Automated in-depth fiber and nuclei typing in cross-sectional muscle images can pave the way to a better understanding of skeletal muscle diseases

In this issue of Acta Physiologica, Lundquist et al.¹ present a new platform for automated image data analysis of skeletal muscle samples. Immunohistochemical analysis can be used to visualize the fiber type composition of different skeletal muscles and their sex-specific adaptations with age, which are individually influenced by genetic predisposition, physical exercise, nutrition, stress, and environmental factors (Figure 1). Because of the correlation of muscle structure and physiology, the assessment can also provide an important contribution to the evaluation of diseased tissue. Over the past 5 years various image data analysis software have been developed that allow automated evaluation of immunohistochemical skeletal muscle cross-sectional images. Manual assessment limits the quantitative result in terms of representativeness, since only a small number of samples can be investigated due to the time required, and in terms of objectivity, which is highly dependent on the qualifications of the evaluator. In the future, automation of image data analysis will be crucial to compare results of skeletal muscle samples of animal models as well as on human biopsies from different research institutions and clinical centers. If the quality of sample preparation and microscopy is ensured, it would be possible to build global databases that could be used, for example, to improve the evaluation and therapy of rare muscle diseases—a worthwhile goal.

Some of the available evaluation routines for skeletal muscle cross-sectional images were programmed directly in MatLab (e.g., MyoView²) or performed with paid software (e.g., Imaris³). However, two open source image processing packages are mainly used in the community to analyze image data. One of these is Fiji/ImageJ, which was developed based on JAVA by an employee of the NIH and is constantly being expanded by a large team. Of the hundreds of plug-ins now written by the user community, Myosight,⁴ MuscleJ,⁵ Myosoft,⁶ MyoVision,⁷ QauntiMus⁸ and one more macro⁹ are available for the analysis of cross-sectional muscle samples. CellProfiler was created at the Broad Institute of MIT and Harvard and is permanently being improved there by a project team in the Cimini Lab. The software, written in Python, is thus also a free, open-source project into which anyone can write their own image processing algorithms as a new module. So far, Muscle2view¹⁰ has been developed to evaluate cross-sectional samples of muscle based on CellProfiler. Lundquist et al. now introduce FiNuTyper (Fiber and Nucleus Typer) as a new module¹ (Figure 2). The question arises, which novel possibilities FiNuTyper opens for the evaluation of muscle cross-section samples compared to the already available image processing packages.

Fiber identification, typing and size determination is available in all the mentioned software modules. By integrating Cellpose,¹¹ an algorithm developed in Python based on machine object recognition, the authors were able to improve the accuracy of fiber detection compared to the already available software algorithms. Furthermore, FiNuTyper is the first tool to allow automatic detection of fiber groupings defined as collections of the same fiber type (Figure 2). This parameter seems to play a role especially in slow-twitch muscle fibers. Here, an increase in such fiber groupings may be a marker for reinnervated muscle fibers and reflect activity-related plasticity.

Whereas differentiation of slow- and fast-twitch muscle fibers has previously been performed using antibodies to detect the distribution of myosin heavy chains (MyHC), Lundquist et al. take a different approach.¹ The authors aimed to simultaneously perform differential nucleotyping, which is not possible via contractile protein labeling. In the entire family of Ca²⁺-ATPases, three work as sarcoplasmic reticulum Ca²⁺ pumps (SERCA).¹² SERCA1 is expressed only in fast-twitch skeletal muscle.¹² SERCA2 is found in fast- and slow-twitch, smooth, and cardiac muscle as well as in non-muscle cells.¹² SERCA3, on the other hand, is found solely in non-muscle cells.¹² The nuclear envelope, which physically separates nucleoplasm from cytoplasm, is interrupted by a complex branched network of invaginations, the nucleoplasmic reticulum. Among many other functions, these structures control the selective bidirectional transport of ions. Ca²⁺ transport is also under the control of SERCA, as has been shown in cardiomyocytes.¹³ Disruption of nucleoplasmic Ca²⁺ signaling appears to trigger the development of hypertrophy and heart failure at an early stage.¹³ With the selection of appropriate SERCA antibodies and detailed comparison to established MyHC1 and MyHC2A immunohistochemistry, the authors convince that SERCA1 is a sensitive and specific marker for type 2 muscle fibers and SERCA2 for type 1. Furthermore, Lundquist et al.¹ were able to demonstrate the quality of this new methodological approach in both healthy and pathological tissue.

The detection of nuclei using SERCA isoforms has the advantage that exclusively myonuclei can be detected and thus distinguished from the mononuclei of blood vessel cells, fibroblasts and satellite cells. Since satellite cells are the resource for fiber growth and the number of myonuclei per myofiber allows statements about the condition of a muscle (e.g., atrophy),¹⁴ this method is clearly superior in its informative value to the established nuclear staining using DAPI. By incorporating another deep learning algorithm, nucleAIzer,¹⁵ the detection of nuclei itself in FiNuTyper¹ is further improved. The distribution of myonuclei within the muscle fiber plays an essential role in their supply and regulation,¹⁶ but mechanisms for their motility, increase, and degradation are poorly understood to date.¹⁶ Since nucleotyping can clearly specify statements about the health status of a muscle in addition to fiber typing, the search for a specific marker has been the subject of research for some time. The protein pericentriolar material 1 has been shown to be a specific marker of post-mitotic myonuclei.¹⁴ However, the detection via SERCA isoforms presented by Lundquist et al.¹ provides two advantages over this labeling. A specific nuclear assignment to the fiber type and the evaluation of myonuclei and fiber type in the same tissue section.

The authors provide an initial promising evaluation of muscle tissue from patients with amyotrophic lateral sclerosis and inclusion body myositis using FiNuTyper.¹ As Lloyd et al.¹⁷ conclude in a review just published in Acta Physiologica, the process of change in muscle fiber type triggered by different endogenous and environmental factors, characterizes the manifestation of a variety of neuromuscular diseases. Slow- and fast-twitch muscle fibers show different susceptibility to stressors^{1, 17} (Figure 1). Due to various routines in the acquisition and evaluation of image data, published results can unfortunately often only be compared inadequately. This makes the characterization of disease patterns difficult. Considering this fact, FiNuTyper¹ could make a significant contribution here. However, only broader application will show whether the platform meets expectations or needs further improvement.

Simone Baltrusch: Conceptualization; writing – original draft.

None.

Research in the authors lab is supported by the DFG (GRK 2676) and Deutsche Diabetes Gesellschaft.

The author has no conflicts of interest to disclose.