Myelofibrosis progression grading based on type I and type III collagen and fibrillin 1 expression boosted by whole slide image analysis

Abstract

Aims: The progression of primary myelofibrosis is characterised by ongoing extracellular matrix deposition graded based on ‘reticulin’ and ‘collagen’ fibrosis, as revealed by Gomori's silver impregnation. Here we studied the expression of the major extracellular matrix proteins of fibrosis in relation to diagnostic silver grading supported by image analysis.

Methods and results: By using automated immunohistochemistry, in this study we demonstrate that the expression of both types I and III collagens and fibrillin 1 by bone marrow stromal cells can reveal the extracellular matrix scaffolding in line with myelofibrosis progression as classified by silver grading. ‘Reticulin’ fibrosis indicated by type III collagen expression and ‘collagen’ fibrosis featured by type I collagen expression were parallel, rather than sequential, events. This is line with the proposed role of type III collagen in regulating type I collagen fibrillogenesis. The uniformly strong fibrillin 1 immune signals offered the best inter‐rater agreements and the highest statistical correlations with silver grading of the three markers, which was robustly confirmed by automated whole slide digital image analysis using a machine learning‐based algorithm. The progressive up‐regulation of fibrillin 1 during myelofibrosis may result from a negative feedback loop as fibrillin microfibrils sequester TGF‐β, the major promoter of fibrosis. This can also reduce TGF‐β‐induced RANKL levels, which would stimulate osteoclastogenesis and thus can support osteosclerosis in advanced myelofibrosis.

Conclusions: Through the in‐situ detection of these extracellular matrix proteins, our results verify the molecular pathobiology of fibrosis during myelofibrosis progression. In particular, fibrillin 1 immunohistochemistry, with or without image analysis, can complement diagnostic silver grading at decent cell morphology.

In situ detection of type‐I and type‐III collagens and fibrillin‐1 extracellular matrix molecules using immunohistochemistry can verify the molecular pathobiology of fibrosis during myelofibrosis progression in parallel with Gomori’s silver impregnation. While the ongoing collagen deposition represent a pro‐fibrotic response, the expression of fibrillin‐1 protein may result from a negative feedback loop response to the elevated pro‐fibrotic TGF‐β levels. Machine learning based digital image analysis of fibrillin‐1 immunosignals showed high correlations both with the eye control scoring of the immunoreactions and the diagnostic silver grading of myelofibrosis.

Introduction

Primary myelofibrosis (PMF) is a clonal stem cell disorder belonging to a group of myeloproliferative neoplasms (MPN), also including polycythaemia vera (PV) and essential thrombocythaemia (ET), which all lack the BCR::ABL1 fusion gene. ¹ Myelofibrosis is characterised by the progressive deposition of extracellular matrix, primarily of collagens, produced by reactive bone marrow stromal cells due to aberrant growth factor and cytokine production by atypical clonal megakaryocytes. ² As a result, uncontrolled stromal cell activation can lead to an ongoing bone marrow fibrosis, which interferes with the residual haematopoiesis and gradually worsens the disease outcome. Therefore, uncontrolled extracellular matrix protein production can be directly linked to myelofibrosis progression.

Elevated production and deposition of types I and III collagens are associated with regenerative organ fibrosis, which pursues a similar pathway to the fibrosis in the bone marrow. ³ , ⁴ Myelofibrosis starts with the formation of a focal and loose network of reticular fibres with rare, perivascular intersections, as defined by the European Consensus set‐up for grade 1 (MF‐1) bone marrow fibrosis using Gomori's silver impregnation. ⁵ The backbone of reticular fibres is made up of homotrimeric type III collagen fibrils. ⁴ , ⁶ , ⁷ These can also contribute to organising and regulating the diameter of type I collagen during fibrillogenesis, ⁸ , ⁹ while osteoblasts (and osteocytes) predominantly deposit type I collagen in the bone matrix. ¹⁰ Advancing myelofibrosis is featured by the increasing density and intersections among reticular and the accumulating type I collagen fibres besides early bundle formation (grade 2/MF‐2). ¹¹ This then accelerates into further concentrations of reticulin and collagen deposition together with osteosclerosis (grade 3/MF‐3).

Fibrillin, an evolutionary conserved extracellular matrix glycoprotein, the major component of 10–12 nm‐diameter microfibril polymers, is best known as a template of elastic fibre assembly. ¹² The microfibrils, contributed by fibrillins 1 and 2 isotypes, ¹³ confer stretching tolerance to elastic tissues such as arteries and ligaments, and mechanical resilience in elastin‐free matrix structures. Their role in fibrotic conditions is underlined by the excessive extracellular matrix, mainly types I and III collagens, ¹⁴ and disorganised fibrillin 1 deposition in scleroderma due to congenital dysfunctions of fibrillin 1 protein. ¹⁵

Although sporadic papers in a small number of cases showed an association between serum procollagen III ¹⁶ or in‐situ type III procollagen levels and reticulin fibrosis, ¹⁷ none have studied fibrosis‐related matrix protein expression in situ in myelofibrosis progression.

Myelofibrosis grading is still determined by revealing progressive reticulin fibrosis and collagen deposition with Gomori's silver impregnation based on the non‐specific catalytic reduction of silver ions by these structures. ¹⁸ Recent development of sensitive, automated immunostaining techniques with antibodies of high specificity and sensitivity offers the in‐situ pathobiological characterisation of bone marrow stromal cell activation pathways and scaffolding, with improved visualisation of cell and tissue architecture compared to silver staining. ¹⁹ Advanced digital microscopy and automated semi‐quantitative image analysis tools offer further accuracy for clinicopathological utilisation of this novel in‐situ molecular approach. ¹⁹ , ²⁰

Here we examined how the in‐situ detection of fibrosis‐related extracellular matrix components, types I and III collagens and fibrillin 1, can verify the pathobiology of matrix deposition during myelofibrosis evolution, and how these markers can be utilised for grading compared to Gomori's silver impregnation. For further validation, the best‐performing immunoreaction, i.e. for fibrillin 1, was analysed using a machine learning‐based automated image analysis on digital slides.

Materials and methods

Bone marrow biopsies

Jamshidi bone marrow biopsies fixed in Schaefer's fixative (4% neutral buffered formaldehyde containing methanol and glucose), decalcified overnight in 10% ethylenediaminetetraacetic acid disodium salt (EDTA‐Na₂) and embedded into paraffin wax, as described previously, were used in this study. ¹⁹ The selected cases were diagnosed between 2016 and 2022 at the Department of Pathology and Experimental Cancer Research, Semmelweis University (Budapest). After an expert haematopathologist (B.T.) established myelofibrosis grades (MF 0–3) according to the recent WHO criteria, ¹ 36 cases were confirmed to be MF‐3 (20 males, 16 females; median age = 66 years; range = 45–86), 16 to be MF‐2 (five males, 11 females; median age = 59 years; range = 31–76), six to be MF‐1 (two males, four females; median age = 46.5 years; range = 29–69), and five were MF‐0 prefibrotic based on atypical megakaryocytes and megakaryocytic and granulocytic proliferation ²¹ (two males, three females; median age = 39 years; range = 15–82). Five cases with normal bone marrow morphology were also included as controls (two males, three females; median age = 52 years; range = 44–79). Janus kinase 2 (JAK2 V617F) or calreticulin (CALR) mutations were detected in 54 of the 63 PMF patients (85.7%). JAK2 V617F mutation was found in 45 patients (71.4%) (two MF‐0, five MF‐1, 12 MF‐2 and 26 MF‐3) and CALR mutations were detected in nine patients (14.3%) (two MF‐0, two MF‐2 and five MF‐3 cases; four patients had type 1, one patient had type 2 and four patients had other CALR mutations). Two patients (3.1%) had no JAK2, CALR or MPL mutations, and seven of the 63 PMF patients (11.1%) were JAK2 V617F‐negative and CALR‐negative with no available residual DNA for sequencing to detect MPL mutation. Our study was performed in line with the regulations of the WMA Declaration of Helsinki and was approved by the National Committee for Research Ethics (ETT TUKEB) under the number IV/129/2022/EKU. Grading of myelofibrosis was based on Gomori's silver impregnation, ¹⁸ as described previously ¹⁹ (and referred to in the Introduction), which reveals reticular and collagen fibres, according to the European Consensus of diagnosing bone marrow fibrosis (as also detailed in the Introduction). ⁵

Immunohistochemistry

Three‐μm thick serial sections mounted on adhesive glass slides were used for immunostaining in a Ventana Benchmark Ultra automated instrument (Roche Diagnostics, Tucson, AR, USA) including antigen retrieval in the high pH CC1 buffer for 40 min, incubation with the primary antibodies for 60 min, detection with the Ultraview system for 40 min and visualisation with DAB/hydrogen peroxide kit. ¹⁹ The following primary antibodies were used: rabbit polyclonal anti‐human type‐I collagen (1:1000, #PA5‐95137, ThermoFisher/Invitrogen, Waltham, MA, USA), anti‐mouse/human type‐III collagen (1:100, #600–401–105‐01, ThermoFisher/Rockland, Rockford, IL, USA); mouse monoclonal anti‐human fibrillin 1 [1:800, immunoglobulin 1 (IgG1), clone:26, MAB2502, Merck/Chemicon, Darmstadt, Germany]; and CD34 (1:100, IgG1, clone: QBEnd/10, Merck‐Sigma‐Aldrich, Burlington, MA, USA) IgGs. After haematoxylin counterstaining, dehydration and clearing the sections were mounted using Pertex medium.

Scoring and image analysis using digital whole slides

All stained slides involved in this study were subjected to whole slide digitalisation using a Pannoramic 1000 Scanner (3DHistech Ltd, Budapest, Hungary), and their visual analysis and scoring was performed within the SlideViewer program (3DHistech). Evaluation of the extracellular matrix immunoreactions followed a similar concept used for assessing Gomori's reticulin staining; score 0: only rare and fragmented fibrils with no intersections; score 1: focal and loose networks with only perivascular intersections; score 2: medium density of positive fibrils with regular intersections and some extrafibrillar deposition; score 3: diffuse and dense intersecting fibril networks and more extrafibrillar antigen deposition the latter was occasionally less extended for type III collagen in advanced MF‐3 cases that in Gomori's reticulin staining.

The image analysis of fibrillin 1 immunoreactions was performed using the PatternQuant semi‐automated machine learning algorithm of the QuantCenter software package (all 3DHistech). Three to five representative, relatively large areas were annotated on each slide and analysed using an image segmentation template which was set up for highlighting all immunostaining from weak but obviously specific signals to strong intensities. The brown immunoreaction was segmented in red, the immune‐negative remaining cells/tissue in green and the cell‐free areas in yellow. The same template was used for all cases to determine and compare the immunopositive tissue area fractions in % (brown/brown + green × 100) within the selected annotations.

Statistics

The R statistical program package (version 4.1.0, R Core Team 2021, Vienna Austria; R‐Studio, Boston, MA, USA) and the Medcalc version 20.113 (Ostend, Belgium) program were used. Categorical variables were tested as proportions, while variables of non‐normal distribution were shown as median and interquartile range. Pairwise inter‐rater agreement of the visual scoring of immunostaining results was tested using Cohen's kappa statistics. ¹⁹ , ²² The correlations between assessors' scoring in pairs were evaluated by Spearman's rank test at 95% confidence intervals (CIs). The discriminating power of immunoscores among MF‐grades set up based on Gomori's silver impregnation was analysed using the non‐parametric Kruskal–Wallis rank‐sum test. The Wilcoxon–Mann–Whitney U post‐hoc test revealed the separating power of immunoscores between pairs of MF‐grades;²² P < 0.05 was considered significant.

Results

Expression of matrix proteins in non‐ and prefibrotic bone marrows

All normal and the five prefibrotic cases showed randomly dispersed fragmented fibrillary positivities for types I and III collagens (Figure 1A,B). Fibrillin 1 reaction was the most evident, delineating blood vessels and adipocyte membranes (Figure 1C). Only one of the five bone marrows with normal morphology was scored MF‐1 based on type I collagen (by two of three assessors) and type III collagen (by one of three assessors) reactions, but this and all other normal and prefibrotic cases proved to be MF‐0 with any of the other markers.

Figure 1.

Immunoperoxidase staining (brown) in serial sections of a prefibrotic bone marrow revealed only fragmented type I (A) and type III (B) collagen fibres, while fibrillin 1 (C) reaction highlighted a more continuous matrix network supporting blood vessels and delineating adipocyte membranes. [Color figure can be viewed at wileyonlinelibrary.com]

Extracellular matrix immunoreactions support myelofibrosis grading

In myelofibrosis, all three immunoreactions decorated the activated stromal cell and matrix scaffolding of bone marrow samples in similar patterns to the reticulin (and collagen) structures revealed by Gomori's silver impregnation. A few cases were up‐scaled from MF‐2 to MF‐3 compared to Gomori's grade based mainly on type I collagen and fibrillin 1 immunoreactions (Figure 2). However, down‐scaling was also observed at least at a similar frequency. The proportions of cases up‐ or down‐scaled by one grade for fibrillin were 7 and 6, for type I collagen 5 and 9 and for type III collagen 4 and 17, respectively (Supporting information, Table S1). In general, type III collagen immune‐positive fibrils were always finer and less intense compared to those revealed by type I collagen and fibrillin 1 reactions. Type I collagen immunostaining also highlighted bone trabecules, including the newly forming bone in osteosclerosis (Figure 3A). However, insufficient pre‐analytics and bone damage in a few cases caused type I collagen to non‐specifically appear in the adjacent megakaryocytes, areas which were neglected at scoring (Figure 3A inset). The uniformly strong fibrillin 1 reactions offered the best visualisation of structural abnormalities of the stromal scaffolding without the risk of non‐specificity. In advanced MF‐3 myelofibrosis, all tested matrix immunoreactions showed close visual correlations with Gomori's silver staining (Figure 3B–E).

Figure 2.

Serial sections of a myelofibrotic bone marrow considered as grade 2 (MF‐2) based on Gomori's silver impregnation (A) and also scored as such after type III collagen immunoreaction (B), but scored for grade 3 (MF‐3) after type I collagen (C) and fibrillin 1 (D) reactions. [Color figure can be viewed at wileyonlinelibrary.com]

Figure 3.

Immunoperoxidase reactions (brown) (A–D) compared to Gomori's silver impregnation (E) in advanced (end‐stage) grade 3 (MF‐3) myelofibrosis. Type I collagen immunostaining revealed bone trabecules and osteosclerosis besides myelofibrosis (A) and occasionally appeared non‐specifically in adjacent megakaryocytes (inset). Higher power shows immune‐positive stromal cells and dense fibrillary arrays of both type I collagen (B) and fibrillin 1 (C) reactions along delicate type III collagen fibrils (D), all in a similar pattern to silver staining (E). Graphs demonstrate the statistical correlations between type I (F) and type III collagen (G) and fibrillin 1 (H) immunoreaction scores after consolidation and Gomori's silver impregnation‐based myelofibrosis grades using both the Kruskal–Wallis test and Wilcoxon's post‐hoc test. [Color figure can be viewed at wileyonlinelibrary.com]

Statistical correlations between matrix immunoreactions and Gomori's silver grades

First, three assessors scored the immunoreactions independently using very similar criteria to Gomori's silver impregnation‐based myelofibrosis grading. Spearman's rank test of the pairwise correlations between assessors' scoring showed the highest rho values between 0.908 and 0.937 for fibrillin, between 0.866 and 0.930 for type I collagen and between 0.857 and 0.915 for type III collagen at P < 0.0001 significance for all three markers (Supporting information, Figure S1). Inter‐rater agreement between pairs of assessors' scores tested using Cohen's kappa statistics resulted in moderate to substantial agreements for fibrillin 1 (κ = 0.721–0.847) and type I collagen (κ = 0.668–0.840) and moderate agreements for type III collagen (κ = 0.593–0.690) (Supporting information, Figure S1).

The discriminating power of immunoreactions among MF silver grades proved to be highly significant for all three immune markers by all assessors when using the Kruskal–Wallis non‐parametric rank test (not shown). When individual scores were consolidated and agreed upon by all assessors the magnitudes of significance levels with the Kruskal–Wallis test did not change, resulting in P = 2.774 e−09 for type I collagen; P = 1.159 e−08 for type III collagen and P = 2.084 e−09 for fibrillin 1 (Figure 3F–H). The Wilcoxon–Mann–Whitney post‐hoc test revealed high pairwise statistical differences between MF‐grades for all three matrix reactions, except for type III collagen between prefibrotic and MF‐1 and between MF‐1 and MF‐2. Detailed significance levels between MF‐grades resulted from the immunoscores of the tested matrix proteins are summarised in graphs as shown in Figure 3F–H.

Additional features of matrix protein expression in myelofibrosis

Fibrillary arrays of peri‐arteriolar fibroblasts (pericytes) and matrix anchoring the external elastic lamina of tunica adventitia showed strong and dense immunostaining for types I and III collagens but relatively sparse staining for fibrillin 1 (Figure 4A–C). Conversely, the folded internal elastic lamina of arterioles stained only for fibrillin 1, while all three reactions were randomly seen among arterial smooth muscle cells of the media. In cell‐rich areas of MF‐3 cases, groups of atypical megakaryocytes were enveloped by the immunopositive fibrillary matrix with all three reactions, but in distal areas type I collagen and fibrillin 1 dominated (Figure 4D–F). In osteosclerosis, bone‐forming osteoblasts and the bone were strongly immunopositive for type I collagen, fibrillin 1 was expressed in osteocytes, while both type III collagen and fibrillin 1 reactions formed rims at the endosteal bone marrow margin without active osteoblasts (Figure 4G–I). Although capillary sinuses were increasing both in number and size in several advanced MF‐3 cases, mature red blood cells infiltrated the remnants of stromal scaffolding as revealed by all three immunoreactions without endothelial lining using a CD34‐specific antibody (Figure 5).

Figure 4.

Additional features of extracellular matrix immunoperoxidase reactions (brown) in advanced MF‐3 myelofibrosis. Serial sections of an arteriole (left column) show dense reactions of pericytes and adjacent matrix for type I (A) and type III (B) collagens, and only scarce reactions at this location but obvious staining of the elastic lamina interna for fibrillin 1 (C). In a cell‐rich myelofibrosis (D–F) all three reactions stained the fibrillar matrix among atypical megakaryocytes, but type III collagen reaction gradually diminished distal from these locations (E). In osteosclerosis the actively bone‐forming osteoblasts and the bone were strongly positive for type I collagen (G), while type III collagen (H) and fibrillin 1 (I) reactions stained an endosteal rim only, with an additional fibrillin 1 reaction in osteocytes. The newly formed bone is seen between the blue rim and the endosteal type III collagen and fibrillin 1 immunoreactions. [Color figure can be viewed at wileyonlinelibrary.com]

Figure 5.

Remnants of stromal scaffolding revealed by immunoperoxidase reactions (brown) for type I (A) and type III (B) collagens and fibrillin 1 (C) with heavy infiltration by red blood cells without endothelial lining (D) in serial sections of an advanced MF‐3 myelosibrosis. Digital differential interference images. [Color figure can be viewed at wileyonlinelibrary.com]

Automated image analysis of fibrillin 1 immunoreactions for myelofibrosis grading

Both the statistical results of visual scoring suggested fibrillin 1 reaction as the most appropriate for automated whole slide image analysis. The brown immunoreaction, the additional negative tissue and the cell‐free areas were highlighted with different colours using the same image segmentation algorithm template for testing all cases (Figure 6A–D). The fibrillin 1‐positive area fractions within the marrow tissues were plotted against myelofibrosis grades defined by Gomori's silver impregnation. The Kruskal–Wallis rank sum test revealed highly significant statistical differences among MF‐grades based on the image analysis results (Figure 6E,F), similar to those gained with eye control scoring (see Figure 3E). Also, the Wilcoxon–Mann–Whitney U post‐hoc test validated the discriminating power of fibrillin 1 immunoreaction quantitative results between pairs of MF‐grades. The visually up‐scaled seven MF cases and down‐scaled six MF cases appeared within the upper and lower ranges, respectively, of their original MF‐grades with image analysis (see Figure 6E).

Figure 6.

Automated image analysis using the PatternQuant machine learning algorithm for measuring the proportions of fibrillin 1 immunoperoxidase reactions (A,C, brown) within representative annotated areas of myelofibrotic bone marrow sections. Image segmentation reveals areas occupied by the immunoreactions (red), the immune‐negative tissue (green) and the cell‐free regions (yellow) (B,D). Higher power confirms the accuracy of segmentation (B,D). Highlighted numbers in (D) show measured areas in μm² and in % proportions. Graphs show the statistical correlations between fibrillin 1 quantitative results and Gomori's silver grades using both the Kruskal–Wallis test and Wilcoxon's post‐hoc test (E). The seven MF cases which were up‐scaled (red triangles) and six MF cases which were down‐scaled (green triangles) at visual scoring segregated to the upper and lower regions, respectively, within their MF categories in the box‐plots. The overlapping distribution curves of quantitative results within MF‐grades (F) confirms the continuous nature of myelofibrosis progression. [Color figure can be viewed at wileyonlinelibrary.com]

Discussion

Although novel prognostic classifications of primary myelofibrosis, e.g. MISPSS70 (http://www.mipss70score.it/), have been developed recently which comprehensively consider gene mutations and karyotyping besides traditional clinical and pathological features, ²³ , ²⁴ histological grading of fibrosis is still a major predictor of outcome in primary myelofibrosis. ²⁵ The ongoing extracellular matrix deposition during disease progression has still been graded using Gomori's silver impregnation, ¹¹ , ¹⁸ which, however, had not been studied systematically at the protein level. Advanced, automated immunostaining allowed us to demonstrate, for the first time, that the expression of both types I and III collagens and fibrillin 1 by bone marrow stromal cells can reveal the extracellular matrix scaffolding, in line with myelofibrosis progression classified by Gomori's silver grading. The uniformly strong fibrillin 1 immune signals offered the best inter‐rater agreements and the strongest statistical correlations with the traditional grading, which was robustly confirmed by automated whole slide image analysis using a machine learning‐based image segmentation algorithm.

Enhanced matrix accumulation in myelofibrosis is considered to follow the pathway of uncontrolled wound‐healing as a result of elevated proinflammatory cytokine and growth factor production of aberrant megakaryocytes, driven mainly by JAK2, CALR or MPL mutations through activating the JAK/STAT pathway. ²⁶ The persistent activation of stromal fibroblasts is mainly mediated by transforming growth factor (TGF)‐β, basic fibroblastic growth factor (bFGF) and platelet‐derived growth factor (PDGF) ²⁷ through their cognate receptors, of which PDGFR‐β expression was shown to be triggered from the early prefibrotic phase. ²⁸ We have recently revealed that the expression of L‐NGFR (low‐affinity nerve growth factor receptor), C‐X‐C motif chemokine 12 (CXCL12) and phospho‐extracellular signal regulated kinase 1–2 (ERK1‐2) may also indicate stromal cell activation during primary myelofibrosis progression. ¹⁹ However, all these reveal the matrix‐producing cells but not the complete matrix structure, which somewhat determines the pathobiology of myelofibrosis progression.

In this study we confirmed that types I and III collagens, together with fibrillin 1, are important matrix proteins of the fibrillary scaffold shown by silver impregnation. The delicate type III collagen fibre meshwork we detected from the early (MF‐1) disease verified that this protein is involved in reticular fibre formation. ⁴ , ⁶ , ⁷ However, the denser fibrillary type I collagen immunoreaction was also revealed in parallel, suggesting that ‘collagen’ fibrosis also starts at the same time, which is in line with the finding that type III collagen fibrils participate in organising type I collagen fibrillogenesis. ⁸ , ⁹ Type I collagen reaction indicated better correlations with silver grades from early on than that of type III collagen, which also underlined that ‘collagen’ fibrosis is not only a late event in myelofibrosis. These collagen isotypes form separate fibrils, as directly confirmed by immunoelectron microscopy, where 200–400 A diameter reticulin fibrils were positive only for type III collagen, and the ~600 A diameter fibrils were decorated only with type I collagen reaction in skin scleroderma samples. ¹⁴ However, it is likely that the fibrils of different isotypes may closely interact when forming collagen fibres. ⁸ , ⁹

Fibrillin 1 immune signal intensities were evenly robust, which offered the best chance of the three markers to recognise early abnormalities of the stromal scaffolding and matrix deposition in association with myelofibrosis grades. This was reflected both by the inter‐rater agreements, the statistical discriminating power between MF‐grades of fibrillin 1 reactions and the fewer numbers of cases where immune score was discrepant from Gomori's silver grading. The ratio of cases up‐ or down‐scaled by one MF‐grade was 7/6 for fibrillin 1, 5/9 for type I collagen and 4/17 for type‐III collagen. Furthermore, the machine learning‐based digital image analysis results of fibrillin 1 immunoreactions strongly verified those gained using eye control scoring. The automated analysis also reflected the continuous nature of myelofibrosis progression biology, as shown by the overlapping quantitative distribution curves of different MF‐grades (see Figure 6F). Fibrillin 1 microfilaments play essential roles both in tissue fibrosis and bone remodelling, as suggested by knock‐out (Fbn1^−/−) mice osteoblasts showing elevated TGF‐β and receptor activator of nuclear kappa‐B ligand (RANKL) production. ²⁹ , ³⁰ Fibrillin 1 microfilaments normally control local TGF‐β availability by anchoring it through latent TGF‐β‐binding proteins (LTBPs). ¹⁵ In line with this, congenital dysfunction of fibrillin 1 in scleroderma can lead to a massive but disorganised fibrillin 1 and a high density of types I and III collagen fibre deposition. ³¹ TGF‐β‐induced elevated RANKL expression is a major promoter of tissue osteclastogenesis. ²⁹ Therefore, the up‐regulated fibrillin 1 expression we detected with immunostaining in proportion with myelofibrosis grades may reflect a rebound effect aiming at sequestering TGF‐β and moderating collagen matrix production. Also, reduced TGF‐β by fibrillin 1 can restrict RANKL production and thus inhibit osteoclastogenesis, ²⁹ which may contribute to osteosclerosis observed in advanced myelofibrosis.

Compared to silver impregnation, immunohistochemistry offered molecular specificity and examining decent cell morphology and potential functional interactions, e.g. of aberrant megakaryocytes with the stromal cell and matrix scaffolding without a non‐specific background (except rarely for type I collagen; see Figure 3A inset). Also, immunohistochemistry clearly defined that many bunches of mature red blood cells characteristic in advanced myelofibrosis infiltrated the remnants of matrix network outside of blood sinuses.

In conclusion, we provide here in‐situ pathobiological evidence of the stromal cell activation related fibrillary matrix production and progressive deposition in line with Gomori's silver impregnation‐based myelofibrosis grades. The parallel up‐regulation of types I and III collagens suggested that ‘reticulin’ and ‘collagen’ fibrosis run side by side. Fibrillin 1 immunoreactions allowed an almost perfect visual discrimination between MF‐grades at strong inter‐rater agreements, which was robustly supported by machine learning‐based image analysis. The in‐situ detection of these extracellular matrix proteins, particularly of fibrillin‐1 with or without image analysis, can complement diagnostic silver grading of myelofibrosis progression at demanding cell morphology.

Conflicts of interest

The authors declare no conflicts of interest in relation to this study.

Supporting information

Table S1. Cases of myelofibrosis showing discrepant results between immunohistochemistry and Gomori' silver granding.

Figure S1. Spearman's rank test of the pairwise correlations between assessors' scoring and inter‐rater agreements between pairs of assessors' scores tested using the Cohen's kappa statistics of type‐I and type‐III collagen and fibrillin‐1 immunohistochemical reactions in primary myelofibrosis progression.

Data availability statement

The data that support the findings of this study are available from the corresponding author upon reasonable request. The data are not publicly available due to privacy or ethical restrictions.