Hypo- and hyper- sensory processing heterogeneity in Autism Spectrum Disorder

Background. Sensory processing atypicalities are part of the core symptoms of autism spectrum disorder (ASD) and could result from an excitation/inhibition imbalance. Yet, the convergence level of phenotypic sensory processing atypicalities with genetic alterations in GABA-ergic and glutamatergic pathways remains poorly understood. This study aimed to characterize the distribution of hypo/hyper-sensory prole among individuals with ASD and investigate the role of deleterious mutations in GABAergic and glutamatergic pathways related genes in sensory processing atypicalities. Method. From the Short Sensory Prole (SSP) questionnaire, we dened and explored a score – the differential Short Sensory Prole (dSSP) - as a normalized and centralized hypo/hypersensitivity ratio for 1136 participants (533 with ASD, 210 rst-degree relatives, and 267 controls) from two independent study samples (PARIS and LEAP). We also performed an unsupervised item-based clustering analysis on SSP items scores to validate this new categorization in terms of hypo and hyper sensitivity. We then explored the link between the dSSP score and the burden of deleterious mutations in a subset of individuals for which whole-genome sequencing data were available. N: number of participants; Age in years; IQ: Intelligence quotient; ADI-R: Autism diagnostic interview-revised; ADOS-2 CSS: Autism diagnoctic observation schedule version calibrated severity score. ADOS-2 CSS Calibrated Severity Scores; Charman et al. 2017


Background
Alongside impairments in social communication and restricted, repetitive behaviours, sensory atypicalities are now considered as being a core de ning diagnostic feature of ASD (DSM-5, American Psychiatric Association, 2013). Reported in up to 87% of individuals with ASD, sensory processing atypicalities have been emphasized as a critical feature for characterizing and understanding ASD (Dellapiazza et al., 2018;Uljarević et al., 2017). However, the sensory pro les described among individuals with ASD are heterogeneous, hampering progress to understand their biological substrates (Charman, To improve the characterization of sensory processing atypicalities in ASD and facilitate the exploration of their biological subtracts, we sought to determine a score summarizing the sensory symptom directions and heterogeneity. We aimed to construct a summarizing score as a normalized and centralized ratio of the hypo-to hyper-sensitivity derived from the Short Sensory Pro le (SSP), a commonly used questionnaire to explore the sensory symptoms frequently reported in individuals with ASD. We hypothesized that this differential SSP score (dSSP), when different from 0, would re ect the disequilibrium of the E/I balance. To achieve the rst goal of this study, we empirically de ned the hypoor hyper-sensory processing type of each item of the SSP questionnaire to calculate the dSSP score. We then tested the relevance of the dSSP score by quantifying and exploring its distribution in two independent samples of individuals with ASD and with typical development from the PARIS and the EU-AIMS study samples (n=1136). We also confronted the empirical construct of the dSSP score to a more data-driven procedure based on a clustering approach of all the SSP questionnaire items. The second step of our study was to explore if the dSSP score could facilitate the exploration of the biological background involved in the sensory peculiarities reported in the individuals with ASD. As a very preliminary study, we aimed to conduct a genetic study exploring in a modest sample of subjects the relationship between the dSSP score and the burden of deleterious mutations affecting genes of the glutamatergic and the GABAergic pathways. We hypothesized the variability of the dSSP score would be correlated with the load of deleterious mutations affecting the glutamatergic and GABAergic equilibrium.

Participants
In total, 1,136 participants were enrolled in the study from two independent study samples. The Paris Autism Research International Sibpair (PARIS) study sample included 165 individuals with ASD, their 210 unaffected rst-degree relatives and 97 individuals with typical development at the Child and Adolescent Psychiatry Department, Robert Debre Hospital, Paris (France). The EU-AIMS Longitudinal European Autism Project (LEAP) sample study was composed of 384 individuals with ASD and 280 individuals with typical development (Loth et al., 2017). The demographic and clinical characteristics of the individuals enrolled in both samples are reported in Table 1. The clinical characterization of the participants from the PARIS and LEAP study were included following the method described elsewhere (Lefebvre et al. 2021;Charman et al., 2017;Loth et al., 2017). As a summary, the diagnosis of ASD was based on DSM-IV-TR/5 criteria and made by summing the information from the Autism Diagnosis Interview-Revised (ADI-R), the Autism Diagnostic Observation Scale -second edition (ADOS-2), and clinical reports from experts in the eld. The cognitive abilities of individuals were also assessed using the Wechsler Intelligence Scales adapted to the age of individuals.

Sensory pro le of individuals included in the study
The sensory pro le of all participants included in the study was assessed with the Short Sensory Pro le (SSP; Tomchek & Dunn, 2007). Each item of this 38-item questionnaire is scored on a 5-point Likert scale (1=always, 2=frequently, 3=occasionally, 4=seldom, and 5=never). The SSP questionnaire includes all the items of the long version of the Sensory Pro le (Dunn & Westman, 1997), which demonstrated the highest discriminatory power of atypical sensory processing. The SSP questionnaire is composed of 7 subscales including tactile sensitivity, taste/smell sensitivity, movement sensitivity, visual/auditory sensitivity, under-responsive/seeks sensation, auditory ltering, and low energy/weak. Building of the dSSP score For the purpose of the study, we constructed a summarizing score as a normalized and centralized ratio of the hypo-to hyper-sensitivity derived from the SSP questionnaire. We rst asked four senior clinical experts in the eld of ASD and blind to the hypothesis of this study to determine which items of the scale were dedicated to the exploration of hyper-, hypo-sensitivity or which seemed to be related to both hyperand hypo-sensitivity. We then split the 38 items into three subgroups, depending on their ability to explore sensory features related to hyper-(n=19), hypo-(n=16) or both (n=3) sensitivity (Supplementary Table 1). We quanti ed inter-observer stability using Kendall rank correlation (Supplementary Table 2). We nally calculated for each individual the dSSP score, a centralized and normalized score re ecting the ratio between the average scores of items related to hypersensitivity (e.g. hyper-sensory score) (n=19) and to hyposensitivity (e.g. hypo-sensory score) (n=16) ( Table 2). A positive dSSP score represented a tendency to hyper-sensitivity and a negative one a tendency of hypo-sensitivity. a The hyper-, b hypo-, and c both sensory scores were rst explored in order to support the construction and use of the dSSP score. We observed differences of mean and variances between groups, with higher hypo and hyper sensory abnormalities in the ASD group in the two cohorts (Table 2).

PARIS cohort LEAP Cohort
Supplementary Figure. Distribution of subjects with ASD with mutations considering the dSSP score.
The mutations were explored in the GABA and glutamatergic pathways. Only the variants with Major Allele Frequency (MAF) <10% were considered.

SUPPLEMENTARY MATERIAL
Supplementary

Clustering distribution of the SSP questionnaire items
To validate the clinical-driven clusters (hypo, hyper and both), we confronted our clusters with data-driven clusters. We explored the link between the dSSP score and the sensory processing modalities using the VARCLUS procedure that divides a set of numeric variables into clusters (JMP software; SAS Institute Inc. 2017). This iterative method extracted oblique components to identify one-dimensional clusters of mutually correlated variables (Woolston et al., 2012). We used the SSP items scores reported by the participants with ASD from the PARIS and the LEAP study samples and then explored the items gathering in each cluster (e.g. item XX from cluster Y was categorized as hyper-or hypo-or both sensitivity). We used a Chi2 test to test for a signi cant relationship between variables and a Cramer's V test, which was a post-hoc test indicating how signi cant this relationship is (scoring from 0 for the low association to 1 for the high association). Using Python script (Python Software Foundation, version 3.7), we calculated the silhouette score (scoring from -1 to 1) (Rousseeuw, 1987) to interpret and validate the consistency within clusters and the Fowlkes-Mallows similarity score (Fowlkes & Mallows, 1983) to assess the similarity between PARIS and LEAP sample clustering (scoring from 0 to 1).

Genetic pro les
To explore if the dSSP score could facilitate the exploration of the biological background involved in sensory atypicalities reported in individuals with ASD, we conducted a genetic study exploring the relationship between the dSSP score and the burden of deleterious mutations affecting the glutamatergic and the GABAergic pathways. We thus performed this exploratory analysis only in a subset of individuals from the PARIS study samples for which whole-genome sequencing (WGS) data were available. For variant calling analysis, the pre-processing steps were as followed: sequence reads were aligned to the human reference genome GRCh37.75 using the Burrows-Wheeler Aligner BWA, then PicardTools was used for removing PCR duplicates, and GATK 3.8.1 was used for small insertion/deletion variants (Indels) realignment and base recalibration. Single Nucleotide Variants (SNV) and Indels were called with the GATK 3.8.1's HaplotypeCaller on each sample alignment le. We produced a Variant Call Format (VCF) le with all the SNV and Indel calls for the cohort. Variants were then functionally annotated with Variant Effect Predictor (VEP) (using Ensembl 92) (McLaren et al., 2016). Additionally, we annotated the variants for their frequency in the population from the gnomAD database version 2.1.1 (Karczewski et al., 2019) and for their Combined Annotation Dependent Depletion score (CADD version 1.3) (Kircher et al., 2014) to evaluate their deleteriousness. We then queried all variants with a minor allele frequency (MAF) ≤ 10% in the gnomAD database, which were either likely gene disruptive (LGD) variants (i.e. stop gain, stop loss, start loss, splice acceptor, splice donor or frameshift) or missense mutations with a CADD score ≥30 (MIS30). SNVs with a CADD PHRED-scaled score > 30 were at the top 0.01% across all potential ∼9 billion SNVs and were therefore considered as having a high likelihood to impact protein structure/function (Rentzsch et al., 2019). To control population structure, we performed a PCA analysis using PLINK 1.9 (Purcell et al., 2007), and we used the rst four components as covariables for all the burden analysis. We used GRAVITY (http://gravity.pasteur.fr), an open-source Cytoscape app that allowed an e cient visualization and analysis of all the exonic variants stored in a database by mapping them on protein-protein interaction (PPI). Variants of interest were manually curated by visualization of aligned sequencing data (BAM les) using IGV (Thorvaldsdóttir et al., 2013). We used admixture to ascertain the ancestry of the participants (Alexander & Lange, 2011). In our sample, 86% of the participants were from European descent. We nally ascertained the genetic variants related to the GABAergic and glutamatergic pathways by using the KEGG database (Kanehisa et al., 2016;Kanehisa & Goto, 2000). We considered the GABAergic synapse pathway (entry I04727) and the Glutamatergic synapse pathway (entIhsa04724). At the end, to explore the link between the dSSP score and the burden of genetic mutations in the GABAergic and/or glutamatergic pathways, we built a linear model with the dSSP score and the carrier of variants within the GABAergic and Glutamatergic pathways. We also used the bootstrap method and reported median p-values and Cohen-d obtained across 2000 re-sampling.

dSSP scores in the PARIS Study and LEAP Study Samples
In accordance with our hypothesis, means of the dSSP scores differed signi cantly between individuals with ASD and controls in the PARIS study samples with a trend for a more hypersensitive sensory pattern in the ASD group than in controls. Due to the moderate number of participants included in the analysis, the intergroup comparisons (patients, rst degree relatives, controls) did not reach signi cance (Table 2).

Items-based Clustering
The VARCLUS procedure based on the item scores of the SSP scale reported in the ASD participants of the PARIS study sample, converged into 9 clusters with 60.5% of variance explained (Supplementary  Table 3; silhouette score=0.11, bootstrap empirical p-value<0.0001). The clusters we obtained were not similarly distributed in items considering the hypo-, hyper-or both sensitivity [Chi 2 (dof=16, N=165)=31.2, p=0.01, Cramer's V=0.64] (Supplementary Table 4). We re-ran the analysis on the LEAP study sample. We found 7 clusters which explained 63% of variance (Supplementary Table 5; silhouette score=-0.08; bootstrap empirical p-value <0.0001). The clusters showed also a statistically signi cant difference in frequency of the items considering the hypo-, hyper-or both sensitivity (Chi 2 (dof=12, N=384)=35.9, p=0.0003, Cramer's V=0.69) (Supplementary Table 6).
We then compared the clustering of the SSP items obtained among the ASD participants of the LEAP study sample to the one obtained on the Paris sample. Clusters derived from both PARIS and LEAP study samples were similar (Fowlkes-Mallows similarity score=0.56, p < 0.0001). Within each cluster, one item was more representative than the others (Supplementary Tables 3 & 5). Interestingly, two SSP items were described as the most representative of their own clusters in both the PARIS and LEAP study samples: item 17 which was related to hypo-sensitivity and item 36 related to hyper-sensitivity. dSSP score correlated with GABA and/or Glutamatergic pathway mutation enrichments in ASD We investigated if differences in dSSP scores were associated with a distinct burden of deleterious variations affecting genes related to the GABA and/or Glutamatergic pathways. We only considered for each subject the Likely Gene Disrupting (LGD) and predicted deleterious missense mutations (CADD>30) ( Table 4; Supplementary Figure). We rst built a linear model for the dSSP score including the burden of gene mutations in GABAergic and/or the glutamatergic pathways, and their interactions. In subjects with ASD (n=135), our analysis reported a trend for an association with a burden of gene mutations in the GABAergic pathway but not in the glutamatergic pathway, nor in the both pathways (R 2 =0.05, F=2.18, p=0.09; GABA: F=3.36, p=0.06, η 2 = 0.02; Glutamate: F=0.08, p=0.77, η 2 = 0.0008; both pathways: F=2.68, p=0.11, η 2 = 0.02) (Supplementary Figure). We then performed a bootstrap analysis -across 2000 resampling-to explore further the relationship between the dSSP score and the burden of deleterious mutations in the GABA pathway. We observed individuals with ASD and with a high dSSP score indeed reported a signi cant enrichment of deleterious mutations in the GABAergic gene pathway (p=0.004, d=1.15). We nally ran a similar analysis only in individuals from European ancestry (based on the results of the admixture analysis) and we obtained a similar trend.

Discussion
Through our study, we aimed to further characterize sensory processing atypicalities in ASD and improve our ability to explore their potential underlying neurobiological mechanisms (Sieman et al., 2020). We thus built a summarizing score -the dSSP score -which showed that, on average, individuals with ASD displayed a trend for a hyper-sensitivity pro le, reaching only signi cance in the LEAP sample (through its power to detect signi cant results). Previous ndings in the literature also reported that a hypersensitivity pro le in children with ASD. This hyper-sensitivity pro le was previously associated with early stages of their developmental trajectory (Green et al., 2012, Ben-Sasson et al., 2019) -as we reported both in the PARIS and the LEAP samples -and with the severity of expressive language impairment . In our study, we also reported a signi cant interaction between dSSP score and chronological age of the subjects. This association may be driven by the load of comorbidity reported in the individuals with ASD, since previous studies reported higher levels of sensory reactivity in those severity of associated comorbidities (Kreiser & White, 2015, Tillmann et al., 2020. This association was observed in subjects with comorbid anxiety or depressive symptoms Rossow et al., 2021) but also somatic complaints (Lefter et al., 2020).
In our study, we also observed a larger dSSP score variability in ASD than in controls which was in line with the heterogeneity reported in ASD in many research areas, such as brain imaging (Masi et al., 2017).
Our results stressed further the need for partial phenotypes beyond categorical diagnosis to help in patients' strati cation and delineate more homogeneous subgroups (Wolfers et al., 2019). The dSSP score we constructed may offer a relevant setting to explore the heterogeneity in ASD (Lombardo et al,  2019). It could also pave new ways to determine the biological mechanisms associated with sensitivity abnormalities in autism. Although our results need to be replicated in larger samples and combined with additional dimensions, the dSSP score may help to uncover new sub-groups with more coherent neurobiological mechanisms (Uljarević et al., 2016;Bruinning et al., 2020). Interestingly, the intermediate dSSP score distribution we observed in the rst-degree relatives between individuals with ASD or those with typical development, suggested further that the dSSP score may be determined by inherited biological substrates (Neufeld et al, 2021), which was in line with our initial hypothesis.
To validate the empirical construct of the dSSP score, we performed a data driven approach of the hypoand hyper-sensory symptoms related items. The cluster analysis revealed a very similar distribution of the items encompassed in the data-driven clusters compared to those included in the two dimensions we empirically built. This result con rmed further the model of two main dimensions summarizing the sensory abnormalities in ASD (hypo-, hyper-sensitivity dimensions) (Baranek et al., 2006) and supported the empirical construct we used to build of the dSSP score. The data-driven analysis revealed that two items emerged as being highly representative of both clusters we identi ed, whatever the PARIS or the LEAP cohort we considered: item 17 (Suppl. Table 1: Item17 -"Becomes overly excitable during movement activity") drove mainly the variability of the hypo-sensitivity related cluster, and item 36 (Suppl. Table 1 : Item 36 -"Is bothered by bright lights after others have adapted to the light") the variability of the hyper-sensitivity related cluster. Beyond the simple dichotomy of sensitivity anomalies in ASD, the dSSP score integrated the personal sensory processing impairment into a ratio score facilitating for example the exploration of the relationship between these symptoms and the E/I imbalance (Pierce et al., 2021).
We nally performed an exploratory analysis to explore the relationship between the dSSP score and the load of deleterious mutations affecting the genes of glutamatergic & GABAergic pathways. We reported that individuals with a high dSSP score i.e., those with a excess of hyper sensory processing atypicalities, displayed a signi cant trend for an enrichment of deleterious gene mutations in the GABAergic pathway.
We speci cally observed mutations affecting the genes as GABARAPL1, GABBR2,…(Supplementary Table   2). Our results were in accordance with numerous reports describing the association between genes affecting directly (such as GABRA4, GRIN1) or indirectly (such CACNA1C, SHANK1-3, CNTN3-6) the GABAergic pathway homeostasis and sensory processing atypicalities in ASD (Leblond et al, 2014;Mercati et al, 2017;Tavassoli et al, 2021, Hartig et al., 2021. The excess load of deleterious mutations affecting the genes of GABAergic pathway reported in our study may reduce the GABAergic tone (Ferguson and Gao, 2018) -as previously showed (Sapey-Triomphe et al., 2019) -and may explained the hypersensitivity observed in ASD.

Limitations
One major limitation of the results we obtained was the di culty of the dSSP score to discriminate subjects with a similar quantum of hypo-and hyper-sensory symptoms resulting in a ratio score of 0 to those with no such symptoms but also resulting in a similar ratio score of 0. Thus, the dSSP score appeared more a potential biomarker of hypo/hyper imbalance than a summarized score of the sensory processing heterogeneity in subjects. We hypothesized that this hypo/hyper imbalance may re ected substantially the E/I imbalance reported in ASD (Sapey-Triomphe et al, 2019, Umesawa et al, 2020. In our study, individuals with ASD displayed a trend for a hyper-sensitivity pro le. This is line with studies showing a brain hyper-excitability mediated for example by a hyper-glutamatergic activity and resulting in increased cortical activity measured with electroencephalography (Bozzi et al, 2018) or measured with the MRI spectrometry (Brown et al., 2013).
On the other hand, our ndings linking the dSSP score and the deleterious mutations in GABAergic genes has to be taken with cautious. Speci cally, the sample size of the molecular analysis was very limited but should be considered as a highlight of the opportunity to use the dSSP score as a tool to dissect the biology of ASD. The lack of power of this sub-analysis in our study requested to use of a MAF below 10% which was not a standard in such similar molecular studies but with larger sample sizes. Obviously, a replication of our ndings showing the association between the dSSP score and the load of mutations in GABAergic genes in ASD was more than requested.

Conclusion
In conclusion, the dSSP score we built in this study may facilitate the identi cation of the underlying neuro-biological conditions related to sensory processing atypicialities in ASD. One further step would be to better estimate the E/I imbalance in ASD by using recently designed electroencephalographic paradigms (Bruinning et al, 2020), and explore its correlation with the ration of hypo-/ hyper-sensitivity processing patterns summarized by the dSSP score. The datasets generated and/or analyzed during the current study are not publicly available due to an embargo period but are available from the corresponding author on reasonable request. Figure 1 Sensory Short Pro le scores within the PARIS sample. A-Distribution of mean total original SPP score per item for participants with ASD (red), their rst-degree relatives (green) and for control participants (blue). A higher SPP score meant stronger anomalies of sensory processing. B-Distribution of the mean differential SPP score on the same participants; here, negative & positive scores respectively represented hypo-and hyper-sensory pro les. Relatives appeared with an intermediate distribution between patients and controls in both A & B. C-Evolution of the differential SPP score with age for the same participants. Linear regression showed an effect of age only for patients: signi cant regression equation of the sensory pro le with age (-0.12 + 0.01 x age, p=0.02, uncorrected) with an R2=0.03 in the group with ASD was obtained. This signi cant regression was not found in either the related group (-0.05 + 0.01 x age, R2=0.01, p=0.1) or the typically developing group (-0.03 + 0.01 x age, R2=0.04, p=0.06).

Figure 2
Sensory Short Pro le scores within the LEAP Cohort. A-Distribution of mean total original SPP score per item for participants with ASD (red) and for participants controls (blue). A higher SPP score meant stronger anomalies of sensory processing. B-Distribution of the mean differential SPP score on the same participants; here, negative & positive scores respectively represented hypo-and hyper-sensory pro les. C-Evolution of the differential SPP score with age for the same participants. Linear regression showed an effect of age only for patients. As observed in the PARIS cohort, we replicated a signi cant regression equation of the sensory pro le with age (-0.28 + 0.02 x age, p=0.0005, uncorrected) with an R2=0.03 in the ASD group. This signi cant regression was not found in the typically developing group (-0.5,15 + 0.0008 x age, R2=0.007, p=0.28).

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.