Cytokine Hub Classification of PASC, ME-CFS and other PASC-like Conditions

doi:10.21203/rs.3.rs-1598634/v1

Background

Post-acute sequelae of COVID-19 (PASC) is a growing healthcare and economic concern affecting as many as 10%-30% of those infected with COVID-19. Though the symptoms have been well-documented, they significantly overlap with other common chronic inflammatory conditions which could confound treatment and therapeutic trials.

Methods

A total of 236 patients including 64 with post-acute sequelae of COVID-19 (PASC), 50 with myalgic encephalomyelitis-chronic fatigue syndrome (ME-CFS), 29 with post-treatment Lyme disease (PTLD), and 42 post-vaccine individuals with PASC-like symptoms (POVIP) were enrolled in the study. We performed a 14-plex cytokine/chemokine panel previously described to generate raw data that was normalized and run in a decision tree model using a Classification and Regression Tree (CART) algorithm. The algorithm was used to classify these conditions in distinct groups despite their similar symptoms.

Results

PASC, ME-CSF, POVIP, and Acute COVID-19 disease categories were able to be classified by our cytokine hub based CART algorithm with an average F1 score of 0.61 and high specificity (94%).

Conclusions

Proper classification of these inflammatory conditions with very similar symptoms is critical for proper diagnosis and treatment.

Immunology

COVID-19

PASC

long COVID

long hauler

PTLD

ME-CSF

Investigation has identified overlapping symptom presentations of PASC with ME/CFS, PTLD and other post-infectious chronic inflammatory disorders^1–3. However, clear etiological and pathophysiological differences exist in these conditions that necessitate precision medicine tailored therapies.

A recent report suggested the use of cytokine hubs to more precisely categorize autoimmune diseases with the stated goal of using the information as therapeutic targets as the expansion of immune based therapy grows⁴. The heterogeneity of immune-mediated inflammatory diseases (IMIDS) described in the prior publication also applies to post-infectious immune-mediated and inflammatory conditions currently in the discussion of post-infectious sequelae of COVID-19 (PASC) and the focus of this report. Unlike the previous publication², PASC-like conditions share many of the same symptoms making diagnosis and, ultimately, treatment more difficult^1–3. Here, we present a machine learning approach to classifying these symptom related conditions.

Patients

After written informed consent was obtained, the immunological lab results were used for the current analysis of 236 patients including 64 with post-acute sequelae of COVID-19 (PASC), 26 with mild-moderate acute COVID-19 (MM), 25 with severe acute COVID-19, 50 with myalgic encephalomyelitis-chronic fatigue syndrome (ME-CFS), 29 with post-treatment Lyme disease (PTLD), and 42 post-vaccine individuals with PASC-like symptoms (POVIP) defined as COVID-negative (nucleoprotein Ab negative, T-cell immunity negative) individuals with PASC-like symptoms 3 months after the last vaccination.

Multiplex Cytokine Quantification

Fresh plasma was used quantification of the following analytes: TNF-α, IL-4, IL-13,IL-2, GM-CSF, sCD40L, CCL5 (RANTES), CCL3 (MIP-1α),IL-6, IL-10, IFN-γ, VEGF, IL-8, and CCL4 (MIP-1β) as previously decribed².

Data Acquisition and Processing

The data [Mild-Moderate acute COVID, Severe acute COVID, PASC, POVIP, ME-CSF and PTLD individuals] consisted of a total of 16 columns represented by an anonymized sample identifier, and the last column was the class or disease state assigned to the individual. The class label was removed from this pandas dataset and assigned into a new variable (target)⁵.

The dataset comprising the cytokine profiles was normalized using L2 (Euclidian) normalization. The L2 nonrealization approach calculates the distance of a given vector of values, such as the cytokine values for each data point or instance from the origin of the vector space. The implementation of the L2 norm, which is a positive value, can be supported by the notion our model is focused on identifying the differences between classes, thus the signal or pattern that characterizes each class and not the effect of the magnitudes on each class⁶. The two datasets (cytokines profiles and targets/labels) were used to train a machine learning classifier using decision trees to define patient’s disease state⁷.

Multiclass classification of disease states using a decision tree model

The decision tree model was based on the Classification and Regression Trees (CART) algorithm in scikit-learn⁸. The model was fine-tuned using GridSearchCV from scikit-learn. To fine-tune the model, we selected the following hyperparameters: maximum number of features, the minimum number of samples to split a node, the minimum number of samples per node, the maximum depth of the tree. Additionally, the impurity criterion was defined as the Gini impurity value. The cross validation was set to 10-fold cross validation with 3 repeats.

The decision tree was used to calculate a confusion matrix in order to visualize the model’s predictive power as well as a “leave one out cross-validation” (LOOCV). The confusion matrix was used to calculate the F1-score which uses the harmonic means of precision and recall; combining them into a performance metric ranging from 0 (poor) to 1 (perfect). The resulting tree was then plotted to visualize the separation of the classes based on the different cytokines.

The symptoms of PASC have been widely reported and significantly overlap with ME-CFS and with PTLD^1-3. In addition, we included patients, of unknown prevalence, who are post-vaccine individuals with PASC-like symptoms (POVIP) in the present study (Figure 1). These patients experience PASC-like symptoms 3 months minimum post-vaccination in the absence of COVID 19 infection.

By using a decision tree classifier, Classification and Regression Trees (CART), we developed an algorithm to propose a simple but powerful predictive model with high interpretability, a characteristic of great importance when attempting to understand differences between disease states. Our model had an average weighted F1 score of 0.61, which was variable due to the stochastic nature of both the model and the dataset. As shown in Table 1 and in the confusion matrix (Supplementary Figure 1), the model was robust in its ability to identify four of the six classes of disease states (e.g. MM, Severe, PASC, and ME-CSF). Some misclassification was demonstrated in the remaining two classes (POVIP and Lyme) that was likely due to overlapping cytokine hubs. The confusion matrix may suggest that the immune contribution to these two states were similar. Clinical data as well as other immunological parameters could potentially separate these two conditions and further increase the model’s predictive power.

The highest performance metrics after fine-tuning was provided by the CART decision tree (Fig. 2) when data were plotted using internal tree plotting functions and python’s matplotlib (Supplementary Fig. 2). As shown in the plotted tree (Fig 2, Supplementary Fig. 2), we demonstrate that the CART algorithm was capable of constructing splitting and terminal hubs with low Gini impurity values and high F1 scores for those classes shown previously in the confusion matrix (Supplementary Fig. 1). In the POVID and PTLD classes, splitting hubs with higher Gini impurity values were observed. The identification of these hubs supports the possibility that the immune profiles of both POVIP and PTLD individuals have similar cytokine patterns.

The disease heterogeneity suggests that the classification need not perfectly match the labels and that individuals within each of these conditions might present different immunological entities potentially requiring differential therapeutics. We demonstrated that for POVIP and MM, three or more distinct cytokine profiles might be important for their classification supporting the presence of different immunological entities within these groups. On the other hand, for ME-CSF, Lyme, and PASC only one or two cytokine classification profiles were found.

The PASC classification highlights the importance of the proinflammatory cytokines IL-2 and IFN-γ as we have previously reported², while in PTLD disease, two classification profiles were identified. Interestingly, both profiles follow a common path including the proinflammatory cytokines GM-CSF and IL-2 in concordance with IL-2 mediated GM-CSF production previously reported⁹. One PTLD profile appears to be driven by IL-8 induced responses while the other mediated by IL-13. PTLD includes a variety of clinical features and pathogenic mechanisms with two different immune clusters¹⁰. Both share features that include T cell receptor signaling and involvement of monocytes/CD4+ T cells. The first cluster characterized by a type 1 inflammatory response associated with post-infectious Lyme arthritis and autoimmune joint disease that are associated with IL-8^10,11. The second cluster includes activation of neutrophils and IL-4/IL-13 signaling¹¹ which aligns more with post-treatment neurological disease¹⁰. These data could explain the two different classification profiles, reported in this study, associated with different clinical manifestations of PTLD.

We agree with the conclusions of Schett et al.¹ that targeting of individual cytokines underlying the immunopathogenesis of these conditions may provide a powerful new tool in the treatment of these immunologically-mediated disorders using precision medicine. Further study may elucidate how pathogen or antigen persistence could contribute to these classifications^12,13.

Ethics:

All the patients/participants provided their written informed consent to participate in this study.

Ethical Review: This study was reviewed and approved by the ethics committee of the Chronic COVID Treatment Center

Decision: Ethical Approval was given.

Data and materials availability:

All requests for materials and raw data should be addressed to the corresponding author

Competing interests:

B.K.P., E.B.F. are employees of IncellDx, Inc. C.B. is a consultant of IncellDx, Inc

Author contributions:

R.Y., P.P. organized the clinical study and actively recruited patients.

B.K.P, E.B.F, J.G-C. performed experiments and analyzed the data.

J.G-C., R.A.M., J.M., C.B. performed the statistics and bioinformatics

B.K.P., J.A.B., G.K., A.K., J.M., E.B.F, J.G-C., R.A.M. wrote and edited the draft of the manuscript and all authors contributed to revising the manuscript prior to submission.

Funding: None

Wong TL, Weitzer DJ. Long COVID and Myalgic Encephalomyelitis/Chronic Fatigue Syndrome (ME/CFS)-A Systemic Review and Comparison of Clinical Presentation and Symptomatology. Medicina (Kaunas). 2021 Apr 26;57(5):418. doi: 10.3390/medicina57050418. PMID: 33925784; PMCID: PMC8145228.
Patterson BK, Guevara-Coto J, Yogendra R, et. al. Immune-based prediction of COVID-19 severity and chronicity decoded using machine learning. Front Immunol. 2021 Jun 28;12:700782. doi: 10.3389/fimmu.2021.700782.
Wong KH, Shapiro ED, Soffer GK. A Review of Post-treatment Lyme Disease Syndrome and Chronic Lyme Disease for the Practicing Immunologist. Clin Rev Allergy Immunol. 2022 Feb;62(1):264-271. doi: 10.1007/s12016-021-08906-w. Epub 2021 Oct 23. PMID: 34687445.
Schett G, McInnes IB, Neurath MF. Reframing Immune-Mediated Inflammatory Diseases through Signature Cytokine Hubs. N Engl J Med. 2021;385(7):628-639.
Mckinney W. Pandas: A Foundational Python Library for Data Analysis and Statistics. http://pandas.sf.net
Brownlee J. Basics of Linear Algebra for Machine Learning: Discover the Mathematical Language of Data in Python.; 2018.
Breiman L, Friedman JH, Olshen RA, Stone CJ. Classification and regression trees. Classif Regres Trees. Published online January 1, 2017:1-358. doi:10.1201/9781315139470
Pedregosa F, Michel V, Grisel O, et al. Scikit-Learn: Machine Learning in Python Vol 12.; 2011. Accessed April 17, 2021. http://scikit-learn.sourceforge.net.
Hartmann FJ, Khademi M, Aram J, Ammann S, Kockum I, Constantinescu C, Gran B, Piehl F, Olsson T, Codarri L, Becher B. Multiple Sclerosis-Associated IL2RA Polymorphism Controls GM-CSF Production in Human TH Cells. Nat Commun. 2014 Oct 3;5:5056. doi: 10.1038/ncomms6056. PMID: 25278028.
Steere AC. Posttreatment Lyme Disease Syndromes: Distinct Pathogenesis Caused by Maladaptive Host Responses. J Clin Invest. 2020 May 1;130(5):2148-2151. doi: 10.1172/JCI138062. PMID: 32281948; PMCID: PMC7190968.
Clarke DJB, Rebman AW, Bailey A, Wojciechowicz ML, Jenkins SL, Evangelista JE, Danieletto M, Fan J, Eshoo MW, Mosel MR, Robinson W, Ramadoss N, Bobe J, Soloski MJ, Aucott JN, Ma'ayan A. Predicting Lyme Disease From Patients' Peripheral Blood Mononuclear Cells Profiled With RNA-Sequencing. Front Immunol. 2021 Mar 8;12:636289. doi: 10.3389/fimmu.2021.636289. PMID: 33763080; PMCID: PMC7982722.
Patterson BK, Francisco EB, Yogendra R, Long E, Pise A, Rodrigues H, Hall E, Herrera M, Parikh P, Guevara-Coto J, Triche TJ, Scott P, Hekmati S, Maglinte D, Chang X, Mora-Rodríguez RA, Mora J. Persistence of SARS CoV-2 S1 Protein in CD16+ Monocytes in Post-Acute Sequelae of COVID-19 (PASC) up to 15 Months Post-Infection. Front Immunol. 2022 doi: 10.3389/fimmu.2021.746021.
Buonsenso D, Piazza M, Boner AL, Bellanti JA. Long COVID: A Proposed Hypothesis-driven Model of Viral Persistence for the Pathophysiology of theSyndrome. Allergy Asthma Proc. 2022 May (in press).

Table 1

Class	Recall		Precision		Specificity		F1 Score
MM	0.8846		0.7419		0.9619		0.8070
Severe	0.6000		0.7143		0.9716		0.6522
PASC	0.8438		0.8308		0.9360		0.8372
Post-Vax	0.8333		0.6140		0.8866		0.7071
ME-CSF	0.7000		0.6604		0.9032		0.6796
Lyme	0.1724		0.5556		0.9807		0.2632
Average		0.6724		0.6862		0.9400		0.6577

Cytokine Hub Classification of PASC, ME-CFS and other PASC-like Conditions

Status:

Version 1

Abstract

Background

Methods

Results

Conclusions

Figures

Introduction

Methods

Patients

Multiplex Cytokine Quantification

Data Acquisition and Processing

Multiclass classification of disease states using a decision tree model

Results And Discussion

Conclusions

Declarations

References

Table

Supplementary Files

Status:

Version 1