Materials and Methods

Subjects
We recruited 1,158 (584 females; age, mean 21.2± S.D. 1.5) Han Chinese undergraduate students at the National University of Singapore to participate in a study of the biological basis of human behavior through advertisement on the Integrated Virtual Learning Environment. At the beginning of the experiment, subjects completed an informed consent form approved by the Institutional Review Board at National University of Singapore. Subsequently, subjects participated in a 2-hour testing session to complete various tasks including the trust game and risk attitude without any feedback in a fixed order using paper and pencil. At the end of the experiment, two out of 20 tasks are randomly drawn to pay the subjects. Several days later, subjects donated 10 to 20 cc of blood for analysis including plasma oxytocin.
10.1371/journal.pone.0051095.t001 Table 1  Regression results for linear and nonlinear relationship between plasma oxytocin and trust.     The table reports coefficient, standard error, and p values. The last row is R-squared.

Behavioral Design
In the trust game, the first player is endowed with SGP$ 20 (about US$16), while the second player is endowed with nothing. In the first stage, the first player decides how much to send (S) to an anonymous and randomly matched second player (20 – S). For every dollar the first player sends, the second player receives three times (3S). In the second stage, the second player decides how much (B) out of 3S to second back to the first player. At the end, the first player receives (20 – S + B) being the amount he/she keeps plus the amount the second player sends back while the second player receives three times the amount sent deducting the amount sent back (3S – B). We use the strategic method [24], in which the second player states his/her response to each of 21 possible choices from the first player. Every participant plays both roles of first and second players without any feedback. At the payment stage with real money, we randomly determine the specific role – first or second mover – for each pair of subjects. The amount sent by first player is used as a measure for trust while the average return amount from the second player is a measure for trustworthiness.
As trust is an inherently risky behavior viz., the trustee may not reciprocate with an act of trustworthiness, we aim to test whether OT is specific to trust in the social interactions or alternatively plasma OT indexes risk in general. We therefore include a risk task using portfolio choice design [25]. In this risk task, subjects are endowed with SGD $20, and decide how much to invest on an experimental stock. For the amount invested, there is 50% chance that it will become 2.5 times, and 50% chance that it will become zero. This design enables us to observe different levels of risk aversion for each subject, and allows ascertainment of the specificity of plasma OT levels towards trust without including a confound i.e. a possible connection between OT and risk attitude.

Assay Procedures
Blood samples for oxytocin assay were collected from the antecubital vein into pre-chilled 5 ml EDTA tubes with 250 KIU of apoprotinin, and refrigerated until processing. Plasma was isolated by centrifugation at 1800 g, 15 minutes, 4°C, and stored in aliquots at −70°C. Oxytocin immunoreactivity levels were quantified in duplicates using a commercial oxytocin ELISA kit (Enzo Life Sciences, NY, USA, formerly Assays Designs, MI, USA), as recommended in previous publications [11]. Thawed samples on ice were diluted 1∶2 times in assay buffer and assayed according to manufacturer's instructions. The oxytocin assay had a sensitivity of 11.7 pg/ml, and inter- and intra-assay coefficient of variations below 15%.
Currently there are differences in opinions surrounding the measurement of oxytocin and particularly concerning the requirement of sample extraction. The commercially available oxytocin EIA kit from Enzo Life (formerly Assay Designs), which has been validated by for linearity, cross reactivity, matrix effects, accuracy, precision and recovery [26], was used in the current study. The experience of some investigators suggests that extraction of the samples leads to significant loss of measureable oxytocin. Importantly, the oxytocin data from non-extracted samples makes biological sense as compared to those from extracted samples, which often gave rise to non-detectable levels of oxytocin. Szeto et al., 2011 confirmed these technical findings; as much as two-three fold of the authentic oxytocin was removed by extraction, and 5% of the extracted samples had non-detectable oxytocin levels [21]. Moreover, it would not be appropriate to measure extremely low levels of target analyte, viz., following extraction in the case of oxytocin, using commercial immunoassays that are insufficiently sensitive, which gives rise to erroneous results. We are also not keen to extract the samples using the solid-phase extraction method, as the procedure requires large volume of samples, and often gives rise to low recovery of analytes, high variability in results and incomplete removal of interferences. Hence, we chose not to extract. During our assay runs, we performed 1∶2 dilutions on the unextracted samples so that the measured oxytocin concentrations fall within the “measurable” portion of the standard curve. Pre-dilution of samples is also a common technical approach used to reduce assay interference due to sample matrix.
Taking into consideration the labile nature of oxytocin in biological matrix [27], we followed strict protocol during sample collection to limit its enzymatic breakdown. All blood samples were collected into pre-chilled EDTA tubes containing protease inhibitor. Processing of samples was performed at 4 deg C. During assay, thawed samples were kept on ice. In Szeto et al., human EDTA plasma/serum were obtained and stored at −80 deg C until assay. There was no mention of proper sample collection and hence, the stability of oxytocin in their plasma samples needs to be examined, which may offer one possible explanation of the degraded oxytocin products found in their samples.
The concentrations of oxytocin in unextracted blood samples are 100-fold more than in extracted samples [21]. Martin-Protean, a biotechnology company that specialises in protein analysis, reported oxytocin values to be higher (at levels of 1000 pg/ml) using novel isolation methods and mass spectrometry, and proposed a new model of explanation that incorporates oxytocin carrier protein neurophysin 1 [28]. In their website (http://martin-protean.com/), “Efforts to quantify oxytocin that capture 0.1% of the oxytocin present would not capture the complete oxytocin story and are likely to be dominated by non-biological variation in experimental procedure”.
Szeto et al. did perform stability tests using tritiated oxytocin added to plasma, and found that oxytocin is stable under different temperatures and after multiple freeze/thaw cycles. On the other hand, they concluded that plasma oxytocin has a short half life of 3–6 minutes and rapidly degrade into products that are more stable underscoring the careful handling procedure employed in our study. While Szeto et al. reported a lack of correlation between oxytocin levels in extracted versus non-extracted plasma samples (r = 0.09), another study reported high correlation (r = 0.89) [29].
Szeto et al. concluded that degradation products of oxytocin are likely to contribute to the measured levels of oxytocin in unextracted samples. However, it is doubtful that these are degraded products of oxytocin as they have molecular masses more than that of oxytocin. Based on their chromatography results, we believe there is a strong possibility that the assay is measuring immunoreactive oxytocin that comprises authentic oxytocin as well as oxytocin prohormones (OX-T) [30] or other forms of oxytocin in the unextracted samples. Recently, a novel form of oxytocin has been described in multiple species of squirrel monkeys, with a substitution of a leucine to a proline in amino acid position 8 [31]. The variant forms of oxytocin cannot be disregarded and appear to be biologically significant, as many other researchers have used the same kit to measure oxytocin in unextracted samples and found a myriad of associations with relevant physiological outcomes.

Statistics
To examine the relationship between plasma OT, Trust and Trustworthiness, we used regression models with robust standard error in STATA 11. In the first model, only plasma OT is included in the analysis to test a linear effect of OT on trust behaviors. Since U-shaped dose response curves are often observed for the action of peptide neuromodulators and steroid hormone actions in the brain [32], [33], we added a quadratic in a second model to test for nonlinear relationships between plasma OT and trust related behaviors. In addition, as pointed out by a recent paper [34], the data in Zak et al [22] might indeed suggest a revered U-shaped relationship between plasma oxytocin and trust. Here, a significantly positive quadratic term would indicate a U-shaped relationship while a significant negative quadratic term would indicate an inverse U-shaped relationship. To examine gender effects, we carried out a separate analysis for each gender. All the statistics reported here are two-tailed.