SARS-CoV-2-specific monoclonal antibody AER002 in Long COVID: an exploratory randomized phase 2a mechanistic trial.

The study was conducted in accordance with all relevant ethical regulations and was approved by the University of California, San Francisco Institutional Review Board. All participants provided written informed consent before enrollment and completion of study procedures. Participants were recruited from the University of California, San Francisco (UCSF)-based Long-term Impact of Infection with Novel Coronavirus (LIINC) cohort ( NCT04362150 ) 69 , which enrolls and prospectively measures adults in the post-acute phase of SARS-CoV-2 infection. All participants were required to enroll in LIINC prior to screening for this trial. Briefly, interested LIINC participants completed a screening visit at which their eligibility for this trial was assessed. The trial enrolled adults (≥18 years) with clinical evidence of Long COVID, defined as at least two moderate or severe symptoms that were new or worsened since the time of SARS-CoV-2 infection and were not known to be attributable to another cause, which had been present for at least 60 days following a test-confirmed SARS-CoV-2 infection that occurred in the period in which AER002 was expected to have neutralizing activity against circulating variants (i.e., prior to August 15, 2022). Our definition was based on the World Health Organization consensus definition 70 ; although not available when the study was planned, our definition is consistent with that which has since been put forth by the National Academies of Sciences, Engineering, and Medicine 1 . We excluded those who had ever previously received a SARS-CoV-2-specific monoclonal antibody, had received COVID-19 convalescent plasma or a SARS-CoV-2 vaccine within the preceding 60 days, who had major surgery within the prior 6 months, had medical comorbidities including active cardiovascular disease (defined as recent myocardial infarction or stroke, or history of congestive heart failure or pulmonary arterial hypertension), viral hepatitis, uncontrolled HIV infection, coagulopathy, severe anemia, moderate or severe immunocompromise according to the 2023 CDC COVID-19 guidelines 71 , and those with a history of anaphylaxis or similar significant allergic reaction. We also excluded those with a diagnosis of ME/CFS or dysautonomia that preceded and remained active prior to their SARS-CoV-2 infection their SARS-CoV-2 infection. The full list of eligibility criteria is in Supplementary Table  9 . Following confirmation of eligibility, participants underwent a series of baseline assessments. On the day of the infusion, they were randomized in a 2:1 ratio to receive a single intravenous infusion of study drug or placebo, which was prepared by an unblinded pharmacist. The participant and study team were blinded to treatment allocation. Participants underwent follow-up evaluations at approximately Day (D) 14, D30, D90, D180, and D360 following the infusion. The study team and participants remained blinded until the database lock following the D180 visit for all participants, after which participants were followed until D360 in an open-label manner to assess long-term safety outcomes. This exploratory phase 2a trial was designed primarily to assess safety and tolerability, with clinical, pharmacokinetic, and mechanistic outcomes included to generate preliminary effect estimates and inform the design of subsequent, adequately powered efficacy trials. The primary safety endpoint was the number of participants in each arm who experienced a grade 3 or greater adverse event that was determined to be definitely, probably, or possibly related to study treatment. The primary efficacy endpoint was the change in the PROMIS-29 Physical Health Summary Score between Baseline and D90. D90 was selected to allow sufficient time for AER002 to affect a putative persistent viral reservoir and for downstream clinical effects to emerge while the long-acting antibody remained at substantial systemic concentrations. Earlier and later assessments were included to characterize the onset and durability of any treatment effect. There were multiple pre-specified secondary endpoints as outlined in Supplementary Table  10 . Pre-specified, exploratory measures included pharmacokinetic measures and analyses of various biomarkers. Participants received a single intravenous infusion of AER002 1200 mg or matching placebo. The 1200 mg dose was selected on the basis of a preceding phase 1 dose-escalation study in healthy adults, in which it was safe and well tolerated and produced prolonged serum exposure 31 . The dose was chosen to maximize the magnitude and duration of exposure in this proof-of-concept trial, as no serum concentration threshold has been established for therapeutic activity against presumed SARS-CoV-2 persistence in Long COVID. The placebo was comprised of the formulation buffer for AER002, which included 20 mM histidine, 8% sucrose, and 0.04% Polysorbate 80. For this study, the assigned product was admixed in 250 mL of normal saline and infused over 30 min. Participants completed the self-administered Patient-Reported Outcomes Measurement Information System (PROMIS)-29 v2.1 survey to assess broad functional ability and quality of life 33 . The PROMIS questionnaires are scaled to fit a T-scale, with a normal distribution including a mean score of 50 and a standard deviation of 10 points. PROMIS-29 scales spanning seven health domains (physical function, fatigue, pain interference, depressive symptoms, anxiety, ability to participate in social roles and activities, and sleep disturbance) were used to generate physical and mental health summary scores 34 , 35 . Higher domain scores indicate a greater level of the construct and can be positive (e.g., physical function) or negative (e.g., fatigue). Quality of life was measured using the interviewer-administered EuroQol EQ-5D-5L measure. The EQ-5D-5L asks respondents to assess five domains of health in the prior week including mobility, self-care, usual activities, pain, and anxiety/depression 46 . The data were summarized as a single index value using the United States EQ-5D-5L value sets 45 , 72 . Participants are also asked to rate their health in the preceding week on a 100-point visual analog scale (VAS) with 100 representing “best imaginable health state” and 0 representing the “worst imaginable health state.” The self-reported Patient Global Impression of Change (PGIC) measure represents a participant’s opinion of the efficacy of an intervention and has been used in the context of pain disorders 49 as well as depression, non-small cell lung cancer, and asthma 73 . Participants were asked to choose from a seven-point scale which phrase best represented their change with regard to activity limitations, symptoms, emotions and overall quality of life, related to Long COVID due to the study intervention. In a similar manner, they were asked the same question with a scale numbered zero to ten with zero representing “much better,” five representing “no change,” and ten representing “much worse.” A free response section was included below to further elaborate on symptom changes, if present. Difference in PGIC was calculated by fitting a proportional odds model with robust Huber–White sandwich variance estimators. The 12-item, interviewer-administered World Health Organization (WHO) Disability Assessment Schedule 2.0 (WHODAS 2.0) is a generic assessment instrument for health and disability covering six domains of function including cognition, mobility, self-care, interactions with others, life activities, and participation in community activities 47 . The simple scoring method according to the WHO manual for administration was utilized for analysis with higher scores signifying greater disability. The Duke Activity Status Index is a 12-item, patient-reported measure of functional capacity, demonstrated to correlate with peak oxygen uptake 36 . It gives a summary score from 0 to 58.2 where higher scores indicate higher functional capacity. Summary scores can be converted into estimates of peak oxygen uptake and maximal metabolic equivalent of tasks (METs). The self-reported Everyday Cognition (ECog) 39-item scale was used to measure daily cognitive functioning 39 . The ECog was modified to assess participants’ current perception of neurocognitive status at the current timepoint compared to before they were ever diagnosed with COVID-19. A total score was calculated by summing the completed items and then dividing by the number of items completed. Higher scores indicate greater cognitive impairment. To utilize the instrument in the context of a clinical trial with frequent administration, we modified the time scale of the questionnaire, specifically assessing the change in participants’ cognitive functioning from before they ever tested positive for COVID-19 to the present day (in contrast to the original measure which compares cognitive functioning 10 years prior to the present day). The COMPASS-31 is a 31-item, patient-reported scale that results in a composite score describing the degree of autonomic symptoms including secretomotor dysfunction, orthostatic intolerance, constipation, diarrhea, and more 41 . Higher composite scores indicate a greater degree of autonomic dysfunction. We modified the COMPASS-31 time scale to assess participants’ autonomic dysfunction over the prior 3 months (instead of 12 months as is included in the standard form) to accommodate the structure of the trial. We used a modified version of the DSQ-PEM to assess PEM 44 . In short, the DSQ-PEM assesses both frequency and severity of five post-exertional malaise symptoms in addition to five supplemental items to determine PEM duration, recovery, and exercise avoidance. Higher scores indicate greater burden of PEM symptoms. In order to use the measure in the context of the trial, we modified the time scale of the questionnaire, asking participants to assess PEM symptoms over the prior 3 months as opposed to the standard six-month time frame. We used standardized questionnaires to inquire about preceding diagnoses and current symptoms related to ME/CFS, hypermobile Ehlers-Danlos Syndrome, and endometriosis at participant screening visits. Participants with a prior diagnosis of ME/CFS preceding and not worsened by their subsequent development of Long COVID were excluded from participation according to pre-specified exclusion criteria. To assess if participants met criteria for ME/CFS after development of Long COVID symptoms, we utilized both the ME/CFS International Consensus Criteria 74 and the Institute of Medicine (IOM) 2015 Diagnostic Criteria 32 and assessed for severity of symptoms using the Bell CFIDS Disability Scale 75 . Hypermobile Ehlers-Danlos Syndrome was evaluated using the 2017 International Classification of the Ehlers-Danlos Syndrome criteria 76 , which includes a “Beighton score” to assess generalized joint hypermobility and includes diagnostic thresholds. To evaluate endometriosis, patients assigned female sex at birth completed the ENDOPAIN 4D Questionnaire 77 which is divided into 21 items across 4 domains; total scores range from 0 to 94 with higher scores indicating greater burden of symptoms. Presence and history of mast cell activation syndrome (MCAS) were evaluated through Coastal Integrative Medicine’s “Mast Cell Activation Questionnaire” 78 , a 30-item assessment evaluating symptoms, medical history, and laboratory measurements; a total score above 8 but less than 14 indicates a pathological activation of mast cells, while a total score of 14 or more indicates a systemic mast cell mediator release syndrome and is considered clinically verified. The Mast Cell Activation Questionnaire was modified for ease of administration in the setting of an outpatient clinical trial. We used CNS Vital Signs 40 , a computer-based neurocognitive testing battery comprised of seven tests: verbal and visual memory, finger tapping, symbol digit coding, the Stroop Test, a test of shifting attention and the continuous performance test. The battery gives a summary neurocognition index (NCI) score averaging five domain scores (Composite Memory, Psychomotor Speed, Reaction Time, Complex Attention, and Cognitive Flexibility) and representing a global score of neurocognition. The six-minute walk test (6MWT) is a simple exercise testing procedure designed to assess physical ability through a six-minute, self-paced walk 37 , 38 . Measures of heart rate, oxygen saturation, blood pressure, and self-reported dyspnea and fatigue are performed before and after the test. Total distance walked in meters at each timepoint is reported. The active standing test is a non-invasive tool to assess orthostatic hypotension (OH) and postural orthostatic tachycardia syndrome (POTS). Procedures were similar to previously reported methods 42 . Blood pressure and heart rate measurements were obtained after 5 min of resting supine and 1, 3, 5, and 10 min of continuous standing. Abnormal active standing test results were defined as those with a decline of >20 mmHg in systolic or >10 mmHg in diastolic blood pressure in at least two consecutive measurements, or those with an increase in heart rate >30 bpm on two consecutive measurements 79 . In addition to safety laboratory tests, C-reactive protein (CRP), erythrocyte sedimentation rate (ESR), fibrinogen, and D-dimer were measured in the clinical laboratory at San Francisco General Hospital. Cytokine levels were measured in plasma using the CorPlex Cytokine Panel (Quanterix). Plasma samples were diluted 4-fold in sample diluent buffer and assays were performed following the CorPlex manufacturer protocols. Each CorPlex Cytokine Panel kit was analyzed by the SP-X Imaging and Analysis System (Quanterix). Anti-SARS-CoV-2 IgG antibodies were measured in plasma samples using the previously described Simoa assays 80 . Antibody levels are presented as normalized average enzymes per bead (AEB), where the measured AEB value is normalized by calibrators obtained from serially diluting a plasma sample collected from a SARS-CoV-2 positive individual. Plasma proteomic measurements were obtained from 1034 unique proteins quantified in singlicate from cryopreserved plasma using the Olink® Reveal platform (Olink® Proteomics, Uppsala, Sweden), a highly sensitive and specific proximity extension assay (PEA)–based technology. 50 For longitudinal analyses, multiple samples from the same donor were placed on the same plate to minimize inter-plate variability. Within each plate, samples were randomized using Olink® Analyze according to the manufacturer’s guidance. After PEA incubation and PCR amplification, DNA libraries were sequenced on an Illumina NovaSeq X Plus (Illumina Inc., San Diego, CA, USA) at the UCSF CAT Core using the Olink_NovaSeqX_10B_8Lib_V3 recipe. Relative protein levels, expressed as NPX values (Normalized Protein eXpression; log₂ scale), were generated using ngs2counts (v5.1.0) and processed with NPX™ Map software (v1.3.0), including plate control normalization. All samples passed quality control, and 100% of data points met QC thresholds. AER002 levels were measured via a validated serum LC-MS/MS method quantifying a unique CDR1-region signature peptide, based on a previously described method 31 . A 50 µL serum aliquot was mixed with 750 µL of water/water saturated with ammonium sulfate (1:1, v/v), vortexed, and incubated at room temperature for ≥15 min. Samples were centrifuged at 15,000 x  g for 15 min at 5 °C, followed by another 15-min incubation at room temperature. The supernatant was discarded, and 100 µL of denaturing buffer (8 M urea in 50 mM ammonium bicarbonate with 0.2 M DTT, 20:1, v/v) was added. Samples were incubated at 60 °C for 1 h, vortexed, then alkylated by adding 20 µL of 500 mM iodoacetamide (final IAA concentration: 40 mM), followed by incubation in the dark at room temperature for 30 min. Excess IAA was quenched by adding 15 µL of 200 mM DTT. Next, 480 µL of 50 mM ammonium bicarbonate digestion buffer was added. Trypsin working solution was prepared by diluting 100 µL of 1 µg/µL trypsin stock in 900 µL buffer, and 50 µL of this solution was added to each sample. Internal standard (IS) solution (60 µL of 5000 ng/mL) was added to all samples except double blanks. Samples were incubated at 37 °C for 6 h, then digestion was quenched by adding 25 µL of water/TFA (9:1, v/v). The samples were centrifuged at 15,000 x  g for 15 min at 5 °C. Supernatants were subjected to solid-phase extraction (SPE) for peptide purification prior to LC–MS/MS analysis. A 10 µL aliquot of each SPE-processed digest was injected into the LC–MS/MS system equipped with a Turbo V electrospray ionization (ESI) source operating in positive ion mode. Quantification of AER002 levels and their stable isotope-labeled internal standard, AER002 [13C6 15N4]FDDYALHWVR, was performed using multiple reaction monitoring (MRM) targeting a unique signature peptide derived from the CDR1 region. The ion source parameters were set as follows: ion spray voltage, 4500 V; source temperature, 500 °C; curtain gas, 30 psi; nebulizer gas (gas 1), 50 psi; heater gas (gas 2), 50 psi; and collision gas, 9 psi. MRM transitions monitored included m/z 441.2 → 530.4 for the unlabeled peptide and m/z 444.5 → 666.5 for the labeled internal standard. Rectosigmoid tissue samples were collected via flexible sigmoidoscopy on an opt-in basis. Up to two collections occurred, once at baseline prior to infusion and another shortly after the 90-day follow-up time point. Tissue samples were fixed in fresh paraformaldehyde (PFA) followed by paraffin embedding approximately 24 h after fixation or were cryopreserved at −180 °C in fetal bovine serum (FBS) and 20% dimethyl sulfoxide (DMSO) according to previously described methods 15 . The manual RNAscope 2.5 HD Duplex assay (Advanced Cell Diagnostics, catalog no. 322430) was used to identify SARS-CoV-2 single-stranded spike RNA [probe-V-nCoV2019-S (catalog no. 848561-C2)] and double-stranded orf1ab RNA [V-nCoV2019-orf1ab-sense (catalog no. 859151)] in situ. Paraffin-embedded tissue blocks were sectioned at 5 μm and mounted onto SuperFrost Plus slides, which were stored in a desiccant chamber prior to staining. Slides were baked in a dry air oven at 60 °C for 1 h, deparaffinized twice in 100% xylene (5 min), and dehydrated twice in 100% ethanol (1 min). To block endogenous peroxidase activity, slides were incubated with hydrogen peroxide for 10 min. Next, heat-induced epitope retrieval was performed for 18 min at 100 °C using Target Retrieval Reagent (ACDBio). Protease digestion was accomplished by treatment with Protease Plus solution (ACDBio) for 30 min at 40 °C. Hybridization was then performed with a duplex cocktail of probe-V-nCoV2019-orf1ab-sense (catalog no. 859151, 1:1 dilution) and probe-V-nCoV2019-S (catalog no. 848561-C2, 1:50 dilution) for 2 hours at 40 °C. Following hybridization, amplifications one through ten, as well as detection of the red and green channels, were performed according to the manufacturer’s original protocol. Following detection of the green channel, slides were counterstained with Gill’s Hematoxylin I (StatLab, catalog no. HXGHE1LT, 1:2 dilution), blued with 0.02% ammonia water (Thermo Scientific Chemicals, catalog no. 1336-21-6), and mounted using VectaMount Permanent Mounting Medium (H-5600-60). Images were captured using the Keyence BZ-X810. FFPE rectal tissues from pre-pandemic participants, as well as FFPE lung tissue from SARS-CoV-2 infected K18-hACE2 mice, were run as negative and positive controls, respectively. Sections from all FFPE blocks were also stained with a human housekeeping gene cocktail (POL2RA, PPIB) to assess block RNA integrity and assay performance. We used the nCounter Host Immune Response RNA expression panel (nanoString) to determine differential expression of 750 human gene transcripts related to viral infection response and inflammation, as well as a custom SARS-CoV-2 RNA panel containing single- and double-stranded genes across the viral genome. Pre-COVID rectosigmoid tissue samples were used to identify false-positive signal and determine thresholds of positivity of SARS-CoV-2 transcript quantitation in the post-COVID samples. nCounter ROSALIND software was used to perform differential gene expression analyses between treatment groups and between baseline and post-infusion timepoints for each study arm. This approach selects the optimal subset of housekeeping probes for normalization using the geNorm algorithm, as implemented in the Bioconductor package. Differential expression between two groups of samples is calculated using the Generalized Linear Model (GLM) for count data assuming a negative binomial distribution and utilizes the raw data and estimates of noise and dispersion across all samples in the experiment to calculate fold-change and P  value for each gene. Prior to DEG analyses, ROSALIND was used to remove genes/probes from differential expression analysis if they are expressed near or at background levels and then to normalize raw data. First, geometric mean of the selected housekeeping genes was calculated for each sample followed by geometric mean calculations across all sample-specific geometric means to generate a global geometric mean. Next, the global geometric mean was divided by the sample-specific geometric mean for each sample to generate a normalization factor; raw counts for each gene were then multiplied by that sample’s normalization factor. Participants were administered intravenously [18 F]F-AraG (~5 mCi) and PET-CT whole-body imaging was carried out for 20 min at approximately 50 min post injection. [18 F]F-AraG was prepared as documented elsewhere 15 . Images were taken from the top of head to mid thighs. Routine urinalysis was performed 7 to 14 days after imaging to ensure proper excretion of [18 F]F-AraG. Standardized uptake values (SUV) in various tissue regions of interest (ROI) from PET-CT data were determined using the OsiriX DICOM viewer software package (Pixmeo). ROI determination was performed in complex structures such as brain sections, heart wall, spleen, and gut wall using two-dimensional isometric ROIs as previously described 15 . For simpler structures such as the spinal cord, bone marrow, and whole lymph nodes, three-dimensional spherical VOIs were used. ROI were placed independently by two individuals blinded to treatment group; ROI determinations were compared across a subcohort to ensure consistency between reviewers. The non-parametric Mann-Whitney U testing (double sided with alpha level of 0.05) was used to compare SUV in cross sectional analyses and the Wilcoxon signed-rank test (two-sided with alpha level of 0.05) was used for paired intra-arm analyses before and after treatment. Strict adjustments for multiple comparisons were not used given the limited sample size as these were hypothesis-generating, exploratory analyses. All PET images and CT images were reviewed by board-certified radiologists for incidental abnormalities that would require clinical follow up. Cardiopulmonary exercise testing (CPET) procedures were performed according to our standard research protocol 81 on an opt-in basis once between screening and infusion, and again shortly after the D90 follow-up time point to avoid biasing or reducing statistical power of the main study analyses. Participants were invited to complete symptom-limited, maximal-effort CPET on an upright cycle ergometer using a metabolic cart (MGC Diagnostics Corporation) to objectively measure exercise capacity by assessing oxygen consumption at peak exercise (peak VO2 relative to body weight, ml/kg/min and percent predicted by the Wasserman equations) and peak power output (Watts). Ramp protocols were determined targeting a 10-min test based on measured maximal voluntary ventilation rounding down to the nearest 5 Watts/minute ramp protocol (minimum 10 Watts/minute); the same protocol was used for both baseline and follow-up tests. CPETs were interpreted blinded to treatment assignment. All randomized participants were included in the main analysis. Linear mixed effects models were used for the primary and secondary patient reported outcomes, with an indicator for randomization group, visit and a visit-by-randomization group interaction. Lab measurement outcomes were log transformed. The visit variable was categorical and corresponded to study visits post-infusion. Models were adjusted for baseline scores and included an unstructured covariance structure. If an outcome was measured at both a screening and baseline visit, an average baseline score was calculated and incorporated in the baseline adjusted models. Baseline adjusted mean scores were obtained and reported for each time point by margining the linear mixed effect models 82 . Baseline adjusted geometric means and percent differences are reported for log-transformed laboratory outcomes. Differences in PGIC between groups were assessed using a proportional odds model with robust Huber–White sandwich variance estimators. A per protocol analysis was conducted excluding participants who had a confirmed or suspected reinfection before the primary endpoint. All statistical analyses were carried out in R (version 4.4.1) 83 . Additional information regarding the statistical analyses can be found in the Statistical Analysis Plan. Responders were defined as individuals reporting PGIC improvements at D90 and/or D180 (the blinded period) of “Moderately Better” or greater. Baseline antibody levels were compared between AER002 and placebo groups, as well as responders and non-responders using the Wilcoxon rank-sum test. Plasma proteomics data were analyzed using linear mixed-effects models fit separately for each protein, with fixed effects for randomization group, visit, and a visit-by-randomization group interaction. A random intercept for each participant was included to account for within-participant correlation. The interaction term was used to assess evidence for differential protein expression. The estimates for the interaction term were used to generate volcano plots, and P values from the models were corrected for multiple testing using Storey’s q-value method 51 . Exploratory gene set enrichment analyses were performed using the clusterProfiler (version 4.16.0) 84 package. Proteins were mapped to Entrez Gene IDs, ranked by fold changes and enrichment was tested against the Gene Ontology Biological Process 85 and Hallmark gene sets 52 . AER002 PK was characterized using noncompartmental and population pharmacokinetic approaches using R (version 4.5) and NONMEM (ICON, plc, version 7.6), respectively. All lower limit of quantitation (LLOQ) values (i.e., <20 mcg/ml) were set to zero. One-, two-, and three-compartment models with first-order elimination were evaluated. Between participant variability was explored on model parameters, alongside assessment of additive, proportional, and combined residual error models. The final model was selected based on goodness-of-fit diagnostics, visual predictive checks, bootstrap estimates, and comparison of objective function values for nested models (-2LL and AIC criterion). Covariate analysis to evaluate impacts of sex, age, and race was carried out using stepwise covariate modeling 86 . As part of a post-hoc exploratory analysis assessing how drug exposure and baseline antibody levels affected response, multiple logistic regression was performed using AER002 AUC (area under the concentration-time curve from noncompartmental analysis) and baseline anti-SARS-CoV-2 levels (high [≥median] vs low [<median] anti-S, anti-S1, anti-RBD, anti-N levels separately) as predictors of whether an individual self-identified as a responder. Further information on research design is available in the  Nature Portfolio Reporting Summary linked to this article.

From July 2023 to April 2024, a total of 460 individuals expressed interest in the trial. Forty-three consented to participation and underwent eligibility assessment and 36 were enrolled and randomized to receive AER002 versus placebo in a 2:1 ratio (Fig.  1 ). Median age was 42 years (interquartile range [IQR]: 32–53); 56% of participants were female; 75% were White, 14% Asian, 8% Hispanic/Latino, and 3% Black (Table  1 ). Pre-existing medical comorbidities included lung problems (asthma, COPD, emphysema, or bronchitis; 17%), autoimmune disease (14%), hypertension (11%), diabetes (2.8%), and HIV (2.8%). Participants with active myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS) preceding their SARS-CoV-2 infection were excluded. Prior to infusion, 5.6% of participants had received a mast cell activation syndrome (MCAS) diagnosis and 8.3% met criteria for hypermobile Ehlers-Danlos Syndrome. Participants of female sex were also evaluated for pain symptoms related to endometriosis (5.6% score ≥60). Fig. 1 CONSORT diagram for the clinical trial. IVIG intravenous immunoglobulin, ITT intention-to-treat, mITT modified intention-to-treat. Per protocol analysis excludes individuals with confirmed or suspected on-study reinfections before the primary endpoint at D90 (AER002 [ n  = 3]; placebo [ n  = 1]). Created in BioRender. Peluso, M. (2026) https://BioRender.com/klsh4ic . Table 1 Demographics, COVID-19 history, and Long COVID history of the study cohort Characteristic Overall N  = 36 a AER002 N  = 24 a Placebo N  = 12 a Age at enrollment (years) 42 (32–53) 42 (33–49) 43 (30–59) Sex at birth  Female 20 (56%) 13 (54%) 7 (58%)  Male 16 (44%) 11 (46%) 5 (42%) Gender  Female 19 (53%) 12 (50%) 7 (58%)  Male 16 (44%) 11 (46%) 5 (42%)     Non-binary 1 (2.8%) 1 (4.2%) 0 (0%) Race/Ethnicity  Asian 5 (14%) 3 (13%) 2 (17%)  Black/African American 1 (2.8%) 0 (0%) 1 (8.3%)  Hispanic/Latino 3 (8.3%) 2 (8.3%) 1 (8.3%)  White 27 (75%) 19 (79%) 8 (67%) Medical comorbidities  Autoimmune disease 5 (14%) 2 (8.3%) 3 (25%)  Cancer b 0 (0%) 0 (0%) 0 (0%)  Diabetes 1 (2.8%) 1 (4.2%) 0 (0%)  Heart attack or heart failure 0 (0%) 0 (0%) 0 (0%)  HIV Infection 1 (2.8%) 0 (0%) 1 (8.3%)  Hypertension 4 (11%) 3 (13%) 1 (8.3%)  Lung problems c 6 (17%) 4 (17%) 2 (17%)  Kidney disease 0 (0%) 0 (0%) 0 (0%) Body mass index (kg/m 2 )  18.0–18.5 1 (2.8%) 1 (4.2%) 0 (0%)  18.5–24.9 21 (58%) 14 (58%) 7 (58%)  25.0–29.9 7 (19%) 4 (17%) 3 (25%)  >=30 7 (19%) 5 (21%) 2 (17%) Active ME/CFS preceding SARS-CoV-2 infection 0 (0%) 0 (0%) 0 (0%) Mast Cell Activation Syndrome 2 (5.6%) 0 (0%) 2 (17%) Beighton score d  0 27 (75%) 21 (88%) 6 (50%)  2–4.9 6 (17%) 1 (4.2%) 5 (41.7%)  >=5 3 (8.3%) 2 (8.3%) 1 (8.3%) ENDOPAIN-4D global score e  0 12 (33%) 6 (25%) 6 (50%)  0.1–29.9 4 (11.1%) 3 (13%) 1 (8.3%)  30–59.9 2 (5.6%) 2 (8.3%) 0 (0%)  >=60 2 (5.6%) 2 (8.3%) 0 (0%) Number of prior known SARS-CoV-2 infections at time of enrollment  1 19 (53%) 13 (54%) 6 (50%)  2 14 (39%) 9 (38%) 5 (42%)  3 3 (8.3%) 2 (8.3%) 1 (8.3%) Index SARS-CoV-2 infection date  March 2020–April 2021 12 (33%) 8 (33%) 4 (33%)  May 2021–November 2021 3 (8.3%) 2 (8.3%) 1 (8.3%)  December 2021–August 2022 21 (58%) 14 (58%) 7 (58%) Time since infection to which Long COVID is attributed (days) 740 (576–1113) 740 (571–1113) 769 (593–1114) Hospitalized for COVID-19 2 (5.6%) 2 (8.3%) 0 (0%) Time since most recent SARS-CoV-2 infection (days) 504 (344–706) 509 (353–709) 504 (280–662) Reported a SARS-CoV-2 re-infection after August 15, 2022 12 (33%) 8 (33%) 4 (33%) Completed primary SARS-CoV-2 vaccination series 36 (100%) 24 (100%) 12 (100%) SARS-CoV-2 infection preceding first SARS-CoV-2 vaccination 11 (31%) 7 (29%) 4 (33%) Time since most recent SARS-CoV-2 vaccination (days) 422 (138–700) 456 (190–700) 327 (99–673) Long COVID symptom count (last 30 days) 14.0 (7.5–18.5) 12.0 (6.0–16.5) 14.5 (11.0–20.0) Symptoms reported (last 30 days) f  Constitutional 33 (92%) 22 (92%) 11 (92%)  Upper respiratory 14 (39%) 9 (38%) 5 (42%)  Cardiopulmonary 29 (81%) 19 (79%) 10 (83%)  Gastrointestinal 22 (61%) 13 (54%) 9 (75%)  Genitourinary 6 (17%) 5 (21%) 1 (8.3%)  Dermatologic 6 (17%) 3 (13%) 3 (25%)  Musculoskeletal 24 (67%) 16 (67%) 8 (67%)  Neurologic 36 (100%) 24 (100%) 12 (100%)  Trouble sleeping 27 (75%) 17 (71%) 10 (83%)  Post Exertional Malaise 31 (86%) 20 (83%) 11 (92%) Met 2015 IOM case definition for ME/CFS g 21 (58%) 14 (58%) 7 (58%) Quality of Life by Visual Analogue Scale at baseline 49 (35–63) 50 (35–64) 44 (35–55) a n (%) Median (IQR). b Cancer requiring treatment within the 2 years before COVID-19. c Asthma, COPD, emphysema, or bronchitis experienced in the 5 years before COVID-19. d A score of 5+ indicates a positive indication for Hypermobile Ehlers-Danlos Syndrome. e Among those with female sex at birth. f Constitutional: fatigue, fever, chills. Upper respiratory: rhinorrhea, sore throat. Cardiopulmonary: cough, shortness of breath, heart palpitations, chest pain, fainting spells. Gastrointestinal: diarrhea, nausea, loss of appetite, abdominal pain, vomiting, constipation . Genitourinary: menstrual cramps, dyspareunia . Dermatologic: rash . Musculoskeletal: back pain, myalgia, joint pain . Neurologic: anosmia, dysgeusia, concentration problems, dizziness, balance problems, neuropathy, vision problems, parosmia. g Participant reports fatigue, post-exertional malaise, and unrefreshing sleep with cognitive impairment and/or dizziness. IVIG intravenous immunoglobulin, ITT intention-to-treat, mITT modified intention-to-treat. Per protocol analysis excludes individuals with confirmed or suspected on-study reinfections before the primary endpoint at D90 (AER002 [ n  = 3]; placebo [ n  = 1]). Created in BioRender. Peluso, M. (2026) https://BioRender.com/klsh4ic . Demographics, COVID-19 history, and Long COVID history of the study cohort Number of prior known SARS-CoV-2 infections at time of enrollment a n (%) Median (IQR). b Cancer requiring treatment within the 2 years before COVID-19. c Asthma, COPD, emphysema, or bronchitis experienced in the 5 years before COVID-19. d A score of 5+ indicates a positive indication for Hypermobile Ehlers-Danlos Syndrome. e Among those with female sex at birth. f Constitutional: fatigue, fever, chills. Upper respiratory: rhinorrhea, sore throat. Cardiopulmonary: cough, shortness of breath, heart palpitations, chest pain, fainting spells. Gastrointestinal: diarrhea, nausea, loss of appetite, abdominal pain, vomiting, constipation . Genitourinary: menstrual cramps, dyspareunia . Dermatologic: rash . Musculoskeletal: back pain, myalgia, joint pain . Neurologic: anosmia, dysgeusia, concentration problems, dizziness, balance problems, neuropathy, vision problems, parosmia. g Participant reports fatigue, post-exertional malaise, and unrefreshing sleep with cognitive impairment and/or dizziness. Most participants (53%) reported a single SARS-CoV-2 infection prior to the study (Table  1 ); the SARS-CoV-2 infection to which Long COVID symptoms were attributed occurred a median of 740 days (IQR: 576–1113) prior to enrollment; few participants (5.6%) had been previously hospitalized for COVID-19. The most recent infection was a median of 504 days (IQR 344–706) prior to enrollment. Notably, 33% of participants had experienced a SARS-CoV-2 reinfection after August 15, 2022, when AER002 was expected to have lost neutralizing capacity. All participants had received at least one SARS-CoV-2 vaccine prior to participation in the study; in 31% of cases, the initial SARS-CoV-2 infection had preceded vaccination. The cohort was highly symptomatic. In the 30 days preceding enrollment, participants reported a median of 14 Long COVID symptoms (IQR: 7.5–18.5). The most common symptoms were constitutional (e.g., fatigue; 92%), neurologic (e.g., “brain fog”, headache; 100%), and cardiopulmonary (e.g., exercise intolerance; 81%); 31 (86%) participants reported post-exertional malaise (PEM; “Symptoms that get worse more frequently than usual after physical or mental effort, such as fatigue, pain, or swollen glands”) at baseline and 21 (58%) met the 2015 Institute of Medicine criteria for ME/CFS 32 . At baseline, the median quality of life on a 100-point visual analogue scale was 49 (IQR 35–63), reflecting the substantial impact of Long COVID on functional status and quality of life. Groups were similar in patient-reported measures of general health (PROMIS-29 33 – 35 ), as well as patient-reported and objective measures of cardiopulmonary function (Duke Activity Status Index [DASI 36 ] and 6-minute walk test 37 , 38 ), neurocognitive function (Everyday Cognition Scale [ECog 39 ] and CNS-Vital Signs 40 neurocognitive battery), and autonomic function (Composite Autonomic Symptom Score [COMPASS-31 41 ] and active stand test 42 ) (Supplementary Table  1 ). Of note, at baseline, participants reported a relatively low overall burden of autonomic symptoms and only one met criteria for autonomic dysfunction by active stand test. Clinical laboratory measures including C-reactive protein (CRP), erythrocyte sedimentation rate (ESR), D-Dimer, and fibrinogen were also similar between groups at baseline (Supplementary Table  1 ). Despite randomization and consistent with chance imbalance in this small study, baseline SARS-CoV-2 antibodies measured by single-molecule array 43 demonstrated that anti-spike (anti-S), anti-spike S1 (anti-S1), and anti-receptor binding domain (anti-RBD) antibody levels were higher in the placebo versus AER002 arms (anti-S [median (IQR): placebo 124 (84.1–165) vs AER002 62.3 (51.2–111)] p  = 0.026, anti-S1 [median (IQR): 224 (144–311) vs 136 (56.4–191)] p  = 0.032, and anti-RBD [median (IQR): 170 (43.1–233) vs 82.6 (30.8–126)] p  = 0.072), while anti-nucleocapsid (anti-N, [median (IQR): 3.50 (1.96–7.33) vs 2.28 (1.79–4.01)]) antibody levels were not different ( p  = 0.54) (Supplementary Fig.  1 ). All participants completed the trial. All participants remained blinded to treatment allocation until every participant had completed the 180-day timepoint; participants were then followed in an open-label manner for safety until the 360-day timepoint. There were 12 confirmed or suspected COVID-19 reinfections during the study period. Notably, 4 reinfections (3 confirmed, 1 suspected) occurred between the infusion and 90-day timepoints, 3 reinfections (2 confirmed, 1 suspected) occurred between the 90-day and 180-day timepoints, and 5 reinfections (5 confirmed, 0 suspected) occurred between the 180-day and 360-day timepoints. AER002 was safe and well-tolerated. A total of 112 adverse events (AEs) among 32 participants were recorded during the study period (Supplementary Tables  2 and 3 ). The most common adverse events were confirmed acute COVID-19 (9%), COVID-19-negative upper respiratory infection (8%), exacerbation of post-exertional malaise (PEM; 6.3%), and diarrhea (3.6%). One grade 3 AE (a planned hospital admission for management of intractable chronic migraine with aura) occurred during the study but was not related. No clinically significant laboratory abnormalities occurred. Based on input from people with lived experience of Long COVID and because of the intensity of the study protocol, we paid particular attention to PEM. Using a modified version of the DePaul Symptom Questionnaire for PEM 44 , 29 participants met criteria for PEM prior to the infusion; of those who had baseline PEM, 76% (22/29) reported the presence of PEM in the two days following infusion and 24% (7/29) reported no PEM in the same timeframe. Of 7 without baseline PEM, none reported the emergence of post-infusion PEM in the two days following the infusion. Additionally, we worked to provide accommodations for PEM symptoms during study participation, which ranged from small adjustments during in-person visits (dimmed lights, frequent breaks, access to examination beds for rest) to administration of questionnaires remotely for those too unwell to complete them in person. In comparison to those receiving placebo, we did not observe a significant improvement in functional status or overall quality of life in those receiving AER002 (Table  2 ). The PROMIS-29 Physical Health Summary Score 33 – 35 improved in both groups between baseline and the 90-day timepoint, with no significant difference between groups (Fig.  2a ). Similarly, the PROMIS-29 Mental Health Summary Score (MHSS) secondary outcome improved in both groups, with no significant difference between groups (Fig.  2b ). There was also no significant difference in quality of life using the 5-Item EuroQol EQ-5D-5L Index Value Score 45 , 46 (Fig.  2c ), the 100-point Visual-Analogue Scale (Fig.  2d ), or the WHODAS 2.0 Simple Score 47 (Fig.  2e ). Similar patterns in these measurements were observed at secondary timepoints (30 days, 180 days, 360 days) (Table  2 ; Supplementary Table  4 ). Fig. 2 Changes in overall health following treatment with AER002 ( n  = 24) versus placebo ( n  = 12). a PROMIS-29 Physical Health Summary Score. b PROMIS-29 Mental Health Summary Score. c EQ-5D-5L Quality of Life Index. d Quality of Life by Visual Analog Scale. e World Health Organization Disability Score (WHODAS) 2.0. f Patient Global Impression of Change (PGIC) within treatment and placebo arms. Raw means are plotted for baseline and estimated marginal means are plotted for subsequent timepoints. Error bars indicate 95% CI. Where applicable, dashed horizontal lines indicate mean score values for the general United States population. Source data are provided as a Source Data file. Table 2 Linear mixed effect model results for continuous patient reported outcomes, performance measures and laboratory measurements through Day 180 Patient Reported Outcome Day AER002 a Placebo a Mean difference 95% CI P  value PROMIS-29 PHSS (T-score) 30 40.7 41.7 -1.0 (−5.5, 3.5) 0.66 90 40.9 45.0 -4.1 (−8.6, 0.47) 0.08 180 41.7 44.4 -2.7 (−7.2, 1.8) 0.24 PROMIS-29 MHSS (T-score) 30 40.3 42.7 -2.4 (−7.3, 2.5) 0.34 90 40.7 45.4 -4.7 (−9.6, 0.2) 0.06 180 41.3 43.8 -2.6 (−7.5, 2.3) 0.30 QoL 100-point VAS 90 53.2 62.6 -9.4 (−19.6, 0.8) 0.07 QoL EQ-5D-5L Index Score 90 0.51 0.66 -0.15 (−0.3, 0.01) 0.07 DASI 90 35.6 40.2 -4.6 (−11.8, 2.7) 0.21 COMPASS-31 90 18.3 13.6 4.7 (−3.9, 13.4) 0.27 WHO-DAS 2.0 90 16.9 12.9 4.0 (−0.06, 8.0) 0.053 ECog-39 90 2.09 1.93 0.16 (−0.2, 0.5) 0.35 6MWT (meters) 90 426 457 -31.3 (−76.0, 13.4) 0.17 CNS-VS NCI 90 99.4 100.8 -1.4 (−6.0, 3.01) 0.51 Lab measurements Day AER002 b Placebo b Percent difference 95% CI P  value CRP (mg/mL) 90 0.81 0.71 13.7 (−41.8, 121.8) 0.70 ESR (mm/hr) 90 8.98 9.48 -5.2 (−38.1, 45.0) 0.80 D-Dimer (mg/L FEU) 90 0.29 0.31 -4.6 (−13.6, 5.3) 0.33 Fibrinogen (mg/dL) 90 309 294 5.0 (−4.9, 16.0) 0.33 a Estimated marginal means. b Estimated marginal geometric means. a PROMIS-29 Physical Health Summary Score. b PROMIS-29 Mental Health Summary Score. c EQ-5D-5L Quality of Life Index. d Quality of Life by Visual Analog Scale. e World Health Organization Disability Score (WHODAS) 2.0. f Patient Global Impression of Change (PGIC) within treatment and placebo arms. Raw means are plotted for baseline and estimated marginal means are plotted for subsequent timepoints. Error bars indicate 95% CI. Where applicable, dashed horizontal lines indicate mean score values for the general United States population. Source data are provided as a Source Data file. Linear mixed effect model results for continuous patient reported outcomes, performance measures and laboratory measurements through Day 180 a Estimated marginal means. b Estimated marginal geometric means. Because of concerns that commonly used patient-reported outcome measurements may not adequately capture symptomatic changes in this patient population 48 , we used the Patient Global Impression of Change (PGIC) scale 49 to directly query participants regarding whether they perceived any clinical changes following treatment and to identify self-reported responders and non-responders. Approximately half of the participants in the treatment arm and three-quarters in the placebo arm reported a perceived response to treatment (odds ratio 0.26 [95% CI, 0.06−1.12; p  = 0.07]; Fig.  2f ). To address symptom variability across participants, we assessed several domains using both patient-reported outcomes (PROs) and objective measures. Measures of post-exertional symptoms were captured throughout the trial. We did not observe a consistent pattern related to post-exertional symptoms exacerbated by the study; there was no difference in the odds of having consistent or resolving PEM between the treatment and placebo groups (Supplementary Fig.  2 ). Among participants with PEM at baseline, five of 19 participants in the AER002 group and 4 of 10 participants in the placebo group reported improvement in PEM at Day 90 ( p  = 0.45). Cardiopulmonary symptoms and performance were also assessed between groups. At Day 90, there were no significant differences in either DASI (Fig.  3a ) or 6-min walk test (Fig.  3b ). Results were similar at secondary timepoints (30 days, 180 days, 360 days). Fig. 3 Changes in patient-reported and objective markers of physical activity, neurocognitive function, and autonomic function following treatment with AER002 ( n  = 24) versus placebo ( n  = 12). a Duke Activity Status Index score (DASI). b 6-minute Walk Test (6MWT) total distance traveled. c Everyday Cognition-39 (ECog-39) Score. d CNS-Vital Signs (CNS-VS) Neurocognitive Index (NCI) Score. e Composite Autonomic Symptom (COMPASS-31) Score. Raw means are plotted for baseline and estimated marginal means are plotted for subsequent timepoints. Error bars indicate 95% CI. Source data are provided as a Source Data file. a Duke Activity Status Index score (DASI). b 6-minute Walk Test (6MWT) total distance traveled. c Everyday Cognition-39 (ECog-39) Score. d CNS-Vital Signs (CNS-VS) Neurocognitive Index (NCI) Score. e Composite Autonomic Symptom (COMPASS-31) Score. Raw means are plotted for baseline and estimated marginal means are plotted for subsequent timepoints. Error bars indicate 95% CI. Source data are provided as a Source Data file. We additionally investigated neurocognitive symptom burden and functioning between AER002 and placebo groups. At Day 90 following the infusion, there were no significant differences in either ECog (Fig.  3c ) or CNS-Vital Signs neurocognitive index score (Fig.  3d ). Results were similar at secondary timepoints (30 days, 180 days, 360 days). Patient-reported and objective measures of autonomic dysfunction were also assessed. At Day 90 following the infusion, there were no significant differences in either COMPASS-31 (Fig.  3e ) or active standing test (one positive). Results were consistent at secondary timepoints (30 days, 180 days, 360 days). We did not observe meaningful changes within groups or differences between groups in systolic blood pressure, diastolic blood pressure, or change in heart rate over the course of the study. We did not identify statistically significant or clinically meaningful differences in targeted blood-based measurements of inflammation (C-reactive protein, erythrocyte sedimentation rate), immune activation (IL-6, TNF-alpha, interferon-gamma, IL-1beta), or clotting (D-dimer, fibrinogen) between groups at timepoints following the intervention (Fig.  4a–h ). Fig. 4 Changes in laboratory-based biomarkers following treatment with AER002 ( n  = 24) versus placebo ( n  = 12). a C-reactive protein (CRP). b D-Dimer. c Fibrinogen. d Erythrocyte sedimentation rate (ESR). e Interleukin-6 (IL-6). f Tumor necrosis factor-alpha (TNF-α). g Interleukin 1-beta (IL-1β). h Interferon-gamma (IFN- γ). Raw means are plotted for baseline and estimated marginal means are plotted for subsequent timepoints. Error bars indicate 95% CI. Source data are provided as a Source Data file. a C-reactive protein (CRP). b D-Dimer. c Fibrinogen. d Erythrocyte sedimentation rate (ESR). e Interleukin-6 (IL-6). f Tumor necrosis factor-alpha (TNF-α). g Interleukin 1-beta (IL-1β). h Interferon-gamma (IFN- γ). Raw means are plotted for baseline and estimated marginal means are plotted for subsequent timepoints. Error bars indicate 95% CI. Source data are provided as a Source Data file. AER002 pharmacokinetics (PK) parameters were investigated throughout the study. AER002 serum concentration was measured via a validated LC-MS/MS method using a unique CDR1-region signature peptide 31 . AER002 PK was characterized using noncompartmental and population pharmacokinetic approaches. A two-compartment population pharmacokinetic model (Supplementary Table  5 ) with between-participant variability for clearance and central volume of distribution was constructed using 114 total serum concentrations from 24 participants obtained over 6 months. Participants of female sex showed a nonsignificant trend toward faster AER002 clearance (terminal half-life 75.5 [median (IQR): 85.4 (65.6−110.3)] vs. 107.8 [median (IQR): 107.1 (81.1−140.1)] days, p  = 0.28). In population PK analyses, male participants were estimated to have 1.96-fold (95% CI: 1.27−2.88) higher peripheral distribution volume than female participants. Plasma proteomic profiling was conducted at multiple timepoints to investigate proteomic signatures among participant groups. Proteomic profiling was performed using the Olink® Reveal platform (Olink® Proteomics, Uppsala, Sweden), a proximity extension assay (PEA)-based technology 50 . A total of 1034 unique proteins were quantified from cryopreserved plasma samples collected at baseline and on the day of infusion prior to administration of study drug (averaged to represent baseline values), and at 14 and 90 days post-infusion 50 . Proteins with unadjusted p  < 0.05 and beta estimate ±0.3 (~1.25-fold change) were considered potentially clinically meaningful, and those that also met significance after multiple-testing correction via Storey’s q-value method 51 . Baseline proteomic profiles were largely comparable between treatment groups, with only minor differences observed for a small number of proteins prior to intervention (Supplementary Fig.  3 ). At Day 14 (Fig.  5a ), treatment was associated with modest proteomic differences relative to placebo, affecting a small subset of markers linked to immune regulatory signaling and tissue remodeling processes (BTLA, SPIK6, MYL3). At Day 90 (Fig.  5b ), we observed a broader pattern of proteomic differences; a larger number of markers showed differences in unadjusted analyses, with the majority demonstrating negative beta estimates indicative of lower levels in the AER002 group compared to placebo. These proteins included multiple markers involved in immune activation, inflammatory signaling, and stress responses pathways, including proteins associated with interferon responses and innate immune responses (GBP2, IGBP1), apoptosis and regulation of cell death (CASP7, BAX, CSDMD, DFFA), and intracellular signaling pathways such as MAP kinase and NF-kappa B signaling (MAP2K1, NFKBIE, IRAK4, AKT1S1). Only a small number of proteins demonstrated increased levels with AER002 (PTPN14, a regulator of cell signaling). However, Molecular Signatures Database (MSigDB) Hallmark gene set collection analyses 52 did not reveal obvious immune or inflammatory pathway trends between groups, likely due to limited sample size. Fig. 5 Plasma proteomics following treatment with AER002 ( n  = 24) versus placebo ( n  = 12). Volcano plots showing differential plasma protein expression for a AER002 vs placebo at Day 14 (D14) and b AER002 vs placebo at Day 90 (D90). Differential expression was assessed using protein-specific linear mixed-effects models. All tests were two-sided, and P values were adjusted for multiple comparisons using Storey’s q-value method. Significant proteins with q < 0.05 are shown in bold red. The horizontal dashed line indicates the unadjusted P value threshold of 0.05. Vertical dashed lines indicate beta estimates of −0.3 and 0.3. Source data are provided as a Source Data file. Volcano plots showing differential plasma protein expression for a AER002 vs placebo at Day 14 (D14) and b AER002 vs placebo at Day 90 (D90). Differential expression was assessed using protein-specific linear mixed-effects models. All tests were two-sided, and P values were adjusted for multiple comparisons using Storey’s q-value method. Significant proteins with q < 0.05 are shown in bold red. The horizontal dashed line indicates the unadjusted P value threshold of 0.05. Vertical dashed lines indicate beta estimates of −0.3 and 0.3. Source data are provided as a Source Data file. We conducted a per-protocol analysis in which we excluded individuals with confirmed and suspected on-study reinfections before the primary endpoint. The results were largely unchanged. There were no significant P  values for all measures reported in detail above including physical health and quality of life outcomes, overall impressions of change, and organ-system specific symptoms and performance measures (Supplementary Table  6 ). As part of the co-enrolled observational cohort study, we assessed patient ascertainment of treatment assignment shortly before unblinding by asking participants to predict their group (AER002 vs placebo) and rate confidence from 1 (no confidence) to 5 (complete confidence). In the AER002 group, 42% predicted their treatment assignment correctly, compared to 50% in the placebo group. For both groups, most rated their confidence as moderate to low (≤3) on the confidence scale. All participants were also offered optional procedures: flexible sigmoidoscopy with colorectal sampling, PET-CT imaging, and cardiopulmonary exercise testing (CPET). These were consented to separately and completed under independent protocols at baseline and approximately three months post-infusion. To avoid biasing primary and secondary outcome assessments, post-infusion optional procedures were scheduled on a separate day following the Day 90 visit. We and others previously demonstrated that gastrointestinal tissue can harbor persistent SARS-CoV-2 single- and double-stranded RNA 6 , 15 , 16 . In a subset of participants ( n  = 17; 12 AER002, 5 placebo), we performed flexible sigmoidoscopy with colorectal sampling a median of 10 days (range 1–29) prior to infusion and again a median of 98 days (range 85–195) following infusion. An additional participant (who received AER002) completed the initial collection but declined repeat sample collection. We used multiplexed chromogenic RNAscope to identify SARS-CoV-2 single-stranded spike RNA and double-stranded orf1ab RNA in formalin-fixed paraffin embedded tissues, with reference to control tissue collected prior to the pandemic (Fig.  6a ). Prior to infusion, we detected single-stranded spike RNA in one participant (718 days from most recent known infection; Fig.  6b ). This participant received AER002 and reported significant improvement over the course of the study; no RNA was detected on their post-infusion biopsy. Following infusion, we detected single- and double-stranded spike and ORF1ab RNA in three participants (all received AER002) who did not have detectable RNA at baseline sampling (Fig.  6c ). Of note, all three had known reinfections following the infusion (biopsies were a median of 21 days from most recent infection); two additional participants (both received placebo) with intervening reinfections did not have signal detected on follow-up biopsies 18 and 57 days after reinfection (Fig.  6c and Supplementary Table  7 ). No definitive SARS-CoV-2 RNA was observed in RNA from bulk colorectal tissue by qPCR (spike, nucleocapsid). Using a custom nCounter assay measuring over 750 human transcripts and nine SARS-CoV-2 transcripts in colorectal tissue, we found no differences in viral-immune-response gene expression between arms (AER002 n  = 9, placebo n  = 2) or between baseline and post-infusion timepoints, though sample sizes were limited by reinfection-related exclusions and specimen quality. Fig. 6 Results from optional colorectal biopsies. a Control (pre-pandemic) colorectal tissue. b Detection of SARS-CoV-2 RNA in one participant at baseline, without detectable RNA at follow-up after receiving AER002. c Detection of SARS-CoV-2 RNA at follow-up in three participants with known re-infections following AER002 infusion and no detection of SARS-CoV-2 RNA at follow-up in one participant with known re-infection following placebo infusion. SS indicates single stranded RNA, DS indicates double stranded RNA. Experiments were performed on multiple adjacent tissue sections from each participant; because the assay consumes the tissue section, exact independent repetition on the same section was not possible. a Control (pre-pandemic) colorectal tissue. b Detection of SARS-CoV-2 RNA in one participant at baseline, without detectable RNA at follow-up after receiving AER002. c Detection of SARS-CoV-2 RNA at follow-up in three participants with known re-infections following AER002 infusion and no detection of SARS-CoV-2 RNA at follow-up in one participant with known re-infection following placebo infusion. SS indicates single stranded RNA, DS indicates double stranded RNA. Experiments were performed on multiple adjacent tissue sections from each participant; because the assay consumes the tissue section, exact independent repetition on the same section was not possible. In addition to flexible sigmoidoscopies, participants could also opt-in to PET-CT imaging before and after infusion. We previously identified differences in whole-body T cell activation states in people with and without prior COVID-19 by non-invasive PET-CT imaging using a novel radiotracer ([18 F]F-AraG) 15 . [18 F]F-AraG selectively accumulates in the mitochondria of activated CD8+ and CD4 + T cells, with substantially lower uptake in myeloid cells, providing a specific marker of T cell activation. Here, we performed [18 F]F-AraG PET imaging in a subset of participants a median of 10 days (range 4−151) prior and 98 days (range 83−179) following study drug (AER002; n  = 6) or placebo ( n  = 4) and compared mean standardized uptake values (SUVmean) across tissue regions of interest (ROI) that we identified in our prior observational study. In cross-sectional analyses prior to infusion, we observed significantly lower cauda equina (lumbar spinal cord) uptake in those who received AER002 compared to placebo but no other significant differences across other ROIs (Fig.  7a ). Following treatment, however, we detected significantly lower tracer uptake in those who received AER002 compared to placebo in several tissues, including temporal brain lobes, thoracic spinal cord, cauda equina, and submandibular glands (all p  < 0.05; Fig.  7a ). Given sample-size limitations and inter-participant variability in tracer uptake, we observed no significant longitudinal differences within either arm, though submandibular uptake trended downward in the AER002 arm and upward in the placebo arm (Fig.  7b, c ). Fig. 7 Results from optional [18 F]F-AraG PET-CT imaging. a Baseline and post-infusion mean standardized uptake values (SUV) in selected regions of interest (ROI) among AER002 ( n  = 6) versus placebo ( n  = 4) recipients. b Within-participant changes in SUVmean in the submandibular glands baseline and post-infusion. c Representative imaging of the submandibular glands from participants in each arm. Bars represent mean SUV and error bars represent 95% confidence interval. Adjusted P values are from two-sided non-parametric Mann-Whitney U testing in cross sectional analyses and two-sided Wilcoxon signed-rank test for paired intra-arm analyses before and after treatment. Alpha <0.05 was considered significant. Strict adjustments for multiple comparisons were not used given limited sample size. All data points are shown. Txt treatment. Source data are provided as a Source Data file. a Baseline and post-infusion mean standardized uptake values (SUV) in selected regions of interest (ROI) among AER002 ( n  = 6) versus placebo ( n  = 4) recipients. b Within-participant changes in SUVmean in the submandibular glands baseline and post-infusion. c Representative imaging of the submandibular glands from participants in each arm. Bars represent mean SUV and error bars represent 95% confidence interval. Adjusted P values are from two-sided non-parametric Mann-Whitney U testing in cross sectional analyses and two-sided Wilcoxon signed-rank test for paired intra-arm analyses before and after treatment. Alpha <0.05 was considered significant. Strict adjustments for multiple comparisons were not used given limited sample size. All data points are shown. Txt treatment. Source data are provided as a Source Data file. Finally, a subset of participants completed opt-in cardiopulmonary exercise testing (CPET) at baseline ( n  = 9) and 3 months ( n  = 5) to evaluate exercise capacity. At baseline, the mean peak oxygen consumption was 25.3 ml/kg/min, or 88% predicted by the Wasserman equations, at a mean power output of 149 Watts. Among those who received AER002 ( n  = 4), all tests were to maximal effort (respiratory exchange ratio >1.16). There were no significant changes in peak oxygen consumption (−0.58 ml/kg/min difference (slightly worse), p  = 0.60), peak oxygen consumption on the percent predicted scale (0.5% increase, p  = 0.92), or peak work (2.25 Watts increase, p  = 0.53). To better understand the potential biological effects of AER002 in people with Long COVID, we conducted pre-specified exploratory analyses of self-identified responders in both treatment arms. We defined “responders” as those who reported an improvement on the PGIC scale of “Moderately Better” or more at the 90-day and/or 180-day timepoints. Twelve participants (6/24 AER002, 6/12 placebo) met criteria at both 90 and 180 days. Four participants (2/24 AER002, 2/12 placebo) met criteria at 90 days only while three participants (2/24 AER002, 1/12 placebo) met criteria at 180 days only (Fig.  2f ). The impact of inflammation, immune activation, and clotting markers on self-reported treatment response was assessed. Baseline fibrinogen levels were higher in responders compared to non-responders in the AER002 group (324 vs 258 mg/dL, p  = 0.048). A similar trend was observed in the placebo group but did not meet statistical significance. We did not identify statistically significant differences in baseline CRP, ESR, or baseline cytokines (i.e., IL-6, TNF-alpha, interferon-gamma, IL-1 beta) between responders and non-responders. Similarly, proteomic signatures of self-reported treatment response were evaluated at several timepoints. Responders to AER002 at Day 90 showed modest, selective proteomic differences from non-responders, primarily lower levels of several signaling and inflammatory proteins (Supplementary Fig  4 ); similar differences were seen in the placebo arm (Supplementary Fig.  5 ), indicating response-associated changes were not unique to AER002 treatment, though the specific proteins and pathways differed. The association between SARS-CoV-2 antibody levels and self-reported treatment response was also examined. As noted above, baseline anti-S, anti-S1, and anti-RBD antibody levels were higher in the placebo arm than in the AER002 arm, due to chance imbalance in randomization. Among the entire study population, responders had significantly higher baseline levels of anti-S (median normalized values for responder vs non-responder: 127 vs 60.5, p  = 0.033), anti-S1 (207 vs 106, p  = 0.036), and anti-RBD (151 vs 65.1, p  = 0.033) antibodies, but not baseline anti-N (2.93 vs 2.05, p  = 0.178) (Fig.  8a ). In both the AER002 and placebo groups, having high levels (classified by baseline levels above median value) of anti-S, anti-S1, anti-RBD, and anti-N antibodies was associated with greater odds of being a responder (in each of 4 different logistic regression models for each antibody type). Treatment-specific odds ratios and 95% confidence intervals are reported in Supplementary Table  8 . Due to low sample size and high correlation between antibody types, we were unable to include more than one antibody in a single model (Supplementary Table  5 ). Fig. 8 Impact of baseline anti-SAR-CoV-2 antibodies and AER002 pharmacokinetics on self-reported response. a Comparison of baseline anti-SARS-CoV-2 antibodies between self-reported responders ( n  = 16) and non-responders ( n  = 20). From left to right: anti-nucleocapsid (N); anti full-length spike (S); anti-S1 subunit (S1); anti-receptor binding domain (RBD). Median values (horizontal line) are shown for each group. b Multiple logistic regression of each baseline antibody level and AER002 exposure (AUC: area under the concentration-time curve) on likelihood of self-reported response displayed in pairs of logistic regression plots (Left: baseline antibody <median, Right: baseline antibody ≥ median) for each antibody (clockwise from top left: anti-N, anti-S, anti-RBD, anti-S1). Placebo – Responder ( n  = 6); Placebo – Non-Responder ( n  = 6); AER002 – Responder ( n  = 10); AER002 – Non-Responder ( n  = 14). Forest plots of odds ratios (OR) derived from each regression analysis depicted below each pair of plots with error bars representing 95% confidence intervals, and symbols representing point estimates (low Ab [reference] n  = 18, high Ab n  = 18). Source data are provided as a Source Data file. a Comparison of baseline anti-SARS-CoV-2 antibodies between self-reported responders ( n  = 16) and non-responders ( n  = 20). From left to right: anti-nucleocapsid (N); anti full-length spike (S); anti-S1 subunit (S1); anti-receptor binding domain (RBD). Median values (horizontal line) are shown for each group. b Multiple logistic regression of each baseline antibody level and AER002 exposure (AUC: area under the concentration-time curve) on likelihood of self-reported response displayed in pairs of logistic regression plots (Left: baseline antibody <median, Right: baseline antibody ≥ median) for each antibody (clockwise from top left: anti-N, anti-S, anti-RBD, anti-S1). Placebo – Responder ( n  = 6); Placebo – Non-Responder ( n  = 6); AER002 – Responder ( n  = 10); AER002 – Non-Responder ( n  = 14). Forest plots of odds ratios (OR) derived from each regression analysis depicted below each pair of plots with error bars representing 95% confidence intervals, and symbols representing point estimates (low Ab [reference] n  = 18, high Ab n  = 18). Source data are provided as a Source Data file. We also assessed an interaction between AER002 area under the concentration-time curve (AUC 0-inf , as calculated based on PK analysis in leveraging data from 175 to 212 days following AER002 administration) and levels of each of the antibody types in these regression models. A greater AER002 AUC was associated with greater odds of being a responder in participants who exhibited low baseline anti-S, anti-S1, and anti-RBD levels (less than median cutoffs) (Fig.  8b ). Specifically, for each 100 mg*day/L AER002 AUC increase, there was a 1.41-fold higher odds of response (95% CI, 1.07−2.50, p  = 0.011) for participants with low baseline anti-S, 1.36-fold higher odds (1.04–2.27, p  = 0.019) for anti-S1, and 1.41-fold higher odds (1.07−2.50, p  = 0.010) for anti-RBD. For those with low anti-N, increasing AER exposure was not associated with greater odds of response (OR 1.17 [0.93−1.58], p  = 0.189). An association between odds of being a responder and AER002 AUC was not found in the subgroup who had high baseline antibodies.

Numerous studies suggest that SARS-CoV-2 can persist in some people after COVID-19 6 , but the precise form, location, and mechanism of persistence remain incompletely understood 17 . Small-molecule antivirals such as a 15-day course of nirmatrelvir/ritonavir have failed to show benefit in Long COVID 21 , 22 , perhaps because of short treatment duration or because persistence may not reflect active viral replication 5 , 17 . In this exploratory phase 2a trial, we hypothesized that a long-acting SARS-CoV-2-specific monoclonal antibody would address these issues and promote symptomatic improvement either through direct neutralization of spike protein or the clearance of infected cells through processes mediated by retained effector function. We found that AER002 was safe and well tolerated but did not produce a detectable group-level improvement in physical health, quality of life, objective physical or cognitive performance, or blood-based biomarkers compared with placebo. Exploratory analyses nevertheless identified potential biological effects on tissue-based immune activation and suggested that greater drug exposure may have been associated with clinical response among participants with lower baseline SARS-CoV-2 antibody levels. These hypothesis-generating findings do not establish efficacy but may inform participant selection, dosing, and mechanistic endpoints in future trials of monoclonal antibodies targeting viral persistence in Long COVID. SARS-CoV-2 persistence has been documented in the post-acute phase of COVID-19 but is not specific to Long COVID. Persistent SARS-CoV-2 may also not be a driver of disease for all patients with the condition; rather, a subset of patients with Long COVID may harbor persistent virus. Estimates of SARS-CoV-2 persistence vary from 10 to 40% in the months following COVID-19; less is known about persistence on the order of years 6 , 17 . The lack of a validated, accessible biomarker to confirm the presence of SARS-CoV-2 persistence presents a major challenge to the conduct of clinical trials targeting this mechanism 5 . No such measure was available at our study’s outset, so we could not recruit based on a persistence metric–a likely contributor to the study not meeting its primary outcome. Tissue measurements have been proposed as a better metric for assessing viral persistence 17 , but only one of the participants who underwent gut biopsy had detection of SARS-CoV-2 RNA in gut tissue at baseline; notably this individual reported improvement. As a result, it is possible that this small study simply did not assess enough individuals with the potential to benefit from this therapeutic approach. Coordinated efforts to develop and validate biomarkers of viral persistence in more accessible matrices like blood are needed; once available, future studies should consider using such measurements to determine eligibility. Participants in this trial had been experiencing Long COVID for about two years, considerably longer than the period in which SARS-CoV-2 persistence has thus far been linked to Long COVID symptoms (3-14 months) 7 – 9 , 53 ; a recent study reported declining plasma antigen detection over two years. 54 It is possible that SARS-CoV-2 persistence is more relevant earlier in the disease course and that, over time, downstream processes such as chronic immune dysfunction, autoimmunity, vascular injury, or mitochondrial dysfunction become increasingly self-sustaining. Longitudinal studies suggest that the biology of ME/CFS, another multi-system disorder often attributed to an infectious trigger, can evolve over time 55 , 56 . Larger studies should consider analyses stratified by time since the Long COVID-triggering SARS-CoV-2 infection to assess whether interventions may have differential benefit in various stages of the disease process. Our population also differed from the three cases previously reported to improve after casirivimab/imdevimab. 29 Those individuals were treated earlier in their Long COVID course and during a subsequent acute reinfection, so their benefit may have reflected reinfection-related immune effects rather than clearance of an established chronic reservoir. We made several observations in our exploratory analyses that may inform future work. Participants in both groups who reported improvement tended to have higher anti-SARS-CoV-2 antibody levels at baseline, suggesting that these measures may represent prognostic markers of subsequent improvement rather than treatment-specific predictors. Consistent with this observation, increasing AER002 exposure was not associated with higher odds of response among participants with high baseline antibody levels. In contrast, among participants with lower baseline anti-SARS-CoV-2 antibody levels, greater AER002 exposure was associated with higher odds of reported improvement. Although hypothesis-generating, these observations raise the possibility that antibody therapy or other immune enhancing approaches may be more relevant in individuals with comparatively weaker endogenous immune responses to the virus. Sustained therapeutic exposure (e.g., higher doses, repeated administration) may be important. Given the small sample size, multiple comparisons, and absence of an overall treatment effect, this finding requires prospective validation. Long COVID is associated with systemic inflammation 57 , and our prior PET imaging work has revealed tissue-based immune activation in multiple organ systems 15 . Although underpowered, exploratory PET imaging showed significantly lower SUV in brain, thoracic spinal cord, and most notably submandibular gland after AER002 versus placebo. Submandibular tracer uptake denoting T cell activation decreased in most AER002 participants and increased in placebo participants, suggesting antibody-mediated blunting of immune activation possibly driven by effects on SARS-CoV-2 reservoirs in oral glandular tissue. Consistent with this, exploratory proteomics analyses suggested AER002 was associated with attenuation of multiple immune and cellular activation pathways, in line with evolving pharmacodynamic effects. While not definitive, these data suggest AER002 had a biological effect on inflammatory and immune-activation pathways relevant to Long COVID. As in other Long COVID trials 21 , 22 , 58 – 60 , we observed a robust response among those who received the placebo. The placebo group exhibited higher baseline SARS-CoV-2 antibody levels than the AER002 group, which we found to be a strong predictor of improvement; this may have driven improvement in the placebo arm and blunted a therapeutic effect in the treatment arm. It is also possible that improvements in both arms were driven by changes due to intensive observation (Hawthorne effect 61 ), the expectancy that the intervention would be beneficial (placebo effect 62 ), or idiosyncratic improvement related to concomitant medication changes. If a placebo effect occurred, it should be noted that these responses are biologically meaningful, engaging neuroimmune networks that influence both symptom perception and inflammatory processes 63 . Such responses can alter pain signaling, autonomic function, and inflammation 64 , 65 . Expectancy-driven mechanisms can modulate disease activity and objective inflammatory markers in inflammatory arthritis 66 . Notably, regardless of cause, few participants reported returning to their pre-COVID baseline. From a clinical standpoint, these effects highlight the interconnectedness of mind, brain, and body in chronic illness 67 . From a design standpoint, strategies such as standardized treatment messaging, placebo run-ins, or crossover designs may help distinguish true therapeutic effects from expectancy-related and temporal effects 68 ; larger, adequately powered studies accounting for these responses are also needed. Strengths of this study include its randomized, placebo-controlled design, detailed clinical phenotyping using both patient-reported and objective measures, extended follow-up, pharmacokinetic analyses, and incorporation of optional procedures to support pathogenesis research. Several limitations should be noted. This exploratory phase 2a trial emphasized safety, tolerability, feasibility, and mechanistic outcomes. Its sample size provided power only to detect a large difference in the PROMIS-29 PHSS and could not address smaller, potentially clinically meaningful treatment effects. The intensive study procedures required frequent, prolonged in-person visits and thus excluded those with the most severe illness, including those who were homebound. The lack of a validated biomarker of SARS-CoV-2 persistence precluded biomarker-enriched enrollment and necessitated the use of a broad clinical definition of Long COVID, resulting in a clinically and potentially biologically heterogeneous cohort. If persistence only drives illness in a subset of individuals, this heterogeneity may have diluted a treatment effect and obscured subgroup-specific signals. Although colorectal tissue sampling was incorporated as an exploratory measure, the sensitivity of a rectosigmoid biopsy is unknown because persistence may be focal, may occur in other tissue compartments, or may be missed by a single sampling procedure or assay. The low frequency of tissue-based viral detection therefore cannot establish that persistence was absent in other participants. Nevertheless, the absence of biomarker-based selection may have reduced the proportion of participants with the biological target of the intervention. Although all participants attributed their symptoms to a preceding test-confirmed SARS-CoV-2 infection and underwent clinician assessment to exclude other etiologies, some symptoms may have arisen from conditions other than Long COVID. Most participants had experienced Long COVID for several years before enrollment. This population remains clinically important because many individuals with severe and persistent illness were infected during the early, pre-vaccine phases of the pandemic and continue to have a substantial unmet need for treatment. However, viral persistence may be more therapeutically relevant earlier in the disease course. Enrollment closer to the Long COVID-triggering infection was constrained by the loss of AER002 neutralizing activity against later variants, which limited eligibility to those with infections occurring before August 15, 2022. Moreover, one third of participants experienced a SARS-CoV-2 reinfection after AER002 was expected to have lost neutralizing activity, raising the possibility of so-called “mixed reservoirs” containing viruses with different susceptibility 17 . We measured binding, not variant-specific neutralizing, antibodies; because the infecting variant was not definitively identified for most participants, we could not determine whether antibody-response quality to the triggering virus modified treatment response. Concomitant medications may have affected some outcome measures and contributed to improvement in both groups despite requests to avoid changes in background medication regimens during the study. In addition, no outcome measures have been specifically validated in Long COVID, and the measures we selected may have failed to capture all relevant dimensions of the illness. Our patient-reported and objective instruments were selected based on expert opinion and the state of the field at the time of trial design, but no outcome has been formally validated in Long COVID. There is broad concern in the field that, without validated, sensitive, and specific instruments capable of capturing the dimensions of illness most relevant to Long COVID, progress in evaluating candidate therapeutics will continue to be limited. Finally, although this study tested the hypothesis that SARS-CoV-2 persistence drives Long COVID, immune dysfunction or other processes could instead promote persistence or sustain disease independently of the virus. In that setting, immunomodulatory or combination approaches may be required 5 . Such studies are now underway. In summary, treatment with AER002 did not show a group-level benefit over placebo for Long COVID. Despite this negative result, several observations suggest promise for monoclonal antibody therapies in certain individuals and should inform future trial design. Future trials targeting viral persistence in this condition would benefit from confirming persistence at enrollment, larger samples to offset heterogeneity and reinfection, agents with broad activity against historical and modern variants, and screening for favorable immune phenotypes (e.g., low baseline antibody levels) and/or dosing strategies that sustain therapeutic levels.

Long COVID is a debilitating infection-associated chronic condition (IACC) characterized by COVID-attributed symptoms that persist for months to years following a SARS-CoV-2 infection 1 . Millions of individuals are affected worldwide 2 and the health and economic impacts are profound 3 . Despite its public health significance, the pathobiology that drives Long COVID is incompletely understood 4 . There is thus an urgent need for rigorous, randomized trials not only to identify promising therapeutics but also to shed light on the underlying pathobiological processes that drive the condition 5 . Although SARS-CoV-2 infection was initially assumed to be transient, there is growing evidence to suggest that components of the virus (RNA and/or protein) can persist for months or years in some individuals 6 . Multiple studies have now demonstrated SARS-CoV-2 antigens circulating in blood for months after acute infection 7 – 9 , although these findings are not universal 10 . Others have identified SARS-CoV-2 RNA or protein in tissue sites, such as gut mucosal biopsies 11 – 16 . The nature of this persistence – whether it represents inert RNA, RNA that can be translated to produce viral proteins, or even low-level viral replication – remains unclear 17 , as does whether it directly causes Long COVID or instead reflects another primary driver such as immune dysregulation 5 . In response to case reports suggesting a benefit of off-label antiviral use in at least some people with Long COVID 18 – 20 , several clinical trials of antivirals have been undertaken. Two trials testing a 15-day course of the protease inhibitor nirmatrelvir/ritonavir (brand name: Paxlovid) failed to show significant benefit in improving symptoms and quality of life among individuals with Long COVID 21 , 22 . However, this strategy has notable limitations, including the short treatment duration and the expectation that drugs like nirmatrelvir would only be effective if Long COVID were driven by active SARS-CoV-2 replication 5 , 17 . SARS-CoV-2-specific monoclonal antibodies have been proposed as another strategy to address viral persistence 4 , 5 , 17 , 23 . Initially developed to treat or prevent SARS-CoV-2 infection, these antibodies neutralize virions and block infection of new target cells 24 . In some cases, monoclonal antibodies can be modified to have an extended half-life 25 – 27 , allowing for a prolonged therapeutic or prophylactic effect. Furthermore, when effector function is retained, such antibodies may also facilitate opsonization of virions and clearance of infected cells expressing spike protein via antibody- or complement-mediated mechanisms, even without neutralization 28 . Several case series have detailed profound improvement in Long COVID symptoms after a single infusion of SARS-CoV-2-specific monoclonal antibodies in a small number of individuals 29 , 30 . Despite this anecdotal evidence, there are no published randomized clinical trials using monoclonal antibodies for Long COVID. Here, we report the results of outSMART-LC ( NCT05877508 ), an exploratory, 2:1 randomized, double-blind, placebo-controlled phase 2a trial of a single infusion of the long-acting SARS-CoV-2-specific monoclonal antibody AER002 31 in people with Long COVID. AER002 is a fully human monoclonal immunoglobulin-G1 (IgG1) mAb that binds to the receptor binding domain (RBD) of SARS-CoV-2 spike protein 31 . The antibody demonstrated in vitro activity against the Omicron variants of concern circulating at the time of trial design (BA.2.12.1 through BA.5), with reduced but retained activity against BQ.1. AER002 included modifications to the fragment crystallizable (Fc) part of the antibody to prolong half-life. Despite its waning efficacy against later SARS-CoV-2 variants, we theorized AER002 would remain effective in clearing SARS-CoV-2 persistence from infections occurring before it lost neutralizing capacity. We therefore designed outSMART-LC as a hypothesis-generating, mechanistic trial, using AER002 as a tool to probe whether clearing persistent SARS-CoV-2 antigen can improve symptoms and quality of life in people with Long COVID and to generate proof-of-concept evidence for this approach.

Source Data Source Data