key: cord-0292720-pl54jz3h
authors: Guo, C.; Che, X.; Briese, T.; Allicock, O.; Yates, R. A.; Cheng, A.; Ranjan, A.; March, D.; Hornig, M.; Komaroff, A. L.; Levine, S.; Bateman, L.; Vernon, S. D.; Klimas, N. G.; Montoya, J. G.; Peterson, D. L.; Lipkin, W. I.; Williams, B. L.
title: Deficient butyrate-producing capacity in the gut microbiome of Myalgic Encephalomyelitis/Chronic Fatigue Syndrome patients is associated with fatigue symptoms
date: 2021-10-28
journal: nan
DOI: 10.1101/2021.10.27.21265575
sha: bea7811e4d63ac69c3f5bfe369017d58bc8e1d1b
doc_id: 292720
cord_uid: pl54jz3h

Abstract Background: Myalgic Encephalomyelitis/Chronic Fatigue Syndrome (ME/CFS) is a complex, debilitating disease of unknown cause for which there is no specific therapy. Patients suffering from ME/CFS commonly experience persistent fatigue, post-exertional malaise, cognitive dysfunction, sleep disturbances, orthostatic intolerance, fever and irritable bowel syndrome (IBS). Recent evidence implicates gut microbiome dysbiosis in ME/CFS. However, most prior studies are limited by small sample size, differences in clinical criteria used to define cases, limited geographic sampling, reliance on bacterial culture or 16S rRNA gene sequencing, or insufficient consideration of confounding factors that may influence microbiome composition. In the present study, we evaluated the fecal microbiome in the largest prospective, case-control study to date (n=106 cases, n=91 healthy controls), involving subjects from geographically diverse communities across the United States. Results: Using shotgun metagenomics and qPCR and rigorous statistical analyses that controlled for important covariates, we identified decreased relative abundance and quantity of Faecalibacterium, Roseburia, and Eubacterium species and increased bacterial load in feces of subjects with ME/CFS. These bacterial taxa play an important role in the production of butyrate, a multifunctional bacterial metabolite that promotes human health by regulating energy metabolism, inflammation, and intestinal barrier function. Functional metagenomic and qPCR analyses were consistent with a deficient microbial capacity to produce butyrate along the acetyl-CoA pathway in ME/CFS. Metabolomic analyses of short-chain fatty acids (SCFAs) confirmed that fecal butyrate concentration was significantly reduced in ME/CFS. Further, we found that the degree of deficiency in butyrate-producing bacteria correlated with fatigue symptom severity among ME/CFS subjects. Finally, we provide evidence that IBS comorbidity is an important covariate to consider in studies investigating the microbiome of ME/CFS subjects, as differences in microbiota alpha diversity, some bacterial taxa, and propionate were uniquely associated with self-reported IBS diagnosis. Conclusions: Our findings indicate that there is a core deficit in the butyrate-producing capacity of the gut microbiome in ME/CFS subjects compared to healthy controls. The relationships we observed among symptom severity and these gut microbiome disturbances may be suggestive of a pathomechanistic linkage, however, additional research is warranted to establish any causal relationship. These findings provide support for clinical trials that explore the utility of dietary, probiotic and prebiotic interventions to boost colonic butyrate production in ME/CFS.

sr-IBS compared with healthy controls without sr-IBS (Mann-Whitney U test, p adj = 0.038 162 and p adj = 0.047, respectively) ( Figure 1C and Supplementary Figure 1C) . No 163 differences were found between stratified groups for observed species (Supplementary 164 Figure 1D ). These results indicate that differences in alpha diversity between ME/CFS 165 and controls may be dependent on comorbid sr-IBS, rather than ME/CFS. 166

To further evaluate the relationship between alpha diversity and ME/CFS, we 167 employed linear regression (Shannon and evenness) and negative binomial regression 168 (observed species) with alpha diversity metrics as outcome variables and ME/CFS and 169 sr-IBS as predictors, adjusting for covariates of site of sampling, sex, BMI, race/ethnicity, 170 age, antibiotic usage 6-12 weeks prior to sample collection, probiotic supplement use, 171

and prebiotic supplement use ( Table 2 ). The interaction term between ME/CFS and sr-172 IBS was explored but was not included in the final model as it was not significant in any 173 of the regression models for alpha diversity. ME/CFS was not a significant predictor of 174 evenness, Shannon diversity, or observed species metrics. However, sr-IBS status was 175 negatively associated with evenness (bEst = -0.03, 95% CI: -0.06-0.00, p = 0.032) and 176 showed a trending negative association with Shannon diversity ((bEst = -0.03, 95% CI: -177 0.06-0.00, p = 0.054). Of other covariates, only subject age showed a significant positive 178 association with all three alpha diversity metrics. These results suggest that differences 179 in alpha diversity are associated with sr-IBS rather than ME/CFS and highlight the 180 controls. Model 2 represented the stratified comparison of ME/CFS without sr-IBS vs. 208 healthy controls without sr-IBS. Model 3 represented the stratified comparison of ME/CFS 209 with sr-IBS vs. healthy controls without sr-IBS. Model 4 represented the stratified 210 comparison of ME/CFS with sr-IBS vs. ME/CFS without sr-IBS (note; Model 4 did not 211 produce any significant taxa differences and is therefore not shown). For each of Models 212 1-3 ME/CFS status was the variable for comparison, while sr-IBS status was the variable 213 for comparison in Model 4, and each model is adjusted for covariates. 214

At the species-level, Model 1 identified four species after adjusting for multiple 215 comparisons that differentiated ME/CFS from healthy controls. Model 2 identified two 216 species that differentiated ME/CFS without sr-IBS from healthy controls without sr-IBS. 217

Model 3 identified fifteen species that differentiated ME/CFS with sr-IBS from healthy 218 controls without sr-IBS (Figure 2A and Supplementary Table 1A ). The relative 219 abundance of two species, Eubacterium rectale and Faecalibacterium prausnitzii were 220 lower in the ME/CFS group in all three models. These results suggest that deficiency of 221 these species in ME/CFS are independent of sr-IBS and other covariates. The most 222 differentially abundant species were found in Model 3, where fifteen species were 223 associated with ME/CFS with sr-IBS compared to healthy controls without sr-IBS. The 224 majority of identified species (11/15) in Model 3 showed no overlap with other models 225 suggesting that sr-IBS has an independent association with changes in the microbiota. 226

The species associated with ME/CFS with sr-IBS included decreased relative abundance 227 of Alistipes putredinis, Dorea longicatena, Odoribacter splanchnicus, and 228

Lachnospiraceae bacterium GAM79, among others, and increased relative abundance of 229

Clostridium bolteae and Flavonifractor plautii. It was also apparent from our analysis of 230 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. ; https://doi.org/10.1101/2021. 10.27.21265575 doi: medRxiv preprint Model 1 that several bacterial species from the same genera such as Eubacterium and 231

Roseburia were significantly associated with ME/CFS before FDR adjustment 232 (Supplementary Table 1A) . Accordingly, we performed the same GAMLSS-BEZI 233 analyses at the genus level. 234

At the genus-level six bacterial genera were significantly associated with ME/CFS 235 from all three models after FDR adjustment ( Figure 2B and Supplementary Table 1B) . 236

The majority of these genera had lower relative abundance in the ME/CFS group 237 compared to healthy controls, including Eubacterium, Faecalibacterium, Dorea, 238

Roseburia and Gemmiger. Only Lachnoclostridium was increased in relative abundance 239 in the ME/CFS groups. 240

At both the species-and genus-level the primary bacteria identified in all three 241 models, and thus associated with ME/CFS independent of sr-IBS, contained the most 242 abundant and common butyrate-producing bacteria (BPB) in the human gut, including the 243 species E. rectale, F. prausnitzii and the genera Eubacterium, Faecalibacterium and 244

While shotgun metagenomics can provide compositional information on individual fecal 248 bacteria, relative abundance of each taxon is dependent on the relative abundance of all 249 bacterial taxa in the microbiota and is not strictly quantitative. In order to assess 250 quantitative differences in fecal BPB, we carried out qPCR analysis using assays 251 targeting the 16S rRNA genes of Roseburia-Eubacterium genera and the species F. 252 prausnitzii to evaluate the quantity of these bacterial taxa per gram of feces. Compared 253 to healthy controls, ME/CFS subjects had significantly fewer 16S rRNA gene copies of 254

Roseburia-Eubacterium genera (Mann-Whitney U test, p = 0.0008) ( Figure 2C ) and the 255 species F. prausnitzii (Mann-Whitney U test, p = 0.004) ( Figure 2D ) per gram of feces. In 256 stratified analyses, Roseburia-Eubacterium 16S copies/gram of feces were lower in both 257 ME/CFS subjects without sr-IBS (Mann-Whitney U test, p adj = 0.029) and ME/CFS 258 subjects with sr-IBS (Mann-Whitney U test, p adj = 0.011) compared to healthy controls 259 without sr-IBS ( Figure 2E) . Similarly, lower quantities of F. prausnitzii 16S/gram of feces 260 were found when comparing ME/CFS without sr-IBS (Mann-Whitney U test, p adj = 0.079) 261

and ME/CFS with sr-IBS (Mann-Whitney U test, p adj = 0.018) compared with healthy 262 controls without sr-IBS, though only a trend was observed in the former after adjusting for 263 multiple comparisons ( Figure 2F) . 264

As total bacterial load in stool has never been evaluated in ME/CFS, we carried 265 out qPCR using a broad range 16S rRNA gene-targeting assay. In contrast to the 266 quantitatively measured deficiencies in BPB in ME/CFS, ME/CFS subjects had higher 267 quantities of total 16S copies/gram of feces compared to healthy controls (Mann-Whitney 268 U test, p < 0.0001) ( Figure 2G ). Stratified analyses revealed that ME/CFS subjects 269 without sr-IBS had higher quantities of total bacterial 16S/gram of feces compared with 270 healthy controls without sr-IBS (Mann-Whitney U test, p adj < 0.0001), while other group 271 comparisons did not differ after adjusting for multiple comparisons ( Figure 2H) . 272

Since we measured both the quantity of BPB 16S and total bacterial 16S, we also 273 calculated the relative abundance of Roseburia-Eubacterium and F. prausnitzii (i.e., 274

Roseburia-Eubacterium 16S/total bacterial 16S). For both Roseburia-Eubacterium and F. 275 prausnitzii, ME/CFS subjects had lower calculated relative abundance compared with 276 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. ; https://doi.org/10.1101/2021. 10 .27.21265575 doi: medRxiv preprint healthy controls (Mann-Whitney U test, p = 0.0007 and p = 0.009, respectively) 277 ( Supplementary Figure 2A, B) . In stratified analyses, Roseburia-Eubacterium 278 calculated relative abundance was lower in ME/CFS subjects without sr-IBS compared 279 with healthy controls without sr-IBS (Mann-Whitney U test, p adj = 0.004), while other 280 comparisons did not reach significance after adjusting for multiple comparisons 281 (Supplementary Figure 2C) . Similar results were obtained for F. prausnitzii calculated 282 relative abundance, although only a trend toward lower F. prausnitzii was observed 283 comparing ME/CFS subjects without sr-IBS with healthy controls without sr-IBS (Mann-284

Whitney U test, p adj = 0.054) (Supplementary Figure 2D) . 285

To further assess the relationships between quantitative measures of BPB and 286 total bacteria (outcome variables) in feces with ME/CFS status (main predictor), 287 generalized linear regression models with a Gamma distribution with log link were fit 288 adjusting for covariates ( Table 3) Table 2) . 299

The findings indicate that, despite having higher quantities of total bacterial 16S in 300 feces, ME/CFS patients had significantly lower quantities of important BPB such as the 301 genera Roseburia and Eubacterium and the species F. prausnitzii. 302 303 Fecal bacterial functional metagenomic pathways differ between ME/CFS and 304 healthy controls 305

Functional metagenomic analysis was carried out using GOmixer, a human gut 306 microbiome-specific metabolic pathway analysis tool [40] . Comparison between ME/CFS 307 subjects and healthy controls revealed nine global metabolic processes that differed 308 between the gut microbiome of ME/CFS subjects and controls ( Figure 3A and 309 Supplementary Table 3A) . Two processes, butyrate and sulfate metabolism, were 310 deficient in ME/CFS. Seven processes, CO2 metabolism, monosaccharide degradation, Table 3B ). The specific module for 315 butyrate production via transferase (MF0116) was deficient in ME/CFS, consistent with 316 deficiencies in BPB we observed based on differential abundance and qPCR analyses. 317

Other modules that were deficient in the ME/CFS microbiome included superoxide 318 reductase (MF0132) and sorbitol degradation (MF0073). Modules that were enriched in 319 ME/CFS included lactate production (MF0119), pyruvate dehydrogenase complex 320 (MF0083), ribose degradation (MF0060), and menaquinone production (MF0133). 321

Analysis of GOmixer gut-brain modules also revealed deficient metagenomic content for 322 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. Table 3C ). Enriched gut-brain modules in ME/CFS included Menaquinone synthesis 325 (vitamin K2) I (MGB040), GABA synthesis III (MGB022), and Isovaleric acid synthesis I 326 (KADH pathway) (MGB034). 327

As differential abundance analysis of taxa, qPCR, and functional metagenomic 328 analysis implicated deficiencies in BPB and functional capacity of the gut microbiome to 329 produce butyrate in ME/CFS subjects relative to healthy controls, we investigated the 330 metagenomic gene content for genes along the four pathways of gut bacterial butyrate 331 production: the acetyl-CoA pathway, the glutarate pathway, the lysine pathway, and the 332 4-aminobutyrate pathway ( Figure 3B) . We aligned bacterial reads to a curated database 333 of genes involved in butyrate production along the four pathways [41] . Comparing the 334 metagenomic content (Counts per Million, CPM) of each gene along the acetyl-CoA 335 pathway showed deficient gene content for nearly every gene along this pathway in 336 ME/CFS relative to healthy controls (Mann-Whitney U test, p < 0.05 for all genes except 337 thl and buk) ( Figure 3C) . Quantitatively, the majority of bacteria in the intestine encode 338 and utilize but (butyryl-CoA:acetate CoA transferase) rather than buk (butyrate kinase) as 339 the terminal gene in the acetyl-CoA pathway to produce butyrate. In fact, most bacteria 340 of the Eubacterium, Roseburia, and Faecalibacterium genera (those we found to be 341 deficient in ME/CFS) only encode the but gene [41] . While nearly all genes of the acetyl-342

CoA pathway differed between ME/CFS and healthy controls, only terminal genes of the 343 glutarate pathway (gcdA and gcdB), and both early and terminal genes of the lysine 344 pathway (kamA, kamD, kamE, atoA, atoD) were deficient in ME/CFS compared to healthy 345 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) Figure 4A-C) . Even in stratified analyses by sr-IBS status, most genes 347 along the acetyl-CoA pathway remained significantly depleted in the microbiome of 348 ME/CFS subjects without sr-IBS compared with healthy controls without sr-IBS (Mann-349 Whitney U test, p adj = 0.071, 0.038, 0.035, 0.049, 0.020 for bhbd, cro, etfA, etfB, and but, 350 respectively) ( Figure 3D ). In contrast, no genes along the glutarate, lysine, or 4-351 aminobutyrate pathways differed significantly among groups in stratified analyses 352

We further evaluated metagenomic gene content for genes along butyrate 354 pathways using generalized linear regression with a Gamma distribution with log link and 355 adjusting for covariates. Even after adjusting for covariates, ME/CFS was a significant 356 predictor of most genes along the acetyl-CoA pathway (all except thl and buk) ( Table 4) . respectively. In addition to ME/CFS status, BMI was associated with only but gene 362 content. In contrast, only a few genes in the glutarate pathway were associated with 363 ME/CFS status (gctA, hgCoAdC and gcdA) in regression analyses (Supplementary 364 Table 4A ). Only the terminal genes, atoA and atoD, were associated with ME/CFS status 365 in the lysine pathway. Interestingly, most genes along the lysine pathway were 366 independently associated with race/ethnicity (Supplementary Table 4B ). None of the 367 genes in the 4-aminobutyrate pathway were significantly associated with ME/CFS status 368 (Supplementary Table 4C) . 369 370

But, as the dominant terminal gene in the bacterial acetyl-CoA pathway for butyrate 372 production, can serve as an indicator gene for the overall butyrate-producing capacity in 373 the human gut [42] . We quantitated the but gene in fecal samples from ME/CFS and 374 control subjects by qPCR. The overall but copies/gram feces were lower in ME/CFS 375 compared to control subjects (Mann-Whitney U test, p = 0.0003) ( Figure 3E ). Further, 376 stratified analyses by sr-IBS status revealed lower but copies/gram feces in ME/CFS 377 without sr-IBS (Mann-Whitney U test, p adj = 0.014) and ME/CFS with sr-IBS (Mann-378 Whitney U test, p adj = 0.008) compared to healthy controls without sr-IBS ( Figure 3F) . No 379 differences were found comparing ME/CFS subjects with and without sr-IBS. These 380 results suggest that deficient but gene quantity in the feces of ME/CFS patients is 381 independent of sr-IBS status. 382

We also assessed the relative abundance of but gene copies to total bacterial 16S 383 in feces based on qPCR. As with the total quantity of but gene/gram of feces, the but 384 gene calculated relative abundance was lower in ME/CFS subjects compared to healthy 385 controls in univariate analyses (Mann-Whitney U test, p = 0.0003) (Supplementary 386 Figure 5A ). In stratified analyses by sr-IBS status, but gene relative abundance was lower 387 observed in ME/CFS subjects with sr-IBS compared with healthy controls without sr-IBS 390 (Mann-Whitney U test, p adj = 0.096) (Supplementary Figure 5B) . 391

To further assess the relationships between quantitative measures of the but gene 392 in feces (outcome variable) with ME/CFS status (main predictor) generalized linear 393 regression models with a Gamma distribution with log link were assessed, adjusting for 394 covariates ( Table 3) . Even after adjusting for covariates in the model, ME/CFS was 395 associated with but gene quantities (FC=0.51, 95% CI: 0.28-0.93, p = 0.028). Of the 396 covariates, only antibiotic usage in the prior 6-12 weeks showed an independent 397 association with but gene quantity. Similarly, lower calculated relative abundance of the 398 but gene was significantly associated with ME/CFS status (FC=0.60, 95% CI: 0.39-0.91, 399 p = 0.017) in regression and only Hispanic race showed an independent association with 400 but gene relative abundance in the model (Supplementary Table 2) . 401

These results confirm quantitative deficiency of the most important terminal gene 402 in the acetyl-CoA pathway of butyrate production in the feces of ME/CFS subjects 403 compared to healthy controls. 404 405

Given the strong evidence indicating reduced abundance and quantity of BPB and 407 reduced metagenomic capacity for producing butyrate, we measured SCFAs in all fecal 408 samples from ME/CFS and healthy control subjects using gas chromatography-mass 409 spectrometry. The fecal concentration of both acetate (Mann-Whitney U test, p = 0.004) 410

and butyrate (Mann-Whitney U test, p < 0.0001) were lower in ME/CFS compared to 411 healthy controls, while propionate was unchanged, in univariate analyses ( Figure 3G) . In 412 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted October 28, 2021. ; https://doi.org/10.1101/2021.10.27.21265575 doi: medRxiv preprint stratified analyses by sr-IBS status, acetate was lower in ME/CFS subjects with sr-IBS 413 compared to healthy controls without sr-IBS (Mann-Whitney U test, p adj = 0.010), but did 414 not differ in ME/CFS subjects without sr-IBS. In contrast, butyrate was lower in both 415 ME/CFS subjects without sr-IBS (Mann-Whitney U test, p adj < 0.0001) and with sr-IBS 416 (Mann-Whitney U test, p adj = 0.002) compared to healthy controls without sr-IBS ( Figure  417 

To further assess the relationships between quantitative measures of SCFAs in 419 feces (outcome variables) with ME/CFS status (main predictor), generalized linear 420 regression models with Gamma distribution with log link were fitted, adjusting for 421 covariates ( Table 5) . As with all our regression analyses, we evaluated the interaction 422 term between ME/CFS and sr-IBS. Unlike our prior regressions, the ME/CFS*sr-IBS 423 interaction term was significant. In order to better understand the nature of the 424 relationships among SCFA levels and ME/CFS and sr-IBS, we evaluated stratified 425 regression models, adjusted for covariates (bottom 3 rows in Table 5 4, comparing ME/CFS subjects with sr-IBS and ME/CFS subjects without sr-IBS, did not 435 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted October 28, 2021. ; https://doi.org/10.1101/2021.10.27.21265575 doi: medRxiv preprint relative abundance of bacterial taxa that were found to be enriched in ME/CFS 459 (Lachnoclostridium) or ME/CFS with sr-IBS (C. bolteae, F. plautii, Flavonifractor). 460

We also evaluated the relationship among our measured variables and fatigue 461 scores based on the five dimensions of the Multidimensional Fatigue Inventory (MFI) on 462 the whole dataset (all cases + all controls, n=197) ( Figure 4B ). In this case, the majority 463 of significant correlations among measured variables and fatigue scores after FDR 464 correction were inverse, as measured variables that tended to be deficient in ME/CFS 465 were associated with higher fatigue scores (more severe fatigue), especially for general 466 and physical fatigue and reduced activity. The relative abundance and qPCR quantitation 467 of bacterial species and the but gene, metagenomic content of genes along the acetyl-468

CoA butyrate pathway, and levels of butyrate were inversely correlated with general 469 fatigue, physical fatigue and reduced activity. 470

As correlations in the whole dataset may reflect reduced levels of measured 471 variables and higher fatigue scores in individuals with ME/CFS vs. healthy controls 472 (Supplementary Table 6 ), we also evaluated associations between measured variables 473 and fatigue scores in ME/CFS subjects alone (n=106) ( Figure 4C ). Although fewer 474 correlations remained significant in the cases after FDR adjustment, various bacterial taxa 475 including F. prausnitzii, the genus Faecalibacterium, Roseburia inulinivorans, Roseburia 476 intestinalis, the genus Roseburia, and the genus Coprococcus correlated inversely with 477 either general fatigue, or physical fatigue, or both. The qPCR quantity for the but gene 478 correlated inversely with general fatigue and the qPCR quantity of F. prausnitzii in stool 479 correlated inversely with general fatigue, physical fatigue and reduced activity ( Figure  480 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted October 28, 2021. ; https://doi.org/10.1101/2021.10.27.21265575 doi: medRxiv preprint 4C). Thus, the lower the abundance or quantity of these BPB, the more severe the fatigue 481 dimension scores in ME/CFS. 482

We also determined whether similar relationships existed among our measured 483 variables and fatigue scores in healthy control subjects alone (n=91). In contrast to 484 ME/CFS subjects, no significant correlations were found in healthy controls after FDR 485 adjustment (as such, no correlogram is shown). 486

As sr-IBS could further influence these relationships, we evaluated the 487 relationships among measured variables and fatigue scores only in ME/CFS patients 488 without sr-IBS (n=71) and only in ME/CFS patients with sr-IBS (n=35). No significant 489 relationships were found for either stratified group after FDR adjustment (as such, no 490 correlograms are shown). 491

Overall, these findings suggest that deficiencies in BPB and their metabolic 492 pathways are correlated with the severity of fatigue symptoms, but these associations are 493 only found in ME/CFS, not healthy controls. Our analyses of the fecal microbiome in ME/CFS subjects and healthy controls 497 matched for age, sex, geography, and socioeconomic status indicated significant 498 differences in composition, function and metabolism. In a systematic review of studies on 499 gut dysbiosis in ME/CFS, Du Preez et al. concluded that intrinsic and extrinsic factors that 500 can influence microbiome composition should be better controlled for in case-control 501 studies [24] . One such factor that may be important in the context of ME/CFS and the 502 microbiome is IBS co-morbidity, which occurs at high prevalence in ME/CFS compared 503 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted October 28, 2021. having an IBS diagnosis (35/106; 33%) than healthy controls (3/91; 3.3%). Both our 506 current and prior fecal shotgun metagenomic study support the notion that IBS co-507 morbidity must be carefully considered in future ME/CFS microbiome studies. In fact, 508 differences in gut microbiome alpha diversity between ME/CFS subjects and healthy 509 controls appears to be largely driven by sr-IBS co-morbidity in ME/CFS, and individuals 510 with ME/CFS and sr-IBS had a range of distinct bacterial species with differential relative 511 abundance compared to healthy controls (i.e., A. putredinis, C. bolteae, F. plautii, D. 512 longicatena) that were not found when comparing ME/CFS without sr-IBS and healthy 513 controls. Thus, at least some microbiome differences in ME/CFS are confounded by IBS 514 co-morbidity. In contrast, microbiota beta diversity differed significantly between ME/CFS 515 and healthy controls; a difference that was independent of sr-IBS status. 516 517 Species and taxa linked to ME/CFS: Relative and Absolute Differences 518 GAMLSS-BEZI models, adjusted for important covariates (including sr-IBS), 519 identified differentially abundant fecal bacterial taxa between ME/CFS subjects and 520 healthy controls. These analyses identified two species (E. rectale and F. prausnitzii) and 521 six genera (Eubacterium, Faecalibacterium, Dorea, Roseburia, Gemmiger, and 522 Lachnoclostridium) that differed in relative abundance in ME/CFS compared to healthy 523 control subjects. Both the species E. rectale and F. prausnitzii and the genera 524

Eubacterium, Faecalibacterium and Roseburia are prominent BPB in the human GI tract 525

[41]. All had lower relative abundance in feces from ME/CFS subjects compared to 526 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted October 28, 2021. prausnitzii. Both the fecal load of these BPB as well as their relative abundances were 529 deficient in fecal samples from ME/CFS patients. 530

An additional, unexpected and novel finding from our qPCR analyses was that 531 ME/CFS patients have higher total bacterial 16S rRNA genes/gram of feces than healthy 532 controls. Total fecal bacterial load is infrequently evaluated in microbiome studies and 533

little is still known about factors that may impact total bacterial load. However, it is well 534 documented that antibiotics can have a dramatic impact on bacterial load [45, 46] . In this 535 study, we controlled for antibiotics, both in our study design and statistical analyses. Thus, 536

it is unlikely that antibiotics can explain the differences in bacterial load. 537

Dietary factors that may impact bacterial load include low fermentable oligo-, di-, 538 mono-saccharides and polyols (FODMAPs) [47] . However, we cannot address this 539 possibility because detailed dietary information was not collected from subjects. While it 540 is possible that subject diets differ between ME/CFS subjects and controls, we note that 541 more ME/CFS subjects reported taking prebiotic fiber supplements (11/106; 10.4%) than 542 healthy controls (2/91; 2.2%). Prebiotic fibers typically stimulate the growth and activity of 543 BPB. The observation that ME/CFS subjects had lower levels of BPB and butyrate and 544 higher total bacterial load even with adjustment of regression models for prebiotic fiber 545 use, suggests that some factor, either associated with the pathobiology or symptoms of 546 ME/CFS, rather than prebiotic fiber intake is selectively influencing these microbiome 547 changes. 548 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted October 28, 2021. Other gastrointestinal disturbances such as severe acute malnutrition with acute 549 diarrhea and inflammatory bowel disease can also alter total bacterial load in feces or 550 intestinal mucosa, respectively [48, 49] . However, subjects in this study were not acutely 551 malnourished and only two cases (2/106) reported a formal diagnosis of IBD; no controls 552 had a diagnosis of IBD. Small intestinal bacterial overgrowth is frequently associated with 553 functional gastrointestinal disorders such as IBS [50] . However, we are not aware of any 554 studies that have shown that fecal bacterial load is increased in IBS, and our findings 555 suggest that increased fecal bacterial load is associated with ME/CFS, independent of sr-556 IBS. Further, despite the significant association of fecal bacterial load and ME/CFS, it 557 remains unclear whether higher fecal bacterial load is representative of bacterial 558 overgrowth in the intestine or a higher rate of bacterial washout or loss of adherent 559 mucosa-associated bacteria. 560 561 Specific butyrate-production deficiency in ME/CFS 562

Our functional metagenomic analysis found that the overall bacterial capacity for 563 butyrate production is deficient in ME/CFS. To delve deeper into the specific pathways of 564 butyrate production that are deficient, we examined the bacterial gene content of the four 565 bacterial butyrate production pathways. Only the gene content for the acetyl-CoA pathway 566 was deficient in ME/CFS. The acetyl-CoA pathway is the dominant pathway of butyrate 567 production in the human gut. Whereas this pathway for butyrate production is fueled by 568 carbohydrates, the other three pathways (glutarate, lysine and 4-aminobutyrate 569 pathways) are fueled by proteins [41] . Counts per million of genes in the acetyl-CoA 570 pathway were substantially higher than for genes in the other three pathways. 571 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted October 28, 2021. ; https://doi.org/10.1101/2021.10.27.21265575 doi: medRxiv preprint

The majority of bacteria that utilize the acetyl-CoA pathway typically encode either 572 the but gene or the buk gene to complete the terminal reaction of butyrate production. 573

The but gene is dominant in the human gut [41] . We found that whereas the but gene was 574 deficient in ME/CFS, the buk gene was not. Several of the bacteria that we found at lower 575 relative abundance and quantity in ME/CFS, including F. prausnitzii, E. rectale, Roseburia 576 spp. and Eubacterium spp. are known to encode the genes for the acetyl-CoA pathway 577 of butyrate [41] . Our functional metagenomic analyses showed that ME/CFS subjects are 578 specifically deficient in BPB that rely on the terminal but gene to produce butyrate. qPCR 579 and metabolomic analyses corroborated our functional metagenomic findings by 580

confirming that the quantity of the bacterially encoded but genes and levels of butyrate 581 are lower in feces of ME/CFS patients than in healthy controls. 582

In addition to butyrate, ME/CFS subjects have reduced quantities of fecal acetate, 583 although the degree of deficiency is less substantial than that of butyrate. Metabolic cross-584 feeding interactions between bacterial groups are likely important determinants of the 585 composition of the intestinal microbial community. Acetate produced by bacterial 586 fermentation of carbohydrates or acetogens is utilized by BPB (those using the acetyl-587

CoA pathway) for butyrate production, and BPB like F. prausnitzii, Roseburia spp. and 588

Eubacterium spp. may grow poorly in the absence of acetate [51-53]. Thus, net acetate 589 deficiency may contribute to deficient BPBs and butyrate and may be associated with 590 known acetate producers such as the genera Dorea and Fusicatenibacter, which were 591 also lower in relative abundance in ME/CFS compared to healthy controls [54, 55] . In 592 contrast to butyrate and acetate, propionate was only reduced in patients with sr-IBS. 593

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. 

The relative abundance and/or quantity of fecal BPB are inversely correlated with 614 magnitude of symptoms, as reflected by the Multidimensional Fatigue Inventory, in 615 ME/CFS subjects. The magnitude of general fatigue and/or physical fatigue was greater 616 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted October 28, 2021. ; https://doi.org/10.1101/2021.10.27.21265575 doi: medRxiv preprint in ME/CFS subjects that had lower relative abundance or quantity of F. prausnitzii, 617

Roseburia spp., Ruminococcus, and Coprococcus. The quantity of fecal F. prausnitzii 618 based on qPCR was inversely correlated with general fatigue, physical fatigue and 619 reduced activity. F. prausnitzii is an important member of the human microbiota, 620

representing 5% of the microbiota in healthy adults. The functional importance of F. 621 prausnitzii in the intestine may extend beyond its role as a BPB to include additional anti-622 inflammatory effects through its production of microbial anti-inflammatory molecule 623 is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. ; https://doi.org/10.1101/2021.10.27.21265575 doi: medRxiv preprint Roseburia species), and ultimately increases the levels of SCFAs [74, 75] . The 640 mechanisms by which physical activity may influence the microbiome are not fully 641 elucidated but such mechanisms may relate to the impact of exercise on blood flow to the 642 intestine, gut barrier integrity, transit time in the large intestine, enterohepatic circulation 643 of bile acids, contraction of skeletal muscles and metabolic flux, or raising of core body 644 temperature [74, 75] . Given the debilitating nature of ME/CFS, we acknowledge that the 645 deficiency in butyrate and BPB may arise as a result of the symptoms of ME/CFS (i.e., 646 fatigue, post-exertional malaise, pain) and the behavioral adjustment to those symptoms 647 (i.e., reduced physical activity). Nonetheless, even if deficient butyrate-producing capacity 648 is a consequence rather than a direct cause of symptoms, such a deficiency could both 649 exacerbate ME/CFS-specific symptoms or potentiate the risk for the development of CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. ; https://doi.org/10.1101/2021.10.27.21265575 doi: medRxiv preprint

The initial cohort consisted of 177 ME/CFS cases and 177 healthy controls prescreened 665 at five geographically-diverse ME/CFS clinics across the USA (Incline Village, NV; Miami, 666 FL; New York, NY; Palo Alto, CA; Salt Lake City, UT) as part of a National Institutes of 667

Health-sponsored R56 study. ME/CFS cases met the requirements of both the 1994 CDC 668

[1] and the 2003 Canadian consensus criteria [9] for ME/CFS. Control participants were 669 matched to ME/CFS cases based on geographical/clinical site, sex, age, race/ethnicity, 670 and date of sampling (±30 days). 671

The CDC Criteria require that cases have fatigue persisting for greater than six 672 months that is clinically-evaluated, persistent or relapsing, and which meets five criteria: 673 is of new onset, is not the result of ongoing exertion, is not alleviated by rest, is made 674 worse by exertion, and results in substantial reduction in previous levels of activity. The 675 CDC criteria additionally require the concurrent occurrence of at least four of the following 676 symptoms for at least six consecutive months: sore throat, tender cervical or axillary 677 lymph nodes, muscle pain, multiple joint pain without swelling or redness, headaches of 678 new or different type, unrefreshing sleep, post-exertional malaise, and impaired memory 679 or concentration. The Canadian consensus criteria impose additional restrictions, 680 requiring at least two neurologic/cognitive manifestations, and at least one clinical feature 681 from two of the following three categories: autonomic manifestations, neuroendocrine 682 manifestations, and immune manifestations. 683

Eligible cases must also have had a diminished or restricted capacity to work, 684

reported a viral-like prodrome prior to onset of ME/CFS, and met a low-score threshold 685 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. ;

for two out of the following three domains measured by the self-reported Short Form-36 686

General Health Survey (SF-36): vitality <35, social functioning <62.5, role-physical <50. 687

Additional exclusion criteria for both cases and controls included disorders or treatments 688 resulting in immunosuppression, and antibiotic use within six weeks prior to the baseline 689 assessment. Based on these screening criteria, we excluded five ME/CFS cases and one 690 control participant prior to the baseline assessment. 691

In the current study, a nested sub-cohort was established for participants with 692 complete survey data collection and biospecimen (stool) collection at the first and fourth 693 (final) timepoints for the overall study. Participants were frequency-matched on key 694 demographic elements to ensure similarity between the nested cohort and full cohort. 695

This sub-cohort is comprised of 106 ME/CFS cases and 91 healthy controls that met 696 these criteria; the derivation of this sub-cohort is outlined in Supplementary Figure 6 . All 697 subjects provided written consent in accordance with study protocols approved by the 698 Columbia University Medical Center Institutional Review Board (IRB). 699 700

All subjects completed standardized screening instruments to assess medical history, 702

family medical history, current medication use, symptom scores, and 703 demographic/lifestyle information, as well as a baseline SF-36. All subjects underwent a 704 screening blood draw to determine that they had normal values in the following three 705 laboratory tests from Quest Diagnostics: complete blood count with differential, 706 comprehensive metabolic panel, thyroid stimulating hormone (TSH). 707 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. On the medical history questionnaire, participants were asked to self-report if they 718 received an IBS diagnosis (sr-IBS) by a physician and the date of diagnosis. In the 719 analytic sub-cohort, 35 out of the 106 (33.0%) ME/CFS cases reported sr-IBS, while only 720 3 out of the 91 control subjects (3.3%) reported sr-IBS. The use of probiotic and prebiotic 721 supplements was specifically included in the "current medication use" data collection 722 instrument, which determined the frequency and recent use of these products. Any 723 subjects that indicated consumption of these supplements daily or a few times a week 724 and reported using them within the last week prior to their study visit would be endorsed 725 for that type of supplement. Participants also reported any antibiotic use that occurred 726 outside of the six-week window for study eligibility. 727

The MFI consists of a 20-item self-reported questionnaire that evaluates five 728 dimensions of fatigue: general, physical and mental fatigue, reduced activity, and reduced 729 motivation [76] . The scoring for this instrument was transformed into a 0-100 scale to 730 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. ; https://doi.org/10.1101/2021.10.27.21265575 doi: medRxiv preprint allow for comparisons between dimensions: a score of 100 was equivalent to maximum 731 disability or severity and a score of zero was equivalent to no disability or disturbance. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. ; https://doi.org/10.1101/2021.10.27.21265575 doi: medRxiv preprint

Shot-gun metagenomics sequencing was carried out on DNA extracts obtained from 197 755 fecal samples (106 ME/CFS cases and 91 healthy controls). For Illumina library 756 preparation, genomic DNA was sheared to a 200-bp average fragment length using a 757

Covaris E210 focused ultrasonicator. Sheared DNA was purified and used for Illumina demultiplexed raw FastQ files were adapter trimmed using Cutadapt [77] . Adaptor 764 trimming was followed by the generation of quality reports using FastQC and filtering with 765

PrinSEQ [78] . Host background levels were determined by mapping the filtered reads 766 against the human genome using Bowtie2 mapper [79] . After the step of host subtraction, 767

25.1 ± 9.0 (mean ± sd) million reads per sample remained on average. Non-host reads 768 were subjected to Kraken2 for taxonomy classification [80] . Kraken2 matches each K-769 mer within a query sequence to the lowest common ancestor (LCA) of all genomes in the 770 database containing the given K-mer. Our Kraken2 local database included all 16,799 771 fully sequenced and representative bacteria species genomes in the RefSeq database 772 (December 2018). The species-level taxonomy abundances were estimated using 773

Bracken, which is recommended to perform a Bayesian estimation of taxonomy 774 abundance after the use of Kraken2 [80, 81] . Structural zeros in the abundance table 775

were further identified using the program Analysis of Microbiome Data in the Presence of 776 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. ; https://doi.org/10.1101/2021.10.27.21265575 doi: medRxiv preprint Excess Zeros version II (ANCOM-II) [82] . The three groups we have compared in our 777 study include ME/CFS with sr-IBS, ME/CFS without sr-IBS, and healthy controls. The 778 taxa presenting as structural zeros in all three categories were eliminated from the 779 dataset. The data was rarefied prior to diversity analyses using a depth of 500,000 reads. 780

Diversity metrics (alpha diversity: Shannon index, pielou's evenness, and observed otus; 781 beta-diversity: Bray-Curtis dissimilarity) were calculated and plotted using the core-782 diversity plugin and the emperor plugin within QIIME2 [83] . were first amplified by conventional PCR from human stool samples. PCR products were 797 run on 1% agarose gels and purified using the QIAgen Gel Extraction kit. Purified products 798

were ligated into pGEM T-Easy vector (Promega), transformed into DH5α competent cells, 799 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. ; https://doi.org/10.1101/2021.10.27.21265575 doi: medRxiv preprint and cultured on Luria Bertani plates with ampicillin. Colonies were inoculated into Luria 800 broth with ampicillin (5 ml), and plasmids were extracted with Pure Link Plasmid 801

Extraction Kit (Invitrogen). After verifying that plasmid sequences had 100% nucleotide 802 similarity to the target bacterial taxa or the bacterial but gene, plasmids were linearized 803 with the restriction enzyme SphI, and 10-fold serial dilutions were generated ranging from 804 5 x 10 6 to 5 copies (note for Eubacterium/Roseburia standards, two plasmids were 805 Eubacterium/Roseburia 16s and Faecalibacterium 16s assays). For but gene the cycle 819 conditions were 50°C for 2 min, 95° C for 10 min, and 45 cycles at 95°C for 15 s, 53°C 820 for 1 min 45 s and 77°C for 30 s (data collection). All samples were run in duplicate and 821 averaged for each assay. Absolute quantity was determined based on the weight of fecal 822 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Kruskal-Wallis test was significant at p < 0.05, then between group significance was 867 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. [88] was used to assess differences in microbiota beta diversity between ME/CFS and 876 healthy controls, and was adjusted for covariates (sr-IBS status, site of sampling, sex, 877 BMI, race/ethnicity, age, antibiotic usage 6-12 weeks prior to sample collection, probiotic 878 supplement use, and prebiotic supplement use). Two-group comparisons were 879 considered significant at p < 0.05. For stratified analyses (more than two groups), the 880 FDR was controlled by Bonferroni correction and an adjusted p-value (p adj ) < 0.05 was 881 considered significant. 882 883 Differential Taxa Abundance, GAMLSS-BEZI models: Prior to differential abundance 884 analysis, PERFect:Permutation was used to remove taxa that are present due to 885 contamination or otherwise are unrelated to ME/CFS [89] . The regression then used to 886 analyze the binary outcome ME/CFS was a Generalized Additive Model for Location, 887

Scale and Shape with a zero-inflated beta distribution (GAMLSS-BEZI) performed using 888 the metamicrobiomeR package [39]. This package was built in order to deal with zero 889 inflated, compositional data such as microbiome relative abundance. Features in the 890 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. ; https://doi.org/10.1101/2021.10.27.21265575 doi: medRxiv preprint abundance table were pre-filtered, using the built-in pre-filtering step in 891 metamicrobiomeR, to remove features with mean relative abundance ≤ 0.1% and with 892 prevalence ≤ 5% across samples. With GAMLSS-BEZI a multivariate regression was 893 performed with ME/CFS status as the variable for comparison and adjusted for covariates 894 (sr-IBS diagnosis, site of recruitment, sex, BMI, race and ethnicity, age, antibiotic use 895 within 6-12 weeks of testing, probiotic supplement use and prebiotic supplement use) for 896 each taxon (species and genera were assessed separately) present in the data. Use of 897

prescription narcotics and antidepressants were tested in each model but did not affect 898 the results and were therefore not included in the final models. Multi-level categorical 899 variables were assigned as dummy variables and continuous variables were scaled using Statistically over/under-represented gut metabolic modules between ME/CFS and 913 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. ; https://doi.org/10.1101/2021.10.27.21265575 doi: medRxiv preprint controls were determined using GOmixer, which applies a Wilcoxon rank-sum test and 914 the Benjamini-Hochberg false discovery rate (FDR) to correct for multiple testing. Data 915 were scaled and the mean differences in pathways and modules were considered 916 significant at an FDR adjusted p-value < 0.1. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. and F. prausnitzii 16S rRNA per gram of feces determined by qPCR between ME/CFS 1037 and healthy controls (C, D) and among stratified groups for healthy controls without (-) sr-1038 IBS, ME/CFS subjects without (-) sr-IBS, and ME/CFS subjects with (+) sr-IBS (E, F). 1039

Box-and-whiskers plots showing the distribution of Total bacterial 16S rRNA genes per 1040 gram of feces determined by qPCR between ME/CFS and healthy controls (G) and 1041 among stratified groups (H). Statistical significance was determined based on two-tailed 1042

p-values from the Mann-Whitney U test (C, D, G). For stratified analyses significance was 1043 first determined based on the Kruskal-Wallis test (K.W., results shown below each figure  1044 in E, F, H). If significant (p < 0.05) based on K.W., then between group significance was 1045 determined based on the Mann-Whitney U test with multiple testing (Bonferroni) 1046 correction (p adj -value). n.s. = not significant; * = p or p adj < 0.05; ** = p or p adj < 0.01; *** = 1047 p or p adj < 0.001; T = trend (p or p adj < 0.1). 1048 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. 

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. significance was determined based on the Mann-Whitney U test with multiple testing 1065

(Bonferroni) correction (p adj -value). n.s. = not significant; * = p or p adj < 0.05; ** = p or p adj 1066 < 0.01; *** = p or p adj < 0.001; **** = p or p adj < 0.0001; T = trend (p or p adj < 0.1).

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. 

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. comparisons were not significant after adjustment). 1083 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. Results for each outcome variable were estimated using generalized linear regression with a Gamma distribution with log link and adjusted for covariates. Fold Change, 95% Confidence Intervals (CI), and p-values (p) are shown for each model. Note: The interaction term ME/CFS*sr-IBS was significant in all three models, therefore the association of each SCFA with ME/CFS and sr-IBS status was evaluated in stratified models (Models 2-4) using generalized linear regression with a Gamma distribution and log link and adjusted for covariates (bottom three rows of the . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. ; https://doi.org/10.1101/2021.10.27.21265575 doi: medRxiv preprint

Supplementary Figure 1 : Gut microbiome alpha diversity in ME/CFS. Box-and-whiskers plots showing the distribution of microbiome alpha diversity (Shannon diversity index and Observed species) between ME/CFS and healthy controls (A, B) and among stratified groups for healthy controls without (-) sr-IBS, ME/CFS subjects without (-) sr-IBS, and ME/CFS subjects with (+) sr-IBS (C, D). Statistical significance was determined based on two-tailed p-values from the Mann-Whitney U test (A, B) . For stratified analyses significance was first determined based on the Kruskal-Wallis test (K.W., results shown below each figure in C, D) . If significant (p < 0.05) based on K.W., then between group significance was determined based on the Mann-Whitney U test with multiple testing (Bonferroni) correction (p adj -value). n.s. = not significant; * = p or p adj < 0.05. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. ; Supplementary Figure 2 : Calculated relative abundance of butyrate producing taxa in feces based on qPCR. Box-and-whiskers plots showing the distribution of calculated relative abundance for fecal Roseburia-Eubacterium 16S rRNA and F. prausnitzii 16S rRNA between ME/CFS and healthy controls (A, B) and among stratified groups for healthy controls without (-) sr-IBS, ME/CFS subjects without (-) sr-IBS, and ME/CFS subjects with (+) sr-IBS (C, D). Statistical significance was determined based on twotailed p-values from the Mann-Whitney U test (A, B) . For stratified analyses significance was first determined based on the Kruskal-Wallis test (K.W., results shown below each figure in D-F). If significant (p < 0.05) based on K.W., then between group significance was determined based on the Mann-Whitney U test with multiple testing (Bonferroni) correction (p adj -value). n.s. = not significant; * = p or p adj < 0.05; ** = p or p adj < 0.01; *** = p or p adj < 0.001; T = trend (p or p adj < 0.1). Box-and-whiskers plots showing the distribution of calculated relative abundance for fecal but gene between ME/CFS and healthy controls (A) and among stratified groups for healthy controls without (-) sr-IBS, ME/CFS subjects without (-) sr-IBS, and ME/CFS subjects with (+) sr-IBS (B). Statistical significance was determined based on two-tailed p-values from the Mann-Whitney U test (A). For stratified analyses significance was first determined based on the Kruskal-Wallis test (K.W., results shown below B). If significant (p < 0.05) based on K.W., then between group significance was determined based on the Mann-Whitney U test with multiple testing (Bonferroni) correction (p adj -value). n.s. = not significant; ** = p or p adj < 0.01; *** = p or p adj < 0.001; T = trend (p or p adj < 0.1). . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. ; https://doi.org/10.1101/2021.10.27.21265575 doi: medRxiv preprint Supplementary Table 1A : Species-level results from GAMLSS-BEZI models for differential abundance of species between: Model 1: All ME/CFS vs healthy controls, Model 2: ME/CFS without sr-IBS vs healthy controls without sr-IBS, Model 3: ME/CFS with sr-IBS vs healthy controls without sr-IBS, and Model 4: ME/CFS without sr-IBS vs ME/CFS with sr-IBS. Models are adjusted for sr-IBS diagnosis (only model 1), site of recruitment, sex, BMI, race and ethnicity, age, antibiotic use within 6-12 weeks of sample collection, probiotic supplement use, and prebiotic supplement use. All species with significant unadjusted p-values are shown for each model. Species highlighted in red are significantly decreased in the ME/CFS groups after FDR adjustment. Species highlighted in blue are significantly increased in the ME/CFS group after FDR adjustment. Odds ratios, 95% CI, unadjusted p-values (P(>|t|), and FDR adjusted p-values are shown. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. ; https://doi.org/10.1101/2021.10.27.21265575 doi: medRxiv preprint Supplementary Table 1B : Genus-level results from GAMLSS-BEZI models for differential abundance of genera between: Model 1: All ME/CFS vs healthy controls, Model 2: ME/CFS without sr-IBS vs healthy controls without sr-IBS, Model 3: ME/CFS with sr-IBS vs healthy controls without sr-IBS, and Model 4: ME/CFS without sr-IBS vs ME/CFS with sr-IBS. Models are adjusted for sr-IBS diagnosis (only model 1), site of recruitment, sex, BMI, race and ethnicity, age, antibiotic use within 6-12 weeks of sample collection, probiotic supplement use, and prebiotic supplement use. All species with significant unadjusted p-values are shown for each model. Species highlighted in red are significantly decreased in the ME/CFS groups after FDR adjustment. Species highlighted in blue are significantly increased in the ME/CFS group after FDR adjustment. Odds ratios, 95% CI, unadjusted p-values (P(>|t|), and FDR adjusted p-values are shown. 

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 28, 2021. ; https://doi.org/10.1101/2021.10.27.21265575 doi: medRxiv preprint Supplementary Table 3B : Functional metagenomic analysis of gut bacterial metabolic modules comparing ME/CFS cases vs. healthy controls using GOmixer analysis. Bacterial metabolic modules that differed between ME/CFS and controls (FDR adjusted p-value < 0.1) are highlighted in red (deficient in ME/CFS) and blue (enriched in ME/CFS). The mean difference, p-value, and FDR adjusted p-value are shown. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted October 28, 2021. ; https://doi.org/10.1101/2021.10.27.21265575 doi: medRxiv preprint Supplementary Table 4C : Generalized linear regression showing the relationship between ME/CFS status and the metagenomic counts (CPM) of genes along the 4-Aminobutyrate pathway of butyrate production.

Results for each gene (outcome variable) were estimated using generalized linear regression with a Gamma distribution and log link and adjusted for covariates. Fold Change, 95% Confidence Intervals (CI), and p-values are shown for each model. Significant predictors (p<0.05) are indicated in bold. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted October 28, 2021. Results for the outcome variable were estimated using generalized linear regression with a Gamma distribution and log link and adjusted for covariates. Fold Change, 95% Confidence Intervals (CI), and p-values (p) are shown for each model. Model 2= ME/CFS(-)sr-IBS vs Controls(-)sr-IBS; Model 3= ME/CFS(+)sr-IBS vs Controls(-)sr-IBS; Model 4= ME/CFS(+)sr-IBS vs ME/CFS(-)sr-IBS. Significant predictors (p<0.05) are indicated in bold. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted October 28, 2021. Results for the outcome variable were estimated using generalized linear regression with a Gamma distribution and log link and adjusted for covariates. Fold Change, 95% Confidence Intervals (CI), and p-values (p) are shown for each model. Model 2= ME/CFS(-)sr-IBS vs Controls(-)sr-IBS; Model 3= ME/CFS(+)sr-IBS vs Controls(-)sr-IBS; Model 4= ME/CFS(+)sr-IBS vs ME/CFS(-)sr-IBS. Significant predictors (p<0.05) are indicated in bold. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted October 28, 2021. Results for the outcome variable were estimated using generalized linear regression with a Gamma distribution and log link and adjusted for covariates. Fold Change, 95% Confidence Intervals (CI), and p-values (p) are shown for each model. Model 2= ME/CFS(-)sr-IBS vs Controls(-)sr-IBS; Model 3= ME/CFS(+)sr-IBS vs Controls(-)sr-IBS; Model 4= ME/CFS(+)sr-IBS vs ME/CFS(-)sr-IBS. Significant predictors (p<0.05) are indicated in bold. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted October 28, 2021. ; https://doi.org/10.1101/2021.10.27.21265575 doi: medRxiv preprint

The chronic fatigue syndrome: a comprehensive approach to its definition and study. International Chronic Fatigue Syndrome Study Group

Myalgic encephalomyelitis: International Consensus Criteria

The National Academies Collection: Reports funded by National Institutes of Health

Prevalence and characteristics of chronic fatigue syndrome/myalgic encephalomyelitis (CFS/ME) in Poland: a cross-sectional study

The prevalence of chronic fatigue syndrome/ myalgic encephalomyelitis: a meta-analysis

Immunological aspects of chronic fatigue syndrome

Systematic review and meta-analysis of the prevalence of chronic fatigue syndrome/myalgic encephalomyelitis (CFS/ME)

The economic impact of ME/CFS: individual and societal costs

Beyond Myalgic Encephalomyelitis/Chronic Fatigue Syndrome: redefining an illness

Insights from myalgic encephalomyelitis/chronic fatigue syndrome may help unravel the pathogenesis of postacute COVID-19 syndrome

Redox imbalance links COVID-19 and myalgic encephalomyelitis/chronic fatigue syndrome

Post-infective and chronic fatigue syndromes precipitated by viral and non-viral pathogens: prospective cohort study

Myalgic encephalomyelitis, chronic fatigue syndrome: An infectious disease. Med Hypotheses

Will COVID-19 Lead to Myalgic Encephalomyelitis/Chronic Fatigue Syndrome? Front Med (Lausanne)

Long-term sequelae following previous coronavirus epidemics

COVID-19 and chronic fatigue syndrome: Is the worst yet to come? Med Hypotheses

Historical Insight into Infections and Disorders Associated with Neurological and Psychiatric Sequelae Similar to Long COVID

Four-Month Clinical Status of a Cohort of 73

Ambulatory monitoring of physical activity and symptoms in fibromyalgia and chronic fatigue syndrome

Exercise and the Gut Microbiome: A Review of the Evidence, Potential Mechanisms, and Implications for Human Health

Interplay Between Exercise and Gut Microbiome in the Context of Human Health and Performance. Front Nutr

The Multidimensional Fatigue Inventory (MFI) psychometric qualities of an instrument to assess fatigue

Cutadapt removes adapter sequences from high-throughput sequencing reads

Quality control and preprocessing of metagenomic datasets

Fast gapped-read alignment with Bowtie 2

Ultrafast and accurate 16S rRNA microbial community analysis using Kraken 2. Microbiome

Bracken: estimating species abundance in metagenomics data

Analysis of Microbiome Data in the Presence of Excess Zeros. Front Microbiol

Reproducible, interactive, scalable and extensible microbiome data science using QIIME 2

Isovaleric acid synthesis I (KADH pathway) 29.33347714 0.0107 0.0357 MGB007: Glutamate synthesis II

Metagenomic gene content of the fecal microbiota for genes in bacterial pathways of butyrate production. Box-and-whiskers plots showing the distribution of gene counts (CPM) for genes in the Glutarate (A, D), Lysine (B, E), and 4-Aminobutyrate (C, F) pathways of butyrate production between ME/CFS and healthy controls (A-C) and among stratified groups for healthy controls without (-) sr-IBS, ME/CFS subjects without (-) sr-IBS, and ME/CFS subjects with (+) sr-IBS (D-F). Statistical significance was determined based on two-tailed p-values from the Mann-Whitney U test for each gene (A-C). For stratified analyses significance was first determined based on the Kruskal-Wallis test for each gene (K.W., results shown below each figure in D-F). If significant (p < 0.05) based on K.W., then between group significance was determined based on the Mann-Whitney U test with multiple testing (Bonferroni) correction (p adjvalue). n.s. = not significant; * = p or p adj < 0.05; ** = p or p adj < 0.01; T = trend (p or p adj < 0.1). Table 3A : Functional metagenomic analysis of gut bacterial metabolic processes comparing ME/CFS cases vs. healthy controls using GOmixer analysis. Bacterial metabolic processes that differed between ME/CFS and controls (FDR adjusted p-value < 0.1) are highlighted in red (deficient in ME/CFS) and blue (enriched in ME/CFS). The mean difference, p-value, and FDR adjusted p-value are shown.

Difference Table 3C : Functional metagenomic analysis of gut bacterial metabolic modules comparing ME/CFS cases vs. healthy controls using GOmixer analysis with gutbrain specific modules. Bacterial gut-brain metabolic modules that differed between ME/CFS and controls (FDR adjusted p-value < 0.1) are highlighted in red (deficient in ME/CFS) and blue (enriched in ME/CFS). The mean difference, p-value, and FDR adjusted p-value are shown. p-values based on two-tailed Mann-Whitney U test