Scoring method for anti-HER2 antibody-drug conjugate therapy

Abstract

A method for predicting how a cancer patient will respond to antibody-drug conjugate (ADC) treatment includes calculating a predicted response score based on the single-cell ADC scores for each cancer cell. The ADC includes an ADC payload and an ADC antibody that targets a protein on each cancer cell, the protein being human epidermal growth factor receptor 2 (HER2). A tissue sample is immunohistochemically stained using a dye linked to a diagnostic antibody that binds to a protein on cancer cells in the tissue sample. Cancer cells are detected in a digital image of the tissue. For each cancer cell, a single-cell ADC score is calculated based on the staining intensity of the dye in the membrane and / or cytoplasm of the cancer cell and / or the membrane and cytoplasm of adjacent cancer cells. The cancer patient's response to ADC treatment is predicted by aggregating all single-cell ADC scores of the tissue sample using statistical operations.

Description

[Technical Field]

[0001] Cross-reference of related applications This application claims priority to U.S. Provisional Patent Application No. 63 / 077,604, filed on 12 September 2020, which is incorporated herein by reference in its entirety.

[0002] Reference to electronically submitted sequence listings The contents of the electronically submitted sequence listing (name: DSADC_400_Seqlisting.txt; size: 24,662 bytes; and creation date: September 9, 2021) are incorporated herein by reference in their entirety.

[0003] The present invention relates to a method for calculating a score indicating how cancer patients respond to therapy using an antibody-drug conjugate having a drug conjugated to an anti-HER2 antibody via a linker structure. [Background technology]

[0004] Assessing the probability of a cancer patient responding to a given treatment is an essential step in determining the treatment regimen for that patient. Such assessments are often based on histological analysis of tissue samples from cancer patients and involve identifying and classifying cancers using standard grading schemes. Immunohistochemical (IHC) staining can be used to distinguish marker-positive cells that express a particular protein from marker-negative cells that do not express that protein. IHC staining typically involves multiple dyes, including one or more dyes linked to protein-specific antibodies and another dye that acts as a counterstain. A common counterstain is hematoxylin, which labels DNA and therefore stains the nucleus.

[0005] Using protein-specific stains or biomarkers, areas of cancer patient tissue that are likely to respond to a given treatment can be identified. For example, biomarkers that stain epithelial cells may help identify suspected tumor areas. Then, other protein-specific biomarkers are used to characterize cells within the cancerous tissue. Cells stained by specific biomarkers can be identified and quantified, and then a pathologist can visually estimate a score indicating the number of positively and negatively stained cells. This score can then be compared to scores calculated in the same way for other cancer patients. If the responses of these other patients to a given cancer treatment are known, a pathologist can predict how likely a cancer patient is to respond to a given treatment based on a comparison of the score calculated for the cancer patient with the scores of other patients. However, visual assessments by pathologists are prone to variability and subjectivity.

[0006] One promising cancer treatment involves antibody-drug conjugates (ADCs) containing a cytotoxic drug conjugated to an antibody on which an antigen is expressed on the surface of cancer cells. The ADC binds to the antigen, undergoes intracellular integration, and selectively delivers the drug to cancer cells, where it accumulates and kills them. There is a need for computer-based methods to generate reproducible objective scores indicating cancer patients' responses to therapies containing therapeutic HER2 antibody-drug conjugates. [Overview of the Initiative] [Means for solving the problem]

[0007] A method for predicting how cancer patients will respond to antibody-drug conjugate (ADC) therapy involves calculating a response score based on a single-cell ADC score for each cancer cell. An ADC comprises an ADC payload and an ADC antibody that targets a protein on each cancer cell. Tissue samples are immunohistochemically stained using a dye linked to a diagnostic antibody that binds to the protein on the cancer cells in the tissue sample. Digital images of the tissue samples are acquired. Image analysis is performed on the digital images, and cancer cells are detected using a convolutional neural network. For each cancer cell, a single-cell ADC score is calculated based on the staining intensity of the dye in the membrane and / or cytoplasm of the cancer cell and / or in the membrane and cytoplasm of other cancer cells closer to the cancer cell than a predetermined distance. By aggregating all single-cell ADC scores from the tissue samples using statistical calculations, a response score is generated that predicts the cancer patient's response to ADC therapy. Patients with a response score above a predetermined threshold are recommended for ADC therapy.

[0008] In one embodiment, a method for generating a survival score for cancer patients treated with an antibody-drug conjugate (ADC) includes calculating a single-cell ADC score. A tissue sample from a cancer patient is immunohistochemically stained using a dye linked to a diagnostic antibody. The ADC comprises an ADC payload and an ADC antibody that targets the human epidermal growth factor receptor 2 (HER2) protein on cancer cells. The diagnostic antibody binds to the HER2 protein on cancer cells in the tissue sample. A digital image of the tissue sample is obtained, and cancer cells in the digital image are detected using image analysis. For each cancer cell, a single-cell ADC score is calculated based on the staining intensity of the dye in the membrane. The single-cell ADC score may also optionally be based on the staining intensity of the dye in the cytoplasm of the cancer cell, as well as the staining intensity of the dye in the membrane and cytoplasm of other cancer cells closer to the cancer cell than a predetermined distance. The resulting quantitative continuous score (QCS) indicating the survival probability of the cancer patient is generated by aggregating all single-cell ADC scores from the tissue sample using statistical calculations. The aggregation of all single-cell ADC scores is performed by determining the mean, determining the median, or determining quantiles at predetermined proportions. In another embodiment, the set of all single-cell ADC scores includes a thresholding operation using a predefined threshold. All cells with a single-cell ADC score greater than a given threshold are labeled as single-cell ADC positive. The set is performed by dividing the number of single-cell ADC positive cells by the total number of cancer cells.

[0009] In one embodiment, a method for predicting a cancer patient's response to an antibody-drug conjugate (ADC) includes detecting cancer cells and calculating a single-cell ADC score for each cancer cell. The ADC comprises an ADC payload and an ADC antibody that targets a protein on the cancer cell. The protein is human epidermal growth factor receptor 2 (HER2). A tissue sample is immunohistochemically stained using a dye linked to a diagnostic antibody. The diagnostic antibody binds to the protein on the cancer cells in the tissue sample. A digital image of the tissue sample is acquired, and cancer cells are detected in the digital image. For each cancer cell, a single-cell ADC score is calculated based on the staining intensity of the dye in the membrane. The single-cell ADC score may also be based on the staining intensity of the dye in the cytoplasm of the cancer cell, and may also be based on the staining intensity of the dye in the membrane and cytoplasm of other cancer cells that are closer to the cancer cell than a predetermined distance. The staining intensity of each membrane is calculated based on the average optical density of the brown diaminobenzidine (DAB) signal in the membrane pixels, and the staining intensity of each cytoplasm is calculated based on the average optical density of the brown DAB signal in the cytoplasmic pixels. Using statistical calculations, the response of cancer patients to ADCs is predicted based on the set of all single-cell ADC scores from tissue samples.

[0010] In another embodiment, a method for identifying cancer patients who exhibit a predetermined response to an antibody-drug conjugate (ADC) includes generating a response score. The ADC comprises an ADC payload and an ADC antibody that targets a protein on cancer cells. Tissue samples from cancer patients are immunohistochemically stained using a dye linked to a diagnostic antibody that binds to a protein on cancer cells in the tissue sample. A digital image of the tissue sample is obtained, and cancer cells in the digital image are detected using a convolutional neural network. For each cancer cell, a single-cell ADC score is calculated based on the staining intensity of the dye in the membrane. The single-cell ADC score may also be based on the staining intensity of the dye in the cytoplasm of the cancer cell, and may also be based on the staining intensity of the dye in the membrane and cytoplasm of other cancer cells that are closer than a predetermined distance to the cancer cell from which the single-cell score is calculated. The QCS score is a response score generated by aggregating all single-cell ADC scores from the tissue sample using statistical calculations. Cancer patients are identified as exhibiting a predetermined response to the ADC based on whether their QCS score exceeds a threshold. Patients with a QCS score greater than a threshold are considered QCS positive (QCS+), and all other patients are considered QCS negative (QCS-). The given response is a reduction in mean tumor size. In another embodiment, the difference between the QCS score and the threshold indicates the probability that a cancer patient is correctly identified as having a given response to ADC. A small difference indicates a low probability, and a large difference indicates a high probability that the cancer patient is accurately identified. In another embodiment, the given response is indicated by the patient's survival, such that the patient's death within a predetermined period is considered a non-response. In one embodiment, the predetermined period is twice the mean survival time of patients treated according to standard care.

[0011] In one embodiment, a method for treating a cancer patient with an antibody-drug conjugate (ADC) includes generating a QCS score in the form of a treatment score. The ADC comprises an ADC payload and an ADC antibody that targets a protein on cancer cells. The protein is human epidermal growth factor receptor 2 (HER2). A tissue sample from the cancer patient is immunohistochemically stained using a dye linked to a diagnostic antibody that binds to the protein on cancer cells in the tissue sample. A digital image of the tissue sample is acquired, and cancer cells are detected in the digital image. For each cancer cell, a single-cell ADC score is calculated based on the staining intensity of the dye in the membrane. The single-cell ADC score may also optionally be based on the staining intensity of the dye in the cytoplasm of the cancer cell, as well as the staining intensity of the dye in the membrane and cytoplasm of other cancer cells that are closer to the cancer cell than a predetermined distance. The treatment score is generated by aggregating all single-cell ADC scores from the tissue sample using statistical calculations. If the treatment score exceeds a predetermined threshold, a treatment including the ADC is administered to the cancer patient. In another embodiment, a cancer patient is given treatment containing an ADC only if the patient's pre-treatment HER2-stained tissue sample is scored as QCS-positive (QCS+).

[0012] In another embodiment, the method of treating cancer patients with an antibody-drug conjugate (ADC) is carried out by a clinician administering the treatment. The ADC comprises an ADC payload and an ADC antibody that targets a protein on cancer cells. The protein is human epidermal growth factor receptor 2 (HER2). The ADC-containing treatment is administered to cancer patients when the response score exceeds a predetermined threshold. The QCS response score was generated by aggregating single-cell ADC scores from cancer patient tissue samples using statistical calculations. Each single-cell ADC score was calculated for each cancer cell based on the staining intensity of a dye in the membrane. The single-cell ADC score may also be based, optionally, on the staining intensity of a dye in the cytoplasm of each cancer cell, and also on the staining intensity of dyes in the membrane and cytoplasm of other cancer cells that are closer than a predetermined distance to each cancer cell from which the single-cell ADC score was calculated. Cancer cells were detected in digital images of cancer patient tissue samples. The tissue samples were immunohistochemically stained using a dye linked to a diagnostic antibody that binds to a protein on cancer cells in the tissue sample.

[0013] Other embodiments and advantages are described in the detailed description below. This summary is not intended to define the invention. The invention is defined by the claims.

[0014] The attached drawings illustrate embodiments of the present invention, where similar reference numerals indicate similar components. [Brief explanation of the drawing]

[0015] [Figure 1] This figure shows the amino acid sequence of the HER2 protein (SEQ ID NO: 1). [Figure 2] This figure shows the amino acid sequence of the heavy chain of the anti-HER2 antibody (SEQ ID NO: 2). [Figure 3] This figure shows the amino acid sequence of the light chain of the anti-HER2 antibody (SEQ ID NO: 3). [Figure 4] This figure shows the amino acid sequence (SEQ ID NO: 4) of the anti-HER2 antibody CDRH1. [Figure 5] This figure shows the amino acid sequence (SEQ ID NO: 5) of the anti-HER2 antibody CDRH2. [Figure 6] This figure shows the amino acid sequence (SEQ ID NO: 6) of the anti-HER2 antibody CDRH3. [Figure 7] This figure shows the amino acid sequence (SEQ ID NO: 7) of the anti-HER2 antibody CDRL1. [Figure 8] This figure shows the amino acid sequence (SEQ ID NO: 8) of the anti-HER2 antibody CDRL2(SAS). [Figure 9] This figure shows the amino acid sequence (SEQ ID NO: 9) of CDRL3 in the anti-HER2 antibody. [Figure 10] This figure shows the amino acid sequence of the heavy chain variable region of the anti-HER2 antibody (SEQ ID NO: 10). [Figure 11] This figure shows the amino acid sequence of the light chain variable region of the anti-HER2 antibody (SEQ ID NO: 11). [Figure 12] This figure shows the amino acid sequence of the heavy chain of the anti-HER2 antibody (SEQ ID NO: 12). [Figure 13] This figure shows trastuzumab deruxtecan, an anti-HER2 antibody-drug conjugate with eight drug-linker units. [Figure 14] This is a flowchart illustrating the process by which an analysis system analyzes digital images of tissue from cancer patients and predicts how likely the cancer patients are to respond to treatments containing anti-HER2 antibody-drug conjugates. [Figure 15] This figure shows a digital image illustrating the image analysis process in step 2 of Figure 14. [Figure 16] Figure 14 shows a digital image from another embodiment of the image analysis process in step 2. [Figure 17] This diagram shows the image analysis process for detecting nuclear objects in cancer cells. [Figure 18] This figure shows an image analysis process for detecting membranes using nuclear objects. [Figure 19] This diagram shows the image analysis process for detecting membrane objects in cancer cells. [Figure 20] This is a screenshot of the results of the image analysis process in an image analysis software environment. [Figure 21] This figure shows the script used for image analysis segmentation and the cell object measurements obtained using the script. [Figure 22] This figure shows the results of sample quantification of staining intensity from image analysis using grayscale values ​​of membrane and cytoplasmic pixels. [Figure 23] Figure 22 is a diagram listing exemplary quantitative staining levels in the membrane and cytoplasm of the image. [Figure 24] This diagram shows the mechanism by which anti-HER2 ADC therapy kills cancer cells. [Figure 25] This figure shows the calculation of single-cell ADC scores for each of the three cells shown in Figure 23, based on cell separation, to illustrate the uptake of the ADC payload into adjacent cells. [Figure 26] This figure shows the formula used to calculate the single-cell score. [Figure 27] This diagram illustrates how pathologists consistently compare and annotate membranes. [Figure 28] This figure compares how membranes detected using the image analysis method in Figure 14 correlate with membranes identified by pathologists. [Figure 29] This figure shows whether each of the 50 patients with HER2 IHC1+ and HER2 IHC2+ / ISH- scores had progressive disease, was in a stable state, or responded to anti-HER2 ADC treatment. [Figure 30] This figure shows the response to tumor reduction in patients classified as "HER2-positive" based on scores from the new scoring method. [Figure 31] This figure shows that a subgroup of patients from the "HER2-negative" category of IHC scoring still demonstrated a favorable objective response rate to anti-HER2 ADC therapy. [Figure 32] Figure 14 shows the stratification of patients from the conventional "HER2-negative" category to "QCS-positive" and "QCS-negative" categories using the new scoring method. [Figure 33]Figure 14 shows the stratification of 151 patients into "QCS positive" and "QCS negative" using the new scoring method. [Figure 34] Figure 34A is a bar graph showing the novel response scores for breast cancer patients, with the conventional HER2 IHC score indicated by a shaded bar. Figure 34B is a table showing the objective response rate (ORR) and median progression-free survival (mPFS) for QCS-positive and QCS-negative patients stratified in Figure 34A. [Figure 35] This is a graph of Kaplan-Meier curves for progression-free survival in two groups of HER2-negative patients identified using a novel embodiment of the method that also considers single-cell ADC score-negative cancer cells adjacent to single-cell ADC score-positive cancer cells. [Figure 36] This graph shows the Kaplan-Meier curves for progression-free survival for the two patient groups across the entire J101 trial, identified using the novel QCS scoring method. [Figure 37] Figure 36 is a table of features used in a novel QCS scoring method to stratify two groups of patients shown in the Kaplan-Meier curve. [Figure 38] This is a table of tumor-infiltrating lymphocyte (TIL)-based and HER2-stained characteristics of a model used to stratify 151 breast cancer patients into those exhibiting longer and shorter progression-free survival (PFS) in response to anti-HER2 ADC therapy. [Figure 39] Figure 39A shows the Kaplan-Meier curves for the TIL density features listed in Figure 38. Figure 39B shows the Kaplan-Meier curves for the HER2+ cell density features listed in Figure 38. Figure 39C shows the Kaplan-Meier curves for the HER2+ neighborhood score features listed in Figure 38. [Figure 40]Figure 40A shows the Kaplan-Meier curves for the TIL density features listed in Figure 38 for exactly 72 patients designated as HER2-positive out of 151 patients in the J101 trial. Figure 40B shows the Kaplan-Meier curves for the HER2+ cell density features for exactly 72 HER2-positive patients out of a total of 151 patients in the J101 trial. Figure 40C shows the Kaplan-Meier curves for the HER2+ neighbor score features for exactly 72 HER2-positive patients out of a total of 151 patients in the J101 trial. [Figure 41] Figure 41A shows the Kaplan-Meier curves for TIL density features for exactly 65 HER2-negative patients out of a total of 151 patients in the J101 trial. Figure 41B shows the Kaplan-Meier curves for HER2+ cell density features for exactly 65 HER2-negative patients out of a total of 151 patients in the J101 trial. Figure 41C shows the Kaplan-Meier curves for HER2+ neighbor score features for exactly 65 HER2-negative patients out of a total of 151 patients in the J101 trial. [Figure 42] This bar graph shows a novel response score for gastric cancer patients, with the conventional HER2 IHC score indicated by the shaded bars. [Figure 43] This is a table of HER2 stain-based features of a model used to stratify gastric cancer patients into those exhibiting longer and shorter progression-free survival (PFS) in response to anti-HER2 ADC therapy. [Figure 44] This is a table of HER2 stain-based features of a model used to stratify gastric cancer patients into those exhibiting longer and shorter overall survival (OS) in response to anti-HER2 ADC therapy. [Figure 45] Figure 43 shows the Kaplan-Meier curves for the feature membOD_density_10, based on PFS in 32 gastric cancer patients in the J101 trial. [Figure 46] Figure 44 shows the Kaplan-Meier curve for the feature membOD_density_10, based on the overall survival (OS) of 32 gastric cancer patients in the J101 trial. [Modes for carrying out the invention]

[0016] The present invention provides a novel method for calculating a score indicating how cancer patients respond to treatment using anti-HER2 antibody-drug conjugates. Another aspect of the present invention relates to a method for calculating a cancer patient score indicating the survival probability of cancer patients treated with ADCs. Another aspect of the present invention relates to a method for predicting a cancer patient's response to ADCs. Another aspect of the present invention relates to identifying cancer patients who show a predetermined response to ADCs. Yet another aspect of the present invention relates to a method for treating cancer patients by administering a treatment containing ADCs if the treatment score exceeds a predetermined threshold.

[0017] I. Definition To facilitate understanding of this invention, numerous terms and expressions are defined below. The term "cancer" is used to have the same meaning as the term "tumor."

[0018] In this invention, "HER2" is synonymous with human epidermal growth factor receptor 2 (which may also be referred to as neu or ErbB-2), and is a transmembrane receptor belonging to the epidermal growth factor receptor (EGFR) subfamily of receptor protein tyrosine kinases, along with HER1 (EGFR or ErbB-1), HER3 (ErbB-3), and HER4 (ErbB-4). HER2 is known to play an important role in cell proliferation, differentiation, and survival in normal and tumor cells by being activated by autophosphorylation of intercellular tyrosine residues through heterodimerization with HER1, HER3, or HER4. In this invention, "HER2 protein" is used interchangeably with HER2. HER2 protein expression can be detected using methods well known to those skilled in the art, such as immunohistochemistry (IHC) or immunofluorescence (IF).

[0019] Figure 1 shows the amino acid sequence of the HER2 protein (SEQ ID NO: 1). In SEQ ID NO: 1, the amino acid sequence consisting of amino acid residues 1-652 is called the "extracellular domain of the HER2 protein," the amino acid sequence consisting of amino acid residues 653-675 is called the "transmembrane domain of the HER2 protein," and the amino acid sequence consisting of amino acid residues 676-1255 is called the "intercellular domain of the HER2 protein."

[0020] In this invention, "anti-HER2 antibody" means an antibody that specifically binds to HER2. The anti-HER2 antibody has the activity to bind to HER2, thereby being internalized into HER2-expressing cells, and after exhibiting HER2-binding activity, the antibody moves into the HER2-expressing cells. The anti-HER2 antibody can target tumor cells, bind to tumor cells, be internalized within the tumor cells, exhibit cytotoxic activity against tumor cells, and conjugate with an antitumor-active drug via a linker to form an antibody-drug conjugate.

[0021] Figure 2 shows the amino acid sequence of the heavy chain of the anti-HER2 antibody (SEQ ID NO: 2), and Figure 3 shows the amino acid sequence of the light chain of the anti-HER2 antibody (SEQ ID NO: 3).

[0022] Figure 4 shows the amino acid sequence of CDRH1 of the anti-HER2 antibody (SEQ ID NO: 4), Figure 5 shows the amino acid sequence of CDRH2 of the anti-HER2 antibody (SEQ ID NO: 5), and Figure 6 shows the amino acid sequence of CDRH3 of the anti-HER2 antibody (SEQ ID NO: 6).

[0023] Figure 7 shows the amino acid sequence of CDRL1 of the anti-HER2 antibody (SEQ ID NO: 7), Figure 8 shows the amino acid sequence of CDRL2(SAS) of the anti-HER2 antibody (SEQ ID NO: 8), and Figure 9 shows the amino acid sequence of CDRL3 of the anti-HER2 antibody (SEQ ID NO: 9).

[0024] Figure 10 shows the amino acid sequence of the heavy chain variable region of the anti-HER2 antibody (SEQ ID NO: 10), and Figure 11 shows the amino acid sequence of the light chain variable region of the anti-HER2 antibody (SEQ ID NO: 11). Figure 12 shows the amino acid sequence of another heavy chain of the anti-HER2 antibody (SEQ ID NO: 12).

[0025] The anti-HER2 antibody in the anti-HER2 antibody-drug conjugate used in the present invention preferably comprises a heavy chain containing CDRH1 (amino acid sequence consisting of amino acid residues 26-33 of SEQ ID NO: 4), CDRH2 (amino acid sequence consisting of amino acid residues 51-58 of SEQ ID NO: 2), and CDRH3 (amino acid sequence consisting of amino acid residues 97-109 of SEQ ID NO: 2), and CDRL1 (amino acid sequence consisting of amino acid residues 27-32 of SEQ ID NO: 3), and CDRL2 (amino acid sequence consisting of amino acid residues 50-52 of SEQ ID NO: 3), which comprises the amino acid sequence represented by SEQ ID NO: 7, and SEQ ID NO: 9 An antibody comprising a light chain containing CDRL3 (an amino acid sequence consisting of amino acid residues 89-97 of SEQ ID NO: 3), more preferably an antibody comprising a heavy chain variable region consisting of the amino acid sequence represented by SEQ ID NO: 10 (an amino acid sequence consisting of amino acid residues 1-120 of SEQ ID NO: 2) and a light chain variable region consisting of the amino acid sequence represented by SEQ ID NO: 11 (an amino acid sequence consisting of amino acid residues 1-107 of SEQ ID NO: 3), even more preferably an antibody comprising a heavy chain consisting of the amino acid sequence represented by SEQ ID NO: 2 and a light chain consisting of the amino acid sequence represented by SEQ ID NO: 3, or an antibody comprising a heavy chain consisting of the amino acid sequence represented by SEQ ID NO: 12 (an amino acid sequence consisting of amino acid residues 1-449 of SEQ ID NO: 2) and a light chain consisting of the amino acid sequence represented by SEQ ID NO: 3.

[0026] In the present invention, the terms "HER2-positive" and "HER2 overexpression" refer to cancer tissue that has been given a score of 3+ for HER2 expression by immunohistochemical method, and cancer tissue that has been given a score of 2+ for HER2 expression by immunohistochemical method and is determined to be positive for HER2 expression by in situ hybridization method. Examples of in situ hybridization methods in the present invention include fluorescence in situ hybridization (FISH) and dual-color in situ hybridization (DISH).

[0027] In the present invention, "HER2-negative" refers to cancer tissue that is given a score of 0 for HER2 expression by immunohistochemical method, cancer that is given a score of 1+ for HER2 expression by immunohistochemical method, and cancer tissue that is given a score of 2+ for HER2 expression by immunohistochemical method and is determined to be negative for HER2 expression by in situ hybridization method.

[0028] In this invention, "HER2-low" refers to cancer tissue that was given a score of 0+ (mean >0 and <1+) for HER2 expression by immunohistochemical method, cancer that was given a score of 1+ for HER2 expression by immunohistochemical method, and cancer tissue that was given a score of 2+ for HER2 expression by immunohistochemical method and was determined to have negative HER2 expression by in situ hybridization method. Tissues classified as 0+ show very weak but noiseless HER2 expression.

[0029] In this invention, the term "QCS-positive" (QCS+) refers to cancer that is likely to respond to anti-HER2 ADC therapy. The term "QCS-negative" (QCS-) refers to cancer that is unlikely to respond to anti-HER2 ADC therapy. The acronym QCS stands for Quantitative Continuous Score. The results of the novel prediction method of this invention are generally called quantitative continuous scores and may be response scores, treatment scores, or indicators of predicted survival. The QCS score is obtained by performing statistical calculations on all single-cell ADC scores obtained for a patient. By applying a predetermined threshold to the QCS score, patients are distinguished as "QCS-positive" and "QCS-negative". By stratifying cancer patients into QCS-positive and QCS-negative groups, it becomes possible to identify QCS+ patients who are likely to benefit from ADC-containing therapy.

[0030] II. Anti-HER2 antibody-drug conjugates. In the present invention, the substructure consisting of a linker and a drug of an anti-HER2 antibody-drug conjugate is called a "drug-linker". The drug-linker can be conjugated to an anti-HER2 antibody via a thioether bond. Therefore, the drug-linker can be bound to thiol groups (sulfur atoms of cysteine ​​residues) formed at interchain disulfide bond sites of the antibody (two sites between heavy chains and two sites between heavy and light chains). In the present invention, it is preferable that the anti-HER2 antibody-drug conjugate used is an anti-HER2 antibody-drug conjugate represented by the following formula: [Formula 1] [ka] In the formula, A represents the connection site with the anti-HER2 antibody.

[0031] The drug linker of the present invention contains exatecan (also represented as IUPAC name: (1S,9S)-1-amino-9-ethyl-5-fluoro-1,2,3,9,12,15-hexahydro-9-hydroxy-4-methyl-10H,13H-benzo[de]pyrano[3',4':6,7]indolidino[1,2-b]quinoline-10,13-dione (chemical name: (1S,9S)-1-amino-9-ethyl-5-fluoro-2,3-dihydro-9-hydroxy-4-methyl-1H,12H-benzo[de]pyrano[3',4':6,7]indolidino[1,2-b]quinoline-10,13(9H,15H)-dione), which is a topoisomerase I inhibitor. Exatecan is the cytotoxic payload of the antibody-drug conjugate and has an antitumor effect. Exatecan is a camptothecin derivative represented by the following formula. [Formula 2] [ka]

[0032] The anti-HER2 antibody-drug conjugate used in the present invention is preferably trastuzumab deruxtecan, also known as DS-8201. The anti-HER2 antibody-drug conjugate used in the present invention may be represented by the following formula. [Formula 3] [ka]

[0033] The drug-linkers shown in square brackets are conjugated to the anti-HER2 antibody via thioether linkage. The meaning of 'n' is the same as the so-called average number of conjugated drug molecules (drug-to-conjugate ratio, DAR), indicating the average number of units of conjugated drug-linkers per antibody molecule.

[0034] The average number of drug-linker conjugate units per antibody molecule in the anti-HER2 antibody-drug conjugate used in the present invention is preferably 2 to 8, more preferably 3 to 8, even more preferably 7 to 8, and still more preferably about 8.

[0035] Figure 13 shows trastuzumab deruxtecan (DS-8201), an anti-HER2 antibody-drug conjugate with eight drug-linker units designated as "DL". The trastuzumab portion of trastuzumab deruxtecan shown in Figure 13 is a humanized anti-HER2 IgG1 mAb, where IgG1 indicates the isotype of the anti-HER2 antibody.

[0036] The anti-HER2 antibody-drug conjugate used in this invention, after being internalized into cancer cells, is cleaved at the linker portion and releases a compound represented by the following formula. [Formula 4] [ka]

[0037] The compounds shown above are the main source of antitumor activity for the anti-HER2 antibody-drug conjugate used in the present invention and have topoisomerase I inhibitory activity.

[0038] The anti-HER2 antibody-drug conjugate used in the present invention also has a bystander effect, in which the anti-HER2 antibody-drug conjugate is internalized in cancer cells expressing the target protein HER2, and subsequently exerts an antitumor effect on adjacent cancer cells that do not express the target protein HER2.

[0039] III. Production of anti-HER2 antibodies. The anti-HER2 antibody-drug conjugate used in the present invention can be manufactured as disclosed in International Publication No. 2015 / 115091.

[0040] The anti-HER2 antibody used in this invention can be obtained by immunizing an animal with any polypeptide selected from HER2 as an antigen or as the amino acid sequence of HER2, recovering the antibody produced in vivo, and then purifying the antibody. The origin of this antigen is not limited to humans; animals can be immunized with antigens derived from non-human animals such as mice, rats, and similar species. In this case, the cross-reactivity of the obtained antibody that binds to the heterologous antigen with the human antigen can be tested to screen for anti-HER2 antibodies applicable to human diseases.

[0041] Alternatively, antibody-producing cells that produce antibodies against an antigen can be fused with myeloma cells to establish a hybridoma, from which monoclonal antibodies can be obtained.

[0042] Antigens can be obtained by genetically engineering host cells to produce genes encoding antigen proteins. Specifically, a vector enabling the expression of the antigen gene is prepared and introduced into host cells to induce gene expression. The antigen thus expressed can then be purified. Antibodies can also be obtained by immunizing animals with the genetically engineered antigen-expressing cells or cell lines expressing the antigen.

[0043] The anti-HER2 antibody used in this invention is either a recombinant antibody artificially modified to reduce heteroantigenicity against humans, such as a chimeric antibody or a humanized antibody, or a human-derived antibody, i.e., an antibody having only the gene sequence of a human antibody. Examples of chimeric antibodies include antibodies derived from different species, such as a mouse-derived or rat-derived antibody in which the variable region is conjugated to a human-derived constant region.

[0044] Regarding humanized antibodies, the antibody may be obtained by incorporating only the complementarity-determining region (CDR) of a heterologous antibody into a human-derived antibody, or by transplanting some of the amino acid residues of the heterologous antibody framework and the CDR sequence of the heterologous antibody into a human antibody using the CDR grafting method (International Publication No. 90 / 07861), or by humanizing the antibody using a gene conversion mutagenesis strategy (U.S. Patent No. 5821337).

[0045] For human antibodies, the antibodies are produced using human antibody-producing mice that possess human chromosome fragments containing the genes for the heavy and light chains of human antibodies. Alternatively, antibodies are obtained by phage display, and the antibodies are selected from a human antibody library.

[0046] In the present invention, modified variants of anti-HER2 antibodies may also be used. Modified variants refer to variants obtained by subjecting the antibody according to the present invention to chemical or biological modification. Examples of chemically modified variants include variants involving linking of a chemical moiety to an amino acid backbone, and variants involving linking of a chemical moiety to an N-linked or O-linked sugar chain. Examples of biologically modified variants include variants obtained by post-translational modification (e.g., N-linked or O-linked glycosylation, N-terminal or C-terminal processing, deamidation, aspartic acid isomerization, or methionine oxidation), and variants in which a methionine residue is added to the N-terminus by expression in prokaryotic host cells. Furthermore, antibodies labeled to be detectable or isolateable anti-HER2 antibodies or antigens used in the present invention, such as enzyme-labeled antibodies, fluorescently labeled antibodies, and affinity-labeled antibodies, are also included in the meaning of modified variants. Such modified variants of anti-HER2 antibodies used in the present invention are useful for improving antibody stability and blood retention, reducing their antigenicity, and for the detection or isolation of antibodies or antigens.

[0047] Furthermore, antibody-dependent cytotoxic activity can be enhanced by controlling the modification of glycans (glycosylation, defucosylation, etc.) linked to the anti-HER2 antibody used in the present invention. Techniques for regulating the modification of antibody glycans are disclosed in International Publication Nos. 99 / 54342, 00 / 61739, and 02 / 31140. However, the techniques are not limited to these. Antibodies with controlled glycan modification can also be used as the anti-HER2 antibody of the present invention.

[0048] It is known that lysine residues at the carboxyl terminus of the heavy chain of antibodies produced in cultured mammalian cells are deleted, and that two amino acid residues (glycine and lysine) at the carboxyl terminus of the heavy chain of antibodies produced in cultured mammalian cells are deleted, and a proline residue newly located at the carboxyl terminus is amidated. However, such deletions and modifications of the heavy chain do not affect the antigen-binding affinity and effector function (complement activation, antibody-dependent cytotoxicity, etc.) of the antibody. Therefore, the anti-HER2 antibodies used in the present invention include antibodies and functional fragments of antibodies that have undergone such modifications, as well as deletion mutants in which one or two amino acids are deleted at the carboxyl terminus of the heavy chain, and mutants obtained by amidation of deletion mutants (for example, heavy chains in which the proline residue at the carboxyl terminus is amidated). The types of deletion mutants having deletions at the carboxyl terminus of the heavy chain of the anti-HER2 antibody used in the present invention are not limited to the above mutants, as long as the antigen-binding affinity and effector function are preserved. The two heavy chains constituting the anti-HER2 antibody used in the present invention may be one selected from the group consisting of full-length heavy chains and the deletion mutants described above, or a combination of two selected from these. The ratio of the amounts of each deletion mutant may be influenced by the type of cultured mammalian cells and culture conditions used to produce the anti-HER2 antibody in the present invention. However, the present invention can also use antibodies in which one amino acid residue at the carboxyl terminus is deleted in both heavy chains of the anti-HER2 antibody.

[0049] IV. Manufacturing of anti-HER2 antibody-drug conjugates. The drug-linker intermediate that can be used in the production of the anti-HER2 antibody-drug conjugate of the present invention is represented by the following formula. [Formula 5] [ka]

[0050] The drug-linker intermediate can be represented by the chemical name N-[6-(2,5-dioxo-2,5-dihydro-1H-pyrrole-1-yl)hexanoyl]glycylglycyl-L-phenylalanyl-N-[(2-{[(1S,9S)-9-ethyl-5-fluoro-9-hydroxy-4-methyl-10,13-dioxo-2,3,9,10,13,15-hexahydro-1H,12H-benzo[de]pyrano[3',4':6,7]indolidino[1,2-b]quinoline-1-yl]amino}-2-oxoethoxy)methyl]glycinamide and can be manufactured in accordance with the disclosure in International Publication No. 2015 / 115091. The anti-HER2 antibody-drug conjugate used in the present invention can be produced by reacting the above-mentioned drug-linker intermediate with an anti-HER2 antibody having a thiol group (also known as a sulfhydryl group).

[0051] Anti-HER2 antibodies containing sulfhydryl groups can be obtained by methods well known to those skilled in the art. For example, by using a reducing agent such as tris(2-carboxyethyl)phosphine hydrochloride (TCEP) in an amount of 0.3 to 3 molar equivalents per interchain disulfide in the antibody and reacting it with an anti-HER2 antibody in a buffer containing a chelating agent such as ethylenediaminetetraacetic acid (EDTA), an anti-HER2 antibody containing sulfhydryl groups with partially or completely reduced interchain disulfides can be obtained.

[0052] Furthermore, by using 2 to 20 molar equivalents of a drug-linker intermediate per anti-HER2 antibody containing a sulfhydryl group, an anti-HER2 antibody-drug conjugate can be produced in which 2 to 8 drug molecules are conjugated per antibody molecule.

[0053] The average number of conjugated drug molecules per antibody molecule in the produced anti-HER2 antibody-drug conjugate can be determined, for example, by a method based on measuring the UV absorbance of the anti-HER2 antibody-drug conjugate and its conjugation precursor at two wavelengths, 280 nm and 370 nm (UV method), or by a method based on quantifying the fragments obtained by treating the antibody-drug conjugate with a reducing agent using HPLC measurement (HPLC method).

[0054] The conjugation of anti-HER2 antibodies with drug-linker intermediates, and the calculation of the average number of conjugated drug molecules per antibody molecule in anti-HER2 antibody-drug conjugates, can be performed in accordance with the disclosures in International Publication No. 2015 / 115091.

[0055] V. Use of anti-HER2 antibody-drug conjugates The antibody-drug conjugate of the present invention can delay the proliferation of cancer cells, suppress their growth, and even destroy them. These actions lead to the alleviation of cancer symptoms, improvement of the quality of life (QOL) of cancer patients, and maintenance of their lives, thus achieving therapeutic effects. Even if the destruction of cancer cells is not achieved, suppressing or controlling their proliferation can lead to longer survival and a higher QOL for cancer patients.

[0056] The antibody-drug conjugate of the present invention is expected to exert therapeutic effects through systemic treatment of patients and local application to cancerous tissue.

[0057] The antibody-drug conjugate may be administered as a pharmaceutical composition containing one or more pharmaceutically compatible components. The pharmaceutically compatible components can be appropriately selected from pharmaceutical excipients commonly used in the art, taking into consideration the dosage and concentration of the anti-HER2 antibody-drug conjugate. For example, the anti-HER2 antibody-drug conjugate used in the present invention may be administered as a pharmaceutical composition containing a buffer such as a histidine buffer, an excipient such as sucrose, and a surfactant such as polysorbate 80. The pharmaceutical composition containing the anti-HER2 antibody-drug conjugate used in the present invention can be used as an injectable preparation, either as an aqueous injectable preparation or a lyophilized injectable preparation, or further as a lyophilized injectable preparation.

[0058] If the pharmaceutical composition containing the anti-HER2 antibody-drug conjugate used in the present invention is an aqueous injection, it can be diluted with a suitable diluent and then administered by intravenous injection. Examples of diluents may include dextrose solution and physiological saline.

[0059] If the pharmaceutical composition containing the anti-HER2 antibody-drug conjugate used in the present invention is a lyophilized injectable preparation, it can be dissolved in injection-grade water, then diluted to the required volume with a suitable diluent, and then administered by intravenous injection. Examples of diluents include dextrose solution and physiological saline.

[0060] Examples of routes of administration that may be used to administer the pharmaceutical composition of the present invention may include intravenous, intradermal, subcutaneous, intramuscular, and intraperitoneal routes.

[0061] The anti-HER2 antibody-drug conjugate used in the present invention can be administered to humans at intervals of 1 to 180 days, at intervals of several weeks, and preferably at intervals of three weeks. The anti-HER2 antibody-drug conjugate used in the present invention can be administered in doses of approximately 0.001 to 100 mg / kg per dose, preferably at doses of 0.8 to 12.4 mg / kg per dose. The anti-HER2 antibody-drug conjugate can be administered at doses of preferably 0.8 mg / kg to 8 mg / kg once every three weeks, and more preferably at doses of 5.4 mg / kg to 6.4 mg / kg once every three weeks.

[0062] VI. Methods for predicting patient response to anti-HER2 antibody-drug conjugates. Figure 14 is a flowchart of steps 1-6 of Method 7, in which an analysis system analyzes digital images of tissue from cancer patients to predict how likely a cancer patient is to respond to treatment including an anti-HER2 antibody-drug conjugate (ADC). In one embodiment, the Method predicts the response to ADCs in patients having cancer selected from the group consisting of breast cancer, gastric cancer, colorectal cancer, lung cancer, esophageal cancer, head and neck cancer, gastroesophageal junction cancer, biliary tract cancer, Paget's disease, pancreatic cancer, ovarian cancer, uterine carcinosarcoma, bladder cancer, prostate cancer, urothelial carcinoma, gastrointestinal stromal tumor, cervical cancer, squamous cell carcinoma, peritoneal cancer, liver cancer, hepatocellular carcinoma, endometrial cancer, kidney cancer, vulvar cancer, thyroid cancer, penile cancer, leukemia, malignant lymphoma, plasmacytoma, myeloma, glioblastoma multiforme, sarcoma, osteosarcoma, and melanoma. In one embodiment, the method predicts the response to ADC in patients with cancer selected from the group consisting of breast cancer, gastric cancer, colorectal cancer, non-small cell lung cancer, esophageal cancer, head and neck cancer, esophagogastric junction adenocarcinoma, biliary tract cancer, Paget's disease, pancreatic cancer, ovarian cancer, uterine carcinosarcoma, bladder cancer, and prostate cancer. In one embodiment, the method predicts the response to ADC in breast cancer patients. In another embodiment, the method predicts the response to ADC in gastric cancer patients. In yet another embodiment, the method predicts the response to ADC in lung cancer patients.

[0063] In the first step, high-resolution digital images are obtained of tissue sections from cancer patients that have been stained using one or more biomarkers or stains.

[0064] To predict the efficacy of ADC therapy, diagnostic biomarkers conjugated with dyes that target the same proteins targeted by ADC therapy are used. In one embodiment, the anti-HER2 ADC comprises an anti-HER2 antibody conjugated to a drug-linker via a thioether linker, represented by the following formula: [Formula 1] [ka] In the formula, A represents the connection site with the anti-HER2 antibody. In some embodiments, the anti-HER2 antibody is an antibody having a heavy chain comprising CDRH1 consisting of the amino acid sequence represented by SEQ ID NO: 4, CDRH2 consisting of the amino acid sequence represented by SEQ ID NO: 5, and CDRH3 consisting of the amino acid sequence represented by SEQ ID NO: 6, and a light chain comprising CDRL1 consisting of the amino acid sequence represented by SEQ ID NO: 7, CDRL2 consisting of the amino acid sequence of amino acid residues 1-3 of SEQ ID NO: 8, and CDRL3 consisting of the amino acid sequence represented by SEQ ID NO: 9.

[0065] In some further embodiments, the anti-HER2 antibody is an antibody comprising a heavy chain variable region consisting of the amino acid sequence represented by SEQ ID NO: 10 and a light chain variable region consisting of the amino acid sequence represented by SEQ ID NO: 11. In some further embodiments, the anti-HER2 antibody is an antibody comprising a heavy chain consisting of the amino acid sequence represented by SEQ ID NO: 12 and a light chain consisting of the amino acid sequence represented by SEQ ID NO: 3. In a particular embodiment, the anti-HER2 antibody is an antibody comprising a heavy chain consisting of the amino acid sequence represented by SEQ ID NO: 2 and a light chain consisting of the amino acid sequence represented by SEQ ID NO: 3. In one embodiment, the anti-HER2 ADC targeted for scoring is trastuzumab deruxtecan, as represented by Formula 3 above. Therefore, in the embodiment, the diagnostic biomarker also targets the HER2 protein.

[0066] In step 2, a pre-trained convolutional neural network processes digital images of cancer patient tissue stained with a diagnostic antibody linked to a dye such as 3,3'-diaminobenzidine (DAB). The staining intensity of the dye in the cancer cell membrane is determined based on the average staining intensity of the dye across all pixels associated with the corresponding segmented membrane object. Furthermore, the staining intensity of the dye in a single pixel is calculated based on the red, green, and blue components of the pixel. The result of the image analysis process is two back-image layers for each pixel in the digital image, representing the probability that the pixel belongs to the cell nucleus and the probability that the pixel belongs to the cell membrane.

[0067] In another embodiment of step 2, two pre-trained convolutional networks process a digital image of the tissue. The result of processing by the first network is a back image layer representing the likelihood that each pixel in the digital image belongs to the cell nucleus. The result of processing by the second network is a back image layer representing the likelihood that each pixel in the digital image belongs to the cell membrane.

[0068] In step 3, individual cancer cells are detected based on heuristic image analysis of the nucleus and the lower layers of the membrane. Cancer cell objects are generated, including cell membrane objects and optionally cytoplasmic objects.

[0069] In step 4, a single-cell ADC score is determined for each cancer cell. The single-cell ADC score is based on a surrogate measurement of (1) the amount of DAB in the cell membrane and (2) the amount of ADC payload uptake. The amount of DAB is determined by the staining intensity of each membrane based on the average optical density of the brown diaminobenzidine (DAB) signal at the membrane pixels. The amount of ADC payload uptake is estimated based on the amount of DAB in the cell membrane and, optionally, in the cytoplasm of the cell, and further optionally, based on the amount of DAB in the membrane and cytoplasm of cells adjacent to the cell being scored. The amount of DAB in the cytoplasm of a cell is determined by the staining intensity of each cytoplasm, which is calculated based on the average optical density of the brown DAB signal at the cytoplasmic pixels. The amount of DAB in adjacent cells is determined for those cancer cells within a given distance of the cell being scored.

[0070] In step 5, a patient score QCS is calculated for the digital image of the tissue based on statistical calculations of the single-cell ADC score of all cancer cells in the digital image. The patient score indicates how cancer patients respond to treatment including anti-HER2 ADCs. The type parameters of the single-cell ADC score in step 4 and the statistical calculations in step 5 are optimized using a training cohort of patients with a known response to ADC treatment. The optimization goal is a low p-value in the Kaplan-Meier analysis of the group of score-positive patients versus score-negative patients in the training cohort.

[0071] In step 6, if the score is greater than a predetermined threshold, treatment including an anti-HER2 ADC is recommended for score-positive patients.

[0072] In some embodiments, a patient is QCS positive if at least 90% of tumor cells have a membrane optical density of 5 or higher. In some embodiments, a patient is QCS positive if at least 90% of tumor cells have a membrane optical density of 6 or higher. In some embodiments, a patient is QCS positive if at least 90% of tumor cells have a membrane optical density of 7 or higher. In some embodiments, a patient is QCS positive if at least 90% of tumor cells have a membrane optical density of 8 or higher. In some embodiments, a patient is QCS positive if at least 90% of tumor cells have a membrane optical density of 8.4 or higher. In some embodiments, a patient is QCS positive if at least 90% of tumor cells have a membrane optical density of 9 or higher. In some embodiments, a patient is QCS positive if at least 90% of tumor cells have a membrane optical density of 10 or higher. In some embodiments, a patient is QCS positive if at least 90% of tumor cells have a membrane optical density of 15 or higher. In some embodiments, a patient is QCS positive if at least 90% of tumor cells have a membrane optical density of 20 or higher. In some embodiments, a patient is QCS-positive if at least 90% of tumor cells have a membrane optical density of 25 or higher.

[0073] In some embodiments, a patient is QCS positive if at least 50% of tumor cells have a membrane optical density of 5 or higher, 8 or higher, or 25 or higher. In some embodiments, a patient is QCS positive if at least 50% of tumor cells have a membrane optical density of 8 or higher. In some embodiments, a patient is QCS positive if at least 60% of tumor cells have a membrane optical density of 5 or higher, 8 or higher, or 25 or higher. In some embodiments, a patient is QCS positive if at least 60% of tumor cells have a membrane optical density of 8 or higher. In some embodiments, a patient is QCS positive if at least 70% of tumor cells have a membrane optical density of 5 or higher, 8 or higher, or 25 or higher. In some embodiments, a patient is QCS positive if at least 70% of tumor cells have a membrane optical density of 8 or higher. In some embodiments, a patient is QCS positive if at least 80% of tumor cells have a membrane optical density of 5 or higher, 8 or higher, or 25 or higher. In some embodiments, a patient is QCS positive if at least 80% of tumor cells have a membrane optical density of 8 or higher. In some embodiments, a patient is QCS positive if at least 90% of tumor cells have a membrane optical density of 5 or higher, 8 or higher, or 25 or higher. In some embodiments, a patient is QCS positive if at least 90% of tumor cells have a membrane optical density of 8 or higher. In some embodiments, a patient is QCS positive if at least 95% of tumor cells have a membrane optical density of 5 or higher, 8 or higher, or 25 or higher. In some embodiments, a patient is QCS positive if at least 95% of tumor cells have a membrane optical density of 8 or higher.

[0074] In some embodiments, the density of tumor cells having at least 8 membrane optical densities within the tumor area is 500 cells / mm². 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor cells having at least 8 membrane optical densities within the tumor area is 1000 cells / mm². 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor cells having at least 8 membrane optical densities within the tumor area is 1250 cells / mm². 2If the above is true, the patient is QCS positive. In some embodiments, the density of tumor cells having a membrane optical density of at least 8 within the tumor area is 1500 cells / mm 2 If the above is true, the patient is QCS positive. In some embodiments, the density of tumor cells having a membrane optical density of at least 8 within the tumor area is 1600 cells / mm 2 If the above is true, the patient is QCS positive. In some embodiments, the density of tumor cells having a membrane optical density of at least 8 within the tumor area is 1670 cells / mm 2 If the above is true, the patient is QCS positive. In some embodiments, the density of tumor cells having a membrane optical density of at least 8 within the tumor area is 1700 cells / mm 2 If the above is true, the patient is QCS positive. In some embodiments, the density of tumor cells having a membrane optical density of at least 8 within the tumor area is 1800 cells / mm 2 If the above is true, the patient is QCS positive. In some embodiments, the density of tumor cells having a membrane optical density of at least 8 within the tumor area is 1900 cells / mm 2 If the above is true, the patient is QCS positive. In some embodiments, the density of tumor cells having a membrane optical density of at least 8 within the tumor area is 2000 cells / mm 2 If the above is true, the patient is QCS positive. In some embodiments, the density of tumor cells having a membrane optical density of at least 8 within the tumor area is 3000 cells / mm 2 If the above is true, the patient is QCS positive.

[0075] In some embodiments, the density of tumor cells having a membrane optical density of at least 20 within the tumor area is 500 cells / mm 2 If the above is true, the patient is QCS positive. In some embodiments, the density of tumor cells having a membrane optical density of at least 20 within the tumor area is 1000 cells / mm 2 If the above is true, the patient is QCS positive. In some embodiments, the density of tumor cells having a membrane optical density of at least 20 within the tumor area is 1500 cells / mm 2If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor cells having at least 20 membrane optical densities within the tumor area is 2000 cells / mm². 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor cells having at least 20 membrane optical densities within the tumor area is 2500 cells / mm². 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor cells having at least 20 membrane optical densities within the tumor area is 2750 cells / mm². 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor cells having at least 20 membrane optical densities within the tumor area is 3000 cells / mm². 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor cells having at least 20 membrane optical densities within the tumor area is 3250 cells / mm². 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor cells having at least 20 membrane optical densities within the tumor area is 3500 cells / mm². 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor cells having at least 20 membrane optical densities within the tumor area is 4000 cells / mm². 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor cells having at least 20 membrane optical densities within the tumor area is 5000 cells / mm². 2 If the above conditions are met, the patient is QCS positive.

[0076] In some embodiments, a patient is QCS positive if the binary spatial proximity score is 90 or higher. In some embodiments, a patient is QCS positive if the binary spatial proximity score is 91 or higher. In some embodiments, a patient is QCS positive if the binary spatial proximity score is 92 or higher. In some embodiments, a patient is QCS positive if the binary spatial proximity score is 93 or higher. In some embodiments, a patient is QCS positive if the binary spatial proximity score is 94 or higher. In some embodiments, a patient is QCS positive if the binary spatial proximity score is 95 or higher. In some embodiments, a patient is QCS positive if the binary spatial proximity score is 96 or higher. In some embodiments, a patient is QCS positive if the binary spatial proximity score is 97 or higher. In some embodiments, a patient is QCS positive if the binary spatial proximity score is 98 or higher. In some embodiments, a patient is QCS positive if the binary spatial proximity score is 99 or higher. In some embodiments, a patient is QCS positive if the binary spatial proximity score is 99.5 or higher. In some embodiments, a patient is QCS positive if the binary spatial proximity score is 99.8 or higher. In the above embodiments, the binary spatial proximity score is calculated as follows: Spatial proximity score = ([Number of tumor cells with OD>8] + [Number of tumor cells with OD<=8 located 50μm away from at least one tumor cell with OD>8]) / [Total number of tumor cells].

[0077] In some embodiments, a patient is QCS positive if the continuous spatial proximity score is 20 or higher. In some embodiments, a patient is QCS positive if the continuous spatial proximity score is 30 or higher. In some embodiments, a patient is QCS positive if the continuous spatial proximity score is 40 or higher. In some embodiments, a patient is QCS positive if the continuous spatial proximity score is 50 or higher. In some embodiments, a patient is QCS positive if the continuous spatial proximity score is 60 or higher. In some embodiments, a patient is QCS positive if the continuous spatial proximity score is 70 or higher. In some embodiments, a patient is QCS positive if the continuous spatial proximity score is 80 or higher. In some embodiments, a patient is QCS positive if the continuous spatial proximity score is 90 or higher. In the embodiments described above, the continuous spatial proximity score is the 10th percentile of the cumulative optical density of each tumor cell and any adjacent tumor cells within 25 μm, weighted by the distance to the adjacent cells.

[0078] In some embodiments, the density of tumor-infiltrating lymphocytes in the stroma is 50 cells / mm³. 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor-infiltrating lymphocytes in the stroma is 100 cells / mm³. 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor-infiltrating lymphocytes in the stroma is 125 cells / mm³. 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor-infiltrating lymphocytes in the stroma is 150 cells / mm³. 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor-infiltrating lymphocytes in the stroma is 160 cells / mm³. 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor-infiltrating lymphocytes in the stroma is 165 cells / mm³. 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor-infiltrating lymphocytes in the stroma is 168 cells / mm³. 2If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor-infiltrating lymphocytes in the stroma is 170 cells / mm³. 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor-infiltrating lymphocytes in the stroma is 175 cells / mm³. 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor-infiltrating lymphocytes in the stroma is 180 cells / mm³. 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor-infiltrating lymphocytes in the stroma is 190 cells / mm³. 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor-infiltrating lymphocytes in the stroma is 200 cells / mm³. 2 If the above conditions are met, the patient is QCS positive.

[0079] In some embodiments, the density of tumor-infiltrating lymphocytes in the stroma is 300 cells / mm³. 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor-infiltrating lymphocytes in the stroma is 400 cells / mm³. 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor-infiltrating lymphocytes in the stroma is 500 cells / mm³. 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor-infiltrating lymphocytes in the stroma is 600 cells / mm³. 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor-infiltrating lymphocytes in the stroma is 700 cells / mm³. 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor-infiltrating lymphocytes in the stroma is 735 cells / mm³. 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor-infiltrating lymphocytes in the stroma is 750 cells / mm³. 2 If the above conditions are met, the patient is QCS positive.

[0080] In some embodiments, the density of tumor-infiltrating lymphocytes in the stroma is 800 cells / mm³. 2If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor-infiltrating lymphocytes in the stroma is 900 cells / mm³. 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor-infiltrating lymphocytes in the stroma is 1000 cells / mm³. 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor-infiltrating lymphocytes in the stroma is 2000 cells / mm³. 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor-infiltrating lymphocytes in the stroma is 3000 cells / mm³. 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor-infiltrating lymphocytes in the stroma is 4000 cells / mm³. 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the density of tumor-infiltrating lymphocytes in the stroma is 5000 cells / mm³. 2 If the above conditions are met, the patient is QCS positive.

[0081] In some embodiments, a) at least 90% of tumor cells have a membrane optical density of 8.4 or higher, and b) the density of tumor cells with a membrane optical density of at least 20 within the tumor area is 3000 cells / mm². 2 If the above conditions are met, the patient is QCS positive. In some embodiments, the patient is score positive if a) at least 90% of tumor cells have a membrane optical density of 8.4 or higher, and b) the binary spatial proximity score is 99.8 or higher. In some embodiments, the patient is score positive if a) at least 90% of tumor cells have a membrane optical density of 8.4 or higher, and b) the continuous spatial proximity score is 50 or higher. In some embodiments, a) at least 90% of tumor cells have a membrane optical density of 8.4 or higher, and b) the density of tumor-infiltrating lymphocytes in the stroma is 735 cells / mm². 2 If the above conditions are met, the patient's score is positive.

[0082] In some embodiments, the patient is defined as having a) at least 90% of tumor cells having a membrane optical density of 8.4 or higher; or b) a tumor cell density of 3000 cells / mm² with at least 20 membrane optical densities within the tumor region. 2 The patient is QCS positive if: a) at least 90% of the tumor cells have a membrane optical density of 8.4 or higher; and c) the binary spatial proximity score is 99.8 or higher, or the continuous spatial proximity score is 50 or higher. In some embodiments, the patient is QCS positive if: a) at least 90% of the tumor cells have a membrane optical density of 8.4 or higher; and b) the density of tumor cells with a membrane optical density of at least 20 in the tumor region is 3000 cells / mm 2 The patient is QCS positive if: a) the above conditions are met; c) the binary spatial proximity score is 99.8 or higher; and d) the continuous spatial proximity score is 50 or higher. In some embodiments, a) at least 90% of tumor cells have a membrane optical density of 8.4 or higher; b) the density of tumor cells with a membrane optical density of at least 20 within the tumor region is 3000 cells / mm². 2 If the above conditions are met; c) the binary spatial proximity score is 99.8 or higher, or the continuous spatial proximity score is 50 or higher; and d) the density of tumor-infiltrating lymphocytes in the stroma is 735 cells / mm³ 2 If the above conditions are met, the patient is QCS positive. In some embodiments, a) at least 90% of tumor cells have a membrane optical density of 8.4 or higher; b) the density of tumor cells with a membrane optical density of at least 20 within the tumor region is 3000 cells / mm². 2 If the above conditions are met; c) the binary spatial proximity score is 99.8 or higher; e) the continuous spatial proximity score is 50 or higher; and d) the density of tumor-infiltrating lymphocytes in the stroma is 735 cells / mm². 2 If the above conditions are met, the patient is QCS positive.

[0083] VII. Examples of prediction and scoring methods. A. Image analysis of stained tissue. Here, we describe the method shown in Figure 14 in relation to specific images of stained cancerous tissue.

[0084] In step 1, the tissue sample is immunohistochemically stained using a dye linked to a diagnostic antibody that binds to the relevant protein on cancer cells in the tissue sample. Figure 15 (upper left image) is a digital image 17 of a portion of the stained tissue obtained in step 1. Image 17 shows tissue from a cancer patient immunohistochemically stained with an anti-Her2 diagnostic antibody linked to a dye. In this example, the diagnostic antibody is Ventana PATHway anti-HER-2 / neu(4B5) rabbit monoclonal primary antibody that targets the protein HER2. In other examples, the diagnostic antibody may be Dako HercepTest primary antibody, a rabbit anti-human antibody that also targets the protein HER2. The anti-Her2 / neu antibody binds to the membrane protein Her2 / neu so that 3,3'-diaminobenzidine (DAB) staining indicates the location of the protein Her2 / neu in the tissue sample.

[0085] In step 2, image analysis is performed on the digital image 17 to generate a back image layer of the nucleus and membrane of cancer cells using a convolutional neural network. The image analysis is used to detect cancer cells and their components, such as the nucleus, membrane, and cytoplasm. Figure 15 shows the image analysis process in step 2. For each pixel of the digital image 17, the convolutional neural network generates a back layer (grayscale image) indicating the probability that each pixel belongs to either the cell nucleus (upper right image in Figure 15) or the membrane (lower left image in Figure 15). High probabilities are shown in black, and low probabilities are shown in white.

[0086] Figure 16 shows another embodiment of how to perform the image analysis in step 2. The convolutional neural network generates a regression layer (grayscale image) for each pixel, showing the distance to the nucleus (upper right image in Figure 16) or the distance to the membrane (lower left image in Figure 16). Larger distances are shown in white, and smaller distances are shown in black. A comparison of Figure 15 and Figure 16 shows that the image analysis of the embodiment in Figure 16 generates thicker membrane objects but smaller nucleus objects than the embodiment in Figure 15.

[0087] In one embodiment, the convolutional neural network includes a series of convolutional layers leading from the input image 17 to a bottleneck layer with a very low spatial size (1-16 pixels), and a series of deconvolutional layers leading to a later layer with the same size as the input image 17. This network architecture is called U-Net. Training of the weights of the convolutional neural network is performed by generating manually annotated layers of nuclei and membranes in multiple training images, and then adjusting the network weights by an optimization algorithm so that the generated later layer most closely resembles the manually generated annotation layer.

[0088] In another embodiment, annotation layers for nuclei and membranes are automatically generated in multiple training images and manually corrected. Epithelial regions and nuclear centers are manually annotated as regions and points, respectively. For each training image, membrane segmentation is automatically generated by seeding by annotated nuclear centers and applying a region growth-like algorithm (e.g., watershed segmentation) constrained by the extent of the annotated epithelial region. Given training images, nuclear segmentation is automatically generated by applying a blob detection algorithm (e.g., by the Maximum Stable Extreme Region MSER algorithm) and selecting only detected blobs containing annotated nuclear centers as nuclei. The automatically generated membrane and nuclear segmentation is visually reviewed and manually corrected as necessary. The correction process includes one of the following methods: rejecting inaccurately segmented membranes or nuclei, explicitly accepting accurately annotated membranes or nuclei, or refining the shape of the membranes or nuclei. For each image with annotated membranes or nuclei, an annotation layer is created. In one embodiment, each pixel in the annotation layer is assigned "1" if it belongs to an annotation object (membrane or nucleus), and "0" otherwise. In another embodiment, the pixels in the annotation layer represent the distance to the nearest annotation object. The network weights are adjusted by an optimization algorithm so that the generated subsequent layers are most similar to the automatically generated membrane and nucleus annotation layers.

[0089] Figure 17 shows step 3, in which individual cancer cell targets, each containing the cell membrane and cytoplasm, are detected. The heuristic image analysis process uses watershed segmentation to segment the cell nuclei using the nuclear backlayer generated by a convolutional neural network. Segmentation generates nuclear objects. Each nuclear object is assigned a unique identifier (UID). Individually identified nuclei are shown as dark objects in Figure 17 (bottom right image). Detected nuclei are also displayed as overlays on the input image 17 (top left image) and the backlayers of the nuclei (top right image) and membrane (bottom left image).

[0090] In one embodiment, watershed segmentation includes thresholding of the nucleus rear layer at a predefined first size threshold. All single connected pixels exceeding the first size threshold are considered to belong to a nucleus object. Nucleus objects with an area less than 16 μm² are discarded. Each nucleus object is assigned a UID. In a subsequent step, the nucleus object is grown toward smaller nucleus rears, where the added nucleus rear pixels must be greater than a second predetermined threshold.

[0091] Figure 18 shows a further segmentation step that uses nuclear objects to detect and improve membrane segmentation. The region growth algorithm uses the detected nuclear objects as seeds to grow up to the membrane ridges (approximate cell boundaries) in the post-membrane layer. The detected membrane (bottom right image) is shown as an overlay on the input image 17 (top left image) and the post-membrane layers of the nucleus (top right image) and membrane (bottom left image).

[0092] Figure 19 illustrates the detection and segmentation of membrane objects, where the boundary pixel region of detected cells is segmented outward to a membrane stochastic layer and a predetermined membrane layer posterior threshold. Thicker boundary regions become membrane objects. Each membrane object is assigned the same UID as the associated nuclear object.

[0093] The space between the membrane and the nucleus is assigned to the cytoplasm using the nuclear UID. For each membrane (see upper left image in Figure 19) and cytoplasm (see upper left image in Figure 19), the average optical density of DAB staining, along with the UID, is exported to a file on the hard drive. For each cell (defined as including the nucleus, cytoplasm, and membrane), the position of the cell's centroid (x, y) in the slide is also exported. The files may reside on a hard disk, solid-state disk, or in a portion of dedicated RAM within a computer system.

[0094] Figure 20 shows the results of image analysis in an image analysis software environment. Figure 20 (top left image) shows segmentation of nuclear and membrane objects as an overlay on the digital image 17. Figure 20 (bottom left image) shows segmentation of nuclear objects as an overlay on the optical density presentation of image 17. Dark optical density pixels are associated with a large amount of DAB, and bright optical density pixels are associated with a small amount of DAB. The DAB optical density of each image pixel is calculated from the red-green-blue representation of the image pixel by transforming the red-green-blue space so that the brown DAB component is an independent color, and then taking the logarithm of that brown color component. When immunofluorescence (IF) imaging is used to determine staining, the HER2 channel is acquired as a 12-bit, 16-bit, or 32-bit grayscale image, in contrast to using the red-green-blue color space. In IF, nuclei are marked using DAPI (2-[4-(aminoiminomethyl)phenyl]-1H-indole-6-carboxyimidoamide hydrochloride) as the first dye. The posterior layer of the nucleus is generated using an image of a first dye as input to a convolutional neural network. A second dye is conjugated to a diagnostic anti-HER2 antibody. The intensity of the fluorescence signal from the second dye corresponds to the optical density (OD) of the DAB. Figures 20 (upper right image) and 21 (top) show the image analysis script used to generate segmented images within the Definiens Developer XD platform. Figures 20 (lower right image) and 21 (bottom) show the exported measurements for all cell membrane and cytoplasmic objects in image 17. In another embodiment, the image analysis script is encoded using another programming language such as C++, C#, Java, Python, or R.

[0095] B. Calculation of the predicted ADC score. A single-cell ADC score is calculated for each cancer cell in the digital image 17 based on the optical density of DAB staining in membrane and cytoplasmic objects. The single-cell ADC score is also based on the staining intensity of the DAB dye in the membrane and cytoplasmic objects of adjacent cancer cells that are closer than a predetermined distance to the cancer cell from which the single-cell ADC score is being calculated. The score predicts the cancer patient's response to anti-HER2 ADC therapy.

[0096] Figure 22 shows exemplary quantitative results of the optical density of staining from the image analysis of steps 2-3 in a schematic diagram using gray values ​​for membrane and cytoplasmic pixels. The heuristic image analysis steps shown in Figures 15-20 are used to obtain exemplary segmentation into the cell nucleus, cell membrane, and cell cytoplasm in Figure 22. Bright gray values ​​in Figure 22 are associated with high DAB optical density, and therefore with a large amount of protein targeted by the diagnostic antibody. Dark gray values ​​are associated with low DAB optical density. Brighter pixels correspond to higher DAB optical density.

[0097] Figure 23 lists exemplary quantitative staining levels on the membrane and in the cytoplasm of the images in Figure 22, which are partially reproduced in Figure 23. The optical densities of the brown DAB signal from the membranes of the first, second, and third cells are 0.949, 0.369, and 0.498, respectively. The optical densities of the brown DAB signal from the cytoplasm of the first, second, and third cells are 0.796, 0.533, and 0.369, respectively. In the schematic image of Figure 23, the first cancer cell 18 expresses a large amount of the target protein HER2 and is very likely to be killed by the ADC payload entering the cell ligated to the ADC antibody (effect 1 in Figure 24). The second cancer cell 19 and the third cancer cell 20 do not express sufficient amounts of the target protein HER2 to be directly killed by the anti-HER2 ADC. However, due to the proximity of the second cancer cell 19 to the first cancer cell 18, the toxic payload released from the first cancer cell also kills the second cancer cell 19 (Effect 3 in Figure 24). The third cancer cell 20 remains active and may be the origin of a drug resistance mechanism that ultimately leads to patient death.

[0098] Figure 24 illustrates the mechanism by which anti-HER2 ADC therapy kills cancer cells. Trastuzumab deruxtecan, for example, also utilizes this mechanism. In the first step, an ADC antibody (e.g., trastuzumab) binds to the target protein HER2, inhibiting its innate function and potentially leading to cell death. In the second step, a payload (e.g., a type I topoisomerase inhibitor) is internalized into the cell, killing it through its toxicity. This payload uptake depends on the amount of target protein on the membrane, as well as the difference between the amounts of target protein on the membrane and in the cytoplasm. After uptake, the payload can be released from the cell into the surrounding tissue. In the third step, the payload can enter nearby cells and kill them. The spatial distribution of the payload within the tissue spreads by passive diffusion.

[0099] Traditional HER2 scoring reflects both the effect of ADC binding on the inhibition of the target protein HER2, and the effect of the cytotoxic payload that enters cancer cells along with the ADC antibody. Therefore, conventional scoring for trastuzumab therapy does not reflect the importance of the presence of the HER2 protein in the cytoplasm, nor the effect of the cytotoxic payload that diffuses into the tissue after being released from the first dead cancer cells. In comparison, the novel predictive ADC score measures the impact of the release of the cytotoxic payload on adjacent cancer cells.

[0100] In step 4 of the method in Figure 14, a single-cell ADC score is determined for each cancer cell. Figure 25 shows the calculation of the single-cell ADC score for each of the three cells shown in Figure 23, incorporating an exponential weighting coefficient based on cell separation to account for the uptake of the ADC payload into adjacent cells. The single-cell score can be calculated based on the formula shown in Figure 26 or the formula described in claim 4. The optical densities listed in Figure 23 for the DAB signals from the cell membrane (0.949, 0.369, and 0.498) and cytoplasm (0.796, 0.533, and 0.369) are inputs to the calculation shown in Figure 25. The first, second, and third cells 18-20 have single-cell scores of 0.145, 0.012, and 0.064, respectively.

[0101] Therefore, the single-cell ADC score incorporates the measurement of the amount of target protein on the cell membrane using DAB optical density and the estimation of ADC payload uptake. As shown in Figure 24, the uptake of the ADC payload to a first cell depends on both the amount of dye in its membrane and cytoplasm, and the amount of dye in the membrane and cytoplasm of a second cell in the vicinity of the first cell. More specifically, the vicinity may be a disk with a predetermined radius around the first cancer cell. In one embodiment, the single-cell ADC score of the first cancer cell is determined by a distance-weighted sum of the DAB optical densities of membrane and cytoplasmic targets where the relevant cancer cell center is closer to the first cancer cell center than a predetermined distance. In one embodiment, the predetermined distance is 50 μm, as used in the calculation in Figure 25. In another embodiment, the distance weighting involves calculating the sum of exponents of scaled negative Euclidean distances from the first cancer cell center to the other cancer cell center. In another embodiment, the powers of the sum are limited to 0, 1, and 2 (constant, linear term, square).

[0102] Figure 26 shows one embodiment of the formula for calculating the single-cell ADC score. The function a in the formula kl The distance from cell j to cell i is |r j -r i | Depends on ODM j This is the DAB optical density of the cell membrane, and ODC. jis the DAB optical density of the cytoplasm of cell j. The constants A_ij, r_norm, and d are the same for all types of cancer. However, the score thresholds for determining whether a patient is eligible for ADC treatment are not the same for different types of cancer.

[0103] In step 5 of the method for demonstrating how a particular cancer patient responds to ADC therapy, a response score is calculated for a digital image 17 of tissue from the cancer patient based on staining for the target protein HER2, and the score indicates how the cancer patient responds to anti-HER2 ADC therapy. The response score is generated by aggregating all single-cell ADC scores from the tissue sample using statistical calculations. Thus, the score is calculated based on the statistics of single-cell ADC scores for all cancer cells detected in the digital image. In one embodiment, the statistic is a predetermined quantile of the estimated ADC payload uptake of all cancer cells in the image 17.

[0104] In step 6 of this method, if the response score is greater than a predetermined threshold, anti-HER2 ADC therapy is recommended for cancer patients. The predetermined threshold in step 6 and the quantile in step 5 are determined by optimizing the positive predictive value, negative predictive value, and prevalence of positive recommendations using a cohort of patients with known single-cell ADC scores and treatment response parameters.

[0105] In another example of calculating the response score in step 5, digital images s of a tissue sample from a cancer patient are acquired and stained with a diagnostic antibody (dAB). Image analysis is performed on the digital images s to identify N cancer cells (c i ) 0≦i≦N It detects [the substance]. The diagnostic antibody (dAB) is Ventana PATHWAY anti-HER-2 / neu(4B5). Cell c i The optical density in the segmented film is

number

number

number

number

[0106] This term

number

number

number

number

number

number

number

[0107] In one example, function

number

number

number

number

number

number

number

number

number

number

number

number

[0108] In Equation 7,

number

[0109] Positive function p(c i One specific example of this is cell c i Associating with, [Formula 13]

number

[0110] C. Verification of a prediction method based on breast cancer patients. The accuracy of the novel predictive ADC score generated according to the method shown in Figure 14 was validated based on a patient study (J101 NCT02564900), a Phase I clinical trial using trastuzumab deruxtecan (DS-8201). The J101 patient study dataset includes stained tissue images and treatment response rates from patients with multiple cancer types, including breast and gastric cancer. The initial validation of the predictive ADC score was based on data from 151 breast cancer patients with varying HER2 expression levels (1+, 2+, 3+). The response score was trained using pathologist annotations, and the performance of the scoring method was validated on unverified data to ensure its generalizability and robustness. The response score was applied blindly to the J101 data. Optical density OD (level of brown DAB staining intensity) was calculated on the detected membrane to derive features relevant to survival prediction. The features of the response score were selected to maximize the objective response rate (ORR) in patients in the positive group and minimize the ORR in patients in the negative group. The objective response rate is the percentage of patients who show either a partial response or a complete response to treatment. Partial response is defined as a tumor reduction of 30% to 100%. Complete response is 100% tumor reduction and elimination.

[0111] The analysis and verification described below demonstrate a high correlation between the optical density values ​​(R=0.993) measured on membranes detected using image analysis according to the method in Figure 14 and the optical density values ​​(R=0.995) measured on reinforced membranes annotated by pathologists. Figure 27 shows how pathologists consistently annotate membranes by comparing them to each other. This figure shows the optical density of membrane stains identified by a first pathologist relative to the optical density identified by a second pathologist. The optical density scale in the figure is listed from 0 to 300, but the actual optical density is measured in the range of 0 to 220. Optical density may be alternatively shown in this application in the range of 0 to 1 by dividing the measured optical density by 220. Figure 28 shows a comparison of how membranes detected using image analysis correlate with membranes identified by pathologists. The comparison of the scatter plots in both figures demonstrates that image analysis using the method in Figure 14 detects membranes almost consistently with pathologists. Furthermore, the response scores obtained from image analysis closely matched the pathologist's HER2 IHC testing scoring, but also showed extensive quantitative overlap between the IHC staining categories and the ISH staining categories. The scoring method shown in Figure 14 demonstrated a direct linear relationship between the objective response rate (ORR) and the increase in HER2 expression across the entire assay range.

[0112] Currently, HER2-targeted therapies are not approved for "low HER2" patients whose cancer is assigned scores for both (i) HER2 expression >0 and <1+ from immunohistochemistry (IHC) staining, (ii) HER2 expression 1+ from immunohistochemistry staining, and (iii) HER2 expression 2+ from immunohistochemistry (ISH) staining and negative (-) from in situ hybridization staining.

[0113] Figure 29 shows a low correlation between the quality of response and having a HER2 IHC2+ score, in contrast to having a HER2 IHC1+ score. Not all breast cancer patients from the J101 trial in the HER2-negative cohort with an IHC2+ score consistently showed a better response than patients in the cohort with an IHC1+ score. Patients represented by longer bars on the right side of the figure showed a greater reduction in tumor size than patients on the left side. Patients on the right whose tumor size shrank by more than 30% are considered to have responded to anti-HER2 ADC treatment. Patients whose tumors shrank by less than 30% and increased by less than 30% are considered to be in a stable state. Patients whose tumor size increased by more than 30% are considered to have progressive disease. Of the 22 patients who responded to treatment as shown in the figure 29, 9 had an IHC1+ score and 13 had an IHC2+ score. Therefore, the HER2 IHC score is not a good predictor of whether a patient will respond to treatment.

[0114] Figure 30 shows the response (change in tumor size) of 168 patients in the DESTINY-Breast01 trial who had a HER2-positive response score. Figure 30 shows that approximately 85% of patients scored as HER2-positive showed a response to treatment (>30% tumor reduction), 15% were stable, and only one patient had progressive disease. Therefore, HER2 IHC scores of 3+ and 2+ by ISH+ were better predictors of response to anti-HER2 ADC therapy and better predictors of no response to anti-HER2 ADC therapy than scores of 2+ and 1+.

[0115] Figure 31 shows that a cohort of 65 out of 151 J101 breast cancer patients included in the HER2-negative category of IHC scoring still demonstrated a favorable objective response rate (ORR) of 42% to anti-HER2 ADC therapy. In this HER2-negative cohort of J101 patients (n=65) for whom HER2-targeted therapy is not currently approved, 42% of patients still responded to anti-HER2 ADC therapy, with a median progression-free survival (mPFS) of 11 months. Therefore, the scoring method in Figure 14 can be used to identify HER2-negative category patients who show a favorable response to anti-HER2 ADC therapy. Figure 31 also shows that the objective response rate was 56% for a cohort of 72 patients classified as HER-positive using IHC scoring, and those patients responded to treatment with an mPFS of 14.1 months.

[0116] Figure 32 shows further stratification of 65 breast cancer patients in the conventional HER2-negative category using the scoring method of Figure 14. By adjusting the cutoff for the novel predictive ADC score, 40 patients from the 65 HER2-negative group were classified as "QCS positive." This cohort of 40 patients had an ORR of 52%, and the patients responded to anti-HER2 ADC treatment with a mPFS of 14.5 months. Therefore, the scoring method of Figure 14 was able to identify a subgroup of HER2-negative patients whose novel response score exceeded a predetermined threshold (based on staining intensity above a predetermined cutoff) indicating a good response to anti-HER2 ADC treatment.

[0117] Figure 33 shows the scoring results for all 151 J101 breast cancer patients using the method in Figure 14. The upper left figure shows an example of the results of QCS image analysis, which detected the cell center (small dot), nuclear boundary, and cell (membrane) boundary of each epithelial cell. The space between the nuclear boundary and the cell boundary is called the cytoplasm.

[0118] The graph in the upper right of Figure 33 shows the distribution of membrane staining for 151 J101 patients. The height of each bar represents the central tumor cell membrane optical density in each patient's HER2-stained tissue image. The shading of the bars represents the pathologist's visual HER2 scoring for each patient. The “cutoff” indicated at >8.04 for membrane optical density was determined based on the patient’s overall survival analysis of the J101 cohort. The graph at the bottom of Figure 33 is a histogram showing the cell OD scores for a single patient and their distribution.

[0119] Figure 34A is a more detailed version of the graph in the upper right of Figure 33, showing the stratification of 151 J101 patients into 120 "QCS-positive" (QCS+) patients and 31 "QCS-negative" (QCS-) patients. The group of 120 patients was scored as "QCS-positive" (QCS+) because at least 90% of the cells in their tissue samples showed an optical density of HER2 membrane staining above a predetermined threshold of 0.0365 (8.04 / 220). The group of 120 "QCS-positive" (QCS+) patients showed an objective response rate (ORR) of 56% and a mean progression-free survival (mPFS) of 14.1 months, as shown in the table in Figure 34B.

[0120] The novel response score also identified a “QCS-negative” group of 31 patients with an objective response rate (ORR) of 26% and a mean progression-free survival (mPFS) of 9 months. Thus, the scoring method in Figure 14 identified a larger proportion of the 151 J101 breast cancer patients who could benefit from anti-HER2 ADC therapy than conventional HER2 IHC scoring performed by pathologists. Nevertheless, the ability to identify patients from the conventional HER2-negative IHC scoring group who could benefit from anti-HER2 ADC therapy is important because of the high unmet need for effective treatment among patients in this HER2-negative group.

[0121] The staining intensity cutoff for the single-cell score used to generate response scores is adjusted to divide J101 patients into a first group with a high ORR and a second group with a low ORR. The best stratification of J101 patients into the first group with the highest ORR and the second group with the lowest ORR was achieved by lowering the staining intensity threshold for the single-cell score to 0.0365 (8.04 / 220) to include the majority of tumor cells expressing at least minimal amounts of HER2. This staining intensity threshold of 8.04 / 220 is in contrast to current clinical guidelines that set higher staining cutoffs to include only a small fraction of cells expressing higher levels of HER2.

[0122] The bar graph in Figure 34A shows the conventional HER2 IHC scores for each of the 151 J101 breast cancer patients, indicated by the shading of the bars. For example, the bars for patients with an IHC3+ score are the second darkest gray, and the bars for patients with an IHC1+ score are white. It is clear that not all patients with the same HER2 IHC score were grouped together, except for patients with the HER2 IHC3+ score corresponding to the highest membrane optical density. However, patients with much lower membrane optical densities were also likely to respond well to anti-HER2 ADC therapy. We define the patient group with an ORR of 56% (n=120) as "HER-positive". Patients in this group were identified as having at least 90% of cells in the tissue slides exhibiting HER2 staining optical density greater than a given intensity threshold of 0.0365 (8.04 / 220), which is lower than the intensity threshold used for the conventional HER2 IHC score of 2+.

[0123] In another embodiment, cells counted as having at least a predetermined threshold of staining intensity also include cells that are in the vicinity of cells with staining above the threshold, but which themselves have staining below the threshold. Thus, the response score also takes into account the spatial heterogeneity of stained cells by characterizing cells as either exhibiting membrane staining above a predetermined optical density threshold (positive cells) or being within a predetermined distance from positive cells.

[0124] Figure 35 is a graph of Kaplan-Meier curves for progression-free survival for two groups of J101 breast cancer patients in a HER2-negative cohort identified using the method in Figure 14, in an embodiment that also considers adjacent stained cells. The method generates a score showing the probability of survival for 65 breast cancer patients in the HER2-negative cohort. The upper curve shows the group of patients with a better survival outcome, which corresponds to patients in whom 95% of cells (i) showed at least the minimum threshold optical density of HER2 staining, or (ii) were located within the minimum distance from cells showing minimum staining.

[0125] In this method, all epithelial cells in each tissue sample that exhibit an average optical density on the membrane greater than 0.04077 (8.97 / 220) are identified, and these cells are called "p1". Next, all epithelial cells in the tissue sample that exhibit an average optical density on the membrane less than 0.04077 but are within a distance of 20 μm (microns) from the p1 cells are identified, and these cells are called "p2". The percentage of "p12" cells is then calculated as (number of p1 + number of p2) / (total number of epithelial cells in the tissue sample). Patients lie on the upper Kaplan-Meier curve if they have a higher survival rate and the percentage of p12 cells is greater than 95%.

[0126] Patients in the upper curve had a mean progression-free survival of 16 months, while patients in the lower curve had a mean progression-free survival of 9 months. Patients in the upper curve, who have longer survival rates and a better response to anti-HER2 ADC therapy, have a more uniform distribution of HER2-positive cells among HER2-negative cells. If the majority of HER2-positive cells are aggregated and not distributed in the tumor tissue within the minimum threshold distance (20 μm in this example) from HER2-negative cells, then anti-HER2 ADCs targeting HER2-positive cells are not approaching HER2-negative cells sufficiently to have a significant bystander effect. The novel predictive QCS score identifies patients with homogeneous HER2 expression who show an improved response from the effect of ADCs on adjacent cancer cells releasing their cytotoxic payload to HER2-positive cells. To achieve a response to anti-HER2 ADC therapy similar to that shown in the upper Kaplan-Meier curve, clinicians would administer anti-HER2 ADC therapy to each patient whose novel response score exceeds 95% of p12 cells in the tissue sample.

[0127] Figure 36 is a graph of Kaplan-Meier curves for progression-free survival for two groups of patients from the entire 151 J101 patients using the method in Figure 14, in an embodiment that also considers p2 cells within the minimum distance of p1 cells. This method generates a score representing the survival probability of the 151 breast cancer patients in the J101 trial. The upper curve represents the group of patients with a better survival outcome, which corresponds to patients in whom 95% of cells either (i) showed an optical density of HER2 staining of at least 0.0481 (10.59244 / 220) or (ii) were located within a minimum distance of 20 μm from cells showing minimal staining. Patients in the upper curve had a mean progression-free survival of 19 months, while patients in the lower curve had a mean progression-free survival of 11 months.

[0128] Figure 37 shows other parameters and features used in the method of Figure 14 to generate a score indicating the survival probability of each cancer patient by aggregating all single-cell ADC scores from tissue samples. The bottom 10 features in Figure 37 are ranked according to the best stratification (measured by log-rank p-value) between two curves obtained using specific parameters of the labeled features.

[0129] The nine features shown above in Figure 37 are ranked according to the mean objective response rate of patients in the curve above, obtained using specific parameters of the labeled features. Feature #1 classifies patients based on the percentage of positive tumor cells, with positivity defined by a membrane optical density greater than 5. Feature #2 ranks patients based on the percentage of positive tumor cells, with positivity defined by a cell optical density greater than 5 (average optical density of both cytoplasm and membrane). Feature #3 is a bystander score that stratifies patients using positive / negative cell classification, where positive cells have a membrane optical density greater than 25, and the score considers all cells within a radius of 100 μm. Feature #4 is a bystander score that classifies patients using positive / negative cell classification, where positive cells have a membrane optical density greater than 10, and the score considers all cells within a radius of 50 μm. Feature #5 is a bystander score that classifies patients using positive / negative cell classification, where positive cells have a membrane optical density greater than 25, and the score considers all cells within a radius of 50 μm. Feature #6 ranks patients by calculating the difference in membrane optical density to cytoplasmic optical density for each tumor cell, and then taking the 15th percentile of the resulting histogram. Feature #7 is a bystander score that classifies patients using positive / negative cell classification, where positive cells have a membrane optical density greater than 10, and the score considers all cells within a radius of 10 μm. Feature #8 stratifies patients by calculating membrane optical density*(membrane optical density - cytoplasmic optical density) for each tumor cell, and then taking the 10th percentile of the resulting histogram. Feature #9 ranks patients by calculating membrane optical density*(max(0, membrane OD - cytoplasmic OD)) for each tumor cell, and then taking the 10th percentile of the resulting histogram.

[0130] The method in Figure 14 for predicting patient response to anti-HER2 ADC therapy may, in some cases, be based on the presence of stromal tumor-infiltrating lymphocytes (TILs) in addition to the staining characteristics of cancer cell targets. The validation described below demonstrates that a modified scoring method in Figure 14, also based on TILs, showed a better correlation between objective response rate (ORR) and increased HER2 expression + TIL prevalence. Tumor-infiltrating lymphocytes (TILs) consist mainly of cytotoxic (CD8+) T cells and helper (CD4+) T cells. CD8+ T lymphocytes correlate with favorable clinical outcomes, and patients with a large number of stromal TILs in breast lesions show better survival when treated with trastuzumab. Combining TIL prevalence with features based on image analysis extracted from HER2 staining improves the accuracy of patient scores generated by the method in Figure 14 for predicting how cancer patients will respond to anti-HER2 ADC therapy, particularly for breast cancer patients with tumors that have low levels of HER2 expression.

[0131] In a further step of the method shown in Figure 14, deep learning-based image analysis is performed on digital images of tissue immunohistochemically stained with a dye-conjugated anti-Her2 diagnostic antibody. A convolutional neural network is trained to detect TILs in HER2-stained IHC images. Although TILs themselves are not HER2-stained, a deep learning-based model was developed to directly detect TILs from HER2-stained IHC images. For the TIL detection model, a separate subset of the region of interest was annotated for TILs. In the absence of functional staining, point annotation was used to indicate TILs, which are characterized as relatively small, round to polygonal cells with little cytoplasm and a uniform nuclear texture. Only TILs located in tumor-associated stroma and intraepithelial TILs were annotated and detected. The TIL detection model was applied to non-epithelial regions within the annotated tumor core region. TIL centers were detected by thresholding the resulting posterior maps and then performing non-maximal suppression.

[0132] In another further step in the method shown in Figure 14, a TIL score is calculated for each HER2-stained IHC image. Thus, this method uses the density of TILs detected in the stained tissue as a biomarker, in addition to the score based on HER2 staining. The model calculates the following three densities: (i) the density of all TILs within the entire annotated tumor area, (ii) the density of TILs in segmented tumor-associated stroma, and (iii) the density of TILs in situ.

[0133] To improve the accuracy of the predicted ADC score, TIL-based features were used in combination with HER2-stained features. The optimal combination of features best divides the patient cohort into groups with longer progression-free survival (PFS) and groups with shorter PFS, as indicated by the log-rank test p-values ​​for the Kaplan-Meier curves of all patients in the cohort. The table below Figure 37 includes three TIL-based features and seven HER2-stained features, ranked by the stratification each feature achieves individually between the two curves of the Kaplan-Meier graph, as indicated by the log-rank p-values.

[0134] Figure 38 is a table of three further features comparing their log-rank test p-values ​​and their predicted objective response rates (ORRs) for patients with longer and shorter survival times. The table in Figure 38 includes features based on TILs, features based on HER2+ tumor cell density, and features based on HER2+ cells in a defined neighborhood calculated on the J101 dataset from a cohort of 151 breast cancer patients. The p-values ​​in the table were determined by cross-validation, where the model cutoff was trained on the training set and then applied to the validation set. The resulting cross-validated model is abbreviated as CM. For example, for the TIL density feature, the cutoff for stratifying patients into high (QCS+)CM and low (QCS-)CM groups is 168. This means that if the density of tumor-infiltrating lymphocytes, measured by dividing the number of TILs in the stroma by the tumor core area (mm^2), exceeds 168 / mm^2, the patient belongs to the CM-high group. For the HER2+ cell density feature, the cutoff is 1672. For this feature, if the density of positive tumor cells, measured by dividing the number of tumor cells with a membrane optical density greater than 15 by the area of ​​the tumor core (mm²), exceeds 1672 / mm², the patient belongs to the high CM group. For the HER2+ neighbor score, the cutoff is 37. The cutoff is first applied by generating a histogram by calculating a continuous bystander score using a radius of 25 μm for all tumor cells. Then, if the 5% quantile of the histogram of all cells is greater than 37, the patient belongs to the high CM group. The TIL density feature achieved the best stratification for 151 patients, yielding a log-rank test p-value of 0.007. From the HER2+ cell density feature, a Kaplan-Meier log-rank p-value of 0.011 was obtained, and from the HER2+ neighbor score feature, a p-value of 0.014 was obtained.

[0135] Figures 39A-C show the Kaplan-Meier curves for the three features listed in the table in Figure 38. Figure 39A is the Kaplan-Meier curve for TIL density. Figure 39B is the Kaplan-Meier curve for HER2+ cell density. Figure 39C is the Kaplan-Meier curve for HER2+ neighbor score. As indicated by the p-values ​​and Kaplan-Meier curves, the TIL density feature provides the best stratification between patients with longer and shorter progression-free survival (PFS) when applied to the entire group of 151 breast cancer patients. The Kaplan-Meier curve in Figure 39C for the HER2+ neighbor score feature is similar to the Kaplan-Meier curve in Figure 36 for the feature bystander_pot_cut1_20, which considers unstained bystander cells within a radius of 20 μm for each HER2-stained tumor cell. The HER2+ neighbor score feature uses a neighbor radius of 25 μm. For comparison, the continuous bystander feature "bystander_pot_cut1_20" in Figure 36 uses a radius of sigma=20um and alpha=2 to determine the weighted sum of the membrane optical densities of all tumor cells around a given cell, including the given cell itself. The distance-dependent weighting is calculated by "exp(-0.5*(dist / sigma)^2)".

[0136] Figures 40A, 40B, and 40C show the Kaplan-Meier curves for the three features listed in the table in Figure 38, applied only to 72 patients designated as HER2-positive out of 151 breast cancer patients. Here again, the TIL density feature achieved the best stratification with a p-value of 0.00095. The p-value for the HER2+ cell density feature was 0.092, and the p-value for the HER2+ neighbor score feature was 0.046.

[0137] Figures 41A, 41B, and 41C show the Kaplan-Meier curves for the three features listed in the table in Figure 38, applied only to 65 patients designated as HER2-negative from a total of 151 breast cancer patients. However, in the HER2-negative cohort, the TIL density feature provided the worst stratification between longer and shorter progression-free survival (PFS). The p-value for the TIL density Kaplan-Meier curve was 0.31. The HER2+ neighbor score achieved the best stratification, with a p-value of 0.0061. The stratification achieved by the HER2+ cell density feature was nearly good, with a p-value of 0.0064. The Kaplan-Meier curve for the HER2+ neighbor score feature in Figure 41C is similar to the Kaplan-Meier curve in Figure 35.

[0138] The Kaplan-Meier curve results in Figures 39–41 show that using TIL-based features in combination with HER2-stained features improves the accuracy of the method in Figure 14 for identifying patients who are likely to respond favorably to anti-HER2 ADC therapy. However, using TIL-based features does not improve the accuracy of identifying patients who are likely to respond well to anti-HER2 ADC therapy from a subgroup of HER2-negative patients. This suggests that tumor-infiltrating lymphocytes (TILs) are not essential for patients to achieve an immediate favorable response to anti-HER2 ADC therapy, but the presence of TILs may delay the overall progression of cancer.

[0139] D. Verification of a prediction method based on gastric cancer patients. The J101 trial, which used trastuzumab deruxtecan (DS-8201), also included patients with gastric cancer. Analysis of the J101 trial dataset showed a low correlation between the quality of response to anti-HER2 ADC therapy and conventional HER2 IHC scoring. Predicted responses using the QCS scoring method shown in Figure 14 deviated significantly from the results of conventional HER2 IHC scoring.

[0140] Figure 42 shows the scoring results for a cohort of 32 J101 gastric cancer patients using the method in Figure 14. In this embodiment, the QCS score is based on the median of the mean membrane optical density of all tumor cells in the sample. The statistical calculation for aggregating single-cell ADC scores is the “median”. Each single-cell ADC score is generated by the average (mean) of the DAB optical density at the cell membrane. Patients are ordered according to their increasing QCS scores. Figure 42 shows that HER2 IHC scores of 0, 1+, 2+, and 3+ do not correspond well with QCS scores. The bar graphs in Figure 42 also show the clinical outcomes of the 32 gastric cancer patients, listed above each bar graph as complete response (CR), partial response (PR), stable disease (SD), and progressive disease (PD). Figure 42 shows the results for gastric cancer patients similar to those for breast cancer patients in Figure 34A.

[0141] Figure 43 is a table of six features used in the method of Figure 14 to generate a score indicating the survival probability of each gastric cancer patient treated with anti-HER2 ADC therapy by aggregating single-cell ADC scores from tissue samples. The table in Figure 43 compares the log-rank test p-values ​​achieved by the features when stratifying patients based on progression-free survival (PFS). This table also shows the predicted objective response rate (ORR) for the cohort of patients with longer survival (patients at the bifurcation on the Kaplan-Meier curve). The ORR is expressed as the positive predictive value (PPV) for the ORR. The positive predictive value (PPV) is the number of patients correctly predicted as responders (by the observed ORR) divided by the total number of responders. The ORR is measured by RECIST using response categories CR, PR (ORR=true) and SD, PD (ORR=false). Figure 43 shows the features used to stratify gastric cancer patients, similar to the features used in Figures 37-38 to stratify breast cancer patients. The feature that provided the best stratification for patients with longer and shorter survival times was membOD_density_10 (density of tumor cells with a membrane optical density (membOD) greater than 10 per square mm of tumor area), which was based on the optical density of membrane staining of epithelial cells. The membOD_density_10 feature achieved stratification with a p-value of 0.00594.

[0142] Figure 44 is a table of seven HER2 stain-based features of a model used to stratify patients treated with anti-HER2 ADC therapy based on overall survival (OS) as opposed to progression-free survival (PFS). Therefore, the list of features is optimized for p-values ​​based on OS, as opposed to PFS. Kurtosis and skewness features provided the best stratification of patients based on overall survival and the achieved p-value of 0.0002.

[0143] Figure 45 shows the Kaplan-Meier curve for the feature membOD_density_10 listed in the table in Figure 43. Based on this feature, patients who are QCS positive and have a tumor cell density with a membrane optical density greater than 10 in the epithelium greater than 810 / mm^2 are more likely to benefit from HER2 ADC therapy. The feature membOD_density_10 was studied by dividing 32 gastric cancer patients based on PFS into a longer survival cohort of 23 patients and a shorter survival cohort of 9 patients, with a p-value of 0.00594.

[0144] Figure 46 shows the Kaplan-Meier curve for feature membOD_density_10, which is listed in the table in Figure 44. Feature membOD_density_10 divided 32 gastric cancer patients based on OS into a longer survival cohort of 23 patients and a shorter survival cohort of 9 patients, with a p-value of 0.00731.

[0145] Although the present invention has been described in relation to specific embodiments for illustrative purposes, the present invention is not limited thereto. Various modifications, adaptations, and combinations of features of the described embodiments can be implemented without departing from the scope of the present invention as described in the claims.

Claims

1. A pharmaceutical composition for use in a method of treating cancer, comprising an antibody-drug conjugate (ADC) payload and an ADC antibody that targets a protein on cancer cells, The aforementioned protein is human epidermal growth factor receptor 2 (HER2), The method includes administering a treatment containing the ADC to cancer patients whose treatment score exceeds a predetermined threshold based on staining intensity. The aforementioned treatment score is, The method involves immunohistochemically staining a tissue sample from a cancer patient using a dye linked to a diagnostic antibody, wherein the diagnostic antibody binds to the protein on the cancer cells in the tissue sample, thereby staining. To acquire a digital image of the aforementioned tissue sample, To detect cancer cells in the aforementioned digital image, For each cancer cell, a single-cell ADC score is calculated based on the staining intensity of the dye in the membrane and cytoplasm of the cancer cell, and the staining intensity of the dye in the membrane and / or cytoplasm of other cancer cells that are closer than a predetermined distance (d) to the cancer cell, and A treatment score is generated by aggregating all single-cell ADC scores from the tissue sample using statistical calculations. Generated by, The staining intensity of each membrane is calculated based on the average optical density of the dye signal in the pixels of the membrane, and / or the staining intensity of each cytoplasm is calculated based on the average optical density of the dye signal in the pixels of the cytoplasm. The single-cell ADC score for cell i is, |r j -r i| < d{a 20 (|r j -r i|) x ODM j 2 + a 11 (|r j -r i|) x ODM j x ODC j + a 02 (|r j -r i|) x ODC j 2 + a 00 (|r j -r i|)} total of all cells j It is calculated as follows: In the formula, ODM j is the optical density of the dye signal in the membrane of cell j, and ODC j is the optical density of the dye signal in the cytoplasm of cell j. The function a kl depends on the distance |r j - ri | between each cell i and each cell j, and the function a kl depends on the distance |r j - ri | between cell i and cell j in the following relationship: a kl (|r j - ri |) = A kl x exp(-|r j - ri | / r normal) by predefined constant coefficients A oo, A 1o, A o1, A 11, A 20, A o2, and The pharmaceutical composition wherein the coefficients A oo, A 1o, A o1, A 11, A 20, A o2, d, and r norm are determined by optimizing the correlation of response scores with the therapeutic response of a cohort of trained patients.

2. The pharmaceutical composition according to claim 1, wherein the detection of cancer cells includes detecting pixels belonging to the membrane and / or pixels belonging to the cytoplasm for each cancer cell.

3. The pharmaceutical composition according to claim 1, wherein the dye signal is a brown diaminobenzidine (DAB) signal in the pixels of the membrane and / or the brown DAB dye signal in the pixels of the cytoplasm.

4. The pharmaceutical composition according to claim 1, wherein the aggregation of all the single-cell ADC scores is selected from the group consisting of determining the mean, determining the median, and determining quantiles having a predetermined percentage.

5. The pharmaceutical composition according to any one of claims 1 to 4, wherein the ADC is trastuzumab deruxtecan (DS-8201).

6. The pharmaceutical composition according to any one of claims 1 to 4, wherein the ADC antibody is trastuzumab.

7. The pharmaceutical composition according to any one of claims 1 to 4, wherein the ADC payload is a topoisomerase I inhibitor.

8. The pharmaceutical composition according to any one of claims 1 to 4, wherein the diagnostic antibody is Ventana anti-HER2 / neu 4B5.

9. The pharmaceutical composition according to any one of claims 1 to 4, wherein the dye is 3,3'-diaminobenzidine (DAB).

10. The pharmaceutical composition according to any one of claims 1 to 9, wherein the cancer patient has a cancer selected from the group consisting of breast cancer, gastric cancer, colorectal cancer, lung cancer, esophageal cancer, head and neck cancer, gastroesophageal junction cancer, biliary tract cancer, Paget's disease, pancreatic cancer, ovarian cancer, uterine carcinosarcoma, bladder cancer, prostate cancer, urothelial carcinoma, gastrointestinal stromal tumor, cervical cancer, squamous cell carcinoma, peritoneal cancer, liver cancer, hepatocellular carcinoma, endometrial cancer, kidney cancer, vulvar cancer, thyroid cancer, penile cancer, leukemia, malignant lymphoma, plasmacytoma, myeloma, glioblastoma multiforme, sarcoma, osteosarcoma, and melanoma.

11. The pharmaceutical composition according to any one of claims 1 to 9, wherein the cancer patient has a cancer selected from the group consisting of breast cancer, gastric cancer, colorectal cancer, and lung cancer.

12. The ADC is an anti-HER2 antibody conjugated to a drug-linker via a thioether bond, and the drug-linker is represented by the following formula: 【Chemistry 1】 In the formula, A represents the connection site with the anti-HER2 antibody. A pharmaceutical composition according to any one of claims 1 to 11.

13. The ADC, A heavy chain comprising CDRH1 consisting of the amino acid sequence represented by SEQ ID NO: 4, CDRH2 consisting of the amino acid sequence represented by SEQ ID NO: 5, and CDRH3 consisting of the amino acid sequence represented by SEQ ID NO: 6, and A light chain comprising CDRL1 consisting of the amino acid sequence represented by SEQ ID NO: 7, CDRL2 consisting of the amino acid sequence of amino acid residues 1-3 of SEQ ID NO: 8, and CDRL3 consisting of the amino acid sequence represented by SEQ ID NO:

9. The pharmaceutical composition according to claim 12, comprising an anti-HER2 antibody containing the above.

14. The ADC, A heavy chain variable region consisting of the amino acid sequence represented by Sequence ID No. 10, and A light chain variable region consisting of the amino acid sequence represented by Sequence ID No. 11, The pharmaceutical composition according to claim 12, comprising an anti-HER2 antibody containing the above.

15. The ADC, A heavy chain consisting of the amino acid sequence represented by Sequence ID No. 12, and A light chain consisting of the amino acid sequence represented by Sequence ID No. 3, The pharmaceutical composition according to claim 12, comprising an anti-HER2 antibody containing the above.

16. The ADC, A heavy chain consisting of the amino acid sequence represented by Sequence ID No. 2, and A light chain consisting of the amino acid sequence represented by Sequence ID No. 3, The pharmaceutical composition according to claim 12, comprising an anti-HER2 antibody containing the above.

17. A method for generating a response score for predicting a cancer patient's response to an antibody-drug conjugate (ADC) comprising an ADC payload and an ADC antibody targeting a protein on cancer cells, wherein the protein is human epidermal growth factor receptor 2 (HER2). The method of immunohistochemically staining a tissue sample using a dye linked to a diagnostic antibody, wherein the diagnostic antibody binds to the protein on the cancer cells in the tissue sample, and stains the sample. To acquire a digital image of the aforementioned tissue sample, To detect cancer cells in the aforementioned digital image, For each cancer cell, a single-cell ADC score is calculated based on the staining intensity of the dye in the membrane and cytoplasm of the cancer cell, and based on the staining intensity of the dye in the membrane and cytoplasm of other cancer cells that are within a predetermined distance of the cancer cell, and The response score is generated by aggregating all single-cell ADC scores from the tissue sample using statistical calculations. Includes, The staining intensity of each membrane is calculated based on the average optical density of the dye signal in the pixels of the membrane, and / or the staining intensity of each cytoplasm is calculated based on the average optical density of the dye signal in the pixels of the cytoplasm. The single-cell ADC score for cell i is, |r j -r i| < d{a 20 (|r j -r i|) x ODM j 2 + a 11 (|r j -r i|) x ODM j x ODC j + a 02 (|r j -r i|) x ODC j 2 + a 00 (|r j -r i|)} total of all cells j It is calculated as follows: In the formula, ODM j is the optical density of the dye signal in the membrane of cell j, and ODC j is the optical density of the dye signal in the cytoplasm of cell j. The function a kl depends on the distance |r j - ri | between each cell i and each cell j, and the function a kl depends on the distance |r j - ri | between cell i and cell j in the following relationship: a kl (|r j - ri |) = A kl x exp(-|r j - ri | / r normal) by predefined constant coefficients A oo, A 1o, A o1, A 11, A 20, A o2, and The method wherein the coefficients A oo, A 1o, A o1, A 11, A 20, A o2, d, and r norm are determined by optimizing the correlation of response scores with the therapeutic response of a cohort of trained patients.

18. A method for generating a survival score for cancer patients treated with antibody-drug conjugates (ADCs), The method involves immunohistochemically staining a tissue sample from a cancer patient using a dye linked to a diagnostic antibody, wherein the ADC comprises an ADC payload and an ADC antibody that targets the human epidermal growth factor receptor 2 (HER2) protein on cancer cells, and the diagnostic antibody binds to the HER2 protein on cancer cells in the tissue sample. To acquire a digital image of the aforementioned tissue sample, To detect cancer cells in the aforementioned digital image, For each cancer cell, a single-cell ADC score is calculated based on the staining intensity of the dye in the membrane and cytoplasm of the cancer cell, and based on the staining intensity of the dye in the membrane and cytoplasm of other cancer cells that are within a predetermined distance of the cancer cell, and By aggregating all single-cell ADC scores from the tissue sample using statistical calculations, a score representing the survival probability of the cancer patient is generated. Includes, The staining intensity of each membrane is calculated based on the average optical density of the dye signal in the pixels of the membrane, and / or the staining intensity of each cytoplasm is calculated based on the average optical density of the dye signal in the pixels of the cytoplasm. The single-cell ADC score for cell i is, |r j -r i| < d{a 20 (|r j -r i|) x ODM j 2 + a 11 (|r j -r i|) x ODM j x ODC j + a 02 (|r j -r i|) x ODC j 2 + a 00 (|r j -r i|)} total of all cells j It is calculated as follows: In the formula, ODM j is the optical density of the dye signal in the membrane of cell j, and ODC j is the optical density of the dye signal in the cytoplasm of cell j. The function a kl depends on the distance |r j - ri | between each cell i and each cell j, and the function a kl depends on the distance |r j - ri | between cell i and cell j in the following relationship: a kl (|r j - ri |) = A kl x exp(-|r j - ri | / r normal) by predefined constant coefficients A oo, A 1o, A o1, A 11, A 20, A o2, and The method wherein the coefficients A oo, A 1o, A o1, A 11, A 20, A o2, d, and r norm are determined by optimizing the correlation of response scores with the therapeutic response of a cohort of trained patients.

19. A method for predicting a cancer patient's response to an antibody-drug conjugate (ADC), comprising an antibody and an ADC payload, wherein the ADC antibody targets a protein on cancer cells. The method of immunohistochemically staining a tissue sample using a dye linked to a diagnostic antibody, wherein the diagnostic antibody binds to the protein on the cancer cells in the tissue sample, and the protein is human epidermal growth factor receptor 2 (HER2), To acquire a digital image of the aforementioned tissue sample, To detect cancer cells in the aforementioned digital image, For each cancer cell, calculate the single-cell ADC score based on the staining intensity of the dye in the membrane, and Using the statistical calculations described in claim 17, predict the cancer patient's response to ADC based on the set of all single-cell ADC scores of the tissue sample. The method, including the method described above.

20. A method for identifying cancer patients for treatment with an anti-HER2 antibody-drug conjugate (ADC) comprising an ADC payload and an ADC antibody that targets a protein on cancer cells, The method of immunohistochemically staining a tissue sample from a cancer patient using a dye linked to a diagnostic antibody, wherein the diagnostic antibody binds to a protein on the cancer cells in the tissue sample, and the protein is human epidermal growth factor receptor 2 (HER2), To acquire a digital image of the aforementioned tissue sample, To detect cancer cells in the aforementioned digital image, For each cancer cell, calculate the single-cell ADC score based on the staining intensity of the dye in the membrane. A response score is generated by aggregating all single-cell ADC scores of the tissue sample using the statistical calculation described in claim 18, and If the response score exceeds the threshold, the cancer patient is identified as one who is likely to benefit from the administration of the ADC. The method, including the method described above.

Images