Chronic inflammation, neutrophil activity, and autoreactivity splits long COVID

Abstract

While immunologic correlates of COVID-19 have been widely reported, their associations with post-acute sequelae of COVID-19 (PASC) remain less clear. Due to the wide array of PASC presentations, understanding if specific disease features associate with discrete immune processes and therapeutic opportunities is important. Here we profile patients in the recovery phase of COVID-19 via proteomics screening and machine learning to find signatures of ongoing antiviral B cell development, immune-mediated fibrosis, and markers of cell death in PASC patients but not in controls with uncomplicated recovery. Plasma and immune cell profiling further allow the stratification of PASC into inflammatory and non-inflammatory types. Inflammatory PASC, identifiable through a refined set of 12 blood markers, displays evidence of ongoing neutrophil activity, B cell memory alterations, and building autoreactivity more than a year post COVID-19. Our work thus helps refine PASC categorization to aid in both therapeutic targeting and epidemiological investigation of PASC.

Fig. 1. Inflammatory protein signatures in PASC.

a–j Blood plasma from 97 PASC patients and 26 CR controls was assessed for 2925 independent protein features. a Left—Principal component analysis of PASC and CR cohorts. Large circles indicate population centroids. Right—Scree plot of the explained sample variance of the first ten principal components. b Feature-wise comparison between PASC and CR cohorts. Proteins of interest are labeled with significant differential abundances (Adj P < 0.001) highlighted in red. c, f, h, j Normalized abundance of indicated proteins with means displayed (red). d Reactome pathway analysis of proteins ranked by Spearman’s correlations with PASC diagnosis. Pathways of interest are labeled. e, g, i Gene set enrichment analysis of indicated pathways. c, f, h, j Two-way unpaired T test. *P < 0.05; **P < 0.01; ***P < 0.001, ****P < 0.0001. Source data are provided as a Source Data file.

Fig. 2. RF modeling identifies PASC features.

a–g Blood plasma from 97 PASC patients and 26 CR controls was assessed for 2925 independent protein features. a Cartoon overview of random forest modeling approach and feature potency assessment. b Left—Receiver-operating characteristic (ROC) plot displaying ten models with randomized train/test splits classifying PASC and CR. Right—probabilistic classification plots for individual patients from the test sets derived from the ten models displayed in the ROC plot. c Feature-wise comparison between PASC and CR cohorts. Proteins of interest are labeled with the 30 features with highest feature potency highlighted. d KEGG pathway analysis of proteins ranked by feature potency in PASC discrimination modeling. e–g Normalized abundance of indicated proteins with means displayed (red). h Correlated but distinct information is provided by EREG and IFI30. Fraction of total samples and % purity indicated for each quadrant. e–g Two-tailed unpaired T test. *P < 0.05; **P < 0.01; ***P < 0.001, ****P < 0.0001. Source data are provided as a Source Data file.

Fig. 3. PASC subsets defined by inflammation.

a–e Blood plasma from 97 PASC patients, 26 CR, 9 SLE, and 28 severe/critical COVID-19 controls was assessed for 2925 independent protein features. a Classification of PASC and CR patients following KNN clustering. b Hierarchical clustering of PASC patients. Major branches are differentially colored. c Quantification of indicated proteins in CR, niPASC, inflPASC, COVID-19, or SLE subjects with median displayed (red). d Generalized linear modeling of individual feature associations with PASC with either nocovariates assigned (left), or DPSO and IDS used as covariates (right). e Quantification of indicated proteins in CR, niPASC, or inflPASC patients more than 89 days post COVID-19 symptom onset with median displayed (red). f Quantification of IL-8 in niPASC and inflPASC patients with mild/moderate COVID-19. g Quantification of IL-8 in niPASC and inflPASC patients more than 89 days post mild/moderate COVID-19 symptom onset with median displayed (red). h–j Clinical blood testing of niPASC and inflPASC patient cohorts. Blue boxes indicate normal testing ranges. h C-reactive protein concentration in inflPASC and niPASC cohorts with mean displayed (red). i Fibrinogen in inflPASC and niPASC cohorts with mean displayed (red). j Neutrophil counts correlated with fibrinogen in PASC (left) and quantified between inflPASC and niPASC patients with mean displayed (red, right). k Neutrophil counts correlated with myeloperixidase expression in the plasma of PASC patients. l Calprotectin and Citrullinated Histone H3 concentration in the plasma of CR, niPASC, or inflPASC patients. c, e One-way ANOVA with multiple comparisons. f–j Two-tailed unpaired T testing. c–j *P < 0.05; **P < 0.01; ***P < 0.001, ****P < 0.0001. Source data are provided as a Source Data file.

Fig. 4. Altered B-cell responses in inflPASC patients.

a–j Flow cytometric analysis B cells from 38 PASC and 11 CR patients. a DN2 B-cell frequency of B-cell-derived cells with median displayed (red). b ASC frequency of B-cell-derived cells with median displayed (red). c SARS-CoV-2 spike specific B-cell frequency of B-cell-derived cells with median displayed (red). d Heatmap of antigen-specific B-cell population frequency z-scores in CR (n = 11), inflPASC (n = 14) or niPASC patients (n = 24). Multivariate clustering of patients by Ward’s method is represented by dendrograms. e Ag-specific DN composition in CR, niPASC, or inflPASC patients. f Ag-specific frequency of non-DN1 (EF-associated) DNs in CR, niPASC, and inflPASC patients with median displayed (red). g Ag-specific memory composition in CR, niPASC, or inflPASC patients. h Ag-specific frequency of IgA or IgG memory in CR, niPASC, and inflPASC patients with median displayed (red). i Representative flow plots of CR (left) or inflPASC (right) ag-specific aN (top) or total memory (bottom) compartments. j (left) Proportion of donors in CR or PASC groups with observable ag-specificity in the aN compartment. Right—Ag-specific aN frequencies of CR or PASC groups from donors with an observable ag-specific aN compartment [from left]. a–c, f, h One-way ANOVA with multiple comparisons. *P < 0.05; **P < 0.01; ***P < 0.001, ****P < 0.0001. Source data are provided as a Source Data file.

Fig. 5. Autoreactive serology in inflPASC.

a–c Plasma from 97 PASC patients was screened for reactivity against SARS-CoV-2. a Serological reactivity against the spike receptor binding domain in inflPASC and niPASC patients, by isotype, with means displayed (red). b Serological reactivity against nucleocapsid in inflPASC and niPASC patients, by isotype, with means displayed (red). c Serological anti-nucleocapsid responses in patients in inflPASC and niPASC cohorts more than 120 days DPSO with means displayed (red). d Normalized abundance of indicated proteins with means displayed (red). e–g Plasma from 96 PASC patients with mixed symptomology were screened by Exagen clinical laboratory for reactivity against 30 clinically relevant autoantigens. e Heatmap of patient results. Each column represents a single patient grouped by the total number of autoreactive positive tests that the patient displayed. Bolded boxes represent clinical positive tests with the color indicating the magnitude of the test result. Scale for each test is documented below the heatmap. f ANA titers in niPASC and inflPASC patients. g Total positive autoreactive tests in niPASC and inflPASC patients. h Trend in ANA titers in (n = 7) niPASC and (n = 8) inflPASC patients 1 year after initial plasma collection and screening. a–c Two-tailed unpaired T testing. d One-way ANOVA with multiple comparisons. *P < 0.05; **P < 0.01; ***P < 0.001, ****P < 0.0001. Source data are provided as a Source Data file.

Fig. 6. Identifying inflPASC in small feature sets.

a Left—Receiver-operating characteristic (ROC) plot displaying ten models with randomized train/test splits classifying inflPASC from all other recoveries. Right—probability plots for individual “test” patients derived from the ten models displayed in the ROC plot. b Feature-wise comparison between PASC and CR cohorts. 12 manually curated proteins for reduced feature inputs are labeled. c Left—Receiver-operating characteristic (ROC) plot displaying ten models with randomized train/test splits classifying inflPASC from all other recovery using the manually curated list of 12 high-potency proteins identified in (b). Right—probability plots for individual “test” patients derived from the ten models displayed in the ROC plot.