Search

JP-7854984-B2 - Cell localization signature and immunotherapy

JP7854984B2JP 7854984 B2JP7854984 B2JP 7854984B2JP-7854984-B2

Inventors

  • リー，ジョージ シー
  • フィッシャー，アンドリュー
  • グレイ，ファロン
  • エドワーズ，ロビン
  • エリー，スコット
  • コーエン，ダニエル エヌ
  • ウォイチク，ジョン ビー
  • バクシー，ビプル エイ
  • パンディア，ディンプル
  • トリリョ－ティノコ，ヒメナ
  • チェン，ベンジャミン ジェイ

Assignees

  • ブリストル－マイヤーズ スクイブ カンパニー

Dates

Publication Date
20260507
Application Date
20210831
Priority Date
20200831

Claims (20)

  1. A computer implementation method for identifying human subjects suitable for immunotherapy to treat human tumors, wherein the method is performed on data processing hardware that causes the operation to take place. Receiving histological images of tumor samples from human subjects; To perform image analysis of histological images and obtain image analysis results that show the abundance of CD8+ T cells in the tumor parenchyma and stroma; A computer implementation method comprising: processing the results of image analysis obtained by performing image analysis of histological images using a trained tumor topology classification model which includes a machine learning feature space which includes boundaries for multiple possible classifications of CD8 localization , to determine the classification of CD8 localization in a tumor sample from the boundaries of the machine learning feature space which includes boundaries for multiple possible classifications of CD8 localization; and generating recommendations for treatment options for a human subject based on the classification of CD8 localization in a tumor sample.
  2. The method according to claim 1, wherein the possible classifications of multiple CD8 localizations include inflammatory, desert, exclusion, and balanced types.
  3. The method according to claim 2, wherein the classification of CD8 localization in tumor samples includes exclusionary types.
  4. The operation is, This further includes determining whether the tumor sample exhibits a negative PD-L1 expression status. Generating recommendations for immunotherapy is based further on the determination that tumor samples exhibit negative PD-L1 expression status. The method according to claim 1.
  5. A trained tumor topology classification model, Receiving multiple training histological images of tumor samples from multiple patients; For each of the training histological images in multiple training histological images, image analysis of the training histological images is performed to obtain the abundance of CD8+ T cells in the tumor parenchyma and stroma in the training histological images; To train a tumor topology classification model using the results of image analysis and the abundance of CD8+ T cells in tumor parenchyma and stroma from each of multiple training histological images; The method according to claim 1, which is trained by a training process comprising: generating a machine learning feature space containing multiple classifications based on training; and identifying boundaries between possible classifications of multiple CD8 localizations.
  6. The method according to claim 5, comprising a graphical representation of the relationship between the abundance of CD8+ T cells and the percentage of stromal CD8+ T cells relative to the total number of T cells present in a histological image.
  7. The operation is, Applying polar coordinate transformation to the graph display to generate polar plots, It further includes, The method according to claim 6, wherein training a tumor topology classification model is further based on using polar plots.
  8. Each of the training histological images is obtained by at least one pathologist and includes a label that provides classification for CD8 localization in the training histological image; and training a tumor topology classification model includes validating the results from the machine learning feature space by comparing the labels of the multiple training histological images. The method according to claim 7.
  9. The method according to claim 1, wherein the treatment options include immunotherapy.
  10. The method according to claim 1, wherein the immunotherapy comprises anti-PD-1/PD-L1 antagonist therapy.
  11. The method according to claim 10, wherein the anti-PD-1/PD-L1 antagonist comprises an antibody ("anti-PD-1 antibody" or "anti-PD-L1 antibody") or an antigen-binding fragment thereof that specifically binds to a target protein selected from programmed death 1 (PD-1) or programmed death ligand 1 (PD-L1).
  12. The method according to claim 10, wherein the anti-PD-1/PD-L1 antagonist comprises an anti-PD-1 antibody.
  13. The method according to claim 12, wherein the anti-PD-1 antibody comprises nivolumab or pembrolizumab.
  14. The method according to claim 10, wherein the anti-PD-1/PD-L1 antagonist comprises an anti-PD-L1 antibody.
  15. The method according to claim 14 , wherein the anti-PD-L1 antibody comprises avelumab, atezolizumab, or durvalumab.
  16. Data processing hardware; and memory hardware that communicates with the data processing hardware and, when executed on the data processing hardware, stores instructions causing the data processing hardware to perform actions to identify human subjects suitable for immunotherapy to treat tumors of human subjects. The operation is, Receiving histological images of tumor samples from human subjects; To perform image analysis of histological images and obtain image analysis results that show the abundance of CD8+ T cells in the tumor parenchyma and stroma; A system comprising: processing the results of image analysis obtained by performing image analysis of histological images using a trained tumor topology classification model that includes a machine learning feature space containing boundaries for multiple possible classifications of CD8 localization , to determine the classification of CD8 localization in a tumor sample from the boundaries of the machine learning feature space for multiple possible classifications of CD8 localization; and generating recommendations for treatment options for a human subject based on the classification of CD8 localization in the tumor sample.
  17. The system according to claim 16, wherein the possible classifications of multiple CD8 localizations include inflammatory, desert, exclusion, and balanced types.
  18. The system according to claim 17, wherein the classification of CD8 localization in tumor samples includes exclusionary types.
  19. The operation is, This further includes determining whether the tumor sample exhibits a negative PD-L1 expression status. Generating recommendations for immunotherapy is based further on the determination that tumor samples exhibit negative PD-L1 expression status. The system according to claim 16.
  20. A trained tumor topology classification model, Receiving multiple training histological images of tumor samples from multiple patients; For each of the training histological images in multiple training histological images, image analysis of the training histological images is performed to obtain the abundance of CD8+ T cells in the tumor parenchyma and stroma in the training histological images; To train a tumor topology classification model using the results of image analysis and the abundance of CD8+ T cells in tumor parenchyma and stroma from each of multiple training histological images; The system according to claim 16, which is trained by a training process that includes generating a machine learning feature space containing multiple classifications based on training, and identifying boundaries between possible classifications of multiple CD8 localizations.

Description

Cross-reference of previously filed applications This PCT application claims the benefit of priority from U.S. Provisional Application No. 63/072,651, filed on 31 August 2020, which is incorporated herein by whole reference. Area of Disclosure: This disclosure provides a method for treating subjects affected by tumors using immunotherapy. Human cancers possess numerous genetic and epigenetic alterations, giving rise to nascent antigens that can be recognized by the immune system (Sjoblom et al., Science (2006) 314(5797):268-274). The adaptive immune system, composed of T and B lymphocytes, possesses potent anti-cancer capabilities, broad ability to respond to diverse tumor antigens, and sophisticated specificity. Furthermore, the immune system exhibits considerable plasticity and memory. By successfully leveraging all these characteristics of the adaptive immune system, immunotherapy is considered unparalleled among all forms of cancer treatment. Over the past decade, intensive efforts to develop specific immune checkpoint pathway inhibitors are beginning to yield new immunotherapeutic approaches to treating cancer. These include the development of antibodies that block the suppressive programmed death 1 (PD-1)/programmed death ligand 1 (PD-L1) pathway, such as nivolumab and pembrolizumab (formerly lambrolizumab, USAN Council Statement, 2013), which specifically bind to the PD-1 receptor, as well as atezolizumab, durvalumab, and avelumab, which specifically bind to PD-L1. The immune system and its response to immunotherapy have been shown to be complex. Furthermore, the effectiveness of anticancer drugs can vary based on the unique characteristics of each patient. Therefore, there is a need for targeted therapy strategies that identify patients who are more likely to respond to specific anticancer drugs, thereby improving clinical outcomes for patients diagnosed with cancer. Figure 1 illustrates exemplary images of tumor tissue samples of various classifications, which were subsequently imaged using CD8+ immunostaining according to an exemplary embodiment.Figure 2 is an exemplary diagram illustrating a method for an image analysis and machine learning-based approach to training a model for tumor topology classification, according to an exemplary embodiment.Figure 3 is another exemplary diagram illustrating a method for classifying tumor topologies using image analysis and machine learning-based approaches, according to an exemplary embodiment.Figure 4 is a flowchart illustrating the process for training a machine learning algorithm for classifying CD8 tumor topologies according to an exemplary embodiment.Figure 5 is a flowchart illustrating the process for classifying the CD8 tumor topology of histological images using a trained machine learning algorithm, according to an exemplary embodiment.Figure 6 is a block diagram of exemplary components of a device according to an exemplary embodiment.Figures 7A–7C are graphical representations of overall survival (OS) in patients with PD-L1-negative (PD-L1 expression <1%) melanoma (Figures 7A–7B) or urothelial carcinoma (Figure 7C) tumors after treatment with either anti-PD-1 antibody (Figures 7A and 7C) or a combination of anti-PD-1 antibody and anti-CTLA-4 antibody (Figure 7B). Patients were stratified by CD8 topology as having either an elimination-type CD8 phenotype (Figures 7A–7C), an inflammatory-type CD8 phenotype (Figures 7A–7C), or a desert-type CD8 phenotype (Figure 7C), as measured using immunohistochemistry and subsequent machine learning analysis as described herein. Patients at risk in each group are shown in Figures 7A–7B. Certain aspects of this disclosure relate to a method for treating a human subject with a tumor, comprising administering an anti-PD-1/PD-L1 antagonist to the subject, wherein a tumor sample obtained from the subject exhibits (i) an exclusionary CD8 localization phenotype and (ii) a negative PD-L1 expression status ("PD-L1 negative"). Other aspects of this disclosure relate to a method for identifying subjects suitable for immuno-oncology (I-O) therapy, such as anti-PD-1/PD-L1 antagonist therapy alone or in combination with anti-CTLA-4 antagonist therapy. In some aspects, the method comprises (i) measuring PD-L1 expression in tumor samples obtained from subjects, and (ii) measuring CD8 expression in tumor samples, where CD8 expression is measured by immunostaining and imaging, and then classifying localized CD8 expression in tumor samples using a machine learning algorithm. In some aspects, the method further comprises administering an anti-PD-1/PD-L1 antagonist to subjects identified as having tumor samples exhibiting (i) an exclusionary CD8 localization phenotype and (ii) a negative PD-L1 expression status ("PD-L1 negative"). In some embodiments, the method further includes administering an additional anticancer agent. In some embodiments, the method further includes administering an anti-CTLA-4 antagonist. I. Terminology To make