The odd one out e Orthographic oddball processing in children with poor versus typical reading skills in a fast periodic visual stimulation EEG paradigm

The specialization of left ventral occipitotemporal brain regions to automatically process word forms develops with reading acquisition and is diminished in children with poor reading skills (PR). Using a fast periodic visual oddball stimulation (FPVS) design during electroencephalography (EEG), we examined the level of sensitivity and familiarity to word form processing in ninety-two children in 2nd and 3rd grade with varying reading skills (n ¼ 35 for PR, n ¼ 40 for typical reading skills; TR). To test children ' s level of “ sensitivity ” , false font (FF) and consonant string (CS) oddballs were embedded in base presentations of word (W) stimuli. “ Familiarity ” was examined by presenting letter string oddballs with increasing familiarity (CS, pseudoword e PW, W) in FF base stimuli. Overall, our results revealed stronger left-hemispheric coarse sensitivity effects ( “ FF in W ” > “ CS in W ” ) in TR than in PR in both topographic and oddball frequency analyses. Further, children distinguished between orthographically legal and illegal ( “

The specialization of left ventral occipitotemporal brain regions to automatically process word forms develops with reading acquisition and is diminished in children with poor reading skills (PR).Using a fast periodic visual oddball stimulation (FPVS) design during electroencephalography (EEG), we examined the level of sensitivity and familiarity to word form processing in ninety-two children in 2nd and 3rd grade with varying reading skills (n ¼ 35 for PR, n ¼ 40 for typical reading skills; TR).
To test children's level of "sensitivity", false font (FF) and consonant string (CS) oddballs were embedded in base presentations of word (W) stimuli."Familiarity" was examined by presenting letter string oddballs with increasing familiarity (CS, pseudoword e PW, W) in FF base stimuli.
Overall, our results revealed stronger left-hemispheric coarse sensitivity effects ("FF in W" > "CS in W") in TR than in PR in both topographic and oddball frequency analyses.
Further, children distinguished between orthographically legal and illegal ("W/PW in FF" > "CS in FF") but not yet between lexical and non-lexical ("W in FF" vs "PW in FF") word forms.Although both TR and PR exhibit visual sensitivity and can distinguish between orthographically legal and illegal letter strings, they still struggle with nuanced lexical distinctions.Moreover, the strength of sensitivity is linked to reading proficiency.Our work adds to established knowledge in the field to characterize the relationship between print Introduction Reading is a key communication skill in today's society and, consequently, learning to read is a crucial educational objective.Children become familiar with letters and script in early childhood due to their abundant presence in their everyday environment.Systematic instruction of the orthographic code and the link between spoken and written language usually starts with the beginning of reading acquisition at school enrollment.It is during this time that children's brains become sensitive to the visual appearance of letters and their combinations in words (Brem et al., 2010;Chyl et al., 2018;Dehaene-Lambertz, Monzalvo, & Dehaene, 2018;Maurer, Brem, Bucher, & Brandeis, 2005;Saygin et al., 2016).This sensitivity of the visual system is an important prerequisite for the efficient processing of written information.Coarse sensitivity to print is reflected in a differential neural response to processing letters and words in comparison to false font, symbol strings, or checkerboards as seen in studies using electroencephalography (EEG), MEG, or fMRI Q2 (Bentin, Mouchetant-Rostaing, Giard, Echallier, & Pernier, 1999;Brem et al., 2010;Cao, Li, Zhao, Lin, & Weng, 2011;Chyl et al., 2018;Maurer et al., 2006;Tong et al., 2016;Zhao et al., 2012).Rudimentary Q3 letter knowledge in kindergarteners or Q4 short graphemeephoneme training is sufficient for such an initial coarse print sensitivity (also sometimes referred to as coarse neural tuning for print/orthography or selectivity for letters (Coch & Meade, 2016;Eberhard-Moscicka, Jost, Raith, & Maurer, 2015;Maurer, Brandeis, & McCandliss, 2005;Wong, Gauthier, Woroch, Debuse, & Curran, 2005) to emerge in alphabetic languages (Brem et al., 2010;Cantlon, Pinel, Dehaene, & Pelphrey, 2011;Centanni, King, Eddy, Whitfield-Gabrieli, & Gabrieli, 2017;Dehaene-Lambertz et al., 2018;Karipidis et al., 2018;Maurer, Blau, Yoncheva, & McCandliss, 2010;Maurer, Brem, et al., 2005;McCandliss, Posner, & Givon, 1997;Pleisch et al., 2019;Yamada et al., 2011).Similarly, in logographic scripts such as Chinese, preschool children processed real characters differently from nonorthographic line-or radical combinations after a 25-min visual identification or writing training (Zhao, Zhao, Weng, & Li, 2018).With increasing practice and refinement of reading skills, the recognition of familiar words becomes fast and automatic and the neural response to words starts to differ from the response to nonwords or consonant strings as indexed by familiarity/lexicality effects (Bruno, Zumberge, Manis, Lu, & Goldman, 2008;Centanni et al., 2017;Eberhard-Moscicka et al., 2015;Tong et al., 2016) sometimes referred to as fine neural tuning for words or use of a fine-grained orthographic code (Adams, 1979;Coch & Meade, 2016;Cohen et al., 2002;Grainger & Holcomb, 2009;Maurer, Brandeis, et al., 2005;Zhao et al., 2014).Importantly, sensitivity and familiarity/lexicality effects can be detected even with tasks during which the stimuli are processed implicitly (Shtyrov, Goryainova, Tugin, Ossadtchi, & Shestakova, 2013).
An alternative approach to the N1 ERP time-domain paradigms applies frequency analysis to examine automatic familiarity processing in the visual domain using a fast periodic visual stimulation (FPVS) EEG paradigm (Lochy, Van Belle, & Rossion, 2015).In this kind of task, a specific stimulus type is presented at a certain periodic rate, which in response induces neuronal activity at the same frequency.Such periodically induced electrophysiological responses are referred to as steady-state visual evoked potentials (SSVEPs) (Montani, Chanoine, Stoianov, Grainger, & Ziegler, 2019).The FPVS oddball task displays a rapid sequence of base items that are periodically intermitted by oddball items.It thereby probes the sensitivity to implicitly discriminate between two classes of visual stimuli (Lochy et al., 2015) or even between stimuli of the same category but differing in their frequency of occurrence (De Rosa, Ktori, Vidal, Bottini, & Crepaldi, 2022).Applying frequency analysis, the periodicity of base and oddball stimulation (i.e., using different frequencies as identifiers for different stimulus categories; "frequency tagging") is exploited to test whether a neural response in relevant brain regions can be observed at the base and the oddball frequency bin in the EEG spectrum.While biological noise is scattered across the EEG bandwidth, the activities of interest are garnered at a single defined frequency each, rendering a high signal-to-noise ratio (SNR) within only a few minutes of stimulation time.
A seminal study in adults compared the visual presentation of W as oddballs with a frequency of 2 Hz embedded in false fonts (FF), nonwords (NW), or pseudowords (PW) as bases with a frequency of 10 Hz (Lochy et al., 2015).Left-lateralized discrimination responses were reported for all contrasts.In addition, responses were graded for similarity (WinFF showed the largest discrimination signal, WinPW the weakest).
A recent study further examined oddball discrimination responses in adult readers in terms of the functional and temporal dynamics of response topographies (Wang et al., 2021).They found two sources with temporally separable time courses for WinFF contrasts (first left vOT, later dorsal parietal), in line with leading models of word processing (e.g.Long et al., 2020;Price & Devlin, 2011).However, they detected different sources for WinPW and WinNW, suggesting different underlying processes for word discrimination depending on the base context (Wang et al., 2021).Moreover, a recent study demonstrated that oddball responses have emerged across various visual categories based on the frequency of occurrence of items within a given category.This finding indicates implicit learning of statistical regularities in visual input streams, as highlighted by De Rosa et al. (2022).
To study automatic visual oddball responses in children, five-year-old pre-readers were studied using the contrasts WinFF, PWinFF, and WinPW with a 6 Hz base and 1.2 Hz oddball frequency (Lochy, Van Reybroeck, & Rossion, 2016).The results showed left-lateralized occipitotemporal responses to W or PW oddballs as compared with a midline occipital FF base response.In contrast to adults, however, the five-year-olds did not show a discrimination response for the WinPW contrast.These results suggest that pre-readers can automatically detect a difference between familiar and unfamiliar character strings, but not between real words and pseudowords.At odds with such previous findings, a recent study found lexical and sublexical tuning already in 7-year-old children, showing that the method of stimulation and analysis influences the detection threshold and that high-level linguistic processing occurs earlier than previously assumed (Wang et al., 2022).
In yet another FPVS study with first graders, W oddballs (in FF bases) were processed bilaterally if initially learned by a whole-word rote-learning approach ("globally taught words") in school.When, however, applying a phonics approach to teach the children word reading, W and PW oddball stimuli (in FF base) were processed predominantly in the left hemisphere, which indicates graphemeephoneme decoding (van  Lochy, 2020b).Additionally, this hemispheric distinction for different teaching methods was modulated by reading skill: children with poor reading skills processed globally taught words more bilaterally than PW or W taught with phonics, while those with good reading skills activated the left hemisphere more strongly for all letter string stimuli.Based on these findings, the authors concluded that children with typical reading skills rely more on their automatized graphemeephoneme mappings even for globally taught words, while those with poor reading skills applied the whole-word method.These results support the previously reported progressive left-lateralization with mastery of reading and specifically highlight the role of graphemeephoneme conversion automaticity therein (e.g., Yoncheva et al., 2010).Moreover, they show that, indeed, the FPVS can be used to compare the processing of different oddball categories (e.g., different learning conditions) in addition to the processing of the classic oddball-base contrasts.
Here, we applied the FPVS oddball paradigm to investigate the effects of reading skills on the automatic neural response to sensitivity and familiarity contrast processing in second and third-graders.Until now, only one study has applied the FPVS paradigm to examine the differences between children with poor and typical reading skills (van de Walle de Ghelcke et al., 2020b).Reading deficiencies might impair automatic lexical differentiation, especially for fine lexical contrasts.The FPVS visual oddball design could highlight such differences between individuals with poor and typical reading skills.
In this study, we aim to better understand the range of sensitivity and familiarity during implicit visual processing in emergent reading children in 2nd to 3rd grade with poor versus typical reading skills.To measure the implicit degree of visual sensitivity to non-word oddballs, we inserted consonant string or false font oddballs in W bases (coarse contrast FFinW versus fine contrast CSinW; Wbase).In addition, we compared W, PW, and CS oddballs appearing in FF bases (WinFF, PWinFF, CSinFF), thus manipulating coarse sensitivity contrast by the degree of familiarity with character strings.This combination of FF-base conditions with the target stimuli as oddballs thus provided us with a gradient of familiarity (W ¼ familiar/lexical, PW ¼ unfamiliar/non-lexical, but orthographically legal; CS ¼ unfamiliar/non-lexical, orthographically illegal).
We expected discrimination responses at the oddball frequency with an occipitotemporal and left-lateralized distribution, as was observed in previous FPVS studies using print stimuli (Lochy et al., 2015(Lochy et al., , 2016)).Furthermore, we predicted the oddball response to be graded for both oddball sensitivity (FFinW > CSinW) and oddball familiarity (WinFF > PWinFF > CSinFF) conditions because response amplitudes in previous studies were stronger when oddball-base differences were larger and expertise towards the stimuli was higher (Collins, Robinson, & Behrmann, 2018;Montani et al., 2019;van de Walle de Ghelcke, Rossion, Schiltz, & Lochy, 2020a).Further, we hypothesized the overall response magnitudes to correlate with children's reading scores (van de Walle de Ghelcke et al., 2020a) and to be more pronounced in children with typical than poor reading skills (group effect) (Lochy, Collette, Schelstraete, Rossion, & Schiltz, 2019).Finally, greater left-lateralized responses to orthographic deviants were expected in children with typical, but not poor reading skills, since previous studies have shown delayed or diminished lateralization of the OT in the latter group (Maurer, Brem, et al., 2005;Maurer et al., 2007;Pleisch et al., 2019).To summarize, our study aims to provide a more detailed insight into the implicit visual processing of different word-like stimuli with regard to print sensitivity and familiarity in early readers with typically and poorly developing reading skills.

Materials and methods
We report all data exclusions, all inclusion/exclusion criteria, whether inclusion/exclusion criteria were established prior to data analysis, all manipulations, and all measures in the study.

Participants and group assignments
In total, 96 native German-speaking children in 2nd to 3rd grade participated in a behavioral and an EEG session which were part of a longitudinal study on the efficacy of a phonics training for children with poor reading skills.One child did not complete the EEG task and the data of another three children had to be excluded due to slightly different experimental settings at the beginning of the study.The final sample of the cross-sectional study consisted of 92 native German-speaking children (M ¼ 8.80 y, SD ¼ .63y).All participants had nonverbal intelligence quotient (IQ) scores >80 (as estimated by the non-verbal intelligence index NIX subpart of the RIAS test; Hagmann-von Arx & Grob, 2014;Reynolds & Kamphaus, 2003), normal or corrected to normal visual acuity and reported no neurological or cognitive impairments, with the exception of dyscalculia (5 parentallyreported cases, 1 diagnosed by a specialist) and Attention Deficit (Hyperactivity) Disorder (AD(H)D, 9 cases with diagnosis, of which 4 under medication), which are known to often be comorbid with DYS (Boada, Willcutt, & Pennington, 2012;German o, Gagliano, & Curatolo, 2010;Kronenberger & Dunn, 2003).For the individuals taking medication, intake was discontinued at least 24 h before each experimental session.Parents gave written informed consent and children gave oral assent.Children received vouchers and presents in return for their participation.The project was approved by the local ethics committee of the Canton of Zurich (No. BASEC No. 2018-01261) and neighboring cantons in Switzerland.In-and exclusion criteria were established prior to data collection.
Based on a set of reading tests that examined reading comprehension (ELFE-II) (Lenhard, Lenhard, & Schneider, 2018) Please cite this article as: Lutz, C. G et al., The odd one out e Orthographic oddball processing in children with poor versus typical reading skills in a fast periodic visual stimulation EEG paradigm, Cortex, https://doi.org/10.1016/j.cortex.2023.12.010 grade, 3 left-handed, M ¼ 8.86 y, SD ¼ .65 y) showed intermediate reading skills (IR) and were thus excluded from group analyses, but included for correlation analyses.Further demographic and behavioral information pertaining to the groups is presented in Table 1.

Cognitive assessments
Children each completed a battery of behavioral assessments.
Table 1 contains a more detailed overview and description of the included tests.Percentile scores of the reading and Table 1 e  ).
1 Used for group comparison. 2 Used for correlation analyses.

EEG task
The present work presents neuroimaging data recorded during the performance of an implicit oddball task denoted as Fast Periodic Visual Stimulation (FPVS).Participants were seated in a ventilated, electromagnetically shielded, and sound-attenuated EEG cabin and viewed the stimuli on a gray background on an LCD monitor with dimensions 60 Â 35 cm 2 , 2560 Â 1440 screen resolution, and 144 Hz refresh rate (at a distance of 92 cm from the display).The font size of the stimuli was twice the size of the fixation cross, and the mean visual angle was 2.76 (s ¼ .033)horizontally and 1.32 (s ¼ .020)vertically.
The task was implicit and required participants to fixate on the centered cross and press a button in response to a rapid transient change in its color (blue to red, 194 msec duration) occurring at random intervals six times per block.The stimuli appearing behind the cross were not explicitly attended.Stimuli were presented using Neurobs Presentation® software (Version 20.1, www.neurobs.com).
The stimulation procedure was similar to previous FPVS EEG studies on print processing (Lochy et al., 2015(Lochy et al., , 2016)).Each condition block commenced with a fixation cross presented for 2 sec.Base stimuli were presented in a continuous sequence at a rate of 6 Hz (one stimulus every 166.67 msec).Oddballs deviating from the base category were inserted at a rate of 1.2 Hz (i.e., every 5th item, 6 Hz/5).Fifty-five different stimuli were thereby repeated five times in a randomized order to achieve a total of 275 stimuli.Sinusoidal contrast modulation of stimuli enabled smooth transitions between the items.To avoid eye movements, a gradual fade-in (20 stimuli, 3.3 sec) and fade-out (15 stimuli, 2.5 sec) preceded and followed the full contrast stimulation sequence (40 sec, 240 stimuli, 48 oddballs, 240*0.167sec ¼ 40 sec).The duration of one block was thus 240*0.167sec þ 2 sec þ 3.3 sec þ 2.5 sec ¼ 47.88 sec.Each condition block was presented twice, with short breaks between each condition block and a longer break between the two parts.If the experimenter noted movement during a condition block, the block was repeated at the end of the sequence of blocks.The procedure and conditions are illustrated in Fig. 1 and the Supplementary video.

Materials
The task comprised four categories of stimuli: words (W), pseudowords (PW), consonant strings (CS), and false font strings (FF  .801).Alternating between words and pseudowords, vowels were systematically replaced with consonants (i.e., always the same consonant inserted for the same vowel) to obtain consonant strings matched to the other categories.The false font was constructed based on the real font in the experiment (Swiss school font "Steinschrift") using FontCreator 11.5 (High-Logic, Utrecht, Netherlands).For each letter, symbol elements were shuffled and re-oriented while maintaining their size, number, and complexity.A balanced number of items from our pool of words, pseudowords, and consonant strings were then written in the false font script to build the false font strings.

EEG data acquisition and preprocessing
EEG data were recorded at a sampling rate of 1000 Hz using a high-density 128-channel EEG system (Net Amps 400, EGI HydroCelGeodesic Sensor Net).A DC filter and anti-aliasing filter were applied and electrode Cz was used as the reference, the electrode just posterior to Cz as the ground (COM).Electrode impedances were kept below 50 kU.Data were preprocessed in BrainVisionAnalyzer 2.1 (BrainProducts GmbH, Munich, Germany).First, data were segmented to exclude blocks that had to be interrupted due to movement and breaks in between sequences.The separate EEG blocks of the same subject from the different conditions were concatenated.A .1 Hz high-pass filter and a 50 Hz Notch filter were then applied.Data were visually inspected and manually marked for bad intervals (to exclude them from the ICA decomposition).Noisy-or artifact-ridden channels were then topographically interpolated (number of channels interpolated per subject: M ¼ 4.15, SD ¼ 3.33, range ¼ 0e12).Due to the purported high signal-to-noise ratio of the FPVS analysis method (Lochy et al., 2015(Lochy et al., , 2016)), we included most of the data.However, sequences that contained roughly more

Frequency analysis
A Fast-Fourier Transform (FFT) was applied to convert the data into the frequency domain.Amplitude spectra were extracted (per single subject, group, and condition) for each electrode and exported for further analyses using R (RCoreTeam, 2021) in-house scripts.We selected literaturebased clusters of electrodes over the left (LOT; E050, E057, E058, E059, E063, E064, E065, E066, E068, E069, E070, E073, E074) and right occipitotemporal cortex (ROT: E082, E083, E084, E088, E089, E090, E091, E094, E095, E096, E099, E100, E101) (Supp.Fig. 1) (Pleisch et al., 2019).Z-scores (amplitude at each frequency minus the average of 20 surrounding bins divided by the standard deviation of the 20 surrounding bins) were computed on the grand average of the spectra per group, condition, and electrode cluster.This was done to assess the response significance at the oddball frequency and each harmonic (multiples of 1.2 Hz, i.e., of the oddball frequency) and thereby determine the number of harmonics to include in the statistical models.Z-scores of 1.96 and above were considered significant.Based on the highest number of consecutive harmonics to exceed this threshold in any group, electrode cluster, or condition, we selected an identical number of harmonics across all groups (including the intermediate group), electrode clusters (averaged before computing Z-scores), and conditions included per statistical model (as described in section 'Statistical Analysis').In total, we included 5 harmonics (H1 to H5, i.e., multiples of 1.2 Hz up to 7.2 Hz, excluding base frequency) for model "familiarity" and 3 harmonics (H1 to H3, i.e., multiples of 1.2 Hz up to 3.6 Hz) for model "sensitivity".For the quantification of the periodic oddball response spread out across harmonics, we then summed the baseline-subtracted amplitudes (i.e., the average voltage amplitude of the 20 surrounding bins subtracted out) at the oddball-and harmonic frequency bins (see e.g.Retter & Rossion, 2016 for validation of procedure).

Statistical analysis
Two linear mixed models (LMM) with the sum of baselinesubtracted amplitudes as the dependent variable were defined and fitted in SPSS® (IBM-Corporation, 2020): i) "sensitivity" and ii) "familiarity"., 2003).Post-hoc pairwise comparisons were corrected using Bonferroni.We report significant values (p .05)and statistical trends (p .1)but restrict our discussion mainly to significant values.To reduce bias by outliers and extreme values, we iteratively removed normalized (z-score) residuals that exceeded a threshold of ±3 (see e.g.Fraga-Gonz alez et al., 2021).In total, 6 data points out of 296 (2.02 % of the data) were removed in model "sensitivity" and 11 out of 444 data points (2.48 % of the data) were removed in model "familiarity".
Linear regression was performed using R (R Core Team, 2021) to analyze the association between neural (baselinesubtracted amplitudes per condition) and behavioral measures of interest (word-and pseudoword reading fluency and text reading comprehension) over LOT (i.e., 15 contrasts were tested).These tests were performed after the exclusion of outliers (1.5 IQR criterion) to exclude spurious findings (see the corresponding sample in the table).
Lastly, plots of scalp topographies of the sum of baselinesubtracted amplitudes at each oddball harmonic (including all oddball harmonics up to the pre-determined number except the base, see description above), and t-maps between conditions and groups were created in EEGlab (Delorme & Makeig, 2004), a Matlab-based toolbox (R2020b, MathWorks, Natick, MA).For the t-maps, we computed electrode-wise pairwise comparisons of the baseline-subtracted oddball amplitude sums: To measure conditions against each other, we used paired two-sided t-tests, and to compare the PR and TR groups, we used independent two-sided t-tests.Further, we applied the R-based DuckDB (Raasveldt & Mu ¨hleisen, 2019) to compute the subject-wise difference between conditionpairs' mean sum of oddball harmonic amplitudes, which were then compared between the groups using independent twosided electrode-wise t-tests.

Oddball discrimination responses
3.1.1.Oddball response amplitudes by reading level 3.1.1.1.PRESENCE OF SIGNIFICANT Z-SCORES OF RESPONSES AT THE ODDBALL FREQUENCY.To test for the presence of discrimination responses in the different conditions, hemispheres, and groups, z-scores at the oddball frequency and its harmonics were examined.Discrimination responses (significant zscores at target oddball and harmonic frequencies) were detected in both models in all conditions in at least one hemisphere and one group except for CSinW (WinFF: H1eH4 in PR, H1eH5 in IR and TR; PWinFF: H1eH3 in PR, H1eH4 in IR and TR; CSinFF: 0 in PR and IR, H1eH4 in TR; FFinW: H1eH3 for all groups).A maximum of three consecutive harmonics for model i "sensitivity" (1.2e3.6 Hz, excluding the base rate at 6 Hz) and five consecutive harmonics (from 1.2 to 7.2 Hz, excluding the base rate at 6 Hz) for model ii "familiarity" were thus found to be significant.More detailed information on the number of consecutively significant harmonics per group, electrode cluster, and condition can be found in Supp.Tables 1 and 2. Additionally, the supplementary information includes an exploratory inspection of the spread of amplitudes across the harmonics (i.e., individually instead of analyzing the sum of consecutively significant harmonics; Supp.Fig. 2).Frequency spectra and topographical scalp maps of the sum of amplitudes at oddball harmonics are shown in Fig. 2.

TOPOGRAPHIC EVALUATION OF RESPONSE AMPLITUDES AT THE
ODDBALL FREQUENCY.To investigate our hypothesis regarding the occipitotemporal and left-lateralized distribution of the oddball discrimination responses, as well as the predicted greater left-lateralization in TR than PR, we computed topographic maps and t-maps (Fig. 2).In TR, the sum of oddball harmonics shows a clear left-lateralized occipitotemporal topographical profile and responses are stronger for oddballs that differ strongly from the base (FFinW > CSinW; sensitivity model) and for more familiar oddballs (W > PW > CS in FF; familiarity model).These effects appear reduced or inconsistent (e.g., right-lateralization for FFinW) in PR.
T-maps for pairwise comparison of the sensitivity conditions show significantly greater activity for FFinW than CSinW in bilateral but right-lateralized occipitotemporal areas in PR, and bilateral left-lateralized occipitotemporal and central regions in TR.A comparison of the groups showed greater activity in left occipitotemporal electrodes in TR than in PR.In CSinW, a right-hemispheric occipitotemporal cluster of electrodes was also observed to show a significant group difference (TR > PR).
T-maps for the familiarity conditions revealed significant left-hemispheric occipitotemporal differences for both WinFF À CSinFF and PWinFF À CSinFF in TR.On the other hand, PR showed bilateral, less focalized posterior differences.In the group comparison per condition, TR showed greater responses than PR over a left posterior, a central, and a right-frontotemporal region in WinFF.In CSinFF, the t-map suggests higher response amplitudes across the (bilateral) posterior scalp for TR than PR.PWinFF showed no notable differences between the groups.PR appear to have displayed a greater difference in response amplitudes between the PWinFF and CSinFF conditions than TR over a right-hemispheric occipital area.For the comparison of WinFF with the other two conditions, TR showed greater amplitude differences than PR over central (WinFF-PWinFF) and over central and right fronto-temporal areas (WinFF-CSinFF).
3.1.1.3.LINEAR MIXED MODEL ANALYSES.Two linear mixed model (LMM) analyses i) "sensitivity" and ii) "familiarity" were performed on the summed baseline-subtracted amplitudes of the specified number of significant harmonics (see Section 2.6.Statistical analysis for description).We used these models to test differences in the level of sensitivity (FFinW > CSinW) and familiarity with (WinFF > PWinFF > CSinFF) the oddball stimulus, depending on the hemisphere and reading group.In the "sensitivity" model (Fig. 3), there was a significant main effect of condition (F  The figure shows frequency spectra and topographies of the sum of baseline subtracted amplitudes across the relevant harmonics (i.e., average amplitude over 3 harmonics) for TR (above) and PR (below).Oddball harmonics that reached significance according to Z-scores >1.96 are marked by a red asterisk in the frequency spectra.Below the topographies, t-maps display the group-wise comparison of topographical scalp maps across conditions (t-tests).Panel B shows the condition-wise comparison of groups (row 1, TR-PR), and the comparison of pairwise condition-differences across the groups ("double-difference", rows 2e3).For the t-maps, red and blue denote significantly greater and lower sums of harmonic amplitudes (respectively) in TR than PR.Activity in OT regions was observed only for FFinW and showed diverging lateralization in the groups.Right panel (Familiarity model): Same as left panel for "familiarity" conditions.Responses are stronger in the left hemisphere, in TR, and for more familiar oddball stimuli.with PR.Main effects were qualified by a condition-hemisphere-group interaction (F 1, 213.73 ¼ 5.07, p ¼ .025).Amplitudes in response to FFinW were stronger over LOT than ROT in TR, but not PR (t 213.77 ¼ 3.19, p Bonferroni ¼ .002).Correspondingly, oddball responses to FFinW were stronger over LOT for TR than PR (t 258.76 ¼ 3.58, p Bonferroni < .001).Lastly, there was a trend for a main effect of the CBCL attention-deficit/hyperactivity covariate (F 1, 68.65 ¼ 3.63, p ¼ .061).No further trends or significant effects were found.In the "familiarity" model (Fig. 4), we found main effects of condition (F 2, 347.68 ¼ 7.09, p < .001),hemisphere (F 1, 346.86 ¼ 31.89,p < .001),and a trend for group (F 1, 66.85 ¼ 3.19, p ¼ .079).Overall, post-hoc pairwise comparisons of conditions showed oddball discrimination amplitudes to be significantly stronger for both WinFF and PWinFF than CSinFF (t 348.63 ¼ 3.57, p Bonferroni < .001and t 346.74 ¼ 2.77, p Bonferroni ¼ .017,respectively).The main effect of hemisphere pointed to stronger oddball amplitudes over LOT than ROT.Finally, we found a trend for a three-way interaction of condition-hemisphere-group (F Models including participants with intermediate reading skills are shown in Supp.Fig. 3 (sensitivity) and Supp.Fig. 4 (familiarity) and largely support the results of the main analyses.

Brainebehavior associations
We computed brain-behavior correlations to clarify whether the response magnitudes of oddball responses are related to children's reading skills.Here, we found a significant association between word reading fluency and FFinW oddball responses and between reading comprehension with FFinW and PWinFF oddball responses (shown in more detail in Table 2 and Fig. 5; only tested for LOT, not ROT, see Methods).These associations indicate that oddball response magnitudes over LOT increase with reading skills.Further trends for brainbehavior associations that did not survive multiple comparison corrections can be found in Table 2.

Discussion
We applied an oddball paradigm to examine potential differences between children with typical and poor reading skills in their ability to implicitly discriminate between orthographic lexical, orthographic non-lexical, and non-orthographic stimuli in a fast-paced visual presentation.First, we used orthographic versus non-orthographic (CS, FF) oddballs in W bases to test fine (CSinW) versus coarse (FFinW) sensitivity,  i.e., the presence and level of neural discrimination responses to print in children.In addition, we were interested in the strength of responses in TR versus PR to oddball items along a gradient of orthographic familiarity and legality (oddballs W > PW > CS) embedded in a sequence of unfamiliar FF base items.Our results show consistent oddball effects for coarse tuning to print (coarse sensitivity; FFinW) in children in 2nde3rd grade and indicate that the strength and lateralization of the coarse tuning responses depend on reading skills.Importantly, linear regression models indicated that the strength of the coarse sensitivity over LOT is positively associated with the reading skills of the children.The effects of the level of familiarity with word forms were less pronounced.
Oddball effects distinguished between orthographically legal and illegal items (W/PW vs CS) but not yet between lexical and non-lexical (W vs PW) word forms.Altogether, effects of familiarity tended to differ between groups and hemispheres, with TR showing familiarity effects mainly over the left hemisphere, while such effects were observed over the right hemisphere in PR.These main findings will be discussed in more detail in the next sections.

Oddball discrimination conditions reveal coarse but no fine sensitivity to print
In agreement with previous reports (Lochy et al., 2016), we found oddball responses to the familiarity conditions in all groups.This suggests that orthographic stimuli triggered brain activity that was distinct from the state evoked by false fonts.The participants were thus able to automatically discriminate between letter-containing items and a false script (WinFF, PWinFF, CSinFF) and therefore show coarse sensitivity to print, as we expected for children of this age (Lochy et al., 2016;Maurer et al., 2006;van de Walle de Ghelcke et al., 2020a).Similarly, significant oddball responses to FFinW, but not to CSinW conditions suggest that coarse print sensitivity has already emerged in all participants, while fine discrimination of words from other letter strings (i.e., familiarity) is not yet detectable with a fast implicit processing task.
Previous studies have reported conflicting findings regarding fine discrimination in young children.In FPVS literature, for example, no PW versus W distinction was found in children (Lochy et al., 2016;van de Walle de Ghelcke et al., 2020a, 2020b; children in these studies were 5e7 years old and French-speaking).Although PW are orthographically legal while CS are not, both PW and CS are orthographic stimuli, and thus both difficult to discriminate from words for early readers.In support of this, a study described 7e14-year-old children's VWFA print "sensitivity" as full-fledged, yet the "specificity" towards letters and words as immature (Centanni et al., 2017).It thus appears that the fundamental skill set for reading is obtained early on in learning (Aghababian & Nazir, 2000, p. 123), but that it takes more time to specialize in word processing (Brem et al., 2006;Coch & Meade, 2016, p. 115) and to become quicker at gathering information from print (Aghababian & Nazir, 2000, p. 123).However, a very recent study found evidence for lexical and sublexical processing in children at early reading stages (kindergarten to second grade) (Wang et al., 2022).The study used an adapted SSVEP task design (alternating between two stimulus types and doing so at a slower pace than previous studies) and reliable component analysis.They compared coarse print tuning, lexical processing, and sublexical orthographic processing using W versus FF, W versus PW, and PW versus nonword (unpronounceable letter combinations with lower orthographic neighborhood and bigram frequencies than PW) contrasts.The authors interpret these methodical modifications to the FPVS as the defining features enabling higher signal detection.In other words, it might be that a lexical discrimination effect in children emerges only at slower presentation rates and more sensitive analysis methods.Further studies are needed to replicate these results and provide more clarity as to the discrepancies between studies.

4.2.
Strong coarse sensitivity and emerging familiarity discrimination in children with typical reading skills over the left hemisphere In our sensitivity model, we found significantly greater oddball responses in TR than in PR, and this group difference was especially pronounced over the left hemisphere.Diminished left occipitotemporal coarse print sensitivity in PR is in line with previous studies using classical ERP designs (Eberhard-Moscicka et al., 2015;Maurer et al., 2007;Pleisch et al., 2019) or FPVS (van de Walle de Ghelcke et al., 2020b), although one previous report did not find differences between reading groups for coarse, but only for lexical and sublexical processing (Wang et al., 2022).The familiarity model supported the results of the sensitivity model in that the degree of discrimination along a gradient of familiar versus unfamiliar and orthographically legal versus illegal character strings tended to differ according to reading skills and hemisphere.Previous literature has described the preferential activity of vOT to familiar print to become predominantly left-lateralized with reading acquisition (Brem et al., 2006;McCandliss et al., 2003;Rossion, Joyce, Cottrell, & Tarr, 2003;Seghier & Price, 2011;Zhao et al., 2012).In our study, we found electrodes over LOT to generally activate more strongly to the oddballs than those over ROT.
In TR, the discrimination response amplitudes to orthographically legal word forms (WinFF and PWinFF) were significantly higher than those to illegal ones (CSinFF) over LOT in the differential t-maps for the contrasts between WinFF-CSinFF and PWinFF-CSinFF, respectively.There was, however, no significant difference for the WinFF-PWinFF tmap contrast.This indicates that TR already differentiate between legal (W, PW) and illegal (CS) orthographic oddballs presented amidst FF base, but not yet between familiar and unfamiliar word forms (PW and W) oddballs amidst FF base over the left-hemispheric vOT (Panda et al., 2022).PR also show significant differences in the t-map comparison of conditions, but the significant electrodes locations were more bilateral, covering large parts of the ventral occipital and posterior temporal sites.Notably, the group-hemispherecondition three-way interaction over LOT and ROT did not reach significance in the "familiarity" LMM.This may be explained by the selection of the a priori defined electrode clusters covering large parts of the OT and inferior parietal cortex.Previous studies have reported asymmetrically (more pronounced in the left hemisphere) greater activation of the vOT/VWFA the more frequent or familiar the letter combinations of word-fragments or the more word-like the stimuli (Vinckier et al., 2007).Therefore, our results suggest that W and PW are processed as more familiar and as more distinguishable from FF than CS by TR, while in PR this lateralization appears less clear.
The results are thus largely in line with our expectations based on previous studies.Although to our knowledge, there is no previous FPVS study comparing W and CS, previous studies have tested the fine contrast of W and PW.For instance, W embedded in PW (WinPW) did not yield a significant discrimination response (Lochy et al., 2016;van de Walle de Ghelcke et al., 2020a) and WinFF versus PWinFF did not differ significantly in children (Lochy et al., 2016;van de Walle de Ghelcke et al., 2020a, 2020b).In adults, WinPW differences indicating lexical differentiation of orthographically legal strings could previously be observed in FPVS paradigms (Lochy et al., 2015), albeit not consistently (Barnes, Petit, Badcock, Whyte, & Woolgar, 2021).It is important to emphasize that the children in the aforementioned studies were pre-schoolers or first-and at most second-graders in the first trimester, thus beginning readers.In more classical visual ERP paradigms, the evidence on fine-tuning in young children remains mixed: while some studies found no significant N1 ERP difference between W and CS (Posner & McCandliss, 1999) or PW and W (Eberhard-Moscicka et al., 2015) in 4-and 7-year-old children or even older children (9e13 years (Arau ´jo et al., 2012); 8e12 yrs (Kast, Elmer, Jancke, & Meyer, 2010)), children with high reading skills may show such fine-tuning between orthographically legal and illegal strings already at an early age (7 yrs) (Zhao et al., 2014).The children in our study were at a more advanced reading stage (2nde3rd grade) and were learning to read an orthographically rather transparent alphabetic language (German).Our finding of an emerging fine sensitivity for orthographically legal versus illegal strings (PWinFF versus CSinFF) in our typically reading children thus supports the rapid and critical changes occurring during the early school years, paralleling the development of sight word reading (Maurer et al., 2011;Wang et al., 2022).

Altered hemispheric patterns in PR may indicate developmental delay or compensatory strategies
Although PR showed significantly stronger discrimination amplitudes to FFinW than CSinW both in the left and right hemisphere, the responses were weaker than in TR and showed no significant lateralization.Topographic plots even indicate a more right-hemispheric OT activation for FFinW in PR.
For the familiarity model, the t-map comparison of pairwise condition differences between groups also showed an increased difference between PWinFF and CSinFF over the right occipitotemporal scalp in PR compared to TR.However, in the linear mixed models, we did not find a significant difference in the processing of PWinFF and CSinFF, but only found trend-level evidence for greater activity to W and PW than to CS oddballs in FF over the right hemisphere of PR.
The reduced and more right-lateralized activation for and differentiation between oddballs could either reflect a delayed development (Maurer et al., 2011) or beginning compensational processes in children with poor reading skills.Indeed, different developmental trajectories of print processing in PR have previously been referred to in literature (Maurer et al., 2011).The lateralization, indicative of print specialization, develops only with intensive training and over time (Eberhard-Moscicka et al., 2015;Maurer, Brem, et al., 2005;van Setten, Maurits, & Maassen, 2019).Several studies report discrepant laterality findings in children.For example, Kast et al. (2010) report missing lateralization in 8-to 12-year-old readers and attribute it to incomplete maturation or reading expertise.Similarly, Spironelli and Angrilli (2009) found greater visual N1 right-lateralization in 10-year-olds.Before reading instruction at school, visual familiarity with print and higher letter knowledge was reflected in a right-lateralized N170 (Maurer, Brem, et al., 2005;Maurer et al., 2006).Similarly, in an artificial script training study with adults, emerging N170 responses to the stimuli were more pronounced over the right rather than the left hemisphere after a 20-min training phase (Maurer et al., 2010).Thus the right hemisphere seems more involved in the beginning stages of reading acquisition before the typical left hemispheric dominance develops (Maurer et al., 2010;Seghier & Price, 2011).Further longitudinal research will be necessary to clarify whether such an activation pattern in the ROT might precede the fine-scaling in the LOT (indicative of a developmental delay in PR).
Another interpretation could be compensatory activity specific to PR.In fMRI (Borghesani et al., 2021;Centanni et al., 2019) studies, researchers found a missing word tuning effect (W > FF) and a hypoactivation to letters and FF in poor compared to TR in the left hemisphere (Ozernov-Palchik & Gaab, 2016).Furthermore, greater functional connectivity among right-hemispheric brain areas has been observed for groups with more severe reading problems (Panda et al., 2022).Right hemispheric activation in PR has been reported as a mechanism to compensate for this left-hemispheric hypoactivation (D emonet et al., 2004;Pugh et al., 2000;Shaywitz et al., 1998;Turker, 2018;Waldie et al., 2013).Compensatory refinement of the ROT could be driven by more pictorial, rotelearned representation of the stimuli since there is evidence for greater involvement of the right vOT in deep orthographies and in the visual appraisal of familiar objects, drawings, and symbols that are not linked to phonology (Lochy et al., 2016;Maurer, Brandeis, et al., 2005;Mei et al., 2013;van de Walle de Ghelcke et al., 2020a).
In summary, the discrepant lateralization patterns between the groups are likely to arise from delayed reading development or compensatory strategies in PR.However, further data sampling time points would be required to attribute our findings to persistent developmental differences or delayed development.

Limitations and outlook
There are several aspects of the current paradigm that may have impacted our results.Further, the paradigm we used was implicit.The attention was therefore not specifically directed to the stimuli, which could affect the engagement of the reading system.Specifically, higher-level reading processes might be mobilized to a greater degree when there is an overt reading-related task requirement in contrast to an implicit task (Maurer et al., 2006;Okumura, Kasai, & Murohashi, 2015;Wang et al., 2021;Yoncheva et al., 2010).Future studies should thus compare implicit and explicit approaches to study this effect of task design.
It is important to note that the present analyses are complex, including three fixed factors and two covariates.A design with many variables could potentially limit our power, despite the large sample of children.However, these variables were important to test our hypotheses relating to group, condition, and lateralization differences and to account for possible confounding effects.We included nonverbal IQ and the attention-deficit/hyperactivity subscore as covariates to account for potential differences related to IQ and attention.This consideration is further justified by the high comorbidity of reading impairments and ADHD (Boada et al., 2012).In the supplementary material (section S5), we also provide an overview of the results of the same models without the covariates.Future studies will be critical to corroborate the present results and conclusions.
An additional factor warranting consideration is statistical regularity.A recent study by De Rosa et al. (2022) noted that participants exhibited an oddball discrimination response not only to stimuli differing in category but also to those differing merely in frequency of occurrence within the study.To provide children with well-matched, highly familiar words, our study also involved repeating stimuli.Importantly, this repetition was consistent across all stimulus categories (W, PW, CS, FF) and all conditions.Moreover, the order of stimulus presentation was randomized, thus controlling for differences in the frequency of repetitions between oddball and base items in each condition and for possible ensuing statistical regularity effects.Consequently, potential statistical responses, if present, would blend into low-frequency background noise, ensuring that our findings remain uncompromised.
For future studies, a longitudinal developmental approach would be of interest.van de Walle de Ghelcke et al. (2020a) did a 1-year follow-up of beginning readers in 1st grade and found an increase in the response strength to letter strings (W or PW) embedded in FF.They also show topographical changes characterized by a transition of peak amplitude from a posterior occipito-medial electrode to a more lateral position (O1 to P7).Their result aligns with descriptions in literature of more posterior VWFA activation in children (Lochy et al., 2016;O. Olulade, D. Flowers, E. Napoliello, & G. Eden, 2015) and more lateral letter string responses in adults (Lochy et al., 2015).Follow-ups across longer time scales or cross-sectional studies across a broader age range with multiple testing time points could capture shifts across reading development and compare to studies using more classical ERP paradigms and focusing on the N1 (Fraga-Gonz alez et al., 2021).Moreover, the effects of reading abilities should be further surveyed within such a longitudinal framework.

Conclusion
The present study assessed visual print processing in 2ndand 3rd-grade PR and TR.The analyses focused on discriminatory responses to oddball stimulus processing over the OT cortex.Our contrasts represented discriminations of oddballs along the levels of familiarity (W vs PW vs CS embedded in FF) and orthographic sensitivity (FF vs CS in W).Our results suggest that both TR and PR show coarse visual sensitivity to print.However, the level of this sensitivity response depended on children's reading skills and was less pronounced in PR.
Children also showed a basic level of familiarity as reflected in the discrimination of orthographically legal and illegal letter strings (W/PW vs CS oddballs), but the more subtle lexicality distinction is still lacking, possibly due to the developmental stage or the rapid visual presentation design.Finally, the differences in the oddball response levels between TR and PR for the conditions and hemispheres may reflect differences in learning and developmental progress between the groups or alternative strategies in PR.The results extend our insights on automatic visual print processing and the influence of reading skills therein.By comparing oddball-base pairs of differing sensitivity (coarse vs fine) and of incrementally increasing familiarity (non-lexical and orthographically illegal, nonlexical but legal, orthographically legal and lexical), it delivers more resolved information about the development of sensitivity versus familiarity in young readers.

Fig. 1 e
Fig. 1 e Experimental Paradigm.Above: Example base and oddball stimuli of the different conditions (grouped by model) are shown.Randomized order of stimuli presented for about 48 sec non-stop.The task was divided into two equal parts with a break in between.Below: Representation of the timeline of one block.

Fig. 2 e
Fig. 2 e Left panel (Sensitivity model): Baseline-subtracted amplitude frequency spectra of the "sensitivity" conditions for typical and poor readers in left (LOT) and right (ROT) occipitotemporal electrode clusters.The figure shows frequency spectra and topographies of the sum of baseline subtracted amplitudes across the relevant harmonics (i.e., average amplitude over 3 harmonics) for TR (above) and PR (below).Oddball harmonics that reached significance according to Z-scores >1.96 are marked by a red asterisk in the frequency spectra.Below the topographies, t-maps display the group-wise comparison of topographical scalp maps across conditions (t-tests).Panel B shows the condition-wise comparison of groups (row 1, TR-PR), and the comparison of pairwise condition-differences across the groups ("double-difference", rows 2e3).For the t-maps, red and blue denote significantly greater and lower sums of harmonic amplitudes (respectively) in TR than PR.Activity in OT regions was observed only for FFinW and showed diverging lateralization in the groups.Right panel (Familiarity model): Same as left panel for "familiarity" conditions.Responses are stronger in the left hemisphere, in TR, and for more familiar oddball stimuli.

Fig. 3
Fig. 3 e "Sensitivity" Linear Mixed Model Results.Mean normalized amplitudes for the different groups, conditions, and hemispheres are shown.A significant three-way interaction between these factors was found.Stars represent significant differences as determined by linear mixed model post-hoc pairwise comparisons: p ≤ .001***,p ≤ .01**,p ≤ .05*.CS ¼ consonant strings, W ¼ words, FF ¼ false fonts; TR ¼ children with typical reading skills, PR ¼ children with poor reading skills, LOT ¼ left occipitotemporal electrode cluster, ROT ¼ right occipitotemporal electrode cluster.

Fig. 4
Fig. 4 e "Familiarity" model.Mean normalized amplitudes for the different groups, conditions, and hemispheres are shown.A trend for a three-way interaction of these factors was found.Asterisks represent differences as determined by linear mixed model post-hoc pairwise comparisons of this interaction: p < .001***,p < .01**,p < .05*,p < .1 þ.CS ¼ consonant strings, PW ¼ pseudowords, W ¼ words, FF ¼ false fonts.TR ¼ children with typical reading skills, PR ¼ children with poor reading skills, LOT ¼ left occipitotemporal electrode cluster, ROT ¼ right occipitotemporal electrode cluster.
c o r t e x x x x ( x x x x ) x x x

Fig. 5 e
Fig. 5 e Brain-behavior association plots of significant linear regression outcomes between reading test scores (raw scores) and baseline-subtracted mean amplitudes for the different conditions in the left hemispheric electrode cluster (LOT).The shaded area around the fitted line shows the 95 % confidence interval.Boxplots to the side of both axes visualize the distribution of the behavioral and neural data values.

Table 2 e
Linear Regression with reading test scores (raw scores) and baseline-subtracted amplitudes for the different conditions over LOT.Sorted top-to-bottom by highest R 2 value.þ: survived Bonferroni multiple comparison corrected p-value of .0033(15 tests, i.e., threshold of .05/15);trend: .1/15¼ .0066.
Please cite this article as: Lutz, C. G et al., The odd one out e Orthographic oddball processing in children with poor versus typical reading skills in a fast periodic visual stimulation EEG paradigm, Cortex, https://doi.org/10.1016/j.cortex.2023.12.010 similar pattern of responses, although amplitudes were larger for the slower presentation rates.Further studies should investigate the impact of such stimulation differences in more detail and in different age groups.
One of them is the stimulation frequency.A recent report explored two frequency variations (2 Hz oddball, 10 Hz base vs 3 Hz oddball, 6 Hz base) for WinFF contrasts (Wang et al., 2021, 2022).The two variations elicited Please cite this article as: Lutz, C. G et al., The odd one out e Orthographic oddball processing in children with poor versus typical reading skills in a fast periodic visual stimulation EEG paradigm, Cortex, https://doi.org/10.1016/j.cortex.2023.12.010 a