A recent article in Scientific American by neuroscientist Nick Turk-Browne is an article entitled "You Don't Remember Being a Baby, But Your Brain Was Making Memories." The article provides no real evidence that brains create memories, and its attempts to support such a claim are mostly references to junk science studies. In a previous post, I documented the untruth of the article's claims that two people could not form memories because of damage to their hippocampus. Let us now look at other aspects of the article that are just as dubious and misleading.
Turk-Browne makes this untrue claim: "Scientists were able to retrieve an otherwise forgotten memory by stimulating neurons in the hippocampus that had been active during an early experience." His only support for this untrue claim is a link to an interview with neuroscientist Tomas Ryan. In the interview Ryan claims, "We found out we could optically stimulate the engrams for forgotten memories -- and the memories were recalled." In the text of the interview that statement by Ryan has a link to Ryan's very low-quality paper "Engram Cells Retain Memory Under Retrograde Amnesia." That is a junk science paper guilty of several examples of bad research practices, such as the use of an unreliable method of trying to judge recall in rodents (the worthless "freezing behavior" method"), and also the use of way-too-small study group sizes such as only 8 mice or 10 mice. Contrary to the groundless boasts of Turk-Browne and Ryan, no evidence was produced by experiments of this type that a forgotten memory can be artificially reactivated.
What goes on in poorly designed experiments of this type is that mice are brain-zapped using light stimulation (optogenetics), and scientists claim the mice are "freezing in fear" because they are recalling a memory of a fearful experience, one "artificially activated" by the optogenetic light stimulation. But it is known that such optogenetic stimulation by itself causes "freezing behavior." A science paper says that it is possible to induce freezing in rodents by stimulating a wide variety of regions. It says, "It is possible to induce freezing by activating a variety of brain areas and projections, including the hippocampus (Liu et al., 2012), lateral, basal and central amygdala (Ciocchi et al., 2010); Johansen et al., 2010; Gore et al., 2015a), periaqueductal gray (Tovote et al., 2016), motor and primary sensory cortices (Kass et al., 2013), prefrontal projections (Rajasethupathy et al., 2015) and retrosplenial cortex (Cowansage et al., 2014).”
So no actual evidence is being produced of an artificial evocation of a memory stored in a brain when you use optogenetic stimulation of a brain region. All that is going on is that mice are allegedly becoming more immobile while they are brain-zapped, which does not tell us anything about memory. I say "allegedly" because there are no standards when it comes to judging whether mice were more immobile when exposed to some stimulus, with the reports of immobility typically being subjective, unreliable ratings by observers who did not follow a blinding protocol and who are free to choose any time span for judging immobility (30 seconds, 60 seconds, 90 seconds, two minutes or three minutes), whichever time span best seems to support a claim of "higher freezing behavior."
I later read a claim in the Scientific American article that shocked me. It was this statement: "My lab has been on a decade long quixotic adventure to study awake infants with functional magnetic resonance imaging (fMRI), a form of brain imaging that can measure activity from regions deep in the brain such as the hippocampus." When we examine the history of MRI scans, we see a history of overconfidence, and authorities dogmatically asserting that "MRI scans are perfectly safe," when they did not actually know whether they were perfectly safe. The 2009 study here ("Genotoxic effects of 3 T magnetic resonance imaging in cultured human lymphocytes") cautions about the use of a high-intensity("3T and above") MRI, and states that "potential health risks are implied in the MRI and especially HF MRI environment due to high-static magnetic fields, fast gradient magnetic fields, and strong radiofrequency electromagnetic fields," also noting that "these results suggest that exposure to 3 T MRI induces genotoxic effects in human lymphocytes," referring to effects that may cause cancer. The experiment discussed below used just such a 3 T MRI scanner.
There is a rule-of-thumb about things that may increase a person's risk of cancer. The rule is that the younger a person is, the more likely some possibly carcinogenic effect is to produce cancer in someone. Some particular stimulus might give you a 1% greater chance of cancer per year. But if you are 75 years, that is very unlikely to cause cancer in you. Conversely, if you are ten years old, that stimulus might have a very substantial chance or likelihood of producing cancer in you, if it increases your chance of getting cancer by 1% per year.
So the rule in medicine is: take the greatest caution to avoid exposing children to any possible cancer risk. We should then be horrified to hear of some neuroscientist subjecting infants to unnecessary fMRI scans. There is not merely a substantial cancer risk from fMRI scans to infants. There are also other risks. Less than 30 days ago someone was killed by an fMRI machine after being magnetically sucked into the machine. In 2001 a six-year-old boy was killed by an fMRI machine after its powerful magnets caused an oxygen cannister to hit his head.
Turk-Browne tells us that he has conducted 400 fMRI sessions, making this sound like 400 experimental sessions with infants. None of the parents who allowed this should have agreed to allow such MRI scans on their infant children, unless there was a medical necessity for such work. Subjecting infants to potentially hazardous experiments sounds like neuroscience research gone far astray, an example of morally reckless inanity.
Turk-Browne makes an unfounded boast, claiming "A team at my lab led by Tristan Yates...used this method to discover that the infant hippocampus can store memories beginning around one year of age." The claim is untrue. The claim has a link to the very low-quality science paper "Hippocampal encoding of memories in human infants," co-authored by Turk-Browne. You can read the paper here. The paper provides not the slightest evidence of any such thing as an encoding or storage of memory in the hippocampus or anywhere else. The study group sizes are only 13, too small for any reliable result to be claimed.
The authors of the paper try to make use of an analysis of a "subsequent memory effect." That is an unreliable neuroscience method that is usually pure pareidolia. It works like this: you scan someone's brain when he is being exposed to some stimulus, and also scan the person's brain when that person is asked to recall that stimulus. (For example, you might scan a person's brain while he is looking at some word pair he is asked to memorize, and then scan the same person's brain when he is asked to recall the second word, given only the first.) Then a neuroscientist looks for some region of the brain that showed "superior activation" both during the exposure to the stimulus, and the recall of the stimulus. This is an example of noise-mining. Regions of the brain randomly undergo tiny fluctuations in activity. So anyone analyzing brain scans can always find some tiny little region that was, say, 1% more active during the exposure to the visual stimulus and the later recall of the stimulus. We would expect you to be able to find such regions even if brains do not store memories. Similarly, if I eagerly analyze rain puddles in Chicago when the New England Patriots are playing football and rain puddles in Chicago when the Kansas City Chiefs are playing football, I may be able to find some little part of Chicago where there was an increased number of rain puddles during both of these types of games. But that does nothing to show that such football games have any causal relation to rain puddles in Chicago.
The paper "Identifying Causal Subsequent Memory Effects" reported that "we are unable to identify any signal that reliably predicts subsequent memory after adjusting for confounding variables, bringing into doubt the causal status of these effects." In other words, when properly analyzed in the fullest way, there was no evidence for any "subsequent memory effect."
Any paper claiming a "subsequent memory effect" with any credibility would have to meet various rigorous standards:
(1) The paper would have to be a pre-registered paper dedicated to trying to confirm some very specific hypothesis related to a subsequent memory effect.
(2) There would have to be a rigorous blinding protocol, to prevent some deal in which the scientists were guilty of "seeing what you are hoping to see."
(3) The subjects would have to be old enough to follow instructions, so that you could reliably tell what seconds were the learning period (or exposure-to-stimulus) and what seconds were the recall-of-stimulus period.
(4) Adequate study groups (probably of at least 30 subjects per study group) would be needed.
(5) A sample size calculation would need to be done to show that the study had used a study group size large enough to produce a high statistical power such as 80%.
None of these things occurred in the low-quality science paper "Hippocampal encoding of memories in human infants," co-authored by Turk-Browne. There was no pre-registration of a hypothesis to be tested. No substantial blinding protocol was followed, but merely a very limited level of blinding subverted by rater suggestions (as described below), with apparently no blinding occurring for the brain scan analysis. The paper has no discussion of a detailed blinding protocol. The study group sizes were only 13. The subjects were not old enough to follow instructions, being mere babies. No sample size calculation was done.
Turk-Browne and his colleagues attempt to derive some "subsequent memory effect" through some bizarre method in which they are trying to judge which stimulus a baby was looking at, and trying to guess whether a baby remembered a previously seen photo, based on which he looked at. It's methodological fumbling. You can't reliably tell which of two photos a baby is looking at, nor does which photo a baby looks at tell you anything about what a baby is recalling.
We read in the Supplemental Information document that judgments about which directions babies were looking in were made by human raters. We read this description of the nonsensical procedure:
"Human coders labeled frames of the video recordings of infant gaze during the encoding and test trials as: looking center, right of center, left of center, off-screen (blinking or looking away), or undetected (out of the camera’s field of view or otherwise not visible, such as being blocked by the infant’s arm). During encoding trial frames, coders were instructed that the infant was 'probably looking at center' because there was only one image on the screen at center. During test trial frames, coders were instructed that the infant was 'probably looking left or right' because there were two images, left and right of center, respectively."
Instead of simply judging objectively about which direction the babies were looking at, these "human coders" were pushed to judge in a particular way, like some baseball umpire getting a message in his earphone telling him "that was probably a strike" or "that was probably a ball." These instructions amount to a form of bias injection, and they mean none of the data about which direction the babies were looking at is reliable. Since none of that data is reliable, none of the claimed data about a "subsequent memory effect" is reliable.
We read in the Supplemental Information file about shady statistical fooling around which sounds like "keep torturing the data until it confesses" shenanigans. Here is only a small part of the impossible-to-justify convoluted rigmarole that went on:
"We used nonparametric bootstrap resampling to test for statistical significance. Specifically, we resampled participants for the contrast of interest with replacement 1000 times and recalculated the average for each iteration. The P value was quantified as the proportion of iterations with the opposite sign of the original subsequent memory effect, doubled for a two-tailed test. This analysis was first performed across the full sample, then separately in median splits of lower and higher average familiarity preference groups and younger and older infant age groups. We similarly quantified age-group differences by: resampling participants from among the younger and older infants, respectively, with replacement; calculating the mean value for each age group; and then subtracting the younger group mean from the older group mean. Again, the P value was the proportion of 1000 iterations that were of the opposite sign as the original group difference, doubled for a two-tailed test. Finally, we tested for a continuous age effect by resampling participants with replacement and recalculating the Spearman’s rank correlation between the subsequent memory effect and age in months over the sampled bivariate data pairs. Again, the P value was calculated as the proportion of resampled coefficients with the opposite sign as the original age effect, doubled for a two-tailed test. In exploratory whole-brain analyses, we performed nonparametric group statistics using the randomise function in FSL."
In the Scientific American article Turk-Browne speaks clearly about his ridiculous methodology when he says this: "If the infant looked longer at the photograph they had seen before, we labeled that image as remembered; otherwise, it was forgotten." That's an absurd procedure. You can't tell whether a baby remembered something by whether he looked longer at a picture on the left or the right, and human raters typically cannot even reliably tell which of two photos placed in front of him a baby is looking longer at. When later in the Scientific American article Turk-Browne claims that "the hippocampus was more active when infants viewed images that they seemed to remember," he is making a groundless statement based on his defective methodology, which provides no reliable data on what infants remembered in his experiments.
The idea of "familiarity preference" made in the paper (the assumption that babies would be more likely to look at a photo they had seen before) is an idea the exact opposite of the "novelty preference" assumption in many mouse experiments done by neuroscientists, which assume that a rodent will be more likely to explore some compartment they have not explored before. The authors confess "we did not observe a familiarity preference at the group level," which is a confession that helps show how misguided the methods and design of the study was. In other words, judging from the whole data set gathered, babies do not have any tendency to look more at a photo they have seen before. This comes as no surprise to me, having co-raised academically high-scoring twin daughters who never paid any attention to anything on the TV screen until they were 20 months old or older. Figure 2 of the paper is laughable, being data gathered from only a single subject.
The whole experiment is just a horrible-methodology mess of reckless infant endangerment inanity. Nothing has been learned about human memory from this nonsense. What has mainly happened is that some infants have been senselessly put at risk for the sake of a junk science paper. What percent (if any) of the babies endangered here will end up getting cancer because of their needless involvement in this methodological nonsense, because of the genotoxic effects of 3T MRI scanners mentioned in the scientific paper quoted above? We will not know for many years, if ever. The general rule of neuroscientists is "scan them and forget them." Neuroscientists do not do long-term tracking of health problems arising over a lifetime from subjects who participated in experimental fMRI scanning.
No comments:
Post a Comment