Automatic quantification method of eosinophilic degranulation in tissues: Application for the study of eosinophilic disorders

The diagnosis of eosinophilic disorders can be a challenge, as tissue eosinophilia must be confirmed and sometimes quantified, and there is a correlation between the level of eosinophil infiltrate and the severity of the disease, such as in the case of eosinophilic oesophagitis (EoE).1 The presence of eosinophilic granules is a marker of activated eosinophils (Eos), which may exert toxic activity in tissues via the release of the contents of their preformed granules, such as eosinophil peroxidase (EPX).2 Moreover, Eos that infiltrate the tissue in more than occasional numbers, and/or group to form aggregates, are always abnormal.3 For a more precise quantification of tissue areas including Eos aggregates and/or clustered granules, which we propose to define as “degranulation surfaces”, we aimed to develop a valuable tool for routine identification of a differential degree of activation in tissues according to the eosinophilic disorder. We retrospectively collected formalin-fixed paraffin-embedded tissues from patients with eosinophilic disorders: bullous pemphigoid (BP, skin, n = 7), chronic rhinosinusitis with nasal polyps (NP, n = 7) and EoE (digestive mucosa, n = 8). Control tissues were skin biopsies of psoriatic skin lesions (n = 7) as an inflammatory skin disorder control and skin from breast plastic surgery as healthy controls (n = 7). The data and tissue collection were approved by the French National Commission for Information Technology and Civil Liberties (CNILs) under the number DEC18-386. Anti-EPX4 (Mayo Clinic Arizona State University, Scottsdale, MM25-82.2.1, dilution 1/750, anti-human mouse antibody) was used to identify intracellular granules (i.e., intact, aggregated Eos), extracellular isolated granules and clustered granules (degranulation surfaces). After staining, the slides were acquired using an automated microscope Axio Scan Z1 slide scanner (Carl Zeiss Micro Imaging Gmbh) with the X20/0.8 objective. This system allows the generation of a virtual slide and analysis of the whole surface of the sample. The automatic quantification of anti-EPX staining was performed using ImageJ software (US National Institutes of Health), in which a macro program (EoJ) was created allowing the automation of multiple commands after calibrating the eosinophil quantification. The macro program (available at https://github.com/antoninolillefacility/ImageJ-Macro.git) was made up of a sequence of commands (Figure 1). In the first step, the entire slide biopsy was delimited and converted into binary by thresholding, where the maximum value of a positive pixel was 255 and the minimum value of a negative pixel was 0. To measure the area of the biopsy, the area of a pixel corresponded to 0.88 μm2. In the second step, the cells, granules and degranulation surfaces were identified according to the degree of staining by colour deconvolution analysis. The 3-channel images were obtained using a specific plugin (Ruifrok, AC & Johnston, DA 2001) allowing us to differentiate the area outside the biopsy, the biopsy and the granules. The detection parameters were manually set on a slide representative of the series of slides to be quantified. In third step, the labelled granules and cells were separated by size. The isolated extracellular granules were defined within a range of 2–8 μm2. The cells were defined within a range of 8 and 100 μm2, and a circularity parameter was added to the macro from 0.6 to 1, the latter value corresponding to a perfect circle. The circularity helped to differentiate isolated cells from clusters of granules or cells. The clusters of granules were not quantified as individual granules. The “degranulation surfaces” represented clustered granules and Eo aggregates. The surfaces were considered beyond a surface of 100 μm2. For comparisons between different conditions, the degranulation ratio was used, calculated as the ratio of the surface of extracellular isolated granule surfaces and degranulation surfaces to total EPX-stained surfaces. The total EPX-stained surface was defined as the total sum of intact cell surfaces, isolated granule surfaces and degranulation surfaces. For comparisons in the same tissue, degranulation ratios were expressed over the total skin biopsy area. Additional information about study methods and findings is available in the following repository: https://doi.org/10.5281/zenodo.7775861. In the first development step, we randomly selected a total of six focused areas (3 mm2) of the highest eosinophilic infiltrates in BP skin sections, and we quantified and compared intact eosinophil counts (Eo density per mm2) and degranulation surface measurements using EoJ by ImageJ software and using hand-writing delimitation of clustered granules by two experts. We showed that degranulation surfaces were measured in the same way by both experts compared to the automatic process (p > .05). The correlation coefficient (95% confidence interval) was 0.96 (0.74–0.99) and 0.78 (0.07–0.97) for experts 1 and 2, respectively, compared with EoJ (p < .05). In the second step, we applied the EoJ algorithm measurement of cell density and degranulation surfaces to the total biopsy area of all samples (EoE, NP, BP, psoriasis, healthy skin). As expected, EoJ confirmed a significantly higher presence of intact Eos in the skin from BP patients compared with that of psoriasis patients and healthy subjects. We then compared three eosinophilic disorders damaging three different organs: EoE, BP and NP (Figure 2). The median intact Eo density on stained whole slides was significantly higher in the EoE group (653.6 ± 168.0 Eos/mm2) compared with the NP group (271.2 ± 136.1 Eos/mm2, p < .01) and BP group (268.6 ± 144.1 Eos/mm2, p < .01). Intact Eo density was similar between the NP and BP groups (p = .97). In addition, the isolated granule density within the whole section (granules/mm2) was higher in the EoE group (481.9 ± 163.2 granules/mm2) than in the NP group (117.7 ± 46.0, p < .001) and BP group (247 ± 167.7, p < .05). Then, we performed an analysis of degranulation surfaces (i.e., clustered granules and Eo aggregate surfaces according to EPX staining); the degranulation ratio was higher in the EoE and BP groups (0.88 [IQR 0.80–0.93], p < .01 and 0.78 [IQR 0.65–0.92], p < .05) compared with the NP group (0.19 [IQR 0.14–0.24]). There was no difference between the BP and EoE groups (p > .99), although the Eo density was significantly higher in the EoE group. Indeed, when we qualitatively analysed the different parameters of the eosinophilic infiltrate, we found that intact Eo density, isolated granule density and degranulation ratio were heterogeneous. Moreover, the presence of a dense infiltrate of intact Eos was not systematically associated with a high-degranulation ratio. We developed an automatic scoring system to detect all histopathological features of Eo tissue infiltrates: intact eosinophils, isolated extracellular eosinophil granules and degranulation area. We observed similar results between the independent manual analysis by two experts and the automatic method for the different histological images showing eosinophil infiltrate in BP skin tissues, confirming the validity of our original quantification approach. First, this automatic method has the advantage of allowing us to assess Eos density precisely, objectively, and reproducibly and in a time-saving manner, and second, it allows “degranulation surface” measurement where the manual quantification of individual granules or altered Eos is not possible due to their clustering. Contrary to available tools in the literature, the detection is not exclusively based on the quantification of the staining surface. The novelty of our method lies in its ability to express the degranulation level (isolated granules, Eo aggregates, aggregated granules) within the total stained area, and whole biopsy slide, leading to a better interpretation of Eos and granules density4, 5 than conventional manual methods, which are restricted to selected small fields and are subjective and time-consuming.6 Using a whole slide could be a better representation of eosinophilic infiltration in the tissue, which can be made up of isolated and dispersed Eos (i.e., in the case of fibrotic NP) or aggregated Eos. Eo aggregates are known to be associated with the severity of NP.7 In the same way, in EoE, physicians are confronted with large Eo aggregates, the quantification of which requires considerable time and effort. In our cohort, we observed that the Eo density was higher in EoE than in BP and NP, which were similar to each other. Conversely, we observed a lower level of Eo degranulation in NP than in EoE and BP. These data suggest that our tool may offer additional information about Eo aggregates that is impossible to obtain with manual approaches and previous automatized approaches. We assumed that Eo degranulation in tissue has particular pathophysiological significance depending on the disease. Moreover, high-individual variability was observed within each group of eosinophilic disorders, which could suggest that the degree of activation and/or functions reflected by degranulation may differ from one patient to another according to potentially different Eo endotypes or the stage of the disease.8 In this work, we demonstrated that if Eos infiltrate the tissue, the density of degranulation is not strictly associated with the Eo density and differs among diseases and even among patients with the same disease. These data encourage the extension of this cohort of patients and the inclusion of other diseases to assess whether the degranulation levels would be associated with specific functional endotypes and/or predictive of response to treatment. Conceptualization: FD, AB, MT, GL, DSS. Data curation: FD, AB, EA. Formal analysis: FD, AB, MT, GL, DSS. Methodology: FD, AB, NVP, JBG, SA, MT, AD. Software: AB, MT, MHG. Supervision: DSS, GL. Validation: DSS, GL. Visualization: FD, EA. Writing Original Draft Preparation: FD. Writing – Review and Editing: GL, DSS, AD, MT, AB. Plateau de microscopie photonique et d'Histologie BICeL - campus santé “US 41 - UAR 2014 - PLBS - UAR 2014 - US 41”. The authors thank E. Jacobsen (Mayo Clinic, Arizona State University, Scottsdale) for the EPX antibody. None. The data and tissue collection were approved by the French National Commission for Information Technology and Civil Liberties (CNILs) under the number DEC18-386. The data that support the findings of this study are available from the corresponding author upon reasonable request.