Stanford Community Symposium 2018: Phair, Metabolic traps, Tryptophan trap

First, in case it's not already clear to anyone, I'm the author of the IDO metabolic trap hypothesis. I think most S4ME members are aware of this, but I just want to be sure.

This is already a long thread. If asked to summarize what Chris has said, I'd list two main points. 1) there are no IDO2 SNPs that rise to genome-wide significance in the UK Biobank cohort of ME/CFS patients, and 2) a new study from the gnomAD group at the Broad Institute has raised the possibility that the Y359STOP mutation is actually a part of an MNV (a multinucleotide variant) - that's two bases in a single codon. The preprint on BioRxiv that Chris cited hasn't been peer-reviewed yet, but it looks (to my relatively amateur genomics eyes) carefully done with appropriate attention to detail. I'm still reading this paper.

@lunarainbows asks, reasonably, how it can be that Chris finds no support in the Biobank data and yet we the authors still think the IDO metabolic trap is a good idea. This is the key question. One of the main reasons scientists publish papers is to find out what other scientists think. Then we try to find the places where we agree and those where we disagree. Then we look for data to try to resolve the disagreements.

Chris will, I'm sure, respond to this question, but I think we are in complete agreement that none of the four known or predicted damaging SNPs is more prominent in ME/CFS patients than in the general CEU population. So I agree with Chris that a GWAS on IDO2 SNVs will not identify a statistically significant association. But since there are multiple damaging mutations, we can ask, instead, do ME/CFS patients have one or more damaging mutations in IDO2 more often than expected based on the allele frequencies in the CEU population and the corresponding Hardy-Weinberg proportions? The simplest way to ask this question statistically is to ask how many ME/CFS patients have no damaging mutations in IDO2? In Ron Davis' Stanford cohort of ME/CFS patients, all diagnosed by physicians who see many such patients, we have 66 patients for whom we have either WGS or targeted sequencing of IDO2 or both. The Hardy-Weinberg proportions predict we should find 8 people with no damaging mutations among those 66. Instead we find zero. I'm writing this paper; in fact I probably should be doing that instead of writing this post, but in my view PR and S4ME are vital forums for this disease, and I welcome the scientific interchange. So, in other words, every ME/CFS patient in our cohort has at least one damaging mutation in IDO2. And Chi-square says that's very unlikely to have happened by chance.

I'm still working on the gnomAD MNV argument about Y359X, but a quick look at our data tells me that the hypothesis would survive loss of this variant as one of our damaging variants.

One more thing. These variants were the hint that led me to the trap idea. There's a lot more depth to the hypothesis so please don't stop reading after Table 1.

 

I think I've read all that Craig has written on catastrophe theory and ME/CFS, but I haven't seen a particular part of cell biology or physiology identified as the site of the catastrophe. There is a lot in common between catastrophe theory and bistability, so if anyone knows what part of biology Craig is considering, please let me know.

 

I hope I have not mislead/ confused. I don't t know if Craig Robertson has pitched this molecularly. It simply strikes me as an explanation for the relapse and remitting aspect and the similarities re catastrophe/ instability are striking, even to a non mathematician.

Interestingly similar mechanisms in climate change models. Even the most basic " daisyworld" by James lovelock in 1980s demonstrated a " switch" rather than gradual change at key points . It seems intrinsic to systems regulation.

 

It's thought that IDO2 is the ancient gene and IDO1 evolved from it by duplication and divergence. They are next to one another on chromosome 8. You are correct that the trap could exist in the absence of IDO2. The trap is inherent in the substrate inhibition of IDO1 and my seeing the high Km for Trp in IDO2 as a kind of back up for IDO1 at high Trp is just the mind of a modeler at work. Still, if the two enzymes are co-expressed in a single cell type, the trap scenario is a natural conclusion. Even relative expression levels could produce the trap scenario.

 

It is intrinsic, but only to nonlinear systems. Of course, as you point out, almost every interesting system is regulated, and typically that makes them nonlinear. There are three interesting features of nonlinear systems that are never seen in linear systems: 1) bistability, 2) sustained oscillations (sometimes called limit cycles), and 3) deterministic chaos.

Indeed, climate models are the origin of everyone's thinking about these issues. Good point.

 

Right. You probably remember that I was looking only for common variants because I felt (and still believe) that common variants are required to account for the existence of ME/CFS epidemics or outbreaks.

 

@Trish It's true that the original group of 20 in Ron Davis' OMF-funded big data study were identified as severely ill. But we now have sequence data for 66 ME/CFS patients covering the entire spectrum of disease severity, and every one of them has at least one damaging mutation in IDO2. The average is 1.8 damaging mutations in IDO2.

 

We've been working on these experiments for a year. They are more difficult than I expected, but they will probably be the subject of our third paper (after the Hardy-Weinberg results). Lots more discussion of this on PR.

 

As I've said elsewhere in this thread, IDO2, despite its numbering, is the ancient gene. It's the one in our evolutionary ancestors. If I had to guess, IDO1 evolved in order to yield a higher affinity (lower Km) for Trp. In other words, IDO1 evolved to compete for Trp.

You're not the only person who thinks IDO2 is being abandoned by evolution. I've read assertions in the literature that it's on its way to elimination from the genome. But the fact that it remains suggests it's helpful under some circumstances.

 

Theoretically, you could find a trap by simulation of differential equation models.

Experimentally, you might observe the consequences of a trap by finding some of your cells in one state, and other cells in another state when all the external conditions were the same. Of course, you'd have to be wise enough not to think the experiment was a failure.

 

What a wonderful breath of fresh air that is, compared to the BPS approach seen for ME/CFS.

 

interesting link to roles and mechanisms of IDO 1 , IDO2 and TDO
https://pubs.rsc.org/en/content/articlelanding/2017/mt/c7mt00105c#!divAbstract

 

Has the role of NO linked gene expression been considered relative to instigatng bistability in IDO1/ IDO2 pathways?
My understanding is that NO inhibits IDO activity by preventing oxidixation of TRP
Is this a link to the sepsis similarities?

 

Thank you @RDP. This is very helpful indeed. I am only now left with a couple of questions: (1) What is the probability that your analysis would have alighted upon *any* gene with this burden of damaging mutations among all that you considered? (In other words, after correcting for multiple tests, what was the q-value from the Chi-squared test?) (2) Have you done, or are you planning, a replication test, using an independent cohort? (As you know, this requires only a single test, so the burden of multiple testing correction, is very small.)

(For others: In essence, my questions are all about whether a single gene, here IDO2, has been pinpointed merely by chance. There are 20,000 genes that could have been pinpointed. At a p-value threshold of 5%, 5% of these (i.e. 1000) would be highlighted by chance; at a threshold of 0.5%, 100 would be highlighted by chance, at p<0.05%, 10 would be highlighted, and so on.)

@RDP My MNV argument is not valid. I was previously put right on this! See previous posts. Thanks all.

 

This reminds me a bit of when I used to work developing PID controllers, and had some involvement in self-tuning of them (but only peripheral - the self-tuning math was too heavy for my capabilities). Even though only two variables were being being tuned (D always being set to I/4), the difficulties were considerable, especially when controlling non-linear processes. An informal visualisation was akin to a ball rolling over a surface with lots of dips/wells in, with a few of them being good places for the tuning to be, and wanting the ball to find and stay in. But so often it can find its way to the wrong dip, akin to very bad tuning, but nonetheless happy to stay there (i.e. trapped in an unhealthy place) without some further influence.

 

@RDP It might be interesting to apply for the 50k exomes' data from UK Biobank. ~200 people with self-reported ME/CFS are expected to have had their exomes sequenced. The preliminary analysis (https://www.biorxiv.org/content/biorxiv/early/2019/03/09/572347.full.pdf) does not report an association between IDO2 and CFS/ME but it might be significant in a single test as a replication of your initial finding. Just a thought.

 

Do you mean a p value? If not, what is a q value??

 

Thanks. A q-value is the expected fraction of 'false discoveries' in the full set of statistical tests used. https://en.wikipedia.org/wiki/Q-value_(statistics)
As Wikipedia puts it: "Just as the p-value gives the expected false positive rate obtained by rejecting the null hypothesis for any result with an equal or smaller p-value, the q-value gives the expected pFDR [positive false discovery rate] obtained by rejecting the null hypothesis for any result with an equal or smaller q-value."
For example, assume that we perform 1000 tests and (as is conventional in single hypothesis testing) apply a significance threshold of 0.05 (p<0.05) then due to random chance we expect 5% of the results to appear significant (P < 0.05), yielding 50 false positives. Instead, the q-value is the false discovery rate (FDR) or the proportion of false positives among all positive results. We want q to be as low as possible. Hope this helps.

 

@RDP

This is very interesting to me, because of how my level of functioning switched from an average of 500-600 steps per day (3 year duration) to my current 4000 - 4500 average steps in a short space of time in early 2016.

With a few minor waves up & down, I have maintained, but not been able to improve, that new level of activity to date. (I am still limited by PEM, by having to walk slowly, by struggling in busy environments and with standing or sitting upright for normal durations of time - but the improvements to my well-being have been life changing for me and my family.)

It is no secret now, that I took an ARV (anti-retroviral) drug for a year. The switch I experienced happened in a very short space of time about 4 months after starting the drug. It was like a brake had been released! My stamina improved dramatically over a two week period, but it took much longer for my strength to catch up to that improved stamina.

This seems to me to support the idea of a trap of some sort being reversed, allowing me to do more. My initial low strength I attribute to the deconditioning of being so physically inactive for 3+ years. (My strength is still not at pre-illness levels.)

It is also obvious to me, that deconditioning was not the limiting factor in my 500 steps days. Once that limit was lifted, I improved and so I did more, until I hit a new limit. Sadly it seems, I still have something else holding my physical abilities back.

So, I’m hoping that I can maintain this level, and that any further switch is not a backward one.

This post is not a recommendation for the ARV I took. I mention it only because of the “switch” I experienced, which I thought might be interesting in the context of this conversation.

PS I have Fitbit stepcount data for this period on file, so the numbers are not guesswork. ;-)