Identification of host–pathogen-disease relationships using a scalable multiplex serology platform in UK Biobank

Certain infectious agents are recognised causes of cancer and other chronic diseases. To understand the pathological mechanisms underlying such relationships, here we design a Multiplex Serology platform to measure quantitative antibody responses against 45 antigens from 20 infectious agents including human herpes, hepatitis, polyoma, papilloma, and retroviruses, as well as Chlamydia trachomatis, Helicobacter pylori and Toxoplasma gondii, then assayed a random subset of 9695 UK Biobank participants. We find seroprevalence estimates consistent with those expected from prior literature and confirm multiple associations of antibody responses with sociodemographic characteristics (e.g., lifetime sexual partners with C. trachomatis), HLA genetic variants (rs6927022 with Epstein-Barr virus (EBV) EBNA1 antibodies) and disease outcomes (human papillomavirus-16 seropositivity with cervical intraepithelial neoplasia, and EBV responses with multiple sclerosis). Our accessible dataset is one of the largest incorporating diverse infectious agents in a prospective UK cohort offering opportunities to improve our understanding of host-pathogen-disease relationships with significant clinical and public health implications.


Supplementary Tables 24
Supplementary proteins as antigens 3,4 . Since 2005, Multiplex Serology has been used in many seroepidemiological studies not only to measure antibodies against HPV but also against a magnitude of other infectious agents. In this work, we transferred the methodology from non-magnetic to magnetic beads to allow for high-throughput testing of large studies such as the UK Biobank (n~500,000 participants). In the following, we describe the changes to the previously used and published protocol for Multiplex Serology on magnetic beads.

Coupling of glutathione-casein (GC) to magnetic Luminex beads
Glutathione-casein (GC) was produced as described 1 , and coupled to magnetic beads (MagPlex microspheres, Luminex) analogous to the previously described manual protocol for coupling of non-magnetic beads (SeroMAP, Luminex). In this work, we transferred the manual coupling protocol for non-magnetic beads to a KingFisher Flex device (Thermofisher 5400640) to allow for automatisation of the GC coupling to magnetic beads. The KingFisher Flex device uses a magnetic tip head to transfer the beads between different deep-well plates containing the coupling reagents. The device supports incubation and mixing phases fully replacing the usage of a centrifuge and shaker during the manual coupling procedure. For each of the 46 bead sets, 4 ml of beads (n= 50*10 6 / bead set) were coupled with GC using the 24-tip-magnetic head, and 24-deep-well plates (KF Flex 24-deep well plate, 94040470 ThermoFisher). The 46 bead sets were coupled with GC in two batches (I n=24, II n=22) on two subsequent days three months before the study samples were analysed with Multiplex Serology. Protocol: Automatic coupling was performed with a KingFisher Flex. This device is able to mix, collect, and transfer magnetic beads between 96-or 24-well plates using a magnetic tip head. For coupling of 4 ml (50*10 6 beads) magnetic beads we used the 24-tip-magnet head and 24 deep-well plates (maximum volume 5 ml). Beads were thoroughly collected when transferred between plates using two collection (30 seconds from the previous plate) and release (5 seconds fast release into next plate) cycles to minimize bead loss. Before collection, the beads were mixed in the previous plate for 1 min at medium speed. The coupling procedure was performed in the dark at room temperature.
1) 4ml of beads (n=50*10 6 beads) were transferred from original Luminex flasks to each one well of a 24deep-well plate. Beads were transferred from plate 1 to plate 2 as described above. 2) Activation 1: Beads were incubated in 3 ml activation buffer (100 mM Na2HPO4, pH 6.2). The program was paused to add activation reagents to the activation buffer already contained in plate 3 (step 3). After dispensing both reagents manually to plate 3, the beads were mixed for 1 min in plate 2, and then transferred from plate 2 to plate 3. 3) Activation 2: Beads were incubated in activation buffer (1.6 ml) plus EDC (200 µl 1-(3-Dimethylaminopropyl)-3-Ethylcarbodiimide (Pierce 22980) at concentration 50 mg/ml) and NHS (200 µl N-hydroxysuccinimide (Pierce 24500), at concentration 50 mg/ml in waterfree DMSO (Fluka 41647)) for 20 min. Every 2 minutes, a mixing step (30 s, medium speed) was conducted to stir up beads and prevent them from forming a pellet. Afterwards the beads were transferred to plate 4 (see step 4). 4) Coupling buffer 1: Beads were mixed for 1 min (medium speed) in 2 ml coupling buffer (50 mM MES, pH 5.0). Afterwards, the beads were transferred to plate 5 for a second washing step. 5) Coupling buffer 2: Beads were mixed for 1 min (medium speed) in 2 ml coupling buffer, and then transferred to plate 6 (step 6). 6) Binding: Beads were incubated with GC (250 µg/ml) in 2 ml coupling buffer for 2 h. Every 2 minutes, the beads were mixed for 30 s with medium speed. The beads were transferred to plate 7 (step 7) 7) Washing: Beads were washed in 2 ml washing buffer (PBS, 0.05% Tween 20, pH 7.4) and mixed for 1 min (medium speed) before they were transferred to plate 8 (step 8). 8) Storage: Beads were transferred into 2 ml storage buffer (PBS, 1 mg/ml Casein, 0.05% Sodium azide, pH 7.4).
Afterwards, the beads were collected in a manual procedure by rinsing each well (3x 2 ml of storage buffer).
Quality and homogeneity of the automatically coupled magnetic beads were confirmed (Supplementary Figure  2) and compared to a manually coupled non-magnetic bead set. No batch effects or systematic differences compared to the non-magnetic bead set were observed. Coefficients of variation of measured median fluorescence intensities (MFI) per bead set ranged between 4% (375 beads per bead set) and 12% (9000 beads per bead set), and coefficients of variation of bead counts ranged between 9% and 10%. Sufficient bead counts (i.e., >100) were measured for all bead concentrations except 375 beads per set. The quality of the automatically coupled magnetic beads was further validated using the pathogen-specific reference panels as described in Section 4. Assay validation.

In situ affinity purification of antigens on GC beads
The recombinant bacterial expression of GST-X-Tag fusion antigens and loading of antigens onto beads have been described in detail before 1, 3,4 . Loading of each antigen onto one magnetic bead set was not automated. In total, 45 bead sets were loaded with pathogen-specific antigens while one bead set was loaded with GST-Tag for background determination and subtraction.

Serum pre-incubation
UKB study sera were shipped to DKFZ on dry ice in a 1:5 serum pre-dilution in preincubation buffer containing Super Chemiblock Heterophile Blocking Agent (Merck Millipore), a proprietary augmenting reagent for removing Rheumatoid Factor and/or HAMA-like interfering antibodies to suppress unspecific signals thereby increasing the signal-to-noise ratio of the serological assay. This reagent pre-absorbs IgM antibodies and strongly reduces IgA antibody signals 5,6 . On the assay day, sera were further diluted to a serum dilution of 1:500 in the same preincubation buffer at DKFZ and incubated on a shaker at room temperature for one hour.
UKB samples were analysed using Multiplex Serology on six assay days in two subsequent calendar weeks. In every assay week, antigen loading onto magnetic beads took place on Monday, each of 20 96-well plates (92 samples, 4 plate controls) were tested on Tuesday-Thursday. Plates were measured on the next day (Wednesday-Friday) using two Luminex 200 readers.

Quality controls
At different stages during assay performance, measurements and data processing the following quality controls were included: 1) On each 96-well assay plate, four wells were used for quality controls. Three wells were used for control sera included on every plate to monitor plate handling (discussed in further detail in Section '5' below). In addition, one well per plate was used as blank (no serum) for background subtraction during data processing. 2) To assess and control for assay drift, two plates (n=184 samples) were processed on each assay day within one assay week. To be able to compare assay performance across both assay weeks, 2 plates (n=184 samples) analysed on the first assay day (Tuesday, week I) were also analysed on the first assay day of week II (Tuesday). For further information see section 'Hardware, software and data processing'. 3) The UKB sera were aliquoted in 96-well plates and further diluted (as described above) by lab personnel at DKFZ using manual multichannel pipettes on assay days. Pipetting errors, also including sera too viscous to ensure pipetting of correct sample volume, were documented. Corresponding sera were later excluded from the final data set and data analysis. 4) For each serum and antigen, at least 100 beads were measured to determine seroreactivity as median fluorescence intensities. When insufficient bead counts occurred during measurements, the corresponding sera were excluded from the final data set. 5) During aliquoting of sera at UKB, lab personnel (independent of DKFZ personnel) interspersed 107 spiked duplicate samples among regular samples. The DKFZ lab and data processing personnel was blinded to the study sample versus spiked duplicate status via pseudonymised IDs, and provided UKB with a complete final data set including the study samples and spiked duplicates. Quality control of the performance of blind spiked duplicates was performed by UKB personnel and included calculation of coefficient of variations within and between batches (see section 5. UK Biobank Multiplex Serology quality control)

Hardware, software and data processing
We used two Luminex 200 devices running with the Luminex XPonent software for measurements of magnetic beads. For each bead set, at least 100 beads / well, i.e. serum were, measured (DD gate: 7000-20,000; 90 seconds sample time out, sample volume 75µl).
Raw data were further processed and analysed using self-implemented scripts (SAS 9.4) for automated background subtraction and assignment of measurements and pseudonymised sample ID. Per antigen (i.e. bead set), mean background reactivity across all plates per assay week were subtracted. Furthermore, one of the 46 included bead set was not loaded with a pathogen-specific GST-X-Tag fusion protein, but only with a GST-Tag fusion protein for determination of serum-specific background (e.g. against E. coli proteins, GST-Tag or polystyrene beads). This background was subtracted from all serum-specific antibody measurements. MFI <0 were set to 1 MFI.
During data processing, sera with documented pipetting or measurement errors were excluded. In addition, for each serum the mean seroreactivity was calculated. In cases where no or insufficient pipetting of sample was suspected, measurement data were checked.
As described above, on each assay day, 184 samples, i.e. bridging panels, were re-measured to assess and account for potential batch effects (inter-assay-day, inter-assay-week. The sero-reactivity in each of these pairwise sets of bridging panels were assessed by scatterplots (slope, correlation) and seroprevalences, and correction factors were calculated.
Quality control analyses were performed using Microsoft Excel 2016. For graphical visualisation, R (version number 3.3.1, June 2016) and Microsoft Excel 2016 were used.

Assay validation
We have adopted the terms "monoplex" (an assay with one or multiple antigen:bead sets all relating to a single infectious agent), and "multiplex" (where multiple antigen:bead sets were present in the same reaction relating to multiple infectious agents) in our assay validation approach to address possible concerns related to assay inter-relatedness.
The validation was performed in a stepwise approach as described in Supplementary Figure 1. First, individual pathogen-specific Monoplex Serology assays were compared to established reference assays based on nonmagnetic beads 7,8,9,10 . Second, based on non-magnetic beads, assay performance of the pathogen-specific assays in both monoplex and multiplex format, i.e. comprising the full UKB Multiplex Serology antigen panel, was performed 4,8,9,11 . The first and second validation steps have been reported previously and are summarised in Supplementary Table 2. In the third step, pathogen-specific assay performance in multiplex format was validated by comparing their performance on non-magnetic versus magnetic beads. To compare pathogenspecific assay performance on non-magnetic and magnetic beads in multiplex format, sensitivity, specificity and Cohen's kappa were calculated in comparison to the reference assay used in Monoplex Serology validation 7,8,9,10 (Supplementary Figure 3). For all pathogen-specific Multiplex Serology assays, similar performance characteristics were observed on non-magnetic and magnetic beads. To further assess agreement of Multiplex Serology on non-magnetic versus magnetic beads in determining seropositivity, Intra-class Correlation Coefficients (ICCs) 12 were calculated using the package 'psych' in R 3.5.0. ICCs directly comparing the performance of Multiplex Serology on magnetic vs. non-magnetic beads were above 0.8 indicating good to excellent reliability, except for Tg where the ICC was 0.48 (Supplementary Table 3). This was likely due to lower specificity but higher sensitivity on non-magnetic versus magnetic beads (Supplementary Figure 2).
A possible concern in multiplex serology assay development is the potential for cross-reactivity between antigen:bead sets in the same reaction, i.e. different intensities measured for the same sample depending on whether the reaction was performed in a monoplex or multiplex format. If that was true, one would expect to observe an increased background in a multiplexed reaction that would alter the number of samples being classified as seropositive or negative. Since we had access to multiple sera with defined 'true' status through using reference assays we used these sera to determine whether the sensitivity and specificity metrics estimated in the monoplex assays were affected by multiplexing 7,8 The only agents found to have significant discrepancies in estimated metrics between monoplex and multiplex were Toxoplasma gondii (Tg) and human T-lymphotropic virus-1 (HTLV-1). For Tg, these differences were likely observed owing to insufficient volumes remaining for 12% of reference sera, i.e. different numbers of sera tested monoplex versus multiplex validation. Furthermore, lower sensitivity of the multiplex assay was expected for Tg given the low signal intensities observed in earlier validation work (at 1:100 serum dilution) and the increased dilution used in the UKB multiplex panel (1:1000) 8 . For HTLV-1, an increase in the background for the Gag antigen was observed for unknown reasons. This was the only antigen which required a significant increase in cut-off to determine seropositivity in the multiplex assay.

UK Biobank Multiplex Serology quality control
The specific methodology used to test the serum samples from 9,695 individuals from UKB is described in the main text, and Supplementary Materials Section 4. Of the total 10,110 serum samples assayed, 29 samples (0.3%) were excluded from analyses (2 highly viscous sera, 8 pipetting errors, 8 were incorrectly diluted and 11 with insufficient bead counts at the reading step). The remaining samples represented either study participants (n=9,695), blind-spiked duplicates (n=107) or repeat assessment encounter samples (n=277) or samples initially destined for repeat assessment samples invalidated due to other errors (n=2). After exclusion of invalid samples (n=29), we observed little evidence of batch effects as demonstrated by comparing antigen-specific MFI values for duplicate samples measured across 2 weeks (Supplementary Figure 4), with a summary of all comparisons shown in Supplementary Table 4. Coefficients of variation (CV) measured for blind spike duplicates for 107 individuals were calculated using the logarithmic method of Bland and Altman 11 . CV was calculated both for all individuals, and among seropositives only. It was hypothesised that the CV calculations including all samples in those infectious agents less prevalent in the population would be more likely to be subject to more random noise, especially those with a median reactivity below the lower limit of quantitation (approx. 30 MFI at 1:1000 serum dilution). Among all samples, CVs ranged between 7.8% (Hp CagA) and 35.4% (Hp GroEL) with a median across all antigens of 17%. Among seropositives only (not available for the HCV and HTLV-1 antigens, as well as HIV-1 Env and HPV-16 E6), CVs ranged between 0.2% (Ct PorB) and 12.0% (Ct pGP3) with median of 3.5% (Supplementary Table 5).
In addition, control sera were included on each assay plate to monitor individual plate handling by assessing inter-plate variance. As an example, seroreactivity for one of the control sera for all antigens and all plates is shown in Supplementary Figure 5. Median intra-day CVs ranged between 11% and 23% (median 16%), median intra-week CVs ranged between 12% and 22% (median 19%) and CVs across all assay days ranged between 16% and 26% (median 21%).
Samples for repeat assessments were available for 277 individuals tested in a blinded fashion. Seroconversion rates within the 3-5 year interval between baseline and repeat assessments ranged from 0%-10.5% but seroreversion rates of 0-9.4% were also observed (Supplementary Table 6). For those infectious agents where a single antigen was used to define seroprevalence, over 70% of samples demonstrating discordant results between repeat samples had MFI values clustering within a single standard deviation around the cut-off for each antigen. Thus, these mismatches are likely a result of the CV of the assay and the remaining 10-30% of discordant samples are likely to represent true seroconverters or reverters. For one measured antigen (Hp CagA) bead-antigen complexing was accidentally omitted in the second week, therefore sero-reactivity and seropositivity estimates for this infectious agent were derived for only 50% of the tested samples.

Multiplex Serology seroprevalence calculations from antigen reactivity data
The final 45 antigens tested in the UKB panel are listed in Supplementary Table 2. The interpretation of the individual antigen positivity data to calculate seroprevalence estimates for each agent was as follows: Herpes simplex virus 1 (HSV-1): Reactivities against gG (1gG; cut-off: 150 MFI) antigen alone were used for prevalence estimates 7 .
Varicella zoster virus (VZV): gE and gI antigens were co-bound to a single bead set and the combined reactivity (cut-off: 100 MFI) was used to calculate seroprevalence estimates 7 .
Human herpes virus 7 (HHV-7): Reactivities against U14 antigen alone were used for prevalence estimates (cutoff: 100 MFI). However, as no reference assay was available for HHV-7, this classification should be considered preliminary. Traditionally, there have been concerns of cross-reactivity between HHV-6 and HHV-7 species antibody responses. We tested for evidence of cross-reactivity and found very little correlation between magnitude of responses against HHV-6 and HHV-7 antigens in our UKB dataset (r<0.2 for all inter-agent comparisons; Supplementary Figure 6).
Hepatitis B virus (HBV): Seropositivity was defined in individuals seropositive against HBc (cut-off: 100 MFI) AND HBe (cut-off: 150 MFI) antigens 8 . HBS antigen was not included and thus measures are likely to represent an unbiased picture of HBV exposure rather than being confounded by vaccination.
HTLV-1: Seropositivity was defined in individuals seropositive against Gag (cut-off: 1500 MFI) AND/OR Env (cut-off: 150 MFI) antigens. Increased reactivities were observed for HTLV-1 Gag responses that could be accounted for by increasing the cut-off 8 .
Hp: Seropositivity was defined if individuals were seropositive against at least two antigens from a total of CagA (cut-off: 400 MFI), VacA (cut-off: 100 MFI), OMP (cut-off: 170 MFI), GroEL (cut-off: 80 MFI), Catalase (cut-off: 180 MFI) and Urease (cut-off: 130 MFI). This classification was only possible for the samples tested in the first week (n=4,871) as discussed above. Therefore all estimates for Hp are only provided for the samples tested in the first week.
For all infectious agents, the calculated seroprevalence estimates were compared to those in the general literature using a search method on PubMed undertaken using the following terms using CMV as an example:

(((cytomegalovirus[Title/Abstract] OR CMV[Title/Abstract]) AND (seroprevalence[Title/Abstract] OR IgG[Title/Abstract] OR antibody[Title/Abstract] OR prevalence[Title/Abstract]) AND (united kingdom[Title/Abstract] OR UK[Title/Abstract]OR europe[Title/Abstract])))
Only studies with more than 500 individuals were considered and studies were excluded if they primarily described individuals with a diagnosis of HIV-1 (except in the case of estimating HIV-1 seroprevalence), prior solid organ or bone marrow transplantation, or those receiving haemodialysis. Reported seroprevalence estimates were prioritised for UK or European adult populations between 40-70 years of age, but other populations were considered if no data from European studies could be identified.

UK Biobank Data
Data was collected from UKB participants using a variety of methods and at separate encounters. Only defined phenotypes thought to be linked with infectious disease exposures or outcomes were curated and tested for association with tested antibody traits. These included: Age: Date of birth collected from NHS central registries and confirmed by the participant at the questionnaire stage of baseline recruitment (Field 33). Age was calculated by subtracting date of birth from the date of first attendance (Field 53).
Sex: Collected from NHS central registries and confirmed by the participant at the questionnaire stage of baseline recruitment (Field 31).

Ethnic Group:
Collected during touch-screen questionnaire at baseline recruitment encounter (Field 21000). Individuals were divided into categories of 'White', 'Asian', 'Black', or 'Other' (if not one of the other three) based on self-report.
Townsend deprivation index (TDI): Calculated immediately prior to participant joining UK Biobank. Based on data from the preceding national census and the participant's postcode area at the time of invitation to join. Since the data was positively skewed, individuals were divided into quintiles (Field 189).

Tobacco smoking status:
Collected during touch-screen questionnaire at baseline recruitment encounter as 'smoking status' question (Field 20116) classifying individuals as a 'Current', 'Previous' or 'Never' smoker.
Alcohol drinking status: Collected during touch-screen questionnaire at baseline recruitment encounter as 'alcohol drinker status' question (Field 20117) classifying individuals as 'Current', 'Previous' or a 'Never' smoker.

Same-sex intercourse ever (sameSI):
Collected during touch-screen questionnaire at baseline recruitment encounter as 'ever had same-sex intercourse' (Field 2159). Individuals were classified into groups dependent on whether they answered 'yes' or 'no' to this question. Disease case: Defined using self-report of diagnoses collected during touch-screen questionnaire (Field 20002) at any patient encounter for multiple sclerosis (MS; code 1261) or coeliac disease (code 1456). Cases of cervical cancer and CIN were defined using ICD9 (2331 and 1808; and 1809 respectively), and ICD10 (D06, D06.1, D06.7; and C53 respectively) codes derived from National Health Service cancer registry data. ICD9 codes were derived from Field 40013 and ICD10 in Field 40006.

Testing for association between demographic variables and infectious agent serostatus
Features such as sex, age, reported ethnicity, number of individuals per household, social deprivation status and lifetime number of sexual partners have been associated with risk of infection with a number of agents. We therefore set out to test associations between such factors as measured in UKB and infectious agent serostatus using a step-wise approach. Firstly, univariate logistic regression models testing association between the measured exposure variable and seropositivity were compared to a null model using the likelihood ratio test. Any comparisons that were found to be significant using this approach (P<0.05) were then taken forwards for testing in a multivariable logistic model including other measured variables as covariates. After inclusion of all other measured variables, no infectious agent serostatus demonstrated any significant association either alone or in a trend with the number of individuals in a household therefore this was not analysed further. Supplementary Tables 7-12 provide statistics of association between antigen or infectious agent seroprevalence and demographic factors including sex, age, ethnicity, TDI and LSP as detailed further in the main text.

Genetic association tests in UKB
Data from direct genotyping and imputation release 3 of UKB data were used in the genetic analyses. Only individuals and SNPs passing basic quality control (QC) steps (missingness or heterozygosity in any batch) were included in the final genetic analyses 23 . Of the 9,695 individuals with demographic and antibody data, 9,611 individuals had genetic data available that was of sufficient quality for analysis after QC. Three genetic variants were tested for association with a variety of traits as described in Supplementary Table 12. Two of these variants were imputed for which we assessed INFO scores as measures of imputation accuracy.
The genetic data was used for three purposes. The first was to attempt to replicate associations between three genetic variants in the major histocompatibility complex reported with JCV and MCV serostatus, and EBV EBNA1 magnitude of antibody response (described as a continuous variable). We tested only variants reported in a single publication performed in a population of European ancestry using association statistics available from Table 2 within that particular manuscript 24 . Since the vast majority of individuals in our UKB dataset were of reported British or European ancestry we used all individuals with genetic data for this analysis. To account for other individuals of non-European ancestry and to account for cryptic relatedness we used two linear mixedmodel association testing software systems. To test association with directly genotyped variants we used GCTA (v1.26.0) 25 . This software calculates a genetic relatedness matrix and includes this matrix as a random effect covariate. We used the entire post-QC dataset of 734,447 variants across the autosomes in addition to age (in years) and sex as fixed effect covariates. For imputed variants we used BOLT-LMM (v2.3.1) to account for imputation probabilities that calculates an estimate of the genetic relatedness matrix from the genotyped variants 26 . In both cases only beta values and standard errors were calculated even if the trait was binary and therefore odds ratios were calculated for UKB traits that were tested in a binary format using the Shiny app (http://cnsgenomics.com/shiny/LMOR/). Secondly, we tested for association with all imputed variants and log10 normalised quantitative responses against EBV VCAp18 using BOLT-LMM (v2.3.1) including age and sex as fixed effect covariates. This antigen was selected as a marker of ever being exposed to the viral capsid protein of EBV rather than a marker of replication that, alone, may be less sensitive. The resultant Manhattan plot from these autosome-wide associations is included in Supplementary Figure 5.
The third set of analyses involved testing for genetic correlation between antibody response traits and disease outcomes (MS and coeliac disease). Firstly, association statistics for HLA alleles were calculated using a large independently collected case-control dataset of MS (International MS case-control) using only data across the MHC region 27 . For the International MS dataset including 17,610 cases and 30,129 controls, genotyped variants across the MHC were used to impute HLA alleles using SNP2HLA (vv1.0.2) 28 and the Type I Diabetes Genomics Consortium reference dataset. The resultant 'hard' called alleles were then tested for association between cases of MS and matched population controls within each individual population using logistic regression in PLINK (v1.9) 29 . The statistics from each population were then combined using a fixed effects meta-analysis in METASOFT (v2.0.1) 30 . These statistics were compared to those equivalent for imputed HLA alleles within the UKB dataset. HLA alleles were imputed for the 9,611 individuals with genotype and antibody data available again using UKB genotype data using SNP2HLA and the T1DGC reference dataset. The alleles were tested for association with quantitative EBNA1 and VCAp18 levels using GCTA with the GRM calculated using genotypes across all autosomes and controlling for sex and age. To formally test for genetic correlation between traits we used a refined dataset of 6,265 unrelated individuals of strict European descent in the UKB subset described here to determine association estimates against antibody responses. Then, using the bivariate GREML analysis toolset in GCTA (v1.26.0) 31 we compared the association of genotyped variants alone against 11,236 individuals of unrelated European descent selected from the remainder of the complete UKB dataset (1,236 cases of MS and 10,000 random controls with no record of MS). This analysis was undertaken using either all autosomal data or only variants across the MHC. The same analysis was undertaken for coeliac disease using an additional total of 11,468 individuals (1,468 cases and 10,000 additional randomly selected European controls).

Supplementary Figures
Supplementary Figure 1: Proposed steps of validation for infectious agents of highlighted relevance to public health.
A) The Multiplex Serology assay for each infectious agent was tested individually (i.e. 'Monoplex') against reference sera defined as positive or negative using established reference assays to estimate performance characteristics. B) All assays were performed simultaneously in a multiplex reaction using the reference sera available from step A) to determine whether performance characteristics were adversely affected by multiplexing. This step was performed on non-magnetic and magnetic beads. C) The Multiplex Serology platform was used for 9,695 samples available from UK Biobank to estimate seroprevalence of the multiple infectious agents across the UK using cut-offs defined in the early validation phase. Simple associations between seroprevalence and lifestyle, environmental and genetic risk factors were then tested using the refined cut-offs. *: Reference sera were available for 11 out of the total 20 infectious agents included in the panel. kappa: Cohen's kappa.
Supplementary Figure 2: Quality control of automatically GC-coupled magnetic bead sets. The 46 automatically coupled bead sets (from left to right, bead set 12 to bead set 90) were compared to a manually coupled non-magnetic bead set (SeroMap bead set 60, rightmost). Per bead set, 375 beads (orange line) to9000 beads (dark blue line) were loaded with recombinantly expressed GST-Tag protein, and incubated with a biotinylated mouse monoclonal antibody directed against the Tag. Bound GST-Tag protein was detected using StrepPE and quantified using a Luminex 200 reader. The red line divides the 46 bead sets into the coupling batches I and II. Upper panel: Measured median fluorescence intensities (MFI) per bead set, indicating homogeneous coupling of GC to the 46 bead sets. Coefficients of variation in reactivity ranged between 4% (375 beads per bead set) and 12% (9000 beads per bead set). Lower panel: Measured bead counts, indicating sufficient bead counts (i.e., >100) for all bead concentrations except 375 beads per set. Coefficients of variation of bead counts ranged between 9% and 10%.

Supplementary Tables
Supplementary

Infectious agent ICC* (95%CI)
HSV  4.4 *GMT: geometric mean titre for 107 individuals with blind spiked duplicate samples available from the same encounter. **coefficient of variation calculated only when both duplicates were defined as seropositive. For some antigens that were observed at low frequencies, these calculations were not possible (NA).