key: cord-0976552-wehyx7ul
authors: Felix Garza, Z. C.; de Jong, S. P. J.; Gibson, J.; Han, A. X.; van Leeuwen, S.; de Vries, R. P.; Boons, G.-J.; van Hoesel, M.; de Haan, K.; van Groeningen, L.; van Willigen, H. D. G.; Wynberg, E.; de Bree, G. J. C.; Matser, A.; van der Hoek, L.; Prins, M.; Kootstra, N. A.; Eggink, D.; Nichols, B. E.; de Jong, M. D.; Russell, C. A.
title: Impacts of the COVID-19 pandemic on future seasonal influenza epidemic
date: 2022-02-06
journal: nan
DOI: 10.1101/2022.02.05.22270494
sha: e23599a4ddef38587a086ccb27b701bb06039b43
doc_id: 976552
cord_uid: wehyx7ul

Seasonal influenza viruses typically cause annual epidemics worldwide infecting 5-15% of the human population. However, during the first two years of the COVID-19 pandemic, seasonal influenza virus circulation was unprecedentedly low with very few reported infections. The lack of immune stimulation to influenza viruses during this time, combined with waning antibody titres to previous influenza virus infections, could lead to increased susceptibility to influenza in the coming seasons and to larger and more severe epidemics when infection prevention measures against COVID-19 are relaxed. Here, based on serum samples from 165 adults collected longitudinally before and during the pandemic, we show that the waning of antibody titres against seasonal influenza viruses during the first two years of the pandemic is likely to be negligible. Using historical influenza virus epidemiological data from 2003-2019, we also show that low country-level prevalence of each influenza subtype over one or more years has only small impacts on subsequent epidemic size. These results suggest that the risks posed by seasonal influenza viruses remained largely unchanged during the first two years of the COVID-19 pandemic and that the sizes of future seasonal influenza virus epidemics will likely be similar to those observed before the pandemic.

The incidence of seasonal influenza has been unusually low since the start of the COVID-19 43 pandemic in early 2020 5,6 , with cases reported to WHO remaining >80% below historical 44 averages as of January 2022 2,7 . This dramatic reduction is likely due to non-pharmaceutical 45

interventions aimed at reducing transmission and spread of SARS-CoV-2 8,9 , which are also 46 effective in limiting exposure to seasonal influenza viruses. The global lull in influenza virus 47 circulation during the past two years and consequent lack of immune stimulation has led to 48 widespread concerns of increased susceptibility to seasonal influenza viruses in the 49 population due to waning immunity, potentially resulting in larger and more severe epidemics 50 in upcoming seasons [9] [10] [11] . Previous studies of antibody titres to seasonal influenza viruses 51

prior to the COVID-19 pandemic showed that antibody tires against influenza A viruses 52 typically wane to half peak levels 3.5-10 years after infection 12-14 . However, evidence is 53 lacking on how antibody immunity against seasonal influenza viruses has changed during the 54 near-absence of seasonal influenza in the COVID-19 pandemic and the impact this could 55

have on future influenza epidemic size. 56

To investigate how the lack of influenza virus circulation since the start of the COVID-19 57 pandemic has impacted antibody levels against seasonal influenza viruses, we measured 58 antibody titres, based on haemagglutination inhibition (HI), to representative strains of there was negligible influenza virus circulation ( Fig. 1a and Extended Data Fig. 2 ). 72

Differentiating the year-on-year individual HI titre distributions by titre rises that are 73

indicative of recent influenza virus infection (≥4-fold increase, ≥2 log2 units), showed that 74 influenza A and B virus infections were most common in individuals with low antibody titres 75

in the year prior to infection ( Fig. 1c and Extended Data Fig. 2) . Overall, the HI titre 76

distributions of the cohort remained largely unchanged over the study period, including 77 during the COVID-19 pandemic. perpetuity.

preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in

The copyright holder for this this version posted February 6, 2022. ; https://doi.org/10.1101/2022.02.05.22270494 doi: medRxiv preprint with waning rates previously reported for adults 12,14 . This indicates that substantial waning of 103 immune protection against seasonal influenza viruses occurs at timescales that are 104 substantially longer than the lull in seasonal influenza virus circulation during the first two 105 years of the COVID-19 pandemic. We also calculated waning rates using HI titres from the 106 same ACS individuals but only for the period after the start of the COVID-19 pandemic, i.e. 107 2020-2021 (Fig. 1d , Extended Data Fig. 2 , and Extended Data Tables 1 and 2 ). There was 108 also no significant waning of HI titres against any of the viruses during this period. 109

To extend our observations beyond the ACS cohort, we also measured antibody titres to the 110 same representative viruses in serum samples collected in mid-2020 and mid-2021 from a 111 longitudinal cohort of adult COVID-19 patients who were confirmed not to be vaccinated 112 against seasonal influenza viruses in 2020 (Extended Data Fig. 1b) corresponding to a one-sided probability of a 2-fold error of approximately 6 −10%. Taken  125 together, these results suggest that there have only been negligible changes in antibody titres 126 to seasonal influenza viruses among adults since the start of the COVID-19 pandemic. 127

The lack of HI antibody titre waning suggests that immunity to seasonal influenza viruses in 128 adults is unlikely to have declined substantially during the first two years of the pandemic. 129

However, previous work showed that waning in children might be different from adults and 130

could have an impact on susceptibility to infection 14 . While not possible to investigate 131 waning in children in our cohorts, historical lulls in circulation of particular influenza virus 132 (sub)types and their impact on subsequent epidemics have the potential to offer insights into 133 how changes in population immunity, or lack thereof, could impact seasonal influenza 134 epidemics in the post-COVID-19 pandemic period. 135

Prior to the COVID-19 pandemic, seasonal influenza virus circulation was highly 137 heterogeneous with influenza epidemics typically being dominated by one or two of the four 138 seasonal influenza viruses. This heterogeneity led to frequent periods of 1-3 years where 139 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in season, we multiplied a country's ILI burden by the proportion of isolates attributed to each 178 type and subtype in each season. We then divided this number by the total ILI burden across 179 all ten seasons and multiplied this number by the total number of seasons to calculate relative 180 epidemic sizes. In these estimates, a relative size of one corresponds to the mean number of 181 influenza virus infections in a single season for a given country, irrespective of type and 182 subtype. Epidemic sizes lower than 0.1, indicating very small or absent epidemics, were 183 observed for 28%, 23%, and 37% of country-seasons for A/H3N2, A/H1N1pdm09 and 184 influenza B viruses, respectively (Fig. 2d) . 185

To investigate the effect of periods of low influenza virus circulation on epidemic (sub)type 186 composition, we calculated the probability of an influenza virus (sub)type's dominance as a 187 function of years since previous dominance, where we defined dominance as a (sub)type 188

accounting for >30% of a season's isolates (Fig. 2e) . The probability of a (sub)type's 189 dominance increased with greater number of years since previous dominance. However, this 190 analysis also implies that there were periods of up to three years where an influenza (sub)type 191 did not dominate in the past, indicating that periods of low to absent circulation of particular 192 seasonal influenza viruses were also common before the COVID-19 pandemic. Mean 193 epidemic sizes for each influenza virus (sub)type increased with time since dominance (Fig.  194 2f). However, these increases were strongly related to probability of (sub)type dominance 195 (Fig. 2e) and epidemic sizes varied substantially since last dominance, suggesting that the 196 overall influence of time since dominance on epidemic size is relatively small. 197

Epidemic sizes of each (sub)type have a negative relationship with incidence of that specific 198

(sub)type in the preceding year with large successive epidemics being rare (Fig. 2g left  199 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for this this version posted February 6, 2022. ; https://doi.org/10.1101/2022.02.05.22270494 doi: medRxiv preprint column). However, this negative relationship largely disappears when taking into account the 200 cumulative incidence of each (sub)type two years into the past (Fig. 2g , right column), with 201 epidemics of high and low incidence both being likely to occur if preceded by years of low-202

to-mid incidence. 203

For each number of years since dominance, there is a striking degree of clustering of 204 epidemic size across countries by season (Fig. 2f) . To investigate this hypothesis, we constructed a Bayesian hierarchical model that uses these 220 effects as predictors of the (sub)type-specific size of seasonal influenza epidemics. 221

Specifically, we considered models that individually include the number of years since 222 previous dominance of that (sub)type, the size of that (sub)type's epidemic in the previous 223

year, or the sum of that (sub)type's incidence in the previous two years as predictors. In these 224 models, the season effects correspond to the predicted 'base size' of a (sub)type's epidemic, 225

given that the previous dominance was in the previous year, given that there was no 226 circulation in the previous year, or given that there was no circulation in the previous two 227 years, respectively. These season effects are modulated by the effects of prior circulation to 228 yield an epidemic's predicted size. Years since dominance, size in the previous year and the 229 sum of previous two seasons' sizes individually had non-trivial effects on epidemic size (Fig.  230 2h). However, between-season differences with regard to season effects were consistently of 231 substantially greater magnitude than any of the predictors related to prior incidence across all 232 model formulations, suggesting that season-specific factors unrelated to the absence or 233

presence of viral circulation in the previous year(s) dominate epidemic size. Previous 234 epidemic size appeared to have a moderate effect on epidemic size. This effect substantially 235 decreased when looking at the sum of the two previous epidemic sizes (Fig. 2h) . 236 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in Because epidemic sizes clustered strongly by season, which might obscure the effect of prior 237 incidence in a model with season effects, we also considered models without season effects. 238

Here, estimates on the impact of absence or presence of circulation in prior years on size 239

were higher, but the differences between seasons with regard to season effects ('base sizes') 240 remained far greater (Fig. 2h) . For example, even when using parameters estimated from a 241 model without season effects, the model predicts that the size of an A/H3N2 epidemic with 242 the mean estimated season effect and previous dominance three years prior is smaller than an 243 epidemic with the largest estimated season effect and A/H3N2 domination in the previous 244 season. Models that included season effects exhibited much better predictive performance 245 than models without season effects (Extended Data Fig. 4a) . Additionally, models that 246 included prior incidence of the opposite subtype had substantial effects of opposite sign, 247

implying that the estimated effects of prior incidence might reflect a combination of prior 248

incidence and effects of heterosubtypic competition (Extended Data Fig. 4b) . These results 249

suggest that inherent season-specific effects have more substantial effects on epidemic size 250 than (sub)type-specific patterns of prior circulation. 251

Taken together, the lack of changes observed in the pattern of measured antibody titres 252 against seasonal influenza viruses and nearly two decades of epidemiological data suggest 253 that the near-absence of seasonal influenza virus circulation during the first two years of the 254 COVID-19 pandemic is unlikely to result in substantially larger influenza epidemics in the 255 years to come. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The haemagglutination inhibition activity of all serum samples was tested in an 377 haemagglutination inhibition assay as described elsewhere 24,25 using two replicates per 378 sample for A/H1N1, B/Yamagata, and B/Victoria, and one single measurement for A/H3N2. 379

Briefly, the haemagglutination titre of each of the four viruses was determined by doing a 380 two-fold serial dilution of 50mL of each virus stock and adding 50mL of PBS and 25mL of 381 1% turkey red blood cells (tRBCs) to each well, followed by one hour incubation at 4°C and 382 the reading of the haemagglutination patterns. The virus stocks were then diluted to a 383 concentration of 4 haemagglutination units (HAU). The diluted viruses were then incubated 384 with 50mL of two-fold serially diluted serum, in a total volume of 75mL for 30 minutes at 385

37°C. The initial dilution used for the serial dilution of the serum was 1:20 of the RDE 386 treated serum. After the incubation step, 25mL of 1% turkey red blood cells were added to 387 the serum-virus mix and incubated at 4°C for one hour. The haemagglutination inhibition 388 patterns were then read out and used for the calculation of antibody titres. Due to the known 389 inefficient agglutination of tRBCs by recent A/H3N2 viruses, we used glycan remodelled 390 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for this this version posted February 6, 2022. ; https://doi.org/10.1101/2022.02.05.22270494 doi: medRxiv preprint turkey red blood cells expressing appropriate receptors for recent A/H3N2 viruses 26 for the 391 implementation of the assay for the A/H3N2 virus stock. 392

Two approaches were used in selecting data from which to determine antibody waning rates. 394

Firstly, we used samples from the RECoVERED cohort for the years 2020 and 2021, where 395 all participants are confirmed to have not received an influenza vaccination between the two 396 sample collections and no natural influenza infection can be safely assumed given the near 397 absence of influenza in the Netherlands during this period. Second, we used the ACS data for 398 the 5 years from 2017 to 2021. Individuals who experienced a 4 or greater fold increase in 399 titre between consecutive visits for a particular strain had their data for the strain removed in 400 order to remove the obscuring effects of vaccination and infection. The advantage of the 401 former approach is the certainty regarding infection and vaccination status, the latter, 402 however, allows a longer period of time over which to observe potentially subtle antibody 403 waning dynamics. 404

True antibody titre log2 HI, " ! , as opposed to that measured by haemagglutination inhibition 406 assay, T i , is a continuous variable which we assume, for every individual, i, decays with time, 407 t, as 408

Where c i are individual specific initial titres and a is the shared waning rate. 410

If serum dilutions could be performed in arbitrarily small increments, we assume the point at 411

which haemagglutination would be observed to cease, Tobs, to be distributed normally about 412 the true value: 413

(2) 414 where ϵ shall be referred to as the "measurement error". Instead, with discrete dilutions in 415 increments of 1, the probability of measuring T ∈ {0, 1, 2...8} is the probability that Tobs falls 416 between T and T − 1. Thus, the measurement probability is given by: 417

(1, " , ) (T, " , ) − ( " − 1, " , ) 1 − (8, " , ) < 1 1 ≤ < 8 ≥ 8

(3) 418

where Φ(x, μ, σ) is the cumulative distribution function of the normal distribution. 419 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in We assumed that the season effects a * , constrained to positive values, are distributed 491 according to a common mean µ . and common standard deviation σ . . 492

We put weakly informative priors on the standard deviations, the mean season effect and the 494 main effect. 495

We also ran a model with relative size as outcome and relative size in the previous year as the 500 predictor (N=180 for each (sub)type) or the sum of the two previous years size (N=160 for 501 each (sub)type). We use the same model specification and priors as for the size-years since 502 dominance model, but we replace the predictor with the relative size in the previous season, 503

or with the sum of relative size in the two previous seasons. For A/H3N2 and 504

A/H1N1pdm09, we also ran these models with time since dominance, previous season 505 epidemic size and sum of two previous seasons epidemics sizes of the other subtype as 506 predictor (N=198, 188 for A/H3N2, A/H1N1pdm09 for time since dominance, N= 180, 160 507 for each subtype for previous year, previous two years' sum, respectively). For all these 508 above models, we also ran the same models without season effects, i.e. with a single value for 509 the intercept, for which the prior is equal to the prior of the mean season effect in the model 510 with season effects. 511

These models were each run for each subtype individually, for 3000 iterations, discarding the 512 first 1000 as burn-in, with four independent chains. All models were fit using MCMC in Stan 513 v2.21.0, with convergence assessed by inspection of Rhat (< 1.05), effective sample size (> 514 200) and the trace plots. We compared models with and without season effects using leave-515 one-out cross-validation 27 . 516

No statistical method was used to predetermine sample size. Data were not randomized nor 518 analysed in a double-blinded manner. The haemagglutination inhibition activity of all serum 519 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for this this version posted February 6, 2022. ; https://doi.org/10.1101/2022.02.05.22270494 doi: medRxiv preprint samples was tested in an haemagglutination inhibition assay using two replicates per sample 520 for A/H1N1, B/Yamagata, and B/Victoria, and one single measurement for A/H3N2. Specific 521 details regarding amount of data are indicated in the figure captions. For parameter estimates 522 95% credible intervals were considered as the significant bounds and were calculated from 523 the 2.5th and 97.5th percentiles of the MCMC traces. For Fig. 2e , error bars correspond to 524 95% confidence interval from an exact two-tailed binomial test for proportions. 525

Further information on research design is available in the Nature Research Reporting 527

Summary linked to this paper. 528

Accession codes for GISAID data used in this paper are provided as supplementary 530

information files. Raw de-identified hemagglutination inhibition data and raw surveillance 531 data downloaded from WHO FluNet and FluID can be found in the project GiHub repository 532

(https://github.com/AMC-LAEB/waning-immunity-to-flu). Biological materials are available 533

for study via the Amsterdam Cohort Studies on HIV infection and AIDS (ACS) and the Viro-534

immunological, clinical and psychosocial correlates of disease severity and long-term 535 outcomes of infection in SARS-CoV-2 -a prospective cohort study (RECoVERED). 536

Custom scripts used for data analysis and modelling are available at GitHub 538 https://github.com/AMC-LAEB/waning-immunity-to-flu. 539 perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for this this version posted February 6, 2022. ; https://doi.org/10.1101/2022.02.05.22270494 doi: medRxiv preprint

The evolution of seasonal influenza viruses

World Health Organization (WHO)

The importance of influenza vaccination during the COVID-19 263 pandemic

A.X.H., Z.C.F.G. and C.A.R. were supported by ERC NaviFlu (No. 818353) . J.G. and C.A.R.