Pupil dilation during visuospatial orienting differentiates between autism spectrum disorder and attention-deﬁcit/hyperactivity disorder

,


Introduction
Attention, the ability to 'take possession of the mind in clear and vivid form' (James, 1890, p. 403), enables us to select and focus on internal and external stimuli, either consciously, but also when salient or unexpected stimuli perpetrate our awareness (Corbetta & Shulman, 2002).Based on behavioral ratings, attention problems are ubiquitous among children referred to child and adolescent psychiatry services, regardless of their actual diagnosis (Schmeck et al., 2001).However, behavioral ratings do not differentiate between cognitive and neural mechanisms underlying attention problems.A considerable body of research within the fields of cognitive psychology and neuroscience demonstrated the functional separation between different attentional modules (Raz & Buhle, 2006).Separate attention networks have been derived, which serve the attention functions of alerting, orienting, and executive control (Petersen & Posner, 2012).
The dorsal frontoparietal attention network supports orienting attention to central, expected, and exploitable stimuli whereas the ventral frontoparietal attention network facilitates the reallocation of attention to peripheral, unexpected, and explorable stimuli (Corbetta, Patel, & Shulman, 2008;Kim, 2014).Increasing alertness after an alerting cue for an upcoming target is regulated by the subcortical locus coeruleus-norepinephrine (LC-NE) system (Aston-Jones, & Cohen, 2005;Petersen, & Posner, 2012).More specifically, through NE-modulated recruitment of the ventral attention network, the LC-NE system modulates arousal-dependent sensory and cognitive processing of salient information (Vazey, Moorman, & Aston-Jones, 2018), such as an alerting cue, and hence plays a critical role in regulating various attention functions during task performance (Sara, & Bouret, 2012).
Task-evoked pupil dilation (PD) has been shown to index phasic LC activity in monkeys (Joshi, Li, Kalwani, & Gold, 2016) and humans (Murphy, O'Connell, O'Sullivan, Robertson, & Balsters, 2014).Task-evoked phasic LC activity modulates NE-induced adaptive gain in synaptic signal transmission, promoting task engagement (Aston-Jones & Cohen, 2005;Gilzenrat, Nieuwenhuis, Jepma, & Cohen, 2010).Continuous measures of PD during task performance (PD progression) hence connect behavioral performance directly to functional indices of brainstem activity.PD during cognitive tasks has previously been implicated as biomarker in ADHD (Wainstein et al., 2017) and ASD (Blaser, Eglington, Carter, & Kaldy, 2014;), but to the best of our knowledge, no study so far compared PD as index of LC activity during attention performance between ADHD and ASD.Comparing PD progression during a cued visuospatial orienting task allows us to gain further insight into subcortical processes underlying atypical attention in ADHD and ASD.The present study compared PD progression, the cue benefit effect, and reaction time measures between children with ADHD without ASD (ADHDÀ), ASD without ADHD (ASDÀ), both disorders (ASD + ADHD), and typically developing children (TD) during a cued visuospatial orienting task.We hypothesized that children with ASD and ASD + ADHD would show atypical phasic PD responses when orienting attention to relatively unexpected spatial locations, and slower reaction times relative to TD and ADHD associated with reflexive orienting (Keehn et al., 2013).Second, we hypothesized that children with ADHDÀ and ASD + ADHD would show atypical cue-evoked PD responses across alerting cues relative to children with ASD-and TD.Finally, the modulatory role of the LC-NE system in atypical attention and task performance was examined exploratory by comparing group differences in the effect of task-evoked PD on reaction time.

Participants and ethical considerations
Twenty-eight children with ADHDÀ, 18 children with ASDÀ, 14 children with ASD + ADHD, and 31 TD were included.The ethical approval was obtained from the Ethical Committee of the Department of Medicine at the Goethe University Frankfurt (46/16).Informed consent and assent were obtained.Participants were between 8 and 18 years old, with estimated full IQ > 70 (HAWIK-IV; Petermann, 2012; WAIS-IV; Petermann & Petermann, 2007; Table 1).TD showed below clinical cutoff values for the Child Behavior Check List (CBCL) total score (Schmeck et al., 2001; Table 1).ADHD and ASD diagnoses were established according to DSM-5 (American Psychiatric Association, 2013) by experienced clinicians.ADHD diagnosis was additionally confirmed by a semistructured diagnostic interview with a primary caregiver (K-SADS-PL, adapted to DSM-5; Kaufman et al., 1997) and ASD diagnosis with the Autism Diagnostic Observation Schedule (ADOS; R€ uhl, B€ olte, Feineis-Matthews, & Poustka, 2004) and the Autism Diagnostic Interview-Revised (ADI-R; B€ olte, R€ uhl, & Schm€ otzer, 2006).Exclusion criteria for all samples were current depressive episodes, bipolar disorder, schizophrenia, and conduct disorder, assessed by the K-SADS-PL (parent interviews).Participants on stimulant medication were asked to withdraw 24 hr before the assessment.

Procedure
Participants completed a visuospatial orienting task in a quiet and dimly lit (110 lux) testing room.Heads were placed on a chin rest to prevent excessive head movements.

Measures
Behavioral ratings.ADHD symptom severity scores were obtained by the ADHD rating scale for parents (FBB-ADHD;D€ opfner & Lehmkuhl, 2000).Parents rated ADHD symptom severity during the past six months on 18 items based on DSM-IV and ICD-10 criteria, scaled from 0 (nonexistent) to 3 (strongly pronounced).
ASD symptom severity scores were obtained using the Social Responsiveness Scale (SRS) (B€ olte & Poustka, 2008).Parents rated ASD symptom severity during the past six months on 65 items based on DSM-IV-TR criteria and other characteristics indicative of ASD, scaled from 0 (never true) to 3 (almost always true).
Cued visuospatial orienting task.PD progression and RT were recorded during a task based on the Posner cueing paradigm (Posner, 1980) (Figure 1).Participants were instructed to press a button as soon as they detected a tadpole (target), but refrain from responding when a fish (distractor) was detected.Following a 500 ms fixation phase to a central cross surrounded by a circle of puddles, either an arrow (specific cue) or circle (nonspecific cue) was presented for 1000 ms (size of.3 9 4°), followed by the distractor or target randomly appearing in any of the eight surrounding puddles (size of 2 9 3°; eccentricity of 6°), with presentation times of either 100 or 300 ms.The arrow indicated the puddle in which either the distractor or the target appeared.Cues and fixation cross had the same luminance.Each task consisted of 120 trials.

Recording
and preprocessing eye-tracking data.PD progression data were recorded using a Tobii X2- 30 binocular eye tracker (Tobii Technology AB, Sweden).A 5point calibration was done.Display resolution was 1024 9 768 pixels.PD data were preprocessed and analyzed in R statistics 3.4.3(R Core Team, 2017).First, raw PD data were controlled for fixations.PD data were only included if corresponding fixations were on screen center during baseline and cue presentations, and within stimulus display area (screen center: AE16.3°) during stimulus presentations.Second, PD data were controlled for sampling variation of the eye tracker (30 Hz AE 2 Hz).PD data were included only if corresponding sampling intervals deviated less than 1.5 SDs (8.3 ms) from the mean (33.3 ms), and if samples were recorded within the respective trial phase (e.g., samples 1-15 within fixation phase).Third, all PD data of implausible size (PD < 2 or >8 mm) and poor validity rating (range: 1-4) were excluded.Finally, absolute PD was calculated as the mean of both eyes.When tracking was unsuccessful for one eye, only data from the successful eye were selected.PD progression gaps smaller than 100 ms were linearly interpolated.Baseline PD was calculated as trial-specific mean during the fixation phase.Relative PD was calculated as absolute PD divided by baseline PD and applied in all analyses.
PD progression metrics.Visual inspection of the PD progression revealed three task-evoked PD responses following cue and stimulus onset, and behavioral response, which were used to calculate amplitude (amp) (1) and latency (lat) PD metrics (2) for each PD response (see Figure 2) following the rationale of previous research (Fan, Miles, Takahashi, & Yao, 2009).
Index k was the within-trial sample (range = 1-90) and was set corresponding to PD responses (cue: k = 21, stimulus: k = 45, and behavioral response: k = 67).Function t retrieved time since trial onset.Time intervals (k-10; k + 10) were chosen by visual inspection of the overall PD progression (see Figure 2).Thus, amplitude and latency metrics refer to intervals 330 ms around respective PD responses.In addition, baseline PD for each trial was calculated as mean PD of the first ten samples (k = 1-10).Amplitude outliers were excluded based on 95% confidence intervals.Latency metrics were calculated for complete observations only.Correlations among PD progression metrics, ASDÀ and ADHD symptom severity and RT are displayed in Table S1.
Statistical analysis.Group differences in PD progression metrics and the effect of PD progression metrics on zstandardized reaction time measures (RT) were analyzed by linear mixed models (LMM) with trial number as random slope varying across participants.RT was corrected for premature responses (x < 200 ms) (Semmelmann & Weigelt, 2017).A backward approach for model selection was implemented.Full models included dummy-coded group as predictors (ASD: yes, no; ADHD: yes, no), cue (specific cue vs. nonspecific cue), and age and sex as covariates.Stimulus type (distractor vs. target) and stimulus duration times (300 ms vs. 100 ms) were additionally included when analyzing group differences in stimulus and behavioral response PD metrics and RT.Reduced models included dummy-coded group main and interaction effects, cue, and significant interaction and covariate effects of full models.PD progression metrics were included as predictor in our exploratory analysis on group differences in the effect of PD progression metrics on RT.Likelihood ratio tests were used to compare goodness of fit of full models to reduced models and a baseline model.p-Values were false discovery rate (FDR) corrected for the number of estimated models (k = 30) (Benjamini & Hochberg, 1995).For the best fitting models, standardized beta coefficients with 95% confidence intervals (CI) of significant predictor and covariate effects are reported.Finally, estimated marginal means or coefficients with corresponding 95% CI were calculated to test for specific ASD-and ADHD group effects in PD metrics, RT, and the effect of PD metrics on RT.

Results
Sample characteristics are reported in Table 1.Groups did not differ in age, sex ratio, and performance IQ scores.Verbal IQ scores were lower in ADHD-compared with TD and ASD + ADHD.All clinical groups scored higher than TD on clinical symptom outcome measures.
Model fit comparison results showed better fit for reduced models compared with full and baseline models in testing group, task, and covariate effects on RT and all PD progression metrics except cueevoked amplitudes, for which a full and reduced model did not fit the data significantly better than a baseline models (Table S2).
Results additionally demonstrated an interaction effect between ASD and cue on stimulus-evoked PD amplitudes, in the absence of an ASD main effect (Table 2).Stronger stimulus-evoked PD amplitude responses were observed in ASD (+ADHD) relative to TD and ADHD-following nonspecific relative to specific cues (Figure 5).

Group differences in the effect of PD progression metrics on RT
Full models estimating group differences in the effect of baseline PD, and response-locked PD amplitudes, and cue latency on RT showed better fit than reduced and baseline models.Reduced models estimating group differences in the effect of the other PD progression metrics on RT showed better fit than full and baseline models (Table S3).
The effect of stimulus-evoked PD amplitudes on RT differed between ASD relative to ADHD and TD for  different cue types (Table S4).Post hoc comparison showed that larger stimulus-evoked PD amplitudes across specific cue trials predicted slower RT in ASD relative to TD and ADHD (coef diff = 1.81, 95% CI: 0.04-2.97).Across nonspecific cue trials, the effect of stimulus-evoked amplitudes on RT did not differ between ASD and TD and ADHD (coef diff = À1.35,95% CI: À0.27-2.97).No other group differences were found (Table S4).

Discussion
Previous cognitive and neuropsychological findings suggested different attention impairments related to visual orienting in ADHD and ASD: the processing of alerting cues in ADHD and processing and orienting to relatively unexpected targets in ASD (Lau-Zhu, Fritz, & McLoughlin, 2019).The LC-NE system is implicated in exploiting alerting cues to increase alertness and detecting task-relevant stimuli (Aston-Jones, & Cohen, 2005;Petersen, & Posner, 2012), and as such in modulating the optimal attentional state during visual orienting.The goal of the present study was to elucidate whether differential subcortical processes underlie visual orienting in ASD and ADHD.Pupil dilation (PD) progression to index LC activity and reaction times (RT) to index response execution were compared between children with ASD-, ADHD-, ASD + ADHD, and TD.Study findings supported moderately slower orienting responses in ASD-and ASD + ADHD to relatively unexpected spatial locations, accompanied by slightly higher PD amplitudes, and slightly shorter cue-evoked PD latencies in ADHD without comorbid ASD.
Our findings of moderately slower RT in ASD and ASD + ADHD when orienting attention to relatively unexpected locations corroborate previous findings of impaired reflexive, rather than voluntary orienting responses in children and adults with ASD (Keehn et al., 2013).Furthermore, our results provide new evidence for slightly increased PD in ASD related to reflexive orienting.Increased PD in ASD has been proposed to reflect a persistent hyperphasic state, promoting an enhanced attentional focus, albeit at the expense of impeding attentional disengagement (Aston-Jones, Iba, Clayton, Rajkowski, & Cohen, 2007;Kaldy, Giserman, Carter, & Blaser, 2016).Increased PD in ASD has been previously observed in a visual search task (Blaser, Eglington, Carter, & Kaldy, 2014), during which participants searched for target stimuli amidst several competitors.In the present study, participants needed to engage and subsequently move their focus of attention either voluntary based on a specific (i.e., spatially directive) In the present study, Increased PD in ASD was furthermore selectively evoked by targets following reflexive orienting.Apart from an enhanced focus, increased PD may index increased task difficulty or cognitive load (Rondeel, van Steenbergen, Holland, & van Knippenberg, 2015).The increased PD responses in ASD observed in the present study may thus reflect a higher effort to control the focus of attention during reflexive orienting in ASD.
In ADHD, slightly shorter cue-evoked latencies were observed relative to all groups, which can be interpreted as faster but also attenuated cue processing (Isabella et al., 2019).Support for faster processing is provided by findings suggesting children with ADHD benefit relatively more from alerting cues to improve task performance (Johnson et al., 2008;Samyn et al., 2017).In the present study, however, cue-evoked latencies were not associated with task performance, and faster cue-evoked latencies were hence unrelated to faster responses.Alternatively, shorter processing times may indicate that children with ADHD invested less attentional resources in processing alerting cues.Attenuated cue processing in ADHD has been observed previously in ERP paradigms (Kratz et al., 2011;Ortega, L opez, Carrasco, Anllo-Vento, & Aboitiz, 2013), and corroborates the hypothesis on poor state regulation in ADHD, that predicts children with ADHD will allocate less attentional resources when processing an alerting cue (van der Meere, 2006).
A limitation of the present study design was that the effect of an alerting cue could not be compared with a baseline condition, because trials without alerting cues were lacking.To further unravel subcortical processes underlying cue processing in ADHD, future studies should compare tonic-, taskevoked PD, and task performance between trials with and without alerting cues.Other limitations of the present study were the small sample of children with comorbid ASD and ADHD (N = 13), which was accounted for in our statistical analysis by including ASD and ADHD as dichotomous predictors.Males were overrepresented in all clinical groups relative to TD (Table 1).Sex was included as covariate in our main analyses, but replication of our findings using more balanced designs is warranted.Finally, it should be emphasized that ADHD-specific differences in cue-evoked PD latencies did not correlate with RT.A previous finding suggests that, in contrast with task-evoked PD amplitudes, task-evoked PD latencies and RT may reflect largely independent processes (Isabella et al., 2019).In the present study, only response-locked PD latencies correlated with RT (Table S1).Correlations between PD metrics and PD amplitudes were thus additionally lacking.Within ASD-, however, cue-and stimulus-evoked PD amplitudes correlated with RT.Taken together, the lack of correlations across groups except for cueand stimulus-evoked PD amplitudes within ASDmay additionally indicate an increased effort to optimize task performance in children with ASD- (Rondeel, van Steenbergen, Holland, & van Knippenberg, 2015).Nevertheless, to further examine how the LC-NE system modulates attention and task performance future studies should examine other, and more sensitive measures of (disorder-specific) response execution in addition to RT (Karalunas et al., 2018).

Conclusion and future directions
In sum, using a rather novel approach to analyse PD progression, the present study provided new evidence for a specific role of the LC-NE system in impaired reflexive orienting responses in ASD (independent of ADHD), indexed by increased PD amplitudes, and atypical processing of alerting cues in ADHD, as indexed by shorter cue-evoked PD latencies.
Future studies may further compare the role of the LC-NE system in atypical attention in ASD and ADHD and other psychiatric disorders, and address the impact of tonic LC activity on phasic discharge during task performance.Furthermore, connecting PD responses to cortical measures of attention could increase our understanding on how the LC-NE dysfunction influences cognitive performance on a neural systems level.Finally, comparing tonic and task-evoked PD across different experimental paradigms will help to unravel during which specific experimental manipulations disorder-specific LC dysregulation can be measured optimally.Unravelling the conditions during which attention problems arise within and across diagnostic boundaries is not only important to differentiate between separate cognitive and neural mechanisms underlying attention problems, but may also help to improve behavioural and pharmacological interventions in clinical practice.

Figure 1 Figure 2
Figure 1 Task design and stimuli [Colour figure can be viewed at wileyonlinelibrary.com] 95% Confidence Intervals; ADHD, Attention-Deficit/Hyperactivity Disorder; ASD, Autism Spectrum Disorder; b, standardized beta regression coefficient.a Full and reduced model fit not better than baseline model.b Not significant in full model, best model fit for reduced model.c Not included in full model because of relevance.

Figure 3 Figure 4
Figure 3 Slower reaction times after nonspecific cue in ASD relative to ADHD-and TD and in ASD + ADHD relative to TD. Error bars represent 95% confidence intervals.ADHD, Attention-Deficit/Hyperactivity Disorder; ASD, Autism Spectrum Disorder; RT, median reaction time; and TD, typically developing

Figure 5
Figure 5 Stronger stimulus-evoked PD amplitude responses in ASD(+ADHD) relative to TD and ADHD after nonspecific relative to specific cues.Error bars represent 95% Confidence Intervals.ADHD, Attention-Deficit/Hyperactivity Disorder; ASD, Autism Spectrum Disorder; PD, pupil dilation; and TD, typically developing

Table 2
Group and task parameter effects on PD progression metrices