Altered Responses to Social Chemosignals in Autism Spectrum Disorder

Articles https://doi.org/10.1038/s41593-017-0024-x

Altered responses to social chemosignals in autism spectrum disorder

Yaara Endevelt-Shapira *, Ofer Perl, Aharon Ravia , Daniel Amir, Ami Eisen, Vered Bezalel, Liron Rozenkrantz, Eva Mishor, Liron Pinchover, Timna Soroka, Danielle Honigstein and Noam Sobel *

Autism spectrum disorder (ASD) is characterized by impaired social communication, often attributed to misreading of emotional cues. Why individuals with ASD misread emotions remains unclear. Given that terrestrial mammals rely on their sense of smell to read conspecific emotions, we hypothesized that misreading of emotional cues in ASD partially reflects altered social chemosignaling. We found no difference between typically developed (TD) and cognitively able adults with ASD at explicit detection and perception of social chemosignals. Nevertheless, TD and ASD participants dissociated in their responses to subliminal presentation of these same compounds: the undetected ‘smell of fear’ (skydiver sweat) increased physiological arousal and reduced explicit and implicit measures of trust in TD but acted opposite in ASD participants. Moreover, two different undetected synthetic putative social chemosignals increased or decreased arousal in TD but acted opposite in ASD participants. These results implicate social chemosignaling as a sensory substrate of social impairment in ASD.

errestrial mammals typically rely on their sense of smell to were asked to identify the odd odor from triplicates containing two read emotions and communicate socially, and there is growing BOs from the same individual and one BO from a different indi- evidence for meaningful social chemosignaling in humans as vidual (Supplementary Fig. 1). We observed no significant differ- T1–6 well . Human social chemosignals can convey information such as ences between TD and ASD participants in performance of this task age7, disease8, aggression9, happiness10 and fear11–14, and they can act (interaction P = 0.72; Fig. 1b). Finally, we collected fear sweat from subliminally to influence brain activity14–16, hormonal state17,18, mate eight men skydiving16 and control sweat from eight men walking selection19, sexual arousal5, infant bonding20 and general psycholog- in a state of calm. Cortisol was significantly higher in the skydiv- ical and emotional state2,3. Living without these signals, a condition ing group, indicating successful fear induction (Methods). We then one might term ‘social anosmia’, may constitute an impairment, but asked 15 TD and 15 ASD male participants to rate these BOs. Fear one, we speculate, with limited implications. This is because visual sweat was rated as less pleasant, more intense and denoting higher and auditory information could make up for the lost chemosignal- fear than control sweat (all P < 0.01), and there were no differences ing21. In contrast, living with these signals distorted rather than between TD and ASD participants in these ratings (all interactions lost, a condition one might term ‘social dysosmia’, could be devas- P > 0.38; Fig. 1c). Having observed that ASD participants spontane- tating. ASD is characterized by impaired social communication22 ously sample chemosignals like TD participants, are equally capable attributed to misreading of emotional cues23, but why individuals of detecting and discriminating BOs and associate a similar explicit with ASD misread emotions remains unknown. We hypothesized perceptual shift with the smell of fear, we next asked how such that a portion of misreading emotions in ASD may be explained by social chemosignals influence autonomic arousal and behavior in social dysosmia. We investigated this hypothesis by comparing the TD and ASD participants. responses of TD participants and cognitively able participants with ASD to the subliminal presentation of social chemosignals. Altered autonomic responses to the undetected smell of fear in ASD. Whether the smell of fear is a chemical cue capitalized on Results by the receiver or a chemical signal capitalized on by the sender, Intact sensory and behavioral substrates for social chemosignal- it remains a bodily odor with meaningful social value, and its suc- ing in ASD. We first set out to establish whether the substrates of cessful detection confers advantage. Exposure to the smell of fear social chemosignaling remain intact in ASD. Humans constantly drives slight but significant alterations in perception of emotionally and mostly subliminally sample conspecific body odors (BO), and salient information24, slight but significant changes in performance one mechanism for this is unwittingly sniffing one’s own hand after of cognitive tasks13, and more robust alterations in autonomic handshaking with a stranger6. We tested whether this persists in 18 arousal and neural activity in the substrates of emotional process- male participants with ASD (Table 1) and observed a doubling of ing14,16. However, the reported effects of the smell of fear mostly post-handshake hand-sniffing instances and quadrupling of post- materializes in women receivers but not in men receivers11. This handshake hand-sniffing duration, i.e., a significant change from poses a potential limitation on our intention to study reactions to baseline (P = 0.012), but not significantly different from the behav- the smell of fear in cognitively able adults with ASD, which is sig- ior of previously observed TD participants6 (χ2 = 0.11, P = 0.74; nificantly more prevalent in men than in women25. Nevertheless, Fig. 1a). Under the assumption that hand-sniffing is a BO sampling we initially set out to test TD and ASD participants using previously mechanism, we next asked whether TD and ASD participants could applied models. equally explicitly detect and discriminate BOs. In a three-alternative We passively exposed 20 TD and 20 ASD male participants to the forced-choice (3AFC) task, 18 TD and 16 ASD male participants volatile bouquet of verified fear sweat (Fig. 2a) or control (collection

Department of Neurobiology, Weizmann Institute of Science, Rehovot, Israel. *e-mail: [email protected]; [email protected]

Table 1 | Thirty-five participants with ASD Participated in experiment ADOS-2 Code Gender Greet Fear Fear AND HEX Fear Discrimination SA RRB Total Severity AQ Age physiological behavioral rating ADOS range response response (years) ASD01 M + + + + + + + 14 4 18 9 30 36–40 ASD02 M + + + + + + + 10 1 11 6 30 32–36 ASD03 M + + + + + + + 8 4 12 7 28 24–28 ASD04 M + + + + + + + 15 4 19 9 19 18–22 aASD05 M + + + + 5 3 8 29 28–30 ASD06 M + + + + + 14 5 19 9 22 43–47 ASD07 M + + + 12 2 14 8 29 31–32 ASD08 M + + + + 5 2 7 3 25 27–29 ASD09 M + + + + 14 3 17 9 21 24–27 ASD10 M + + + + + + + 10 7 17 9 29 19–22 aASD11 M + ** + + 6 3 9 14 32–35 ASD12 M + + + + + + + 13 6 19 9 18 19–23 ASD13 M + + + + + + 10 4 14 8 18 28–32 ASD14 M + + + 11 2 13 7 26 27–31 ASD15 M + + + + + + 8 3 11 6 35 24–28 ASD16 M + + 19 6 25 10 21 20 ASD17 M + 7 1 8 4 17 19 ASD18 M + + 17 3 20 10 35 32 ASD19 M + + + + + + 9 2 11 6 21 28–31 ASD20 M + 20 5 25 10 21 ASD21 M + + 6 2 8 4 20 19 ASD22 M + 25 25 ASD23 M + + 16 2 18 9 17 34–35 ASD24 M + + + 7 1 8 4 36 23–24 ASD25 M + + + + 12 7 19 9 26 21–22 ASD26 M + + + 3 3 6 3 33 27 ASD27 M + + + + 17 5 22 10 26 19–20 ASD28 M + + + + 10 6 16 9 35 26–27 ASD29 M + + + + 5 4 9 5 21 22–23 ASD30 M + + + + 7 3 10 6 26 25–26 ASD31 M + + 11 18 ASD32 M + + 7 2 9 5 14 20–21 ASD33 M + + 14 3 17 9 28 27–31 ASD34 F + 10 6 22 10 28 25 ASD35 F + 10 3 13 7 26 37 All participants with ASD were assessed by experienced clinicians independent of the present study and met Diagnostic & Statistical Manual of Mental Disorders (DSM) diagnostic criteria for an ASD. This table details the gender, age, AQ, ADOS module 4 scores and autism severity27 of the participants, as well as their participation in the different experiments and rating tasks. Note that age appears as a range because these different experiments were conducted across about 4 years. SA, social affect; RRB, restricted and repetitive behaviors. aTwo subjects have ADOS-2 scores calculated using the previous algorithm (subscales: Social Affect and Communication). pads alone) while performing two successive cognitive–emotional Postexperimental debriefing revealed that only 3 TD and 3 ASD tasks previously tested with sweat odors: the Faces task, in which participants noted a change in odor. Thus, consistent with our participants estimated the fearfulness of 27 face images12, and intention of subliminal manipulation, most participants were not the emotional Stroop task, in which participants determined the consciously aware of the odor condition. As expected, we failed to color of rapidly presented words differing in emotional content9 observe a behavioral effect for the smell of fear in the Faces and (Methods and Supplementary Fig. 2). Our primary measure of Stroop tasks in this all-male cohort (all P > 0.14). Nevertheless, we interest, obtained concurrently with the tasks, was electrodermal observed pronounced autonomic responses to the undetected smell activity (EDA), a particularly strong indicator of autonomic arousal. of fear, some common and some dissociated. A response that was We also measured nasal airflow and heart rate. Fear and control common to TD and ASD participants (interaction P = 0.20) was a conditions were counterbalanced for order, and participants were marked reduction in nasal inhalation in the presence of the smell blind to condition. of fear (P = 0.000293; Fig. 2b and Supplementary Fig. 3). In turn, we

112 Nature Neuroscience | VOL 21 | JANUARY 2018 | 111–119 | www.nature.com/natureneuroscience © 2017 Nature America Inc., part of Springer Nature. All rights reserved. NATure NeuroScience Articles observed a remarkable dissociation whereby the smell of fear sig- ab70 1 TD ASD nificantly increased event-related EDA (ER-EDA) in TD but not in 60 0.9 ASD participants (interaction P 0.001, Cohen’s d 1.2; Fig. 2c). A = ′= 50 0.8 similar dissociation was evident in the Stroop task (Supplementary 0.7 40 Fig. 4). Although these experiments were designed with ER-EDA as 0.6 30 the major measure of interest, to avoid a selective approach we also 0.5 generated a classifier using all collected data. For each parameter 20 0.4 we included the fear, control and delta values. We found that a dis- 10 0.3 criminant-analysis classifier using a covariance matrix estimate fol- Proportion correct duration post greet (s) 0 lowed by a leave-one-out cross-validation correctly labeled 16 of 20 Change in face-touching 0.2 TD and 12 of 20 ASD participants, i.e., 4 false positives and 8 false –10 0.1 negatives (70.0% accuracy, bootstrap P 0.002). Finally, we asked –20 0 = TD ASD whether the extent of chemosignal impact was related to the extent Men BO Women BO of social impairment by correlating the results with AQ (autism c 1 26 27 quotient) and ADOS (Autism Diagnostic Observation Schedule) 0.9 *** *** *** *** * scores. In the Faces task, we observed a modest negative correla- 0.8 ASD control ASD fear tion with AQ scores, such that lower ER-EDA responses to fear were 0.7 TD control associated with higher AQ scores across all participants (r = −0.33, 0.6 TD fear P = 0.043; Supplementary Fig. 5). In the Stroop task, we observed a 0.5 27 similar correlation with ADOS severity (r = −0.62, P = 0.024) and 0.4 ADOS total score (r = −0.71, P = 0.0065; note that because 15 par- Explicit rating 0.3 ticipants had Stroop ER-EDA yet two lacked ADOS, this correlation 0.2 reflects 13 participants; Fig. 3a). To conclude this experiment, mod- 0.1 els that previously revealed behavioral effects in women but not in 0 men again failed to uncover behavioral effects in men. In turn, auto- Fearfulness Pleasantness Intensity nomic arousal in men was significantly impacted by the undetected smell of fear; this impact significantly dissociated TD participants Fig. 1 | Intact sensory and behavioral substrates for social chemosignaling from ASD participants and was correlated with ASD severity. in ASD. a, Increase in time of right hand at nose after handshake in n = 18 ASD participants compared to previously published6 18 Altered behavioral responses to the undetected smell of fear in TD participants (ASD participants’ hand-sniffing duration before ASD. The autonomic measures implied that the smell of fear was handshake, mean = 2.9 ± 8.2 s; after handshake, mean = 13.1 ± 21.4 s, doing something different in TD and ASD participants, but we did corrected for baseline behavior in greets without handshaking: t17 = 2.8, not know how that was reflected in behavior. Tasks that include P = 0.012). b, Probability of detecting body odors where chance is 0.33 realistic type behaviors and action may provide for a more power- (n = 16 ASD participants, 18 TD participants). ANOVA with conditions of ful tool than cognitive tasks posed to a stationary participant. To group (TD or ASD) and odor (men or women) revealed no main effect this end, we modified a set of lifelike realistic manikins, endowing of group (F1,32 = 0.1, P = 0.76), a significant main effect of odor (F1,32 = 4.1, them with recorded natural voices, sensors detecting the proxim- P = 0.05) and no group × odor interaction (F1,32 = 0.14, P = 0.72). This ity of behaving participants, and the ability to precisely emit body reflected that both TD and ASD participants could equally successfully odors in a controlled fashion (Fig. 4a–d and Supplementary Fig. 6). discriminate between the smells of different men (TD = 56 ± 32%, On each of 64 trials in a spatial location target detection task, 20 ASD = 56 ± 26%; both groups from chance (33%): both t > 2.9, both TD and 20 ASD male participants were instructed to approach one P < 0.01), yet both TD and ASD participants equally failed to discriminate of two identical looking manikins (‘Chris’ and ‘Steve’) to receive between the smells of different women (TD = 43 ± 32%, ASD = 38 ± 32%; a spoken hint regarding the location of the upcoming target (for both groups from chance: both t < 1.3, both P > 0.23). c, Explicit perceptual example, “I am confident that the target is going to appear on the ratings (n = 15 ASD participants, 15 TD participants) applied to body odors right side of your screen”). Upon approach, the manikin also emit- obtained from individuals in a state of calm (control) or fear. Two-way ted, from its nostrils, an undetected body odor. One manikin emit- repeated-measures ANOVA with conditions of condition (fear or control ted the undetected smell of fear, and the other emitted a control sweat), parameter (pleasantness, intensity or fear) and a categorical odor (sports sweat) obtained from the very same donors (fear and independent factor of group (ASD or TD) revealed a significant interaction control manikins counterbalanced across participants). Participants −8 of parameter × condition (F2,56 = 38.5, P < 10 ), indicating that both were told in advance that one manikin gives better hints, but they groups perceived fear sweat as less pleasant (control = 0.58 ± 0.11 visual were not told which manikin it was. In reality, both manikins gave −5 analog scale (VAS), fear = 0.4 ± 0.11 VAS, t29 = 7.7, P < 10 ), more intense correct hints 70% of the time. After receiving the hint, the partici- (control = 0.40 ± 0.15 VAS, fear = 0.49 ± 0.14 VAS, t29 = 2.85, P = 0.008) pant walked back to a standing-height monitor and used a mouse and with higher fear (control = 0.33 ± 0.12 VAS, fear = 0.50 ± 0.13 VAS, to direct a cursor and click on the target once it appeared, on the −5 t29 = 5.7, P < 10 ) than control sweat. No interactions were observed (all left or right of the screen. We hypothesized that increased trust in F < 0.8, all P > 0.38). All tests were two-tailed, all centers reflect mean and one of the manikins would materialize in reduced time-to-target for all error bars reflect s.e.m. *P < 0.05; ***P < 0.001. correct hints given by that manikin. This is because if one trusts the hint, one can move the cursor in the correct direction before the target appears. Notably, this is a genuinely double-blind experiment ASD participants had increased trust in the fear-smell manikin. given that a manikin doesn’t know what odor it is emitting. This materialized in several measures. First, time-to-target was Postexperimental debriefing revealed that none of the ASD and faster for TD participants following hints from the control mani- only three of the TD participants noticed that the manikins had kin, yet ASD participants were faster following hints from the fear smells. Thus, consistent with our intention of subliminal manipula- manikin (interaction P = 0.003; nonparametric reanalysis: U = 88, tion, nearly all participants were not consciously aware of the odor Z = −2.87, P = 0.0035, Cohen’s d = 1.0; Fig. 4e,f). Moreover, analy- condition. Nevertheless, we observed a dissociation whereby TD sis of the mouse trajectory recorded 1 s after participants initiated participants had increased trust in the control-smell manikin, yet the trial revealed that TD participants preferentially moved the

a b c ** ** *** 0.9 * Control 2 Control 0.7 Fear Fear 0.8 0.6 1.6 0.7 0.6 0.5 1.2 0.5 0.4 0.4 0.3 0.8 0.3 response (NLPM) Cortisol ( μ g/dL) ough-to-peak normalized μ S)

0.2 r Sni 0.2 (t 0.4 0.1 0.1 EDA

0 0.0 0 Before After TD ASD TD ASD dive dive

d TD e ASD 1 1 1 Fear 1 Fear Control Control 0.8 P value 0.8 P value 0.8 0.8 0.6 0.6 P value 0.6 P value 0.6 0.4 0.4

0.2 0.2 0.4 0.4 ED A (normalized μ S) ED A (normalized μ S) 0 0 02468101214 0.2 0.2 –0.2 –0.2 02 468101214 –0.4 0 –0.4 0 Time (s) Time (s)

Fig. 2 | Altered autonomic responses to the undetected smell of fear in ASD. a, Levels of salivary cortisol in sweat donors before (0.29 ± 0.14 μg/dL) and after (0.51 ± 0.18 μg/dL) skydiving (n = 8 TD participants, t7 = 3.4, P = 0.011, W = 34, P = 0.02). b, Nasal airflow during the Faces task (n = 20 ASD participants, 20 TD participants). ANOVA revealed a main effect of condition (F1,38 = 15.9, P = 0.00029) but no effect of group (F1,38 = 0.7, P = 0.41) or interaction (F1,38 = 1.7, P = 0.20). This reflected that both TD and ASD subjects significantly reduced the vigor of their nasal inspirations when exposed to fear sweat (control = 0.37 ± 0.16 normalized liters per minute (NLPM), fear = 0.29 ± 0.15 NLPM). c, Extent of ER-EDA in the Faces task (n = 19 TD participants, 17 ASD participants, quantified results of the data in d and e). ANOVA revealed no group main effect (F1,34 = 0.16, P = 0.69) but a main effect of condition (F1,34 = 5.5, P = 0.025) and an interaction of group × condition (F1,34 = 12.8, P = 0.001, Cohen’s d′ = 1.2). Exposure to the smell of fear relative to control significantly increased ER-EDA in TD participants (normalized trough-to-peak, control = 0.75 ± 0.40 normalized μS, fear = 1.27 ± 0.28 normalized μS, t18 = 4.6, P = 0.0002) yet had no effect on ER-EDA in ASD participants (normalized trough-to-peak, control = 1.0 ± 0.36 normalized μS, fear = 0.92 ± 0.37 normalized μS, t16 = 0.80, P = 0.43). d, Average TD ER-EDA (27 events per participant) during fear (red) and control (blue) conditions. Gray line reflects point-by-point t test with values on the right y axis; dashed line reflects significance (n = 19 TD participants). e, Average ASD ER-EDA (n = 17 ASD participants). All tests were two-tailed, all centers reflect mean and all error bars reflect s.e.m. *P < 0.05; **P < 0.01; ***P < 0.001. cursor as advised by the control manikin yet ASD participants did cannot rule out a contribution of perceptual differences between not (interaction P = 0.0428; Fig. 4g–k). Finally, a simple nonpara- fear sweat and control to the dissociation between TD and ASD. metric analysis of explicit trustworthiness ratings applied to the With this in mind, we set out to measure responses to a puta- manikins at the end of the experiment revealed that whereas 16 of 20 tive chemosignal perceptually obscured in an olfactory carrier TD participants found the control manikin more trustworthy (bino- and to compare this to responses to the olfactory carrier alone. mial, P = 0.0046), this was true for only 9 of 20 ASD participants Androstadienone (4,16-androstadien-3-one; AND) is a sweat- (binomial, P = 0.7), reflecting a significant dissociation between bound molecule implicated in impacting arousal28 (note that we do TD and ASD participants (χ2 = 5.2, P = 0.023). We did not observe not equate AND with the smell of fear; we merely use it as a putative any relation between behavior in this task and AQ or ADOS scores. chemosignal that impacts arousal). In a separate 3AFC experiment, Nevertheless, these dissociated behaviors and perceptions pro- we first established a concentration of AND diluted in the odor- vide a behavioral expression of what was evident in the autonomic ant eugenol that was perceptually indiscriminable from eugenol measures of the previous experiment: individuals with ASD perceive alone by both TD and ASD participants (Fig. 5a and Supplementary the smell of fear, but they interpret it differently and act on it differ- Fig. 7). After obscuring AND in the olfactory mask, we measured ently, sometimes displaying behavior opposite that of TD individuals. its impact using a widely applied assay for human chemosignaling29, in which 23 TD and 17 ASD participants returned to the lab on Altered autonomic responses to undetected putative synthetic consecutive days at the same time of day and watched a series of social chemosignals in ASD. In the above experiments we used nat- emotional videos concurrent with psychophysiological monitor- ural sweat, which contains a large number of volatile components. ing, once after exposure to the odor carrier alone and once after Although presentation was nonexplicit and mostly subliminal, we exposure to the odor carrier embedded with undetected AND.

Experiments were counterbalanced for order and double-blind to a 1.0 condition (Supplementary Fig. 8). The physiological measure most commonly influenced by AND 0.5 is ongoing skin conductance30, i.e., nonspecific (in contrast to event related) alterations in electrodermal activity that manifest in 0 changes in the standard deviation (s.d.) of the ongoing recording 0102030 (NS-EDA). We again observed a dissociation whereby AND tended –0.5 to increase arousal in TD participants yet significantly reduced –1.0 arousal in ASD participants (interaction P = 0.007, nonparametric (normalized μ S) reanalysis: U = 73, Z = 3.2, P = 0.001, Cohen’s d′ = 0.98; Fig. 5b,c). –1.5

Similar and significant effects were evident in a composite measure ∆ ER-EDA (fear vs. control) of arousal (Supplementary Fig. 9) and in levels of testosterone mea- –2.0 sured in saliva (Supplementary Fig. 10). Although the AND experiment was designed with NS-EDA as the major measure of interest, –2.5 Total ADOS to avoid a selective approach we also generated a classifier using all collected data. For each parameter we included the AND, control b 3 and delta values. We found that a discriminant-analysis classifier using a covariance matrix estimate followed by a leave-one-out 2 cross-validation correctly labeled 18 of 23 TD and 13 of 17 ASD participants, i.e., 5 false positives and 4 false negatives (77.5% accuracy, 1 1,000-repetition bootstrap P = 0.002). We did not observe any relation between responses and AQ or ADOS scores. Taken together, these measures imply that, like natural fear sweat, undetected levels 0 5152535 of AND had opposite effects in TD and ASD participants. The above experiments uncovered chemosignal-induced –1 e ects of chemosiganls increases in arousal in TD participants that were not evident in ASD Combined physiological participants. To probe the persistence of this dissociation we sought –2 a chemosignaling effect that entails reduced rather than increased arousal in TD participants. The long-chain aliphatic aldehyde hexadecanal (HEX) has been implicated as a chemosignal mediat- –3 AQ ing social buffering in rodents31. HEX activates the OR37 olfactory receptor, which is highly conserved across mammals, and therefore Fig. 3 | Relation between altered social chemosignaling and ASD severity. the function of HEX may be conserved as well32. HEX can be found a, The relation between changes in ER-EDA during the Stroop task and in human feces, breath and skin33. Therefore, we set out to test the ADOS total score (n = 13 ASD participants, r = −0.71, P = 0.0065). b, The hypothesis that HEX acts as a social-buffering arousal-reducing relation between chemosignal-induced changes in autonomic arousal (a social chemosignal in humans and that its influence will differ in combination of the key measure in each experiment; see Methods) and TD and ASD participants. AQ scores for all participants (n = 30 ASD participants, 51 TD participants, In a separate 3AFC experiment, we first established a concentra- r = −0.38, P = 0.0005). Each dot represents a single participant; both tests tion of HEX in the odorant eugenol that was perceptually indiscrim- survive Bonferroni corrections. inable from eugenol alone by both TD and ASD participants (Fig. 5d and Supplementary Fig. 11). Next, we measured the impact of unde- designed with startle response as the major measure of interest, to tected HEX on the acoustic startle response. The startle response is avoid a selective approach we also generated a classifier using all col- a brainstem aversive reflex that occurs after a startling event such lected data. For each parameter we included the HEX, control and as a loud noise34. If HEX promotes social buffering in humans, we delta values. We found that a discriminant-analysis classifier using would expect it to reduce startle. The startle response is a particu- a covariance matrix estimate followed by a leave-one-out cross-vali- larly appealing model here because it is automatic and nonverbal, dation correctly labeled 16 of 17 TD and 11 of 17 ASD participants, rendering it less susceptible to differences between TD and ASD i.e., 1 false positive and 6 false negatives (79.4% accuracy, bootstrap participants that are unrelated to the process of interest. Moreover, P = 0.005). We did not observe a correlation between the extent the startle response is generally intact in individuals with ASD35 and of altered response and the severity of ASD as indicated by either modulated by social chemosignals in TD individuals36. Therefore, ADOS or AQ scores. To summarize, as in the case of fear sweat and we measured the startle response to a ~90-dB sound (repeated 20 AND, a novel putative chemosignal differently impacted arousal in times) after exposure to HEX or control (eugenol alone) in 17 TD (2 TD and ASD participants despite going undetected by both groups. women) and 17 ASD (2 women) participants. Each participant was tested on separate days at the same time, once with HEX and once Discussion with control, counterbalanced for order, with participants blind to Experiments with subliminal presentation of a natural stimulus condition (Supplementary Fig. 12). In addition to measuring the (smell of fear) and two different synthetic putative social chemo- startle response, we measured nasal airflow, EDA, heart rate, sali- signals (AND and HEX) converged to imply altered autonomic and vary hormones and self-reported mood before and after the study. behavioral responses to social chemosignals in participants with We observed that, consistent with previous studies35, the overall ASD. The models we used were mostly nonverbal, and our mea- startle response in the control condition was intact in participants sures were mostly not performance-based. Such measurement of with ASD. Moreover, we again observed a dissociation whereby automatic responses guards against unrelated sources of variance the addition of undetected HEX significantly reduced startle in in participants with ASD. One may note that merely finding a dif- TD participants but not in those with ASD (interaction P = 0.009, ference between TD and ASD populations on a given task is to be Z = 2.3, P = 0.021, Cohen’s d′ = 0.98; Fig. 5e–h). We observed mostly expected and does not necessarily help in understanding ASD. We similar results in all the measures we obtained in this experiment argue, however, that this particular case is different on two impor- (Supplementary Figs. 13–15). Although the HEX experiment was tant counts. First, in some of the tasks there was in fact no baseline

a b

f e ** g h * 1.2 0.1 30 TD ASD 240 TD ASD 20 0.05 190 10 140 0.7 0 0

∆ RT (s) 90 Fear RT (s) –10 –0.05

40 ∆ Trajectory (pixels) –20 Fear trajectory (pixels )

0.2 –0.1 –10 –30 0.2 0.7 1.2 TD ASD –104090140 190 240 TD ASD Control RT (s) Control trajectory (pixels) i

j k 50 50 TD ASD

0 0 trajectories trajectories ∆ Accumulate d ∆ Accumulate d –50 –50 300 100 0100 300 300100 0100 300 ∆Pixels (fear – control) ∆Pixels (fear – control)

Fig. 4 | Altered behavioral responses to the undetected smell of fear in ASD participants. a, Front and back views of the manikins. Back view is undressed to expose the in-torso olfactometer hatch. b, Close-up view of the manikin nostrils from which the undetected odors were emitted. c, A schematic representation of the in-torso olfactometer component. d, Proximity sensor obscured at the manikin’s navel. e, Response time to target following correct hints (n = 19 ASD participants, 20 TD participants). Each circle (gray circles, TD participants that were aware of the smell) represents the response time of a single participant following fear (y axis) and control (x axis). The diagonal line reflects the unit slope line (x = y) such that if points accumulate above the line then values are greater for fear and if they accumulate under the line then values are greater for control. f, Change in response time (quantified results of the data in e). ANOVA with conditions of hint (correct vs. incorrect), condition (fear vs. control sweat) and independent factor of group −8 (ASD vs. TD) revealed a significant main effect of hint (F1,37 = 79.2, P < 10 ) reflecting higher reduced time-to-target (RT) following the incorrect hints (correct = 0.50 ± 0.17 s, incorrect = 0.85 ± 0.23 s) and a significant interaction of hint × condition × group (F1,37 = 4.6, P = 0.038). ANOVA applied on each of the hint types separately uncovered a significant interaction of condition × group (F1,37 = 10.1, P = 0.003, nonparametric reanalysis: U = 88, Z = −2.87, P = 0.0035, Cohen’s d = 1.0) only in the correct hints, reflecting that within the TD group, RT was higher in fear compared to control (fear = 0.47 ± 0.15 s, control = 0.44 ± 0.14 s, t19 = 2.6, P = 0.0176), yet in the ASD group, RT was higher in control compared to fear (fear = 0.53 ± 0.19 s, control = 0.57 ± 0.19 s, t18 = 2.1, P = 0.050). g, Mouse trajectory at 1 s following task onset (n = 19 ASD participants, 20 TD participants). Each circle (gray, TD participants that were aware of the smell) represents the cursor location of a single participant following fear (y axis) and control (x axis). The diagonal line reflects the unit slope line (x = y) such that if points accumulate above the line then values are greater for fear, and if they accumulate under the line then values are greater for control. ANOVA revealed no main effects (F1,37 < 1.7, P = 0.20) yet a significant interaction of group × condition (F1,37 = 4.4, P = 0.0428), reflecting that participants in the TD group moved more toward the side suggested by the control-smell manikin relative to the fear-smell manikin (control = 114.9 ± 59.2 pixels, fear = 96.8 ± 53.4 pixels, t19 = 2.1, P = 0.049), yet no such effect was evident in the ASD group (control = 76.5 ± 71.3 pixels, fear = 85.5 ± 70.3 pixels, t18 = 0.92, P = 0.37). h, Change in cursor location (quantified results of the data in g). i, A schematic of the monitor during the task, depicting a central starting point for the mouse cursor (black cross in red box) and two potential target locations on the left and right (black-bordered clear boxes), with the target here appearing on the left. j,k, Heat maps reflecting mouse trajectories for TD participants (j) and ASD participants (k) during performance of the task. Data were collapsed, such that the left and right now reflect moving consistent with (trusting) the hint (left) or inconsistent with (distrusting) the hint (right). Color code reflects location following advice from the fear manikin minus advice from the control manikin. The full derivation of this panel is detailed in Methods and in Supplementary Fig. 19. All tests were two-tailed, all centers reflect mean and all error bars reflect s.e.m. *P < 0.05; **P < 0.01.

116 Nature Neuroscience | VOL 21 | JANUARY 2018 | 111–119 | www.nature.com/natureneuroscience © 2017 Nature America Inc., part of Springer Nature. All rights reserved. NATure NeuroScience Articles difference, yet differences emerged after the addition of putative We acknowledge several limitations of the current study: First, social chemosignals that were not consciously detected by either consistent with the gender asymmetry in ASD25, an overwhelming group. Second, unlike many sensory differences between TD and majority of our ASD cohort was men. While this was not an active ASD participants, the mechanism we are probing is inherently a selection bias on our side, it remains a notable limitation, specifi- part of social interaction and in this relates to the core behavioral cally because of the significant gender differences in ASD47 com- phenotype of ASD. bined with potential gender-specific aspects of responses to social Our results suggest social dysosmia but not social anosmia in chemosignals48. Indeed, if we separately analyze the results from ASD. ASD participants spontaneously sampled chemosignals and the two women who volunteered for the HEX study, we observe were able to explicitly detect them and perceive fearfulness when significantly stronger results in the same direction (Supplementary explicitly requested to do this (Fig. 1c). However, when the stim- Fig. 16). Thus, we will go as far as stating that any conclusions from ulus was subliminal, although it registered with ASD participants this study can currently be applied only to men. Second, individuals (for example, reduced nasal inspiration for fear sweat), it had a with autism can be categorized into those that appear over-sensitive profoundly different impact than in TD participants. Because such to sensory stimuli and those that appear under-sensitive49. One can social chemosignals are typically subliminal, this distortion is not speculate that such divergent sensory profiles may present divergent subject to any correction through cognitive appraisal. Notably, responses to social chemosignals, yet this important potential source altered responses to subliminal signals in participants with ASD of variability was not explored here. Finally, although many of the is consistent with evidence in other modalities37. However, unlike effects we observed were quite robust, they were mostly not signifi- in other modalities, the influence of odors on behavior is, in fact, cantly correlated with ADOS or AQ scores. To generate an overall greater when the odors are subliminal than when they are con- sense of this relationship, we combined the key physiological mea- sciously perceived38,39. Thus, whereas altered responses to sublimi- sures across experiments (ER-EDA, startle electromyograph and nal auditory or visual information may render individuals with ASD NS-EDA) and plotted this against AQ scores, providing for 81 obser- unaware of socio-emotional nuance, they can remain cognizant of vations from a total of 30 participants with ASD and 51 TD partici- major auditory or visual emotional tone. In contrast, in chemosig- pants. We observed a modest but significant correlation (r = −0.38, naling, the subliminal information is itself the major tone. P = 0.0005; Fig. 3b). This modest correlation is consistent with our A previous study with children found that imbuing objects with a impression of the data: the nature of the response to chemosignals is child’s mother’s body odor had no impact on TD children but signif- associated with AQ, but this correlation is underwhelming. In other icantly altered imitation toward the objects in children with ASD40. words, this study uncovered pronounced group differences, but the Thus, the notion of dissociated responses to social chemosignals in relation of these measures to individual patterns along the autism TD persons is evident in children as well. Moreover, the children's spectrum as they are reflected in AQ and ADOS remain unclear. study implied that added social chemosignals might shift behav- Whereas the above limitations relate to all five experiments, each ior of those with ASD toward that of TD individuals. Our experi- individual experiment also had its unique limitations. Rather than ments contained only one case of chemosignal-impacted behavior, further detailing them here, however, we would like to end this and here too we observed a similar indication: response-time in the section on a related strength of this study, namely the convergence manikin task with control sweat was significantly longer in ASD across experiments: different models using different stimuli all con- versus TD participants (ASD = 0.57 ± 0.19 s, TD = 0.44 ± 0.14 s, sistently converged to imply a dissociated response to undetected two-tailed t37 = 2.6, P = 0.013), yet fear sweat abolished this differ- social chemosignals in participants with ASD. Notably, for the sake ence (ASD = 0.53 ± 0.19 s, TD = 0.47 ± 0.15 s, two-tailed t37 = 1.05, of clarity and brevity, we concentrated in the body of this manuscript P = 0.30). Although these studies are utterly different and cannot be on the key markers we hypothesized, such as ER-EDA and startle. directly compared, together they support the tantalizing possibility However, as detailed in the Supplementary Figs., additional mark- of using chemosignals toward behavioral modification in individu- ers such as changes in mood (Supplementary Fig. 17) and levels of als with ASD (notably, ordinary odors have been used for behav- salivary hormones (Supplementary Fig. 10) repeatedly converged ioral therapy in the past, but later studies suggested that the effects to also imply dissociation between TD and ASD participants in the were not olfaction-specific41). Our study, however, only hints in this response to undetected chemosignals (Supplementary Fig. 18). direction, and is nowhere near application. Finally, although this study focused on the altered response to Given the effects we observed, one may ask how much of the social chemosignals in participants with ASD, it included two addi- ASD phenotype can be attributed to altered chemosignaling. tional developments. One was the manikin behavioral model for Notably, poor reading of fear signals in ASD is evident in vision42 studying social chemosignaling and the other was the uncovering and audition43 as well. Thus, we do not think that chemosignaling of a potential novel human social chemosignal, namely HEX. As alone explains all of the variance in misreading emotions. Moreover, to the former, we are very enthusiastic about this model. Typical ASD is a spectrum, and in its more severe presentation includes models used by us and others in this field are utterly unnatural. This profound impairment. We do not think and are not claiming that is not to say that taking advice from a manikin is totally natural severe motor impairments or extreme cognitive limitations can behavior, but the reality of approaching a life-sized realistic looking be attributed to social dysosmia. In turn, we do not think that the figure and then acting out a behavior (this is a walking rather than mechanisms of social dysosmia are isolated from the mechanisms of sitting experiment) together gave rise to meaningful effects. This severe impairment: our underlying speculative hypothesis relates to uncovered a behavioral response to fear sweat in men, a behavioral the function of olfactory receptors. We hypothesize that these recep- marker that was previously unavailable. Moreover, this result goes tors play a significant role not only in olfaction and chemosignaling beyond the question of fear-sweat alone to imply that body odor but also in the guidance of neurodevelopment44. Thus, significantly can be used to modify human–machine interactions. This raises the altered function in olfactory receptors (as we speculate in severe enticing possibility of imbuing robots with body odors to impact ASD) would lead to significantly altered neurodevelopment, poten- interactions with them. tially impacting motor, cognitive and social chemosignaling abili- In the second off-topic finding of this study we uncovered poten- ties. In turn, marginally altered function in olfactory receptors (as tial chemosignaling properties for HEX. Undetected HEX drove a we speculate in mild ASD) could alter only some aspects of ordinary reduction in startle response in TD participants but not in partici- olfaction45,46 and social chemosignaling alone. The above, however, pants with ASD. This calming effect of undetected HEX is consis- remains a speculative hypothesis that is consistent with our results tent with its predicted function in mammalian social buffering31. but not proven or negated by this study. That said, this one test is of course nowhere near satisfying criteria

a b c g TD ** 35 1 1 1 0.15 TD Control x = y 0.9 0.9 30 HEX ASD P values 0.1 0.8 0.8 0.5 25 0.7 0.7 0.05 0.6 P value 20 0.6 0 0.5 0.5 0 –1 –0.5 0 0.5 1 15

EMG (mV) 0.4 0.4 –0.05 10 0.3 Δ NS-EDA ( μ S) AND Proportion correct 0.3 –0.5 0.2 0.2

Δ NS-EDA ( μ S) AND – control 5 –0.1 0.1 0.1 –1 0 0 0 –0.15 ΔNS-SCR (μS) control TD ASD 050 100 150 TD ASD Time (ms)

d e f h ASD 35 1 10 0.8 100 TD ** x = y Control 0.9 ASD 8 30 HEX 0.7 P values 0.8 80 6 0.6 4 25 0.7 0.5 2 0.6 P valu e 60 20 0.4 0 0.5 40 –2 15 0.3 EMG (mV) 0.4 HEX EMG (mV)

Proportion correct –4 0.2 10 0.3 20

Δ EMA (mV) HEX – control –6 0.1 0.2 –8 5 0.1 0 0 –10 TD ASD 020406080 100 TD ASD 0 0 EMG (mV) control 050 100 150 Time (ms)

Fig. 5 | Altered autonomic responses to undetected putative synthetic social chemosignals in ASD. a, Probability of detecting 0.5 mM AND in 10% eugenol, where chance is 0.33 (n = 16 ASD participants, 18 TD participants; detection by TD participants = 37 ± 21%, detection by ASD participants = 47 ± 27%; both groups from chance: both t < 2.1, both P > 0.05; groups from each other: t32 = 1.3, P = 0.203). b, The influence of AND on ongoing changes in skin conductance (NS-EDA). Each circle represents the change in NS-EDA of a single participant following AND (y axis) and control (x axis). The diagonal line reflects the unit slope line (x = y) such that if points accumulate above the line then values are greater for AND, and if they accumulate under the line then values are greater for control (n = 22 TD participants, 17 ASD participants). c, Change in NS-EDA (quantified results of the data in b). ANOVA on NS-EDA with factors of condition (AND vs. control), time (before vs. after compound administration) and group (ASD vs. TD) −6 revealed main effects of group (F1,37 = 4.9, P = 0.033, ASD = 0.23 ± 0.18 μs, TD = 0.13 ± 0.09 μs) and time (F1,37 = 34.4, P < 10 ; before = 0.11 ± 0.13 μs, after = 0.24 ± 0.19 μs), as well as a significant interaction of condition × time × group (F1,37 = 8.0, P = 0.007, nonparametric reanalysis: U = 73, Z = 3.2, P = 0.001, Cohen’s d′ = 0.98). This reflected that exposure to AND relative to control significantly decreased NS-EDA in ASD participants (NS-EDA, Δcontrol = 0.21 ± 0.18 μs, ΔAND = 0.14 ± 0.16 μs, t16 = 2.5, P = 0.024) but strongly trended toward the opposite influence in TD participants (NS-EDA, Δcontrol = 0.06 ± 0.12 μs, ΔAND = 0.15 ± 0.21 μs, t21 = 2.0, P = 0.054). d, Probability of detecting 0.083 M HEX in 10% eugenol, where chance is 0.33 (n = 16 ASD participants, 18 TD participants; detection by TD participants = 32 ± 22%, detection by ASD participants = 36 ± 27%; both groups from chance: both t < 0.44, both P > 0.66; groups from each other: t32 = 0.49, P = 0.63). e, The influence of HEX on startle response (EMG). Each circle represents the change in startle of a single participant following HEX (y axis) and control (x axis). The diagonal line reflects the unit slope line (x = y) such that if points accumulate above the line then values are greater for HEX and if they accumulate under the line then values are greater for control (n = 16 TD participants, 16 ASD participants). f, Change in startle EMG (quantified results of the data in b). One-way repeated-measures ANOVA on startle EMG amplitudes with a factor of condition (HEX vs. control) and a categorical independent factor of group (TD vs. ASD) revealed a main effect of group (F1,30 = 6.5, P = 0.016) and a significant condition × group interaction (F1,30 = 7.7, P = 0.009, Z = 2.3, P = 0.021, Cohen’s d′ = 0.98). g, Average event- related EMG for TD participants (~20 events per participant) during HEX (red) and control (blue) conditions. Gray line reflects point-by-point t test with values on the right y axis; dashed line reflects significance (n = 16 TD participants). Undetected HEX significantly reduced the startle response in TD participants (amplitude: control = 32.3 ± 11.6 mV, HEX = 26.0 ± 10.3 mV, t15 = 2.8, P = 0.014). h, Average event-related EMG in ASD participants (n = 16 ASD participants). Undetected HEX did not significantly impact startle in ASD participants (amplitude: control = 45.9 ± 26.1 mV, HEX = 48.7 ± 28.0 mV, t15 = 1.2, P = 0.25). All tests were two-tailed, all centers reflect mean and all error bars reflect s.e.m. **P < 0.01. for definitively labeling HEX a human social chemosignal. This as well as further investigation of other OR37 ligands. Such inves- will require extensive additional studies. Thus, we should clearly tigation will further resolve the role chemosignaling may play in state that we do not definitively claim that HEX is a human social human behavior, in both neurotypical and atypical development. chemosignal, nor does this study rely on the notion that it is. The We conclude in reiterating that a series of experiments con- key contribution of this one experiment was to provide an instance ducted here in a total of 81 TD and 35 ASD participants converge to of decreased rather than increased arousal in the TD response. paint a picture of dissociated autonomic and behavioral responses Finally on this front, the predominant reliance on AND in the study to undetected social chemosignals in ASD. This may give rise to a of human social chemosignaling has recently been criticized50. condition we term social dysosmia. We speculate that social dysos- Although we do not agree with this criticism, we hope that the mia may underlie part of the impaired reading of emotional cues in current identification of an additional biologically plausible social ASD. This speculation suggests novel paths of research, diagnosis chemosignal will drive investigation into its behavioral significance, and treatment.

Methods 29. Frumin, I. & Sobel, N. in Pheromone Signaling 373–394 (Humana Press, Methods, including statements of data availability and any asso- Totowa, NJ, USA, 2013). 30. Bensaf, M. et al. Sex-steroid derived compounds induce sex-specifc efects ciated accession codes and references, are available at https://doi. on autonomic nervous system function in humans. Behav. Neurosci. 117, org/10.1038/s41593-017-0024-x. 1125–1134 (2003). 31. Klein, B. et al. Activation of the mouse odorant receptor 37 subsystem Received: 17 August 2017; Accepted: 19 October 2017; coincides with a reduction of novel environment-induced activity within the Published online: 27 November 2017 paraventricular nucleus of the hypothalamus. Eur. J. Neurosci. 41, 793–801 (2015). References 32. Hoppe, R., Lambert, T. D., Samollow, P. B., Breer, H. & Strotmann, J. 1. Lübke, K. T. & Pause, B. M. Always follow your nose: the functional Evolution of the “OR37” subfamily of olfactory receptors: a cross-species signifcance of social chemosignals in human reproduction and survival. comparison. J. Mol. Evol. 62, 460–472 (2006). Horm. Behav. 68, 134–144 (2015). 33. de Lacy Costello, B. et al. A review of the volatiles from the healthy human 2. McClintock, M. K. Reproduction in Context (MIT Press, Cambridge, MA, body. J. Breath Res. 8, 014001 (2014). USA, 2000). 34. Lang, P. J., Bradley, M. M. & Cuthbert, B. N. Emotion, attention, and the 3. de Groot, J. H., Smeets, M. A., Kaldewaij, A., Duijndam, M. J. & Semin, G. R. startle refex. Psychol. Rev. 97, 377–395 (1990). Chemosignals communicate human emotions. Psychol. Sci. 23, 35. Bernier, R., Dawson, G., Panagiotides, H. & Webb, S. Individuals with autism 1417–1424 (2012). spectrum disorder show normal responses to a fear potential startle 4. Wysocki, C. J. & Preti, G. Facts, fallacies, fears, and frustrations with human paradigm. J. Autism Dev. Disord. 35, 575–583 (2005). pheromones. Anat. Rec. A Discov. Mol. Cell. Evol. Biol. 281, 1201–1211 (2004). 36. Prehn, A., Ohrt, A., Sojka, B., Ferstl, R. & Pause, B. M. Chemosensory 5. Gelstein, S. et al. Human tears contain a chemosignal. Science 331, anxiety signals augment the startle refex in humans. Neurosci. Lett. 394, 226–230 (2011). 127–130 (2006). 6. Frumin, I. et al. A social chemosignaling function for human handshaking. 37. Senju, A. Atypical development of spontaneous social cognition in autism eLife 4, e05154 (2015). spectrum disorders. Brain Dev. 35, 96–101 (2013). 7. Mitro, S., Gordon, A. R., Olsson, M. J. & Lundström, J. N. Te smell of age: 38. Li, W., Moallem, I., Paller, K. A. & Gottfried, J. A. Subliminal smells can perception and discrimination of body odors of diferent ages. PLoS One 7, guide social preferences. Psychol. Sci. 18, 1044–1049 (2007). e38110 (2012). 39. Cecchetto, C., Rumiati, R. I. & Parma, V. Relative contribution of odour 8. Olsson, M. J. et al. Te scent of disease: human body odor contains an early intensity and valence to moral decisions. Perception 46, 447–474 (2017). chemosensory cue of sickness. Psychol. Sci. 25, 817–823 (2014). 40. Parma, V., Bulgheroni, M., Tirindelli, R. & Castiello, U. Body odors promote 9. Mutic, S., Brünner, Y. F., Rodriguez-Raecke, R., Wiesmann, M. & Freiherr, J. automatic imitation in autism. Biol. Psychiatry 74, 220–226 (2013). Chemosensory danger detection in the human brain: body odor 41. Woo, C. C., Donnelly, J. H., Steinberg-Epstein, R. & Leon, M. Environmental communicating aggression modulates limbic system activation. enrichment as a therapy for autism: a clinical trial replication and extension. Neuropsychologia 99, 187–198 (2017). Behav. Neurosci. 129, 412–422 (2015). 10. de Groot, J. H. et al. A snif of happiness. Psychol. Sci. 26, 684–700 (2015). 42. Pelphrey, K. A. et al. Visual scanning of faces in autism. J. Autism Dev. Disord 11. de Groot, J. H. B. & Smeets, M. A. M. Human fear chemosignaling: evidence 32, 249–261 (2002). from a meta-analysis. Chem. Senses 42, 663–673 (2017). 43. Philip, R. C. et al. Defcits in facial, body movement and vocal emotional 12. Zhou, W. & Chen, D. Fear-related chemosignals modulate recognition of fear processing in autism spectrum disorders. Psychol. Med. 40, 1919–1929 (2010). in ambiguous facial expressions. Psychol. Sci. 20, 177–183 (2009). 44. Secundo, L., Snitz, K. & Sobel, N. Te perceptual logic of smell. Curr. Opin. 13. Chen, D., Katdare, A. & Lucas, N. Chemosignals of fear enhance cognitive Neurobiol. 25, 107–115 (2014). performance in humans. Chem. Senses 31, 415–423 (2006). 45. Ashwin, C. et al. Enhanced olfactory sensitivity in autism spectrum 14. Wudarczyk, O. A. et al. Chemosensory anxiety cues enhance the perception conditions. Mol. Autism 5, 53 (2014). of fearful faces - an fMRI study. Neuroimage 143, 214–222 (2016). 46. Rozenkrantz, L. et al. A mechanistic link between olfaction and autism 15. Sobel, N. et al. Blind smell: brain activation induced by an undetected spectrum disorder. Curr. Biol. 25, 1904–1910 (2015). air-borne chemical. Brain 122, 209–217 (1999). 47. Halladay, A. K. et al. Sex and gender diferences in autism spectrum disorder: 16. Mujica-Parodi, L. R. et al. Chemosensory cues to conspecifc emotional stress summarizing evidence gaps and identifying emerging areas of priority. Mol. activate amygdala in humans. PLoS One 4, e6415 (2009). Autism 6, 36 (2015). 17. Wyart, C. et al. Smelling a single component of male sweat alters levels of 48. Savic, I., Berglund, H., Gulyas, B. & Roland, P. Smelling of odorous sex cortisol in women. J. Neurosci. 27, 1261–1265 (2007). hormone-like compounds causes sex-diferentiated hypothalamic activations 18. Preti, G., Wysocki, C. J., Barnhart, K. T. & Sondheimer, S. J. & Leyden, J.J. in humans. Neuron 31, 661–668 (2001). Male axillary extracts contain pheromones that afect pulsatile secretion of 49. Hazen, E. P., Stornelli, J. L., O’Rourke, J. A., Koesterer, K. & McDougle, C. J. luteinizing hormone and mood in women recipients. Biol. Reprod. 68, Sensory symptoms in autism spectrum disorders. Harv. Rev. Psychiatry 22, 2107–2113 (2003). 112–124 (2014). 19. Jacob, S., McClintock, M. K., Zelano, B. & Ober, C. Paternally inherited HLA 50. Wyatt, T. D. Te search for human pheromones: the lost decades and the alleles are associated with women’s choice of male odor. Nat. Genet. 30, necessity of returning to frst principles. Proc. Biol. Sci. 282, 20142994 (2015). 175–179 (2002). 20. Lundström, J. N. et al. Maternal status regulates cortical responses to the Acknowledgements body odor of newborns. Front. Psychol. 4, 597 (2013). We thank H. Breer and J. Strotmann for suggesting that we investigate hexadecanal in 21. Lemogne, C. et al. Congenital anosmia and emotion recognition: A humans. We thank Ziv and all the instructors and management at Paradive for their case-control study. Neuropsychologia 72, 52–58 (2015). gracious hospitality and help. This work was supported by ISF grant #1379/15, ERC 22. Association, A. P. Diagnostic and Statistical Manual of Mental Disorders (5th Advanced grant #670798 SocioSmell, grant #712254 from the US Air Force Office of edn.) (American Psychiatric Association Publishing, Arlington, VA, Scientific Research Program on Trust and Influence and by the McEwen Fund. USA, 2013). 23. Baron-Cohen, S. Social and pragmatic defcits in autism: cognitive or Author contributions afective? J. Autism Dev. Disord. 18, 379–402 (1988). Developed the idea: Y.E.-S. and N.S. Ran experiments: Y.E.-S., D.A., A.E., V.B., L.R., 24. Pause, B. M., Ohrt, A., Prehn, A. & Ferstl, R. Positive emotional priming of E.M., L.P., T.S. and N.S. Developed devices: O.P., D.H., and N.S. Analyzed data: Y.E.-S., facial afect perception in females is diminished by chemosensory anxiety N.S., A.R. and O.P. Wrote the paper: Y.E.-S., N.S. and O.P. signals. Chem. Senses 29, 797–805 (2004). 25. Werling, D. M. & Geschwind, D. H. Sex diferences in autism spectrum Competing interests disorders. Curr. Opin. Neurol. 26, 146–153 (2013). The authors declare no competing financial interests. 26. Baron-Cohen, S., Wheelwright, S., Skinner, R., Martin, J. & Clubley, E. Te autism-spectrum quotient (AQ): evidence from Asperger syndrome/ high-functioning autism, males and females, scientists and mathematicians. Additional information J. Autism Dev. Disord 31, 5–17 (2001). Supplementary information is available for this paper at https://doi.org/10.1038/ 27. Hus, V. & Lord, C. Te autism diagnostic observation schedule, module 4: s41593-017-0024-x. revised algorithm and standardized severity scores. J. Autism Dev. Disord. 44, Reprints and permissions information is available at www.nature.com/reprints. 1996–2012 (2014). 28. Grosser, B. I., Monti-Bloch, L., Jennings-White, C. & Berliner, D. L. Correspondence and requests for materials should be addressed to Y.E.-S. or N.S. Behavioral and electrophysiological efects of androstadienone, a human Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in pheromone. Psychoneuroendocrinology 25, 289–299 (2000). published maps and institutional affiliations.

Methods The handshake, Faces, Stroop and HEX experiments were single-blind, i.e., Location. All experiments were conducted in stainless-steel coated experiment participants were blind to experimental conditions but experimenters were rooms subserved by HEPA and carbon fltration, designed specifcally to prevent not. Notably, experimenters were blind at data reduction of these experiments. cross-contamination across experiments and conditions in human olfaction Finally, this study reports five independent experiments conducted over 4 years. studies. To investigate the possibility of any unintentional reporting bias in this effort, we analyzed the distribution of P values across all experiments. We observed a 57 Participants. A total of 35 (2 female) cognitively able adult participants with significantly right-skewed distribution that points against selective reporting ASD (Table 1) and 81 (2 female) TD matched controls (Supplementary Table 1) (Supplementary Fig. 18). The details of the overall statistical approach are also participated in the reported experiments after providing written informed available in the Life Sciences Reporting Summary. consent to procedures approved by the Weizmann Institute IRB committee (TD participants) and the Asaf Harofe Medical Center (Zrifin, Israel) Helsinki Composite figure generation. Fig. 3b. We correlated autism-spectrum quotient Committee (ASD participants). All relevant ethical regulations were followed. (AQ) scores with the most pronounced physiological responses in each of All individuals in the ASD group were assessed by experienced clinicians the three physiological experiments: effect of HEX on startle, effect of AND independent of the present study and met DSM (mostly Fourth edition) diagnostic on NS-EDA and effect of fear on ER-EDA in the Faces task. The scores of criteria for an ASD. The TD and ASD cohorts did not differ in age (ASD participants that participated in more than one experiment were averaged, such that each participant had only one data point representation in the final participants = 26.3 ± 6.0 years, TD participants = 27.3 ± 3.5 years; t114 = 1.1, P = 0.28) but significantly differed in years of education (ASD participants = 13.2 ± 2.6 correlation. We found a significant relation whereby higher AQ (more autism- like traits) was associated with less chemosignal-induced physiological effects years, TD participants = 14.7 ± 1.6 years, t111 = 3.8, P < 0.001) and, as intended, significantly differed in autism quotient (AQ) score (ASD participants = 24.7 ± 6.5, (r = −0.38, P < 0.001; Fig. 3b). In other words, different effects of chemosignals on physiological arousal were evident not only between typical and atypical TD participants = 16.5 ± 4.7, t113 = 7.5, P < 0.0001). Notably, for each separate experiment we recruited TD participants to match the ASD cohort in age and development but also within both groups together as a function of social skills. gender such that they did not differ on these fronts. Fig. 4j,k. This figure contains a lot of data that we tried to compress into only two panels. Here we detail the path from raw data to those two panels Participant inclusion/exclusion criteria. For ASD. Inclusion criteria for ASD (Supplementary Fig. 19). Supplementary Software contains Matlab code for participants was diagnosis by an experienced clinician independent of the present generating the heat maps in Fig. 4. First, trials were split according to the hint given study who determined the individual met DSM diagnostic criteria for an ASD. by the manikins to include only ‘good’ or correct hint trials. Next, we extracted Participants were also screened for an intact sense of smell. mouse trajectory coordinates at 1,000 ms from trial onset of all trials according to ‘hint content’ (‘go left’ or ‘go right’), smell (fear-smell manikin or control-smell For TD. Inclusion for TD was AQ under 30 and intact sense of smell. This resulted manikin) and group (ASD or TD). To facilitate visualization, we collapsed the in the exclusion of two (2) TD participants with AQ ≥ 30 (no ASD participants hint content condition by superimposing and flipping the heat maps to result in were excluded). a new figure depicting ‘towards cue’ trajectories, regardless of left/right screen orientation. In this view, ‘towards hint’ is centered on the left side of the figure. Data exclusion. Exclusions were according to previously published criteria, as Lastly, heat map images were smoothed using a 2D Gaussian filter with a kernel of detailed below. 1 SDs. To compare the effect of fear sweat across groups, heat maps were subtracted (fear–control) and the product was smoothed with a 2D Gaussian filter with a For EDA. We applied exclusion criteria according to Green et al.51. This included kernel of 10 SDs. This resulted in two separate heat maps, depicting the mouse- exclusion of participants with no EDA reactivity at all, i.e., EDA under 0.02 μS, trajectories of ASD and TD individuals to the smell of fear. or with excessive motion. This resulted in the exclusion of 1 TD and 3 ASD participants from ER-EDA in the Faces task in Experiment 2 and 5 ASD Detailed methods per experiment. Handshake experiment. Methods were 6 participants from ER-EDA in the Stroop task in Experiment 2. In experiment 3, identical to Frumin et al. . Participants were frst led to a room where they were one TD participant had no EDA recording due to technical fault. requested to sit and wait. About 3 min later a cosmetics-free experimenter entered the room, introduced himself using a fxed greeting (20 ± 8 s duration) either For startle. We applied exclusion criteria according to Pause52, Prehn6 and with or without a handshake and concluded by telling the participant that they Blumenthal53. The startle response is typically observed in a time window of would soon return to start the experiment. Tese ~20 s are referred to as the ‘greet’. 30–90 ms following the acoustic stimulus36,52. Events in which the maximum Te participant was then again lef alone in the room for an additional 3 min. amplitude in this temporal window was higher than the maximum amplitude Te entire interaction was flmed with hidden cameras. Te flm data were then within the 100-ms prestimulus baseline by a factor of 2.5 were considered scored for potential olfactory hand-sampling behavior. Criterion for scoring was responses. Participants that had at least 5 responses in each condition were any application of a hand to the face, as long as touching was under the eyebrows considered responders52. According to this criterion, 1 TD participant was and above the chin. Lef (nonshaking) and right (shaking) hands were scored excluded from further analysis. One ASD participant was excluded because his separately. Next, for each participant we summed the time each hand spent at the mean EMG amplitude exceeded more than 3 s.d. from mean EMG amplitude of all vicinity of the nose (i.e., under the eyebrows and above the chin only) across 1 min participants53. before ( + greet event time) and 1 min afer ( + greet event time) the greeting event Finally, for reasons that are not totally clear to us, 9 of 17 ASD participants in (which culminated at ~80 ± 16 s, given the added time of the greet event itself). We Experiment 4 refused to spit into a salivette. This prevented meaningful analysis of tested 28 cognitively able ASD participants in this experiment, 18 with handshake hormones in saliva in this experiment. and 10 without handshake, to establish the no-handshake baseline. Afer each experiment, participants were ofered the right to destroy the photographic data, Statistical analysis. To estimate sample size, we used power analysis where we or in turn provide specifc consent for its use in research and/or publication. Te had previous estimates of relevant variance. Thus, for the study of autonomic handshake experiment could only be conducted across participants, not within responses to chemosignals we conducted power analyses using G*Power software54 participants (as you cannot meet a person for the frst time twice). Tis renders applied to Prehn et al.36, which suggested at least 12 participants per group in a need for a relatively large sample size, one that is hard to obtain with ASD 6 within-participants analyses. For experiments using methods without previous participants. In the original published handshaking study , efects were observed application, no statistical methods were used to predetermine sample sizes, but with groups of 18 participants in the handshake condition and an additional 18 our sample sizes are larger than those reported in previous publications comparing participants to establish the no-handshake baseline. Here we had the sought- behavior and autonomic responses between TD adults and cognitively able ASD for 18 participants for the important handshake condition but failed to recruit adults, in which cohorts of 10 participants served for reaction-time comparisons55 an additional 18 for the no-handshake baseline, instead basing this value on 10 and 11 participants for a pupillometry task56. Each analysis began with an participants. Tis reduced number of participants in the baseline condition reduces estimation of data distribution. Normally distributed data were analyzed using the power of our observation. More specifcally, given the trend in the data toward ANOVA followed by planned two-tailed t tests. When data were not normally increased hand-snifng in ASD, it reduces the power of the conclusion of ‘no distributed (P < 0.05 using Shapiro–Wilk normality test) or the variances were not diference’ between TD and ASD. Tat said, in terms of the question being asked homogeneous among groups (P < 0.05 using Levene’s homogeneity of variance here, namely, ‘do individuals with ASD sample the stimuli in question?’, then the test), we also conducted a Mann–Whitney test and report this throughout the answer remains yes, and perhaps even more so than in TD individuals. text. In all cases the nonparametric reanalysis was consistent with the parametric approach. Body odor discrimination task: body odor collection. Twelve donors (6 female, mean age 30.5 ± 6.1, range 23–46 years) were provided with new 100% cotton white Randomization. Trial-orders were randomized within tasks. Conditions were T-shirts. The donors were instructed to wear the shirt for two consecutive nights. counterbalanced in order across participants. The donors were further instructed to avoid eating ingredients that can alter body odor (fenugreek, asparagus, curry, etc.) for at least 2 d prior to body odor sampling. Blinding. The AND experiment and the manikin experiment were double-blind, In addition, during the sampling days donors were asked not to use soap, shampoo, i.e., both participants and experimenters were blind to experimental conditions. conditioner or deodorant. Between the two nights, the T-shirts were kept inside a

Nature Neuroscience | www.nature.com/natureneuroscience © 2017 Nature America Inc., part of Springer Nature. All rights reserved. NATure NeuroScience Articles closed glass jar that was stored in the donors’ home freezer. After the jar containing Cortisol Immunoassay kit (Salimetrics, CA, USA). After completion of the T-shirt was brought back to the lab, it was stored at −20 °C to prevent bacterial growth. immunoassay, the absorbance of the fluorescent cortisol conjugate–antibody complex in the wells were obtained at 450 nm and corrected at 490 nm with a Shirt sniffing device (SSD). On the morning of the experiment, shirts were thawed microplate reader. Standard dilutions of cortisol (0, 0.012, 0.037, 0.111, 0.333, 1.0 inside the jars to avoid humidity condensation. Using sterile scissors, shirts were and 3.0 μg/dL) were used along a nonspecific binding well in the first two columns then cut into two longitudinal pieces, such that each half of the shirt contained one of the kit for calibration. Defined high and low control concentrations were used axillary area. Each half shirt was then placed inside an SSD: a glass jar covered by as quality controls for each column of the plate. The absolute salivary cortisol a cap with an air filter, inhalation mask and a one-way flap valve (Supplementary concentration was estimated from the fluorescence of the hormone conjugate– Fig. 1). The shirts were replaced by new shirts from the same donors after two antibody complex by computing the inverse value on a four-parameter sigmoid fit sessions to avoid odor attenuation and hence participants’ ability to judge by odor obtained with the standard values. intensity rather than by discrimination. Two of the SSDs contained a shirt that originated from the same donor and the Manikin experiment. We devised a task leading participants to interact in close third contained a shirt from a different donor. Trials were not time-limited. The proximity with two identical manikins, one emitting the smell of fear and the trials were randomly ordered and were composed of three trios of male donor with other emitting control odor (pooled from four donors; control sweat obtained male distracter and three trios of female donor with female distracter. Each one from the very same scent-donors during mild activity; Supplementary Fig. 6). of the trios was of a different combination of donors and was repeated only once The odors were emitted by olfactometer at 1 L/min from the manikin’s nostrils, throughout a session. Each participant completed the five-trial task twice, once and each manikin was placed under a high-flow ceiling exhaust vent, assuring no with odors from women and once with odors from men. contamination across manikins. Fear and control were counterbalanced between manikins across participants. Each manikin had a unique nametag (Chris or Chemosignal detection. Each trial contained sequential presentation of three Steve) and a unique voice, all counterbalanced across conditions. Participants jars (counterbalanced for order), two containing the carrier alone (100 µL were told that they were participating in an experiment investigating the influence eugenol, CAS# 97-53-0, 10% in propylene glycol) and one containing the of tone of voice on behavioral interactions. To avoid added sources of variance, carrier + chemosignal: either hexadecanal (100 µL, CAS# 629-80-1, 200 mg in we wanted the different voices used by the manikins to be perceived as equally 10 mL propylene glycol + 100 µL eugenol) or androstadienone (100 µL, 0.5 mM trustworthy. To achieve this we conducted an online experiment inviting raters to androsta-4,16,-dien-3-one dissolved in 10% eugenol; note that it is important rate the trustworthiness of 6 different voices saying the same text later used by the to periodically replace the hexadecanal in use, as it oxidizes). Participants were manikins. Based on 198 ratings, we selected two voices (#3 and #4) that were rated allowed to take one 2-s-long sniff at each odor presentation and were then asked to equal in their trustworthiness (Supplementary Fig. 20). pick out the jar that contained the dissimilar odor. Each participant completed five At trial onset, a participant stood in front of a monitor, with his back to the repetitions with each chemosignal. manikins, which stood ~2 m away and ~1.5 m apart. The monitor contained a ‘ready’ button in the middle, flanked by two potential target locations (marked Fear sweat experiments. Sample collection. We applied absorbing pads to the as empty squares), one on the extreme right and one on the extreme left of the washed underarms of tandem skydivers from about 2 h before their frst dive, screen. Participants were instructed by an on-screen script to approach either Steve at which time we also collected saliva. We collected the pads and a second or Chris to obtain a hint. When the proximity sensor sensed that the participant saliva sample directly afer they landed. Samples were stored at −20 °C and was within 30 cm of the manikin, the manikin uttered the hint, for example, “I thawed to room temperature for use. For the 3AFC task we used sweat from 8 am confident that the target is going to appear on the right side of your screen.” skydivers (age = 29.1 ± 5.7 years) and 8 controls (age = 34.5 ± 4.69 years) who The manikin voice was emitted from the manikin mouth, and was intentionally signifcantly difered in levels of cortisol (fear cortisol = 0.32 ± 0.19 μg/dL, control set at low volume, requiring participants to keep their head close to the manikin cortisol = 0.1 ± 0.06 μg/dL, Mann–Whitney Z = 2.73, P = 0.006). For the manikin head. This setup assured maximal exposure to the chemosignals. After receiving experiment, we applied the pads on the lower back rather than armpit. Although the hint, participants returned to the monitor and used an above-waist-height this location does not have a high density of apocrine glands, it is also less mouse to press the ‘ready’ button. Two seconds later (± 0.5 s random jitter) a target susceptible to contamination by personal hygiene products. Te control pads were (large blue star) appeared on either the left or right of the screen. Participants applied in the same location during nonstressed outdoor activity on a hot and were instructed to be as fast and accurate as they could at using the mouse to click humid day. on the target. After completing 64 trials, participants were also asked to rate the trustworthiness of Chris and Steve. Procedures. Stroop and Faces experiments. Participants were ftted with two bipolar electrodermal fnger electrodes. During the experiment, participants were alone Experiments with AND and HEX. Procedures with AND. We used a commonly in the room. Tey frst watched a 2-min nature video for equilibrium. Next, in the applied within-participants design, in which participants returned on separate Faces task participants used an on-screen scale to rate the fearfulness of 27 faces, days, at the same time of day, to be tested once with eugenol alone (control) and each presented briefy (250 ms), presentation triggered by nasal inhalation, with once with AND masked in eugenol (0.5 mM androsta-4,16,-dien-3-one dissolved ~30 s between faces. Next, in the emotional Stroop task9, participants used four in 10% eugenol). Participants and experimenters were blind to condition. A keys of a keyboard to denote the color of emotional and neutral words presented baseline period for equilibrium was followed by odorant exposure of 10 timed in rapid succession. Finally, an emotionally neutral nature video was presented for snifs, minimum inter-snif interval = 40 s, and each snif was followed by 5 min without concurrent odor exposure before repeating the above experiment pleasantness, intensity and familiarity estimates. Tis was followed by placing an with the second odor (either fear or control, counterbalanced for order). Te adhesive pad on the participant’s upper lip containing 30 µL of the experimental experimental timeline of the fear sweat experiment is in Supplementary Fig. 2. compound for continued exposure throughout the experiment. Tis pad allowed Odors, either fear sweat or control (empty pads), were provided through a nasal volatiles into the nose but prevented transdermal difusion. Tis was followed mask subjects were misled to believe was used only for recording respiration. by three counterbalanced sections, each containing an emotionally neutral flm Te reasons we chose pads alone as control were as follows: previous studies followed by a validated5,58 emotion-evoking flm, one negative, one positive and have established that the efects of the smell of fear are more pronounced when one erotic. Sections were interdigitated with mood questionnaires and saliva compared to an empty control than to a sweat control12. Given that our question samples. Autonomic nervous system parameters (electrocardiogram (ECG), here is not whether there is a response to the smell of fear in these particular tasks electrodermal activity (EDA), nasal respiration (NR) and skin temperature (ST)) (this has been established9,11,12) but rather whether the response difers between TD were concurrently sampled at 1 kHz using hardware and sofware as in the previous and ASD participants, we opted for the potentially strongest efects. To rephrase: experiment. Te experimental timeline for the experiment with AND are in here we are asking a question about the potential statistical interaction, not about Supplementary Fig. 8. the potential main efect. Procedures with HEX (startle experiment). We used a within-participants design, in Measures. Data were collected at 400 Hz through an instrumentation amplifier which participants returned on separate days, at the same time of day, to be tested (PowerLab 16SP) using LabChart7 software (ADInstruments, New South Wales, once with eugenol alone (control) and once with HEX masked in eugenol (HEX, Australia). Event-related electrodermal activity (ER-EDA) analysis was conducted CAS# 629-80-1, diluted to 0.083 M). Participants were blind to condition. Baseline after bandpass-filtering the data (0.05–35 Hz) to remove drift and zeroing at event and odor-exposure procedures were identical to the experiment with AND. This onset. Data were normalized through dividing each participant’s timeline by their was followed by a 20-min neutral nature film with earphone-delivered startling maximal value. Twenty TD and 20 ASD participants participated in the task. sounds starting 5 min into the film and recurring 20 times at an interstimulus interval (ISI) that was varied semirandomly between 5 and 60 s. Each startling Cortisol in saliva. Saliva collection was by unstimulated passive drooling. Samples sound was a broadband white noise of 50 ms at ~90 dB presented binaurally were stored at −20 °C and were thawed and centrifuged before testing. The saliva through Sennheiser hd280 headphones. The experimental timelines for the from each tube was assayed in triplicate wells. Tubes from a given participant were experiment with HEX is in Supplementary Fig. 12. all assayed on the same plate, and tubes from different visits obtained at a given time were assayed on the same column of the 96-well plate to avoid systematic Autonomic nervous system parameters. Electrocardiogram (ECG), electrodermal errors between conditions. We used the Extended Range High Sensitivity Salivary activity (EDA), nasal respiration (NR) and skin temperature (ST) measurements

Nature Neuroscience | www.nature.com/natureneuroscience © 2017 Nature America Inc., part of Springer Nature. All rights reserved. Articles NATure NeuroScience were sampled at 1 kHz and recorded using a Power-Lab 16SP Monitoring System TD participants), the EMG startle amplitude data were entered into a repeated- (ADInstruments, Australia). Data were later displayed, stored, reduced and measures ANOVA with conditions of odor (hexadecanal/control) and group (ASD/ analyzed using LabChart 7 software (ADInstruments). TD) as independent factors. ECG was measured using three circular Ag/AgCl conductive adhesive Self-reported mood was measured with a commonly applied59 17-item electrodes (0.9 cm diameter). Electrodes were placed under the left and right questionnaire. Participants rated how strongly they were experiencing each of the ribcage, and a ground electrode was placed on the fore of the right foot. Heart rate 17 different emotions using a visual analog scale (VAS) ranging from ‘not at all’ to variability (HRV, SDNN) was measured during the different movies. ‘very strongly’ for each descriptor. The order of descriptors was random. For the EDA was measured through two finger electrodes placed on the second experiment with AND we analyzed the corresponding descriptors to the mood- phalanx of the index and the third digit of the nondominant hand. EDA was inducing films: sadness, happiness and sexual arousal58. For the experiment with measured by applying a 0.5-µA/cm2 AC current. The EDA amplifier (GSR Amp. HEX these descriptors were divided into four categories: positive mood (calm, FE116 ADInstruments) was fully isolated with low voltage, 75-Hz (~40-mV) AC content, confident, happy, interested and amused), high-arousal negative mood excitation. The s.d. of EDA (Std-EDA) was used to estimate ongoing long-term (contemptuous, angry, anxious, embarrassed, disgusted and stressed), low-arousal changes in nonspecific EDA (NS-EDA). The values were obtained following a negative mood (annoyed, bored and sad) and sexually aroused58. bandpass filter of 0.05–35 Hz. Nasal airflow was measured using a nasal cannula (1103, Teleflex Life Sciences Reporting Summary. Further information on experimental design is medical) placed at the nares and attached to a spirometer (Spirometer FE141 available in the Life Sciences Reporting Summary. ADInstruments). Skin surface temperature (ST) was measured using a small ceramic- Data availability. All the data of this manuscript are available in a Supplementary encapsulated metal oxide semiconductor placed directly below the axilla. The data Dataset, which also includes a sheet providing the raw data of each of the main were reduced to max–min temperature. experiments reported in this study.

Hormones in saliva. Saliva collection was by unstimulated passive drooling. Code availability. Available in the Supplementary Software online is a Matlab code Samples were stored at −20 °C and were thawed and centrifuged before testing. for generating the heat maps in Fig. 4 of the manuscript. The saliva from each tube was assayed in triplicate wells. Tubes from a given participant were all assayed on the same plate, and tubes from different visits References obtained at a given time were assayed on the same column of the 96-well plate to 51. Green, S. R., Kragel, P. A., Fecteau, M. E. & LaBar, K. S. Development and avoid systematic errors between conditions. Free testosterone from the supernatant validation of an unsupervised scoring system (Autonomate) for skin was tested using the Extended Range Salivary Testosterone Immunoassay kit. After conductance response analysis. Int. J. Psychophysiol. 91, 186–193 (2014). completion of the immunoassay, the absorbance of the fluorescent testosterone 52. Pause, B. M., Adolph, D., Prehn-Kristensen, A. & Ferstl, R. Startle response conjugate–antibody complex in the wells were obtained at 450 nm and corrected potentiation to chemosensory anxiety signals in socially anxious individuals. at 490 nm with a microplate reader. Standard dilutions of testosterone (0, 6.1, 15.4, Int. J. Psychophysiol. 74, 88–92 (2009). 38.4, 96, 240 and 600 pg/mL) were used along a nonspecific binding well in the first 53. Blumenthal, T. D. et al. Committee report: guidelines for human startle two columns of the kit for calibration. Defined high and low control concentrations eyeblink electromyographic studies. Psychophysiology 42, 1–15 (2005). were used as a quality control for each column of the plate. The absolute salivary 54. Faul, F., Erdfelder, E., Lang, A. G. & Buchner, A. G*Power 3: a fexible testosterone concentration was estimated from the fluorescence of the hormone statistical power analysis program for the social, behavioral, and biomedical conjugate–antibody complex by computing the inverse value on a four-parameter sciences. Behav. Res. Methods 39, 175–191 (2007). sigmoid fit. 55. Harris, H. et al. Perceptual learning in autism: over-specifcity and possible remedies. Nat. Neurosci. 18, 1574–1576 (2015). Startle response analysis. The startle response is typically observed in a time 56. Lawson, R. P., Mathys, C. & Rees, G. Adults with autism overestimate the window of 30–90 ms following the acoustic stimulus36,52. Events in which the volatility of the sensory environment. Nat. Neurosci. 20, 1293–1299 (2017). maximum amplitude in this temporal window were higher than the maximum 57. Simonsohn, U., Nelson, L. D. & Simmons, J. P. P-curve and efect size: amplitude within the 100-ms prestimulus baseline by a factor of 2.5 were Correcting for publication bias using only signifcant results. Perspect. Psychol. considered responses. Participants who had at least 5 responses in each condition Sci. 9, 666–681 (2014). were considered responders52. According to this criterion, 1 TD participant was 58. Bensaf, M., Brown, W. M., Khan, R., Levenson, B. & Sobel, N. Snifng excluded from further analysis. The percent of rejected trials (21.6%) among human sex-steroid derived compounds modulates mood, memory and the remaining participants was not different between groups or conditions (all autonomic nervous system function in specifc behavioral contexts. Behav. P > 0.37). We then measured the maximum amplitude in the 30- to 90-ms temporal Brain Res. 152, 11–22 (2004). window following the acoustic stimuli. One ASD participant was excluded 59. Levenson, R. W., Ekman, P. & Friesen, W. V. Voluntary facial action generates because his mean EMG amplitude exceeded more than 3 s.d. from mean EMG emotion-specifc autonomic nervous system activity. Psychophysiology 27, amplitude of all participants53. For the remaining participants (16 ASD and 16 363–384 (1990).

Corresponding author(s): Yaara Endevelt-Shapira and Noam Sobel Initial submission Revised version Final submission Life Sciences Reporting Summary Nature Research wishes to improve the reproducibility of the work that we publish. This form is intended for publication with all accepted life science papers and provides structure for consistency and transparency in reporting. Every life science submission will use this form; some list items might not apply to an individual manuscript, but all fields must be completed for clarity. For further information on the points included in this form, see Reporting Life Sciences Research. For further information on Nature Research policies, including our data availability policy, see Authors & Referees and the Editorial Policy Checklist.

` Experimental design 1. Sample size Describe how sample size was determined. For the cases where we had previous examples of similar research we conducted power analyses using G*Power software (Faul, F., Erdfelder, E., Lang, A. G. & Buchner, A. G*Power 3: a flexible statistical power analysis program for the social, behavioral, and biomedical sciences. Behavior research methods 39, 175-191 (2007)) applied to the previous studies. In cases of novel paradigms tested here for the first time we used sample sizes optimized for power at normal distribution, typically 40 participants (20 per group) in a within-subjects repeated measures design. In all cases we estimated and reported Cohens d' as a measure of resultant power. Finally, we also conducted p_curve analysis to guard against biased reporting (Simonsohn, U., Nelson, L. D. & Simmons, J. P. P-curve and effect size: Correcting for publication bias using only significant results. Perspectives on Psychological Science 9, 666-681 (2014))

Notably, our cohort of 35 participants with ASD overall, and ~20 participants per individual experiment is relatively large in comparison to typical studies collecting autonomic physiology measures from adults with ASD. 2. Data exclusions Describe any data exclusions. Exclusions were according to previously published criteria. For EDA: We applied exclusion criteria according to Green et al. This included exclusion of participants with no EDA reactivity at all, i.e., EDA under 0.02 μS, or with excessive motion. This resulted in the exclusion of 1 TD and 3 ASD participants from ER-EDA in the Faces task in Experiment 2, and 5 ASD participants from ER- EDA in the Stroop task in Experiment 2. In experiment 4, one TD participant had no EDA recording due to technical fault. For startle: We applied exclusion criteria according to Pause, Prehn, and Blumenthal. The startle response is typically observed in a time window of 30-90 ms following the acoustic stimulus. Events in which the maximum amplitude in this temporal window was higher than the maximum amplitude within the 100 ms prestimulus baseline by a factor of 2.5 were considered as responses. Participants that had at least 5 responses in each condition were considered as responders. According to this criterion, 1 TD participant was excluded from further analysis. One ASD participant was excluded because his mean EMG amplitude exceeded more than 3 SD from mean EMG amplitude of all participants.

3. Replication Describe whether the experimental findings were For the revision we replicated two paradigms (Faces and Stroop) in 8 additional reliably reproduced. subjects with striking replicability (Figure in Reply to Referee document) June 2017 4. Randomization Describe how samples/organisms/participants were For subject selection, no randomization, nor preselection was used. This is because allocated into experimental groups. we have one group of TD and one of ASD, so we obviously cannot assign across groups. We did, however, randomize trial order within experiments, and counterbalanced conditions across subjects.

1 5. Blinding nature research | life sciences reporting summary Describe whether the investigators were blinded to All subjects were blind to condition in all experiments. group allocation during data collection and/or analysis. Experimenters were also blind to condition in experiment with AND and in the behavioral fear Experiment (the Manikin Experiment). Note: all studies involving animals and/or human research participants must disclose whether blinding and randomization were used.

6. Statistical parameters For all figures and tables that use statistical methods, confirm that the following items are present in relevant figure legends (or in the Methods section if additional space is needed). n/a Confirmed

The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement (animals, litters, cultures, etc.) A description of how samples were collected, noting whether measurements were taken from distinct samples or whether the same sample was measured repeatedly A statement indicating how many times each experiment was replicated The statistical test(s) used and whether they are one- or two-sided (note: only common tests should be described solely by name; more complex techniques should be described in the Methods section) A description of any assumptions or corrections, such as an adjustment for multiple comparisons The test results (e.g. P values) given as exact values whenever possible and with confidence intervals noted A clear description of statistics including central tendency (e.g. median, mean) and variation (e.g. standard deviation, interquartile range) Clearly defined error bars

See the web collection on statistics for biologists for further resources and guidance.

` Software Policy information about availability of computer code 7. Software Describe the software used to analyze the data in this For general statistical analysis we used Statistica for Windows and Matlab. We study. used G*Power for power analysis, and Chart 7 for mac (AD Instruments) for analyzing autonomic measures.

For manuscripts utilizing custom algorithms or software that are central to the paper but not yet described in the published literature, software must be made available to editors and reviewers upon request. We strongly encourage code deposition in a community repository (e.g. GitHub). Nature Methods guidance for providing algorithms and software for publication provides further information on this topic.

` Materials and reagents Policy information about availability of materials 8. Materials availability Indicate whether there are restrictions on availability of No restrictions unique materials or if these materials are only available for distribution by a for-profit company. 9. Antibodies Describe the antibodies used and how they were validated No antibodies used in the study for use in the system under study (i.e. assay and species). 10. Eukaryotic cell lines a. State the source of each eukaryotic cell line used. No eukaryotic cell line used in the study

b. Describe the method of cell line authentication used. NA June 2017 c. Report whether the cell lines were tested for NA mycoplasma contamination.

d. If any of the cell lines used are listed in the database NA of commonly misidentified cell lines maintained by ICLAC, provide a scientific rationale for their use.

2 ` Animals and human research participants nature research | life sciences reporting summary Policy information about studies involving animals; when reporting animal research, follow the ARRIVE guidelines 11. Description of research animals Provide details on animals and/or animal-derived NA materials used in the study.

Policy information about studies involving human research participants 12. Description of human research participants Describe the covariate-relevant population A total of 35 (2F) cognitively able adult participants with ASD (Table 1) and 81 (2F) characteristics of the human research participants. TD matched controls participated in the reported experiments. All individuals in the ASD group were assessed by experienced clinicians independent of the present study and met DSM (mostly -IV) diagnostic criteria for an ASD. The TD and ASD cohorts did not differ in age (TD = 27.3 ± 3.3 years; ASD = 26.3 ± 6.0 years, t(99) = 1.1, p = 0.29) only slightly but significantly differed in years of education (ASD = 13.3 ± 2.6, TD = 14.8 ± 1.6, t(96) = 3.6, p < 0.001), and as intended, significantly differed in autism quotient (AQ) score (ASD = 24.7 ± 6.5, TD = 16.5 ± 4.7, t(98) = 7.2, p < 0.0001). June 2017