Fourteen adult psychopathic males (mean age: 43.14 ± 11.52 years, all right handed) with a history of violent and/or sexual offences participated in the study. All of them were long-time detained in one of two cooperating German maximum security forensic psychiatric institutions. Exclusion criteria were an age below 18 or over 55 years, an IQ below 80 or health problems. The participants were informed about the aim of the study and gave written informed consent. They were paid 20€ in agreement with the forensic institutions. All subjects were scored using Hare's Psychopathy Checklist Revised (PCL-R) (Hare, 2003), resulting in a mean score of 30.14 ± 2.77 (range = 25–34). Nine out of fourteen participants reached the cut-off score of 30, which is commonly used in American studies to classify psychopaths. However, all participants exceeded the psychopathy score of German and European norms (Cooke et al., 2004). Regarding the two main factors of the PCL-R, the mean scores were 11.79 ± 2.51 for factor 1 and 15.86 ± 1.56 for factor 2. According to the four facet model the mean scores were: Interpersonal facet: 5.50 ± 1.65; affective facet: 6.29 ± 1.35; lifestyle facet 7.00 ± 1.35; antisocial facet 15.86 ± 1.56. All participants met the full criteria for a diagnosis of antisocial personality disorders in line with the DSM-IV criteria (Apa, 2000). The study was approved by the Ethics Committee of the Medical Faculty of the University of Tübingen. The classical conditioning experiment presented here was conducted in the context of a comprehensive study on criminal psychopaths.

A classical delayed fear conditioning paradigm was applied, using a modified version of the design of Birbaumer et al. (2005). Two neutral grey-scale male faces were used as conditioned stimuli (Schneider et al., 1994). With pseudo-random assignment between participants, one of the faces was paired with the unconditioned stimulus (CS+), while the other face was never followed by the unconditioned stimulus (CS−). The (US) consisted of an electric shock and was administered using a Digitimer DS5 Isolated Bipolar Constant Current Stimulator (Digitimer Ltd., United Kingdom). The two electrodes were attached over the distal and proximal phalanx of the right thumb. The intensity of the electric shock was individually adjusted in a pre-experiment, immediately before the actual experiment, at a level where the participants estimated the electric shock as unpleasant (“8” on a visual analog scale in which “0” indicates not at all unpleasant and “10”, extremely unpleasant). The faces were presented for 5 s and the US was administered after 4 s of the picture presentation and lasted 500 ms.

A 50% partial reinforcement schedule was chosen, indicating that only 50% of the CS+ were followed by US. The conditioning procedure consisted of a habituation phase (6 CS+, 6 CS−, 4 US), an acquisition phase (32 CS+, 32 CS−) separated in an early and late acquisition block, and finally an extinction phase (16 CS+, 16 CS−). During habituation and extinction, the Inter-Trial-Interval (ITI) varied between 4.5 and 7.5 s, during acquisition between 7.5 and 12 s.

Subjective ratings of emotional valence and arousal were collected on a 9-point scale using the Self Assessment Manekin (Bradley and Lang, 1994) four times (after habituation, early acquisition, late acquisition and extinction). Based on the inherent implicit learning mechanism in classical conditioning it was not intended to inform the subjects about the CS-UCS contingency. The only information that was provided was that male faces will be shown and electric shocks will be applied from time to time. However, the contingency of US and CS was assessed after both acquisition phases and after the extinction phase using a visual analogue scale (“How likely is it that an electric stimulus will follow this face?” (ranging from “1,” indicating no association between CS and US, to “9,” indicating an absolute certainty that US follows CS).

Physiological data were acquired using a Theraprax Neurofeedbacksystem (NeuroConn GmbH, Ilmenau, Germany). Electroencephalography (EEG), electrooculography (EOG), and skin conductance response (SCR) were recorded using a sampling rate of 128 Hz and 40 Hz high cut-off-filtering. EEG activity was measured using four recording electrodes, located at Fz, FCz, Cz, and Pz according to the 10–20 system. Left mastoid was used as the reference and right mastoid as ground. SCR activity was recorded from the intermediate phalanx of the index and ring finger of the non-dominant hand with strap electrodes.

Skin conductance data were analyzed using the Matlab-based software Ledalab 3.4 (Benedek and Kaernbach, 2010). The recordings were decomposed into their tonic and phasic components which resulted in phasic activity timelines with zero baselines. SCR was then extracted for each single trial using the integrated phasic SCR between 1 and 4 s after picture onset and averaged over paired and unpaired trials in the different conditioning phases. This novel approach allows an unbiased estimation of sympathetic fear-related activity compared to the conventional baseline to peak computation using minimum amplitude criteria for classifying SCR.

The EEG data were processed using BrainVision Analyzer Professional 2.01 (BrainProducts GmbH, Gilching, Germany). A notch filter at 50 Hz was applied. Thereafter the signals were filtered using a 0.1 Hz highpass and a 15 Hz lowpass filter. Ocular artifacts were adjusted using an eye-blink artifact correction method (Gratton et al., 1983). The data was then segmented in epochs of 5.5 s duration (−0.5 to 5 s relative to the onset of the CS). On the segmented data, a time interval of 300 ms was used to detect and reject remaining artifacts exhibiting (a) gradient changes more than 15 μV/ms, (b) voltage differences of more than 100 μV, or c) signal amplitudes over ±80 μV. Baseline correction was performed using a 200 ms interval before trial onset. ERPs were extracted using the following parameters: For N100, amplitude and latency of the highest voltage peak between 100 and 220 ms was registered, while 300–800 ms was used for P300. CNV areas under the curve were calculated for the time intervals 800–1500 ms (iCNV) as well as 3500–4000 ms (tCNV).

The self-report and SCR data were analyzed by means of repeated-measures analyses of variance (ANOVA) with phase (habituation, acquisition and extinction) and CS-type (CS+ vs. CS−) as within-group factors. Contingency ratings were analyzed using the acquisition phase (early vs. late) and the CS-type as within subject factor. For each EEG parameter separate repeated measures ANOVAs were performed using electrode position (FCz, Fz, Cz, and Pz), and CS-type as within-subject factors during habituation and extinction. During acquisition, the first and second half of the phase (early and late acquisition block) were selected as an additional within-subject factor. Greenhouse-Geisser correction was applied if the sphericity assumption was violated. In order to improve signal-to-noise ratio in subsequent analysis, EEG amplitudes during early and late acquisition phase were averaged over corresponding CS type. Differences between the CS types during early and late acquisition as well as both acquisition blocks combined were calculated. The differential values were correlated with the individual total PCL-R score as well as with its two factors and underlying facets (Pearson's bivariate correlation). Apart from simple correlations a partial correlation approach was conducted using one facet as independent variable and the respective other facets as control variables. In addition, a forward stepwise regression analysis using the four facets as independent variables and the differential (CS+ vs. CS−) subjective, peripheral and EEG parameters as dependent variables. For the inclusion we used F-probability of 0.05 and for the exclusion a probability of 0.10. Furthermore, a multiple regression analysis with all four facets at once was computed to determine the account of variance. Because of problems during SCR recordings, data were available only for 11 participants. One participant was excluded from the EEG analyses due to artifacts during data acquisition.

The correlation analyses revealed a close relationship between the differential SCR responses and N100 and P300 amplitudes during early and late acquisition. Reduced anticipatory fear responses were accompanied with degraded early attention to the CS+ compared to the CS−, as reflected in the N100 waveforms. Late attentional processes, as reflected in the P300 potential showed the opposite effect. In a similar vein, a smaller conditioned SCR was correlated with a more negative rating of CS+ faces (see Table 1).

Repeated measurement ANOVAs with the factors phase (habituation, early acquisition, late acquisition and extinction) and CS type revealed no significant main effects or interactions for valence and arousal as well as for SCRs during fear conditioning. However, a significant difference in contingency ratings between CS types [F_{(1, 13)} = 45.66, p < 0.0001] was found indicating, that processing of the paired and unpaired stimuli was intact on a cognitive level. The correlation analysis revealed that during early acquisition high total PCL-R scores were negatively associated with differential SCR amplitudes (r =−0.757, p = 0.007). Regarding the different facets, the effect was most prominently induced by the affective facet of psychopathy (r = −0.663, p = 0.036; see Figure 1, right). Partial correlation analysis using the affective facet as dependent variable and the other facets as control variable confirmed the observed association, although it became not significant due to the reduced degrees of freedom relative to the number of control variables (r = −0.604, p = 0.112). A stepwise regression analysis revealed that the reduced differential SCR responses during the early acquisition phase were solely predicted by the affective facet [R² = 0.439, [F_{(1, 9)} = 7.04, p = 0.026]. The other facets did not explain additional variance. Concerning the differential valence ratings, we found the opposite effect. Participants scoring high on the affective facet rated the paired stimulus more negative than the unpaired stimulus in the early acquisition phase (r = −0.589, p = 0.027; see Figure 1 left). This effect was confirmed using partial correlation analysis (r = 0.608, p = 0.050). During late acquisition phase, no significant correlations between subjective measures, SCR and PCL-R scores were found. Contingency ratings revealed that the CS-UCS pairing was correctly identified during the early [t₍₁₃₎ = 4.63, p < 0.001] and late [t₍₁₃₎ = 7.43, p < 0.001] acquisition phase. All zero-order correlations and partial correlations for the subjective and SCR measures are depicted in Table 2.

During habituation and extinction, no significant effects were found. During acquisition, a significant electrode effect [F_{(3, 36)} = 6.545, p = 0.004] with increased negativity at fronto-central electrode position was observed. However, neither CS type nor acquisition blocks yielded statistically significant effects. The correlation analyses revealed a positive relationship between the CS type related iCNV differentiation (CS+ minus CS−) with the antisocial facet (Fz: r = 0.733, p = 0.004; FCz: r = 0.716, p = 0.006; Cz: r = 0.623, p = 0.023) as well as with the original factor 2 of the PCL-R (Fz: r = 0.583), and in particular during the early acquisition phase (Fz: r = 0.747, p = 0.003). High scores on the antisocial facet were associated with smaller negative and even positive shifts of brain activity in response to CS+ compared to the CS−. Moreover, the interpersonal facet of the PCL-R covaried negatively with the CS+/CS− differentiation (Fz: r = −0.561, p = 0.036; Pz: r = 0.602, p = 0.029). Thus, participants with high interpersonal deficiencies had larger negative shifts in their brain activity in response to paired (CS+) relative to unpaired (CS−) stimuli (see Figure 4). A partial correlation analysis revealed that the antisocial facet showed the strongest association (r = 0.698), just as the stepwise regression analysis yielded [R² = 0.537; F_{(1, 12)} = 12.77, p = 0.004]. During extinction, a negative correlation between scores in the antisocial facet and CS+/CS− iCNV differentiation was observed (r = −0.588, p = 0.035).

ANOVA revealed no significant effects during habituation or extinction. During the acquisition phase, there was only a tendency of an electrode effect [F_{(3, 36)} = 2.64, p = 0.064] with pronounced tCNV at Cz and the smallest effect at Pz, but no CS type or blocks effect. Resembling the iCNV results, the correlation analyses showed that the CS+/CS− differentiation in the terminal CNV was raised by the antisocial and interpersonal psychopathy facet. While high scores of the antisocial facet correlated positively with CS+/CS− tCNV differentation at fronto-central sites (Fz: r = 0.645, p = 0.017; FCz: r = 0.654, p = 0.015, Cz: r = 0.595, p = 0.032), the interpersonal facet revealed a negative correlation centrally (Cz: r = 0.610, p = 0.027). Thus, high antisocial scores were again associated with smaller negative or even positive shifts in brain activity in CS+ trials, while an opposite pattern was found in participants with pronounced interpersonal deficits. A step-wise regression analysis showed that a model including the interpersonal facet only, explained most variance [R² = 0.373, F_{(1, 12)} = 6.53, p = 0.027]. During extinction, a negative correlation between scores in the lifestyle facet and the tCNV difference between CS+ and CS− was observed (r = −0.566, p = 0.044).

The present study aimed to explore associations of psychopathy and the different facets in Hare's psychopathy construct with subjective, peripheral-physiological and cortical measurements in a classical fear conditioning paradigm. On the group level, without accounting for the individual PCL-R scores, we did not find a conditioning effect in the subjective and peripheral measures in these highly psychopathic criminals. The results of our study corroborate previous findings, demonstrating a psychopathy related deficiency in developing a conditioned fear response to aversive or fearful stimuli (Flor et al., 2002; Veit et al., 2002; Birbaumer et al., 2005; Rothemund et al., 2012). In addition, the correlation analyses indicated that the deficiency in fear conditioning is linearly modulated by total PCL-R scores. Participants with extremely high psychopathy scores showed weaker or absent conditioned electrodermal fear responses compared to lower scorers. A closer inspection revealed that this effect was most prominently modulated by the affective facet. The opposed pattern was found using the differences in valence ratings between CS+ and CS− as a subjective measure of successful conditioning. Interestingly, the correlation between the differential valence ratings and SCR responses revealed that subjects with profound fear deficit rated the CS+ faces more negative than the CS−. We postulate that participants scoring high on the affective facet either tried to mimic normal emotional behavior by responding more negatively to the paired stimuli or that they are indeed perfectly able to evaluate the expected “cognitive” dimension of fear. This matches the observation that core psychopaths are masters of deception and/or are cognitively quite aware of the contingency between face stimuli and painful electric shock. Recently, Lopez et al. (2013) assessed self-reported psychopathy (PPI-R, The Psychopathic Inventory Revised), (Lilienfeld and Widows, 2005) in a student sample and showed that high scores in the “fearless dominance” subscale, but not in the “impulsive antisociality” subscale, were associated with deficient fear conditioning. This is similar to what we found in highly criminal psychopaths. The diminished emotional responsiveness is a key finding in psychopathy research and it has been shown that the reduced startle response during presentation of negative emotions is closely related to factor 1 of the PCL-R (Patrick et al., 1993; Patrick, 1994; Vaidyanathan et al., 2009b). Those findings fit perfectly to the elaborated theoretical framework of Patrick et al. (2007), which describes a bifactor conceptualization of psychopathic syndromes with different underlying etiological mechanisms. Nonetheless, it is not yet clear whether the affective or the interpersonal facet contributes to the fear deficit or to which extent they might contribute. Based on the anticipatory skin conductance responses, our data indicate that in a fear conditioning paradigm, the affective facet modulates the deficit in fear reactivity.

Regarding the EEG measures, we found decreased N100 amplitudes to the CS+ compared to the CS− during the early acquisition phase in participants with pronounced affective deficits. The same association was found in participants with distinct interpersonal deficits, when regarding the early and late acquisition phase combined. This result might reflect a different early attention status, presumably arising from the conditioning procedure in psychopaths scoring high on the superordinate factor 1 of the PCL-R. Our findings are in contrary to Flor et al. (2002) who revealed increased N100 amplitudes in the psychopaths to CS+ compared to CS− trials in the early conditioning phase, but in line with Rothemund et al. (2012) who showed overall lower N100 amplitudes in psychopaths compared to healthy controls. In context of the response modulation theory, Newman et al. (2010) proposed attentional abnormalities in psychopathy as an alternative fear deficit explanation. Furthermore, in a recent study deficient fear responses in highly psychopathic individuals were only found when the attention was shifted to irrelevant information at early stages prior to the onset of fear-relevant stimuli (Baskin-Sommers et al., 2011). We could show that the phase specific CS type differentiation in the N100 amplitude was directly related to the differential SCR responses in the early and late acquisition phase. Higher N100 amplitudes to the CS+ compared to the CS− were accompanied with enhanced electrodermal reactivity to the CS+, supporting the importance of early sensory processing of conditioned stimuli in fear conditioning (Miskovic and Keil, 2012). Although a direct comparison between our findings and the results of Newman et al. (2010) and Baskin-Sommers et al. (2011) is not possible, due to critical differences ranging from the experimental paradigm (implicit vs. explicit learning) up to the data collection (anticipatory SCR's and ERP's vs. fear potentiated startle response), our findings highlight the influence of early attentional processes in fear related learning that might be different in participants with affective/interpersonal deficits.

Regarding the P300 component, correlational analyses revealed larger CS type differentiation in the P300 amplitudes in high PCL-R scorers during the early acquisition phase. Considering the late acquisition phase, however a negative correlation was found with the PCL-R total scores as well as with factor 1 and the P300 responses. Flor et al. (2002) found a comparable positive CS+/− differentiation at frontal leads only in the psychopathic group during the early acquisition period, while Rothemund et al. (2012) showed a CS type differentiation in both psychopathic and non-psychopathic individuals. Our findings are consistent with the view that high psychopathic individuals exhibit intact attentional processes in particular during the early conditioning phase, when the meaning of the situation must be conceived. With respect to the psychopathy facets we showed that the interpersonal facet was negatively correlated with the CS+/− evoked P300 responses throughout the acquisition, while the antisocial facet modulated the P300 responses in the opposite way. It has to be emphasized that the P300 does not reveal a uniform pattern in psychopathy and most studies reporting reduced P300 amplitudes in psychopathy (Kiehl et al., 1999, 2000, 2006; Gao and Raine, 2009) used paradigms that influence P300 pattern selectively. In a similar vein, Patrick et al. (2006) reported a strong association between the reduction in P300 amplitude and externalizing dimensions such as antisocial behavior, pathological gambling, drug abuse and disinhibition in a visual oddball paradigm. Accordingly, this approach explains the decreased ERPs in psychopaths, scoring high on the interpersonal but not on the antisocial facet. However, it is important to mention that the P300 potential is not a measure for successful or unsuccessful Pavlovian fear conditioning, instead it measures selective attention and expectancy, modulating emotional learning (Verleger et al., 1994). Regardless of psychopathy scores, we demonstrated that enhanced P300 to the conditioned stimuli was associated with diminished electrodermal fear responses, suggesting cognitive top-down modulation in affective learning (Olofsson et al., 2008).

Concerning the conditioned CNV responses related to expectancy, orientation (iCNV) and preparation (tCNV), we found increased cortical negativity in the CS+ compared to the CS− condition in frontal and central sites for participants scoring high on the interpersonal facet. On the contrary, high scorers on the antisocial facet showed increased negativity in CS− compared to CS+ trials. Augmented iCNV was observed in some studies investigating psychopaths (Forth and Hare, 1989; Flor et al., 2002), while others reported no differences (Raine and Venables, 1987) or even reduced CNV responses (Walter et al., 1964; McCalloum, 1973). Rockstroh et al. (1982) emphasizing cognitive rather than emotional aspects as main sources of the CNV. Therefore, the enhanced iCNV during CS+ compared to CS− trials in participants with pronounced interpersonal deficits reflected heightened attention or interest in the conditioned face stimuli, while the antisocial facet showed an inverse effect. With respect to the tCNV, the increased tCNV differentiation in subjects scoring high on the interpersonal facet might be interpreted in the context of preparedness, cognitive appraisal and contingency evaluation and can be considered as a further proof of superior cognitive processing without affecting emotional fear learning.

In line with previous studies, we found deficient fear conditioning in incarcerated, highly psychopathic offenders as indicated by diminished SCR differentiation between types of conditioned stimuli. Stepwise regression analysis revealed that only the affective facet is responsible for the low fear responses. Socialization is to a high degree based on the learning of stimulus-response (classical conditioning) and stimulus-reinforcement (instrumental conditioning) associations to adequately adapt behavior. A weak SCR differentiation during fear conditioning can therefore be an indication of maladaptive learning and failed socialization. Indicated by the event related potentials, we found a rather inferior early (N100) attentional but superior late (P300) attentional processing in subjects scoring high on the affective facet. Regarding the CNV responses, supposedly reflecting cognitive processing (Rockstroh et al., 1982), the interpersonal facet was associated with stronger CNV responses to the CS+ compared to the CS−, while the antisocial facet revealed the opposite effect. Such a “diaschisis” between the emotional and cognitive processing was often proposed in descriptive (Cleckley, 1941/1982), legal (Sommer et al., 2006) and artistic (Musil, 1930-1943) accounts and explanations of psychopathic criminals. This discrepancy between emotional and cognitive processing is not only mirrored by the reported discrepancy between the cognitive and emotional awareness of the aversive stimulus, but also obvious in empathy tasks with psychopathic individuals. Psychopaths demonstrate a complete failure to experience emotional empathy (Vollm et al., 2006) and this dysfunction was particularly evident in psychopaths with affective/interpersonal deficits (Decety et al., 2013). The psychopathic lack of emotional empathy seems to relate to disrupted affective processing and production. On the other hand, psychopaths are able to complete theory of mind tasks that require perspective taking without much difficulty (Richell et al., 2003).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.