Microbiome-driven effects of non-nutritive sweeteners on human glucose tolerance

Lately, I have been experimenting with the "grazing diet". Instead of eating three meals a day one would eat a dozen or more really small meals; basically snacks.

The motivation was that the glucose response profile of my insulin was poorly matched with the glucose response profile of the foods I eat. My insulin (Humalog), has a profile lasting three to four hours, while most of my foods have a much shorter profile, some with a profile of an hour or less. By spreading my carbohydrate intake over more time I get a better match to my insulin response. For example, I spend two or more hours eating lunch. So far, it seems to be helping noticeably.

Listen, the term noise is probably the worst term for this data. Because we don’t know if it’s noise until we examine to see if it is noise. So until we know it’s noise we can’t call it noise. It’s like you’re walking into a crowded room and you’re trying to hear one thing but there is too much “noise”. This assumes we know what we’re looking for in the first place.

The Segal and Elinav labs are powerhouses in computational and gnotobiotic microbiome study. They've published several highly cited papers around interactions between diet, the microbiome, and various host parameters. A couple highlights include this [1] 2015 Cell paper predicting host glycemic response from microbial and dietary information, and this [2] 2014 Nature paper identifying artificial sweeteners as a source of glucose intolerance (mediated through the microbiome).

In the current work, they show that two non-nutritive sweeteners (NNS, saccharin and sucralose) impair glucose tolerance, and that the microbiome of individuals most susceptible to NNS-induced glucose-intolerance can transmit some of the phenotype to mice. These data are generated with human cohorts of good size (n=20 per sweetener) over a reasonable time frame (2 weeks of daily NNS administration). Importantly, the levels of NNS that are administered are well below the acceptable daily intake (ADI). For example, sucralose is given at 102 mg/day, about 34% of the ADI of 5 mg/kg, and reasonably close to an estimate of 1.6 mg/kg as average daily consumption in humans (reference in [3]). The strongest data for the paper is with sucralose (and saccharin). The researchers show that consumption of sucralose causes a shift in glycemic response: participants consuming sucralose had higher glucose excursion in a glucose tolerance test (GTT) than those consuming either control diet (Fig 2A, E, F). In addition to GTT changes, sucralose-consuming participants had altered level of 9 identified metabolites (Fig 4B-D).

After establishing these baseline results, the researchers search for mechanism by stratifying the sucralose cohort in the top and bottom responders. These are, respectively, the individuals who show most change from baseline in GTT at the end of intervention (2 weeks) and those that show the least. There are differences in metabolites, as well as the biochemical pathways those metabolites come from (TCA cycle) in these two groups (Fig 4E, F). There are correlations between changes in the microbiome (both specific taxa and functional gene categories) and the changes in measured metabolite levels as well (Fig 5).

In figure 6, the researchers present their strongest data. The researchers inoculate groups of germ-free mice with fecal samples from the sucralose participants from either the baseline or end of intervention. 4 groups of mice receive the feces of the top 1-4 responders (most perturbed GTT), 3 groups receive feces from the bottom 1-3 responders, and each of these is compared to a group receiving baseline feces. The purpose of this test is to see if the microbiome, altered by sucralose administration, can cause impaired glucose tolerance in mice that have never been exposed to sucralose. The researchers show that indeed there is significant glucose-tolerance impairment in mice that receive post-sucralose feeding feces, though interetingly they show that both bottom- and top-responder feces causes this (Fig 6A, G). They show that baseline samples from bottom- and top-responders do not cause differences in glucose tolerance (Fig 8C), showing that something about the sucralose treatment changes microbial composition to promote glucose intolerance. The researchers attempt to find a mechanistic explanation for the differences by comparing groups of mice colonized with top- and bottom-responder (grouped by baseline or end of intervention) feces.

Ultimately, this is an extremely impressive paper representing a lot of work (and many storied I haven't recapped). Like many microbiome papers, I think it oversells the mechanistic and physiologically relevant aspects of the research.

1. The data is presented in ways that maximize statistical significance with very little reference to the scale of the actual change. a. Fig 4 B-D and Fig 5 B, D, F show significant metabolite differences but give no reference to actual changes in blood concentration (also Fig 5 metabolites not significant after FDR correction). Without isotope-dilution mass spectrometry (which this is not), it's hard to tell how large the changes are in the blood metabolites. The relationship between concentration of a metabolite and measured area on a mass spec is a power function (for different metabolites exponents can be less or greater than 1), and so this data may represent a lot of change in concentration or very little. In addition, the authors rely on GTT differences to tell the story, but what is the scale of these changes? It is not clear that there is physiological relevance to this scale of change. b. Many of the metrics used are hard to relate to physiologically relevant quantities and allow researcher degrees of freedom. As an example Fig 3 ordinates the participant samples using a principal components projection of the microbial gene annotations. The loadings determining the ordination - and selected for highlight are shown in Fig 3 G-J. The researchers group several of these loadings into super pathways (e.g. purine metabolism) but it's very hard to tell if this kind of difference reflects a functional capacity change in the microbiome (and certainly gives no data about the actual transcription of these genes). Any of these groupings could be highlighted, allowing a lot of flexibility in the storytelling with no penalization for multiple hypotheses. c. Fig 8J "Spearman correlation of sucrose degradation pathway fold change abundance (day 21/baseline) with fold difference in GTT-AUC of each of the conventionalized mouse groups." This is so far away from physiology it's hard to say what it means.

2. Both Eran Segal and Eran Elinav are co-founders of the company DayTwo - a personalized microbiome company that helps diabetics manage their symptoms with microbiome based analytics and treatments. The paper feels like it explores the 'personalization angle' and the expense of other mechanistic studies. For example, the importance of osmolarity on the microbiome and how phenotypes around osmostress might be contributing to the resulting host phenotypes.

[1] https://www.sciencedirect.com/science/article/pii/S009286741... [2] https://www.nature.com/articles/nature13793?tdc_uid=921043 [3] https://foodinsight.org/everything-you-need-to-know-about-su...