Preprint Seven replicated genomic associations of myalgic encephalomyelitis/chronic fatigue syndrome: a biobank study, 2026, Slaughter, Ponting et al.

[wigglethemouse]

I see it as a positive that this work was done and published. We know previous work on UK biobank data only identified one variant for females. It seems to matter how you identify the patient cohort. What surprised me was replication in a completely unrelated study in the US. 5 of the replicated variants in this paper were from the All of Us program data.

There are details on patient and control selection in the Stanford Symposium video so look out for that when published on youtube for those of us that can't fully digest a paper..

I'd be interested to learn more from @Chris Ponting about the comparison of using the 5x10-8 level vs the FDR level for analysing genes. I didn't quite understand that in the presentation given that Chris has always talked about the importance of meeting the 5x10-8 p value threshold.

 

[forestglip]

What surprised me was replication in a completely unrelated study in the US. 5 of the repilcated variants in this paper were from the All of Us program data.

I was surprised too. They started from a discovery case cohort less than a tenth the size of DecodeME, and presumably with a more heterogenous cohort (though maybe not?) and were able to detect significant variants that replicated in a different country.

 

[ScoutB]

Same. Wondering now if this could all be self consistent if the cohorts in this new study are people genetically predisposed to a less severe illness and one with an older onset, thus these variants are *less* aggressively selected against evolutionarily and were easier to find even in a smaller gwas. (mostly inspired by this paper jnmaciuch shared a while ago on complex traits being selected against in a way that makes them hard to find in studies)

 

[forestglip]

CSMD1 was a significant gene in two PrecisionLife studies:

Genetic Risk Factors for ME/CFS Identified using Combinatorial Analysis, 2022, Das et al
(In supplementary table 7, disease signature 68.)


Identification of Novel Reproducible Combinatorial Genetic Risk Factors for [ME] in [DecodeME Cohort] and Commonalities with [LC], 2026, Sardell et al
(CSMD1 was one of 259 candidate core genes.)

CSMD1 might play a role in complement regulation, schizophrenia, and synaptic regulation:

The Schizophrenia-Associated Gene, CSMD1, Encodes a Brain-Specific Complement Inhibitor (Baum, 2018, Dissertation)

a) a genome-wide significant association [with schizophrenia] signal localizes to CSMD1,

b) the gene encodes a protein composed almost entirely of protein domains conserved in inhibitors/regulators of the complement cascade, i.e., CUB (Complement, Urchin-EGF, Bone-Morphogenic protein) and Sushi (aka CCP or Complement Control Proteins) domains,

c) protein fragments of CSMD1 can directly and indirectly inhibit the activation of C4 and C3 in vitro,

d) CSMD1 mRNA is brain-enriched.

Despite these intriguing observations, however, little was known about the about the normal functions of CSMD1 in neural tissues. Here, we show that CSMD1 protein is highly brain enriched, is most pronounced in cortex, present at synapses, and is expressed predominantly by neurons.

Using a human stem cell line and a mouse each genetically lacking CSMD1, we present evidence that CSMD1 regulates complement activation on neural cells and in vivo, and that loss of Csmd1 abrogates the development of a complement-and-pruning-dependent neural circuit.

Together, these data corroborate CSMD1 as a brain-specific complement inhibitor.

PubMed results for CSMD1 include findings related to neurodevelopmental disorders, cancer progression, schizophrenia, mammary duct formation, fertility, malaria, and more.

 

[Chris Ponting]

I see it as a positive that this work was done and published. We know previous work on UK biobank data only identified one variant for females. It seems to matter how you identify the patient cohort. What surprised me was replication in a completely unrelated study in the US. 5 of the replicated variants in this paper were from the All of Us program data.

There are details on patient and control selection in the Stanford Symposium video so look out for that when published on youtube for those of us that can't fully digest a paper..

I'd be interested to learn more from @Chris Ponting about the comparison of using the 5x10-8 level vs the FDR level for analysing genes. I didn't quite understand that in the presentation given that Chris has always talked about the importance of meeting the 5x10-8 p value threshold.

Thank you for this question. The p<5x10-8 threshold is, indeed, the "industry-standard" used for GWAS. The reason why we did not apply this threshold is that it was set using results from more conventional GWAS methods, specifically linear/logistic models. TarGene is different (in ways described here: https://pmc.ncbi.nlm.nih.gov/articles/PMC12479317/) so we should not unthinkingly apply the same threshold. Thus, we decided to apply a conventional false discovery rate threshold, and then seek to replicate associations meeting this threshold in independent cohorts. Hope this helps!

 

[Evergreen]

Congrats to Slaughter et al., including @Chris Ponting and @Simon M, on this publication.

My question is...will there be a blog, @Simon M?

 

[Nightsong]

A fascinating result.

I haven't had as good a read through yet as I'd like, but just flagging that both the Sharepoint sumstats link and the paper-specific GitHub repo links are broken.

 

[ME/CFS Science Blog]

Interesting analysis, thanks to the authors!

I find it strange that the analysis of 1,268 ME/CFS cases found more significant hits (SNPS with p < 5 * 10^-8) than DecodeME with 15,579 cases (and double the amount of controls). The effective sample size was only around 5000. This doesn't fit with the low heritability that DecodeME found.

My first guess would be that a genomic inflation factor of 1.09 might not be modest for this tiny sample size and that might include some bias. Would be interesting to hear what the LDSC intercept was.

Another potential reason might be that the 15 significant SNPs were quite rare, with a MAF of around 1%, so they might depend on just a small number of individuals.

 

[forestglip]

Another potential reason might be that the 15 significant SNPs were quite rare, with a MAF of around 1%, so they might depend on just a small number of individuals.

Maybe low minor allele frequency is the reason. I just did a quick calculation of MAF for cases and controls from supplementary table 1:

Spoiler: Calculated MAF

I couldn't fit all 176 variants in the post, so this is only the 100 most significant. The first four columns are copied from Supplementary Table 1, and the MAF columns are calculated by me from the genotype counts.

rsID	p-value	case_geno_counts	ctrl_geno_counts	Case MAF	Control MAF
rs115349603	7.2E-10	A/A:1250 A/G:15 G/G:0	A/A:110030 A/G:2946 G/G:9	0.006	0.013
rs4525595	2.6E-09	G/G:1248 G/A:17 A/A:1	G/G:109283 G/A:3474 A/A:21	0.008	0.016
rs117290148	3.4E-09	G/G:1252 G/C:13 C/C:0	G/G:110099 G/C:2794 C/C:20	0.005	0.013
rs75552905	3.5E-09	A/A:1247 A/G:18 G/G:0	A/A:109651 A/G:3252 G/G:29	0.007	0.015
rs36069412	4.5E-09	A/A:1231 A/G:34 G/G:0	A/A:106566 A/G:6209 G/G:88	0.013	0.028
rs79025431	5.3E-09	T/T:1213 T/C:52 C/C:1	T/T:104770 T/C:8008 C/C:146	0.021	0.037
rs4271043	6.7E-09	G/G:1251 G/A:14 A/A:0	G/G:109998 G/A:2845 A/A:15	0.006	0.013
rs116623105	1.2E-08	G/G:1248 G/T:16 T/T:0	G/G:109246 G/T:3234 T/T:0	0.006	0.014
rs77608528	1.6E-08	T/T:1250 T/G:17 G/G:0	T/T:109826 T/G:3255 G/G:23	0.007	0.015
rs28552103	1.6E-08	G/G:1237 G/A:19 A/A:1	G/G:109091 G/A:3311 A/A:26	0.008	0.015
rs73117935	1.7E-08	C/C:1247 C/T:19 T/T:1	C/C:109694 C/T:3268 T/T:20	0.008	0.015
rs117742180	3.3E-08	A/A:1200 A/G:65 G/G:0	A/A:103162 A/G:9628 G/G:208	0.026	0.044
rs7914	3.6E-08	A/A:805 A/G:424 G/G:34	A/A:67460 A/G:39534 G/G:5831	0.195	0.227
rs115569355	3.8E-08	A/A:1238 A/C:25 C/C:1	A/A:108666 A/C:4178 C/C:35	0.011	0.019
rs72926310	4.9E-08	A/A:1239 A/G:25 G/G:0	A/A:108314 A/G:4518 G/G:0	0.010	0.020
rs56114267	6.3E-08	C/C:1240 C/G:26 G/G:1	C/C:108280 C/G:4686 G/G:52	0.011	0.021
rs12118405	8.0E-08	T/T:1234 T/C:28 C/C:0	T/T:108011 T/C:4703 C/C:49	0.011	0.021
rs73046799	8.3E-08	A/A:1252 A/G:15 G/G:0	A/A:110022 A/G:2994 G/G:20	0.006	0.013
rs80011338	1.2E-07	A/A:1241 A/C:25 C/C:0	A/A:108673 A/C:4164 C/C:34	0.010	0.019
rs1207120	1.4E-07	G/G:889 G/A:360 A/A:16	G/G:78259 G/A:31393 A/A:3117	0.155	0.167
rs78613483	1.5E-07	A/A:1253 A/G:13 G/G:0	A/A:110622 A/G:2457 G/G:12	0.005	0.011
rs75647405	1.5E-07	A/A:1246 A/C:15 C/C:0	A/A:109626 A/C:3117 C/C:20	0.006	0.014
rs72769308	1.9E-07	G/G:1195 G/A:27 A/A:0	G/G:104346 G/A:4424 A/A:43	0.011	0.021
rs12551209	2.0E-07	A/A:948 A/G:297 G/G:13	A/A:80761 A/G:29375 G/G:2638	0.128	0.154
rs6502314	2.3E-07	A/A:895 A/G:356 G/G:15	A/A:77748 A/G:31845 G/G:3091	0.152	0.169
rs78303500	2.4E-07	T/T:1252 T/C:15 C/C:0	T/T:109829 T/C:3119 C/C:19	0.006	0.014
rs116048825	2.5E-07	T/T:1126 T/C:17 C/C:0	T/T:99037 T/C:3133 C/C:21	0.007	0.016
rs73164132	2.9E-07	A/A:1240 A/G:24 G/G:0	A/A:108528 A/G:4396 G/G:42	0.009	0.020
rs117553493	2.9E-07	T/T:1239 T/C:26 C/C:1	T/T:108411 T/C:4629 C/C:45	0.011	0.021
rs10491334	3.0E-07	T/T:821 T/C:417 C/C:28	T/T:71806 T/C:36477 C/C:4755	0.187	0.203
rs2360997	3.0E-07	C/C:14 C/G:367 G/G:881	C/C:3097 C/G:31056 G/G:78674	0.156	0.165
rs41554714	3.3E-07	C/C:1250 C/T:15 T/T:0	C/C:110146 C/T:2751 T/T:24	0.006	0.012
rs80303305	3.9E-07	G/G:1252 G/A:14 A/A:0	G/G:110198 G/A:2782 A/A:11	0.006	0.012
rs7166097	4.1E-07	C/C:38 C/T:441 T/T:787	C/C:6079 C/T:39859 T/T:66918	0.204	0.230
rs35758227	4.3E-07	G/G:1235 G/A:30 A/A:0	G/G:108418 G/A:4522 A/A:38	0.012	0.020
rs80204084	4.6E-07	C/C:1246 C/A:16 A/A:1	C/C:110205 C/A:2542 A/A:16	0.007	0.011
rs79030614	5.1E-07	C/C:1240 C/T:25 T/T:0	C/C:108837 C/T:4048 T/T:31	0.010	0.018
rs75312050	5.4E-07	G/G:1251 G/A:15 A/A:0	G/G:110495 G/A:2513 A/A:36	0.006	0.011
rs116706541	5.8E-07	T/T:1235 T/C:13 C/C:1	T/T:109585 T/C:2316 C/C:17	0.006	0.010
rs75318300	5.9E-07	C/C:878 C/A:362 A/A:16	C/C:79247 C/A:30269 A/A:2929	0.157	0.161
rs72741654	6.5E-07	T/T:1234 T/C:29 C/C:0	T/T:107826 T/C:4718 C/C:58	0.011	0.021
rs1534442	6.7E-07	C/C:743 C/T:482 T/T:37	C/C:67352 C/T:39657 T/T:5876	0.220	0.228
rs73195242	6.8E-07	T/T:1237 T/C:23 C/C:0	T/T:108347 T/C:4244 C/C:71	0.009	0.019
rs116738745	7.4E-07	G/G:1250 G/A:16 A/A:0	G/G:110035 G/A:2855 A/A:16	0.006	0.013
rs11606204	7.5E-07	A/A:1181 A/G:83 G/G:3	A/A:101824 A/G:10937 G/G:272	0.035	0.051
rs117849226	7.5E-07	G/G:1245 G/A:20 A/A:1	G/G:109417 G/A:3513 A/A:29	0.009	0.016
rs73198318	8.3E-07	A/A:1243 A/G:23 G/G:0	A/A:109154 A/G:3758 G/G:34	0.009	0.017
rs2647545	8.4E-07	A/A:906 A/G:345 G/G:14	A/A:78863 A/G:30994 G/G:3075	0.147	0.164
rs66558288	8.7E-07	G/G:855 G/A:388 A/A:21	G/G:74393 G/A:34569 A/A:3929	0.170	0.188
rs118000645	9.5E-07	C/C:1250 C/T:14 T/T:0	C/C:110356 C/T:2511 T/T:11	0.006	0.011
rs77435165	1.0E-06	A/A:1245 A/G:14 G/G:0	A/A:110398 A/G:2270 G/G:10	0.006	0.010
rs11825381	1.1E-06	G/G:804 G/A:428 A/A:24	G/G:73068 G/A:34897 A/A:4214	0.189	0.193
rs62385806	1.2E-06	T/T:1234 T/C:29 C/C:1	T/T:108094 T/C:4888 C/C:43	0.012	0.022
rs116937777	1.2E-06	T/T:1248 T/A:16 A/A:0	T/T:110183 T/A:2725 A/A:12	0.006	0.012
rs7517260	1.2E-06	A/A:58 A/G:542 G/G:665	A/A:8379 A/G:44120 G/G:60323	0.260	0.270
rs117447397	1.4E-06	G/G:1247 G/T:18 T/T:0	G/G:110470 G/T:2536 T/T:21	0.007	0.011
rs6664253	1.7E-06	A/A:882 A/G:359 G/G:19	A/A:75568 A/G:33637 G/G:3640	0.158	0.181
rs116924494	1.8E-06	A/A:1245 A/G:20 G/G:0	A/A:109492 A/G:3400 G/G:28	0.008	0.015
rs12121687	1.8E-06	T/T:1242 T/G:25 G/G:0	T/T:108622 T/G:4251 G/G:47	0.010	0.019
rs117430951	1.8E-06	C/C:1237 C/T:26 T/T:0	C/C:108613 C/T:4177 T/T:14	0.010	0.019
rs77080860	1.9E-06	C/C:1248 C/T:15 T/T:0	C/C:110049 C/T:2830 T/T:12	0.006	0.013
rs143786057	2.0E-06	C/C:1254 C/T:13 T/T:0	C/C:110176 C/T:2690 T/T:22	0.005	0.012
rs77893777	2.1E-06	A/A:1244 A/G:21 G/G:1	A/A:109281 A/G:3617 G/G:29	0.009	0.016
rs17611784	2.1E-06	T/T:1232 T/C:30 C/C:0	T/T:108134 T/C:4537 C/C:38	0.012	0.020
rs16822632	2.1E-06	A/A:1251 A/G:15 G/G:0	A/A:110675 A/G:2307 G/G:12	0.006	0.010
rs3130602	2.2E-06	T/T:1241 T/C:25 C/C:0	T/T:108891 T/C:4147 C/C:44	0.010	0.019
rs11735109	2.2E-06	G/G:1230 G/A:27 A/A:1	G/G:108176 G/A:4431 A/A:44	0.012	0.020
rs78771625	2.2E-06	T/T:1241 T/G:26 G/G:0	T/T:108864 T/G:4190 G/G:40	0.010	0.019
rs72939083	2.3E-06	G/G:1249 G/A:17 A/A:0	G/G:110146 G/A:2879 A/A:17	0.007	0.013
rs117675373	2.4E-06	C/C:1246 C/A:15 A/A:0	C/C:110237 C/A:2604 A/A:10	0.006	0.012
rs118141754	2.4E-06	A/A:1248 A/G:16 G/G:0	A/A:109867 A/G:3048 G/G:13	0.006	0.014
rs150321190	2.4E-06	T/T:1126 T/C:19 C/C:0	T/T:99196 T/C:3256 C/C:23	0.008	0.016
rs62044679	2.4E-06	T/T:1250 T/C:16 C/C:0	T/T:110506 T/C:2426 C/C:12	0.006	0.011
rs80288492	2.4E-06	T/T:1209 T/G:56 G/G:0	T/T:104817 T/G:8038 G/G:158	0.022	0.037
rs11800918	2.4E-06	T/T:1211 T/C:56 C/C:0	T/T:104772 T/C:8045 C/C:175	0.022	0.037
rs57900343	2.4E-06	G/G:1198 G/A:22 A/A:0	G/G:105161 G/A:3329 A/A:21	0.009	0.016
rs78934565	2.5E-06	A/A:1249 A/C:17 C/C:0	A/A:109999 A/C:3065 C/C:9	0.007	0.014
rs1534443	2.5E-06	G/G:749 G/A:480 A/A:37	G/G:67429 G/A:39689 A/A:5879	0.219	0.228
rs73064070	2.6E-06	A/A:1247 A/G:20 G/G:0	A/A:109846 A/G:3205 G/G:23	0.008	0.014
rs3128933	2.7E-06	A/A:1239 A/G:25 G/G:0	A/A:108752 A/G:4162 G/G:10	0.010	0.019
rs9276969	2.7E-06	A/A:1242 A/C:25 C/C:0	A/A:108863 A/C:4153 C/C:47	0.010	0.019
rs12529462	2.8E-06	A/A:1143 A/G:117 G/G:6	A/A:97810 A/G:14733 G/G:534	0.051	0.070
rs78616260	2.9E-06	A/A:1199 A/G:62 G/G:3	A/A:104675 A/G:8222 G/G:159	0.027	0.038
rs116960995	2.9E-06	T/T:1239 T/C:24 C/C:1	T/T:108899 T/C:3985 C/C:28	0.010	0.018
rs77968392	3.1E-06	T/T:1239 T/C:28 C/C:0	T/T:108564 T/C:4427 C/C:33	0.011	0.020
rs17144566	3.1E-06	T/T:809 T/C:431 C/C:26	T/T:73700 T/C:35085 C/C:4241	0.191	0.193
rs150621117	3.2E-06	G/G:1253 G/A:13 A/A:1	G/G:110655 G/A:2394 A/A:15	0.006	0.011
rs117755521	3.4E-06	G/G:1231 G/A:28 A/A:0	G/G:107698 G/A:4330 A/A:29	0.011	0.020
rs72721736	3.4E-06	T/T:1243 T/C:24 C/C:0	T/T:109458 T/C:3605 C/C:27	0.009	0.016
rs34748721	3.5E-06	G/G:1172 G/T:84 T/T:6	G/G:100894 G/T:11506 T/T:311	0.038	0.054
rs74764125	3.5E-06	G/G:1232 G/A:17 A/A:0	G/G:108814 G/A:3047 A/A:1	0.007	0.014
rs117370067	3.6E-06	A/A:1237 A/G:27 G/G:0	A/A:108566 A/G:4308 G/G:31	0.011	0.019
rs74963073	3.7E-06	T/T:1246 T/C:21 C/C:0	T/T:109834 T/C:3213 C/C:24	0.008	0.014
rs7654969	3.8E-06	G/G:1249 G/T:13 T/T:0	G/G:110151 G/T:2358 T/T:11	0.005	0.011
rs77489603	3.8E-06	T/T:1249 T/C:17 C/C:0	T/T:110233 T/C:2739 C/C:8	0.007	0.012
rs77602254	3.8E-06	A/A:1193 A/G:66 G/G:1	A/A:103484 A/G:9119 G/G:214	0.027	0.042
rs396567	3.9E-06	C/C:76 C/A:353 A/A:830	C/C:5427 C/A:38675 A/A:68544	0.201	0.220
rs79893826	3.9E-06	T/T:1225 T/C:35 C/C:2	T/T:107376 T/C:5338 C/C:94	0.015	0.024
rs10520406	4.0E-06	C/C:1200 C/T:63 T/T:2	C/C:104128 C/T:8624 T/T:193	0.026	0.040
rs12145346	4.1E-06	A/A:942 A/C:293 C/C:13	A/A:90280 A/C:19834 C/C:1082	0.128	0.099

[TR]

Out of 176 significant variants, ~54% had an MAF < 0.02 based on allele counts of cases and controls combined.

I tried comparing to DecodeME by filtering to all the variants that cross a similar p-value threshold as this study (I used p<1.5e-5). Out of 3645 variants below this p-value in DecodeME, only 6 variants have an MAF < 0.02 (0.1%).

So maybe these are mostly artifactual findings related to low MAF. Here is a paper talking about this:

Data quality control in genetic case-control association studies (Anderson et al., 2010)

The final step in per-marker QC is to remove all SNPs with a very low MAF. Typically, a MAF threshold of 1–2% is applied, but studies with small sample sizes may require a higher threshold. The small sizes of the heterozygote and rare-homozygote clusters make these variants difficult to call using current genotype-calling algorithms; they frequently present as false positives in case-control association tests. Furthermore, even when accurately called, association signals observed at these rare SNPs are less robust because they are driven by the genotypes of only a few individuals. Given that the power to detect association with rare variants is so low23, their removal does not overly affect the study.

 

[ScoutB]

Interesting. How does that square with studies like sequenceME that specifically look for rare variants? I guess the genotyping used will be more reliable than what that quote is discussing for one thing.

Still really interesting there was a such a difference at all between the two studies.

only 6 variants have an MAF < 0.02 (0.1%)

does a MAF of 0.01 mean 0.1% or 1%?

 

[forestglip]

I find it strange that the analysis of 1,268 ME/CFS cases found more significant hits (SNPS with p < 5 * 10^-8) than DecodeME with 15,579 cases (and double the amount of controls).

Is that right? I count 287 variants in DecodeME with p<5e-8. Based on the supplementary table in this study, there are only 15 variants that pass this threshold.

Edit: The 15 doesn't include the imputed variants in this study. Looking at Supplementary Figure 2, it looks like that brings it up to around 23 variants.

 

[forestglip]

Interesting. How does that square with studies like sequenceME that specifically look for rare variants? I guess the genotyping used will be more reliable than what that quote is discussing for one thing.

Yes, more reliable genotyping, plus rare variants within a single gene are typically grouped together to increase statistical power. So the statistical test is per gene, not per variant.

does a MAF of 0.01 mean 0.1% or 1%?

Sorry, that wasn't clear. In DecodeME, 6 out of 3645 variants with p<1.5e-5 (about 0.1% of variants below this threshold) had an MAF less than 0.02 (or 2%).

 

[ScoutB]

Ah cool, makes sense

 

[ME/CFS Science Blog]

Is that right? I count 287 variants in DecodeME with p<5e-8. Based on the supplementary table in this study, there are only 15 variants that pass this threshold.

I was thinking about (FUMA-defined independent) regions because many of those 287 variants were highly correlated and represented the same signal. It seems that there were 12 independent loci in this discovery cohort reaching the significance threshold, while there were only 6 in the main DecodeME analysis

 

[forestglip]

I was thinking about (FUMA-defined independent) regions because many of those 287 variants were highly correlated and represented the same signal. It seems that there were 12 independent loci in this discovery cohort reaching the significance threshold, while there were only 6 in the main DecodeME analysis

Oh ok, makes sense. Yes, that's interesting.

Edit: Maybe the fact that the independent hits have few variants within each locus suggests that the most of the ones that passed the threshold were artifacts, because otherwise, I think we'd expect many other variants in the region that are in LD with each top hit to also pass the threshold, as we see in DecodeME, where there were more total SNPs per independent locus.

 

[wigglethemouse]

My takeaway from the presentation at the Stanford symposium was that the replicated findings are of most interest.

Does anyone have an inkling what the statistics of getting a 1% MAF hit by random in one dataset replicated in another dataset are? Gut feel tells me it should be rather rare.

 

[forestglip]

My takeaway from the presentation at the Stanford symposium was that the replicated findings are of most interest.

Does anyone have an inkling what the statistics of getting a 1% MAF hit by random in one dataset replicated in another dataset are? Gut feel tells me it should be rather rare.

Here are the MAFs I calculated for just the seven replicated variants listed in Table 1:

rsID	Nearest gene	MAF (Full cohort)
rs115186419	ETV5	0.021
rs73175505	CSMD1	0.024
rs261902	BICD1	0.192
rs117553493	CLYBL	0.020
rs72741654	RORA	0.017
rs74963073	GRIN2A	0.013
rs76847656	SBK1	0.013

6 out of 7 of these have an MAF right around 1-2%. So I wonder if the reason most of the replicated variants had low MAFs could be because the low MAF variants were where most of the false positives happened.

---

Separately, maybe I'm misunderstanding something, but Table 1 and the supplementary spreadsheet say the rs261902 variant had log10p = 4.85. But Supplementary Figure 7 seems to put it at around log10p = 2.

 

[Nightsong]

One small observation I'd make from reading through so far - I haven't looked at the data itself - is that the control cohort wasn't merely defined by an absence of ME/CFS associated signals; it was also defined by them providing

a ‘good’ or ‘excellent’ overall health rating

or, for AoU, providing

a rating of ‘Excellent’, ‘Very Good’, or ‘Good’

to survey questions on physical & mental health. That could result in loci relating to morbidity in a broader sense becoming salient; I note a locus associated with ETV5 (which in turn was previously linked to adiposity), was found. I wonder if the authors considered performing a sensitivity analysis with controls lacking in specific ME/CFS associated markers but not otherwise selected for >= good health?

For rs76847656, I wonder why the discussion didn't touch on the nearest gene, SBK1? Emphasising TUFM's eQTL & immune-related biology is understandable, although I notice that SBK1 appeared in a small NK expression paper from 2016:

Dysregulation of Protein Kinase Gene Expression in NK Cells from Chronic Fatigue Syndrome/Myalgic Encephalomyelitis Patients

 

[forestglip]

I noticed that in 174 out of the 176 significant variants in supplementary spreadsheet S1, the minor allele is protective against ME/CFS (lower minor allele frequency in ME/CFS). Maybe the consistency is an indication of a batch effect.

 

[ME/CFS Science Blog]

the control cohort wasn't merely defined by an absence of ME/CFS associated signals; it was also defined by them providing "a ‘good’ or ‘excellent’ overall health rating"

I was wondering about this too because the goal is to compare ME/CFS cases agains the rest of the population, not necessarily those with good or excellent health. So it could have caused an additional selection bias.

I assume it was to make sure that none of the controls accidentally had ME/CFS for example undiagnosed cases or people who forgot to mention CFS in the interview. But given how uncommon ME/CFS already is, those cases would have formed only a very small percentage of the control group and would likely have little influence on the results.

On the other hand requiring controls to have good or excellent overall health might have changed and shifted the control population in all sorts of ways (e.g higher educated, richer, less bad luck in life, more optimistic, etc), that didn't occur in the ME/CFS population. The result may be that DNA difference not only reflect ME/CFS but those other differences as well.