Diagnosis of [ME/CFS] With Partial Least Squares Discriminant Analysis: Relevance of Blood Extracellular Vesicles, 2022, González-Cebrián et al

Andy

Senior Member (Voting rights)
Full title: Diagnosis of Myalgic Encephalomyelitis/Chronic Fatigue Syndrome With Partial Least Squares Discriminant Analysis: Relevance of Blood Extracellular Vesicles

Myalgic Encephalomyelitis/Chronic Fatigue Syndrome (ME/CFS), a chronic disease characterized by long-lasting persistent debilitating widespread fatigue and post-exertional malaise, remains diagnosed by clinical criteria. Our group and others have identified differentially expressed miRNA profiles in the blood of patients. However, their diagnostic power individually or in combinations seems limited.

A Partial Least Squares-Discriminant Analysis (PLS-DA) model initially based on 817 variables: two demographic, 34 blood analytic, 136 PBMC miRNAs, 639 Extracellular Vesicle (EV) miRNAs, and six EV features, selected an optimal number of five components, and a subset of 32 regressors showing statistically significant discriminant power. The presence of four EV-features (size and z-values of EVs prepared with or without proteinase K treatment) among the 32 regressors, suggested that blood vesicles carry relevant disease information. To further explore the features of ME/CFS EVs, we subjected them to Raman micro-spectroscopic analysis, identifying carotenoid peaks as ME/CFS fingerprints, possibly due to erythrocyte deficiencies. Although PLS-DA analysis showed limited capacity of Raman fingerprints for diagnosis (AUC = 0.7067), Raman data served to refine the number of PBMC miRNAs from our previous model still ensuring a perfect classification of subjects (AUC=1). Further investigations to evaluate model performance in extended cohorts of patients, to identify the precise ME/CFS EV components detected by Raman and to reveal their functional significance in the disease are warranted.

Open access, https://www.frontiersin.org/articles/10.3389/fmed.2022.842991/full
 
So, I'm just getting my head around this. I think this study is using data from a previous study
Almenar-Pérez E, Sarría L, Nathanson L, Oltra E. Assessing diagnostic value of microRNAs from peripheral blood mononuclear cells and extracellular vesicles in Myalgic Encephalomyelitis/Chronic Fatigue Syndrome. Sci Rep. (2020)
That study was hampered from finding significant results by the fact that the sample size was small and the range of variables being looked at was enormous.

So, they seem to have pulled out some fairly heavy duty stats to overcome that problem, at least in their minds, and have created a diagnostic model by slicing and dicing the data from just 15 people with severe ME/CFS and 15 healthy controls various ways.

I'm going to have to take a break before I get to the good part, the discussion. But I think they concluded that the size distribution of extracellular vesicles and the contents of them (as identified by the Raman spectroscopy), among other things, are helpful in discriminating people with severe ME/CFS from healthy controls.


is this promising?
So, maybe it's promising, maybe it's just statistical torture of the data. It will take at least another study with a much bigger sample size zeroing in on the variables they think are important to confirm or negate their ideas. I think there was a recent study of EVs that found that EV size wasn't very helpful in diagnosing ME/CFS.

(Just to reiterate, I haven't yet read the whole study and I'm at the point where it is hard to hang on to my thoughts long enough to do something with them. )
 
Last edited:
(This was the recent study on EVs. It also had a problem finding something significant after taking into account the large number of variables. They were more focused on markers on the membranes of the EVs.)
Comparative Analysis of Extracellular Vesicles in Patients with Severe and Mild MECFS, Bonilla et al, 2022

Our study revealed a significant correlation between severe ME/CSF and levels of EVs bearing the B cell marker CD19 and the platelet marker CD41a, though these changes were not significant after correction for multiple comparisons.
 
Authors:
Alba González-Cebrián1, Eloy Almenar-Pérez2,3, Jiabao Xu4, Tong Yu4, Wei E. Huang4, Karen Giménez-Orenga5, Sarah Hutchinson3, Tiffany Lodge3, Lubov Nathanson6,7, Karl J. Morten3, Alberto Ferrer1 and Elisa Oltra2,8*

1Grupo de Ingeniería Estadística Multivariante, Departamento de Estadística e Investigación Operativa Aplicadas y Calidad, Universitat Politècnica de València, Valencia, Spain
2Department of Pathology, School of Health Sciences, Universidad Católica de Valencia San Vicente Mártir, Valencia, Spain
3Nuffield Department of Women's and Reproductive Health, The Women Centre, University of Oxford, Oxford, United Kingdom
4Department of Engineering Science, University of Oxford, Oxford, United Kingdom
5Escuela de Doctorado, Universidad Católica de Valencia San Vicente Mártir, Valencia, Spain
6Kiran C. Patel College of Osteopathic Medicine, Nova Southeastern University, Fort Lauderdale, FL, United States
7Institute for Neuro Immune Medicine, Nova Southeastern University, Fort Lauderdale, FL, United States
8Centro de Investigación Traslacional San Alberto Magno, Universidad Católica de Valencia San Vicente Mártir, Valencia, Spain
 
CK levels being a clinical feature that had been previously reported as a potential biomarker of ME/CFS for showing significant reduced levels in an expanded cohort of patients (48 "This is the first report that serum CK concentrations are markedly reduced in severe ME/CFS, and these results suggest that serum CK merits further investigation as a biomarker for severe ME/CFS."). Highly expressed in muscle, heart, and brain the CK enzyme holds a key role in ATP homeostasis. The low levels found by Nacul et al., possibly reflecting energy dysregulation in these tissues, may be linked to the profound fatigue found in ME/CFS patients with the severe having the lowest levels.

Regardless of EV composition differences we were interested in exploring if the Raman spectroscopic data was sufficient to efficiently distinguish ME/CFS cases from HCs. Despite its potential discriminatory capacity of ME/CFS body fluid components (Figures 4, 5), in good agreement with the disease “plasma factor” hypothesis reported by Ron Davis' group at Standford University (63), which is also supported by differences in proteins or lipid plasma levels (64, 65), the diagnostic value of Raman data seems limited when compared to our PLS-DA model including analytic variables, PBMC miRNA profiles and EV features (Figures 1, 2).
My bolding and underlining.

I do hope they get the chance to build on this work and see if the Raman Spectrometer can produce a bio marker.

I had my yearly blood work last month. I'm a diabetic. I'm bitter that ME/CFS doesn't have a test like HbA1c.
 
The thread on the Nacul study that found lower CK concentrations in severe ME/CFS is here
Evidence of Clinical Pathology Abnormalities in People with ME/CFS from an Analytic Cross-Section (2019) Nacul et al.

In that thread, doubts were raised that the finding is significant, as lower levels of creatine kinase is associated with lower activity levels.
Creatine kinase mirrors activity to a significant extent. Going to a disco could double it. A level of 50 is very normal, just at the lower end. Pathology is associated with raised levels and not low levels by and large.
The authors of the Nacul study themselves acknowledged that the lower levels of activity of people with severe ME/CFS may well have confounded the finding of lower CK.

The EV study that is the subject of the thread used blood CK levels in their model to identify the people with ME/CFS in their sample:
Gonzales-Cebrian study said:
It is worth mentioning that among the blood analytic group of variables the iterative PLS-DA modeling process selected, blood creatine phosphokinase (CK, labeled as cpkbloodb, please see Supplementary Table 1key tab for variable nomenclature used) level was a feature retrieved with and without the inclusion of Raman data
So, with CK being a marker for activity levels, this part of their model is probably essentially predicting whether people have ME/CFS or not based on their activity level. It's not very helpful.
 
ME Association

Professor Elisa Oltra (University of Valencia) has kindly provided us with a lay summary:

In the article “Diagnosis of Myalgic Encephalomyelitis/Chronic Fatigue Syndrome With Partial Least Squares Discriminant Analysis: Relevance of Blood Extracellular Vesicles”, published by Frontiers in Medicine on April 1st 2022, the authors take a step further by examining the biomarker value of over 800 variables in the blood of severely affected ME/CFS patients, some obtained from their former studies.

The samples obtained from the UK ME Biobank (UKMEB), London, UK, revealed that the extracellular vesicles present in plasma, together with a set of microRNAs, and few additional blood features including increased mean corpuscular haemoglobin concentration, lower creatine phosphokinase levels and eosinophil counts, serve to faithfully discriminate ME/CFS patients from healthy subjects.

Although the method represents the most accurate to date for the diagnosis of severe ME/CFS, its translation to the clinic seems hampered by the heterogeneity of variables demanded which translates into a technically complex method. Nevertheless, the reduction of over 800 variables to thirty-two should simplify its validation in other groups of patients.

Moreover, the work introduces a relatively quick simple method (Raman spectroscopy) as a potential useful tool for an initial triage diagnosis stage. The differences in patient's vesicles detected with Raman are in good agreement with the findings of altered deformability of red blood cells, high contributors of blood vesicles, by Ron Davis group.

Top cellular functions associating to the highly discriminant microRNAs were immunity, neuroinflammation, and metabolism. Dr CS MEA

 
Although the method represents the most accurate to date for the diagnosis of severe ME/CFS
Although I think there are probably some important metabolites identified in this study, it's not correct to suggest that the 32 variables add up to a method of diagnosing severe ME/CFS. At the moment, the 32 variables are required to separate the 15 severe ME/CFS from the 15 healthy controls. Given creatine kinase is one of the variables, and that is likely to relate to activity levels, and they had over 800 variables to choose from, this isn't yet a particularly remarkable achievement.

They need to take their potential diagnostic model and apply it to a large numbers of people with ME/CFS and healthy controls in a blinded way and get good results before they can begin to claim that it is useful for diagnosis.

Researchers need to stop making this claim of having made a diagnostic model when they haven't tested it on an independent sample. I don't know why they do it; surely they know that a model made in this way counts for little until validated.
 
Last edited:
"Press release on our new study. We can separate severe ME/CFS patients from Healthy controls with 100% accuracy. What about less severe cases ? This is the next step.
This had me staring at the claim for a while, with my mouth agape. I'm repeating what I said in my post above - it's wrong that researchers would make this claim on the basis of an ability to separate 15 patients from 15 controls using 32 variables cherry-picked from a pool of over 800 variables.

I don't know if they think saying that brings hope to people with ME/CFS, or helps the research team win funding, or wins them admiration. I don't think it does any of those things, certainly it should not. To me, it just suggests that these researchers have drunk their own Kool-aid and have a tenuous grasp of science.

Edit - and that's a shame, because the team did do some interesting work.
 
Last edited:
If I understand correctly they tested 800 variables in a sample of 15 patients and did not test the 32 variables that made up their model on a new cohort. Unfortunately, I think that means the results are likely not very meaningful.
What then would be the appropriate methodology to further this study? Testing the 32 identified variables in a larger sample? (and cross your fingers it can be funded and replicated?)
 
Researchers need to stop making this claim of having made a diagnostic model when they haven't tested it on an independent sample. I don't know why they do it; surely they know that a model made in this way counts for little until validated.

What then would be the appropriate methodology to further this study? Testing the 32 identified variables in a larger sample? (and cross your fingers it can be funded and replicated?)

They say in the Oxford Press release:
https://www.wrh.ox.ac.uk/news/diagnostic-test-could-offer-new-hope-for-me-cfs-patients
Their next step is to apply this approach to mild and moderately affected ME/CFS patients with different levels of disability and compare to other disease groups as well as healthy controls. This will determine if we have a potential panel of biomarkers which could be used to developed a diagnostic test

So maybe someone else who didn't understand how preliminary this work is wrote the headlines. It's clearly just a first step and will need much larger studies on different groups before they can make any claim of having a diagnostic test. And that will need funding...
 
So maybe someone else who didn't understand how preliminary this work is wrote the headlines.
Clearly, someone who didn't understand how preliminary this work is wrote the headlines. Maybe that someone wasn't an author. But, there should be internal processes that ensure the authors sign off on what is said on their behalf on social media about the paper. And, when the paper itself says this:

this paper said:
In summary, this work describes for the first time an ME/CFS model based on PLS-DA of 32 analytical variables capable of diagnosing the disease with perfect sensitivity and specificity (AUC=1), further confirming the biologic nature of this disease and highlighting the relevance of patient EV features for their diagnosis.

then it's understandable why someone would assume that it's ok to say "We can separate severe ME/CFS patients from Healthy Controls with 100% accuracy."

If I understand correctly they tested 800 variables in a sample of 15 patients and did not test the 32 variables that made up their model on a new cohort. Unfortunately, I think that means the results are likely not very meaningful.
Yes, what they did is a bit more complicated than that, but essentially amounts to that. I am yet to pick through it to find out exactly what they did. They did at one point make a model using data from 12 patients, and, I think, tested it using data from the full 15 patient cohort. But, it is clear that data used never came from more than 15 patients with ME/CFS.

I think there may be some inconsistency with the number of variables reported to have been used in the final model - 32, 34, 35? But, there were a number of models made, so it may be ok. I don't find the description of what was done very clear on a preliminary read.
edit - sorry, fixed a quote problem
 
Last edited:
Back
Top Bottom