In Search of an Adult Attachment Stress Provocation to Measure Effect on the Oxytocin System

Design

The goals of this exploratory study, which we consider to be pilot work for future research, were to assess feasibility of the protocol and gather preliminary test-of-concept and effect size data. This study was appended to an NSF-funded social psychology experimental project (Award Number 0820609, PI, S. L. Brown, Physiological Effects of Helping Others) that involved repeated blood draws for oxytocin across a series of conditions. A subset (n = 16) volunteered to extend their participation an additional half hour in order to complete the AI challenge protocol. One participant’s intravenous access could not be maintained, so the final pilot sample included 15 women. Each woman served as her own control, with the bonding segment and abandonment segment considered in relation to her baseline (on arrival to the lab) and in relation to each other.

Recruitment and Participants

The parent study recruited a convenience sample at the University of Michigan using flyers and a research engagement webpage. It was advertised that participants would be paid. The study was explained to be a scientific study on social interactions. The parent study limited participation to females between the ages of 18 and 35. Due to the effects of steroid hormones on oxytocin levels, inclusion criteria were nonpregnant women on days 1 to 14 of their menstrual cycle. Exclusion criteria were smoking, using hormonal medications (i.e., birth control or hormonal replacement therapy), and having a major endocrine disorder (i.e., Cushing’s disease, diabetes, or Addison’s disease). In addition, all participants were asked to refrain from using nonprescription drugs and alcohol for 24 hours prior to the study, to eat breakfast prior to arriving at the laboratory, and to be well hydrated. Due to the possibility that women who are mothers might react differently to the AI challenge, we invited only childless, nonpregnant adult women. Due to the potential for blood or needle phobias to increase stress, volunteers with these fears were asked to refrain from participating.

Overview of Procedures

During the recruitment email contact, enrolled women were instructed to complete a set of online questionnaires using Qualtrics. On arrival, research assistants reviewed the informed consent information, and participants gave written informed consent. The study nurse inserted an indwelling intravenous catheter and collected repeated specimens across the protocols using a needleless stopcock mechanism kept patent with infusions of an anticoagulant (heparin). Prior to testing, all participants were exposed to a 15-minute closeness induction procedure with a confederate, intended to prime the “caregiving system.” This was followed by a 15-minute stress period involving a reaction time test and modified Trier Social Stress test (Fredrickson, Mancuso, Branigan, & Tugade, 2000). Volunteers who continued into the AI challenge component had a 10 to 15 minute recovery period after the end of the parent-study protocol described above and then underwent this AI challenge test by viewing the half-hour film segment. Blood specimens used for this add-on study were from three time points: (a) at the baseline time point of the parent study, which was 20-minutes into the post–catheter insertion recovery period; (b) 1 minute after the film’s “bonding scene”; and (c) 1 minute after the film’s “abandonment scene.”

Film Protocol and Segment Description

After the parent study, the participant remained in the room alone with the lights on to view the film on a 22-inch TV screen from a distance of 7 feet. She was told only that she would be watching a segment of the movie Artificial Intelligence. A nurse entered the room at 5 minutes into the segment (1 minute after the bonding scene), to draw a blood sample through the stopcock. The nurse then left the room while the participant continued to watch the film segment. The nurse returned to the room after 30 minutes to collect the final blood sample 1 minute after the abandonment scene.

The segment used for this protocol is that described by Wirth and Schultheiss (2006). It begins with a mother programming a robotic boy to recognize her as his mother. After initial programming, the robotic child calls her “mother.” As the scene proceeds, the robotic boy becomes more attached to his mother. Five minutes into the scene, the robotic boy tells his mother that he loves her. This moment was identified as a time of bonding and marked the time point for the bonding scene blood specimen to be collected. As the story continues, the mother’s real son comes home from the hospital and the robotic boy has a series of increasingly dangerous interactions with the human son and other children, resulting in a plan to dismantle him. The final scene portrays the woman bringing the robotic boy into the forest where she abandons him in hopes that he will survive. When the robotic boy realizes what is happening he pleads for his “mother” not to leave him. This moment was identified as a stressful moment portraying abandonment and marked the time point for the abandonment scene blood draw.

Survey Measures

As explained above, the parent study was a social psychology experiment done on a convenience sample, so we were able to select by gender, but we were not able to select participants based on psychiatric status. We included measures in the survey that would permit us to characterize the volunteers on factors important to our theory that oxytocin response to an attachment system stress challenge would differ if the woman has posttraumatic spectrum psychiatric disorders. In a sample size of 15 college students, we did not expect many cases of such disorders to occur, so we used self-administered symptom reports that yield dimensional, interval-level variables for analysis. We assessed for history of childhood abuse and neglect and for symptoms of PTSD, depression, dissociation, somatization, and interpersonal sensitivity.

Twenty-eight relevant items from the Life Stressor Checklist were used to gather history of trauma exposure (Wolfe & Kimmerling, 1997). The Life Stressor Checklist was used because it includes nonlegal language, behaviorally specific questions, and a comprehensive list of traumatic events specific to women’s life experiences, including traumatic reproductive and care giving experiences (Kilpatrick et al., 1994). For the purposes of this study, we created a total trauma exposure sum for descriptive purposes (range = 0-28). We classified a woman as having a childhood maltreatment history if she reported physical neglect, emotional abuse, physical abuse, and sexual abuse involving contact only or penetration prior to age 16.

The PTSD Checklist–Civilian Version (PTSD-PCL; Weathers, Litz, Herman, Huska, & Keane, 1993) is a 17-item self-report checklist used to identify the 17 symptoms of PTSD as outlined in the fourth edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM-IV; American Psychiatric Association, 1994). The checklist prompts participants to select the level of distress ranging from 1 = not at all to 5 = extremely to describe each reported PTSD symptom over the prior 30 days. The sum of these items is then computed with a potential total ranging from 17 to 85. PTSD diagnosis was established with a cutoff score of 44 (Blanchard, Jones-Alexander, Buckley, & Forneris, 1996; Ruggiero, Del Ben, Scotti, & Rabalais, 2003). Initial psychometric testing by Weathers et al. (1993) demonstrated a test–retest reliability of .96, with validity established with a kappa of .64 in comparison to the Structured Clinical Interview. More recently, the PTSD-PCL was used with 392 college students and demonstrated a Cronbach’s alpha reliability of .94 for the total scale, with a test–retest reliability of .92 for immediate retesters, .88 for participants with 1-week intervals, and .68 for 2-week interval retesters (Ruggiero et al., 2003).

We measured depression symptoms with the Beck Depression Inventory (BDI; Beck, Ward, Mendelson, Mock, & Erbaugh, 1961). The BDI includes 21 items that are scored from 0 to 3 in terms of intensity (Beck, Steer, & Garbin, 1988). The BDI has demonstrated adequate reliability scores with means of .86 and .81 for psychiatric and nonpsychiatric patients, respectively, and acceptable validity testing compared with the Hamilton Psychiatric Rating Scale for Depression (Beck et al., 1988). The cutoff point indicating likely major depressive disorder is 30 (Beck et al., 1988).

Dissociation is an associated feature of PTSD and considered to be a component of complex PTSD or disorders of extreme stress not otherwise specified (DESNOS; Herman, 1992, 1997). Dissociative disorders also can occur in the absence of PTSD. The Dissociative Experiences Scale–Taxon (DES-T) is an 8-item measure that has been identified as a sensitive measure to identify pathological dissociation (Waller, Putnam, & Carlson, 1996). The DES-T is a self-report measure in which participants indicate the frequency of various dissociative experiences using percentages from 0 to 100 (Waller & Ross, 1997). The scores are then averaged to create a total score. Per Waller and Ross (1997), “In non-clinical samples, DES-T scores that are greater than 35 provide strong evidence that the individual belongs to the pathological dissociation taxon” (p. 506). In their nonclinical study, the cut-point of 35 was very conservative, with a sensitivity of .57, specificity of 1.0, and 100% positive predictive value for a dissociative disorder.

Somatization and interpersonal sensitivity are two additional associated features of PTSD or elements of complex PTSD or DESNOS. We used subscales of the Symptom Checklist–90 (SCL-90) to measure these symptoms (Derogatis, Lipman, & Covi, 1973). Although somatization and interpersonal sensitivity can reach diagnostic thresholds for somatoform and borderline personality disorder diagnoses, the SCL-90 subscales do not have validated cut-points. These subscales assess the presence of symptoms within the last 7 days on a 5-point rating scale from 0 = not at all to 4 = extremely. The somatization and interpersonal subscales have previously demonstrated adequate reliability, including an internal consistency of .86 for both subscales and validity as established by adequate correlations with the Minnesota Multiphasic Personality Index (Derogatis, Rickels, & Rock, 1976).

Hormonal Measures

Blood specimens were drawn with a 5-mL Luer-lock syringe from the stopcock mechanism after a 1 to 2 mL waste to clear the heparin flush. Blood was immediately deposited into a 6-mL vacutainer tube containing ethylenediaminetetraacetic acid and 400 KIU of the protease inhibitor Trasylol (Fisher BioReagents, Pittsburgh, PA) per 1 mL of blood. The vacuum seal was broken so that transfer could occur without hemolysis. Blood samples were kept ice-chilled for up to 2.5 hours before being centrifuged at 4°C at 1,000 rpms for 15 minutes. All plasma samples were separated into aliquots and stored in a freezer at −80°C until assayed.

Plasma Oxytocin Assay

Oxytocin enzymimmunoassay was performed with a commercially available kit obtained from Assay Designs, Inc/Enzo Biomedical (Ann Arbor, MI) using duplicate samples from each aliquot. The plasma was diluted in assay buffer 1:2, to give results reliably within the linear portion of the standard curve. The intra-assay coefficient of variance was 9.1%, and the interassay coefficient of variance was 6.0%. The values used in this analysis are the mean of the duplicates at each time point.

Analysis Plan

Data were analyzed using IBM SPSS Statistics version 20.0. Preliminary analyses included descriptive statistics for demographics, trauma history, psychiatric symptom scores, and oxytocin values in relation to limits of detection and assessment of normality, homoscedasticity, and linearity of relationships among variables. Oxytocin concentration values are depicted in natural units (pg/mL), but statistical tests and regressions were conducted using log-transformed values.

For Research Question 1, we tested Hypotheses 1 through 4 with a repeated-measures analysis of variance (RM-ANOVA) and its associated post hoc analyses. We explored Research Question 2 that psychopathology would be associated with the pattern of oxytocin response using correlation coefficients (Pearson’s r). Strong differences in correlations across different psychiatric measures led to a post hoc analysis of six selected cases to answer Research Question 3. Due to the exploratory, pilot nature of this experiment, we prefer to risk a Type I error, so we accepted P < .10 as the level of statistical significance for interpretation. Given that we were able to specify directional hypotheses, we used one-tailed tests of significance.

Participants

Participants (n = 15) were female university students who ranged in age from 18 to 27 years. All reported themselves to be within 14 days of the onset of their last menstrual period. They were predominantly White and characterized their family income as above average (Table 1). All the participants reported they were not of Hispanic ethnicity. Four participants reported a history of childhood abuse.

OPEN IN VIEWER

Table 1.

Characteristics of the Study Sample.

Demographics
Mean age, years (SD)	21.7 (2.6)
Race, % (n)
Caucasian	46.7 (7)
African America	13.3 (2)
Asian or Pacific Islander	40.0 (6)
Income, % (n)
Below average	13.3 (2)
Average	26.7 (4)
Above average	60.0 (9)
Trauma history
Mean number of trauma exposures (SD)	2.9 (2.3)
History of child abuse, % (n)
Positive for physical, sexual, or emotional	26.7 (4)
Psychiatric symptom scores, mean (SD)
Posttraumatic stress disorder symptoms	32.1 (13.4)
Dissociation symptoms	6.7 (13.2)
Interpersonal sensitivity symptoms	19.7 (6.0)
Somatization symptoms	18.4 (6.9)
Depression symptoms	30.8 (11.6)
Oxytocin levels (pg/mL), mean (SD)
Baseline	183.6 (77.5)

Oxytocin Values in Response to the AI Challenge

Figure 1 displays the mean oxytocin values with standard error bars for the whole sample (n = 15) in pg/mL, with natural log values depicted on a right-hand y axis. The first hypothesis that oxytocin level would change from the baseline to bonding scene was not supported (p = .393). The post hoc analysis for differences between the baseline and abandonment oxytocin values was significant (p = .010), supporting Hypothesis 2. Hypothesis 3 was supported with a significant post hoc analysis demonstrating a difference between oxytocin levels at the bonding and abandonment scenes (p = .041). RM-ANOVA (Table 2) was also conducted to assess Hypothesis 4, assessing the impact of time on oxytocin values across the three time periods (baseline, bonding, and abandonment). There was a significant main effect of time (i.e., effect of provocation), F(2, 28) = 3.61, p = .02. This difference explains 21% of variance by partial eta squared and has an observed power of .62.

OPEN IN VIEWER

Figure 1.

Mean oxytocin values across three protocol time points.

OPEN IN VIEWER

Table 2.

Oxytocin Changes Across Protocol for the Sample (Using the Natural Log of Oxytocin Level).

	Whole sample (n = 15)
	F(df = 2, 28)	Partial eta squared	Observed power
Time point	Mean difference a	Standard error a	p b
RM-ANOVA	3.61, p = .02	.21	.62
Baseline to bonding	.032	.113	.393
Baseline to abandonment	−.309	.116	.010
Bonding to abandonment	−.278	.148	.041

Note. RM-ANOVA = repeated-measures analysis of variance.

a.

Based on estimated marginal means.

b.

Significance determined using p < .10, one-tailed.

Research Question 2 was about the extent to which posttraumatic spectrum psychopathology that can manifest subsequent to early relational trauma or childhood maltreatment and attachment problems might affect the oxytocin response. Our convenience sample size of 15 limited us to exploratory analyses with interval-level variables. To first assess the relationship of psychopathology symptoms to oxytocin levels (a diagnosis-centered exploration), we used a correlation matrix with the aggregate (Table 3). This correlation matrix demonstrated that there was a significant moderate relationship between symptoms of depression, dissociation, somatization, and interpersonal sensitivity and an increase in oxytocin level at the bonding scene time point. Only PTSD symptom count was significantly associated with a decrease in oxytocin level at the abandonment scene time point.

OPEN IN VIEWER

Table 3.

Correlations of Psychopathology With Oxytocin Level Across the Protocol.

	Baseline a	Bonding scene a	Abandonment scene a
Child abuse total b	r = .034, p = .452	r = −.109, p = .349	r = −.204, p = .233
PTSD score	r = −.025, p=.464	r = .190, p = .249	r = −.352, p = .099
Depression score	r = −.126, p = .328	r = .449, p = .047	r = −.113, p = .345
Dissociation score	r = −.021, p = .471	r = .545, p = .018	r = −.045, p = .436
Somatization score	r = −.042, p = .441	r = .593, p = .010	r = −.189, p = .250
Interpersonal sensitivity score	r = .131, p = .321	r = .354, p = .098	r = −.130, p = .322

Note. PTSD = posttraumatic stress disorder.

a.

All p values are one-tailed.

b.

Child abuse total is the sum of types of abuse: physical neglect, emotional, physical, and contact and penetration sexual abuse.

This pattern in the diagnosis-focused correlations led to a post hoc person-focused examination via six selected cases (Table 4) to answer research question 3. We selected six women representing the extremes within this small convenience sample by the following process. We sorted the data file in rank order by their PTSD symptom score on the PTSD Checklist, which can range from 17 (no symptoms) to 85 (all 17 symptoms rated at the highest level), and which has a validated cut-point of 44 corresponding to probable PTSD diagnosis (Blanchard et al., 1996; Ruggiero et al., 2003). Four of the 15 young women had a PTSD score >44. One of these women had a dissociation score of 54, well above the clinical cut-point of 35 (Waller & Ross, 1997). We separated these into one complex PTSD case and three PTSD-only cases. There were two women with no PTSD symptoms and no dissociation symptoms who we considered as healthy controls. We plotted the oxytocin curves of these six cases (Figure 2). The strongest pattern of response is within the woman distinguished by the high level of dissociation, whose oxytocin level increased by 155% at the bonding scene. This contrasts with the PTSD-only trio whose oxytocin levels all fell at the bonding scene, although to a less dramatic extent.

OPEN IN VIEWER

Table 4.

Depiction of Mental Health, Trauma History, and Oxytocin Response Across Six Influential participants.

	ID
	17	37	75	113	28	30
Trauma history	No trauma reported	Index trauma was prior traumatic miscarriage or abortion. Also reported accident, emotional abuse, witnessing parent violence, physical abuse as child and adult, sexual harassment, severe money problems, death of a loved one.	Index trauma was emotional abuse. Also reported family member jailed, witnessing parent violence.	No “worst” trauma. Reported sudden death of a loved one, death of a loved one.	No trauma reported.	No “worst” trauma. Reported sudden death of a loved one.
PTSD (17-85)	54	50	50	48	17	17
Dissociation (0-100)	50	0	20	6	0	0
Interpersonal sensitivity (0-40)	30	28	18	22	10	17
Somatization (0-52)	41	18	14	17	16	14
Depression (0-63)	63	35	38	24	21	26
Mental health	PTSD, dissociation, depression	PTSD only	PTSD only	PTSD only	Healthy	Healthy
Direction	↑	↓	↓	↓	↑	No change
Change (pg/mL)	279	−24	−20	−112	+38	0
Change from baseline	+155%	−10%	−11%	−54%	+35%	0%

OPEN IN VIEWER

Figure 2.

Oxytocin values and psychopathology for six selected cases.

Test of Concept Conclusions

In relation to the Research Question 1, Hypotheses 2 through 4 were supported in this sample, which demonstrated modest changes in oxytocin levels across the three time points. There was a decrease from the baseline to abandonment time point and a decrease from the bonding to abandonment time point. Although the hypothesized increase from baseline to bonding scene did not reach significance, Hypothesis 4 was supported when the RM-ANOVA showed significant differences in oxytocin across the baseline, bonding, and abandonment time points.

In relation to Research Question 2, dissociation and somatization correlated most strongly with higher levels at the bonding scene, and PTSD correlated most strongly with lower levels at the abandonment scene. Although we did not expect to find cases of severe psychopathology in the form of PTSD or dissociation, there were participants with distinct psychiatric symptom patterns, allowing us to consider them as a preliminary indication of what a clinical sample might show. Consistent with what we found in another study of oxytocin, the woman with high dissociative symptoms had the strongest response to the provocation (Seng et al., 2013). Since dissociation psychopathology is strongly associated with early relational trauma (van der Kolk, 2005), and since oxytocin has been theorized to be dysregulated in the face of early relational trauma (Schore, 2003; Seng, 2010; Teicher et al., 2002), the correlation of dissociation symptom level and oxytocin level found here, as well as the pattern seen in this sentinel case, warrant further consideration. The overall lower level of the women with PTSD without dissociation also is consistent with findings from our previous research (Seng et al., 2013). This distinction makes some sense conceptually because PTSD is characterized as an anxiety disorder with hypervigilence (i.e., mistrust), both manifestations that would be more expected in the context of low levels of oxytocin. Whereas dissociation is a form of self-anesthesia in a context of threat and, in the context of maltreatment, may help the young child preserve trust in an untrustworthy parent in the interest of survival, manifestations that would be more expected in the context of high levels of oxytocin (Freyd, 1996). This distinction may also have a biological basis. The hypothalamic–pituitary–adrenal axis and oxytocin system are mutually regulating, so differences in cortisol levels may oppose oxytocin to different extents. Cortisol is sometimes high in PTSD, but has also been found to be low when the PTSD is chronic or from long-ago trauma exposures such as childhood maltreatment, or in women (Meewisse, Reitsma, de Vries, Gersons, & Olff, 2007; Trickett, Noll, Susman, Shenk, & Putnam, 2010). There is also evidence from one study that an enzyme that lyses oxytocin (prolyl endopeptidase) is more active in people with PTSD (Maes et al., 1999), suggesting that the breakdown of the hormone would be faster and amounts available in plasma lower. Current literature (Lanius et al., 2010; Seng, D’Andrea, & Ford, in press; Seng et al., 2013; Stein et al., 2012; Wolf et al., 2012) suggests that there may well be a dissociative subtype of PTSD. Thus, the most interesting finding of this small study might be the hint that oxytocin may be a biomarker with utility for distinguishing these subtypes at the physiological level. Thus, we conclude that this protocol specifically, and oxytocin as a biomarker more generally, warrant further studies that characterize participants on a full range of posttraumatic psychopathology. To verify that the mechanism of the provocation is indeed a relational stressor affecting people differently according to their capacity or propensity to experience healthy attachment, measures of adult attachment should be included. As with any hormone, study of the hormone available in the peripheral circulation is only a rough proxy for its action on target tissues, and in this instance, as a neurotransmitter, so designs are needed that elucidate more elements of how the oxytocin system is functioning under provocation and across subtypes.

The effect sizes observed in the RM-ANOVA are modest but suggest that the protocol provokes a response that is discernible, on average, in a small group of unselected college women. Our post hoc exploration suggests that in a sample selected to contrast women with and without psychopathology, effect sizes might be larger. Thus, we conclude that further testing is needed to state that this protocol is a useful provocation for social psychology research on oxytocin in nonclinical samples. We recommend large enough healthy-control samples in future studies to determine utility and effect sizes within these normal samples compared with people in patient groups.

Feasibility Conclusions

This protocol seemed well suited to our purpose because the science fiction story includes two “tear jerker” scenes, one in which the robot boy tells the human woman “I love you” for the first time, and one in which she reluctantly abandons him in the forest to prevent the need to dismantle him. There are other films we could have used, but they have disadvantages. For example, Sophie’s Choice (Director: Alan Pakula, 1982) depicts abandonment of a child, but the context of the Nazi Holocaust might be an additional provocation. The Kids Are Alright (Director: Lisa Cholodenko, 2010) depicts mothers dropping their daughter off at her dormitory at the start of her first year in college, but this context suggests a normal developmental milestone that might have as much positive valance as negative. Thus, AI as a science fiction film seemed to have the advantages of being free of contextual intensity and of including both positive and negative manifestations of attachment.

The AI challenge protocol to assess plasma oxytocin reactivity is feasible. The film was acceptable to all participants. The data collection protocol worked well, with only 1 of 16 women (6%) losing venous access and no major departures from the established procedures. The specimen processing was hampered by the need to collect plasma in one building, holding it on ice for nearly 3 hours from the baseline draw in the parent study through the last draw in the AI protocol, and transporting it on ice to another location for cold centrifuging and storage. The levels observed in this pilot, however, are consistent with those seen in another report where 2 hours elapsed (Feldman, Gordon, Schneiderman, Weisman, & Zagoory-Sharon, 2010), and in other studies as well (Gordon et al., 2008). As a stand-alone protocol, the AI challenge protocol would take less than 1 hour. If the protocol took place in a clinical or academic facility with a cold centrifuge and freezer at hand, the theoretical risk of deterioration would be decreased.

Strengths, Limitations, and Recommendations for Further Research

There are numerous limitations to this study, many of which are inherent to preliminary studies. First, the sample size was small and not selected for psychopathology. Observed power in the RM-ANOVA was .62, confirming that we were underpowered to draw solid conclusions. Our sample only included women in the follicular phase of the menstrual cycle and is therefore not generalizable to other populations including men, children, and menopausal and pregnant or mothering women. Our specimens may have deteriorated due to the amount of time on ice. This limitation would have affected all 15 women in our sample equally, but it means that the values overall in this study may be low. That said, the lowest oxytocin levels observed are those measured at the end of the protocol, during the abandonment scene, where the delay from blood draw to freezer was the shortest and within the timeframe customary in other oxytocin studies. We did not measure attachment style or problems, so we do not have an indicator to use to verify the mechanism of this film segment as a stressor to the attachment system. Finally, this protocol was an add-on to a social psychology experiment that involved a closeness-induction and stress protocol that should have primed or activated the caregiving system. Our baseline levels represent a basal oxytocin level because it was drawn prior to these parent study manipulations. However, the manipulations occurring between baseline and the AI challenge protocol may have affected the response to the film. We do not know if the priming and stressing in the parent protocol would have the effect of making pilot participants more reactive or less reactive, possibly due to fatigue. For example, this priming may have activated the attachment system prior to our provocation, which could have potentially blunted the effects of the AI bonding scene. Therefore, testing the AI challenge as a stand-alone protocol is an essential next step.

There were strengths to this study as well. The team is experienced at complex social research protocols involving manipulations and blood specimen collection. The Carter laboratory is very experienced at oxytocin assaying and analysis of the raw data. Using an unselected convenience sample of female college students is a conservative first approach since it would usually be considered a homogenous sample, and the amount of variance observed would be considered a likely minimum.