Automated interpretation of ANCA patterns - a new approach in the serology of ANCA-associated vasculitis

Antineutrophil cytoplasmic antibodies (ANCA)-associated systemic small vessel vasculitis (AAV) comprising granulomatosis with polyangiitis (GPA, previously known as Wegener's granulomatosis, microscopic polyangiitis (MPA), and eosinophilic granulomatosis with polyangiitis (EGPA), previously known as Churg-Strauss syndrome, is a group of related autoimmune disorders characterized by microvascular inflammation, tissue necrosis, and circulating ANCA [1‐6]. According to the recommendations for ANCA diagnostics, positive findings of standard screening tests by indirect immunofluorescence (IIF) on ethanol-fixed neutrophils (ethN) need to be confirmed with antigen-specific enzyme-linked immunosorbent assays (ELISAs) [4]. Dependent on ethN IIF pattern, ANCA can be subclassified into cytoplasmic ANCA (C-ANCA) and perinuclear ANCA (P-ANCA) patterns. Non-C/P-ANCA patterns are usually reported as atypical ANCA, which have been found in particular in patients with inflammatory bowel disease [7‐9]. The majority of C-ANCA recognizes proteinase 3 (PR3) and a positive C-ANCA pattern confirmed by an anti-PR3-ANCA ELISA is pathognomonic for GPA [1, 3]. In contrast, the main autoantigenic target of P-ANCA is myeloperoxidase (MPO) and such ANCA have been demonstrated in patients with MPA, EGPA and less frequently in Goodpasture's syndrome patients. Furthermore, the titer of both anti-PR3-ANCA and anti-MPO-ANCA is strongly associated with the active and inactive state of GPA and MPA, respectively. Due to the observations that anti-MPO-ANCA and antinuclear antibodies (ANAs) may demonstrate similar IIF patterns on ethN, IIF on formalin-fixed neutrophils (formN) is employed for their discrimination [10].

Pattern interpretation of ANCA is characterized by human bias and high variability due to methodological issues such as differing fixation protocols for neutrophils and fluorescence microscopy components (for example, lamps, filters, objectives) [11]. Remarkably, computer-based image analysis of IIF patterns by pattern recognition algorithms has recently been successfully applied for automated analysis of ANA by HEp-2 cell-based assays [12‐14], of dsDNA antibodies by Crithidia cell-based assays and of ANCA by neutrophil cell-based assays [15, 16]. However, the study of Melegari et al. [16] published as a review covered a small number of samples and only positive/negative discrimination between manual and automated ANCA pattern interpretation. Interestingly, Boomsma et al. reported earlier an IIF method for the quantitative image analysis of anti-PR3 antibody positive GPA patients [17]. The study did not reveal major differences between quantitative image analysis and the other techniques including ELISA and titration by manual IIF in their capacity to predict relapses of disease activity. However, no comprehensive approach using pattern recognition algorithms for automated ANCA pattern interpretation like in the present study has been reported so far. Furthermore, we provide for the first time variability data of an automated ANCA IIF pattern interpretation in the present study. In particular, a novel pattern recognition algorithm software module for ANCA pattern analysis has been established on the automated reading system AKLIDES™ and was compared to conventional routine interpretation of ANCA by IIF on ethN and formN.

Testing for ANCA plays a pivotal role in the serological diagnosis of AAV [22]. In accordance with international guidelines, IIF and independent detection techniques like ELISA should be used for the assessment of ANCA [4, 23]. Interestingly, IIF has been reported to be used as a screening method due to its high sensitivity followed by anti-PR3-ANCA and anti-MPO-ANCA ELISAs as confirmatory tests under routine laboratory conditions due to its high sensitivity and negative predictive value [24]. In contrast to the ELISA technique, IIF is prone to human bias and ANCA IIF image interpretation has been not automated until recently [16]. Thus, the recommended combination of IIF and confirmatory testing requires a high level of expert knowledge and lacks standardization [25]. Several studies tried to demonstrate that ELISA technology is better adaptable for standardization of ANCA assessment and has a lower lab-to-lab variability compared to IIF [26‐28]. However, most commercially available ELISAs have been reported to be inferior to IIF in terms of sensitivity [29]. Therefore, automated interpretation of ANCA IIF patterns as demonstrated for ANA detection by novel pattern recognition algorithms could provide a cost-efficient and standardized alternative approach for ANCA screening [13‐16]. In particular, regarding routine laboratories with large numbers of ANCA determinations, automated interpretation would overcome shortcomings of IIF such as high level of manual work and exceeding data management [13, 16].

Several studies have confirmed the usefulness of automated IIF systems for objective ANA pattern interpretations recently, providing the basis for the employment of IIF as gold standard for ANA testing as required by the American College of Rheumatology even under differing routine conditions [13, 14, 16, 30].

In contrast to ANA detection on HEp-2 cells, polymorphonuclear granulocytes are characterized by varying shapes of the nucleus, which is usually lobed into three segments. Therefore, algorithms for identification of granulocyte staining patterns proved to be more complex compared to those for HEp-2 cells. Furthermore, ethN and formN show different boundary characteristics with DAPI staining. Consequently, the type of fixation needs to be taken into account assessing signal intensity and pattern classification to adapt the algorithms to the changed morphology.

Due to polymorphic nucleus and cytoplasmic ANCA staining patterns, signal assessment and pattern classification cannot be performed simply inside or outside of the DAPI positive area as described for ANA pattern detection [12, 21]. Cytoplasmic and perinuclear ANCA patterns need to be detected and identified in border regions of the nucleus, rendering detection of ANCA more complex and challenging for automation.

Moreover, this is the first study on ANCA IIF pattern reading, reporting quantitative data for interassay variability. As shown for anti-dsDNA and ANA detection by IIF recently, this approach allows determining a functional assay sensitivity giving the lowest fluorescence intensity with an interassay variability of equal or less than 20% [15, 31, 32]. This creates the opportunity for a standardized cutoff determination improving the test accuracy for borderline samples and might provide a better reporting of ANCA IIF results to clinicians in order to support the progress in the treatment of AAV [33]. Relying on quantitative data for ANCA IIF reading may even give the opportunity to report findings for different test-result intervals as proposed recently for ANCA testing by ELISA [34].

Novel ANCA pattern recognition algorithms are a very useful tool for the automated reading of ANCA patterns on ethN and formN and can further improve the standardization efforts for the detection of ANCA. The automated evaluation is prone to render the comparison of diagnostic data possible. This is most important for clinical studies but also for diagnostic purposes and in clinical research. Automated pattern interpretation demonstrated high diagnostic performance for the assessment of ANCA and revealed no difference to visual reading. Further studies are warranted to evaluate this novel fully automated approach for automated software-based screening of ANCA in patients with suspected AAV.