Subscribe (free) on , , , , or by .
Ashley's Biggest Takeaways
- RNA Seq of the metabolome has revealed significant overlap in the differential between Autism Spectrum Disorder and neurotypical controls in both microbial and human pathways.
- Understanding how the human body itself functions as a whole is key to figuring out the crosstalk between host and microbe.
- Importantly, clinical trials are often limited by small sample size.
- AI is being used to bridge this gap, distill large amounts of information and help put things in context.
Featured Quotes
What is ASD?
One of the main challenges of autism is trying to figure out how to phenotypically characterize it. Right now, the way it is done is from a clinical perspective, in terms of behavioral assays and just watching the kid and their behavior. It's very hard to quantify.So I'll say that autism isn't a single disorder, it's a whole spectrum. That's why it's called Autism Spectrum Disorder. And that spectrum encompasses a multitude of different symptoms, and it turns out about 50% of autistic children have GI symptoms—diarrhea, constipation and much more.
There's a lot that goes into this. It includes sleep, dietary issues, and a lot of these kids have restrictive diets. How all this connects to behavior and learning ability—these are all major unknowns.
Omics Studies Indicate Significant Host-Microbe Crosstalk
We are trying to understand how all these omics levels are tied together—so we have the microbiome and the metabolome across [various] body sites, urine, blood, fecal—as well as trying to understand the connection to diet, the connection to the immune system and the connection to brain function. This is in terms of not just the behavioral assays, but also, in our case, actually mapping back to RNA Seq extracted from brain tissues.I can give a hint of what we discovered. There's a lot to unpack. There are 4 groups of microbes that stand out. So there's Bacteroides fragilis, which has been shown to improve symptoms in mice. When you feed B. fragilis as a probiotic, those pop up in our analyses as protective.
We have Bifidobacterium, like bifiduma, and these are bugs that are associated with early development and childhood. So these are amongst the first bugs that colonize the gut; they help digest milk. This signature also pops up, and exactly what it means is unclear. Whether it's protective, or if it's if it's representing a delay in gut maturation, that's currently not well understood.
Other bugs that pop up like Desulfovibrio, which is a sulfate reducer, and Prevotella copri, which is a particularly interesting one, because it has been shown to have immense dietary diversity. So, I'll give you a little bit context. In previous studies that showed that P. copri has enormous diversity, but then you can see differences in this bug with respect to Western diets. So, its hypothesized that these bugs have evolved to digest certain complex carbohydrates, and some of them have specialized in Western diets.
So basically, what we've done is differentiated ASD children from age, sex matched neurotypical controls. This turned out to be critical for tweezing apart the signal. So, the way that this worked is, after computing all those microbial differences across pairs, across studies, we can extract putative gene function. So, figuring out what genes differentiate between different groups of bugs, those bugs that are more elevated in ASD children, and those bugs more elevated in neurotypical controls.
So, from those genes, you get metabolic information. You get an idea about what enzymes are present that differentiate these bugs and the enzymes that break down certain byproducts. This gives you an idea how to link back to diet, as well as linking back to brain metabolism.
So, one thing that made this interesting is that we looked at 4 different RNA Seq studies. These are RNA Seq as a measure of the brain transcriptome activity, so the intermediates from translating DNA to protein. And you can map those back to metabolic pathways as well. And once you map everything down to metabolism, now you have a unified language to kind of figure out how everything can be connected.
One thing we noticed is that when you focus on pathways, our differential between ASD and neurotypical controls in both microbial pathways and human pathways there are substantial overlap. So we're talking about on the order of like 150 some pathways. Really raising the question how much crosstalk is there?
I think answering this question requires us to confront the elephant in the room: causation. How can you figure out the causal relationship of microbes to host response in the host phenotype? And I think clinical trials are going to play an increasingly important role in that.
The next question is figuring out how can we bridge that gap, going from cross-sectional and longitudinal studies to those involving interventions, those involving investigating patient stratification, as well as deeply phenotyping these cohorts. This is where the bridge to AI can happen.
The Bridge to AI
One thing AI is good as figuring out how to distill information. One of the big challenges when you're talking about human population, particularly clinical trials, is the small sample size. And traditionally, this has been very difficult from a statistical perspective.AI can help with trying to put things in context. AI can build complex models that distill knowledge from large datasets. But then the question is, what where do we get large datasets? And and there's a lot of research going on in the epidemiological literature on leveraging cross sectional cohorts. Observational cohorts borrow information from larger scale studies involving human populations to inform the clinical trials.
AI needs tremendously large datasets to understand population heterogeneity. And to figure out how to distill and combine multiple data layers, right now, this is a hard problem. In AI, trying to integrate multiple datasets, you are seeing some advances in the context of image and text processing. If you've heard of stable diffusion and generating artwork from text, those are some of the cutting edge techniques and machine learning. But figuring out how to scale that to multiple modalities, that's a big, outstanding question.
Families [impacted by ASD] are desperate. And there are no good solutions. So, figuring out how to make how to make impact, that's very important. In terms of communicating [the benefits of this research]. Being honest is important, figuring out how to get past the jargon and connecting.
Where Do We Start to Make a Difference?
I actually prefer to think about where do you end? Because if you think about that, it can help you prioritize. And this is honestly one of the big motivations for AI. These systems are infinitely complex, and it's very difficult to figure out where to even start. But if you have a good end point, namely improvement in GI symptoms, behavioral symptoms, sleep—80% of autistic children struggle with sleep. Sleep is one the lowest hanging fruits. We know that slips linked to brain developed. We know that most of these families [impacted by ASD] don't sleep. The parents too.I think those are very real attainable objectives. If you can start thinking about measuring those end outcomes and trying to try to improve those measurements, while designing clinical trials, I think that would help with prioritizing.
Links for the Episode
- Take the
