| Microbiome researcher here. As usual, when a microbiome paper reaches HN, I emerge to urge caution. You should be extremely skeptical of the larger conclusions drawn from this paper. The biology of Alzheimer's disease (AD) is poorly understood, and the contribution of dietary and microbiome factors are extremely unclear. I think you should treat the evidence from this paper (and shared in this thread) as more like "fails to converge" than "I will use this data to update my priors". I want to note, first, that the researchers should be commended on clearly communicating their procedures, crafting a well written paper, and doing some nice experiments (in figure 4 particularly). The main criticisms I have relate to the conclusions drawn, and then echoed in this HN thread. 1. The behavioral analysis was not conducted in a blinded way - the researchers doing the behavioral analyses knew if the rats were AD or non-AD groups. From supplement "The experimenter was
blinded to the FMT human donor but not to the rat group." The degree to which rodent behavior reflects human behavior is unclear (I would argue very little) but you should definitely be skeptical of measures with significant subjectivity done without blinding. 2. The paper has many materially important assertions not backed by the data. > However, at phylum level, Alzheimer’s patients had a higher abundance of Bacteroidetes (Fig. 1C) reported to comprise many pro-inflammatory species A phylum is an extremely diverse group of microbes - the phylum Bacteroidetes contains microbes that likely diverged hundreds of millions of years ago (from their LCA). Claiming that an entire phylum is "pro-inflammatory" is not even wrong. The different Bacteroidetes found in humans can be "pro-inflammatory" or "anti-inflammatory" depending on a truly staggering number of factors (diet, what other microbes are there, host immune status, host epithelial integrity). To claim that it's clear that AD patients have more of a "bad phylum" exemplifies the strong claim X weak evidence of this paper. > ...and a lower abundance of the phyla Firmicutes and Verruocomicrobiota, reported to produce beneficial metabolites. Ahh yes, the beneficial metabolites of my favorite Firmicutes - Clostridium botulinum (most toxic metabolite known to man), Clostridium difficile (kills >20,000/year in the US), etc... Again, you can't paint with this broad a brush and have the conclusion mean anything. > Importantly, a positive correlation was observed between the abundance of the health-associated SCFA producer Coprococcus and the MMSE score, and inverse correlations were detected between the abundance of the disease-associated pathobionts Desulfovibrio, Dialister and the MMSE score, supporting a microbiome signature for cognitive performance in Alzheimer’s disease. Desulfovirbio has evidence of potential linkage to AD (a quick google found correlational, but no causal evidence). On the other hand, a quick google showed that other groups disagree about Dialister - Vogt et al 2017 (PMID: 29051531) find that Dialister is reduced in AD patients and the less Dialister you have, the worse your symptoms! Again, claims stronger (and in this willfully ignoring) than evidence. 3. The main data from this paper is generated in a standard fecal transplant experiment. In short, feces from AD patients is given to some rats, and feces from healthy donors (HD) is given to other rats. Figure 2-4 rely on this setup. These are not independent verifications or tests of the hypothesis "Do AD bacteria --> AD?". Any systematic difference between the feces with the AD label and the HD label could be causal for the results the authors see. This is a major confounder in all microbiome work. Behavioral changes associated with AD are going to change food intake and bowel habit, which in turn will substantially change microbial characteristics of the gut. Other than the microbes, there are a huge number of human derived proteins, metabolites, signaling molecules, viruses, etc in the stool. The researchers don't eliminate these, and their data would be consistent with this as an explanation just as much as the microbes. Even the best studies cannot create a design that blocks every confounder, but FMT paradigm suffers from a really large number of them and the authors don't address this. There are some experiments you can do - for example, you could culture the individual microbes you believe to be responsible for the AD pathology, introduce them into rats, and show the _microbe alone_ is enough to cause the disease (this resolves the last confounder cited). The authors don't do this. There are many other considerations in designing FMT experiments that I could quibble with (like why not germ-free animals - that is the standard rather than the microbiota depletion they do), but the main point is: any systematic difference in AD vs HD feces could cause the differences in rat physiology they observe. 4. Figure 2 is an excellent example of a broader phenomenon in microbiome papers: statistically significant results with extremely small effect sizes. Consider 2D-I. The authors show (for instance) that the mean water content of stool differs significantly in their AD- vs. HD-colonized rats. The effect size is maybe ~12%. Everything in 2G-I is smaller than this. Do we think that losing 12% of the water from feces is a part of the physiological cascade that leads to AD? While this could be true (or this could be evidence of some other underlying change being caused by the AD microbes) the effects are extremely small. I think it's much more likely that some other systematic difference in the feces (as outlined in 3) is causal for these small-scale changes seen. This is similar to the data in Figure 3. Consider 3B - the AD microbiota induces a (significant) ~0.23% change in the "discrimination index". What size is that effect? It might be large, but considering there is no comparison to differences in whatever the human equivalent of this index is (if it even exists) I am going to guess it's more on the 'too small to be meaningful' side. > These observations were further corroborated by correlating the clinical human donor profile to the Alzheimer’s behavioral readouts of the recipient rats.
They test 50 different metadata correlations, don't correct for multiple hypotheses, and find one correlation at p<0.05 and one at p<0.08. This does not match the tone of that sentence. 5. Quoting from the paper: "Rats colonized with faecal material from Alzheimer’s donors exhibited no change in locomotor parameters in the Open Field Test (Supplementary Fig. 6A), no change in anxiety-related behaviours in the Elevated Plus Maze (EPM) (Supplementary Fig. 6B), or in antidepressant-like behaviour in the Forced Swim test (FST) (Supplementary Fig. 6C), indicating no specific effects of the Alzheimer’s human gut microbiota on comorbid features of Alzheimer’s disease in rats."
So clearly whatever the effect of the microbiota is, it's not enough to trigger these measures of pathology. Does that suggest that the microbiota causes only some AD features? If you read their abstract "Our findings reveal for the first time, that Alzheimer’s symptoms can be transferred to a healthy young organism via the gut microbiota, confirming a causal role of gut microbiota in Alzheimer’s disease" they certainly don't qualify the result this way. 6. Figure 4 - there are some nice experiments here. They show differences in neuronal survival conditioned on AD or HD feces. There are microbial metabolites that affect neuronal survival that are certainly not linked to AD causally (e.g. the short chain fatty acids). Evaluate this in the context of point 3 above: the authors show differences in neuronal survival, but whether that is due to changes in AD patients microbiome that are _caused_ by the disease (e.g. they eat weird and get weird microbes that don't make short chain fatty acids) rather than _casual for_ the disease cannot be evaluated from this data. 7. Figure 5 has no evidence of microbial cause. The authors show that serum from AD patients injected into rats caused reduced neurogenesis. As outlined in point 3 above - this could be (and seems much more liekly to be) from any of the human-derived factors floating in the blood. They try to address this by saying "gut microbiota composition explains up to 58% of the variance of individual plasma metabolites" and citing a paper (PMID: 36151114) but this is a very misleading citation. First, the authors of the cited paper corrected that figure to 46%. Second, that figure is for a SINGLE metabolite, not the whole plasma metabolome. Quoting from the cited paper "We detected the largest variance explained (46%) for an uncharacterized common metabolite with the provisional identifier X-11850.". The gut microbiome makes a lot of chemicals, but your body makes many many more. The idea that gut-derived metabolites are the predominant thing in the bloodstream is...not correct. The authors similarly try to suggest with figure 6 that because the AD-transplanted rats show a different plasma metabolome, that means the serum injection experiment points to the microbes. This is not causal evidence, and it isn't a test of what they say it is. |
It's a bit sad that this comment is in the lower half of the thread, while the comments in the higher half are at the same time: clearly uninformed on the topic and still making strong statements.