Method for classifying cells

Abstract

Herein is reported a method for classifying cells in a mixture of cells wherein the mixture comprises T cells and B cells, the method comprising the steps of: first, applying a labeled antibody binding to at least F-actin, MHCII and CD3 to the mixture of cells to obtain a labeled mixture of cells, wherein the antibodies are each labeled with a dye wherein the dyes have different (non-overlapping) emission wavelengths, second, acquiring at least one image of the mixture of cells, and third, classifying the cells in the mixture of cells as isolated cells, if the cells are single cells, are F-actin positive, if the cells are F-actin positive, and if the cells are not F-actin positive, the cells are not F-actin positive. And is MHCII positive and CD3 negative, or is MHCII negative and CD3 positive; or a diad or concatemer of cells, if the cell is an aggregate of two or three cells, the cell is F-actin positive, MHCII positive and CD3 positive, or a diad or concatemer of the cells, if the cell is an aggregate of two or three cells, is F-actin positive, MHCII positive and CD3 positive.

Description

[0001] The present invention belongs to the field of analytical technology. In more detail, this article reports a method for classifying cells in a mixture of B cells and T cells into isolated cells, cell concatemers without signaling, and cell concatemers with signaling based on differential markers. This classification allows the characterization of therapeutic methods that interfere with cell signaling formation. Background Art

[0002] Therapeutic antibodies are widely used to treat serious diseases. Most of them modify immune cells and act within the immunological synapse; an important cell-cell interaction that directs the humoral immune response. Although many antibody designs have been generated and evaluated, high-throughput tools for systematic antibody characterization and functional prediction are lacking.

[0003] The formation of the immune synapse is the first event of the adaptive immune response induced by the interaction of T cells with their corresponding antigen presenting cells (APCs). This rapidly formed cell-cell interface is initiated by the recognition of peptide-loaded MHC complexes by the T cell receptor (TCR). It involves the rearrangement of actin filaments of the cytoskeleton and the recruitment of signaling, co-stimulatory, co-inhibitory and adhesion molecules to the nascent synapse [1,2]. This process is crucial for triggering and fine-tuning T cell responses and ensuring a complete immune response. Dysfunctional immune synapse formation has been observed in several immune-related disorders [3-8] and has therefore been considered a potential target for stimulating or inhibiting immune responses by regulating their assembly or function [9-11]. For example, various therapeutic antibodies that alter immune synapse formation to treat cancer and autoimmune diseases have been developed [12-15]. Although significant progress has been made in the development of immune synapse targeting agents in recent years [9], there is still a need to further improve the compounds, especially to improve their efficacy. It has been identified that antibody size and format[16,17], dose, and target expression

[18] may be key parameters for immune synapse formation and its effects on T cell function.

[0004] However, no methods have been reported to systematically quantify and characterize immune synapse morphology, study its relevance to T cell responses, or identify properties that predict antibody efficacy in vitro.

[0005] A key technology for high-throughput data acquisition for this purpose is imaging flow cytometry (IFC), which combines the benefits of traditional flow cytometry with deep multichannel imaging at the single-cell level. IFC has recently been successfully applied to visualize and quantify immune synapses in primary human T:APC cell conjugates [19-21], however, none of these studies investigated immune synapse formation in the context of T cell function.

[0006] Recent studies have demonstrated the potential of machine learning algorithms for more robust and accurate analysis of high-throughput imaging data, and this approach has been shown to overcome the limitations of conventional gating strategies [22–24]. The use of machine learning for IFC data analysis has also enabled the identification of morphological patterns in cells, combined analysis of RNA and protein data, and the implementation of predictive models [22–26]. Although limited open source software implementations designed for IFC data analysis are available [26,27], they either rely on additional software, increasing the complexity of the analysis pipeline, or focus only on predictive performance and lack interpretability.

[0007] Immune synapses have previously been studied using high-content cellular imaging of human cell lines and primary cells using an artificial APC system that utilizes plate-bound ICAM-1 and stimulatory antibodies

[37] . Although German et al. convincingly demonstrated the power of their pipeline by profiling immune synapses, they did not investigate whether these profiles could be used to predict drug efficacy

[37] . In other studies, the potential of synapse formation for CAR T-cell therapy has also been investigated, where researchers used the average intensity of staining (such as F-actin and P-CD3ζ per cell), clustering of tumor antigens, and polarization of perforin-containing particles as measures of the quality of synapse formation. These features vary between different CAR T cells and correlate with their in vitro and in vivo efficacy and clinical outcomes [39,40].

[0008] M. Chen et al. disclosed that heparin-binding EGF-like growth factor regulates the bidirectional activation of CD4+ T cells and dendritic cells independently of the epidermal growth factor receptor (Am. J. Resp. Crit. Care, 2018, Meeting Abstracts. A5826).

[0009] BH Hosseini et al. disclosed that immune synapse formation determines the interaction force between T cells and antigen presenting cells measured by atomic force microscopy (Proc. Natl. Acad. Sci USA 106 (2009) 17852-17857).

[0010] F. Ahmed et al. disclosed that numbers are important in the quantitative and dynamic analysis of immune synapse formation using imaging flow cytometry (J. Immunol. Meth. 347 (2009) 79-86).

[0011] G. Wabnitz et al. disclosed that influx microscopy of human leukocytes is a tool for quantitative analysis of actin rearrangement in immune synapses (J. Immunol. Meth. 423 (2015) 29-39).

[0012] US2021 / 270812 discloses a method for analyzing immune cells. Summary of the invention

[0013] Here we report a method for sorting cells in a cell mixture using single-cell imaging flow cytometry.

[0014] We further report a method for classifying cells in cell mixtures using a combination of single-cell imaging flow cytometry and artificial intelligence (scifAI) for preprocessing, feature engineering, and interpretable predictive machine learning of imaging flow cytometry (IFC) data. Fig.40 The algorithm flow chart is presented in .

[0015] Utilizing the method according to the invention, the class frequencies and morphological changes under different immune stimulations can be analyzed. The applicability of the method according to the invention has been demonstrated by analyzing T cell factor production across multiple donors and therapeutic antibodies. These properties have been quantitatively predicted in vitro, thereby linking morphological features to function and showing the potential to significantly influence antibody design.

[0016] The method according to the invention is generally applicable to IFC data and, given its modular architecture, can be directly incorporated into existing workflows and analysis pipelines, e.g. for rapid antibody screening and functional characterization.

[0017] Therefore, the present invention encompasses the following embodiments:

[0018] 1. A method for classifying cells in a cell mixture, wherein the mixture comprises T cells and activated B cells or antigen presenting cells, the method comprising the following steps:

[0019] a) applying labeled antibodies that bind to at least F-actin, MHCII and CD3 to the cell mixture to obtain a labeled cell mixture, wherein the antibodies are each labeled with a dye, wherein the dyes have different (optionally non-overlapping) emission wavelengths,

[0020] b) acquiring at least one image of the cell mixture, and

[0021] c) classifying the cells in the cell mixture into

[0022] i) the isolated cells, if the cells are single cells, are positive for F-actin, and

[0023] - MHCII positive and CD3 negative, or

[0024] - MHCII negative and CD3 positive,

[0025] ii) Doublets or multiplets of cells, if the cells are aggregates of two or three or more cells, are F-actin positive, MHCII positive and CD3 positive.

[0026] 1a. A method for classifying cells in a cell mixture, wherein the mixture comprises T cells and activated B cells or antigen presenting cells, the method comprising the following steps:

[0027] a) applying labeled antibodies that bind to at least F-actin, MHCII and CD3 to the cell mixture to obtain a labeled cell mixture, wherein the antibodies are each labeled with a dye, wherein the dyes have different (optionally non-overlapping) emission wavelengths,

[0028] b) acquiring at least one image of the cell mixture, and

[0029] c) Classifying the images as follows:

[0030] The image contains

[0031] i) isolated cells, if said cells in said image are positive for F-actin, and

[0032] - MHCII positive and CD3 negative, or

[0033] - MHCII negative and CD3 positive,

[0034] ii) Doublets or multiplets of cells, if the cells in the image are aggregates of two or three or more cells, and the aggregates are F-actin positive, MHCII positive and CD3 positive.

[0035] 2. The method according to embodiment 1, wherein

[0036] Step a) is: applying labeled antibodies that bind to at least F-actin, MHCII, CD3 and P-CD3ζ to the cell mixture, wherein the antibodies are each labeled with a dye, wherein the dyes have different (non-overlapping) emission wavelengths,

[0037] And step c) is: classifying the cells in the cell mixture into

[0038] i) a single B cell or antigen presenting cell, if the cell is F-actin positive, MHCII positive, CD3 negative and P-CD3ζ negative,

[0039] ii) a single T cell without signaling, if the cell is F-actin positive, MHCII negative, CD3 positive and P-CD3ζ negative,

[0040] iii) a single T cell with signaling, if the cell is F-actin positive, MHCII negative, CD3 positive and P-CD3ζ positive,

[0041] iv) a doublet of a B cell or antigen presenting cell and a T cell that forms a synapse without signaling, if the doublet is positive for F-actin, positive for MHCII, positive for CD3 and negative for P-CD3ζ,

[0042] v) a doublet or multimer of one or two or more B cells or antigen presenting cells and one or two or more T cells that forms a signaling synapse, if the doublet or multimer is F-actin positive, MHCII positive, CD3 positive and P-CD3ζ positive.

[0043] 2a. The method according to embodiment 1a, wherein

[0044] Step a) is: applying labeled antibodies that bind to at least F-actin, MHCII, CD3 and P-CD3ζ to the cell mixture, wherein the antibodies are each labeled with a dye, wherein the dyes have different (non-overlapping) emission wavelengths,

[0045] And step c) is: classify the image as follows:

[0046] The image contains

[0047] i) a single B cell or antigen presenting cell, if said cell in said image is F-actin positive, MHCII positive, CD3 negative and P-CD3ζ negative,

[0048] ii) a single T cell without signaling, if the cell in the image is F-actin positive, MHCII negative, CD3 positive and P-CD3ζ negative,

[0049] iii) a single T cell with signaling, if the cell in the image is F-actin positive, MHCII negative, CD3 positive and P-CD3ζ positive,

[0050] iv) a doublet of a B cell or antigen presenting cell and a T cell that forms a synapse without signaling, if the cell doublet in the image is positive for F-actin, positive for MHCII, positive for CD3 and negative for P-CD3ζ,

[0051] v) a doublet or multimer of one or two or more B cells or antigen presenting cells and one or two or more T cells that form a signaling synapse, if the cell doublet or cell multimer in the image is F-actin positive, MHCII positive, CD3 positive and P-CD3ζ positive.

[0052] 3. The method according to any one of embodiments 1 to 2a, wherein

[0053] Step b) is: acquiring an image of the cell mixture using an imaging flow cytometer.

[0054] 4. The method according to any one of embodiments 1 to 3, wherein the acquired images are images each showing a single cell or isolated doublets or multiplets.

[0055] 5. A method for classifying cells in a cell mixture, wherein the mixture comprises T cells and activated B cells or antigen presenting cells, the method comprising the following steps:

[0056] a) acquiring at least one image of the cell mixture,

[0057] b) generating a feature extraction pipeline to derive biologically interpretable features from the at least one image,

[0058] c) predicting that the cell belongs to one of the following categories based on the biologically interpretable features obtained

[0059] i) Single B cells or antigen presenting cells,

[0060] ii) Single T cells without signaling,

[0061] iii) Single T cell with signaling,

[0062] iv) formation of a B cell or antigen presenting cell and T cell duplex without signaling synapse,

[0063] v) Doublets or multimers of one or two or more B cells or antigen presenting cells and one or two or more T cells forming a signaling synapse.

[0064] 5a. A method for classifying cells in a cell mixture, wherein the mixture comprises T cells and activated B cells or antigen presenting cells, the method comprising the following steps:

[0065] a) acquiring at least one image of the cell mixture,

[0066] b) generating a feature extraction pipeline to derive biologically interpretable features from the at least one image,

[0067] c) predicting the cell in the image as being based on the derived biologically interpretable features

[0068] i) Single B cells or antigen presenting cells,

[0069] ii) Single T cells without signaling,

[0070] iii) Single T cell with signaling,

[0071] iv) formation of a B cell or antigen presenting cell and T cell duplex without signaling synapse,

[0072] v) Doublets or multimers of one or two or more B cells or antigen presenting cells and one or two or more T cells forming a signaling synapse.

[0073] 6. A method according to any one of embodiments 1 to 5a, wherein the cell doublets and multimers (in the image) are classified as synapses based on morphology, labeling intensity, co-localization of labeling, texture and synaptic characteristics.

[0074] 7. The method of any one of embodiments 1 to 6, wherein the cell doublets and multimers (in the image) are classified as synapses based on one or more of the following additional features:

[0075] - Co-localization of CD3 and MHCII markers, and / or

[0076] - Colocalization of MHCII and P-CD3ζ markers, and / or

[0077] - Texture of MHCII, and / or

[0078] - CD3-marked textures, and / or

[0079] -P-CD3ζ labeling intensity.

[0080] 8. The method of any one of embodiments 1 to 7, wherein the characteristics of each cell or doublet or multiplet or image are determined based on the ratio of the marker signal intensity in the synaptic region to the entire cell.

[0081] 9. The method according to any one of embodiments 1 to 4 and 6 to 8, wherein the labeled cell mixture is an intracellular labeled cell mixture obtained by fixing the cells, permeabilizing the cells and applying the labeled antibody.

[0082] 9a. The method according to any one of embodiments 1 to 9, wherein the method further comprises the following step d):

[0083] d) counting the number of cells or cell images in each category and calculating the relative frequency of the cells in each category in the cell mixture.

[0084] 10. The method according to any one of embodiments 1 to 9a, wherein dead cells, deformed cells or cropped cells (aggregates of more than three cells) are removed before step b), or wherein images of dead cells, deformed cells or cropped cells (aggregates of more than three cells) are not recorded, or wherein images of dead cells, deformed cells or cropped cells and unfocused images are removed before step c), or wherein images of dead cells, deformed cells or cropped cells and unfocused images are not analyzed in step c), or wherein images of dead cells, deformed cells or cropped cells and unfocused images are not counted in step d).

[0085] 11. The method according to any one of embodiments 1 to 10, wherein step b) comprises the following additional sub-steps:

[0086] b-1) Gating on live +CD3+MHCII+ cells within the focus,

[0087] b-2) selecting images showing a single CD3+ T cell and a single MHCII+ B cell or antigen presenting cell from the population obtained in step b-1) using area and aspect ratio features,

[0088] b-3) determining the signal intensity of labeled CD3 within the synaptic mask (the synaptic mask is defined as the combination of the morphological CD3 and MHCII masks with a dilation rate of 3) and gating synapses showing CD3 signal in the mask, and

[0089] b-4) Exclusion of T cells and B cells or antigen presenting cells in a layer by using the height and area features of bright field (BF).

[0090] 12. The method according to any one of embodiments 5 to 11, wherein the method comprises the following step c):

[0091] c) Build a model based on the biologically interpretable features obtained using the XGBoost classifier, and predict that the cell or image belongs to one of the following categories based on the model

[0092] i) Single B cells or antigen presenting cells,

[0093] ii) Single T cells without signaling,

[0094] iii) Single T cell with signaling,

[0095] iv) formation of a B cell or antigen presenting cell and T cell duplex without signaling synapse,

[0096] v) Doublets or multimers of one or two or more B cells or antigen presenting cells and one or two or more T cells forming a signaling synapse.

[0097] 13. The method of any one of embodiments 1 to 12, wherein the mixture comprises B cells or antigen presenting cells to T cells at a cell ratio of about 4:3.

[0098] 14. The method according to any one of embodiments 1 to 13, wherein the T cells are CD4-positive memory T cells or CD8-positive T cells or a mixture thereof.

[0099] 15. The method of any one of embodiments 1 to 14, wherein the T cells are CD4-positive memory T cells.

[0100] 16. The method of any one of embodiments 1 to 15, wherein the cell mixture is centrifuged after mixing.

[0101] 17. The method of any one of embodiments 1 to 16, wherein step b) further comprises compensating the image using a compensation matrix derived from stained single cells.

[0102] 18. A method for classifying cells in a cell mixture, wherein the mixture comprises T cells and activated B cells or antigen presenting cells, the method comprising the following steps:

[0103] a) preparing a labeled cell mixture by:

[0104] a-1) dividing the cell mixture into at least two aliquots,

[0105] a-2) applying to a first aliquot of the mixture an antibody that binds to one or more cell surface targets present on one or two cells of the mixture, and applying to a second aliquot of the mixture an antibody that has the same structure as the antibody applied to the first aliquot but does not bind to one or more cell surface targets present on one or two cells of the mixture,

[0106] a-3) incubating the aliquot obtained in step a-2),

[0107] a-4) applying labeled antibodies that bind to at least F-actin, MHCII and CD3 to the incubated aliquot of the cell mixture obtained in step a-3) to obtain a labeled cell mixture, wherein the antibodies are each labeled with a dye, wherein the dyes have different (non-overlapping) emission wavelengths,

[0108] b) acquiring at least one image of each aliquot of the labeled cell mixture, and

[0109] c) classifying the cell images in each aliquot of the cell mixture to include

[0110] i) isolated cells, if the image contains cells that are single cells, positive for F-actin, and

[0111] - MHCII positive and CD3 negative, or

[0112] - MHCII negative and CD3 positive,

[0113] ii) Doublets or multiplets of cells, if the image contains cells that are aggregates of two or three cells that are F-actin positive, MHCII positive and CD3 positive.

[0114] d) for each aliquot, counting the number of images of cells in each category and calculating the relative frequency of the cells in each category in the cell mixture, and

[0115] e) determining the difference in class frequencies between the first aliquot and the second aliquot.

[0116] 19. The method of embodiment 18, wherein

[0117] Step a-4) is: applying labeled antibodies that bind to at least F-actin, MHCII, CD3 and P-CD3ζ to the incubated aliquot of the cell mixture, wherein the antibodies are each labeled with a dye, wherein the dyes have different (non-overlapping) emission wavelengths,

[0118] And step c) is: classifying the cells in the cell mixture into

[0119] i) a single B cell or antigen presenting cell, if the cell is F-actin positive, MHCII positive, CD3 negative and P-CD3ζ negative,

[0120] ii) a single T cell without signaling, if the cell is F-actin positive, MHCII negative, CD3 positive and P-CD3ζ negative,

[0121] iii) a single T cell with signaling, if the cell is F-actin positive, MHCII negative, CD3 positive and P-CD3ζ positive,

[0122] iv) a doublet of a B cell or antigen presenting cell and a T cell that forms a synapse without signaling, if the doublet is positive for F-actin, positive for MHCII, positive for CD3 and negative for P-CD3ζ,

[0123] v) a doublet or multimer of one or two or more B cells or antigen presenting cells and one or two or more T cells that forms a signaling synapse, if the doublet or multimer is F-actin positive, MHCII positive, CD3 positive and P-CD3ζ positive.

[0124] 19a. The method of embodiment 18, wherein

[0125] Step a-4) is: applying labeled antibodies that bind to at least F-actin, MHCII, CD3 and P-CD3ζ to the incubated aliquot of the cell mixture, wherein the antibodies are each labeled with a dye, wherein the dyes have different (non-overlapping) emission wavelengths,

[0126] And step c) is: classify the image as follows:

[0127] The image contains

[0128] i) a single B cell or antigen presenting cell, if said cell in said image is F-actin positive, MHCII positive, CD3 negative and P-CD3ζ negative,

[0129] ii) a single T cell without signaling, if the cell in the image is F-actin positive, MHCII negative, CD3 positive and P-CD3ζ negative,

[0130] iii) a single T cell with signaling, if the cell in the image is F-actin positive, MHCII negative, CD3 positive and P-CD3ζ positive,

[0131] iv) a doublet of a B cell or antigen presenting cell and a T cell that forms a synapse without signaling, if the doublet in the image is positive for F-actin, positive for MHCII, positive for CD3 and negative for P-CD3ζ,

[0132] v) a doublet or multiplet of one or two or more B cells or antigen presenting cells and one or two or more T cells that forms a signaling synapse, if the doublet or multiplet in the image is F-actin positive, MHCII positive, CD3 positive and P-CD3ζ positive.

[0133] 20. The method of any one of embodiments 18 to 19a, wherein

[0134] Step b) is: acquiring an image of the cell mixture using an imaging flow cytometer.

[0135] 21. The method according to any one of embodiments 18 to 20, wherein the acquired images are images each showing a single cell or isolated doublets or multiplets.

[0136] 22. A method according to any one of embodiments 18 to 21, wherein the cell doublets and multimers or images thereof are classified as synapses based on morphology, labeling intensity, co-localization of labeling, texture and synaptic features.

[0137] 23. The method of any one of embodiments 18 to 22, wherein the cell doublets and multimers in the image are classified as synapses based on one or more of the following additional features:

[0138] - Co-localization of CD3 and MHCII markers, and / or

[0139] - Colocalization of MHCII and P-CD3ζ markers, and / or

[0140] - Texture of MHCII, and / or

[0141] - CD3-marked textures, and / or

[0142] -P-CD3ζ labeling intensity.

[0143] 24. The method of any one of embodiments 18 to 23, wherein the characteristics of each cell or doublet or multiplet or image are determined based on the ratio of the marker signal intensity in the synaptic region to the entire cell.

[0144] 25. The method according to any one of embodiments 18 to 24, wherein the labeled cell mixture is an intracellular labeled cell mixture obtained by fixing the cells, permeabilizing the cells and applying the labeled antibody.

[0145] 26. The method of any one of embodiments 18 to 25, wherein dead cells, deformed cells, or cropped cells (aggregates of more than three cells) are removed prior to step b), or wherein images of dead cells, deformed cells, or cropped cells (aggregates of more than three cells) are not recorded, or wherein images of dead cells, deformed cells, or cropped cells and unfocused images are removed prior to step c), or wherein images of dead cells, deformed cells, or cropped cells and unfocused images are not analyzed in step c), or wherein images of dead cells, deformed cells, or cropped cells and unfocused images are not counted in step d).

[0146] 27. The method of any one of embodiments 18 to 26, wherein step b) comprises the following additional sub-steps:

[0147] b-1) Gating on live +CD3+MHCII+ cells within the focus,

[0148] b-2) selecting images showing a single CD3+ T cell and a single MHCII+ B cell or antigen presenting cell from the population obtained in step b-1) using area and aspect ratio features,

[0149] b-3) determining the signal intensity of labeled CD3 within the synaptic mask (the synaptic mask is defined as the combination of the morphological CD3 and MHCII masks with a dilation rate of 3) and gating synapses showing CD3 signal in the mask, and

[0150] b-4) Excluding T cells and B cells or antigen presenting cells in a layer by using the height and area features of bright field (BF).

[0151] 28. The method of any one of embodiments 18 to 27, wherein the mixture comprises B cells or antigen presenting cells to T cells at a cell ratio of about 4:3.

[0152] 29. The method according to any one of embodiments 18 to 28, wherein the T cells are CD4-positive memory T cells or CD8-positive T cells or a mixture thereof.

[0153] 30. The method of any one of embodiments 18 to 29, wherein the T cells are CD4-positive memory T cells.

[0154] 31. The method of any one of embodiments 18 to 30, wherein the cell mixture is centrifuged after mixing.

[0155] 32. The method of any one of embodiments 18 to 31, wherein step b) further comprises compensating the image using a compensation matrix derived from stained single cells.

[0156] 33. A method for ranking antibodies in a large number of antibodies, the method comprising the steps of:

[0157] 1) performing the method according to any one of embodiments 18 to 31 on each antibody of the plurality of antibodies individually or on all antibodies together, wherein each antibody is applied to a separate aliquot of the mixture,

[0158] 2) ranking the antibodies based on the frequency change of (the images of) one or more of the following categories:

[0159] i) Single B cells or antigen presenting cells,

[0160] ii) Single T cells without signaling,

[0161] iii) Single T cell with signaling,

[0162] iv) doublets of B cells or antigen presenting cells and T cells that form synapses without signaling, and

[0163] v) Doublets or multimers of one or two or more B cells or antigen presenting cells and one or two or more T cells forming a signaling synapse.

[0164] 34. The method of embodiment 33, wherein the antibodies in the plurality of antibodies are ranked by decreasing stimulation of an immune response.

[0165] 35. A method according to any one of embodiments 33 to 34, wherein the antibodies in the plurality of antibodies are sorted by reducing the frequency of cells or images that are classified as doublets or multimers of one or two or more B cells or antigen presenting cells and one or two or more T cells that form a signaling synapse.

[0166] 36. The method of any one of embodiments 33 to 35, wherein ranking the antibodies in the plurality of antibodies is performed by further reducing the signal intensity of labeled F-actin, labeled P-CD3ζ, and labeled MHCII in the synaptic region.

[0167] 37. The method of embodiment 33, wherein the antibodies in the plurality of antibodies are ranked by decreasing inhibition of an immune response.

[0168] 38. A method according to any one of embodiments 33 and 37, wherein the antibodies in the plurality of antibodies are sorted by reducing the frequency of cells or images that are classified as doublets or multimers of one or two or more B cells or antigen presenting cells and one or two or more T cells that form a signaling synapse.

[0169] 39. The method of any one of embodiments 33 and 37 to 38, wherein ranking the antibodies in the plurality of antibodies is performed by further increasing the frequency of cells or images classified as single T cells without signaling.

[0170] 40. The method of any one of embodiments 33 and 37 to 39, wherein ranking the antibodies in the plurality of antibodies is performed by further reducing the frequency of cells or images classified as single T cells with signaling.

[0171] 41. A method according to any one of embodiments 33 and 37 to 40, wherein the antibodies in the plurality of antibodies are sorted by further increasing the frequency of the average signal intensity of labeled F-actin, by increasing the signal of the intensity of labeled P-CD3ζ within the synaptic region, and by de-clustering the signal of the T cell receptor.

[0172] 42. Use of the method according to any one of embodiments 1 to 41 for characterizing the morphology of synapses formed between T cells and B cells or antigen presenting cells.

[0173] 43. Use of the method of any one of embodiments 33 to 41 for determining the correlation between antibody concentration and T cell response.

[0174] 44. Use of the method according to any one of embodiments 33 to 41 for predicting the therapeutic mode of an antibody.

[0175] 45. Use of the method according to any one of embodiments 33 to 41 for predicting the efficacy of an antibody.

[0176] 46. ​​Use of F-actin, MHCII and CD3 for sorting T cells in a mixture comprising T cells and B cells or activated B cells or antigen presenting cells.

[0177] 47. Use of F-actin, MHCII and CD3 for sorting B cells in a mixture comprising T cells and B cells or activated B cells or antigen presenting cells.

[0178] 48. The use according to any one of embodiments 46 to 47, wherein the image comprises or the cell is classified as an isolated cell, i.e., the classification of the cell is that the cell is an isolated cell if the cell is a single cell, is F-actin positive, and

[0179] - MHCII positive and CD3 negative, or

[0180] - MHCII negative and CD3 positive.

[0181] 49. The use according to any one of embodiments 46 to 47, wherein the image comprises or the cells are classified as doublets or multiplets of cells, i.e., the classification of the cells is that the cells are doublets or multiplets of cells if the cells are aggregates of two or three cells that are F-actin positive, MHCII positive and CD3 positive.

[0182] 50. The use according to any one of embodiments 46 to 49, further comprising β-CD3ζ.

[0183] 51. The use according to any one of embodiments 46 to 50, wherein the image comprises or the cell is classified as a single B cell or an antigen presenting cell, i.e., the classification of the cell is that the cell is a single B cell or an antigen presenting cell if the cell is F-actin positive, MHCII positive, CD3 negative and P-CD3ζ negative.

[0184] 52. The use according to any one of embodiments 46 to 51, wherein the image comprises or the cell is classified as a single T cell without signaling, i.e., the cell is classified as a single T cell without signaling if the cell is F-actin positive, MHCII negative, CD3 positive and P-CD3ζ negative.

[0185] 53. The use according to any one of embodiments 46 to 52, wherein the image comprises or the cell is classified as a single T cell with signaling, i.e., the classification of the cell is that the cell is a single T cell with signaling if the cell is F-actin positive, MHCII negative, CD3 positive and P-CD3ζ positive.

[0186] 54. The use according to any one of embodiments 46 to 53, wherein the image comprises or the cell is classified as a doublet of a B cell or an antigen presenting cell and a T cell that forms a synapse without signal transduction, i.e., the classification of the cell is that the cell is a doublet of a B cell or an antigen presenting cell and a T cell that forms a synapse without signal transduction if the doublet is F-actin positive, MHCII positive, CD3 positive and P-CD3ζ negative.

[0187] 55. The use according to any one of embodiments 46 to 54, wherein the image contains or the cells are classified as a doublet or multimer of one or two or more B cells or antigen presenting cells and one or two or more T cells that form a synapse with signal transduction, that is, the classification of the cells is that the cells are a doublet or multimer of one or two or more B cells or antigen presenting cells and one or two or more T cells that form a synapse with signal transduction if the doublet or multimer is F-actin positive, MHCII positive, CD3 positive and P-CD3ζ positive.

[0188] 56. The method or use of any one of embodiments 1 to 55, wherein the concatemer is a concatemer of two B cells or antigen presenting cells and one T cell.

[0189] 57. The method or use of any one of embodiments 1 to 55, wherein the concatemer is a concatemer of one B cell or antigen presenting cell and two T cells.

[0190] In addition to the various embodiments depicted and claimed, the disclosed subject matter also relates to other embodiments having other combinations of features disclosed and claimed herein. Thus, the specific features presented herein may be combined with each other in other ways within the scope of the disclosed subject matter, such that the disclosed subject matter includes any suitable combination of features disclosed herein. For purposes of illustration and description, the foregoing description of specific embodiments of the disclosed subject matter has been presented. It is not intended to be exhaustive or to limit the disclosed subject matter to those embodiments disclosed. DETAILED DESCRIPTION

[0191] General Definition

[0192] It must be noted that, as used herein and in the appended claims, the singular forms "a", "an", and "the" include plural references unless the context clearly dictates otherwise. Thus, for example, reference to "a cell" includes a plurality of such cells and equivalents thereof known to those skilled in the art, and so forth. Likewise, the terms "a", "one or more", and "at least one" may be used interchangeably herein. It should also be noted that the terms "comprising", "including", and "having" may be used interchangeably.

[0193] The term "about" means a range of + / -20% of the value that follows. In certain embodiments, the term "about" means a range of + / -10% of the value that follows. In certain embodiments, the term "about" means a range of + / -5% of the value that follows.

[0194] As used herein, the terms "comprises," "including," "having," "having," "may," "containing," and variations thereof are intended to be open-ended transitional phrases, terms, or words that do not exclude the possibility of additional actions or structures. The term "comprising" also encompasses the term "including..." The present disclosure also contemplates other embodiments "comprising," "consisting of," and "consisting essentially of" the embodiments or elements set forth herein, whether or not explicitly set forth.

[0195] Antibody

[0196] General information on the nucleotide sequences of human immunoglobulin light and heavy chains is given in: Kabat, EA et al., Sequences of Proteins of Immunological Interest, 5th Edition, Public Health Service, National Institutes of Health, Bethesda, MD (1991).

[0197] The term "antibody" herein is used in the broadest sense and encompasses various antibody structures, including but not limited to full length antibodies, monoclonal antibodies, multispecific antibodies (eg, bispecific antibodies), and antibody-antibody fragment fusions and combinations thereof.

[0198] The term "natural antibody" refers to naturally occurring immunoglobulin molecules with different structures. For example, a natural IgG antibody is a heterotetrameric glycoprotein of about 150,000 daltons, consisting of two identical light chains and two identical heavy chains bonded by disulfide bonds. From the N-terminus to the C-terminus, each heavy chain has a heavy chain variable region (VH), followed by three heavy chain constant domains (CH1, CH2 and CH3), whereby the hinge region is positioned between the first heavy chain constant domain and the second heavy chain constant domain. Similarly, from the N-terminus to the C-terminus, each light chain has a light chain variable region (VL), followed by a light chain constant domain (CL). The light chain of an antibody can be classified into one of two types based on the amino acid sequence of its constant domain, which are called kappa (κ) and lambda (λ).

[0199] The term "full-length antibody" refers to an antibody having a structure substantially similar to that of a natural antibody. A full-length antibody comprises two full-length antibody light chains and two full-length antibody heavy chains, each full-length antibody light chain comprising a light chain variable region and a light chain constant domain in the N-terminal to C-terminal direction, and each full-length antibody heavy chain comprising a heavy chain variable region, a first heavy chain constant domain, a hinge region, a second heavy chain constant domain, and a third heavy chain constant domain in the N-terminal to C-terminal direction. In contrast to natural antibodies, a full-length antibody may comprise additional immunoglobulin domains, such as one or more additional scFvs or heavy or light chain Fab fragments or scFabs conjugated to one or more ends of different chains of a full-length antibody, but each end is only conjugated to a single fragment. These conjugates are also covered by the term full-length antibody.

[0200] The "class" of an antibody refers to the type of constant domain or constant region (preferably Fc region) possessed by the heavy chain of the antibody. There are five major classes of antibodies: IgA, IgD, IgE, IgG, and IgM, and some of them can be further divided into subclasses (isotypes), e.g., IgG1, IgG2, IgG3, IgG4, IgA1, and IgA2. The heavy chain constant domains corresponding to the different classes of immunoglobulins are called α, δ, ε, γ, and μ, respectively.

[0201] The term "heavy chain constant region" refers to the region containing constant domains in the immunoglobulin heavy chain, i.e., CH1 structure, hinge region, CH2 domain and CH3 domain. In certain embodiments, the human IgG constant region extends from Ala118 to the carboxyl terminus of the heavy chain (numbered according to the Kabat EU index). However, the C-terminal lysine (Lys447) of the constant region may be present or absent (numbered according to the Kabat EU index). The term "constant region" refers to a dimer comprising two heavy chain constant regions, which can be covalently linked to each other via hinge region cysteine ​​residues to form an interchain disulfide bond.

[0202] The term "heavy chain Fc region" refers to the C-terminal region of an immunoglobulin heavy chain, which contains at least a portion of a hinge region (middle and lower hinge region), a CH2 domain, and a CH3 domain. In certain embodiments, the human IgG heavy chain Fc region extends from Asp221 or from Cys226 or from Pro230 to the carboxyl terminus of the heavy chain (numbered according to the Kabat EU index). Therefore, the Fc region is smaller than the constant region but consistent with it in the C-terminal portion. However, the C-terminal lysine (Lys447) in the heavy chain Fc region may be present or absent (numbered according to the Kabat EU index). The term "Fc region" refers to a dimer comprising two heavy chain Fc regions, which can be covalently linked to each other by hinge region cysteine ​​residues to form an interchain disulfide bond.

[0203] The constant region of an antibody, more precisely the Fc region (and this is also true for the constant region), is directly involved in complement activation, C1q binding, C3 activation and Fc receptor binding. Although the effect of an antibody on the complement system depends on certain conditions, binding to C1q is caused by a defined binding site in the Fc region. Such binding sites are known in the prior art and are described, for example, in the following literature: Lukas, TJ, et al., J. Immunol. 127 (1981) 2555-2560; Brunhouse, R., and Cebra, JJ, Mol. Immunol. 16 (1979) 907-917; Burton, DR, et al., Nature 288 (1980) 338-344; Thommesen, JE, et al., Mol. Immunol. 37 (2000) 995-1004; Idusogie, EE, et al., J. Immunol. 164 (2000) 4178-4184; Hezareh, M., et al., J. Virol. 75 (2001) 12161-12168; Morgan, A., et al., Immunology 86 (1995) 319-324; and EP 0 307 434. Such binding sites are, for example, L234, L235, D270, N297, E318, K320, K322, P331 and P329 (numbered according to the Kabat EU index). Antibodies of subclasses IgG1, IgG2 and IgG3 generally exhibit complement activation, C1q binding and C3 activation, while IgG4 does not activate the complement system, does not bind C1q and does not activate C3. "Fc region of an antibody" is a term well known to the skilled person and is defined based on the cleavage of an antibody by papain.

[0204] The term "monoclonal antibody" as used herein refers to an antibody obtained from a substantially homogeneous antibody population, that is, except for possible variant antibodies (e.g., containing naturally occurring mutations or produced during the production of monoclonal antibody preparations, such variants are usually presented in small amounts), each antibody comprising the population is identical and / or binds to the same epitope. Contrary to polyclonal antibody preparations that typically include different antibodies for different determinants (epitopes), each monoclonal antibody in a monoclonal antibody preparation is directed to a single determinant on an antigen. Therefore, the modifier "monoclonal" indicates that the characteristic of an antibody is obtained from a substantially homogeneous antibody population, and should not be interpreted as requiring antibodies to be produced by any particular method. For example, monoclonal antibodies can be prepared by a variety of techniques, including but not limited to hybridoma methods, recombinant DNA methods, phage display methods, and methods utilizing transgenic animals comprising all or part of a human immunoglobulin locus.

[0205] The term "valency" as used in this application indicates the presence of a specified number of binding sites in an antibody. Thus, the terms "bivalent", "tetravalent", and "hexavalent" indicate the presence of two binding sites, four binding sites, and six binding sites in an antibody, respectively.

[0206] "Monospecific antibody" refers to an antibody with a single binding specificity, i.e., specifically binding to one antigen. Monospecific antibodies can be prepared as full-length antibodies or antibody fragments (e.g., F(ab')2) or combinations thereof (e.g., full-length antibodies plus additional scFv or Fab fragments). Monospecific antibodies do not need to be monovalent, i.e., monospecific antibodies may contain more than one binding site that specifically binds to one antigen. For example, natural antibodies are monospecific but bivalent.

[0207] "Multispecific antibody" means having binding specificity for at least two different epitopes or two different antigens on the same antigen. Multispecific antibodies can be prepared as full-length antibodies or antibody fragments (e.g., F(ab')2 bispecific antibodies) or combinations thereof (e.g., full-length antibodies plus additional scFv or Fab fragments). Multispecific antibodies are at least bivalent, i.e., contain two antigen binding sites. In addition, multispecific antibodies are at least bispecific. Therefore, bivalent bispecific antibodies are the simplest form of multispecific antibodies. Engineered antibodies with two, three or more (e.g., four) functional antigen binding sites have also been reported (see, e.g., US2002 / 0004587).

[0208] In certain embodiments of all aspects and embodiments of the invention, the antibody is a multispecific antibody, such as at least one bispecific antibody. In certain embodiments, one of the binding specificities is directed against a first antigen, and the other is directed against a different second antigen. In certain embodiments, a multispecific antibody can bind to two different epitopes of the same antigen. Multispecific antibodies can also be used to localize cytotoxic agents to cells expressing one or more antigens.

[0209] Multispecific antibodies can be prepared as full-length antibodies or antibody-antibody fragment fusions.

[0210] Techniques for making multispecific antibodies include, but are not limited to, recombinant co-expression of two immunoglobulin heavy chain-light chain pairs of different specificities (see Milstein, C. and Cuello, AC, Nature 305 (1983) 537-540, WO 93 / 08829, and Traunecker, A. et al., EMBO J. 10 (1991) 3655-3659) and “knob-in-hole” engineering (see, e.g., US 5,731,168). Multispecific antibodies can also be prepared by engineering electrostatic manipulation effects for preparing antibody Fc-heterodimer molecules (see, e.g., WO 2009 / 089004); cross-linking two or more antibodies or fragments (see, e.g., US 4,676,980, and Brennan, M. et al., Science, 229 (1985) 81-83); using leucine zippers to produce bispecific antibodies (see, e.g., Kostelny, S. A. et al., J. Immunol. 148 (1992) 1547-1553); using common light chain technology to avoid light chain mispairing problems (see, e.g., WO 98 / 50431); using specific techniques for preparing bispecific antibody fragments (see, e.g., Holliger, P. et al., Proc. Natl. Acad. Sci. USA 90 (1993) 6444-6448); and trispecific antibodies were prepared as described in Tutt, A. et al., J. Immunol. 147 (1991) 60-69.

[0211] Also included herein are engineered antibodies with three or more antigen binding sites, including, for example, "octopus antibodies" or DVD-Ig (see, for example, WO 2001 / 77342 and WO 2008 / 024715). Other examples of multispecific antibodies with three or more antigen binding sites can be found in WO 2010 / 115589, WO 2010 / 112193, WO 2010 / 136172, WO 2010 / 145792 and WO 2013 / 026831. Bispecific antibodies or their antigen-binding fragments also include "double-acting Fab" or "DAF" (see, for example, US2008 / 0069820 and WO 2015 / 095539).

[0212] Multispecific antibodies can also be provided in an asymmetric format with domain crossing, i.e. by exchanging VH / VL domains (see, e.g., WO 2009 / 080252 and WO 2015 / 150447), CH1 / CL domains (see, e.g., WO 2009 / 080253) or complete Fab arms (see, e.g., WO 2009 / 080251, WO 2016 / 016299, see also Schaefer et al., Proc. Natl. Acad. Sci. USA 108 (2011) 1187-1191, and Klein et al., MAbs 8 (2016) 1010-1020) in one or more binding arms with the same antigen specificity. In certain embodiments of all aspects and embodiments of the invention, the bispecific antibody comprises a Cross-Fab fragment. The term "Cross-Fab fragment" means a Fab fragment in which the variable or constant regions of the heavy and light chains are exchanged. The cross-Fab fragment contains a polypeptide chain consisting of a light chain variable region (VL) and a heavy chain constant region 1 (CH1), and a polypeptide chain consisting of a heavy chain variable region (VH) and a light chain constant region (CL). Asymmetric Fab arms can also be engineered by introducing charged or uncharged amino acid mutations into the domain interface to guide the correct pairing of Fab heavy chain fragments and cognate light chains. See, for example, WO 2016 / 172485.

[0213] The antibody or fragment may also be a multispecific antibody as described in WO 2009 / 080254, WO 2010 / 112193, WO 2010 / 115589, WO 2010 / 136172, WO 2010 / 145792, or WO 2010 / 145793.

[0214] The antibody or fragment thereof may also be a multispecific antibody as disclosed in WO 2012 / 163520.

[0215] Various other molecular formats of multispecific antibodies are known in the art and can be produced using the cells according to the invention (see, for example, Spiess et al., Mol. Immunol. 67 (2015) 95-106).

[0216] Bispecific antibodies are generally antibody molecules that specifically bind to two different, non-overlapping epitopes on the same antigen or to two epitopes on different antigens.

[0217] In certain embodiments of all aspects and embodiments, the antibody is a composite (multispecific) antibody selected from the group of (composite) (multispecific) antibodies consisting of:

[0218] Full-length antibodies with domain swapping

[0219] (i.e., a multispecific IgG antibody comprising a first Fab fragment and a second Fab fragment, wherein in the first Fab fragment

[0220] a) only the CH1 domain and the CL domain are replaced with each other (i.e., the light chain of the first Fab fragment comprises the VL domain and the CH1 domain, and the heavy chain of the first Fab fragment comprises the VH domain and the CL domain); b) only the VH domain and the VL domain are replaced with each other (i.e., the light chain of the first Fab fragment comprises the VH domain and the CL domain, and the heavy chain of the first Fab fragment comprises the VL domain and the CH1 domain); or

[0221] c) the CH1 and CL domains are replaced with each other and the VH and VL domains are replaced with each other (i.e. the light chain of the first Fab fragment comprises the VH and CH1 domains, and the heavy chain of the first Fab fragment comprises the VL and CL domains); and

[0222] The second Fab fragment comprises a light chain and a heavy chain, the light chain comprises a VL and a CL domain, and the heavy chain comprises a VH and a CH1 domain;

[0223] The full length antibody with domain exchange may comprise a first heavy chain comprising a CH3 domain and a second heavy chain comprising a CH3 domain, wherein the two CH3 domains are engineered in a complementary manner by corresponding amino acid substitutions so as to support heterodimerization of the first heavy chain with the modified second heavy chain, e.g., as disclosed in WO 96 / 27011, WO 98 / 050431, EP1870459, WO 2007 / 110205, WO 2007 / 147901, WO 2009 / 089004, WO 2010 / 129304, WO 2011 / 90754, WO 2011 / 143545, WO 2012 / 058768, WO 2013 / 157954 or WO 2013 / 096291 (incorporated herein by reference);

[0224] Full-length antibody with domain swapping and additional heavy chain C-terminal binding site (BS)

[0225] (i.e., a multispecific IgG antibody comprising

[0226] a) a full-length antibody comprising two pairs each of a full-length antibody light chain and a full-length antibody heavy chain, wherein the binding site formed by each of the pairs of full-length heavy chain and full-length light chain specifically binds to a first antigen, and

[0227] b) an additional Fab fragment, wherein the additional Fab fragment is fused to the C-terminus of one heavy chain of the full-length antibody, wherein the binding site of the additional Fab fragment specifically binds to a second antigen,

[0228] wherein the additional Fab fragment that specifically binds to a second antigen i) comprises a domain crossover such that a) the light chain variable domain (VL) and the heavy chain variable domain (VH) are replaced with each other, or b) the light chain constant domain (CL) and the heavy chain constant domain (CH1) are replaced with each other, or ii) is a single chain Fab fragment);

[0229] One-armed single-chain antibody

[0230] (i.e., an antibody comprising a first binding site that specifically binds to a first epitope or antigen and a second binding site that specifically binds to a second epitope or antigen, wherein the individual chains are as follows

[0231] -Light chain (variable light chain domain + light chain kappa constant domain)

[0232] - Combined light chain / heavy chain (variable light chain domain + light chain constant domain + peptide linker + variable heavy chain domain + CH1 + hinge + CH2 + CH3 with knob mutation)

[0233] -Heavy chain (variable heavy chain domain + CH1 + hinge + CH2 + CH3 with hole mutation); two-armed single-chain antibody

[0234] (i.e., an antibody comprising a first binding site that specifically binds to a first epitope or antigen and a second binding site that specifically binds to a second epitope or antigen, wherein the individual chains are as follows

[0235] - Combined light chain / heavy chain 1 (variable light chain domain + light chain constant domain + peptide linker + variable heavy chain domain + CH1 + hinge + CH2 + CH3 with hole mutation)

[0236] - Combined light chain / heavy chain 2 (variable light chain domain + light chain constant domain + peptide linker + variable heavy chain domain + CH1 + hinge + CH2 + CH3 with knob mutation));

[0237] Common light chain bispecific antibodies

[0238] (i.e., an antibody comprising a first binding site that specifically binds to a first epitope or antigen and a second binding site that specifically binds to a second epitope or antigen, wherein the individual chains are as follows

[0239] -Light chain (variable light chain domain + light chain constant domain)

[0240] -Heavy chain 1 (variable heavy chain domain + CH1 + hinge + CH2 + CH3 with hole mutation)

[0241] - Heavy chain 2 (variable heavy chain domain + CH1 + hinge + CH2 + CH3 with knob mutation));

[0242] T cell bispecific antibodies (TCB)

[0243] (i.e., a full-length antibody having: an additional heavy chain N-terminal binding site with domain swapping comprising

[0244] - a first Fab fragment and a second Fab fragment, wherein each binding site of the first Fab fragment and the second Fab fragment specifically binds to a first antigen,

[0245] - a third Fab fragment, wherein the binding site of the third Fab fragment specifically binds to the second antigen, and wherein the third Fab fragment comprises a domain crossover such that the variable light chain domain (VL) and the variable heavy chain domain (VH) are replaced with each other, and

[0246] - an Fc region comprising a first Fc region polypeptide and a second Fc region polypeptide,

[0247] wherein the first Fab fragment and the second Fab fragment each comprise a heavy chain fragment and a full-length light chain,

[0248] wherein the C-terminus of the heavy chain fragment of the first Fab fragment is fused to the N-terminus of the first Fc region polypeptide,

[0249] wherein the C-terminus of the heavy chain fragment of the second Fab fragment is fused to the N-terminus of the variable light chain domain of the third Fab fragment, and the C-terminus of the CH1 domain of the third Fab fragment is fused to the N-terminus of the second Fc region polypeptide);

[0250] Antibody-polymer fusion

[0251] (i.e., a multimeric fusion protein comprising

[0252] (a) an antibody heavy chain and an antibody light chain, and

[0253] (b) a first fusion polypeptide comprising, in the N-terminal to C-terminal direction, a first portion of a non-antibody multimeric polypeptide, an antibody heavy chain CH1 domain or an antibody light chain constant domain, an antibody hinge region, an antibody heavy chain CH2 domain, and an antibody heavy chain CH3 domain; and a second fusion polypeptide comprising, in the N-terminal to C-terminal direction, a second portion of the non-antibody multimeric polypeptide and, in the case where the first polypeptide comprises an antibody heavy chain CH1 domain, an antibody light chain constant domain, or in the case where the first polypeptide comprises an antibody light chain constant domain, an antibody heavy chain CH1 domain,

[0254] in

[0255] (i) the antibody heavy chain of (a) and the first fusion polypeptide of (b), (ii) the antibody heavy chain of (a) and the antibody light chain of (a), and (iii) the first fusion polypeptide of (b) and the second fusion polypeptide of (b) are each independently covalently linked to each other via at least one disulfide bond,

[0256] in

[0257] The variable domains of the antibody heavy chain and antibody light chain form a binding site that specifically binds to an antigen).

[0258] "Knob-in-hole" dimerization modules and their use in antibody engineering are described in Carter P., Ridgway JBB, Presta LG: Immunotechnology, Vol. 2, No. 1, February 1996, pp. 73-73(1).

[0259] The CH3 domain in the antibody heavy chain can be changed by the "knob-into-holes" technology, which is described in detail with several examples in, for example, WO 96 / 027011, Ridgway, JB et al., Protein Eng. 9 (1996) 617-621 and Merchant, AM et al., Nat. Biotechnol. 16 (1998) 677-681. In this method, the interaction surface of the two CH3 domains is changed to increase the heterodimerization of the two CH3 domains, and thereby increase the heterodimerization of the polypeptides containing them. One of the two CH3 domains (of the two heavy chains) can be a "knob" and the other a "hole". The introduction of disulfide bridges further stabilizes the heterodimers (Merchant, AM et al., Nature Biotech. 16 (1998) 677-681; Atwell, S. et al., J. Mol. Biol. 270 (1997) 26-35) and increases the yield.

[0260] The mutation T366W in the CH3 domain (of the antibody heavy chain) is denoted as a "knob mutation" or "mutated knob", while the mutations T366S, L368A, Y407V in the CH3 domain (of the antibody heavy chain) are denoted as "hole mutations" or "mutated hole" (numbering according to the Kabat EU index). Additional interchain disulfide bridges located between the CH3 domains can also be used by introducing the S354C mutation into the CH3 domain of the heavy chain with a "knob mutation" (denoted as "knob-cys-mutation" or "mutated knob-cys") or by introducing the Y349C mutation into the CH3 domain of the heavy chain with a "hole mutation" (denoted as "hole-cys-mutation" or "mutated hole-cys") (numbering according to the Kabat EU index) (Merchant, AM et al., Nature Biotech. 16 (1998) 677-681).

[0261] As used herein, the term "domain crossing" means that in an antibody heavy chain VH-CH1 fragment and its corresponding cognate antibody light chain pair, i.e. in antibody Fab (fragment antigen binding), the domain sequence deviates from the sequence in the native antibody in that at least one heavy chain domain is replaced by its corresponding light chain domain, or vice versa. There are three general types of domain crossings: (i) crossings of CH1 and CL domains, which results from domain crossings in the light chain to produce a VL-CH1 domain sequence, and from domain crossings in the heavy chain fragment to produce a VH-CL domain sequence (or a full-length antibody heavy chain having a VH-CL-hinge-CH2-CH3 domain sequence); (ii) domain crossings of VH and VL domains, which results from domain crossings in the light chain to produce a VH-CL domain sequence, and from domain crossings in the heavy chain fragment to produce a VL-CH1 domain sequence; and (iii) domain crossings of a complete light chain (VL-CL) and a complete VH-CH1 heavy chain fragment ("Fab crossing"), which results from domain crossings to produce a light chain having a VH-CH1 domain sequence, and from domain crossings to produce a heavy chain fragment having a VL-CL domain sequence (all of the aforementioned domain sequences are presented in the N-terminal to C-terminal direction).

[0262] As used herein, the term "replacement of one another" with respect to the corresponding heavy chain domain and light chain domain refers to the aforementioned domain intersection. Thus, when the CH1 domain and the CL domain "replacement of one another", it refers to the domain intersection mentioned under item (i) and the resulting heavy chain and light chain domain sequences. Thus, when VH and VL "replacement of one another", it refers to the domain intersection mentioned in item (ii); and when the CH1 and CL domains "replacement of one another" and the VH and VL domains "replacement of one another", it refers to the domain intersection mentioned in item (iii). For example, bispecific antibodies comprising domain exchange are reported in WO 2009 / 080251, WO 2009 / 080252, WO 2009 / 080253, WO 2009 / 080254 and Schaefer, W. et al., Proc. Natl. Acad. Sci USA 108 (2011) 11187-11192. Such antibodies are often referred to as CrossMab.

[0263] In certain embodiments of all aspects and embodiments of the invention, the multispecific antibody comprises at least one Fab fragment comprising a domain intersection of CH1 and CL domains, or a domain intersection of VH and VL domains, or a domain intersection of VH-CH1 and VL-VL domains. In a multispecific antibody with domain intersection, Fabs that specifically bind to the same antigen are constructed to have the same domain sequence. Therefore, in the case where more than one Fab with domain intersection is included in the multispecific antibody, the Fabs specifically bind to the same antigen.

[0264] A "humanized" antibody refers to an antibody comprising amino acid residues from non-human HVRs and amino acid residues from human FRs. In certain embodiments, a humanized antibody will substantially comprise at least one variable domain, typically two variable domains, wherein all or substantially all HVRs (e.g., CDRs) correspond to HVRs of non-human antibodies, and all or substantially all FRs correspond to FRs of human antibodies. A humanized antibody may optionally comprise at least a portion of an antibody constant region derived from a human antibody. An antibody in "humanized form," such as a non-human antibody, refers to an antibody that has undergone humanization.

[0265] As used herein, the term "recombinant antibody" refers to all antibodies (chimeric, humanized and human antibodies) prepared, expressed, created or isolated by recombinant means, such as using cells according to the invention. This includes antibodies isolated from recombinant cells, such as NS0, HEK, BHK, amniocytes or CHO cells modified according to the invention.

[0266] As used herein, the term "antibody fragment" refers to a molecule other than an intact antibody, which comprises a portion of an intact antibody and binds to the same epitope on the same antigen as the intact antibody, i.e., it is a functional fragment. Examples of antibody fragments include, but are not limited to, Fv; Fab; Fab'; Fab'-SH; F(ab')2; bispecific Fab, diabody, linear antibody, single-chain antibody molecule (e.g., scFv or scFab).

[0267] Embodiments of the method according to the invention

[0268] Here, we report scifAI, a machine learning framework for efficient and interpretable analysis of high-throughput imaging data based on a modular implementation.

[0269] It has been shown herein that the methods according to the invention have the potential to: (i) predict the frequencies of immunologically relevant cell classes, (ii) perform systematic morphological profiling of immune synapses, (iii) study inter-donor and inter- and intra-experimental variability, and (iv) characterize the mode of action of therapeutic antibodies, and (v) predict their functionality in vitro.

[0270] The present invention is based at least in part on the discovery that high-throughput imaging of immune synapses using IFCs combined with specific data pre-processing and machine learning allows for screening of novel antibody candidates and improved evaluation of lead molecules in terms of functionality, mode of action insights, and antibody characteristics such as affinity, avidity, and format.

[0271] The present invention is illustrated below using specific antibodies and techniques. This is presented only as an example of the working of the present invention and should not be construed as limiting. The true scope of the present invention is set forth in the appended claims.

[0272] A comprehensive multichannel imaging flow cytometry dataset of the immune synapse

[0273] Using high-throughput IFC, a comprehensive dataset has been generated for the systematic analysis of the immune synapse of T cell / B cell conjugates (TB conjugates) (see Figure 1 and 2 Human memory CD4+ T cells isolated from peripheral blood of different donors were co-cultured with EBV-transformed lymphoblastoid B cells (B-LCLs) that expressed high levels of co-stimulatory molecules CD86 and CD80 or were untreated with superantigen (Staphylococcus aureus enterotoxin A, SEA) pulsed ( Figure 3-6 P-CD3ζ(Y142) was chosen as a readout of early T cell activation, the highest titrated concentration of SEA (100 ng / mL) and a time point of 45 min to investigate functional immune synapses ( Figure 7 and 8A total of nine donors from four independent experiments were screened ( Figure 2 ), and acquired 1,182,782 images (±SEA, Figure 3 ).

[0274] It has been found that a suitable multichannel set for analysis and classification consists of bright field (BF), F-actin (cytoskeleton), MHCII, CD3 and P-CD3ζ (TCR signaling). This allows the capture of a wide range of biologically driven properties of the immune synapse ( Figure 1 ).

[0275] Use a multi-step pipeline to remove dead, deformed, unfocused or clipped cells (see Examples).

[0276] In addition, a set of 5221 images from seven randomly selected donors were labeled by expert immunologists into nine categories at two levels. Figure 3 and Fig. 9 ). The first level indicates the number of existing cells in the image: singletons (n=1), doublets (n=2) and multitons (n>2). The second level characterizes the type of cells, the interactions between cells and the presence of TCR signaling. Singletons consist of the categories "single B-LCL", "single T cell signaling" and "single T cell with signaling". In the following, "no" is represented as "w / o" and "yes" is represented as "w / ". Doublets contain the categories "T cells with small B-LCLs", "B-LCLs and T cells in a layer", "synapses without signaling", "synapses with signaling" and "no cell-interactions". The category "multiple synapses" contains more than two cells and at least one B-LCL and T cell.

[0277] Without being bound by this theory, it is assumed that the categories "T cells with small B-LCL" and "no cell-cell interaction" are experimental artifacts. However, they were annotated to enhance the predictive power of the classification model and were subsequently filtered out and not used for further analysis (see Examples).

[0278] ScifAI: An explainable AI framework for analyzing multichannel imaging flow cytometry data

[0279] Here we report a single-cell imaging flow cytometry AI (scifAI) module.

[0280] Generally applicable to single-cell imaging projects, the module provides functionality for importing and preprocessing input data, several feature engineering pipelines (including a set of biologically driven features and implementations of autoencoder-generated features, see Examples), and efficient and meaningful feature selection methods.

[0281] In addition, the module implements several machine learning and deep learning models for training supervised image classification models, for example, for predicting cell configurations (such as immune synapses). Following the principles of multiple instance learning, the module also implements the function of regressing a set of selected images for downstream continuous readouts (such as cytokine production).

[0282] Extensive documentation and example code in the form of Jupyter notebooks are available online at https: / / github.com / marrlab / scifAI / and https: / / github.com / marrlab / scifAI-notebooks.

[0283] ScifAI for high-throughput profiling of immune synapses

[0284] To characterize immune synapses in an unbiased manner, we first designed and computed a set of biologically driven interpretable features using the scifAI module. These features are based on morphological, intensity, colocalization, texture, and synaptic features extracted from 5 sets of stained images and their corresponding masks (see Examples and Figure 10-11 ). Synaptic features are implemented based on the ratio of the signal intensity of each fluorescent channel in the synaptic region to the whole cell. Taking advantage of the large amount of unlabeled data, a multi-channel autoencoder is implemented to learn a second set of data-driven features from the image in an unsupervised manner

[24] . The autoencoder is designed to encode the image into a 128-dimensional abstract feature space by reconstructing the input image (see example).

[0285] Subsequently, scifAI was used to decompose the supervised machine learning pipeline to classify 5221 annotated images across nine immunologically relevant cell classes. A series of supervised machine learning models for predicting all nine categories using both interpretable feature spaces and abstract autoencoder features across all donors and experimental conditions were trained and benchmarked. These models included XGBoost classifiers for interpretable features and multi-class logistic regression (LR) for interpretable and data-driven features. To pre-select features and reduce dimensionality, a feature pre-selection pipeline was implemented using a collection of different methods (see Examples and Figure 12-13). For comparison, several convolutional neural network (CNN) architectures (such as Resnet18, ResNet34, DeseNet121, and DeepFlow) were trained, which have previously been shown to be successful in the classification task of imaging flow cytometry data [22, 24, 28]. CNN architectures essentially learn feature representations based on the input image and its corresponding labels. All models were trained on a stratified subset of 2923 (70%) annotated images. To benchmark the classification models and feature space combinations, the macro F1 scores on the remaining images were compared as a hold-out test set containing 1567 (30%) annotated images (see Examples). The XGBoost model using the interpretable feature set performed best among all classifiers (F1 macro = 0.93 ± 0.01, mean ± std bootstrap, n = 1000).

[0286] Thus, in certain embodiments of all aspects and embodiments of the method according to the present invention, an XGBoost model utilizing an interpretable feature set is used.

[0287] The XGBoost model was followed by the convolutional neural network ResNet34 (0.92 ± 0.01), ResNet18 (0.91 ± 0.01), DeepFlow (0.90 ± 0.01), DenseNet121 (0.90 ± 0.02), multi-class logistic regression using an interpretable feature set (0.89 ± 0.02), and logistic regression using a data-driven feature set (0.83 ± 0.02).

[0288] It was found that the XGBoost model provided the best trade-off between performance and interpretability. Therefore, the XGBoost model was chosen as the final classifier for the labeled expansion to the full dataset ( Figure 4 ).

[0289] Investigation of the model confusion matrix on the holdout set revealed that misclassifications occurred primarily within the signaling properties of the cell classes, while all other classes showed good overall agreement (see Fig.14 ).

[0290] After training the XGBoost classifier, it has been explored which underlying features drive the class prediction. Therefore, these features have been sorted by their corresponding Gini index (see Figure 5 ). The most predictive features were based on the colocalization of CD3 and MHCII, the colocalization of MHCII and P-CD3ζ, the texture of MHCII and CD3, and the intensity of P-CD3ζ.

[0291] In certain embodiments according to all aspects and embodiments of the methods of the present invention, in addition to synapses, cell doublets and multimers are classified based on one or more of the following:

[0292] - the (correlation) distances of MHCII and CD3 markers,

[0293] - the distance between the centers of MHCII and CD3 markers,

[0294] - Manders overlap coefficient of MHCII and P-CD3ζ markers,

[0295] - homogeneity of CD3 marker,

[0296] -Comparison of MHCII markers,

[0297] - Kurtosis intensity of P-CD3ζ labeling,

[0298] - Manders overlap coefficients for MHCII and CD3 markers, and / or

[0299] - Maximum intensity of CD3 labeling.

[0300] In a preferred embodiment of all aspects and embodiments of the method according to the present invention, in addition to synapses, cell doublets and multimers are classified based on one or more of the following:

[0301] - Co-localization of CD3 and MHCII markers, and / or

[0302] - Colocalization of MHCII and P-CD3ζ markers, and / or

[0303] - Texture of MHCII, and / or

[0304] - CD3-marked textures, and / or

[0305] -P-CD3ζ labeling intensity.

[0306] Without being bound by this theory, it is assumed that based on the definition of features and classes, (i) the texture of CD3 and MHCII markers can be used to detect the presence of T- and B-LCL cells in the image, (ii) the colocalization of CD3 and MHCII markers can be used to detect different doublet types, and (iii) the intensity of P-CD3ζ markers and the colocalization of MHCII and P-CD3ζ markers can be used to detect whether it is a signaling T cell (see Fig.15 ).

[0307] The annotated data subset and available IFC channels are sufficient to achieve high classification performance

[0308] We further investigated how many annotated samples are needed to achieve reasonable classification performance. Therefore, we repeatedly trained the model using stratified subsets of the training data and evaluated the F1 macro on the test set. The results showed that by using 1500 images (45% of the training data), we could achieve 90% F1 macro on the test set ( Fig.16 ).

[0309] Furthermore, it has been determined which channels are sufficient to achieve high performance. Therefore, the BF channel is retained and the model is trained using all possible combinations of fluorescent channels.

[0310] It was found that channels BF, MHCII, and P-CD3ζ were sufficient to achieve a similar result to the F1 macro using all channels ( Fig.17 ).

[0311] Characterizing the effects of therapeutic antibodies on synapse formation

[0312] We further investigated the effects of therapeutic antibodies on immune synapse formation and better characterized their morphological profiles. This analysis included the study of latent class frequency changes and feature differences.

[0313] Two types of antibodies have been used in research: an immune response activator and an immune response suppressor. T cell activating bispecific (TCB) antibodies are designed to target CD3 and CD19 (co-receptors for B cells)

[29] (see Fig.18 Inhibitory antibodies (ie, Teplizumab) are described as binding only to CD3 (see Fig.19 ) and has been shown to attenuate T cell responses [30,31]. For each antibody, appropriate controls (Ctrl-TCB and isotype, respectively) were run in the same experiment and donor. Since teplizumab requires an existing immune response for subsequent suppression, T cells were first stimulated with SEA (see Fig.18 The same settings were also used for isotype controls. CD19-TCB was measured in six donors across two experiments and Teplizumab in seven donors across three experiments (see Figure 20-23 To determine the class frequency changes between antibodies and their controls, the previous XGBoost classifier (see Figure 4 and 15 ) predicts the categories of all images based on interpretable features (see examples and Fig.24). To ensure that the previously trained XGBoost model was transferable from ±SEA to the antibody experiments, experts annotated a randomly selected subset of 396 images of CD19-TCB and 227 images of Teplizumab. The high agreement between the expert annotations and the XGBoost predictions for the new experiments (macro F1 score = 0.86 for TCB and 0.85 for Teplizumab) confirmed that the trained model was generalizable and could therefore be used for further analysis (see Fig.25 ).

[0314] To compactly represent class frequency changes, log2 fold change values ​​between antibodies and their corresponding controls were calculated.

[0315] Focusing on the differences in the features of synapses stimulated by antibodies, images predicted as "signaling synapses" for each donor were selected and interpretable features from the fluorescence channels alone (including texture, synaptic features, morphology, intensity, and colocalization between antibodies) were compared with their controls. The BF channel was not included because its intensity was difficult to interpret and its morphological properties were captured by other fluorescence channels (see Methods).

[0316] In certain embodiments according to all aspects and embodiments of the invention, the method is used to determine changes in class frequency in the presence of a therapeutic antibody, wherein the number or frequency of doublets and multimers of synapses signaling in the absence and presence of a therapeutic antibody is determined, and / or wherein interpretable features from a fluorescent channel (including texture, synaptic features, morphology, intensity, and co-localization between antibodies) are compared to a control thereof.

[0317] CD19-TCB increases the formation of stable immune synapses

[0318] CD19-TCB stimulated immune response resulting in a significant increase in doublet and multiplet frequencies. The "synapses with signaling" category thus showed the highest increase (median log_2 (CD19-TCB / Ctrl-TCB) = 2.6, n = 6 donors, p = 0.036), followed by "multiples of two B cells and one T cell forming synapses with signaling" (median = 1.94, p = 0.036), "B-LCL and T cells in one layer" (median = 1.77, p = 0.036), and "doublets or multiples of one or two B cells and one T cell forming synapses without signaling" category (median = 0.44, p = 0.036). For monosomes, the overall trend was a decrease in the frequency of the categories "single B-LCL" (median = -0.29, p = 0.036) and "single T cell without signaling" (median = -0.78, p = 0.036) (see Fig.26 ).

[0319] In addition, the feature differences of CD19-TCB-induced synapses were analyzed (see Methods), comparing 210 interpretable features from all fluorescence channels. It was found that among the 210*6=1260 possible combinations of features and donors, 210 features were significantly increased and 163 features were significantly decreased (see Fig. 27 All donors showed largely similar responses to stimulation with CD19-TCB. An average of 27 ± 4 features per donor were significantly decreased, and 33 ± 7 features were significantly increased ( Fig. 27 From these features, a number of features with similar variations in at least 4 of the 6 donors were identified ( Fig. 27 and Table 1).

[0320] Table 1: Significant features induced by CD19-TCB. Shown are features that were significantly altered in at least four donors after addition of CD19-TCB. This table represents Fig. 27 1 indicates a significant increase ( Fig. 27 red in the figure), -1 indicates a significant decrease ( Fig. 27 ), and 0 indicates no significant change ( Fig. 27 gray in the image).

[0321]

[0322]

[0323] Furthermore, similar to SEA stimulation, an increase in the mean intensity of P-CD3ζ was identified, with a higher enrichment within synaptic regions (see Figure 28-29 and Table 1).

[0324] In addition, a stronger enrichment of F-actin and MHCII to synapses has been found ( Figure 31-34 ).

[0325] Thus, the addition of therapeutic antibodies resulted in an increase in doublet and multimer frequency and (stronger) enrichment of F-actin and MHCII in synaptic regions. Without being bound by theory, it is hypothesized that this indicates enhanced formation of tight immune synapses, which translates into efficient TCR signaling. These observations are consistent with the generally described mode of action of TCBs, promoting stable interactions between tumor cells and T cells [32-33].

[0326] Teplizumab alters synapse formation and TCR signaling

[0327] The presence of teplizumab significantly reduced the frequency of doublets and multiplets compared with CD19-TCB ( Fig.35). The highest decrease was observed for the category of "doublets of one B cell and one T cell forming a synapse with signaling" (median log_2(teplizumab / isotype) = -0.75, n = 7, p = 0.018), followed by "multiples of two B cells and one T cell forming a synapse with signaling" (median = -0.51, p = 0.031), "doublets of B cells and T cells forming a synapse without signaling" (median = -0.44, p = 0.018), and "B-LCLs and T cells in one layer" (median = -0.18, p = 0.018). Thus, "single T cell with signaling" (median = 0.66, p = 0.018) and "single B-LCL" (median = 0.07, p = 0.018) were significantly increased compared to isotype. Surprisingly, the frequency of the category “T cells without signaling” was significantly reduced (median = -0.35, p = 0.018), likely due to a significant increase in “single T cells with signaling” (see Fig.35 ).

[0328] The differences in features of synapses induced by Teplizumab in seven donors were analyzed. 132 features based on F-actin, MHCII, and P-CD3ζ and their colocalization were extracted from the images of "doublets of B cells and T cells forming synapses without signaling". CD3 features could not be included for analysis due to interference from the binding of Teplizumab and anti-CD3 staining antibodies. Therefore, anti-CD4 staining antibodies were used to identify T cells. Among 132*7=924 possibilities, 131 significantly increased features and 169 significantly decreased features were identified ( Fig.30 and Table 2 ).

[0329] Table 2: Significant features induced by Teplizumab. Shown are features that were significantly altered in at least six donors after the addition of Teplizumab. This table indicates Fig.30 1 indicates a significant increase ( Fig.30 red in the figure), -1 indicates a significant decrease ( Fig.30 ), and 0 indicates no significant change ( Fig.30 gray in the image).

[0330]

[0331]

[0332] In particular, Teplizumab resulted in a significant decrease in an average of 25 ± 8 features per donor and a significant increase in 19 ± 17 features ( Fig.30Donor 6 showed the least number of changes, with five features significantly increased. In contrast, Donor 4 showed the greatest number of increased features with 50 features. A set of features was identified that were significantly increased or decreased in at least 5 of the 7 donors ( Figure 36-37 A decrease in the average intensity of F-actin was identified, while donors 2 and 4 showed a significant increase ( Fig.36 and 38 This opposite response of the two donors could also be detected in other F-actin-related features (Table 2). In addition to the changes in F-actin features, a significant reduction in the intensity of P-CD3ζ within synapses was detected, and a stronger clustering of TCR signaling around the entire T cell periphery was observed ( Fig.37 and 39 ).

[0333] Thus, using the method according to the invention, new insights into the immunosuppressive mode of action of Teplizumab were obtained due to the reduction in the number of synapses, and changes in F-actin reorganization and P-CD3ζ signaling towards synapses were identified.

[0334] in conclusion

[0335] Using the method according to the invention, the mode of action of therapeutic antibodies can be analyzed.

[0336] In addition, using the method according to the present invention, the function of therapeutic antibodies can be predicted in vitro.

[0337] The present invention is based, at least in part, on the generation of morphological profiles of the immune synapse that allow characterization of the mode of action of therapeutic antibodies early after the initiation of an immune response. From this, the associated downstream T cell responses can be predicted.

[0338] These findings and methods differ from previous work in that they did not take into account inter-experimental effects on synapse formation [21,35].

[0339] The present invention is based at least in part on the detection and determination of changes in immune synapses. This has been achieved at least in part by incorporating interpretable features extracted from fluorescence images into a machine learning framework. It has thus been achieved that the method according to the present invention is scalable, provides reproducible results and facilitates deployment into existing workflows, unlike previous work that used a combination of methods for each stage of the analysis [27, 36, 37].

[0340] Without being bound by this theory, it is hypothesized that a combination of interpretable features and illustrative machine learning allows identification of relevant morphological classes (such as immune synapses) with accuracy equal to or even better than state-of-the-art methods. The morphological spectrum of immune synapses can thereby be analyzed in an unbiased manner and the mode of action of antibodies can be characterized in a biologically relevant context.

[0341] Therefore, the method according to the present invention is an improvement compared to known methods that mainly focus on performance rather than interpretability [26,38].

[0342] The power of the method according to the invention has been shown by analyzing the effects of two therapeutic antibodies, CD19-TCB and Teplizumab, on the immune synapse, both of which bind to CD3 and have been described to activate and inhibit T cell responses, respectively [29-31].

[0343] It has been found that using the method according to the invention, more stable immune synapses are formed in the presence of CD19-TCBs, as indicated by a stronger enrichment of MHCII and F-actin within the synapse, which is paralleled by a higher intensity of P-CD3ζ labeling.

[0344] It has been found that, using the method according to the invention, in the presence of Teplizumab, a reduction in synapse formation and prevention of F-actin reorganization and localization of P-CD3ζ towards synapses occurs. These observations provide new insights into the immunosuppressive mode of action of Teplizumab, which has been little studied in vitro to date [30,31].

[0345] Without being bound by this theory, it is hypothesized that the reduced P-CD3ζ intensity in synaptic regions and the nonpolarized distribution of P-CD3ζ signals observed throughout the T cell periphery may indicate altered TCR signaling, which could translate into reduced T cell effector function. It has been reported that large peripheral P-CD3ζ microclusters are self-reactive T cells with altered synapse formation and aberrant T cell responses [6].

[0346] Surprisingly, the method according to the invention allows the identification of features within synaptic classes, revealing inter-donor variability upon stimulation with different antibodies.

[0347] Thus, the method according to the invention enables rapid in vitro screening of responders and preselection of suitable patients for clinical trials.

[0348] In conclusion, by applying the method according to the present invention, the mode of action of therapeutic antibodies can be thoroughly investigated based on salient features and also more insights can be gained into inter-donor variability that could potentially translate into different functional outcomes in vivo.

[0349] With the method according to the present invention, state-of-the-art methods have been improved by incorporating biologically driven features such as texture, intensity statistics, and synapse-related features.

[0350] For the first time, interpretable features of the immune synapse have been used to predict the effectiveness of therapeutic antibodies on T cell factor production. It is thus possible to predict the functional outcome of unseen antibodies and pinpoint the drivers required for prediction.

[0351] For example, for TCB, the intensity of MHCII and the morphology of F-actin have been found to be the most prominent features predicting cytokine readouts.

[0352] The ability to predict unseen antibodies allows for the study of various antibody formats to better understand mechanistically how different formats affect T cell responses and help guide format selection.

[0353] The methods according to the present invention encompass data acquisition and analysis, which can be tailored to investigate various hypotheses and develop different applications based on imaging flow cytometry data.

[0354] For example, although in the current example, memory CD4+ T cells were analyzed because they are expected to exhibit a more rapid immune response and higher synaptic propensity compared to naive T cells

[49] , imaging and analysis of CD8+ T cells, the main players in cytotoxicity, could similarly elucidate how synaptic features are associated with the killing efficiency of therapeutic antibodies against tumor cells.

[0355] The method according to the invention can also be used to design IFC experiments, optimizing the number and type of staining and the total number of images per donor to be acquired.

[0356] The method according to the invention can be used to improve the quality and speed of antibody development, for example to give new insights into the mode of action of a particular candidate molecule, or to predict high-throughput efficacy in vitro. The identification of lead molecules and better prioritization in terms of epitope, affinity, avidity and antibody format will have a huge impact on the decision-making process.

[0357] Besides this, the method according to the invention may even help to identify responders in a patient population and predict their clinical outcome.

[0358] ***

[0359] All references cited herein are incorporated by reference in their entirety.

[0360] ***

[0361] The following examples and drawings are provided to aid the understanding of the present invention, the true scope of the invention being set forth in the appended claims.It should be understood that modifications may be made to the procedures set forth without departing from the spirit of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0362] Figure 1Schematic diagram of the data generation and analysis pipeline. To systematically analyze the immune synapses of TB cell conjugates, 1,182,782 images were acquired using imaging flow cytometry. Afterwards, they can be manually classified (right), or scifAI (left) can be used to extract morphological features, train machine learning models, profile immune synapses, and characterize the functionality of therapeutic antibodies.

[0363] Figure 2 Gating strategy for identification of single interacting TB-LCL synapses using IDEAS software for imaging flow cytometry.

[0364] Figure 3 Experts manually annotated a subset of 5221 images into nine immunologically relevant categories that can be grouped into singletons (B cells or T cells), doublets (having one B cell and one T cell), and multitons (containing more than 2 cells). Cell images show bright field (BF, scale bar = 2.4 μm), F-actin (cytoskeleton), MHCII, CD3, and P-CD3ζ (P-CD3ζ) (a marker for TCR signaling).

[0365] Figure 4 Six different methods for training predictive machine learning models to identify immunologically relevant classes were benchmarked, combining different classification algorithms and feature engineering strategies. These methods included interpretable (interp.) features combined with an interpretable classifier, an autoencoder for generating data-driven features, an interpretable classifier, and three convolutional neural networks. Interpretable features combined with an XGBoost classifier achieved the best trade-off between interpretability and classification performance.

[0366] Figure 5 List of donors used in this study, their age, sex, and experiment number.

[0367] Figure 6 List of experiments and donors with and without SEA.

[0368] Figure 7Use conventional FACS test assay conditions.In the presence of different concentrations of SEA (0.1-100ng / mL) or untreated (-SEA), primary memory CD4+T cells separated from PBMC of healthy donors were stimulated with B-LCL cells.The frequency of P-CD3ζ+ (P-CD3ζ positive), TNF-ɑ+ (TNFα positive) and CD69+CD4+ (CD69 and CD4 positive) T cells was determined at different time points.The small FACS histogram in the bar graph shows the expression levels of three kinds of markers by comparing the highest concentration (100ng / mL) of SEA after 60 minutes with untreated control (-SEA).The data shown represent an experiment using T cells from three different donors.

[0369] Figure 8 After 45 and 120 minutes, the percentage of single TB-LCL synapses and P-CD3ζ+CD4+ (P-CD3ζ and CD4 positive) T cells was measured by imaging flow cytometry between two different SEA concentrations (10 and 100 ng / mL). Data are representative of two donors.

[0370] Fig. 9 The number of expert data labels for each donor.

[0371] Fig.10 Visual representation of each multi-channel image and the corresponding mask. The masks are exported along with the images in the IDEAS software.

[0372] Fig.11 List of interpretable features. Morphology, intensity statistics, texture Synaptic features based on one channel. Colocalization features based on two channels. ScifAI automatically detects existing channels and generates the specified features.

[0373] Fig.12 Feature pre-selection pipeline to reduce the dimensionality of the feature space and eliminate multicollinearity. First, highly correlated features are discarded. Then a collection of different classifiers are trained on the data and their top k features are selected. Finally, hierarchical clustering is performed on top of the feature union to resolve multicollinearity.

[0374] Fig.13 The number of selected features before passing them to the XGBoost classifier. To obtain the best top-k, the data selection pipeline + XGBoost is trained on stratified randomly selected 85% of the training set and tested on the remaining 15%.

[0375] Fig.14 Confusion matrix of data selection pipeline (top k = 211) + XGBoost based on predictions on test set.

[0376] Fig.15 Based on the Gini index, the first eight features detected for cell class were ranked. These features include co-localization, texture, and intensity of MHCII, CD3, and P-CD3ζ (P-CD3ζ). Exemplary images were obtained from donor 7, which sampled the 5th, 50th, and 95th percentiles of each feature distribution.

[0377] Fig.16 Number of annotated images versus classification performance.

[0378] Fig.17 The XGBoost model was used to train the classifier. A 5-fold cross validation was used and the training data was used for this evaluation. At each step, features based on the selected channel were used to train the classifier. Since brightfield (BF) is a stain-free channel, it was always kept in the data. The combinations were ranked based on the F1 macro.

[0379] Fig.18 Schematic representation of the mode of action of teplizumab.

[0380] Fig.19 Schematic representation of the CD19-TCB mode of action.

[0381] Fig. 20 Donors and their corresponding experiments used for category frequency analysis and feature difference analysis.

[0382] Fig.21 Donors used for category frequency analysis and feature difference analysis and their corresponding experiments.

[0383] Fig. 22 List of experiments and donors using TCB and its controls.

[0384] Fig.23 List of experiments and donors using teplizumab and isotype controls.

[0385] Fig.24 Class frequencies of the main categories as described in Methods. Each dot represents a donor and is color coded by experiment.

[0386] Fig.25 Confusion matrices for CD19-TCB and Teplizumab classification based on 396 and 227 expert-annotated images, respectively. The previously trained model ( Figure 4 ) achieves macro F1 scores of 0.86 and 0.85 on the two datasets, respectively.

[0387] Fig.26 Class frequency differences are depicted as log2 fold change between CD19-TCB and its corresponding control (Ctrl TCB). Each point represents a donor, and the colors are as follows Fig.21The vertical black lines are the median values ​​across donors for each category.

[0388] Fig. 27 Systematic comparison of 210 relevant features between CD19-TCB and Ctrl-TCB whose images were predicted as "signaling synapses" across six donors. Each line represents a feature and each column represents a donor. For each donor, significantly increased features are depicted in dark gray / black and significantly decreased features are depicted in light gray. Donors are sorted based on the number of significantly changed features. The bottom bar graph shows the counts of features that were increased or decreased for each donor.

[0389] Fig.28 Statistical and visual inter-donor comparison of the representative feature “Mean Intensity P-CD3ζ” between CD19-TCB and Ctrl-TCB. For visualization purposes, the feature was mapped between zero and one for each donor, respectively.

[0390] Fig.29 Visual representation of the representative feature "Mean Intensity P-CD3ζ" randomly sampled from donor 9 and found to be consistent with the statistical results (scale bar = 2.4 μm).

[0391] Fig.30 Systematic comparison of 132 relevant features between Teplizumab and isotypes whose images were predicted as "signaling synapses" in all six donors. Each line represents a feature, and each column represents a donor. For each donor, significantly increased features are depicted in dark gray / black, and significantly decreased features are depicted in light gray. Donors are sorted based on the number of significantly changed features. The bottom bar graph shows the counts of features that were increased or decreased for each donor.

[0392] Fig.31 Statistical and visual inter-donor comparison of the representative feature “F-actin enrichment in synapses” between CD19-TCB and Ctrl-TCB. For visualization purposes, the feature was mapped between zero and one for each donor, respectively.

[0393] Fig.32 Visual representation of the representative feature "F-actin enrichment in synapses" randomly sampled from Ctrl-TCBs and CD19-TCBs from donor 9 and found to be consistent with the statistical results (scale bar = 2.4 μm).

[0394] Fig.33 Statistical and visual inter-donor comparison of the representative feature “MHCII enrichment in synapses” between CD19-TCB and Ctrl-TCB. For visualization purposes, the feature was mapped between zero and one for each donor, respectively.

[0395] Fig.34 Visual representation of the representative feature "MHCII enrichment in synapses" randomly sampled from Ctrl-TCB and CD19-TCB from donor 9 and found to be consistent with the statistical results (scale bar = 2.4 μm).

[0396] Fig.35 Class frequency differences are depicted as log2 fold change between Teplizumab and its corresponding control (isotype). Each point represents a donor, and the colors are as follows Fig. 22 The vertical black lines are the median values ​​across donors for each category.

[0397] Fig.36 Statistical and visual inter-donor comparison of characteristic F-actin between teplizumab and its isoforms.

[0398] Fig.37 Statistical and visual inter-donor comparison of characteristic P-CD3ζ between teplizumab and its isoforms.

[0399] Fig.38 Visual representation of characteristic F-actin of both isotype and Teplizumab were randomly sampled from donor 3 and found to be consistent with the statistical results (scale bar = 2.4 μm).

[0400] Fig.39 Visual representation of characteristic P-CD3ζ of both isotype and teplizumab were randomly sampled from donor 3 and found to be consistent with the statistical results (scale bar = 2.4 μm).

[0401] Fig.40 Algorithm flow chart of a method for classifying cells in a cell mixture using a combination of single-cell imaging flow cytometry and artificial intelligence (scifAI) according to the present invention.

[0402] Materials and methods

[0403] Cell culture

[0404] EBV-transformed B lymphoblastoid cell line (B-LCL) of donor 333 was obtained from Astarte Biologics (No. 1038-3161JN16) and cultured in RPMI-1640 medium (PAN-Biotech; Catalog No. P04-17500) with 10% FBS (Anprotec; Catalog No. AC-SM-0014Hi) and 2mM L-glutamine (PAN-Biotech; Catalog No. P04-80100). Z138 (MCL, a gift from the University of Leicester) and Nalm-6 (ALL, DSMZ ACC 128) tumor cells were cultured in RPMI1640 containing 10% FBS and 1% Glutamax (Invitrogen / Gibco No. 35050-038).

[0405] Immune synapse formation and imaging flow cytometry

[0406] To analyze immune synapses, human memory CD4+T cells were isolated from PBMCs of nine healthy human donors using the negative selection EasySep enrichment kit (Catalog No. 19157) from STEMCELL Technologies. Live / dead staining of T cells and B-LCL cells was performed at RT for 15 minutes (eBioscience; Catalog No. 65-0865-14) using the fixable viability dye eF780. The cells were then resuspended in RPMI-1640 medium supplemented with 10% FBS (Anprotec; Catalog No. AC-SM-0014Hi), 5% penicillin-streptomycin (Gibco; Catalog No. 15140-122) and 2mM L-glutamine (PAN-Biotech; Catalog No. P04-80100). B-LCL cells were then transferred to wells of a 96-well round-bottom plate (300,000 cells per well) and pre-incubated with superantigen Staphylococcal enterotoxin A (SEA) (Sigma-Aldrich; Catalog No. S9399) for 15 min at 37°C or left untreated. mem The cells were added to the previously prepared B-LCL cells (250,000 cells per well) to produce a final ratio of 4:3 (B-LCL:T mem ), and then the appropriate in-house prepared compounds (10 μg / mL of isotype Ctrl or Teplizumab and 1 μg / mL (5 nM) of Ctrl-TCB or CD19-TCB) were added to the B-LCL-T memIn cell co-culture. In order to strengthen the conjugate formation between B-LCL cells and T cells, they were centrifuged at 300xg for 30sec and then directly transferred to a 37°C incubator for 45min. Afterwards, the medium in each well was carefully aspirated with a pipette, and the cells were immediately fixed at RT for 12 minutes, followed by permeabilization using the Foxp3 / transcription factor staining buffer set (Catalog No. 00-5523-00) from eBioscience.

[0407] Intracellular staining was performed for 40 min at 4°C in permeabilization buffer containing fluorescently labeled antibodies: CD3-BV421 (clone UCHT1, Biolegend; catalog number 300433), HLA-DR-PE-Cy7 (clone L243, Biolegend; catalog number 307616), Phalloidin AF594 (ThermoFisher; catalog number A12381), and P-CD3ζY142-AF647 (K25-407.69, BD catalog number 558489).

[0408] After washing, cells were suspended in FACS buffer (PBS supplemented with 2% FBS) and acquired on an Amnis ImageStreamX Mark II imaging flow cytometer (Luminex) equipped with five lasers (405, 488, 561, 592, and 640 nm). On average, approximately 55,000 images were collected for each sample at 60x magnification at low speed settings. IDEAS software (version 6.2.187.0, EMD Millipore) was used for data analysis and cell labeling.

[0409] To identify immune synapses using IDEAS software, a Figure 2 Gating strategy in. First, cell gating is performed on live +CD3+MHCII+ cells in focus. In this population, images showing single CD3+T cells and single MHCII+B-LCL cells are selected using area and aspect ratio features. Next, in order to exclude non-interacting cells, the CD3 intensity in the self-created synaptic mask is determined. As used herein, the term "mask" refers to the outer silhouette superimposed by all images (BF and all labeled antibodies) obtained for cells or doublets or multiplets. The synaptic mask is defined as a combination of morphological CD3 and MHCII masks, with an expansion rate of 3. Only synapses showing CD3 signals in the mask will be gated. Finally, T+B-LCL cells in a layer are excluded by utilizing the height and area features of the bright field (BF), and single TB-LCL synapses are analyzed.

[0410] Intracellular staining of cytokines using conventional flow cytometry

[0411] For intracellular cytokine staining, cells were first treated with GolgiPlug (BD Biosciences; catalog number 555029) and GolgiStop (BD Biosciences; catalog number 554724) for at least 2-4 h before staining. After incubation, live / dead staining was performed using the fixable viability dye eF780 at 4°C for 20 min (eBioscience; catalog number 65-0865-14). Cells were then fixed and permeabilized using the Foxp3 / transcription factor staining buffer set from eBioscience (catalog number 00-5523-00) as described in the synapse formation assay. Intracellular staining was performed at 4°C for 30 min in permeabilization buffer containing fluorescently labeled antibodies: TNFα-APC (clone MAb11, BD Biosciences; catalog number 554514), IFN--PE (clone B27, BD Biosciences; catalog number 554701), and granzyme B-PE-Cy7 (clone QA16A02, Biolegend; catalog number 372214). Finally, cells were suspended in FACS buffer (PBS supplemented with 2% FBS and 1 mM EDTA) and acquired on a FACS Celesta from BD Biosciences.

[0412] Tumor cell lysis assay (in vitro)

[0413] PBMCs depleted of B cells from healthy donor blood were prepared using standard density-gradient separation, followed by depletion of B cells with CD20 microbeads (Miltenyi; catalog number 130-091-104). Then, in the presence or absence of CD19-TCB, PBMCs depleted of B cells were incubated with tumor targets (Z-138 or Nalm-6) at a ratio of 5: 1 for 24 hours. Tumor cell lysis was calculated based on LDH release (LDH cytotoxicity detection kit from Roche Applied Science) and normalized to spontaneous release (PBMC+untreated target=0% tumor cell lysis) and maximum release (using Triton X-100 to lyse tumor targets=100% lysis).

[0414] Quantification of CD19 expression

[0415] Anti-human CD19-AF647 (Biolegend No. 302220) antibody and the corresponding isotype control muIgG2b (Biolegend No. 400330) were used according to the manufacturer's instructions and the Quantum TM Alexa 647MESF kit (Cat. No. 647) determined CD19 expression on B-LCL cells. To quantify CD19 molecules on tumor target cell lines Nalm-6 and Z-138, the QiFi kit (Cat. No. K0078) from Dako was performed using anti-human CD19 purified (BD No. 555410) antibody and corresponding isotype control muIgG2b (BD No. 557351) according to the manufacturer's instructions.

[0416] Preparing imaging datasets for analysis

[0417] A total of 2,899,575 flow cytometry images were recorded. The dataset consisted of nine different donors across four independent experiments. Donor 1 and donor 2 were used twice. Different conditions were measured, including -SEA (total images = 625,001), +SEA (557,781), Ctrl-TCB (330,000), CD19-TCB (324,020), isotype (405,000), and Teplizumab (403,375). Images contained bright field (BF), F-actin, MHCII, CD3, P-CD3ζ, and live-dead staining. Live-dead staining was used only to filter out dead cells. For each experiment, images were compensated using a compensation matrix derived from a single cell stained. After compensation, the original images (16 bits) and their corresponding channel segmentation masks were exported from the IDEAS software and saved in HDF5 format. To achieve parallelization, each image and its corresponding mask were saved separately.

[0418] Interpretable feature engineering from images

[0419] A set of 296 biologically driven features were extracted to study immune synapses. These features include morphology, intensity, colocalization, texture, and synaptic correlation values ​​(see Figure 5 , 6, 9, 22, 32). Morphological features were calculated based on the segmentation mask of each channel. Features included “area”, “bounding box area”, “convex area”, “eccentricity”, “equivalent diameter”, “Euler number”, “extent”, “maximum Feret diameter”, “minimum Feret diameter”, “filling area”, “major axis length”, “minor axis length”, “Hu moment”, “orientation”, “perimeter”, “Crofton perimeter”, “solidity”, and “weighted Hu moment”. All morphological features were extracted using the scikit-image library

[41] . For intensity features, cells were first segmented using their corresponding masks. Intensity features included “minimum”, “sum”, “mean”, “standard deviation”, “skewness”, “kurtosis”, “maximum”, and “Shannon entropy”. In addition, percentiles of intensity values ​​were calculated, including “10th percentile”, “20th percentile”, …, “90th percentile”. All intensity features were calculated based on NumPy

[42] and SciPy

[43] functions. For colocalization features, “dice distance” and “Jaccard distance” were implemented using the SciPy

[43] library to calculate the mask overlap between two channels. In addition, “correlation distance”

[43] , “Euclidean distance”

[43] , “Manders overlap coefficient”

[44] , “intensity correlation quotient”

[44] , “structural similarity”

[41] , and “Hausdorff distance”

[41] were calculated. For texture features, we used gray-level co-occurrence matrix (GLCM) features

[45] , including “contrast”, “dissimilarity”, “homogeneity”, “ASM”, “energy”, and “correlation”. Synaptic-related features were defined as “Ch enrichment (mean)” = (h) / (h), “Ch enrichment (sum)” = (h) / (h), and “Ch enrichment (maximum)” = (h) / (h)

[35] . Finally, “background average” and “gradient RMS” were implemented for image quality control. All of these functions were implemented using NumPy (version = 1.18.5), Pandas (1.1.5), SciPy (1.8.0), scikit-image (0.19.2), and scikit-learn (1.0.2)

[46] .

[0420] Autoencoder feature extraction

[0421] To utilize the large amount of unlabeled data, we implemented and trained a multi-channel autoencoder

[24] . This autoencoder consists of a separate encoder for each channel. The encoder is designed to map each channel to a 32-dimensional vector. The concatenation of these vectors forms a 5*32-dimensional space. These features are then mapped to a 128-dimensional feature vector. A decoder is implemented on top of the concatenated vectors to reconstruct the original image. The “L2 norm” is used as the reconstruction loss. The enhancements used to train the autoencoder include random rotation, random scaling, random flipping, and random Gaussian noise.

[0422] Feature pre-selection

[0423] Considering the large number of features, a feature pre-selection pipeline was implemented to select the most relevant features based on the work of Haq et al.

[47] (see Figure 12-14 ). First, the Pearson correlation between the features is measured. If at least two features are highly correlated (|corr|>0.95), only one of them is retained (randomly) and the rest are eliminated. In the next step, the features are ranked using six different methods. These methods include mutual information, linear support vector machine, logistic regression with L1 regularization, logistic regression with L2 regularization, random forest, and XGBoost. The top k (hyperparameters to be chosen) features from each method are selected and their union is used. After this reduction, the Spearman correlation matrix between the features is calculated and spectral clustering is performed on the correlations. Then, m clusters are created and a feature is randomly selected for each cluster. The last step is performed to account for multicollinearity between the features.

[0424] Classification

[0425] There are three main approaches for training: supervised learning algorithms, feature-based methods, and deep learning.

[0426] Classic supervised learning model

[0427] Two different algorithms were used to train machine learning models. A boosting method called XGBoost

[48] was used, which used an ensemble of trees (n_trees = 100) on the data. The second model was a logistic regression. The advantage of using these models is that they provide interpretability after training.

[0428] Convolutional Neural Networks

[0429] To train supervised deep learning models, well-known architectures from the field of computer vision were used, including ResNet18, Resnet34, ResNet50, ResNet152, DeseNet121, and DeepFlow [22, 24, 28]. All models were pre-trained on ImageNet. Considering that the models were designed for three-channel input, the first convolutional layer with three input channels to six input channels was removed. In addition, the classification layer also needed to be adjusted to nine categories. However, the rest of the network remained as is, using their pre-defined ImageNet weights. We used multi-class cross entropy loss for training. The learning rate (lr) was set to 0.001, with an adaptive strategy of gradually decreasing it within 10 epochs. The augmentations used to train the autoencoder included random rotation, random scaling, random flipping, and random Gaussian noise.

[0430] Classification Feature Importance

[0431] Feature pre-selection filtering is used to reduce the number of features, and then the XGBoost classifier is trained on the annotated data. Although XGBoost can provide feature importance using the Gini index, these importances may be biased for different reasons (such as correlation between pre-selected features, number of features, pre-selection process, outliers, etc.). To address this issue, the training data is randomly divided into 5 parts (stratified) and the XGBoost classifier is trained five times, using 4 of the 5 parts each time. This process was repeated 100 times, resulting in 500 different models. In each training, a random number of pre-selected features (top k) ranging from 30 to 200 features are used. Finally, for each feature, we obtain a series of Gini indices. The median Gini index of each feature is used to rank the features.

[0432] Importance of Classification Staining

[0433] To determine which coloring contributes most to the prediction, we used recursive channel elimination. In each run, BF was kept color-free. Then, Explainable Features + XGBoost was trained based on the features of the selected channels.

[0434] Category frequency analysis

[0435] For each donor, we first used the trained XGBoost classifier to predict the class of each image. We then excluded images using this data cleaning scheme:

[0436] 1. Use "Average Live-Dead Intensity" >= "Average Live-Dead Intensity (90th Percentile)" to filter out images containing dead cells (using Live-Dead staining)

[0437] 2. Filter out dead cells (using live-dead staining) using "Average Live-Dead Intensity" > "Average Live-Dead Intensity (90th percentile)"

[0438] 3. Filter out unfocused images using these conditions: “Gradient RMS BF” > “Gradient RMS BF (2nd percentile)” and “Gradient RMS BF” < “Gradient RMS BF (90th percentile)”

[0439] 4. Filter out images based on high entropy using XGBoost predictions (entropy > 1.0). Entropy is calculated using the SciPy package. This step is performed to ignore images that the classifier is most uncertain about in its predictions.

[0440] 5. Filter out images predicted as “B-LCL” and with “MHCII mean intensity” < “MHCII mean intensity (5th percentile)”. This step ensures that images predicted as “B-LCL” contain the minimum MHCII intensity.

[0441] 6. Filter out images predicted as “B-LCL” and with “MHCII area” < “MHCII area (10th percentile)”. This step ensures that images predicted as “B-LCL” contain cells of appropriate size and reduce artifacts.

[0442] 7. Filter out images predicted as “T cells” with “CD3 mean intensity” < “CD3 mean intensity (1st percentile)”. This step ensures that images predicted as “T cells” contain the minimum CD3 intensity.

[0443] 8. Filter out images predicted as "B-LCL and T cells in one layer" and "MHCII area" < "MHCII area (20th percentile)". Perform this step to omit "B-LCL and T cells in one layer" with smaller "B-LCL".

[0444] 9. Filter out images based on Isolation Forest Outlier Detection. It uses n_estimator = 100, max_samples = "auto", contamination = "auto" and max_features = 20 as main parameters. To reduce running time, only the first 30 features based on the Gini index from XGBoost training are used.

[0445] 10. Filter out images based on Uniform Manifold Approximation and Projection (UMAP).

[0446] First, using UMAP, all images are z-transformed into 2D space. The features are standardized using the mean and standard deviation of each feature. In order to reduce the running time, only the first 30 features based on the Gini index from XGBoost training are used. Then the DBSCAN algorithm is run with eps=0.09 and minimum_sample=5. If (number of images in the cluster) / (total number of images)<0.0001, the clustered results are filtered out.

[0447] All these steps are completed based on the scikit-learn implementation plan. Unless otherwise stated, all parameters are set using the default values ​​of scikit-learn. After data cleaning, for each condition of each donor, the frequency of each category is calculated using "F_C = (number of images predicted to be C) / (total number of images)". In order to deal with the compositional nature of the data, log_2 (F_C_antibody / F_C_control) is used to compare the frequency change fold. The advantage of this transformation is that the sum of frequencies is not equal to a constant value. After calculating the log_2 change fold, the Wilcoxon rank sum test is used to analyze the effect of antibodies on category frequencies. The Wilcoxon rank sum test whether two samples may be derived from the same population. In order to consider multiple testing, Benjamini-Hochberg corrections were used for +SEA / -SEA, CD19-TCB / control TCB, and Teplizumab / isotype, respectively. Since the experiments were performed independently, only each comparison was corrected individually.

[0448] Feature difference analysis

[0449] The effects of perturbations of signaling synapses by the presence of CD19-TCB and teplizumab, respectively, were analyzed.

[0450] First, images predicted to be “signaling synapses” were selected. Bright field (BF) features were removed because the intensity of BF does not contain biological meaning. In addition, the morphological features of BF were already captured based on the F-actin mask. Therefore, this information is redundant. Without being bound by the theory, it is assumed that this feature reduction is also necessary because it reduces the number of tests and increases the chances of finding a meaningful p-value after multiple testing correction.

[0451] This procedure generated 210 features to compare SEA and TCB based on F-actin, MHCII, CD3, and P-CD3ζ.

[0452] For Teplizumab, the feature reduction was even greater. Without being bound by this theory, this reduction was required due to the use of CD4 in recording Teplizumab images instead of CD3 for CD19TCB. Therefore, a meaningful CD3-based comparison between Teplizumab and its control was not feasible. Therefore, 132 features extracted from F-actin, MHCII, and P-CD3ζ were analyzed.

[0453] After feature selection, the Mann-Whitney U test was used to compare features of each condition and its control.

[0454] To understand the direction of change, the difference in the median value of the characteristic for each condition and its control was used. To account for multiple testing, the Benjamini-Hochberg procedure with α = 0.05 was used. Since the conditions were independent, the p-values ​​for each condition and its control were corrected separately.

[0455] Granzyme B prediction and feature ranking

[0456] To predict granzyme B, images predicted as “synapses without signaling” and “synapses with signaling” for each condition were used. Without being bound by the theory, it was assumed that synapses would lead to cytokine production. Considering that for each donor and condition, thousands of images were available. An aggregation pipeline was used to create a feature vector corresponding to each donor and condition. To reduce the number of features, only consistent feature changes for CD19-TCB were used ( Figure 3 ). For each donor and condition, features were aggregated using the 5th, 50th, and 95th percentiles to capture both extreme values ​​and mean values ​​for each feature.

[0457] After deriving the aggregated features, a linear regression model was trained with LassoLars using leave-one-donor cross-validation. The most important features were based on the magnitude of the coefficients.

[0458] Visualization

[0459] For plotting and images, matplotlib (version = 3.3.2) and seaborn (0.11.2) in Python were used.

[0460] References

[0461] 1. Huppa, JB, and Davis, MM, Nat. Rev. Immunol. 3, 973–983 (2003).

[0462] 2. Dustin, ML, Cancer Immunol. Res. 2, 1023–1033 (2014).

[0463] 3. Kearney, CJ et al., Crit. Rev. Immunol. 35, 325–347 (2015).

[0464] 4. Kallikourdis, M. et al., Front. Immunol. 6, 433 (2015).

[0465] 5. Laura, RD et al., Jci Insight 3, e120757 (2018).

[0466] 6. Schubert, DA et al., J. Exp. Med. 209, 335–352 (2012).

[0467] 7. Calvez, R. et al., Haematologica 96, 1415–1423 (2011).

[0468] 8. Alan, GR, et al., J. Clin. Invest. 118, 2427–2437 (2008).

[0469] 9. Francesca, F. and Baldari, CT, Pharmacol. Res. 134, 118–133 (2018).

[0470] 10. Tai, Y. et al., Front. Pharmacol. 9, 642 (2018).

[0471] 11. Scott, M., S.-T. et al., Ann. NY Acad. Sci. 1183, 123–148 (2010).

[0472] 12. Wetzel, SA et al., J. Immunol. 169, 6092–6101 (2002).

[0473] 13. Cremasco, F. et al., PLoS One 16, e0241091 (2021).

[0474] 14. Amita, J. et al., J. Clin. Pharmacol. 46, 10–20 (2006).

[0475] 15. Lee, J. Y., et al., Nat. Commun. 7, 13354 (2016).

[0476] 16. Dickopf, S. et al., Comput. Struct. Biotechnol. J. 18, 1221–1227 (2020).

[0477] 17. Cartwright, ANR, et al., Nat. Commun. 5, 5479 (2014).

[0478] 18. Vyver, AJVD et al., Mol. Cancer. Ther. 20, 357–366 (2021).

[0479] 19. Wabnitz, GH, et al., J. Immunol. Meth. 423, 29–39 (2015).

[0480] 20. Wabnitz, GH, et al., Eur. J. Immunol. 41, 3157–3169 (2011).

[0481] 21. Ahmed, F. et al., J. Immunol. Meth. 347, 79–86 (2009).

[0482] 22. Lippeveld, M. et al., Cytom. Part A 97, 308–319 (2020).

[0483] 23. Eulenberg, P. et al., Nat. Commun. 8, 463 (2017).

[0484] 24. Kranich, J. et al., J. Extracell. Vesicles 9, 1792683 (2020).

[0485] 25. Chlis, N.-K. et al., Nucleic Acids Res. 48, gkaa926-(2020).

[0486] 26. Doan, M. et al., Nat. Protoc. 16, 3572–3595 (2021).

[0487] 27. Hennig, H. et al., Methods, San Diego Calif 112, 201–210 (2017).

[0488] 28. Eulenberg, P. et al., Biolxiv 081364 (2017) doi:10.1101 / 081364.

[0489] 29.Leclercq, G. et al., J. Immunother. Cancer 10, e003766 (2022).

[0490] 30. Xu, D. et al., Cell Immunol. 200, 16–26 (2000).

[0491] 31. Herold, KC et al., J. Clin. Invest. 111, 409–418 (2003).

[0492] 32. Cremasco, F. et al., Plos One 16, e0241091 (2021).

[0493] 33. Bacac, M. et al., Clin. Cancer Res. 22, 3286–3297 (2016).

[0494] 35. Wabnitz, G. et al., J. Vis. Exp. (2019) doi:10.3791 / 55345.

[0495] 36. Carpenter, AE et al., Genome Biol. 7, R100–R100 (2006).

[0496] 37. German, Y. et al., Cell Reports 36, 109318 (2021).

[0497] 38. Perakis, A. et al., Arxiv (2021) doi:10.1007 / 978-3-030-87589-3_58.

[0498] 39. Naghizadeh, A. et al., Plos Comput. Biol. 18, e1009883 (2022).

[0499] 40. Xiong, W. et al., Mol. Ther. 26, 963–975 (2018).

[0500] 41. Walt, S. van der, et al., Peerj 2, e453 (2014).

[0501] 42. Harris, CR et al., Nature 585, 357–362 (2020).

[0502] 43. Virtanen, P. et al., Nat. Methods 17, 261–272 (2020).

[0503] 44. Adler, J. and Parmryd, I., Cytom. Part A 77A, 733–742 (2010).

[0504] 45. Haralick, RM et al., Ieee Transactions Syst. Man. Cybern. SMC-3, 610–621 (1973).

[0505] 46. ​​Buitinck, L. et al. Arxiv (2013).

[0506] 47.Haq, AU et al., Ieee Access 7, 151482–151492 (2019).

[0507] 48. Krishnapuram, B. et al., Proc. 22nd Acm. Sigkdd Int. Conf. Knowl. Discov. Data Min. 785–794 (2016) doi:10.1145 / 2939672.2939785.

Claims

1. A method for classifying cells in a cell mixture, wherein the mixture comprises T cells and activated B cells or antigen presenting cells, the method comprising the following steps: a) applying labeled antibodies that bind to at least F-actin, MHCII and CD3 to the cell mixture to obtain a labeled cell mixture, wherein the antibodies are each labeled with a dye, wherein the dyes have different emission wavelengths, b) acquiring at least one image of the cell mixture, and c) classifying the cells in the cell mixture into i) the isolated cells, if the cells are single cells, are positive for F-actin, and - MHCII positive and CD3 negative, or - MHCII negative and CD3 positive, ii) Doublets or multiplets of cells, if the cells are aggregates of two or three or more cells, are F-actin positive, MHCII positive and CD3 positive.

2. The method according to claim 1, wherein Step a) is: applying labeled antibodies that bind to at least F-actin, MHCII, CD3 and P-CD3ζ to the cell mixture, wherein each of the antibodies is labeled with a dye, wherein the dye has a different emission wavelength, And step c) is: classifying the cells in the cell mixture into i) a single B cell or antigen presenting cell, if the cell is F-actin positive, MHCII positive, CD3 negative and P-CD3ζ negative, ii) a single T cell without signaling, if the cell is F-actin positive, MHCII negative, CD3 positive and P-CD3ζ negative, iii) a single T cell with signaling, if the cell is F-actin positive, MHCII negative, CD3 positive and P-CD3ζ positive, iv) a doublet of a B cell or antigen presenting cell and a T cell that forms a synapse without signaling, if the doublet is positive for F-actin, positive for MHCII, positive for CD3 and negative for P-CD3ζ, v) a doublet or multiplet of one or two B cells or antigen presenting cells and one T cell that forms a signaling synapse, if the doublet or multiplet is positive for F-actin, MHCII, CD3 and P-CD3ζ.

3. The method according to any one of claims 1 to 2, wherein Step b) is: acquiring an image of the cell mixture using an imaging flow cytometer. 4 . The method according to claim 1 , wherein the acquired images are images each showing a single cell or isolated doublets or multiplets.

5. The method according to any one of claims 1 to 4, wherein the mixture comprises B cells or antigen presenting cells and T cells at a cell ratio of about 4:

3.

6. The method according to any one of claims 1 to 5, wherein the T cells are CD4-positive memory T cells.

7. The method according to any one of claims 1 to 6, wherein the cell mixture is centrifuged after mixing.

8. Use of F-actin, MHCII and CD3 for sorting cells in a mixture comprising T cells and B cells or activated B cells or antigen presenting cells, wherein the classification is that the cell is an isolated cell, if the cell is a single cell, is F-actin positive, and - MHCII positive and CD3 negative, or - MHCII negative and CD3 positive, or The classification is that the cells are doublets or multiplets of cells, if the cells are aggregates of two or three cells, are F-actin positive, MHCII positive and CD3 positive.

Images