Abstract
Reading acquisition involves the integration of auditory and visual stimuli. Thus, low-level audiovisual multisensory integration might contribute to disrupted reading in developmental dyslexia. Although dyslexia is more frequently diagnosed in males and emerging evidence indicates that the neural basis of dyslexia might differ between sexes, previous studies examining multisensory integration did not evaluate potential sex differences nor tested its neural correlates. In the current study on 88 adolescents and young adults, we found that only males with dyslexia showed a deficit in multisensory integration of simple nonlinguistic stimuli. At the neural level, both females and males with dyslexia presented smaller differences in response to multisensory compared to those in response to unisensory conditions in the N1 and N2 components (early components of event-related potentials associated with sensory processing) than the control group. Additionally, in a subsample of 80 participants matched for nonverbal IQ, only males with dyslexia exhibited smaller differences in the left hemisphere in response to multisensory compared to those in response to unisensory conditions in the N1 component. Our study indicates that deficits of multisensory integration seem to be more severe in males than females with dyslexia. This provides important insights into sex-modulated cognitive processes that might confer vulnerability to reading difficulties.
Significance Statement
Although dyslexia is more frequent in males and the neural basis of dyslexia might differ to some extent between sexes, previous studies examining multisensory integration impairment did not evaluate potential sex differences. We observed that only males with dyslexia showed a deficit in multisensory integration of simple nonlinguistic stimuli. At the neural level, both females and males with dyslexia presented smaller differences in response to multisensory compared to those in response to unisensory conditions in the N1 and N2 components than the control group. Our findings are the first to indicate sex differences in deficits of multisensory integration in dyslexia, which seem to be more severe in males.
Introduction
Learning to read relies on forming multisensory associations between units of spoken language and arbitrary visual symbols. Several years of instruction are needed to master this process. However, around 10% of individuals with developmental dyslexia show persistent reading difficulties that cannot be explained by lower intellectual level, lack of educational opportunities, sensory impairments, or co-occurring neurological disorders (World Health Organization, 2019). In accordance with the multiple deficit model of neurodevelopmental disorders (Pennington, 2006; McGrath et al., 2020), dyslexia is recognized to arise from an interplay of various risk factors collectively contributing to compromised reading skills, rather than being attributed solely to a single deficit. Since acquiring reading involves the integration of auditory and visual stimuli, one such risk factor could be a deficit in multisensory integration, which might contribute to disrupted reading in dyslexia (Hahn et al., 2014).
Males are diagnosed with dyslexia more frequently than females (Quinn and Wagner, 2015) due to males’ lower mean and more variable reading performance (Arnett et al., 2017). Emerging research suggests potential sex differences in the neural basis of dyslexia (Altarelli et al., 2014; Evans et al., 2014; Müller-Axt et al., 2022), which may lead to variations in cognitive deficits. While the origins of these differences remain unknown, some theories implicate female sex hormones in protecting against disruptions in brain development (Geschwind and Galaburda, 1985). Additionally, there is evidence of genetic variants related to a lower risk of dyslexia in females but not in males (Gu et al., 2018). Although sex is often overlooked a potential explanatory factor for the heterogeneity of findings in dyslexia (Ramus et al., 2018) both at the behavioral and neural levels (Krafnick and Evans, 2019), no prior study on multisensory integration took into account potential sex differences.
Regarding general differences in nonlinguistic multisensory integration in dyslexia, previous research mostly focused on the evaluation of the temporal window of integration using temporal order judgment (TOJ) or simultaneity judgment (SJ) tasks. In both tasks, weaker temporal abilities are reflected in longer interstimulus intervals (ISIs) needed for correct classification (García-Pérez and Alcalá-Quintana, 2012; Colonius and Diederich, 2020). Individuals with dyslexia performed worse in audiovisual (TOJ task; Hairston et al., 2005), audiotactile (both TOJ and SJ tasks), and visuotactile (SJ task) conditions (Laasonen et al., 2002), suggesting that multisensory integration impairment might be generalized to various sensory modalities. However, the reported results are not always consistent, as it was also argued that differences in the temporal window of integration in dyslexia can be explained by disrupted processing of unisensory stimuli, rather than by the multisensory integration impairment (Gori et al., 2020).
Less affected by temporal processing abilities, the redundant-target effect (RTE) task relies on a simple detection of stimuli in multisensory compared to unisensory conditions. Faster reaction times (RTs) in multisensory conditions might stem from an early integration of these stimuli in sensory processing, which underlies speeded behavioral response (Miller, 1982). This paradigm until now was applied only once in participants with dyslexia, who presented smaller facilitation in RTs in multisensory conditions compared to those in controls, supporting the notion of disrupted multisensory integration (Harrar et al., 2014).
Event-related potentials (ERPs) studies on multisensory integration typically compare the neuronal response to the multisensory condition and the sum of the unisensory conditions in the P1-N1-P2-N2 complex, that is, early components related to sensory processing (Brandwein et al., 2011; Molholm et al., 2020). For syllables, the P1 component with a sub-additivity effect (sum of unisensory conditions eliciting greater neuronal response than multisensory condition) was found in the dyslexic but not in the control group (Kronschnabel et al., 2014). However, this difference could be driven by late neuronal responses to unimodal auditory and visual stimuli in the dyslexic group, rather than by intergroup differences in the multisensory integration (Kronschnabel et al., 2014). At the behavioral level, no group differences in mean RTs were observed, while facilitation of the RTs in the multisensory condition was not evaluated. In contrast, Francisco (2017) did not find any ERP differences between dyslexic and control groups using both the standard McGurk effect and the SJ task (with syllables), although behaviorally the dyslexic group performed worse in the visual condition in the McGurk task and had a wider temporal window of integration in the SJ task.
In a choice reaction task involving nonlinguistic stimuli (1,000 Hz tones and flashes, occurring alone or simultaneously) where male participants determined the simultaneous presentation of stimuli, males with dyslexia exhibited delayed P2 and P3 components in visual-alone and multisensory conditions (Breznitz and Meyler, 2003). However, direct comparisons of neural responses between the multisensory and the sum of unisensory conditions were not conducted. In terms of behavior, dyslexic males showed slower mean RTs than control males in the multisensory condition, although facilitation in RTs for the multisensory compared to unisensory conditions was not assessed (Breznitz and Meyler, 2003).
Inconsistencies between studies might be related to the differences in tasks, stimuli, or group composition (male/female ratio). In fact, in autism spectrum disorder (ASD), another neurodevelopmental disorder with a higher prevalence in males, severe multisensory speech processing deficits were found to be dependent on participants’ sex, with females less affected than males (Ross et al., 2015). Although sex differences in multisensory integration in dyslexia have not been directly tested before, a meta-analysis of unisensory and multisensory temporal processing indicated that studies in which participants were matched for sex yielded smaller effect sizes than studies that did not match for sex (Meilleur et al., 2020). Hence, in the current study, we aim to investigate these potential sex differences. To achieve this, we employ the RTE task, which has previously highlighted deficits in dyslexia (Harrar et al., 2014). Building on previous ASD research (Ross et al., 2015), we anticipate sex-dependent multisensory integration deficits, with dyslexic males showing pronounced impairments compared to females and typical readers of both sexes. These patterns should be also reflected in EEG signals.
Materials and Methods
Participants
We recruited 88 Polish adolescents and young adults—44 diagnosed with dyslexia (22 females, 22 males) aged between 15.2 and 24.89 years (M = 19.41, SD = 3.45) and 44 typical readers (22 females, 22 males) aged 15.09–24.95 years (M = 19.56, SD = 3.18). All participants were right-handed, born at term, without a history of neurological and/or psychiatric diagnosis and treatment, without hearing impairment, with normal or corrected-to-normal vision, and had IQ higher than 80 as assessed by the Polish version of the Abbreviated Battery of the Stanford-Binet Intelligence Scale-Fifth Edition (SB5; Roid et al., 2017). Only participants with a clinical diagnosis of dyslexia performed by the psychological and pedagogical counseling centers were qualified for the dyslexic group.
The study was approved by the institutional review board at the University of Warsaw, Poland (reference number 2N/02/2021). All participants (or their parents in the case of underaged participants) provided written informed consent and received monetary compensation (400 PLN, approximately 80 Euro) for their participation in the study.
A power analysis was performed in the G*Power (Faul et al., 2007), and based on the previous behavioral study on multisensory integration in dyslexia, which employed the RTE task (Harrar et al., 2014), a sample of 44 participants (22 per group) turned out to be required to obtain statistical power at 0.85 level (α = 0.05) to detect a large effect size (d = 0.83). We increased the sample to 88 participants to evaluate the interaction between group and sex (resulting in 22 participants in each subgroup).
The dyslexic and control groups were matched for age and family socioeconomic status based on the mother's and the father's years of education (Table 1). A univariate ANOVA revealed a significant effect of group [F(1,84) = 195.57, p < 0.001, η2p = 0.700] with higher scores in the dyslexic (M = 51.95, SD = 10.25) than those in the control group (M = 24.86, SD = 7.62) in the Polish version of the Adult Reading History Questionnaire (ARHQ-PL; Bogdanowicz et al., 2015) in which higher score implies a higher risk of dyslexia. Although all participants had typical IQ, there was a significant effect of group in the IQ scale [F(1,83) = 13.33, p < 0.001, η2p = 0.138], with lower IQ in the dyslexic (M = 102.95, SD = 11.87) than those in the control group (M = 111.52, SD = 9.97). Also, in the nonverbal subscale only, there was a significant effect of group [F(1,84) = 5.50, p = 0.021, η2p = 0.061], with lower scores in the dyslexic (M = 10.20, SD = 2.92) than those in the control group (M = 11.57, SD = 2.56). The effect of sex, as well as the interaction between group and sex, was not significant for any measure.
Due to the variance in nonverbal IQ between the dyslexic and control groups, we conducted a repetition of all analyses utilizing a subsample of 80 participants, which were rigorously matched for nonverbal IQ. The outcomes have been documented in the section titled “Results for a subsample of 80 participants (out of 88) matched in nonverbal IQ.” We chose to present results from the entire sample in the first place, followed by those from the nonverbal IQ-matched subsample for two reasons: (1) to provide complete transparency regarding our dataset and (2) based on a power analysis, which indicated that a sample size of 88 participants was necessary to detect the expected effect size. Thus, to ensure sufficient statistical power, in the first step, we conducted the analyses on the entire sample, despite the lack of matching in nonverbal IQ among participants.
Reading and reading-related tasks
We assessed participants’ reading skills using a wide variety of paper–pencil tasks. Reading speed was tested with a number of words and pseudowords accurately read in 1 min (Szczerbiński and Pelc-Pękała, 2013). Rapid automatized naming was evaluated with subtests of object, color, digit, and letter naming (Fecenec et al., 2013). Reading comprehension was assessed by 26 short sentences (e.g., “Lemons are yellow”, “A year has 7 months”), which participants read silently and had to mark whether a given sentence was true or false with time to complete the task as the outcome measure. Phonological processing was tested using a phoneme deletion task (Szczerbiński and Pelc-Pękała, 2013) in which participants had to say a word without a given phoneme (e.g., “farm” without “f”), and spoonerism tasks—switching phonemes and syllables between two words (Bogdanowicz et al., 2016). Orthographic awareness was assessed with 28 pairs of pseudowords in which one was written according to the Polish spelling rules, while the other was not, and the task was to choose which one was written correctly (Awramiuk and Krasowicz-Kupis, 2014). The outcome measure in this task was the accuracy–time ratio. We also assessed participants’ nonverbal selective attention and perception speed by the task requiring crossing out target digits (6 and 9) embedded among nontarget digits in 3 min (Ciechanowicz and Stańczak, 2006), as well as short-term and working memory using forward and backward conditions from the Digit Span subtest from the Wechsler Adult Intelligence Scale-Revised (WAIS-R; Wechsler, 1981).
Redundant-target effect task
Participants sat in a chair with their heads on a chin-rest in a dark, sound-attenuated, and electrically shielded room and performed a simple reaction time task, while continuous EEG was recorded. The task was prepared in the Presentation® software (Version 20.1, Neurobehavioral Systems, www.neurobs.com) and consisted of visual-alone, auditory-alone, and audiovisual conditions. In the visual-alone condition, a white flash was presented centrally on a black background (60 ms), while in the auditory-alone condition, a 1,000 Hz tone was presented binaurally through sound-isolating earphones (60 ms). The audiovisual condition consisted of the simultaneous presentation of auditory and visual stimuli. Participants were instructed to press a button as quickly as possible after a stimulus presentation with their right index finger. The same response button was used for all three conditions.
Stimuli for each condition were presented in a random order in 8 blocks of 48 trials, with 16 trials per condition in each block. Additionally, in each block, two catch trials with no stimulus presentation were randomly included to prevent anticipations. At the beginning of each trial, a white fixation cross was presented centrally for 500 ms followed by a blank screen with ISIs ranging randomly and equiprobably between 1,000 and 2,500 ms. After a stimuli presentation (60 ms), a blank screen was presented for 1,000 ms before starting the next trial. In total, the task consisted of 400 trials (128 trials per condition and 16 catch trials), and it took ∼25 min to complete it. To reduce fatigue and maintain alertness, 1 min breaks between blocks were introduced.
Electroencephalography acquisition
EEG was recorded from 62 scalp electrodes and 2 ear electrodes using the Brain Products system (actiCHamp Plus, Brain Products). Data were recorded in BrainVision Recorder Software (BrainVision Recorder, Ver. 1.22.0002, Brain Products) with a 500 Hz sampling rate. Electrodes were positioned in line with the extended 10-20 system, and electrode Cz served as an online reference. All electrodes’ impedances were kept below 10 kΩ.
Statistical analyses
Redundant-target effect task
Firstly, to evaluate the differences in RTs between conditions, median RTs from the task were analyzed. Next, Miller's race model inequality (RMI; Miller, 1982) was employed to assess the effect of multisensory integration. According to the RMI, behavioral facilitation in the multisensory compared to unisensory conditions can be explained in terms of simple probability summation of two unisensory inputs. Namely, faster RTs in the multisensory condition might arise from more inputs inducing a response from which the fastest one “wins the race.” In that case, the race model holds, and behavioral facilitation can be explained without interaction between sensory inputs. However, when the race model is violated, that is, RTs in the multisensory condition are faster than predicted by the RTs in unisensory conditions, it is assumed that two inputs interacted and multisensory integration emerged (Miller, 1982). In the analysis of the RMI, the upper limit of cumulative probability (CP) for RTs at a given latency in the multisensory condition is placed based on CP from RTs in unisensory conditions, and the race model is violated when the CP value in the multisensory condition exceeds the sum of CP values in unisensory conditions.
Data were analyzed following a step-by-step procedure described by Mahoney and Verghese (2019). Briefly, we did not apply any data-trimming procedures, and all omissions as well as RTs slower than 1,000 ms were set to infinity instead of excluding them from the analyses. Also, the race model significance testing was applied only to the violated portions of the difference wave between the actual and predicted values in our data, instead of a priori defined percentile range of RTs. The RMI permutation test with a kill-the-twin correction (which increases the power of detecting violation by taking into account responses in catch trials) was applied to determine whether there is a statistically significant violation of the race model over the range of RTs identified in our dataset (Inequality 8 R script available at Gondan and Minakata, 2016). Next, for each participant, area under the curve (AUC) was calculated for the previously determined significantly violated portion of the percentile bins of RTs. The AUC was used as a measure of the magnitude of the multisensory integration, with greater values indicating a greater magnitude of multisensory integration.
Due to a technical error, one participant (male from the control group) did not perform all task trials, and his logfiles were not recorded. However, as the vast majority of trials were presented (98 visual-alone, 100 auditory-alone, and 103 multisensory trials), we decided to include his data in the analyses. Behavioral data from that person were obtained from the EEG file based on the event markers’ timing written in the recorded signal. Responses in trials that were not presented were set to infinity.
EEG data
Offline, the EEG signal was analyzed in the EEGLAB (Delorme and Makeig, 2004) and ERPLAB (Lopez-Calderon and Luck, 2014). The data were filtered between 1.6 and 45 Hz [FIR filter, with transition bandwidth 1.6 Hz and cutoff frequencies (−6 dB) 0.8 and 45.8 Hz] and re-referenced to the average of both ear electrodes. The high-pass filter was set at 1.6 Hz based on previous ERP studies on multisensory integration (McCracken et al., 2019; Molholm et al., 2020) to avoid pre-stimulus slow anticipatory waves that would be otherwise represented twice in the SUM (visual-alone + auditory-alone) condition (Teder-Sälejärvi et al., 2002). Chunks of data recorded during breaks between blocks as well as bad channels were manually rejected. The number of rejected channels ranged between 0 and 4 (M = 0.33, SD = 0.78). Next, independent component analysis (ICA) artifact rejection was applied. Components were automatically labeled by ICLabel (Pion-Tonachini et al., 2019), and those identified as eye blinks, muscle activity, and channel noise were excluded.
Data were epoched between −100 ms and 500 ms to the stimuli onset, and an automatic rejection criterion of ±100 μV (peak-to-peak) was applied to exclude epochs with excessive amplitudes. Trials with omitted responses and with RTs slower than 1,000 ms were also excluded. The number of epochs retained in the analysis ranged between 53 and 128 (M = 113.45, SD = 17.90) in the visual-alone condition, 32–128 (M = 109.65, SD = 24.30) in the auditory-alone condition, and 34–128 (M = 110.98, SD = 22.71) in the multisensory condition. Previously rejected bad channels were interpolated using the nearest neighbor spline (Perrin et al., 1987, 1989).
ERPs were computed by averaging epochs according to the stimuli condition (visual-alone, auditory-alone, multisensory), and the SUM condition was created by summing ERPs in the visual-alone and auditory-alone conditions. The difference wave between multisensory and SUM conditions (multisensory—SUM) was also calculated. Comparison of responses to multisensory and SUM conditions is typical in ERP studies on multisensory integration (Teder-Sälejärvi et al., 2002; Brandwein et al., 2011; McCracken et al., 2019; Molholm et al., 2020) and is based on the principle of linear summation of electrical fields. Thus, any divergence between neural responses in multisensory and SUM conditions indicates that the inputs were processed differently when presented simultaneously or not.
Global field power (GFP) was calculated for the grand average of all conditions, that is, visual-alone, auditory-alone, and multisensory, to choose appropriate time windows for the analysis. GFP measures the spatial standard deviation of the scalp potential across the electrodes. A local maximum of the GFP corresponds to a given distribution of electrical activity over the scalp (a microstate); a pass between the maxima in the curve corresponds to a reorganization of the distribution; thus, the occurrence of GFP maxima in time indicates latencies of evoked potentials (Skrandies, 1990). For each time window determined from the GFP, we identified the region where the microstate, averaged across conditions, had a maximal amplitude. Next, we checked for the differences between conditions at regions of interest showing the highest activity on average. For every time window, AUC and 50% fractional area latency were calculated separately for visual-alone, auditory-alone, multisensory, and SUM conditions. Moreover, AUC was also calculated for the difference wave (multisensory—SUM). We decided to quantify ERPs amplitude based on the AUC instead of mean amplitude since it allows us to find the area for either positive or negative regions at a given time window, which prevents from positive and negative effects canceling each other out at broad measurement windows (Lopez-Calderon and Luck, 2014).
For some participants, the algorithm did not find the specific component in a given time window (typically the P1 component in unisensory conditions); thus, it was impossible to calculate the 50% fractional area latency. In the reported analyses, it is reflected in a lower number of degrees of freedom.
Results
Behavioral results
Reading and reading-related tasks
Results from reading and reading-related tasks are provided in Table 2. For every task, we performed separate univariate ANOVA and tested the effect of group, sex, and the interaction between group and sex. In all tasks, there was a significant effect of group with the dyslexic group performing worse than the control group. Furthermore, there was a significant effect of sex in the orthographic awareness task [F(1,84) = 6.06, p = 0.016, η2p = 0.067] with females (M = 0.47, SD = 0.17) achieving higher scores than those of males (M = 0.40, SD = 0.16). Nevertheless, this difference did not retain significance following correction for multiple comparisons (Table 2). The interaction between group and sex was not significant in any task.
Redundant-target effect task
In terms of median RTs from the task, one participant (female from the control group) was identified as an outlier based on median RTs exceeding three SD in all conditions and thus was excluded from this analysis. A 3 × 2 × 2 (condition, group, sex) repeated measure ANOVA revealed a main effect of condition [F(1.69, 140.14) = 198.41, p < 0.001, η2p = 0.705, Greenhouse–Geisser corrected] with post hoc comparisons indicating that RTs in multisensory condition (M = 239.03, SD = 37.70) were faster compared to those in both auditory (M = 269.62, SD = 50.25, p < 0.001, Bonferroni corrected) and visual (M = 281.02, SD = 42.22, p < 0.001, Bonferroni corrected) conditions. Additionally, RTs in the auditory condition were faster than RTs in the visual condition (p < 0.001, Bonferroni corrected). The effects of group, sex, and any interactions were not significant.
In terms of the RMI, data from all participants were included as the RMI does not depend on median RTs. Also, all participants had accuracy above 70% in every condition; thus, no one was excluded from the analyses based on high omission rates (Mahoney and Verghese, 2019). In the whole sample, the race model turned out to be violated within the first four quantiles (15%) of the distribution of the RTs (Fig. 1), and this violation was significant (p < 0.001) as assessed by the RMI permutation test with the kill-the-twin correction (Inequality 8; Gondan and Minakata, 2016).
Next, for each participant, AUC was calculated for the first four quantiles (15%) of RTs. The AUC was used as a measure of the magnitude of the multisensory integration, with greater values indicating a greater magnitude. Since age turned out to be positively correlated with AUC (r = 0.38, p < 0.001), we performed a univariate ANCOVA with age as a covariate and tested the effect of group and sex. Results indicated significant effects of age [F(1,83) = 15.75, p < 0.001, η2p = 0.159], group [F(1,83) = 4.46, p = 0.038, η2p = 0.051] with a lower magnitude of multisensory integration in the dyslexic (M = 0.04, SD = 0.16) than that in the control group (M = 0.12, SD = 0.17), and an interaction between group and sex [F(1,83) = 4.55, p = 0.036, η2p = 0.052]. Post hoc comparisons indicated that males from the dyslexic group had a lower magnitude of multisensory integration (M = −0.01, SD = 0.11) than males from the control group (M = 0.13, SD = 0.22, p = 0.004, Bonferroni corrected), while the difference between females from the dyslexic (M = 0.10, SD = 0.19) and control (M = 0.10, SD = 0.12) groups was not significant (p = 0.989, Bonferroni corrected). Moreover, the difference between females and males was significant in the dyslexic group (p = 0.020, Bonferroni corrected), but not within the control group (p = 0.525, Bonferroni corrected). This suggests that the observed reduced level of multisensory integration among dyslexic males, as opposed to dyslexic females, cannot be solely attributed to control males outperforming control females. The differences between actual and predicted values separately for females and males are presented in Figure 2.
EEG results
Based on the GFP curve (Fig. 3), three time windows for the analysis were chosen: 56–82 ms, 98–196 ms, and 196–384 ms. For each time window, the following electrodes were selected from the greatest mean amplitude topography (Fig. 3) and averaged for the analyses: for the 56–82 ms (P1 component), three central electrodes (FC1, FCz, FC2); for the 98–196 ms (N1 component), eight electrodes in the left hemisphere (F5, FC5, FC3, FC1, C3, C5, CP1, CP5) and corresponding eight electrodes in the right hemisphere (F6, FC6, FC4, FC2, C4, C6, CP2, CP6); and for the 196–384 ms (N2 component), six electrodes in the left hemisphere (AF7, AF3, F7, F5, F3, F1) and corresponding six in the right hemisphere (AF8, AF4, F8, F6, F4, F2).
Firstly, separate univariate ANOVAs were conducted in the visual-alone and auditory-alone conditions for the AUC and 50% fractional area latency in the P1 component (area of the positive region specified) to test the effect of group and sex. In the N1 and N2 components, separate repeated measures ANOVAs were conducted for the AUC and 50% fractional area latency (area of the negative region specified) to test the effect of group, sex, and hemisphere. Next, to analyze ERPs in multisensory and SUM conditions, repeated measure ANOVAs were conducted to test the effect of condition, group, and sex in the P1 component and the effect of condition, group, sex, and hemisphere for the N1 and N2 components separately for the AUC and 50% fractional area latency. Additionally, for the AUC, similar repeated measures ANOVAs were also conducted for the difference wave (multisensory—SUM) with effects of group, sex, and hemisphere being tested.
Visual-alone condition
56–82 ms (P1 component)
There were no significant effects or interactions either for AUC or 50% fractional area latency.
98–196 ms (N1 component)
For the AUC, there were no significant effects or interactions. For the 50% fractional area latency, there was a significant effect of sex [F(1,84) = 5.25, p = 0.024, η2p = 0.059] with the N1 component occurring later in males (M = 150.77, SD = 11.92) compared to females (M = 145.12, SD = 11.17). Any other effects or interactions were not significant.
196–384 ms (N2 component)
For the AUC, there was a significant effect of hemisphere [F(1,84) = 12.76, p < 0.001, η2p = 0.132] with greater AUC in the left (M = 0.90, SD = 0.83) than that in the right hemisphere (M = 0.84, SD = 0.74). Any other effects or interactions were not significant.
For the 50% fractional area latency, there was a significant effect of hemisphere [F(1,82) = 16.13, p < 0.001, η2p = 0.164] with the N2 component occurring later in the left (M = 291.36, SD = 22.60) than in the right hemisphere (M = 286.55, SD = 23.29) and the effect of group [F(1,82) = 4.96, p = 0.029, η2p = 0.057] with the N2 component occurring later in the dyslexic (M = 294.24, SD = 22.79) compared to the control group (M = 283.68, SD = 20.68). Any other effects or interactions were not significant.
Auditory-alone condition
56–82 ms (P1 component)
For the AUC, there were no significant effects or interactions. For the 50% fractional area latency, there was a significant effect of sex [F(1,76) = 5.54, p = 0.021, η2p = 0.068] with the P1 component occurring later in males (M = 69.88, SD = 4.95) than in females (M = 67.32, SD = 5.06). Any other effects or interactions were not significant.
98–196 ms (N1 component)
For the AUC, there were no significant effects or interactions. For the 50% fractional area latency, there was a significant effect of hemisphere [F(1,84) = 4.35, p = 0.040, η2p = 0.049] with the N1 component occurring later in the left (M = 144.42, SD = 10.63) than in the right hemisphere (M = 142.97, SD = 9.86). Any other effects or interactions were not significant.
196–384 ms (N2 component)
For the AUC, there was a significant effect of hemisphere [F(1,84) = 22.90, p < 0.001, η2p = 0.214] with greater AUC in the left (M = 0.50, SD = 0.38) than that in the right hemisphere (M = 0.42, SD = 0.31). Any other effects or interactions were not significant. For the 50% fractional area latency, there were no significant effects or interactions.
Multisensory and SUM conditions
ERPs for multisensory and SUM conditions separately for the dyslexic and control groups in four selected electrodes (F5, F6, FC1, FC2) are presented in Figure 4.
56–82 ms (P1 component)
For the AUC, there was a significant effect of condition [F(1,84) = 4.64, p = 0.034, η2p = 0.052] with greater AUC in the SUM (M = 0.05, SD = 0.04) than that in the multisensory condition (M = 0.04, SD = 0.03). Any other effects or interactions were not significant.
For the 50% fractional area latency, there was a significant effect of condition [F(1,73) = 22.83, p < 0.001, η2p = 0.238] with P1 component occurring later in the SUM (M = 70.87, SD = 4.95) than in the multisensory condition (M = 67.85, SD = 4.80). Any other effects or interactions were not significant.
98–196 ms (N1 component)
For the AUC, there was a significant effect of condition [F(1,84) = 43.27, p < 0.001, η2p = 0.340] with greater AUC in the SUM (M = 0.84, SD = 0.33) than that in the multisensory condition (M = 0.75, SD = 0.30), the effect of sex [F(1,84) = 6.88, p = 0.010, η2p = 0.076] with greater AUC in females (M = 0.88, SD = 0.31) compared to that in males (M = 0.71, SD = 0.29), and an interaction between group and condition [F(1,84) = 8.31, p = 0.005, η2p = 0.090]. Post hoc comparisons indicated that the greater AUC in the SUM compared to that in the multisensory condition was found both in the dyslexic (MSUM = 0.80, SDSUM = 0.35, Mmultisensory = 0.75, SDmultisensory = 0.35, p = 0.011, Bonferroni corrected) and the control group (MSUM = 0.88, SDSUM = 0.32. Mmultisensory = 0.75, SDmultisensory = 0.25, p < 0.001, Bonferroni corrected). The difference between groups was not significant either in the multisensory (p = 0.980, Bonferroni corrected) or SUM conditions (p = 0.260, Bonferroni corrected).
The analysis for the difference wave (multisensory—SUM) for the AUC revealed the significant effect of group [F(1,84) = 5.99, p = 0.016, η2p = 0.067] with a greater difference in AUC between multisensory and SUM conditions in the control (M = 0.21, SD = 0.12) compared to that in the dyslexic group (M = 0.15, SD = 0.08). Any other effects or interactions were not significant.
For the 50% fractional area latency, there was a significant effect of condition [F(1,84) = 9.50, p = 0.003, η2p = 0.102] with the N1 component occurring later in the multisensory (M = 147.02, SD = 9.43) than in the SUM condition (M = 145.20, SD = 8.76) and the effect of sex [F(1,84) = 8.36, p = 0.005, η2p = 0.090] with the N1 component occurring later in males (M = 148.69, SD = 9.17) compared to females (M = 143.54, SD = 7.42). Any other effects or interactions were not significant.
196–384 ms (N2 component)
For the AUC, there was a significant effect of condition [F(1,84) = 18.15, p < 0.001, η2p = 0.178] with greater AUC in the SUM (M = 1.23, SD = 1.02) compared to that in the multisensory condition (M = 1.08, SD = 0.92), the effect of hemisphere [F(1,84) = 36.18, p < 0.001, η2p = 0.301] with greater AUC in the left (M = 1.25, SD = 1.06) than that in the right hemisphere (M = 1.06, SD = 0.86), and interactions between hemisphere and condition [F(1,84) = 5.67, p = 0.019, η2p = 0.063] as well as between group and condition [F(1,84) = 6.26, p = 0.014, η2p = 0.069]. For the hemisphere*condition interaction, post hoc comparisons indicated that the greater AUC in the SUM than that in the multisensory condition was found both in the left (MSUM = 1.32, SDSUM = 1.12, Mmultisensory = 1.19, SDmultisensory = 1.04, p < 0.001, Bonferroni corrected) and right hemispheres (MSUM = 1.15, SDSUM = 0.93, Mmultisensory = 0.97, SDmultisensory = 0.82, p < 0.001, Bonferroni corrected). Also, there was a greater AUC for both SUM and multisensory conditions in the left compared to the right hemisphere (p < 0.001, Bonferroni corrected). For the group × condition interaction, post hoc comparisons indicated that the greater AUC for the SUM (M = 1.34, SD = 1.09) compared to that in the multisensory condition (M = 1.09, SD = 0.97, p < 0.001, Bonferroni corrected) was found in the control group, while it was not significant in the dyslexic group (MSUM = 1.13, SDSUM = 0.94, Mmultisensory = 1.07, SDmultisensory = 0.88, p = 0.217, Bonferroni corrected). The difference between groups was not significant either in the multisensory (p = 0.897, Bonferroni corrected) or SUM (p = 0.350, Bonferroni corrected) conditions.
The analysis for the difference wave (multisensory—SUM) for the AUC revealed the significant effect of hemisphere [F(1,84) = 5.75, p = 0.019, η2p = 0.064] with a greater difference in AUC between multisensory and SUM conditions in the right (M = 0.37, SD = 0.27) compared to that in the left hemisphere (M = 0.34, SD = 0.25) and the effect of group [F(1,84) = 5.80, p = 0.018, η2p = 0.065] with a greater difference in the control (M = 0.42, SD = 0.29) than that in the dyslexic group (M = 0.29, SD = 0.20). Any other effects or interactions were not significant.
For the 50% fractional area latency, there was a significant effect of group [F(1,84) = 5.85, p = 0.018, η2p = 0.065] with the N2 component occurring later in the dyslexic (M = 294.44, SD = 21.47) compared to the control group (M = 284.12, SD = 18.40) and interaction between condition, hemisphere, and sex [F(1,84) = 4.28, p = 0.042, η2p = 0.048]. However, post hoc comparisons did not reveal any significant differences. Any other effects or interactions were not significant.
Correlations between reading speed and multisensory integration
In the last step of the analysis, we evaluated correlations between reading speed, the magnitude of multisensory integration at the behavioral level, and ERP components related to multisensory processing. We employed pseudowords read per minute as the sole measure for reading speed due to its robust and consistent ability to predict reading difficulties in adults surpassing other reading and reading-related measures (Reis et al., 2020; Carioti et al., 2021; Brèthes et al., 2022). For the sake of clarity, we averaged values obtained from the left and right hemispheres in N1 and N2 components. Apart from two-tailed correlations performed in the whole sample (Table 3), we also performed one-tailed correlations separately for females and males (Table 4) to validate relationships obtained in the whole sample. One-tailed statistics were employed in separate correlations for each sex, as the correlation directions had been already established through the two-tailed correlations conducted on the entire sample. Nonparametric Spearman's correlations were performed due to violations of the normal distribution. We presented correlation matrices using both uncorrected p-values and values adjusted for multiple correlations at p < 0.0056. These adjustments were made for nine planned comparisons between reading speed and the behavioral and neural indices of multisensory integration.
In the entire sample, it turned out that the magnitude of multisensory integration was positively correlated with pseudowords read in 1 min [r(86) = 0.22, p = 0.040] and negatively with 50% fractional area latency in the N1 component in the multisensory condition [r(86) = −0.22, p = 0.041]. Additionally, the latency of the N2 component in the multisensory condition displayed a negative correlation with pseudowords read in 1 min [r(86) = −0.31, p = 0.004]. However, upon adjusting the p-value for multiple correlations involving both behavioral and neural measures, the only significant result that remained was the negative correlation between N2 latency and pseudowords read in 1 min.
In female participants, the magnitude of multisensory integration negatively correlated with the AUC of the difference wave in the N2 component [r(42) = −0.25, p = 0.050]. Moreover, the 50% fractional area latency of the N2 component in the multisensory condition showed a negative correlation with pseudowords read in 1 min [r(42) = −0.41, p = 0.003], and this correlation withstood correction for multiple comparisons. On the other hand, in male participants, the magnitude of multisensory integration positively correlated with pseudowords accurately read in 1 min [r(42) = 0.29, p = 0.027] and negatively with the 50% fractional area latency in the N2 component in the multisensory condition [r(42) = −0.29, p = 0.029]. Moreover, in males, the AUC of the difference wave in the N1 component positively correlated with pseudowords accurately read in 1 min [r(42) = 0.33, p = 0.015]. However, in males, none of these relationships retained significance after adjusting the p-value for multiple correlations involving both behavioral and neural measures.
Results for a Subsample of 80 Participants (Out of 88) Matched in Nonverbal IQ
Participants
Similar to the entire sample, the dyslexic and control groups were equated in terms of age, sex, and family socioeconomic status. Additionally, groups were also matched for nonverbal IQ (Table 5).
Behavioral results
Reading and reading-related tasks
Parallel to the entire sample, the dyslexic group performed worse than the control group in all reading and reading-related tasks (Table 6). In contrast to the entire sample, the difference between groups in a perception speed task did not retain significance after correction for multiple comparisons.
Redundant-target effect task
All results were replicated. In terms of median RTs from the task, similarly to the analysis on the whole sample, one participant (female from the control group) was excluded from this analysis based on median RTs exceeding three SD in all conditions. A repeated measure ANOVA revealed a main effect of condition [F(1.72, 129.03) = 188.64, p < 0.001, η2p = 0.716, Greenhouse–Geisser corrected]. RTs in multisensory condition (M = 238.92, SD = 37.08) were faster than those in auditory (M = 269.45, SD = 49.05, p < 0.001, Bonferroni corrected) and visual (M = 281.36, SD = 41.96, p < 0.001, Bonferroni corrected) conditions. Additionally, RTs in auditory condition were faster than those in visual condition (p < 0.001, Bonferroni corrected). The effects of group, sex, and interactions were not significant.
In terms of the RMI, a univariate ANCOVA with age as a covariate revealed significant effects of age [F(1,75) = 13.14, p < 0.001, η2p = 0.149] and group [F(1,75) = 3.97, p = 0.050, η2p = 0.050] with a lower magnitude of multisensory integration in the dyslexic (M = 0.05, SD = 0.17) than that in the control group (M = 0.12, SD = 0.18). Additionally, an interaction between group and sex [F(1,75) = 4.66, p = 0.034, η2p = 0.059] indicated that males from the dyslexic group had a lower magnitude of multisensory integration (M = −0.01, SD = 0.12) than males from the control group (M = 0.13, SD = 0.23, p = 0.004, Bonferroni corrected). The difference between females from the dyslexic (M = 0.10, SD = 0.20) and control (M = 0.10, SD = 0.12) groups was not significant (p = 0.905, Bonferroni corrected). The difference between females and males was significant in the dyslexic group (p = 0.022, Bonferroni corrected), while not in the control group (p = 0.478, Bonferroni corrected).
EEG Results
Visual-alone condition
56–82 ms (P1 component)
Similar to the entire sample, no significant effects or interactions either for AUC or 50% fractional area latency were found.
98–196 ms (N1 component)
We replicated the results from the entire sample. For the AUC, there were no significant effects or interactions. For the 50% fractional area latency, a significant effect of sex [F(1,76) = 5.81, p = 0.018, η2p = 0.071] was found with the N1 component occurring later in males (M = 151.72, SD = 11.93) compared to females (M = 145.37, SD = 11.54). Any other effects or interactions were not significant.
196–384 ms (N2 component)
All results were replicated. For the AUC, a significant effect of hemisphere [F(1,76) = 12.70, p < 0.001, η2p = 0.143] was found with greater AUC in the left (M = 0.90, SD = 0.80) than that in the right hemisphere (M = 0.83, SD = 0.72). Any other effects or interactions were not significant.
For the 50% fractional area latency, a significant effect of hemisphere [F(1,74) = 12.71, p < 0.001, η2p = 0.147] was found with the N2 component occurring later in the left (M = 289.51, SD = 20.82) than in the right hemisphere (M = 285.38, SD = 20.84) and the effect of group [F(1,74) = 7.16, p = 0.009, η2p = 0.088] with the N2 component occurring later in the dyslexic (M = 293.41, SD = 18.77) compared to the control group (M = 281.49, SD = 20.02). Any other effects or interactions were not significant.
Auditory-alone condition
56–82 ms (P1 component)
We replicated the results from the entire sample. For the AUC, no significant effects or interactions were found. For the 50% fractional area latency, a significant effect of sex [F(1,68) = 7.49, p = 0.008, η2p = 0.099] was found with the P1 component occurring later in males (M = 70.06, SD = 5.16) than in females (M = 66.85, SD = 4.96). Any other effects or interactions were not significant.
98–196 ms (N1 component)
In contrast to the analysis on the entire sample, which revealed a significant effect of hemisphere for both AUC and 50% fractional area latency, no significant effects or interactions were found either for AUC or 50% fractional area latency.
196–384 ms (N2 component)
Similar to the entire sample, for the AUC, there was a significant effect of hemisphere [F(1,76) = 21.09, p < 0.001, η2p = 0.217] with greater AUC in the left (M = 0.48, SD = 0.36) than that in the right hemisphere (M = 0.41, SD = 0.31). Any other effects or interactions were not significant. For the 50% fractional area latency there were no significant effects or interactions.
Multisensory and SUM conditions
56–82 ms (P1 component)
In contrast to the outcomes obtained from the entire sample, the effect of condition for the AUC was not significant [F(1,76) = 2.06, p = 0.155, η2p = 0.026]. Instead, a significant interaction between group and condition was observed [F(1,76) = 5.31, p = 0.024, η2p = 0.065], with greater AUC in the SUM condition (M = 0.05, SD = 0.04) compared to that in the multisensory condition (M = 0.03, SD = 0.03, p = 0.010, Bonferroni corrected) significant in the dyslexic group, though not in the control group (MSUM = 0.04, SDSUM = 0.04, Mmultisensory = 0.04, SDmultisensory = 0.04, p = 0.540, Bonferroni corrected). The difference between groups was not significant either in the multisensory (p = 0.394, Bonferroni corrected) or SUM conditions (p = 0.292, Bonferroni corrected). Any other effects or interactions were not significant.
For the 50% fractional area latency, we replicated the results from the entire sample. There was a significant effect of condition [F(1,65) = 16.85, p < 0.001, η2p = 0.206] with the P1 component occurring later in the SUM (M = 70.82, SD = 5.14) than in the multisensory condition (M = 67.97, SD = 4.92). Any other effects or interactions were not significant.
98–196 ms (N1 component)
For the AUC, all effects were replicated. There was a significant effect of condition [F(1,76) = 38.37, p < 0.001, η2p = 0.335] with greater AUC in the SUM (M = 0.83, SD = 0.34) than that in the multisensory condition (M = 0.74, SD = 0.31), the effect of sex [F(1,76) = 5.78, p = 0.019, η2p = 0.071] with greater AUC in females (M = 0.87, SD = 0.31) compared to that in males (M = 0.70, SD = 0.30), and an interaction between group and condition [F(1,76) = 6.18, p = 0.015, η2p = 0.075]. Post hoc comparisons indicated that the greater AUC in the SUM compared to that in the multisensory condition was found both in the dyslexic (MSUM = 0.79, SDSUM = 0.36, Mmultisensory = 0.74, SDmultisensory = 0.36, p = 0.011, Bonferroni corrected) and the control group (MSUM = 0.87, SDSUM = 0.32. Mmultisensory = 0.74, SDmultisensory = 0.24, p < 0.001, Bonferroni corrected). The difference between groups was not significant either in the multisensory (p = 0.941, Bonferroni corrected) or SUM conditions (p = 0.286, Bonferroni corrected).
The analysis for the difference wave (multisensory—SUM) for the AUC also replicated the effect of group [F(1,76) = 4.47, p = 0.038, η2p = 0.056] with a greater difference in AUC between multisensory and SUM conditions in the control (M = 0.21, SD = 0.12) compared to that in the dyslexic group (M = 0.16, SD = 0.08). Additionally, in contrast to the analyses on the whole sample, an interaction between hemisphere, group, and sex [F(1,76) = 4.68, p = 0.034, η2p = 0.058] was found. Post hoc comparisons indicated that males from the control group (M = 0.21, SD = 0.15) had a greater difference in AUC than males from the dyslexic group in the left hemisphere (M = 0.13, SD = 0.08, p = 0.028, Bonferroni corrected; Fig. 5), while in the right hemisphere this difference did not reach significance (MCON = 0.20, SDCON = 0.12, MDYS = 0.13, SDDYS = 0.09, p = 0.052, Bonferroni corrected). The differences between females from the control and the dyslexic group were not significant either in the left (MCON = 0.19, SDCON = 0.13, MDYS = 0.19, SDDYS = 0.10, p = 0.874, Bonferroni corrected) or the right hemisphere (MCON = 0.22, SDCON = 0.12, MDYS = 0.18, SDDYS = 0.07, p = 0.151, Bonferroni corrected). Moreover, only females from the control group had a greater difference in AUC in the right than that in the left hemisphere (p = 0.034, Bonferroni corrected). Any other effects or interactions were not significant.
For the 50% fractional area latency, all results were replicated. There was a significant effect of condition [F(1,76) = 6.66, p = 0.012, η2p = 0.081] with the N1 component occurring later in the multisensory (M = 147.05, SD = 9.32) than in the SUM condition (M = 145.43, SD = 8.71) and the effect of sex [F(1,76) = 8.49, p = 0.005, η2p = 0.100] with the N1 component occurring later in males (M = 148.93, SD = 9.16) compared to females (M = 143.55, SD = 7.11). Any other effects or interactions were not significant.
196–384 ms (N2 component)
We replicated all results. For the AUC, there was a significant effect of condition [F(1,76) = 16.77, p < 0.001, η2p = 0.181] with greater AUC in the SUM (M = 1.21, SD = 0.97) compared to that in the multisensory condition (M = 1.06, SD = 0.91), the effect of hemisphere [F(1,76) = 32.16, p < 0.001, η2p = 0.297] with greater AUC in the left (M = 1.23, SD = 1.03) than that in the right hemisphere (M = 1.04, SD = 0.84), and interactions between hemisphere and condition [F(1,76) = 6.14, p = 0.015, η2p = 0.075] as well as and between group and condition [F(1,76) = 5.64, p = 0.020, η2p = 0.069]. For the hemisphere × condition interaction, post hoc comparisons indicated that the greater AUC in the SUM than that in the multisensory condition was found both in the left (MSUM = 1.30, SDSUM = 1.06, Mmultisensory = 1.16, SDmultisensory = 1.02, p < 0.001, Bonferroni corrected) and the right hemisphere (MSUM = 1.13, SDSUM = 0.90, Mmultisensory = 0.96, SDmultisensory = 0.81, p < 0.001, Bonferroni corrected). Also, there was a greater AUC for both SUM and multisensory conditions in the left compared to that in the right hemisphere (p < 0.001, Bonferroni corrected). For the group × condition interaction, post hoc comparisons indicated that the greater AUC for the SUM (M = 1.26, SD = 1.00) compared to that in the multisensory condition (M = 1.02, SD = 0.92, p < 0.001, Bonferroni corrected) was found in the control group, while it was not significant in the dyslexic group (MSUM = 1.16, SDSUM = 0.95, Mmultisensory = 1.10, SDmultisensory = 0.91, p = 0.228, Bonferroni corrected). The difference between groups was not significant either in the multisensory (p = 0.728, Bonferroni corrected) or SUM (p = 0.640, Bonferroni corrected) conditions.
The analysis for the difference wave (multisensory—SUM) for the AUC also replicated the results from the entire sample. There was a significant effect of hemisphere [F(1,76) = 5.46, p = 0.022, η2p = 0.067] with a greater difference in AUC in the right (M = 0.37, SD = 0.27) compared to that in the left hemisphere (M = 0.34, SD = 0.24) and the effect of group [F(1,76) = 4.53, p = 0.037, η2p = 0.056] with a greater difference in the control (M = 0.41, SD = 0.28) than that in the dyslexic group (M = 0.29, SD = 0.20). Any other effects or interactions were not significant.
For the 50% fractional area latency, parallel to the results from the entire sample, a significant effect of group [F(1,76) = 5.18, p = 0.026, η2p = 0.064] was observed with the N2 component occurring later in the dyslexic (M = 293.12, SD = 21.92) compared to the control group (M = 282.82, SD = 18.29) and an interaction between condition, hemisphere, and sex [F(1,76) = 4.80, p = 0.032, η2p = 0.059]. However, post hoc comparisons did not reveal any significant differences. Any other effects or interactions were not significant.
Correlations between reading speed and multisensory integration
For both females and males (Table 7), a positive correlation between pseudoword reading speed and the magnitude of multisensory integration [r(78) = 0.23, p = 0.044] and a negative correlation between pseudoword reading speed and the latency of the N2 component in the multisensory condition [r(78) = −0.29, p = 0.009] were replicated. However, unlike the results obtained from the entire sample, the latter correlation did not retain significance upon adjusting the p-value for nine planned correlations at p < 0.0056. Furthermore, in contrast to the results obtained in the whole sample, the correlation between latency in the N1 component in the multisensory condition and the magnitude of multisensory integration was not significant [r(78) = −0.19, p = 0.085].
Within the female subgroup (Table 8), a negative relationship between pseudoword reading speed and the latency of the N2 component in the multisensory condition [r(38) = −0.39, p = 0.006] was replicated. Nonetheless, this relationship did not withstand correction for multiple comparisons. Also, a correlation between the magnitude of multisensory integration and the AUC of the difference wave in the N2 component was not significant [r(38) = −0.26, p = 0.052].
Among the male participants only (Table 8), all results were replicated. The magnitude of multisensory integration positively correlated with pseudowords reading speed [r(38) = 0.31, p = 0.028] and negatively correlated with the latency of the N2 component in the multisensory condition [r(38) = −0.28, p = 0.038]. Additionally, the AUC of the difference wave in the N1 component positively correlated with pseudoword reading speed [r(38) = 0.39, p = 0.006]. Also, none of these relationships retained significance after adjusting the p-value for multiple correlations.
Discussion
The aim of the current study was to assess the effect of sex on multisensory integration in dyslexia using simple nonlinguistic stimuli. As the neural basis of dyslexia differs to some extent between sexes (Altarelli et al., 2014; Evans et al., 2014; Müller-Axt et al., 2022), this could translate to the differences in manifestations of cognitive deficits. We expected to find greater impairment in multisensory integration in males with dyslexia compared to that in other groups, a pattern previously observed in ASD (Ross et al., 2015).
When evaluating gain in RTs in multisensory relative to unisensory conditions according to the RMI (Miller, 1982), we found that the dyslexic group benefited less than the control group from multisensory input. In line with our hypothesis, an interaction between group and sex indicated that this effect was driven by males with dyslexia who showed impairment in multisensory integration. Smaller facilitation of RTs in the multisensory condition in participants with dyslexia of similar age was previously reported by Harrar et al. (2014), even though they followed a different approach to analyzing RMI. Instead of analyzing AUC in a previously determined violated range of RTs, as recommended by Mahoney and Verghese (2019), they assessed the number of violated percentile bins over the entire RTs distribution. They also did not evaluate the interaction between group and sex probably due to a considerably smaller sample size (9 females and 8 males in the dyslexic group and 10 females and 9 males in the control group). Such a sample size would not provide sufficient statistical power to detect a 2 × 2 interaction. In terms of median RTs, the multisensory condition was faster than unisensory conditions in line with previous RTE studies (Molholm et al., 2004; Mahoney et al., 2011; Harrar et al., 2014); however, we did not find any group differences, as opposed to Harrar et al. (2014) who observed generally slower RTs in the dyslexic group irrespective of condition.
To our knowledge, the current study is the first to imply sex differences in multisensory integration in dyslexia. However, it might partially explain why studies on unisensory and multisensory temporal processing in dyslexia indicated smaller effect sizes when participants were matched for sex, compared to those in studies in which participants were not matched (Meilleur et al., 2020). This suggests the importance of further investigation of sex differences in multisensory integration in dyslexia with other age groups, stimuli, and task requirements to better understand a range of deficits in the processing of multisensory information. Replication of these results in other age groups is also needed since we found a positive correlation between age and the magnitude of multisensory integration. Enhancement of multisensory integration with age was already reported during childhood (Brandwein et al., 2011), in young adults compared to adolescents (Ostrolenk et al., 2019) and in elderly compared to younger adults (Laurienti et al., 2006; Mahoney et al., 2011) suggesting that the older we get, the more we benefit from multisensory information. It would be interesting to investigate whether sex differences revealed in our study are already present in younger children or whether females with dyslexia “catch up” as they get older. Importantly, in the current study, participants with dyslexia did not report any comorbid neuropsychiatric disorders, thus it is unlikely that obtained results are driven by the co-occurring diagnosis, such as attention-deficit hyperactivity disorder. Moreover, at the behavioral level, the only significant interaction between group and sex was found in the multisensory integration task, while not in any of the reading and reading-related tasks. This suggests that the deficit in multisensory integration found in males with dyslexia cannot be explained by general differences in reading skills between females and males with dyslexia.
At the neural level, we systematically observed the differences between responses to multisensory and the sum of unisensory conditions in all analyzed time windows. More specifically, we found a sub-additivity effect (SUM condition eliciting greater neural response than the multisensory condition). This effect was previously reported in ERP studies on multisensory integration (Klucharev et al., 2003; Kronschnabel et al., 2014); however, a reversed, super-additivity effect was also observed (Santangelo et al., 2008; Molholm et al., 2020). Additionally, in the P1 component, the SUM condition had delayed latency compared to that in the multisensory condition, while in the N1 an opposite pattern was observed.
In the N1 component, both the dyslexic and control groups presented greater amplitude in the SUM compared to that in the multisensory condition, indicating that in both groups, stimuli were processed differently when presented simultaneously or not. However, the control group presented a greater amplitude difference between the two conditions than the dyslexic group. Additionally, in the N2 component, the difference in the amplitude between conditions was significant only in the control group, while the dyslexic group had delayed component latency. Yet, the latter results should be taken with caution since differences appearing after 200 ms post-stimuli could be associated with motor responses represented twice in the SUM condition (Giard and Besle, 2010). Nonetheless, as differences were observed also in the N1 component, which cannot be explained by motor responses, they rather suggest atypical processing of multisensory stimuli in the dyslexic group. Previous ERP studies examining multisensory integration in dyslexia either found no group differences (Francisco, 2017) or reported differences in the P1 component driven by late neuronal responses to unisensory stimuli in the dyslexic group (Kronschnabel et al., 2014). Here, the only significant between-group difference in unisensory conditions was the delayed latency of the N2 component in the visual-alone condition in the dyslexic group. Hence, it is unlikely that obtained differences between multisensory and SUM conditions are driven solely by atypical responses to unisensory stimuli by the dyslexic group, as suggested by Kronschnabel et al. (2014). However, those studies used linguistic stimuli; thus, the reference to our results has a limited scope. While delayed latency in the P2 and P3 components for visual and multisensory conditions was previously reported among dyslexic males, in response to nonlinguistic stimuli (Breznitz and Meyler, 2003), this delay was also associated with slower RTs in the multisensory condition. Here, we did not observe the effects of either sex or dyslexia status on median RTs. Consequently, our findings do not support the hypothesis of a generally reduced speed of processing low-level stimuli within the dyslexic group. Again, correspondence between studies is limited due to differences in task demands and the analyzed components.
In the whole sample of 88 participants, we did not observe an interaction between group and sex either in amplitude or latency of ERPs at any time window. This might suggest that both females and males with dyslexia exhibit atypical processing of multisensory information at the neural level. Alternatively, although our sample size was relatively big for an ERP study [compared to 13 participants with dyslexia and 22 controls in a study by Kronschnabel et al. (2014) and 20 participants with dyslexia and 17 controls in a study by Francisco (2017)], it still could be too small to detect the interaction between group and sex. Nevertheless, in a subsample of 80 participants matched in nonverbal IQ, we found a significant interaction between hemisphere, group, and sex in the N1 component. Namely, in the left hemisphere, the sub-additivity effect was lower in males with dyslexia compared to those in control males, while no differences were present in females. This result would be in agreement with behavioral effects, emphasizing that deficits of multisensory integration seem to be more severe in males than females with dyslexia. Future studies should more closely match the groups for nonverbal IQ, as it has been shown that multisensory gains in simple detection tasks can predict nonverbal IQ in children (Denervaud et al., 2020). Irrespective of dyslexia status, we also found sex differences in the N1 component with females presenting greater amplitudes and shorter latencies in multisensory and SUM conditions compared to those in males. Stronger and faster N1 responses in females were also observed in previous ERP studies using linguistic stimuli (for review, see Sato, 2020), suggesting more general sex differences in the N1 ERP component.
The magnitude of multisensory integration was positively correlated with pseudowords reading efficiency, which might support the hypothesis of a low-level multisensory processing deficit underlying reading difficulties in dyslexia. This is consistent with results reported by Harrar et al. (2014), who found a correlation of a similar magnitude between literacy and multisensory integration. However, when we evaluated correlations separately in females and males, this relationship was significant only in male participants. In the whole sample, the magnitude of multisensory integration was also correlated with the earlier latency of the N1 component in the multisensory condition. A correspondence between behavioral and neurophysiological indices of multisensory integration was previously reported for the amplitude of the difference wave in the N1 component, though the latency was not analyzed (Brandwein et al., 2011). Moreover, earlier latency of the N2 component in the multisensory condition correlated with better pseudoword reading performance. The effect was further confirmed in females. This might suggest the existence of sex differences underlying relationships between neural processes related to multisensory integration and reading skills. However, the only relationship that retained its significance after correcting for multiple comparisons was the negative correlation between the latency of the N2 component in the multisensory condition and reading speed (both in the entire sample and among females exclusively). As a result, the interpretation of the other relationships should be approached with caution and necessitates further replication with larger sample sizes. Additionally, in a subsample of 80 participants matched for nonverbal IQ, none of these correlations withstood the correction for multiple comparisons indicating weak associations between variables.
To sum up, the current study indicates sex differences in multisensory integration of simple nonlinguistic stimuli in dyslexia, as only males with dyslexia presented a behavioral deficit in the multisensory integration. At the neural level, both females and males with dyslexia presented atypical processing of multisensory stimuli in N1 and N2 components. However, in a subsample matched for nonverbal IQ, we found evidence of suboptimal multisensory processing in the N1 component, specifically among males with dyslexia. These results suggest that sex might modulate cognitive skills underlying reading skills and emphasize the need for future research characterizing the contribution of sex to the etiology of neurodevelopmental disorders.
Data availability statement
The study was not preregistered. Behavioral data, raw, and processed EEG data required to reproduce the reported analyses can be found at OSF: https://osf.io/k92sj/.
Footnotes
This work was supported by the National Science Centre Grant 2019/35/B/HS6/01763.
The authors declare no competing financial interests.
- Correspondence should be addressed to Katarzyna Jednoróg at k.jednorog{at}nencki.edu.pl.