A highly specific and sensitive nanoimmunosensor for the diagnosis of neuromyelitis optica spectrum disorders

A precise diagnosis for neuromyelitis optica spectrum disorders (NMOSD) is crucial to improve patients’ prognostic, which requires highly specific and sensitive tests. The cell-based assay with a sensitivity of 76% and specificity of 100% is the most recommended test to detect anti-aquaporin-4 antibodies (AQP4-Ab). Here, we tested four AQP4 external loop peptides (AQP461–70, AQP4131–140, AQP4141–150, and AQP4201–210) with an atomic force microscopy nanoimmunosensor to develop a diagnostic assay. We obtained the highest reactivity with AQP461–70-nanoimunosensor. This assay was effective in detecting AQP4-Ab in sera of NMOSD patients with 100% specificity (95% CI 63.06–100), determined by the cut-off adhesion force value of 241.3 pN. NMOSD patients were successfully discriminated from a set of healthy volunteers, patients with multiple sclerosis, and AQP4-Ab-negative patients. AQP461–70 sensitivity was 81.25% (95% CI 56.50–99.43), slightly higher than with the CBA method. The results with the AQP461–70-nanoimmunosensor indicate that the differences between NMOSD seropositive and seronegative phenotypes are related to disease-specific epitopes. The absence of AQP4-Ab in sera of NMOSD AQP4-Ab-negative patients may be interpreted by assuming the existence of another potential AQP4 peptide sequence or non-AQP4 antigens as the antibody target.

assay (CBA) recommended by the International Panel for NMO Diagnosis 1,8 . CBA has an average sensitivity of 76% and specificity of 100%, thus failing to detect AQP4-Ab in 24% of the patients with NMOSD clinical manifestations 8 . This failure could be caused by: (i) undetectable serological levels of AQP4-Ab; (ii) reactivity with a different AQP4 sequence; or (iii) non-AQP4 antigen recognition [9][10][11][12] . New approaches with more sensitive methods are therefore needed for NMOSD diagnosis, which may include nanoimmunosensors such as those developed for detecting a biomarker for demyelinating diseases [13][14][15][16][17] . Indeed, sensors exploiting atomic force microscopy (AFM) may be sufficiently sensitive to diagnose patients for which AQP4-Ab is not detectable 18 . In this paper, we report an AFM nanoimmunosensor to detect interaction forces between samples of patients and AQP4 peptides, rather than with antigens as in the CBA method. Using peptides brings a series of advantages, mostly related to the simplicity in nanoimmunosensor assembly since it does not require protein expression on a cell surface. Also, the use of peptides allows for epitope mapping 19,20 . These molecules including immunogenic AQP4 peptides located on the astrocyte surface, known as loop A, loop C, and loop E, in NMOSD-related nomenclature were explored to identify AQP4 epitopes [20][21][22] . In order to verify which peptide would be specific for AQP4-Ab, we screened four AQP4 peptides from the extracellular loop (AQP4 61-70 , AQP4 131-140 , AQP4 141-150 , and AQP4 201-210 ). These peptides were chosen because they are located in the extracellular regions of AQP4 protein 6 , where the interaction between AQP4-Ab and AQP4 protein is more likely to occur. The nanoimmunosensor assay permitted to identify AQP4 61-70 as highly specific to distinguish NMOSD patients tested positive for AQP4-Ab from subjects, who were either seronegative for the AQP4-Ab or were not diagnosed with NMOSD.

Results
AFM screening of the peptide panel. The first objective was to verify with atomic force spectroscopy (AFS) measurements whether one (or more) of the peptides had a specific interaction with the serum samples of NMOSD AQP4-Ab-positive patients tested with CBA. By specific interaction we mean a kind of Ag-Ab interaction, in contrast to nonspecific interactions deriving from nonspecific bindings that may include weak forces (e.g. hydrogen bonds or van der Waals forces) 23 and covalent bonds in amide bond formation 24 . In principle, two factors can be used to distinguish between the two types of interaction: the intensity of the adhesion force and the shape of the approach-retracting curves. The adhesion forces for eight NMOSD AQP4-Ab positive patients were mapped with Force Volume, i.e. images where each pixel contains an approach-retracting force curve (in a total of 256 force curves per image). Two distinct patterns of force curves were identified, which differ especially in the retracting curve, as shown schematically in Fig. 1a: (i) retracting curves with only one slope are attributed to nonspecific interactions 25 ; (ii) retracting curves with multiple repeated slopes are typical of specific interactions such as the Ag-Ab complex formation owing to hydrogen bonds and van der Waals forces 26,27 . Patterns of specific interaction were only observed when the AFM tip was coated with the AQP4 61-70 peptide. This can be inferred from the interactive document mapping (IDMAP) plots 28 in Fig. 1b, where a clear distinction between data for the retracting and approaching curves could be seen for AQP4 61-70 , but not for the other peptides. In this analysis, each curve was transformed into a single data point, marked in blue for the approaching and red for the retracting curves. The reason for the difference is that the approaching and retracting curves almost coincide for the nonspecific interactions, apart from a small region where there is a nonzero attractive force, while the differences are larger for the force curves associated with specific interactions (see Fig. 1a). Indeed, with specific interaction there is stretching or elongation of molecules throughout the retracting line, in contrast to the sharp detachment with nonspecific interactions. Another difference between the two types of force curves is in the intensity of the adhesion force. The Force Volume maps in Fig. 1b indicate a smaller force for the specific interactions with AQP4 61-70 . A more quantitative analysis was performed by selecting fifty spectra for each peptide, all of which had the overall behaviour for each class. This procedure of employing only part of the spectra was adopted because of the heterogeneity of the serum samples since not all the force curves presented the typical behaviour of their class. The boxplot graphs in Fig. 1c show that the median adhesion forces are practically indistinguishable for the peptides AQP4 131-140 , AQP4 141-150 and AQP4 201-210 peptides. In contrast, the median adhesion forces were significantly distinct for AQP4 61 24 , which are prevented by the Ag-Ab type of interaction due to bioaffinity of this complex (affinity and avidity) 29,30 in the case of AQP4 61-70 . Furthermore, the adhesion force values measured for AQP4 61-70 are consistent with those reported between antigens and antibodies using the AFS technique 18 . Because a smaller adhesion force for a specific interaction seems counterintuitive, we performed a series of subsidiary AFS experiments at various pHs, whose results are given in the Supplementary Information. The analysis confirms the hypothesis above to explain the stronger adhesion forces for the peptides with nonspecific interactions.  Figure 2c presents the ROC curve in which the AQP4 61-70 -nanoimmunosensor was effective in discriminating AQP4-Ab-positive from controls (healthy volunteers and MS) with an AUC value of 1.0, with p < 0.0001. The assay sensitivity was tested with ROC curve to verify if AQP4-Ab-negative would be distinguished from AQP4-Ab-positive NMOSD patients. Figure 2d displays the ROC curve resulting in an AUC value of 0.82, with p = 0.0078, which proved that these groups are distinct. The cut-off value of 241.3 pN was determined from the ROC curve in Fig. 2c, at 100% of specificity (95% CI 63.06-100) with the AQP4 61-70 -nanoimmunosensor assay. The sensitivity of this diagnostics assay was 81.25% (95% CI 56.50-99.43) (Fig. 2d). Therefore, adhesion forces below this cut-off threshold using the AQP4 61-70 -nanoimmunosensor indicate the presence of AQP4-Ab.

Discussion
The peptide sequence AQP4 61-70 (GTEKPLPVDM) from the extracellular loop of AQP4 was found to bind specifically to the serum samples of NMOSD patients tested positive for AQP4-Ab, which is in contrast to Kampylafka et al. 20 who reported major AQP4-Ab reactivity against AQP4 intracellular loops. However, this specific binding www.nature.com/scientificreports www.nature.com/scientificreports/ is consistent with recent studies where extracellular loops referred to as loop A, loop C, and loop E were found as disease-specific epitopes for NMOSD diagnosis 19,21,22 . AQP4 61-70 is contained in loop A, thus demonstrating the role of this sequence as an epitope of AQP4 protein due to its high reactivity with AQP4-Ab. Although some studies with whole protein reported that AQP4 conformation can interfere in the recognition by AQP4-Ab 31,32 , other approaches using peptides seem to be promising to understand the heterogeneity of NMOSD AQP4-Ab-negative patients with regard to which sequence of AQP4 or other non-AQP4 antigens is responsible for the pathology. The performance of the AQP4 61-70 -nanoimmunosensor is higher than for most published sensors 33 , including those with the CBA assay. The sensitivity of CBA can be related to the intrinsic sensitivity of the method or to the AQP4-Ab absence against the AQP4 61-70 sequence in the NMOSD AQP4-Ab-negative patient group. With the nanoimmunosensor strategy reported here, it is possible to identify new AQP4 peptide sequences (see Fig. S2; Supplementary  Information) to expand the AQP4 peptide panel and address a crucial issue involving CBA: should the researchers try to improve the assay sensitivity or patients absolutely do not have AQP4-Ab? This issue has attracted interest because the meaning of the AQP4-Ab absence is still unknown. It is possible that AQP4-Ab-negative patients have an antigen typical of another pathology instead of MS or NMOSD, for example that could bind to the antibody against the myelin oligodendrocyte glycoprotein (MOG-Ab). A percentage of acute disseminated encephalomyelitis (ADEM) and NMOSD AQP4-Ab-seronegative patients were seropositive to MOG-Ab 34 . Though approximately 20% of NMOSD AQP4-Ab-negative were seropositive to MOG-Ab 11 and there was evidence that this antibody was related to relapses 35 , there is another hypothesis, as follows. Based on Nakashima 36 , it would be inappropriate to include AQP4-Ab-negative and MOG-Ab-positive patients in NMOSD or ADEM categories. Moreover, the www.nature.com/scientificreports www.nature.com/scientificreports/ NMOSD AQP4-Ab-negative phenotype may refer to other antigens targeted by a distinct mechanism 37 , as observed in the autoimmune disorder myasthenia gravis (MG) for which the acetylcholine receptor (AchR) antibody was found to be the disease's biomarker. Other targets were found as disease-specific epitopes for distinct MG phenotypes, as the muscle-specific kinase (MuSK) protein and the lipoprotein-related protein 4 (LRP4). Antibodies against these proteins were identified in MG AChR-seronegative patients 38 . The same may apply to AQP4-Ab-negative patients, i.e. different disease-specific targets might exist in the mechanisms responsible for NMOSD. One should also stress an implication of the findings here. Sequences of amino acids, e.g. peptides, can participate in autoimmune diseases as immunogenic sequences, with binding sites composed of linear epitopes 39 . Then, forces were measured using the nanoimmunosensor made with AQP4 61-70 peptide for eight AQP4-Ab-negative NMOSD patients, MS, and healthy volunteers. Measurements were analysed using Nanoscope Analysis 7.30 and Origin 8.0 software. The Force Volume technique was applied to obtain 256 measurements ( Fig. S1; Supplementary Information) of each serum sample and then fifty measurements from specific interactions were selected to be analysed quantitatively, according to the method reported by Bizzarri and Cannistraro 27 .

Statistical analysis.
Results were analysed with the boxplot graph due to the nonparametric characteristic of our data. The U-test Mann Whitney was applied to determine p values for assessing statistical differences. The nanoimmunosensor accuracy was analysed with the receiver operating characteristic (ROC) curve, which determines p value, cut-off, and area under the ROC curve (AUC).
In addition, ROC was used to analyse sensitivity and specificity of the nanoimmunosensor, i.e., the nanoimmunosensor efficiency in distinguishing typical NMOSD patients from a set of AQP4-Ab-negative, MS patients, and healthy volunteers (n = 25 measured in triplicate), as well as the presence or absence of AQP4-Ab in the patients' serum samples. (2019) 9:16136 | https://doi.org/10.1038/s41598-019-52506-w www.nature.com/scientificreports www.nature.com/scientificreports/ Data treatment with information visualisation. Raw adhesion force (pN) vs. position (nm) spectra were analysed with multivariate data analysis using the PEx-Sensors software. The dissimilarities between the samples were converted to Euclidean distances. Because of the high dimensionality of the data (462 dimensions), they were reduced to a two-dimensional representation with the algorithm Fastmap and further improved with the Force Scheme algorithm using 500 iterations to recover some of the lost precision during data reduction. Mapping was performed with the Interactive Document Map (IDMAP) technique 45 , which has been successful in the analysis of biosensing data 46-48 . Surface plasmon resonance. Surface plasmon resonance (SPR) measurements were carried out via the BioNavis SPR Navi 200 system with a sensing device (50 nm-thick gold layer covered glass slides) previously cleaned in a mixture of 5H 2 O:1H 2 O 2 :1NH 4 OH (v/v) for 10 min at 85 °C.   Table 1.
The molecules persistence on the surface after washing with Milli-Q ® water flow produce Δ θ values 51,52 , as observed here. According to Janmanee et al. 53 , each adsorption step occurs by covalent linkages. Here, in the first step adsorption was due to amide II formation between NH 2 group of Cys and COOH group of PEG. The same amide II group was formed between NH 2 of PEG and COOH of AQP4 61-70 peptide. The increase in the angle in the sensorgram when comparing Δ θ of the reference channel with the detection channel pointed to AQP4-Ab binding to AQP4 61-70 peptide, as expected from other studies [54][55][56] .

Data availability
All data that were generated or analysed during this study and that supports the reported findings are included in this paper and additionally provided as supplementary information.