Both high and low spatial frequencies are critical for visual consciousness in autism: Evidence of an emotional attentional blink paradigm.

High Spatial Frequencies (HSF - conveying local information) may serve a critical role in 34 visual consciousness. Despite an HSF bias during visual perception in autism, autistic 35 individuals demonstrate impairments in face processing. Our aim was to investigate the 36 respective role of HSF and Low Spatial Frequencies (LSF - conveying coarse information) on 37 visual consciousness in autism. Thirty-two autistic adults and 35 typically developing (TD) 38 controls performed an emotional attentional blink paradigm with spatially filtered distractors. 39 TD participants showed reduced T2 accuracy (i.e., accuracy for the second target given the 40 correct report of the first target T1) after unfiltered and HSF distractors compared to LSF 41 distractors. In the autistic group, we observed lower T2 accuracy than controls after HSF and 42 LSF distractors but not after unfiltered distractors. Results suggest the importance of HSF 43 for visual consciousness in TD participants whereas, both LSF and HSF seem important in 44 autism. 45


Introduction
Efficient face processing play an important role in successful navigation of the social 54 world (see Oruc et al., 2019) and people demonstrate the remarkable ability to process human 55 faces and to identify emotion on faces (Tanaka, 2001). This ability would partly rely on the 56 treatment of low-level information such as orientation and spatial frequencies (Balas and 57 Verdugo, 2018; Oruc et al., 2019). In a visual stimulus, spatial frequency refers to the energy 58 distribution derived from the Fourier transform, and expressed in the number of cycle per 59 degree of visual angle (Park et al., 2012). Coarse information would be conveyed by Low 60 Spatial Frequencies (LSF -below two cycles per degree) mainly by the dorsal stream, while 61 fine information would be conveyed by High Spatial Frequencies (HSF -above six cycles per 62 degree) by the ventral stream (Skottun, 2015). Addressing specifically the question of the 63 interaction between spatial frequency and consciousness during face processing, De Gardelle 64 & Kouider (2010) demonstrated that HSF information, more than LSF, correlates with 65 consciousness. Previous behavioral studies also showed that HSF components would 66 dominate conscious perception when the exposition time is long enough (Schyns and Oliva,67 1994), i.e., more than 100 ms. These findings are corroborated by neurobiological evidence 68 from functional magnetic resonance imaging as the processing of HSF information from faces 69 involved the inferior occipital and fusiform gyri (Iidaka et al., 2004;Vuilleumier et al., 2003), 70 regions implied in conscious perception. Indeed, the recording of single neuron response in 71 the medial temporal lobe showed firing related to conscious perception (Quiroga et al., 2008). 72 Electrophysiological studies also demonstrated that conscious face perception correlates with 73 a boost in the activity of the N170 in the occipito-temporal cortex (Navajas et al., 2013). In 74 sum, there is a body of evidence that HSF would be related to visual consciousness of faces 75 due to the cortical structure underlying their processing. 76 Hamilton, 2013). These impairments may significantly contribute to impaired social 80 interaction, one of the core symptoms of Autism Spectrum Disorders 1 (ASD). It has been 81 suggested that discrepancy between autistic and typically developing (TD) individuals in face 82 processing could be partly explained by a different use of LSF and HSF. Autistic persons 83 exhibited a perceptual bias toward local information while processing either non-social 84 (Caplette et al., 2016;Kéïta et al., 2014;Mottron et al., 2006) or social visual stimuli such as 85 emotional faces (Deruelle et al., 2008(Deruelle et al., , 2004Jemel et al., 2006). Given the HSF bias in autism 86 and the critical role of HSF for visual consciousness of faces, the difficulties encountered by 87 autistic individuals in face processing and emotion recognition are surprising. One 88 explanation could be a reduced sensitivity to HSF in regions associated with consciousness in 89 autism. In line with this hypothesis, Corradi-Dell'Acqua and al. (2014) showed a reduced 90 sensitivity to HSF in a part of the fusiform gyrus (the fusiform face area) in autism compared 91 to control, whereas the response to LSF was intact. Interestingly, this result could also explain 92 that the fusiform face area has often been found hypoactivated in autism during face 93 processing (Golarai et al., 2006). As the fusiform is associated with visual consciousness and 94 belongs to the social brain network, a primary fusiform deficit might disturb several 95 mechanisms relying on face perception, such as emotion recognition (Dawson et al., 2005). 96 However, this result has not been reproduced and moreover, it does not explain the previously 97 reported better use of HSF in autism for categorizing emotional face. Thus, it is unclear which 98 spatial frequencies play a critical role in visual consciousness of emotional faces in autism. 99 Answering this question would help us to understand if there are atypical low-level visual 100 processing that could be linked to atypical visual consciousness of faces in autism and that 101 could explain some difficulties in face processing. We therefore considered that the 102 attentional blink paradigm would be useful for this investigation. 103 Rapid Serial Visual Presentation paradigms have been utilized to study visual 104 consciousness by manipulating the time course of attention. During this type of tasks, 105 participants are asked to detect two visual targets (T1 and T2) embedded among a stream of 106 distractors presented at a frequency of about 10 items per second (Shapiro et al., 1997). 107 People typically show high accuracy of detecting the first target T1 (Olivers and Meeter, 108 2008). However, the conscious perception and the subsequent report of the second target T2 is 109 inconsistent depending on the lag between the two targets (Shapiro et al., 1997). When the lag 110 between T1 and T2 is very short (below 200 ms), the detection of the second target T2 is 111 good, often better than the detection of the first target T1, which is called the Lag-1 sparing 112 phenomenon (Olivers and Meeter, 2008). On the contrary, a second target T2 occurring 113 between 200 ms and 500 ms after T1 often escape from conscious perception (Olivers and 114 Meeter, 2008), which is called the attentional blink (AB). The boost and bounce theory of 115 on emotional stimuli. On the contrary, high similarity between targets and distractors can 149 account for an increased AB as shown in TD (Müsch et al., 2012;Visser et al., 2004). 150 The aim of our current study was to investigate the link between visual consciousness of 151 emotional face stimuli, operationally defined by the AB, and fundamental low-level visual 152 processing, operationally defined by using spatial frequencies, in autism compared to TD 153 participants. We created an AB task, where participants had to detect two unfiltered happy 154 faces (T1 and T2) in a stream of angry faces distractors, which were either in LSF, HSF or 155 unfiltered. We chose happy and angry emotions because it is easy to differentiate them. 156 Happiness is not confused with other emotions and the salient and distinctive smile of happy 157 faces is easily identified, probably because the recognition of facial expression relies more on 158 perceptual processing than on affective dimensions (Calvo and Nummenmaa, 2016). Happy 159 faces are usually well recognized by autistic participants (similarly as control; for a meta-160 analysis, see Uljarevic & Hamilton, 2013), but this might be the result of efficient 161 compensatory strategies based on cognitive and linguistic resources (Harms et al., 2010). 162 Hence, using happy faces enabled us to focus on the role that HSF and LSF played in visual 163 consciousness of emotional faces in autism, while keeping the task feasible by autistic 164 participants. We capitalized on the finding that spatial-frequency filtered distractors would 165 allow us to manipulate similarity between targets and distractors: the more similar the target 166 and the distractors would be, the greater the AB. We reasoned that, manipulating spatial 167 frequency filtering of the distractors, more than targets, was an interesting way to investigate 168 incidentally which type of low-level information is important for visual consciousness, while 169 keeping a target ecologically intact. In this experiment, we tested three main hypotheses. 170 Firstly, we predicted a strong AB effect in both groups, as distractors have a high degree of 171 similarity with targets because they are all faces. Secondly, given a pivotal role that HSF 172 plays in visual consciousness during face processing, we expected a stronger AB after HSF 173 and unfiltered distractors (as it also contains HSF) compared to LSF distractors in both 174 groups. Lastly, due to the HSF bias for emotional face recognition in ASD, autistic 175 participants would be significantly more disturbed by HSF distractors than TD participants. 176 Similar to other AB studies, we controlled for T1 accuracy. We also assessed the Full investigating spatial frequencies in a AB task allowing a proper power analysis. Thirty-three 186 autistic adults (15 females, 15 males, 2 transgenders female-to-male) and 35 TD control adults 187 (17 females, 17 males, and 1 transgender female-to-male) were recruited for this study. We 188 recruited a similar number of males and females on purpose, as a second task was performed 189 by the participants after the AB task, in which sex differences were investigated. One autistic 190 man was excluded from the analysis as the task was too difficult for him and he skipped all Disease 10 th revision (OMS and Collectif, 1992) by asking them to show us the written report 202 of their diagnosis. We collected the scores of the Autism Diagnostic Observation Schedule 203 (Lord et al., 1989), the Autism Diagnostic Interview-Revised (Lord et al., 1994) and IQ when 204 they were available. Their age at diagnosis was between 10 and 45 years old (mean = 28.5, 205 SD = 10.1). Three of them also received a diagnosis of attention deficit with or without 206 hyperactivity disorder (one of them was treated with methylphenidate); two received a 207 diagnosis of anxiety disorder and two had a history of traumatic brain injury. Two were 208 receiving neuroleptic medication and six were used to take antidepressant when they were 209 enrolled in the study but were stabilized with the treatment. Twenty-one autistic participants 210 had neither other diagnosis nor treatment. For those who could not provide available IQ data 211 (8 individuals), an estimation of their FSIQ, VIQ, and PIQ was performed using four selected 212 subtests (Vocabulary, Similarities, Block Design and Matrix) of the WAIS-IV (Grégoire and 213 Wierzbicki, 2009;Wechsler, 2008). Note that 10 ASD participants scored below the cut-off of 214 32 on the AQ. We kept them as AQ does not fully capture autism symptoms in autistic 215 persons who have poor insight (Bishop and Seltzer, 2012), showed by a weak correlation with 216 ADOS scores sometimes (Ashwood et al., 2016). Indeed, AQ scores were primary collected 217 to remove TD participants with high AQ but was not aimed to exclude ASD participants as 218 they have been diagnosed by expert professionals. Detailed information and scores for each 219 participant can be found in the following repository : 220 https://osf.io/maqvz/?view_only=3002780bdf8247ecb3669e9870b4c00d. 221 TD adults were recruited via advertisements and mailing list. They didn't have any 222 neurological, neurodevelopmental or psychiatric diagnosis. As far as possible, groups were 223 paired on sex, age and education. An estimation of their FSIQ, VIQ, and PIQ was performed. 224 After having checked assumptions, we performed two-sample t-test. Groups did not 225 differ on age, education, VIQ, and FSIQ. Nevertheless, it is worth noting that differences 226 between groups on FSIQ is approaching the significance threshold and may be related to the 227 significant difference between groups on PIQ. As predicted, groups differed on the AQ. All 228 relevant statistics regarding groups description are set out in Table 1. 229    We calculated the correct response rate (accuracy) for T2 given that T1 was correctly reported 323 (Gaigg and Bowler, 2009;Shapiro et al., 1997;Yerys et al., 2013) for each of the 324 experimental conditions. This latter is abbreviated "T2 accuracy" in the following text. Figure  325 3 represents the T2 accuracy for each group (ASD, TD) as a function of Lag (3, 5, and 7) and 326 Distractor (BSF, HSF, and LSF). All means and standard errors for T2 accuracy are reported 327 in Table 2. We conducted a mixed-design ANOVA with Group (ASD vs TD participants) as a 328 between subject factor, and Lag (3, 5 or 7) and Distractor (BSF, LSF or HSF) as within-329 subject factors. Assumptions of sphericity were tested with Mauchly Test. As there was no 330 violation of sphericity, no correction was required. The normality assumption on residuals 331 were visually inspected with histogram and qq-plot, and tested with a Shapiro-Wilk test. 332 The analysis revealed three significant main effects. The main effect of Group 333 (F (1, 65) = 4.14, p = .04, η 2 = .04), indicated that overall T2 accuracy of ASD participants 334 was significantly lower than overall T2 accuracy of TD participants. A main effect of Lag 335 (F (1, 85) = 104.8, p < .001, η 2 = .17) was also found. Post-hoc paired t-test with Tukey 336 adjustment showed that T2 accuracy at Lag 3 was worse than T2 accuracy at Lag 5 337 The data set out in Figure 3 suggests that the lack of significant difference between BSF and 356 LSF at Lag 7, which was not expected, could be explained by a different pattern in each 357 group. This led us to perform within-group paired t-test with Tukey adjustment although the 358 foregoing analysis provided no interaction involving the group factor (Group × Distractor: LSF distractors at longer lags (5 and 7). These differences between groups could therefore 370 explain the lack of significance previously found at Lag 7 between BSF and LSF in the main 371 effect and in the interaction. Moreover, the analysis showed a difference in T2 accuracy 372 between BSF and HSF at Lag 7 (t (390) = 2.39, p = .04, d = 0.30) for the ASD group, 373 suggesting greater AB after HSF distractors compared to BSF distractors at the longest lag. 374 Finally, we tested our main hypothesis: HSF distractors would produce a greater AB than LSF 375 distractors for the ASD group as compared to the control group. We performed between-376 group paired t-tests with Tukey adjustment for each distractor type and within each Lag. The 377 expected between-group difference on T2 accuracy after HSF distractors was only found at 378 Lag 7 (t (161) = − 2.21, p = .03, d = − 0.46). Differences between groups on T2 accuracy 379 were also found after LSF distractors at Lag 5 (t (161) = − 2.31, p = .02, d = − 0.55) and at 380 Lag 7 (t (161) = − 3.00, p = .003, d = − 0.73). These unexpected findings indicate that LSF 381 distractors also produce a greater AB for the ASD group as compared to the TD group at 382 longer lags (5 and 7). No between-group differences in T2 accuracy were found after BSF 383 distractors. These results are represented on Figure 3. 384 385 386

387
We calculated a correct response rate for T1 for each participant. The latter is abbreviated by 388 "T1 accuracy" in the text. As variances were unequal, we performed a Mann-Whitney-389 Wilcoxon test. The test was not significant, indicating that T1 accuracy of autistic participants 390 respectively. We did not find any significant correlation between T2 accuracy and AQ, FSIQ, 397 nor VIQ. A significant negative correlation between age and T2 accuracy was found for the 398 TD group only (r = − .43, p < .01), indicating that the older the TD participants, the stronger 399 the AB. We also found a significant negative correlation between age and T1 accuracy in the 400 TD group (r = − .35, p < .05), revealing that the older TD participants are, the lower is their 401 T1 accuracy. Finally, we found a significant positive correlation between T1 accuracy and 402 FSIQ in the TD group (r = .41, p < .05) and in the ASD group (r = .39, p < .05), indicating 403 that the higher FSIQ is, the better is the T1 accuracy. We also found significant positive 404 correlations between T1 accuracy and PIQ in the TD group (r = .46, p < .01) and in the ASD 405 group (r = .40, p < .05), revealing that the higher the PIQ is, the better is the T1 accuracy. Although the timings of Lag 5 (665 ms) and 7 (931 ms) would usually be considered outside 421 of the AB timeframe, we suggest that the AB is observed in longer lags here due to the task 422 difficulty. Indeed, the AB magnitude can be modulated by factors such as task demands 423 (Elliott and Giesbrecht, 2010;Shore et al., 2001). 424 We also found a stronger AB after HSF distractors and after BSF distractors as 425 compared to LSF distractors, at Lag 5 and at Lag 7 in both groups taken together, which is 426 consistent with our second hypothesis. This result provides new evidence that HSF could play 427 a primary role in visual consciousness, particularly of emotional faces, as previously 428 suggested (De Gardelle and Kouider, 2010; Schyns and Oliva, 1994). As HSF are mainly 429 conveyed by the ventral stream, our findings could support the idea that visual consciousness 430 emerged, at least in part, from the temporal pathway processing (Navajas et al., 2013;Quiroga 431 et al., 2008;Sergent et al., 2005). It should be noted that the difference between BSF and LSF 432 at Lag 7 was only marginally significant. This could be explained by differences between 433 groups on LSF distractors, which will be further discussed. No difference between distractors 434 at Lag 3 can be explained by the strength of the AB at that lag. groups on T2 accuracy according to distractor types at Lag 3, probably due to the strong AB 441 effect at Lag 3 for both groups, group differences emerged at Lags 5 and 7. Of particular 442 importance is lower T2 accuracy for autistic participants, compared to TD, after HSF 443 distractors at Lag 7, which partially support our hypothesis. Interestingly, at Lags 5 and 7, our 444 analysis also revealed that, in the ASD group, T2 accuracy after BSF were not lower than T2 445 accuracy after LSF, contrary to the control group and T2 accuracy for the ASD group after 446 LSF distractors was unexpectedly lower as compared to the TD group. Additionally, we did 447 not observe differences between groups on BSF. The latter finding indicates that the main 448 effect of group observed on T2 accuracy in our experiment could be more related to the 449 filtered nature of the distractors than to the emotional content, which could have reduced the 450 AB in TD participants compared to ASD as it has been observed by Gaigg & Bowler (2009). to study brain reactions to an unexpected event occurring among regular events and can be 495 seen as an electrophysiological signal of perceptual prediction errors (Stefanics et al., 2014). 496 Autistic people often elicit atypical reactions to deviant stimuli in Mismatch paradigm 497 (Gomot and Wicker, 2012). In our experiment, the second target appeared after a stream of 498 distractors containing the same low-level characteristics (i.e., the same type of filtering was 499 used for each distractor). It is reasonable to assume that participants became used to low-level 500 characteristics of the distractors (at Lag 5 and 7) and made strong predictions for the next 501 image, based on regularity. When distractors were low-pass or high-pass filtered, the 502 appearance of the unfiltered second target T2 could violate the expectation generated by the 503 statistical regularity of the stream (based on low level information) leading to prediction 504 errors. The disruption of the statistical regularity of the stream may impact autistic 505 participants more, which leads to impaired T2 accuracy. In other words, individuals with 506 autism would struggle to disengage from their prior as they expect to see the same low-level 507 information. The more effortful adaptation to the change in low-level features in T2 would 508 prevent them to memorize the face, subsequently impairing their accuracy. By contrast, TD 509 can more flexibly adapt the disruption of the statistical regularity and could stay concentrated 510 on the task, without being disturbed by low-level changes. In the case of BSF distractors, 511 targets and distractors had the same low-level content, generating no prediction error related 512 to low-level characteristics of the stimuli (i.e., spatial frequencies filtering), which may 513 explain the absence of between-group differences for BSF distractors. This Mismatch 514  Raymond et al., 1995;Yerys et al., 2013). In addition, the answer choice in our task was 527 presumably harder as compared to other similar tasks and may affect performances. Indeed, 528 our participants have to identify the T1 face between three happy faces (and the same for T2). 529 Finding a target face among three faces, without answering by chance, is more difficult than a 530 dichotomous yes/no answer on a question such as "Did you see a face?". Additionally, we 531 gave the instruction not to answer by chance, giving the possibility to choose "I don't know". 532 The general repartition of answers on T2 was the following: 38.6% "I don't know", 0.5 % No 533 answer, 33.5% correct answers and 27.4% incorrect answers. Thus, considering only the trials 534 when participants answer (i.e., the participant has a slight to strong degree of certainty 535 regarding his answer), participants chose the correct answer in more than 50% (chance level 536 at 33%). To better account for the confidence of participants, it could be interesting to employ 537 a method used in Eiserbeck and Abdel Rahman (2020) in future research instead of offering 538 the possibility to answer "I don't know". They used an AB paradigm with faces using 539 objective as well as subjective criterion (i.e., degree of certainty of the participant regarding 540 T2) as an index of T2 detection (Eiserbeck and Abdel Rahman, 2020). Whereas correct T2 541 report with a "slight impression" was between 70 % and 74 % (which is in accordance to 542 usual T2 accuracy in AB task), correct T2 report with a "strong impression" was between 543 33 % and 36 % (which is similar to T2 accuracy in our task). 544 The task difficulty probably impaired T1 recognition as well. Indeed, we observed 545 apparent poorer T1 accuracy in our experiment with greater variability as compared to T1 546 accuracy usually observed in other experiments. T1 accuracy was similar in both group, in 547 accordance with intact happy faces identification in autism (Uljarevic and Hamilton, 2013). A 548 correlation exists between T1 accuracy and FSIQ, as well as between T1 and PIQ for both 549 groups, which could be in line with previous studies showing that higher IQ can enhance 550 emotion recognition in both TD and ASD participants (Jones et al., 2011;Wright et al., 2008). 551 Nevertheless, when we removed the few outliers in term of IQ performances (2 ASD and 2 552 TD participants), the correlations between T1 accuracy and FSIQ and PIQ were not 553 significant anymore, although T1 accuracy still not significantly differed between groups. 554 Correlation between FSIQ and T2 accuracy was not significant neither, which is congruent 555 with previous research (Colzato et al., 2007) and contrary to Gaigg & Bowler (2009), we did 556 not find any correlation between VIQ and T2 accuracy in ASD (nor in TD), which can 557 probably be explained by the non-verbal nature of our task. Taken together, these findings 558 support the fact that IQ variations among our subjects have not impaired our results. Finally, 559 we found a significant negative correlation between accuracy (for T1 and for T2) and age in 560 TD, performances becoming weaker with age, but this is not found in ASD. AB performances 561 tend to improve between 18 and 39 years old, and then to decline (Georgiou-Karistianis et al., 562 2007). Hence, the negative correlation could be at odds with the fact that only 9 TD 563 participants are aged 39 and above. However, a fast decrease of performances after 39 might 564 explain this result. Surprisingly, this decline is not observed in ASD participants despite they 565 are a bit older than TD participants in our study (ASD age rank = 19 -47, median = 33.5; TD 566 age rank = 19 -44, median = 31.0). Two hypotheses can be brought up. First, it may be 567 related to high IQ of ASD participants as compared to TD in our study. Indeed, higher IQ is 568 usually associated with less cognitive decline (Steffener and Stern, 2012). Second, ASD might 569 also partially protect against a cognitive decline related to age (Lever and Geurts, 2015), 570 which is yet to be determined. 571 572

573
Our study has some limitations pertaining to participants. Firstly, autistic participants 574 were adults, mainly late diagnosed and had very high intellectual abilities, thus representing a 575 small subset of autistic adults. Hence, our results cannot be generalized to the whole 576 spectrum. Secondly, there is a high variability in performances of autistic participants, 577 probably reflecting the highly heterogeneous autism spectrum or some other aspects such as, 578 the degree of alexithymia. Data driven analysis might be conducted to distinguish autistic 579 subgroups (see Latinus et al., 2019), but would require larger sample size. Thirdly, the 580 correlation found with age could potentially have a slight impact on the result and should 581 conduct to lower the maximal age in further studies. Finally, we were not able to collect 582 ADOS data from all participants (we collected ADOS scores of 19 participants) although it 583 would have been interesting to study correlation between these scores and performances in 584 addition to the AQ, as the later have sometimes weak correlation with ADOS (Ashwood et al., 585

2016). 586
There is another limitation pertaining to the experiment. Our experiment included happy 587 targets only because it has been suggested that people, even with autism, can recognize happy 588 emotion easily. This selection was intentional to focus on low level visual processing, which 589 is the main research hypothesis. Further studies need to determine if the effect of spatial 590 frequencies reported here is specific to happy faces, or could be observed with other 591 emotional facial expression, with non-social stimuli (e.g. International Affective Picture 592 System) and non-emotional stimuli (e.g. gender, objects, natural scenes). We thank all participants and their families for their help in this study. We also thank the 631 Expertise Center for Asperger of Grenoble, the Savoyard Center for Autism Evaluation, Nelly 632 Coroir and Jérôme Ecochard for their help with volunteer recruitment. Finally, we thank Eric 633 Guinet for his invaluable assistance with screen tests and calibration. 634

Disclosure statement 635
The authors declare that they have no competing interests. 636

Data availability statement 637
The datasets generated and/or analyzed during the current study are available in the Open 638