nature immunology

Article https://doi.org/10.1038/s41590-025-02135-5

Identification of soluble biomarkers that

associate with distinct manifestations of
long COVID

Received: 23 May 2024 Yu Gao1,13, Curtis Cai 1,13, Sarah Adamo 1,2,13, Elsa Biteus3,4, Habiba Kamal3,
Lena Dager5, Kelly L. Miners6, Sian Llewellyn-Lacey6, Kristin Ladell 6,
Accepted: 14 March 2025
Pragati S. Amratia6, Kirsten Bentley 6, Simon Kollnberger6, Jinghua Wu1,
Published online: 30 April 2025 Mily Akhirunnesa1, Samantha A. Jones7, Per Julin8,9, Christer Lidman3,
Richard J. Stanton 6, Paul A. Goepfert10, Michael J. Peluso 11,
Check for updates
Steven G. Deeks11, Helen E. Davies 7,13, Soo Aleman3,5,13, Marcus Buggert 1,13 &
David A. Price 6,12,13

Long coronavirus disease (COVID) is a heterogeneous clinical condition

of uncertain etiology triggered by infection with severe acute respiratory
syndrome coronavirus 2 (SARS-CoV-2). Here we used ultrasensitive
approaches to profile the immune system and the plasma proteome in
healthy convalescent individuals and individuals with long COVID, spanning
geographically independent cohorts from Sweden and the United Kingdom.
Symptomatic disease was not consistently associated with quantitative
differences in immune cell lineage composition or antiviral T cell immunity.
Healthy convalescent individuals nonetheless exhibited higher titers of
neutralizing antibodies against SARS-CoV-2 than individuals with long
COVID, and extensive phenotypic analyses revealed a subtle increase in the
expression of some co-inhibitory receptors, most notably PD-1 and TIM-3,
among SARS-CoV-2 nonspike-specific CD8+ T cells in individuals with long
COVID. We further identified a shared plasma biomarker signature of disease
linking breathlessness with apoptotic inflammatory networks centered
on various proteins, including CCL3, CD40, IKBKG, IL-18 and IRAK1, and
dysregulated pathways associated with cell cycle progression, lung injury
and platelet activation, which could potentially inform the diagnosis and
treatment of long COVID.

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has symptomatology, including immune dysregulation, ongoing inflam-
left a pernicious legacy of global ill health, commonly known as long mation and tissue damage, and viral persistence3–9. The reactivation
coronavirus disease (COVID)1. The etiology of this heterogeneous con- of latent herpesviruses may also contribute to the pathogenesis of
dition remains obscure, but common symptoms include breathless- long COVID3,10,11.
ness, cognitive impairment, often described as ‘brain fog’, fatigue and A handful of ‘omics approaches have been used to probe the mole­
pain, alongside a host of other clinical manifestations indicating the cular intricacies of long COVID. For example, affinity proteomics studies
involvement of different organ systems in the body2,3. Several hypoth- have identified distinct inflammatory phenotypes and enrichment of
eses have been proposed to account for such diverse and persistent the NF-κB and type 2 interferon (IFN) signaling pathways as correlates

A full list of affiliations appears at the end of the paper. e-mail: [email protected]; [email protected]

Nature Immunology | Volume 26 | May 2025 | 692–705 692

Article https://doi.org/10.1038/s41590-025-02135-5

of disease, highlighting associations with various soluble biomar­ Table 1 | Key features of the primary cohort recruited from
kers, such as IFNγ, interleukin-1 (IL-1), IL-6 and tumor necrosis factor the United Kingdom
(TNF), which are typically upregulated in acute COVID-19 (refs. 12–14).
Similar findings have been described using conventional approaches Characteristic Controls (n = 70) Cases (n = 70)
to cytokine quantification15. A longitudinal multiomics study fur- Age (years), median (range) 43 (21–79) 45 (20–74)
ther reported that various autoantibodies, altered cytomegalovirus Female (%) 54 (77.1) 52 (74.3)
(CMV)-specific and SARS-CoV-2-specific CD8+ T cell dynamics, and
Race (%)
Epstein–Barr virus (EBV) and SARS-CoV-2 viremia were associated with
the emergence of particular subtypes of long COVID10. More recently, White 58 (82.9) 62 (88.6)
another multiomics study found that elevated herpesvirus-specific Black 0 (0.0) 1 (1.4)
antibody titers, immune cell perturbations, and decreased cortisol
Asian 9 (12.9) 4 (5.7)
levels were distinguishing features of persistent illness after infection
with SARS-CoV-2 (ref. 16). By contrast, hypothesis-driven approaches Mixed 2 (2.9) 2 (2.9)
and low-resolution proteomics have identified dysregulation of the Other 1 (1.4) 1 (1.4)
complement system, which is known to drive inflammation, as a con- BMI > 30 kg m–2 (%) 21/66 (31.8) 32/69 (46.3)
sistent feature of long COVID17,18. These observations suggest that
Coexisting conditions (%)
multiple factors could be associated with the development of discrete
symptom complexes and patterns of disease onset within the clinically Yes 13 (18.6) 35 (50.0)
diverse spectrum of long COVID. No 57 (81.4) 35 (50.0)
In this study, we used a variety of multidimensional approaches Employment status (%)
and integrative data analysis pipelines to profile the immune system
Pre-COVID-19
and the plasma proteome in healthy convalescent individuals with
a molecularly confirmed history of infection with SARS-CoV-2 and Employed 45/49 (91.8) 59/66 (89.4)
individuals with long COVID, spanning geographically independent Employed, altered duties 0/49 (0.0) 0/66 (0.0)
cohorts from Sweden and the United Kingdom. We found higher titers
Employed, sick leave 0/49 (0.0) 0/66 (0.0)
of SARS-CoV-2 spike-specific neutralizing antibodies in healthy conva-
lescent individuals than in individuals with long COVID. By contrast, Unemployed 0/49 (0.0) 1/66 (1.5)
minimal intergroup differences were apparent in immune cell line- Retired 1/49 (2.0) 4/66 (6.1)
age composition and virus-specific CD4+ and CD8+ T cell immunity, Student 3/49 (6.1) 2/66 (3.0)
although some co-inhibitory receptors, especially PD-1 and TIM-3, were
Post-COVID-19
relatively overexpressed among SARS-CoV-2 nonspike-specific CD8+
T cells in individuals with long COVID. We also detected a unique array Employed 46/49 (93.9) 35/66 (53.0)
of soluble biomarkers in the plasma proteome that correlated directly Employed, altered duties 0/49 (0.0) 11/66 (16.7)
with the clinical manifestation of breathlessness in individuals with
Employed, sick leave 0/49 (0.0) 11/66 (16.7)
long COVID. Network and pathway analyses linked these biomarker
signatures with apoptotic processes and inflammation, highlighting Unemployed 0/49 (0.0) 2/66 (3.0)

key roles for signaling cascades involving ceramide, FAS, NF-κB and Retired 1/49 (2.0) 5/66 (7.6)
TNF. Moreover, core network components, including CCL3, CD40 and Student 2/49 (4.1) 2/66 (3.0)
IL-18, were identified as potential contributors to persistent inflamma-
Date of infection with SARS-CoV-2 (%)
tion in individuals with long COVID. These results provide a mechanis-
tic framework to unravel the complex etiology and pathogenesis of March 2020 to August 2020 11 (15.7) 14 (20.0)
ongoing symptomatic disease triggered by infection with SARS-CoV-2. September 2020 to June 2021 23 (32.9) 25 (35.7)
July 2021 to October 2022 36 (51.4) 31 (44.3)
Results
COVID-19 vaccination status, median (IQR)
Clinical characterization
The primary cohort included healthy convalescent individuals (con- Number of vaccinations before infection 1.5 (0–3) 0 (0–2)
trols; n = 70) and individuals with long COVID (cases; n = 70) recruited Total number of vaccinations 3 (3–3) 3 (3–3)
from University Hospital Llandough (Table 1 and Supplementary
Symptoms, median (IQR)a
Table 1). All participants had a clearly defined episode of symptomati-
cally mild acute COVID-19 confirmed via direct molecular evidence of Breathlessness 0 (0–0) 3 (2–6)
infection with SARS-CoV-2. Intergroup comparisons revealed largely Fatigue 0 (0–2) 6 (4–8)
equivalent distributions for age (cases, median = 45 years; controls, Musculoskeletal 0 (0–0) 3.5 (0–6)
median = 43 years), body mass index (BMI; cases, median = 29.8 kg m–2;
Neuropsychiatric 0 (0–0) 4 (1–6)
controls, median = 28.5 kg m–2), race (cases, White = 88.6%; controls,
White = 82.9%), sex (cases, female = 74.3%; controls, female = 77.2%), Pain 0 (0–0) 4 (2–5)
time since initial reported infection (cases, median = 416 days; con- Ability to maintain self-care, median (IQR) 0 (0–0) 0 (0–3)
trols, median = 268 days), and vaccination against SARS-CoV-2 (cases,
Ability to maintain daily tasks, median (IQR) 0 (0–0) 5 (3–8)
median number of vaccinations = 3; controls, median number of vacci-
nations = 3; Fig. 1a,b and Table 1). All baseline medical evaluations were Overall general health, median (IQR)b 0 (−1 to 0) −4 (−2 to −6)

normal in individuals with long COVID. Symptom scores are depicted IQR, interquartile range. aDifference in numeric rating scale score (0 = no symptom, 10 = worst
possible symptom) before versus after COVID-19. bDifference in numeric rating scale score
in Fig. 1c. Breathlessness was further assessed using the Dyspnea-12
(0 = worst possible, 10 = best possible) before versus after COVID-19.
questionnaire, scored out of 36, and the Nijmegen questionnaire,
scored out of 64, which provide a metric for hyperventilation (Fig. 1d). cohort included healthy convalescent individuals (controls; n = 30) and
Pain was most commonly localized to the chest (31%), joints (26%) and individuals with long COVID (cases; n = 95) recruited from Karolinska
muscles (16%) in individuals with long COVID (Fig. 1e). The secondary University Hospital (Table 2).

Nature Immunology | Volume 26 | May 2025 | 692–705 693

Article https://doi.org/10.1038/s41590-025-02135-5

a Sex b
Female Male
77% 23%

Time from acute COVID-19 (days)

60 0.11 0.01
100 0.44 1,000
HC
80 800
40

BMI (kg m–2)

Age (years)
60 600
Male
Female 40 400
26%
74% 20

LC 20 200

0 0 0
HC LC HC LC HC LC

c Numeric rating scale score d Breathlessness score e Main site of pain

HC LC HC LC Muscle/back
Unsure Chest
1%
13% 31%
–13 –7 –13 –9 –7 –8 –7 –16 –16
2.3 × 10 1.1 × 10 3.8 × 10 1.0 × 10 1.4 × 10 1.1 × 10 1.1 × 10 1.3 × 10 6.7 × 10
15
60 Abdominal pain
1.5 × 10–12 4%

Head
10 9%
40
2.6 × 10–13
Score

LC
Score

5 20

0 0
16%
Muscle 26%
Dyspnea 12 (of 36) Nijmegen (of 64)
ue

in

on

re

lth
ss

es
ty

y
et

Joint
Pa

ca
ne

ili

iti
tig

si

ea
xi
ob

es

iv
ss

al

lh
An
Fa

ct
pr
M
le

on

ra
la
th

De

rs

ne
ua
ea

Pe

ge
Us
Br

ll
ra
ve
O

f 0.005 g 8
h 100 10
0.34
10
7 6 × 10
0.03 0.33
Normalized max. CD107a (%)

6 0.96
10
SARS-CoV-2 IgG (U ml–1)
Neutralization (log NT50)

ADNKA (log AUC)

5
10 8 1
4 × 10
4
10
10
3
10
2 × 10
8
0.1
2
10
1
10
0
10 0 1 0.01
HC LC LC HC LC HC LC HC LC
al

d
de
in
rig

en
O

t
Ex

Fig. 1 | Cohort overview and humoral immunity in healthy convalescent highest plasma dilution that achieved a 50% reduction in plaque formation
individuals and individuals with long COVID recruited from the United (NT50; HC, n = 70; LC, original n = 70 and extended n = 146). g, Scatter dot plot
Kingdom. a, Ring charts and scatter dot plots showing sex, age and BMI for showing total SARS-CoV-2 spike-specific immunoglobulin titers (HC, n = 52;
healthy convalescent individuals (HC, n = 70) and individuals with long COVID LC, n = 57). h, Scatter dot plots showing maximum CD107a mobilization (left)
(LC, n = 70). b, Scatter dot plots showing the corresponding time to sampling quantified as percentage values relative to the corresponding positive controls
from the initial diagnosis of acute COVID-19. c, Violin plots showing the and normalized ADNKA (right) quantified as a function of degranulation
corresponding distribution of clinical symptom numeric rating scale scores. (CD107a+) among viable NK cells (Aqua−CD3−CD56+) with potent cytotoxic
d, Scatter dot plots showing breathlessness scores as assessed using the activity (CD57+; HC, n = 55 and n = 66, respectively; LC, n = 40 and n = 66,
Dyspnea-12 and Nijmegen questionnaires (HC, n = 49 and n = 49, respectively; respectively); AUC, area under the curve. Horizontal bars represent median
LC, n = 66 and n = 62, respectively). e, Ring chart highlighting the anatomical values (a–d and f–h). Significance was evaluated using a two-tailed
distribution of pain experienced by individuals with long COVID. f, Scatter dot Mann–Whitney U-test (a–d and f–h).
plot showing SARS-CoV-2-specific neutralization activity quantified as the

Neutralizing antibody titers are suboptimal in long COVID activity in standard plaque reduction assays than individuals with long
To evaluate the humoral immune system, we measured total SARS-CoV-2 COVID (Fig. 1f), despite equivalent overall titers of antibodies targeting
spike-specific immunoglobulin titers, virus neutralization activity, and the spike protein of SARS-CoV-2 (Fig. 1g). This finding was confirmed
antibody-dependent natural killer (NK) cell activation (ADNKA) in across a larger number of donors from the same cohort17, achieving
plasma samples obtained from donors in the United Kingdom. Healthy even greater significance (Fig. 1f). By contrast, no such intergroup dif-
convalescent individuals exhibited substantially better neutralization ferences were apparent for ADNKA measured as a cumulative metric

Nature Immunology | Volume 26 | May 2025 | 692–705 694

Article https://doi.org/10.1038/s41590-025-02135-5

Table 2 | Key features of individuals with long COVID of CD69 and 4-1BB (CD137) after peptide stimulation directly ex vivo22,23.
recruited from Sweden PBMCs were stimulated with individual peptide pools spanning the
major immunogenic proteins from SARS-CoV-2 (spike, nucleocapsid,
Characteristic Borg CR10 ≤ 2 (n = 60) Borg CR10 > 2 (n = 35) combined membrane and envelope, ORF1a, ORF1b and ORF3–ORF10)
Age (years), 46 (21–64) 50 (23–66) and selected immunogenic proteins from CMV (IE-1, IE-2 and pp65) and
median (range) EBV, the latter segregated according to lytic (BRLF1, BZLF1, BMLF1 and
Female (%) 88.7% 75.6% BARF1) and latent phases (EBNA1, EBNA2, EBNA3A, EBNA3B, EBNA3C
and LMP2) of the viral life cycle (Extended Data Fig. 2b). The frequencies
BMI > 30 kg m–2 (%) 15/41 (22.9) 19/45 (27.2)
of antiviral CD4+ and CD8+ T cells were statistically indistinguishable
across all of these specificities in healthy convalescent individuals
against all expressed viral target proteins using healthy donor cell and individuals with long COVID recruited from the United Kingdom
preparations with a surrogate marker of potential cytotoxicity (Fig. 1h), (Fig. 3a). By contrast, the frequencies of CD4+ T cells targeting the
namely CD57 (ref. 19). SARS-CoV-2 nucleocapsid protein and the EBV latent proteins and the
Collectively, these findings identify a qualitative deficit in the frequencies of CD8+ T cells targeting the SARS-CoV-2 spike protein, the
humoral immune response against SARS-CoV-2, specifically impacting CMV proteins and the EBV lytic proteins were higher in individuals with
neutralization activity in individuals with long COVID. long COVID than in healthy convalescent individuals recruited from
Sweden (Extended Data Fig. 3e).
Immune cell perturbations are limited in long COVID In further experiments, we measured the expression of immu-
To evaluate the cellular immune system, we first conducted a multi- nophenotypic markers related to activation, memory, effector function
dimensional flow cytometric analysis of the major lineages typically and exhaustion among CD4+ and CD8+ T cells targeting defined proteins
present among peripheral blood mononuclear cells (PBMCs), focusing from SARS-CoV-2, CMV or EBV. In the primary cohort, no significant
initially on donors recruited from the United Kingdom (Fig. 2a). Using intergroup differences in expression intensity were observed for CD28,
dimensionality reduction and Gaussian mixture models, we identified CD39, CD71, CD95, CX3CR1 or PD-1, but some markers of activation
clusters that corresponded to the major lineages of monocytes, B cells, (CD38 and HLA-DR), exhaustion (TIGIT) and stemness (CD127) were
NK cells and T cells (Fig. 2b and Extended Data Fig. 1), but these analyses variably downregulated among some antiviral CD4+ and CD8+ T cell
were unable to differentiate between some other immune cell subsets, populations in the context of long COVID (Fig. 3b). More profound dif-
such as basophils and plasmacytoid dendritic cells (pDCs; Fig. 2b), ferences were apparent in the secondary cohort, potentially reflecting
and were also unable to differentiate between healthy convalescent the limited number of healthy convalescent individuals relative to the
individuals and individuals with long COVID (Fig. 2c). We therefore number of individuals with long COVID (Extended Data Fig. 3f). Line-
interrogated the data conventionally using a manual flow cytometric age analysis further revealed comparable expression of CD38, CD69,
gating strategy (Extended Data Fig. 2a). HLA-DR and PD-1 among global CD4+ and CD8+ effector memory T cells
In the adaptive lymphocyte compartment, similar proportions of and terminally differentiated effector memory T cells in healthy con-
naive B cells, total B cells, naive T cells, total T cells, naive CD4+ T cells, valescent individuals and individuals with long COVID recruited from
total CD4+ T cells, naive CD8+ T cells and total CD8+ T cells were iden- the United Kingdom (Fig. 3c).
tified in healthy convalescent individuals and individuals with long Collectively, these results demonstrate that experimental findings
COVID (Fig. 2d), and in the innate lymphocyte compartment, similar are not necessarily transferable across geographically distinct cohorts
proportions of immature NK cells (CD16−CD56bright), mature NK cells of individuals with long COVID, likely reflecting differences in clinical
(CD16+CD56dim), total NK cells (including CD16−CD56dim) and total characterization and sample size. Our findings nonetheless align with
innate lymphoid cells (CD127+) were identified in healthy convalescent the notion that circulating antiviral CD4+ and CD8+ T cell populations
individuals and individuals with long COVID (Fig. 2e). A comparable are largely equivalent in healthy convalescent individuals and individu-
pattern was observed for classical monocytes (CD14+), intermedi- als with long COVID24.
ate monocytes (CD14+CD16+) and conventional DCs (CD11c+CD123−)
in the myeloid cell lineage, whereas the proportions of nonclassical SARS-CoV-2-specific CD8+ T cells are phenotypically variable
monocytes (CD16+) were relatively increased in healthy convalescent To refine our phenotypic analyses, which were potentially confounded
individuals (Fig. 2f), and the proportions of basophils (CD123+HLA-DR−) by alterations in surface marker expression arising as a consequence
and pDCs (CD123+HLA-DR+) were relatively increased in individuals of antigen-induced activation, we used peptide–HLA class I tetramers
with long COVID (Fig. 2g). Hierarchical clustering confirmed these directly ex vivo to identify and characterize unperturbed CD8+ T cells
differences within an otherwise rather uniform immune cell landscape targeting specific epitopes from SARS-CoV-2, CMV, EBV or influenza
(Fig. 2h,i). By contrast, no such perturbations were apparent in the A virus (IAV)25,26. For this purpose, we selected healthy convalescent
secondary cohort of donors recruited from Sweden, although the individuals (n = 17) and individuals with long COVID (n = 15) from
proportions of classical monocytes were relatively decreased and the the primary cohort based on the expression of HLA-A*02:01 and/or
proportions of intermediate monocytes were relatively increased in HLA-B*07:02. As a means to calibrate our findings against CD8+ T cells
individuals with long COVID (Extended Data Fig. 3a–d). with known features of exhaustion27, we also performed similar analy-
Collectively, these data indicate that immune cell perturbations ses using samples from untreated individuals infected with human
are quantitatively subtle and, despite intercohort variability, generally immunodeficiency virus type 1 (HIV-1), extending the range of spe-
confined to the myeloid compartment in individuals with long COVID. cificities to include epitopes restricted by HLA-A*24:02, HLA-B*08:01
and HLA-B*57:01 (Fig. 4a).
T cell immunity remains largely unaltered in long COVID CD8+ T cells targeting spike epitopes from SARS-CoV-2 expressed
To extend these findings, we quantified CD4+ and CD8+ memory T cell CD38 and HLA-DR more frequently than CD8+ T cells targeting non-
responses against SARS-CoV-2 and the persistent herpesviruses CMV spike epitopes from SARS-CoV-2 (Fig. 4b and Extended Data Fig. 4a),
and EBV, exposure to which has been differentially linked with the devel- likely as a consequence of repeated subunit vaccination. Similarly,
opment of long COVID10,16,20,21. We used activation-induced marker (AIM) CD8+ T cells targeting viral epitopes associated with persistent (CMV,
assays for this purpose, enumerating functional antigen-specific CD4+ EBV and HIV-1) or recurrent antigen exposure (IAV) expressed CD38
T cells by assessing the upregulation of CD69 and CD40L (CD154) and and HLA-DR more frequently than CD8+ T cells targeting nonspike
functional antigen-specific CD8+ T cells by assessing the upregulation epitopes from SARS-CoV-2, and a comparable pattern was observed

Nature Immunology | Volume 26 | May 2025 | 692–705 695

Article https://doi.org/10.1038/s41590-025-02135-5

a b c
General Lymphoid Myeloid CD14+ HC
monocytes 1
CD45 CD3 CD86 10 CD16+
Basophils
and pDCs
LC
HLA-DR 6

Count (× 104)
CD34 CD19 monocytes

Viability CD4 CD14

CD8 CD16 NK cells 4
5
CD56 CD123 CD14+ B cells 2
CD11c monocytes 2

UMAP 2
UMAP 2
0 CD4+ T cells 0
(Teff)
es ) lls lls s ) e) ) 1 2
Memory Activation yt RA ce ce DC T CM iv (T eff tes es
CD45RA PD-1 oc T EM NK B d p nd (na lls cy cyt
CD8+ T cells
on nd an e a lls ce no no
CCR7 CD38 –5 (Teff and TEMRA) + m a ils iv ce + T+ m+o mo
CD27 CD69 D 16 (T eff
s o ph (na + T D4 14 4
C ell s ls 8 C D D1
CD127 CD71 UMAP 1 c Ba cel CD C C
CD8+ T cells + T + T
CX3CR1 CD83 –10 CD4+ T cells
(naive) 8 4
(naive and TCM)
C D D
C
–10 –5 0 5 10 15
UMAP 1
d Naive B cells B cells T cells Naive CD4+ T cells CD4+ T cells Naive CD8+ T cells CD8+ T cells

Percentage of CD4+ T cells

Percentage of CD8+ T cells

Percentage of total B cells

Percentage of CD45+ cells

Percentage of CD45+ cells

Percentage of CD45+ cells

Percentage of CD45+ cells

0.13
100 50 100 100 100 100 50 0.84
0.93 0.97
0.38
80 40 80 80 80 80 40
0.22
60 30 0.46 60 60 60 60 30

40 20 40 40 40 40 20

20 10 20 20 20 20 10

0 0 0 0 0 0 0

HC LC HC LC HC LC HC LC HC LC HC LC HC LC

e NK cells ILCs f CD14+ monocytes Intermediate monocytes CD16+ monocytes

Percentage of CD45+ cells

Percentage of CD45+ cells

0.11
50 0.6 100
Percentage of total

Percentage of total

Percentage of total
0.92 20 20 0.037
40 0.49 80 0.47
monocytes

monocytes

monocytes
15 15
30 0.4 60
10 10
20 40
0.2
5 5
10 20

0 0 0 0 0

HC LC HC LC HC LC HC LC HC LC

Immature NK cells Mature NK cells g Basophils cDCs pDCs

Percentage of CD45+ cells

Percentage of CD45+ cells

Percentage of CD45+ cells

0.0018
Percentage of NK cells

Percentage of NK cells

50 100 0.87 6
4 0.013 4
5
40 80 0.52
3 3
4
30 0.74 60
2 2 3
20 40
2
1 1
10 20 1
0 0 0 0 0

HC LC HC LC HC LC HC LC HC LC
h
i
0.2
Group Group
Basophils HC
NK cells LC 0.1
Mean z score

Mature NK cells
CD14+ monocytes z score 0
pDCs
ILCs 4
–0.1
B cells
Naive B cells
2 –0.2
Naive CD4+ T cells
+
Naive CD8 T cells
Immature NK cells 0
ai r T s
C NK lls

oc lls
es
ils

ai oc s
B es

cD ls
ia re c s
ai m K s
C no lls

M CD T c s

T lls
B lls
lls
C C T LCs

m T c ls
N atu 8 + ell
C

ed tu NK C
N te N ell

e
l

6 + 4 + el

yt
ph

ve e ce

on e
ve ty

+ t
ce

ve o ce

+ e
ce
ce
on pD

D8 cy

D4 c

D1 D c
I

T cells
so
Ba

CD4+ T cells
m

–2
+

N
4

cDCs
D1

rm m
te m
C

+
In I

CD8 T cells
CD16+ monocytes
Intermediate monocytes

Fig. 2 | Immune cell lineages in healthy convalescent individuals and individuals of innate lymphocytes gated manually; ILCs, innate lymphoid cells. f, Scatter dot
with long COVID recruited from the United Kingdom. a, List of surface markers plots showing the frequencies of monocytes gated manually. g, Scatter dot plots
used to characterize immune cell lineages in the periphery. b, Uniform manifold showing the frequencies of basophils and DCs gated manually; cDCs, conventional
approximation and projection (UMAP) representation of immune cell lineages DCs. h, Heat map showing hierarchically clustered z scores derived from the
identified via dimensionality reduction of marker expression values; Teff, effector T frequencies of immune cell subsets gated manually. i, Bar plot showing mean
cells; TEMRA, terminally differentiated effector memory T cells; TCM, central memory z scores for each immune cell subset gated manually for individuals with long
T cells. c, Distribution of cells by group of origin in UMAP space (left) or within COVID (healthy convalescent individuals, n ≤ 70; individuals with long COVID,
UMAP clusters (right). d, Scatter dot plots showing the frequencies of naive and n ≤ 70; b–i). Horizontal bars represent median values (d–g). Significance was
total B and T cells gated manually. e, Scatter dot plots showing the frequencies evaluated using a two-tailed Mann–Whitney U-test (d–g).

Nature Immunology | Volume 26 | May 2025 | 692–705 696

Article https://doi.org/10.1038/s41590-025-02135-5

a Spike Nucleocapsid Mem + Env

0.19 0.22
100 10 10

Percentage of T cells

Percentage of T cells
Percentage of T cells 0.19 0.72
0.9
10 0.9 1 1
1
0.1 0.1
0.1
0.01 0.01
0.01

0.001 0.001 0.001

CD69+ CD40L+ CD69+ 4-1BB+ CD69+ CD40L+ CD69+ 4-1BB+ CD69+ CD40L+ CD69+ 4-1BB+
ORF1a ORF1b 1 ORF3–10

Percentage of T cells
10 0.09 0.81
Percentage of T cells

10

Percentage of T cells
0.34 0.41
0.92 0.93
1 1 0.1
HC
0.1 0.1 LC
0.01
0.01 0.01

0.001 0.001 0.001

CD69+ CD40L+ CD69+ 4-1BB+ CD69+ CD40L+ CD69+ 4-1BB+ CD69+ CD40L+ CD69+ 4-1BB+
CMV EBV latent EBV lytic
0.25 0.61 0.16
0.79
10 1 10
0.9
Percentage of T cells

Percentage of T cells

Percentage of T cells
1 1 0.9
0.1
0.1 0.1
0.01
0.01 0.01

0.001 0.001 0.001

CD69+ CD40L+ CD69+ 4-1BB+ CD69+ CD40L+ CD69+ 4-1BB+ CD69+ CD40L+ CD69+ 4-1BB+

b CD4 T cells + +
CD8 T cells
Spike * **
Nucleocapsid *
Mem + Env **
ORF1a * log2 (fold change)

ORF1b 3
2
ORF3–10 1
0
CMV

EBV latent
**
EBV lytic *
8

9
8

R
1
R1

-1

C 71
R1

-1
5

5
27
IT

IT

27
D7
D2

D3

D3

D3
-D

D2

D3

-D
D9

D9
PD

PD
G

G
C

C
D1

D1
LA

LA
C

C
TI

TI
X3

X3
C

C
C

C
C
C

C
C

C
H

H
C

c CD4+ T cells CD8+ T cells

PD-1 HLA-DR CD69 CD38 PD-1 HLA-DR CD69 CD38
100 50 50 100 50 50 50
Percentage of TEM
Percentage of TEM

10
0.5 0.061 0.049
80 40 40 80 40 40 40
CD4+ T cells

CD8+ T cells

0.44 0.48
8
60 30 30 60 30 30 30
6 0.48 0.35 0.73
40 20 4 20 40 20 20 20

20 10 2 10 20 10 10 10

0 0 0 0 0 0 0 0

HC LC HC LC HC LC HC LC HC LC HC LC HC LC HC LC

100 100 50 100 100 50 50 50

Percentage of TEMRA

Percentage of TEMRA

0.79
80 80 40 80 80 0.34 40 40 40
CD4+ T cells

CD8+ T cells

0.41 0.83 0.065

0.89
60 60 30 60 60 30 0.68 30 30
0.37
40 40 20 40 40 20 20 20

20 20 10 20 20 10 10 10

0 0 0 0 0 0 0 0

HC LC HC LC HC LC HC LC HC LC HC LC HC LC HC LC

Fig. 3 | T cell immunity in healthy convalescent individuals and individuals population among individuals with long COVID versus healthy convalescent
with long COVID recruited from the United Kingdom. a, Scatter dot plots individuals; *P < 0.05 and **P < 0.01. c, Scatter dot plots showing the frequencies
showing the frequencies of functional CD4+ and CD8+ T cells targeting defined of functional CD4+ and CD8+ effector memory T (TEM; top) cells and terminally
proteins from SARS-CoV-2, CMV or EBV; Mem, membrane; Env, envelope. b, Heat differentiated effector memory T (TEMRA; bottom) cells expressing the indicated
map summarizing the phenotypic attributes of functional CD4+ and CD8+ T cells activation markers; healthy convalescent individuals, n ≤ 70; individuals with
targeting defined proteins from SARS-CoV-2, CMV or EBV. Data are shown for long COVID, n ≤ 70 (a–c). Horizontal bars represent median values (a and c).
each marker as the log2-transformed fold change in percent positive for each Significance was evaluated using a two-tailed Mann–Whitney U-test (a–c).

Nature Immunology | Volume 26 | May 2025 | 692–705 697

Article https://doi.org/10.1038/s41590-025-02135-5

for expression of the cytotoxic serine protease granzyme B (Fig. 4b). individuals than in individuals with long COVID (Fig. 4k). Moreover,
Co-inhibitory receptor expression also varied as a function of viral CD8+ T cells targeting lytic epitopes from EBV expressed CXCR3 more
specificity, typically paralleling the likely frequency of antigen expo- frequently, granzyme B less frequently, and TCF-1 more intensely in
sure (Fig. 4c). Of particular note, we found that CD8+ T cells targeting healthy convalescent individuals than in individuals with long COVID
spike epitopes from SARS-CoV-2 expressed co-inhibitory receptors (Fig. 4l and Extended Data Fig. 4c). No such differences were observed
more intensely than CD8+ T cells targeting nonspike epitopes from for CD8+ T cells targeting epitopes from CMV or CD8+ T cells targeting
SARS-CoV-2, based on a combined score for PD-1, TIM-3, LAG-3 and latent epitopes from EBV (Extended Data Fig. 4d,e).
TIGIT (Fig. 4d). No such differences were observed for the transcrip- Collectively, these findings suggest a possible role for cumulative
tion factors TCF-1, T-BET or EOMES (Fig. 4e). However, CD8+ T cells viral antigen exposure in the pathogenesis of long COVID, potentially
targeting epitopes from CMV or HIV-1 expressed T-BET more intensely accompanied by suboptimal immune control of EBV.
than CD8+ T cells targeting nonspike epitopes from SARS-CoV-2, and
CD8+ T cells targeting epitopes from CMV, EBV or HIV-1 expressed Plasma proteomic signatures of breathlessness in long COVID
EOMES more intensely than CD8+ T cells targeting nonspike epitopes To explore disease pathogenesis more systematically, we used a
from SARS-CoV-2 (Fig. 4e). data-driven approach to select healthy convalescent individuals
Collectively, these observations support the premise that antigen (n = 51) and individuals with long COVID (n = 51) from the primary
exposure drives the expression of activation markers and co-inhibitory cohort for plasma proteome characterization using a Proximity Exten-
receptors as a function of viral specificity and further suggest that such sion Assay (Olink Explore 3072). Briefly, immune cell subset propor-
encounters are not sufficiently frequent in the convalescent phase to tions were summarized via principal component analysis (PCA),
induce exhaustion among CD8+ T cells targeting nonspike epitopes and outlier samples were excluded based on the greatest deviation
from SARS-CoV-2, irrespective of progression to long COVID. from the origin along PC1 to PC4. Target proteins were grouped into
eight panels under the following broad themes: cardiometabolic
SARS-CoV-2-specific CD8+ T cell phenotypes in long COVID (n = 2), inflammation (n = 2), neurology (n = 2) and oncology (n = 2).
To determine if any of these phenotypic attributes segregated with PCA revealed that donors could not be separated by disease status
disease, we visualized our flow cytometry data using the dimensionality (Fig. 5a) but could be separated to some extent by BMI (Extended
reduction technique uniform manifold approximation and projec- Data Fig. 5a). We then performed a differential expression analysis,
tion (UMAP), focusing on CD8+ T cells targeting nonspike epitopes which revealed a skewed upregulation of many proteins in individuals
from SARS-CoV-2. A largely overlapping distribution was observed with long COVID (Supplementary Table 2), although most fell below
for healthy convalescent individuals and individuals with long COVID the threshold for significance after multiple-hypothesis correction
(Fig. 4f). Phenograph analysis further revealed seven clusters, most (Fig. 5b and Extended Data Fig. 5b). A gene set enrichment analysis
of which displayed an even representation (Fig. 4g). However, clus- (GSEA) further showed that several pathways, including those related
ters 3 and 7 were more obviously represented among healthy con- to ceramide, platelet-derived growth factor receptor-β (PDGFRB)
valescent individuals, and cluster 5 was more obviously represented and HIV-1 Nef, were associated with this proteomic signature of long
among individuals with long COVID (Fig. 4g). Of note, cluster 5 exhib- COVID (Extended Data Fig. 5c).
ited the highest expression intensities of co-inhibitory receptors, To extend our analyses beyond a simple binary classification, we
including PD-1 (Fig. 4g and Extended Data Fig. 4b). In line with this stratified donors into three groups for each clinical symptom, irrespec-
observation, we found that CD8+ T cells targeting nonspike epitopes tive of the initial categorization as healthy convalescent individuals or
from SARS-CoV-2 expressed co-inhibitory receptors more intensely in individuals with long COVID (Fig. 5c and Extended Data Fig. 5d). Using
individuals with long COVID than in healthy convalescent individuals, this approach, we found that breathlessness was strongly associated
reaching significance for TIM-3 (Fig. 4h). No such differences were with differential protein expression (Extended Data Fig. 5d). Donors
observed for CD8+ T cells targeting spike epitopes from SARS-CoV-2 with severe breathlessness (score of 6–10) segregated from donors with
(Fig. 4h). Moreover, CD8+ T cells targeting nonspike epitopes from no (score of 0) or mild breathlessness (score of 1–5) via PCA (Fig. 5d)
SARS-CoV-2 displayed higher co-inhibitory scores in individuals with and exhibited distinct patterns of protein upregulation (Fig. 5e and
long COVID than in healthy convalescent individuals, suggesting a link Supplementary Table 3). Moreover, GSEA confirmed that severe breath-
between antigen exposure and disease (Fig. 4i). It was also notable that lessness was associated with the enrichment of several pathways that
co-inhibitory scores varied across specificities within the nonspike characterized the proteomic signature of long COVID, including those
repertoire (Fig. 4j). related to ceramide and HIV-1 Nef (Extended Data Fig. 5e). Of note,
In further analyses, we found that CD8+ T cells targeting nonspike breathlessness and other symptom scores were largely independent
or spike epitopes from SARS-CoV-2 expressed TCF-1, a key determi- of age and correlated only weakly with BMI (Extended Data Fig. 5f,g),
nant of memory formation, more intensely in healthy convalescent which is known to impact the plasma proteome28.

Fig. 4 | Phenotypic characteristics of virus-specific CD8+ T cells in healthy intensities of co-inhibitory receptors among tetramer+CD8+ T cells targeting
convalescent individuals and individuals with long COVID recruited from nonspike or spike epitopes from SARS-CoV-2. i, Scatter dot plots showing
the United Kingdom. a, Schematic representation of the experimental co-inhibitory scores among tetramer+CD8+ T cells targeting nonspike or spike
design. b, Scatter dot plots showing the expression frequencies of HLA-DR epitopes from SARS-CoV-2. j, Scatter dot plots showing co-inhibitory scores
and CD38 or granzyme B (GZMB) among tetramer+CD8+ T cells. c, Scatter among tetramer+CD8+ T cells targeting nonspike epitopes from SARS-CoV-2
dot plots showing the expression intensities of co-inhibitory receptors restricted by HLA-A*02:01 or HLA-B*07:02; NC, nucleocapsid. k, Scatter dot
among tetramer+CD8+ T cells. d, Scatter dot plot showing co-inhibitory plots showing the expression intensities of transcription factors among
scores, calculated as the cumulative normalized expression intensities of the tetramer+CD8+ T cells targeting nonspike or spike epitopes from SARS-CoV-2.
co-inhibitory receptors shown in c, among tetramer+CD8+ T cells. e, Scatter l, Scatter dot plots showing the phenotypic characteristics of tetramer+CD8+
dot plots showing the expression intensities of transcription factors among T cells targeting lytic epitopes from EBV; healthy convalescent individuals,
tetramer+CD8+ T cells. f, UMAP visualization summarizing the phenotypic n = 17; individuals with long COVID, n = 15; untreated individuals infected
characteristics of tetramer+CD8+ T cells targeting nonspike epitopes from with HIV-1, n = 14 (b–l). Horizontal bars represent median values (b–e and h–l).
SARS-CoV-2. Individual marker representations are colored by expression Significance was evaluated using a two-tailed Mann–Whitney U-test
intensity. g, Phenograph clustering (top) and cluster distribution of (b–e and h–l); gMFI, geometric mean fluorescence intensity.
tetramer+CD8+ T cells (bottom). h, Scatter dot plots showing the expression

Nature Immunology | Volume 26 | May 2025 | 692–705 698

Article https://doi.org/10.1038/s41590-025-02135-5

–11

a b
3.6 × 10
HLAs Specificities –13 0.00032
5 × 10 150 0.02

HLA-DR CD38 (%)

–8
A*02:01 A*24:02 B*07:02 B*08:01 B*57:01 Spike Nonspike IAV CMV EBV HIV 90
3 × 10
–9 0.8 HC
5.8 × 10
LC

+
0.026

GZMB (%)
–6
2.4 × 10 100
–8
60 4.6 × 10

+
50
30

0 0

e
-1

-1
V

V
ik

ik

ik

ik
IA

EB

IV

IA

EB

IV
sp

Sp

sp

Sp
C

C
H

H
on

on
N

N
c 3.1 × 10
0.006
–10
0.28
0.019
1.6 × 10
9.9 × 10
–6
–5
1.8 × 10
3.1 × 10
–6
–10

2,000 –7 2,000 2,000 –5

3.6 × 10 3,000 0.0065 0.026 1.8 × 10
–5
0.002 4.5 × 10 0.00013 0.028 HC

LAG-3 (gMFI)
1,500 1,500

TIM-3 (gMFI)
1,500

TIGIT (gMFI)
–5
PD-1 (gMFI)

4.9 × 10 0.00094 0.0014 0.19

2,000 LC
1,000 1,000 1,000
1,000
500 500 500
0
0 0 0
–1,000
e

e
-1

-1

-1

-1
V

V
ik

ik

ik

ik

ik

ik

ik

ik
IA

EB

IV

IA

EB

IV

IA

EB

IA

EB

M
IV

IV
sp

Sp

sp

Sp

sp

Sp

sp

Sp
C

C
H

H
on

on

on

on
N

N
d 1.2 × 10
–15
e 0.15
2.3 × 10
–7
7.5 × 10
–5
3.9 × 10
–7

2 × 10
–8 0.043 0.019
0.87 20,000
15 0.87 –10
Co-inhibitory score

2.6 × 10
–12
2,000 4.2 × 10
0.35 0.42

EOMES (gMFI)
0.00013 0.98 HC

T-BET (gMFI)
TCF-1 (gMFI)

0.057 15,000 0.63 800

5.9 × 10
–6 1,500 0.13
10 LC
1,000 10,000 400
5 500 5,000
0
0 0
0
e

e
-1

-1

-1

-1
V

V
ik

ik

ik

ik

ik

ik

ik

ik
IA

EB

IV

IA

EB

IA

EB

M
IV

IA

EB

IV

IV
sp

Sp

sp

Sp

sp

Sp

sp

Sp
C

C
H

H
on

on

on

on
N

N
f CD45RA CCR7 CD95 CD127 CD27 CD38 g
Expression level

Cluster 1
Cluster 2
Cluster 3
Cluster 4
HC Cluster 5
LC Cluster 6
Cluster 7

PD-1 TIM-3 LAG-3 TIGIT CD39 HLA-DR

Cluster 1 Cluster 5
57 43 76 24

CXCR3 CX3CR1 GZMB Ki-67 TCF-1 T-BET EOMES Cluster 2 Cluster 6

58 42 43 57

Cluster 3 Cluster 7
31 69 20 80

h Nonspike Spike
0.099 0.0048 0.71 0.8 0.86 0.56 0.28 0.43
1,200 800

500 1,000 600

600 900 600 Cluster 4
LAG-3 (gMFI)

LAG-3 (gMFI)
TIM-3 (gMFI)

800
TIGIT (gMFI)

TIM-3 (gMFI)

TIGIT (gMFI)
PD-1 (gMFI)

200
PD-1 (gMFI)

48 52
500
400
HC
250 600 400
400 400 LC
0
100 200
200
0 0 300
200
0
0

i Nonspike Spike j Nonspike HLA-A*02:01

(LLL + ALS + LLY)
NC HLA-B*07:02
(SPRWYFYYL)
k Nonspike

0.0027 0.31 0.0071 0.023 0.61 0.59

0.035 1,500 3,000
8
5
Co-inhibitory score
Co-inhibitory score
Co-inhibitory score

Co-inhibitory score

HC 3 HC
EOMES (gMFI)

6
HC
TCF-1 (gMFI)

T-BET (gMFI)

200
4 1,000
10 LC LC 2,000
LC
4 2
3
0
2
2 1 500 1,000
5
0
1
0

l EBV lytic Spike

0.03 0.036 0.059 0.2 0.016 0.25 0.92
6
100 600
Co-inhibitory score

5,000
600
60
EOMES (gMFI)
TFC-1 (gMFI)

200
HC
TCF-1 (gMFI)

T-BET (gMFI)

75 500 4,000
4
HC
CXCR3 (%)

GZMB (%)

500
40
50
LC 400 3,000
LC
400
2 2,000 100
300
20
25 200 1,000
300
0 0 0
100 0

Nature Immunology | Volume 26 | May 2025 | 692–705 699

Article https://doi.org/10.1038/s41590-025-02135-5

a Group
b Cardiometabolic LC
c 30
d Breathlessness
LC Inflammation Group Breathlessness None (0) Severe (6–10)
HC 25 Mild (1–5) Not reported
Significance Neurology LC None (0)
50 HC Low (1–5) 50

No. donors
P < 0.05 Oncology 20
Severe (6–10)
Adjusted P < 0.05 0 40 80 120 15
25 25

PC 2
PC 2

Cardiometabolic HC 10
0 0
Inflammation
5
Neurology
–25 –25
Oncology 0
–50 0 50 0 1 2 3 4 5 6 7 8 9 10 –50 0 50
0 40 80 120
PC 1 Breathlessness score PC 1
No. positive differentially
expressed Olink proteins

e Breathlessness (severe) f Pearson = 0.44, P = 3.53 × 10

–5
Pearson = 0.41, P = 1.3 × 10
–4
versus Adjusted P < 0.05
3 4
Breathlessness (none)
2 3

RPS10
2

IDI2
4 MAP2K6 1
4 Cardiometabolic ANK2
Inflammation CLIP2 RABEP1 1
SPART NFAT5 0
EEF1D STX8 RBPMS 0
BCAT1 DAAM1
NEXN –1
HPSE ECI2 CORO1A VTI1A MYO9B –1
DTD1 PRKG1 0 5 10 0 5 10
HCG22
EIF2AK3 3
3 BNIP2 FRMD4B –4 –4
GGCT SELPLG Pearson = 0.403, P = 1.72 × 10 Pearson = 0.402, P = 1.79 × 10
SKIV2L PKD2
4
–log10 (P)

–log10 (P)

LRRC59 GAPDH
POMC ATP1B2 3 4

SNAPIN
SPESP1 2 CA8

SPRR3
2 MUC2
CST6 BOC 2 3
PPIF PGR SMPD3
POSTN MAN1A2
CD70
SCGB1A1
1 2
LEPR MET
IGDCC4 ACADSB
ENPP5 NTF3 0 1
1 1
0 5 10 0 5 10
–4 –4
Pearson = 0.395, P = 2.44 × 10 Pearson = 0.392, P = 2.67 × 10
5 3
0 0
4 2

MYO6
STAM
–0.5 0 0.5 1.0 –1 0 1 3
1
log2 (fold change) log2 (fold change) 2
1 0
4 AZI2 STAM 0 5 10 0 5 10
Neurology REEP4 EIF4G3
4 Oncology TRIM25 CENPJ Pearson = –0.338, P = 1.92 × 10
–3
Pearson = –0.328, P = 2.6 × 10
–3
SPRR3 TAX1BP1
BLOC1S3 YARS1 CDC42BPB 1.5 2
DLG4
CAPS DNMBP RAB2B
1.0
MZT1 1
3 MICALL2
ENPP5

TBCB PDXDC1 0.5

OMP
MAP3K5
3
MYOC
ARHGEF10 0 0
PRTG
WASF3 –0.5
–log10 (P)

–log10 (P)

STX3 –1
2
PMVK
SLC27A4 CTSV –1.0
OMP IL18RAP 2 0 5 10 0 5 10
CNGB3
FH MORN4
SCN2A DCTN2 –3 –3
LBR HRAS Pearson = –0.317, P = 3.69 × 10 Pearson = –0.308, P = 4.81 × 10

1 1 3
1

MORN4
MYOC

2
0 1
0 0 0
–1

–0.5 0 0.5 1.0 –0.5 0 0.5 1.0 0 5 10 0 5 10

log2 (fold change) log2 (fold change) Breathlessness score

Fig. 5 | Dysregulation of the plasma proteome associated with breathlessness expressed plasma proteins from each panel versus the highest and lowest
in healthy convalescent individuals and individuals with long COVID breathlessness score tiers, irrespective of clinical assignation. Significance was
recruited from the United Kingdom. a, PCA of plasma protein concentrations evaluated using a two-tailed Mann–Whitney U-test with (red) or without (gray)
colored by donor group for healthy convalescent individuals (n = 51) and Benjamini–Hochberg correction. The dashed line indicates P = 0.05.
individuals with long COVID (n = 51). b, Bar plots showing the corresponding f, Correlation dot plots showing the highest (n = 6) and lowest ranked plasma
numbers of differentially upregulated plasma proteins from each panel. proteins (n = 4) in terms of normalized expression versus breathlessness scores
Significance was evaluated using a two-tailed Mann–Whitney U-test with (red) or for healthy convalescent individuals (n = 51) and individuals with long COVID
without (gray) Benjamini–Hochberg correction. c, Stacked histogram showing (n = 51), irrespective of clinical assignation; IDI2, isopentenyl-diphosphate
the distribution of breathlessness scores for healthy convalescent individuals δ-isomerase-2; SPRR3, small proline-rich protein 3; ENPP5, ectonucleotide
(n = 34) and individuals with long COVID (n = 48). d, PCA of plasma protein pyrophosphatase/phosphodiesterase family member 5; OMP, olfactory marker
concentrations colored by breathlessness score tiers for healthy convalescent protein. Significance was evaluated using the two-tailed Pearson coefficient.
individuals (n = 51) and individuals with long COVID (n = 51), irrespective of Shading indicates the 95% confidence interval for each regression line.
clinical assignation. e, Volcano plots showing the corresponding differentially

To identify specific proteins associated with breathlessness, we phosphodiesterase family member 5 and olfactory marker protein
performed a correlation analysis without prior stratification based (Fig. 5f). We then used the list of proteins ranked according to cor-
on symptom severity. The concentrations of almost all plasma pro- relation with breathlessness to perform a GSEA, which showed that
teins were skewed toward a positive correlation with breathlessness dysregulation of the plasma proteome was associated with related
score (Extended Data Fig. 6a and Supplementary Table 4). The most phenotypes, such as atelectasis (lung collapse) and tachypnea (rapid
positively correlated proteins included isopentenyl-diphosphate breathing), and further revealed an enrichment for pathways linked to
δ-isomerase 2 and small proline-rich protein 3, and the most nega- cell cycle progression (for example, RhoA), inflammation (for example,
tively correlated proteins included ectonucleotide pyrophosphatase/ TNF) and platelet activation (for example, PDGFRB and thromboxane

Nature Immunology | Volume 26 | May 2025 | 692–705 700

Article https://doi.org/10.1038/s41590-025-02135-5

A2 (TXA2); Extended Data Fig. 6b). In a further step designed to identify PD-1 and TIM-3, were relatively overexpressed among SARS-CoV-2
proteins with outsized roles in the breathlessness signatures associated nonspike-specific CD8+ T cells in individuals with long COVID. Our data
with inflammation, we performed network analyses using Cytoscape. also revealed an informative plasma biomarker signature linking per-
The output highlighted a complex protein network centered around sistent respiratory symptoms with apoptotic inflammatory networks
CD40 in a module that also included CCL3, CCL4, IKBKG and IL-18 and pathway dysregulation indicative of cell cycle progression, lung
(Extended Data Fig. 6c). injury and platelet activation in individuals with long COVID.
To validate these findings, we performed similar analyses using Donor groups in our primary cohort were carefully matched
plasma samples from donors in the secondary cohort, focusing on for age, BMI, race, sex, time since infection, and vaccination against
individuals with long COVID classified as low (0–2; n = 60) or moderate SARS-CoV-2, thereby minimizing the impact of confounding factors
(3–7; n = 35) according to the Borg CR10 scale, which measures that could potentially bias comparative analyses of healthy convales-
perceived exertion during physical activity (Fig. 6a)29. Differential cent individuals and individuals with long COVID. Women were over-
expression analysis revealed upregulated protein expression among represented as a consequence30–32. The predominant symptoms were
individuals with a moderate score, although no markers achieved breathlessness, fatigue, pain, mobility issues, anxiety and depression,
significance after multiple-hypothesis correction (Fig. 6b and Sup- which align with the known clinical spectrum of long COVID3. Pain was
plementary Table 5). GSEA of the ranked list of proteins nonetheless localized primarily to the chest, joints and muscles, again consistent
identified enrichment of signaling pathways observed in the primary with distributions reported in other individuals with long COVID33,34.
cohort (Extended Data Fig. 5e), including MET and PI3K/AKT/MTOR In contrast to influenza virus and other acute respiratory pathogens,
(Fig. 6c). Unbiased analyses further revealed that high Borg CR10 which predominantly exacerbate localized symptoms during and after
scores were correlated with pathways enriched among individuals in infection, these diverse postacute sequelae likely reflect an underlying
the primary cohort with severe breathlessness, including those associ- etiological complexity, which mandates a systematic approach to the
ated with ceramide, syndecan-4 and TXA2 (Extended Data Fig. 7a,b and diagnosis and management of individuals with long COVID35.
Supplementary Table 6). In contrast to a recent study16, we found that healthy convalescent
To unify these data, we correlated protein expression across the individuals were better able to neutralize SARS-CoV-2 than individuals
primary and secondary cohorts as a function of symptom severity. A with long COVID. This observation suggests a qualitative difference in
total of 275 proteins were differentially expressed among individuals antibody induction, potentially reflecting the fact that healthy conva-
with severe breathlessness and individuals with a moderate Borg CR10 lescent individuals were vaccinated more frequently before infection
score (Fig. 6d and Supplementary Table 7). Of these, all but three were than individuals with long COVID, which could help mitigate the risk of
consistently upregulated in both cohorts among donors with greater persistent disease36. No such differences were detected with respect to
symptom severity, suggesting a shared signature of plasma proteome overall SARS-CoV-2 spike-specific IgG titers or ADNKA. This latter find-
dysregulation in individuals with long COVID. Network analysis of dif- ing could be explained by the functional equivalence of antibodies tar-
ferentially expressed inflammatory markers in the secondary cohort geting the spike protein37 and/or by the availability of nonspike targets
further identified a major hub centered around CD40 (Extended Data expressed on the cell surface after infection with SARS-CoV-2 (ref. 38).
Fig. 7c), reminiscent of the primary cohort pattern (Extended Data Systemic immune perturbations are thought to play a role in the
Fig. 6c). A similar analysis of significantly upregulated proteins from pathogenesis of long COVID3,21. For example, innate immune cell acti-
the inflammation panel (n = 82) spanning both cohorts revealed that vation and a paucity of naive B and T cells have been described in one
the most confident network connections were centered around CCL3, cohort of individuals with long COVID8, whereas a relative abundance of
CD40, IKBKG, IL-18 and IRAK1 (Fig. 6e). Many of these proteins are asso- highly cytotoxic CD8+ T cells and NK cells has been described in another
ciated with the NF-κB pathway. In addition, overrepresentation analy- cohort of individuals with long COVID39. We found only nuanced differ-
sis of upregulated proteins spanning all panels across both cohorts ences between the immune cell lineage profiles of healthy convalescent
identified apoptosis as the most significant hit among all gene sets in individuals and individuals with long COVID. In the primary cohort,
the Hallmark Collection and ceramide and FAS as highly significant these differences were limited to nonclassical monocytes, which were
hits in the Pathway Interaction Database (Fig. 6f). Of note, proteins relatively overrepresented in healthy convalescent individuals, and
associated with these pathways, including apoptosis-inducing factor basophils and pDCs, which were relatively overrepresented in indi-
mitochondria-associated 1, caspase-3, caspase-7 and IL-18, were specifi- viduals with long COVID, whereas in the secondary cohort, these dif-
cally upregulated among donors with severe breathlessness recruited ferences were limited to classical monocytes, which were relatively
from the United Kingdom (Fig. 6g). overrepresented in healthy convalescent individuals, and intermediate
Collectively, these results identify dysregulated plasma proteins monocytes, which were relatively overrepresented in individuals with
that could serve as biomarkers of persistent breathlessness after infec- long COVID. Such inconsistencies likely reflect a number of factors,
tion with SARS-CoV-2, potentially facilitating the diagnosis and treat- including comorbidities and disease heterogeneity, and underscore
ment of long COVID. the importance of cross-validation in studies of long COVID3,10,40,41.
SARS-CoV-2 proteins can be detected in many tissues long after
Discussion the acute infectious event and could potentially engender a state
Long COVID continues to pose medical challenges with unmet diagnos- of chronic immune activation linked with the development of long
tic and therapeutic needs that reflect the elusive mechanistic nature COVID6,42,43. It is also known that CD4+ and CD8+ memory T cells pro-
of a symptomatically heterogeneous disease. In this study, we used vide durable immunity against SARS-CoV-2 (refs. 22,23,25,44). These
high-dimensional flow cytometry and plasma proteomics to seek bio- cells are exquisitely poised to mount anamnestic responses and
markers that could inform the pathogenesis of long COVID. Quantita- would likely proliferate and shift to an activated and/or exhausted
tive differences in immune cell lineage composition and virus-specific phenotype under conditions of recurrent antigen stimulation asso-
CD4+ and CD8+ T cell immunity were minimal and nonreproducible ciated with a failure to clear residual viral products and/or ongoing
across two geographically distinct cohorts in direct comparisons of viral replication, thereby feasibly becoming immunopathogenic
healthy convalescent individuals and individuals with long COVID. rather than protective in the context of long COVID10,45. In line with
Antibody neutralization activity was nonetheless significantly higher in this notion, one study reported sustained SARS-CoV-2-specific CD4+
healthy convalescent individuals than in individuals with long COVID, T cell responses during late recovery in individuals with long COVID46,
despite comparable SARS-CoV-2 spike-specific IgG titers and equiva- and another study reported enhanced expression of the exhaustion
lent levels of ADNKA, and some co-inhibitory receptors, especially markers CTLA-4 and PD-1 among SARS-CoV-2-specific CD8+ T cells

Nature Immunology | Volume 26 | May 2025 | 692–705 701

Article https://doi.org/10.1038/s41590-025-02135-5

a Group b Borg CR10 (moderate)

25 LC versus
Borg CR10 Borg CR10 (low)
20
Frequency Low (0–2)
Cardiometabolic QDPR CCL18 Inflammation CFP
15 Moderate (3–7) AFM SHMT1
3 TCTN3 PLPBP NIT2 3 TF ACYP1
10 LCAT TBCA
PDAP1
CYP24A1 MXRA8 SNCA
5
2

–log10(P)

–log10(P)
2 HLA–DRA
APLP1 MYH4 MMP15
0
SAT1 ITGA11 TLR1
0 1 2 3 4 5 6 7 8 9 10 FGF19
Borg CR10 1 1

c Enriched in moderate versus low Borg CR10

0 0
MET pathway (PID) P
0.015 –0.4 0 0.4 –0.5 0 0.5
TCPTP pathway (PID) log2 (fold change) log2 (fold change)
0.010
RET pathway (PID)
0.005 Neurology BRME1 Oncology KDR
PI3KCI pathway (PID) DBI
3
CYB5R2 SCLY DYNLT3
ERBB1 receptor IGF2R NUDT5 3
proximal pathway (PID) MITD1 GRPEL1
BHLHE40 CIAPIN1
BST1 SRC
Oxidative phosphorylation (H) ADCYAP1R1
GABRA4

–log10(P)

–log10(P)
2 ARHGEF5 GSTT2B IL9 HAO1
2
PI3K AKT MTOR signaling (H) MDGA1 MRPL28 TMED1
ATXN2 DCC FUT1 CA11 YARS1
Mitotic spindle (H)
1 1
Protein secretion (H)

MTORC1 signaling (H)

0 0

0 0.4 0.8 1.2 1.6 2.0 –0.4 0 0.4 0.8 –0.5 0 0.5
NES log2 (fold change) log2 (fold change)

d e PDE5A
–16
DBNL
Spearman = 0.44, P < 2.2 × 10
CRKL
ANGPT1 VASP
P < 0.05 PRKG1
(UK breathlessness) NEXN
PDGFB PDLIM7
(moderate versus low Borg CR10)

P < 0.05 F2R

EGF
0.5 (SE Borg CR10) TRIM5
CD40 IKBKG PPP1R9B
P < 0.05
log2 (fold change)

(both cohorts) STX7

VAMP8 MAP2K6
P > 0.05 CCL3
IRAK1
ARHGEF12

0 JAM3

VTI1A IL18 XIAP AMOT

TTR
PIKFYVE STAT2 HEXIM1
PLA2G4A
SNCA SERPINF1
SAMD9L
–0.5
PTGES2 MPIG6B

NRGN EIF4E
BNIP3L
DNAJB2 TREML1
–1.0 –0.5 0 0.5 1.0 1.5 FIS1
EIF4G1
log2 (fold change) MARS1 SHMT1
(severe versus no breathlessness)
DNAJA2 BAG4
GPI

f Hallmark gene sets PID pathways

g CASP3
0.012
CASP7
0.014
AIFM1
0.013
IL-18
0.0073
0.012 0.013 0.011 0.0085
5
8 0.63
7 0.54 0.36 0.98
UV response (up) p38 α/β 4
6
Oxidative 6 3
FAS 5
phosphorylation 4
NPX

NPX

NPX

NPX

4 2
mTORC1 signaling Ceramide 2
3 1
Reactive oxygen 2
α-Synuclein 0
species pathway 0
Apoptosis p75NTR 0 1 –1
–2

0 1 2 3 0 1 2
e

M L re

M L e

M L re

M L re

e
SeLow

er w

SeLow

er w

SeLow

er w

SeLow

er w
on

at

on

at

on

at

on

at
ve

ve
od o

ve

ve
od o

od o

od o

–log10(P) –log10(P)
N

Breathless- Breathless- Breathless- Breathless-

ness Borg CR10 ness Borg CR10 ness Borg CR10 ness Borg CR10

Fig. 6 | Shared signatures of plasma proteome dysregulation associated with tiers (primary cohort, x axis) and Borg CR10 score tiers (secondary cohort, y
respiratory symptoms in healthy convalescent individuals and individuals axis). Significance was evaluated using a two-tailed Spearman rank test. Proteins
with long COVID. a, Bar plot showing the distribution of Borg CR10 scores for are colored according to significance without Benjamini–Hochberg correction.
individuals in the secondary cohort with long COVID (n = 95). b, Volcano plots e, Network analysis showing differentially expressed plasma proteins from the
showing differentially expressed plasma proteins from each panel versus Borg inflammation panel across both cohorts depicted using Cytoscape. Nodes and
CR10 score tiers for individuals in the secondary cohort with long COVID (n = 95). edges represent proteins and functional relevance, respectively. Edge thickness
Significance was evaluated using a two-tailed Mann–Whitney U-test. The dashed represents the level of confidence. f, Overrepresentation analysis of significantly
line indicates P = 0.05. No proteins achieved adjusted P < 0.05 after Benjamini– upregulated plasma proteins across both cohorts showing the five top terms
Hochberg correction. c, GSEA showing differentially expressed plasma proteins from the Hallmark Collection and the Pathway Interaction Database. Significance
by rank versus the Borg CR10 score tiers for individuals with long COVID (n = 95), was evaluated using a hypergeometric test. g, Comparison of plasma protein
irrespective of clinical assignation. The top five terms from the Hallmark (H) and concentrations between cohorts split by symptom severity. Horizontal bars
Pathway Interaction Database (PID) gene sets (Molecular Signatures Database represent median values. Significance was evaluated using a two-tailed
(MSigDB) Collections) are shown. Significance was evaluated using the GSEA Mann–Whitney U-test; NPX, normalized protein expression; AIFM1, apoptosis-
method without correction; NES, normalized enrichment score. d, Scatter dot inducing factor mitochondria-associated 1; CASP, caspase.
plot showing individual plasma protein fold change across breathlessness score

Nature Immunology | Volume 26 | May 2025 | 692–705 702

Article https://doi.org/10.1038/s41590-025-02135-5

in individuals with long COVID14. The converse scenario in terms of primarily to hypercoagulability and thromboinflammation18, poten-
response magnitude has also been described for IFNγ-producing CD8+ tially accompanied by amyloid fibrin microclot deposition63, endothe-
T cells targeting the nucleocapsid protein of SARS-CoV-2 (ref. 47). In our lial dysfunction64 and vasculoproliferation65, which collectively impair
primary cohort, SARS-CoV-2-specific CD4+ and CD8+ T cell responses oxygen exchange and lead to the sensation of breathlessness in indi-
were comparable in magnitude across the entire viral proteome in viduals with long COVID.
healthy convalescent individuals and individuals with long COVID, One limitation of our study was that the matching process for the
whereas in our secondary cohort, relatively elevated frequencies primary cohort was not entirely accurate, such that healthy convalescent
of SARS-CoV-2 nucleocapsid-specific CD4+ T cells and SARS-CoV-2 individuals were sampled earlier after infection (median = 268 days)
spike-specific CD8+ T cells were observed in individuals with long than individuals with long COVID (median = 416 days). This discrepancy
COVID. However, more refined analyses of the primary cohort revealed combined with a preferential loss of functionally optimal antibodies
altered memory profiles and enhanced co-inhibitory scores among could partially explain the relative paucity of neutralization activity
SARS-CoV-2 nonspike-specific CD8+ T cells in individuals with long in individuals with long COVID. A minority of healthy convalescent
COVID, indicating a relatively greater cumulative history of exposure individuals also reported breathlessness as a symptom, likely attribut-
to antigens derived from SARS-CoV-2. An alternative possibility is that able to other pathologies affecting the respiratory system, which were
immune exhaustion facilitates viral persistence, but further studies not assessed clinically. Moreover, our approach was limited to sam-
are required to determine the protective versus reactive properties ples acquired from the vascular circulation, which emerging evidence
of SARS-CoV-2-specific CD4+ and CD8+ T cells in relation to the patho- suggests is a highly specialized immunological niche66. In addition, the
genesis of long COVID. origins and roles of proteins detected in plasma samples are open to
Many factors can affect immune responses against SARS-CoV-2, interpretation, providing only indirect evidence for any given underly-
including genetic background, infection history and vaccination ing pathology. Comparative analyses of disease-relevant tissue samples
status7,48–50, and many factors beyond immune responses against will therefore be required to validate the localized pathology associ-
SARS-CoV-2 have been linked with the pathogenesis of long COVID3,51, ated with our reported systemic cellular and molecular signatures of
including reactivation of the herpesviruses CMV and/or EBV3,10,11,52,53. long COVID.
Most of these latter associations have been defined serologically16,20,54. In summary, our findings suggest that lung damage associated
We addressed the same issue by interrogating CD4+ and CD8+ T cells tar- with the canonical symptom of breathlessness can be identified via
geting immunodominant regions of CMV or EBV. In the primary cohort, the systemic upregulation of multiple apoptotic, cardiovascular and
no intergroup differences in response magnitude were detected for any inflammatory biomarkers in the presence of a largely unperturbed
specificity, whereas in the secondary cohort, the frequencies of CD4+ cellular immune system, indicative of localized tissue pathology and
T cells targeting EBV latent proteins and the frequencies of CD8+ T cells ongoing but minimal exposure to viral antigens potentially facilitated
targeting CMV proteins or EBV lytic proteins were relatively elevated by suboptimal humoral immunity in individuals with long COVID.
in individuals with long COVID. Phenotypic analyses focused on the
primary cohort further revealed high co-inhibitory scores among Online content
CD8+ T cells targeting epitopes from CMV, which overexpressed PD-1, Any methods, additional references, Nature Portfolio reporting sum-
and a terminally differentiated profile among CD8+ T cells targeting maries, source data, extended data, supplementary information,
lytic epitopes from EBV in individuals with long COVID. In line with acknowledgements, peer review information, details of author contri-
these observations, we found that SARS-CoV-2 spike-specific CD8+ butions and competing interests, and statements of data and code avail-
T cells overexpressed various activation markers, including CD38 ability are available at https://doi.org/10.1038/s41590-025-02135-5.
and HLA-DR, and various co-inhibitory receptors spanning PD-1, TIM-
3, LAG-3 and TIGIT, likely reflecting recurrent antigen exposure as References
a consequence of repeated subunit vaccination55. Accordingly, our 1. Nalbandian, A. et al. Post-acute COVID-19 syndrome. Nat. Med.
data align with the notion that bystander viral reactivation frequently 27, 601–615 (2021).
accompanies the development of persistent disease after infection 2. Subramanian, A. et al. Symptoms and risk factors for long
with SARS-CoV-2 (refs. 10,54,56). COVID in non-hospitalized adults. Nat. Med. 28, 1706–1714
Plasma proteomics has emerged as a useful strategy to help (2022).
decipher the molecular basis of various diseases via the identi- 3. Davis, H. E., McCorkell, L., Vogel, J. M. & Topol, E. J. Long COVID:
fication of systemic biomarkers indicative of tissue-localized major findings, mechanisms and recommendations. Nat. Rev.
pathology12,57–59. Using a high-throughput platform in conjunction with Microbiol. 21, 133–146 (2023).
a symptom-targeted approach, we found that severe breathlessness 4. Chen, B., Julg, B., Mohandas, S. & Bradfute, S. B. RECOVER
was associated with extensive dysregulation of the plasma proteome. Mechanistic Pathways Task Force. Viral persistence, reactivation,
It should be noted that our approach was focused on a curated panel and mechanisms of long COVID. eLife 12, e86015 (2023).
of proteins spanning a targeted fraction of the entire proteome, such 5. Zadeh, F. H., Wilson, D. R. & Agrawal, D. K. Long COVID:
that we potentially failed to identify some biomarkers and pathways complications, underlying mechanisms, and treatment strategies.
characteristic of long COVID. Our findings nonetheless align broadly Arch. Microbiol. Immunol. 7, 36–61 (2023).
with other strands of evidence indicating that chronic inflammation 6. Castanares-Zapatero, D. et al. Pathophysiology and mechanism
is a cornerstone of long COVID8,12,14,15,60. Moreover, network analyses of long COVID: a comprehensive review. Ann. Med. 54, 1473–1487
identified connections centered around CD40, incorporating various (2022).
caspases (CASP2 and CASP7) and kinases (IKBKG and MAP2K6), col- 7. Stein, S. R. et al. SARS-CoV-2 infection and persistence in the
lectively linking breathlessness with inflammatory apoptosis and/or human body and brain at autopsy. Nature 612, 758–763 (2022).
cell death, which could feasibly reflect ongoing exposure to antigens 8. Phetsouphanh, C. et al. Immunological dysfunction persists for 8
derived from SARS-CoV-2 (refs. 61,62). Similar inflammatory profiles months following initial mild-to-moderate SARS-CoV-2 infection.
have been identified previously in individuals with long COVID12,13. Nat. Immunol. 23, 210–216 (2022).
Pathway analyses further identified dysregulated proteins associated 9. Proal, A. D. et al. SARS-CoV-2 reservoir in post-acute sequelae of
with cell cycle progression (for example, RhoA) and platelet activa- COVID-19 (PASC). Nat. Immunol. 24, 1616–1627 (2023).
tion (for example, PDGFRB and TXA2). In sum, these observations fit 10. Su, Y. et al. Multiple early factors anticipate post-acute COVID-19
with a dynamic process of lung damage and remodeling attributable sequelae. Cell 185, 881–895 (2022).

Nature Immunology | Volume 26 | May 2025 | 692–705 703

Article https://doi.org/10.1038/s41590-025-02135-5

11. Sherif, Z. A. et al. Pathogenic mechanisms of post-acute sequelae 34. Al-Aly, Z., Bowe, B. & Xie, Y. Long COVID after breakthrough
of SARS-CoV-2 infection (PASC). eLife 12, e86002 (2023). SARS-CoV-2 infection. Nat. Med. 28, 1461–1467 (2022).
12. Talla, A. et al. Persistent serum protein signatures define an 35. Xie, Y., Choi, T. & Al-Aly, Z. Long-term outcomes following hospital
inflammatory subcategory of long COVID. Nat. Commun. 14, admission for COVID-19 versus seasonal influenza: a cohort study.
3417 (2023). Lancet Infect. Dis. 24, 239–255 (2024).
13. Suhre, K. et al. Identification of robust protein associations with 36. Catala, M. et al. The effectiveness of COVID-19 vaccines to
COVID-19 disease based on five clinical studies. Front. Immunol. prevent long COVID symptoms: staggered cohort study of data
12, 781100 (2021). from the UK, Spain, and Estonia. Lancet Respir. Med. 12, 225–236
14. Yin, K. et al. Long COVID manifests with T cell dysregulation, (2024).
inflammation and an uncoordinated adaptive immune response 37. Grant, M. D. et al. Combined anti-S1 and anti-S2 antibodies
to SARS-CoV-2. Nat. Immunol. 25, 218–225 (2024). from hybrid immunity elicit potent cross-variant ADCC against
15. Schultheiss, C. et al. The IL-1β, IL-6, and TNF cytokine triad is SARS-CoV-2. JCI Insight 8, e170681 (2023).
associated with post-acute sequelae of COVID-19. Cell Rep. Med. 38. Fielding, C. A. et al. SARS-CoV-2 host-shutoff impacts innate NK
3, 100663 (2022). cell functions, but antibody-dependent NK activity is strongly
16. Klein, J. et al. Distinguishing features of long COVID identified activated through non-spike antibodies. eLife 11, e74489 (2022).
through immune profiling. Nature 623, 139–148 (2023). 39. Galan, M. et al. Persistent overactive cytotoxic immune response
17. Baillie, K. et al. Complement dysregulation is a prevalent in a Spanish cohort of individuals with long-COVID: identification
and therapeutically amenable feature of long COVID. Med. 5, of diagnostic biomarkers. Front. Immunol. 13, 848886 (2022).
239–253 (2024). 40. Taquet, M. et al. Incidence, co-occurrence, and evolution of
18. Cervia-Hasler, C. et al. Persistent complement dysregulation with long-COVID features: a 6-month retrospective cohort study of
signs of thromboinflammation in active long COVID. Science 383, 273,618 survivors of COVID-19. PLoS Med. 18, e1003773 (2021).
eadg7942 (2024). 41. Davis, H. E. et al. Characterizing long COVID in an
19. Chattopadhyay, P. K. et al. The cytolytic enzymes granyzme A, international cohort: 7 months of symptoms and their impact.
granzyme B, and perforin: expression patterns, cell distribution, EClinicalMedicine 38, 101019 (2021).
and their relationship to cell maturity and bright CD57 expression. 42. Raveendran, A. V., Jayadevan, R. & Sashidharan, S. Long COVID:
J. Leukoc. Biol. 85, 88–97 (2009). an overview. Diabetes Metab. Syndr. 15, 869–875 (2021).
20. Peluso, M. J. et al. Chronic viral coinfections differentially affect 43. Crook, H., Raza, S., Nowell, J., Young, M. & Edison, P. Long
the likelihood of developing long COVID. J. Clin. Invest. 133, COVID—mechanisms, risk factors, and management. BMJ 374,
e163669 (2023). n1648 (2021).
21. Altmann, D. M., Whettlock, E. M., Liu, S., Arachchillage, D. J. & 44. Tarke, A. et al. SARS-CoV-2 vaccination induces immunological
Boyton, R. J. The immunology of long COVID. Nat. Rev. Immunol. T cell memory able to cross-recognize variants from Alpha to
23, 618–634 (2023). Omicron. Cell 185, 847–859 (2022).
22. Gao, Y. et al. Immunodeficiency syndromes differentially impact 45. Wong, L. R. & Perlman, S. Immune dysregulation and immuno­
the functional profile of SARS-CoV-2-specific T cells elicited by pathology induced by SARS-CoV-2 and related coronaviruses—are
mRNA vaccination. Immunity 55, 1732–1746 (2022). we our own worst enemy? Nat. Rev. Immunol. 22, 47–56 (2022).
23. Gao, Y. et al. Ancestral SARS-CoV-2-specific T cells 46. Files, J. K. et al. Duration of post-COVID-19 symptoms is
cross-recognize the Omicron variant. Nat. Med. 28, 472–476 associated with sustained SARS-CoV-2-specific immune
(2022). responses. JCI Insight 6, 151544 (2021).
24. Altmann, D. M. et al. Persistent symptoms after COVID-19 are 47. Peluso, M. J. et al. Long-term SARS-CoV-2-specific immune and
not associated with differential SARS-CoV-2 antibody or T cell inflammatory responses in individuals recovering from COVID-19
immunity. Nat. Commun. 14, 5139 (2023). with and without post-acute symptoms. Cell Rep. 36, 109518
25. Sekine, T. et al. Robust T cell immunity in convalescent individuals (2021).
with asymptomatic or mild COVID-19. Cell 183, 158–168 (2020). 48. Silva Andrade, B. et al. Long-COVID and post-COVID health
26. Adamo, S. et al. Memory profiles distinguish cross-reactive and complications: an up-to-date review on clinical conditions and
virus-specific T cell immunity to mpox. Cell Host Microbe 31, their possible molecular mechanisms. Viruses 13, 700 (2021).
928–936 (2023). 49. Cheung, C. C. L. et al. Residual SARS-CoV-2 viral antigens
27. Buggert, M. et al. T-bet and Eomes are differentially linked to detected in GI and hepatic tissues from five recovered patients
the exhausted phenotype of CD8+ T cells in HIV infection. PLoS with COVID-19. Gut 71, 226–229 (2022).
Pathog. 10, e1004251 (2014). 50. Alwan, N. A. The road to addressing long COVID. Science 373,
28. Zaghlool, S. B. et al. Revealing the role of the human blood 491–493 (2021).
plasma proteome in obesity using genetic drivers. Nat. Commun. 51. Astin, R. et al. Long COVID: mechanisms, risk factors and
12, 1279 (2021). recovery. Exp. Physiol. 108, 12–27 (2023).
29. Williams, N. The Borg Rating of Perceived Exertion (RPE) scale. 52. Ahamed, J. & Laurence, J. Long COVID endotheliopathy:
Occup. Med. 67, 404–405 (2017). hypothesized mechanisms and potential therapeutic approaches.
30. Takahashi, T. et al. Sex differences in immune responses that J. Clin. Invest. 132, e161167 (2022).
underlie COVID-19 disease outcomes. Nature 588, 315–320 53. Vojdani, A., Vojdani, E., Saidara, E. & Maes, M. Persistent
(2020). SARS-CoV-2 infection, EBV, HHV-6 and other factors may
31. Ganesh, R. et al. The female-predominant persistent immune contribute to inflammation and autoimmunity in long COVID.
dysregulation of the post-COVID syndrome. Mayo Clin. Proc. 97, Viruses 15, 400 (2023).
454–464 (2022). 54. Gold, J. E., Okyay, R. A., Licht, W. E. & Hurley, D. J. Investigation of
32. Bai, F. et al. Female gender is associated with long COVID long COVID prevalence and its relationship to Epstein–Barr virus
syndrome: a prospective cohort study. Clin. Microbiol. Infect. 28, reactivation. Pathogens 10, 763 (2021).
611.e9–611.e16 (2022). 55. Cai, C. et al. SARS-CoV-2 vaccination enhances the effector
33. Choutka, J., Jansari, V., Hornig, M. & Iwasaki, A. Unexplained qualities of spike-specific T cells induced by COVID-19. Sci.
post-acute infection syndromes. Nat. Med. 28, 911–923 (2022). Immunol. 8, eadh0687 (2023).

Nature Immunology | Volume 26 | May 2025 | 692–705 704

Article https://doi.org/10.1038/s41590-025-02135-5

56. Peluso, M. J. & Deeks, S. G. Early clues regarding the pathogenesis 65. Iosef, C. et al. Plasma proteome of long-COVID patients indicates
of long-COVID. Trends Immunol. 43, 268–270 (2022). HIF-mediated vasculo-proliferative disease with impact on brain
57. Shu, T. et al. Plasma proteomics identify biomarkers and and heart function. J. Transl. Med. 21, 377 (2023).
pathogenesis of COVID-19. Immunity 53, 1108–1122 (2020). 66. Buggert, M., Price, D. A., Mackay, L. K. & Betts, M. R. Human
58. Al-Nesf, M. A. Y. et al. Prognostic tools and candidate drugs circulating and tissue-resident memory CD8+ T cells. Nat. Immunol.
based on plasma proteomics of patients with severe COVID-19 24, 1076–1086 (2023).
complications. Nat. Commun. 13, 946 (2022).
59. Gu, X. et al. Probing long COVID through a proteomic lens: Publisher’s note Springer Nature remains neutral with
a comprehensive two-year longitudinal cohort study of regard to jurisdictional claims in published maps and
hospitalised survivors. EBioMedicine 98, 104851 (2023). institutional affiliations.
60. Peluso, M. J. et al. Markers of immune activation and inflammation
in individuals with postacute sequelae of severe acute respiratory Open Access This article is licensed under a Creative Commons
syndrome coronavirus 2 infection. J. Infect. Dis. 224, 1839–1848 Attribution 4.0 International License, which permits use, sharing,
(2021). adaptation, distribution and reproduction in any medium or format,
61. Krishnamachary, B. et al. Extracellular vesicle-mediated as long as you give appropriate credit to the original author(s) and the
endothelial apoptosis and EV-associated proteins correlate with source, provide a link to the Creative Commons licence, and indicate
COVID-19 disease severity. J. Extracell. Vesicles 10, e12117 (2021). if changes were made. The images or other third party material in this
62. Andre, S. et al. T cell apoptosis characterizes severe COVID-19 article are included in the article’s Creative Commons licence, unless
disease. Cell Death Differ. 29, 1486–1499 (2022). indicated otherwise in a credit line to the material. If material is not
63. Kell, D. B., Laubscher, G. J. & Pretorius, E. A central role for amyloid included in the article’s Creative Commons licence and your intended
fibrin microclots in long COVID/PASC: origins and therapeutic use is not permitted by statutory regulation or exceeds the permitted
implications. Biochem. J. 479, 537–559 (2022). use, you will need to obtain permission directly from the copyright
64. Boccatonda, A., Campello, E., Simion, C. & Simioni, P. Long-term holder. To view a copy of this licence, visit http://creativecommons.
hypercoagulability, endotheliopathy and inflammation following org/licenses/by/4.0/.
acute SARS-CoV-2 infection. Expert Rev. Hematol. 16, 1035–1048
(2023). © The Author(s) 2025

1
Center for Infectious Medicine, Department of Medicine Huddinge, Karolinska Institutet, Stockholm, Sweden. 2Laboratory of Translational
Immuno-Oncology, Department of Biomedicine, University and University Hospital Basel, Basel, Switzerland. 3Division of Infectious Diseases and
Dermatology, Department of Medicine Huddinge, Karolinska Institutet, Stockholm, Sweden. 4Center for Clinical Research Sörmland, Uppsala University,
Uppsala, Sweden. 5Department of Infectious Diseases, Karolinska University Hospital, Stockholm, Sweden. 6Division of Infection and Immunity, Cardiff
University School of Medicine, University Hospital of Wales, Cardiff, UK. 7Department of Respiratory Medicine, University Hospital Llandough, Penarth,
UK. 8Post-COVID Policlinic, Karolinska University Hospital, Stockholm, Sweden. 9Department of Neurobiology, Care Sciences and Society, Karolinska
Institutet, Stockholm, Sweden. 10Division of Infectious Diseases, Department of Medicine, University of Alabama at Birmingham, Birmingham, AL, USA.
11
Department of Medicine, University of California, San Francisco, San Francisco, CA, USA. 12Systems Immunity Research Institute, Cardiff University
School of Medicine, University Hospital of Wales, Cardiff, UK. 13These authors contributed equally: Yu Gao, Curtis Cai, Sarah Adamo, Helen E. Davies,
Soo Aleman, Marcus Buggert, David A. Price. e-mail: [email protected]; [email protected]

Nature Immunology | Volume 26 | May 2025 | 692–705 705

Article https://doi.org/10.1038/s41590-025-02135-5

Methods Cells and viruses

Study design A549 and VeroE6 cells expressing human angiotensin-converting
The objective of this study was to characterize the immunological and enzyme 2 (ACE2) and transmembrane serine protease 2 (TMPRSS2)
proteomic features of long COVID. SARS-CoV-2 spike-specific antibody were used to support viral entry and propagation67. Antibody function-
titers were measured using an enzyme-linked immunosorbent assay. ality assays were performed using the England-2 strain of SARS-CoV-2
Antibody neutralization activity and ADNKA were quantified against (ref. 38).
the England-2 strain of SARS-CoV-2. Immune cell lineages were profiled
via multidimensional flow cytometry. Antigen-specific CD4+ and CD8+ Peptides
T cells were enumerated functionally using a flow cytometric AIM SARS-CoV-2 peptides were manufactured as 15-mers overlapping by
assay. Antigen-specific CD8+ T cells were further identified physically 11 amino acids spanning the spike protein (Peptides & Elephants) or
using peptide–HLA class I tetramers to enable detailed phenotypic as 20-mers overlapping by 10 amino acids spanning the nucleocapsid,
analyses via multidimensional flow cytometry. Plasma proteomes combined membrane and envelope, ORF1a, ORF1b and ORF3–ORF10
were analyzed using a targeted affinity platform. Clinical symptoms proteins (Sigma-Aldrich). EBV peptides were manufactured as 15-mers
were integrated with the frequencies and phenotypic attributes of overlapping by 11 amino acids spanning the BRLF1, BZLF1, BMLF1 and
immune cells to delineate plasma biomarkers and signaling pathways BARF1 proteins (lytic pool) and the EBNA1, EBNA2, EBNA3A, EBNA3B,
associated with long COVID. EBNA3C and LMP2 proteins (latent pool; JPT Peptide Technologies).
CMV peptides were manufactured as 15-mers overlapping by 11 amino
Donors acids spanning the combined IE-1, IE-2 and pp65 proteins ( JPT Peptide
The primary cohort included healthy convalescent individuals (con- Technologies). Lyophilized peptides were reconstituted at a stock
trols; n = 70) and individuals with long COVID (cases; n = 70) recruited concentration of 10 mg ml–1 in DMSO and further diluted to 100 μg ml–1
from University Hospital Llandough (Table 1 and Supplementary in phosphate-buffered saline (PBS).
Table 1). All participants had a clearly defined episode of symptomati-
cally mild acute COVID-19 confirmed via direct molecular evidence Tetramers
of infection with SARS-CoV-2. None required hospitalization. Cases Peptide–HLA class I complexes were generated and tetramerized
were diagnosed according to the National Institute for Health and with fluorescent tags as described previously68,69. The following
Care Excellence guideline NG188 (https://www.nice.org.uk/guidance/ specificities were used in this study: CMV pp65 HLA-A*02:01 NLVP-
ng188). Groups were matched as closely as possible for age, BMI, MVATV (BV421), CMV pp65 HLA-B*07:02 TPRVTGGGAM (PE), EBV
race, sex, time since infection, and vaccination against SARS-CoV-2 BMLF1 (lytic) HLA-A*02:01 GLCTLVAML (PE), EBV EBNA3A (latent)
(Fig. 1a,b and Table 1). Eligible individuals were men and nonpreg- HLA-B*07:02 RPPIFIRRL (BV421), HIV-1 p2p7p1p6 Gag HLA-A*02:01
nant women over the age of 18 years with no alternative explanatory FLGKIWPSHK (PE), HIV-1 p17 Gag HLA-A*02:01 SLYNTVATL (BV421),
disease and symptoms that persisted for at least 12 weeks after the HIV-1 Pol HLA-A*02:01 ILKEPVHGV (PE), HIV-1 p17 Gag HLA-A*24:02
initial diagnosis of acute COVID-19. One persistent symptom was KYKLHIVW (BV421), HIV-1 Nef HLA-A*24:02 RYPLTFGW (PE), HIV-1 p24
sufficient for the diagnosis of long COVID. All individuals underwent Gag HLA-B*07:02 GPGHKARVL (BV421), HIV-1 p24 Gag HLA-B*08:01
a comprehensive medical evaluation, including chest radiography, EIYKRWII (PE), HIV-1 p24 Gag HLA-B*57:01 KAFSPEVIPMF (PE), HIV-1
electrocardiography, lung function tests (spirometry with gas transfer p24 Gag HLA-B*57:01 QASQEVKNW (BV421), IAV matrix protein M1
as indicated and measurement of exhaled nitric oxide), and standard HLA-A*02:01 GILGFVFTL (BV421), IAV nucleoprotein HLA-B*07:02 LPFD-
blood tests (autoantibody screens; bone, liver and kidney function; KTTVM (BV421), SARS-CoV-2 spike HLA-A*02:01 YLQPRTFLL (BV421),
coagulation screens; full blood count; markers of nutrition). Symp- SARS-CoV-2 nucleocapsid HLA-A*02:01 LLLDRLNQL (PE), SARS-CoV-2
toms were scored individually using a numeric self-rating scale from 0 ORF3 HLA-A*02:01 ALSKGVHFV (PE), SARS-CoV-2 ORF3 HLA-A*02:01
(no symptom) to 10 (worst possible symptom). Overall general health LLYDANYFL (PE) and SARS-CoV-2 nucleocapsid HLA-B*07:02 SPRW-
was scored similarly on an inverse scale from 0 (worst possible) to 10 YFYYL (PE).
(best possible). The secondary cohort included healthy convalescent
individuals (controls; n = 30) and individuals with long COVID (cases; Antibody quantification
n = 95) recruited from the Karolinska University Hospital (Table 2). SARS-CoV-2 spike-specific antibody titers were measured using a
All participants in the primary cohort were recruited between March SARS-CoV-2 Spike (Trimer) Ig Total ELISA Kit (Thermo Fisher Scientific).
and August 2022, and all participants in the secondary cohort were Samples were assayed in duplicate and calibrated against a standard
recruited between June and October 2022. PBMCs from donors with curve. Data were analyzed using Prism version 9.5.0 (GraphPad).
untreated chronic HIV-1 infection (n = 14) were obtained from the
University of Alabama at Birmingham or the University of California, Neutralization assay
San Francisco. Antibody neutralization activity was quantified as described previ-
ously38. Briefly, serial dilutions of plasma were mixed in duplicate with
Samples 600 plaque-forming units of England-2, incubated for 1 h at 37 °C, and
PBMCs were isolated via standard density gradient centrifugation and added to VeroE6 cells expressing ACE2 and TMPRSS2. After 48 h, cell
cryopreserved in fetal bovine serum (Thermo Fisher Scientific) con- monolayers were fixed in 4% paraformaldehyde (Thermo Fisher Scien-
taining 10% dimethyl sulfoxide (DMSO; Sigma-Aldrich). EDTA plasma tific), permeabilized with 0.5% NP-40 (Merck), and blocked with PBS
samples were stored at −80 °C. containing 0.1% Tween-20 (PBST) and 3% nonfat milk for 1 h at room
temperature (RT). The primary antibody (anti-SARS-CoV-2 nucleocap-
Ethics sid protein, clone 1C7, Stratech Scientific) was diluted 1:500 in PBST
All participants provided written informed consent in accordance containing 1% nonfat milk and added to the cell monolayers for 1 h at RT.
with the principles of the Declaration of Helsinki (2013). The primary Cells were then washed with PBST. The secondary antibody (anti-mouse
study was approved by the Cardiff University School of Medicine IgG-HRP, polyclonal, Jackson ImmunoResearch) was diluted 1:3,000
Research Ethics Committee (21/55) and the Health Research Authority in PBST containing 1% nonfat milk and added to the cell monolayers
and Health and Care Research Wales (20/NW/0240), and the second- for 1 h at RT. Cells were then washed again with PBST. Assays were
ary study was approved by the Swedish Ethical Review Authority developed using SIGMAFAST OPD (Sigma-Aldrich) and analyzed at an
(2022-00100-01). optical density of 450 nm using a CLARIOstar Plus Microplate Reader

Nature Immunology
Article https://doi.org/10.1038/s41590-025-02135-5

(BMG Labtech). Control wells contained no sample, a standardized sam- incubated for 12 h at 37 °C. Negative-control wells contained equiva-
ple with moderate neutralization activity, or no SARS-CoV-2. The neu- lent DMSO. After incubation, cells were washed with PBS, labeled with
tralization titer for each sample was calculated as the highest plasma LIVE/DEAD Fixable Aqua (Thermo Fisher Scientific) for 10 min at RT,
dilution that achieved a 50% reduction in plaque formation (NT50). washed with FACS buffer, and stained with anti-CCR4–BB700 (clone
1G1, BD Biosciences), anti-CCR6–BUV737 (clone 11A9, BD Biosciences),
ADNKA anti-CCR7–APC-Cy7 (clone G043H7, BioLegend), anti-CX3CR1–PE
ADNKA was quantified as described previously38,70. Briefly, target A549 (clone 2A9-1, BioLegend) and anti-CXCR3–AF647 (clone G025H7, BioLe-
cells expressing ACE2 and TMPRSS2 were infected overnight with gend) for 10 min at 37 °C. Cells were then stained further with anti-CD3–
England-2 (multiplicity of infection = 5), collected using TrypLE Express BUV805 (clone UCHT1, BD Biosciences), anti-CD4–BUV496 (clone SK3,
Enzyme (Thermo Fisher Scientific), mixed with healthy donor PBMCs BD Biosciences), anti-CD8–BUV395 (clone RPA-T8, BD Biosciences),
at a ratio of 1:10, and incubated with serial dilutions of plasma in the anti-CD14–BV510 (clone M5E2, BioLegend), anti-CD19–BV510 (clone
presence of anti-CD107a–FITC (clone H4A3, BioLegend) and GolgiStop HIB19, BioLegend), anti-CD28–BUV563 (clone CD28.2, BD Biosciences),
(0.7 μl ml–1; BD Biosciences) for 5 h at 37 °C. Cells were then washed with anti-CD38–APC-R700 (clone HIT2, BD Biosciences), anti-CD39–BV711
cold PBS, stained with anti-CD3–PE-Cy7 (clone UCHT1, BioLegend), (clone A1, BioLegend), anti-CD45RA–BV570 (clone HI100, BioLegend),
anti-CD56–BV605 (clone 5.1H11, BioLegend), anti-CD57–APC (clone anti-CD69–BV650 (clone FN50, BioLegend), anti-CD71–BUV661 (clone
HNK-1, BioLegend) and LIVE/DEAD Fixable Aqua (Thermo Fisher Sci- M-A712, BD Biosciences), anti-CD95–PE-Dazzle594 (clone DX2, BioLe-
entific) for 30 min at 4 °C, washed again with cold PBS, and fixed in 4% gend), anti-CD127–PE-Cy5 (clone A019D5, BioLegend), anti-CD137–
paraformaldehyde (Thermo Fisher Scientific). Control wells contained PE-Cy7 (clone 4B4-1, BioLegend), anti-CD154–BV421 (clone 24-31,
a seronegative sample, uninfected target cells, or a standardized sam- BioLegend), anti-HLA-DR–BV605 (clone G46-6, BD Biosciences),
ple that elicited moderate ADNKA. Data were acquired using an Attune anti-PD-1–BUV615 (clone EH12.1, BD Biosciences) and anti-TIGIT–BV786
NxT Flow Cytometer (Thermo Fisher Scientific). Activation was quan- (clone 741182, BD Biosciences) for 30 min at RT in the presence of Bril-
tified as a function of degranulation (CD107a+) among viable NK cells liant Stain Buffer Plus (BD Biosciences; Supplementary Table 9). Stained
(Aqua−CD3−CD56+) with potent cytotoxic activity (CD57+) using FlowJo cells were washed twice with FACS buffer, fixed in Cytofix Fixation
version 10.9.0 (FlowJo) and normalized to the standardized sample Buffer (BD Biosciences), and acquired using a FACSymphony A5 (BD
via area under the curve analyses in Prism version 9.5.0 (GraphPad). Biosciences). Data were analyzed using FlowJo version 10.9.0 (FlowJo).

Immune cell lineage analysis Tetramer staining and phenotypic analysis

PBMCs were thawed quickly, resuspended in RPMI 1640 Complete PBMCs were thawed quickly, resuspended in RPMI 1640 Complete
Medium (Sigma-Aldrich) supplemented with DNase I (10 U ml –1; Medium (Sigma-Aldrich) supplemented with DNase I (10 U ml–1;
Sigma-Aldrich), and seeded at 1 × 106 cells per well in 96-well U-bottom Sigma-Aldrich), and seeded at 2 × 106 cells per well in 96-well U-bottom
plates (Corning). Cells were incubated first with Human TruStain plates (Corning). Cells were incubated first with dasatinib (50 µM;
FcX (BioLegend) for 10 min at RT and then with LIVE/DEAD Fixa- STEMCELL Technologies) for 10 min at RT and then with the relevant
ble Aqua (Thermo Fisher Scientific) for 10 min at RT. Anti-CCR7– peptide–HLA class I tetramers (each at 1 µg per stain) for 20 min at
APC-Cy7 (clone G043H7, BioLegend) and anti-CX3CR1–PE (clone RT (Supplementary Table 10). After incubation, cells were washed
2A9-1, BioLegend) were added for 15 min at 37 °C. Cells were then with PBS, labeled with LIVE/DEAD Fixable Aqua (Thermo Fisher Sci-
stained with anti-CD3–BV650 (clone OKT3, BioLegend), anti-CD4– entific) for 10 min at RT, washed with FACS buffer, and stained with
PE-Cy5.5 (clone S3.5, Thermo Fisher Scientific), anti-CD8–BUV396 anti-CCR7–APC-Cy7 (clone G043H7, BioLegend), anti-CX3CR1–BUV615
(clone RPA-T8, BD Biosciences), anti-CD11c–BB515 (clone B-ly6, BD (clone 2A9-1, BD Biosciences) and anti-CXCR3–PE-Cy5 (clone G025H7,
Biosciences), anti-CD14–PE-Cy5 (clone 61D3, Thermo Fisher Scien- BioLegend) for 10 min at 37 °C. Cells were then stained further with
tific), anti-CD16–BUV496 (clone 3G8, BD Biosciences), anti-CD19– anti-CD3–BUV805 (clone UCHT1, BD Biosciences), anti-CD4–PE-Cy5.5
BUV563 (clone HIB19, BD Biosciences), anti-CD27–BV786 (clone (clone RM4-5, Thermo Fisher Scientific), anti-CD8–BUV395 (clone
O323, BioLegend), anti-CD34–BB660 (clone 581, BD Biosciences), RPA-T8, BioLegend), anti-CD14–BV510 (clone M5E2, BioLegend),
anti-CD38–APC (clone HB7, BD Biosciences), anti-CD45–BUV805 anti-CD19–BV510 (clone HIB19, BioLegend), anti-CD27–BV786 (clone
(clone HI30, BD Biosciences), anti-CD45RA–BV570 (clone HI100, O323, BioLegend), anti-CD38–BUV496 (clone HIT2, BD Biosciences),
BioLegend), anti-CD56–BUV615 (clone NCAM16.2, BD Biosciences), anti-CD39–BV711 (clone A1, BioLegend), anti-CD45RA–BV570 (clone
anti-CD69–BUV737 (clone FN50, BD Biosciences), anti-CD71–BUV661 HI100, BioLegend), anti-CD95–BB700 (clone DX2, BD Biosciences),
(clone M-A712, BD Biosciences), anti-CD83–BB790 (clone HB15e, BD anti-CD127–BB630 (clone HIL-7R-M21, BD Biosciences), anti-HLA-DR–
Biosciences), anti-CD86–BB630 (clone 2331 (FUN-1), BD Biosciences), BV650 (clone G46-6, BD Biosciences), anti-LAG-3–BUV661 (clone
anti-CD123–PE-Cy7 (clone 7G3, BD Biosciences), anti-CD127–BV421 3DS223H, Thermo Fisher Scientific), anti-PD-1–BUV737 (clone EH12.1,
(clone A019D5, BioLegend), anti-HLA-DR–BV605 (clone G46-6, BD BD Biosciences), anti-TIGIT–PE-Dazzle594 (clone A15153G, BioLegend)
Biosciences) and anti-PD-1–R718 (clone EH12.1, BD Biosciences) for and anti-TIM-3–BV605 (clone F38-2E2, BioLegend) for 20 min at RT,
30 min at RT (Supplementary Table 8). Stained cells were washed washed twice with FACS buffer, fixed/permeabilized using a FoxP3
twice with FACS buffer (PBS containing 2% fetal bovine serum and Transcription Factor Staining Buffer Set (Thermo Fisher Scientific),
2 mM EDTA), fixed in Cytofix Fixation Buffer (BD Biosciences), and and stained intracellularly with anti-EOMES–EF660 (clone WD1928,
acquired using a FACSymphony A3 (BD Biosciences). Data were ana- eBioscience), anti-granzyme B–BB790 (clone GB11, BD Biosciences),
lyzed using FlowJo version 10.9.0 (FlowJo). anti-Ki67–AF700 (clone B56, BD Biosciences), anti-T-BET–PE-Cy7 (clone
4B10, eBioscience) and anti-TCF-1–AF488 (clone C63D9, Cell Signaling
AIM assay Technology) for 30 min at RT (Supplementary Table 11). Stained cells
PBMCs were thawed quickly, resuspended in RPMI 1640 Complete were washed twice with FACS buffer and acquired using a FACSym-
Medium (Sigma-Aldrich) supplemented with DNase I (10 U ml–1; phony A3 (BD Biosciences). Data were analyzed using FlowJo software
Sigma-Aldrich), and rested at 1 × 106 cells per well in 96-well U-bottom version 10.9.0 (FlowJo).
plates (Corning) for 3 h at 37 °C. The medium was then supplemented
with unconjugated anti-CD40 (clone HB14, Miltenyi Biotec) and Plasma proteomics
anti-CXCR5–BB515 (clone RF8B2, BD Biosciences), followed 15 min A data-driven approach was used to select healthy convalescent
later by the relevant peptides (each at 0.5 μg ml–1), and the cultures were individuals (n = 51) and individuals with long COVID (n = 51) for

Nature Immunology
Article https://doi.org/10.1038/s41590-025-02135-5

plasma proteome characterization via a Proximity Extension Assay References

(Olink Proteomics). Immune cell subset proportions were summarized 67. Rihn, S. J. et al. A plasmid DNA-launched SARS-CoV-2 reverse
using a PCA. Outlier samples were excluded based on the greatest devia- genetics system and coronavirus toolkit for COVID-19 research.
tion from the origin along PC1 to PC4. Plasma samples were analyzed PLoS Biol. 19, e3001091 (2021).
in two batches using Explore 3072 (Olink Proteomics). Sixteen bridge 68. Melenhorst, J. J. et al. Detection of low avidity CD8+ T cell populations
samples were included for quality control purposes in each batch. with coreceptor-enhanced peptide-major histocompatibility
complex class I tetramers. J. Immunol. Methods 338, 31–39 (2008).
General statistics 69. Price, D. A. et al. Avidity for antigen shapes clonal dominance in
Differences between groups were assessed using a two-tailed Mann– CD8+ T cell populations specific for persistent DNA viruses. J. Exp.
Whitney U-test. Raw P values are shown. Correlations were evaluated Med. 202, 1349–1361 (2005).
using the two-tailed Pearson coefficient or a two-tailed Spearman rank 70. Vlahava, V. M. et al. Monoclonal antibodies targeting
test. Significance was assigned at P < 0.05. Basic statistical analyses nonstructural viral antigens can activate ADCC against human
were performed using Prism version 9.5.0 (GraphPad). cytomegalovirus. J. Clin. Invest. 131, e139296 (2021).
71. Doncheva, N. T., Morris, J. H., Gorodkin, J. & Jensen, L. J.
Flow cytometry data analysis Cytoscape StringApp: network analysis and visualization of
Samples acquired for immune cell lineage analysis were gated to the proteomics data. J. Proteome Res. 18, 623–632 (2019).
single-cell/viable/CD45+ population and subsequently exported to 72. Cai, C., Gao, Y. & Buggert, M. Olink data for ‘Identification of soluble
contain only 3,000 events using the FlowJo Plugin DownSample version biomarkers that associate with distinct manifestations of long
3. Exported fcs files were loaded into R using flowCore version 2.6.0. All COVID’. Zenodo https://doi.org/10.5281/zenodo.14772494 (2025).
data were concatenated into a single matrix with compensated markers
(excluding viability, CD34 and CD45). Data for each marker were scaled Acknowledgements
and centered for analysis using umap version 0.2.10.0. Clustering We express our gratitude to all donors, health care personnel, study
was performed using a Gaussian mixture model (maxNumCompo- coordinators, administrators and laboratory managers involved in
nents = 10) implemented in mclust version 6.0.0. Data were visualized this work. The findings reported here represent independent research
using ggplot2 version 3.4.2. Antigen-specific CD4+ and CD8+ T cell fre- funded in part by the National Institute for Health and Care Research
quencies assessed via the AIM assay were calculated after background in response to the emergence of long COVID (COV-LT2-0041). The
subtraction. Samples acquired for detailed phenotypic characteriza- views expressed in this publication are those of the authors and
tion were excluded below a threshold of five tetramer+ CD8+ T cells per not necessarily those of the National Institute for Health and Care
specificity. The expression of each marker was then normalized to the Research or The Department of Health and Social Care (United
average geometric mean fluorescence intensity across all samples and Kingdom). Additional support was provided by the SciLifeLab/
specificities and used to calculate the co-inhibitory score, representing KAW National COVID-19 Research Program, the Swedish Research
the summed data for PD-1, TIM-3, LAG-3 and TIGIT. Statistical analyses Council (2021-06534) and the PolyBio Research Foundation (Balvi
were performed using R version 4.2.1. B43). Y.G. was supported by the Åke Wibergs Stiftelse (M22-0099)
and the Magnus Bergvalls Stiftelse (2022-307). C.C. was supported
Plasma proteome data analysis by the Swedish Society for Medical Research (PG-22-0432-H-01).
Bridge sample data were normalized using the olink_normalization S. Adamo was supported by a Swiss National Science Foundation
function implemented in OlinkAnalyze version 3.4.1. Differential Postdoc Mobility Grant (P500PB_211069). S. Aleman was supported
expression analyses were performed using a Wilcoxon rank-sum/ by the Swedish Research Council (2021-06534). M.B. was supported
Mann–Whitney U-test with Benjamini–Hochberg correction imple- by the Swedish Research Council (2018-02330, 2020-06121, 2021-
mented via the olink_wilcox function in OlinkAnalyze version 3.4.1. 01141, 2021-04779 and 2022-01313), the Knut and Alice Wallenberg
GSEA was performed using fgsea version 1.20.0 incorporating lists of Foundation (2021.0136), the European Research Council (101041484),
all analyzed proteins ordered by correlation coefficient or fold change. the Swedish Society for Medical Research (CG-22 0009), the Swedish
Gene sets were downloaded from the MSigDB using msigdb version Cancer Society (22 2237 Pj), the Karolinska Institutet (2019-00969,
7.5.1. Overrepresentation analysis was performed using the fora func- 2021-00513 and 2022-01719), the Åke Wibergs Stiftelse (M20-0190)
tion implemented in fgsea version 1.20.0 incorporating all measured and the Center for Innovative Medicine (FoUI-988204).
proteins as the ‘universe’. At least five proteins were required in each
gene set for consideration. Significance was evaluated using a hyper- Author contributions
geometric test. Correlations were calculated using the cor.test func- Y.G., C.C., S. Adamo, H.E.D., S. Aleman, M.B. and D.A.P. conceptualized
tion implemented in stats version 4.1.3. PCAs were performed using the study. Y.G., C.C., S. Adamo, S.L.-L., P.S.A., K.B. and J.W. performed
the prcomp function implemented in stats version 4.1.3. Data were experiments. Y.G., C.C. and S. Adamo analyzed and visualized data.
visualized using ggplot2 version 3.4.2 and pheatmap version 1.0.12. Y.G., E.B., H.K., L.D., K.L.M., S.L.-L., K.L., S.K., M.A., S.A.J., P.J., C.L.,
All analyses were performed using R version 4.2.1. Network analyses P.A.G., M.J.P., S.G.D., H.E.D., S. Aleman, M.B. and D.A.P. provided
of plasma proteins that were differentially expressed as a function of resources and/or collected samples. H.E.D., S. Aleman, M.B. and D.A.P.
symptom severity were performed using the stringApp in Cytoscape acquired funding for the project. R.J.S., H.E.D., S. Aleman, M.B. and
version 3.10.3 (ref. 71). D.A.P. supervised the work. Y.G., C.C., S. Adamo, S. Aleman, M.B. and
D.A.P. wrote the paper. M.B. and D.A.P. are joint senior authors and
Reporting summary codirected the study. All authors approved the final draft of the paper
Further information on research design is available in the Nature and concurred with the decision to submit for publication.
Portfolio Reporting Summary linked to this article.
Funding
Data availability Open access funding provided by Karolinska Institute.
Raw proteomics data are available via Zenodo (https://doi.org/10.5281/
zenodo.14772494)72. Any additional information required to reanalyze Competing interests
the data reported in this paper is available from the corresponding S. Aleman has received honoraria for educational events and lectures
authors upon reasonable request. unrelated to this work from Gilead, AbbVie, Biogen and MSD and

Nature Immunology
Article https://doi.org/10.1038/s41590-025-02135-5

reports grants from Gilead and AbbVie. M.B. is a consultant for Bristol Correspondence and requests for materials should be addressed to
Myers Squibb, Mabtech, Pfizer, Oxford Immunotec and MSD. The other Marcus Buggert or David A. Price.
authors declare no competing interests.
Peer review information Nature Immunology thanks Randy Cron
Additional information and the other, anonymous, reviewer(s) for their contribution to the
Extended data is available for this paper at peer review of this work. Primary Handling Editor: P. Jauregui, in
https://doi.org/10.1038/s41590-025-02135-5. collaboration with the Nature Immunology team.

Supplementary information The online version contains supplementary Reprints and permissions information is available at
material available at https://doi.org/10.1038/s41590-025-02135-5. www.nature.com/reprints.

Nature Immunology
Article https://doi.org/10.1038/s41590-025-02135-5

CCR7 CD38 CD11c CD3

CD8 CD86 CD16 CD123

CD19 CD56 CD69 CD71

CD127 CD45RA HLA-DR

High
CD27 CD14 CD4

Low

CX3CR1 CD83 PD−1

UMAP2

UMAP1
Extended Data Fig. 1 | Expression of immune cell lineage markers measured via flow cytometry. UMAP representation of individual immune cell lineage markers
among peripheral blood mononuclear cells after dimensionality reduction.

Nature Immunology
Article https://doi.org/10.1038/s41590-025-02135-5

a 56.8 97.4

98.3 99.2

SSC-A

SSC-A

SSC-A
FSC-A
SSC-W FCS-W FCS-A Viability

0.075

CD34–BB660
97.6

SSC-A
CD45–BUV805 CD45–BUV805

Naive Plasma B cells TCM Naive

CD8+ T 23.6 75.6 9.98
73.7 0.51 12.3 14.9 51.1

CCR7–APC-Cy7
CD19–BUV563

CD8–BUV396

CD27–BV786
CD38–APC

Memory TEFF TEMRA

24.7 37.8 T cells 49.7 CD4+ T 72.6 20.7 13.2 4.59 9.83

CD27–BV786 CD3–BV650 CD4–PE-Cy5.5 CD45RA–BV570 CX3CR1–PE

Immature Basophils 6.65 TCM Naive 95.5 1.44

2.60 CD56− 71.5
48.3 44.8

CCR7–APC-Cy7
pDCs

CD123–PE-Cy7
CD56+
CD56–BUV615

Mature
2.15

CD27–BV786
81.8 26.3

CD16−CD56dim TEFF TEMRA

SSC-A

13.6 Other 88.3 6.79 0.11 2.97 0.11

CD16–BUV496 CD56–BUV615 HLA-DR–BV605 CD45RA–BV570 CX3CR1–PE

HLA-DR+ 98.1 CD16 +

CD38high
6.24 HLA-DRhigh
Int.
HLA-DR–BV605

HLA-DR–BV605

HLA-DR–BV605
CD16–BUV496

61.9
CD86–BB630

ILCs 4.09
33.5 cDCs
CD14+ 89.2
HLA‐DR− 1.63 5.14 82.1

CD127–BV421 HLA-DR–BV605 CD14–PE-Cy5 CD11c–BB515 CD38–APC

CD69+ HLA-DR+
PD-1+ CD38+
CCR7–APC-Cy7

CCR7–APC-Cy7

CCR7–APC-Cy7

CCR7–APC-Cy7
2.91 2.55
2.63 16.4
CD4 T cell
+

PD-1–R718 CD69–BUV737 HLA-DR–BV605 CD38–APC

PD-1+ CD69+ HLA-DR+ CD38+

CCR7–APC-Cy7

CCR7–APC-Cy7

CCR7–APC-Cy7

CCR7–APC-Cy7

11.8 3.09 2.49 3.70

CD8+ T cell

PD-1–R718 CD69–BUV737 HLA-DR–BV605 CD38–APC

b Single cells Single cells

FSC-W

SSC-W
SSC-A

Lymphocytes

FSC-A FSC-A SSC-A

CD3+ 0.22
Memory
CCR7–APC-Cy7
CD3–BUV805

CD69–BV650

Dump–BV510 CD45RA–BV570 CD137–PE-Cy7

CD8 +
Memory
CCR7–APC-Cy7
CD8–BUV395

CD69–BV650

CD4+ 0.27

CD4–BUV496 CD45RA–BV570 CD154–BV421

Extended Data Fig. 2 | See next page for caption.

Nature Immunology
Article https://doi.org/10.1038/s41590-025-02135-5

Extended Data Fig. 2 | Flow cytometric gating strategies for the identification for the identification of antigen-specific CD4+ and CD8+ T cells via upregulation
of immune cell lineages and activation-induced marker upregulation. (a) Flow of the activation-induced markers CD69 and CD154 or CD69 and CD137,
cytometric gating strategy for immune cell lineage characterization. Numbers respectively. Numbers indicate percentages in the drawn gates.
indicate percentages in the drawn gates. (b) Flow cytometric gating strategy

Nature Immunology
Article https://doi.org/10.1038/s41590-025-02135-5

a T cells B cells Naive B cells

b NK cells ILCs
0.39 0.64
100 50 100 1.2
50
0.071 1
80 40 80
% of CD45+ cells

% of CD45+ cells

% of CD45+ cells

% of CD45+ cells
% of total B cells
40
0.14 0.8
60 30 60 30 0.11
0.6
40 20 40 20 0.4
20 10 20 10 0.2
0 0 0 0 0

HC LC HC LC HC LC HC LC HC LC

Naive CD4+ T cells Naive CD8+ T cells CD4+ T cells CD8+ T cells Immature NK cells Mature NK cells

100 100 100 50 0.14

0.65 50 100
0.31 0.28
% of total CD4+ cells

% of total CD8+ cells

40

% of CD45+ cells

% of CD45+ cells
80 80 80 40 80

% of NK cells

% of NK cells
0.22
60 60 60 30 30 60
0.27
40 40 40 20 20 40

20 20 20 10 10 20

0 0 0 0 0 0
HC LC HC LC HC LC HC LC HC LC HC LC

c d
CD14+ monocytes Intermediate monocytes CD16+ monocytes Basophils cDCs pDCs
0.003
0.61
100 0.00021 4 6
20 10 4
0.11 0.44 5 0.1
% of monocytes

% of monocytes

% of monocytes

% of CD45+ cells

% of CD45+ cells

% of CD45+ cells
80 3
15 8 3
4
60 6
10 2 2 3
40 4 2
5 1 1
20 2 1
0 0 0 0 0
0
HC LC HC LC HC LC HC LC HC LC HC LC

e Spike Nucleocapsid Mem + Env ORF3–10

0.32 0.86 0.34 10
10 0.0307 10 10
0.0142 0.11
0.21
1 1 1
% of T cells

% of T cells

% of T cells

1 % of T cells 0.06
0.1 0.1 0.1

0.1
0.01 0.01 0.01

0.01 0.001 0.001 0.001

CD69+CD40L+ CD69+4-1BB+ CD69+CD40L+ CD69+4-1BB+ CD69+CD40L+ CD69+4-1BB+ CD69+CD40L+ CD69+4-1BB+

CMV EBV latent EBV lytic

0.0093
10 0.35 10 10 0.0117
0.0228
0.0298
1 1 1 0.089 HC
% of T cells

% of T cells

% of T cells

0.1 0.1 0.1

LC
0.01 0.01 0.01

0.001 0.001 0.001

CD69+CD40L+ CD69+4-1BB+ CD69+CD40L+ CD69+4-1BB+ CD69+CD40L+ CD69+4-1BB+

f
CD4+ T cells CD8+ T cells
Spike

Nucleocapsid

Mem + Env Log2 (fold change)

3
ORF3–10 2
1
CMV 0

EBV latent

EBV lytic
IT

IT
28

38

C 71

-1

39

95

28

38

-1

39

95

7
R

12

12
-D

-D
G

G
PD

PD
D

D
C

C
TI

TI
D

D
LA

LA
C

C
X3

X3
C

C
H

Extended Data Fig. 3 | See next page for caption.

Nature Immunology
Article https://doi.org/10.1038/s41590-025-02135-5

Extended Data Fig. 3 | Immune cell lineages and T cell immunity in healthy SARS-CoV-2, CMV, or EBV. (f) Heatmap summarizing the phenotypic attributes
convalescent individuals and patients with long COVID recruited from of functional CD4+ and CD8+ T cells targeting defined proteins from SARS-
Sweden. (a) Scatter dot plots showing the frequencies of naive and total B and CoV-2, CMV, or EBV. Data are shown for each marker as the log2-transformed
T cells gated manually. (b) Scatter dot plots showing the frequencies of innate fold change in percent positive for each population among patients with long
lymphocytes gated manually. (c) Scatter dot plots showing the frequencies of COVID (LC) versus healthy convalescent individuals (HC). *P < 0.05, **P < 0.01.
monocytes gated manually. (d) Scatter dot plots showing the frequencies of HC, n ≤ 20; LC, n ≤ 56 (a, b, c, d, e and f). Horizontal bars represent median values
basophils and dendritic cells gated manually. (e) Scatter dot plots showing the (a, b, c, d and e). Significance was evaluated using a two-tailed Mann–Whitney U
frequencies of functional CD4+ and CD8+ T cells targeting defined proteins from test (a, b, c, d, e and f).

Nature Immunology
Article https://doi.org/10.1038/s41590-025-02135-5

a
0.032 Spike
53.8

Tetramer–BV421
CD8–BUV395
Dump–BV510
Non-spike
99.2
SSC-A

FSC-H
63.4 75.8 6.21E−3

FSC-A FSC-A CD3–BUV805 CD4–PE-Cy5.5 Tetramer–PE

Spike+ Spike+ Memory spike+ Memory spike+ Memory spike+

28.0 24.4
50.0 1.39 80.6
CD45RA–BV570

HLA-DR–BV650
5.56

CD8–BUV395

CD8–BUV395
CD95–BB700

37.8 9.76 50.0

CCR7–APC-Cy7 CCR7–APC-Cy7 CD38–BUV496 CXCR3–PE-Cy5 GzmB–BB790

b 7
6
5

Cluster
Count

4
3
2
1
PD-1 TIM-3 LAG-3 TIGIT

c EBV (lytic)
0.48 0.11 0.11 0.61 0.48 0.14
250 1000 2500 700
1000 400
EOMES (gMFI)

600
HC
LAG-3 (gMFI)

T-BET (gMFI)

200
TIGIT (gMFI)
TIM-3 (gMFI)
PD-1 (gMFI)

800
800
150
300 2000 500
LC
600 600 400
100
1500 300
400 50 200 400
200
0 100
200 200 1000

d CMV
0.15 0.84 0.51 0.32 0.81 0.41 0.53 0.68
800 600
600 1250 1000 4000
Coinhibotory score

7.5
LAG-3 (gMFI)

EOMES (gMFI)
TIM-3 (gMFI)

900
PD-1 (gMFI)

T-BET (gMFI)
TCF-1 (gMFI)
TIGIT (gMFI)

1000 750
3000
400 400
600 400 750
5.0
500
HC
2000
LC
0 500 2.5 250 200
300 1000
200
250 0
0.0 0
0 0

e EBV (latent)
0.2 0.81 0.2 0.71 0.23 0.31 0.83 0.81
2000 1000 2500 600
1000
Coinhibotory score

700
EOMES (gMFI)

750 9
LAG-3 (gMFI)

T-BET (gMFI)
TCF-1 (gMFI)

1000 2000
TIM-3 (gMFI)

TIGIT (gMFI)

1500
PD-1 (gMFI)

750 400

1000
500 6
600
1500 HC
500 500
250
500 200 LC
3 1000
250 500 400
0 0 0
0 500
300

Extended Data Fig. 4 | See next page for caption.

Nature Immunology
Article https://doi.org/10.1038/s41590-025-02135-5

Extended Data Fig. 4 | Additional phenotypic characteristics of virus-specific CD8+ T cells targeting lytic epitopes from EBV. (d) Scatter dot plots showing
CD8+ T cells in healthy convalescent individuals and patients with long COVID the expression intensities of selected markers among tetramer+ CD8+ T cells
recruited from the UK. (a) Flow cytometric gating strategy for the identification targeting epitopes from CMV. (e) Scatter dot plots showing the expression
of tetramer+ CD8+ T cells directly ex vivo. Numbers indicate percentages in the intensities of selected markers among tetramer+ CD8+ T cells targeting latent
drawn gates. (b) Flow cytometry histograms showing the expression patterns epitopes from EBV. Healthy convalescent individuals (HC), n = 17; patients with
of coinhibitory receptors among clusters of tetramer+ CD8+ T cells targeting long COVID (LC), n = 15 (b, c, d and e). Horizontal bars represent median values
nonspike epitopes from SARS-CoV-2 identified using Phenograph. (c) Scatter dot (c, d and e). Significance was evaluated using a two-tailed Mann–Whitney U test
plots showing the expression intensities of selected markers among tetramer+ (c, d and e). gMFI, geometric mean fluorescence intensity.

Nature Immunology
Article https://doi.org/10.1038/s41590-025-02135-5

a b HC vs. LC
adj. P < 0.05
50
BMI Cardiometabolic Inflammation
Normal EEF1D S100A13 SCRIB SHMT1 RBPMS
25 EIF2AK3 DMD WAS
3 3 MYO9B
Overweight CLIP2 NFAT5
PC2

COPB2 ENOX2 CORO1A

Obese SPARCL1 KIF1C PRKG1
0 MET

−log10 (P)

−log10 (P)
Not reported PDZD2 CLSTN2
2 BOC 2
IGFBP1 LMOD1
CKAP4
−25 FGL1
TSLP PON3
1 1
−50 0 50
PC1
0 0

c Enriched in HC
PID pathways
Enriched in LC
−0.4 0.0 0.4
log2 (fold change)
0.8 −0.5 0.0 0.5
log2 (fold change)
1.0

P HIV NEF pathway Neurology Oncology

0.4 Ceramide pathway MYO6 AZI2 GSAP
MYOC
SCRIB
4
0.3 3 SPTLC1 ANGPT2
TCR pathway NAPRT EIF4G3 SCRIB
0.2 PTEN MAVS
NUBP1
0.1 3
PDGFRB pathway FBN2 ISLR2

−log10 (P)

−log10 (P)
2 TMEM25 CNTN4 WASF3 DUT
UPA/UPAR pathway EPB41L5 CLEC14A 2 PCDHB15
KLK4 NBL1
Integrin1 pathway CLEC10A DDR1 ITGB7 KLK8
1 TP73 OMP
1 UPK3BL1 ITIH5
HNF3B pathway

Integrin5 pathway 0 0

−2.0 −1.2 −0.4 0.4 1.2 2.0 −0.4 0.0 0.4 0.8 0.0 0.5
NES log2 (fold change) log2 (fold change)

d Breathlessness
Score = 6–10
BMI
Score >30
Usual activities
Score = 6–10
Anxiety
Score = 6–10
Mobility
Score = 6–10
e
Oncology Enriched in Enriched in
no breathlessness severe breathlessness
Neurology
Inflammation
Cardiometabolic Apical surface (H)
Score = 0 Score <25 Score = 0 Score = 0 Score = 0 PI3K AKT MTOR
signaling (H)
Oncology
Neurology
Mitotic spindle (H)
Inflammation
PI3KCI AKT
Cardiometabolic pathway (PID)
0 40 80 120 160 200 0 40 80 120 160 200 0 40 80 120 160 200 0 40 80 120 160 200 0 40 80 120 160 200
Oxidative
Number of positive differentially expressed Olink proteins phosphorylation (H) P

Fatigue Pain Personal care Depression Dyspnea-12 Ceramide pathway (PID)

Score = 6–10 Score = 6–10 Score = 6–10 Score = 6–10 Score = 19–36 0.02
Oncology TCPTP pathway (PID)
0.01
Neurology
Inflammation HIV NEF pathway (PID)
Cardiometabolic
MET pathway (PID)
Score = 0 Score = 0 Score = 0 Score = 0 Score = 0–3
Oncology TXA2 pathway (PID)
Neurology
Inflammation −2.0 −1.2 −0.4 0.4 1.2 2.0
Cardiometabolic NES
0 40 80 120 160 200 0 40 80 120 160 200 0 40 80 120 160 200 0 40 80 120 160 200 0 40 80 120 160 200 H: Hallmark
Number of positive differentially expressed Olink proteins PID: Pathway interaction database

f −1 −0.5 0
Spearman
0.5 1
g
Spearman = 0.25
P = 0.02
n.s. **** **** **** **** **** **** **** **** **** **** Overall health (1–10)
n.s. *** n.s. n.s. n.s. n.s. n.s. n.s. n.s. ** n.s. Age
**** *** *** **** ** * ** ** n.s. * n.s. BMI 40
**** n.s. *** **** **** **** **** **** *** **** **** Fatigue
**** n.s. **** **** **** **** **** **** **** **** **** Usual activities 36
**** n.s. ** **** **** **** **** **** **** **** **** Dyspnea-12 (/36)
Breathlessness
**** n.s. * **** **** **** **** **** *** **** ***
32
BMI

**** n.s. ** **** **** **** **** **** *** **** **** Pain
Anxiety
**** n.s. ** **** **** **** **** **** **** * *
28
**** n.s. n.s. *** **** **** *** *** **** * ** Depression
Mobility
24
**** ** * **** **** **** **** **** * * ****

**** n.s. n.s. **** **** **** *** **** * ** **** Personal care
Overall health (1–10)

BMI
Fatigue
Usual activities
Dyspnea-12 (/36)
Breathlessness
Pain
Anxiety
Depression
Mobility
Personal care
Age

20
0 2 4 6 8 10

Breathlessness score

Extended Data Fig. 5 | See next page for caption.

Nature Immunology
Article https://doi.org/10.1038/s41590-025-02135-5

Extended Data Fig. 5 | Dysregulation of the plasma proteome associated with individuals (n = 34) and patients with long COVID (n = 48), irrespective of clinical
clinical assignation and symptomatology in healthy convalescent individuals assignation. Significance was evaluated using a two-tailed Mann–Whitney U test
and patients with long COVID recruited from the UK. (a) Principal component with (red) or without Benjamini-Hochberg correction (gray). (e) GSEA showing
analysis of plasma protein concentrations colored by body mass index (BMI) for the corresponding differentially expressed plasma proteins by rank versus the
healthy convalescent individuals (n = 51) and patients with long COVID (n = 51). highest and lowest breathlessness score tiers, irrespective of clinical assignation.
(b) Volcano plots showing the corresponding differentially expressed plasma (f) Heatmap showing correlations among clinical scores for healthy convalescent
proteins from each panel versus clinical assignation. Significance was evaluated individuals (n = 51) and patients with long COVID (n = 51), irrespective of clinical
using a two-tailed Mann–Whitney U test with (red) or without Benjamini- assignation. (g) Correlation dot plot showing the corresponding breathlessness
Hochberg correction (gray). The dashed line indicates P = 0.05. (c) Gene set scores versus BMI, irrespective of clinical assignation. Significance was evaluated
enrichment analysis (GSEA) showing the corresponding differentially expressed using the GSEA method without correction (c and e) or a two-tailed Spearman
plasma proteins by rank versus clinical assignation. (d) Bar plots showing rank test (f and g). H, Hallmark; NES, normalized enrichment score; PID, Pathway
the numbers of differentially upregulated plasma proteins from each panel Interaction Database.
versus the highest and lowest symptom score tiers for healthy convalescent

Nature Immunology
Article https://doi.org/10.1038/s41590-025-02135-5

a Protein expression correlated with breathlessness score

b Atelectasis
HPO PID

RHOA pathway
Panel
Lymphedema PDGFRB pathway
Cardiometabolic
150
Inflammation Elevated hepatic
TXA2 pathway
Number of proteins

transaminase P
Neurology
Tachypnea ERBB1 internalization 0.03
100
pathway
Oncology 0.02
−1 0 1 2
NES MET pathway
0.01
Hallmark pathways
50
Fatty acid metabolism FAS pathway

ENPP5 RPS10 Mitotic spindle

TNF pathway
0 Oxidative phosphorylation
PTP1B pathway
−0.5 −0.4 −0.3 −0.2 −0.1 0.0 0.1 0.2 0.3 0.4 0.5 MTORC1 signaling
Pearson correlation coefficient −1 0 1 2 −1 0 1 2
NES NES

c DCTD
NT5C3A
CRELD2

NEXN
DBNL
PPP1R9B PDE5A MYO9B MANF
ARHGEF12

GLOD4 DAAM1
PDLIM7 ENAH GIT1 SPART
DBN1
TRIM5

JAM3 PRKCQ
SNX15 CLIP2
CD226 SELPLG WAS SCRIB
PNLIP
CD84 CRKL
VASP
IRAK1 AMOT DGKA
TNFRSF13C PRKG1 LATS1
IKBKG C1QTNF5
TNFRSF17 ITGA2
HEXIM1
CD70 MAP2K6 F2R
YY1 GLA
TNFRSF14 CD40 TOP2B
CD244 CD40LG GPI
CXCL8
NRGN GNPDA2
CXCL3 SHMT1
LY9 IL18

CD300A CCL4 MARS1

COL18A1 STAT2 SPINT2
CCL3 TGFB1
PPBP
DNAJB2 XIAP SAMD9L
TREML1
IL7
IL1B CASP9
ADD1
MPIG6B PRR5
CXCL6
GAPDH
PDGFB CSF3R EIF4E
DAPK2 ANGPT1 SERPINF1
BAG4
NTF3 PRDX3 PRKAB1
FKBP1B EGF
SPRY2 EIF4G1
TSC1
TXN HSPA1A
PLA2G4A GLRX5
SPRED2 CASP2
PGR SNCA DNAJA2
VAMP8 ST13
PTGES2 DNAJB6
CSNK1D TTR
STX5
NUDC
SCGB1A1 FIS1
PIKFYVE ERP29 CACYBP
STX7
STX8 NDUFA5
RBPMS
BNIP3L

VTI1A

Extended Data Fig. 6 | Dysregulation of the plasma proteome associated by correlation with breathlessness scores for healthy convalescent individuals
with breathlessness in healthy convalescent individuals and patients (n = 34) and patients with long COVID (n = 48), irrespective of clinical assignation.
with long COVID recruited from the UK. (a) Stacked histogram showing the Significance was evaluated using the GSEA method without correction. NES,
distribution of correlation coefficients from pairwise comparisons of plasma normalized enrichment score. (c) Network analysis showing the corresponding
protein concentrations versus breathlessness scores for healthy convalescent differentially expressed plasma proteins from the inflammation panel depicted
individuals (n = 34) and patients with long COVID (n = 48), irrespective of using Cytoscape. Each node represents a protein. Node color indicates protein
clinical assignation. Significance was evaluated using the two-tailed Pearson concentration, and node size indicates significance. Red denotes overexpression
coefficient. (b) Gene set enrichment analysis (GSEA) showing selected terms in patients with long COVID, and blue denotes underexpression in patients with
from Human Phenotype Ontology (HPO), the Pathway Interaction Database long COVID. Each edge represents the functional relevance between a pair of
(PID), and the Hallmark Collection. Plasma protein concentrations were ranked proteins, and line thickness represents the level of confidence.

Nature Immunology
Article https://doi.org/10.1038/s41590-025-02135-5

a b
Protein expression correlated with Borg CR10 score PID

200 Syndecan-4 pathway

Panel
Cardiometabolic TXA2 pathway
Inflammation P
Glypican pathway
150
Number of proteins

Neurology
VEGFR1/2 pathway 0.04
Oncology
EPO pathway 0.03
100 0.02
TAp63 pathway
Ceramide pathway 0.01

50 ILK pathway
Smad2/3 nuclear pathway

0 −1.6 −1.2 −0.8 −0.4 0.0 0.4 0.8 1.2 1.6 2.0
NES
−0.3 −0.2 −0.1 0.0 0.1 0.2 0.3 Proteins negatively Proteins positively
Pearson correlation coefficient correlated with Borg CR10 correlated with Borg CR10

c TPD52L2
MTDH
PRSS8

DFFA SPINT2 RIDA

GLOD4
PAXX SHMT1
PSIP1
HEXIM1
GPI
VTI1A TRIM5 NUB1 LCAT

MARS1 ATP5IF1
BTD PIKFYVE
EIF4G1
VAMP8 EIF4E C1RL
PHYKPL STX7 KYNU
PTGES2 PLA2G4A AMOT
MGMT AFM
VNN1 XIAP YWHAQ
PARP1 ITIH1
PSTPIP2 BAG4
TTR
IRAK1 CFI ATRN
IKBKG GAPDH
TNFRSF11B MAP2K6
MERTK FGA
DAAM1 SERPINA1
GOPC TNFRSF11A RAP1A
CD83 CFP
IRAK4 APCS
SERPINA5
RABEP1 SLC9A3R1 CD40 TREM2 C3 DAG1
BCHE TF CFH
TREML1
BCR TRAF3 DECR1
IL18 KLKB1
DBNLPDZK1 PROS1 GP5
A1BG
ARHGEF12 TNFRSF14 CCL7 PCBD1
SNCA EGF PLG CPB2
CXCL9
PDIA3 CCL3
NEXN
PPP1R9B ANGPT1
CRKL GCHFR
MANF
MPIG6B PDGFB
NFATC1 CCL26 SERPINF1
FIS1
CCL13 MMP1 FGF19
LAT
STAT2 PTPN6 CCL28 GHR

ERP29 SAMD9L F2R

DNAJB2 PGF PLXNA4
TLR1 VASP
NRGN ITGA6
HLA-DRA
PIK3AP1 BNIP3L ADAMTS1
PRKG1 BSG
PDLIM7 PRDX5
ITGA11
DAPP1 JAM3 PDE5A DNAJA2
TNFAIP8L2 GLRX5 NPPC LYVE1

SKAP2

Extended Data Fig. 7 | Dysregulation of the plasma proteome associated with method without correction. NES, normalized enrichment score. (c) Network
perceived exertion in patients with long COVID recruited from Sweden. analysis showing the corresponding differentially expressed plasma proteins
(a) Stacked histogram showing the distribution of correlation coefficients from from the inflammation panel depicted using Cytoscape. Each node represents
pairwise comparisons of plasma protein concentrations versus Borg CR10 score a protein. Node color indicates protein concentration, and node size indicates
for patients with long COVID (n = 95). Significance was evaluated using the significance. Red denotes overexpression in patients with a moderate Borg CR10
two-tailed Pearson coefficient. (b) Gene set enrichment analysis (GSEA) showing score, and blue denotes underexpression in patients with a moderate Borg CR10
plasma protein concentrations ranked by correlation with Borg CR10 scores for score. Each edge represents the functional relevance between a pair of proteins,
patients with long COVID (n = 95). Significance was evaluated using the GSEA and line thickness represents the level of confidence.

Nature Immunology
 nature portfolio | reporting summary
Corresponding author(s): Marcus Buggert, David A Price
Last updated by author(s): Feb 26, 2025

Reporting Summary
Nature Portfolio wishes to improve the reproducibility of the work that we publish. This form provides structure for consistency and transparency
in reporting. For further information on Nature Portfolio policies, see our Editorial Policies and the Editorial Policy Checklist.

Statistics
For all statistical analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods section.
n/a Confirmed
The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement
A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly
The statistical test(s) used AND whether they are one- or two-sided
Only common tests should be described solely by name; describe more complex techniques in the Methods section.

A description of all covariates tested

A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons
A full description of the statistical parameters including central tendency (e.g. means) or other basic estimates (e.g. regression coefficient)
AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g. confidence intervals)

For null hypothesis testing, the test statistic (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted
Give P values as exact values whenever suitable.

For Bayesian analysis, information on the choice of priors and Markov chain Monte Carlo settings
For hierarchical and complex designs, identification of the appropriate level for tests and full reporting of outcomes
Estimates of effect sizes (e.g. Cohen's d, Pearson's r), indicating how they were calculated
Our web collection on statistics for biologists contains articles on many of the points above.

Software and code

Policy information about availability of computer code
Data collection BD FACSDiva version 9.1
Olink Explore version 3.0.5

Data analysis R version 4.2.1

FlowJo version 10.9.0
Prism version 10
CytoScape version 3.10.3
For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors and
reviewers. We strongly encourage code deposition in a community repository (e.g. GitHub). See the Nature Portfolio guidelines for submitting code & software for further information.
April 2023

1
 nature portfolio | reporting summary
Data
Policy information about availability of data
All manuscripts must include a data availability statement. This statement should provide the following information, where applicable:
- Accession codes, unique identifiers, or web links for publicly available datasets
- A description of any restrictions on data availability
- For clinical datasets or third party data, please ensure that the statement adheres to our policy

Raw proteomics data are available via Zenodo (doi.org/10.5281/zenodo.14772494). This paper does not report original code. Data from the Molecular Signatures
Database (MSigDB) used version 7.5.1. Any additional information required to reanalyze the data reported in this paper is available from the corresponding authors
upon request.

Research involving human participants, their data, or biological material

Policy information about studies with human participants or human data. See also policy information about sex, gender (identity/presentation),
and sexual orientation and race, ethnicity and racism.
Reporting on sex and gender Biological sex is reported in Figure 1a, Table 1 and 2.

Reporting on race, ethnicity, or Ethnicity/race is reported in Table 1 and 2.

other socially relevant
groupings

Population characteristics Population characteristics are summarized Table 1 and 2.

Recruitment Participants were recruited from two distinct sites as healthy convalescent individuals or patients with long COVID.

Ethics oversight Cardiff University School of Medicine Research Ethics Committee

Health Research Authority and Health and Care Research Wales
Swedish Ethical Review Authority

Note that full information on the approval of the study protocol must also be provided in the manuscript.

Field-specific reporting
Please select the one below that is the best fit for your research. If you are not sure, read the appropriate sections before making your selection.

Life sciences Behavioural & social sciences Ecological, evolutionary & environmental sciences
For a reference copy of the document with all sections, see nature.com/documents/nr-reporting-summary-flat.pdf

Life sciences study design

All studies must disclose on these points even when the disclosure is negative.
Sample size No formal sample size calculation was performed. The sample size was determined based on feasibility, ethical considerations, and prior
studies investigating similar research questions.

Data exclusions No data were excluded from the flow cytometry analyses. A few samples were not selected for proteomic analyses due to cost limitations.
Samples not selected for proteomic analyses were considered to be outliers based on circulating immune cell lineage composition (Figure 2).

Replication Replication was not performed for individual samples due to limited material availability.

Randomization Randomization was applied where relevant in this study. Specifically, for the OLINK Proteomics analysis, sample preparation followed the
manufacturer’s recommendations, which included a randomization step during sample loading onto the 96-well plate to minimize potential
batch effects.

Blinding Investigators were not blinded. All samples were treated identically. Each experimental protocol was performed identically for each sample.

Reporting for specific materials, systems and methods

April 2023

We require information from authors about some types of materials, experimental systems and methods used in many studies. Here, indicate whether each material,
system or method listed is relevant to your study. If you are not sure if a list item applies to your research, read the appropriate section before selecting a response.

2
Materials & experimental systems Methods

nature portfolio | reporting summary

n/a Involved in the study n/a Involved in the study
Antibodies ChIP-seq
Eukaryotic cell lines Flow cytometry
Palaeontology and archaeology MRI-based neuroimaging
Animals and other organisms
Clinical data
Dual use research of concern
Plants

Antibodies
Antibodies used CCR7 APC-Cy7 353212 G043H7 BioLegend
CD38 APC 567144 HB7 BD Biosciences
PD-1 R718 566974 EH12.1 BD Biosciences
CD11c BB515 564491 B-ly6 BD Biosciences
CD86 BB630 Custom 2331(FUN-1) BD Biosciences
CD34 BB660 Custom 581 BD Biosciences
CD83 BB790 Custom HB15e BD Biosciences
CD16 BUV496 612945 3G8 BD Biosciences
CD19 BUV563 741361 HIB19 BD Biosciences
CD56 BUV615 613002 BD Biosciences
CD69 BUV737 612817 FN50 BD Biosciences
CD45 BUV805 612891 HI30 BD Biosciences
CD127 BV421 351310 A019D5 BioLegend
CD3 BV650 317323 OKT3 BioLegend
CD27 BV786 302832 O323 BioLegend
CD14 PE-Cy5 15-0149-41 61D3 Thermo Fisher Scientific
CD4 PE-Cy5.5 MHCD0418 S3.5 Thermo Fisher Scientific
CD123 PE-Cy7 560826 7G3 BD Biosciences
CD38 APC-R700 564979 HIT2 BD Biosciences
CXCR3 AF647 353712 G025H7 BioLegend
CCR4 BB700 566475 1G1 BD Biosciences
CXCR5 BB515 564624 RF8B2 BD Biosciences
CD137 PE-Cy7 309818 4B4-1 BioLegend
CD127 PE-Cy5 351324 A019D5 BioLegend
CD95 PE-Dazzle594 305634 DX2 BioLegend
CX3CR1 PE 341604 2A9-1 BioLegend
TIGIT BV786 747838 741182 BD Biosciences
CD39 BV711 328228 A1 BioLegend
CD69 BV650 310934 FN50 BioLegend
HLA-DR BV605 562845 G46-6 BD Biosciences
CD45RA BV570 304132 HI100 BioLegend
CD14 BV510 301842 M5E2 BioLegend
CD19 BV510 302242 HIB19 BioLegend
LIVE/DEAD Fixable Aqua L34966 Thermo Fisher Scientific
CD154 BV421 310824 24-31 BioLegend
CD3 BUV805 612895 UCHT1 BD Biosciences
CCR6 BUV737 612780 11A9 BD Biosciences
CD71 BUV661 750651 M-A712 BD Biosciences
PD-1 BUV615 612991 EH12.1 BD Biosciences
CD28 BUV563 741392 CD28.2 BD Biosciences
CD4 BUV496 612936 SK3 BD Biosciences
CD8 BUV395 563795 RPA-T8 BD Biosciences
CX3CR1 BUV615 751513 2A9-1 BD Biosciences
CXCR3 PE-Cy5 353756 G025H7 BioLegend
PD-1 BUV737 612791 EH12.1 BD Biosciences
LAG-3 BUV661 376-2239-42 3DS223H Thermo Fisher Scientific
CD38 BUV496 612946 HIT2 BD Biosciences
CD27 BV786 302832 O323 BioLegend
HLA-DR BV650 564231 G46-6 BD Biosciences
TIM-3 BV605 345018 F38-2E2 BioLegend
CD95 BB700 566543 DX2 BD Biosciences
CD127 BB630 Custom HIL-7R-M21 BD Biosciences
CD4 PE Cy5.5 35-0042-82 RMA-4.5 Thermo Fisher Scientific
April 2023

TIGIT PE-Dazzle 372716 A15153G BioLegend

Granzyme B BB790 Custom GB11 BD Biosciences
TCF-1 AF488 6444S C63D9 Cell Signaling
T-BET PE-Cy7 25-5825-82 4B10 eBioscience
KI67 AF700 561277 B56 BD Biosciences

3
 EOMES EF660 50-4877-42 WD1928 eBioscience
Anti-mouse IgG HRP 115-005-003 polyclonal Jackson ImmunoResearch

nature portfolio | reporting summary

Anti-SARS-CoV-2 nucleocapsid protein BSM-41411M 1C7 Stratech Scientific

Validation All primary antibodies used in this study were commercially purchased and routinely tested by the respective manufacturers for
validation in flow cytometry.

Eukaryotic cell lines

Policy information about cell lines and Sex and Gender in Research
Cell line source(s) A549 (ATCC#CCL-185) and VeroE6 cells (ATCC#CRL-1586) expressing human angiotensin-converting enzyme 2 (ACE2) and
transmembrane serine protease 2 (TMPRSS2).

Authentication The A549 (ATCC# CCL-185) and VeroE6 (ATCC# CRL-1586) cell lines were authenticated through morphology assessment, STR
(short tandem repeat) profiling. No additional authentication was performed beyond the manufacturer’s validation. Cells
were used at low passage numbers and routinely monitored for morphology and growth characteristics.

Mycoplasma contamination None detected testing routinely.

Commonly misidentified lines No commonly misidentified cell lines were used in this study.
(See ICLAC register)

Plants
Seed stocks Not applicable.

Novel plant genotypes Not applicable.

Authentication Not applicable.

Flow Cytometry
Plots
Confirm that:
The axis labels state the marker and fluorochrome used (e.g. CD4-FITC).
The axis scales are clearly visible. Include numbers along axes only for bottom left plot of group (a 'group' is an analysis of identical markers).
All plots are contour plots with outliers or pseudocolor plots.
A numerical value for number of cells or percentage (with statistics) is provided.

Methodology
Sample preparation PBMCs were isolated via standard density gradient centrifugation and cryopreserved in fetal bovine serum (Thermo Fisher
Scientific) containing 10% dimethyl sulfoxide (Sigma-Aldrich).

Instrument BD FACSymphony™ A3 or BD FACSymphony™ A5.

Software BD FACSDiva version 9.1

FlowJo version 10.9.0

Cell population abundance No cell sorting experiments were performed in this study.

Gating strategy Gating strategies are shown in Extended Data Figures 2a, 2b, 4b.

Tick this box to confirm that a figure exemplifying the gating strategy is provided in the Supplementary Information.
April 2023