Method for determining a biological parameter of a subject and associated methods and devices

Abstract

The ageing of the population is a key issue in the modern societies. The inventors have therefore searched how to determine parameters linked with the biological age of the subject. This has led them to find that characterizing mitochondria in an area of the subject enables to deduce such parameters. For this, the inventors has developed a tool providing with characterizing parameters of a mitochondrion, and notably a morphology parameter of the mitochondrion and an ultrastructure parameter of the mitochondrion. This opens the way of exploiting the characterizing parameters of mitochondria for multiples applications, such as therapy, medicine screening or clinical follow-up for aging related disorders or mitochondrial disorders.

Description

[0001] METHOD FOR DETERMINING A BIOLOGICAL PARAMETER OF A SUBJECT AND ASSOCIATED METHODS AND DEVICES

[0002] FIELD OF THE INVENTION

[0003] The present invention concerns a method for determining a biological parameter of a subject. The invention concerns a method for predicting the biological age of any given patient. The invention also concerns a method for predicting that a subject is at risk of suffering from a disorder, the disorder being an aging disorder or a mitochondrial disorder. The invention also relates to a method for diagnosing said disorder. The invention also concerns a method for identifying a therapeutic target for preventing and / or treating said disorder. The invention also relates to a method for identifying a biomarker, the biomarker being a diagnostic biomarker of the disease, a prognostic biomarker of the disease or a predictive biomarker in response to the treatment of the disease. The invention also concerns a method for screening a compound useful as a medicine, the compound having an effect on a known therapeutical target, for preventing and / or treating the disease. The invention also relates to the associated computer program products and a computer readable medium.

[0004] BACKGROUND OF THE INVENTION

[0005] Societies in both developed and underdeveloped countries are adapting to population aging due to significant increases in life expectancy, which now represents a major challenge for health care systems around the world. To alleviate the economic burden, it is necessary to reduce the dependency of the aging population and promote healthy aging.

[0006] Indeed, aging of the body is accompanied by a progressive loss of cellular function and deterioration of many tissues, leading to impaired function and increased vulnerability to death.

[0007] More specifically, frailty syndrome is a clinical consequence of functional decline associated with age, which clinically results in muscle loss. Sarcopenia is a progressive, systemic skeletal muscle disorder involving accelerated loss of muscle mass and function, associated with increased negative outcomes including falls, functional decline, frailty, and mortality.

[0008] Several molecular mechanisms have been described as potential causes of the etiology of sarcopenia. Mechanisms linked in particular to hormonal function (e.g. IGF-1 and insulin), muscle fiber composition and neuromuscular training, proliferation and differentiation of myo-satellite cells or dysregulation of proteostasis have been proposed to play a crucial role. However, none of these potential causes has enabled to predict sarcopenia with a good accuracy and even less to obtain a biological parameter of a subject, and notably the aging of the body.

[0009] SUMMARY OF THE INVENTION

[0010] There is therefore a need for a method for determining a biological parameter of a subject, in particular a parameter linked with the biological age of the subject, which is more accurate.

[0011] To this end, the specification describes a method for determining at least one biological parameter of a subject, the method being computer-implemented and comprising:

[0012] - a step of characterizing mitochondria in an area of the subject, to obtain characterizing parameters, the characterizing parameters of a mitochondrion comprising a morphology parameter of the mitochondrion and an ultrastructure parameter of the mitochondrion, and

[0013] - a step of deducing the at least one biological parameter based on the characterizing parameters.

[0014] According to further aspects of the method for determining, which are advantageous but not compulsory, the cleaning system might incorporate one or several of the following features, taken in any technically admissible combination:

[0015] - one biological parameter is a parameter linked with the biological age of the subject, notably the chronological age of the subject.

[0016] - a Feret diameter is defined for each mitochondrion, a morphology parameter being the Feret diameter.

[0017] - a mean intensity and a normalized intensity variation are defined for each mitochondrion, an ultrastructure parameter being the mean intensity or the normalized intensity variation.

[0018] - the step of characterizing comprises:

[0019] - an operation of receiving electronic microscopy images of the area of the subject, and

[0020] - an operation of applying a numerical model on the electronic microscopy images, to obtain the characterizing parameters.

[0021] - the numerical model is a neural network trained with a database comprising images from only three species.

[0022] - the step of characterizing comprises obtaining mutations of loci nd1 , nd3 and nd5. The specification also relates to a method for predicting that a subject is at risk of suffering from a disorder, the disorder being an aging related disorder or a mitochondrial disorder, the method for predicting at least comprising the step of:

[0023] - carrying out the steps of a method for determining at least one biological parameter of a subject, to obtain at least one determined parameter, and

[0024] - predicting that the subject is at risk of suffering from a disorder based on the at least one determined parameter.

[0025] The specification also concerns a method for diagnosing a disorder, the disorder being an aging related disorder or a mitochondrial disorder, the method for diagnosing at least comprising the step of:

[0026] - carrying out the steps of a method for determining at least one biological parameter of a subject, to obtain at least one determined parameter, and

[0027] - diagnosing the disorder based on the at least one determined parameter.

[0028] The specification also relates to a method for identifying a therapeutic target for preventing and / or treating a disorder, the disorder being an aging related disorder or a mitochondrial disorder, the method for identifying comprising the steps of:

[0029] - carrying out the steps of a method for determining at least one biological parameter a first subject, to obtain at least one first determined parameter,

[0030] - carrying out the steps of a method for determining at least one biological parameter a second subject, to obtain at least one second determined parameter, and

[0031] - selecting a therapeutic target based on the comparison of the first and second determined parameters.

[0032] The specification also concerns a method for identifying a biomarker, the biomarker being a diagnostic biomarker of a disorder, a prognostic biomarker of a disorder or a predictive biomarker in response to the treatment of a disorder, said disorder being an aging related disorder or a mitochondrial disorder, the method for identifying comprising the steps of:

[0033] - carrying out the steps of a method for determining at least one biological parameter a first subject, to obtain at least one first determined parameter,

[0034] - carrying out the steps of a method for determining at least one biological parameter a second subject, to obtain at least one second determined parameter, and

[0035] - selecting a biomarker based on the comparison of the first and second determined parameters.

[0036] The specification also relates to a method for screening a compound useful as a probiotic, a prebiotic or a medicine, the compound having an effect on a known therapeutical target, for preventing and / or treating a disease, said disorder being an aging related disorder or a mitochondrial disorder, the method comprising the steps of:

[0037] - carrying out the steps of a method for determining at least one biological parameter a first subject, to obtain at least one first determined parameter,

[0038] - carrying out the steps of a method for determining at least one biological parameter a second subject, to obtain at least one second determined parameter, and

[0039] - selecting a compound based on the comparison of the first and second determined parameters.

[0040] The specification also concerns a computer program comprising instructions for carrying out the steps of a method as previously described when said computer program is executed on a suitable computer device.

[0041] The specification also relates to a computer readable medium having encoded thereon a computer program as previously described.

[0042] The specification also concerns a device for determining at least one biological parameter of a subject, the device for determining comprising a calculator adapted to carry out a method as previously described.

[0043] BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The invention will be better understood on the basis of the following description which is given in correspondence with the annexed figures and as an illustrative example, without restricting the object of the invention. In the annexed figures:

[0045] - figure 1 is a schematic representation of a device adapted to determine the value of a biological parameter of a subject, and

[0046] - figures 2 to 22 represent results of experiments enabling to show that the device efficiently determines at least one biological parameter of the subject.

[0047] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0048] A device for determining 10 is illustrated on figure 1.

[0049] The device for determining 10 is a device adapted to determine the value of a biological parameter of a subject.

[0050] The subject should here be construed broadly as encompassing any subject comprising mitochondria.

[0051] According to the embodiment, the subject is an invertebrate and a vertebrate. It can be a fish.

[0052] Advantageously, the subject is a mammal, preferably a mouse, a rat, a human, a pig or a goat, and even more preferably a human being. A biological parameter is a parameter characterizing the subject.

[0053] A specific class of biological parameters is parameters linked with the biological age, simply named age parameters in the rest of the specification

[0054] An age parameter is, by definition, a parameter relative to a physical value varying with the ageing of the subject.

[0055] For instance, the age parameter is the chronological age.

[0056] The chronological age is the person’s age in terms of days, months or years.

[0057] Another example of age parameter is the loss of lean mass that corresponds to a loss of muscle mass.

[0058] Still another example of age parameter is the loss of mobility, which essentially refers to a loss of the ability to move around freely and without pain.

[0059] A parameter representative of the muscle integrity or a parameter relative to the muscle mitochondrial respiration are other examples of age parameters.

[0060] The device for determining 10 is adapted to carry out a method for determining a biological parameter of a subject, which will be described hereinafter.

[0061] For this, according to the example of figure 1 , the device for determining 10 comprises an acquisition unit 12 and a calculator 14.

[0062] The acquisition unit 12 is adapted to acquire information on mitochondria in an area of a subject.

[0063] A mitochondrion is an organelle found in the cells of most eukaryotes, such as animals, plants and fungi.

[0064] Mitochondria have a double membrane structure and use aerobic respiration to generate adenosine triphosphate, which is used throughout the cell as a source of chemical energy.

[0065] The area is any biological zone of the subject, which comprises mitochondria.

[0066] As a specific example, the area is a part of a skeletal muscle of the subject.

[0067] In the present case, the information obtained by the acquisition unit 12 are images.

[0068] The acquisition unit 12 is here an electronic microscope.

[0069] An electron microscope is a microscope that uses a beam of electrons as a source of illumination.

[0070] The electronic microscope is adapted to image a sample, here a sample of the subject.

[0071] In the present example, the acquisition unit 12 is, for instance, acquiring images of a biopsy of a skeletal muscle of the subject.

[0072] This implies that the acquisition unit 12 operates ex vivo, rendering the method for determining also ex vivo. According to the example of figure 1 , the calculator 14 comprises a characterizing module 16 and a deducing module 18.

[0073] The characterizing module 16 is adapted to carry out a step of characterizing of a method for determining, which will be described hereinafter.

[0074] In the present case, the characterizing module 16 comprises two sub-modules, which are a receiving sub-module 20 adapted to implement a receiving operation and an applying sub-module 22 adapted to implement an applying operation.

[0075] The deducing module 18 is adapted to carry out a step of deducing of a method for determining, which will be described hereinafter.

[0076] The calculator 14 is an electronic circuit designed to manipulate and / or transform data represented by electronic or physical quantities in registers of the calculator 14 and / or memories into other similar data corresponding to physical data in the register or memory.

[0077] As specific examples, the calculator 14 is produced in the form of a programmable logic component, such as an FPGA (Field Programmable Gate Array), or even an integrated circuit, such as an ASIC (Specific Integrated Circuit).

[0078] Alternatively, when the method is carried out in the form of one or more software programs, that is to say in the form of a computer program, also called a computer program product, it is also capable of being recorded on a medium, not shown, readable by computer. The computer-readable medium is, for example, a medium capable of storing electronic instructions and of being coupled to a bus of a computer system. For example, the readable medium is an optical disk, a magneto-optical disk, a ROM memory, a RAM memory, any type of non-volatile memory (for example FLASH or NVRAM) or a magnetic card. A computer program comprising software instructions is then stored on the readable medium.

[0079] The same remarks relative to the physical implementation of the calculator 14 stand for each of the module or sub-module, which are part of the calculator 14.

[0080] The operating of the device for determining 10 is now described in reference to an example of carrying out a method for determining.

[0081] For the sake of exemplification, it is assumed here said method for determining aims at determining the value of an age parameter of the subject.

[0082] The method for determining comprises a step of characterizing and a step of deducing.

[0083] During the step of characterizing, the calculator 14 characterizes parameters of mitochondria in the area of the subject.

[0084] This means that the calculator 14 obtains characterizing parameters of mitochondria in the area of the subject. According to the embodiments, the characterizing parameters are obtained for each mitochondria or a portion of the mitochondria in the area.

[0085] For instance, so as to limit the calculation burden, the characterizing parameters of less than 50% of the mitochondria in the area are obtained.

[0086] These characterizing parameters are according to the described example a morphology parameter of the mitochondrion and an ultrastructure parameter of the mitochondrion.

[0087] A morphology parameter is a parameter characterizing the shape of the mitochondrion.

[0088] An example of morphology parameter is the Feret diameter of a mitochondrion.

[0089] The Feret diameter is defined as the largest distance between two points of the mitochondria.

[0090] In practice, this means that a boundary is defined for the mitochondrion and that the Feret diameter is the distance between the points of the boundary, which are the most remote one from another.

[0091] Another example of morphology parameter is the area of the mitochondrion.

[0092] For instance, the area is defined as the area of the surface encompassed by the boundary defined for the mitochondrion.

[0093] Still another example of morphology parameter is the perimeter of the mitochondrion

[0094] For instance, the perimeter is defined as the length of the boundary.

[0095] The uttrastructure parameter is linked with the content of the mitochondrion and more specifically with its ultrastructure.

[0096] In particular, the ultrastructure parameter provides information relative to the density and the shape of the crista.

[0097] A crista is a fold in the inner membrane of a mitochondrion. The crista gives the inner membrane its characteristic wrinkled shape, providing a large amount of surface area for chemical reactions to occur on. This aids aerobic cellular respiration, because the mitochondrion requires oxygen. Cristae are studded with proteins, including ATP synthase and a variety of cytochromes.

[0098] According to a specific example, the ultrastructure parameter is the mean intensity of the mitochondrion.

[0099] The mean intensity is defined as the average pixel intensity in the area identified as the mitochondrion in the image taken from electronic microscopy.

[0100] Such pixel intensity is representative of the ultrastructure of the mitochondrion. An ultrastructure parameter is thus here to be construed as a statistical value representative of an intensity in an area identified as being a mitochondrion in an image of the mitochondrion acquired by electronic microscopy.

[0101] However, other statistical values representative of the intensity in the area identified as the mitochondrion may be used.

[0102] For instance, the ultrastructure parameter is the standard deviation of the mean intensity (parameter Intensity SD), the sum of the intensities in each pixel of the area identified as the mitochondrion or the normalized standard deviation of the mean intensity (parameter Intensity SD (Mean%)) in each pixel.

[0103] The normalized standard deviation of the mean intensity represents the degree of variation of intensities in each pixel normalized by the the mean Intensity.

[0104] Other examples can be found in table I, which can be found in the experimental section.

[0105] Still another set of examples is given in table II at the end of said experimental section.

[0106] In the present case, the step of characterizing is carried out by implementing an operation of receiving and an operation of applying.

[0107] During the operation of receiving, the calculator 14 receives electronic microscopy images of the area of the subject from the acquisition unit 12

[0108] During the operation of applying, the calculator 14 applies a numerical model on the obtained electronic microscopy images, to obtain the characterizing parameters.

[0109] The numerical model is a segmentation algorithm.

[0110] Such algorithm is a deep-learning algorithm trained to obtain the characterizing parameters based on microscopy images in input.

[0111] Any algorithm may be used in this context.

[0112] In the present example, the algorithm is a neural network.

[0113] From a very schematic point of view, a classic neural network comprises an ordered succession of layers of neurons, each of which takes its inputs from the outputs of the previous layer.

[0114] More precisely, each layer includes neurons taking their inputs from the outputs of the neurons of the previous layer, or from the input variables for the first layer.

[0115] Alternatively, more complex neural network structures can be considered with a layer that can be connected to a layer further away than the immediately preceding layer.

[0116] Each neuron is also associated with an operation, that is to say a type of processing, to be carried out by said neuron within the corresponding processing layer.

[0117] Each layer is connected to the other layers by a plurality of synapses. A synaptic weight is associated with each synapse, and each synapse forms a connection between two neurons. It is often a real number, which takes positive and negative values. In some cases, the synaptic weight is a complex number.

[0118] Each neuron is capable of carrying out a weighted sum of the value(s) received from the neurons of the previous layer, each value then being multiplied by the respective synaptic weight of each synapse, or connection, between said neuron and the neurons of the previous layer, then to apply an activation function, typically a non-linear function, to said weighted sum, and to deliver at the output of said neuron, in particular to the neurons of the next layer which are connected to it, the value resulting from the application of the activation function. The activation function makes it possible to introduce non-linearity into the processing carried out by each neuron. The sigmoid function, the hyperbolic tangent function, the Heaviside function are examples of activation functions.

[0119] As an optional complement, each neuron is also able to apply, in addition, a multiplicative factor, also called bias, to the output of the activation function, and the value delivered at the output of said neuron is then the product of the bias value and the value from the activation function.

[0120] In the present case, the neural network is a ll-net network.

[0121] The training phase of the algorithm uses known technics of supervised training.

[0122] According to an embodiment, the numerical model is a model trained based on images of the same species of the subject.

[0123] Said model is therefore named specie-specific.

[0124] By the expression “trained based on”, it is meant that the database used to train the numerical model comprises images of the same species of the subject.

[0125] In another embodiment, the numerical model is a generalist model.

[0126] Said generalist model has been trained on different species and is used for any subject, whether or not the subject belongs to one of the species on which the training was carried out.

[0127] Said generalist model is trained on at least two species, one invertebrate and a vertebrate.

[0128] Advantageously, the generalist model is trained based on images of only three species.

[0129] In the present case, the three species are a fly, a mouse and a fish.

[0130] More specifically, the fly is a Drosophilia melanogaster, the mouse is a Mus musculus and the fish is a Brachydanio rerio (Zebrafish or Z-fish).

[0131] Therefore, by applying a specific model or a generalist model, the calculator 14 obtains characterizing parameters of mitochondria in the area.

[0132] As apparent from the results, the generalist model provides with similar results. During the step of deducing, the calculator 14 determines the age parameter based on the characterizing parameters.

[0133] For instance, according to a specific embodiment, the calculator 14 calculates the distribution of the characterizing parameters and determines to which predefined subpopulation the calculated distribution belongs.

[0134] The predefined sub-populations are clusters of subjects sharing the same kind of distribution and having a range of chronological ages.

[0135] More precisely, it appears that it is possible to distribute a population of subjects in sub-populations with different ranges of chronological age and characterize each subpopulation by specific distributions of the characterizing parameters.

[0136] This means that the characterizing parameters of the mitochondria in the area is a signature of the belonging to specific sub-population.

[0137] By determining the sub-population, the calculator 14 determines that the age parameter is in the range of chronological age associated with the sub-population.

[0138] Examples with four sub-populations are provided in the experimental section.

[0139] According to another embodiment, the calculator 14 determines the age parameter by using a trained numerical model.

[0140] Such method enables here to predict the chronological age with a precision of 90% for the fish and 72% for the mouse.

[0141] Such method makes it possible to consider new therapeutic angles to combat the progression of muscle aging, and more broadly the disorders associated with aging (aging related disorder).

[0142] Similarly, the method is also interesting for mitochondrial pathologies, and more broadly to mitochondrial disorders.

[0143] Aging related disorders designate according to the WHO, a gradual decrease in physical and mental capacity, a growing risk of disease and ultimately death. These changes are neither linear nor consistent, and they are only loosely associated with a person’s age in years. Examples of aging related disorder are cognitive alterations, hearing loss, cardiovascular diseases, muscle sarcopenia or cancers.

[0144] Mitochondrial disorders designate are a series of often hereditary disorders characterized by dysfunction of the mitochondria, organelles in all cells. Examples of mitochondrial disorders include ptosis, external ophthalmoplegia, proximal myopathy and exercise intolerance, cardiomyopathy, sensorineural deafness, optic atrophy, pigmentary retinopathy, diabetes mellitus, diabetes mellitus and deafness.

[0145] More specifically, the method for determining may be used advantageously in other methods, the adaptation to these methods being immediate. Notably, the method for determining may also be adapted for a method for diagnosing an aging related disorder (respectively a mitochondrial disorder), a method for identifying a therapeutic target for preventing and / or treating a an aging related disorder (respectively a mitochondrial disorder), a method for identifying a biomarker, the biomarker being a diagnostic biomarker of an aging related disorder (respectively a mitochondrial disorder), a prognostic biomarker of an aging related disorder (respectively a mitochondrial disorder)or a predictive biomarker in response to the treatment of an aging related disorder (respectively a mitochondrial disorder) and a method for screening a compound useful as a probiotic, a prebiotic or a medicine, the compound having an effect on a known therapeutical target, for preventing and / or treating an aging related disorder (respectively a mitochondrial disorder).

[0146] Other embodiments of the method for determining benefiting from said advantages of accuracy may be considered.

[0147] The method for determining may be used for other parameters than an age parameter. First, some examples of age parameter may be considered as biological parameter. For instance, the loss of locomotion may be due to an illness and not linked with the age.

[0148] Beside biological age prediction and mitochondria diseases, mitochondria morphometries may be used in diseases where mitochondria play a central role, such as:

[0149] - chemotherapy response, and may help to distinguish between chemo responder and non responder,

[0150] - metabolic diseases progression such as type II diabetes or obesity,

[0151] - neurogenerative diseases,

[0152] - xenobiotic responses, and

[0153] - diet based intervention response.

[0154] On all these diseases, mitochondria morphometries might be addressed using a blood sample, a skin or skeletal muscle biopsies or any given tissues.

[0155] In variant or in complement, the method for determining may also use other characterizing parameters for mitochondria.

[0156] For instance, the quantity of mitochondria in the area is another interesting characterizing parameter.

[0157] Indeed, the Applicant has demonstrated that during aging, in both invertebrates and vertebrates, the loss of lean mass associated with a loss of locomotion is accompanied by a decrease in the quantity of mitochondria within skeletal muscles.

[0158] The adaptation of the numerical model of the characterizing module 16 is quite straightforward in so far as the Applicant has developed a versatile numerical tool. Indeed, the numerical tool has been trained to obtain several metrics as outputs (see table 1).

[0159] Such metrics are chosen to enable a user of the numerical tool to calculate each characterizing parameter that exists.

[0160] This renders the numerical model adapted to carry out a method for determining characterizing parameters of a mitochondrion.

[0161] Such numerical model is therefore useful in any context necessitating metrics relative to mitochondria in an area.

[0162] The device for determining 10 may also vary according to the embodiments.

[0163] In the present case, the characterizing module 16 and the deducing module 18 are separate.

[0164] However, they may be fused by a using a unique numerical model, which takes as inputs the images coming from the acquisition unit 12 and outputs the biological parameter.

[0165] Indeed, one advantage of the numerical model developed for the characterizing module 16 is that it can be easily completed by another numerical model using its outputs as inputs to predict other value of interest for the user.

[0166] According to another embodiment, the characterizing module 16 is not part of the calculator 14 but integrated in the acquisition unit 12.

[0167] This leads to an electronic microscope provided with the functionality of outputting the characterizing parameters of mitochondria in a sample observed by the electronic microscope.

[0168] Indeed, the method for determining is, in the described example, a method for postprocessing images.

[0169] As such, the post-processing can be made in any locations.

[0170] For instance, it may be considered that the acquiring is made in a first location, the characterizing of the mitochondria in a second location and the deducing in a third location.

[0171] Such versatility enables to consider other embodiments wherein the acquiring phase and the characterizing are very different from applying an image analysis technique on a set of images to obtain characterizing parameters relative to the mitochondria.

[0172] For instance, during the acquiring phase, it is obtained mutations of loci nd1 , nd3 and nd5 by a dedicated acquiring unit.

[0173] Then, the characterizing is based on the obtained mutations.

[0174] For instance, another numerical tool can be used, said numerical tool taking as inputs the mutation of loci nd1 , nd3 and nd5 and outputting the searched characterizing parameters. The embodiments and alternative embodiments considered here-above can be combined to generate further embodiments of the invention.

[0175] EXPERIMENTAL SECTION

[0176] Experiments were carried out by an Applicant and are illustrated by the figures 2 to 28. The details of the figures are given at the end of the experimental section.

[0177] These experiments are organized in two part, a first part relative to the numerical tool developed to obtain the characterizing parameters and a second part showing that the deducing step enables to obtain biological parameters.

[0178] First part: the numerical tool developed to obtain the characterizing parameters

[0179] In the present study, the Applicant established a pipeline named EMito-Metrix (Figure 2) that aims at, using electronic microscopy pictures, fully characterize mitochondria morphology and ultrastructure of skeletal muscle using an automated and unsupervised workflow. CellPose algorithm Cyto2 has been pre-trained on pictures from different species, from invertebrates (Drosophilia melanogaster (Fly)) to vertebrates (Brachydanio rerio (zebrafish or Z-Fish), Nothobranchius furzeri (Killifish or K-Fish) and Mus musculus (Mouse)), and a generalist and a specie-specific (specialist) model have emerged, able to segment mitochondria with a very high accuracy across all species. As the Applicant’s approach generate multiple metrics per mitochondria (e.g 21 metrics per mitochondria), the Applicant has created a user-friendly graphical interface packages in a Fiji plugin, that contains a list of customizable graphs to optimize data visualization using dimensionality reduction (UMAP or PCA) and more conventional depiction of data distribution (density curves, histograms, violin plots or star plots) (Figure 2). Additionally, a machine learning predictive analytics model has also been included, in order to evaluate using unsupervised prediction if any given experimental set up has an impact (positive or negative) on mitochondria morphology. To illustrate the relevance of the tool, the Applicant has run an in-depth comparison of mitochondria morphology across species such as fly, fish and mouse. This analysis helped to identify mitochondria metrics that are typical for each specie, but also some traits that are commonly observed in all species. Interestingly, while the prediction accuracy is very high in fly and fish, the accuracy score is very low in mouse, suggesting that mitochondria in mouse share morphological traits with other species. Overall, the plugin will certainly allow any scientists without coding skills to (i) extract substantial information using pre-trained models and graphs from his own dataset (ii) and by using the prediction tool to address whether any intervention has a beneficial or detrimental effect on mitochondria morphology.

[0180] Results

[0181] Successful and automated mitochondria segmentation on EM pictures

[0182] Despite spectacular progress in the resolution of confocal microscopy, the best way to assess and study mitochondria morphometry is the electron microscopy, where a single object sizing around few nanometers can be imaged and analyzed. But currently, there is hardly any existing tool or pipeline that allow to successfully segment the mitochondria out of a black and white contrasted-picture. With the recent publication of learning algorithms, such as CellPose2.0, it offers a solid option to use this tool using learned-based strategy and be able to robustly detect the mitochondria after proper training. In this purpose the Applicant has trained CellPose-Cyto2 algorithm on electron microscopy pictures of a highly mitochondria-enriched tissue represented across species, the “skeletal muscle”, from invertebrate (Fly) to vertebrate (Fish and Mouse) models (Figure 3). First off, the Applicant separately trained the algorithm in specie-specific manner (Specific model-SM), as the Applicant could not rule out a difference in the mitochondria contrast related to speciespecificity (Figure 3A-D). For each specie Fly, Z-Fish and Mouse, half of the dataset was used to train the algorithm, and the other half was employed to test the accuracy of the trained algorithm to detect mitochondria. To evaluate the proper detection, the Applicant calculated the average precision (AP) using an loll (intersection over union) threshold on 1000 to 3000 Region of interest (with 1 ROI=1 mitochondria). Using the Cyto2 model implemented in Cellpose, the Applicant first segmented all detectable objects - mitochondria and squeletale fibers - on every field, regardless the species (Figure 3A-C) with a poor APiou>0,5. But after only few rounds of training, the Applicant gradually improved the accuracy of segmentation using over 1000 ROI, and ended up detecting correctly, 78% in Fly, 74% in and Fish and 58% in Mouse of the total mitochondria. Of note, out of 71% detected objects, there might have some undetected objects on the pictures, but there are no false positive objects, which is a very important point to consider. In parallel, the Applicant has worked on a generalist model (GM), able to detect any given mitochondria regardless of the specie of origin (Figure 3E-F and Figure 7A-E). As opposed to the speciespecific model, only 100 ROI were needed to meet a satisfactory level of segmentation. As compared to SM, GM reaches equivalent performances with 80% in Fly, 71% in Z-Fish and 56% in Mouse of properly detected mitochondria (Figure 3E-F and Figure 7). To firmly validate the accuracy of the mitochondria segmentation, the Applicant challenged the trained algorithm with EM pictures from a separate dataset on another vertebrate model, the African Turquoise Killifish (K-Fish). On this additional model, pretrained Cellpose, from scratch analysis showed a robust detection of mitochondria with the SM, but again the GM outperformed the SM as already demonstrated on Fly, Z-Fish and Mouse (not shown). These results indicate that the pretrained CellPose is very accurate for mitochondria segmentation, and that the diversity of the training dataset matters and helped to generate a Generalist model as robust as the Species_specific models.

[0183] Comparison of performance between pretrained CellPose2.0 and other available applications

[0184] Multiple applications or plugins are already available in the field to determine various mitochondria morphology metrics, albeit being very performing / relevant, the majority of these do not meet the need. And only a few are usable on 2D EM pictures with unbiased and unsupervised and no human-in the-loop procedures, and among them the Applicant has used MitoNet, DeepImageJ and compared segmentation performances with Cyto2- Cellpose and pretrained Generalist and Specialist Models. Thus, the Applicant has run a performance comparison between these three apps and the trained-algorithm either as the specie-specific or generalist model, on a set of 10 ROIs from Fly, 35 ROIs from Z-Fish and 20 ROIs from Mouse (Figure 4). To evaluate and compare the performance of each model, the Applicant has measured the APi0u>o,s, while across species DeepImageJ and llastik have an AP between 2-5% and 10-25% respectively, both of the pretrained models outperformed llastik and DeepImageJ with an APIoll>0,5 averaging between 60-80% of correct predictions. Of note the model where algorithm specifically trained on each specie display robust AP score between 58-78%, while the generalist model still out-performed the other models, with a similar score compare to the SM with an AP score between 56-80%. Altogether, using the pretrained CellPose algorithm either on SM or GM model is outperforming other unbiased / unsupervised app for mitochondria segmentation on EM pictures and represent a clear added-value for any user.

[0185] Data computation of all variables extracted from EMito-Metrix

[0186] Many available apps or plugins help every user to segment the mitochondria, and sometimes certain app give the option to export (give name) all measured metrics on a excel file. However, when hundreds / thousands of objects are segmented, it might be a lot of data to compute, and beside plotting data on a simple graph, and without the help of data scientist, extracting information out the dataset might be very challenging. To further help the user / biologist to compute the metrics in multiple ways, the Applicant proposed different options containing a list of customizable graphs to optimize data visualization using dimensionality reduction (LIMAP or PCA) and more conventional depiction of data distribution (density curves, histograms, violin plots or starplots) (Figure 5 and Figures 10 and 11). To further illustrate the potentiality of the workflow, as an example, the Applicant will compare datasets of mitochondria metrics from Fly and Fish (Figure 5). As the pretrained algorithm is packaged in a FIJI plugin, the user will have access to 21 metrics found in the default “Set Measurements” box usable (listed and explained at https: / / imagej.nih.gov / ij / docs / menus / analyze.html). Among them, in this example, eight metrics are considered as relevant with a low redundancy (Area, Perimeter, Total Intensity, Circularity, Roundness, Solidity, Aspect Ratio, Feret Diameter- See Table 1 for definition / explanation). As these metrics are a direct reflect of mitochondria morphology, the Applicant has also calculated morpho-metrics which give insights on mitochondria ultrastructure, based on contrast and intensity values that are a direct indication of cristae localization, number, shapes or density. Using “Mean Intensity” (Mean I nt) average levels of gray inside each ROI (per mitochondria), the Applicant calculated “Total intensity” for the sum of gray levels in each segmented mitochondria. These two metrics helped to generate the Standard Deviation of the Intensity (I nt SD) which reflect through the variation of gray levels in every single mitochondria, and the Int SD (%Mean Int), which represent the degree of variation of gray levels normalized by the the mean Intensity. Both values embodied a certain extent of heterogeneity in terms of ultrastructure. After a very accurate mitochondria segmentation on Fly and Fish EM pictures (Figure 5A), mitochondria metrics and notably Feret diameter and Intensity have been extracted and displayed in Figure 5B-E (with the nine other metrics displayed in Figures 8 and 9). Using density curves, histograms and violin plots, Fly and Fish exhibit drastic differences in both metrics, that the three depictions clearly highlight. To further explore the dataset, star plot is usable (Figure 5E) as it is a very practical / relevant layout to compare multivariate datasets. Impressions gotten from only two variables are confirmed, with mitochondria morphology of the Fly being very different from those in the Z-Fish on a larger scale when 9 more metrics are added for the comparison (Figure 5E). Next, to integrate all possible variable in the analysis (11 metrics per mitochondria for 8844 and 2319 mitochondria respectively in Z-Fish and Fly) the Applicant also propose to use graphs with dimensionality reduction (such as UMAP or PCA- Figure 5F and Figure 9K). UMAP and PCA depict a very clear difference between Fly and Z-Fish, and the clustering of each specie is exacerbated, underlying the added-value of UMAP and PCA to analyze high dimensional dataset. Overall, this example illustrates a user-friendly interface with a list of customizable graphs to optimize data visualization and distribution that will certainly help every biologist to drawn new hypothesis and help the community to move forward.

[0187] Cross species comparison of mitochondria metrics variables

[0188] To exemplify the relevance and the potency of this application, the Applicant compared mitochondria morphometries in four different model organisms used routinely in labs, and show how using very simple commands, any user might be able to extract a large number of informations. For instance, the Applicant analyzed datasets extracted from a cross-species comparison, where mitochondria were segmented using electron microscopy pictures of skeletal muscles of from fly, zebrafish, killifish and mouse (Figures 6, 10 and 11). To begin the cross-species analysis, the Applicant ran LIMAP or PCA on a dataset covering at least three biological replicates from each specie, with an average of > 2200 events / species, and a total of 15503 mitochondria plotted (Figure 6A-B). LIMAP separated mitochondria populations into distinct clusters more than the PCA, notably splitting mitochondria originated from Z-Fish, K-Fish and Fly in three different clusters. Interestingly, mitochondria from mouse does not form a unique cluster, separated from the others, suggesting that mitochondria from mouse display features represented in other species. However, when using a paired-analysis (Figure 6C-D), LIMAP showed that mouse mitochondria are quite distinct from mitochondria originated from Fly forming two different clusters (Figure 6C), but looked very similar to those from Z-Fish (Figure 6D). Other LIMAP paired-comparison (Figure 10A-C) mapped other inter-species differences, like notably a clear clusterization between Fly and K-Fish. Additionally, when this multivariate dataset is displayed using the star plot (Figure 11 E), it is easier to determine which morphology feature is creating the difference. While for the fly perimeter, area and ferret diameter stand out as “specie-specific” metrics, Intensity SD for the K-Fish, and Mean Intensity for the Z-Fish are also potential metrics “specie-specific”. To further strengthen this analysis, all these metrics are also represented “individually” to confirm potential difference between species (Figure 5B-D and Figure 10D-F), and again Feret Diameter (Figure 10D) is clearly higher in Fly compare to other species. Next, to further explore the dataset, the Applicant has implemented in the current plugin several machine learning tools, that would help the user to extract even more informations and solidify conclusions. Here in the example based on the interspecies comparison, the question is whether the difference in mitochondria morphology are such, that the Applicant would be able to distinguish between the species based on the mitochondria metrics. To test this hypothesis, the Applicant has used / tested different two classes of machine learning algorithms: tree-based models (LR parser, Random Forests and XGBoost), and a neural network (Multi Layer Perceptron, MLP), the Applicant split the data set into training and tests datasets (80% and 20% respectively) (Figure 6F and Figure 11) and the Applicant compared the performance of each algorithm in terms of good prediction (Accuracy score). On the dataset, the MLP algorithm displayed the best performance, with an average accuracy score of 73% of right predictions (Figure 10F-G), but XGBoost reached the best prediction of 93% for the Z-Fish (Figure 6F and Figure 11 A). When averaging four algorithm prediction, species such as Z-Fish and Fly have the best prediction rate with 87% and 79,8 % respectively, and some like mouse display a very poor performance with only 28%, confirming conclusions drawn from the UMAP analysis, where species displaying unique clusters, have a robust prediction percentage. To define the contribution of each variable in specie prediction, the Shapley Additive exPlanations (SHAP) was applied on the XGBoost algorithm. The summary plot shows the ranking by the mean absolute value of global SHAP contribution for each variable by decreasing importance (Figure 6G-J). Despite having a certain extent of uniformity in mitochondria, there are, remarkably, various levels of intensity and intensity-related metrics that are the biggest contributor in the specie determination (Figure 6G-J). As intensity metrics relate to ultrastructural features like cristae shapes and density (see Table 1), it suggests that there is a pattern of intensity that is typical for each specie, implying that mitochondrial bioenergetic might be a crucial feature that has evolved differently through the phylogenetic tree. At this stage, it is just speculations that has been partially explored yet and will need to be confirmed experimentally. Overall, this short example shows the potentiality of the workflow that can be easily applied to any given experimental set up where mitochondria morphology could be compared in several conditions.

[0189] Discussion

[0190] Hereby the Applicant has developed a new open-source tool, able to segment automatically mitochondria on EM pictures in an unsupervised and unbiased manner, the Applicant has trained the Cyto2 version of CellPose 2.0 algorithm, that has shown a robust and reliable detection across species with an APIoU>0.5 averaging, across species, 70% (SM) and 69% (GM). This version of pretrained Cyto2 is usable under Fiji, which allows any user to extract 17 morphology metrics with four extra ultrastructural metrics calculated through CellPose interface, meaning 21 potential metrics per segmented mitochondria (see Tablel). As this app generates large datasets (e.g., 21 metrics per segmented mitochondria and for the cross-species study the Applicant has segmented over 15000 mitochondria), to allow any biologist without coding skills to extract valuable information from his own dataset, it was mandatory to add a module where all data could be computerized in various manners including some graphs using dimensionality reduction. Additionally, and to make this plugin unique, the Applicant has added an extra module using two classes of machine learning algorithms a tree-based models (LR parser, Random Forests and XGBoost), and a neural network (Multi Layer Perceptron, MLP). These algorithms will help, by prediction, to determine whether there is a difference in the datasets based on mitochondria morphology metrics. For example, in the cross-species study presented here, RF or XGB algorithms were able to predict correctly the specie based on the mitochondria metrics with more than 90% of accuracy for Z-Fish. The Applicant can extrapolate that using these machine learning tools any biologist will be able to evaluate the impact of any treatment or gene mutation based on the mitochondria morphology using an unbiased methodology.

[0191] The morphology of mitochondria, and especially the mitochondria ultrastructure, is studied from a very long time, and multiple solutions are already available in the community, most of them being helpful but not always meeting the need of common biologist / scientist, either by generating bias as it requires human intervention. Some recent pipelines, are simply remarkable in terms of potentialities but very complex or even impossible to run without coding skills, some others are adapted to 3D-EM, or live imaging requiring rare / expansive equipment that only few labs may afford. This is the reason why the Applicant has created an app that fills this gap, with an automatic, unsupervised and unbiased workflow usable with very classical EM pictures. Moreover, by adding a module where all generated data are depicted in multiple ways is undoubtedly a great added value compare to other existing solutions.

[0192] One of the questions in the field of mitochondria / muscle is to study the different populations of mitochondria in the muscle fibers where subsarcolemmal (SSM) and intermyofibrillar (IMF) mitochondria co-exist. Many hypotheses in the field regarding the different functions hold by the SSM and IMF have been drawn, and the precise morphology / ultrastructural difference between both populations are still under investigation. One can assume that the pipeline could help to better characterize the difference in terms of morphology between SSM and IFM, but this will possible to train the algorithm to segment one or the other sub-populations will certainly require a very large number of EM-pictures and lots of training rounds. But the Applicant has good hope that with the tool, the Applicant might be able to propose to discriminate using pretrained Cellpose both SSM and IMF. With constant progress in deep learning-based imaging analysis, it is also very likely that a new solution could be available, able to distinguish invisible difference between two objects.

[0193] The workflow presented here has been validated for skeletal muscle. As the goal was to propose a pipeline to the largest community possible, the Applicant decided to start off with a workflow usable for the most represented model organisms in labs from invertebrates to vertebrates. The Applicant choose to work with skeletal muscle pictures as it is the main tissue represented across these model organisms. However, as the training of the algorithm was performed to properly segment over 15000 mitochondria over 5000 pictures, one can assume that a training with such diversity should almost guarantee a descent accuracy of the mitochondria segmentation in other tissues. The accurate mitochondria segmentation on the K-Fish confirmed this assumption as K-Fish EM pictures were not part of the training dataset. Additionally, the Applicant also has tested liver, brain and intestine of Z-Fish where the APIoll>0,5 reached 0,8 (not shown). Compare to ground truth, these scores are very good and encouraging. But in order to reach an APIoll>0,5 of 1 , and generate a pan- tissue / pan-species generalist model, a trained algorithm able to segment accurately any mitochondria from any tissues from any species, it would simply require lots of effort by training CellPose2.0 algorithm on a larger dataset of EM pictures from the Applicant’s own database but also including EM pictures part of remarkable databanks recently generated. Of note, building EM pictures dataset has constituted a major challenge in the field, and it has been achieved and exemplified by MitoEM and CEM500K dataset. Using such a datasets and other algorithm, a step forward has been achieved recently, when mitochondria segmentation using a generalist model able to segment mitochondria in large variety of EM-3D pictures. This progress has been really impressive, although only few labs can still perform 3D electron microscopy, and mitochondria segmentation across species using this pipeline still need to be performed.

[0194] Materials and Methods (or Methods) Animal tissues Worm

[0195] Worms or C. elegans from the N2 (WT) strain were grown on OP50 and maintained at 22.5°C at the CBI (Toulouse).

[0196] Fly

[0197] All Fly or Drosophila melanogaster stocks and genetic crosses were grown using standard medium at 25°C at the RESTORE Institute. The strains used in the study were Oregon.

[0198] Fish

[0199] Zebrafish, Brachydanio rerio and African Turquoise Killifish, Nothobranchius furzeri were maintained at the RESTORE Institute in accordance with institutional guidelines for animal research and were approved by the Animal Care and Use Ethics Committee US006 CREFRE-CEEA-122 (protocol 17 / 1048 / 03 / 20). Mice

[0200] Breeding and experimental procedures were performed in accordance with institutional guidelines for animal research and were approved by the Animal Care and Use Ethics Committee US006 CREFRE-CEEA-122 (protocol 17 / 1048 / 03 / 20).

[0201] Tissue preparation

[0202] All samples (a 5-millimeter biopsy), from all species of were fixed with 2% glutaraldehyde in Sorensen buffer (see below) immediately after dissection. Biopsies were performed on the central part of the worm, on the thorax of fly, on the trunk of the zebrafish, and on the quadriceps of the mouse.

[0203] Tissue transmission electron microscopy

[0204] Sample was fixed with 2% glutaraldehyde in Sorensen buffer (0.1 M, pH = 7.4) for 1 hours, washed with the Sorensen phosphate buffer (0.1 M) for 12 hours. Then were post fixed with 1% OsO4 in Sorensen buffer (Sorensen phosphate 0.05 M, glucose 0.25 M, OsO4 1%) for 1 hours, washed twice with distilled water and prestained with 2% uranyl acetate aqueous solution for 12 hours. Samples were dehydrated in an ascending ethanol series until ethanol 100° and then with propylene oxide. Sample was embedded in epoxy resin (E- NMA Hard)). After 48h of polymerization at 60°C, ultrathin sections (70nm) were mounted on 100 mesh collodion-coated copper grids and poststained with 3% uranyl acetate in 50% ethanol and with 8.5% lead citrate. Observations by HT 7700 Hitachi transmission electron microscope at an accelerating voltage of 80 KV.

[0205] Image Analysis

[0206] Analysis of transmission electron microscopy images was carried out using Fiji (Schindelin, J.; Arganda-Carreras, I. & Frise, E. et al. (2012)) and Cellpose (Stringer, C. et al. (2021)).

[0207] 1 - Cellpose training

[0208] Training of the various Species-specific and Generalist models was carried out using Cellpose 2.0 software (Pachitariu, M. & Stringer, C. (2022)). For each species, the images acquired by TEM were previously separated into 2 distinct groups, one - training dataset - used for training the 2 models, and the second - validation dataset - for model validation. For each image in the training group, mitochondria were first detected using the Cyto2 model implemented in Cellpose. To re-train the Cyto2 model with his images, the Applicant used Cellpose 2.0's Human-in-the-loop training function and manually annotated, one by one, mitochondria not detected or only partially detected by the Cyto2 model (False Negative). For each image, the Applicant also took care to remove all false detections (False Positive). Thus correctly annotated, these images were used again in the Cellpose 2.0 software to re- train the initial model (Cyto2) and generate the 4 models.

[0209] For the Zebrafish-specific model, 31 images were trained and 2875 mitochondria manually annotated. For the Drosophila-specific model, training was carried out on 11 images, for a total of 965 annotated mitochondria. For the Mouse-specific model, 25 images and 2001 mitochondria were manually annotated. Finally, for the Generalist model, the Applicant combined all images and annotations from the 3 Species-specific models, for a total of 67 images and 4809 annotated mitochondria.

[0210] 2- Model validation

[0211] Each of the 4 trained models was then validated using different images from the validation group (35, 10, 20 and 65 images, respectively for Zebrafish-specific, Drosophilaspecific, Mouse-Specific and Generalist). Using Cellpose 2.0 software, the Applicant subjected each image to its 2 corresponding models - Species-specific and Generalist - and obtained a mitochondrial segmentation prediction map (P) for each. In parallel, for each validation image, the Applicant manually annotated the mitochondria to obtain a Ground Truth map (GT).

[0212] To assess the performance of the 3 Species-specific models and the Generalist model, the Applicant compared the prediction and ground truth maps for each mitochondrion, and measured various performance indicators using Fiji software:

[0213] - Intersection map (P D GT): corresponds to the area of colocalization between a mitochondrion predicted by the model and its nearest ground truth (annotated mitochondrion)

[0214] - Union map (P u GT): corresponds to the projection of a predicted mitochondrion and its nearest ground truth.

[0215] - Intersection over Union or loU: corresponds to the ratio between intersection (P D GT) and union (P u GT). This score is calculated for each mitochondrion, and represents a measure of a model's accuracy for an object detection spot. Its value ranges from 0 to 1 , with 1 for a perfect object prediction. Note that a score greater than 0.5 is considered by consensus to be the threshold value for good prediction.

[0216] For each image in the validation group, the Applicant then used these calculated loU values to measure various performance metrics: - True Positives TP: number of objects in the prediction map with an loll value greater than 0.5. This metric corresponds to the number of mitochondria correctly predicted by the model.

[0217] - False Positive FP: number of objects in the prediction map with an loU value less than or equal to 0.5. This metric corresponds to the number of erroneous predictions.

[0218] - False Negative FN: the number of mitochondria not predicted by the model. It corresponds to the difference between the number of ground-truth mitochondria and the number of true positives (GT - P).

[0219] - Precision (TP I (TP+FP)). Indicates the extent to which the objects predicted by the model are real (i.e. are mitochondria). The calculated score ranges from 0 to 1.

[0220] - Sensitivity (TP I (TP+FN)). Indicates the model's ability to predict all or some of the real objects (mitochondria). The calculated score ranges from 0 to 1.

[0221] - Average Precision AP (TP I (TP+FP+FN)). Performance indicator that takes into account the 2 criteria of specificity and precision. The calculated score ranges from 0 to 1.

[0222] For each trained model (species-Specific or Generalist), the mean Average Precision (mAP) is calculated by averaging the AP score calculated for each image in the validation group.

[0223] 3- Comparison with other prediction models

[0224] For each species, the Applicant compared the Species-specific and Generalist models with 3 other models available to the scientific community:

[0225] -Cyto2 model implemented in Cellpose & Cellpose 2.0 software

[0226] (https: / / cellpose.readthedocs.io / en / latest / models.html)

[0227] - The Neuron segmentation in 2D EM model (Falk et al. Nature Methods (2019); Ronneberger et al. arXiv (2015); Lucas von Chamier et al. biorXiv (2020)) available via the bioImage Zoo (https: / / bioimage.io / )

[0228] - MitoNet, a model trained on 2D unlabeled cellular EM database (https: / / www.sciencedirect.com / science / article / abs / pii / S240547122200494X)

[0229] For each of these 3 models, the Applicant generated the Prediction map and GroundTruth map for each of the images in the validation group. To compare the performance of these 3 models with the Species-specific and Generalist models, we measured the mean Average Precision (mAP) as described above (see part 2).

[0230] For each species, the Applicant tested the different calculated mAP values 2 by 2 using a paired Student's t test.

[0231] 4- Batch segmentation & Morphological analysis The various stages of the analysis pipeline - image processing, segmentation, morphological measurements - were carried out using Fiji software (Schindelin, J.; Arganda- Carreras, I. & Frise, E. et al. (2012)). The raw TEM images were first amplitude normalized (normalization to the minimum and maximum gray level value) to attenuate acquisition- related gray level variations observed between images. Automatic mitochondrial segmentation was then performed using the Cellpose wrapper implemented in the BIOP (BioImaging And Optics Platform) plugin. The parameters taken as input to the plugin and set by the experimenter are the type of prediction model (Species-specific or Generalist) and the diameter (in pixels) of the objects to be segmented. The mitochondrial segmentation maps (or labelled maps) generated at the output of Cellpose were then used in Fiji to calculate the various individual (per mitochondrion) or average (per image) metrics.

[0232] Data computation

[0233] Data collected from the image analysis is used as input for a data analysis pipeline. All the output files are concatenated together, along with the species category. Information about the position of the mitochondria in the image is discarded, as well as empty variables. The python package pandas is used for data manipulation. Then, the packages umap-learn and sklearn decomposition module were used respectively for LIMAP and PCA dimensionality reduction analysis. Packages matplotlib and seaborn were used for plotting graphs.

[0234] For model training, package sklearn was used to scale and split the data, do the gridsearch cross-validation, and evaluate the models

[0235] Table 1. Fiji Mitochondria Metrics- Set Measurements Box

[0236] Second part: the deducing step enables to obtain biological parameters

[0237] Summary Understanding the biological and molecular bases of human muscle aging remains an open and fascinating question. Multiple mechanisms have been described and proposed, but a growing number of elements argue in favor of a major role of mitochondria in the aging process. Mitochondria exist within the cell as a very heterogeneous entity, but the biological significance of this extreme diversity is still poorly understood, and this is what the Applicant tried to elucidate here during this work. The Applicant demonstrates here that during aging, in both invertebrates and vertebrates, the loss of lean mass associated with a loss of locomotion is accompanied by a decrease in the quantity of mitochondria within skeletal muscles. An in-depth analysis of the morphology of mitochondria in skeletal muscle by electron microscopy shows an extreme diversity of mitochondria in all species, with an overall trend towards more enlarged and contrasting mitochondria with aging. An automatic analysis of images at very high resolution established, among multiple parameters, that the diameter and the average intensity of the mitochondria are those which weigh the most in the explainability of age. Further analysis by flow cytometry confirmed the imaging explorations and revealed the coexistence of four subpopulations of mitochondria, stratified by their size, whose respective proportions change with age, with small mitochondria being the pool, most abundant at young ages. Phenotyping by flow cytometry showed functional differences between the subpopulations, particularly on the bioenergetic level with greater metabolic flexibility in large mitochondria compared to small ones. Sequencing of the mitochondrial DNA of each subpopulation demonstrated that there is a greater number of mutations in large mitochondria, including the presence of hotspots in the coding regions for nd1 , nd3 and nd5. All of these results suggest the existence of physiological heterogeneity in the basal state, but which seems to evolve with age, towards physiopathological heterogeneity. A better understanding of the biological properties of these subpopulations of mitochondria will make it possible to consider new therapeutic angles to combat the progression of muscle aging, and more broadly the pathologies associated with aging.

[0238] Introduction

[0239] Societies in both developed and underdeveloped countries are adapting to population aging due to significant increases in life expectancy, which now represents a major challenge for health care systems around the world. To alleviate the economic burden, it is necessary to reduce the dependency of the aging population and promote healthy aging. Frailty syndrome is a clinical consequence of functional decline associated with age, which clinically results in muscle loss. Sarcopenia is a progressive, systemic skeletal muscle disorder involving accelerated loss of muscle mass and function, associated with increased negative outcomes including falls, functional decline, frailty, and mortality. Several molecular mechanisms have been described as potential causes of the etiology of sarcopenia. Mechanisms linked in particular to hormonal function (e.g. IGF-1 and insulin), muscle fiber composition and neuromuscular training, proliferation and differentiation of myo-satellite cells or dysregulation of proteostasis have been proposed to play a crucial role, but recently, mitochondrial function has attracted much attention in this field, as a growing body of experimental evidence supports the major role of mitochondria in the aging process.

[0240] Mitochondria are essential organelles that have been studied extensively for over a century. Mitochondria are affected by aging (and vice versa) to very different degrees, depending on the tissue and species 1. First considered and stereotyped as the cellular energy powerhouse, coupling respiration to the production of 'ATP, the energy currency of the cell, it is now well established that mitochondria influence cell fate through various sensing and signaling pathways such as ROS production, calcium buffering, control of epigenetic programs or the determination of the sternness of cells 2,3. Mitochondria are heterogeneous and highly dynamic organelles, and their functions are subject to complex regulations through the modulation of their biogenesis, their bioenergetics, their dynamics and their elimination (mitophagy) within cells4. Mitochondrial dysfunction has previously been associated with the progression of sarcopenia, as evidenced by mitophagy defects, increased ROS production, and heteroplasmy. The function of mitochondria in energy production (ATP synthesis) and regulation of cell death is well characterized, but other aspects of mitochondrial biology are poorly described. Among them, the morphological heterogeneity that exists within mitochondria populations, and in particular the link that exists between structure and function, is not really understood. With the corollary, the question of the causal relationship between morphology and function? This morphological heterogeneity is very vast with the coexistence of (i) large round mitochondria, part of a network, connected to the endoplasmic reticulum or isolated, combined with (ii) an astonishing ultra-structural diversity which is closely linked to the activity of the oxidative phosphorylation system (OXPHOS) and (iii) huge differences in membrane potential. To better understand / define the link between mitochondrial morphology and function, the Applicant performed an in-depth analysis of the heterogeneity of mitochondria in skeletal muscle, which is a tissue (i) where mitochondria are present in very large numbers, (ii) which is primarily affected by aging. Using a combined approach with image analysis and flow cytometry on skeletal muscles from three different fish models ( zebrafish / zebrafish (Z- Fish), and African Turquoise Killifish (K-Fish) and mouse), the Applicant shows here that during aging, in these three vertebrates whose aging kinetics are different (K-Fish lives up to 6 months, while Z-Fish and mice can live up to 36 months) there is a reduction in locomotion associated with overall muscle loss which should be compared with the overall alterations, observed elsewhere, that mitochondria undergo in skeletal muscle. Using an electron microscopy image-on-image analysis approach, the Applicant discovered that, in all species, several pools of mitochondria coexist in healthy tissue from a young age under physiological conditions, and that the proportion of each sub- pool evolves during aging with a decrease in small mitochondria and an increase in large mitochondria during aging progression. These morphological changes are so unique that they make it possible, through the use of analysis tools based on artificial intelligence, to predict chronological age with an average accuracy of 90% in fish and 72% in the mouse. After sorting each of the subpopulations, ex vivo bioenergetic evaluations showed that large diameter mitochondria have greater membrane potential and ROS production than mitochondria of smaller diameters. The Applicant also showed that the decrease in mitochondrial DNA (mtDNA) during aging is accompanied by a distribution dependent on the size of mitochondria which evolves during aging. After deep sequencing, the Applicant were able to show that large mitochondria presented more mutations, specifically in certain areas such as the loci coding for nd3, nd5. It can be assumed that these mutations are the cause of the swelling, morphological and functional change of the mitochondria. Experiments are currently underway to try to understand the in vivo biological role of subpopulations. Overall, the data reveal that the mitochondrion should be considered a very complex system composed of a myriad of co-existing entities with different biological properties. This work may make it possible to better design therapeutic interventions or anti-aging strategies, or other potential mitochondrial pathologies, based on the transfer / transplantation of mitochondria.

[0241] Results

[0242] The decrease in locomotion and muscle mass during aging

[0243] The decrease in motor skills in elderly individuals is first characterized by a reduction in locomotion. In order to characterize locomotion in zebrafish (Z-Fish), killifish (K-Fish) and mice during aging, the Applicant first determined the activity, the distance traveled as well as the immobility time, zebrafish, killifish and aged mice compared to young individuals. As indicated in the diagram representing the monitoring of fish over 10 minutes (red line), the Applicant observed an alteration in the activity of aged Z-Fish and K-Fish compared to young individuals characterized by a reduction in the surface explored in the swimming tank in older individuals compared to young ones (Figure 12a-b). Interestingly, aged Z-Fish and K-Fish swim significantly less and remain more immobile than young fish, with a decrease in distance traveled of 45% and 35% respectively (Figure 12d-e). Similarly, in old mice, the distance traveled and the distance traveled in the training wheel is significantly decreased by approximately 80% compared to young mice (Figure 12c-f). These results suggest that locomotion, all species combined, in old individuals is considerably disrupted compared to young individuals.

[0244] To go further and understand if an alteration of muscle tissue could be the cause of this reduction in locomotion, the Applicant determined the size of the skeletal muscle fibers involved in animal locomotion. The Applicant made cross sections of muscle from Z-Fish, K-Fish and young and old mice and quantified the area of muscle fibers using the Cellpose algorithm capable of detecting different muscle fibers (Figure 12g-L). The Applicant observed a significant decrease in fiber size in aged fish (Z-Fish and K-Fish), as well as in aged mice compared to their young controls (Figure 12j-l). These data suggest that aged fish and mice exhibit marked muscle atrophy compared to their young controls. Muscle aging is characterized by muscle atrophy, in other words a reduction in the average diameter of fibers.

[0245] In the literature, there are a large number of publications which link the reduction in lean mass to mitochondrial dysfunction. Interestingly, the Applicant quantified by Electron Microscopy on skeletal muscle biopsies the quantity of mitochondria by species over time, and observed a significant decrease in the number of mitochondria during aging in Z-Fish, K-Fish and the mouse (data not shown). These data demonstrate that during aging, zebrafish, Killifish and mice exhibit alterations in the locomotion process associated with muscular atrophy as well as a reduction in the chondrioma, a probable reflection of a reduction in biogenesis, and / or alteration of clearance mechanisms. To properly characterize the type of mitochondrial damage during muscle aging, the Applicant began by exhaustively studying the mitochondria via their morphology as well as the ultrastructure by Electron Microscopy.

[0246] The diameter and ultrastructure of mitochondrial cristae is a strongly discriminating factor during muscle aging.

[0247] To study the morphology and functions of mitochondria, the best approach remains electron microscopy (EM), but black and white images represent a significant technical obstacle for automated and unsupervised analysis. Using mitochondria segmentation based on an off-the-shelf pipeline the Applicant developed, conventional ME images of muscle biopsies from the three models Z-Fish, K-Fish and mouse were processed using a pre-trained CellPose 2.0 algorithm (EMito-Metrix: Morin E, Doumard E et al 2023, in preparation). Using the Fiji interface, 20 metrics related to mitochondria morphology are extracted. Thanks to this ready-to-use pipeline, executed in fully automatic unsupervised mode, the Applicant were able to study in detail the morphology of mitochondria during aging (Figure 13).

[0248] Clearly, ME images of muscle biopsies from older individuals show mitochondria with enlarged shapes, unstructured and swollen cristae compared to young individuals where the mitochondria are rounder with aligned and structured cristae (Figure 13a). After detection and segmentation of mitochondria (Figure 13b), and using the EMito-Metrix pipeline, the Applicant can visualize what are the discriminating morphometric parameters depending on the species and age (young in blue and old in orange) on the radar plot (Figure 13c). Interestingly, the Applicant can easily see that the Feret diameter which corresponds to the distance between the two most distant points on the boundaries of the object is a parameter which varies significantly depending on age in the Z-Fish and the mouse. In addition, other parameters directly associated with Feret diameter such as the area and perimeter of mitochondria are parameters which vary with age. In addition to the global morphology parameters, the Applicant obtained ultrastructure parameters of the mitochondria, more precisely the content of the mitochondria (Mean Int, Total Int, Int SD, and Int SD % Mean Int) and interestingly, the Applicant observed that the average intensity of mitochondria was a parameter which increased significantly compared to the young condition, especially in K-Fish (Figure 13c). Looking in more detail at the histograms of the most impacted parameters, it can be seen that in the Z-Fish, the Feret diameter seems larger in old fish compared to young fish. In addition, the average intensity seems higher in older individuals compared to young people with a large number of mitochondria with a higher average intensity in both age classes and exacerbated in the elderly (Figure 13d). On the K-Fish histograms, the Applicant sees that the diameter of the mitochondria only seems to increase slightly with age while the average intensity seems to be drastically increased in old fish, with a distribution that shifts strongly to the left, associated with higher SD intensity in old individuals compared to young fish (Figure 13e). Finally, in mice the Applicant observes a slight increase in diameter in the aged condition accompanied by an increase in the average mitochondria intensity and SD intensity compared to the young condition (Figure 13f). These data suggest that morphology and ultrastructure are drastically impacted during muscle aging in all species.

[0249] Morphometric parameters of mitochondria predict the chronological age of Z- Fish, K-Fish and mice

[0250] Then and to go further in understanding the evolution of the morphological parameters of mitochondria during age, the Applicant used the EMito-Metrix tool, for an unsupervised multivariate analysis. The Applicant was thus able to determine whether, species by species, the morphometry of the mitochondria could make it possible to discriminate the age of the animals (figure 14a-c). On the UMAPs, while in the mouse the mitochondria from young or old animals are not distinguished (Figure 14c), in the K-Fish there is formation of two clear clusters composed of young mitochondria on one side and old mitochondria on the other (Figure 14b). On the other hand, although part of the same species (fish), in Z- Fish, the picture is more nuanced and only a small part of the old mitochondria are distinguished from the rest of the young mitochondria (Figure 14a). To go further in the interspecies comparison, the Applicant represented on the UMAPs the morphological parameters of aged mitochondria (Figure 14d-f). It appears that aged K-Fish mitochondria stand out from those of Z-Fish and mouse (Figure 14d;f), while the morphometric differences of aged mouse mitochondria and Z-Fish (Figure 14e) are minor. These results suggest that mitochondria evolve differently morphologically depending on the species. It is important to note that the K-Fish is a fish with a compressed lifespan, since it only lives 6 months, which stands out strongly from the Z-Fish and the mouse whose aging is much more gradual and it is certain that this compressed lifespan has a heavy impact on the function of mitochondria and their morphology.

[0251] As previously shown by way of example, the EMito-Metrix pipeline is capable of identifying the different species, based on the morphology of the mitochondria, using an analytical prediction module. In view of these results and the impact that the morphology of the mitochondria would seem to have on the prediction of age, the Applicant therefore determined whether the morphological and ultrastructure parameters of the mitochondria were discriminating enough to predict the chronological age of the mitochondria, individuals for each species (Figure 14g-l). For this the Applicant tested using two classes of learning algorithms: tree model (LR parser, Random Forests and XGBoost), and one neural network (Multi Layer Perceptron, MLP). The Applicant split the dataset in two with a training dataset and a test dataset (respectively 80% and 20%), and the Applicant compared the performances of each algorithm via the correct age prediction ( precision score or Accuracy score). The most efficient algorithm in this case is XGBoost, which is capable of correctly predicting the age of the animal in both fish models (Figure 14g-h) from images of mitochondria. However, the accuracy score is lower in mice with only 72% of correct predictions, which confirms the conclusions drawn from UMAP representations (Figure 14c; i), the morphology of mitochondria between young and old does not allow clustering in depending on age. In order to define the contribution of each variable in the prediction of the chronological age for each species, the Applicant was able to evaluate the values of SHAP (the Shapley Additive exPlanations (SHAP) (Figure 14j-l) applied to XGBoost. The graphs show the hierarchical and decreasing classification by the mean for the weight of each variable in the prediction. Like this, you can see that in an old Z-Fish, the area, the variability of the intensity or the total intensity contribute strongly to the prediction (Figure 14j-l). In other words, these 3 morphological parameters are very present in the aged mitochondria compared to the young mitochondria. This pattern of SHAP values is really specific to each species, since for the aged K-Fish and the aged mouse, the ranking of the morphological parameters of the mitochondria is really different. Taken all together, the Applicant could see that the morphometric parameters such as the area or the average intensity and the variability of the intensity, corresponding to the content of the mitochondrion, and therefore its ultrastructure (density and shape of the cristae), are parameters which change with age and which could have a significant impact on function.

[0252] Development of a flow cytometry approach to sort mitochondria based on their morphological parameters

[0253] Given the importance of the morphological parameters of mitochondria during aging, and to better study the link between the morphology and functions of mitochondria, the Applicant sought a technique allowing us to extract the mitochondria from their tissue, and to be able to sort them according to their size / structure. The only technology that can allow us to answer this question is flow cytometry. Even if this technique is trivial for whole cells, it took several months of development to be able to sort and analyze the functions of the mitochondria without their function being too altered. The first step consists of isolating mitochondria from a fish muscle biopsy. After cell lysis, the mitochondria are marked with an anti-TOM22 antibody coupled to magnetic beads, then after passing through a magnet and elution, the Applicant recovered the purified mitochondria. After isolation, the Applicant can label the mitochondria with mitochondria-specific fluorescent probes like MitotrackerGreen (MTG) or MitotrackerRed (MTR) so that the sample can be detected and analyzed by flow cytometry (Figure 15a). Given the scale of the task, for the rest of the work, the Applicant only used the two fish models.

[0254] In order to analyze mitochondria by size, the Applicant calibrated the cytometer using fluorescent beads of known sizes ranging from 0.24um to 1.35um (Figure 15b). After passing the sample of mitochondria through the cytometer and selecting the mitochondria marked positively with MTG (Figure 15c) which appear in the form of a continuum provided, the Applicant can segment the MTG+ population into different subpopulations ranging from P1 to P4 corresponding to different sizes of mitochondria (Figure 15d). Finally, the Applicant can apply the template to the different age conditions and analyze the distribution by size range of mitochondria (Figure 15e). As shown in the event density graphs, where a point represents a mitochondrion, the Applicant can observe a high density in the gates of mitochondria of small sizes P1 and P2 in the young condition while the Applicant observed a high density in the gates corresponding to the large P3 and P4 mitochondria in the aged condition (Figure 15e).

[0255] Once the workflow was established, after isolation, labeling and reading of the mitochondria samples with a cytometer, the Applicant studied the distribution by size range of mitochondria from skeletal muscle from Z-Fish (Figure 15f-h) and K-Fish (Figure 15i-k) using a pan-mitochondria marker MTG. To verify mitochondria viability / function, the Applicant performed TMRM labeling, which is an indicator of mitochondrial membrane potential. In the event of damage to the internal and / or external mitochondrial membranes, the TMRM will no longer be able to bind the electrons present in the matrix and therefore a reduction or even a total loss of the signal will be observed. As shown in the graph (Figure 48f; 15i), in the young and aged condition the mitochondria present a double MTG+ and TMRM+ labeling of around 80%, suggesting that the integrity of the majority of the mitochondria has been preserved whether in the Z-Fish or the K-Fish. In order to analyze the distribution by size range of mitochondria, the Applicant analyzed the mitochondria of 2-month-old Z-Fish by applying size segmentation which tells us about the percentage of mitochondria in each size range. Interestingly, the Applicant notices a large size heterogeneity in a 2-month-old Z-Fish, with a greater proportion of small mitochondria (P1) than large mitochondria (P4) (Figure 15g). By comparing the distribution by size range of an “old” Z-Fish aged 24 months, the Applicant notices a great diversity in size with a proportion of small mitochondria (P1) significantly reduced from 25% in young Z-Fish to 10% in aged Z-Fish, accompanied by a greater proportion of large mitochondria (P4) in aged fish compared to young ones increasing from 10% to 5% respectively (Figure 15h).

[0256] In order to verify whether this result is found in another fish species, the Applicant analyzed K-Fish skeletal muscle mitochondria by applying the pipeline from mitochondria isolation to flow cytometry analysis. As for the Z-Fish the Applicant carried out a double MTG / TMRM marking in the young as well as the aged condition. The mitochondria double labeled MTG+ T+MRM are around 80%, suggesting that here too the majority of mitochondria have been functionally preserved (Figure 15i). The Applicant then applied segmentation by size. The Applicant noticed a great heterogeneity in size in a young K-Fish aged 4 weeks, with a greater proportion of small mitochondria (P1 ) than large mitochondria (P4) (Figure 15j). By comparing the distribution by size range of a 20-week-old K-Fish, the Applicant noticed a great diversity in size with an abundance of small mitochondria (P1) significantly reduced, going from 35% in young K-Fish to 20%. in aged K-Fish, accompanied by a greater proportion of large mitochondria (P4) in aged fish (10%) compared to young fish (3%) (Figure 15k).

[0257] All these results demonstrate that the use of such a pipeline makes it possible, on the one hand, to isolate and analyze viable mitochondria by size range and, secondly, to highlight the existence of great heterogeneity, size / structure of mitochondria, which is impacted during muscle aging. Furthermore, these results are found in two different fish species, with mitochondria of larger size during aging in skeletal muscle, reproduce the results obtained by the unsupervised analysis of electron microscopy images, suggesting that the extraction mitochondria from their tissue of origin did not drastically modify the morphology, or at least their size (results not shown). Coexistence of mitochondria with heterogeneous metabolic capacities during aging in Z-Fish.

[0258] In order to determine whether these different identified mitochondria populations supported different functions during muscle aging, the Applicant undertook to precisely characterize each subpopulation in terms of bioenergetics, ROS production, as well as their mitochondrial DNA content. Firstly, the Applicant marked the mitochondria with the cationic probe TMRM, which will bind the electrons present at the mitochondrial matrix, an indicator of the mitochondrial membrane potential; high TMRM labeling is indicative of a high mitochondrial membrane potential. The Applicant then treated the mitochondria with FCCP, which is a widely used uncoupling agent; its activity has the effect of reducing the membrane potential by uncoupling the electron transport chain.

[0259] After double MTG / TMRM labeling and treatment with 1 mM FCCP, the Applicant observed a significant decrease in TMRM labeling in the young and aged condition following treatment with FCCP (Figure 16a). Looking at the average fluorescence intensity of TMRM labeling per subpopulation at different ages, the Applicant first noticed an exponential increase in TMRM labeling as a function of mitochondria size, with for each subpopulation a tendency towards an increase. TMRM labeling in “aged” mitochondria (Figure 16b). The FCCP treatment leads to a similar decrease in the membrane potential of the subpopulations (Figure 16b-d) but be careful, the effectiveness of the FCCP treatment is minimized by the use of the logarithmic scale. Overall, and independent of age, larger mitochondria have a higher membrane potential, suggesting a probably more active ETC. To go further in the characterization of subpopulations of mitochondria, the Applicant wanted to analyze the production of ROS using the MitoSox probe marking ROS. The Applicant did not observe differences in MitoSox marking in young vs. old bulk analysis (figure 16e - left panel) but the analysis of MFI by subpopulations shows that the production of ROS would be increased with the size of the mitochondria and that this phenomenon is exacerbated with age (Figure 16e-right panel). Then, and in order to continue the in-depth characterization of mitochondria subpopulations, the Applicant questioned the content of mtDNA, known to be drastically impacted during aging. The Applicant therefore labeled the mitochondria with PicoGreen, which is a fluorescent DNA intercalator. In total analysis the Applicant were able to observe a trend towards a decrease in DNA content (Fgure 16f-left panel). In analysis disaggregated by subpopulation the Applicant observed an increase in the quantity of mtDNA in large mitochondria but with a significant reduction with age in picoGreen labeling for P3 and P4 (Figure 16f-right panel). By analyzing the picogreen labeling, it appears clear that many mitochondria are devoid of mtDNA. The Applicant were able to observe that each class of mitochondria has MitotrackerDeepred marking marking all the mitochondria lacking PicoGreen marking. The value of A(%MTDR+ - %MTDR+ PicoGreen+) reflects the mitochondria without DNA in both age classes (figure 16g). On the histogram representing this delta, the Applicant observes that the small mitochondria (P1 and P2) present a greater number of mitochondria devoid of mtDNA compared to the large size (P3 and P4), an observation which does not change with the 'age. To complete this analysis, the Applicant wanted to look at the abundance of proteins involved in mitochondrial dynamics, particularly in mitochondrial fusion, with Mitofusin 2 (Mfn2) and OPA1 (Figure 16i-k). Unfortunately at this stage, the Applicant has not been able to obtain reliable and robust data using fission markers such as DRP1 , the specificity of the antibody used raising questions. In group analysis, mitochondria from aged Z-Fish have the same Mfn2 and OPA1 marking as young fish (Figure 16i). However, looking by subpopulation, MFN2 marking increases with the size of the mitochondria until P3 then decreases at P4 in the young. But the Applicant noted an increase in MFN2 in the P1-Aged and a decrease in the P4-aged. (Figure 16j). Interestingly, OPA1 follows the same trend, with an increase in the number of mitochondria expressing OPA1 as a function of size. With OPA1 seems more expressed in P3 and less in P4 (Figure 16k), suggesting that perhaps disturbances of the fusion mechanisms, via the decrease in expression of MFN2 and OPA1 , do not participate in the increase in the size of mitochondria.

[0260] Coexistence of mitochondria with heterogeneous metabolic capacities during aging in K-Fish.

[0261] To determine whether the observations made in the Z-Fish are robust and reproducible, the Applicant conducted the same experiments in the K-Fish. The experimental strategy is therefore modeled on that used for Z-Fish, and the characteristics of mitochondria isolated from young (1 month) and old (5 months) K- Fish were compared (Figure 17). Bioenergetically, membrane potential increases with mitochondria size without age-related differences. The flexibility of the CTE is also increased in P4 compared to other subpopulations, again without any impact of age (Figure 17a-d). Lipid peroxidation increases with age, and this peroxidation is also increased with the size of mitochondria in young K-Fish (Figure 17e). But for P4, peroxidation is more important in old K-Fish compared to young P4, which could be linked to a more active CTE (Figure 17e). Concerning the mtDNA content, it is interesting to note that unlike the Z-Fish, there is no decrease in mtDNA with age (Figure 17f), and that the mtDNA content increases with the size of the mitochondria without age effects. Similar to Z-Fish, although at a lower amplitude, a proportion of each mitochondria subpopulation is devoid of mtDNA (Figure 17g). Remarkably, large populations (P3 and P4) lacking mtDNA are present in greater quantities during aging (Figure 17h). To complete this analysis, the Applicant wanted to look at the expression of proteins involved in mitochondrial dynamics, particularly in mitochondrial fusion, with Mitofusin 2 (MFN2) and OPA1 (Figure 17i-k). In pooled analysis, the expression of MFN2 and OPA1 is similar between young and aged K-Fish (Figure 17i). Looking by subpopulation, MFN2 and OPA1 markings increase from P1 to P3, but their detection decreases in P4 in young and old K-Fish (Figure 17). Overall, the heterogeneity of mitochondria in K-Fish is quite similar to that of Z-Fish, highlighting the robust nature of the observations in two vertebrate models with very different aging kinetics.

[0262] Mitochondrial heterogeneity is physiological and present from the first days of life in Z-Fish

[0263] The mitochondrion is therefore a plural entity both morphologically and functionally in a young adult fish of two months. This heterogeneous distribution of mitochondria size seems maintained in a young organism where mitochondrial biogenesis or quality control mechanisms are still efficient, suggesting that each of these populations plays a biological role. The next question was to know if this mitochondrial heterogeneity is identical to earlier stages in the Z-Fish, and in particular in the 7-day-old larva (Figure 18). After dissociation of the larva as a whole, the mitochondria can be stratified between P1 and P4 with proportions different from those observed in the 2-month-old Z-Fish, and in particular a small proportion of P1 (Figure 18a), but the whole subpopulations show good viability (Figure 18b). The TMRM marking also seems robust, since the FCCP treatment causes the marking to be lost on the entire chondriome (Figure 18c), but strong disparities appear between the subpopulations when the Applicant compares the MFI of the TMRM marking (Figure 18d) where the P4 have the highest membrane potential but a lower FCCP effect.

[0264] Concerning the mtDNA content, here too the marking shows a great disparity with the P3 population which presents the greatest percentage of mitochondria containing mtDNA (Figure 18e). But when quantifying MFI, P4 mitochondria show greater mtDNA content on average (Figure 18f-g).

[0265] Taken all together, these results show that in a young and developing organism, the subpopulations of mitochondria are already in place, with different proportions than in adulthood, but with very similar characteristics (membrane potential, and mtDNA content). This suggests that the presence of these subpopulations of mitochondria is physiological, and that each of them with their different morphology plays different biological roles with their intrinsic characteristics.

[0266] Regulation of the proportion and functions of mitochondrial subpopulations by intermittent young

[0267] After showing that the coexistence of subpopulations of mitochondria was physiological in the larva and young adult Z-Fish, the Applicant wondered if in the aged Z- Fish, there was the possibility of regulating the abundance of subpopulations, of mitochondria through a nutritional approach capable of stimulating mitochondrial biogenesis and mitophagy mechanisms. To do this, the Applicant subjected the 8-month-old Z-Fish to a 4-week intermittent fasting protocol (IF), a nutritional intervention known for its positive effects on chondrioma. First of all, on a quantitative level, the relative abundance of the subpopulations is similar in the aged Control group (fed Ad Libitum “AL”) and the aged IF group (Figure 19a).

[0268] Nevertheless, subpopulations of small mitochondria like P1 and P2 show an increase in the percentage of double-labeled MTR+ TMRM+ events, suggesting that there are more mitochondria in P1 and P2 showing an increase in mitochondrial respiration (Figure 19b - vs). But when the TMRM labeling is analyzed by the MFI, the Applicant can see that the MFI for the TMRM of the P1-P4 subpopulations increases significantly, only in the IF condition, an increase which is entirely reduced with the FCCP (Figure 19c-f). Considering the increase in mitochondrial respiration under IF, the Applicant measured ROS production in parallel (Figure 19g-i) in AL and IF conditions. As expected, the quantity of ROS is generally reduced in the chondriome of Z-Fish subjected to IF (Figure 19g), this intervention being known for its anti-oxidant virtues in particular through more active mitochondrial biogenesis. The proportion of mitochondria as a function of the quantity of ROS remains similar in AL and IF (Figure 19h), but the MFI analysis shows greater labeling in P4 in IF condition, suggesting that the increase in mitochondrial respiration observed in P4, is accompanied by an increase in ROS (Figure 19i).

[0269] The Applicant measured the impact of IF on the mtDNA content of mitochondria. The bulk analyzes show a trend towards an increase in the quantity of mtDNA under IF, but the mtDNA is distributed in the same way in each subpopulation compared to the control diet (Figure 19j-k) . On the other hand, by the analysis of MFI, the quantity of mtDNA is increased in the P4 subpopulation of Z-Fish fish in IF (Figure 191).

[0270] Taken all together, these results suggest that without changing the proportions of the P1-P4 subpopulations, IF made it possible to restart mitochondrial respiration, particularly in P3 / P4, with the concomitant effect of an increase in mtDNA storage and production, of ROS (especially in P4).

[0271] Copy number and high mtDNA mutation burden are associated with mitochondrial morphology and age

[0272] The results show that in addition to the morphological heterogeneity of mitochondria, other indices, such as mtDNA content which seem to vary greatly within subpopulations, must be taken into account to further complicate this heterogeneity. First of all and to complete the analyzes carried out in flow cytometry with picogreen, the Applicant carried out a quantitative evaluation of the mtDNA content of the P1 and P4 subpopulations of Z- Fish by digital PCR (Figure 20a). As published elsewhere and shown in the work (Figure 16f), the Applicant observed that in the young Z-Fish P4 contains approximately 10 times more mtDNA than the P1 population, which shows to what extent mtDNA is not distributed evenly within the subpopulations of mitochondria (Figure 20a). Furthermore, with age, the amount of mitochondrial DNA decreases, and furthermore this decrease is only significant for the P4 population (Figure 20a). Using a very sensitive and quantitative technique the Applicant confirms the results obtained by flow cytometry, which solidifies the conclusions and interpretations. After this quantitative analysis, and to go further, the Applicant decided to sequence the mitochondrial DNA contained in the P1 and P4 subpopulations isolated from four young (2 months) and old (24 months) Z-Fish (Figures 20 and 21). The quantity of mtDNA being quite low at the outlet of the sorter, the Applicant went through a PCR amplification step, where each mtDNA molecule is amplified by the use of two overlapping PCRs (PCR1 :9kb and PCR2:11kb). This step also allows us to avoid possible contamination of the nuclear DNA. After validation of quality control, the Applicant were therefore able to sequence the 16 samples (4xP1 2M, 4xP1 24M, 4xP42M and 4x P424M). For each condition, approximately one hundred thousand sequences were obtained, filtered and aligned to the Zv11 reference genome. The sequences cover all of mtDNA. Sequence variants were then detected (variant calling) and normalized. Certain sequence variants were found in all the samples analyzed with an allelic frequency greater than 99%, and made it possible to define a “French zebrafish” haplotype. For the rest, the Applicant will talk about sequence variants as mutations. As expected, the number of mutations present on the mtDNA molecule is greater in animals aged 24 months compared to those aged 2 months (Figure 20b-d). Qualitatively, these mutations are more present on loci which code for proteins (Figure 20b), these mutations are in the vast majority transitions (Figure 20c) which result in non-synonymous mutations, although a large quantity of Mutations remain associated with synonymous changes (Figure 20d). Quite similarly, when the analysis is stratified on the mutation subpopulations (P1 and P4), the P4 subpopulation has the same mutational profile as the 24 months, which suggests that the P4 subpopulation is the one, among the subpopulations, which has the highest amount of mutations on the mtDNA molecule (Figure 20e-g). When the analysis is stratified by time and subpopulations, the results show that the P4-24M subpopulation is the one carrying the overwhelming majority of mutations (Figure 20h-j). Schematically, a hierarchy of mtDNA preservation seems to emerge, with from the most native to the most mutated P1-2M, P1-24M, then P4-2M and P4-24M.

[0273] High penetrance of mutations in the nd1, nd3 and nd5 loci is associated with the morphology of large mitochondria in aged individuals

[0274] The rest of the analysis consisted of identifying the genes or loci which present a high frequency of mutations, depending on the size of the mitochondria and their age (Figure 21). Mutation frequencies (allele frequency: AF) were averaged over sliding windows of 0.1 kb, over the entire 16 kb of mtDNA. The significant FAs were represented on a heatmap (Figure 21a) and a Manhattan plot (Figure 21 b), which makes it possible to identify the top loci where the frequency of mutations is high, depending on age and subpopulations. First of all, the unsupervised clustering of the heatmap shows us a rapprochement of the subpopulations independently of age, in other words the mutation profile of P4-2M is very close to P4-24M. This is confirmed by the presentation in Manhattan plot, where the mutation profiles, especially for large AF, or P1 and P4 independently of age, stand out. These first results suggest that the P1 and P4 populations could constitute two pools of mitochondria independent of each other. In P4-24M, which is the pool with the most mutations, several loci coding for proteins seem to present a high frequency of mutations, such as nd3, nd1 , nd5 and, to a lesser extent, co3 and cytb (Figure 21a-b), confirming the results presented in Figure 20. The gene-by-gene analysis regardless of their class once again confirms a large quantity of mutations in the protein-coding loci in the P4-24M population, and in particular the nd5 locus. T o functionally validate that the mutation of these genes causes a change in mitochondria morphology, it would be necessary to be able to make targeted mutations of the mtDNA at the identified loci. Currently it is still difficult to use CRISPR technology specifically in mtDNA, although a few teams seem to have promising work. To circumvent this technical obstacle, the Applicant is working on the injection, into Z-Fish embryos at the one-cell stage, of restriction enzyme mRNAs carrying a site for addressing the mitochondria, which therefore only allow the cleavage of mtDNA (the only one accessible by the restriction enzyme). Furthermore, the chosen restriction enzyme will present a single restriction site in the mtDNA in order to avoid the deletion of large fragments and promote the formation of more discrete mutations following DNA repair. The Applicant proposes that the generation of mutations in the sequences can lead to “nonsense” mutations in the targeted locus, and lead to an alteration in the morphology of mitochondria as observed in elderly individuals. Thus, the Applicant hopes to be able to demonstrate a cause and effect link between the mutation of these loci and the morphology / structure of “aged” mitochondria. These validation experiments are currently underway.

[0275] Only mitochondria from young individuals are capable of supporting embryonic development in Z-Fish

[0276] To complete this study, and try to understand, beyond the characteristics of subpopulations, the biological function of mitochondria, the Applicant tried to set up an in- vivo model where the Applicant could evaluate the biological functions in the broad sense of the term, subpopulations of P1 and P4 mitochondria isolated from young and old fish. Before the Applicant could get there, the Applicant sets up a “RhoZero-like” mosaic embryo model in which the mtDNA of certain cells is eliminated (Figure 22). To do this, at the onecell stage, the Z-Fish embryo is injected with an mRNA coding for the restriction enzyme Ncol addressed to the mitochondrion (where the GFP mRNA) (Figure 22a) which was chosen to cut the Z-Fish mtDNA in 3 distinct positions, the cytb and tRNA 10 and tRNA18 loci (Figure 22a). Note that this enzyme was chosen so as not to be able to cut the mtDNA of K-Fish, and avoid damaging the mitochondria which will be injected for a possible “rescue” of embryonic development. After injection of mtNcol mRNA, more than half of the embryos present developmental defects (DD: developmental defect), and another part of approximately 30% have comparable embryonic development (ND: no defect) to the control ( CTL: GFP mRNA injection) (Figure 22b-e). When the Applicant look at the digestion pattern of a PCR product flanking the 3 loci targeted by Ncol under CTL and ND conditions, the PCR products are completely digested generating 2 bands, on the other hand in the case of DD embryos, a 3rd band appears in the digestion profile suggesting that following the action of mtNcol, “nonsense” type repair took place (Figure 22g-i). The model therefore seems to work as expected, and the Applicant therefore used the same system for the rescue manipulations, where the mRNA of the mtNcol enzyme is co-injected with bulk mitochondria isolated from young or old K-Fish. In this context, the Applicant can see that after injection of young mitochondria, no embryo dies and the number of DD decreases drastically (Figure 22f). Whereas, conversely, the co-injection of old mitochondria even leads to an increase in the number of dead embryos, and a decrease in the number of ND embryos (Figure 22f). Without going into the study of subpopulations yet, it appears clear that the bulk of young or old mitochondria have very different properties, and only young mitochondria are capable of supporting embryonic development. The deleterious effects of aged mitochondria are probably multiple between bioenergetic deficit, fragility and pro- inflammatory action, numerous hypotheses are possible. But the main thing for the rest of the project, and that this “RhoZero-Embryos” protocol seems to work (Figure 22j-m), since certain cells present mitochondria marked in red without any DAPI marking after injection of mtNcol, in addition the mitochondria co-injected and marked in green are identified in the cytoplasm of certain cells. These are the cells which notably present DAPI marking of the cytoplasm suggesting the presence of mtDNA. The Applicant will therefore be able to very quickly test the impact of P1 and P4 derived from young and old K-Fish on the rescue of “Rho-Zero- Embryos” from Z-Fish, and in particular measure proliferation, cell migration or respiration, mitochondrial (SeaHorse).

[0277] Discussion :

[0278] The work has made it possible to confirm certain concepts already described in the literature, this time using a new unsupervised approach, and further details what mitochondrial morphological heterogeneity represents in the physiological state but also physiopathologically during aging. The work has made it possible to establish the coexistence of subpopulations of mitochondria according to their morphology in an organism just after birth or as a young adult, this is what the Applicant calls physiological heterogeneity. Although there is still work for an exhaustive characterization, the Applicant can say that large mitochondria (P3 / P4) have a higher membrane potential than small ones (P1 / P2), even when weighted by mitochondria size, with a better response to the FFCP. P3 / P4 mitochondria contain more mitochondrial DNA than P1 / P2, even after normalizing for size, and unlike P1 / P2, almost all P3 / P4 contain mtDNA. With age, through mechanisms that are still poorly identified, the proportion of small mitochondria (P1 and P2) decreases, while the number of large mitochondria increases, this is what the Applicant could call the physiopathological heterogeneity. Aged P3 / P4 show greater membrane potential than young P3 / P4, but lower mtDNA content. Even when aged, mitochondria from P1 to P4 are capable of responding to nutritional intervention such as intermittent fasting, known to improve metabolic flexibility. The mtDNA content decreases with age in P4, but the quantity of mutations increases over time, particularly at protein-coding loci such as nd1 or nd3. Many questions emerge from this work, which sometimes challenge certain dogmas. First of all, it seems that from a very young age, mitochondria are plural and present very broad morphological heterogeneity. The results demonstrate that large mitochondria exist from the beginning of life, and are perfectly functional. These mitochondria even have a high membrane potential, again emphasizing their vitality. This questions the idea that large mitochondria could correspond to dysfunctional mitochondria, which persist in cells with affected mitochondrial biodynamics and / or altered mitophagy pathways 12,13. Along the same lines, another result also questions other dogmas, and in particular the significant proportion of mitochondria which seem to be devoid of DNA molecules in the basic state, a phenomenon which is also accentuated by age. In a healthy cell or organism, where the Applicant knows that mitochondrial biogenesis is active, and the quality control mechanisms (mitophagy) are effective, the Applicant can hypothesize that these mitochondria are not eliminated and play a role, biological role which remains to be determined. However, the Applicant can question the experimental design, where the use of PicoGreen may present limits. Indeed, picogreen remains a very sensitive probe, but it needs to cross a double membrane to mark the mtDNA, and the fluorescent signal also needs to cross this double lipid membrane to be detected. The Applicant can legitimately wonder if the proportion of mitochondria lacking mtDNA is not due to technical problems. To complete these results, the Applicant are working on strategies combining immunodetection and electron microscopy. As already used by the group of Nils-Goran Larsson 14, the Applicant could use an anti-TFAM antibody to detect the nucleoid, and evaluate the percentage of mitochondria lacking TFAM signal. The advantage of this technique would also be to associate the morphology of the mitochondrion with the quantity of mtDNA. This surprising observation has also been described by other teams who were able to describe the presence of this type of mitochondria devoid of mtDNA, not always in a pathological context 15,16. Depletion of mtDNA has often been linked to mitopathies or serious mitochondrial dysfunctions 17,18,19, it is possible that this is true in a certain context where a majority of the chondrioma is affected, but in basal conditions, these mitochondria lacking mtDNA could play a biological role which remains to be defined, as do those like P4 which “store” large quantities of mtDNA.

[0279] One of the most burning questions raised by the work concerns the potential connection between small and large mitochondria, in other words are P1 and P4 related? Indeed at this stage the Applicant can consider two options, (i) either P1 / P2 play their role within the expected time window, and then slowly drift towards larger mitochondria thanks to defective clearance mechanisms and become P3 / P4. The opposite scenario can also be envisaged with P3 / P4 which, thanks to fission mechanisms, could produce P1 / P2 type mitochondria. The second option (ii) could also be that P1 / P2 and P3 / P4 are completely independent entities from their genesis until their degradation. In light of the results, and in particular those of mtDNA sequencing, the Applicant believes that the second option is potentially the most credible. Indeed, one way to connect subpopulations with each other is to compare the profile of mutations. Logically if P3 / P4 derive from P1 / P2 (or vice versa), their mutation profile should be at least partly superimposable. The results show a real difference between the mutation patterns between P1 and P4, and the unsupervised clustering (Figure 21a) highlights a difference between the overall profile of the two subpopulations, and even under the effect of age, the two subpopulations evolve differently, and here too the heatmap and the Manhattan plot (Figure 21 b) show that the mutation patterns do not overlap / converge. It therefore seems that these two populations evolve distinctly from a very young age, depending on their biological properties and the stress to which they are subjected. The Applicant can think in particular of oxidative stress which is more important in P4 mitochondria than in P1 , which could explain the difference in the quantity of mutations (Figure 16e).

[0280] The regulation of the proportions of P1 / P2 and P3 / P4 during life is also a very intriguing question. The pathways regulating the abundance of mitochondria are mitochondrial biogenesis on one side and mitophagy on the other. Could the disruption of one of these pathways (or both) during aging contribute to the respective proportion of mitochondrial subpopulations? To already see if these pathways are engaged in the system, the Applicant subjected the aged fish to 4 weeks of intermittent fasting which is known to revitalize mitochondrial biogenesis and mitophagy mechanisms 20,21. In terms of proportion, only the P1 and P2 populations increased in fish subjected to IF, and no change was noted for P3 / P4 (Figure 19b). It appears that P1 / P2 mitochondria are dependent on biogenesis. In the same context under IF, the P3 / P4 population does not vary, even in the context of aging, which suggests that P3 / P4 does not owe its existence to a loss of efficiency of degradation systems. And even without varying in terms of abundance, P3 / P4 mitochondria, under IF, seem to be able to improve their respiration with an increase in the quantity of mtDNA (Figure 19f-i). But be careful, the P3 / P4 pool may have been completely renewed and replaced by a de novo pool with better performance or the P3 / P4 by mechanisms. In any case, it also seems clear that small and large mitochondria do not respond in the same way following a reactivation of the mitochondria production / degradation mechanisms. To see more clearly, these pathways could also be challenged by pharmacological treatments, such as Urolithin A which is known to activate mitophagy pathways 22 or SIRT1 or AMPK agonists known to stimulate mitochondrial biogenesis 23. These molecules could be tested in aged fish, and see to what extent the proportions of small and large mitochondria will be affected. But all the results, following IF treatment, lead us to believe that P1 / P2 could be renewed regularly over time, but that P3 / P4 could have a much longer lifespan, which is supported by the quantity of mutations which is much greater in P3 / P4 compared to P1 / P2. At this stage it is only a hypothesis, and numerous experiments must be carried out to support this concept.

[0281] One of the next challenges of the study will consist of going further in the functional characterization of P1 / P2 / P3 and P4 mitochondria. One of the pitfalls the Applicant faces is technical, since the only strategy used in this study to sort the mitochondria is flow cytometry. Given the pressure in the sorting system and the length of the tubing, the Applicant recovers very few mitochondria per subpopulation, and their low viability prevents us from considering exploratory experiments, or transfers into Rho Zero cells, or zebrafish embryos. Recently the Applicant was able to carry out some tests using equipment (the Tyto-Miltenyi sorter) which allows mitochondria to be sorted at very high throughput without tubing, and without any pressure. The idea would now be to sort the subpopulations of mitochondria on Tyto, and to be able to characterize them functionally either by SeaHorse or oxygraphy, evaluate the components of the respiratory chain, and evaluate their biological function. For example after transfer to RhoZero of P1-2M / P4-2M / P1-24M and P4- 24M, the Applicant could measure their proliferation and migration capacity. The Applicant could measure the differential impact of subpopulations of mitochondria on the host cell, notably by RNA sequencing or metabolomic phenotyping. Similar experiments are planned on zebrafish embryos, which at the one-cell stage will be injected with P1-2M / P4-2M / P1- 24M and P4-24M. Thus the Applicant will be able to go further in the characterization of biological properties, and plan on the possibility of using certain subpopulations of mitochondria to fight against aging.

[0282] Material and methods :

[0283] Housing and handling of animals

[0284] Zebrafish and Killifish:

[0285] Zebrafish and killifish were handled in a facility certified by the French Ministry of Agriculture. Every effort has been made to minimize the number of animals used and their suffering, in accordance with the guidelines of the European Directive on the protection of animals used for scientific purposes. In accordance with the guidelines of the European Directive on the protection of animals used for scientific purposes (2010 / 63 / EU) and the guiding principles of French Decree 2013-118. For flow cytometry analyses, the Applicant used zebrafish aged 2 and 24 months and killifish aged 1 month and 5 months. The animals were subjected to a 14h / 1 Oh photoperiod (Apafis: Z-Fish n : 2368, K-Fish: 43902) The fish were maintained at 27°C in collective tanks (Z-Fish) or individual tanks (K-Fish)

[0286] Mouse :

[0287] The mice are housed in accordance with the principles and guidelines of the medical research institute and under aoafis n: °21-3U1301-CD-01.

[0288] Isolation of purified mitochondria from zebrafish and killifish skeletal muscles:

[0289] Approximately 20 milligrams of skeletal muscle from Z-Fish, or K-Fish, was removed and finely chopped on a board kept on ice. The muscle was homogenized in 1 ml of ice- cold lysis buffer from the Miltenyi Biotec Mitochondria Isolation Mouse Tissue Kit (Ref 130- 096-946) supplemented with a Halt Protease / Phosphatase Inhibitor Cocktail (Pierce), with 10 passages in a homogenizer in 2 mL glass (Sigma). The Teflon piston was smoothed to avoid any frictional forces between the piston and the Potter glass. The homogenate was diluted with 10 mL of 1X separation buffer and mixed by inversion, then 50 p L of antibodies coupled to anti-TOM22 magnetic microbeads were added. Samples were incubated on a rocker at 4°C for 1 hour in the dark. Lysates were applied to LS columns pre-moistened with cold 1X separation buffer. The columns were rinsed 3 times each with 2 mL of 1X separation buffer, before elution in 1 mL of IBM2 respiration buffer containing; 120mM Glucose, 50mM kCL 20mM Tris HCL 4mM KH2PO4 2mM Mgcl2 1mM EGTA 10mM Pyruvate 5mM Glutamate 5mM Malate, pH 7.4. The lysates are then distributed into different tubes for flow cytometry markings.

[0290] FACS labeling and analysis of mitochondria

[0291] Calibration of the cytometer with beads of known sizes:

[0292] The cytometer was calibrated with a mix of beads purchased from Spherotech (ref: NFPPS-52-4K) and Biocytex (ref: 7803) to compose a size range from 0.2um to 1.35um detailed below.

[0293] Mitochondria gating and sorting strategy:

[0294] In order to be able to highlight the populations of mitochondria of different sizes, the Applicant first analyzed beads of predefined sizes (Megamix-Plus SSC BioCytex, Marseille- France). Megamix-Plus SSC is a mixture of fluorescent beads of various diameters (0.16 pm, 0.20 pm, 0.24 pm, 0.5 pm, 0.88 pm and 1.35 pm) dedicated to flow cytometry using side detection (SSC) as a size-related parameter. Acquiring beads according to the procedure allows the cytometer to be adjusted to study mitochondria in a region of constant size. Data acquisition was performed using logarithmic scales for the instrument scatter, forward scatter (FSC) and side scatter (SSC) parameters, respectively associated with the size and internal complexity of the instrument, event analyzed. The windows of sizes P1 to P4 are thus defined after calibration of the cytometer P1 grouping together the small-sized events and P4 those of the large-sized ones. Freshly isolated mitochondria are selected based on MitotrackerGreen marking. Thanks to the windows of sizes P1 to P4, the different populations of mitochondria are thus highlighted according to their size. Data acquisition was carried out using the “BD FACS Diva software” on the FACS ARIA III (Becton Dickinson ). The device parameters, size windows and acquisition speed and the number of events collected were fixed for all specimens. The analyzes are carried out with Kaluza software version 1.2 (Beckman Coulter). The mitochondria were sorted according to the subpopulations of interest positive for MTG+ markings.

[0295] Intermittent youth protocol:

[0296] 8-month-old Z-Fish whose food was provided ad libitum (AL) until 7 months of age were used in this study. At this time, fish were assigned to one of two groups: AL, fed ad libitum; IF, having access to food every other day for 4 weeks, with at least 3 animals per group, and the animals were sacrificed. All procedures were approved by INSERM. mtDNA extraction and sequencing

[0297] The sorted mitochondria populations were centrifuged at 10,000g 10' 4°C. The DNA of the different mitochondrial subpopulations was extracted using the DNeasy Blood & Tissue kits (Qiagen) after 16 hours of incubation in a buffer, lysis supplemented with 0.2 mg / ml proteinase K and RNAseA (Qiagen). 1ng of DNA is used to amplify the mtDNA with TaKaRa LA Taq® DNA Polymerase (TAKARA) into 2 PCR fragments of 9 and 11kb respectively. Libraries were generated from an equimolar preparation of the two PCR products using the XT DNA Library Prep Kit (Illumina) and sequenced on Miseq (Illumina).

[0298] Construction of the plasmid

[0299] The mtNcol gene was synthesized by PCR assembly of seven pairs of complementary oligonucleotides of 46 to 100 nucleotides which have an overlap of 20 to 25 nucleotides between the fragments. The mtNcol gene is composed of the SP6 promoter, the human cox8 mitochondrial targeting signal, the Ncol coding sequence and the SV40 polyadenylation site. Injections into the embryo

[0300] 250 ng / ul of GFP and mtNcol mRNA synthesized by SP6 mMessage Machine (Lifetechnology). Isolated mitochondria are concentrated by centrifugation, resuspended in a minimal amount of IB2M buffer, and kept on ice before injection.

[0301] Genotyping

[0302] For genotyping, DNA was extracted by incubating embryos in NaOH solution for 15 min. To measure the efficiency of mtNcol, PCR and Ncol digestion were performed for each putative mtDNA mtNcol target site.

[0303] Embryo imaging

[0304] High-resolution images of 3hpf embryos were taken using the x63 objective on the LSM880 Fast Airy Scan confocal microscope (Zeiss). Confocal images of cross sections of zebrafish muscles were taken using the *20 objective on the Operetta microscope (Perkin).

[0305] The area of fiber cross sections was automatically detected by Cellpose software and each area was measured by Fiji software.

[0306] List of primers

[0307] Immunostaining and sectioning

[0308] - for cutting fish muscles: Z-Fish aged 6 and 36 months and K-Fish aged 1 and 6 months were used. After euthanasia by ice bath, the caudal part of the fish was mounted in agarose and cut into slices using a vibroslice. Immunofluorescence was carried out according to the protocol described in Batut et al., 2011 with the A4.1025 antibody (all muscle fibers) and the F310 primary antibodies (slow muscle fibers).

[0309] -for mouse muscle cuts:

[0310] The muscles (here gastrocnemius) are removed to perform cryosections and measure the size of the fibers following Hemalun-Eosin marking. Cross sections (10 mm) were cut perpendicular to the muscles in a cryostat. Cross sections were incubated in hematoxylin solution (HHS16, Sigma-Aldrich, Munich, Germany) for 2 minutes, rinsed with distilled water, incubated in eosin solution (HT-110-1-16, Sigma- Aldrich, Munich, Germany) for 2 minutes and finally rinsed with 95% ethanol, and mounted on an inverted microscope (Nikon inverted Eclipse) connected to a CCD camera (VCC- 2972, SANYO Electric Co.). The cross-sectional area (CSA) of 70 to 100 muscle fibers per individual, regardless of the fiber type, was measured semi-automatically by a detection algorithm developed by M Vigneau.

[0311] Analysis of TEM images:

[0312] The analysis of the images acquired by transmission electron microscopy was carried out using the software Fiji (Schindelin, J.; Arganda-Carreras, I. & Frise, E. et al. (2012)) and Cellpose (Stringer, C. et al. (2021)).

[0313] 1 - Cellpose training

[0314] The training of the different Species-specific and Generalist models was carried out using the Cellpose 2.0 software (Pachitariu, M. & Stringer, C. (2022)). For each species, the images acquired in TEM were previously separated into 2 distinct groups, one - training dataset - used for training the 2 models, and the second - validation dataset - for validating the models. For each image of the training group, a first detection of mitochondria was carried out with the Cyto2 model implemented in Cellpose.

[0315] To re-train the Cyto2 model with the images, the Applicant used the Human-in-the- loop training function of Cellpose 2.0 and manually annotated, one by one, the mitochondria not detected or partially detected by the Cyto2 model (False Negative). For each image, the Applicant also took care to remove all erroneous detections (False Positive). Thus correctly annotated, these images were used again in the Cellpose 2.0 software to re-train the initial model (Cyto2) and thus generate the 4 models.

[0316] For the Zebrafish-specific model, 31 images were trained and 2875 mitochondria annotated manually. For the Drosophila-Specific model, training was carried out on 11 images, for a total of 965 annotated mitochondria. For the Mouse-specific model, 25 images and 2001 mitochondria were manually annotated. Finally, for the Generalist model, the Applicant combined all the images and annotations from the 3 Species-specific models, for a total of 67 images and 4809 annotated mitochondria. 2 - Model validation

[0317] Each of the 4 trained models was then validated using the different images from the validation group (35, 10, 20 and 65 images, respectively for Zebrafish-specific, Drosophilaspecific, Mouse-Specific and Generalist). Using Cellpose 2.0 software, the Applicant submitted each image to its 2 corresponding models - Species-specific and Generalist - and obtained for each a mitochondria segmentation prediction map (Prediction map). In parallel, for each validation image, the Applicant manually annotated the mitochondria in order to have a “ground truth” map of the mitochondria (Ground Truth map).

[0318] In order to evaluate the performance of the 3 Species-specific models and the Generalist model, the Applicant compared the prediction and ground truth maps for each mitochondrion, and measured different performance indicators using the Fiji software:

[0319] - Intersection map (ROlp D ROIGT): corresponds to the colocalization surface between a mitochondrion predicted by the model and its closest ground truth (annotated mitochondrion)

[0320] - Union map (ROlp u ROIGT): corresponds to the combined surface of a predicted mitochondrion and its closest ground truth

[0321] - Intersection over Union or loU: corresponds to the ratio between the intersection (P D GT) and the union (P u GT). This score is calculated for each mitochondrion, and represents a measure of the accuracy of a model for an object detection task. Its value is between 0 and 1. Note that a score greater than 0.5 is consensually considered to be the threshold value for a good prediction.

[0322] For each image in the validation group, the Applicant then used these calculated loU values to measure different performance metrics:

[0323] - Number of True Positives (True Positive TP): number of objects in the image having an loU value greater than 0.5. This metric corresponds to the number of mitochondria correctly predicted by the model

[0324] - Number of False Positives (False Positive FP): number of objects in the image having an loU value less than or equal to 0.5. This metric corresponds to the number of incorrect predictions.

[0325] - Number of False Negatives (False Negative FN): this is the number of mitochondria not predicted by the model. It corresponds to the difference between the number of ground truth mitochondria and the number of True Positives (GroundTruth - TruePositive).

[0326] - Precision (TP I (TP+FP)). Indicates to what extent the objects predicted by the model are real (i.e. are mitochondria). The calculated score varies between 0 and 1.

[0327] - Sensitivity (TP I (TP+FN)). Indicates the model's ability to predict all or part of the real objects (the mitochondria). The calculated score varies between 0 and 1. - Average Precision AP (TP I (TP+FP+FN)). Performance indicator which takes into account the 2 criteria of specificity and precision. The calculated score varies between 0 and 1.

[0328] For each prediction model, the mean Average Precision (mAP) is calculated by averaging the Average Precision calculated for each image in the validation group.

[0329] 3 - Comparison with other prediction models

[0330] For each species, the Applicant compared the Species-specific and Generalist models with 2 other models made available to the scientific community:

[0331] - The Cyto2 model implemented in the Cellpose & Cellpose 2.0 software (https: / / cellpose.readthedocs.io / en / latest / models.html)

[0332] - The Neuron segmentation in 2D EM model (Falk et al. Nature Methods (2019); Ronneberger et al. arXiv (2015); Lucas von Chamier et al. biorXiv (2020)) available via the bioImage Zoo (https: / / bioimage.io / )

[0333] For each of these 2 models, the Applicant generated the prediction maps (Prediction map) and ground truth maps (GroundTruth map) for each of the images in the validation group. In order to compare the performances of these 2 models with the Species-specific and Generalist models, the Applicant measured the different indicators described above (see part 2), namely the mean Average Precision (mAP), Precision and Sensitivity.

[0334] For each species, the Applicant tested the different calculated mAP values 2 by 2 using a paired Student t test.

[0335] 4 - Batch segmentation & Morphological analysis

[0336] The different stages of the analysis pipeline - image processing, segmentation, morphological measurements - were carried out using the Fiji software (Schindelin, J.; Arganda-Carreras, I. & Frise, E. et al. (2012 )). The raw images obtained in TEM were first normalized in amplitude (normalization on the minimum and maximum gray level value) in order to attenuate the variations in gray level observed between the images, and linked to the acquisition. Automatic segmentation of mitochondria was then carried out using the Cellpose wrapper implemented in the BIOP (BioImaging And Optics Platform) plugin. The parameters taken as input to the plugin and set by the experimenter are the type of prediction model (Species-specific or Generalist) and the diameter (in pixels) of the objects to segment. The mitochondrial segmentation maps (or labeled maps) generated as output from Cellpose were then used in Fiji to calculate the different individual (per mitochondrion) or average (per image) metrics. Acquisition and analysis of CSA images

[0337] After immunofluorescence, images were acquired using a Perkin Elmer Operetta microscope with x20 objectives and Harmony High-Content imaging and analysis software. Image processing was automated with the help of Mathieu VIGNEAU from Restore. The Cellpose algorithm was used on the images to detect individual muscle fibers. The cross- sectional area and fluorescence intensity of GFP were measured with FIJI software.

[0338] Study of locomotion

[0339] Fish:

[0340] Z-Fish aged 6 and 36 months and K-Fish aged 1 and 6 months were used. To assess spontaneous locomotion, individuals were placed in separate tanks and monitored for 1 hour using EthoVision XT software. Using this same software, the average distance and immobility time of each fish were calculated.

[0341] Mouse :

[0342] Mice aged 3 months and 21 months were habituated for 2 weeks, one mouse per cage and the analysis was carried out over 1 month in DVC cages marketed by Techniplast.

[0343] FACS analysis of mitochondria in whole blood

[0344] Blood samples were taken from the Z-Fish and K-Fish after sacrifice, from the cardinal vein with previously heparinized syringes. The blood collection site is located in the axis of the body and behind the anus, in the region of the dorsal aorta in the posterior cardinal vein. Mouse blood samples were collected after the mice were sacrificed at the jugular vein. Human blood samples were provided by EFS. All blood samples were labeled with 250 nM MitotrackerGreen then the samples were filtered and analyzed with FACS according to the FACS protocol used for the analysis of mitochondria described above. Half of the samples were labeled with the APC anti-human CD9 Antibody (V P018) purchased from BioLegend.

[0345] Statistical analysis

[0346] Data are expressed as means ± standard error of the means (SEM) of at least 4 independent experiments. Statistical analyzes were performed using GraphPad Prism 6 for Windows (GraphPad Software, San Diego, CA, USA). To compare means between groups, the Applicant used one-way analysis of variance (ANOVA) followed by Dunnett's multiple comparisons or Student's t test. P values <0.05 were considered significant. Other examples of parameters which may be used

[0347] The following table II provides with another list of parameters corresponding to mitochondria morphometries, which may be used in the context of the present invention: Table II: Description of mitochondria morphometries

[0348] Among all ultrastructure parameters previously mentioned, the following combinations have been found of interest:

[0349] • Intensity SD and Skewness,

[0350] • Intensity SD and Meanjntensity,

[0351] • Totallnt, Medianlnt and Meanjntensity,

[0352] • Mito_Medianlnt_Norm and Intensity SD,

[0353] • CristaOrientation_Minor and Mito_Medianlnt_Norm, and

[0354] • Mito_Medianlnt and Intensity SD. When considering all these combination, it appears that the Intensity SD parameter is a relevant ultrastructure parameter. This is the preferred one among all the parameter relative to the crista’s density within the mitochondrion.

[0355] This also shows that a statistical value of the intensity is interesting, notably median and average, more preferably the median.

[0356] A parameter relative to the asymmetry of the mitochondrion’s gray values may be used in combination or for replacing one of the previous parameters.

[0357] Similarly, a measurement of the crista’s orientation may be used in combination or for replacing one of the previous parameters.

[0358] Therefore, advantageously, the ultrastructure parameters are a combination of the four kinds of parameters:

[0359] • a parameter relative to the crista’s density within the mitochondrion,

[0360] • a statistical value of the intensity of the mitochondrion,

[0361] • a parameter relative to the asymmetry of the mitochondrion’s gray values, and

[0362] • a parameter of the crista’s orientation.

[0363] More preferably, the ultrastructure parameters are a combination of two parameters: a parameter relative to the crista’s density within the mitochondrion with a statistical value of the intensity of the mitochondrion or with a parameter relative to the asymmetry of the mitochondrion’s gray values.

[0364] As a specific embodiment of this case, the ultrastructure parameters are a combination of Intensity SD and Skewness or Intensity SD and Meanjntensity.

[0365] List of figures

[0366] Figure 2: Overview of the entire workflow. Through the processing of EM pictures, mitochondria are segmented (step1-2), and mitochondria morphometries are extracted (step 3) and plotted using multiple graphs (step4).

[0367] Figure 3: Successful mitochondria segmentation using a pretrained CellPose 2.0 algorithm- specialist model. (A-C) Rounds of training on each specie (Fly, Z-Fish and Mouse) for the CellPose2.0 algorithm to generate a specie-specific model (SM). (D) Performance comparison (Mean average precision (blue) and mean precision (red)) for the SM on every specie depending number of ROI used for training. (E-F) Performance comparison for each model (SM vs GM) on every specie with APIoll>0,5 score plotted with the ROI used for the pretraining.

[0368] Figure 4: Performance comparison with other existing models. Comparison of mitochondria segmentation on (A) fly, (B) fish and (C) mouse between llastik, DeepImageJ, CellPose2.0 Cyto2, Trained CellPose 2.0 on Specie-Specific model, and Trained CellPose 2.0 GM model. (D) Performance evaluation for the mitochondria segmentation of every model using the Average Precision Score (AP). Red dots: Ground truth mitochondria, Blue dots: prediction of mitochondria and white dots: Overlap between ground truth and predicted mitochondria.

[0369] Figure 5: Mitochondria morpho-metrics are computed in multiple customizable graphs. After mitochondria segmentation on (A) fly (red) and fish (blue), mitochondria metrics such as Feret diameter and Intensity are layout using (B) density curves, (C) histograms, (D) violin plots and (E) star plot. (F) LIMAP projections of the dataset for fly (red) and fish (blue).

[0370] Figure 6: An example of EMitoMetrix application: Study of mitochondria morphology through a cross specie analysis. (A-B) LIMAP and PCA projections of mitochondria metrics for each specie (K-Fish-orange; Fly-red; Z-Fish-blue; Mouse-green). (C-D) paired LIMAP projections to compare (C) fly and mouse or (D) fish and mouse datasets. (E) Star plot displaying all mito metrics for each specie. (F) Confusion matrix generated by MLP algorithm depicting accuracy score on the test dataset of the specie prediction. (G-J) Global explainability of the specie prediction model for the 11 mitometrics for each specie (G) worm, (H) fly, (I) fish and (J) mouse (in order of importance based on the mean of absolute SHAP values). Each point color encodes the SHAP value of each variable for each individual; red and blue colors indicate high and low values of the variable, respectively.

[0371] Figure 7: Successful mitochondria segmentation using a pretrained CellPose 2.0 algorithm-generalist model. (A-C) Rounds of training on each specie (Fly, Z-Fish and Mouse) for the CellPose2.0 algorithm to generate a generalist model (SM). (D) Performance comparison (Mean average precision (blue) and mean precision (red)) for the GM on every specie depending number of ROI used for training. (E) Comparison of AP Score in Fly, Z- Fish and Mouse using the Generalist model.

[0372] Figure 8: All morphometries displayed as Density curves and Histograms. After mitochondria segmentation on fly (red) and fish (blue), mitochondria metrics such as (A&L) Area, (B&M) Perimeter, (C&N) Total Intensity, (D&O) Circularity, (E&P) Roundness, (F&Q) Solidity, (G&R) Intensity Standard Deviation, (H&S) Intensity Standard Deviation (%mean intensity), (l&T) Aspect Ratio are layout using (A-l) density curves, (L-T) histograms.

[0373] Figure 9: All morphometries displayed as violin plots, star plot and PCA. After mitochondria segmentation on fly (red) and fish (blue), mitochondria metrics such as (A) Area, (B) Perimeter, (C) Total Intensity, (D) Circularity, (E) Roundness, (F) Solidity, (G) Intensity Standard Deviation, (H) Intensity Standard Deviation (%mean intensity), (I) Aspect Ratio are layout using (A-l) violin plots. (J) Star plot and (K) Principal Component Analysis (PCA) projections on mitochondria morphometries. Figure 10: Mitochondria morphometries on the cross-species analysis. (A-C) LIMAP projections of (A) worm and fly, (B) worm and fish and (C) worm and mouse datasets. Mitochondria metrics such as Feret diameter is layout using (D) density curves, (E) histograms and (F) violin plots

[0374] Figure 11 : Performance comparison of all algorithm on the train and test datasets. Confusion matrix generated by (A) MLP, (B) LR, (D) Random Forest, (F) XGBoost algorithm depicting accuracy score on the train dataset of the specie prediction. Confusion matrix generated by (C) LR, (E) Random Forest, (G) XGBoost algorithm depicting accuracy score on the test dataset of the specie prediction.

[0375] Figure 12: Aged Z-fish, K-fish and Mouse exhibit an altered locomotion associated with muscle atrophy. Swim track recording of (a) Z-fish, (b) K-fish and (c) locomotion track recording of mouse, young (upper panel) and old (lower panel), 10 minutes tracking for one fish per conditions and 5-8 mice per conditions. Quantification of distance moved (cm) (upper panel) and immobility time in seconds (lower panel) in (d) Z-Fish and (e)K-Fish. mean ± s.e.m, Student’s t-test n = 4-8 adult for Z-fish and 10-15 adult for K-fish. Quantification of distance moved (Km) (f) in mouse, and distance moved in the wheel (Km) for young (blue) and old (grey) animals. mean ± s.e.m. Student’s t-test n = 4-8 adult mice. Immunofluorescence on cross section area of skeletal muscle and cell-pose detection in (g) Z-fish, (h) K-fish. H&E staining of cross section area of gastrocnemius from mouse (i) with young animals (upper panel) and old (lower panel). Scale bar 200um. Quantification of muscle fiber area in Z-fish (j), K-fish(k) and mouse (I) mean ± s.e.m. Student’s t-test n = 4- 8 adult animals per conditions.

[0376] Figure 13: The diameter and ultrastructure of mitochondrial cristae is a highly discriminating factor in muscle ageing. High-magnification electron micrograph of mitochondria on muscle biopsies from (a) Z-Fish, K-fish and mouse in young and old animals, (b) Cell-pose mask detection of mitochondria on TEM biopsies form young and old animals. (c)Star plot displaying all mitometrics for each species, histogram representing mitochondria density according to mitometrics in young versus old animal such as Feret diameter(d), Mean intensity(e) and Intensity SD(f).

[0377] Figure 14: Mitochondrial morphometries predict chronological age in Z-Fish, K-Fish and mouse. Paired UMAP projections of mitochondria metrics of (a) young and old Z-fish, (b) young and old K-fish (c) young and mouse. UMAP projections of (d) old Z-fish and old K-fish (e) old mouse and old Z-fish (f) old mouse and old K-fish. Confusion matrix generated by MLP algorithm depicting accuracy score on the test dataset of the age prediction for Z- fish (g), K-Fish (h) and mouse (i) Global explainability of the age prediction model for the 11 mitometrics for each specie (j) Z-Fish, (k) K-fish, (I) mouse (in order of importance based on the mean of absolute SHAP values). Each point color encodes the SHAP value of each variable for each individual; red and blue colors indicate high and low values of the variable, respectively.

[0378] Figure 15: Development of a flow cytometry approach to sort mitochondria on the basis of their morphological parameters, (a) Experimental setup: isolation of mitochondria from muscle biopsies, labeling of mitochondria with specific fluorescent probes, MitotrackerGreen (MTG) and analysis of samples by flow cytometry, (b-e) Cytometer calibration with fluorescent (FITC positives) beads of known sizes and segmentation of mitochondrial samples by size range, (f) Analysis of double positives events for MTG and TMRM as percentages in young versus old Z-fish. Size-range distribution of mitochondria isolated form young Z-fish(g), and old Z-fish (h). Analysis of double positives events, MTG and TMRM as percentages in young versus old K-fish (i). Size-range distribution of mitochondria isolated form young K-fish (j), and old K-fish (h). Mean ± s.e.m. Student’s t- test for bulk analysis and two-way ANNOVA for sub populations comparison. n = 5 adult animals per conditions.

[0379] Figure 16: Coexistence of mitochondria with heterogeneous metabolic capacities during aging in Z-Fish. Flow cytometry analysis of percentage of double positive events for MitotrackerGreen and TMRM on (a) bulk mitochondria in young versus old conditions and with 750um FCCP. (b) MFI of MitotrackerGreen (MTG) and TMRM positives events by mitochondria sub-populations for young and old conditions followed by FCCP treatment (c). Representation of Delta MFI TMRM representing the part of uncoupled mitochondria (d). Flow cytometry analysis of ROS production using MitoSox MFI, gated on MTG+ events (e) on bulk mitochondria and sub-populations. Flow cytometry analysis of mtDNA content using PicoGreen MFI gated on MitoDeepRed (MitoDR) events on bulk mitochondria (left) and sub-populations (right) (f). Flow cytometry analysis of percentage of all mitochondria (red histogram) superimposed with percentage of mitochondria positives for mtDNA staining (PicoGreen), for young (left panel) and young conditions (g). Representation of the percentage mitochondria mtDNA-null in young (blue) and old (grey) conditions (h). Flow cytometry analysis of percentage of events MFN2 and OPA1 positives events gated on MTG+ events on bulk mitochondria (i) and on mitochondria subpopulations for OPA1 (j) and MFN2 (k) in young versus old conditions. Mean ± s.e.m. Student’s t-test for bulk analysis and two-way ANNOVA for subpopulations comparison, n = 4 adult animals per conditions.

[0380] Figure 17: Coexistence of mitochondria with heterogeneous metabolic capacities during aging in K-Fish. Flow cytometry analysis of percentage of double positive events for MitotrackerGreen and TMRM on (a) bulk mitochondria in young versus old conditions and with 1 uM FCCP. (b) MFI of MitotrackerGreen (MTG) and TMRM positives events by mitochondria sub-populations for young and old conditions followed by FCCP treatment (c). Representation of Delta MFI TMRM representing the part of uncoupled mitochondria (d). Flow cytometry analysis of lipid peroxidation MFI, gated on MTG+ events (e) on bulk mitochondria and sub-populations. Flow cytometry analysis of mtDNA content using PicoGreen MFI gated on MitoDeepRed (MitoDR) events on bulk mitochondria and subpopulations (f) Flow cytometry analysis of percentage of all mitochondria (red histogram) superimposed with percentage of mitochondria positives for mtDNA staining (PicoGreen), for young (left panel) and young conditions (g). Representation of the percentage mitochondria mtDNA-null in young (blue) and old (grey) conditions (h). Flow cytometry analysis of percentage of events MFN2 and OPA1 positives events gated on MTG+ events on bulk mitochondria (i) and on mitochondria subpopulations for OPA1 (j) and MFN2 (k) in young versus old conditions. Mean ± s.e.m. Student’s t-test for bulk analysis and two-way ANNOVA for subpopulations comparison, n = 4 adult animals per conditions.

[0381] Figure 18: Mitochondrial heterogeneity is physiological and present from the first days of life in Z-Fish. Flow cytometry analysis of mitochondria size distribution in percentage of Mitotracker DeepRed (MitoDR) of three different group containing 10 larvae of 7days post fertilization (dpf) (a). Percentage of doubles positives events for MitoDR and TMRM by mitochondria sub-populations (b). TMRM MFI representation of bulk mitochondria (blue) after FCCP treatment (green) (c) and on mitochondria sub-populations (d)mtDNA content analysis using PicoGreen, percentage of double positives events for MitoDR and PicoGreen (e) or MFI (f,g). Mean ± s.e.m. Student’s t-test for bulk analysis and two-way ANNOVA for subpopulations comparison, n = 10 larvae per group.

[0382] Figure 19: Regulation of the proportion and function of mitochondrial subpopulations by intermittent fasting. Flow cytometry analysis of mitochondria size distribution in percentage of Mitotracker DeepRed (MitoDR) of old animals ad libitum (AL) fed (green) or in intermittent fasting (IF)(blue) feeding (a). Percentage of doubles positives events for MitoDR and TMRM of mitochondria sub-populations in Al versus IF group (b), and MFI (c). TMRM MFI representation of bulk mitochondria (dark color) and after FCCP treatment (light color) (d) and on all mitochondria sub-populations (e), for P1 and P2 and P3(f). mtDNA content analysis using PicoGreen, MFI of double positives events for MitoDR and PicoGreen in AL or IF group on bulk mitochondria (g) and percentage of double positives on sub-populations (h) and MFI (i). Analysis of lipid peroxidation using MitoPE, MFI of double positives events for MitoDR and PicoGreen in AL or IF group on bulk mitochondria (j) and percentage of double positives on sub-populations (k) and MFI (I). Mean ± s.e.m. Student’s t-test for bulk analysis and two-way ANNOVA for subpopulations comparison, n = 3 animals per group. Figure 20: Copy number and high mutation load of mtDNA are associated with mitochondrial morphology and age. (a) Graph depicting the copy number of mtDNA per mitochondria determine by dPCR from FACs sorted mitochondria (P1 or P4) of 2 and 24 months old zebrafish, (b, e, h) Comparison the number of mutation per gene biotype (rRNA, tRNA or coding genes) between 2M and 24M (b), P1 and P4 (e), and all conditions together (h). (c, f, i) Comparison the mutation number according to their type (insertion / deletion, transition or transvertion) between 2M and 24M (c), P1 and P4 (f), and all conditions together (j). (d, g, j) Comparison the number of mutation according to their consequences (frameshift, nonsense, non-synonymous, or synonymous) between 2M and 24M (d), P1 and P4 (g), and all conditions together (j).

[0383] Figure 21 : High penetrance of mutations in the nd1 , nd3 and nd5 loci is associated with large mitochondrial morphology in older individuals (a) Graph depicting the distribution along the mtDNA of the mean count of mutation from FACs sorted mitochondria (P1 or P4) of 2 and 24 frefremonths old zebrafish, (b) Manhattan plots for each condition representing significantly enriched hotspot for genomic slide window of 1 kb along the mtDNA. Allele frequency (AF) is represented on Y axis and p-value is correlated to circle size and color. Map of coding region along the mtDNA is represented under plots, (c) Heatmap for significantly enriched hotspot for genomic slide window of 0,1 kb. For significantly enriched hotspot for the P4-24M condition, genes present on genomic slide window of interest are depicted on the right side of the heatmap.

[0384] Figure 22: Development of Zebrafish embryos mtDNA-null. a. Experimental design: Either GFP mRNA or mtNcol mRNA is injected into one cell stage zebrafish embryo. mtNcol is a restriction enzyme targeted to the mitochondria to induce mtDNA clivage at 3 different sites (cyb, tRNAI O and 18). For the rescue, isolated mitochondria from young and old killifish are subsequently injected into pre-injected embryo, b- c. Transmitted light picture from GFP mRNA (b) or mtNcol mRNA (c) injected embryos at 1 dpf. d-e. Transmitted light picture at higher magnification of mtNcol mRNA injected embryos regroup according to the presence or not of morphological defects. ND- No defect. DD- Developmental defect, f. Graph showing proportion of dead embryos, embryos presenting developmental defect or no defect in the different context of injections, g-i. Ncol digestion of PGR amplification of mtNcol putative target site in cyb (g), tRNAI O (h) and tRNA18 (i). The different PGR products are majoritairly digested by Ncol in control injected embryos (CTL) and in mtNcol mRNA injected embryos presenting no defect (ND). In mtNcol mRNA injected embryos with developmental defect (DD), the Applicant can detect some nondigested PGR products as indicated by red arrows, j-m. Fast Airy scan confocal section of 3hpf embryos, j. All mitochondria appears in red after labelling with mitotracker deepred for 1 hour. k. Nuclear DNA (asteriks) and mtDNA are labelled by DAPI and appear in blue. I. Isolated mitochondria from Killifish are labelled by Pkmito and appear in green, m. Merge. Cells presenting mitochondria and mt DNA are indicated by white arrow head, while the cell with mitochondria lacking mtDNA is pinpointed by the yellow arrow.

Claims

CLAIMS1.- Method for determining at least one biological parameter of a subject, the method being computer-implemented and comprising:- a step of characterizing mitochondria in an area of the subject, to obtain characterizing parameters, the characterizing parameters of a mitochondrion comprising a morphology parameter of the mitochondrion and an ultrastructure parameter of the mitochondrion, and- a step of deducing the at least one biological parameter based on the characterizing parameters.2.- Method for determining according to claim 1 , wherein one biological parameter is a parameter linked with the biological age of the subject, notably the chronological age of the subject.3.- Method for determining according to claim 1 or 2, wherein a Feret diameter is defined for each mitochondrion, a morphology parameter being the Feret diameter.4.- Method for determining according to any one of the claims 1 to 3, wherein a mean intensity and a normalized intensity variation are defined for each mitochondrion, an ultrastructure parameter being the mean intensity or the normalized intensity variation.5.- Method for determining according to any one of the claims 1 to 4, wherein the step of characterizing comprises:- an operation of receiving electronic microscopy images of the area of the subject, and- an operation of applying a numerical model on the electronic microscopy images, to obtain the characterizing parameters.6.- Method for determining according to claim 5, wherein the numerical model is a neural network trained with a database comprising images from only three species.7.- Method for determining according to any one of the claims 1 to 6, wherein the step of characterizing comprises obtaining mutations of loci nd1 , nd3 and nd5.8.- A method for predicting that a subject is at risk of suffering from a disorder, the disorder being an aging related disorder or a mitochondrial disorder, the method for predicting at least comprising the step of:- carrying out the steps of a method for determining at least one biological parameter of a subject, to obtain at least one determined parameter, the method for determining being according to any one of claims 1 to 7, and- predicting that the subject is at risk of suffering from a disorder based on the at least one determined parameter.9.- A method for diagnosing a disorder, the disorder being an aging related disorder or a mitochondrial disorder, the method for diagnosing at least comprising the step of:- carrying out the steps of a method for determining at least one biological parameter of a subject, to obtain at least one determined parameter, the method for determining being according to any one of claims 1 to 7, and- diagnosing the disorder based on the at least one determined parameter.10.- A method for identifying a therapeutic target for preventing and / or treating a disorder, the disorder being an aging related disorder or a mitochondrial disorder, the method for identifying comprising the steps of:- carrying out the steps of a method for determining at least one biological parameter a first subject, to obtain at least one first determined parameter, the method for determining being according to any one of claims 1 to 7 and the first subject being a subject suffering from the disorder,- carrying out the steps of a method for determining at least one biological parameter a second subject, to obtain at least one second determined parameter, the method for determining being according to any one of claims 1 to 7 and the second subject being a subject not suffering from the disorder, and- selecting a therapeutic target based on the comparison of the first and second determined parameters.11.- A method for identifying a biomarker, the biomarker being a diagnostic biomarker of a disorder, a prognostic biomarker of a disorder or a predictive biomarker in response to the treatment of a disorder, said disorder being an aging related disorder or a mitochondrial disorder, the method for identifying comprising the steps of:- carrying out the steps of a method for determining at least one biological parameter a first subject, to obtain at least one first determined parameter, the method fordetermining being according to any one of claims 1 to 7 and the first subject being a subject suffering from the disorder,- carrying out the steps of a method for determining at least one biological parameter a second subject, to obtain at least one second determined parameter, the method for determining being according to any one of claims 1 to 7 and the second subject being a subject not suffering from the disorder, and- selecting a biomarker based on the comparison of the first and second determined parameters.12.- A method for screening a compound useful as a probiotic, a prebiotic or a medicine, the compound having an effect on a known therapeutical target, for preventing and / or treating a disease, said disorder being an aging related disorder or a mitochondrial disorder, the method comprising the steps of :- carrying out the steps of a method for determining at least one biological parameter a first subject, to obtain at least one first determined parameter, the method for determining being according to any one of claims 1 to 7 and the first subject being a subject suffering from the disorder and having received the compound,- carrying out the steps of a method for determining at least one biological parameter a second subject, to obtain at least one second determined parameter, the method for determining being according to any one of claims 1 to 7 and the second subject being a subject suffering from the disorder and not having received the compound, and- selecting a compound based on the comparison of the first and second determined parameters.13.- A computer program comprising instructions for carrying out the steps of a method according to any one of claims 1 to 12 when said computer program is executed on a suitable computer device.14.- A computer readable medium having encoded thereon a computer program according to claim 13.15.- A device for determining (10) at least one biological parameter of a subject, the device for determining (10) comprising a calculator (14) adapted to carry out a method according to any one of claims 1 to 12.

Images