Methods of predicting treatment responses in multiple myeloma using spatial signatures

Abstract

The present disclosure relates to computer-implemented methods for classifying cell neighborhoods in a bone marrow sample obtained from a subject suffering from multiple myeloma (MM) and for defining significant cell neighborhoods in that sample that are associated with treatment outcomes in subjects with MM who receive a T cell-redirecting therapy. Image processing systems that can be used to analyze bone marrow samples and identify cell neighborhoods are also provided. Moreover, methods for predicting a treatment response to a T cell-redirecting therapy in a subject suffering from MM are disclosed.

Description

METHODS OF PREDICTING TREATMENT RESPONSES IN MULTIPLE MYELOMA USING SPATIAL SIGNATURESCROSS-REFERENCE TO RELATED APPLICATION

[0001] This application claims benefit of U.S. Provisional Application No. 63 / 716,000 filed November 4, 2024, the entire contents of which is incorporated herein by reference.FIELD

[0002] The present disclosure relates to computer-implemented methods for classifying cell neighborhoods in a bone marrow sample obtained from a subject suffering from multiple myeloma (MM) and for defining significant cell neighborhoods in that sample that are associated with treatment outcomes in subjects with MM who receive a T cellredirecting therapy. Image processing systems that can be used to analyze bone marrow samples and identify cell neighborhoods are also provided. Moreover, methods for predicting a treatment response to a T cell-redirecting therapy in a subject suffering from MM are disclosed.BACKGROUND

[0003] T cell-redirecting treatments such as chimeric antigen receptor T cell (CAR- T) therapy has shown remarkable response rates in patients with relapsed / refractory multiple myeloma (RRMM). However, patients ultimately experience disease progression partially due to tumor-intrinsic alterations and an immunosuppressive microenvironment. Previous research primarily utilized sequencing and flow cytometry to investigate relapse mechanisms, but the influence of bone marrow (BM) spatial organization remains underexplored.SUMMARY

[0004] Provided are computational methods of identifying cellular neighborhoods in the bone marrow (BM) that are indicative of the prognostic outcome for patients with RRMM. Specifically, by identifying immune cells that included various CD4+ and CD8+ T cell subtypes including CD4+ and CD8+ memory T cells and granzyme B-positive (GZMB+) CD8+ T cells as well as immunosuppressive cells such M2-like macrophages and myeloid- derived suppressor cells (MDSCs), and myeloma cells within the BM microenvironment, itwas possible to identify cell neighborhoods that are associated with the success or failure of T cell-redirecting treatments such as CAR-T therapy. The computational methods described herein enable the reliable identification of such cell neighborhoods and as such can provide spatial signatures that are useful for prognostication. They involve classifying each cell found in a section of a BM sample obtained from a subject based on a feature vector of the cell and reference feature vectors defining each cell neighborhood. The feature vector of each cell is indicative of a proportion of the various cell types including the above-mentioned immune cells found in a region surrounding the cell.

[0005] In one aspect, there is provided a computer-implemented method of classifying cell neighborhoods in a bone marrow sample obtained from a subject suffering from multiple myeloma (MM), the method comprising: receiving an image of a plurality of cells in the bone marrow sample; receiving label data for the image, wherein the label data is indicative of a cell type of each cell of the plurality of cells and wherein the cell type is selected from the list comprising: adipocytes; B cells; exhausted CD4+ T cells; CD4+ memory T cells; CD8+ GZMB+ T cells; exhausted CD8+ T cells; CD8+ memory T cells; endothelial cells; Ml -like macrophages; M2 -like macrophages; myeloid-derived suppressor cells (MDSCs); megakaryocytes; monocyte-derived dendritic cells (moDCs); monocytes; myeloma cells; natural killer cells; regulatory CD4+ T cells; naive CD4+ T cells; and naive CD8+ T cells, segmenting the image to produce a segmented image identifying the location of each cell of the plurality of cells in the image; determining, based on the segmented image and the label data, a feature vector for each cell of the plurality of cells, wherein the feature vector is indicative of a proportion of each cell type found in a region surrounding the cell, wherein the region surrounding the cell extends a first distance from the cell; receiving a definition of one or more cell neighborhoods, wherein each of the one or more cell neighborhoods is defined by a reference feature vector; and classifying each cell of the plurality of cells based on the feature vector of the cell and the one or more reference feature vectors, wherein each cell is classified as belonging to one of the one or more cell neighborhoods, or belonging to a remainder class.

[0006] The described methods have been used to identify cell neighborhoods that are indicative of the prognostic outcome for subjects with MM undergoing T cell-redirecting therapy.

[0007] Accordingly, in another aspect, there is provided a method of predicting a treatment response to T cell-redirecting therapy in a subject suffering from multiple myeloma (MM), the method comprising: (A) receiving the proportions determined in the method of claim 2 for a bone marrow sample obtained from the subject; (B) comparing, for each cell neighborhood, the proportion to a corresponding reference value; and (C) predicting, based on the comparing, the treatment response of the subject to immunotherapy.

[0008] In a further aspect, there is provided a computer-implemented method of defining one or more significant cell neighborhoods for responsiveness of subjects suffering from multiple myeloma (MM) to a T cell-redirecting therapy, the method comprising: receiving a plurality of images, each image being an image of a plurality of cells in a bone marrow sample obtained from one of a plurality of subjects suffering from multiple myeloma (MM); label data for the plurality of images, wherein the label data is indicative of a cell type of each cell of the plurality of cells and wherein the cell type is selected from the list comprising: adipocyte cells; B cells; exhausted CD4+ T cells; CD4+ memory T cells; CD8+ GZMB+ T cells; exhausted CD8+ T cells; CD8+ memory T cells; endothelial cells; Ml-like macrophages; M2-like macrophages; MDSCs; megakaryocytes; monocyte-derived dendritic cells (moDCs); monocytes; myeloma cells; natural killer cells; regulatory CD4+ T cells; naive CD4+ T cells; and naive CD8+ T cells, segmenting the images to produce a plurality of segmented images, each segmented image identifying the location of each cell of the plurality of cells in the corresponding image; determining, for each of the plurality of images and based on the corresponding segmented image and the label data, a feature vector for each cell of the plurality of cells, wherein the feature vector is indicative of a proportion of each cell type found in a region surrounding the cell, wherein the region surrounding the cell extends a first distance from the cell; clustering the feature vectors to generate K cell neighborhoods; determining the proportion of each cell neighborhood of the K cell neighborhoods for each subject; determining a relationship between the proportion of each cell neighborhood and a treatment response of the subjects to T cell-redirecting therapy; and identifying one or more significant cell neighborhoods, wherein a cell neighborhood is identified as a significant cell neighborhood if it has a statistically significant relationship with the treatment response of the subjects; and outputting a definition of each of the one or more significant cell neighborhoods.

[0009] In a further aspect, there is provided an image processing system comprising: a processor, configured to receive: an image of a plurality of cells in a bone marrow sample obtained from a subject suffering from multiple myeloma (MM); a definition of one or more cell neighborhoods, wherein each of the one or more cell neighborhoods is defined by a reference feature vector; and label data for the image, wherein the label data is indicative of a cell type of each cell of the plurality of cells and wherein the cell type is selected from the list comprising: adipocytes; B cells; exhausted CD4+ T cells; CD4+ memory T cells; CD8+ GZMB+ T cells; exhausted CD8+ T cells; CD8+ memory T cells; endothelial cells; Ml-like macrophages; M2-like macrophages; MDSCs; megakaryocytes; moDCs; monocytes; myeloma cells; natural killer cells; regulatory CD4+ T cells; naive CD4+ T cells; and naive CD8+ T cells, the processor being configured to: segment the image to produce a segmented image identifying the location of each cell of the plurality of cells in the image; determine a feature vector based on the segmented image and the label data, wherein the feature vector is indicative of a proportion of each cell type found in a region surrounding the cell, and wherein the region surrounding the cell extends a first distance from the cell; classify each cell of the plurality of cells based on the feature vector of the cell and the one or more reference feature vectors, wherein each cell is classified as belonging to one of the one or more cell neighborhoods, or belonging to a remainder class.

[0010] In a further aspect, there is provided an image processing system comprising: a processor, configured to receive: a plurality of images, each image being an image of a plurality of cells in a bone marrow sample obtained from one of a plurality of subjects suffering from multiple myeloma (MM); and label data for the plurality of images, wherein the label data is indicative of a cell type of each cell of the plurality of cells and wherein the cell type is selected from the list comprising: adipocytes; B cells; exhausted CD4+ T cells; CD4+ memory T cells; CD8+ GZMB+ T cells; exhausted CD8+ T cells; CD8+ memory T cells; endothelial cells; Ml-like macrophages; M2-like macrophages; MDSCs; megakaryocytes; moDCs; monocytes; myeloma cells; natural killer cells; regulatory CD4+ T cells; naive CD4+ T cells; and naive CD8+ T cells, the processor being configured to: segment each of the plurality of images to produce a plurality of segmented images, each segmented image identifying the location of each cell of the plurality of cells in the corresponding image; for each cell in each image, determine a feature vector based on the corresponding segmented image and the label data, wherein the feature vector is indicative of a proportion of each cell type found in a region surrounding the cell, and wherein theregion surrounding the cell extends a first distance from the cell; cluster the cells into K cell neighborhoods based on the feature vectors; determine the proportion of each cell neighborhood of the K cell neighborhoods for each subject; determine, for each subject, a relationship between the proportion of each cell neighborhood and a treatment response of the subjects to T cell-redirecting therapy; and identify one or more significant cell neighborhoods, wherein a cell neighborhood is identified as a significant cell neighborhood if it has a statistically significant relationship with the treatment response of the subjects; and output a definition of each of the one or more significant cell neighborhoods.

[0011] In a further aspect, there is provided a processing system configured to perform the methods provided herein.

[0012] In a further aspect, there is provided a non-transitory computer-readable storage medium having stored thereon computer readable code configured to cause a computer to perform the methods provided herein when the code is run on the computer.

[0013] Other features, objects, and advantages are apparent in the detailed description, drawings and embodiments that follow. It should be understood, however, that the detailed description, the drawings, and the embodiments, while indicating embodiments, are given by way of illustration only, not limitation. Various changes and modifications will become apparent to those skilled in the art.BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Further description, by way of example, is provided with reference to the following drawings.

[0015] FIG. 1A shows progression-free survival (PFS) for 50 RRMM patients following treatment with the BCMA-directed CAR-T cell therapies idecabtagene vicleucel (ide-cel; 39 patients) or ciltacabtagene autoleucel (cilta-cel; 11 patients).

[0016] FIG. IB shows overall survival (OS) for 50 RRMM patients following treatment with the BCMA-directed CAR-T cell therapies ide-cel (39 patients) or cilta-cel (11 patients).

[0017] FIG. 2 shows PFS of patients following treatment with BCMA-directed CAR- T cells. The proportions of the pre-treatment cell neighborhoods (preCNs) at baseline (i.e. before CAR-T cell therapy) which could be linked to PFS are shown. FIG. 2A: preCNO; FIG.2B: preCNl l; FIG. 2C: preCN8; FIG. 2D: preCN9. Patients having high proportions of preCNs 0, 11 and 8 had reduced PFS survival, while patients having high proportions of preCN9 had increased PFS survival. PreCN areas were normalized, and patients were separated into high or low groups based on the median value for each preCN. PFS was analyzed using the Kaplan-Meier method. A log-rank test was used for statistical analysis.

[0018] FIG. 3A shows the proportions of different CD8+ T cell populations across preCNs. FIG. 3A-1 : pSTAT- / + CD8+ GZMB+ memory T cells; FIG. 3A-2: pSTATl- / + CD8+ memory T cells; FIG. 3A-3: naive CD8+ T cells; FIG. 3A-4: exhausted CD8+ T cells. Black diamonds denote preCNs 0, 8, 11 and 9.

[0019] FIG. 3B shows the proportions of different CD4+ T cell populations across preCNs. FIG. 3B-1 : pSTATl- / + non-Thl7 CD4+ memory T cells; FIG. 3B-2: pSTATl- / + Thl7 CD4+ memory T cells; FIG. 3B-3: naive CD4+ T cells; FIG. 3B-4: exhausted CD4+ T cells. Black diamonds denote preCNs 0, 8, 11 and 9.

[0020] FIG. 3C shows the proportion of Tregs across preCNs. Black diamonds denote preCNs 0, 8, 11 and 9.

[0021] FIG. 4 shows the proportion of the top 10 cell types that myeloma cells are co-localized with in preCNs 0, 11, 8 and 9.

[0022] FIG.5 shows chord diagrams visualizing interactions between myeloma cells, known immunosuppressive cells (MDSCs, M2-like macrophages) and CD4+ T cell subpopulations (non-Thl7 CD4+ memory T cells, Thl7 CD4+ memory T cells, naive CD4+ T cells, exhausted CD4+ T cells and regulatory T cells) for preCNO (A), preCNl l (B), preCN8 (C) and preCN9 (D).

[0023] FIG. 6 shows chord diagrams visualizing interactions between myeloma cells, known immunosuppressive cells (MDSCs, M2-like macrophages) and CD8+ T cell subpopulations (CD8+ memory T cells, GZMB+ CD8+ memory T cells, naive CD8+ T cells and exhausted CD8+) for preCNO (A), preCNl 1 (B), preCN8 (C) and preCN9 (D).

[0024] FIG. 7 shows box plots showing the distance between each myeloma cell and the nearest M2-like macrophage (A), regulatory T cell (B), CD8+ memory T cell (C), GZMB+ CD8+ memory T cell (D), CD4 memory T cell (E), MDSC (F), naive T cell (G), exhausted CD8+ T cell (H) and exhausted CD4+ T cell (I) in each of preCNs 0, 11, 8 and 9.

[0025] FIG. 8 shows the proportion of Ki-67+ (A) or PD-L1+ (B) myeloma cells across preCNs.

[0026] FIG. 9 shows PFS of patients, following treatment with BCMA-directed CAR-T cells, having high or low proportions of Ki-67+ (A) or PD-L1+ (B) myeloma cells at baseline (i.e. before CAR-T cell therapy). Patients were separated into high or low groups based on the median normalized value for each myeloma cell type. PFS was analyzed using the Kaplan-Meier method. Log-rank test was used for statistical testing.

[0027] FIG. 10 is a flowchart illustrating an example of a method of defining one or more significant cell neighborhoods.

[0028] FIG. 11 is a flowchart illustrating an example of a method of classifying cell neighborhoods and predicting a treatment response of a subject.

[0029] FIG. 12 shows a computer system in which a processing system according to an embodiment is implemented.DETAILED DESCRIPTION

[0030] Unless otherwise defined herein, technical and scientific terms have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs and as commonly used in the art to which this application belongs. Exemplary methods and materials are described below, although methods and materials similar or equivalent to those described herein can also be used in the practice or testing of the present disclosure. In case of conflict, the present specification, including definitions, will control. To facilitate ready understanding, certain terms used herein are first defined below.

[0031] Unless otherwise required by context, singular terms shall include pluralities and plural terms shall include the singular. Thus, the singular forms “a”, “an”, and “the” include plural referents unless the context clearly dictates otherwise. For example, “a T cell” is understood to represent one or more T cell(s) or a population of T cells. As such, the terms “a” (or “an”), “one or more”, and “at least one” can be used interchangeably herein.

[0032] “And / or” is to be taken as specific disclosure of each of the two specified features or components with or without the other. Thus, the term “and / or” as used in a phrase such as “A and / or B” herein is intended to include “A and B”, “A or B”, “A” (alone), and “B” (alone). Likewise, the term "and / or" as used in a phrase such as “A, B, and / or C” is usedinterchangeably with “A and / or B and / or C” and is intended to encompass each of the following aspects: A, B, and C; A, B, or C; A or C; A or B; B or C; A and C; A and B; B and C; A (alone); B (alone); and C (alone).

[0033] The words “have” and “comprise”, or variations such as “has”, “having”, “comprises”, or “comprising” will be understood to imply the inclusion of a stated integer or group of integers but not the exclusion of any other integer or group of integers. It is further understood that wherever aspects are described herein with the language “comprising” or “having” or grammatical equivalents thereof, otherwise analogous aspects described in terms of “consisting of’ and / or “consisting essentially of’ are also provided. In other words, if a composition comprising A, B and C is recited, a composition consisting essentially of A, B and C is also contemplated as is a composition consisting of A, B and C.

[0034] The term “about” refers to an interval of accuracy that a person skilled in the art will understand to still ensure the technical effect of the feature in question. The term indicates a deviation from the indicated numerical value of ±10, ±5%, or ±1%.

[0035] The term “substantially” refers to the qualitative condition of exhibiting total or near-total extent or degree of a characteristic or property of interest. One of ordinary skill in the biological arts will understand that biological and chemical phenomena rarely, if ever, go to completion and / or proceed to completeness or achieve or avoid an absolute result. The term “substantially” is therefore used herein to capture the potential lack of completeness inherent in many biological and chemical phenomena.

[0036] The term “exhausted” in relation to T cells refers to T cells that are losing or have lost their functional capability. Thus, the term "exhausted" may be used interchangeably with "dysfunctional". For example, exhausted T cells may have an impaired ability to respond to infected cells or tumors. In some instances, the term "exhausted" may more specifically refer to cells expressing the markers CD3, CD45RO, PD-1, and TIM-3. Additionally or alternatively, exhausted T cells may express the marker TIGIT.

[0037] The term “co-localization” refers to the spatial association between two or more cells. For example, two cells may border each other. Co-localization can be determined by calculating the distance between cell contours in an image. Cells whose contours are within 0-15 pm, e.g., 10 pm, of each other, are considered to co-localize. Physical proximity between cells may imply that the cells are interacting.

[0038] The term "spatial signature" refers to a distinct cellular composition and pattern of cellular interactions within a cell neighborhood of a BM sample. In some instances, the term "spatial signature" may more specifically refer to the col-localization between an immune cell and a myeloma cell within a cellular neighborhood.

[0039] The term "baseline spatial signature" refers to a spatial signature in a BM sample that is taken from a subject prior to starting a T cell-redirecting therapy, such as CAR T-cell therapy.

[0040] Generally, techniques of cell and tissue culture, molecular biology, virology, immunology, microbiology, genetics, analytical chemistry, synthetic organic chemistry, medicinal and pharmaceutical chemistry, and protein and nucleic acid chemistry and hybridization described herein are those well-known and commonly used in the art. Enzymatic reactions and purification techniques are performed according to manufacturer’s specifications, as commonly accomplished in the art or as described herein. Further, nomenclature used in connection with these technology areas herein is as commonly used in the art as can be seen by reference to, e.g., http / / www.informatics.j ax.org / mgihome / nomen / gene. shtml and https: / / www.genenames.org / .

[0041] All publications and other reference materials referenced herein are hereby incorporated by reference in their entirety. Although a number of documents are cited herein, this citation does not constitute an admission that any one of these documents forms part of the common general knowledge in the art.Sample preparation

[0042] The methods described here are performed on images of a plurality of cells in a bone marrow (BM) sample obtained from a subject suffering from multiple myeloma (MM). Before the BM sample can be imaged, various processing steps are undertaken to prepare the sample for staining with a set of antibodies that enable identification of various cell types within the plurality of cells.Preparation o f bone marrow samples

[0043] The BM sample may be fixed using formalin and paraffin and stored prior to being sectionized to glass slides for antibody staining. To prepare such a formalin-fixedparaffin-embedded (FFPE) BM sample for antibody staining, the glass slide comprising the sample is typically deparaffinized, rehydrated, and then subjected to antigen-retrieval.

[0044] To deparaffinize the BM sample, it can be heated. The sample may be heated for about 45 minutes to about 3 hours, e.g. for about 1 hour until the paraffin coat is melted. The sample may be heated at a temperature of about 58°C to about 64°C, e.g. about 62°C. To remove the paraffin, the sample may be dewaxed using an organic solvent, such as xylene. The sample may be dewaxed for about 20 minutes to about 40 minutes, e.g., for about 30 minutes.

[0045] The deparaffinized BM sample can then be rehydrated. The sample may be rehydrated using an alcohol gradient series, in which the sample is incubated in solutions with progressively decreasing alcohol content. The sample may be incubated in each solution for about 5 minutes. The alcohol gradient series may comprise aqueous solutions comprising about 100%, about 95%, about 90%, about 80%, about 70%, about 60%, about 50%, about 40%, about 30%, about 20%, about 10% and / or about 0% alcohol (e.g., ethanol). For example, the alcohol gradient series may comprise solutions comprising 100%, 95%, 80%, 70% and 0% ethanol in deionized water.

[0046] Antigen-retrieval and permeabilization can be performed on the hydrated BM sample. For antigen-retrieval, the sample may be heated in an antigen-retrieval solution, such as Target Retrieval Solution (pH 9) from Agilent (Santa Clara, CA). The sample may be heated at about 94°C to about 98°C, e.g. at about 96°C. The sample may be heated in the antigen-retrieval solution for about 35 minutes to about 45 minutes, e.g. for about 40 minutes.

[0047] For permeabilization, a permeabilization buffer may be applied to the sample.The permeabilization buffer may comprise a detergent, e.g., polysorbate 20, and / or a surfactant, e.g. Triton X-100. The buffer may be TBS-T comprising 0.2% Triton-X 100. The sample may be soaked in the buffer twice, for about 10 minutes each time.

[0048] Following permeabilization, the sample may be subjected to a blocking step. Blocking reduces non-specific antibody binding. Blocking may be performed by incubating the sample in a blocking buffer. An example of a suitable blocking buffer is 1% Human Trustain FcX (Biolegend #422301) in 3% BSA / TBS-T. The sample may be incubated in the blocking buffer for at least about 45 minutes, e.g., for about 1 hour.Cell type identification

[0049] The described methods receive an image of a plurality of cells in the BM sample as well as label data for the image. The label data is indicative of a cell type of each cell of the plurality of cells. The cell type comprise adipocytes, B cells, exhausted CD4+ T cells, CD4+ memory T cells, CD8+ GZMB+ T cells, exhausted CD8+ T cells, CD8+ memory T cells, endothelial cells, Ml-like macrophages, M2-like macrophages, myeloid-derived suppressor cells (MDSCs), megakaryocytes, monocyte-derived dendritic cells (moDCs), monocytes, myeloma cells, natural killer cells, regulatory CD4+ T cells, naive CD4+ T cells, and naive CD8+ T cells.

[0050] In order to identify cells and their cell types within the processed BM sample, the sample is stained with a panel of antibodies. A suitable panel of antibodies may include antibodies that specifically bind to aSMA, CD14, TIM-3, CD16, CD163, CDl lb, CD31, FoxP3, CD4, CD68, CD20, CD8, CD56, PD-1, CD138, Granzyme B (GZMB), CD3, CCR7, HLA-DR, and CD45RO. This panel can be used to identify the above-mentioned cell types as shown in Table 1.

[0051] It is understood that a skilled person is able to identify alternative panels of antibodies that can be used in a similar manner to provide the label data for the cell types that are analyzed by the disclosed methods.Table 1. Exemplary markers for cell type identification

[0052] In addition, a suitable panel of antibodies may include antibodies that specifically bind to a protein in the nucleus (e.g., histone H3) and in the cell membrane (e.g., Na+ / K+ ATPase, CD45, and / or MHC class I, namely HLA-A, HLA-B, and HLA-C) to identify cell shapes and contours. For example, Standard Biotool’s cell segmentation kit includes antibodies ICSK1, ICSK2, and ICSK3 that bind to markers expressed on the cell membrane of human cells. Having information about the cell contours is useful to determine co-localization of cells and potential interactions between them. Such information can be employed to identify spatial signatures of immune cells and myeloma cells that have prognostic value.

[0053] The panel of antibodies may also include one or more antibodies that can be used to further characterize particular cell types, e.g., by their function, or, in the case of myeloma cells, their malignancy. Such one or more antibodies may specifically bind to Granzyme K, e.g., to identify Granzyme K-positive (GZM K+) CD8+ memory T cells; CCR4; CCR6; CD127; CD45, e.g., to identify nucleated haemopoietic cells; CD27, e.g., to further characterize B cells (e.g., as memory B cells); pSTATl, e.g., to determine T cellactivation; TIGIT, e.g., to further characterize dysfunctional or exhausted T cells; and Ki-67 and PD-L1, e.g., to determine the malignancy of myeloma cells.

[0054] To reach desired concentrations, the antibodies may be diluted in blocking buffer. The samples may be incubated with the antibodies overnight, e.g., at 4°C.

[0055] Following staining, the samples are typically washed, e.g. with deionized water, and air-dried prior to imaging.

[0056] Typically, each of the antibodies in the panel are conjugated to a distinct detectable moiety such as a metal tag or fluorescent dye. Different images of the BM sample that include the staining patterns resulting from each antibody provide the label data that can be used to label each cell in the image.Image acquisition

[0057] Various imaging methods may be used to acquire an image of a plurality of cells in the processed and stained BM sample. An exemplary method used herein is imaging mass cytometry (IMC; Giesen et al., Nat Methods. 2014; 11, 417-22).

[0058] Other suitable imaging methods include multiplexed ion beam imaging (MIBI; Angelo et al., Nat Med. 2014; 20, 436-442), cyclic immunofluorescence (CyCIF; Lin et al., Nat Commun. 2015; 6, 8390), tissue-based cyclic immunofluorescence (t-CyCIF; Lin et al., eLife. 2018; 7, e31657), iterative indirect immunofluorescence imaging (4i; Gut et al., Science. 2018; 361, 7042), co-detection by indexing (CODEX; Goltsev et al., Cell. 2018; 174, 968-981), and multiplex immunofluorescence (MxIF; Gerdes et al., PNAS. 2013; 110, 11982-11987; McKinley et al., JCI Insight. 2017; 2, e93487). The skilled person will be aware of other imaging techniques that can be used to obtain an image of a plurality of cells in the processed and stained BM sample.Imaging mass cytometry

[0059] For example, imaging mass cytometry (IMC) may be used to perform highly multiplexed imaging and is suited to profiling selected areas of tissues across many bone marrow biopsy samples.

[0060] For image acquisition, a microscopy slide comprising the processed and stained BM sample is mounted on a precise motor-driven stage inside the ablation chamber of an IMC instrument. A high-energy UV laser is focused on the tissue, and each individuallaser shot ablates tissue from an area of roughly 1 gm2. The energy of the laser is absorbed by the tissue, resulting in vaporization followed by condensation of the ablated material.

[0061] The ablated material from each laser shot is transported in the gas phase into the plasma of the mass cytometer, where first atomization of the particles, and then ionization of the atoms, occurs. The ion cloud is then transferred into a vacuum, and all ions below a mass of 80 m / z are filtered using a quadrupole mass filter. The remaining ions (mostly those used to tag antibodies) are analyzed in a time-of-flight mass spectrometer to obtain an accumulated mass spectrum from all ions that correspond to a single laser shot. This spectrum can be regarded as the information underlying a 1 pm2pixel.

[0062] With repetitive laser shots (e.g., at 200 Hz) and a simultaneous lateral sample movement, a tissue can be ablated pixel by pixel. Ultimately, an image will be reconstructed from each pixel mass spectrum.

[0063] IMC utilizes metal-tagged antibodies to detect proteins or other metal-tagged molecules in biological samples. Antibody-metal conjugation may be performed using commercially available kits, such as the Maxpar labeling kit from Standard BioTools (South San Francisco, CA). Suitable metal tags include 141Pr, 142Nd, 143Nd, 144Nd, 145Nd, 146Nd, 147Sm, 149Sm, 150Nd, 151Eu, 152Sm, 153Eu, 154Sm, 155Gd, 156Gd, 158Gd, 159Tb, 160Gd, 161Dy, 162Dy, 163Dy, 164Dy, 165Ho, 166Er, 167Er, 168Er, 169Tm, 170Er, 171Yb, 172Yb, 173Yb, 174Yb, 175Lu, 176Yb, 195Pt, and 196Pt.Image processing

[0064] Once an image of a plurality of cells and the label data from the antibody staining of the BM sample have been acquired, the provided computational methods segment the image to produce a segmented image identifying the location of each cell of the plurality of cells in the image. Based on the segmented image and the label data, the methods determine feature vectors for each cell of the plurality of cells. A feature vector is indicative of a proportion of each cell type found in a region surrounding the cell.

[0065] In one aspect, a computer-implemented method of defining one or more significant cell neighborhoods for responsiveness of subjects suffering from multiple myeloma (MM) to a T cell-redirecting therapy is provided. The method comprises (i) receiving a plurality of images, each image being an image of a plurality of cells in a bone marrow sample obtained from one of a plurality of subjects suffering from multiple myeloma(MM), (ii) receiving label data for the plurality of images, wherein the label data is indicative of a cell type of each cell of the plurality of cells and wherein the cell type is selected from the list comprising adipocyte cells; B cells; exhausted CD4+ T cells; CD4+ memory T cells; CD8+ GZMB+ T cells; exhausted CD8+ T cells; CD8+ memory T cells; endothelial cells; Ml-like macrophages; M2-like macrophages; myeloid-derived suppressor cells (MDSCs); megakaryocytes; monocyte-derived dendritic cells (moDCs); monocytes; myeloma cells; natural killer cells; regulatory CD4+ T cells; naive CD4+ T cells; and naive CD8+ T cells, (iii) segmenting the images to produce a plurality of segmented images, wherein each segmented image identifies the location of each cell of the plurality of cells in the corresponding image; (iv) determining, for each of the plurality of images and based on the corresponding segmented image and the label data, a feature vector for each cell of the plurality of cells, wherein the feature vector is indicative of a proportion of each cell type found in a region surrounding the cell, wherein the region surrounding the cell extends a first distance from the cell; (v) clustering the feature vectors to generate K cell neighborhoods; (vi) determining, for each subject, a relationship between the proportion of each cell neighborhood for the subjects and a treatment response of the subjects to T cell-redirecting therapy; and (vii) identifying one or more significant cell neighborhoods, wherein a cell neighborhood is identified as a significant cell neighborhood if it has a statistically significant relationship with the treatment response of the subjects; and (vii) outputting a definition of each of the one or more significant cell neighborhoods.Defining significant cell neighborhoods

[0066] An exemplary method 100 is described in detail below, with reference to FIG. 10. In this method, data is gathered from a group of subjects and used to identify cell neighborhoods having predictive power for treatment outcomes (significant cell neighborhoods).

[0067] In step 105, a processor receives a plurality of images. Each image is an image of a plurality of cells in a bone marrow sample. A bone marrow sample, and corresponding image, is obtained from each of 50 subjects suffering from multiple myeloma.

[0068] In step 110, the processor receives label data for the plurality of images. The label data is data that indicates the cell type of each cell in each image. In the present example, the images are obtained using IMC as described above, and so each image is encoded with mass spectrum data. As explained above, each pixel of an image constructed using IMCcontains mass spectrum data that indicates which markers were present in a 1 pm2area of the sample. As the type and quantity of markers taken up by a cell is a function of the type of cell, the mass spectrum data for the part of an image corresponding to one cell indicates the cell type of that cell, and serves as label data.

[0069] In general, the label data indicates that each cell is one of: an adipocytes; a B cell; an exhausted CD4+ T cell; a CD4+ memory T cell; a CD8+ GZMB+ T cell; a exhausted CD8+ T cell; a CD8+ memory T cell; an endothelial cell; a Ml-like macrophage; a M2-like macrophage; a myeloid-derived suppressor cell (MDSC); a megakaryocyte; a monocyte- derived dendritic cell (moDC); a monocyte; a myeloma cell; a natural killer cell; a regulatory CD4+ T cell; a naive CD4+ T cell; and a naive CD8+ T cell.

[0070] In the present example, the label data is not indicative of adipocytes.

[0071] In step 115, the processor segments each of the plurality of images, producing a plurality of segmented images. Each segmented image contains a segmentation mask that identifies the location of each cell in the image. In some examples, some of the segmenting can be performed manually by a human. For example, pixels belonging to megakaryocytes may be identified manually, based on the occurrence of the marker CD31 in the image.

[0072] In the present example, the segmenting is based on the label data. For example, markers such as NA+ / K+ ATPase, CD45, and / or MHC class I in the label data may be used to identify cell membranes, and a marker such as histone H3 may be used to identify the nucleus. Standard Biotool’s cell segmentation kit includes antibodies ICSK1, ICSK2, and ICSK3 that bind to markers expressed on the cell membrane of human cells, and one or more (e.g., two or more) of these markers in the label data can likewise be used to identify cell membranes. Based on this information, the images can be segmented to identify which pixels belong to cells in each image. In other examples, such as where the images are optical images obtained through magnification, the segmenting may be based on other known image processing methods.

[0073] As mentioned above, in the present example the label data does not identify adipocytes. The locations of adipocytes in the images are estimated as described below.

[0074] For ease of explanation, the method of estimating the location of adipocytes will be described for a single image. In step 120, the segmented image is smoothed byapplying a Gaussian blur, to remove noise. In other examples a different smoothing method may be used.

[0075] In step 121, the smoothed segmented image is binarized. By this it is meant that the mask in the segmented image is converted to a binary mask, where foreground pixels corresponding to cells are represented by a first value (such as 1) and background pixels are represented by a second value (such as 0). As the label data does not indicate adipocytes, any adipocytes present in the sample will be background pixels of the image.

[0076] In step 122, the binarized segmented image is inverted to produce an inverted segmented image. The inverting swaps or otherwise changes the first and second values mentioned above, such that pixels identified as foreground pixels in the segmented image are now identified as background pixels in the inverted segmented image, and vice versa.

[0077] In step 123, individual objects are identified in the inverted segmented image by performing watershed segmentation. In other examples, individual objects can be identified using other known techniques, such as edge detection.

[0078] In step 124, the objects are filtered by size. This size filtering excludes objects that are too large or too small to be one (or a group of) adipocytes. The filtering selects obj ects that are in the correct size range to be one (or a group of) adipocytes. For example, the filtering excludes intercellular spaces and areas in the image in which no tissue is present. In the present example, objects having an area in the range 5 pm2to 10000 pm2are selected, with all other objects being excluded. In other examples, other ranges may be used. In some examples, different ranges may be applied for each inverted segmented image, to account for differences in tissue structure between samples.

[0079] In step 125, the selected objects are eroded. The erosion of the selected objects breaks thin connections within objects, separating those objects into two or more constituent objects. The constituent objects are identified as being distinct (separate) objects. The erosion enables adipocytes that are in contact with each other through a bottleneck (a small point of contact) to be separated into distinct objects.

[0080] In step 126, the eroded objects are dilated. The dilation restores the objects to their original size. Accordingly, the method can identify objects in the correct size range to be adipocytes and can separate at least some contacting adipocytes into distinct objects.

[0081] In step 127, the (now dilated) eroded objects are added to the segmented image. By this it is meant that the data of the pixels in the segmented image that correspond to the locations of the eroded objects are replaced by data of the corresponding pixels of the inverted segmented image. In this way, the segmented image now additionally identifies the location of adipocytes in the sample.

[0082] In step 128, the label data for the image is updated to indicate that the eroded objects added to the image are adipocyte cells.

[0083] Steps 120-126 and 127 can be performed as part of a pre-processing operation, performed by a separate device before the processor receives the images and label data. Accordingly, the label data received by the processor may include data relating to adipocyte cells.

[0084] In step 130, the processor determines a feature vector for each cell in each of the images. The feature vectors are multidimensional vectors that represent each cell in a feature space. The cell being described by a feature vector will be referred to herein as the ‘central’ cell, because the feature vector for the central cell identifies the composition of the cells in a region surrounding the central cell. In the present example, the region surrounding the central cell is a region that extends 10pm from the surface of the central cell (that is, the outer edge of the central cell). Any cell that overlaps with the region (that it, has at least one pixel within the region) is within the region and will contribute to the feature vector for the central cell. All other cells are excluded from consideration. By way of example, the feature vector for a central cell having two myeloma cells and three natural killer cells within the region surrounding the cell could be [. . ., 2, 3,. . .].

[0085] In step 140, the processor clusters the cells in the images into clusters (known as cell neighborhoods) based on the feature vectors of the cells. The clustering is implemented by K means clustering, with K set to 12 in the present example. In other examples K can be set to a different value, such as any integer in the range 10-20 and / or other known clustering methods can be used to cluster the cells. In some examples, the clustering may be performed using mini-batch K means clustering.

[0086] In step 150, the proportion of each cell neighborhood is determined for each subject. By proportion, it is meant the relative size of each cell neighborhood. This can be measured in multiple ways, such as the number of cells classified into each cellneighborhood, or the percentage of the image for that subject occupied by each cell neighborhood. For example, for a first subject it might be determined that cell neighborhood 0 (CNO) makes up 50% of the area of the image for that subject, while the remaining 11 cell neighborhoods (CN1-CN11) each make up 4.5% of the area of the image.

[0087] In step 160, a relationship between the proportion of cell neighborhoods and treatment response of the subjects to T cell-redirecting therapy is determined through statistical analysis. In the present example, each cell neighborhood is analyzed as follows.

[0088] The cell neighborhood proportions are normalized between subjects using an isometric log-ratio transformation, to provide normalized proportions / values for each cell neighborhood.

[0089] The N subjects (e.g., wherein N is about 50 or more) are dichotomized into two groups based on whether each subject has a greater-than median normalized proportion / value of the cell neighborhood or a less-than median normalized value / proportion of the cell neighborhood. The incidence of progression-free survival (PFS) for the two groups is tracked over a period of months (e.g., about 12 months or more, for instance, about 20 months or more) following T cell-redirecting therapy (in the present example, CAR T cell therapy), and can be used to generate survival curves (e.g., Kaplan-Meier curves) for each group.

[0090] The curves can be compared using a log-rank test to determine whether there is a statistically significant relationship between the normalized proportion of each cell neighborhood and treatment outcomes. A cell neighborhood is identified in step 170 as significant (having predictive power) if the relationship between the normalized proportion of the cell neighborhood and the treatment outcome has a significance level less than 0.1. In other examples, the threshold for significance may be set to a different value, such as 0.05.

[0091] In the present example, four significant cell neighborhoods were identified 170 (CNO, CN8, CN9 and CN11, described in more detail below). The centroid of each of the significant cell neighborhoods is output 180 as a feature vector defining that cell neighborhood. In some examples, definitions of each of the K cell neighborhoods may be output. Where definitions of only the significant cell neighborhoods are output, they may be accompanied by one or more boundary surfaces in feature space enclosing the definitions.The one or more boundary surfaces indicate the boundaries at which a point in feature space is closer to a non-significant cell neighborhood than to a significant cell neighborhood.

[0092] In the present example, the thresholds used to dichotomize the subjects into two groups were the median normalized proportions / values of the cell neighborhoods. Other thresholds may be used, such as mean proportions. The thresholds may also be provided as an output.

[0093] In the example described above, the proportions of the cell neighborhoods were normalized using an isometric log-ratio transformation. In other examples the method may not include a normalization step. The normalization step is less important, the higher the number of subjects (N) is.

[0094] The definitions provided in method 100 may be used to aid in making treatment decisions for subjects, as is described below.Use of significant cell neighborhood definitions

[0095] In another aspect, there is provided a computer implemented method of classifying cell neighborhoods in a bone marrow sample obtained from a subject suffering from multiple myeloma (MM). The method comprises: (i) receiving an image of a plurality of cells in the bone marrow sample; (ii) receiving label data for the image, wherein the label data is indicative of a cell type of each cell of the plurality of cells and wherein the cell type is selected from the list comprising: adipocyte cells; B cells; exhausted CD4+ T cells; CD4+ memory T cells; CD8+ GZMB+ T cells; exhausted CD8+ T cells; CD8+ memory T cells; endothelial cells; Ml -like macrophages; M2 -like macrophages; myeloid-derived suppressor cells (MDSCs); megakaryocytes; monocyte-derived dendritic cells (moDCs); monocytes; myeloma cells; natural killer cells; regulatory CD4 T cells; naive CD4 T cells; and naive CD8 T cells, (iii) segmenting the image to produce a segmented image identifying the location of each cell of the plurality of cells in the image; (iv) determining, based on the segmented image and the label data, a feature vector for each cell of the plurality of cells, wherein the feature vector is indicative of a proportion of each cell type found in a region surrounding the cell, wherein the region surrounding the cell extends a first distance from the cell; (v) receiving a definition of one or more cell neighborhoods, wherein each of the one or more cell neighborhoods is defined by a reference feature vector; and (vi) classifying each cell of the plurality of cells based on the feature vector of the cell and the one or more referencefeature vectors, wherein each cell is classified as belonging to one of the one or more cell neighborhoods, or belonging to a remainder class.

[0096] The method may further comprise: (A) receiving the classification of each cell; and (B) determining the proportion of each cell neighborhood.

[0097] In the described methods, the region typically extends a first distance from the edge of the cell. By edge, it is meant the outer surface of the cell. In other examples, the region may extend from within the cell, such as from a center of mass of the cell.

[0098] The first distance can be from 5 pm up to and including 15 pm, e.g., 10 pm. In some examples, such as where the region extends from a point within the cell, the first distance may be greater. For example, the first distance may be from 20 pm up to and including 40 pm.

[0099] An exemplary method 200 is described below with reference to FIG. 11 for a single image of a plurality of cells obtained for a subject suffering from MM. In some examples, multiple images may be obtained for a sample, with each image analyzed as explained below.

[0100] The method described below makes use of the cell definitions determined in method 100 to assist in making treatment decisions for the subject. By determining the proportions of the cell neighborhoods defined in method 100 for a new subject, the status of the new subject as belonging to low or high groups for the significant cell neighborhoods can be identified and their treatment outcomes predicted.

[0101] Steps 205-230 of method 200 are the same as steps 105-130 of method 100, adapted for application to a single image. As with the example described above, the image and label data of the present example are obtained from IMC. In other examples different techniques may be used to obtain the image and / or label data.

[0102] In step 240, a processor receives definitions of a plurality of cell neighborhoods. In the present example, the processor receives a definition for each one of the K cell neighborhoods. In other examples, the processor might only receive a definition for the significant cell neighborhoods and the one or more boundary surfaces.

[0103] In step 250, the processor classifies each cell into one of the cell neighborhoods based on the feature vector of the cell and the definitions of the cell neighborhoods. In the present example, the classifying compares the distance between thefeature vector of a cell and each of the definitions. A cell is classified into the cell neighborhood to which its feature vector is closest in the feature space.

[0104] In step 260, the classification of each cell is received and, in step 270, the proportion of each cell neighborhood is determined. As mentioned above for step 150, in the present example this is achieved by determining the percentage of the area of the image occupied by each cell neighborhood. The proportions may be normalized using an isometric log-ratio transformation. As will be explained below, this proportion data is useful in predicting treatment outcomes for the subject. For example, the classification of the cells may be used to predict a treatment response to a T cell-redirecting therapy (e.g., CAR T cell therapy). In examples in which multiple images are obtained for a subject, the proportions of the cell neighborhoods may be determined based on the total proportion of each cell neighborhood across all of the images.

[0105] Accordingly, in a related aspect, a method of predicting a treatment response to a T cell-redirecting therapy in a subject suffering from multiple myeloma (MM) is provided, the method comprising: (A) receiving the proportions of cell neighborhoods determined for a bone marrow sample obtained from the subject; (B) comparing, for each of the one or more cell neighborhoods, the proportion to a corresponding reference value; and (C) predicting, based on the comparing, the treatment response of the subject to T cellredirecting therapy. Comparison to a reference value allows the new subject to be identified as having a similar cellular environment to the subjects of method 100 belonging to the low or high group, for the significant cell neighborhoods. As explained above for method 100, these two groups have different treatment outcomes. Based on this knowledge, a treatment outcome of the new subject can be predicted.

[0106] In step 310 the proportions of the cell neighborhoods determined above are received, for example by a computing system or by a medical professional. More specifically, in the present example the proportions of the significant cell neighborhoods are received. The proportions of the other cell neighborhoods may or may not be received in other examples.

[0107] In step 320, the proportions of each of the significant cell neighborhoods are compared to a reference value. In the present example, the reference value is the threshold proportion previously relied on to determine whether the cell neighborhood was a significant cell neighborhood (the median proportion of that cell neighborhood for the N subjects, e.g.,50 subjects, mentioned above). The method may comprise obtaining the thresholds, such as from a memory or from a network device.

[0108] In step 330, the treatment response of the subject is predicted based on whether the proportions of the significant cell neighborhoods are greater than or less than the reference values. More specifically, if the proportion of a significant cell neighborhood that is related to a positive treatment outcome (in the present example, a high / higher incidence of PFS) is greater than the threshold, the subject is predicted to be responsive to the treatment. If the proportion of a significant cell neighborhood that is related to a negative treatment outcome (in the present example, a low / lower incidence of PFS) is greater than a threshold, the subject is predicted to be unresponsive to the treatment.

[0109] A decision as to whether a subject is to receive a T cell-redirecting therapy may be based on the prediction, with subjects receiving a T cell-redirecting therapy if they are predicted to be responsive, and subjects receiving an alternative therapy if they are predicted to be unresponsive. Alternatively, if the subject is predicted to be unresponsive to a T cell-redirecting therapy based on a method disclosed herein (e.g., because the proportion of a significant cell neighborhood that is related to a negative treatment outcome is greater than a pre-determined threshold), T cell-redirecting therapy may be combined with an immune-stimulative therapy (e.g., immune checkpoint inhibitor therapy including treatment with one or more of nivolumab, pembrolizumab, cemiplimab, atezolizumab, avelumab, durvalumab, relatlimab, and ipilimumab). In some instances, the T-cell redirecting therapy may be followed by maintenance therapy (e.g., treatment with immunomodulatory drugs like lenalidomide or pomalidomide, or combinations of immunomodulatory drugs with monoclonal antibodies like daratumumab or isatuximab) to reduce the likelihood of recurrence.

[0110] Conversely, subj ects who are predicted to be responsive to the treatment based on a method disclosed herein (e.g., because the proportion of a significant cell neighborhood that is related to a positive treatment outcome is greater than a pre-determined threshold) may not receive any additional therapy. For example, in such circumstances, cell-redirecting therapy may not be combined with an immune-stimulative therapy (e.g., immunomodulatory drugs like lenalidomide or pomalidomide, immune checkpoint inhibitor therapy (including treatment with one or more of nivolumab, pembrolizumab, cemiplimab, atezolizumab,avelumab, durvalumab, relatlimab, and ipilimumab), or monoclonal antibodies like daratumumab and isatuximab) to reduce the risk of toxicities or toxicity-related death.[OHl] In the examples described above, the identification of significant cell neighborhoods is based on a relationship between the proportion of the cell neighborhoods and incidence of PFS after undergoing a T cell-redirecting treatment such a CAR-T therapy.

[0112] In some examples, the label data may include data indicative of adipocytes. Consequently, in some examples the methods 100 and 200 need not comprise steps 120-128 and 220-228.

[0113] In some examples the label data may or may not be encoded in the images. In some examples, methods 100 and 200 may comprise a step of labelling each cell in the image based on the label data. The labelling may be performed by a machine learning model, such as a decision tree-based algorithm. Training the model may comprise providing the model with images of cells and label data as training input data, and images of cells including cell labels as ground truth output data.

[0114] In the exemplary methods described above, IMC was used to obtain images of cells and label data. The skilled person will be aware that other imaging techniques can be used to obtain images of cells and / or label data. For example, label data can be obtained using other imaging techniques including MIBI, CyCIF, t-CyCIF, 4i, CODEX, and MxIF.

[0115] In the example described above, data from a population of 50 subjects suffering from multiple myeloma was used to determine the significant cell neighborhoods. In some examples, the population may be greater than 50 or may be less than 50.

[0116] In the example described above, four significant cell neighborhoods were identified in the method 100, and the definitions of those four significant cell neighborhoods were relied on in the method 200. In other examples, a different number of cell significant cell neighborhoods may be identified and / or a different number may be relied on in the method 200. In general, one or more significant cell neighborhoods may be identified in the method 100 and used it the method 200.

[0117] In the example described above, in method 200 each cell was classified into one of the 12 cell neighborhoods based on its proximity to the centroids of the cell neighborhoods. More specifically, each cell was classified into the cell neighborhood having the closest centroid in feature space to the feature vector of that cell. The 8 ‘non-significant’cell neighborhoods can be referred to collectively as a remainder class. Cells classified as belonging to one of these cell neighborhoods can be described as belonging to the remainder class.

[0118] As mentioned above, in some examples definitions might only be obtained for the significant cell neighborhood(s), and the definition(s) may be accompanied by one or more boundary surfaces. Cells can be classified as belonging to one of one or more significant cell neighborhoods, or as belonging to the remainder class. A cell can be classified into the remainder class if it is outside of each of the one or more boundary surfaces (that is, the one or more boundary surfaces are between the cell and each of the one or more centroids of the significant cell neighborhoods).

[0119] In the example described above for method 100, the incidence of PFS was tracked over a median follow-up time (e.g., 12 months or more, for instance, about 20 months or more) in and N subjects (e.g., about 50 or more). Data on treatment outcomes can be obtained over different follow-up times and with a different number of subjects. Different data sets may result in the definition of significant cell neighborhoods that differ from the cell neighborhoods identified in the Examples.Computer system

[0120] FIG. 12 of the accompanying drawings schematically illustrates an exemplary computer system 1000 upon which a computer program according to an embodiment may run. The exemplary computer system 1000 comprises a computer-readable storage medium 1020, a memory 1040, a processor 1060 and one or more interfaces 1080, which are all linked together over one or more communication busses 1100. The exemplary computer system 1000 may take the form of a conventional computer system, such as, for example, a desktop computer, a personal computer, a laptop, a tablet, a smart phone, a server, a mainframe computer, and so on.

[0121] The computer-readable storage medium 1020 and / or the memory 1040 may store one or more computer programs (or software or code) and / or data (including but not limited to: the images, the label data, the definitions of the cell neighborhoods and the thresholds). The computer programs stored in the computer-readable storage medium 1020 and / or the memory 1040 may include computer programs that, when executed by the processor 1060, cause the processor 1060 to carry out a method according to an embodiment.The computer-readable storage medium 1020 and / or the memory 1040 may be a non- transitory computer readable storage medium. The computer-readable storage medium 1020 and / or the memory 1040 may be one or more memory chips or memory blocks implemented within the processor, magnetic media such as hard disk or floppy disks, or optical media such as for example DVD and the data variants thereof, CD.

[0122] The processor 1060 may be any data processing unit suitable for executing one or more computer readable program instructions, such as those belonging to computer programs stored in the computer-readable storage medium 1020 and / or the memory 1040. The processor 1060 may include one or more of: general purpose computers, special purpose computers, microprocessors, digital signal processors (DSPs), application specific integrated circuits (ASIC), gate level circuits and processors based on multi-core processor architecture, as non-limiting examples. As part of the execution of one or more computer-readable program instructions, the processor 1060 may store data to and / or read data from the computer-readable storage medium 1020 and / or the memory 1040. The processor 1060 may comprise a single data processing unit or multiple data processing units operating in parallel or in cooperation with each other. The processor 1060 may, as part of the execution of one or more computer readable program instructions, store data to and / or read data from the computer-readable storage medium 1020 and / or the memory 1040.

[0123] The one or more interfaces 1080 may comprise a network interface enabling the computer system 1000 to communicate with other computer systems across a network. In some examples, the computer system 1000 may obtain the images, the label data, the definitions of the cell neighborhoods and / or the reference values via the network. The network may be any kind of network suitable for transmitting or communicating data from one computer system to another. For example, the network could comprise one or more of a local area network, a wide area network, a metropolitan area network, the internet, a wireless communications network, and so on. The computer system 1000 may communicate with other computer systems over the network via any suitable communication mechanism / protocol. The processor 1060 may communicate with the network interface via the one or more communication busses 1100 to cause the network interface to send data and / or commands to another computer system over the network. Similarly, the one or more communication busses 1100 enable the processor 1060 to operate on data and / or commandsreceived by the computer system 1000 via the network interface from other computer systems over the network.

[0124] The interface 1080 may alternatively or additionally comprise a user input interface and / or a user output interface. The user input interface may be arranged to receive input from a user, or operator, of the system 1000. The user may provide this input via one or more user input devices (not shown), such as a mouse (or other pointing device, track-ball or keyboard. The user output interface may be arranged to provide a graphical / visual output to a user or operator of the system 1000 on a display (or monitor or screen) (not shown). The processor 1060 may instruct the user output interface to form an image / video signal which causes the display to show a desired graphical output. The display may be touch-sensitive enabling the user to provide an input by touching or pressing the display.

[0125] In some embodiments, the interface 1080 may alternatively or additionally comprise an interface to a measurement system, for obtaining images and / or label data for a sample obtained from a subject (for example, an IMC device).

[0126] In general, the various examples may be implemented in hardware or special purpose circuits, software, logic or any combination thereof. For example, some aspects may be implemented in hardware, while other aspects may be implemented in firmware or software which may be executed by a controller, microprocessor or other computing device, although these are not limiting examples. While various aspects may be illustrated as block diagrams, it is well understood that these blocks-may be implemented in, as non-limiting examples, hardware, software, firmware, special purpose circuits or logic, general purpose hardware or controller or other computing devices, or some combination thereof.

[0127] A single processor or other unit may fulfil the functions of several items recited in the claims. The functions may be performed in a single integrated electronic device, or the functions may be distributed across different discrete devices. For example, some functions may be performed by a remote service accessed via a wired or wireless network connection. A computer program may be stored / distributed on a suitable medium, such as an optical storage medium or a solid-state medium supplied together with or as part of other hardware, but may also be distributed in other forms, such as via the Internet or other wired or wireless telecommunication systems.Baseline spatial signatures

[0128] Using the methods described herein, it was discovered that subjects suffering from MM with a high proportions of certain pre-treatment cellular neighborhoods (preCNs) in their BM at baseline (i.e., before being administered a T cell-redirecting therapy) had reduced PFS (see FIG. 2). Moreover, it was also discovered that subjects with a high proportion of another preCN had increased PFS. The identified preCNs are characterized by spatial signatures of certain immune cell types in the vicinity of myeloma cells.

[0129] Accordingly, in a further aspect, baseline spatial signatures are provided that can be used to determine whether a subject suffering from MM is likely to benefit from receiving a T cell-redirecting treatment such as CAR-T therapy - in particular in terms of PFS. For example, the disclosed methods may be used to select subjects that are likely to benefit from a T cell-redirecting treatment such as CAR-T therapy.Baseline spatial signature 1

[0130] Using the methods described herein, it was identified that subjects with a high proportion of the bone marrow covered by preCNO at baseline (i.e. before CAR-T infusion) had reduced PFS. preCNO is characterized by baseline spatial signature 1, and subjects with a high proportion of this spatial signature in their BM are less likely to benefit from a T cellredirecting treatment such as CAR-T therapy.

[0131] The cell composition of baseline spatial signature 1 comprises: (a) a high proportion of naive CD4+ T cell subpopulation across the CD4+ T cell compartments, (b) a high proportion of exhausted CD4+ / CD8+ T cell subpopulations across the CD4+ / CD8+ T cell compartments, (c) a low proportion of GZMB+ CD8+ memory T cells across the CD8+ T cell compartment, and (d) a low proportion of CD4+ memory T cells across the CD4+ T cell compartment.

[0132] The cell composition of baseline spatial signature 1 may comprise about 13- 23%, e.g., about 17-19%, of naive CD4+ T cell subpopulation across the CD4+ T cell compartments. The cell composition of baseline spatial signature 1 may comprise about 14- 24%, e.g., about 18-20%, of exhausted CD4+ T cell subpopulations, and about 8-18%, e.g., about 12-14%, of exhausted CD8+ T cell subpopulations across the CD4+ / CD8+ T cell compartments. The cell composition of baseline spatial signature 1 may comprise 9-19%, e.g., about 13-15%, of GZMB+ CD8+ memory T cells across the CD8+ T cell compartment,and 32-42%, e.g. about 36-38% of CD4+ memory T cells across the CD4+ T cell compartment.

[0133] A myeloma cell of baseline spatial signature 1 may co-localize with other myeloma cells, B cells, MDSCs, naive T cells, CD4+ memory T cells, CD8+ memory T cells, monocytes, exhausted CD4+ T cells, M2-like macrophages, and / or MKs. Myeloma cells may show about 10-20% co-localization with other myeloma cells. Myeloma cells may show about 10-20% co-localization with B cells. Myeloma cells may show about 5-15% colocalization with MDSCs. Myeloma cells may show about 5-15% co-localization with naive T cells. Myeloma cells may show about 5-10% co-localization with CD4+ memory T cells. Myeloma cells may show about 1-2% co-localization with CD8+ memory T cells. Myeloma cells may show about 5-10% co-localization with monocytes. The myeloma cells may show about 5-10% co-localization with exhausted CD4+ T cells. Myeloma cells may show about 5-10% co-localization with M2-like macrophages. Myeloma cells may show about 5-10% co-localization with MKs.

[0134] In baseline spatial signature 1, a high degree of co-localization between immunosuppressive cells (e.g., M2-like macrophages and MDSCs) and CD4+ T cell subpopulations is observed. For example, more than about 50%, e.g., 55-75% of CD4+ memory T cells and Thl7 CD4+ memory T cells may co-localize with immunosuppressive cells. Less than about 30 %, e.g., 5-25%, of CD4+ memory T cells and Thl7 CD4+ memory T cells may co-localize with exhausted CD4+ T cells. Less than about 30%, e.g., about 5- 25%, of CD4+ memory T cells and Thl7 CD4+ memory T cells may co-localize with myeloma cells.

[0135] High levels of co-localization are also observed between immunosuppressive cells and CD8+ T cell subpopulations in baseline spatial signature 1. For example, more than about 50%, e.g., about 60-70%, of GZMB+ CD8+ memory T cells may co-localize with immunosuppressive cells. Less than about 40%, e.g., about 25-35%, of GZMB+ CD8+ memory T cells may co-localize with exhausted CD8+ T cells. Less than about 10%, e.g., about 5%, of GZMB+ CD8+ memory T cells may co-localize with myeloma cells.Baseline spatial signature 2

[0136] Using the methods described herein, it was identified that subjects with a high proportion of the bone marrow covered by preCN 11 at baseline (i.e., before CAR-T infusion)had reduced PFS. preCNl 1 is characterized by baseline spatial signature 2, and subjects with a high proportion of this spatial signature in their BM are less likely to benefit from a T cellredirecting treatment such as CAR-T therapy.

[0137] The cell composition of baseline signature 2 comprises: (a) a high proportion of PD-L1+ myeloma cells across myeloma subpopulations, (b) a high proportion of naive CD4+ / CD8+ T cell subpopulation across the CD4+ / CD8+ T cell compartments, (c) a high proportion of regulatory CD4+ T cell subpopulation across the CD4+ / CD8+ T cell compartments, and (d) a low proportion of non-Thl7 CD4+ memory T cells across the CD4+ T cell compartment.

[0138] The cell composition of baseline spatial signature 2 may comprise about 10- 20%, e.g., about 14-16%, of PD-L1+ myeloma cells across myeloma subpopulations. The cell composition of baseline spatial signature 2 may comprise about 10-20%, e.g., aboutl4- 16%, of naive CD4+ T cell subpopulation, and about 34-44%, e.g. about 38-40%, of naive CD8+ T cell subpopulation across the CD8+ T cell compartments. The cell composition of baseline spatial signature 2 may comprise about 22-32%, e.g. about 26-28%, of regulatory CD4+ T cell subpopulation across the CD4+ / CD8+ T cell compartments. The cell composition of baseline spatial signature 2 may comprise about 32-42%, e.g. about 36-38%, of non-Thl7 CD4+ memory T cells across the CD4+ T cell compartment.

[0139] A myeloma cell of baseline spatial signature 2 may co-localize with other myeloma cells, MDSCs, monocytes, M2-like macrophages, NK cells, regulatory T cells, B cells, naive T cells, exhausted CD4+ T cells, and / or CD4+ memory T cells. In particular, a myeloma cell of baseline spatial signature 2 may co-localize with other myeloma cells.

[0140] Myeloma cells may show about 55-65% co-localization with other myeloma cells. Myeloma cells may show about 10-20% co-localization with MDSCs. Myeloma cells may show about 1-10% co-localization with monocytes. Myeloma cells may show about 1- 10% co-localization with M2-like macrophages. Myeloma cells may show about 1-10% colocalization with NK cells. Myeloma cells may show about 1-10% co-localization with regulatory T cells. Myeloma cells may show about 1-5% co-localization with B cells. Myeloma cells may show about 1-5% co-localization with naive T cells. Myeloma cells may show about 1-5% co-localization with exhausted CD4+ T cells. Myeloma cells may show about 1-5% co-localization with CD4+ memory T cells.

[0141] In baseline spatial signature 2, a high degree of co-localization between immunosuppressive cells (e.g., M2-like macrophages and MDSCs) and CD4+ T cell subpopulations is observed.

[0142] For example, more than about 50%, e.g., about 50-95%, of CD4+ memory T cells and Thl7 CD4+ memory T cells may co-localize with immunosuppressive cells.

[0143] High levels of co-localization are also observed between immunosuppressive cells and CD8+ T cell subpopulations in baseline spatial signature 2.

[0144] For example, more than about 50%, e.g. about 75-95%, of CD8+ memory T cells may co-localize with immunosuppressive cells. Less than about 30%, e.g., about 5-15%, of CD8+ memory T cells may co-localize with myeloma cells.Baseline spatial signature 3

[0145] Using the methods described herein, it was identified that subjects with a high proportion of the bone marrow covered by preCN8 at baseline (i.e. before CAR-T infusion) had reduced PFS. preCN8 is characterized by baseline spatial signature 3, and subjects with a high proportion of this spatial signature in their BM are less likely to benefit from a T cellredirecting treatment such as CAR-T therapy.

[0146] The cell composition of baseline spatial signature 3 comprises: (a) a high proportion of Ki-67+ myeloma cells across myeloma subpopulations, (b) a high proportion of exhausted CD4+ / CD8+ T cell subpopulations across the CD4+ / CD8+ T cell compartments, (c) a low proportion of Ki-67- and PD-L1- myeloma cells across myeloma subpopulations, (d) a low proportion of naive CD4+ / CD8+ T cell subpopulations across the CD4+ / CD8+ T cell compartments, and (e) a low proportion of Thl7+ CD4+ memory T cells across the CD4+ T cell compartment.

[0147] The cell composition of baseline spatial signature 3 may comprise about 14- 24%, e.g., about 18-20%, of Ki-67+ myeloma cells across myeloma subpopulations. The cell composition of baseline spatial signature 3 may comprise about 23-33%, e.g., about 27-29%, of exhausted CD4+ T cell subpopulations, and about 11-21%, e.g., about 15-17%, of exhausted CD8+ T cell subpopulations across the CD4+ / CD8+ T cell compartments. The cell composition of baseline spatial signature 3 may comprise about 72-82%, e.g., about 76-78%, of Ki-67- and PD-L1- myeloma cells across myeloma subpopulations. The cell composition of baseline spatial signature 3 may comprise about 1-10%, e.g., about 3-5%, of naive CD4+T cell subpopulations, and about 13-23%, e.g., about 18-20%, of naive CD8+ T cell subpopulations across the CD4+ / CD8+ T cell compartments. The cell composition of baseline spatial signature 3 may comprise about 1-5%, e.g., about 2-4%, of Thl7+ CD4+ memory T cells across the CD4+ T cell compartment.

[0148] A myeloma cell of baseline spatial signature 3 may co-localize with M2-like macrophages, monocytes, myeloma cells, MDSCs, CD4+ memory T cells, exhausted CD4+ T cells, CD8+ memory T cells, MKs, B cells, and / or GZMB+ CD8+ memory T cells. In particular, a myeloma cell of baseline spatial signature 3 may co-localize with M2-like macrophages.

[0149] Myeloma cells may show about 40-50% co-localization with M2-like macrophages. Myeloma cells may show about 15-25% co-localization with monocytes. Myeloma cells may show about 10-20% co-localization with other myeloma cells. Myeloma cells may show about 5-15% co-localization with MDSCs. Myeloma cells may show about 1-5% co-localization with CD4+ memory T cells. Myeloma cells may show about 0-5% colocalization with exhausted CD4+ T cells. Myeloma cells may show about 0-5% colocalization with CD8+ memory T cells. Myeloma cells may show about 1-5% colocalization with MKs. Myeloma cells may show about 1-5% co-localization with B cells. Myeloma cells may show about 1-5% co-localization with GZMB+ CD8+ memory T cells.

[0150] In baseline spatial signature 3, a high degree of co-localization between immunosuppressive cells (e.g. M2-like macrophages and MDSCs) and CD4+ T cell subpopulations is observed.

[0151] For example, more than about 60%, e.g., about 65-85%, of CD4+ memory T cells and Thl7 CD4+ memory T cells may co-localize with M2-like macrophages. Less than about 20%, e.g. about 1-10%, of CD4+ memory T cells and Thl7 CD4+ memory T cells may co-localize with myeloma cells.

[0152] High levels of co-localization are also observed between immunosuppressive cells and CD8+ T cell subpopulations in baseline spatial signature 3.

[0153] For example, more than 60%, e.g., about 65-90%, of CD8+ memory T cells may co-localize with M2 -like macrophages. Less than about 10%, e.g., about 1-5%, of CD8+ memory T cells may co-localize with myeloma cells.Baseline spatial sisnature 4

[0154] Using the methods described herein, it was identified that subjects with a high proportion of the bone marrow covered by preCN9 at baseline (i.e., before CAR-T infusion) had increased PFS. preCN9 is characterized by baseline spatial signature 4, and subjects with a high proportion of this spatial signature in their BM are more likely to benefit from a T cellredirecting treatment such as CAR-T therapy.

[0155] The cell composition of baseline spatial signature 4 comprises a high proportion of Thl7 CD4+ memory T cells across the CD4+ T cell compartment.

[0156] For example, the cell composition of baseline spatial signature 4 may comprise about 22-32%, e.g., 26-28% of Thl7 CD4+ memory T cells across the CD4+ T cell compartment.

[0157] A myeloma cell of baseline spatial signature 4 may co-localize with other myeloma cells, MDSCs, monocytes, M2 -like macrophages, CD4+ memory T cells, B cells, adipocytes, CD8+ memory T cells, naive T cells, and ECs. In particular, a myeloma cell of baseline spatial signature 4 may co-localize with other myeloma cells.

[0158] Myeloma cells may show about 45-55% colocalization with other myeloma cells. Myeloma cells may show about 10-20% co-localization with MDSCs. Myeloma cells may show about 5-10% co-localization with monocytes. Myeloma cells may show about 1- 10% co-localization with M2-like macrophages. Myeloma cells may show about 1-10% colocalization with CD4+ memory T cells. Myeloma cells may show about 1-10% colocalization with B cells. Myeloma cells may show about 1-5% co-localization with adipocytes. Myeloma cells may show about 1-5% co-localization with CD8+ memory T cells. Myeloma cells may show about 1-5% co-localization with naive T cells. Myeloma cells may show about 1-5% co-localization with ECs.

[0159] In baseline spatial signature 4, a high degree of co-localization between myeloma cells and CD4+ T cell subpopulations is observed.

[0160] For example, more than about 50%, e.g., about 55-80%, of CD4+ memory T cells and Thl7 CD4+ memory T cells may co-localize with myeloma cells. Less than about 40%, e.g., about 20-354%, of CD4+ memory T cells and Thl7 CD4+ memory T cells may co-localize with immunosuppressive cells.

[0161] High levels of co-localization are also observed between myeloma cells and CD8+ T cell subpopulations in baseline spatial signature 4.

[0162] For example, more than about 50%, e.g., about 60-80%, of CD8+ memory T cells may co-localize with myeloma cells. Less than about 50%, e.g., about 20-40%, of CD8+ memory T cells may co-localize with immunosuppressive cells.T cell-redirecting therapy

[0163] The described methods have been used to identify cell neighborhoods that are indicative of the prognostic outcome for MM subjects undergoing T cell-redirecting therapy, such as CAR T cell therapy.

[0164] T cell-redirecting therapy (TCRT) involves the targeted redirection of T cells against pathological cells, such as cancer cells. TCRTs include bispecific T cell engagers (BiTEs), which direct the cytotoxic activity of T cells against cancer cells; bi- and tri-specific antibodies, which facilitate the killing of target cells by bringing said target cells into contact with T cells; and chimeric antigen receptor (CAR) and T cell receptor (TCR) T cells, which have been genetically modified to target specific antigens expressed on cancer cells.

[0165] Accordingly, in another aspect, there is provided a method of predicting a treatment response to immunotherapy in a subject suffering from multiple myeloma (MM), the method comprising: (A) receiving the proportions determined in the method of claim 2 for a bone marrow sample obtained from the subject; (B) comparing, for each cell neighborhood, the proportion to a corresponding reference value; and (C) predicting, based on the comparing, the treatment response of the subject to immunotherapy.

[0166] The proportion may be expressed as a percentage of the area of the image occupied by that cell neighborhood.Combination therapy

[0167] It may be advantageous to combine a T cell-redirecting therapy with a conventional chemotherapeutic drug or antibody treatment. For example, treatment with an immune modifying drug such as lenalidomide, thalidomide or pomalidomide may be combined with a T cell-redirecting therapy. Alternatively or additionally, treatment with an immune modifying drug may be used as a maintenance therapy after the T cell redirecting therapy. Similarly, treatment with a monoclonal antibody like daratumumab or isatuximab directed against myeloma cells or an immune-stimulatory antibody such as an immunecheckpoint inhibitor (e.g., nivolumab, pembrolizumab, cemiplimab, atezolizumab, avelumab, durvalumab, relatlimab, or ipilimumab) may be combined with a T cellredirecting therapy.

[0168] Accordingly, treatments given in combination with a T cell redirecting therapy may be given before, concurrently to or after a T cell-redirecting therapy, e.g., to increase or maintain a treatment response to the T cell-redirecting therapy.EXAMPLES

[0169] The following examples are included for illustrative purposes only and are not intended to limit the scope of the invention. Those of ordinary skill in the art may be aware of materials and methods similar or equivalent to those described below and any of these can be used to practice or test the provided methods.Example 1. IMC analysis of bone marrow architecture in patients with RRMM

[0170] To assess the influence of bone marrow (BM) architecture on survival and toxicity outcomes following BCMA-directed CAR-T therapy, BM samples obtained from 50 RRMM patients treated with either ide-cel (39, 78%) or cilta-cel (11, 22%) were analyzed. Among these, 44 baseline BM samples were collected at a median of 2.2 months (range 0.2- 4.5) before CAR-T infusion, and 17 samples were obtained one month post-treatment, including 11 matched baseline samples. IMC data were integrated with clinical and laboratory data from medical records. The study population exhibited a response rate of 84%, with a (stringent) complete remission in 58% of patients. At a median follow-up of 21.4 months, the median progression-free survival (PFS) was 13 months (95% CI, 10.8-20.5 months; FIG. 1A), while the median overall survival (OS) was not reached (95% CI, 21.8— NR months; FIG. IB). High-risk features were highly prevalent, with 13 patients (26%) presenting extramedullary disease (EMD) and 23 patients (46%) harboring high-risk cytogenetic abnormalities.

[0171] To characterize the immune architecture of the BM, IMC was used with a 36- antibody panel targeting immune cell phenotypes and functional markers, including Ki-67, PD-L1, granzyme B (GZMB), and phosphorylated STAT1 (pSTATl) (Table 2). Antibodymetal conjugation was conducted with the Maxpar labeling kit (Standard BioTools, South San Francisco, CA) at the CyTOF core of Brigham and Women’s Hospital. Followingconjugation, the concentration was assessed with a NanoDrop (Thermo Fischer Scientific, Waltham, MA). Antibody concentration and specificity were evaluated by visual inspection of IMC images with an experienced board-certified pathologist.Table 2. Antibody panel

[0172] Tissues were stained as previously described (Schapiro et al., Nat Methods. 2017; 14, 873-876). Formalin-fixed paraffin-embedded (FFPE) slides were dewaxed in xylene for 30 minutes at room temperature after deparaffinized at 62°C on heat blocks for 1 hour, followed by rehydration in an alcohol gradient series (ethanol: deionized water 100:0, 95:5, 80:20, 70:30, 0: 100; 5 minutes each). Heat-induced epitope retrieval was conducted in a water bath at 96°C in Target Retrieval Solution at pH9 (Agilent, Santa Clara, CA) for 40 minutes. After immediate cooling, tissues were washed twice with deionized water (5 minutes each), followed by TBS-T twice (5 minutes each) and TBST / 0.2% Triton X-100 (10 minutes each). Then, tissues were blocked with 3% BSA plus 1% human TruStain FcX (Biolegend, San Diego, CA) in TBST (blocking buffer) for 1 hour at room temperature. Next, tissues were incubated overnight at 4°C with antibodies diluted in the blocking buffer. Tissues were then washed three times with deionized water (5 minutes each), and they were dried before imaging mass cytometry measurement.

[0173] The abundance of bound antibody was quantified with a Hyperion XTi Imaging Mass Cytometer (Standard BioTools). Tissue was laser-ablated. Ablated tissue aerosol was transported to a CyTOF mass cytometer (Standard BioTools) for quantification, as previously described [https: / / www.nature.com / articles / nmeth.2869]. Images were acquired from each tissue at different resolutions according to the analysis objectives: cell mode (1 pm resolution), tissue mode (7 pm resolution), and preview mode (28 pm resolution).

[0174] Based on initial whole-slide images obtained using the preview mode with a resolution of 28 pm, and additional sections with conventional H&E stain, regions of interest (ROIs) were identified as BM areas that showed myeloma cell infiltration. These ROIs were then further analyzed with the high-resolution cell mode imaging, achieving a single-cell resolution.Example 2. Automated cell annotation

[0175] The .med files from Hyperion XTi (Standard BioTools) were converted into single channel images using Steinbock pipeline (vO.16.1) (Windhager et al., Nat Protoc. 2023; 18, 3565-3613) and used as input for CellProfiler (v4.2.6) (Stirling et al., BMCBioinformatics. 2021; 22, 433) to perform cell segmentation. The markers used to identify the different cells were the cell membrane markers bound by ICSK1 and ICSK2, Histone H3 for the nucleus, and CD31 to identify megakaryocytes and endothelial cells. Due to megakaryocytes’ higher dimensions and irregular borders, each image was manually reviewed, selecting megakaryocytes and blood vessels as separate objects.

[0176] Unlike other single-cell methods, the smaller number of markers and their noisy nature made UMAP and Leiden clustering unsuitable for identifying cell types on the whole dataset. To overcome this issue, one-tenth of the dataset was subsampled (53,046 cells) to perform a clustering-based cell-type annotation on an easier-to-handle dataset. Clusters were computed using the R package Rphenoannoy (vO. l.O) (Levine et al., Cell. 2015; 162, 184-97) based on the expression of following markers: CD4+5, CD45RO, CD3, CD4+, CD8+, CD20, CD27, CD138, CD14, CD16, CD56, CD68, CDl lc, CDl lb, CD31, aSMA. Clusters were assigned as B cells, endothelial cells, megakaryocytes, monocytes, myeloma cells, natural killer (NK) cells, plasma cells, T cells, and unclassified cells depending on their marker expression.

[0177] To explore the correlation between all the markers, the expression values were imported in Python (v3.11.6) (Van Rossum & Drake, Python 3 Reference Manual. 2009) and used the ,corr() function from Pandas (v2.1.1) (McKinney et al., Proceedings of the 9th Python in Science Conference. 2010; 445, 51-56). To make clear the marker’s role in cell type assignment, those with more than 0.8 correlation were removed. To annotate the whole dataset, a machine learning (ML) approach was applied to the selected markers. The decision tree-based algorithm XGBoost (v2.1.1) (Chen & Guestrin, Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. 2016; 785- 794) was used to obtain a classifier able to perform the classification on the whole dataset. The choice of a decision tree-based algorithm was driven by the nature of cell type annotation, where cells are labelled based on the higher or lower expression of specific markers, a process that mirrors the hierarchical structure of a decision tree. Parameters of the classifier are specified in Table 3.Table 3. Parameters of the XGBoost classifier utilised for cell type annotation

[0178] A finer annotation of the labels obtained from the classifier were obtained using a Gaussian Mixture Model (GMM) (Reynolds et al., Springer. 2009; 659-663). To optimize the number of clusters and the model type, the BIC function from the R package mclust (v6.0.0) (Scrucca et al., The R Journal. 2016; 8, 289-317) was used. When working in Python, the same function was implemented using sklearn GaussianMixture (vl.3.1) (Pedregosa etal., JMLR. 2011; 12, 2825-2830). GMM-based clustering on selected features was run for the detailed cell type annotation. After cell type annotation, for each cell type, those cells were removed whose area was considered an outlier. To define the outliers, the interquartile range was multiplied by a factor of 1.5. The total number of cells retrieved after outlier removal was 511,756.

[0179] Since standard BM sample processing leads to the loss of adipocytes, making them undetectable via IMC, a computational approach was developed to extrapolate adipocyte locations based on the 'hole' morphology left by their absence. This approach included the following steps:• Extract segmentation mask from the single-cell object;• Apply Gaussian filter (blurs the image and allows division of adipocytes from the background);• Binarize segmentation mask;• Invert segmentation mask (1 — > 0, 0 — > 1);• Apply watershed segmentation;• Filter objects by size (remove intercellular spaces and areas without tissue);• Apply binary expansion on the identified objects (reduce and expand the objects allowing division of continuous adipocytes in contact with each other); and• Add the identified adipocytes to the original segmentation mask

[0180] The methodology has been developed using the Python libraries skimage (v0.22.0) (Walt et al., PeerJ. 2014; 2, 453) and Numpy (vl.24.4) (Harris et al., Nature. 2020; 585, 357-362) and implemented in a Python module named AdipoFinder.

[0181] In total, 511,756 single cells were identified across all samples. A cell annotation algorithm, leveraging protein expression profiles, identified 12 major cell types, including adipocytes, B cells, CD4++ T cells, CD8++ T cells, endothelial cells (ECs), macrophages, myeloid-derived suppressor cells (MDSCs), megakaryocytes (MKs), monocyte-derived dendritic cells (moDCs), monocytes, plasma cells, and natural killer (NK) cells, which were further classified into 32 cell subtypes. Notably, a cell population with universal marker expression accounted for 3.53% of all cells but could not be further classified. These cells are most likely derived from misallocated marker signals from overlapping cell boundaries and predominantly located at tissue edges, were excluded from subsequent analyses.

[0182] To validate our cell type annotations, IMC-derived plasma cell infiltration data were compared with pathologist reports. A strong correlation was found (Pearson r = 0.72, p < 0.001), indicating high concordance between IMC-based and pathologist-based cell classification.Example 3. Identification of cellular neighborhoods with prognostic significance

[0183] The example illustrates that the computational methods described in Examples 1 and 2 can be used to identify cellular neighborhoods with prognostic significance in subjects with MM.

[0184] Emerging evidence from solid tumors suggests that specific cellular neighborhoods within, adjacent to, or outside the tumor site can influence therapeutic responses. To identify such cellular neighborhoods in the BM samples, spatial clustering algorithms were applied to the IMC data. Each cell was described with a vector representing the proportions of cell types in its direct neighbors. A MiniBatchKMeans clustering wasperformed with skleam to identify the clusters with similar cell type composition. The number of clusters (k) was identified through a grid search and the correlation of the CNs compositions and areas to clinical features such as PFS and CRS.

[0185] To compute a neighborhood graph able to consider all the neighboring cells depending on their borders instead of their centroids, thus avoiding the expulsion of adipocytes and megakaryocytes, the Python module BorderGraph was developed. BorderGraph is composed of a two-step approach. In the first step, the coordinates of the pixels falling in the border of each object of the segmentation mask are retrieved using the function find contours() of skimage. This step was applied to each image separately. The second step calculated the minimum Euclidean distances between the contour pixels of each pair of objects in each image. The pairs of objects for which the distance is below the defined threshold will result as connected in the neighborhood graph. The Python module BroderGraph was developed using AnnData (vO.10.8) (Virshup etal., biorxiv. 2021), Scanpy (vl.10.2) (Wolf et al., Genome Biology. 2018; 19), Scipy (vl.11.3) (Virtanen et al., Nat Methods. 2020; 17, 261-272), and skimage. It integrates with the Scanpy and Squidpy (Palla et al., Nat Methods. 2022; 19, 171-178) pipelines.

[0186] Computed with a distance threshold of 10pm, the neighborhood graph was utilized to ensure considering physical interactions. Cell-cell interactions were inferred from the Border Neighborhood Graph using the function interaction matrix() from Squidpy (vl.6.0). To obtain CNs-specific interactions, for each CN, the Neighborhood Graph was filtered and the interactions computed separately.

[0187] These algorithms defined neighborhoods based on a 10 pm radius from each cell border and utilized 18 distinct cell types: adipocytes, B cells, exhausted CD4+ T cells (CD4+ Tex), CD4+ memory T cells (CD4+ Tmem), GZMB+ CD8+ T cells (CD8+ GZMB+), exhausted CD8+ T cells (CD8+ Tex), CD8+ memory T cells (CD8+ Tmem), ECs, Ml-like macrophages, M2-like macrophages, MDSCs, MKs, moDCs, monocytes, myeloma cells, NK cells, naive T cells (Tnaive), and regulatory T cells (Tregs). This approach revealed 12 distinct cellular neighborhoods (CNs) in both pre-treatment (preCNl-12) and post-treatment (postCNl-12) samples, with an additional neighborhood (classified as NaN), representing cells not clearly assigned to any neighborhood. While certain CNs correlated with expected BM microenvironments, other CNs were less distinct. For example, preCN3 was predominantly composed of plasma cells (approximately 80%), resembling regions of plasmacell infiltration, while preCN4 likely represented blood vessels, as it was enriched for endothelial cells.

[0188] To evaluate the prognostic significance of the identified CNs for CAR-T outcomes, the distribution of preCNs was normalized across patients using a log-ratio transformation. Patients were stratified into high and low groups based on the median value for each preCN. Kaplan-Meier survival analysis revealed that patients with higher levels of preCNO, preCNl 1, and preCN8 in their BM samples had significantly worse PFS compared to those with lower levels (preCNO high: 6.5 months vs. preCNO low: 18.9 months, p=0.0007; preCNl l high: 7.9 months vs. preCNl l low: 18.9 months, p=0.019; preCN8 high: 6.3 months vs. preCN8 low: 15.4 months, p=0.017), while an opposite trend was observed for preCN9 (preCN9 high: 14.6 months vs. preCN9 low: 6.5 months, p=0.1; FIG. 2). Other preCNs were not significantly associated with PFS. Except for a trend in preCN8 (p=0.085), there was no association with OS.Example 4. Spatial interactions between immune cells in cellular neighborhoods

[0189] This example illustrates the spatial interactions between immune cells as well as between immune cells and neighboring myeloma cells in the identified cellular neighborhoods with prognostic significance.

[0190] The findings in Example 3 could not be explained solely by the medium cell type composition. Therefore, further analyses were performed incorporating cellular subtypes, spatial interaction metrics, and distance measures to better characterize these cellular neighborhoods and their potential impact on CAR-T efficacy.

[0191] Focusing on T cell compartments as markers of the immunosuppressive microenvironment affecting CAR-T outcomes, CD4+ and CD8+ T cells were further classified into naive (Tnaive), exhausted (Tex), and pSTATl -positive or negative effector memory cells (Tmem), differentiating by granzyme B positivity in CD8+ T cells and Thl7 expression in CD4+ T cells. Comparing CD8+ and CD4+ T cells across preCNs revealed substantial variability. PreCNs associated with inferior PFS showed increased immunosuppressive T cell subpopulations.

[0192] For example, as can be seen from FIG. 3 A and FIG. 3B, despite a higher percentage of GZMB+ Tmem cells (30.5%), preCN8 also had the highest amounts of exhausted CD8+ (15.8%) and CD4+ T cells (27.8%), as well as the lowest amounts of naiveCD8+ (18.4%) and CD4+ T cells (4.1%) and Thl7 T cells (1.5%). Similarly, preCNO was characterized by low to moderate levels of CD4+ / CD8+ Tmem cells but high amounts of exhausted T cells and naive CD4+ T cells. PreCNl 1 was enriched in naive CD8+ and CD4+ T cells with low to moderate percentages of CD4+ / CD8+ Tmem cells but with 27.3% the second highest percentage of regulatory T cells (FIG. 3C).

[0193] In contrast, preCN9, which was associated with improved PFS, displayed a mixed T cell landscape. As illustrated by FIG. 3 A and FIG. 3B, the proportion of Th 17 CD4+ T cells was 12.5% in preCN9, the second highest among all preCNs, with only moderate amounts of CD8+ Tmem and naive T cells.

[0194] The compositional information was complemented by determining spatial cell co-localization as an indicator for cell-cell interactions across preCNs. Calculating all neighboring cells within a 10 pm radius for each cell, the number of specific cell types surrounding each cell was quantified. Assessing the myeloma microenvironment, the distribution of cell types neighboring myeloma cells was analyzed, revealing high variability across different preCNs. The data are summarized in FIG. 4.

[0195] In preCNO, the myeloma microenvironment was highly heterogeneous, with myeloma cells, B cells, MDSCs, and various T cell subsets, each comprising 6-16% of neighboring cells. In contrast, preCNs 11 and 9 showed that myeloma cells were predominantly neighbored by other myeloma cells (48-59% of neighboring cells), followed by MDSCs (13-17%), monocytes (4-8%), and macrophages (4-6%). Other neighboring cells in preCNl l mainly consisted of NK cells, Tregs, Tnaive, and exhausted CD4+ T cells, whereas in preCN9, myeloma cells were more frequently neighbored by CD4+ and CD8+ Tmem cells. PreCN8 displayed a distinct profile, with a predominance of M2-like macrophages comprising up to 44% of neighboring cells (FIG. 4).

[0196] To visualize the co-localization, chord plots were created showing 100% of cell-cell interactions for each subpopulation, with link thickness representing interaction numbers. Incorporating all subpopulations resulted in highly complex figures. Therefore, to investigate the spatial impact of known immunosuppressive cells on myeloma and T cells, the data were subsetted to include myeloma cells, M2-like macrophages, MDSCs, and either the CD4+ T cell compartment (FIG. 5A [preCNO], FIG. 5B [preCNl 1], FIG. 5C [PreCN8], FIG. 5D [PreCN9]) or the CD8+ T cell compartment (FIG. 6A [preCNO], FIG. 6B [preCNl l], FIG. 6C [PreCN8], FIG. 6D [PreCN9]). Interactions betweenimmunosuppressive cells (such as M2-like macrophages and MDSCs) and T cells or myeloma cells, and between T cells and myeloma cells, were displayed. Interactions with exhausted T cells were included as potential spatial indicators of ongoing dysfunctional processes. This analysis revealed a high degree of spatial co-localization between immunosuppressive cells and T cell subpopulations, with a predominance of MDSCs and Tregs in preCNO and preCNl 1, and a predominance of M2-like macrophages in preCN8.

[0197] In preCNO, for instance, more than half of cells involving CD4+ Tmem and Thl7 CD4+ T cells were co-localized with M2-like macrophages and MDSCs (63.4-69.8%), in addition to 11-22% which were in spatial relation to exhausted T cells (see FIG. 5A). In contrast, only 8.6% of CD4+ Tmem and 25.7% of Thl7 T cells were related to myeloma cells (id.). For the CD8+ T cell compartment of preCNO, the majority of GZMB+ CD8+ Tmem cells were co-localized with M2-like macrophages and MDSCs (65.6%) and exhausted CD8+ T cells (30%), whereas a minor fraction was spatially related to myeloma cells (4.5%) - see FIG. 6A. Patterns for preCNl 1 showed similar results (see FIG. 5B and FIG. 6B). Notably, there was no spatial relation between GZMB+ CD8+ Tmem cells and myeloma cells.

[0198] PreCN8 exhibited a unique interaction profile, with M2-like macrophages accounting for up to 80% of contacts with the CD4+ and CD8+ T cell compartments as well as the myeloma compartment (see FIG. 5C and FIG. 6C). In contrast, preCN9 had significantly fewer spatial contacts between M2 -like macrophages and MDSCs and the CD4+ and CD8+ T cell compartments (maximum 31.9%), and a higher proportion of spatial relations to myeloma cells (68-74%) - see FIG. 5D and FIG. 6D). Notably, exhausted CD4+ / CD8+ T cells were mainly co-localized with myeloma cells and less directly with other T cell populations, suggesting that their exhaustion may derive from antitumor activity (id.).

[0199] Hypothesizing that in neighborhoods associated with inferior survival, effector cells are further from myeloma cells and closer to immunosuppressors, distances were measured between cell types. To measure distances between different cell types, pixel length (one pixel equals one micrometer) was utilized and the distances of various cell types to a cell of interest were calculated across preCN8, preCN9, preCNl 1, and preCNO. Distances between cell types were calculated using the contours from BroderGraph, as distance metric we used Euclidean distance between the two closest pixels between contour pairs. To calculate the cell-cell distances for each cell type pair, the analysis to cells wasrestricted within the same CN and on the same slide. Distances between cells from different slides or CNs were excluded from the calculation. Using this approach made it possible to focus on intra-CN, intra-slide interactions, avoiding confounding effects from inter-sample comparisons. The results of this analysis are shown in FIG. 7A (myeloma cells and M2 -like macrophages), FIG. 7B (myeloma cells and Tregs), FIG. 7C (myeloma cells and CD8+ Tmem cells), FIG. 7D (myeloma cells and GZMB+ CD8+ Tmem cells), FIG. 7E (myeloma cells and CD4+ Tmem cells), FIG. 7F (myeloma cells and MDSCs), FIG. 7G (myeloma cells and Tnaive cells), FIG. 7H (myeloma cells and CD8+ Tex cells), and FIG. 71 (myeloma cells and CD4+ Tex cells).

[0200] Analyzing the distances from the closest T cell subpopulations to myeloma cells, preCN9 exhibited the shortest distance between CD4+ Tmem cells and myeloma cells across all baseline neighborhoods (see FIG. 7E). Among CD8+ effector cells, distances did not vary significantly (see FIG. 7C and FIG. 7D). However, exhausted CD4+ and CD8+ T cells and regulatory and naive T cells showed the smallest distance to myeloma cells in preCNl 1 (see FIG. 71, FIG. 7H, FIG. 7B, and FIG. 7G, respectively). In preCNO, T cell subpopulations tended to be further from myeloma, including GZMB+ CD8+ Tmem, Treg cells, CD8+ Tex, and Tnaive (see FIG. 7D, FIG. 7B, FIG. 7H, and FIG. 7G, respectively).Example 5. Association between myeloma cell subpopulations and patient survival

[0201] This example illustrates the association between myeloma cell subpopulations and patient survival.

[0202] To further characterize the baseline cellular neighborhoods preCN8, preCN9, preCNl l, and preCNO, detailed phenotypic and functional markers were utilized to determine the distribution of cell subpopulations within each neighborhood. This approach allowed the categorization of myeloma cells into three subpopulations: Ki-67 positive (Ki- 67 -), PD-L1 positive (PD-L14-), and Ki-67- / PD-Ll- myeloma cells.

[0203] Analyzing the distribution of these subpopulations among all myeloma cells, it was observed that preCN8 was enriched with Ki-674- myeloma cells (18.8%) compared to other preCNs (see FIG. 8A), whereas preCNl l exhibited a high proportion of PD-L14- myeloma cells (15%; see FIG. 8B). Notably, a higher-than-median presence of Ki-674- and PD-L14- myeloma cells was significantly associated with poorer PFS (Ki-674- myeloma high:6.5 months vs. Ki-67+ myeloma low: 19.3 months, p=0.0005; PD-L1+ myeloma high: 18.9 months vs. PD-L1+ myeloma low: 6.5 months, p=0.0028, FIG. 9A and FIG. 9B), whereas the association was less with Ki-67- / PD-Ll- myeloma (p=0.032, id.).Example 6. Identification of distinct spatial signatures with prognostic significance

[0204] This example illustrates that the identified cellular neighborhoods with prognostic significance in subjects with MM are characterized by spatial signatures of certain immune cell types in the vicinity of myeloma cells. Identifying such spatial signatures may provide additional prognostic value to stratify subjects with MM.

[0205] Based on the results shown in Examples 3 and 4, spatial signatures 1, 2, 3 and 4 were defined for preCNO, preCNl l, preCN8 and preCN9, respectively.Baseline spatial signature 1: preCNO

[0206] Compared to the cell compositions of other cell neighborhoods, the cell composition of preCNO comprised a high proportion of naive CD4+ T cell subpopulation across the CD4+ T cell compartments (17.6%) and exhaustedCD4+ / CD8+ T cell subpopulations across the CD4+ / CD8+ T cell compartments (CD4+ Tex: 19.2%, CD8+ Tex: 12.7%), and a low proportion of GZMB+ CD8+ Tmem cells across the CD8+ T cell compartment (13.7%) and CD4+ Tmem cells across the CD4+ T cell compartment (37.2%).

[0207] Compared to other cell neighborhoods, mixed cell types co-localized with myeloma cells within a 10 pm radius in preCNO. The top ten cell types in preCNO that colocalized with myeloma cells were: myeloma cells (15.6%), B cells (14.9%), MDSCs (9%), Tnaive cells (8%), CD4+ Tmem cells (7.4%), CD8+ Tmem cells (0.7%), monocytes (6.9%), CD4+ Tex cells (6.4%), M2 -like macrophages (4.3%), and MKs (4%).

[0208] Analysis of the spatial co-localization patterns between MDSCs, M2-like macrophages, myeloma and the CD4+ T cell compartment in preCNO showed that more than half of CD4+ Tmem and Thl7 CD4+ Tmem cells co-localized with immunosuppressive cells (63.4-69.8%) and a high proportion of CD4+ Tmem and Thl7 CD4+ Tmem cells colocalized with exhausted CD4+ Tex cells (11-22%). Conversely, a low proportion of CD4+ Tmem and Thl7 CD4+ Tmem cells co-localized with myeloma cells (8.6-25.7%)

[0209] Analysis of the spatial co-localization patterns between MDSCs, M2-like macrophages, myeloma and the CD8+ T cell compartment in preCNO showed that more than half of GZMB+ CD8+ Tmem cells co-localized with immunosuppressive cells (65.6%) anda high proportion of GZMB+ CD8+ Tmem cells co-localized with exhausted CD8+ Tex cells (30%). Conversely, a low proportion of GZMB+ CD8+ Tmem cells co-localized with myeloma cells (4.5%).

[0210] For cells within a 100 pm radius, there was a larger physical distance between myeloma cells and each of the closest GZMB+ CD8+ Tmem and Tnaive cells in preCNO compared to other cell neighborhoods.Baseline spatial signature 2: preCNll

[0211] Compared to the cell compositions of other cell neighborhoods, the cell composition of preCNl l comprised high proportions of PD-L1+ myeloma cells across myeloma subpopulations (15%), naive CD4+ / CD8+ T cell subpopulation across the CD4+ / CD8+ T cell compartments (CD4+ Tnaive: 14.6%, CD8+ Tnaive: 39.4), and regulatory CD4+ T cell subpopulation across the CD4+ / CD8+ T cell compartments (27.3%). Conversely, the cell composition of preCNl l showed low proportions of non-Thl7 CD4+ Tmem cells across the CD4+ T cell compartment (36.8%).

[0212] Compared to other cell neighborhoods, myeloma cells in preCNl l mainly colocalized with other myeloma cells within a 10 pm radius. The top ten cell types in preCNl l that co-localized with myeloma cells were: myeloma cells (58.6%), MDSCs (13.1%), monocytes (4.1%), M2 -like macrophage (4.1%), NK cells (4.1%), Treg cells (3.2%), B cells (2.5%), Tnaive cells (2.2%), CD4+ Tex cells (1.9%), and CD4+ Tmem cells (1.6%).

[0213] Analysis of the spatial co-localization patterns between MDSCs, M2-like macrophages, myeloma and the CD4+ T cell compartment in preCNl l showed that more than half of CD4+ Tmem and Thl7 CD4+ Tmem cells co-localized with immunosuppressive cells (50-83.3%).

[0214] Analysis of the spatial co-localization patterns between MDSCs, M2-like macrophages, myeloma and the CD8+ T cell compartment in preCNl l showed that more than half of (GZMB+ / -) CD8+ Tmem cells co-localized with immunosuppressive cells (81.1- 87.6%). In contrast, a low proportion of (GZMB+ / -) CD8+ Tmem cells co-localized with myeloma cells (0-10.8%).

[0215] For cells within a 100 pm radius, there was (1) a short physical distance between myeloma cells and each of the closest CD4+ / CD8+ Tex and Tnaive cells, and (2) alarger physical distance between myeloma cells and each of the closest (GZMB+ / -) CD8+ Tmem cells in preCNl 1 compared to other cell neighborhoods.Baseline spatial signature 3: preCN8

[0216] Compared to the cell compositions of other cell neighborhoods, the cell composition of preCN8 comprised higher proportions of Ki-67+ myeloma cells across myeloma subpopulations (18.8%) and exhausted CD4+ / CD8+ T cell subpopulations across the CD4+ / CD8+ T cell compartments (CD4+ Tex: 27.8%, CD8+ Tex: 15.8%). Conversely, the cell composition of preCN8 showed lower proportions of Ki-67- and PD-L1- myeloma cells across myeloma subpopulations (77.1%), naive CD4+ / CD8+ T cell subpopulations across the CD4+ / CD8+ T cell compartments (CD4+ Tex: 4.1%, CD8+ Tex: 18.4%), and Thl7+ CD4+ Tmem cells across the CD4+ T cell compartment (Thl7 CD4+ Tmem: 1.5%).

[0217] Compared to other cell neighborhoods, myeloma cells in preCN8 mainly colocalized with M2 -like macrophages within a 10 pm radius. The top ten cell types in preCN8 that co-localized with myeloma cells were: M2-like macrophages (44.3%), monocytes (17.8%), myeloma cells (15.6%), MDSCs (8.2%), CD4+ Tmem cells (3.2%), CD4+ Tex cells (1.9%), CD8+ Tmem cells (1.7%), MKs (1.4%), B cells (1.2%), and GZMB+ CD8+ Tmem cells (1.0%).

[0218] Analysis of the spatial co-localization patterns between MDSCs, M2-like macrophages, myeloma and the CD4+ T cell compartment in preCN8 showed that more than half of CD4+ Tmem cells and Thl7 CD4+ Tmem cells co-localized with M2-like macrophages (74-75.2%). Conversely, a low proportion of CD4+ Tmem cells and Thl7 CD4+ Tmem cells co-localized with myeloma cells (2.7-6.2%).

[0219] Analysis of the spatial co-localization patterns between MDSCs, M2-like macrophages, myeloma and the CD8+ T cell compartment in preCN8 showed that more than half of (GZMB+ / -) CD8+ Tmem cells co-localized with M2-like macrophages (76.6-80.3%). In contrast, a low proportion of (GZMB+ / -) CD8+ Tmem cells co-localized with myeloma cells (1.3-2.1%).

[0220] For cells within a 100 pm radius, there was a short-to-medium physical distance between myeloma cells and each of the closest CD8+ Tex cells in preCN8 compared to other cell neighborhoods.Baseline spatial sisnature 4: yreCN9

[0221] Compared to the cell compositions of other cell neighborhoods, the cell composition of preCN9 comprised a high proportion of Thl7 CD4+ Tmem cells across the CD4+ T cell compartment (26.7%).

[0222] Compared to other cell neighborhoods, myeloma cells in preCN 11 mainly colocalized with other myeloma cells within a 10 pm radius. The top ten cell types in preCNl 1 that co-localized with myeloma cells were: myeloma cells (47.9%), MDSCs (16.9%), monocytes (8.3%), M2-like macrophages (5.9%), CD4+ Tmem cells (3.9%), B cells (3.6%), adipocytes (2.3%), CD8+ Tmem cells (2.1%), Tnaive cells (1.8%), and ECs (1.8%).

[0223] Analysis of the spatial co-localization patterns between MDSCs, M2-like macrophages, myeloma and the CD4+ T cell compartment in preCN9 showed that a high proportion of CD4+ Tmem cells and Thl7 CD4+ Tmem cells co-localized with myeloma cells (67.8-70.1%), while a low proportion of CD4+ Tmem cells and Thl7 CD4+ Tmem cells co-localized with immunosuppressive cells (28.5-29.3%).

[0224] Analysis of the spatial co-localization patterns between MDSCs, M2-like macrophages, myeloma and the CD8+ T cell compartment in preCN9 showed that a high proportion of (GZMB+ / -) CD8+ Tmem cells co-localized with myeloma cells (65.2-73.7%), while a low proportion of (GZMB+ / -) CD8+ Tmem cells co-localized with immunosuppressive cells (23.2-32.1%).

[0225] For cells within a 100 pm radius, there was (1) a short physical distance between myeloma cells and the closest CD4+ Tmem cells, and (2) a larger physical distance between myeloma cells and the closest CD8+ Tex cells in preCN9 compared to other cell neighborhoods.

[0226] Overall, as shown in Example 3, patients with more area within the bone marrow covered by spatial signatures 1, 2 and 3 (preCNO, 11 and 8, respectively) at baseline (i.e. before CAR-T infusion) showed inferior survival. Conversely, patients with more area covered by spatial signature 4 (preCN9) within the bone marrow at baseline showed superior survival.

[0227] It should be understood that the details provided herein are given by way of illustration only, not limitation. Other features, objects, and advantages are apparent from theabove detailed description, drawings and examples. Various changes and modifications will be apparent to those skilled in the art.

[0228] All patents, patent publications, and non-patent publications referenced herein are indicative of the level of skill of those skilled in the art to which this invention pertains. All these publications are herein incorporated by reference to the same extent as if each individual publication were specifically and individually indicated as being incorporated by reference.

Claims

CLAIMS1. A computer implemented method of classifying cell neighborhoods in a bone marrow sample obtained from a subject suffering from multiple myeloma (MM), the method comprising: receiving an image of a plurality of cells in the bone marrow sample; receiving label data for the image, wherein the label data is indicative of a cell type of each cell of the plurality of cells and wherein the cell type is selected from the list comprising:Adipocytes;B cells;Exhausted CD4+ T cells;CD4+ memory T cells;CD8+ granzyme B+ (GZMB+) T cells;Exhausted CD8 T cellsCD8 memory T cells;Endothelial cells;Ml -like macrophages;M2 -like macrophages;Myeloid-derived suppressor cells (MDSCs);Megakaryocytes;Monocyte-derived dendritic cells (moDCs);Monocytes;Myeloma cells;Natural killer cells;Regulatory CD4+ T cells;Naive CD4+ T cells; andNaive CD8+ T cells; segmenting the image to produce a segmented image identifying the location of each cell of the plurality of cells in the image; determining, based on the segmented image and the label data, a feature vector for each cell of the plurality of cells, wherein the feature vector is indicative of a proportion ofeach cell type found in a region surrounding the cell, wherein the region surrounding the cell extends a first distance from the cell; receiving a definition of one or more cell neighborhoods, wherein each of the one or more cell neighborhoods is defined by a reference feature vector; and classifying each cell of the plurality of cells based on the feature vector of the cell and the one or more reference feature vectors, wherein each cell is classified as belonging to one of the one or more cell neighborhoods, or belonging to a remainder class.

2. The method of claim 1, further comprising:A receiving the classification of each cell; andB determining the proportion of each of the one or more cell neighborhoods.

3. The method of claim 1 or 2, further comprising labelling each cell in the image based on the label data, and wherein determining the feature vectors is based on the segmented image and the labels.

4. The method of any one of the preceding claims, wherein the region extends from the edge of the cell.

5. The method of any one of the preceding claims, wherein the first distance is from 5 pm up to and including 15 pm, preferably 10pm.

6. The method of any one of the preceding claims, further comprising, before determining the feature vectors: smoothing the segmented image; inverting the segmented image to produce an inverted segmented image identifying the location of a background portion of the image of the plurality of cells; identifying individual objects in the inverted segmented image; filtering the objects by size, to select objects having a size in the range 5 square pm to 10000 square pm; eroding the selected objects to produce eroded objects; dilating the eroded objects to restore the original sizes of the eroded objects;adding the portions of the inverted segmented image that correspond to the eroded objects to the segmented image; indicating in the label data that the portions of the inverted segmented images added to the segmented image are adipocyte cells.

7. The method of claim 6, further comprising binarizing the segmented image before inverting the segmented image.

8. The method of claim 6 or 7, wherein identifying individual objects in the inverted segmented image comprises applying watershed segmentation to the inverted segmented image.

9. The method of any one of claims 6-8, wherein smoothing the segmented image comprises applying a Gaussian blur to the segmented image.

10. The method of any one of claims 6-9, wherein eroding the selected objects and dilating the eroded objects is implemented by binary expansion.

11. A method of predicting a treatment response to a T cell-redirecting therapy in a subject suffering from multiple myeloma (MM), the method comprising:A receiving the proportions determined in the method of claim 2 for a bone marrow sample obtained from the subject;B comparing, for each of the one or more cell neighborhoods, the proportion to a corresponding reference value; andC predicting, based on the comparing, the treatment response of the subject to the T cell-redirecting therapy.

12. The method of claim 11, wherein the proportion is expressed as a percentage of the area of the image occupied by that cell neighborhood.

13. A computer implemented method of defining one or more significant cell neighborhoods for responsiveness of subjects suffering from multiple myeloma (MM) to a T cell-redirecting therapy, the method comprising:receiving a plurality of images, each image being an image of a plurality of cells in a bone marrow sample obtained from one of a plurality of subjects suffering from multiple myeloma (MM); receiving label data for the plurality of images, wherein the label data is indicative of a cell type of each cell of the plurality of cells and wherein the cell type is selected from the list comprising:Adipocytes;B cells;Exhausted CD4+ T cells;CD4+ memory T cells;CD8+ GZMB+ T cells;Exhausted CD8+ T cellsCD8+ memory T cells;Endothelial cells;Ml -like macrophages;M2 -like macrophages;Myeloid-derived suppressor cells (MDSCs);Megakaryocytes;Monocyte-derived dendritic cells (moDCs);Monocytes;Myeloma cells;Natural killer cells;Regulatory CD4+ T cells;Naive CD4+ T cells; andNaive CD8+ T cells; segmenting the images to produce a plurality of segmented images, each segmented image identifying the location of each cell of the plurality of cells in the corresponding image; determining, for each of the plurality of images and based on the corresponding segmented image and the label data, a feature vector for each cell of the plurality of cells, wherein the feature vector is indicative of a proportion of each cell type found in a region surrounding the cell, wherein the region surrounding the cell extends a first distance from the cell; clustering the feature vectors to generate K cell neighborhoods;determining the proportion of each cell neighborhood of the K cell neighborhoods for each subject; determining a relationship between the proportion of each cell neighborhood for the subjects and a treatment response of the subjects to a T cell-redirecting therapy; and identifying one or more significant cell neighborhoods, wherein a cell neighborhood is identified as a significant cell neighborhood if it has a statistically significant relationship with the treatment response of the subjects; and outputting a definition of each of the one or more significant cell neighborhoods.

14. The method of claim 13, wherein the clustering is K means clustering and wherein K is in the range 10-20, e.g., 12.

15. The method of claim 13 or 14, wherein the T cell-redirecting therapy is chimeric antigen receptor (CAR) T cell therapy.

16. The method of any one of claims 13-15, wherein determining the relationship between the proportion of each cell neighborhood and the treatment response of the subjects to immunotherapy comprises, for each cell neighborhood: dichotomizing the subjects into a high group and a low group, wherein the high group have a greater than median proportion of that cell neighborhood and the low group have a less than median proportion of that cell neighborhood; and generating a Kaplan-Meier curve for each group; comparing the Kaplan-Meier curves for each group to determine a significance level for the cell neighborhood; and identifying the cell neighborhood as a significant cell neighborhood if the significance level is less than 0.1.

17. The method of claim 16, wherein comparing the Kaplan-Meier curves for each group comprises applying a log-rank test to the Kaplan-Meier curves.

18. A processing system configured to perform the method of any of claims 1-17.

19. A non-transitory computer-readable storage medium having stored thereon computer readable code configured to cause a computer to perform the method of any of claims 1-17 when the code is run on the computer.

20. An image processing system comprising: a processor configured to: receive an image of a plurality of cells in a bone marrow sample obtained from a subject suffering from multiple myeloma (MM); receive a definition of one or more cell neighborhoods, wherein each of the one or more cell neighborhoods is defined by a reference feature vector; and receive label data for the image, wherein the label data is indicative of a cell type of each cell of the plurality of cells and wherein the cell type is selected from the list comprising:Adipocytes;B cells;Exhausted CD4+ T cells;CD4+ memory T cells;CD8+ GZMB+ T cells;Exhausted CD8+ T cellsCD8+ memory T cells;Endothelial cells;Ml -like macrophages;M2 -like macrophages;Myeloid-derived suppressor cells (MDSCs);Megakaryocytes;Monocyte-derived dendritic cells (moDCs);Monocytes;Myeloma cells;Natural killer cells;Regulatory CD4+ T cells;Naive CD4+ T cells; andNaive CD8+ T cells;segment the image to produce a segmented image identifying the location of each cell of the plurality of cells in the image; for each cell in the image, determine a feature vector based on the segmented image and the label data, wherein the feature vector is indicative of a proportion of each cell type found in a region surrounding the cell, and wherein the region surrounding the cell extends a first distance from the cell; and classify each cell of the plurality of cells based on the feature vector of the cell and the one or more reference feature vectors, wherein each cell is classified as belonging to one of the one or more cell neighborhoods, or belonging to a remainder class.

21. The system of claim 20, wherein the processor is further configured to determine the proportion of each of the one or more cell neighborhoods.

22. The system of claim 20 or 21, wherein the processor is further configured to label each cell in the image based on the label data, and wherein the processor is configured to determine the feature vector for each cell based on the segmented image and the labels.

23. The system of any one of claims 20-22, wherein the processor is configured to: smooth the segmented image; invert the segmented image to produce an inverted segmented image identifying the location of a background portion of the image of the plurality of cells; identify individual objects in the inverted segmented image; filter the objects by size, to select objects having a size in the range 5 square pm to 10,000 square pm; erode the selected objects to produce eroded objects; dilate the eroded objects to restore the original sizes of the eroded objects; add the portions of the inverted segmented image that correspond to the eroded objects to the segmented image; indicate in the label data that the portions of the inverted segmented images added to the segmented image are adipocyte cells.

24. An image processing system comprising: a processor, configured to:receive a plurality of images, each image being an image of a plurality of cells in a bone marrow sample obtained from one of a plurality of subjects suffering from multiple myeloma (MM); and receive label data for the plurality of images, wherein the label data is indicative of a cell type of each cell of the plurality of cells and wherein the cell type is selected from the list comprising:Adipocytes;B cells;Exhausted CD4+ T cells;CD4+ memory T cells;CD8+ GZMB+ T cells;Exhausted CD8+ T cellsCD8+ memory T cells;Endothelial cells;Ml -like macrophages;M2 -like macrophages;Myeloid-derived suppressor cells (MDSCs);Megakaryocytes;Monocyte-derived dendritic cells (moDCs);Monocytes;Myeloma cells;Natural killer cells;Regulatory CD4+ T cells;Naive CD4+ T cells; andNaive CD8+ T cells; segment each of the plurality of images to produce a plurality of segmented images, each segmented image identifying the location of each cell of the plurality of cells in the corresponding image; for each cell in each image, determine a feature vector based on the corresponding segmented image and the label data, wherein the feature vector is indicative of a proportion of each cell type found in a region surrounding the cell, and wherein the region surrounding the cell extends a first distance from the cell; cluster the cells into K cell neighborhoods based on the feature vectors;determine the proportion of each cell neighborhood of the K cell neighborhoods for each subject; determine a relationship between the proportion of each cell neighborhood and a treatment response of the subjects to a T cell-redirecting therapy; and identify one or more significant cell neighborhoods, wherein a cell neighborhood is identified as a significant cell neighborhood if it has a statistically significant relationship with the treatment response of the subjects; and output a definition of each of the one or more significant cell neighborhoods.

Images