Understanding the aetiology of Myalgic encephalomyelitis
Gemma Louise Samms
Background
ME/CFS is a devastating condition that lacks effective treatments. Those affected experience debilitating symptoms and little healthcare or societal support. There are diagnostic challenges associated with ME/CFS as it lacks diagnostic molecular biomarkers, and case definitions have evolved over time.
Current diagnostic criteria require post-exertional malaise (PEM), the cardinal symptom of ME/CFS, but this is not always applied in research. A replication crisis in the field resulting from small-scale studies, or poorly defined cases, has resulted in a poor understanding of the aetiology of ME/CFS, thus well-powered research to gain a deeper understanding of this condition is required.
Methods
I used NHS-DigiTrials data capturing ~62-million individuals across England to examine the prevalence of ME/CFS (ICD-10:G93.3 codes) overall, and stratified by age, self-reported gender, index of multiple deprivation, self-reported ethnicity, general practice and integrative care board (ICB) (Chapter 2).
Subsequently, I used UK Biobank (UKB) data on ~500,000 individuals for three purposes. First, to evaluate overlap and discordance between four sources of ME/CFS data, with the aim of building a well-defined UKB-ME/CFS cohort for future research (Chapter 3).
Second, to explore whether there were differences in inherited chromosomally integrated (iciHHV-6) carrier status, and the seroprevalence and antibody binding to antigens for the following human herpes viruses (HHVs: HHV-6 overall, HHV-6A, HHV-6B, HHV-7, CMV and EBV) in those with ME/CFS versus controls (Chapter 4). For this, I defined UKB-ME/CFS cohorts and controls based on the work presented in Chapter 3.
Thirdly, I used UKB’s Experience of Pain questionnaire (PQ), to perform genome-wide association study (GWAS) and post-GWAS analyses (FUMA, MAGMA and colocalisation analysis) on ME/CFS, and on a commonly comorbid condition, Fibromyalgia syndrome (FMS) (Chapter 5).
The ME/CFS trait was based on self-reported diagnoses, and I attempted replication in a disjoint set of individuals with an ICD-10:G93.3 code. FMS-related traits included self-reported FMS, classification as per the American College of Rheumatology (ACR2016) criteria, and the total and component scores of ACR2016.
Results
100,055 individuals (0.16%) were linked to an ICD-10:G93.3 code among the NHS-DigiTrials population, but there were striking differences in prevalence across subgroups and locations.
For example, across ICBs, prevalence ranged between 0.086-0.82% and 0.024-0.21% for females and males, respectively, and those of other-than-white ethnicity were five-fold less likely to receive a diagnosis than those of white ethnicity. This ethnicity bias was strongest for ME/CFS, compared to a range of other health conditions.
Additionally, diagnosis was lower for those living in the more deprived postcodes, compared to the least deprived, and 176 general practices with zero patients linked to an ICD-10:G93.3 code were also disproportionately located in more deprived postcodes. To account for unequal diagnosis, I used the highest rate of diagnosis (Cornwall and The Isles of Scilly) and extrapolated this to the UK population and estimated that about 404,000 individuals in the UK (0.6%) have ME/CFS symptoms.
Next, within the UKB, I identified 5,354 individuals with at least one piece of evidence for ME/CFS, however there were inconsistencies between data sources (only 36% (n=1,922) had more than one piece of evidence for ME/CFS). ICD-10:G93.3 was the most supported data source within UKB, with 72% of this cohort having further evidence of ME/CFS. However, ICD-10:G93.3 codes identified only 26% of UKB-ME/CFS cases.
Furthermore, I identified that 61% of one UKB-ME/CFS cohort did not self-reported PEM-like symptoms, and 28% of another UKB-ME/CFS cohort reported ‘good’ or ‘excellent’ overall health.
I discuss the limitations and inconsistencies among these four UKB-ME/CFS cohorts and recommended filtering steps to improve confidence in case status. Additionally, ME/CFS diagnosis as per primary care records that lacked inconsistencies were extrapolated to estimate a ME/CFS prevalence of ~410,000 in the UK, accounting for those who could not partake in UKB due to being house- or bed-bound.
Additionally, I identified a lower carrier status of iciHHV-6 (0.36% cases versus 1.35% controls, p=0.0095), a higher seroprevalence of HHV-6B (96.88% versus 79.09% controls, p=0.0082), and lower antigen binding to three CMV antigens (pp 28: p=0.0087, pp 52: p=0.0080, pp150: p=0.0080), among some UKB-ME/CFS cohorts.
Furthermore, a single novel genome-wide significant (GWS, p<5x10-8) locus was identified for ME/CFS (TNNT1), but this failed replication in a disjoint set of individuals with an ICD-10:G93.3 code.
A further 20 GWS loci were identified for FMS-related traits, and colocalisation analysis revealed 54 ‘tier 1’ genes (PPH4≥0.75). MAGMA gene-based test also identified six of these ‘tier 1’ genes plus 18 additional genes, and a brain tissue enrichment was observed for FMS pseudo-continuous traits.
Five genes implicated in FMS-related GWAS (DCC/SDCCAG8/CTNNA2/FAM120A/EXD3) have previously been associated with FMS, or pain-related traits, albeit with partially overlapping UKB data.
Overall, FMS-related traits implicated genes involved in neural function, immunity, inflammation, oxidative stress, mitochondrial function and DNA repair.
Conclusion
Through use of two large datasets, I attained two converging UK ME/CFS prevalence estimates that are much higher (~60%) than previously estimated. This revised estimate of ~400,000 is now used by the UK Government. However, the large variation in ICD-10:G93.3 codes suggest unequal diagnosis, which may reflect higher barriers to diagnosis among certain groups, for example those of other-than-white ethnicity or those living in more deprived postcodes.
It is essential that these barriers to diagnosis are overcome so that diagnosis is accessible to all, and this highlights the need for the development of a diagnostic test for ME/CFS. Additionally, the relevance of ICD-10:G93.3 as a proxy for ME/CFS was demonstrated in the UKB, however this code is insufficient in detecting most ME/CFS cases and therefore accurate codes (such as ICD-10:G93.32, that is currently unavailable in the UK) are required for more accurate case ascertainment.
I identified inconsistencies among UKB-ME/CFS cohorts and recommended filtering steps to improve confidence in case status. This work resulted in publication of paper on ME/CFS biomarkers, and a genetics paper that is in preparation.
This work also aided my interpretation of the HHV findings, although further work investigating HHVs in larger and better-defined cohorts is required because this data was only available on 2% of UKB participants and therefore, filtered UKB-ME/CFS cohorts were very small (n<34), and consequently, underpowered.
Finally, the genetic analyses revealed numerous associations for FMS, however replication is a critical next step. The FMS analyses benefitted from the use of the ACR2016 classification which helped to define cases, resulting in a more homogeneous population that was better powered to detect genetic associations. In contrast, the cohort used in the ME/CFS GWAS was small and contained poorly defined cases, and thus, larger, better-defined cohorts are required to increase statistical power.
Overall, this work demonstrates the importance of large and well-defined cohorts in ME/CFS research, and it is hoped this paves the way for future work.
Web | DOI | Thesis: University of Edinburgh | Under embargo until Oct. 28, 2027
Gemma Louise Samms
Background
ME/CFS is a devastating condition that lacks effective treatments. Those affected experience debilitating symptoms and little healthcare or societal support. There are diagnostic challenges associated with ME/CFS as it lacks diagnostic molecular biomarkers, and case definitions have evolved over time.
Current diagnostic criteria require post-exertional malaise (PEM), the cardinal symptom of ME/CFS, but this is not always applied in research. A replication crisis in the field resulting from small-scale studies, or poorly defined cases, has resulted in a poor understanding of the aetiology of ME/CFS, thus well-powered research to gain a deeper understanding of this condition is required.
Methods
I used NHS-DigiTrials data capturing ~62-million individuals across England to examine the prevalence of ME/CFS (ICD-10:G93.3 codes) overall, and stratified by age, self-reported gender, index of multiple deprivation, self-reported ethnicity, general practice and integrative care board (ICB) (Chapter 2).
Subsequently, I used UK Biobank (UKB) data on ~500,000 individuals for three purposes. First, to evaluate overlap and discordance between four sources of ME/CFS data, with the aim of building a well-defined UKB-ME/CFS cohort for future research (Chapter 3).
Second, to explore whether there were differences in inherited chromosomally integrated (iciHHV-6) carrier status, and the seroprevalence and antibody binding to antigens for the following human herpes viruses (HHVs: HHV-6 overall, HHV-6A, HHV-6B, HHV-7, CMV and EBV) in those with ME/CFS versus controls (Chapter 4). For this, I defined UKB-ME/CFS cohorts and controls based on the work presented in Chapter 3.
Thirdly, I used UKB’s Experience of Pain questionnaire (PQ), to perform genome-wide association study (GWAS) and post-GWAS analyses (FUMA, MAGMA and colocalisation analysis) on ME/CFS, and on a commonly comorbid condition, Fibromyalgia syndrome (FMS) (Chapter 5).
The ME/CFS trait was based on self-reported diagnoses, and I attempted replication in a disjoint set of individuals with an ICD-10:G93.3 code. FMS-related traits included self-reported FMS, classification as per the American College of Rheumatology (ACR2016) criteria, and the total and component scores of ACR2016.
Results
100,055 individuals (0.16%) were linked to an ICD-10:G93.3 code among the NHS-DigiTrials population, but there were striking differences in prevalence across subgroups and locations.
For example, across ICBs, prevalence ranged between 0.086-0.82% and 0.024-0.21% for females and males, respectively, and those of other-than-white ethnicity were five-fold less likely to receive a diagnosis than those of white ethnicity. This ethnicity bias was strongest for ME/CFS, compared to a range of other health conditions.
Additionally, diagnosis was lower for those living in the more deprived postcodes, compared to the least deprived, and 176 general practices with zero patients linked to an ICD-10:G93.3 code were also disproportionately located in more deprived postcodes. To account for unequal diagnosis, I used the highest rate of diagnosis (Cornwall and The Isles of Scilly) and extrapolated this to the UK population and estimated that about 404,000 individuals in the UK (0.6%) have ME/CFS symptoms.
Next, within the UKB, I identified 5,354 individuals with at least one piece of evidence for ME/CFS, however there were inconsistencies between data sources (only 36% (n=1,922) had more than one piece of evidence for ME/CFS). ICD-10:G93.3 was the most supported data source within UKB, with 72% of this cohort having further evidence of ME/CFS. However, ICD-10:G93.3 codes identified only 26% of UKB-ME/CFS cases.
Furthermore, I identified that 61% of one UKB-ME/CFS cohort did not self-reported PEM-like symptoms, and 28% of another UKB-ME/CFS cohort reported ‘good’ or ‘excellent’ overall health.
I discuss the limitations and inconsistencies among these four UKB-ME/CFS cohorts and recommended filtering steps to improve confidence in case status. Additionally, ME/CFS diagnosis as per primary care records that lacked inconsistencies were extrapolated to estimate a ME/CFS prevalence of ~410,000 in the UK, accounting for those who could not partake in UKB due to being house- or bed-bound.
Additionally, I identified a lower carrier status of iciHHV-6 (0.36% cases versus 1.35% controls, p=0.0095), a higher seroprevalence of HHV-6B (96.88% versus 79.09% controls, p=0.0082), and lower antigen binding to three CMV antigens (pp 28: p=0.0087, pp 52: p=0.0080, pp150: p=0.0080), among some UKB-ME/CFS cohorts.
Furthermore, a single novel genome-wide significant (GWS, p<5x10-8) locus was identified for ME/CFS (TNNT1), but this failed replication in a disjoint set of individuals with an ICD-10:G93.3 code.
A further 20 GWS loci were identified for FMS-related traits, and colocalisation analysis revealed 54 ‘tier 1’ genes (PPH4≥0.75). MAGMA gene-based test also identified six of these ‘tier 1’ genes plus 18 additional genes, and a brain tissue enrichment was observed for FMS pseudo-continuous traits.
Five genes implicated in FMS-related GWAS (DCC/SDCCAG8/CTNNA2/FAM120A/EXD3) have previously been associated with FMS, or pain-related traits, albeit with partially overlapping UKB data.
Overall, FMS-related traits implicated genes involved in neural function, immunity, inflammation, oxidative stress, mitochondrial function and DNA repair.
Conclusion
Through use of two large datasets, I attained two converging UK ME/CFS prevalence estimates that are much higher (~60%) than previously estimated. This revised estimate of ~400,000 is now used by the UK Government. However, the large variation in ICD-10:G93.3 codes suggest unequal diagnosis, which may reflect higher barriers to diagnosis among certain groups, for example those of other-than-white ethnicity or those living in more deprived postcodes.
It is essential that these barriers to diagnosis are overcome so that diagnosis is accessible to all, and this highlights the need for the development of a diagnostic test for ME/CFS. Additionally, the relevance of ICD-10:G93.3 as a proxy for ME/CFS was demonstrated in the UKB, however this code is insufficient in detecting most ME/CFS cases and therefore accurate codes (such as ICD-10:G93.32, that is currently unavailable in the UK) are required for more accurate case ascertainment.
I identified inconsistencies among UKB-ME/CFS cohorts and recommended filtering steps to improve confidence in case status. This work resulted in publication of paper on ME/CFS biomarkers, and a genetics paper that is in preparation.
This work also aided my interpretation of the HHV findings, although further work investigating HHVs in larger and better-defined cohorts is required because this data was only available on 2% of UKB participants and therefore, filtered UKB-ME/CFS cohorts were very small (n<34), and consequently, underpowered.
Finally, the genetic analyses revealed numerous associations for FMS, however replication is a critical next step. The FMS analyses benefitted from the use of the ACR2016 classification which helped to define cases, resulting in a more homogeneous population that was better powered to detect genetic associations. In contrast, the cohort used in the ME/CFS GWAS was small and contained poorly defined cases, and thus, larger, better-defined cohorts are required to increase statistical power.
Overall, this work demonstrates the importance of large and well-defined cohorts in ME/CFS research, and it is hoped this paves the way for future work.
Web | DOI | Thesis: University of Edinburgh | Under embargo until Oct. 28, 2027