A novel, high-performance, low-volume, rapid luciferase immunoprecipitation system (LIPS) assay to detect autoantibodies to zinc transporter 8

Abstract Background Zinc transporter 8 autoantibodies (ZnT8A) are thought to appear close to type 1 diabetes (T1D) onset and can identify high-risk multiple (≥2) autoantibody positive individuals. Radiobinding assays (RBA) are widely used for ZnT8A measurement but have limited sustainability. We sought to develop a novel, high-performance, non-radioactive luciferase immunoprecipitation system (LIPS) assay to replace RBA. Methods A custom dual C-terminal ZnT8 (aa268-369; R325/W325) heterodimeric antigen, tagged with a NanoluciferaseTM (Nluc-ZnT8) reporter, and LIPS assay was developed. Assay performance was evaluated by testing sera from new onset T1D (n = 573), healthy schoolchildren (n = 521), and selected first-degree relatives (FDRs) from the Bart’s Oxford family study (n = 617; 164 progressed to diabetes). Results In new-onset T1D, ZnT8A levels by LIPS strongly correlated with RBA (Spearman’s r = 0.89; P < 0.0001), and positivity was highly concordant (94.3%). At a high specificity (95%), LIPS and RBA had comparable assay performance [LIPS pROC-AUC(95) 0.032 (95% CI: 0.029–0.036); RBA pROC-AUC(95) 0.031 (95% CI: 0.028–0.034); P = 0.376]. Overall, FDRs found positive by LIPS or RBA had a comparable 20-year diabetes risk (52.6% and 59.7%, respectively), but LIPS positivity further stratified T1D risk in FDRs positive for at least one other islet autoantibody detected by RBA (P = 0.0346). Conclusion This novel, high-performance, cheaper, quicker, higher throughput, low blood volume Nluc-ZnT8 LIPS assay is a safe, non-radioactive alternative to RBA with enhanced sensitivity and ability to discriminate T1D progressors. This method offers an advanced approach to current strategies to screen the general population for T1D risk for immunotherapy trials and to reduce rates of diabetic ketoacidosis at diagnosis.


Introduction
Type 1 diabetes (T1D) results from progressive autoimmunemediated destruction of insulin-producing beta-cells in pancreatic islets.The most effective biomarkers used to predict the development of T1D are four major islet autoantibodies that recognize beta-cell antigens with high specificity: autoantibodies to endogenous insulin (IAA), glutamic acid decarboxylase 65 (GADA), islet antigen 2 (IA-2A), and zinc transporter 8 (ZnT8A).Using these four markers, >90% of islet autoimmunity can be detected at the clinical onset of T1D [1,2], but these islet autoantibodies can be detected many years before diagnosis [3].During the prodrome of T1D, the development of islet autoantibodies appears to occur sequentially, spreading from one beta-cell antigen to multiple antigens [4][5][6].The risk of T1D correlates with the number of islet autoantibodies [7] and is highest in individuals with multiple autoantibodies (≥2), which confers up to 80% risk within 10-15 years in childhood [3].Despite the well-described islet autoantibody staging of T1D risk [8], progression rates can vary from months to decades [9].
Identified in 2007, ZnT8A are among the most recently discovered islet autoantibody markers and are present in ~60-80% of new-onset T1D, dependent on age at onset [10].Data from some birth-cohort studies of European descent have shown that in preclinical T1D, ZnT8A alongside IA-2A, are often not present at primary seroconversion and appear closer to onset, in children already positive for either GADA and/or IAA [11,12].Therefore, ZnT8A can be used to identify individuals at the greatest risk of progressing to T1D independent of other islet autoantibodies [11,[13][14][15].Thus, accurate detection of ZnT8A is pivotal in identifying high-risk individuals for enrolment in secondary intervention clinical trials aimed at delaying progression to T1D.
Current methods to detect ZnT8A use C-terminal ZnT8 (aa268-369) which includes a common single nucleotide polymorphism (SNP) site at aa325 that encodes arginine (R325) or tryptophan (W325).This polymorphic variant forms two major epitopes of ZnT8A giving rise to three specificities: R325-specific (ZnT8RA), W325-specific (ZnT8WA), and non-specific ZnT8A that bind ZnT8 independent of this SNP site [16].Antigen-specific radiobinding assays (RBAs) are well described and widely utilized to measure islet autoantibodies with individual laboratories electing to immunoprecipitate ZnT8A in serum utilizing monomeric (ZnT8-R325 or ZnT8-W325) or heterodimeric (ZnT8-R325 + ZnT8-W325) [ 35 S]-methionine labelled ZnT8 peptides.Whilst RBAs are still often regarded as the "gold-standard" method for islet autoantibody measurement, the use of radioisotopes is costly, labour-intensive, tightly regulated, has limited long-term sustainability, and cannot be applied to large-scale population studies or be easily implemented into clinical practice.As > 80% of new cases occur outside of T1D-affected families [17], an active part of current research is to develop highperformance, low-volume, high throughput, cost-effective, and non-radioactive assays with the capacity for automation to enable screening for T1D in the general population.
We previously developed a NanoLuciferase TM (Nluc)-IAA LIPS method which not only showed high concordance with RBA but had enhanced sensitivity in new-onset T1D and predictive utility in first-degree relatives (FDRs) that subsequently progressed to T1D [18].Therefore, we optimized a novel Nluc-tagged ZnT8-R325 + ZnT8-W325 dual heterodimer (Nluc-ZnT8) LIPS fluid-phase immunoassay using the superior Nluc/furimazine reaction [19] and, evaluated its performance and predictive utility for T1D in a large population-based UK T1D family cohort.

New-onset type 1 diabetes and first-degree relatives from the Bart's-Oxford family study
The well-characterized population-based Bart's Oxford (BOX) family study, established in 1985, has recruited and prospectively followed individuals under 21 years of age with newly diagnosed T1D and their FDRs within the former Oxford Regional Health Authority in the UK [20].The BOX study is currently approved by the South Central-Oxford C. Research Ethics Committee.Participants provided informed, written consent and the study was performed according to the principles of the Declaration of Helsinki.
Serum samples from BOX participants were selected based on available historical data from ZnT8-R325 and ZnT8-W325 monomeric RBAs and sufficient sample volume for testing Table 1.A total of 573 new-onset T1D samples (<3 months of diagnosis) were selected where sera were available to evaluate whether Nluc-ZnT8 LIPS had improved sensitivity [RBA positive n = 388 (67.7%);RBA negative n = 185 (32.3%); 320 male (55.9%); median age at onset 11.3 years (range 1.0-54.9)].FDRs of people with T1D are followed annually for the development of diabetes by questionnaire and were predominantly tested for ZnT8A in individuals positive for at least one other autoantibody [GADA/IA-2A/IAA by RBA and/or islet cell antibody (ICA; >20 Juvenile Diabetes Foundation Units (JDF-U) assay, previously described [21][22][23][24]].

The Nluc-ZnT8 LIPS assay methodology
Filtered Nuc-ZnT8 dual heterodimer antigen was further diluted to 4.0 × 10 6 (±0.2 × 10 6 ) LU/25 µl in TBST-0.15%Tween-20 (TBST-0.15%).In duplicate, 1 µl of serum was plated into a 96-deep well plate (Sarstedt, Nümbrecht, Germany) and incubated with Nluc-ZnT8 dual heterodimer antigen for 2.5 h at RT shielded from light.The remainder of the LIPS methodology, previously described [27], was followed before the detection of bioluminescence.To a final volume of 30 µl, 40 µl/well of Nano-Glo® Furimazine substrate assay reagent (1:50 NanoGlo® Furimazine substrate further diluted to a final concentration of 1:150 with TBST-0.15%) was injected into each well immediately before LU determination by a Bertold Centro XS3 luminometer using a standardized detection protocol; inject 40 µl/well, shake 5 s/ well, detect 2 s/well.The in-house logarithmic standard curve developed for the monomeric RBAs was used to determine and express ZnT8A binding as AUs.In the IASP2020 workshop, the AS95 was 78%.

Data transformation and statistical analysis
Due to the presence of all ZnT8A specificities [16], the maximum AU derived from both monomeric RBA was used to compare Nluc-ZnT8 LIPS and RBA by Spearman's rank correlation and Bland-Altman analysis.The sensitivity and specificity of the methods were assessed by the total area under the curve receiver operator curve (ROC-AUC) and partial ROC-AUC at 95% specificity (pROC-AUC(95)) analysis utilizing the new-onset T1D and healthy schoolchildren cohorts (pROC package for the R software [28]).Kaplan-Meir survival curve analysis was used to assess the predictive utility of the methods in FDRs and the Mantel-Cox log-rank test was used to compare survival between categories.Diabetes survival (%) between 5 and 20 years of study follow-up was reviewed.Unless stated otherwise, all graphs and statistical analysis were performed using the Prism v6 software (GraphPad Software, Inc., CA, USA; v. 9.1.0).An alpha value P < .05 was considered significant in all analyses.
Levels of ZnT8A detected by RBA and Nluc-ZnT8 LIPS were strongly correlated and ZnT8A positivity was highly concordant.S1.Overall, Bland-Altman analysis revealed that only 39 new-onset T1D (6.8%) were outside the 95% CI of agreement with Nluc-ZnT8 LIPS reporting lower ZnT8A levels compared with RBA (Supplementary Fig. S6A  and S6B).
In 521 healthy schoolchildren, levels of ZnT8A were also correlated between RBA and Nluc-ZnT8 LIPS [r: 0.46 (95% CI: 0.39-0.53),P < 0.0001; Fig. 1B] with ZnT8A status concordant in 96.0%(n = 500/521) between methods.Of 21 discrepant samples, 2.3% (12/521) were found positive by RBA but negative by Nluc-ZnT8 LIPS, and 1.7% (9/521) were found negative by RBA but were positive by Nluc-ZnT8 LIPS.None of the 21 discrepant samples were found positive for other islet autoantibodies and Bland-Altman analysis revealed that only 2 of all 521 (0.4%) healthy schoolchildren were outside the 95% CI of agreement (Supplementary Fig. 6C/D).The threshold for positivity for Nluc-ZnT8 LIPS was placed at the 97.5th percentile of this cohort (0.22AU).Nluc-ZnT8 LIPS and RBA were highly correlated and concordant, and at high specificities, assay sensitivity was comparable.Scatterplots comparing levels of ZnT8A binding (arbitrary units; AU) between Nluc-ZnT8 LIPS and monomeric RBAs in 573 newly diagnosed T1D patients (A) and 521 healthy schoolchildren (B) show that levels of ZnT8A detected by both assays were strongly correlated and ZnT8A positivity was highly concordant.
There was strong evidence to suggest a difference in AUC-ROC between the methods (P = 0.0007), offering an assay sensitivity at 95% specificity of 71.2% versus 68.6% in RBA.However, pROC-AUC(95) between methods were com- In relatives that were positive for both Nluc-ZnT8 LIPS and monomeric RBA, the 20-year diabetes risk was 59.8% (95% CI 44.7-74.4)(Fig. 3A; Table 2).There were only four relatives that the monomeric RBAs found positive that the LIPS assay did not identify, and, of these, two (50%) relatives slowly progressed to diabetes over a 20-year follow-up.Due to limited numbers, these two curves could not be robustly compared.In relatives that tested negative by both monomeric RBAs, Nluc-ZnT8 LIPS identified a small subset of 18 relatives that had a 20-year diabetes risk of 32.2% (95% CI −1.3 to 51.2) [Fig.3B; Table 2; 5/18 relatives later progressed to diabetes (range 6.1-24.1 years after sampling)], but this was not different from relatives found negative by both methods [20-year diabetes risk of 25.9.% (95% CI: 21.4-29.8),P > 0.05].
Across all Kaplan-Meir survival analysis, there was no clear difference in the performance of LIPS and RBA across all 5-year increments of study follow-up, except in relatives found single autoantibody positive by LIPS or GADA/IA-2A/ IAA RBAs (Table 2).

Discussion
A novel, low-volume, rapid, and high-performance fluidphase Nluc-ZnT8 LIPS method to detect ZnT8RA/ZnT8WA simultaneously was developed and validated in a large cohort of healthy schoolchildren, new-onset T1D, and FDRs to replace the conventionally used fluid-phase RBA.Nluc-ZnT8 LIPS and RBA were highly correlated and concordant, and at high specificities, assay performance was similar.Overall diabetes risk was comparable between relatives who were found positive or negative by either method.When stratified by RBA status, Nluc-ZnT8 LIPS did identify a small additional subset of 18 relatives, but diabetes risk was similar to relatives found ZnT8A negative by both methods.Nluc-ZnT8 LIPS positivity was useful in further stratifying diabetes risk in relatives found positive for other islet autoantibodies measured by RBA across a wide age spectrum.Furthermore, Nluc-ZnT8 LIPS ranked among the top assays for ZnT8A measurement in IASP2020 and showed good inter-laboratory concordance (unpublished data).Accordingly, the data presented in this study provides strong evidence that the Nluc-ZnT8 LIPS assay is a suitable replacement for the RBA in the context of T1D.
The Nluc-ZnT8 LIPS methodology described in this study after several optimization steps resulted in a high-performance non-radioactive assay that has a one-day turnaround time, requires very low sample volume (2 µl/duplicate), and has Diabetes risk (%; 95% confidence interval (CI)) at 5-year increments of study follow-up.A maximum of 20 years follow-up was reviewed where ≥100 FDRs remained.Across all Kaplan-Meier survival analysis, there was no clear difference in the performance of LIPS and RBA across all 5-year increments of study follow-up, except in relatives found single autoantibody positive by LIPS or by GADA/IA-2A/IAA RBAs.LIPS and RBA had the highest 20-year risk at 59.8%.Only four FDRs found positive by RBA were found negative by Nluc-ZnT8 LIPS but only two of these (50%) progressed to diabetes within 20 years of follow-up.Due to limited numbers, these two curves could not be robustly compared.(B) Nluc-ZnT8 LIPS status in FDRs found negative by RBA.Nluc-ZnT8 LIPS identified a small subset of additional FDRs that progressed to diabetes than RBA (20-year diabetes risk of 32.2%), but this was of comparable risk to relatives found negative by both methods (20-year diabetes risk of 25.9.%) a comparable methodology to the RBA with widely available reagents.This allows most laboratories who are running RBA or other similar immunoassay formats to easily implement the method with limited re-training requirements to dispense radiation use more readily.Nluc-ZnT8 LIPS assay optimization resulted in a quicker, cheaper, and more sustainable method than the RBA.The cost of reagents was reduced by decreasing the amount of protein A sepharose required, switching to an alternative method for antigen preparation, and further diluting the NanoGlo® Furimazine substrate.Currently, LIPS is approximately two-thirds cheaper than RBA and it is likely to become more cost-effective as the cost of radionuclides continues to increase.It is worth noting that ZnT8A levels were not comparable (but were highly correlated) between Nluc-ZnT8 LIPS and RBA.This probably indicates that the sera used to generate the standard curve behaved differently between methods compared with other samples tested, which resulted in a lower positivity threshold (0.22AU vs. 1.8AU in RBA at the 97.5th percentile).The Nluc-ZnT8 LIPS assay using a ZnT8-R325 + W325 dual heterodimer instead of monomeric ZnT8-R325/W325 peptides also allowed for simultaneous ZnT8A measurement without loss of sensitivity or specificity, which will also have cost-and time-saving implications to other laboratories with similar ZnT8A testing strategies.Collectively, these practical and cost-effective advantages did not negatively impact assay performance.
LIPS assay to detect GADA, IA-2A, ZnT8A, IAA, and autoantibodies to glycoprotein tetraspanin protein family member 7 (TSPAN7A) in T1D have been reported [18,[29][30][31][32].These studies have utilized Renilla luciferase, Gaussia luciferase (Gluc), or Nluc and report either comparable or increased sensitivity compared with widely used assays [RBA/enzyme-linked immunosorbent assay (ELISA)].Only one study has explored the measurement of ZnT8A using a ZnT8-R325 + W325 heterodimer by LIPS [31].Comparable to the present study, Ustinova et al. (2014) describe a liquidphase LIPS assay that can be conducted within 1 day and uses the same amount of neat sample, but there are a number of technical differences, particularly regarding the design and preparation of the antigen.First, the two designs of ZnT8 construct varies by both the luciferase and placement of the luciferase ZnT8-W325 + ZnT8-R325-Gluc heterodimer (Gluc-ZnT8) vs.
ZnT8-R325 + ZnT8-W325-Nluc-ZnT8-R325 + ZnT8-W325 dual heterodimer (Nluc-ZnT8).Second, the Gluc-ZnT8 was overexpressed in an insect cell line (Tn5), which required a 55-h incubation and subsequent procedures to exclude insect-derived proteins before use.Whereas, this Nluc-ZnT8 method used a cell-free in vitro expression system which took ~2 h before use.While the Gluc-ZnT8 protocol produced comparably high antigen yields to the Nluc-ZnT8 protocol (>1 × 10 9 LU/µl), the antigen was only stable for ~2 months at −80°C, but we observed stability ≥ 1-year at −80°C.Additionally, when the liquid-phase Gluc-ZnT8 LIPS assay was compared to the commercially available solid-phase RSR TM Limited ELISA (Cardiff, UK) in new-onset T1D and controls, assay concordance was agedependent (lower in children) and lower specificity (68-78%) was observed, even when a high antigen concentration (10-15 × 10 6 LU) was required to discriminate the populations.Age cut-offs in these populations were not reported; but using the RSR TM Limited ELISA, age-related positivity cut-offs for ZnT8A have been described in another study [33].We did not observe age-dependent effects, but the majority of individuals with new-onset T1D studied were aged <21 years and healthy schoolchildren were of adolescent age.Nevertheless, our study includes one of the largest studies of ZnT8A prevalence in new-onset T1D (within 3 months of diagnosis) [1] and healthy control populations reported, in which we did not observe low sensitivity or specificity.This could be due to the many advantages of the NanoGlo® luciferase/substrate coupled system (Promega) over other bioluminescent reactions such as Gluc [19]; lower autoluminescence, brighter glow-type bioluminescent signal with >2-h half-life, enhanced physical/chemical stability, and improved expression in mammalian cells with little evidence of protein/antigen modifications.
We found that for T1D risk prediction, the ZnT8A RBA and LIPS assay had comparable performance.However, in comparison to other islet autoantibody markers measured by RBA (GADA/IA-2A/IAA ± ICA), Nluc-ZnT8 LIPS positivity further stratified T1D risk in relatives found positive for other islet autoantibodies that included rapid (<5 years) and slow (>10 years) progressors who developed diabetes.This confirmed previous findings that ZnT8A detection is useful in assessing diabetes risk in relatives positive for at least one other islet autoantibody, increasing the discrimination of high-risk individuals [5,11,13,14].A limitation of the present study is that the populations studied were pre-selected and therefore, we cannot predict how Nluc-ZnT8 LIPS would perform as an initial screening test.Other limitations include the high prevalence of diabetes in autoantibody-negative relatives as well as, the collection of diabetes diagnosis by questionnaire.However, the high prevalence of diabetes diagnosis can be largely explained by the increasing diagnosis of T2D with ageover-study follow-up (80.0% of ZnT8A negative relatives who progressed to diabetes reported a T2D diagnosis).Nevertheless, Nluc-ZnT8 LIPS was able to discriminate diabetes risk in both single and multiple autoantibody-positive relatives, which is highly representative of a T1D diabetes classification.Similar to many other studies in T1D, another limitation of the present study, is that by selecting autoantibody-positive relatives, individuals who may have developed ZnT8A at seroconversion, or those with single ZnT8A responses, would not be accounted for.Based on current literature, there is a low prevalence of single ZnT8A responses at diagnosis, and before diagnosis in relatives appears rare [1,14].They are also more likely to be associated with lower diabetes risk, like other single autoantibody responses [3,21].However, the prevalence of ZnT8A in populations of lower T1D genetic risk is not clear.There are also emerging data of additional epitopes other than the C-terminal of ZnT8, that may represent an earlier stage of ZnT8A autoimmunity, preceding GADA/IAA responses [34].This requires further investigation.
While this study focused on T1D risk prediction in FDRs, emerging general population studies suggest that multiple autoantibodies are present in ~0.3% of subjects [35] versus ~3% in FDRs: data from >180 000 aged 2.5-45 years participating in TrialNet [36].The difference in autoantibody prevalence has tremendous cost and feasibility implications for whole population screening of islet autoimmunity.For the last 20-30 years, RBAs have dominated islet autoantibody detection and remain one of the most commonly performed assays.Whilst the RBAs are low-volume and highly sensitive, RBAs are not only unsustainable in the long-term but they are also costly and cannot be readily adapted to the demands of population screening.Emerging immunoassays have sought to increase the viability of such studies.Among LIPS assays other examples include ELISA, electrochemiluminescence (ECL), and agglutination-PCR (ADAP).There have been LIPS, ELISA, and ECL methods described for ZnT8A detection, and when compared to RBA or ELISA, show comparable or improved assay performance [31,37,38].All methods have inherent strengths and limitations that may benefit or restrict their use in research or clinical settings.A detailed discussion of these assays is beyond the scope of this manuscript, but automation of assays may be critical for large population screening.
Full automation is not possible for the Nluc-ZnT8 LIPS in its present format.A future avenue of this work would be to further adapt the assay to a plate-bound bridge-type ELISA for IgG/IgA/IgM detection, assess assay performance, and validate its compatibility in robotic systems.To achieve a plate-bound LIPS assay, large-scale ZnT8 protein production is needed which has been problematic, requiring stringent protocols for other immunoassay formats [39].This could increase experimental costs but remains a key step for highthroughput assays.Integration of multiple antigens for simultaneous detection of multiple autoantibodies (multiplex) in either fluid-or plate-bound LIPS format is also viable, with the possibility of integrating different luciferases with distinct emission spectra to discriminate autoantibody specificity in a single test (e.g.Nluc with Firefly luciferase).Multiplex formats of ELISA, ECL, and ADAP assays in the context of T1D have been described [40][41][42], with ECL and ADAP able to distinguish autoantibody specificity.
The development of next-generation high-performing nonradioactive immunoassays, like this one, and licencing of immunotherapeutic agents [43,44] in recent years, makes general population screening for early detection of at-risk individuals, reduced rates of individuals presenting with diabetic ketoacidosis at diagnosis, and primary/secondary intervention therapy to slow the rate of T1D progression an imminent reality.

FFBFigure 2 .
Figure 2. Measurement of ZnT8A by Nluc-ZnT8 LIPS or RBA has comparable diabetes risk in FDRs.§ Reference category for Mantel-Cox log-rank tests.****P < 0.0001.Positivity for ZnT8A measured by Nluc-ZnT8 LIPS (A) or RBA (B) had a much higher diabetes risk than those found ZnT8A negative by either method (P < 0.0001).Overall diabetes risk is comparable in FDRs found positive or negative by either Nluc-ZnT8 LIPS or RBA however, Nluc-ZnT8 LIPS identified a greater number of at-risk individuals at first sampling or first ZnT8A positive sample (67 vs. 53)

Figure 3 .
Figure 3. Nluc-ZnT8 LIPS ZnT8A status does not identify FDRs with higher diabetes risk when stratified by RBA ZnT8A status.§ Reference category for Mantel-Cox log-rank tests.NS: not significant.(A) Nluc-ZnT8 LIPS status in FDRs found positive by RBA.FDRs identified as positive in both Nluc-ZnT8 LIPS and RBA had the highest 20-year risk at 59.8%.Only four FDRs found positive by RBA were found negative by Nluc-ZnT8 LIPS but only two of these (50%) progressed to diabetes within 20 years of follow-up.Due to limited numbers, these two curves could not be robustly compared.(B) Nluc-ZnT8 LIPS status in FDRs found negative by RBA.Nluc-ZnT8 LIPS identified a small subset of additional FDRs that progressed to diabetes than RBA (20-year diabetes risk of 32.2%), but this was of comparable risk to relatives found negative by both methods (20-year diabetes risk of 25.9.%)

Table 1 .
New-onset T1D and FDRs selected from the BOX family study

Table 2 .
Diabetes risk in FDRs over 20-years of follow-up in the BOX study