The Underestimated Significance of Conditioning in Placebo Hypoalgesia and Nocebo Hyperalgesia

Placebo and nocebo effects are intriguing phenomena in pain perception with important implications for clinical research and practice because they can alleviate or increase pain. According to current theoretical accounts, these effects can be shaped by verbal suggestions, social observational learning, and classical conditioning and are necessarily mediated by explicit expectation. In this review, we focus on the contribution of conditioning in the induction of placebo hypoalgesia and nocebo hyperalgesia and present accumulating evidence that conditioning independent from explicit expectation can cause these effects. Especially studies using subliminal stimulus presentation and implicit conditioning (i.e., without contingency awareness) that bypass the development of explicit expectation suggest that conditioning without explicit expectation can lead to placebo and nocebo effects in pain perception. Because only few studies have investigated clinical samples, the picture seems less clear when it comes to patient populations with chronic pain. However, conditioning appears to be a promising means to optimize treatment. In order to get a better insight into the mechanisms of placebo and nocebo effects in pain and the possible benefits of conditioning compared to explicit expectation, future studies should carefully distinguish both methods of induction.


Introduction
Placebo and nocebo e ects are prevalent topics in current research, especially in the domain of pain, where they can be investigated comparatively easily and serve as a model for other systems (e.g., immune, motor, and respiratory systems [1]). While placebo hypoalgesia refers to decreased pain sensitivity due to an inert treatment (e.g., sham procedure and inert substance), its counterpart nocebo hyperalgesia is de ned as increased pain sensitivity attributable to an inert treatment [2]. Due to their capacity to improve or worsen symptoms and well-being, placebo and nocebo e ects are highly relevant not only in research but also in clinical practice. ey contribute to most therapeutic e ects [3], occur regularly in doctor-patient interactions [4], and are assumed to play a role in the development and maintenance of chronic pain and other diseases [5,6]. Placebo and nocebo e ects also signi cantly in uence the outcome of randomized placebo-controlled clinical trials, for example, leading to an underestimation of clinical e ects of novel drugs [7][8][9]. Although this clinical relevance and research on placebo and nocebo e ects have prospered during the past decade, many issues remain unresolved or poorly understood. One of these issues that deserve our attention concerns the causing mechanisms of placebo and nocebo e ects and the signi cance of conditioning in this realm. Only if we understand how these e ects emerge, we can systematically take advantage of placebo and avoid nocebo e ects, especially in the clinical context.
After giving a brief overview over current evidence and models on the development of placebo hypoalgesia and nocebo hyperalgesia including conditioning and expectation approaches, we will turn to studies that stress the signicance of conditioning and demonstrate that conditioning e ects might have been misunderstood and/or underestimated. We will highlight the need to better understand conditioned placebo and nocebo e ects and propose experimental designs that allow a conclusive investigation of the speci c mechanisms at work.

Development of Placebo and Nocebo Effects
Placebo and nocebo e ects can be induced by di erent means. Empirical evidence shows that verbal suggestion [10], classical conditioning [11][12][13], and observational learning [14,15] can lead to placebo hypoalgesia and nocebo hyperalgesia. Verbal suggestions in terms of written or spoken statements on the alleged e ect of the placebo or nocebo are thought to induce an explicit expectation that directly mediates the e ect [16,17]. In contrast, direct experience of a pain-increasing or -relieving e ect is essential in conditioning of placebo hypoalgesia and nocebo hyperalgesia [18]. Here, a placebo/nocebo serves as conditioned stimulus and is repeatedly coupled to a pain stimulus (unconditioned stimulus), in which the intensity is reduced or increased without the person's knowledge compared to before placebo/nocebo administration or to a simultaneous stimulation without placebo/nocebo. Subsequently, the conditioned e ect is tested by retracting the surreptitious change in stimulation intensity while still administering the placebo/nocebo. Observational learning of placebo and nocebo e ects has been investigated only recently. It is usually carried out by showing videos of or live models who are demonstrating reduced or increased pain sensations upon administration of a placebo/nocebo [15,[19][20][21]. It is assumed that the observation serves as an unconditioned stimulus [22].
Typically, conditioning and verbal suggestions are both implemented in experimental studies to increase placebo/nocebo e ects in pain [23][24][25][26]. Accordingly, metaanalyses come to the conclusion that the combination of conditioning and verbal suggestion leads to larger e ects than verbal suggestion alone [27,28]. When inducing placebo hypoalgesia or nocebo hyperalgesia by verbal suggestions alone, results are inconsistent, sometimes resulting only in weak [24] or even no e ects [25,29,30] and at other times leading to robust pain reduction [10,31] or increase [10,32]. Some studies indicate that learning is less important in nocebo hyperalgesia compared to placebo hypoalgesia, but direct comparisons are rare [24,33]. Interestingly, while placebo e ects induced by expectation can be antagonized by the opioid antagonist naloxone, conditioned placebo e ects can be antagonized by naloxone when the pharmacological conditioning had been performed with an opioid and by cannabinoid receptor antagonists when it had been performed with ketorolac, a nonsteroidal anti-in ammatory drug [34,35]. ese ndings highlight that placebo and nocebo e ects in pain di er depending on whether they were induced by verbal suggestion or conditioning.
In a series of three studies, Benedetti et al. [10] showed that (a) placebo and nocebo e ects in hormone secretion (growth hormone and cortisol) were a ected by pharmacological conditioning, but not by verbal suggestions, (b) placebo and nocebo e ects in pain were induced by pharmacological conditioning and by verbal suggestions, but pharmacologically conditioned placebo hypoalgesia was overridden by opposing verbal suggestions, and (c) motor performance in Parkinsonian patients depended on verbal suggestion after repeated deactivation of implanted stimulating electrodes. Based on these observations, the authors developed an in uential model stating that conditioning can directly lead to placebo and nocebo e ects in unconscious processes (e.g., hormone secretion) and that expectation can lead to placebo and nocebo e ects in conscious processes (e.g., pain and motor performance) and that e ects of conditioning on conscious processes are necessarily mediated by explicit expectation. Furthermore, it is assumed that expectation cannot directly a ect unconscious processes [10]. Although these conclusions t to the results of the presented studies, we assert that the model underestimates conditioned e ects, as outlined below.
Previous research on the mechanisms of placebo and nocebo e ects might be biased due to a number of reasons: (1) studies that investigate the e ects of conditioning without verbal suggestion are scarce [30,36,37]. In order to estimate the contribution and investigate the mechanisms of conditioning to a given e ect, it is likely not as simple as to subtract the e ect size from a verbal suggestion group from that of a verbal suggestion plus conditioning group. It could well be that conditioning and verbal suggestion do not interact in an additive manner, although this seems to be assumed in most studies. (2) When investigating the e ects of a conditioning procedure without verbal suggestions, oftentimes a medically connoted placebo or nocebo (e.g., ointment, pill, and sham acupuncture) is used [13,29,38,39]. However, these placebos and nocebos most probably induce expectations from the outset independent of the experimental manipulation. Participants might have been preconditioned by previous experiences with similar medical devices and procedures. As a consequence, unintended additional expectations will develop that complicate disentangling expectation-and conditioning-induced e ects. Evidence supporting this assumption comes from a study by Montgomery and Kirsch [39], in which a small placebo e ect emerged after applying an inert tincture without giving a verbal suggestion and before conditioning.
(3) Another pitfall of many studies testing conditioned effects is the implementation of the conditioning procedure itself. Due to credibility, a medically connoted placebo or nocebo, like an ointment, will only be applied once in the beginning of the study, for example, on the right arm, and then tested against a spot without ointment, for example, on the left arm [12,13,[39][40][41][42]. In other cases, the placebo/nocebo is applied only very few times [11,43]. Proper conditioning, however, relies on repeated pairings of the conditioned stimulus (i.e., the placebo or nocebo) and the unconditioned stimulus (i.e., reduced or increased pain) (e.g., [44]). It is known from the conditioning literature that conditioned e ects are stronger when more pairings of CS and US are applied [45]. By applying a placebo/nocebo only once, it cannot be ensured that the pairing is processed as intended. To summarize, most previous studies do not allow disentangling of placebo/nocebo e ects induced by conditioning and/or explicit expectations or do not use adequate conditioning procedures to test for conditioned e ects so that conditioned placebo and nocebo e ects can hardly be evaluated.

Mediation by Expectation Hypothesis
An important aspect of the above-cited model [10] concerns the necessary mediation of conditioning e ects in pain via explicit expectation. Colloca and Miller [22] developed a learning perspective on placebo responses by applying Peirce's theory of signs and re ning Benedetti's model [10]. In addition to Benedetti's model, they state that signs in the form of indices (i.e., conditioned stimuli), symbols (i.e., communication), and icons (i.e., observations) are detected and processed, resulting in the formation of expectations, which thereby contribute to placebo and nocebo responses. In line with Benedetti's model, bodily functions that are consciously accessible, as pain relief, are mediated by expectation, but "an event that cannot be experienced and perceived by human cognition (e.g., growth factor secretion) appears not to be in uenced by self-cognition" (p. 1865). Although the authors endorse the possibility of conditioning without awareness, and accordingly unconscious expectations, they argue that it is circumstantial in creatures with higher phylogenetic level. In the following, we will show that conditioned placebo hypoalgesia and nocebo hyperalgesia are not necessarily mediated by explicit expectations, emphasizing the assumption that the conditioned e ect can exist independent of explicit expectation.
Many studies report correlations between pain and expected pain ratings [17,39,46,47], sometimes on a trialby-trial basis [26], after conducting a conditioning procedure. Using a mediation analysis, it has been shown that conditioning is highly related to expectancy and expectancy predicts the placebo e ect, while the direct e ect of conditioning on the placebo e ect is no longer signi cant [38].
is result suggests that conditioning is mediated by expectation. However, the robustness of this outcome seems somewhat uncertain, as a bias-corrected bootstrap approach was not signi cant. A study of Montgomery and Kirsch [39] can serve as another example for mediation by expectation. ey compared a conditioned group (uninformed pairing) to a group with informed pairing (i.e., participants underwent a conditioning procedure but were informed about the intensity reduction on placebo trials), and a control group that was not conditioned. e strongest placebo e ect appeared in the uninformed pairing group, but when including expectancy as a covariate, group di erences disappeared and expectancy was signi cantly related to placebo e ects. Furthermore, some studies report that conditioned e ects can be blocked by opposite verbal suggestions [10,39]. However, this is not always the case, as there are studies showing conditioned e ects despite opposite verbal suggestions [11,12] or expectancy [48]. Also, conditioned e ects have been shown to be mediated by expectancy only partially [49], and expectation ratings do not necessarily predict conditioned e ects [37,50] or correspond to pain ratings [29]. In a study of De Pascalis et al. [41], two placebo creams were administered with supposedly "strong" and "weaker" dosage. Subsequently, only after application of the "strong placebo," the intensity of electric pain stimuli was surreptitiously reduced. Although expected pain level varied according to the manipulation (i.e., strong placebo led to higher expectation of pain relief compared to weak placebo), pain ratings did no di er after strong and weak placebo, highlighting the dissonance between expectations and measured placebo hypoalgesia.
Studies on open-label placebo administration give another important insight into the mediating role of expectation in placebo hypoalgesia. Typically, open-label placebo studies try to boost expectation of a positive e ect despite openly administering an inert substance [51,52]. However, a highly intriguing study in healthy participants showed that conditioned placebo hypoalgesia persisted after revealing the inert nature of a placebo intervention (cream) independent of the participants' expectation. Other than in the previously mentioned open-label placebo studies, participants in this study were not encouraged to believe in a positive e ect from the placebo cream. Although participants no longer expected a hypoalgesic e ect, the placebo e ect remained [50]. In a recent systematic review and meta-analysis on open-label placebo studies, it was hypothesized that the mechanism driving such e ects is classical conditioning [53].
In an editorial, Wager [54] begs interesting questions on the role of expectation as a mediator of placebo e ects. He argues that, despite the predictive value of expectation for placebo e ects, a causal relationship is not necessarily implied. Besides a direct in uence of expectation on pain experience, he proposes three alternative explanations for the observed e ects: expectancy could a ect reported pain ratings (1) directly, (2) via a kind of social contract that is initiated after giving an expectancy rating, and (3) as a consequence of the third variable that a ects both expectancy and pain rating, like demand characteristics, personality traits, treatment history, or situational factors. Supporting these ideas, a study of de Jong et al. [29] can be consulted, in which three di erent groups were investigated: the experimental group underwent a conditioning procedure with additional verbal suggestions, control group 1 underwent the same conditioning procedure but was told that stimulus intensities would be halved during trials, in which the placebo was applied, and that an inert substance was used, and control group 2 received verbal suggestions only. Despite similar di erential expectations of the experimental group and control group 2, only the experimental group showed placebo hypoalgesia. Although no placebo e ect was found in control group 1, pain ratings did not di er between the experimental group and control group 1, suggesting a relative independence from verbally induced expectations. Yet, expectation was found to correlate with the placebo hypoalgesia, however, irrespective of the experimental manipulation. e authors speculate that this correlation re ects an interaction between individual traits and general characteristics of the placebo used, thus possibly serving as an example for Wager's third case.
In conclusion, mediation by expectation of conditioned placebo hypoalgesia and nocebo hyperalgesia can occur but Pain Research and Management 3 in many cases does not. We therefore now turn to studies that demonstrate the signi cance of conditioning e ects independent of explicit expectation.

Evidence for the Significance of Classical Conditioning
Evidence of di erent research areas highlights the signicance of conditioning of placebo hypoalgesia and nocebo hyperalgesia that is independent of explicit expectation. It has been shown that besides humans, animals, such as rodents, can develop conditioned placebo hypoalgesia and nocebo hyperalgesia [55][56][57]. Furthermore, a limited number of studies implemented placebo or nocebo conditioning procedures without additional verbal suggestions and used meaningless cues as conditioned stimuli (e.g., red and green lights), instead of medically connoted substances or procedures, bypassing some of the abovementioned limitations of previous studies. Results are inconsistent: while recent studies found placebo hypoalgesia and nocebo hyperalgesia after conditioning without additional verbal suggestions that were not predicted by expectancy ratings [37], or without contingency awareness [58], other studies reported no placebo hypoalgesia or nocebo hyperalgesia after conditioning without additional verbal suggestions [30,36]. A line of research that strongly supports the existence of conditioned placebo and nocebo e ects in pain that are independent of explicit expectation uses subliminal cues and implicit conditioning, that is, conditioning without contingency awareness. Some studies showed that placebo hypoalgesia and nocebo hyperalgesia that had been conditioned with supraliminally presented cues (i.e., explicit conditioning) can be activated by subliminally presented cues [21,59]. Jensen et al. [60], for instance, conditioned healthy participants with the display of faces and activated placebo hypoalgesia and nocebo hyperalgesia by supraliminal as well as subliminal face presentation [60]. ese results were replicated in a functional magnetic resonance imaging (fMRI) study of the same working group [61].
One attempt to implicitly condition placebo hypoalgesia by using a tactile cue (direction in which a placebo cream was applied to the skin) was not successful in inducing placebo hypoalgesia [43], but a recent study indicated that contingency awareness is not necessary to induce a nocebo e ect in heat-pain perception [58]. So far, there is only one study that used a subliminal stimulus presentation also during the acquisition phase of a conditioning experiment in order to rule out explicit expectation in conditioning [44]. e authors assigned healthy participants to one of four groups, each including an acquisition phase and a test phase and varying subliminal and supraliminal stimulus presentations. Overall, F tests indicated that there was no di erence in placebo hypoalgesia and nocebo hyperalgesia between the di erent types (i.e., subliminal/supraliminal) of the acquisition or test phase, strongly suggesting the presence of implicitly conditioned e ects.
In order to gain more insight into the role of conditioned e ects on placebo and nocebo e ects in pain, research on implicit conditioning should be expanded and diversi ed, as previous studies mostly did not directly test for contingency awareness and typically focused on subliminal conditioning, which has some pitfalls (e.g., intra-and interindividually varying threshold for subliminal stimulus presentation). Although accumulating evidence points to an important role of conditioning of placebo and nocebo e ects in pain independent of explicit expectations, most current theoretical accounts do not yet incorporate these aspects su ciently [10,22]. One point of criticism regarding prevalent models concerns the classi cation of physiological processes into being either conscious or unconscious. Evidence suggests that it is conceivable that most, if not all, bodily functions, including perception and behavior, have "unconscious" (i.e., outside a person's awareness; cf., e.g., blindsight and implicit operant conditioning [62][63][64]) and "conscious" (i.e., verbally represented) portions. Pain, especially, is known as a multidimensional phenomenon that incorporates sensory, a ective, behavioral, and physiological levels that can all be accessed on aware and unaware levels [63][64][65][66]. An exception provides a theoretical framework of Haug [67], which, however, has not received much attention. He proposes that placebo/nocebo e ects are mediated by socalled aliefs, which are subdoxastic states, that is, cognitive states "with consciously inaccessible content which is inferentially isolated from the subject's large network of beliefs (and which may directly cause behavior)" (p. 690). Aliefs can be consciously or unconsciously activated in humans and nonhumans by the internal or external environment. ey are associative, automatic, arational, a ect-laden, and action generating and might be a better candidate for a common nal path than expectation in Colloca and Miller's [22] model because they can explain placebo and nocebo e ects that have been induced by verbal suggestion as well as (implicit) conditioning.

Clinical Context
While the majority of studies on placebo hypoalgesia and nocebo hyperalgesia examined healthy participants, only few investigated pain patients and even less focused on clinical pain [68]. Nonetheless, the available studies give some insights into the mechanisms of placebo and nocebo e ects in a clinical pain context.
Verbal suggestion of pain relief seems to be highly effective for acute procedural (large e ect with Hedges' g � 1.03) as well as chronic pain (small e ect with Hedges' g � 0.25 [68,69]) even after open-label placebo administration in patients with irritable bowel syndrome (IBS [51]) and chronic low back pain [52]. Furthermore, large placebo e ects (comparable to the e ect of the local anesthetic Lidocain) have been induced by using verbal suggestions and application of a placebo (lubricant) during the induction of experimental pain (rectal distention) in patients with IBS [70,71]. While explicit expectations accounted for large amounts of the variance in experimental visceral pain during placebo and Lidocain administration, it was not predictive for a cutaneous pain model despite placebo hypoalgesia being present [70]. In a similar study, expectation did predict the placebo and Lidocain e ect in the late phase of experimental visceral pain, but not in the early phase [71].
Placebo e ects have also been shown in studies that combined verbal suggestion and conditioning in patients with IBS [72], knee osteoarthritis [73], atopic dermatitis [40], musculoskeletal pain [74], and chronic low back pain [69]. Results of a meta-analysis show that, in pain patients, medium-sized placebo hypoalgesia (Hedges' g � 0.65) was induced with a combination of verbal suggestion and conditioning; however, for this analysis, only three studies were available, and only experimental pain was investigated [68].
In a recent study, placebo e ects in experimental as well as chronic pain of patients with musculoskeletal pain did not di er when induced by verbal suggestion or a combination of verbal suggestion and conditioning. Yet, larger placebo responses in chronic pain were found in responders with more negative treatment history [74], indicating that prior experience plays an important moderating role, which, however, could be mediated by expectations. Investigating patients with atopic dermatitis with experimental electric pain, placebo hypoalgesia upon administration of a placebo cream was observed after verbal suggestion, after conditioning without verbal suggestion, and after a combination of both. Without conditioning, however, the e ect quickly diminished, suggesting that conditioning compared to verbal suggestion leads to longer lasting placebo hypoalgesia in patients [40]. Finally, Klinger et al. [69] tested four di erent groups with chronic low back pain in clinical pain, in an experimental electric pain model, in self-rated functional capacity, and in a behavioral test using time needed to perform standardized daily activities. Patients received a sham opioid solution and were either instructed that they received a placebo (PI) or a pain-reducing opioid solution (OI). Furthermore, half of the participants underwent a conditioning procedure, resulting in two more groups (PI + Cond and OI + Cond). Opioid instruction led to large placebo e ects on clinical and experimental pain, to decreased time needed for the exercises (small e ect), and to increased self-rated functional capacity (medium effect). e combination with conditioning (OI + Cond) led to larger e ects than OI on all outcome variables. e placebo-instructed group with conditioning showed a small yet signi cant e ect for the time needed to perform exercises.
ese results support the notion that conditioning incrementally contributes to placebo hypoalgesia in patients with low back pain because the e ects grew larger when a combination of conditioning and verbal suggestion was used, and the behavioral outcome measure showed placebo e ects even in the absence of verbal suggestions.
e available studies indicate that both verbal suggestion and conditioning are powerful determinants of placebo hypoalgesia in clinical populations. However, not enough data have been gathered yet to identify the relative signicance or an independent share of conditioning compared to explicit expectation in the clinical context. From a theoretical perspective, it can be assumed that conditioning constitutes an important aspect in chronic pain [75]. Especially, extinction has shown to be de cient in patients with chronic pain [76]. Furthermore, phenomena like latent inhibition (i.e., poorer learning as an e ect of ine ective preexposure) or blocking e ects (i.e., ine ective responding to a second conditioned stimulus) increase the risk of negative treatment experiences. Repeated experience of therapy failures, which is common in chronic pain, might lead to weakened placebo or persistent nocebo e ects potentially contributing to the maintenance of chronic pain. However, due to ethical constraints, nocebo hyperalgesia cannot be investigated as rigorously as placebo hypoalgesia in clinical samples, and most knowledge in the realm is derived from studies on the disclosure of possible adverse e ects in clinical trials [77]. On the other hand, there is limited evidence that conditioned placebo e ects are longer lasting compared to e ects that had been induced by explicit expectation [40,78], and this could be an opportunity for the treatment of clinical pain conditions.

Conclusion and Future Directions
In sum, conditioning usually leads to the development of explicit expectation or (more technically) contingency awareness. Furthermore, it is not surprising that such an expectation is able to enhance placebo hypoalgesia and nocebo hyperalgesia as numerous studies have shown when comparing the e ects after verbal suggestion and conditioning with verbal suggestion, respectively [27,28]. However, this does not exclude the existence and impact of conditioned e ects that do not depend on explicit expectation. Accordingly, Amanzio and Benedetti [34] elegantly showed that placebo hypoalgesia that had been conditioned with ketorolac and enhanced by a verbal suggestion was only partly reversed by the opioid antagonist naloxone, that is, naloxone only reversed the placebo e ect induced by the expectation part, as the other experimental groups indicated. It is conceivable that conditioned e ects can be compensated by or at least interact with contrary explicit expectations [10,39]. is does not mean, however, that an expectationindependent, conditioned e ect does not exist. We assert that there is not only one way to elicit placebo and nocebo e ects in pain that is mediation by explicit expectations. Rather, classical conditioning on its own can generate these e ects, which can be substantial in size, as available evidence shows for experimental as well as clinical pain.
One possibility to avoid confounding between induction by expectation and conditioning and to advance our understanding of the mechanisms causing placebo hypoalgesia and nocebo hyperalgesia would be the implementation of implicit conditioning designs. Here, an involvement of explicit expectations is excluded, and e ects purely caused by conditioning can be investigated. Compared to conditioning with subliminally presented cues, implicit conditioning has some advantages, as discussed above. Another way would be the development of experimental designs, in which contrary placebo or nocebo e ects are induced by conditioning and explicit expectation, and these e ects are measured on independent variables. Pain Research and Management 5