The Breadth of the Neutralizing Antibody Response to Original SARS-CoV-2 Infection is Linked to the Presence of Long COVID Symptoms

Background: The associations between longitudinal dynamics and the breadth of SARS-CoV-2 neutralizing antibody response with various Long COVID (LC) phenotypes prior to vaccination are not known. The capacity of antibodies to cross neutralize a variety of viral variants may be associated with ongoing pathology and persistent symptoms. Methods: We measured longitudinal neutralizing and cross-neutralizing antibody responses to pre- and post-SARS-CoV-2 Omicron variants in participants infected during the early waves of the COVID-19 pandemic, prior to wide-spread rollout of SARS-CoV-2 vaccines. Cross sectional regression models adjusted for various clinical covariates and longitudinal mixed effects models were used to determine the impact of the breadth and rate of decay of neutralizing responses on the development of Long COVID symptoms in general, as well as LC phenotypes. Results: We identified several novel relationships between SARS-CoV-2 antibody neutralization and the presence of LC symptoms. Specifically, we show that, although neutralizing antibody responses to the original, infecting strain of SARS-CoV-2 were not associated with LC in cross-sectional analyses, cross-neutralization ID50 levels to the Omicron BA.5 variant approximately 4 months following acute infection was independently and significantly associated with greater odds of LC and with persistent gastrointestinal and neurological symptoms. Longitudinal modeling demonstrated significant associations in the overall levels and rates of decay of neutralization capacity with LC phenotypes. A higher proportion of participants had antibodies capable of neutralizing Omicron BA.5 compared with BA.1 or XBB.1.5 variants. Conclusions: Our findings suggest that relationships between various immune responses and LC are likely complex but may involve the breadth of antibody neutralization responses.


INTRODUCTION 70
Many individuals experience post-acute sequelae of SARS-CoV-2 infection (PASC), which can affect 71 quality of life and return to health [1][2][3]. The etiologic drivers of Long COVID (LC), a form of PASC 72 defined by ongoing, often debilitating, symptoms, are poorly understood and likely involve multiple 73 mechanisms [2,4,5]. Proposed mechanisms include aberrant autoreactive immune responses, 74 microvascular dysregulation, and reactivation of latent human herpesviruses which may lead to the 75 systemic inflammatory responses now identified in individuals with Long COVID compared to those 76 who fully recovered [6][7][8][9][10][11]. Furthermore, there is growing evidence that SARS-CoV-2 subgenomic RNA 77 and proteins are present in the tissues of at least a subset of immunocompetent individuals with LC [12-78 14]. Although those with persistent symptoms tend to have higher levels of SARS-CoV-2 Spike-specific 79 antibody levels [10,[15][16][17][18], we and others have previously demonstrated that LC is associated with 80 adaptive immune dysregulation and exhaustion [15,18]. 81 SARS-CoV-2 infection leads to rapid development of robust antibody responses, although neutralizing 82 capacity wanes more quickly than total Spike IgG levels over time [17,[19][20][21]. A higher initial viral 83 burden or persistence of viral antigens may lead to observed dysregulated immune phenotypes and 84 higher antibody levels. However, there is a paucity of information regarding the associations between 85 longitudinal dynamics or the breadth of the neutralizing antibody response with various LC phenotypes 86 with some data showing that weaker antibody responses over time being associated with LC [22]. 87 Recent pre-print data suggests that an expanded antibody response against the prior OC43 endemic 88 coronavirus may be associated with Long COVID [23]. This suggests that the breadth of the response 89 to initial infection may play an important role in the development of LC. Given that the rapid emergence 90 of Omicron variants that evade neutralization result from infection from older SARS-CoV-2 strains (e.g., 91 ancestral SARS-CoV-2, Alpha and Delta variants) as well as to 25], there is 92 strong interest in determining the relationship between the breadth and durability of the initial antibody 93 responses and the presence of Long COVID symptoms. The rapid emergence of novel variants and 94 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted March 31, 2023. ;https://doi.org/10.1101https://doi.org/10. /2023 generating HIV-1 pseudovirions that express the SARS CoV-2 Spike protein as previously described 119 [20,[27][28][29]. The pseudovirus is prepared by co-transfecting HEK293 producer cells with an HIV-1 120 genomic vector that contains a firefly luciferase reporter gene together with a SARS CoV-2 Spike 121 protein expression vector. Neutralizing antibody activity is measured by assessing the inhibition of 122 luciferase activity in HEK293 target cells expressing the ACE2 receptor and TMPRSS2 protease 123 following pre-incubation of the pseudovirions with serial dilutions of the serum specimen. ID50 values 124 was generated for the original SARS-CoV-2 Spike protein as well as the following variants: Alpha 125 is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted March 31, 2023. ;https://doi.org/10.1101https://doi.org/10. /2023 Breadth of antibody neutralization to the original infecting SARS-CoV-2 variant and subsequent 168 viral variants. Using the PhenoSense assay (Monogram Biosciences), we measured the inhibitory 169 serum dilutions at which 50% neutralization occurred (ID50) using pseudoviruses expressing SARS-170 CoV-2 Spike protein from the original strain (with which a majority of our participants were infected) and CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted March 31, 2023. ; https://doi.org/10.1101/2023.03.30.23287923 doi: medRxiv preprint demographic characteristics, including age, sex, hospitalization during acute infection, body mass index 193 (BMI), and a pre-existing history of diabetes mellitus for the most clinically relevant variants in our study 194 population (ancestral SARS-CoV-2, Delta and Omicron sub variants; Figure 2). Overall, antibody 195 neutralization ID50 decreased over time for all strains (p<0.01) with the exception of the Omicron BA.1 196 variant, (p=0.16) for which initial levels were substantially lower than to the other variants. When 197 stratifying by age greater than 50 years, we identified no difference in antibody neutralization to all 198 variants tested across all time points in the mixed effects models. In contrast, male sex was associated 199 with higher viral neutralization for original SARS-CoV-2, Delta, and Omicron BA.5 variants, but not for 200 Omicron BA.1, across all time points. These observed sex-based differences were similar between 201 original and Delta variants, each approximately 0.41 and 0.33 log 10 higher, respectively, observed in 202 males compared to females recovering from  For those hospitalized during acute infection, neutralization levels were significantly higher across all 204 strains with the exception of BA.1. The magnitude of the difference in responses for those hospitalized 205 versus not hospitalized was diminished with those strains with lower overall responses: 1.18 and 0.94 206 log 10 higher responses for the original virus and Delta variant versus 0.33 and 0.28 log 10 higher for 207 omicron BA.1 and BA.5 responses, respectively. Individuals with pre-existing diabetes also had 208 differential neutralization across variants, with higher initial levels and longitudinal levels over all time 209 points, and showing a more rapid decline over time compared to those without diabetes. The 210 magnitude of wild-type SARS-CoV-2 responses for those with a BMI >30 was higher across all time 211 points for the original and delta viral strains (All P<0.001) and declined more rapidly for the original 212 Delta and Omicron BA.5 variants (All P>.012; Supplemental Figure 2). In contrast, neutralization 213 ID50s from participants with a pre-existing history of diabetes mellitus were overall lower across all time 214 points for the original SARS-CoV-2 (All P <0.05) and ID50s to the original strain and Delta variant 215 declined less rapidly (P<0.05). 216 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Breadth of the neutralizing antibody responses is associated with increased odds of Long 217
COVID. In order to determine whether neutralization capacity was related to Long COVID, we 218 performed logistic regression modeling with either the presence of any Long COVID symptom at a 219 given sample time point or with specific Long COVID symptom phenotype (neurocognitive, 220 cardiopulmonary, gastrointestinal, musculoskeletal and fatigue) at the three main collection time points: 221 1 month (N=69; median 33 days), 2 months (N=115; median 59 days) and 4 months (N=119; median 222 120 days) following acute infection. Data from only one time point per participant within each time 223 period was included to avoid oversampling of specific individuals. Specifically, the sample time closest 224 to 30 days within a 21-45 day window, 60 days within a 56-75 day window, and 120 days within a 100-225 150 day window were included. Factors included in the first model ( Figure 3A) included the neutralizing 226 antibody ID50 (continuous variable), prior hospitalization during acute COVID-19, female sex, and age 227 greater than 50 years of age. Overall, there were no significant differences between neutralization ID50 228 to any strain and the presence of Long COVID in general or any specific Long COVID phenotype at 1 229 and 2 months following acute infection (all P>0.05). As shown in Figure 3A, the neutralization ID50 of 230 the ancestral SARS-CoV-2 (the infecting strain in this study population), as well as Alpha and Delta 231 variants, were not significantly associated with the presence of any Long COVID symptom or specific 232 Long COVID phenotype approximately 4 months after acute infection. However, cross-neutralization 233 ID50s to Omicron BA.5 were significantly and positively associated with neurocognitive and 234 gastrointestinal symptoms (i.e. higher odds of having symptoms within these phenotypes). There were 235 no significant associations between BA.5 neutralization ID50 and fatigue and cardiopulmonary 236 symptoms or as shown in Supplementary Figure 3). Regression analyses including ID50s to both 237 wild-type SARS-CoV-2 and Omicron BA.3 were also performed as in Supplementary Figure 4A. 238 Including ID50s to both wild type and BA.5 strains led to similar results, with cross-neutralization to 239 Omicron BA.5 being significantly associated with having any Long COVID symptom and neurocognitive 240 symptoms, whereas ancestral SARS-CoV-2 ID50 were not significantly associated with Long COVID or 241 any symptoms cluster. 242 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Finally, we assessed neutralizing antibody responses against the original infecting SARS-CoV-2 strain 261 by individual Long COVID phenotype (i.e. non mutually exclusive symptom cluster) compared to those 262 without any Long COVID symptoms or those with or without symptoms but not within the specific Long 263 COVID symptom cluster (Figure 4A & B, respectively). Overall, differences in levels across all time 264 points or changes over time were similar, with high interpatient variation in neutralization ID50s 265 observed. Nonetheless, we found that those with gastrointestinal and cardiopulmonary symptoms had 266 0.27 and 0.43 log 10 higher neutralization ID50 compared to those without any Long COVID symptom 267 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted March 31, 2023. ; https://doi.org/10.1101/2023.03.30.23287923 doi: medRxiv preprint across all data points over time (P=0.04 and <0.001, respectively; Figure 4A). Decay in neutralization 268 ID50 was faster (i.e. more negative slope in mixed linear effects model) in participants with 269 cardiopulmonary and musculoskeletal symptoms compared to those without any symptoms (P=0.01 270 and 0.047, respectively). Compared to those with or without persistent symptoms, but no symptoms in 271 the specified phenotype cluster, those with cardiopulmonary or musculoskeletal symptoms were overall 272 higher across all time points (both P<0.001), and those with musculoskeletal, cardiopulmonary and 273 neurocognitive symptoms declined more rapidly than those without those specific symptoms (all P < 274 0.05; Figure 4B). increased or decreased risks of Long COVID, but it will be very difficult to identify variant-specific 305 effects in the current era of widespread and variable vaccination, reinfection, and antiviral treatment 306 uptake. 307 We also evaluated the longitudinal relationships between antibody neutralization responses, various 308 clinical factors, and Long COVID phenotypes. The mixed effects models allowed us to determine 309 differences between these variables across all data points over time (including multiple time points for 310 each individual) and changes (i.e. decay) over time. These analyses revealed relationships between 311 neutralization responses that were not observed in the cross-sectional analyses, such as significantly 312 higher SARS-CoV-2 neutralizing responses to the original infection SARS-CoV-2 across all time points 313 for those with gastrointestinal and cardiopulmonary Long COVID symptoms. Regardless of the overall 314 higher levels, a faster decay of neutralizing ID50 was observed for these phenotypic clusters, in 315 addition to those with musculoskeletal symptoms. Of note, neutralization titers decay more rapidly on a 316 whole than common epitope antibody responses which have been associated with various PASC 317 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted March 31, 2023. ;https://doi.org/10.1101https://doi.org/10. /2023 symptoms in longitudinal analyses [16,17,20], highlighting importance of temporal immune dynamics 318 in the study of this condition. Together, these and the cross-sectional data suggest that an overall 319 higher neutralization response that wanes more quickly, or ones that remain broad over time, are 320 associated with LC. While the causes of these dynamics are unknown, one possibility is that persistent 321 SARS-CoV-2 antigen presentation in tissues, which has been proposed as a potential mechanism of 322 Long COVID, may lead to overall higher antibody neutralization over time, and potentially to a broader 323 response to subsequent variants. While speculative, our findings suggest that relationships between 324 various immune responses and Long COVID are likely complex, and different approaches to data 325 analyses may yield different, but complementary information. 326 Strengths of this study include the use of highly characterized samples from the pre-vaccine and pre-327 Omicron era , before reinfections became common. This allowed for a more straightforward analysis of 328 neutralization dynamics in the absence of these complex confounding factors. In addition, both those 329 with and without Long COVID were recruited and assessed in an identical manner, addressing potential 330 biases that might occur when comparison groups are derived from different cohorts as has been 331 common in studies of Long COVID. As in similar analyses where we have been able to evaluate 332 mechanisms according to distinct Long COVID phenotypes [9], we leveraged our high degree of 333 symptom characterization to analyze different case definitions of Long COVID. This approach is 334 informative, especially since the case definition remains controversial and it is possible that different 335 phenotypes are driven by different mechanisms. Limitations of the study include a lack of participants 336 infected with more recent variants, preventing us from extending our observations into more recent 337 waves of the pandemic. The cohort was a convenience sample, and although this allows for valid 338 inferences regarding Long COVID biology comparing people with and without the phenotype of interest 339 within the cohort, extrapolation to all individuals with prior COVID-19 must be done cautiously. The 340 neutralization assay used Spike protein pseudoviruses rather than intact, replication-competent virus, 341 but these pseudovirus assays have been shown to have comparable results in several studies [27,29,342 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted March 31, 2023. ; https://doi.org/10.1101/2023.03.30.23287923 doi: medRxiv preprint 31-33]. Nonetheless, we believe that these results suggest at least one potential contributor to Long 343 COVID, although more work will be necessary to validate these observations in other cohorts, including 344 those derived from later waves of the pandemic in the setting of vaccination or reinfection. 345 Chen, Aidan Donovan, Carrie Forman, and Rania Ibrahim for assistance with data entry and 361 review. We thank the UCSF AIDS Specimen Bank for processing specimens and maintaining the LIINC 362 biospecimen repository. We are grateful to Elnaz Eilkhani and Monika Deswal for regulatory support. 363 We are also grateful for the contributions of LIINC leadership team members: Isabelle Rodriguez-364 Barraquer, and Rachel Rutishauser. 365 366 367 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted March 31, 2023. ;https://doi.org/10.1101https://doi.org/10. /2023  is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted March 31, 2023. ;https://doi.org/10.1101https://doi.org/10. /2023  CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted March 31, 2023. ;https://doi.org/10.1101https://doi.org/10. /2023   CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted March 31, 2023. ;https://doi.org/10.1101https://doi.org/10. /2023