A Conversation with Mark Baxter about the Gilman et al (2014) Paper

Yesterday, I had the pleasure of discussing the Gilman et al (2014) paper with Mark Baxter, Ph.D. Mark is a neuroscientist interested in brain mechanisms of learning and memory.

Mark and I had a brief back-and-forth on Twitter recently about the Gilman et al (2014) paper, which claimed to show that even casual cannabis consumption leads to structural abnormalities in the amygdala and nucleus accumbens. We decided that this issue warranted a longer conversation, so we agreed to talk about the paper on Google chat. Below is a slightly abridged version of that conversation. Where we reference ideas and papers that are not fully discussed here, I have included links to articles that should help make sense of what we’re talking about.

You can find the original Gilman et al (2014) paper here. You can follow Mark on Twitter here.

Mark hastened to add a disclaimer after our conversation that he is not an expert on addiction neuroscience.

Daniel: Maybe let’s begin by discussing some of the problems I have with the study (both the actual study itself and the media storm it created), and then we can get to your research?

Mark: Certainly.

Daniel:  I’ll begin by admitting that I’m by no means an expert on statistics or fMRI. My issues with this paper did not begin by doubting its study design or analysis methods, but rather by the media reports on the study. Having then read through the study itself, as well as other peoples’ responses to it, I have mostly stuck with my initial suspicion. My first problem has to do with the subjects.

The paper’s criterion for recruiting subjects into the marijuana group is that they smoke at least one joint per week. This, I think, is what fueled the claim that even “casual” marijuana use can cause the structural changes reported in the study. But if you look at the details on the subjects reported in the paper itself (in Table 1), you’ll see that these subjects are anything but casual users of marijuana. These subjects smoke an average of 11.2 joints per week, 1.8 joints/occasion, smoke on an average of 1.8 occasions/day, have on average been smoking since they were 16.6 years old, have been smoking for an average 6.21 years, and smoke an average of 3.83 days per week. This, to me, seems like fairly heavy consumption.

Table 1
A portion of Table 1. The data from the marijuana group is on the right. Means are highlighted in yellow and standard deviations are highlighted in purple. The zeros in the middle are the values of the control (non-marijuana group).

So, assuming that everything else in the study is valid, we should at least get rid of the claim that even casual marijuana usage leads to structural changes in the nucleus accumbens and the amygdala. But I don’t think that the consumption level is the only problem here.

Mark:  I need to look at what the criteria for marijuana dependence are in the DSM-IV. I think the “casual” use bears on the absence of dependence, but I agree the level of use described appears to be beyond “casual.” It’s unfortunate they give means and standard deviations instead of ranges though (in Table 1). The standard deviation for the number of joints is almost as high as the mean which suggests substantial variation.

Daniel:  For most of the values, that’s true. But, importantly, there isn’t a lot of variance (a standard deviation of 2.13) for age of onset. So all of the subjects began consuming marijuana when they were teenagers. Granted, some of the media reports did emphasize that casual marijuana use does cause structural brain change particularly in adolescent users, but this shouldn’t surprise anyone. We know, for example, that early onset of marijuana consumption can cause deficits in reward processing (which might be due to the changes in the nucleus accumbens reported in this study), but I don’t think there’s much evidence that cannabis consumption causes a similar behavioral deficit for those who begin smoking as adults.

So, I’d like to see whether you find these structural brain changes in people who began consuming cannabis as adults.

Mark:  That’s definitely true. I wasn’t paying attention to the media reports whether they stressed use in adolescence or not. That is definitely critical.

Daniel:  Most of them did not even mention adolescent use, as far as I remember.

Mark:  Wow, that’s significant. I would very much doubt you would see similar changes in people that start that level of cannabis use as adults. It’s an empirical question but a firm prediction.

Daniel:  I would agree, just based off of what we know about the behavioral changes that cannabis causes in adolescents versus adults.

You provided a nice segue-way into my next point when you mentioned the fact that these subjects were deemed not to be dependent on cannabis, which was based on a Structured Clinical Interview for the DSM-IV.

Mark:  I just looked up those criteria and it’s hard to see how anyone would ever get a marijuana dependence diagnosis. It requires 3/4 of: tolerance (need for more to get same effect or diminished effect of same amount), withdrawal (withdrawal syndrome or use of substance or closely-related substance to avoid withdrawal), taking substance in larger amounts or for longer period of time than intended, and having persistent desire or unsuccessful efforts to cut down or control use.

As I understand it there is little tolerance to cannabis (even “reverse tolerance” as levels build up over time) and if people are not motivated to curtail use, they can’t get a dependence diagnosis. But the clinical situation is outside of my experience. Other articles claim 4.3% prevalence of marijuana dependence.

Daniel:  Interesting. Then you’re probably right to say that it’d be hard to get diagnosed as dependent on cannabis.

Mark:  Abuse seems more likely (that is one or more of life problems related to substance use: poor work/home/school performance, legal problems, use in hazardous situations, use despite social/interpersonal problems). So 4.3% may include dependence + abuse (it’s a DSM-IV statistic and I don’t have a DSM-IV handy).

Daniel:  But, importantly, none of the subjects had any current or lifetime Axis I disorders. So we can assume that these subjects were psychologically healthy (unless they had personality disorders, which would fall under Axis II, but I don’t think that’s so relevant). And I find this to be problematic for the most common media interpretation of the study, which is that cannabis consumption causes brain damage, rather than just structural brain changes.

Mark:  Yes. Although, notably, absence of Axis I diagnoses doesn’t mean “normal” or neurocognitively unimpaired. But absence of neurocognitive/neuropsychological data from these subjects is also notable.

Yeah mentions of “brain damage” were crazy. It was increases in gray matter!

I found it interesting on some variables that there was a greater range in the controls than in the cannabis subjects (more separation/linearity in the shape measures).

Daniel:  Exactly. But it’s easy to blame media misrepresentation, when the authors of the study make similar claims. Hans Breiter, one of the authors of the study, said in a Society for Neuroscience press release that, “this study raises a strong challenge to the idea that casual marijuana use isn’t associated with bad consequences.” So clearly they’re interpreting these brain changes to indicate brain damage.

Mark:  Well, a hypertrophy of a brain area could be construed as a “bad consequence.” I wouldn’t want my nucleus accumbens or amygdala to grow. If he’d said “casual marijuana use is associated with brain damage” I would nail him on that.

But I agree it’s ambiguous and I think he should have qualified his statement – in view of no direct demonstration of a deleterious effect of the morphological changes.

Daniel:  Exactly. It could very well be that these structural changes indicate brain damage, but I’d need to see a lot more evidence first before I can accept that claim.

Mark:  Well “brain damage” should be confined to neuronal atrophy/loss, which clearly does not seem to be the situation here.

Daniel:  Right.

Mark:  I’ll take “abnormality” (as loaded with value judgment as that term is) but I agree that “bad consequences” is probably a little deceptive. But media sees “brain difference” and hears “brain damage”

Daniel:  I’m quite certain that the authors of the study were aware of that fact before they chose the words with which to describe their findings.

Mark:  Well, the paper never really says they’re bad (just abnormal). It was the press release/interview that gets Hans in trouble on that one (“bad consequences”). I am sure he believes that the changes are bad i.e. will lead to functional impairment/compromise/vulnerability to psychopathology.

Daniel:  That’s true. But what I’m really concerned with here is how this study gets interpreted. And I think that the researchers behind the study are responsible for discussing their findings in such a way that best ensures that they don’t get used as backing for political/legal measures that are not in fact consistent with their results.

In the context of today’s political environment and the debates surrounding the legalization of medical marijuana, I think that Breiter’s comments are particularly irresponsible.

Mark:  Trying (and failing) to find the SFN press release. Found a Boston Globe article that is interesting if you read it carefully.

Daniel:  Here’s the SFN press release: http://www.sfn.org/Press-Room/News-Release-Archives/2014/Brain-Changes-Are-Associated-with-Casual-Marijuana-Use-in-Young-Adults.

Mark: Ah, thank you. Yeah, shame on Hans for “bad consequences.” That’s too much editorializing.

But the increases in gray matter reported in the study are interesting and consistent with the most notable effect of adolescent cannabis in rodents, which is increased sensitivity to other drugs of abuse (adolescent but not adult rats are much more interested in heroin after cannabis).

Daniel:  Well again that speaks to the importance of emphasizing that this effect holds mostly for adolescents (presumably).

Mark:  It would have been so easy to include a delay discounting or some other task and look for a functional correlate of accumbens shape. (I guess he couldn’t have offered them heroin and asked them how much they liked it.)

Daniel: Well, maybe some variant of the marshmallow task for adults.

Mark:  They cite the animal data on cannabis increasing dendritic length and branching in nucleus accumbens, which would be consistent with increased gray matter/shape changes.

Daniel:  Right, the Kolb et al. paper. Do you know if the increase in dendritic length and branching in the nucleus accumbens corresponded to noticeable behavioral changes?

Mark: Kolb et al don’t provide very specific data on how old their rats are but they appear to be postadolescent for sure. The only behavioral data are whether they are sensitized to THC 30 days after initial administration.

Daniel:  Which I’m betting they weren’t, based on what we know about THC consumption patterns in human adults.

But it’s interesting to note that they found these changes in postadolescent rats.

Mark:  Indeed, they were not. If they’d given them cocaine or amphetamine, I expect they would have gotten significant sensitization.

Daniel:  Maybe because we have endogenously occurring cannabinoids?

Mark: Yeah the function of the endocannabinoids is still very mysterious. The CB receptors are highly developmentally regulated.

Daniel:  I’ll defer to your expertise on this one. I’ve read a bit about endocannabinoid research, but I couldn’t make heads or tails of it. Probably because we don’t know enough.

Mark:  Well no one is really certain still what the endogenous ligands for the cannabinoid receptors are. CB1 receptors are decreased in schizophrenia in prefrontal cortex, which I think has attracted attention (and is something I’ve thought about trying to pursue in a primate model).

But back to the Breiter paper. I did think they got a little harshly judged for Bonferroni corrections, especially on the drug use measures which are all very highly correlated (a point addressed in the paper). 

Daniel:  Exactly what I wanted to talk about next!

Mark:  And this was all coming from computational biologists who mostly deal with data sets that are all noise and so they have to Bonferroni correct the hell out of it to convince people there is anything there.

Daniel:  I think we’ve both read Lior Pachter’s blog post, then.

Mark:  Yep! I wouldn’t have done the bold-uncorrected asterisk-corrected thing they did because if you’re using p-values and significance cutoffs it’s either significant or nonsignificant.

Daniel:  That’s exactly what I wanted to mention. While Pachter’s criticism regarding how liberal their Bonferroni corrections were makes sense to me, again the fact that Gilman et al boldfaced their p-values before correcting for multiple comparisons and that they put tiny asterisks next to the p-values that remain significant after Bonferroni correction seems irresponsible. Irresponsible in the sense that it makes their results look more significant than they really are, which is problematic considering the political context of the study. And, even if you don’t think that the researchers needed to have corrected for every cluster of voxels measured (as Pachter suggests), I still think they should have corrected for the fact that the drug use measures were highly correlated.

Mark:  If they wanted to be really strict about the measures of use they could have thrown it in a principal component analysis and gotten a unitary measure of “frequency of use” for correlation and they would have found basically the same thing. I mean, only one Bonferroni-corrected significant correlation would allow you to say that the morphological change correlates with frequency of use.

It’s clear the frequency measures are more related to shape than to gray matter density or volume too, which is interesting. If they had four totally independent measures of use and failed to correct, I’d be more concerned that they had gone fishing for a relationship.

Daniel:  Well they want to say that the frequency measure is correlated with left nucleus accumbens volume (Figure 1C), but, as Pachter points out, if you remove that outlier it’s likely that the effect will flatten out.

Figure 1C
Figure 1C from the study, plotting the relationship between the average number of joints smoked on every occasion and the volume of left nucleus accumbens. I have drawn a red circle around the outlier – if you take out this outlier, the correlation between cannabis consumption and left nucleus accumbens volume will certainly be insignificant.

Mark:  Figure 1C mystifies me. Why show a nonsignificant correlation?

Actually, if you take out the 0 smoking occasions/day (the controls) a lot of the drug use relationships to gray matter and volume look quadratic to me in Figure 1.

Then their shape correlations which are much “more significant” and more interesting are shown in tiny scatterplots in Figure 2. I think Figure 1B and 1C were rather poor judgment about presentation and they should have expanded Figure 2, as shape seems to be the more sensitive measure.

Figure 1B
Figure 1B from the study, which plots the relationship between various smoking measures and gray matter density. Mark points out that without the data from the control groups (the x=0 points on the left of each plot), the relationship between cannabis consumption and gray matter density in the figures above actually looks quadratic. The same holds for the plot of cannabis consumption versus left nucleus accumbens volume (Figure 1C). This also undermines the supposed linearity of the relationship between cannabis consumption and the brain measures reported.
Figure 2
Figures 2A and 2B from the study, which show the results of the shape analyses. Here, you do actually see that cannabis consumption is correlated with a difference in shape in the left and right amygdalae and left and right nucleus accumbens.


Daniel:  Well this goes back to your point about your concern over the potential hypertrophy of a brain area. Change in the shape of an ROI doesn’t concern me as much as a change in volume, and clearly they don’t have strong evidence of a change in the volume of their ROIs. Change in shape (or connectivity) is to be expected with anything that causes any perceptual or memory changes, even if those changes aren’t negative.

Mark:  I don’t think there’s any strong a priori expectation about the mode (volume / gray matter density / shape) or direction (increase/decrease) of a plastic change in brain structure.

Consider the London taxi drivers (an overworked example): I think they have a relative hypertrophy of the anterior hippocampus and an atrophy of the posterior hippocampus (I might have that backwards).** 

Daniel:  Fair. 

Mark:  But it is associated with performance increases in certain kinds of spatial tasks.

Daniel:  Well that goes back to my main point, which is that none of these changes – volume, shape, or density changes – indicate brain damage. But I was just bringing back your earlier point that though you wouldn’t consider an increase in the volume of a brain area as brain damage, you wouldn’t want your nucleus accumbens or amygdala to grow.

Mark:  As a general rule I want to avoid changes in brain morphology!

Daniel:  Well that’s too vague a rule. Education causes changes in brain morphology. But you don’t want to avoid that. The key is to identify what morphological changes cause behavioral deficits. And, as I think we both agree, the changes cited in this paper likely do not correspond to behavioral deficits.

Mark:  Well. Changes in brain morphology now. Frankly I’d be a little worried if I engaged in any single activity to the extent it produces a persistent change in brain structure. But I may be overgeneralizing.

As far as the changes in this paper, I would allow for these specific changes to be a possible substrate for increased sensitivity to reward value/rewarding stimuli and/or increased impulsivity. But this was not shown. I’m not even aware that anyone’s looked at individual differences in impulsivity or reward sensitivity and amygdala/accumbens morphology in a “normal” adult sample.

Daniel:  Right, since those are the general functions associated with these brain areas. And we do know that cannabis consumption can lead to these behavioral changes in adolescents. But no one is talking about making cannabis legal for teenagers so, again, I don’t think this paper contributes anything meaningful to the current political debate surrounding the legality of cannabis.

Mark:  Oh I totally agree. This is orthogonal to the legalization of cannabis.

I note that they include a line about legalization of marijuana in their discussion though! They may believe that (that their findings warrant further research that “will help inform discussion about the legalization of marijuana”) but I disagree with them.

Daniel:  Well that has been the primary motivating factor for me, at least, in calling out the issues with this study. If this paper hadn’t been released just before multiple states were poised to re-evaluate the legal status of marijuana, I wouldn’t be as concerned with its faulty methodology or the irresponsible reporting of the study.

Mark:  That makes a lot of sense.

Daniel:  And, again, considering who funded this study, this just seems like an effort on the part of the federal government to push people away from legalizing medical marijuana/general use of marijuana in state elections this November.

Mark:  I think it’s interesting from a scientific point of view and I think it is not as technically flawed as it’s been made out to be, but I also don’t think it should be used as a basis for public policy. For me, I evaluate everything against caffeine/ethanol. 

Daniel:  Well it’s certainly less damaging than ethanol!

Mark:  Ethanol use is horrible, neurotoxic, produces persistent changes in brain function, has abuse liability, bad for adolescents, but still super legal and a great source of government revenue!

Daniel:  I completely agree. The fact that studies like this are published and widely circulated on social media while everyone continues to drink multiple times a week makes me really pessimistic about both our drug policies and the power that the federal government has to affect public opinion on drugs. But I think that ultimately reason will win out – this study wouldn’t have been so hotly debated if people weren’t beginning to think more critically about drug policy.


After that, we discussed drug policy a little more, made some comments about how much we enjoy discussing an issue with someone who is both open-minded and well informed, and then we wrapped up our conversation.

I would like to again thank Mark for discussing this paper with me. This experience really drove home for me what critical thinking is all about: it’s about questioning everything, it’s about sticking to the evidence, and, most importantly, it’s about keeping the conversation going.

What do you think about the Gilman et al (2014) study? Leave a comment below.

**  He did have it backwards. The original Maguire et al (2000) paper found that London taxi drivers had a relative increase in volume of the posterior hippocampus and a relative decrease in volume of the anterior hippocampus.

One thought on “A Conversation with Mark Baxter about the Gilman et al (2014) Paper

Leave a Reply