A/H1N1 hemagglutinin antibodies show comparable affinity in vaccine-related Narcolepsy type 1 and control and are unlikely to contribute to pathogenesis

An increased incidence of narcolepsy type 1 (NT1) was observed in Scandinavia following the 2009–2010 influenza Pandemrix vaccination. The association between NT1 and HLA-DQB1*06:02:01 supported the view of the vaccine as an etiological agent. A/H1N1 hemagglutinin (HA) is the main antigenic determinant of the host neutralization antibody response. Using two different immunoassays, the Luciferase Immunoprecipitation System (LIPS) and Radiobinding Assay (RBA), we investigated HA antibody levels and affinity in an exploratory and in a confirmatory cohort of Swedish NT1 patients and healthy controls vaccinated with Pandemrix. HA antibodies were increased in NT1 patients compared to controls in the exploratory (LIPS p = 0.0295, RBA p = 0.0369) but not in the confirmatory cohort (LIPS p = 0.55, RBA p = 0.625). HA antibody affinity, assessed by competition with Pandemrix vaccine, was comparable between patients and controls (LIPS: 48 vs. 39 ng/ml, p = 0.81; RBA: 472 vs. 491 ng/ml, p = 0.65). The LIPS assay also detected higher HA antibody titres as associated with HLA-DQB1*06:02:01 (p = 0.02). Our study shows that following Pandemrix vaccination, HA antibodies levels and affinity were comparable NT1 patients and controls and suggests that HA antibodies are unlikely to play a role in NT1 pathogenesis.

, as well as a lower affinity of antibodies from NT1 patients' sera recognizing the NP-PR1934 (the influenza nucleoprotein variant present in the Pandemrix vaccine) 20 . However, it is acknowledged that hemagglutinin (HA) is the major antigenic determinant during an influenza infection. Suggestions of T cell mediated molecular mimicry between HA and hypocretin have been investigated with conflicting conclusions [23][24][25][26][27][28][29] . In this study we analysed the antibody response to HA with respect to Pandemrix vaccination in NT1 patients and healthy vaccinated Swedish residents in an exploratory cohort of samples collected in 2012, and a second replicatory cohort of samples from additional NT1 patients and their family members collected in 2015-2017.
We measured HA antibodies using the Luciferase Immunoprecipitation System (LIPS) and Radiobinding Assay (RBA), two alternative immunoassay formats widely adopted for the study of autoantibodies, i.e. assays that are more likely to detect antibodies to conformational epitopes [30][31][32] . We previously showed that an RBA test, using 35 S-radiolabled HA expressed in vitro by coupled transcription, detected increased HA antibodies levels among NT1 patients in the exploratory cohort as well as an association of HA antibodies with HLA-DQB1*06:02 18 . In light of the different antigen expression system adopted by LIPS, which yields a trimeric post-translationally modified HA, we used the two assays to explore: first, if the HA antibody association to NT1 was found only in samples collected close to the time of vaccination with Pandemrix, or persisted also in confirmatory cohort of patients collected at a later time-point; second, if the antibody affinity of the HA response might differ between NT1 patients and control; and third, if systematic differences in results could be found and traced to the type recombinant antigens used in the two assay formats.

Results
Antibody levels. A/H1N1 HA antibody levels were determined in patients and controls from both the exploratory and confirmatory study cohorts. Comparisons were performed separately for LIPS and RBA and antibody titres were found to be non-normally distributed. A/H1N1 HA antibodies were similar in range but with modestly increased median levels among narcolepsy patients compared to control subjects in the exploratory cohort both in LIPS (median arbitrary units (

Age.
A/H1N1 HA antibody levels measured by LIPS were investigated in relation to age. In the exploratory cohort, we observed a modest inverse correlation of HA antibody levels with age in both cases and controls (R 2 = 0.092 and 0.16, respectively). This finding was replicated in the confirmatory cohort ((R 2 = 0.24 and 0.16, respectively). Our prior findings 33 suggested that differences in antibody levels were most prominent in children aged less than 13 years. Accordingly, we stratified the study subjects into two age groups: children (< 13 years) and adolescents/adults (≥ 13 years). Upon stratification, we observed that HA antibody levels measured by LIPS in healthy controls were higher in younger vs older subjects in both study cohorts (Mann-Whitney p = 0.00386 and p = 0.0037) ( Table 1). HA antibody levels were higher in younger vs older NT1 patients but this difference reached statistical significance only in the confirmatory cohort (p = 0.020). Compared to controls belonging to the same age group, HA antibody levels in NT1 patients were higher only in the exploratory cohort (Mann-Whitney p = 0.0419 not corrected). LIPS vs. RBA comparisons. Antibody levels were compared between LIPS and RBA. In the exploratory cohort, antibody titres correlated both in the patients (rho = 0.370, p = 0.0151) as well as the controls (rho = 0.303, p = 0.0152). In the confirmatory cohort, antibodies correlated significantly only in patients (rho = 0.664, p = 6.98e-05) but not in controls (rho = 0.203, p = 0.141) (Supplementary Figure S1).
HA antibody affinity. The affinity of HA antibodies was evaluated in LIPS and RBA in a selection of sera from the exploratory cohort. Kd50 was determined by displacement of antibody binding using increasing concentrations of Pandemrix vaccine (Fig. 2). Since HA antibody titres were often near or at the saturation limit of our assays, all selected samples were diluted to achieve comparable levels of binding within the linear range of the assays. Kd50 were comparable between patients and controls in the LIPS (NT1: 48 ng/ml vs. control: 39 ng/ ml, p = 0.357) and the RBA (NT1: 472 ng/ml vs. control: 491 ng/ml, p = 0.632) (  Fig. 3b).

Discussion
The report of increased narcolepsy cases following a nationwide vaccination programs with the Pandemrix flu vaccine in Sweden and Finland emerged in August 2010, with the full scope of its association with influenza vaccination surfacing in 2011-2012. The vaccine had never received approval by the FDA for distribution in the USA and the concern for the safety of its administration led to the effective discontinuation of all vaccination The mechanistic association between narcolepsy and Pandemrix vaccination has proven contentious. In light of the predominant hypothesis that narcolepsy has an autoimmune origin, the specific formulation of inactivated viral components and adjuvant in the Pandemrix vaccine has been suspected of triggering the disease in genetically predisposed young individuals 34 via the induction of autoimmunity to self-proteins via molecular mimicry. However, some of the evidence previously published regarding this point has not withstood scrutiny 35 , leaving essentially unresolved the link between Pandemrix vaccination and narcolepsy development.
To contribute towards a better understanding of the immune response in patients that developed narcolepsy after a Pandemrix vaccination, in this and previous studies we explored whether the antibody response to the flu antigens included in the vaccine formulation was quantitatively or qualitatively different from controls.
In this work in patients that developed the disease soon after the 2009 nationwide program in Sweden, we not only expand the number investigated of these rare subjects but also addressed the concern that methodological differences might impact the evaluation of flu antibodies. In particular, we explored how the characteristics of recombinant antigen used in two different immunoassays affected the measurement of antibodies to HA, a Figure 1. Raincloud plots of A/H1N1 HA antibody levels in subjects from the exploratory and the confirmatory study cohorts, determined using LIPS and RBA. Antibody levels in each subject are shown as circles with their overall probability density estimates in narcolepsy patients (magenta fill) or healthy controls (light blue fill). Boxplots represent the interquartile range (IQR) and median A/H1N1 HA antibody levels (solid black lines) measured in LIPS and in RBA. Panels show results from the exploratory (a) and the confirmatory (b) study cohorts. P-values are indicated between the compared groups. www.nature.com/scientificreports/ protein that in its native state has trimeric quaternary structure and is subjected to important glycosylation as a post-translational modification. Using a novel LIPS immunoassay based on a recombinant trimeric HA antigen expressed in eukaryotic cells, in the original exploratory cohort we were able to replicate our previous observation of a broadly similar range of HA antibodies with modestly increased median levels in narcoleptic patients compared to control. In addition, HA antibodies measured by LIPS were higher in subjects carrying the HLA-DQB1*06:02:01 allele irrespective of disease status. In the second confirmatory cohort instead, we found no statistically significant difference neither in range, distribution and median HA antibody levels in NT1 patients compared to family member Table 1. A/H1N1 HA antibody levels in relation to age. LIPS and RBA results are reported according to age groups: childhood patients (exploratory study, n = 11; confirmatory study, n = 11), childhood controls (exploratory study, n = 21; confirmatory study, n = 10), adult patients (exploratory study n = 32; confirmatory study, n = 20), adult controls (exploratory study, n = 43; confirmatory study, n = 44). AU and U/ml are referred to LIPS and RBA results, respectively. Indicated p-values refer to Mann-Whitney U tests comparing A/H1N1 HA antibodies levels between: patients and age-matched controls (p1), narcolepsy childhood and adolescent/ adult patients (p2), healthy childhood and adolescent/adult controls (p3).   In addition, the enrichment of HLA-DQB1*06:02:01 alleles in the family members vs the general population (72 vs 28%, respectively) might have affected the comparison HA antibodies between case and control. Overall, while observing a range of HA antibodies broadly similar in their distribution using both immunoassays, the partial correlation of binding between LIPS and RBA was suggestive of the likely detection of partially different epitopes by the two systems. It is known that HA has a complex expression profile, with different versions of the protein present during an infection cycle, as well as major differences in glycosylation patterns that affect immune responses. The HA 3D-structure on viral particles guides antibody responses to some epitopes while hinder the exposure of others, for example in the HA stem-regions. Most studies and diagnostic procedures often prioritize simplicity and inexpensiveness of antibody assays over a thorough assessment of how immunoglobulins (vaccine-or infection induced) that bind recombinant proteins recapitulate the humoral response recognizing native antigens. In fact, Table 2. Affinity analysis of A/H1N1 HA antibodies. Calculated Kd50 values (a) and maximal displacement (b) of A/H1N1 HA antibodies in a subset of subjects from the exploratory cohort described in Fig. 2  www.nature.com/scientificreports/ the choice and type of recombinant protein adopted in immunoassay can have a major impact results when measuring antibodies including those associated with virus induced responses in a disease 36 or vaccination setting. In our study, while the impact of the addition of a nanoluciferase reporter on the protein structure was not assessed, both the adopted expression system and the addition of a trimerization foldon domain to the chimeric antigen were previously shown to lead to a post-translationally modified, trimeric, structurally stable recombinant HA protein that was similar to its native counterpart. Moreover, in addition to an improved antigen proteinfolding, the intrinsic high luciferase activity of the NanoLuc reporter was likely associated with the observed increased sensitivity of the HA antibody LIPS compared to RBA.
Using both immunoassays in experiment of binding displacement using Pandemrix vaccine as a competitor, we determined that HA antibodies showed a comparable affinity in NT1 patient and control sera. This was in contrast to our previous study in which we observed a lower affinity of NP antibodies in NT1 patients using RBA and non-PTM NP radiolabeled antigen. Importantly, in the previous study commercially available NP proteins were used to displace antibodies instead of the Pandemrix vaccine itself 20 as in the current study. Since Pandemrix comprises polymorphic and structurally altered NPs 21 , vaccination might have driven the emergence of antibodies that would be more difficult to displace using commercial NP antigens produced in alternative systems.
Among the limitations of our study we can count: first, the lack of information regarding the HA antibody status prior to the Pandemrix vaccination in both study cohorts, and second, the fact that patients in the confirmatory cohort were sampled in 2015-2017 i.e. a few years later compared to the exploratory cohort sampled in 2012. In the absence of additional NT1 samples collected soon after the Pandemrix vaccination, we resorted to use these samples to analyse a larger number of these very rare patients. The difference in timing of collection might have impacted on the correlation between HA antibodies and disease. In particular, the possible exposure to the pandemic flu virus in the intervening years after the vaccination might have constituted a potential confounder of immunological results. It cannot be excluded that the characteristics of dominating flu strains circulating in the years following the vaccination, including their antigenicity and sequence divergence, might have been shifted in between the sample collection of the exploratory and confirmatory cohorts. For this reason and in the absence of sequential samples from the same subjects, as mitigatory measure, we enrolled in the confirmatory cohort patients' family members as control, reasoning that any potential exposure to a flu virus infection in NT1 patients would have been at least more likely to have affected control samples from the same family and to be synchronous. Re-immunization due to additional vaccination in both patients and family members was instead deemed unlikely, also because of the absence of state sponsored flu vaccination programs in Sweden before or after the 2009 pandemic except for the elderly.
Nevertheless, the comparable results obtained using two alternative immunoassays, of which one is ideally suited to the measurement of antibodies to conformational epitopes, make us we believe that our study main outcomes are valid: (1) quantitative differences of HA antibodies in NT1 patients compared to control were very modest and observed only in samples obtained relatively close to the vaccination (2) the affinity of HA antibodies did not differentiate NT1 patients from control.
In conclusion, while our results suggest that the HA antigen is unlikely to play a role in the association between Pandemrix vaccination and increased incidence of NT1, the mechanistic explanation of this relationship remains elusive.

Methods
Study populations. The current project included two narcolepsy cohorts: an exploratory and a confirmatory study group. All patients developed disease following the 2009-autumn Pandemrix vaccination and before the end of sample collection in 2012. The exploratory cohort consisted of 43 Pandemrix vaccinated NT1 patients and 64 matched population controls collected in 2012, as previously described 20,33 . The median age of patients in October of 2009 (start of vaccination) was 17 years (range 6-69 years, inter-quartile-range 12-27 years). Compared to our previous study, that included 47 patients and 80 controls, we excluded 4 patients and 16 controls who were not vaccinated with Pandemrix. The confirmatory cohort consisted of 31 Pandemrix vaccinated NT1 patients and 54 first-degree relatives as control, collected in 2015-2017 in collaboration with the Swedish Narcolepsy Association as previously described 18,37 . Patients were asked by questionnaire for year and month of acquired disease. Based on 22 answers, the median age was 15.5 years (range 6-70 years, inter-quartile-range11,75-19.5 years). No patients or controls were shared between the study-cohorts. In contrast to our previous study, that included 31 patients and 66 first-degree relatives, we now excluded 12 controls who were not Pandemrix vaccinated. The study was approved by the Regional Ethical Review Boards in Lund (Sweden) and Stockholm (Sweden), and conducted in accordance with the current guidelines. The informed consent was obtained from all participating subjects and/or their legal guardians.
Determination of antibody affinity. To quantify the affinity of HA antibodies we used serum samples from the exploratory cohort, of these 60 were analysed conducting binding displacement experiments in LIPS and 41 in RBA. The discrepancy in the tested subjects is explained by the lower antibody levels measured by RBA that precluded the conduction of displacement experiments since at the lower displacement concentration the signal would be comparable to background noise. One control-sample was excluded from calculations in the RBA as displacement did not reach 50% as required for determination of Kd50.
Immunocomplexes were captured with 2.5 μl of blocked 39 rProtein A (GE Healthcare Europe GmbH, Freiburg, Germany) for 1 h at 4 °C with shaking. Plates were washed 5 times by dispensing 750 μl/well of TBST, centrifugation at 500 g for 3′ at 4 °C, and removal of supernatant using a micro-plate plate washer (BioTek Instruments Inc., Winooski, VT, USA). Resin pellets were transferred to an OptiPlate 96-well plate (PerkinElmer, Waltham, MA, USA) and the luciferase activity was measured after addition of 40 μl/well of Nano-Glo substrate (Promega), followed by a 2 s/well readout in a Berthold Centro XS3 Luminometer (Berthold Technologies GmbH & Co. KG, Bad Wildbad, Germany). Raw data were converted to Arbitrary Units (AU) using a positive serum as index. Sera that bound recombinant H1N1 above the linear range of the assay were serial diluted and re-tested until binding fell into the linear range, and calculated AU were corrected by multiplying for the corresponding dilution factor.
Antibody binding displacement experiments were performed by adding to test reactions Pandemrix vaccine in increasing amounts, corresponding to final concentrations of 0, 0.004, 0.008, 0.04, 0.4 and 4 μg/ml.

In vitro transcription translation and RBA. The influenza A virus (A/California/04/2009/(H1N1))
HA gene (GeneBank accession: FJ966082) was subcloned into the pTnT-vector to enable in vitro transcription translation for expression of [ 35 S]-methionine radiolabelled antigen as previously described 33,40 . The TnT SP6coupled reticulocyte lysate system was used for protein-synthesis, and the protocol included 90 min incubation at 30 °C followed by size-restricted selection of protein-product using Nap-5 columns.
Displacement experiments were performed using RBA as described above but with additional analyses included with increasing concentration of Pandemrix vaccine (0, 0.004, 0.008, 0.04, 0.4 and 4 μg/ml). The vaccine was diluted to twice the desired concentration in assay buffer, and then mixed 50/50 with radiolabelled proteins (800 cpm/μl) to obtain final concentration.
HLA-DQB1*06:02 association in the confirmatory study were determined using HLA high resolution sequencing with Illumina MiSeq as previously described 18 . Statistical analysis. All statistical analyses were performed using the R software 41 (https ://www.R-proje ct.org). Comparisons of median antibody titres across groups were performed using the Mann Whitney U test. For age stratification, ages in the confirmatory cohort were re-calculated back to October 2009, i.e. the initiation of Pandemrix vaccination in Sweden. The correlation between LIPS and RBA antibody titres was assessed using the Spearman's coefficient of rank correlation test (ρ (rho)). Affinity was calculated through the determination of the half-maximal binding Kd50 (ng/ml). A four-parameter log-logistic function was constructed with y = D + (A − D)/(1 − (x/C)^D) as previously described using the drc package 20,42 (https ://CRAN.R-proje ct.org/ packa ge=drc). The significance threshold (a) was set to 0.05 for all tests.

Data availability
The datasets analysed in the current study are not publicly available due to privacy and ethical reasons but are available from the corresponding author on reasonable request.