A rapid antibody screening haemagglutination test for predicting immunity to SARS-CoV-2 variants of concern

Background Evaluation of susceptibility to emerging SARS-CoV-2 variants of concern (VOC) requires rapid screening tests for neutralising antibodies which provide protection. Methods Firstly, we developed a receptor-binding domain-specific haemagglutination test (HAT) to Wuhan and VOC (alpha, beta, gamma and delta) and compared to pseudotype, microneutralisation and virus neutralisation assays in 835 convalescent sera. Secondly, we investigated the antibody response using the HAT after two doses of mRNA (BNT162b2) vaccination. Sera were collected at baseline, three weeks after the first and second vaccinations from older (80–99 years, n = 89) and younger adults (23–77 years, n = 310) and compared to convalescent sera from naturally infected individuals (1–89 years, n = 307). Results Here we show that HAT antibodies highly correlated with neutralising antibodies (R = 0.72–0.88) in convalescent sera. Home-dwelling older individuals have significantly lower antibodies to the Wuhan strain after one and two doses of BNT162b2 vaccine than younger adult vaccinees and naturally infected individuals. Moverover, a second vaccine dose boosts and broadens the antibody repertoire to VOC in naïve, not previously infected older and younger adults. Most (72–76%) older adults respond after two vaccinations to alpha and delta, but only 58–62% to beta and gamma, compared to 96–97% of younger vaccinees and 68–76% of infected individuals. Previously infected older individuals have, similarly to younger adults, high antibody titres after one vaccination. Conclusions Overall, HAT provides a surrogate marker for neutralising antibodies, which can be used as a simple inexpensive, rapid test. HAT can be rapidly adaptable to emerging VOC for large-scale evaluation of potentially decreasing vaccine effectiveness.


Plain language summary
The aim of this study was to rapidly investigate the immune responses after SARS-CoV-2 infection and vaccination of younger adults and the elderly. Antibodies are proteins produced by the immune system that are released into the bloodstream and help fight infections. A simple method using red blood cells obtained from blood was developed and used to detect antibodies to SARS-CoV-2. This test was able to measure protective antibodies to several variants of concern. The elderly had lower antibody responses after vaccination. Two vaccinations induced a broader antibody response to viral variants, similar to the response induced following Covid-19. This antibody detection method could be used as a finger prick test to rapidly detect specific antibodies to emerging variants and enable quick identification of individuals who might benefit from a booster vaccination.
T here is increasing evidence that neutralising antibodies to the receptor binding domain (RBD) on the severe acute respiratory syndrome coronavirus-2 virus (SARS-CoV-2) spike protein represent an immunological correlate of protection 1 . SARS-CoV-2 evolution has been rapid with the ancestral virus and emerging variants of concern (VOC) straining global health care systems. These VOC (alpha (B.1.1.7) 2 , beta (B.1.351) 3 , gamma (P.1) 4 , delta (B.1.617.2) 5 and the recent omicron (B.1.1.529) 6,7 show increased transmissibility, can escape pre-existing immunity and reduce vaccine effectiveness [8][9][10] , with breakthrough infections reported in COVID-19 vaccinees with low neutralising antibodies 11 . There is a need for a rapid low-cost surrogate neutralisation assay, which can be used at a low biosafety level. This assay could be used for large-scale screening to identify vaccinees potentially susceptible to emerging VOC and who would benefit from a booster vaccine dose.
The neutralisation assay with live native virus is the gold standard for evaluating antibodies to VOC 1 . However, neutralising assays are difficult to standardise across laboratories, are time consuming, expensive and require high containment. Therefore, antibody binding and pseudotype virus assays are widely used to study antibody responses [12][13][14] , but still require specialised laboratory facilities.
Here, we correlate the low-cost rapid hemagglutination test (HAT) 15 with neutralisation of the ancestral Wuhan-like strain in two large independent cohorts of infected patients. Further, we confirm the correlation between HAT and neutralising antibodies to VOC. In the HAT assay, the RBD domain is linked to a monomeric anti-erythrocyte single domain nanobody. When polyclonal serum antibodies bind to the RBD they cross-link and agglutinate the erythrocytes, which can be read visually after one hour. The HAT has a specificity of >99% for detection of convalescent antibodies after polymerase chain reaction (PCR) confirmed infection 15,16 . For influenza, a correlate of protection (COP) has been defined as a haemagglutination inhibition (HAI) titre of 40 for 50% protection from infection. If a similar COP could be established for HAT, it would allow simple standardised evaluation of susceptibility to SARS-CoV-2 infection and waning vaccine responses to VOC to guide public health policies.
Initially, we establish that HAT titres correlate with neutralising antibodies. We then use the HAT to investigate the antibody responses in 719 individuals consisting of home-dwelling older vaccinees (80-99-year-olds) and younger adults, in both those vaccinated with mRNA (BNT162b2) and in naturally infected individuals to the Wuhan-like virus. With VOC HAT we confirm that HAT titres can be used as a surrogate marker for neutralising antibody titres in vaccinated or infected individuals. Finally, we show that the HAT is readily adapted to finger prick testing.

Study participants
Norwegian vaccine and infection cohorts. A cohort of convalescents 415 infected individuals was prospectively recruited during the first Wuhan (pre-alpha) and delta pandemic waves in Bergen, Norway to compare the serological assays used in this study as described in [17][18][19] . For the comparison of vaccine and infection cohorts in Bergen Norway, we prospectively recruited two different age groups (home dwelling older and healthy younger adults) who received two doses of BNT162b2 mRNA COVID-19 vaccine at a 3-week interval during January 2021, and compared them to a group of 307 naturally infected individuals infected (1-89, median 47 years) with the Wuhan-like virus (D614G spike mutation) in February to April 2020 17,19 (Table 1). The older vaccinee group consisted of 96 home-dwelling elderly (80-99 years, median 86), 89 (92.7%) of whom were seronegative and 7 had previous SARS-CoV-2 infection with detectable prevaccination antibodies. The younger adult group consisted of 316 vaccinees (23-77 years, median 38) of whom 309 adults had no history of confirmed SARS-CoV-2 PCR test. Four younger vaccinees were not vaccinated on day 21; they received their second vaccination at day 19 (n = 1), or day 23 (n = 2) or day 24 (n = 1). Seven younger individuals had previous SARS-CoV-2 infection and pre-existing antibodies. This study is compliant with all relevant ethical regulations for work with humans and conducted according to the principles of the Declaration of Helsinki (2008) and the International Conference on Harmonization (ICH) Good Clinical Practice (GCP) guidelines. All Bergen subjects provided written informed consent before inclusion in the study, which was approved by the Western Norway Ethics committee (#118664 and #218629, NIH ClinicalTrials.gov Identifier: NCT04706390). Demographics (gender, age), PCR test results and COVID-19-like symptoms were recorded in an electronic case report form (eCRF) in (REDCap ® (Research Electronic Data Capture) (Vanderbilt University, Nashville, Tennessee). Clotted blood samples were collected on the day of vaccination, 3 weeks after receiving the first and 3-5 weeks (mean 55 days, standard deviation ± 5 days) after the second vaccine doses or 3-10 weeks after confirmed infection. Sera were separated and stored at −80°C and heat-inactivated for one hour at 56°C before use in the serological assays.
UK convalescent cohort. Informed signed consent was obtained from 420 blood donor in the NHS Blood and Transplant cohort for purposes of clinical audit, to assess and improve the services and the research, and specifically to improve knowledge of the donor population. The use of these anonymised donor samples to assess neutralising antibody levels using different assays was approved by the National Blood Supply Committee for Audit and Research Ethics of National Health Service Blood and Transplant Research and Audit Committee (BS-CARE; BSCR20047 and BSCR20051).
Finger-prick and venous blood comparison. For the comparison of finger-prick and venous blood, participants were recruited from Oxford University Hospitals NHS Foundation Trust when they were attending the research clinic with the Oxford Protective T Cell Immunology for COVID-19 (OPTIC) Clinical Team. Written informed consent was obtained from participants with different past infection and vaccination status. Seventy-eight paired finger-prick blood and venous blood in EDTA tubes were taken at the same time and analysed on the same day by the HAT assay. Human study protocols were approved by the research ethics committee at Yorkshire & The Humber-Sheffield (GI Biobank Study 16/YH/ 0247).
Haemagglutination test (HAT). The haemagglutination test (HAT) 15  2) (125 ng/ well) were pre-mixed and 50 µl added per well. Negative controls (PBS) and positive controls (monoclonal antibodies CR3022 and EY-6A) were included in each run. Plates were incubated to allow red blood cells to settle for 1 hr and were read by tilting the plate for 30 s and photographing. Positive wells agglutinated and the HAT titre is defined as the last well in which the teardrop did not form. Partial teardrops were scored as negative.
The IH4-RBD reagents for each VOC were standardised by showing that agglutination of red cells occurred at the same endpoint dilution (~16 ng/well) of the well characterised human monoclonal antibody EY6A 15,20 for each VOC at a working dilution of IH4-RBD of 2 ug/ml (100 ng/well in 50 ul). All the RBDs of the VOC share the conserved class IV epitope recognised by EY6A.
RBD positive sera were run in an additional ELISA, where the ELISA plates were coated with SARS CoV-2 spike protein (Wuhan, 100 ng/well). Sera were serially diluted in duplicate in a 5-fold dilution, starting from a 1:100 dilution, and the ELISA plates were incubated with diluted serum for 2 h at room temperature. Bound IgG antibodies were detected and measured as described for the RBD screening ELISA.
Positive controls were serum from a hospitalized COVID-19 patient with the pre-alpha virus and CR3022 22 , whereas pooled pre-pandemic sera (n = 128) were used as a negative control 18 . The mean endpoint titre was calculated for each sample. Samples with no detectable antibodies were assigned a titre of 50 for calculation purposes.
Pseudotype-based neutralisation assay. The pseudotype-based neutralisation assay was performed in biosafety level 2 laboratory in Bergen, Norway. The SARS-CoV-2 pseudotype virus was generated by co-transfection lentiviral vectors pHR'CMV-Luc, pCMVRΔ8.2, and pCMV3 construct encoding the Wuhan or delta spike protein into HEK293T cells as previously described 23 . The protease TMPRSS2 and human ACE2 encoding constructs were transfected into HEK293T to make target cells for the neutralisation assay. The lentiviral vectors and TMPRSS2-encoding constructs were a kind gift from Dr. Paul Zhou, Institute Pasteur of Shanghai, China. The ACE2-encoding construct was a kind gift from Dr. Nigel Temperton, University of Kent, UK. The SARS-CoV-2 Wuhan and delta spikeencoding constructs were purchased from Sino Biological. Serum samples were heat inactivated at 56°C for 60 min, analysed in serial dilutions (duplicated, starting from 1:10). The SARS-CoV-2 pseudotype viruses corresponding to 20,000 to 200,000 relative luciferase activity (RLA) were mixed with diluted sera in 96-well plates and incubated at 37°C for 60 min. Afterwards, ACE2-TMPRSS2 cotransfected HEK293T cells were added into 96-well plates and cultured for 72 h. RLA was measured by a BrightGlo Luciferase assay according to the manufacturer's instructions (Promega, Madison, WI, USA). The pseudotype-based neutralization (PN) titres (IC 50 and IC 80 ) were determined as the reciprocal of the sera dilution giving 50% and 80% reduction of RLA, respectively. Negative titres (<10) were assigned a value of 5 for calculation purposes.
Virus strains. The Wuhan-like strain used in the microneutralisation and virus neutralisation assays in Bergen Norway was the clinical isolate; SARS-CoV-2/Human/NOR/Bergen1/2020 (GISAID accession ID EPI_ISL_541970) and at Public Health England, UK the isolate England/02/2020 24   Cells were further incubated with biotinylated goat anti-rabbit IgG (H+L) and horseradish peroxidase (HRP)-streptavidin (Southern Biotech). The reactions were developed with o-Phenylenediamine dihydrochloride (OPD) (Sigma-Aldrich).
The MN titre was determined as the reciprocal of the serum dilution giving 50% inhibition of virus infectivity. Negative titres (<20) were assigned a value of 5 for calculation purposes. The MN assay for the 420 convalescent UK samples was conducted in a certified Biosafety Level 3 as previously described at Public Health England (PHE), UK 24 . using the virus England/ 02/2020. Sera were heat inactivated at 56°C for 60 min, before analyses in duplicate serial dilutions (starting from 1:20), and mixed with 100 TCID 50 viruses in 96-well plates and incubated for 1 h at 37°C. Then, the cell suspension was added to the virus/ antibody mixture 24 and incubated at 37°C for 22 h. Cells were fixed and permeabilized before staining for NP antibodies, then biotinylated goat anti-rabbit IgG, followed by Extravidinperoxidase. The reaction was developed with OPD and MN titres calculated, as described above.
At Oxford, UK the detection of antibodies to the Wuhan-like and VOC (alpha, B.1.1.7 and beta, B.1.351) used the method described in 25 . Briefly, quadruplicate serial dilutions of serum were preincubated with appropriate SARS-CoV-2 for 30 min at room temperature, then Vero CCL81 cells were added and incubated at 37°C, 5% CO 2 for 2 h. A carboxymethyl cellulosecontaining overlay (1.5%) was added, monolayers were fixed and stained for the nucleocapsid (N) antigen or spike (S) antigen using EY2A and EY6A monoclonal antibodies, respectively. After development the number of infectious foci were counted by ELISpot reader. Data were analysed using four-parameter logistic regression (Hill equation) in GraphPad Prism 8.3.
Virus neutralisation assay. The virus neutralisation (VN) assay was performed in a certified Biosafety Level 3 facility in Bergen, Norway 18 . Serum samples were tested against a clinically isolated virus: SARS-CoV-2/Human/NOR/Bergen1/2020 as previously described 18 . Briefly, serum samples were heat inactivated at 56°C for 60 min, analysed in serial dilutions (duplicated, starting from 1:20), and mixed with 100 TCID 50 viruses in 96-well plates and incubated for 1 h at 37°C. Mixtures were transferred to 96-well plates seeded with Vero cells. The plates were incubated at 37°C for 4-5 days, all wells were examined under microscope for cytopathic effect (CPE). The VN titre was determined as the reciprocal of the highest serum dilution giving no CPE. Negative titres (<20) were assigned a value of 5 for calculation purposes.
The delta virus neutralisation assay was performed at the University of Oxford, UK as previously described 26 . Briefly, serial two-fold serum dilutions from 1:20 were incubated with 50 TCID 50 virus in 96-well plates for 1 h before addition of 20,000 Vero E6 TMPRSS2 cells per well. Plates were incubated for 3 days before staining with amido black and CPE read by eye. Negative titres (<20) were assigned a value of 5 for calculation purposes.
Statistics and reproducibility. The two-tailed Mann-Whitney U test with 95% confidence level was used to compare ranks in HAT titres between the older and adult vaccinees. The non-parametric two-tailed Spearman R correlation with 95% confidence interval was used to investigate the correlation between the antibody titres from different serological assays. All analyses were conducted in GraphPad Prism version 9.20.

Results
Previous studies have shown that the HAT titre correlates with SARS-CoV-2 RBD binding and ACE2 blocking antibodies 15,27,28 , and identified high titre (>100) neutralising sera with a sensitivity of 76.5% 28 . First, we used the World Health Organisation (WHO) approved human SARS-CoV-2 standards panel to confirm the relationship between spike specific binding and neutralising assay to the HAT assay (Table 2). Second, we investigated the relationship between endpoint HAT titres and neutralising antibodies using three neutralisation assays in convalescent sera from SARS-CoV-2 infected individuals from the first pandemic wave (pre-alpha) and the ongoing delta-wave in Bergen, Norway [17][18][19] . We then confirmed the results in an independent UK cohort 24 .
We extended these results by comparisons to a pseudotype neutralisation (PN) assay, and a classical live virus neutralisation (VN) assay with complete inhibition of Cytopathic Effect (100% CPE) as its endpoint (Fig. 1a, d, e) 18 . The correlation of HAT and PN titres were significant (p < 0.0001) at 50% (R = 0.79) and 80% (R = 0.78) IC (Fig. 1a, d). Confirming our previous results, the VN titres correlated with HAT titres (R = 0.74, p < 0.0001) (Fig. 1e). HAT titres ≥ 40 detected 100% of samples with VN titres ≥ 20, but the PPV fell to 54% consistent with the classical VN assay having the more rigorous endpoint.
Confirmation of correlation between neutralisation and HAT titres. As interlaboratory variation has been reported for neutralisation assays, we confirmed the significant correlation between HAT and MN titres (R = 0.88, p < 0.0001) in an independent UK collection of 420 convalescent samples (Fig. 1c, Table 1, Supplementary Fig. 2). In close agreement with the Bergen cohort, a positive HAT ≥ 40 detected 98% of samples with MN IC 50 titre ≥ 20, with PPV of 87%. Similarly, for identification of high titre sera a HAT > 480 identified 75% of sera with MN IC 50 > 100 with PPV 86%. In summary, the HAT titres highly correlated with neutralisation titres in two independent laboratories showing the utility of HAT as a rapid and inexpensive surrogate for the neutralisation test.
Evaluation of HAT antibody responses in older and younger vaccinees. Older adults have carried the burden of COVID-19 throughout the pandemic with increased risk of hospitalizations and death, and are prioritised for vaccination, although most vaccine licensure trials have excluded the oldest (>85 years old) 29 .
As proof of principle, we used the HAT to investigate the Wuhanlike antibody responses in seronegative healthy younger adults (n = 309, median 37 years) and older home-dwelling adults (n = 89, 80-99 years, median 86 years) after the BNT162b2 mRNA COVID vaccine and in individuals naturally infected with the Wuhan-like strain (n = 307, median 47 years) 19 (Table 1, Fig. 2). A HAT titre of ≥40 was used as a cut-off to assess the propor of vaccine responders to the Wuhan-like virus 15 . Only 31% of older subjects responded after the first vaccination compared to 74% of younger vaccinees (Fig. 2a, Table 3). After the second dose, 78% of the older vaccinees had HAT titre of ≥40 compared to 94% of infected individuals and 99% younger vaccinees. Older people also had a significantly lower magnitude of response than younger adult vaccinees after both the first and second vaccine doses, with the exception after second dose against alpha (Table 4). In summary, the older adults had a blunted response after one dose of mRNA vaccine and required the second dose to increase the magnitude of the response.
Development of variant of concern reagents for HAT. Variants of concern have amino acid changes in their spike protein, and importantly in their RBD which may allow escape from neutralising antibodies. The alpha variant rapidly became the dominant strain in early 2021 2 with beta and gamma dominating in some geographical areas, and was subsequently replaced by the highly transmissible delta variant in 2021 30 . We developed HAT reagents for the VOC as they arose and confirmed a strong correlation between the alpha (R = 0.79, p < 0.0001) and beta (R = 0.89, p < 0.0001) in a UK set of naturally infected and vaccinated donors 25 (Fig. 1h). As the delta variant dominates in Norway, we collected convalescent sera from from 37 infected individuals to confirm the relationship between HAT and pseudotype and virus neutralisation assays. A good correlation was observed for both pseudotype (R = 0.82, p < 0.0001) and virus neutralisation assays (R = 0.72, p < 0.0001) (Fig. 1f, g).
HAT antibodies to variants of concern in older vaccinees. We then investigated the breadth of the VOC response in vaccinees and infected subjects. Older vaccinees had the lowest number of responders and lower cross-reactivity after both one and two vaccinations. The second vaccination boosted the number of responders in older adults, from 20-29% to 58-76% to VOC (Fig. 2e, i, m, o). In older and younger vaccinees that responded to VOC, there was good cross-reactivity to alpha and delta, but less so to the beta and gamma in all groups after two doses of vaccine or infection. A similar but higher response pattern to different VOC was observed in infected individuals, with 92% to alpha, 82% to delta, 75% to beta and 68% to gamma compared to responses in 96-100% of younger adult vaccinees. In summary, two doses of mRNA vaccine or natural infection induced higher responses in younger adults to VOCs than in older vaccinees.
Vaccine response in previously infected younger and older individuals. Natural infection induces higher titres of SARS-CoV-2 specific antibodies in older individuals than in younger adults 17 . Previously infected older subjects (n = 7, median age 87 years), none of whom had been hospitalised, had higher prevaccination HAT titres to the Wuhan-like virus than previously infected younger adults (n = 7, median age 38 years). Previously infected older and younger adults developed high Wuhan-like and alpha cross-reactive antibody titres after one vaccine dose (Fig. 2c, d), although lower responses to other VOC. Crossreactive titres were boosted in some of the older and healthy vaccinees after the second vaccination (Fig. 2g, h, k, l, o, p, s, t). In summary, previously infected older individuals develop high antibody titres after one vaccine dose comparable to healthy younger adults, which contrasts with the suboptimal antibody responses in SARS-CoV-2 naïve older vaccinees.
The use of HAT as a point of care fingerprick test. For the HAT to be implemented at low biosafety level and in resource limited settings, a fingerprick test using autologous patient erythrocytes could be used to rapidly identify populations with low titres to Wuhan-like and VOC SARS CoV-2 viruses. As advised by the Infectious Diseases Society of America guideline on serological   Table 4 The haemagglutination test (HAT) antibody response to the Wuhan-like virus and variants of concern after one and two doses of mRNA vaccine and after SARS-CoV-2 infection in seronegative younger and older adults.   testing 31 , we confirmed the comparability of fingerprick and venous blood for reactivity to all VOC (Fig. 3). The HAT can thus be rapidly adapted to test for antibodies to emerging VOC for large-scale screening of fingerprick blood samples with autologous erythrocytes.

Discussion
The rapid evolution of SARS-CoV-2 VOC, particularly delta and omicron, with increased transmissibility and the possibility of escape from vaccine induced immunity, represents a considerable threat. There is a need for a low-cost rapid serological assay which can be used for large-scale screening globally without requiring specialised laboratory equipment to rapidly identify populations susceptibility to VOC. HAT is simple to perform, requires no special equipment, and can be done at the point-of-care in virtually any setting using a fingerprick sample. The HAT IH4-RBD VOC reagents are freely available for research 15 . Inter-laboratory comparability can be guaranteed by including HAT titrations on WHO approved standard sera (as shown in Table 2). We demonstrate the versatility of the HAT in analysing susceptibility to VOC in home-dwelling older vaccinees showing the importance of two vaccine doses to achieve good cross-reactive antibody titres in older adults who have not been previously infected. Older and high-risk individuals were prioritised in the very first rounds of vaccination early in 2021 in Europe and America. Depending upon the decay in antibody titres over time, the HAT could be used to rapidly identify individuals who may need a booster vaccine dose to mount efficient antibody responses to VOC. We found the oldest age group had a decreased breadth of cross-reactive antibodies to VOC after the first vaccine dose, particularly to the beta and gamma viruses, in agreement with escape from neutralising antibodies 10 . Although vaccination induced cross-reactive antibodies against delta, milder break through infection with this variant in vaccinated subjects is becoming an increasing problem 11 . Reports of very high viral load during delta infections 32,33 may necessitate higher antibody titres to provide sterilising immunity and prevent infection.
In previously infected adults, only one dose of vaccine seems to be required to produce high levels of cross-reactive antibodies against the VOC 16,27 . Extending these findings, we found that in previously SARS-CoV-2 infected older adults, only one dose of vaccine was required to mount strong anamnestic responses, similar to younger vaccinees 34-40 . Caveats to our study are that most convalescent blood samples from our naturally infected cohort were collected during the first SARS-CoV-2 Wuhan-like wave, but we did include a small subset (n = 37) of delta infected individuals. Strengths are that we have confirmed the relationship between HAT and several neutralisation assays in two large cohorts in independent laboratories, showing that the relationship holds for VOCs, and included 719 individuals either infected and/or vaccinated, aged up to 99 years old. To our knowledge, this is the first study reporting antibody cross-reactivity to four VOC in this older age group.
Neutralising titres of between 10 and 30 in humans 1 , depending upon the assay, have been reported to predict 50% protection from symptomatic infection, and much lower levels to protect against severe infection. Although the absolute HAT titres correlating with protection are not yet known, we demonstrated that the HAT titres correlated with neutralisation titres, and thus provide a surrogate test for neutralising antibodies. We suggest that a positive HAT titre of 40-80, equivalent to 1:40 dilution of whole blood obtained by fingerprick, would correlate with neutralising titres 10-30, and would predict protection. A prospective study to test this predicted relationship between HAT titres and protection is now warranted. The HAT may also aid in evaluating and licensing of new COVID vaccines.
We predict that the lower HAT titres against VOC will lead to a more rapid decline in protective efficacy against variants, thus requiring booster vaccinations. The emergence of the highly infectious and transmissible delta and more worrying omicron VOC which have caused breakthrough infections in vaccinees highlights the importance of real-time cross-reactivity studies. Monitoring of population susceptibility of both previously infected subjects and vaccinees to VOC with increased transmissibility through simple serological assays can guide public health policy.

Data availability
Source data behind figures are available in the Supplementary Data file.