Analysis method for mouse retina ganglion cell morphological classification

Abstract

The invention relates to an analysis method for mouse retinal ganglion cell morphological classification, which realizes single-cell-level accurate analysis and high-throughput integration: carrying out whole-cell patch post-clamp dendritic structure reconstruction on retinal ganglion cells, objectively quantifying parameters such as dendritic area, layering depth, dendritic asymmetry and the like, and analyzing the morphological classification of the mouse retinal ganglion cells. The subjective experience dependence of traditional Golgi staining and the resolution limitation of immunohistochemistry are thoroughly eliminated; a hierarchical classification rule is established through an automatic algorithm, manual interpretation deviation is eliminated, and the classification standardization degree is improved; the method breaks through the low-flux bottleneck of a genetic marker technology, does not need to depend on time-consuming gene model construction, synchronously analyzes the morphological diversity of cells through single cell sequencing and morphological reconstruction, and reveals that cells with similar dendritic layering positions can present functional differentiation due to transcriptome differences. The morphological analysis system established by the invention has single cell precision, high-throughput coverage and multi-dimensional relevance, and provides technical support for systematically analyzing the morphological development mechanism and subtype specificity of RGCs.

Description

Technical Field

[0001] This invention relates to the field of retinal nerve research, and in particular to an analytical method for classifying the morphology of mouse retinal ganglion cells. Background Technology

[0002] Morphological analysis of retinal ganglion cells (RGCs) is crucial for understanding visual information processing mechanisms, neuronal developmental regulation, and related disease pathology. As the only output neurons transmitting visual signals from the retina to the brain, the morphological diversity of RGCs (such as dendritic layering, area, and branching patterns) directly determines their functional specificity and forms the structural basis for analyzing complex visual processing pathways, including key "ON" / "OFF" light responses and orientation selectivity. Existing morphological analysis techniques, based on morphological observation and combined with molecular markers and genetic manipulation, have constructed a technical system from qualitative description to quantitative analysis: early methods relied on Golgi staining and immunohistochemistry to achieve whole-cell morphological observation and molecular markers; later, genetically guided sparse labeling techniques were developed, using conditional reporter genes to achieve spatiotemporally specific labeling at the single-cell level. In terms of morphological parameter measurement, Z-axis layered images are acquired using optical microscopy, and software such as Neuromantic is used to quantify dendritic area, layering depth, and branching complexity. Cluster analysis is then used to distinguish between single-layered (e.g., "OFF") and double-layered (e.g., ON-OFF orientation selectivity) subtypes.

[0003] Traditional Golgi staining uses random silver staining to label a small number of RGCs. This labeling process relies on the randomness of tissue processing (labeling only 1-5% of cells), and the interpretation of results is highly dependent on researcher experience. The determination of dendritic complexity and stratification boundaries (such as distinguishing between single and double strata) lacks quantitative standards and is easily influenced by subjective judgment. While immunohistochemistry achieves subtype enrichment labeling using specific antibodies (such as Brn3b and Melanopsin), it is limited by antibody affinity, tissue penetration, and fluorescence imaging resolution (usually subcellular rather than single-cell level). This makes it difficult to clearly distinguish the complete dendritic morphology of individual cells in dense RGC layers, especially for morphologically similar subtypes: such as ON-type and OFF-type directional selective ganglion cells in humans and other primates. Classification errors often occur due to large overlap of dendritic fields or blurred signals between ganglion cells. Furthermore, it cannot dynamically track in vivo morphological development, and the labeling range is limited to known molecular targets. While genetic marker technologies (such as the Cre-loxP system) can achieve subtype-specific markers, gene model construction takes 1-2 years, requires multiple generations of hybridization, and suffers from unstable marker efficiency (5-30%), a tendency for non-specific recombination, and difficulty in analyzing population characteristics due to sparse markers. Furthermore, batch markers cannot resolve single-cell details due to overlapping dendrites, and each model targets only a single molecular marker, making repeated construction costly. These limitations have confined early research to a few known targets, hindering the systematic revelation of subtype diversity and molecular regulatory mechanisms. These technological limitations have severely restricted the development of RGC morphological analysis. Summary of the Invention

[0004] To address the problems existing in the prior art, the present invention aims to provide an analytical method for classifying the morphology of mouse retinal ganglion cells. This method involves recording whole-cell data and reconstructing the morphology, followed by Neurolucida software combined with confocal microscopy Z-stack scanning to achieve precise three-dimensional reconstruction of dendritic structures at the single-cell level. Anticholine acetyltransferase antibodies are used to label starburst-free cell layers (SACs) to locate the boundaries of the inner plexiform layer (IPL), quantifying 21 morphological parameters (such as dendritic area, normalized layer depth, and asymmetry index) to solve the problems of subjective interpretation and layer ambiguity in traditional methods. A customized algorithm is developed to automate the calculation of morphological parameters and subtype classification, achieving a consistency score of 0.93±0.16 and an efficiency >90%, significantly improving analytical efficiency and objectivity, and constructing a deeply coupled analytical system.

[0005] The purpose of this invention is to provide an analytical method for classifying the morphology of mouse retinal ganglion cells, comprising the following steps: S1. Sample preparation and fixation: After dark adaptation, mice were sacrificed, the retina was quickly separated and placed in Ames medium to maintain its activity, and full-thickness retinal slices were prepared and fixed. S2, Immunofluorescence staining; S3, Laser confocal microscopy: First, the retinal slide was placed under a 10x microscope to locate RGCs, the target cells were transferred to the center of the field of view, and then the microscope was switched to 20x for imaging. The parameters are set as follows: aperture size is set to 1 AU, high-resolution 3D images are acquired (1024×1024 pixels in both x and y directions), the Z-axis scanning depth is determined according to the thickness of the retina, and the scanning starts from the GCL layer and ends at the inner kernel layer (INL); the images are scanned in z-stack mode, with an interlayer spacing of 0.25 μm. If the dendrite area is large, the stitching function is applied to capture the entire dendritic region completely. S4 and RGCs morphological structure reconstruction: S4.1 Cell body reconstruction steps: Import the acquired neuronal confocal images into Neurolucida software, load the original images using the "Image Stack" mode, and perform precise calibration according to the actual scale of the confocal images. In the images, green fluorescence is generated by the binding of Streptavidin 488 conjugate antibody and intracellular Biocytin dye, marking the neuronal cell bodies, axons, and dendrites to be reconstructed; red fluorescence is displayed by the binding of Alexa Fluor 555 and anti-ChAT antibody, marking the GCL and INL layers where starburst acanthocytosis (SAC) cell bodies are located, as well as the ON starburst acanthocytosis (ON SAC) and OFF starburst acanthocytosis (OFF SAC) layers composed of their dendrites. Select the plane with the clearest and largest surface area of ​​the cell body in the Z-axis direction and depict it according to the actual cell body outline. S4.2 Axon and Dendrite Reconstruction: Following the natural extension order of axons and dendrites, open the image layer by layer and under the "Axon" and "Dendrites" functions, perform fine depiction following the actual axon and dendrite directions of RGCs, while marking the start point, secondary branch point and end point of each branch; S4.3, GCL / INL layer localization: Switch to different layers of the confocal image, find the clearest cross-section of the GCL layer and INL layer, and use the "Contour" function to identify them respectively, where Contour 1 is defined as the GCL layer and Contour 2 as the INL layer; S5, RGCs Morphological Feature Extraction and Morphological Classification: S5.1 Morphological feature extraction: MorphKit, a program designed for RGCs, is used to extract morphological features from the reconstructed image file. The extraction includes, but is not limited to, a series of feature parameters such as dendrite asymmetry, dendrite area, number of dendrite branch segments, and number of branch points. For detailed parameters and corresponding explanations, please refer to Table 1. S5.2 Morphological Classification Criteria: The reconstructed ganglion cells are classified using the latest international classification standards from the Eyewirers database. High-level clusters are further subdivided using a decision tree, including the following feature thresholds: 1) Layered contours: percentiles (e.g., 25th, 70th IPL), multimodal distribution; 2) Cell body area; 3) Dendritic morphology: density, complexity, asymmetry. Naming Rules: Using the SAC cell bodies in the GCL and INL layers as reference points, the INL layer position is set to 0, and the GCL layer position to 1, dividing the thickness between the two layers into 10 intervals. The first letter of the cell morphology name is determined based on the interval where the maximum dendritic density peak is located. If the peak is located in the first interval (0 - 0.1), the first letter of the name is 1; and so on. For bilayered dendritic cells, the second letter reflects the interval where the second highest dendritic density peak is located. "s" (short) and "t" (high) in the cell name represent dendritic thickness, "n" (narrow) and "w" (wide) represent the planar expansion width of the dendrites. Additionally, "a" (asymmetry) is used. This indicates the asymmetrical distribution of dendritic branches, with "o" (outer side) and "i" (inner side) referring to the relative positions of the areas with the highest dendritic density. S5.3 Cell classification method: First, extract the morphological characteristics of 381 neurons of 47 types from the Eyewirers database, and then calculate 21 important parameters based on the reconstructed RGCs (see Table 1). First, through dendritic stratification profile correlation analysis, the linear correlation coefficient and Euclidean distance between the cells to be classified and each representative cell type in the database are calculated: Among them, (1) linear correlation coefficient: following the standard formula of Pearson correlation coefficient, combined with the logic of the MATLAB function corrcoef, the specific explanation is as follows: for two cell morphology parameter variables (vectors) of length n, X(x1, x2,..., x n (i.e., dendritic layer profile data) and Y(y1, y2, ..., y n (i.e., the dendritic layer profile data of the i-th template), the formula for calculating its linear correlation coefficient r is: r = = (∑x i ) / n is the mean of variable X; = (∑y i ) / n is the mean of variable Y; The molecule is the "covariance" of X and Y, reflecting the direction and strength of their linear correlation; The denominator is the product of the standard deviations of X and Y, used for normalization to limit the range of the correlation coefficient to ([-1, 1]), where 1 represents perfect positive correlation, -1 represents perfect negative correlation, and 0 represents no linear correlation. (2) Euclidean distance: A measure of the straight-line distance between two cell types in n-dimensional space, used to calculate the similarity between dendritic profiles of retinal ganglion cells. The formula is as follows: d(x,y) = x = dendritic density profile vector of the cells to be classified; y = dendritic density profile vector of template cell; x i , y i = The dendrite density value at the i-th depth bin; n = the number of depth bins (usually 100 IPL depth levels). Next, thresholds are set based on data-driven methods to screen candidate templates. Then, a hierarchical decision tree is used for preliminary classification, and the final classification result is determined by multi-indicator weighted scoring. The threshold setting adopts a data-driven method. Cells with clear molecular markers are first screened out. The correlation coefficient threshold is the square of the lower quartile of the correlation of this type of cells, and the distance threshold is directly the lower quartile of the distance of this type of cells. When no specific threshold value is given in the experiment, its specific size needs to be determined by calculating the corresponding lower quartile and following the above rules based on the correlation distribution and distance distribution of cells with known molecular markers. Finally, cross-validation using templates of known cell types ensures the accuracy of the algorithm on the basic dataset. Next, confidence thresholds are set for cells with ambiguous morphology, and manual verification is combined to handle biological variations. Crucially, the consistency between morphological classification and functional characteristics (such as ON / OFF electrophysiological characteristics) is verified by combining neuronal electrophysiological recordings to strengthen biological relevance.

[0006] Further preferably, S1 further includes the following steps: after injecting biotin into retinal ganglion cells using whole-cell patch-clamp technology, immediately remove the retinal slice from the recording slot, place it into a well of the cell culture plate, remove excess Ames solution, add 1 ml of cold, fresh 4% PFA fixative, cover the bottom of the culture dish with a layer of plastic wrap to form an airtight seal, and fix it in a 4°C refrigerator for 12 hours.

[0007] More preferably, S2 includes the following steps: S2.1 Washing: Gently remove the 4% PFA fixative from the tissue and replace it with 0.01 M PBS buffer. Place the 24-well plate containing the tissue on a shaker and shake to wash three times, 10 minutes each time. S2.2 Blocking: Remove the PBS, mix 5% BSA and 0.3% Triton X-100 in 0.01 M PBS as the blocking solution, add about 400 μl to each well, and block the tissue at room temperature for 2 h; S2.3 Primary antibody incubation: After pouring out the blocking solution, add 400 μl of primary antibody dilution solution to each well. The entire process should be carried out in the dark. Wrap the 24-well plate with aluminum foil and incubate overnight in a shaker at 4°C. S2.4, Wash again: Remove the primary antibody and wash three times with 0.01 M PBS, shaking for 10 min each time; S2.5 Secondary antibody incubation: Remove PBS, add 400 μl of secondary antibody to each well: Alexa Fluor 555-labeled goat anti-Goat secondary antibody (working concentration 1:1000) and DAPI staining solution (working concentration 1:1000), and dilute with 0.01 M PBS solution; incubate at room temperature for 2 h; S2.6 Final washing: Rinse three times with 0.01 M PBS, shaking, for 10 min each time; S2.7. Mounting: Transfer the tissue onto a glass slide, ensuring the retinal ganglion cell layer (GCL) is facing upwards and flat. Add a fluorescent antiquenching agent, cover with a coverslip to seal, and then image under a confocal microscope. Be careful not to create air bubbles during mounting.

[0008] More preferably, the primary antibody diluent in step S2.3 is Streptavidin, Alexa Fluor™ 488 conjugate: working concentration 1:800; ChAT antibody: working concentration 1:200, both diluted with 3% BSA in 0.01 M PBS.

[0009] More preferably, the 21 parameters for reconstructing RGCs in step S5.3 are: asymmetry, size, num_path_segments, num_branchpoints, num_irreducible_nodes, max_branch_order, average_nodal_angle_deg, average_nodal_angle_rad, average_local_angle_deg, average_local_angle_rad, average_tortuosity, real_length_sum, real_length_mean, real_length_median, real_length_min, real_length_max, euclidean_length_sum, euclidean_length_mean, euclidean_length_median, euclidean_length_min, and euclidean_length_max.

[0010] Compared with the prior art, the beneficial effects of the present invention are as follows: Firstly, this invention digitizes the three-dimensional morphology of reconstructed neurons, extracts 21 key morphological parameters (such as total dendritic area, IPL layering depth, and dendritic asymmetry index), and combines Neurolucida software tracking with customized algorithm automatic calculation to achieve three-dimensional reconstruction of dendritic structures (error rate <10%). This is more objective than the subjective layering interpretation of traditional Golgi staining (which relies on researcher experience), revealing morphological abnormalities in pathological states, while traditional techniques struggle to capture such subtle differences.

[0011] Secondly, this invention references the hierarchical classification framework of the Eyewires database, classifying reconstructed cells into various morphological types (e.g., 1wt, 37*, 7i / 7o), each corresponding to a clearly defined IPL layer location, dendritic density, and branching pattern. For example, C42_AlphaOFFS type RGCs are all single-layered, with a dendritic area >50,000 μm², and are layered in the outer layer of the IPL (OFF sublayer). Precise positioning of the boundaries of each IPL sublayer and determination of the intervals containing the maximum dendritic density peak and the secondary dendritic density peak ensures accurate dendritic location measurement (interlayer spacing 0.25 μm), providing clearer results than traditional Golgi and DAB staining (low resolution, strong background interference).

[0012] Thirdly, this invention develops an automated algorithm to calculate morphological parameters, reducing the time spent on manual interpretation. Attached Figure Description

[0013] Figure 1 This is a technical roadmap of the present invention; Figure 2 Image of the confocal image of the cells; Figure 3 The image shows the cell morphology reconstructed from an example cell using Neurolucida software. Figure 4 This is a reconstructed cell morphology image after extracting cell morphology parameters using a script. Figure 5 The image shows some feature analysis results after morphological parameters were extracted from the example cells. Detailed Implementation

[0014] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments: like Figures 1-5 As shown, an analytical method for classifying the morphology of mouse retinal ganglion cells, such as... Figure 1 As shown, it consists of the following steps: 1. Sample preparation and fixation C57BL / 6J wild-type mice (purchased, with approved animal ethics documentation) were sacrificed after dark acclimatization. The retinas were rapidly isolated and placed in Ames medium to maintain viability, and full-thickness retinal slices were prepared. Using whole-cell patch-clamp technology (equipment: HEKA PatchMaster), biotin was injected into retinal ganglion cells. Immediately afterward, the retinal slices were removed from the recording chamber and placed in one well of a 24-well cell culture plate. Excess Ames solution was removed, and approximately 1 ml of cold, fresh 4% PFA fixative was added. A layer of plastic wrap was placed under the culture dish lid to create an airtight seal, and the plate was fixed at 4°C for 12 hours.

[0015] 2. Immunofluorescence staining Washing: Gently remove the 4% PFA fixative from the tissue and replace it with 0.01 M PBS buffer. Place the 24-well plate containing the tissue on a shaker and wash three times for 10 min each time.

[0016] Blocking: Remove the PBS, mix 5% BSA and 0.3% Triton X-100 in 0.01 M PBS as the blocking solution, add about 400 μl to each well, and block the tissue at room temperature for 2 h.

[0017] Primary antibody incubation: After discarding the blocking solution, add 400 μl of primary antibody dilution buffer (Streptavidin, Alexa Fluor™ 488 conjugate: working concentration 1:800; ChAT antibody: working concentration 1:200, both diluted with 3% BSA in 0.01 M PBS) to each well. Protect the plate from light throughout the process, wrap the 24-well plate with aluminum foil and incubate overnight in a shaker at 4°C.

[0018] Wash again: Remove the primary antibody and wash three times with 0.01 M PBS, shaking for 10 min each time.

[0019] Secondary antibody incubation: Remove PBS and add 400 μl of secondary antibody to each well: Alexa Fluor 555-labeled goat anti-Goat secondary antibody (working concentration 1:1000) and DAPI staining solution (working concentration 1:1000), diluted with 0.01 M PBS solution. Incubate at room temperature for 2 h.

[0020] Final washing: Rinse three times with 0.01 M PBS, shaking for 10 min each time.

[0021] Mounting: Transfer the tissue onto a glass slide, ensuring the GCL layer is flat and facing upwards. Add a drop of fluorescence quencher, cover with a coverslip for sealing, and then image under a confocal microscope. Take care to avoid creating air bubbles during mounting.

[0022] Confocal shooting results as follows Figure 2 As shown, green fluorescence indicates the complete morphology of the cells to be reconstructed; red fluorescence marks SAC cells, including the GCL and INL layers where the cell bodies are located, and the ON and OFF SAC layers formed by their dendrites. The left image is an XY view image after tomographic scanning, which clearly shows the cell bodies, axons, and dendrites; the right image is a YZ view image after tomographic scanning, which shows that the cell bodies are located in the GCL layer; the axons are located above the GCL layer; and the dendrites are divided into two layers between the GCL and INL, one layer concentrated in the ON SAC layer near the GCL side, and the other layer concentrated in the OFF SAC layer near the INL side.

[0023] 3. Laser confocal microscopy Imaging was performed using a Zeiss LSM980 laser confocal microscope.

[0024] First, the retinal slide was localized using a 10x scope (Plan-Apochromat, 10×, 0.8 NA). The target cells were then moved to the center of the field of view, and the image was taken using a 20x scope (Plan-Apochromat, 20×, 0.8 NA).

[0025] The settings are as follows: aperture size is set to 1 AU, and high-resolution 3D images are acquired (1024×1024 pixels in both the x and y directions). The Z-axis scanning depth is determined according to the retinal thickness, starting from the GCL layer and ending at the INL layer. Images are scanned in z-stack mode with a spacing of 0.25 μm between adjacent layers. If the dendrite area is extensive, the stitching function is used to capture the entire dendritic region completely.

[0026] 4. Morphological and structural reconstruction of RGCs Cell body reconstruction steps: Import the acquired neuronal confocal images into Neurolucida software. Load the original images using "Image Stack" mode and perform precise calibration according to the actual scale of the confocal images. Figure 3 In the image, the left image shows a planar view of the cell (showing the X and Y axes), and the right image shows a lateral view of the cell (showing the Z axis). White represents the cell body, red represents the axon, and green represents the dendrites, clearly showing the IPL layering of the dendrites. Green fluorescence, generated by the binding of Streptavidin 488 conjugate antibody to intracellular Biocytin dye, identifies the neuronal cell body, axon, and dendrites to be reconstructed. Red fluorescence, generated by the binding of Alexa Fluor 555 to anti-ChAT antibody, marks the GCL and INL layers where the SAC cell body is located, as well as the ON and OFF SAC layers formed by its dendrites. The plane with the clearest and largest surface area along the Z-axis of the cell body was selected and depicted according to the actual cell body outline. The reconstructed morphology of representative cells is shown below. Figure 4 As shown.

[0027] Axon and dendrite reconstruction: Following the natural extension order of axons and dendrites, the image is opened layer by layer and finely depicted under the "Axon" and "Dendrites" functions, following the actual axon and dendrite directions of RGCs, while marking the start point, secondary branch point and end point of each branch.

[0028] GCL / INL layer localization: Switch to different layers of the confocal image, find the clearest cross-section of the GCL layer and INL layer, and use the "Contour" function to identify them respectively, where Contour 1 is defined as the GCL layer and Contour 2 as the INL layer.

[0029] 5. Morphological feature extraction and morphological classification of RGCs Morphological feature extraction: MorphKit, a program designed for RGCs, is used to extract morphological features from the reconstructed image files. The extraction includes, but is not limited to, a series of feature parameters such as dendrite asymmetry, dendrite area, number of dendrite branch segments, and number of branch points. Detailed parameters and corresponding explanations are shown in Table 1. Morphological classification criteria: The reconstructed ganglion cells were classified using the latest international classification standards from the Eyewirers database. High-level clusters were further subdivided using a decision tree, including the following feature thresholds: 1) Layer profile: percentiles (e.g., 25th, 70th IPL), multimodal distribution; 2) Cell body area; 3) Dendritic morphology: density, complexity, asymmetry. Naming rules: Using the SAC cell bodies in the GCL and INL layers as reference points, the INL layer position was set as 0, and the GCL layer position as 1, with the thickness between the two layers divided into 10 intervals. The first letter of the cell morphology name was determined based on the interval where the maximum dendritic density peak was located. If the peak was located in the first interval (0 - 0.1), the first letter of the name was 1; and so on. For bilayer dendritic cells, the second letter reflected the interval where the second highest dendritic density peak was located. In the cell name, "s" (short) and "t" (high) indicate the dendritic thickness, and "n" (narrow) and "w" (wide) indicate the dendritic extension width. In addition, "a" (asymmetric) indicates the asymmetric distribution of dendritic branches, and "o" (outer) and "i" (inner) refer to the relative positions of the parts with the highest dendritic density.

[0030] like Figure 5 As shown, the left image is a dendrite polarity map, with coordinates representing a set of concentric circles centered on the cell body. The polarity distribution of dendrites around the cell body is calculated, reflecting the cell's symmetry. The middle image is a dendrite density distribution map; the denser the dendrite distribution, the higher the score. The right image shows the dendrite IPL layering depth. With the INL layer defined as 0 and the GCL layer as 1, and the IPL divided into 10 layers, the OFFSAC layer is 0.28, and the ON SAC layer is 0.62. The red lines represent the dendrite density distribution curves in each IPL layer. The legend shows that the dendrites of the example cells are densely distributed near the ON SAC and OFF SAC layers, respectively. Combining the polarity and density maps, this indicates a bilayered asymmetric cell, consistent with the confocal results.

[0031] Cell classification method: First, morphological features of 381 neurons of 47 types were extracted from the Eyewirers database. Then, based on the 21 parameters of the reconstructed RGCs calculated beforehand, the parameters are as follows: asymmetry, size, num_path_segments, num_branchpoints, num_irreducible_nodes, max_branch_order, average_nodal_angle_deg, average_nodal_angle_rad, average_local_angle_deg, average_local_angle_rad, average_tortuosity, real_length_sum, real_length_mean, real_length_median, real_length_min, real_length_max, euclidean_length_sum, euclidean_length_mean, euclidean_length_median, euclidean_length_min, and euclidean_length_max. The definitions are shown in Table 1. Table 1 provides the definitions and explanations of each parameter. First, through dendritic layer profile correlation analysis, the linear correlation coefficient and Euclidean distance between the cells to be classified and each representative cell type in the database are calculated: (1) Linear correlation coefficient: following the standard formula of Pearson correlation coefficient, combined with the logic of the MATLAB function corrcoef, the specific explanation is as follows: for two cell morphology parameter variables (vectors) of length n X(x1, x2, ..., x n (i.e., dendritic layer profile data) and Y(y1, y2, ..., y n (i.e., the dendritic stratification profile data of the i-th template), the formula for calculating its linear correlation coefficient r is: r = = (∑x i ) / n is the mean of variable X; = (∑y i ) / n is the mean of variable Y; The molecule is the "covariance" of X and Y, reflecting the direction and strength of their linear correlation; The denominator is the product of the standard deviations of X and Y, used for normalization to limit the range of the correlation coefficient to ([-1, 1]), where 1 represents perfect positive correlation, -1 represents perfect negative correlation, and 0 represents no linear correlation. (2) Euclidean distance: A measure of the straight-line distance between two cell types in n-dimensional space, used to calculate the similarity between dendritic profiles of retinal ganglion cells. The formula is as follows: d(x,y) = x = dendritic density profile vector of the cells to be classified; y = dendritic density profile vector of template cell; x i , y i = The dendrite density value at the i-th depth bin; n = the number of depth bins (usually 100 IPL depth levels). Next, thresholds are set based on data-driven methods to screen candidate templates. Then, a hierarchical decision tree is used for preliminary classification, and the final classification result is determined by multi-indicator weighted scoring. The threshold setting adopts a data-driven method. Cells with clear molecular markers are first screened out. The correlation coefficient threshold is the square of the lower quartile of the correlation of this type of cells, and the distance threshold is directly the lower quartile of the distance of this type of cells. When no specific threshold value is given in the experiment, its specific size needs to be determined by calculating the corresponding lower quartile and following the above rules based on the correlation distribution and distance distribution of cells with known molecular markers. Finally, cross-validation using templates of known cell types ensures the accuracy of the algorithm on the basic dataset. Next, confidence thresholds are set for cells with ambiguous morphology, and manual verification is combined to handle biological variations. Crucially, the consistency between morphological classification and functional characteristics (such as ON / OFF electrophysiological characteristics) is verified by combining neuronal electrophysiological recordings to strengthen biological relevance.

[0032] Cell screening examples and specific parameter values ​​are shown in Table 2: Table 2 shows the results of various parameters for the example cells. Table 2 shows the results of various parameters for 5 example cells: the first row is the cell name; the second row is the actual classification of cell morphology; the third row is the predicted classification of cell morphology; and the fourth to twenty-fourth rows are the values ​​of 21 parameters of cell morphology.

[0033] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions conceived without inventive effort should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope defined in the claims.

Claims

1. A method for analyzing the morphological classification of mouse retinal ganglion cells, characterized in that, Includes the following steps: S1. Sample preparation and fixation: After dark adaptation, mice were sacrificed, the retina was quickly separated and placed in Ames medium to maintain its activity, and full-thickness retinal slices were prepared and fixed. S2, Immunofluorescence staining; S3, Laser confocal microscopy: First, the retinal slide was placed under a 10x microscope to locate RGCs, the target cells were transferred to the center of the field of view, and then the microscope was switched to 20x for imaging. The parameters are set as follows: the aperture size is set to 1 AU, high-resolution three-dimensional images are acquired (1024×1024 pixels in both x and y directions), the Z-axis scanning depth is determined according to the thickness of the retina, and the scanning starts from the GCL layer and ends at the kernel layer (INL); The images were captured using z-stack mode with a spacing of 0.25 μm between adjacent layers. If the dendrites of the cells were extensive, the stitching function was used to capture the entire dendritic region. S4 and RGCs morphological structure reconstruction: S4.1 Cell body reconstruction steps: Import the acquired neuronal confocal images into Neurolucida software, load the original images using the "Image Stack" mode, and perform precise calibration according to the actual scale of the confocal images. In the images, green fluorescence is generated by the binding of Streptavidin 488 conjugate antibody and intracellular Biocytin dye, marking the neuronal cell bodies, axons, and dendrites to be reconstructed; red fluorescence is displayed by the binding of Alexa Fluor 555 and anti-ChAT antibody, marking the GCL and INL layers where starburst acanthocytosis (SAC) cell bodies are located, as well as the ON starburst acanthocytosis (ON SAC) and OFF starburst acanthocytosis (OFF SAC) layers composed of their dendrites. Select the plane with the clearest and largest surface area of ​​the cell body in the Z-axis direction and depict it according to the actual cell body outline. S4.2 Axon and Dendrite Reconstruction: Following the natural extension order of axons and dendrites, open the image layer by layer and under the "Axon" and "Dendrites" functions, perform a detailed depiction following the actual axon and dendrite directions of RGCs, while marking the start point, secondary branch point and end point of each branch; S4.3, GCL / INL layer localization: Switch to different layers of the confocal image, find the clearest cross-section of the GCL layer and INL layer, and use the "Contour" function to identify them respectively, where Contour 1 is defined as the GCL layer and Contour 2 as the INL layer; S5, RGCs Morphological Feature Extraction and Morphological Classification: S5.1 Morphological feature extraction: MorphKit, a program designed for RGCs, is used to extract morphological features from the reconstructed image file. The extraction includes, but is not limited to, a series of feature parameters such as dendrite asymmetry, dendrite area, number of dendrite branch segments, and number of branch points. For detailed parameters and corresponding explanations, please refer to Table 1. S5.2 Morphological Classification Criteria: The reconstructed ganglion cells are classified using the latest international classification standards from the Eyewirers database. High-level clusters are further subdivided using decision trees, including the following feature thresholds: 1) Layered contours: percentiles (e.g., 25th, 70th IPL), multimodal distribution; 2) Cell body area; 3) Dendritic morphology: density, complexity, asymmetry. Naming Rules: Using the SAC cell bodies in the GCL and INL layers as reference points, the INL layer position is set to 0, and the GCL layer position to 1, dividing the thickness between the two layers into 10 intervals. The first letter of the cell morphology name is determined based on the interval where the maximum dendritic density peak is located. If the peak is located in the first interval (0 - 0.1), the first letter of the name is 1; and so on. For bilayered dendritic cells, the second letter reflects the interval where the second highest dendritic density peak is located. "s" (short) and "t" (high) in the cell name represent dendritic thickness, and "n" (narrow) and "w" (wide) represent the planar expansion width of the dendrites. Additionally, "a"... (Asymmetric) indicates the asymmetric distribution of dendritic branches, with "o" (outer side) and "i" (inner side) referring to the relative positions of the areas with the highest dendritic density; S5.3 Cell classification method: First, extract the morphological characteristics of 47 types and 381 neurons from the Eyewirers database, and then calculate 21 important parameters based on the reconstructed RGCs (see Table 1). First, through dendritic stratification profile correlation analysis, the linear correlation coefficient and Euclidean distance between the cells to be classified and each representative cell type in the database are calculated: Among them, (1) linear correlation coefficient: following the standard formula of Pearson correlation coefficient, combined with the logic of the MATLAB function corrcoef, the specific explanation is as follows: for two cell morphology parameter variables (vectors) of length n, X(x1, x2,..., x n (i.e., dendritic layer profile data) and Y(y1, y2, ..., y n (i.e., the dendritic stratification profile data of the i-th template), the formula for calculating its linear correlation coefficient r is: r = = (∑x i ) / n is the mean of variable X; = (∑y i ) / n is the mean of variable Y; The molecule is the "covariance" of X and Y, reflecting the direction and strength of their linear correlation; The denominator is the product of the standard deviations of X and Y, used for normalization to limit the range of the correlation coefficient to ([-1, 1]), where 1 represents perfect positive correlation, -1 represents perfect negative correlation, and 0 represents no linear correlation. (2) Euclidean distance: A measure of the straight-line distance between two cell types in n-dimensional space, used to calculate the similarity between dendritic profiles of retinal ganglion cells. The formula is as follows: d(x,y) = x = dendritic density profile vector of the cells to be classified; y = dendritic density profile vector of template cell; x i , y i = The dendrite density value at the i-th depth bin; n = the number of depth bins (usually 100 IPL depth levels). Next, thresholds are set based on data-driven methods to screen candidate templates. Then, a hierarchical decision tree is used for preliminary classification, and the final classification result is determined by multi-indicator weighted scoring. The threshold setting adopts a data-driven method. Cells with clear molecular markers are first screened out. The correlation coefficient threshold is the square of the lower quartile of the correlation of this type of cells, and the distance threshold is directly the lower quartile of the distance of this type of cells. When no specific threshold value is given in the experiment, its specific size needs to be determined by calculating the corresponding lower quartile and following the above rules based on the correlation distribution and distance distribution of cells with known molecular markers. Finally, cross-validation was performed using templates of known cell types to ensure the accuracy of the algorithm on the basic dataset. Next, confidence thresholds were set for cells with ambiguous morphology and combined with manual verification to handle biological variations. Crucially, by combining neuronal electrophysiological recordings to verify the consistency between morphological classification and functional characteristics (such as ON / OFF electrophysiological characteristics), the biological relevance is enhanced.

2. The method for analyzing the morphological classification of mouse retinal ganglion cells according to claim 1, characterized in that, The S1 further includes the following steps: using whole-cell patch-clamp technique, after injecting biotin into retinal ganglion cells, immediately remove the retinal slice from the recording slot, place it into one well of the cell culture plate, remove excess Ames solution, add 1 ml of cold, fresh 4% PFA fixative, cover the bottom of the culture dish with a layer of plastic wrap to form an airtight seal, and fix it in a 4°C refrigerator for 12 hours.

3. The method for analyzing the morphological classification of mouse retinal ganglion cells according to claim 2, characterized in that, S2 includes the following steps: S2.1 Washing: Gently remove the 4% PFA fixative from the tissue and replace it with 0.01 M PBS buffer. Place the 24-well plate containing the tissue on a shaker and shake to wash three times, 10 min each time. S2.2 Blocking: Remove the PBS, mix 5% BSA and 0.3% Triton X-100 in 0.01 M PBS as the blocking solution, add about 400 μl to each well, and block the tissue at room temperature for 2 h; S2.3 Primary antibody incubation: After pouring out the blocking solution, add 400 μl of primary antibody dilution solution to each well. The entire process should be carried out in the dark. Wrap the 24-well plate with aluminum foil and incubate overnight in a shaker at 4°C. S2.4, Wash again: Remove the primary antibody and wash three times with 0.01 M PBS, shaking for 10 min each time; S2.5 Secondary antibody incubation: Remove PBS, add 400 μl of secondary antibody to each well: Alexa Fluor 555-labeled goat anti-Goat secondary antibody (working concentration 1:1000) and DAPI staining solution (working concentration 1:1000), and dilute with 0.01 M PBS solution; incubate at room temperature for 2 h; S2.6 Final washing: Rinse three times with 0.01 M PBS, shaking, for 10 min each time; S2.

7. Mounting: Transfer the tissue onto a glass slide, ensuring the retinal ganglion cell layer (GCL) is facing upwards and flat. Add a fluorescent antiquenching agent, cover with a coverslip to seal, and then image under a confocal microscope. Be careful not to create air bubbles during mounting.

4. The method for analyzing the morphological classification of mouse retinal ganglion cells according to claim 3, characterized in that, The primary antibody diluent in step S2.3 is Streptavidin, Alexa Fluor™ 488 conjugate: working concentration 1:800; ChAT antibody: working concentration 1:200, both diluted with 3% BSA in 0.01 M PBS.

5. The method for analyzing the morphological classification of mouse retinal ganglion cells according to claim 4, characterized in that: The 21 parameters of the reconstructed RGCs in step S5.3 are: asymmetry, size, num_path_segments, num_branchpoints, num_irreducible_nodes, max_branch_order, average_nodal_angle_deg, average_nodal_angle_rad, average_local_angle_deg, average_local_angle_rad, average_tortuosity, real_length_sum, real_length_mean, real_length_median, real_length_min, real_length_max, euclidean_length_sum, euclidean_length_mean, euclidean_length_median, euclidean_length_min, euclidean_length_max.

Images