The behavioral component of sexual inhibition and its relation with testosterone levels: An fMRI study in transgender and cisgender individuals

transgender individuals


Introduction
Transgender individuals are people who experience an incongruence between their gender identity and sex assigned at birth.While transgender identities are not defined by sexuality, feelings of gender incongruence and variables such as the related body image concerns and gender affirming hormone treatment (GAHT) are likely to affect sexual behavior in complex ways (Holmberg et al., 2019).In some transgender individuals, these interactions can lead to the avoidance of sexual activity.For instance, Gil-Llario et al. (2021) report that 59.2% of transgender women (female gender identity, male sex assigned at birth) and 37% of transgender men (male gender identity, female sex assigned at birth) who are waiting for gender affirming medical treatment (GAMT) avoid sexual intercourse.Even after starting GAMT, transgender individuals might still show an inhibited sexual response profile, as having difficulties with initiating and seeking sexual contacts has been found to be one of the most common struggles with sexuality in transgender individuals (Kerckhof et al., 2019).
These findings, related to avoidance of sexual activity in transgender individuals, rely on questionnaire studies, where participants explicitly reply to questions about their own behavior.However, such explicit measures can be greatly affected by social desirability and a participant's conscious and easily accessible beliefs about their own behavior (De Houwer, 2006), while sexual behavior relies on responding processes that can operate without conscious control (Bargh and Barndollar, 1996;Dewitte, 2016).In order to measure avoidance behavior without the distortion of conscious processes, researchers have developed the implicit Approach Avoidance Task (AAT).In this task, participants are instructed to approach or avoid certain stimuli as quickly and accurately as they can, after which their reaction times are analyzed to make inferences about their underlying avoidance tendencies.The task was first developed in the context of fear of spiders (Rinck and Becker, 2007), where it was found that the AAT results predicted fear-related behavior independently from questionnaire scores.Since then, the paradigm has been successfully applied to various fields of study such as emotion regulation (Eisma et al., 2015;Heuer et al., 2007) and substance abuse (Cousijn et al., 2011;Watson et al., 2013;Wiers et al., 2009).
The AAT has also been adapted to the field of sex research (Dewitte, 2016;Hofmann et al., 2009) and has recently been framed as measuring the behavioral (as opposed to cognitive) component of sexual inhibition, and to rely on a distinct neural system.For instance, in cisgender men, the behavioral component of sexual inhibition recruits the inferior frontal gyrus and inferotemporal cortex, and is further characterized by a hypoactivation in the anterolateral prefrontal cortex (Rodriguez-Nieto et al., 2019).The complexity of this inhibitory system has been illustrated by the suggestion that testosterone might play an important role as modulator of the neural activity underlying behavioral sexual inhibition, as testosterone levels correlate positively with activity in the ventrolateral prefrontal cortex, putamen, middle temporal gyrus, and precentral gyrus during the avoidance of sexual stimuli in cisgender men (Rodriguez-Nieto et al., 2017).This involvement of testosterone is of relevance for transgender individuals, as many (but not all) of them might at some point consider GAHT, by which they increase or decrease their testosterone levels (Coleman et al., 2022).Human sexuality and sexual responding are complex phenomena, affected by many interacting determinants related to biology, psychology, and social variables (Althof et al., 2005;Rosen and Barsky, 2006).Whenever transgender individuals avoid sexual encounters, this can be caused by (neuro)biological, psychological, and social variables, and most likely an interaction of these.However, to date, there are no studies investigating the behavioral component of sexual inhibition, its possible underlying neuronal mechanisms, and a possible modulatory effect of testosterone in transgender individuals.
In this study, we aim to offer an exploration of the association between the approach and avoidance of sexual stimuli, brain activation, and testosterone levels in transgender and cisgender individuals, in order to build a biopsychological framework of sexual responding in transgender individuals.Our main objectives are to 1) investigate whether the neural mechanisms underlying behavioral sexual inhibition reported in cisgender men (Rodriguez-Nieto et al., 2019) can be replicated and generalized to cisgender women and transgender individuals, 2) investigate whether these neural mechanisms differ between transgender and cisgender individuals, and 3) explore whether testosterone can explain these differences by investigating the modulating effect of testosterone on brain activation during the AAT.

Participants
We included both cisgender and transgender participants.All participants had to be at least 18 years old, fluent in either English or Dutch, and in the ability to travel to Maastricht University.Exclusion criteria included medical disorders related to sex hormones (e.g., congenital adrenal hyperplasia), neurological disorders (e.g., prosopagnosia, aphasia), and features not considered safe in the MRI scanner environment (e.g., blood vessel clips, metal prosthetic implants, permanent make-up).Transgender participants were recruited via social media and online forums; via the personal network of the main researcher; via snowballing; and via Ziekenhuis Oost-Limburg at Genk, Belgium, which is a hospital that offers gender affirming care.Towards those recruited via the hospital, it was clearly communicated that no medical data would be linked to the study data, and that no individual outcomes would be shared with the clinicians.Cisgender participants were recruited via the SONA system of Maastricht University (through which students can earn credits by participating in scientific research) and via personal contacts of the main researcher.In total, 80 individuals participated in the MRI study.Given that during twelve of those sessions we encountered technical problems related to the Approach-Avoidance Task, and one additional participant made mistakes on more than 20% of the trials, the final sample presented in this study includes participants.

Study procedure
The study was approved by the Ethics Review Committee Psychology and Neuroscience (ERCPN) of Maastricht University (approval code: 213_05_10_2019).Additionally, we gathered feedback from the transgender community on our research design via focus groups (output to be published elsewhere).Individuals expressing interest in the study first received an extensive information letter containing an outline of the study procedure and information about the study aim and participants' rights (e.g., the right to discontinue participation at any time during the study without having to provide a reason).If individuals still expressed interest in participating after having received this information, the main researcher would check for inclusion and exclusion criteria via email, and schedule the MRI session with the participant.One week before the MRI session, participants would receive a link to an online questionnaire to be completed before the session.This questionnaire took approximately 30-45 min.Furthermore, they received instructions regarding personalized stimuli (which were used for an MRI task not included in this manuscript, so we will currently not further report on this).
MRI sessions were conducted at Scannexus, Maastricht, the Netherlands.Upon arrival, participants provided written informed consent, received an elaborate explanation of the study procedure and had the opportunity to ask questions.Then, a saliva sample was collected to measure testosterone levels.After this, participants would go into the 3 T MRI scanner.The scanning session consisted of an anatomical scan (approx.7 min), two resting-state scans (each approx.10 min; data not reported in this manuscript), an Approach Avoidance Task (approx.22 min), and a Face Adaptation Task (approx.18 min; data not reported in this manuscript).After the scanning session, participants completed a final behavioral task on a computer (approx.min; data not reported in this manuscript).Finally, participants were debriefed with more background information about the research questions and methods, potential support resources in case they felt upset after the study, and how to follow up on study outcomes.Participants received €30 as a reimbursement for their participation, and if applicable, travel reimbursement.
Saliva samples were frozen and stored at − 20 degrees Celsius until analysis (which was performed in Dresden, Germany).After thawing, samples were centrifuged at 3000 rpm for 5 min, which resulted in a clear supernatant of low viscosity.Salivary concentrations were measured using commercially available chemiluminescence immunoassay with high sensitivity (LOQ 1.05 pg/ml; Tecan -IBL International, Hamburg, Germany).For each sample, a duplicate analysis was performed (inter-assay CV 5.1%), and the measures reported in this article represent the average of those two outcomes per participant.For some cisgender women, testosterone values were too low to be detectable.In those cases, their value was set to 0.

MRI acquisition
MRI data were acquired with a 3 T Siemens Prisma Scanner.After a short localizer scan, T1-weighted images were acquired in the sagittal plane, TR = 2300 ms, TE = 2.98 ms, TI = 900 ms, 192 slices, FOV = 256 mm, matrix sixe = 256 × 256, 1 mm isotropic voxel.During the Approach Avoidance Task, a total of 1022 T2 * -weighted echo planar images were acquired with TR = 1300 ms, TE = 32.60 ms, flip angle = 70 deg, slice thickness = 2 mm, FOV = 208 mm, 60 slices, multi-band acceleration factor = 4, and a total scan time of 22 min and 8 s 2.3.Materials 2.3.1.Online questionnaires 2.3.1.1.Demographics.A short demographic questionnaire assessed participants' age, country of residence, first language, and number of children using open questions.Multiple choice questions assessed educational level, living situation (e.g., living with partner, family, or alone), and whether participants had a romantic relationship (and if so, the duration of the relationship).
Sex assigned at birth was assessed using a multiple choice question presenting the options Male, Female, or Other (please specify).The latter option was selected by none of the participants.Gender identity was assessed using a multiple choice question presenting the options Man, Woman, Non-binary, or Other (please specify).We divided participants into five groups based on their answers on these two questions: cisgender men (CM; Male + Man), cisgender women (CW; Female + Woman), transgender men (TM; Female + Man), transgender women (TW; Male + Woman), and non-binary (NB; when selected Non-binary or Other on the second question).Furthermore, participants were asked whether they identified as transgender and whether they had ever received a diagnosis of Gender Incongruence or Gender Dysphoria via Yes/No questions.Transgender participants also indicated around what age they became aware of their transgender identity via a multiplechoice question.Sexual orientation was briefly assessed using two sliding scales, one to indicate attraction to men and one to women, ranging from 0 to 100.
Participants with a female sex assigned at birth additionally answered two questions about whether they were taking hormonal contraception, and about their menstrual phase (indicating in which phase they were in case they had a menstrual cycle).Finally, all participants indicated whether they currently had a mental health diagnosis (and if so, which one(s)), and whether they would be taking medication at the time of the MRI session.

Gender affirming medical treatment.
Transgender participants were presented questions regarding GAMT.First, they indicated whether they were on any waiting lists for GAMT (for an intake consultation, GAHT, or GAS).They then indicated whether they were receiving GAHT, and if so, for how many months already, how satisfied they were with the treatment, and what had been their main motives for starting this treatment.In case they did not receive GAHT (yet), they were asked whether they were planning to initiate this in the future, and what their main motives for (not) wanting this were.Similarly, participants were asked whether they had undergone any GAS procedures, and if so, how many months ago this was, how satisfied they were with the outcomes, and what had been their main motives.They were also asked whether they planned to undergo any (further) GAS, and indicated their main motives for (not) wanting this.

Gender dysphoria.
Gender dysphoria was assessed using the Utrecht Gender Dysphoria Scale -Gender Spectrum (UGDS-GS; McGuire et al., 2020).This questionnaire consists of 18 statements that participants rate on a five point scale ranging from Completely disagree to Completely agree.Items include statements such as 'The bodily functions of my assigned sex are distressing for me (i.e., erection, menstruation)' and 'I hate my birth assigned sex'.For every participant, ratings from all items were averaged, resulting in scores ranging from 1 (low gender dysphoria) to 5 (high gender dysphoria).The scale showed high reliability in this sample (Mc Donald's ω = .962).

Sexual esteem.
The measure of sexual esteem, which is defined as a person's self-evaluation of worth as a sexual being, was taken from a broader questionnaire on sexual self-concept (Buzwell and Rosenthal, 1996;adapted by Deutsch et al., 2014).All items were statements that had to be rated on a five-point scale ranging from Strongly disagree to Strongly agree.The scale consists of 24 items belonging to four subscales (behavior, body perception, conduct, and attractiveness), which we combined into one general Sexual Esteem measure (Mc Donald's ω = .913).A score ranging from one to five was calculated by averaging all the item scores.
2.3.1.5.Sexual satisfaction.We used the Global Measure of Sexual Satisfaction (GMSEX; Lawrance and Byers, 1995) to assess sexual satisfaction.The scale presents five sets of antonyms to describe one's sex life, such as Good vs. Bad and Satisfying vs. Unsatisfying.Participants are asked to indicate which antonym applies most to them using a seven point scale.The ratings on the items are summed, resulting in scores ranging from 7 (low sexual satisfaction) to 35 (high sexual satisfaction).The scale showed high reliability in this sample (Mc Donald's ω = .947).
2.3.1.6.General life satisfaction.General life satisfaction was assessed using the Satisfaction With Life Scale (SWLS; Diener et al., 1985).The scale consists of five items such as 'In most ways my life is close to my ideal', which participants rate on a seven point scale ranging from Strongly disagree to Strongly agree.Item sum scores range from 7 (low general life satisfaction) to 35 (high general life satisfaction), and the scale showed high reliability in the current sample (Mc Donald's ω = .897).

Anxiety and depression. The Hospital Anxiety and Depression
Scale (HADS; Zigmund and Snaith, 1983) was used to assess anxiety and depression.The scale interleaves seven items regarding anxiety (e.g., 'I get a sort of frightened feeling as if something awful is about to happen') with seven items regarding depression (e.g., 'I look forward with enjoyment to thing', reverse item).Participants indicated for each item how applicable it is to them using four options that are scored from 0 to 3. Sum scores range from 0 to 21 for both anxiety and depression, with higher scores indicating higher anxiety/depression.The anxiety scale (Mc Donald's ω = .775)and depression scale (Mc Donald's ω = .840)each showed sufficient reliability in our sample.

Other questionnaires.
In addition to the questionnaires described above, participants completed several other questionnaires that will not be analyzed in the current article.These included questionnaires on self-concept discrepancies (Higgins et al., 1986;Lynch et al., 2009) and sexual attitudes and sexual self-efficacy (Deutsch et al., 2014).

Approach avoidance task
During the scanning session, participants performed an Approach Avoidance Task (AAT) (Fig. 1) in the scanner, during which they were instructed to respond to sexual stimuli and non-sexual stimuli with a joystick.Participants were assigned one of three sexual stimuli sets, based on their own indication of what they preferred/found most arousing.They could choose between a set depicting a woman and a man; two women; or two men.Stimuli were found on the internet and were selected on the basis of presenting couples with various body types and ethnicities engaging in naked kissing and caressing, manual or oral sex, and penetrative sex; not including thumbnails; having the couple as the main focus of the picture; and not including sexual practices that would potentially be found aversive or extreme by participants (e.g., BDSM).The set of non-sexual stimuli was identical for each participant, and presented a man and a woman dancing (taken from Rodriguez et al., 2018).These pictures of dancing couples were chosen as control stimuli to match the sexual stimuli in terms of visual complexity (body composition, color,.).Both stimulus sets consisted of 48 pictures, of which 24 were presented vertical (257.6 ×400 pix) and 24 horizontal (337.9 ×272.5 pix).
The AAT consisted of four blocks, each with 48 randomized trials (24 sexual and 24 non-sexual stimuli).In two blocks, participants were instructed to approach sexual stimuli and avoid non-sexual stimuli, and vice versa in the other two blocks.For each condition (approaching sexual stimuli and avoiding sexual stimuli), one block included vertical stimuli and the other horizontal stimuli, resulting in four unique blocks (approach sex with vertical stimuli; approach sex with horizontal stimuli; avoid sex with vertical stimuli; and avoid sex with horizontal stimuli) which were randomized across participants.In each trial, a stimulus was shown for 1300 ms during which participants had to make an approach or avoidance response.Approaching a stimulus entailed pulling the joystick towards them, while avoiding entailed pushing it away.During the next 1300 ms, the stimulus would change size depending on the participant's response, increasing in size after an approach response (to 515.2 ×800 pix for vertical stimuli, 675.8 ×545 pix for horizontal stimuli), decreasing in size after an avoidance response (to 128.8 ×200 pix for vertical stimuli, 168.95 ×135.25 pix for horizontal stimuli), and not changing in size when no response was made or the reaction time was above 1300 ms.Next, a fixation cross was shown during the inter trail interval, which was jittered around 2600, 3900, and 5200 ms.Including instructions and breaks between blocks, task duration totaled to 22 min and 8 s.For behavioral and fMRI analyses, only correct trials were analyzed.Trials were considered incorrect when participants made a fault response (e.g., making an approach response when an avoid response was required), or when they responded after more than 1300 ms.

Questionnaire and behavioral analyses
All statistical analyses were performed using the software JASP (JASP Team, 2020).We applied a significance threshold of p = .05for all analyses.For the AAT, an Approach Avoidance Index was calculated by subtracting reaction times in blocks where sexual stimuli had to be avoided from reaction times in blocks where sexual stimuli had to be approached.A higher index hence indicates a stronger control over sexual approach behavior.Group differences between the transgender and cisgender groups on the questionnaires and behavioral AAT outcomes were analyzed using independent sample t-tests in case of continuous variables, and chi-square tests of independence in case of categorical variables.When we were interested in differences between the five groups (CM, CW, TM, TW, and NB), we applied one-way ANOVAs for continuous variables.In case of a significant ANOVA effect, post-hoc comparisons were analyzed applying the Tukey HSD test for multiple comparisons.Paired t-tests were applied to make within-subject comparisons regarding behavioral AAT outcomes (to check for differences between conditions within subjects).All correlations reported were calculated using Pearson product-moment correlation coefficients.

fMRI analyses
The neuroimaging data were pre-processed and analyzed with Brain Voyager QX Version 21.4 (Brain Innovation, Maastricht, Netherlands).The images were motion-corrected (trilinear / sinc interpolation and aligned to the first functional volume) and corrected for slice timing skew using cubic spline interpolation.A temporal high pass filter (two cycles) was applied.Images were co-registered to the individual T1 weighted images and normalized to Talairach stereotaxic space.Volume time courses were spatially smoothed using a 6 mm full width halfmaximum Gaussian kernel.
We conducted a random-effects general linear model (GLM) analysis with correction for serial correlations to analyze AAT effects.Each condition was entered as a regressor in the design matrix (Approach Sex, Avoid Dance, Approach Dance, Avoid Sex).Task effects were analyzed by contrasting Approach Sex with Approach Dance trials to investigate activation during the approach of sexual stimuli, and by contrasting Avoid Sex with Avoid Dance trials to investigate activation during the avoidance of sexual stimuli.In order to investigate differences in brain activation during the AAT, a random-effects GLM analysis was performed with the four conditions as within-subjects factor and cisgender vs. transgender on the between-subjects level.These analyses were all performed on the whole-brain level.The resulting maps were corrected for multiple comparisons by means of cluster threshold level estimation (p set at <.001, 1000 Monte Carlo simulation iterations; Forman et al., 1995), and the nomenclature of the cluster peak values was defined with the software tool Talairach Client (Lancaster et al., 1997(Lancaster et al., , 2000)).Finally, we performed ROI analyses investigating the relation between testosterone levels and beta values in the regions found to show significant Fig. 1.Design of the Approach Avoidance Task.The timeline presents two trials in a block in which participants were instructed to approach non-sexual and avoid sexual stimuli.When seeing a non-sexual picture, participants had to pull the joystick towards themselves, after which the picture would become larger.When seeing a sexual picture, participants had to push the joystick away, after which the picture would become smaller.The opposite reaction was required in blocks where participants were instructed to approach sexual and avoid non-sexual stimuli.While in this figure stimuli are presented as black and white drawings, during the actual task, they were pictures in colors.
activation differences between cisgender and transgender participants during the AAT in order to investigate a possible modulating effect of testosterone during this task.This relation was investigated by regressing testosterone and sex assigned at birth (to control for the latter) on beta values during the approach and avoidance of sexual stimuli in the ROIs.

Demographics
The sample included 17 cisgender men (CM), 17 cisgender women (CW), 8 transgender men (TM), 14 transgender women (TW), and 11 non-binary individuals (NB; 7 with a female sex assigned at birth, 4 with a male sex assigned at birth).The average age of the sample was 27.72 (SD = 8.89, range 18-56), with cisgender participants (M = 24.85,SD = 5.53) being significantly younger than transgender participants (M = 30.67,SD = 10.66)(t(65) = − 2.813, p = .006).To further inspect this group difference on age, we conducted an additional ANOVA with five instead of two groups, which also indicated a significant group difference (F(4, 62) = 4.283, p = .004)that post-hoc tests revealed to be driven by the TW group being older than the CW (p = .002)and CM groups (p = .035).Most participants' current country of residence was the Netherlands (N = 52, 78%) or Belgium (N = 12, 18%), and the most common first languages were Dutch (N = 38, 57%) and German (N = 14, 21%).Table 1 presents descriptive statistics and group differences on several categorical variables.This table indicates that compared to cisgender participants, transgender participants were more likely to have received a mental health diagnosis and to be taking medication at the time of the scanning session.However, this difference was mostly due to transgender participants with a diagnosis of Gender Incongruence or Gender Dysphoria, and due to transgender participants listing their GAHT as medication.Furthermore, the table indicates that there was a slight difference in distribution according to living situation and number of children, with cisgender participants more often indicating living with (a) friend(s) and not with a partner, and having no children compared to transgender participants.This likely follows from the age difference reported above.
Cisgender participants rated their attraction to men 49.52 on average (SD = 42.98), and transgender participants 46.03 (SD = 33.64),which was not significantly different (t(64) = 0.367, p = .715).There was a difference, however, in how the two groups rated their attraction to women (t(65) = − 3.726, p < .001),with the cisgender group having lower ratings (M = 39.27,SD = 40.41)than the transgender group (M = 72.24,SD = 31.31).This was mostly driven by CW showing much lower attraction to women than the other groups.In this group, 7 CW indicated taking hormonal contraception and 9 CW indicated not doing so (1 nonresponse).None of the TM or NB individuals with a female sex assigned at birth indicated taking hormonal contraception.In terms of menstrual phase, the CW not on hormonal contraception showed a varied profile, with 4 participants indicating they were in the luteal phase; 2 indicating being in the ovulation phase; 2 indicating they were in the menstrual phase; and 1 non-response.This profile was less varied in the TM and NB (female sex assigned at birth) groups, with 6 TM indicating they did not menstruate and 1 indicating he was in the menstrual phase (1 nonresponse), and 5 NB individuals indicating they did not menstruate and 2 indicating they were in the luteal phase.
Of 33 transgender participants, 21 indicated having received a diagnosis of Gender Incongruence or Gender Dysphoria (64%), and most had become aware of their transgender identity before adolescence (N = 23, 70%).GAMT profiles are presented in Table 2, which illustrates that while there is great heterogeneity among the sample, most participants were receiving GAHT at the time of the scanning session.

Questionnaires
Table 3 presents descriptive statistics for the cisgender and transgender groups for the questionnaires.Transgender participants scored significantly higher on gender dysphoria and lower on general life satisfaction and sexual esteem compared to cisgender participants.There were no group differences on sexual satisfaction, anxiety, or depression.

Testosterone measures
Descriptive statistics for the testosterone levels measured in saliva are presented in Table 4.An ANOVA test indicated that testosterone values differed significantly between the five groups (F(62,4) = 3.176, p = .019).Post-hoc tests indicated that this was driven by TM having significantly higher values than CW (p = .015)and than TW (p = .031).As Table 4 indicates, there was high variability in testosterone levels, especially in transgender participants with a female sex assigned at  Numbers indicate the number of participants indicating having the described GAMT profile.GAHT = gender affirming hormone treatment, GAS = gender affirming surgery, SD = standard deviation.
birth.This was likely driven by the GAHT most of these participants were receiving (e.g., shortly after receiving a testosterone shot).
Values are expressed in pg/ml and represent averages from duplicate analyses.M = mean, SD = standard deviation.

Behavioral approach avoidance task outcomes
The cisgender and transgender groups differed significantly on which of the three stimulus sets they chose to be shown during the AAT (X 2 (2, N = 67) = 16.345,p < .001),with most transgender participants choosing the set depicting two women, and most cisgender participants choosing the set depicting a man with a woman.Participants made an average of 8.96 misses/mistakes during the task (which had a total of 192 trials), with a standard deviation of 7.57 (range 0 to 39).Transgender and cisgender participants did not differ on the amount of mistakes they made (t(65) = 1.722, p = .090),and participants were not inclined to make more or less mistakes on blocks in which they had to approach sexual stimuli compared to blocks in which they had to avoid sexual stimuli (t(66) = − .183,p = .855).Furthermore, transgender and cisgender participants did not differ on the Approach Avoidance Index (t (65) = − .228,p = .820),indicating that both groups had equal control over the behavioral component of sexual inhibition.To explore whether behavioral AAT outcomes were directly related with neuroendocrine or sexual self-concept measure, we calculated the correlations between the Approach Avoidance Index and testosterone level (r = − .014,p = .913)and between the index and sexual esteem (r = − .152,p = .219),which were both non-significant.
Table 5 presents the average reaction times according to reaction and stimulus type.A repeated measures ANOVA with two within-subjects variables (reaction, 2 levels: Approach vs. Avoid; stimulus type, 2 levels: Sex vs. Dance) and one between-subjects variable (group, 2 levels: cisgender vs. transgender) indicated a main effect of stimulus type (F(65, 1) = 7.381, p = .008)and a significant interaction between reaction and stimulus type (F(65, 1) = 13.825,p < .001).Post-hoc tests indicated that participants were faster to react to sexual compared to non-sexual trials, and that they were quicker in approaching sexual trials compared to approaching non-sexual trials (p < .001)(which is consistent with previous studies: see Rodriguez-Nieto et al., 2019;Hofmann et al., 2009;Dewitte 2016).
Reaction times are presented in ms.

Activation during approach and avoidance of sexual stimuli
Aiming to identify the brain activation profiles during the AAT in the entire sample, we contrasted Approach Sex trials with Approach Dance trials in order to capture activation during the approach of sexual stimuli, and Avoid Sex trials with Avoid Dance trials in order to capture activation during the avoidance of sexual stimuli (Table 6).The approach of sexual stimuli resulted in activation of clusters in the right middle occipital gyrus, left and right precentral gyrus, and left inferior parietal lobule (Fig. 2).Furthermore, there was a decrease in activation in the left and right insular cortices, right inferior parietal lobule, and right inferior and middle frontal gyri.The avoidance of sexual stimuli activated clusters in the left inferior occipital gyrus, and left anterior and posterior cingulate, and deactivated clusters in the right and left transverse temporal gyri, right inferior parietal lobule, right medial frontal gyrus, and left middle frontal gyrus (Fig. 3).

Differences between cisgender and transgender groups
In order to investigate whether cisgender and transgender participants showed differential neural activation during the AAT, we performed a random effects GLM analysis with condition (4 levels: Approach Sex, Avoid Dance, Approach Dance, Avoid Sex) as withinsubjects factor and cisgender/transgender (2 levels: cisgender, transgender) as between-subjects factor (Table 7).To further inspect these   X, y, and z present cluster peak coordinates in standardized Talairach space.R = right, L = left.
interactions, we performed ROI analyses on the significant clusters, performing a repeated-measures ANOVA for each separate cluster.The results are presented in Fig. 4, which displays various patterns, that were further investigated using post-hoc tests (see Appendix A for statistics).
In the cisgender group, there was higher activation in the left medial frontal gyrus during the avoidance of non-sexual stimuli compared to the approach of non-sexual stimuli, which was not the case in the transgender group.Similarly, the left precentral gyrus showed higher activation during the avoidance of non-sexual trials compared to the approach of sex trials in the cisgender but not the transgender group.Additionally, the left inferior parietal lobule showed higher activation during both types of non-sexual trials compared to trials in which sexual stimuli were approached in the cisgender group, but not the transgender group.The right anterior cingulate and left middle temporal gyrus showed differential activation according to response rule in the transgender group, which was not the case in the cisgender group.Finally, transgender participants showed higher activation in the left fusiform gyrus when approaching compared to avoiding non-sexual trials, which was not the case in the cisgender group.

Testosterone analyses
Given that, due to their GAHT, four transgender participants showed a very high testosterone value (> 100 pg/ml), these participants were excluded from the current analyses.We investigated whether testosterone modulated the neural response during the AAT in the six ROIs that showed significant differences between cisgender and transgender participants.In order to control for sex assigned at birth, we regressed testosterone levels and sex assigned at birth (dummy coded) on average cluster beta values when approaching sexual stimuli (Approach Sex > Approach Dance) and avoiding sexual stimuli (Avoid Sex > Avoid Dance).There were no significant outcomes (all p > .001,see Appendix A), indicating that the neural response in these ROIs does not seem to be modulated by testosterone.
Additionally, we performed an ANCOVA analysis (with testosterone as covariate) on the GLM maps for approaching and avoiding sexual stimuli on the whole-brain level as an explorative analysis.In the transgender group, testosterone was positively correlated with activation levels during the approach of sexual stimuli (Approach Sex > Approach Dance) in the right claustrum (cluster peak TAL coordinates: x = 34, y = 11, z = 1; r = 0.645, p = .000157;Fig. 5).In the cisgender group, testosterone correlated negatively with activation levels during the approach of sexual stimuli in the left amygdala (cluster peak TAL coordinates: x = − 25, y = − 2, z = − 13; r = − 0.661, p = .000021;Fig. 6).Finally, during the avoidance of sexual stimuli (Avoid Sex > Avoid Dance), there was a negative correlation between testosterone and activation levels in the right postcentral gyrus (cluster peak TAL coordinates: x = 52, y = − 25, z = 44; r = − 0.591, p = .000236;Fig. 7) in the cisgender group, and there were no significant clusters in the transgender group.

Discussion
In this study, we investigated the associations between brain regions underlying the behavioral component of sexual inhibition and testosterone levels in transgender and cisgender participants.Functional MRI analyses of an Approach Avoidance Task indicated that the approach of sexual stimuli resulted in the activation of clusters in the right middle occipital gyrus, bilateral precentral gyrus, and left inferior parietal lobule, and a deactivation in the bilateral insular cortices, right inferior parietal lobule, and right frontal regions.The avoidance of sexual Fig. 2. Brain activation at the group level during the approach of sexual stimuli (Approach Sex > Approach Dance trials, cluster level threshold corrected at p < .001).Top row: anterior perspective, coronal view from anterior perspective, posterior perspective.Middle row: superior perspective, transverse view from superior perspective, inferior perspective.Bottom row: right lateral perspective, left medial perspective, left lateral perspective.
stimuli activated the left inferior occipital gyrus, and left anterior and posterior cingulate, and deactivated clusters in the bilateral transverse temporal gyri, right inferior parietal lobule, and bilateral frontal regions.Activational differences between transgender and cisgender participants were found in the left medial frontal and middle temporal gyri, left inferior parietal lobule, left precentral and fusiform gyri, and right anterior cingulate.When investigating whether brain activation in these regions during the AAT was related to testosterone levels, this was found not to be the case.However, on the whole brain level, testosterone levels modulated activation in the right claustrum during the approach of sexual stimuli in the transgender group, and in the left amygdala in the cisgender group.Finally, testosterone levels were negatively correlated to activity in the right postcentral gyrus in the cisgender group during the avoidance of sexual stimuli.Below, we will discuss these findings and their implications, as well as study strengths and limitations.

Neural activations underlying approach and avoidance of sexual stimuli: consistent with previous studies
When reviewing the extant literature on the neural bases of general and sexual inhibition, our findings regarding task activation during the AAT are very much in line with previous outcomes.For instance, we replicated findings by Rodriguez-Nieto et al. (2019), who reported deactivation of the right inferior parietal lobule and frontal regions during the avoidance of sexual stimuli, and activation of the right middle occipital gyrus during the approach of sexual stimuli.Furthermore, all regions that showed diminished activation during the approach of sexual stimuli in our study (bilateral insula, right inferior parietal lobule, right inferior and middle frontal gyri) have previously been associated with general response inhibition (Dambacher et al., 2014(Dambacher et al., , 2015;;Chambers et al., 2009;Swick et al., 2011).This confirms that the AAT was able to activate behavioral sexual inhibitory processes in our sample.
Additionally, the task recruited brain regions that are not typically associated with inhibitory processes.The bilateral precentral gyri activation seen when approaching sexual versus non-sexual stimuli is likely related to the faster reaction times in blocks where sexual stimuli had to be approached compared to avoided, indicating a faster or stronger motor response when approaching sexual stimuli relative to approaching non-sexual stimuli.Furthermore, both the approach and the avoidance of sexual versus non-sexual stimuli activated occipital regions, which is likely related to the fact that these regions have been found to process information on the sexual arousing character of such stimuli  X, y, and z present cluster peak coordinates in standardized Talairach space.R = right, L = left.(Stoléru et al., 2012).

Minimal differences in neural activation profile between transgender and cisgender participants
Analyses investigating group differences between transgender and cisgender participants revealed a few brain regions that showed differential activation between groups during the AAT.Follow-up analyses indicated that for some of these regions, these differences were only related to conditions including non-sexual stimuli and not sexual stimuli.We will not further discuss these differences, and will only focus on brain regions showing group differences related to sexual stimuli.
The right anterior cingulate cortex showed higher activation during blocks in which participants were instructed to avoid sexual stimuli   Scatterplot (left) presenting the relation between testosterone levels (x-axis, pg/ml) and activity in the right claustrum (right) during the approach of sexual stimuli (y-axis, beta values corrected for serial correlations) in the transgender group (black regression line represents entire group).In each subgroup (colored regression lines), the correlation went into the same (positive) direction, indicating that it is unlikely the overall group results is based on a spurious correlation (r = .344,p = .229in transgender women; r = 0.987, p = .002in transgender men; and r = .761,p = .010in non-binary participants).F = transgender women, M = transgender men, NB = non-binary participants.Fig. 6.Scatterplot (left) presenting the relation between testosterone levels (x-axis, pg/ml) and activity in the left amygdala (right) during the approach of sexual stimuli (y-axis, beta values corrected for serial correlations) in the cisgender group (black regression line represents entire group).Note the bimodal distribution of testosterone based on gender (F = cisgender women, M = cisgender men).When calculating the correlation within each subgroup (colored regression lines), the association went in the same (negative) direction in cisgender men (r = − .407,p = .105),but not in cisgender women (r = .354,p = .325).Fig. 7. Scatterplot presenting the relation between testosterone levels (x-axis, pg/ml) and activity in the right postcentral gyrus during the avoidance of sexual stimuli (y-axis, beta values corrected for serial correlations) in the cisgender group (black regression line represents entire group).Note the bimodal distribution of testosterone based on gender (F = cisgender women, M = cisgender men).When calculating the correlation within each subgroup (colored regression lines), the association went in the same (negative) direction in cisgender men (r = − .546,p = .024),but not in cisgender women (r = .019,p = .943).
relative to blocks in which they were instructed to approach sexual stimuli in the transgender group only.This is not surprising, as this brain region has previously been shown to be active during general and sexual response inhibition (Beauregard et al., 2001;Xue et al., 2018).It is intriguing that this neural response is more pronounced in transgender participants than it is in cisgender participants.This could indicate that the AAT required more neural 'effort' from transgender participants (although more neural activation cannot always straightforwardly be interpreted as such, especially in the absence of behavioral group differences in reaction times as is the case here).Similarly, the left middle temporal gyrus showed differential activation based on task response rule in the transgender and the cisgender group.Interestingly, this region has previously been associated with a perceptual Go/No-Go task (also assessing inhibition processes), in which left hemispheric activation was mostly linked to task instructions based on what participants saw (while right hemispheric activation in this regions was more related to task instructions based on when participants saw a stimuli) (Talari and Hirsch, 2005).Therefore, activation in these brain regions is likely related to the decision process related to the AAT performance, which is (similarly to the pattern in the right anterior cingulate cortex) more pronounced in the transgender compared to cisgender group.
However, the opposite pattern was present in the left inferior parietal lobule.In this brain region, brain activity differed based on condition and response rule in the cisgender group, but not the transgender group.The inferior parietal lobule is associated with the execution of the AAT in our study and previous studies (Rodriguez-Nieto et al., 2019), and in this case seems to be recruited more pronouncedly in cisgender compared to transgender participants.Together, these findings related to group differences in brain activation during the AAT indicate that while transgender and cisgender show similar response patterns on the task, they might arrive there via slightly different neural processes.

No evidence for a modulatory role of testosterone in the approach and avoidance of sexual stimuli
Given previous reports that neural activation during behavioral sexual inhibition is modulated by testosterone in cisgender men (Rodriguez-Nieto et al., 2017), and indications that testosterone influences neural responses underlying sexual arousal in transgender individuals (Kim et al., 2016), we investigated whether task activations in the regions showing differential activation between cisgender and transgender participants were related to testosterone levels.Contrary to our expectations, we did not find any such associations.One possible explanation for this is that Rodriguez-Nieto et al. (2017) found these associations in regions partly different from our ROIs, such as the ventrolateral prefrontal cortex and putamen.However, they did find a modulatory effect of testosterone in the middle temporal gyrus and precentral gyrus, which were included as ROIs in our analyses.Perhaps the findings of Rodriguez-Nieto et al. (2017) were too preliminary, representing a small sample of only thirteen participants, or there was too much variability in our sample with regards to gender identity, testosterone level, or factors such as age.Therefore, including more participants could provide more insights.Alternatively, our lack of significant findings might simply indicate that testosterone levels are not implicated in the neural differences between cisgender and transgender people during sexual inhibition, and that other biological or psychological/social variables are the main drivers of these differences.
However, we did find associations between testosterone levels and activity during the AAT in brain regions that did not show any differences between cisgender and transgender participants, and that have not previously been indicated in sexual inhibition.In the transgender group, testosterone correlated positively with activity in the right claustrum during the approach of sexual stimuli.This brain region has previously been associated with sexual arousal (Stoléru et al., 2012), indicating that transgender participants with higher testosterone levels might have experienced greater arousal at the approach of sexual stimuli than those with lower levels, or that an approach reaction lead to more arousal in this group.Interestingly, ours is not the first study to find an association between testosterone and claustrum activation during a task involving sexual stimuli, as Redouté et al. (2005) found the same association during a passive viewing task.Even more intriguing is that their sample consisted of hypogonadal patients, of whom some were treated (and thus had higher testosterone levels induced by hormonal treatment) and others were not.Taken together with the fact that we did not find this association in our cisgender sample, this suggests that the modulatory effect of testosterone on brain activation during the appraisal of sexual stimuli is only (or mostly) present when these testosterone levels are induced by hormone therapy, which offers interesting insights for clinical practice.Alternatively, this lack of an effect in the cisgender sample is a matter of statistics, as there is fewer variability in testosterone levels in the cisgender group, with all cisgender women showing very low values.
In the cisgender group, testosterone was negatively correlated with activity in the left amygdala during the approach of sexual stimuli, and with activity in the right postcentral gyrus during the avoidance of sexual stimuli.Both brain regions are thought to be involved in the emotional component of appraising sexual stimuli, and the postcentral gyrus is additionally thought to be involved in the regulation of endocrine responses to sexual stimuli (Stoléru et al., 2012).It is likely that these results reflect a spurious correlation, since cisgender men and cisgender women differ highly in testosterone level, and have also been shown to differ in cerebral activation levels in the amygdala and postcentral gyrus during sexual arousal (Stoléru et al., 2012).

Strengths and limitations
This is the first fMRI study assessing sexual responding in transgender individuals using an active paradigm contrary to passive viewing.The validity of our AAT paradigm is supported by the fact that the neural activation we observed is in line with previous studies, and is likely further strengthened by the fact that participants were shown stimuli with a certain degree of individualization.Furthermore, to our knowledge, this is the first MRI study including non-binary transgender participants, expanding generalizability beyond the gender binary.Nevertheless, we would briefly like to describe some study limitations.First of all, the transgender sample was very heterogeneous in terms of GAMT profile and testosterone levels, possibly complicating some analyses, especially those with relation to testosterone.Similarly, the cisgender women in our sample showed a heterogeneous picture in terms of oral contraceptive use and menstrual phase.Furthermore, we did not perform any analyses based on sex assigned at birth or gender identity (as this was not the focus of our objectives), although these factors are also likely to influence results.With regards to the testosterone samples, we did not manage to take blood samples, which are considered to be more reliable than saliva samples.However, we attempted to mitigate this issue by performing duplicate analyses of each sample, and by opting for LC-MS/MS analyses instead of immune-assay/enzyme-assay analyses.Finally, the transgender sample was significantly older than the cisgender sample, which was mainly driven by the transgender women being older than some other groups.However, we do not expect age to significantly affect brain activity in regions typically associated with our current paradigm.

Conclusion
In this study, we investigated which brain regions underlie the behavioral component of sexual inhibition in transgender and cisgender participants using an fMRI paradigm including a sexual Approach Avoidance Task.We further investigated whether these neural bases differ between the two groups, and whether they are related to testosterone levels.The brain regions recruited during our task were mainly consistent with previous literature on general and sexual inhibition, involving areas such as the bilateral insula, left anterior cingulate cortex, and frontal regions.Transgender and cisgender generally showed similarities in neural activation during the task, except for a few regions involved in decision making processes (left middle temporal gyrus) and sexual response inhibition (right anterior cingulate cortex and left inferior parietal lobule).This indicates that, while the groups are mostly similar in their neurobiological response to the AAT, there are slight differences that suggest that transgender individuals are more likely to recruit brain areas associated with decision making (the left middle temporal gyrus) than cisgender individuals.In none of these regions was the response to the AAT modulated by testosterone levels, although we did find that in the transgender group, testosterone levels correlated positively with cerebral activation in the right claustrum (a region associated with sexual arousal) during the approach of sexual stimuli.The fact that this association was lacking in the cisgender sample, indicates that this association might be unique to GAHT affected testosterone levels.Overall, these findings indicate that transgender and cisgender individuals mostly show similarities in their neural response to a sexual inhibition task, and that exogenous testosterone levels are unlikely to play an important role.This suggests that previous clinical findings related to greater avoidance of sexual activity in transgender individuals relative to cisgender individuals are more likely to be explained by social and psychological factors.However, as the transgender and cisgender groups seemed to arrive at similar behavioral outcomes via slightly different neurobiological routes (with the transgender group recruiting brain regions related to decision making, contrary to the cisgender group), this might indicate that approaching and avoiding sexual stimuli might have different meanings for the two groups.Furthermore, the findings suggest that a change in testosterone level via GAHT might affect neural responses underlying the approach and avoidance of sexual stimuli only slightly, and the lack of behavioral group differences in our sample indicate that it is unlikely that these slight changes ultimately affect sexual responding.

Fig. 3 .
Fig. 3. Brain activation at the group level during avoidance of sexual stimuli (Avoid Sex > Avoid Dance trials, cluster level threshold corrected at p < .001).Top row: anterior perspective, coronal view from anterior perspective, posterior perspective.Middle row: superior perspective, transverse view from superior perspective, inferior perspective.Bottom row: right lateral perspective, left medial perspective, left lateral perspective.

Fig. 4 .
Fig. 4. Interaction between Approach Avoidance Task conditions and having a cisgender or transgender identity.ApS = Approach Sex, AvD = Avoid Dance, ApD = Approach Dance, AvS = Avoid Sex.

Fig
Fig.5.Scatterplot (left)  presenting the relation between testosterone levels (x-axis, pg/ml) and activity in the right claustrum (right) during the approach of sexual stimuli (y-axis, beta values corrected for serial correlations) in the transgender group (black regression line represents entire group).In each subgroup (colored regression lines), the correlation went into the same (positive) direction, indicating that it is unlikely the overall group results is based on a spurious correlation (r = .344,p = .229in transgender women; r = 0.987, p = .002in transgender men; and r = .761,p = .010in non-binary participants).F = transgender women, M = transgender men, NB = non-binary participants.

Table 1
Descriptive statistics and group differences on categorical variables education, employment, relationship, living situation, number of children, having received a mental health diagnosis, and taking medication at the time of the scanning session.

Table 2
Group profiles in terms of received gender affirming medical treatment (GAMT).

Table 3
Descriptive statistics for gender dysphoria, sexual esteem, sexual satisfaction, general life satisfaction, anxiety, and depression.

Table 4
Descriptive statistics for testosterone levels measured in saliva samples.

Table 5
Average reaction times Approach Avoidance Task.

Table 6
Neural activation profiles during the approach and avoidance of sexual stimuli (n = 67).

Table 7
Brain regions containing clusters showing differential activations between cisgender (n = 34) and transgender (n = 33) participants during the Approach Avoidance Task.