This post was originally published in Neuropsychotherapist magazine January 2015
and written by
Coherence Psychology Institute
Reconsolidation Process Is Triggered by Reactivating a Target Learning or Memory
As noted earlier, in the reconsolidation discovery studies of 1997 to 2000, a state of deconsolidation was found to exist only soon after the target learning had been reactivated by a suitable cue or reminder. This observation was interpreted by the researchers to mean that each reactivation of a target learning deconsolidates its neural circuits, launching the reconsolidation process. That conclusion may have been sensible based on the initial few studies, but it turned out to be incorrect. Pedreira, Pérez-Cuesta, and Maldonado (2004) were first to show that reactivation alone does not bring about deconsolidation and reconsolidation. They concluded, “at odds with the usual view, retrieval per se is unable to induce labilization of the old memory” (p. 581), and they demonstrated that what the brain requires to trigger the reconsolidation process is re- activation plus another critical experience, described below. Subsequently, this same two-step requirement has been demonstrated in at least 22 other studies that I have tallied as of this writing.
They are listed in Table 1.
In the discovery studies of 1997 to 2000, researchers had fulfilled this two-step requirement without awareness of doing so, as shown later in this section. The early interpretation that reactivation by itself produces deconsolidation spread widely among both neuroscientists and science journalists and became a reconsolidation meme. Despite the post-2004 piling up of decisive evidence revealing that this original conclusion was incorrect, it has continued to be asserted in new writings by not only science journalists but also by some prominent researchers who were involved in the original studies, as well as by many later reconsolidation researchers. As of this writing, more than 10 years since the mismatch requirement was first detected and published, new research articles continue to be published that lack any consideration of the mismatch requirement’s role in the reported results (e.g., Wood et al., 2015).
It is perhaps understandable that science journalists would latch on to and continue to spread the misconception that reactivation in itself destabilizes the reactivated learning, if they were unaware of what the ongoing research was revealing. It is less clear why the error would continue to be voiced by researchers. From my point of view as a clinician observer witnessing this situation unfold for almost a decade, I cannot escape the impression that many reconsolidation researchers appear unaware of sizable amounts of research published in their own area of specialization. Some of the more significant reconsolidation research articles, such as that of Schiller et al. (2010), assert that reactivation induces reconsolidation and reference none of the studies in Table 1 that have shown that view to be incorrect. Commenting here on this situation is hopefully warranted by the importance of assuring that research findings critically important for clinical application are not obscured.
What, then, is the second step that must accompany reactivation?
Pedreira et al. (2004), followed by all of the studies listed in Table 1, have shown that in order to induce reconsolidation, reactivation must be accompanied or followed soon by what researchers term a mismatch experience or prediction error experience. This is an experience of something distinctly discrepant with what the reactivated target memory “knows” or expects—a surprising new learning con- sisting of anything from a superfluous but salient nov- elty element to a direct contradiction of what is known according to the target learning. It makes sense from an evolutionary perspective that deconsolidation and reconsolidation, being the brain’s process for updating learnings and memories, would be triggered only by new information that is at odds with the contents of an existing learning (Lee, 2009). Lee wrote, “reconsolidation is triggered by a violation of expectation based upon prior learning, whether such a violation is qualitative (the outcome not occurring at all) or quantitative (the magnitude of the outcome not being fully predicted)” (p. 417). It would be biologically costly, with no benefit, if the brain launched the complex neurochemical process of reconsolidation when there is no new knowledge requiring a memory update. The studies listed in Table 1 have shown that the brain evolved so as to launch de/reconsolidation only when an experience of something discrepant with a reactivat- ed, learned expectation or model of reality signals the need for an update of that existing knowledge. This empirical finding of a critical role of mismatch or prediction error can be regarded as a neurobiological valida- tion of a central feature of the learning models of both Piaget (1955) and Rescorla and Wagner (1972). Thus, what shifts a particular learning into a deconsolidated, destabilized state, allowing its expression to be modified or erased by new learning during an approximately five-hour window, is not simply reactivation of that learning, but the experience of that reactivated learning encountering a mismatch or prediction error. As stated by Agren (2014) in reviewing research on reconsolidation of emotional learnings in humans, “it would appear that prediction error is vital for a reactivation of memory to trigger a reconsolidation process” (p. 73). Likewise, Delorenzi et al. (2014) commented, “strong evidence supports the view that reconsolidation depends on detecting mismatches between actual and expected experiences” (p. 309). Exton-McGuinness, Lee, and Reichelt (2015) review the role of prediction errors in memory reconsolida- tion studies and sum up their position by stating, “We propose that a prediction error signal . . . is necessary for destabilisation and subsequent reconsolidation of a memory” (p. 375). That is the research finding that translates into major advances for the psychotherapy field (Ecker, 2011; Ecker et al., 2012, 2013a,b).
For those advances to materialize, it is necessary for clinicians to understand well what the brain regards as an experience of mismatch or prediction error. Misconceptions abound on this point as well. The follow- ing example shows the meaning of mismatch at the basic level of classical conditioning in the laboratory, as demonstrated by Pedreira et al. (2004) and other studies listed in Table 1.
Clinically relevant learnings are often far more complex, and the guiding of mis- match experiences in psychotherapy looks very different, as a rule, from the laboratory instances described in this article, but the principles of mismatch are usefully clarified at this basic level. Consider a target learning that was created by several repetitions of turning on a blue light and delivering a mild electric shock several seconds later, during the last half-second of the light being on. Subsequently, if the blue light is turned on again, the learned expectation of the shock is reactivated immediately, along with fear and the physiological expressions of fear, such as a mouse’s freezing or a human’s change of skin conductance. However, this reactivation does not deconsolidate and destabilize the memory circuits of this learned association of light and shock, because no mismatch experience has occurred as yet. While the blue light stays on without any shock being delivered, a mismatch or prediction error has not occurred because the shock might still occur.
The target learning is in a state of expectancy of the shock. Mismatch occurs when the blue light is turned off with no shock having been experienced. Only then are perceptions discrepant with what the target learning “knows.” Now the synapses encoding the target learning unlock into a modifiable state, because now it is definite that no shock occurred as expected while the blue light was on.
Understanding the mismatch requirement allows us to interpret correctly the results of various studies that were misinterpreted by the researchers because they analyzed their studies without reference to the mismatch requirement.
The simple logic of the situation, as stated by Agren (2014), is that “the studies that have shown effects of reconsolidation . . . must somehow have induced a prediction error” (p. 80). Ecker et al. (2012) articulated the same principle: “Whenever the markers of erasure of a learning are observed, both reactivation and a mismatch of that learning must have taken place, unlocking its synapses, or erasure could not have resulted. This logic can serve as a useful guide for identifying the critical steps of process in both the experiments of researchers and the sessions of psychotherapists” (p. 23).
Therefore, identifying the presence or absence of mismatch in each of the many published studies of reconsolidation that lacks consideration of the mismatch requirement is an exercise necessary for bring- ing the field of reconsolidation research to maturity from its present fragmented condition. The remainder of this section begins that unifying exercise by describing several key studies, analyzing the presence or absence of mismatch in them, and reinterpreting their results accordingly. This analysis of mismatch in published studies yields instructive insights into how mismatch may function. The study by Nader, Schafe, and LeDoux (2000), which repeated the basic design of some other early studies (Przybyslawski et al., 1997, 1999; Roullet et al., 1998), is often regarded as the one that brought the initial research to a tipping point of establishing the reconsolidation phenomenon conclusively. Nader et al. used the same classical conditioning procedure as described in the example just above, but with an audible tone rather than a blue light. They taught rats to expect a shock during the last half-second of a 30-s tone. Later, their procedure accomplished mem- ory reactivation with the onset of the 30-s tone, and it accomplished memory mismatch with the offset of the tone with no shock occurring, triggering destabi- lization of the target learning and launching the re- consolidation process.
However, the researchers were unaware of the mismatch requirement (which was discovered four years later by Pedreira et al., 2004) or of the crucial role of this mismatch in triggering de-consolidation of the target learning. It was by chance that their procedure happened to include the needed mismatch. Memory erasure resulted from anisomycin administered soon after that mismatch experience (but not when administered 6 hr later, when the reconsolidation window was no longer open), confirming that memory destabilization (deconsolidation) had occurred, because anisomycin destroys only non-consolidated synapses. Understandably but erroneously, Nader et al. con- cluded that memory reactivation was sufficient for triggering destabilization. If their design had includ- ed reactivation by the tone together with the expected shock, eliminating the mismatch of expectations, no deconsolidation or erasure would have occurred. Such failure to achieve destabilization of a reactivated target learning has been reported in many studies (e.g., Bos, Becker, & Kindt, 2014; Cammarota, Bevilaqua, Medina, & Izquierdo, 2004; Hernandez & Kel- ley, 2004; Mileusnic, Lancashire, & Rose, 2005; Wood et al., 2015), and we can now recognize that this failure was due to an absence of mismatch or prediction error in the procedure used. (For example, as reported by Hernandez and Kelley in 2004, a rat’s memory that pressing a certain lever brings a sugar reward was indeed reactivated when the rat was once again placed in the chamber with the lever, pressed it, and received a sugar pellet, but this reactivation provided the expected reinforcement and entailed no experience of prediction error, so memory destabilization did not occur.)
In these studies, too, the researchers made no mention of a mismatch or prediction error requirement in their interpretation of results. Instead, they concluded incorrectly that the particular type of memory under study was not subject to reconsolidation. Subsequently, other studies successfully demonstrated reconsolidation for those types of memory (see, e.g., Wang, Ostlund, Nader, & Balleine, 2005).
All 23 studies listed in Table 1 have shown that reactivation alone does not launch the reconsolidation process, but reactivation plus mismatch does. This point was particularly emphasized by Forcato, Arg- ibay, Pedreira, and Maldonado (2009) in titling their article, “Human Reconsolidation Does Not Always Occur When a Memory Is Retrieved,” and by Seven- ster, Beckers, and Kindt (2012), who titled theirs “Retrieval Per Se Is Not Sufficient to Trigger Reconsol- idation of Human Fear Memory.”
The latter authors characterized their next published study by stating, “we show in humans that prediction error is (i) a necessary condition for reconsolidation of associative fear memory and (ii) determined by the interaction between original learning and retrieval” (Sevenster, Beckers, & Kindt, 2013, p. 830). Reconsolidation can also be triggered by a mismatch of when events are expected to occur, with no change in what occurs, as demonstrated by Díaz- Mataix, Ruiz Martinez, Schafe, LeDoux, and Doyère (2013).
On Day 1 in their study, rats heard a 60-s tone and received a momentary electrical shock at the 30-s point, midway through the tone. For each rat this was repeated 10 times to create a reliable conditioned response of fear to the tone. On Day 2, each rat heard the tone and received the shock again just once, re-activating the learned association of tone and shock. The shock occurred at the same 30-s point for some rats, but for others it occurred at the 10-s point. Im- mediately after this reactivation experience, researchers administered a chemical agent (anisomycin) that disrupts nonconsolidated memory circuits. On Day 3, the tone was played again for each of the rats five times with no accompanying shock, and the strength of fear responses was measured. Rats that had un- changed shock timing on Day 2 reacted with fear on Day 3 fully as strongly as they had done on Day 2, indicating that anisomycin had no effect and, therefore, that the reactivation without mismatch on Day 2 had not destabilized the target learning. In contrast, rats whose shock timing had been changed on Day 2 reacted on Day 3 with only half as many fear responses as on Day 2, indicating that anisomycin had signifi- cantly impaired the target learning and, therefore, that the reactivation with timing mismatch on Day 2 had indeed destabilized the target learning.
This important finding that temporal mismatches trigger reconsolidation will figure significantly in other discussions later in this article. Díaz-Mataix et al. did identify the prediction error that played a critical role in their procedure, and they concluded from their observations that new information must accompany reactivation in order to destabilize the target learning. That conclusion corroborates what was demonstrated in at least sixteen prior studies listed in Table 1, so it unclear why Díaz-Mataix et al. describe their finding as though it is a new discovery and cite only one of prior studies (Sevenster et al., 2012). A target learning that has been destabilized by mismatch can be erased not only by chemical agents, but also by a counterlearning experience with no use of chemical agents. It is this endogenous approach that is most desirable for psychotherapeutic use and which has been applied extensively in that context (Ecker et al., 2012).
In laboratory studies, endogenous erasure or modification of a target learning has been demonstrated with both animal and human subjects (e.g., Galluccio, 2005; Liu et al., 2014; Monfils, Cowansage, Klann, & LeDoux, 2009; Schiller et al. 2010; Steinfurth et al., 2014; Walker et al., 2003; Xue et al., 2012).
Monfils et al. (2009) used three pairings of a 20 second audio tone (the conditioned stimulus, CS) and half-second footshock (unconditioned stimulus, US), with 3 min between pairings, to train rats to respond to the tone with fear. One day later, the target learning was reactivated by the CS/tone, but there was no accompanying US/shock, which is a mismatch of the expectation of the US. So far, the procedure is basically the same as that of Nader et al. (2000), described above, but rather than disrupt the target learning chemically at this point, Monfils et al. continued to present the CS without US repeatedly. CS2, the second tone, was presented 10 min or 1 hr after the first one, but then additional CS tones came at 3-min intervals, for a to- tal of 19 CSs. That procedure successfully and robustly erased the rats’ learned fear of the tone. Note that if the initial 10-min or 1-hr interval had been a 3-min period like all of the ensuing intervals, the repetitive CS counterlearning procedure would have been a standard multitrial extinction training, which is well known not to bring about erasure. Thus the longer interval between CS1 and CS2 was critically important for achieving erasure through reconsolidation rather than suppression through extinction. The fact that erasure occurred implies that the target learn- ing was destabilized and erasable during the series of CSs, which in turn implies that the longer interval from CS1 to CS2 resulted in a mismatch of expected and actual timing. (The discussion of results provided by Monfils et al. does not refer to the concept of mismatch or prediction error, however.) The key role of a temporal mismatch in inducing destabilization in both Monfils et al. (2009) and Díaz-Mataix et al. (2013) makes it clear that the brain learns the temporal features of new emotional experiences no less than it learns other characteristics, and that mismatches of timing can be highly effective for inducing reconsolidation in cases where the target learning has distinct temporal structure. In other recent research, networks of dedicated “time cells” in the hippocampus have been found to measure and remember time intervals (Jacobs, Allen, Nguyen, & Fortin, 2013; MacDonald, Lepage, Eden, & Eichen- baum, 2011; Naya & Suzuki, 2011; Paz et al., 2010). The important observations made by Monfils et al. (2009) will be revisited and utilized later in this article to address fundamental questions of what governs whether reconsolidation or extinction occurs and why extinction fails to produce erasure. It is noteworthy too that the erasure procedure used by Monfils et al. was subsequently adapted for use with human subjects by Schiller et al. (2010), who demonstrated the first endogenous erasure of a fear learning in hu- mans in a controlled study. The Appendix to this article provides a detailed examination of the mismatches involved in their procedure. Fulfillment of the mismatch requirement is evident in the successful inducing of reconsolidation in a wide range of experimental procedures. For example, an associative fear learning can be triggered into reconsolidation by a reexperiencing of only the un- conditioned stimulus (US) without the conditioned stimulus (CS) (Díaz-Mataix et al., 2011; Liu et al., 2014). The target learning, consisting of experiencing first the CS followed by the US, is mismatched if the US occurs without the CS. That mismatch consists of both the absence of the expected CS and also, importantly, a change in the expected temporal sequence of events, because the target learning expects the US to occur after the CS, not without the prior occurrence of the CS. Another example is the case of having two different co-occurring CSs, both of which have been paired with the same US. Debiec, Díaz- Mataix, Bush, Doyère, and LeDoux (2013) showed that reexposure to either one of the CSs can trigger the reconsolidation of the memory of the other. Here the expected co-occurrence of both CSs is mismatched when only one CS is presented. None of the authors referenced in this paragraph explained their results in terms of the mismatch requirement, however. They discussed their results as though the triggering of re- consolidation can be attributed to reactivation alone. Even researchers who are well aware of the mis- match/prediction error requirement can overlook the occurrence of mismatch in their own procedures.
For example, Pine et al. (2014) provided an ingenious and intricate demonstration that reconsolidation occurs for complex, unconscious emotional learnings in humans—and in doing so they have supplied the strongest empirical support to date for the anecdotal clinical observations reported by Ecker et al. (2012, 2013a,b)—but they commented, “Our results seem to counter a recent theory that new learning (or the generation of a prediction error) is required during reactivation in order to trigger reconsolidation. . . . Here, no new learning took place during the reminder” (p. 11).
However, the “reminder” (reactivation) that they used for triggering reconsolidation on Day 2 of their procedure contained three distinct temporal mismatches relative to the original learning on Day 1: a reversal of overall sequence, an overall duration of the series of trials that was one-fourth as long, and the introduction of a 10-min delay within the over- all sequence. Thus, due to these temporal mismatches of the original learning, its reactivation was actually accompanied by an abundance of new learning, triggering destabilization and reconsolidation. As noted above, we know from Díaz-Mataix et al. (2013) that even a single temporal mismatch can be an effective destabilizer.
As this section’s final example of how the mismatch requirement can account for diverse reconsolidation phenomena, there have been several studies of how the age or strength of a target learning effects the triggering of memory destabilization (Boccia, Blake, Acosta, & Baratti, 2006; Clem & Huganir, 2010; De- biec, LeDoux, & Nader, 2002; Eisenberg & Dudai, 2004; Frankland et al., 2006; Inda et al., 2011; Milekic & Alberini, 2002; Steinfurth et al., 2014; Suzuki et al., 2004; Winters, Tucci, & DaCosta-Furtado, 2009).
Reviewing all results of these studies is beyond the scope of the present article, other than to summarize that, as a rule, stronger reactivation is required in order to destabilize stronger or older target learnings. Some of the studies in this area successfully destabilized both young and older target learnings (up to 8 weeks after acquisition), but others failed to destabilize older memories. Lee (2009) commented, “it is also possible that all memories undergo reconsolidation regardless of their age, but that previous studies have failed to use sufficiently intense memory reactivation condi- tions for older memories” (p. 416). The results found by Suzuki et al. (2004) can serve to illustrate the further research possibilities that become apparent as a result of asking the question: If the mismatch requirement is responsible for experimental observations, what are those observations showing about how mismatch functions under various circumstances? Suzuki et al. taught rats to fear a test chamber (context/CS) by placing each rat in the chamber for 2.5 min and then administering a 2-s footshock (US). Rats in one group received just one shock; those in another group received three shocks separated by 30 s. All rats were removed from the context/CS 30 s after their final footshock. Then, either 1 day, 1 week, 3 weeks, or 8 weeks later, immediately after administration of anisomycin, rats were placed in the context/ CS for various amounts of time and then removed with no shock, in order to reactivate the fear learning and disrupt it if it had been destabilized by the reactivation. One day later their fear level was measured during a 3-min reexposure back in the context/CS.
This procedure resulted in the following findings:
For fear memory created by a single context-shock pairing, a 1-min shock-free reexposure to the context did not destabilize the fear learning, but a 3-min reexposure did destabilize it if memory age was 1 day, 1 week, or 3 weeks. The implication is that a 1-min shock-free re- exposure did not create a mismatch experience, but a 3-min reexposure did create a mismatch. This is suggestive of a temporal structure in the target learning. The 2.5-min period of initial exposure to the context/ CS fits that possibility well, because relative to that learned 2.5-min period, the 1-min reexposure could have been too short to be experienced as a nonreinforcement, so it would not create a mismatch experi- ence, but the 3-min reexposure would. If memory age was 8 weeks, a 3-min reexposure no longer caused destabilization, but a 10-min reexposure did destabilize. This possibly implies that memory of the 2.5-min period lost definiteness over time and therefore required a longer reexposure for decisive nonreinforcement and mismatch to be experienced. For fear memory created by three context-shock pairings instead of one, a 3-min reexposure no longer caused destabilization, but a 10-min reexposure did de- stabilize. Here the challenge is to understand how a stronger fear training would alter the timing memory. Three 2-s shocks coming every 30 s is a grueling minute that might feel to a rat much longer than a minute spent sniffing around curiously in a harmless place, just as a human also experiences time periods very differently depending upon the presence or absence of pain. The prior 2.5-min duration may have been distorted or blurred retroactively by this long, traumatic minute, such that the longer 10-min reexposure was necessary for decisive non-reinforcement and mis- match to be experienced. The interpretations sketched above are not the only possible ways in which the mismatch requirement could have resulted in the observations made by Suzuki et al. (2004).
They are offered here heuristically, by way of showing how the mismatch requirement can be logically applied to illuminate how experimental procedures interact with the inherent properties of the brain’s memory systems. The experimental procedures discussed in this section in relation to the mismatch requirement illustrate a principle that is critical for understanding reconsolidation phenomena: What does, or does not, constitute a mismatch experience depends entirely on the specific makeup of the target learning at the time of mismatch. That is a principle that I will refer to henceforth as mismatch relativity. It is essential for understanding the effects of reconsolidation procedures used in both laboratory studies and therapy sessions. In the small minority of reconsolidation research articles that do address the mismatch requirement, I have never seen mismatch relativity articulated explicitly; rather it is either tacitly assumed or asserted in an abstract manner (as in Bos et al., 2014, and Sevenster et al., 2013, 2014; for example, Bos et al. state, “The experience of a prediction error upon reactivation critically depends on the interaction between the original learning of the fear association and the memory retrieval” [p. 6]).
Mindfulness of mismatch relativity is critical for consistent outcomes in utilizing reconsolidation in psychotherapy to bring about transformational change.
Only by attending closely to the specific elements of a symptom-generating emotional learning can a psychotherapist reliably guide mismatch experiences that disconfirm those specific elements, as is necessary for their nullification and dissolution. A question often asked by clinicians learning about reconsolidation is: When my panicky therapy client drives on the highway and the feared terrible fiery crash doesn’t happen, that seems to be a mismatch experience, as needed to launch reconsolidation, yet it doesn’t unlock or erase the learned fear. Doesn’t this show that the model is incorrect?
To clarify this, we need to apply the mismatch relativity principle and examine whether or not a mismatch experience actually took place. That begins with examining the detailed makeup of the target learning in question. In this case, the target learning is not that a car crash happens on every drive; rather it is that a crash might happen unpredictably on any drive. That learning is not mismatched or disconfirmed by an accident not happening on any one drive or on any number of drives. A safe, uneventful drive creates no prediction error and therefore does not induce deconsolidation, so the target learning is not revised and the model has not failed to apply.
This example naturally raises the question:
For that target learning, what would be a mismatch experience?
The knowledge that a crash might happen unpredictably on any drive is true as a recognition of existential reality, so no mismatch or disconfirmation of that knowledge is possible. However, that knowledge is not the entire learning maintaining the panicky dread of a fiery car crash. Some other learning is responsible for that emotional intensity, and it is for elements of that learning that mismatches can be created. The most common form of this other learning, though not the only possibility (see Ecker, 2003, or Ecker & Hulley, 2000, for an account of diverse learnings underlying anxiety and panic symptoms), is suppressed traumatic memory of the same or a similar kind, such as a car crash, a fiery explosion, the death of high school classmates in a head-on collision, a terrible scare from skidding on ice on a mountain road or from being pulled along very fast at 3 years old in a little wagon tied to the bicycle of an older sibling, and so forth. The suppressed state of the traumatic memory preserves its emotionally raw, unprocessed quality, including desperate fear and helplessness. Desuppression of the memory (in small enough steps to be tolerable) reveals a set of specific elements, each of which is a particular learning. It is these component learnings that can now be subjected to a mismatch experience.
For example, the helplessness felt and learned in the original situation can in many cases encounter a mismatch experience through the technique of empowered reenactment, which is widely used in trauma therapy to create a vivid experience of potent self-protection in the original scene. For a detailed clinical example of that kind, see Ecker et al. (2012, pp. 86–91).
In summary of this section, the research findings on memory reconsolidation represent a nontheoretical set of instructions for bringing about transformational change in a target learning. These instructions specify that in order for a target learning to become destabilized and susceptible to being unlearned and nullified, it must be both reactivated and subjected to a mismatch or prediction error experience. The mismatch relativity principle has been introduced here, within the exercise of analyzing the occurrence of mismatch in published studies, to emphasize that what is, and what is not, a mismatch experience is always defined in relation to the specific elements of the target learning and what the target learning “knows” or expects. This exercise of examining the role of mismatch in published studies will continue in each of the next two sections. (For numerous examples of creating mismatch experiences in psychotherapy, see Ecker et al., 2012, Chapters 3–6.)
For all the references made in this post regarding reprogramming the mind using memory reconsolidation.