Algorithmic automation of bone research with StrataQuest
Throughout life, bone undergoes a constant process of remodeling in which bone is resorbed by osteoclasts and rebuilt by osteoblasts.1,2 An imbalance in these two processes can result in diseases like osteoporosis, rheumatoid arthritis, or multiple myeloma.1,2 There is consequently an interest in compounds that can modulate the balance between osteoclasts and osteoblasts and, therefore, may have the potential to treat such diseases.1
Osteoclasts, the cells responsible for bone resorption, are multinucleated cells that form from the differentiation of precursor cells in a process called osteoclastogenesis.1,3 This process can be visualized under light microscopy through a challenging but well-established protocol that uses tartrate-resistant acid phosphatase (TRAP)-staining. This takes advantage of the increased levels of this compound in differentiated osteoclasts.1
The current gold standard to quantify osteoclast formation is to count the number of TRAP-positive cells with at least three nuclei.1 This process is currently performed manually using image processing software1,3,4 – a time-consuming process that precludes large-scale experiments and is subject to substantial intra- and inter-user variability.3,4 And while counting the number of osteoclasts provides valuable information, other information that could be relevant, such as the number of cells in a culture, cell area, or the number of nuclei, is not reliably retrieved with this manual approach.4,5 These properties could help researchers better assess the impact of hormones or therapeutic agents on osteoclast biology.5 Therefore, there is a need for automated methods that can reliably and efficiently quantify osteoclasts in culture and could be used to screen potential therapeutic agents for bone-related diseases.1,2
To this end, a team from the Medical University of Vienna, Austria, developed an algorithm to enable biologists and medical scientists to automatically detect and quantify osteoclasts within commercially available automated image analysis software.5
Origins of an Algorithm
To develop their algorithm the team began by establishing a staining protocol. This included labeling of the nuclei with DAPI, using an antibody against the macrophage-antigen F4/80 – whose low expression is a marker for differentiation into mature osteoclasts – and staining of the membrane and cytoskeleton to make all cells “visible.” Cells were fixed in formaldehyde, and images were acquired using TissueGnostics TissueFAXS microscopy platform.5
The team then developed their algorithm using a four-stage process. This began with illumination correction to correct for uneven illumination that may have been introduced during the image acquisition process, which could interfere with the automated analysis. They then performed segmentation to separate pixels into foreground and background and set thresholds to enable this process to be automated. The team chose adaptive thresholding and computed local thresholds for each subregion of the image. This step was followed by post-processing to clean up images and remove artifacts and unwanted cells. The final step was to label the remaining cells, enabling measurements to be computed directly at the level of single cells. The team then trialed their algorithm to explore the effect of melatonin, a putative therapeutic agent, on osteoclast formation in vitro.5
Putting it into Practice
The researchers incubated a culture of murine bone marrow-derived macrophages with RANKL and M-CSF, which are both essential for osteoclast precursor differentiation to functionally active osteoclasts.1,5
They applied melatonin at three different concentrations (1, 0.1, and 0.01 μM) for the duration of the cultivation period. Melatonin is a hormone produced by the pineal gland that has long been thought to have a beneficial effect on bone structure and has been proposed as a potential therapy against osteoporosis in pre- and post-menopausal women. It is already well established to have a stimulatory influence on osteoblast activity, but a direct action on osteoclasts has not previously been demonstrated.5
The team then performed automated analysis using the osteoclast detection algorithm within StrataQuest (TissueGnostics). They measured a range of parameters including mean intensity values of labeled proteins, number of nuclei per cell, total number of cells and mean cell area.
Example of image anaylsis performed by IF Cell Culture - Osteoclast App.
By discriminating between multinucleated cells with low F4/80 intensity (osteoclasts) and mononucleated cells with higher F4/80 intensity (precursors), the team was able to show that melatonin had no impact on osteoclast number. However, at a concentration of 10nM, the mean area of osteoclasts after treatment with melatonin increased 130-200% compared with a control concentration of 1 μM and the mean number of nuclei/osteoclasts increased by up to 140%. These effects were not observed for precursor cells, indicating that melatonin had a stimulating effect on the differentiation of precursors into mature osteoclasts.5
This is the first time that melatonin has been shown to directly influence osteoclast activity and this finding would not be possible with the traditional manual approach. In the future, they suggest that combined with an automated staining system, automated osteoclast quantification could be incorporated into a fully automated medium- to high-throughput workflow, perhaps even with application in a clinical setting.5
Innovation Comes Integrated
This research was made possible with TissueGnostics StrataQuest software. StrataQuest is state-of-the-art software that among other functionalities automatically detects, quantifies, and analyzes the interactions between different cell types as well as structures (e.g. blood vessels, colon crypts, immune aggregates). This was coupled with TissueFAXS, which provides a convenient workflow to acquire up to eight slides automatically and comes equipped with automatic image stitching, which is essential when studying cultures containing large cells like osteoclasts.
The algorithm that the team developed has since been incorporated into the StrataQuest software and is available as an integrated app (The IF Cell Culture - Osteoclast App), allowing you to perform automated segmentation of nuclei and the identification of cultured multinucleated osteoclasts, as well as one or two additional markers.
Apart from the Osteoclast App, multiple Apps developed for automated analysis of commonly used stains in bone research are available: Safranin O, Goldner and Von Kossa. As an example, the Bone Mineralization App separates Safranin O-stained bone tissue into its morphological substructures (cartilage, mineralized cartilage, bone marrow, and mineralized bone). Measurements assessed with this App include TV (trabecular bone volume), BV (total bone volume), MCV (mineralized cartilage), CV (cartilage volume), and bone marrow (BM).
Example of image anaylsis performed by Bone Mineralization App.
If you would like a specialized workflow tailored to your bone research-related project, we offer customized Apps, developed to perform a specific task and adapted to your needs. To find out more about StrataQuest and how it could be incorporated into your workflow, get in touch with a member of the TissueGnostics team today.
References
- Cohen-Karlik E, Awida Z et al. Quantification of Osteoclasts in Culture, Powered by Machine Learning. Front Cell Dev Biol. 2021;9:674710. doi: 10.3389/fcell.2021.674710.
- Kohtala S, Nedal TMV, Kriesi C, et al. Automated Quantification of Human Osteoclasts Using Object Detection. Front Cell Dev Biol. 2022;10:941542. doi: 10.3389/fcell.2022.941542.
- Davies BK, Hibbert AP, Roberts SJ et al. A Machine Learning-Based Image Segmentation Method to Quantify In Vitro Osteoclast Culture Endpoints. Calcif Tissue Int. 2023;113(4):437-448. doi: 10.1007/s00223-023-01121-z.
- Heindl A, Schepelmann M, Stumberger R et al. Automated detection, quantification and characterization of osteoclasts in cultures using a combined image-processing and machine-learning strategy. Bone 2011. 48(2):S125-S126. Doi: 10.1016/j.bone.2011.03.242.
- Heindl, A. et al. (2012). Towards the Automated Detection and Characterization of Osteoclasts in Microscopic Images. In: Pietschmann, P. (eds) Principles of Osteoimmunology. Springer, Vienna. https://doi.org/10.1007/978-3-7091-0520-7_2.
