Monday, December 30, 2013

How Can We Forget?

** This post is meant to be read in tandem with its more complimentary cousin, Electroconvulsive Therapy Impairs Memory Reconsolidation, at The Neurocomplimenter. **

spECTrum 5000Q® ECT device (MECTA)

Bad memories haunt a significant number of people with serious mental illnesses, such as chronic major depression and post-traumatic stress disorder (PTSD). If it were possible to undergo an experimental procedure that selectively impairs your memory for an extremely unpleasant event, would you do it? If this sounds like the plot of Eternal Sunshine of the Spotless Mind, you're not alone.

A pet peeve of mine is reference to this excellent but far-fetched film in scientific journals and popular media coverage of “memory erasure.” The idea that it's possible to selectively remove a complex autobiographical memory that has become intimately entwined with the fabric of our constructed selves is utter science fiction.

At some level, even Michel Gondry knew it. One incident in Eternal Sunshine is suggestive of how memories might actually be stored. It was after one of the main characters (Joel) had his memories of his ex-girlfriend Clementine erased, and he couldn't remember who Huckleberry Hound was. He had associated the cartoon character and the song "Darling Clementine" with her. That resembles a semantic network, where an overlapping network of neurons and synapses code different but semantically related things. Take out all episodic and semantic memories of Clementine, and knowledge of Huckleberry Hound goes with it.

The latest incarnation of this particular memory erasure meme was provoked by publication of a paper (Kroes et al., 2013) that examined the process of memory reconsolidation in depressed patients administered a course of electroconvulsive therapy (ECT). Here are some of the headlines:
Zapping the brain can help to spot-clean nasty memories

Absolutely shocking: electrocuting brain can wipe unpleasant memories

Unwanted Memories Erased in Electroconvulsive Therapy Experiment

Shocking Memories Away

My companion post at The Neurocomplimenter reviews the literature on memory reconsolidation and describes the experiment of Kroes et al. (2013) in some detail. What I'd like to do here is to point out possible weaknesses in the results that could undermine the authors' conclusions. I'll also discuss a much earlier ECT study which did not support the notion that reactivated memories are especially vulnerable to disruption (Squire et al., 1976).

To briefly reiterate the methods used in the new paper (Kroes et al., 2013), the participants were 39 patients with moderate to severe major depression. They were either at the end of an acute treatment cycle or receiving maintenance ECT. The study used a between-subjects design with three different experimental conditions, with patients randomly assigned to one of the three groups (n=13 in each). The within-subjects factor was whether or not the patients received a reminder of previously learned material before treatment.

All participants learned two different emotionally charged slide stories with audio narration, each consisting of 11 images. In one, a boy is in an accident that severs his feet, which are reattached at the hospital. In the other, two sisters leave their home at night, and one is kidnapped at knife point and attacked by an escaped convict.

Memory for one of the stories was reactivated a week later by presenting part of the first slide, and then giving a test for this slide. Only four minutes later, Groups A and B were anesthetized and received ECT. Group C received their ECT treatment at a later date. The final memory test for Groups A and C was 24 hrs after the reminder, while Group B was tested as soon as they woke up from the procedure (mean = 104 min later). The final test consisted of 40 multiple choice questions about each of the stories.

The basic idea is that reconsolidation of the reactivated story isn't complete within 104 min, so Group B's test performance should be the same for the two stories. In contrast, reconsolidation is complete by 24 hrs, so for Group A the disruptive effect of ECT should selectively impair memory for the transiently reactivated story, which is in a labile state (relative to the "consolidated" story learned 7 days earlier).

Is that what was observed? Statistically speaking, yes. But two patients in Group B (out of 13 total) performed very well on the reactivated story (see purple box in the figure below).

Fig. 1 (modified from Kroes et al. 2013). ECT disrupts reconsolidation. Memory scores on the multiple choice test are expressed as percentage correct (y axis). Memory for the reactivated story shown in solid bars and non-reactivated story in open bars. Each circle is the score for an individual patient. The horizontal dotted line is chance performance. Group A is in red, Group B in blue, and Group C in orange. Edited to add: The purple box highlights 2 outliers in Group B who could be driving the major effect.

If these two individuals are omitted, would the difference between Groups A and B still be significant? This is the key finding of the paper, that memory for the reactivated story is no better than chance if ECT disrupts reconsolidation (a time-dependent process). Hence all the Eternal Sunshine / “memory erasure” headlines.

The purple boxes show that (1) the Reactivation x Group interaction squeaks in at just barely significant (p=.049); and (2) the Group A vs. Group B comparison for the reactivated story is p=.042. Clearly, it would be nice to include twice as many patients in each group. But it took the authors 3.5 years to recruit their final total of n=39.

This type of study is not easy to pull off, which is why I applaud the authors (and the patients) for such an ambitious undertaking. I thought it was a very clever idea as well, but not an original one as it turns out.

In the 1970s and 80s, Dr. Larry Squire and his colleagues published a series of papers on ECT and memory. The one I'll describe here takes a similar approach to Kroes et al. by testing previously learned material after ECT, and by giving a memory reminder just before the treatment (Squire et al., 1976).

Squire et al. (1976) used a completely within-subjects design (n=12) that manipulated the pre-ECT learning interval (14-18 hrs vs. 3-10 min). The third condition presented a memory reminder 3-10 min before ECT for material learned 14-18 hrs previously. Completely different stimuli were used each time, and the order of conditions was counterbalanced. The memory tests were recognition memory for a set of 32 previously learned items (common objects, common words, yearbook photos, and nonsense drawings), and paired associates (producing the correct target for 18 previously learned cue-target pairs). In all conditions, retention was tested 6-10 hrs after ECT (compare to 104 min and 24 hrs in Kroes et al.).

A separate group of patients (n=9) was tested on their remote memories for old TV shows under three conditions: (1) 6-10 hrs after ECT; (2) 14-18 hrs before  ECT and again with the same questions 6-10 hr after ECT; (3) Less than 10 min before ECT and again with the same questions 6-10 hrs after ECT (the reminder procedure).

The critical result is that the memory reactivation procedure did not impair performance (bar graphs A vs. R below). In both of these conditions, material was learned 14-18 hrs before ECT. This is in contrast to the findings of Kroes et al. (2013).

Fig. 1 (modified from Squire et al., 1976). Results from (A) 32-item recognition memory test, and (B) paired-associate learning test, under three conditions in conjunction with the patients' first 3 ECT sessions. Retention was significantly impaired in Condition B (initial learning 3-10 min before ECT). The reminder procedure (Condition R) caused no impairment in performance relative to Condition A.

Squire and colleagues (1976) concluded that “...the results provide no evidence that the presentation of previously learned material just prior to ECT increases its vulnerability to disruption.” Similar results were observed in the patient group tested on their knowledge of old TV shows: “the results clearly indicate that amnesia for remote memory did not occur when remote memory was evoked prior to ECT.”

The final conclusions [clairvoyantly] throw cold water on the study published 37 years later:
The present findings also have important clinical implications. The reactivation phenomenon described in experimental animals has raised the possibility that it might be therapeutically advantageous to evoke depressive ideation just prior to treatment, in order to produce amnesia for this ideation. The results reported here strongly suggest that this procedure would be ineffective.

However, you'll probably notice some differences between the two studies. Kroes et al. pointed out that their effect was observed at a 24 hr retention interval, while Squire et al. only tested at 6-10 hrs (perhaps not long enough to disrupt reconsolidation). The Squire stimuli were neutral in valence, whereas the Kroes stimuli were emotional (and perhaps more susceptible to disruption). There were also differences in the patient groups (Squire's were younger, mean=39 yrs), anesthesia used, electrode locations, and ECT parameters (likely to be way more potent in the earlier study, which would predict worse amnesia). An unfortunate side effect of ECT is memory impairment, although other studies claim the opposite.

Certainly, subjective cognitive complaints after ECT are very common. For a first hand look at some of the more devastating effects, watch the powerful video below. For a lighthearted and critically acclaimed look at fictional memory erasure, watch Eternal Sunshine of the Spotless Mind.


Kroes MC, Tendolkar I, van Wingen GA, van Waarde JA, Strange BA, & Fernández G (2013). An electroconvulsive therapy procedure impairs reconsolidation of episodic memories in humans. Nature neuroscience PMID: 24362759

Squire LR, Slater PC, & Chace PM (1976). Reactivation of recent or remote memory before electroconvulsive therapy does not produce retrograde amnesia. Behavioral biology, 18 (3), 335-43 PMID: 1016174

Liz Spikol (of The Trouble With Spikol fame) tries to explain the confusion and the loss of self she felt after waking up from ECT.

"After the ECT, I did not know how to use a toothbrush. And that lasted for months."
- Liz Spikol (at 9:28)

Subscribe to Post Comments [Atom]


At December 30, 2013 7:47 PM, Blogger CCA said...

I am a clinical psychiatrist in charge of busy ECT Clinic. This entry is excellent, to say the least. I am impressed by the originality of the writer and by his sharpness. I´ll use this blog entry in my teaching-journal clubs/etc.

At December 31, 2013 12:05 AM, Blogger The Neurocritic said...

Thank you!

At December 31, 2013 6:17 AM, Anonymous Anonymous said...

NC, I completely agree about the issue with those two outliers. I have not read the original paper, but given that those two participants clearly look like outliers, the authors should have at least done some statistical outlier tests and reported the results. Do you want to bet that if the authors had embraced the no-difference hypothesis, those two outliers would have been excluded? :)

PS. Happy New Year Neurocritic!

At December 31, 2013 6:40 PM, Blogger The Neurocritic said...

Thanks, Anonymous. And Happy New Year to you too!

At January 01, 2014 5:07 PM, Anonymous Anonymous said...

Long ago it was hypothesized that ECT "works" by making depressed people forget why they were depressed in the first place!

At January 05, 2014 2:32 AM, Anonymous Anonymous said...

I would be interested to know which ECT parameters you think would be likely to be way more potent in the earlier study.
I haven't read either study, but in general sine-wave current and bilateral electrode placement are thought to cause more memory impairment that brief-pulse current and unilateral electrode placement. Nowadays brief-pulse is used more frequently than it was in the 70s. Unilateral electrode is still the minority electrode placement (although I don't know which electrode placement was used in these particular studies).
Other parameters have if anything gone in the direction of more potency. Voltage is similar, amperage is similar or higher, and there has been an increase in the duration of the electric shock. This means that the amount of electrical charge (millicoulombs) is if anything greater for today's patients than for those in earlier times and helps to explain why some people undergoing ECT today report similar memory loss to those who had it many years ago.

At January 05, 2014 12:24 PM, Blogger The Neurocritic said...

Thanks for your interesting observations. I originally wanted to include an appendix with more detailed info to compare patients and ECT parameters. The earlier study did use sine-wave current and bilateral electrode placement, and the later one used brief-pulse and bilateral or unilateral placements.

Squire et al. 1976

PATIENTS: N = 19 (15 had never received ECT, the others had not in the past year). Within-subject design with 3 different treatment conditions. n=12 for memory tests #1 and #2, n=9 for memory test #3 (2 of the same plus 7 new ones). Condition order counterbalanced to fullest extent possible. Mean age=39 (23-60).

ECT PARAMETERS: bilateral, 0.6-0.75 sec, 130-150 V.

ANESTHETICS: atropine** (0.8-1.0 mg), methohexital sodium (40-80 mg), succinylcholine (30-60 mg).

Kroes et al. 2013

PATIENTS: final N = 39 (either at the end of an acute treatment cycle or receiving maintenance ECT). BETWEEN-subject design with 3 different treatment conditions, with patients randomly assigned to Groups A, B, C with n=13 in each. Mean age=57 (34-84).

ECT PARAMETERS: Right unilateral or bifrontotemporal with a brief pulse (0.25–0.5 ms), constant current (0.9 A) apparatus (Thymatron System IV, Somatics). Applied charges for each group (means) = 321.78; 428.40; 292.71 mC (millicoulomb). Dosage set at 2.5-fold initial seizure threshold (i.e. stimulus dosage that elicited a motor seizure of at least 20 sec, as established during first ECT session) in bifrontotemporal and at sixfold initial seizure threshold in right unilateral ECT.

ANESTHETICS: IV etomidate** (0.2–0.3 mg/kg) followed by succinylcholine (1.0 mg/kg) and positive pressure oxygen (100%).

** has amnestic properties

At January 08, 2014 10:32 AM, Anonymous Anonymous said...

Thank you. So Squire gives volts and seconds and the Dutch team give other parameters but not volts and seconds. But looking at the specifications for the Thymatron machine
you would need a shock of 7.2 seconds to get those millicoulombs, which is about ten times the duration of the shock in Squire's experiment. Who got the more powerful shocks? Difficult to say, but I don't think it is a foregone conclusion that it is the patients in the older experiment.

At March 13, 2014 11:10 AM, Blogger Unknown said...

Dear Neurocritic,

I am afraid the figures got lost in my post. If interested, please contact me and I will send you the figures.



At March 13, 2014 1:20 PM, Blogger Unknown said...

One’s study being discusses on both the Neurocomplimenter and Neurocritic can only be taken as a compliment. Once again the Neurocritic/complimenter has done an excellent job in describing a study, placing it into context, and highlighting potential pitfalls and outstanding issues.

The Neurocritic describes a possible weakness in our study by pointing out two potential outliers in Group B (purple box Fig 1 of the Neurocritics entry), an issue that can be especially problematic for studies with a relative low number of subjects. However, a critical aspect of our study is that the reactivation of memory was a within-subjects variable, i.e. subjects learned two stories, one week later memory for only one of the two stories was reactivated, and subsequently memory for both stories was tested. This means that for each subject we have a memory score for the reactivated and non-reactivated story. The two suspect outliers pointed out by the Neurocritic score high on the reactivated story, but these same two subjects also score high on the non-reactivated story (See Fig. 1 below), i.e. their overall memory performance is good.

Fig 1. Memory scores for each group with potential outliers as suggested by the Neurocritic in purple and green boxes.

The analyses of possible outliers should thus be done taking the memory scores of both stories into account. In order to do so we calculated the difference scores in memory performance for both stories (Non-reactivated memory score – Reactivated memory score, Fig 2 below). As one can clearly see the two subjects’ difference scores are not extreme, and overall there appear to be no outliers.

Fig 2. Difference in memory scores for the Non-Reactivated story – Reactivated story for each subject in Group B. The two possible outliers are identified by the purple boxes (note that these scores are in portions).

Yet appearances don’t mean much in science. Thus we calculated the mean of the difference scores in group B (M = -0.0002, SD: 0.02936) and the 5% trimmed mean (-0.022). The trimmed mean hardly differs from the mean and is within one standard deviation. Indeed a SPSS boxplot (Fig. 3 below) shows no outliers.

Fig 3. SPSS boxplot of the difference scores in group B does not reveal any outliers.

We are thus unable to identify any outliers, and find no reason to assume that outliers might have driven our results. We therefore stand by our conclusion that our results provide evidence for ECT to disturb reconsolidation of episodic memories in humans.

Of course this is a proof-of-concept under highly experimentally controlled conditions. As the Neurocritic/complimenter discusses there are critical outstanding questions to be addressed before this can lean to practical clinical applications: Can we disturbed reconsolidation of older memories? Of real-world highly traumatic memories? Do these memory alterations persist over prolonged periods of time? And can we find alternative, less invasive, methods to alter memory that have less side effects.

An interesting future for this research field lies ahead, and hopefully new publications on reconsolidation will again lead to excellent entries by the Neurocritic/complimenter.

At February 18, 2017 3:24 PM, Anonymous Anonymous said...

Hi Neurocritic, towards the beginning of your post, you say that "The idea that it's possible to selectively remove a complex autobiographical memory that has become intimately entwined with the fabric of our constructed selves is utter science fiction". In light of the recent work by Sheena Josselyn on eliminating the neurons representing specific memories, do you think think that this still holds true?

Here's my references:

At February 19, 2017 10:05 PM, Blogger The Neurocritic said...

Hi Anonymous, thanks for your interest. The work of Josselyn and colleagues is very impressive. But I still believe that an artificially instilled fear memory in a mouse is quite different from complex human autobiographical memories (as in the science writers' favorite, Eternal Sunshine of the Spotless Mind). The fear memory in these rodent experiments is relatively simple: the association of a sound with a shock. The neural pathways are pretty well known. It's not an episodic memory that includes rich contextual details like where you went to dinner with your partner, what you ate, what you talked about, where you parked, what you wore. And why you argued.

From the Quanta article:

"The key to her team’s success was a set of molecular tools designed to mark the brain cells that formed a new memory — the neurons that fired when the mice learned that a specific sound predicted an imminent electric shock. The researchers then used these molecular tags to wipe out only those brain cells, eradicating that memory but leaving others intact."

I'm not sure how you would target the specific cells involved in encoding all the details surrounding that dinner date and the engram of what ultimately led to your breakup that night.

I'm not familiar with the “Cocaine Engram” study, but the translational potential seems greater to me.

At June 23, 2017 9:08 PM, Anonymous Anonymous said...

Hi again neurocritic. In this study, the researchers have claimed that their results demonstrated memory erasure.

In these kinds of studies, I'm always confused what they mean by memory. In this study, do they mean that the fear component of the memory is attenuated, or do they mean that they erased the actual memory of the event. That seems to be what the futurism article is implying towards the end of the article.

At July 04, 2017 7:59 AM, Anonymous Anonymous said...

Sorry, I mean't to phrase that as a question, but I somehow forgot the question mark. Does this snail experiment show anything particularly new about memory erasure? The researchers talk about removing both associative and non-associative memory. But do snails have anything resembling a complex autobiographical memory?

At September 28, 2017 2:57 PM, Anonymous James E said...

I'm not a neuroscientist and my understanding of the brain is layman level, but would the silencing of genes that are important to the maintenance of memories also be subject to the limitations of memory erasure that you discuss in this article? I ask this because I recently read (!po=35.2778 ) that interfering with DNA methyltransferase can erase a long term "memory" in snails. In the linked study, it is stated that "identifying genes that can suppress memory might allow us to reduce some of the harmful effects of traumatic experience."


Post a Comment

<< Home

eXTReMe Tracker