency discrimination and oral or written language in this group? Following Bishop et al. [48], we predicted that like children with SLI, children with dyslexia would show an MMN response that was no different from that of typically developing controls, but that their LDN would be reduced in amplitude, and that this would be particularly pronounced for the small deviants. Further, we predicted that children with dyslexia would show normal phase locking but reduced desynchronization of activity relative to controls. Finally, based on past research, it was difficult to predict whether any of the EEG measures would correlate with frequency discrimination or language/literacy [23, 48], but we hypothesised that any significant correlations would be restricted to between frequency discrimination and phase locking to small deviants, and between event-related desynchronization and nonword repetition.

The current study asked: Do children with dyslexia show differences in their mismatch responses to frequency deviants during the MMN time window using (1) conventional amplitude measures and (2) time-frequency analysis?, and (3) is performance on any of these measures linked to behavioural measures of frequency discrimination, language, and literacy? We found that children with dyslexia did not differ from controls in their MMN responses to the frequency deviants and, surprisingly, that they showed greater levels of ITC (phase locking) in the time window corresponding to the MMN, particularly in the older group. In contrast, for the LDN time window, the younger children with dyslexia showed a smaller LDN response to small deviants, whilst the older children with dyslexia showed a reduction in their event-related desynchronization to frequency deviants over the same timescale, in the form of a less negative ERSP. Neither the conventional nor time-frequency analyses scores correlated with performance on a behavioural frequency discrimination task. However, higher ITC scores for the MMN were associated with poorer performance on a test of nonword repetition.

Our failure to find a difference between the dyslexic and control groups on the average amplitude measure of the MMN adds to a growing number of studies that have reported similar results [28‐32]. However, as outlined earlier, the literature is by no means consistent as several studies have reported evidence for a reduced MMN to tone deviants in adults with a history of dyslexia (both compensated and noncompensated [25, 26]), whilst Lachmann and colleagues [37] reported the same in a group of children with dyslexia, but only in those who showed difficulties with word but not nonword reading (c.f. [24]). In contrast, Hugdahl et al. [27] found that the MMN was enhanced in a group of 25 children with dyslexia. Bishop [40] undertook an extensive review of the literature to better understand the reasons for inconsistencies in findings in the literature. She identified a number of relevant factors including differences in statistical power, stimuli, time windows, and presentation rates (see also [39] for a review). Our findings reinforce Bishop's [40] conclusions. First, regarding power, with a sample size of 20 in each group and 20 in each age band, our study was powered at 0.69 to detect a strong effect size, so it is possible a small effect could have been missed. However, the fact that we saw a trend for the dyslexic children to show a larger MMN than controls suggests that if such an effect did exist in our data, it was likely to be in the opposite direction than would be expected were deficits to exist at this stage of processing. Second, our findings support an increasing consensus that when significant group differences are detected, they are likely to arise from studies where the frequency difference between standards and deviants is small (<10%; [39, 40]). Although we did not find an effect of deviant size on the MMN, the younger dyslexic group showed a reduction in their LDN response to small deviants only. This brings us on to the question of time windows. To the extent that our study provided evidence for differences in late-stage auditory processing in dyslexia (as measured by the LDN and associated ERSP), it may be tempting to conclude that those studies reporting a difference between dyslexic and control samples may have been conflating the MMN and LDN time windows (see [48] for similar argument). However, close examination of the time windows reported in the above cited studies yields no evidence that this is the case. Nevertheless, that we were interested in the time window associated with the LDN may have contributed indirectly to our null results. Indeed, Bishop [40] noted a strong trend for significant differences in the MMN to tone deviants to arise from those studies using a stimulus onset asynchrony of 500 ms or less. Because we were also interested in late-stage processing, we did not do this. However, it is possible that differences in the MMN to tone deviants between dyslexic and control groups only arise when the frequency difference between standard and deviant stimuli is small and the presentation rate is fast.

Together with the MMN result, our finding that the dyslexic group showed an enhanced ITC to frequency deviants relative to controls suggests that children with dyslexia have no difficulty with the initial detection or discrimination of sound differences, nor in the precision of synchronization of cortical oscillatory activity corresponding to these abilities. This deduction is of particular interest in light of two recent theories which have attributed the reading difficulties associated with dyslexia to difficulties in cortical phase locking [2, 3]. Specifically, Goswami [3] argued that dyslexia arises from a reduction in oscillatory phase locking in auditory cortex to slower temporal modulations, in particular to delta and theta ranges (0.5–4 and 4–8 Hz, respectively). Giraud and Ramus [2] instead implicated a deficient low-gamma steady state response in the left auditory cortex. We could not ascertain whether the children in our sample had deficits in the phase locking of their gamma response, owing to the application of an online notch filter in our study. However, in as much as we showed evidence for enhanced phase locking at 4–7 Hz, our study indicates that children with dyslexia are unlikely to have a deficit in generating phase-locked oscillations in the theta range in response to steady-state (nonmodulating) auditory stimuli. It is also worth noting that our findings are not incompatible with those of two recent MEG studies, both of which examined phase-locking activity to non-steady-state (modulating) noise in two separate groups of adults with a childhood history of dyslexia [70, 71]. Lehongre et al. [71] found that their dyslexic group showed a reduction in cortical phase locking in the left hemisphere to acoustic modulations in the low gamma range (25–35 Hz). Poorer phase locking over this frequency range was linked to greater deficits in phonological processing and rapid naming. In contrast, Lehongre et al. [71] also found evidence for enhanced cortical entrainment to high gamma rates (>50 Hz) in their dyslexic sample, which was linked to a verbal memory deficit. Hämäläinen et al. [70] reported evidence for a differential pattern of phase locking at 2 Hz in the cortices of their dyslexic group; phase locking was reduced in the right hemispheres of their dyslexic subjects relative to those of controls, and the dyslexic group showed a more bilateral distribution of responses than normal readers. However, Hämäläinen et al. [70] failed to find any group differences at 4 Hz and, moreover, also reported evidence for enhanced phase locking in the left hemispheres of their dyslexic group at 10 and 20 Hz. Together with the findings of Hämäläinen et al. [70] and Lehongre and colleagues [71], our results suggest that if cortical phase locking is impaired in individuals with dyslexia or with a history of the disorder, this impairment is unlikely to be in the theta range.

The results of our study also shed further light upon what stage(s) of auditory processing is (are) likely to be impaired in dyslexia. Our analyses of the LDN and corresponding ERSP indicate that the difficulties children with dyslexia experience in processing deviations in auditory information may arise at a relatively late stage of processing (i.e. following initial detection and discrimination). However, our findings were not clear cut. Indeed, whereas the younger children with dyslexia showed a reduction in their LDN response to small deviants, the older children with dyslexia did not differ from controls in their LDN amplitude to either small or large deviants. Instead, this same group showed a reduced ERSP relative to the controls during the same time window, which was absent in the younger children with dyslexia. Clearly, these findings need to be replicated, and preferably over a wider age range than that studied here, in order to understand these developmental effects. However, the fact that we observed a reduction of broad, low frequency event-related desynchronization over this time window in the older children with dyslexia suggests that this group may maintain subtle differences in their late-stage auditory processing that can no longer be detected in the difference wave.

Inasmuch as our findings suggest that the auditory deficit in dyslexia occurs in late—but not early—stage processing, they are consistent with other recent findings from the literature [72]. Neuhoff et al. [72] measured the MMN and LDN to speech signals in children with dyslexia, their unaffected siblings, and controls [72]. They found that whilst the MMN was not significantly different between groups, both the children with dyslexia and their unaffected siblings showed an LDN that was significantly reduced relative to that of controls, a finding that was used to argue that the LDN may represent a neurophysiological endophenotype for dyslexia. Our findings are also consistent with those of Bishop et al. [48], who showed that children with SLI showed a reduction in their LDN response to small deviants and failed to show the event-related desynchronization during the LDN time window that was characteristic of age- and nonverbal IQ-matched controls. However, how far our findings indicate that this pattern may extend to children with reading but not oral language difficulties remains uncertain. Indeed, the children with dyslexia in our study had poorer scores than the controls on both the parental report questionnaire of communication (CCC-2) and on a test of nonword repetition, suggesting that at least some of them had oral language problems in addition to poor reading. This is hardly surprising, as many children with dyslexia also meet criteria for SLI, and vice versa [51, 73, 74], and indeed several theories purport a shared causal mechanism underlying both disorders (e.g. [3, 5, 50]). Nonetheless, our findings suggest that the pattern of reduced desynchronization of neural activity during late-stage processing at least extends to (older) children who have difficulties with learning to read in addition to oral language problems and that this may be a marker for developmental disorders of oral and written language.

Finally, our results add to an increasingly complicated picture regarding the relationship between auditory processing abilities (measured both behaviourally and electrophysiologically) and language and literacy outcomes in children. First, we failed to replicate the findings of Bishop et al. [23] for a positive correlation between ITC to small deviants and performance on psychophysical measures of frequency discrimination, in either the dyslexic or the control groups. Nevertheless, because the analyses were within-group, our sample sizes were small. However, it is noteworthy that Bishop et al. [48] equally did not find evidence for correlations between frequency discrimination thresholds and any of the electrophysiological measures they obtained in groups of typically developing children and children with SLI. We also failed to replicate the finding of Bishop et al. [48] for a significant correlation between nonword repetition and event-related desynchronization. Rather, the only correlation that remained significant in our study after controlling for multiple comparisons was between precision of phase locking during the MMN time window and nonword repetition, although this was driven by the combination of both higher ITC scores and lower nonword repetition scores in the dyslexia group.

In so far as auditory processing as measured in this study was not strongly associated with behavioural measures of oral and written language, our results question the causal link that has been purported to exist between auditory processing deficits, in this case frequency discrimination, and developmental disorders of language and literacy (e.g. [2‐5]). At the same time, it is increasingly clear that both children and adults with dyslexia (and indeed many with SLI) do experience difficulties with auditory processing, as measured using both behavioural and electrophysiological, as well as imaging techniques (for reviews, see [9, 40]). We have previously argued that the link between auditory processing deficits and developmental disorders of language may be noncausal, perhaps driven by the existence of a third factor [75]. However, another possibility that has been put forward is that rather than being a cause of reading/language impairments, deficits in auditory processing might instead be a consequence [76]. Using structural equation modeling, Bishop and colleagues [76] found that the relationship between auditory processing, family history, and phonological processing in SLI was best predicted by a model where family history fed into phonological processing, which then impacted upon auditory processing. According to this model, genetic risk factors may affect a child's ability to develop phonological categories, leading to changes in the way that sound is represented in the brain, as measured by the ERP. This would predict that auditory processing deficits and difficulties with language and literacy should co-exist (as they indeed do). However, it would not require poorer auditory processing to be associated with more severe behavioural symptoms, as a number of different genetic risk factors could lead to different profiles of behavioural strengths and weaknesses (see also [74]). Clearly, further research is needed to support this hypothesis. Large-scale longitudinal studies with a priori predictions would go a long way towards achieving this.

The current study demonstrates that children with dyslexia may show unimpaired cortical processing regarding the initial detection and/or discrimination of sound differences in the frequency domain. Rather, such auditory processing deficits may instead manifest in dyslexia at a relatively late stage of processing. These findings are consistent with the notion that a reduction in the LDN, or the broad, low-frequency event-related desynchronization associated with the LDN, may represent a neurophysiological endophenotype of dyslexia.