Patent Application
20240105283
Systems and methods relate to using expression levels for a set of genes in order to identify a phenotype of a tumor and to identify a treatment candidate based on the phenotype (e.g., a treatment candidate that includes an anti-TGFβ agent when the phenotype is immune-excluded). The set of genes can include genes predictive of digital-pathology characteristics of CD8+T cells (e.g., in terms of quantity and/or spatial location).
Clinical success of cancer immunotherapies such as immune checkpoint inhibitors has revolutionized traditional cancer treatment. By targeting the immune checkpoint regulators including CTLA-4 and the PD-1/PD-L1 axis, these immunotherapies promote cytotoxic killing of cancer cells by enhancing the function of effector T cells. Despite impressive efficacy demonstrated in subsets of patients with melanoma, NSCLC, urothelial bladder cancer, and renal cell cancer, significant challenges still exist in this field. Dramatic and durable responses were mainly observed in subsets of patients with a pre-existing T cell immunity in tumors. As such, other steps in the tumor immunity cycle may influence the effectiveness of immunotherapies based on checkpoint blockade. These include antigen presentation and T cell priming, capacity of tumor infiltration by functional CD8+T effector cells, as well as accumulation of immunoregulatory mechanisms that evolved to protect tissue integrity from exuberant immune responses. Overcoming mechanisms that impede immune activation may thus enhance the potential of cancer immunotherapy.
CD8+T cells are the main players in eradicating cancer cells in most of the immunotherapy settings. CD8+T cells recognize tumor-associated antigens through the MHC class I/T cell receptor complex and mediate cytotoxic killing of tumor cells. Given that effective cytotoxic killing requires direct contact between CD8+T cells and tumor cells, it has been increasingly recognized that different CD8+T cell distributions in the tumor microenvironment (TME) may elicit different responses to immunotherapies.
Three basic tumor-immune phenotypes have been described previously, including 1) the inflamed/infiltrated phenotype in which CD8+T cells infiltrate the tumor epithelium; 2) the immune excluded phenotype in which infiltrating CD8+T cells accumulate in the tumor stroma rather than the tumor epithelium, and 3) the immune desert phenotype in which CD8+T cells are either absent or present in very low numbers. These histologically established tumor-immune phenotypes provided a useful framework to profile immune contexture in solid tumors. However, it remains challenging to systematically define the tumor-immune phenotype of most cancer patients due to the highly heterogeneous and complex nature of immune cell infiltration and distribution. Further, the molecular features and mechanisms that shape spatial distribution of tumor-infiltrating CD8+T cells are not well understood.
In some embodiments, systems and methods use a machine-learning approach to classify and molecularly characterize tumor-immune phenotypes. This approach can be used to detect previously undiscovered molecular features that are associated with distinct immune phenotypes. More specifically, a classifier can be configured to receive a data set that includes expression levels corresponding to a pre-identified set of genes and to output a label that corresponds to a tumor-immune phenotype. The classifier can use the Prediction Analysis of Microarrays. The pre-identified set of genes may include at least 1, at least 10, at least 50, at least 100 or at least 120 of the genes in Table 1. The tumor-immune phenotype can include one of: immune-desert phenotype, an immune-excluded phenotype or an inflamed/infiltrated phenotype.
The pre-identified set of genes can include and/or can contain genes for which expression levels are specifc to and/or significantly related to CD8+T-cell characteristics detectable by using pathology images. The CD8+T-cell characteristics can include a quantity of CD8+T cells and/or can correspond to locations of CD8+T cells (e.g., a quantity of CD8+T cells in the tumor epithelium, a quantity of CD8+T cells in the stroma, a proportion of CD8+T cells in the tumor epithelium, a proportion of CD8+T cells in the stroma, etc.). A machine-learning model (e.g., regression and/or random-forest model) may be used to determine which gene expression levels are related to CD8+T-cell characteristics.
The label identified by the classifier for a given data set can be used to identify a treatment candidate. For example, the treatment candidate may include anti-TGFβ (and potentially also a checkpoint inhibitor, such as anti-PD-L1) when the phenotype is identified as an immune-excluded phenotype. As another example, the treatment candidate may include a checkpoint inhibitor (and lack anti-TGFβ) when the phenotype is identified as an inflamed/infiltrated phenotype. The treatment candidate can be identified by performing a look-up process using an identifier of the phenotype. In some instances, multiple treatment candidates are identified.
An output can be generated to include an identification of the particular phenotype, the treatment candidate(s) and/or an identification of a subject associated with the new expression-level data set. The output can be presented locally and/or transmitted to another device.
In some instances, the machine-learning approach can include performing an additional clustering (e.g., consensus clustering) using some or all of the training data in order to detect molecular features of individual phenotypes. The additional clustering may include accessing a data set that includes, for each of a set of subjects, an expression level of each of multiple gene determined to be specific to a quantity or spatial distribution of CD8+T cells. The additional clustering may be configured such that there are more clusters than there are phenotype labels. Each of the clusters may be nonetheless associated with a given phenotype label (e.g.,and used to generate a molecular profile (based on expression levels associated with the cluster) for the cluster. Thus, for a given phenotype label, the additional clustering can generate one or more molecular profiles that can be used identify (for example) treatment candidates for the phenotype (e.g., which may be generally associated with the phenotype label or may be associated with a specific cluster).
In some embodiments, a method of treatment is provided that includes targeting the TGFβ pathway. It has been discovered, through implementation of the machine-learning approach, that the cytokine, TGFβ is a molecular mediator in promoting CD8+T cell exclusion and immune suppression via a crosstalk with both tumor cells and tumor stroma (at least in some contexts, such as for ovarian cancer). Thus, targeting the TGFβ pathway may overcome T cell exclusion from tumors and improve subjects' response to cancer immunotherapy.
In some embodiments, a computer-implemented method is provided that includes accessing gene expression data for a predefined set of genes, the gene expression data corresponding to a subject. For each gene in the predefined set of genes, an expression level of the gene may have been identified as being informative of a quantity of CD8+cells associated with a tumor of the subject and/or a spatial distribution of CD8+cells. The method includes generating a cluster assignment using the gene expression data; determining that the cluster assignment corresponds to a particular phenotype; and outputting a result based on the particular phenotype.
The spatial distribution of CD8+cells may be computed from a first quantity of CD8+cells located in a tumor epithelium in the subject and a second quantity of CD8+cells located in a tumor stroma in the subject, each of the first quantity and the second quantity having been determined based on an assessment of one or more digital pathology images. The particular phenotype may include an immune-desert phenotype, immune-excluded phenotype or an inflamed/infiltrated phenotype. The predefined set of genes may have been identified using a machine-learning model (e.g., a regression model or a random-forest regression model). The method may further include selecting one or more treatment candidates based on the particular phenotype, wherein the result identifies the one or more treatment candidates. The particular phenotype may include an immune-excluded phenotype, and the one or more treatment candidates may include anti-TGFβ. The predefined set of genes may include at least one of GZMA, GZMB, GMZH, CD40LG, TAPBP, PSMB10, HLA-DOB, FAP, TDO2, LRRTM3, ASTN1, SLC4A4, UGT1A3, UGT1A5, and UGT1A6. The predefined set of genes may include at least five genes identified in Table 1. The predefined set of genes includes at least one gene identified in rows 1-56 of Table 1, at least one gene identified in rows 57-244 of Table 1 and/or at least one gene identified in rows 245-346 of Table 1. The result may identify the particular phenotype.
In some embodiments, a method of treatment is provided that includes identifying a subject with a tumor; determining that the tumor corresponds to an immune excluded phenotype; and prompting administration of anti-TGFβ to the subject (or administering anti-TGFβ to the subject). An amount of anti-TGFβ administered may be sufficient to result in a promotion of MHC class I expression in cancer cells of the tumor. An amount of anti-TGFβ administered may be sufficient to result in suppression of extracellular matrix production by cancer-associated fibroblasts associated with the tumor. An amount of anti-TGFβ administered may be sufficient to result in suppression of production of immunosuppressive molecules by cancer-associated fibroblasts associated with the tumor. The method may further include prompting administration of (or administering) a checkpoint inhibitor to the subject, where an amount of anti-TGFβ administered and an amount of checkpoint inhibitor administered may be collectively sufficient to enhance cytotoxic activity of effector T cells in the subject. The checkpoint inhibitor includes anti-PD-L1. Determining that the tumor corresponds to the immune excluded phenotype may include: accessing one or more digital pathology images corresponding to the subject; determining, based on the one or more digital pathology images, a first quantity of CD8+cells located in a tumor epithelium in the subject; determining, based on the one or more digital pathology images, a second quantity of CD8+cells located in a tumor stroma in the subject; generating a distribution metric based on the first quantity and second quantity; and determining that the distribution metric exceeds a predefined threshold. Determining that the tumor corresponds to the immune excluded phenotype may include: accessing gene expression data for a predefined set of genes, the gene expression data corresponding to the subject; generating a cluster assignment using the gene expression data; and determining that the cluster assignment corresponds to the immune excluded phenotype. The predefined set of genes may include at least five genes identified in Table 1.
In some embodiments, a system is provided that includes one or more data processors and a non-transitory computer readable storage medium containing instructions which, when executed on the one or more data processors, cause the one or more data processors to perform part or all of one or more methods disclosed herein.
In some embodiments, a computer-program product is provided that is tangibly embodied in a non-transitory machine-readable storage medium and that includes instructions configured to cause one or more data processors to perform part or all of one or more methods disclosed herein.
Some embodiments of the present disclosure include a system including one or more data processors. In some embodiments, the system includes a non-transitory computer readable storage medium containing instructions which, when executed on the one or more data processors, cause the one or more data processors to perform part or all of one or more methods and/or part or all of one or more processes disclosed herein. Some embodiments of the present disclosure include a computer-program product tangibly embodied in a non-transitory machine-readable storage medium, including instructions configured to cause one or more data processors to perform part or all of one or more methods and/or part or all of one or more processes disclosed herein.
The terms and expressions which have been employed are used as terms of description and not of limitation, and there is no intention in the use of such terms and expressions of excluding any equivalents of the features shown and described or portions thereof, but it is recognized that various modifications are possible within the scope of the invention claimed. Thus, it should be understood that although the present invention as claimed has been specifically disclosed by embodiments and optional features, modification and variation of the concepts herein disclosed may be resorted to by those skilled in the art, and that such modifications and variations are considered to be within the scope of this invention as defined by the appended claims.
The patent or application file contains at least one drawing executed in color. Copies of this patent or patent application publication with color drawing(s) will be provided by the Office upon request and payment of the necessary fee.
The present disclosure is described in conjunction with the appended figures:
In the appended figures, similar components and/or features can have the same reference label. Further, various components of the same type can be distinguished by following the reference label by a dash and a second label that distinguishes among the similar components. If only the first reference label is used in the specification, the description is applicable to any one of the similar components having the same first reference label irrespective of the second reference label.
Systems and methods disclosed herein can generate and use quantitative metrics for characterizing immune phenotypes. The metrics can characterize a quantity and spatial distribution of a given cell type as determined by processing immunohistochemistry images. A particular use case is to generate and use these metrics to characterize ovarian cancer.
In some embodiments, gene-expression data are accessed (e.g., received from a computing system associated with a laboratory or care-provider office) and used to predict a tumor-immune phenotype. The prediction may be generated using a computing system that is co-located with and/or includes the computing system associated with the laboratory or care-provider office and/or using a computing system that is remote from the computing system associated with the laboratory or computing system. For example, the prediction may be generated using a cloud computing system (e.g., that includes one or more servers, one or more processors and/or one or more memories).
The phenotype prediction can be generated using a machine-learning model, such as a classifier. The gene-expression data can identify expression levels of one or more genes in Table 1 (e.g., at least 1, at least 10, at least 50, at least 100 or at least 120 of the genes in Table 1) and/or one or more genes for which expression levels correlate with and/or are predictive of a quantity, spatial distribution and/or locations of CD8+T cells. The gene-expression data can identify expression levels for a set of genes. The set of genes may include one or more genes (e.g., or 5 or more, 10 or more, 20 or more or 50 or more) for which expression levels are correlated with, predictive of, and/or informative as to CD8+T cell spatial distribution. The set of genes may include one or more genes (e.g., or 5 or more, 10 or more, 20 or more or 50 or more) for which expression levels are correlated with, predictive of, and/or informative as to CD8+T cell quantity. The set of genes can include at least 1 gene, at least 10 genes, at least 20 genes or at least 50 genes of genes identified in Rows 1-56 of Table 1; at least 1 gene, at least 10 genes, at least 20 genes or at least 50 genes of genes identified in Rows 57-244 of Table 1 and/or at least 1 gene, at least 10 genes, at least 20 genes or at least 50 genes of genes identified in Rows 245-346 of Table 1.
In some instances, the machine-learning model (and/or another machine-learning model) may identify one or more genes that are represented in the gene-expression data. For example, a set of parameters (e.g., weights) that are learned and/or fit by the machine-learning model may represent a degree to which expression of various genes are predictive of a quantity and/or location of CD8+T cells, and at least one of the one or more genes may be determined based on the parameters (e.g., using an absolute or relative threshold). As another example, a pre-configured input data set may be used to interpret the model and to decipher whether and/or an extent to which expression of various genes influence phenotype predictions, and at least one of the one or more genes can be identified based on the interpretation.
A tumor-immune phenotype can correspond to a presence, density and/or location of CD8+T cells. For example, a tumor-immune phenotype can include 1) an inflamed/infiltrated phenotype in which CD8+T cells infiltrate the tumor epithelium; 2) an immune excluded phenotype in which infiltrating CD8+T cells accumulate in the tumor stroma rather than the tumor epithelium, and 3) an immune desert phenotype in which CD8+T cells are either absent or present in very low numbers. It will thus be appreciated that a tumor-immune phenotype may include one traditionally identified by analyzing one or more digital-pathology images.
Thus, in some instances, one or more genes for which expression levels are used to predict a phenotype may be determined by training a machine-learning model to learn the extent to which expression levels of individual genes are predictive of a phenotype determined (e.g., using a computer algorithm and/or manual annotation) by analyzing digital pathology images. The machine-learning model may be configured to learn the extent to which expression levels of various genes are predictive of traditional phenotypes (e.g., inflamed/infiltrated, immune excluded or immune desert phenotypes).
The machine-learning model may alternatively or additionally be configured to learn the extent to which expression levels of various genes are predictive of one or more novel and/or non-traditional phenotypes. For example, the machine-learning model may classify various gene-expression data sets into distinct clusters, and each of some or all of the clusters may be associated with a phenotype (e.g., corresponding to a potential label output of the machine-learning model). The clustering can include a connsensus clustering. The phenotype associated with each phenotype may be determined based on (for example) CD8+T cell characteristiccs (e.g., quantity and/or spatial distribution) associsted with training data associated with the cluster.
A tumor-immune phenotype can be used to inform treatment decisions and/or generate predictions as to whether and/or a degree to which a particular subject will respond to a particular treatment. For example:
A computer system may use one or more rules and/or a look-up table to identify a recommended treatment based on a predicted phenotype. An output of the computing system (e.g., that is locally presented and/or transmitted to another device) may include a predicted phenotype, a recommended treatment and/or expression levels of one or more genes (e.g., used to generate the predicted phenotype).
Digital pathology system 105 can be configured to generate one or more digital images corresponding to a particular sample. For example, an image can include a stained section of a biopsy sample. As another example, an image can include a slide image (e.g., a blood film) of a liquid sample.
Some types of samples (e.g., biopsies, solid samples and/or samples including tissue) can be processed by a fixation/embedding system to fix and/or embed the sample. The sample can be infiltrated with a fixating agent (e.g., liquid fixing agent, such as a formaldehyde solution) and/or embedding substance (e.g., a histological wax). For example, a fixation sub-system can fixate a sample by exposing the sample to a fixating agent for at least a threshold amount of time (e.g., at least 3 hours, at least 6 hours, or at least 12 hours). A dehydration sub-system can dehydrate the sample (e.g., by exposing the fixed sample and/or a portion of the fixed sample to one or more ethanol solutions) and potentially clear the dehydrated sample using a clearing intermediate agent (e.g., that includes ethanol and a histological wax). An embedding sub-system can infiltrate the sample (e.g., one or more times for corresponding predefined time periods) with a heated (e.g., and thus liquid) histological wax. The histological wax can include a paraffin wax and potentially one or more resins (e.g., styrene or polyethylene). The sample and wax can then be cooled, and the wax-infiltrated sample can then be blocked out.
A sample slicer can receive the fixed and embedded sample and can produce a set of sections. The sample slicer can expose the fixed and embedded sample to cool or cold temperatures. The sample slicer can then cut the chilled sample (or a trimmed version thereof) to produce a set of sections. Each section may have a thickness that is (for example) less than 100 μm, less than 50 μm, less than 10 μm or less than 5 μm. Each section may have a thickness that is (for example) greater than 0.1 μm, greater than 1 μm, greater than 2 μm or greater than 4 μm. The cutting of the chilled sample may be performed in a warm water bath (e.g., at a temperature of at least 30° C., at least 35° C. or at least 40° C.).
An automated staining system can facilitate staining one or more of the sample sections by exposing each section to one or more staining agents. Each section may be exposed to a predefined volume of staining agent for a predefined period of time. In some instances, a single section is concurrently or sequentially exposed to multiple staining agents. The multiple staining agents may include (for example) haematoxylin and a primary antibody (e.g., CD8 immunohistochemistry).
Each of one or more stained sections can be presented to an image scanner, which can capture a digital image of the section. The image scanner can include a microscope camera. The image scanner may be further configured to capture annotations and/or morphometrics identified by a human operator.
In some instances, a section is returned to the automated staining system after one or more images are captured, such that the section can be washed, exposed to one or more other stains and imaged again. When multiple stains are used, the stains may be selected to have different color profiles, such that a first region of an image corresponding to a first section portion that absorbed a large amount of a first stain can be distinguished from a second region of the image (or a different image) corresponding to a second section portion that absorbed a large amount of a second stain.
It will be appreciated that one or more components of digital pathology system 105 may, in some instances, operate in connection with human operators. For example, human operators may move the sample across various sub-systems (e.g., of a fixation embedding system or of an image-generation system) and/or initiate or terminate operation of one or more sub-systems, systems or components of digital pathology system 105.
Further, it will be appreciated that, while various described and depicted functions and components of digital pathology system 105 pertain to processing of a solid and/or biopsy sample, other embodiments can relate to a liquid sample (e.g., a blood sample). For example, digital pathology system 105 may be configured to receive a liquid-sample (e.g., blood or urine) slide, that includes a base slide, smeared liquid sample and cover. The image scanner can then capture an image of the sample slide.
The digital pathology images may be processed at digital pathology system and/or at a remote system. In some instances, image processing can include aligning multiple images corresponding to a same sample. For example, multiple images may correspond to a same section of a same sample. Each image may depict the section stained with a different stain. As another example, each of multiple images may correspond to different sections of a same sample (e.g., each corresponding to a same stain or for which different subsets of the images correspond to different stains). For example, alternating sections of a sample may have been stained with different stains. Section alignment can include determining whether and/or how each image is to be translated, rotated, magnified and/or warped such that images corresponding to a single sample and/or to a single section are aligned. An alignment may be determined using (for example) a correlation assessment (e.g., to identify an alignment that maximizes a correlation).
Image processing can further include automatically detecting depictions of objects (e.g., biological objects) of one or more particular types in each of the aligned images. Object types may include types of cells or types of biological structures. For example, a first set of objects may correspond to a particular (e.g., labeled) cell type, such as T cells or CD8+T cells, and a second set of objects may correspond to a tumor region. In some instances, at least one type of object is identified via manual annotations. For example, input from a human annotator may identify a border of a tumor region, and automated cell detection may identify locations (e.g., borders or point locations) of CD8+T cells. In some instances, all objects are detected via automated detection (e.g., where tumor epithelium are distinguished from stroma epithelium using an algorithm that distinguishes shape and size of tumor nuclei from stroma nuclei). Cells may be detected using the counterstain signals, and a primary protein of interest may be evaluated using the haematoxylin signal. A DAB intensity statistic (e.g., mean DAB intensity) may be calculated for each nucleus.
In some instances, objects of different types are detected within a same image. In some instances, objects of a first type are detected within a first image, and one or more objects of a second type are detected within a second image (associated with a same or different sample slide).
Object detection may use static rules and/or a trained model to detect and characterize objects. Rules-based object detection can include (for example) detecting one or more edges, identifying a subset of edges that are sufficiently connected and closed in shape and/or detecting one or more high-intensity regions or pixels. A portion of an image may be determined to depict an object if (for example) an area of a region within a closed edge is within a predefined range and/or if a high-intensity region has a size within a predefined range. Detecting object depictions using a trained model may include employing a neural network, such as a convolutional neural network, a deep convolutional neural network and/or a graph-based convolutional neural network. The model may have been trained using annotated images that included annotations indicating locations and/or boundaries of objects. The annotated images may have been received from a data repository (e.g., a public data store) and/or from one or more devices associated with one or more human annotators.
Rules-based object detection and trained model object detection may be used in any combination. For example, rules-based object detection may detect depictions of one type of object while a trained model is used to detect depictions of another set of object. Another example may include validating results from rules-based object detection using objects output by a trained model, or validating results of the trained model using a rules-based approach. Yet another example may include using rules-based object detection as an initial object detection, then using a trained model for more refined object analysis, or applying a rules-based object detection approach to an image after depictions of an initial set of objects are detected via a trained network.
Object detection can also include (for example) pre-processing an image to (for example) transform a resolution of the image to a target resolution, apply one or more color filters, and/or normalize the image. For example, a color filter can be applied that passes colors corresponding to a color profile of a stain used to stain a sample. Rules-based object detection or trained model object detection may be applied to a pre-processed image.
For each detected object, a single representative location of the depicted object (e.g., centroid point or midpoint), a set of pixels or voxels corresponding to an edge of the depicted object and/or a set of pixels or voxels corresponding to an area of the depicted object may be identified and stored as object data. This object data can be stored with an identifier of the object (e.g., a numeric identifier), an identifier of a corresponding image, an identifier of a corresponding subject and/or an identifier of the type of object.
Gene-expression detection system 110 can be configured to detect the expression level of each of a set of genes. Gene expression levels may represent the extent to which DNA is converted to a functional product, such as a protein. Gene-expression detection system 110 can determine gene-expression levels by measuring mRNA that corresponds to a precursor for a protein or by measuring proteins directly. Exemplary techniques that may be used by gene-expression detection system 110 include Northern blotting, Western blotting, RT-qPCR, flow cytometry, and RNA-Seq.
Northern blotting involves separating a sample of RNA on an agarose gel. The RNA sample can be radioactively labeled to generate RNA that is complementary to a target sequence. The radioactively labeled RNA can then be detected by an autoradiograph to determine size and sequence information about the mRNA. Labelling may also be performed using digoxigenin and biotin substances.
Western blotting involves a similar process as Northern blotting, but Western blotting measures protein levels instead of mRNA levels. During Western blotting, electrophoresis is performed on the protein sample to separate individual proteins into distinct bands. The proteins can then be transferred to a treated piece of paper. The paper is incubated with an antibody for the target protein so that the antibody binds to the target protein.
In RT-qPCR, a complementary DNA (cDNA) template is generated for an mRNA sample during reverse transcription. Then, during quantitative PCR, the cDNA is amplified. A labeled hybridization probe or dye with a known fluorescence may be used during the amplification. A measurement of the number of copies of original mRNA can be determined using a standard curve. RT-qPCR provides the ability to detect a single mRNA molecule, but the process can be expensive depending on the probe or dye used.
Flow cytometry involves analyzing gene expression at a single-cell level. A biological sample containing DNA is injected into a flow cytometer and cells flow one at a time through a channel. A beam of light illuminates the cells and detectors record an intensity and duration of a signal of scattered light by each cell. Fluorophore labels, dyes, and stains with a known emission signal can be attached to an antibody of a target protein to quantify protein levels in each cell of the sample. In addition to providing quantification at the single-cell level, flow cytometry allows multiple proteins to be targeted at a time, reducing time involved in analysis.
During RNA-Seq, cDNA fragments are generated from RNA molecules. The cDNA molecules are then sequenced using high-throughput techniques. The reads can be aligned to a reference genome or reference transcripts to determine gene expression levels. RNA-Seq allows the entire transcriptome (e.g., mRNA, rRNA, tRNA) to be analyzed. RNA-Seq is not limited to genes that encode proteins, and thus, detects genes that do not encode proteins. However, RNA-Seq is relatively easy to perform and provides accurate quantification of gene expression levels.
Gene-expression detection system 110 may perform normalization (e.g., to counts per million), filtering (e.g., to remove lowly expressed genes), and/or transformations. Outliers may be removed, such as by using a component analysis technique (e.g., principal component analysis).
Each samples processed by digital pathology system 105 may have been collected from a subject. One or more different users (e.g., one or more physicians, laboratory technicians and/or medical providers) may have initiated the collection of the sample, initiated the processing of the sample and/or may receive results of processing of the sample. An associated user can include a person who ordered a test or biopsy that produced a sample being imaged and/or a person with permission to receive results of a test or biopsy. For example, a user can correspond to a physician or a subject (from whom a sample was taken) him/herself. A user can use one or one user devices 120 to (for example) initially submit one or more requests (e.g., that identify a subject) that a sample be processed by digital pathology system 105.
In some instances, each of digital pathology system 105 and/or gene-expression detection system 110 transmits results directly to expression-based phenotype classification system 115. In some instances, each of digital pathology system 105 and/or gene-expression detection system transmits results to user device 120, which can initiate automated processing of the results by expression-based phenotype classification system 115.
Expression-based phenotype classification system 115 can include a label generator 120 that can assign one or more labels to each subject's data in a training set based on objects detected within the subject's digital pathology images. The labels may include a first “quantity” label characterizing a quantity of depictions of a particular object type (e.g., CD8+T cells) and a second “spatial-distribution” label characterizing a spatial distribution of depictions of a particular object type (CD8+T cells). The quantity label may include and/or may be based on a count (e.g., raw or normalized count, such as a density) of depictions of the object type within one or more regions. For example, the quantity label may be defined to be the sum of depictions of CD8+T cells in stroma versus tumor regions or the square root of the sum of the square of the count of CD8+T cells in the stroma regions and the square of the count of CD8+T cells in the tumor regions.
The spatial-distribution label may be based on a difference, ratio and/or angle between a count (e.g., raw or normalized count, such as a density) of depictions of the object type within a first region and a count of depictions of the object type within a second region. For example, the spatial-distribution label may be defined to be the arctangent of the ratio of a count of CD8+T cells in the stroma regions relative to a count of the CD8+T cells in the tumor regions. Thus, if all of the CD8+T cells are in the tumor regions, the spatial-distribution label would be 0.
Expression-based phenotype classification system 115 further includes a gene-significance detector 125 that uses the gene-expression data and the labels to determine, for each gene (of a set of genes for which expression levels were measured), whether the gene is specific to a quantity prediction (predicting a quantity of CD8+T cells), spatial-distribution prediction (predicting a distribution of CD8+T cells across tumor versus stroma cells), both or neither. Gene-significance detector 125 may, for each of the set of genes, fit or train a model using the labels and gene-expression data from the training data. The model may include (for example) a regression model and/or random-forest regression model. Gene-significance detector 125 may characterize a gene as being specific to a quantity prediction (or spatial-distribution prediction) when an increase in a mean-square error of the quantity prediction (or spatial-distribution prediction) was above a predefined threshold (e.g., a bottom threshold of a fourth quartile). In some instances, a given gene is specific both to a quantity prediction and to a spatial-distribution prediction. In some instances, a given gene is not specific both to a quantity prediction or to a spatial-distribution prediction.
A phenotype clustering controller 130 can use expression levels from the training data for the genes determined to be specific to quantity predictions and for genes specifc to spatial-distribution predictions to perform a clustering analysis (e.g., consensus clustering). In some instances, training data pertaining to genes determined to be specific both to quantity and spatial-distribution predictions were further used for the clustering analysis. For example, the immune desert phenotype may be associated with smaller quantity predictions (predicting fewer CD8+T cells), an immune infiltrated phenotype may be associated with a spatial-distribution prediction predicting presence of CD8+T cells in tumor regions, and an immune excluded phenotype may be associated with a spatial-dstribution prediction predicting relatively few CD8+T cells in tumor regions and more CD8+T cells in stroma regions.
The clustering analysis may implement a constraint on a number of clusters. Phenotype clustering controller 130 can assign each of the clusters to an immune phenotype based on the labels associated with the clusters. Immune phenotypes to which a cluster may be assigned may include immune desert, immune excluded or immune infiltrated.
Thus, multiple machine-learning models may be used to identify the genes that are specific to T-cell quantity and distribution and to characterize how expression of those genes are associated with immune phenotypes.
While digital-pathology images can be used to identify particular genes that are informative and/or predictive as to immune phenotype and can also be used to identify genetic profiles associated with immune phenotypes, the particular genes and genetic profiles may then be used to support predicting immune-phenotype prediction without relying on digital pathology images. Thus, phenotype clustering controller 130 may be configured to receive a new data set of gene-expression levels corresponding to a particular subject from gene-expression detection system 110 (which may be a same or different system as one contributing to training data) and may assign the data set to a particular cluster and to thus predict a phenotype associated with the cluster for the particular subject.
Each component and/or system depicted in
Blocks 210-220 may be performed (e.g., at expression-based phenotype classification system 115) for each subject in the set of subjects. At block 210, a set of CD8+T cell depictions in the digital pathology image(s) corresponding to the subject can be identified. For example, each digital pathology image may have been subjected to CD8+IHC staining and mjr fyt}
At block 215, each detected CD8+T cell is assigned to a category to indicate whether it is within a tumor region or a stroma region. In some instances, a human annotator may have identified each of one or more tumor and/or stroma regions within the image (or another version thereof), and a mapping may be used for the categorization. In some instances, an automated processing is used to predict which portions of the image correspond to tumor (versus stroma regions). For example, hematoxylin signals may be predictive of whether a given cell is within a tumor region, as nuclei in tumors may have greater asymmetry and size outliers. A neighbor, cluster, convolution-network or other approach may then be used to process nuclei assignments to predict tumor/stroma regions.
At block 220, a quantity label and spatial distribution label can be generated for the subject based on the CD8+T cell detections and classifications. The quantity label may be based on (for example) a total number of detected CD8+T cells, a (normalized or unnormalized) number of CD8+T cells detected in each stroma region, a (normalized or unnormalized) number of CD8+T cells detected in each tumor region, a square of a number of CD8+T cells detected in each stroma region, and/or a square of a number of CD8+T cells detected in each tumor region. For example, the quantity label can be defined to be a square root of a sum of a square of a number of CD8+T cells detected in each stroma region and a square of a number of CD8+T cells detected in each tumor region. The spatial-distribution label may be based on (for example) a difference between, a ratio or and/or an angle between a (normalized or unnormalized) number of CD8+T cells detected in each stroma region and a (normalized or unnormalized) number of CD8+T cells detected in each tumor region. In some instances, the quantity label and the spatial-distribution label can be configured to be represented as polar coordinates.
At block 225, a regression model may be used (e.g., by expression-based phenotype classification system 115) to identify which genes of the set of genes represented in the expression data are specific to CD8+T cell quantity and/or CD8+T cell spatial distribution. For each of the set of genes, a first model may be trained and/or a first function may be fit to determine an extent expression of the gene is predictive of and/or informative of (e.g., in terms of entropy reduction) values of the quantity label. Similarly, a second model may be trained and/or a second function may be fit to determine an extent expression of the gene is predictive of and/or informative of (e.g., in terms of entropy reduction) values of the spatial-distribution label. The first and second models and/or functions may be of a same or different type. The first and/or second models and/or functions may include a regression function and/or a random forest regression model. Training a model and/or fitting a function may result in determining one or more parameters and/or weights, which may then be compared to a threshold to assess specificity. The threshold may include an absolute threshold or relative threshold (e.g., defined based on the parameters and/or weights identified across the set of genes). A subset of the set of genes determined to be sufficiently specific may be determined based on the threshold analysis. In some instances, the subset includes genes within the set of genes determined to be sufficiently specific for the quantity variable or for the spatial-distribution variable. In some instances, the subset includes genes within the set of genes determined to be sufficiently specific for the quantity variable and/or for the spatial-distribution variable.
At block 230, a cluster analysis is performed using expression values for genes determined to be sufficiently specific. The cluster analysis may include using a component analysis, such as principal component analysis or independent component analysis. The cluster analysis may limit a number of clusters (e.g., to 3, 4, 5, 6, 7, 8, etc.). The cluster analysis may be unsupervised and/or performed only based on quantity and spatial-distribution values.
At block 235, each of the clusters may be assigned to an immune phenotype based on quantity and/or spatial-distribution labels associated with data points (associated with subjects) assigned to the cluster. The immune-phenotype assignment may be based on whether cluster-associated quantity labels were low or high and/or whether cluster-associated spatial-distribution labels were indicative of CD8+T cell enrichment in the stroma versus in tumors. Potential immune-phenotype assignments include immune desert, immune excluded or immune infiltrated. For example, the immune desert phenotype may be associated with low CD8+T cell quantity labels; the immune excluded phenotype may be associated with high CD8+T cell quantity labels and spatial distribution labels indicating stroma concentration; and the immune infiltrated phenotype may be associated with high CD8+T cell quantity labels and spatial distribution labels indicating tumor concentration.
At block 240, cluster data is stored. The cluster data may indicate how the clusters are differentiated from each other (e.g., via one or more hyperplanes, weight assessments, principal components, ranges of quantity and/or spatial-distribution values, etc.). The cluster data may further identify, for each cluster, to which immune phenotype the cluster corresponds.
At block 310, a cluster assignment is generated using the new gene-expression data and cluster data (e.g., that was stored at block 240 in process 200). For example, each of the expression levels in the new gene-expression data may be weighted and/or transformed (e.g., using one or more components) to generate a set of coordinates in a representative space. A distance between the coordinates and each of a set of reference coordinates (corresponding to multiple clusters) may be calculated to identify a cluster associated with a minimum distance. In some instances, a cluster assignment is generated using a nearest-neighbor or K-means approach.
At block 315, it is determined that the cluster assignment corresponds to a particular immune phenotype. The determination may be made using a look-up from data in the cluster data (e.g., that was stored at block 240 in process 200).
At block 320, a result is output based on the particular immune phenotype. The result may identify the particular immune phenotype, a treatment predicted to be effective for the particular immune phenotype, a predicted efficacy of a particular treatment given the predicted particular immune phenotype, etc. The result may further be accompanied by (for example) some or all of the new gene-expression data (or a processed version thereof).
In some instances, a prediction of a molecular subtype of a tumor is generated based on a predited immune phenotype. For example, it may be predicted that a particular subject has an immunoreactive molecular subtype of ovarian cancer when it is predicted that genetic expression data for the subject corresponds to an infiltrated immune phenotype. As another example, it may be predicted that a particular subject has a mesenchymal molecular subtype of ovarian cancer when it is predicted that genetic expression data for the subject corresponds to an excluded immune phenotype. As yet another example, it may be predicted that a particular subject has either a differentiated molecular subtype or a proliferative molecular subtype when it is predicted that genetic expression data for the subject corresponds to an immune desert phenotype.
In some instances, immune phenotype predictions may be used to investigate and identify pathways and immune features of a particular immune phenotype (e.g., an excluded phenotype). More specifically, an immune phenotype may be predicted based on expression levels of multiple genes (e.g., in accordance with process 300), and in situ analysis may be performed to detect whether and/or an extent to which a particular phenotype is associated with one or more particular types of upregulation or downregulation. For example, as further detailed in Section IV.D. below, phenotype predictions and transcriptional analysis can be used to predict that the immune excluded phenotype is associated with upregulation of TGFβ and stromal activation and the loss of antigen presentation on tumor cells. As another example, phenotype predictions and transcriptional analysis can be used to predict that the immune excluded phenotype and a subset of the immune desert phenotype are associated with a downregulation of HLA-A.
It will be appreciated that treatments may be informed, selected and/or provided based on the immune phenotype predictions and/or based on predicted pathways and/or immune features of particulaar immune phenotypes. For example, it may be inferred or determined that a tumor of a subject has an immunosuppressive microenvironment (e.g., by processing a sample to assess immunoactivity or based on gene-expression data). A treatment of an inhibitor of TGFβ may then be provided to the subject. As another example, it may be inferred or determined that a subject has a medical condition associated with reduced expression of HLA-A relative to healthy subjects. The medical condition may include an immune excluded phenotype of cancer. A treatment including an inhibitor of TGFβ can then be provided to the subject. As another example, it may be inferred or determined that a subject has a medical condition associated with reduced expression of HLA-A relative to healthy subjects. The medical condition may include an immune excluded phenotype of cancer. A treatment including an IFNγ treatment and a EZH2 or DNMT inhibitor.
IV.A. Technique for Processing Immunohistochemistry Images to Generate CD8 T Cell Quantity and/or Distribution Metrics
Digital pathology images were accessed, which depict stained samples. More specifically, CD8 immunohistochemistry with a haematoxylin counter-staining was performed on each of a set tissue samples collected from a set of subjects in the ICON7 trial having ovarian cancer (n=155). Cell-type detection was performed. Each detected cell was assigned to a category (e.g., a tumor epithelium cell or stromal cell). The assignment was based on a size and shape of a nucleus. CD8+T cell densities in the tumor epithelium and/or CD8+T cell densities in the stroma compartment were calculated based on the categorizations.
Metrics were defined to include a total CD8+T cell count, a CD8+T cell count per tumor epithelium and/or a CD8+T cell count stroma area (See
These two digitally defined quantitative metrics were used to profile the immune phenotype of each tumor using a two-dimensional map (
IV.B. Machine-Learning Processing of CD8 T Cell Quantity and/or Distribution Metrics to Identify Tumor-Immune Phenotype
A gene expression-based molecular classifier was generated using a machine learning approach to characterize tumor-immune phenotypes.
As indicated in blocks 1 and 2 of
In an exemplary case, a training data set was defined to include data from 155 samples from the ICON7 trial. By assessing the learned data, 352 genes were identified for which expression of the gene was significantly related to the quantity (R) and/or spatial distribution of CD8+T cells (θ) (See
Focusing on the 159 genes that are associated with either the quantity or spatial distribution of CD8+T cells, consensus clustering was performed on the training data. Six clusters were detected with distinct molecular profiles (
A 157-gene classifier was built to distinguish these three tumor-immune phenotypes, by applying the Prediction Analysis of Microarrays (PAM) approach to the training set (
Four clinically and biologically relevant molecular subtypes, i.e. immunoreactive (IMR), mesenchymal (MES), proliferative (PRO) and differentiated (DIF), have been previously identified in ovarian cancer. The relationship between the tumor-immune phenotypes defined in this study and the predicted molecular subtypes based on previously developed classifier (as described in Verhaak, R. G. et al. Prognostically relevant gene signatures of high-grade serous ovarian carcinoma. J Clin Invest 123, 517-525, (2013) and Tothill, R. W. et al. Novel molecular subtypes of serous and endometrioid ovarian cancer linked to clinical outcome. Clin Cancer Res 14, 5198-5208, (2008), which are hereby incorporated by reference in their entireties for all purposes, was assessed. As shown in
Finally, the results indicate a significant association of the tumor-immune phenotypes with clinical outcome in ovarian cancer. A Cox proportional hazards analysis was performed on the dataset from 172 patients enrolled in a chemo-control arm of the ICON7 clinical trial with uniform follow-up. As shown in
Molecular features associated with the two quantitative metrics defining distinct immune phenotypes were identified.
In order to gain a more comprehensive understanding of the biology underlying these tumor-immune phenotypes, pathway enrichment analysis was performed on the full transcriptome of the 370 ICON7 samples. Based on two databases, KEGG (Antigen processing and presentation and Chemokine signaling) and Hallmark (IFNγ response, WNT-β-catenin signaling, TGFβ signaling and Angiogenesis), molecular pathways significantly enriched in each tumor-immune phenotype were summarized in
Furthermore, pathway analysis revealed additional molecular features characterizing the distinct tumor-immune phenotypes. The pathways characterizing the infiltrated and desert phenotypes are represented in
To evaluate in more detail which specific immune and stromal cell types are associated with a given immune phenotype, a cell type enrichment analysis was performed using xCell, a gene signature-based deconvolution method, on the bulk RNAseq datasets of ICON? study (n=370). The deconvolution analysis confirmed many findings from the machine learning and pathway enrichment analyses, including a high overall immune score in infiltrated and excluded tumors, and the highest overall stromal score in the excluded tumors (
Genetic components, such as tumor mutation burden (TMB), neo-antigen burden, and high genomic instability including microsatellite instability high (MSI-H) and deficient mismatch repair (dMMR), have been shown to associate with increased T cell infiltration and better responses to checkpoint inhibitors in some cancer types. To investigate the impact of genetic components in ovarian cancer in the context of tumor-immune phenotypes, the published ovarian cancer TCGA dataset (n=412) was accessed. Both bulk RNAseq and whole exome sequencing data are available for this dataset. Using the RNAseq data, a tumor-immune phenotype was predicted for each of 412 ovarian tumor samples in the TCGA dataset by applying the 157-gene molecular classifier (
Integrated digital pathology and transcriptional analysis can be used to uncover biological pathways and immune features underlying the T cell excluded phenotype, including the upregulation of FAP, a marker of activated stroma and downregulation of antigen presentation genes. To validate these findings and distinguish which cell compartment underwent these molecular changes, in situ analysis was performed on an independent ovarian tumor collection of 84 samples. RNAseq transcriptome analysis was performed on these samples and their tumor-immune phenotypes were predicted based on the 157-gene classifier developed in this study (
Consistent with the findings from the ICON7 dataset, infiltrated and excluded tumor-immune phenotypes have similar abundant quantities of CD8+T cells by in situ analysis (
Further, these in situ analyses identified specific cell compartments contributing to these observed modulations. For example, the downregulation of MHC class I in the excluded tumors was restricted to the tumor compartment. In contrast, the infiltrated tumors exhibited strong and homogenous MHC class I staining on tumor cells. On the other hand, the desert tumors exhibited both intra-tumor and inter-tumor heterogeneity in MHC class I expression. This heterogeneity was reflected by an intermediate H-Scores for MHC class I in the tumor epithelium (
Assessments were performed to determine the mechanism of downregulation of MHC class I expression in the ovarian tumor cells. Defects of antigen presentation machinery in tumor cells by downregulation of MHC class I expression via genetic mutations or epigenetic suppression have been shown to represent an important mechanism of immune escape in multiple cancers_ENREF_23. The detection of somatic mutations in the HLA genes has been previously studied in different TCGA cohorts including the ovarian cohort. Unlike colon and head and neck cancer, mutations in HLA genes are rare in ovarian cancer samples, indicating loss of MHC-I is not likely due to genetic mutations.
Further assessments were performed to determine whether the loss of MHC class I expression is due to epigenetic regulation. To specifically detect the methylation on tumor cells, DNA methylation profiles were generated for a panel of 48 ovarian cancer cell lines using the Infinium Human Methylation 450K Chip. A strong anti-correlation was observed between the methylation level of the promoter region of the HLA-A gene (beta value) and its expression level (Log2(RPKM+1)) (
More specifically, in ovarian cancer cell lines with hypermehtylation of HLA-A promoter, treatment with demethylating agent 5-aza-2′-deoxycytidine, a DNA methyltransferase (DNMT) inhibitor, was shown to be able to significantly induce the expression of MHC class I protein at the tumor cell surface (
Parallel to the downregulation of MHC-I in tumor cells, another primary feature of the excluded tumors is the upregulation of TGFβ/reactive stroma genes. TGFβ has been shown to downregulate MHC class I on uveal melanoma cells in vitro and TGFβ1 null mice exhibited an aberrant expression of MHC-I and MHC-II in tissues. To determine whether TGFβ might play a direct role in downregulation of the expression of MHC class I on ovarian tumor cell, two MHC-Ihigh-expressing ovarian cancer cell lines were etreaeted with TGFβ1. Flow cytometry analysis revealed that TGFβ1 decreased the surface expression of MHC-I by 37.7±3.2% in SK-OV-3 and 40.45±14.2% in OVCA-420 compared to the untreated cells. Further, in the presence of Galunisertib, a small molecule TGFβ inhibitor targeting the TGFβRI, MHC class I expression was restored to the untreated level (
In addition to loss of MHC class I expression on tumor cells, other features associated with the T cell excluded tumors include enriched TGFβ expression and signaling (
In addition, the data also suggests that TGFβ may contribute to an overall immunosuppressive tumor microenvironment in the T cell excluded tumors. Supporting this notion, TGFβ specifically induced the expression of several immune-modulatory molecules in the fibroblast cells, including tumor promoting cytokines, IL11, and TNFAIP6, a potent anti-inflammatory molecule previously reported to inhibit the recruitment of neutrophils and shift pro-inflammatory vs. anti-inflammatory protein profiles in macrophages to elicit immune suppression (
Finally, supporting the findings from the in vitro studies, the data indicates that many of the TGFβ induced ECM and immune-modulatory genes in vitro, were also specifically enriched in the T cell excluded tumors in the ICON7 dataset (
IV.G. Anti-TGFβ Enhances Anti-Tumor Activity in Combination with PD-L1 in Ovarian Cancer Mouser Model
Thus, TGFβ may have a central role in mediating CD8+T cell exclusion and immune suppression in ovarian cancer. To determine whether blocking TGFβ signaling can provide synergy to checkpoint inhibitors in ovarian cancer mouse model, immunocompetent mice subcutaneously implanted with BrKrasX1.3 ovarian cancer cells were treated (approximately 13 days after tumor inoculation) with the isotype control, anti-PD-L1, anti-TGFβ or a combination of anti-PD-L1 and anti-TGFβ antibodies according to the schedule shown in
To further investigate the underlying mechanisms of action, pharmacodynamic changes of anti-TGFβ, anti-PD-L1, alone or in combination were characterized in the BrKrasX1.3 ovarian cancer mouse model at day 8 post the initiation of the treatment, while no difference of tumor mass was noticeable between the groups (
Collectively, these results provided pre-clinical proof of concept and potential mechanisms of action for targeting the TGFβ pathway as a novel therapeutic strategy to overcome T cell exclusion and immune suppression, and ultimately improve the patient response to cancer immunotherapy.
In the present embodiments, a novel digital image analysis algorithm was developed to quantify the quantity and spatial distribution of CD8+T cells in the tumor microenvironment. Coupling this digital pathology algorithm with transcriptome analysis in a large cohort of archival tumor tissues from the ICON7 Phase III clinical trial, a random forest machine learning algorithm was built to classify tumor-immune phenotypes in ovarian cancer. This approach yielded a set of high-dimensional quantitative metrics to define tumor-immune phenotypes. The described Example provides the first proof of concept of classifying tumor-immune phenotypes based on a gene expression classifier. The novel approach developed in this study may enable systematic characterization of tumor-immune phenotypes in large clinical trials and translational studies, in which availability of CD8 IHC image analysis are often limited. With additional validation and optimization, the molecular classifier developed in this study may be widely applicable to classify tumor-immune phenotypes in other solid tumor types.
Although a computational framework, Tumor Immune Dysfunction and Exclusion (TIDE), can be used to identify factors that predict cancer immunotherapy response. The study represents the first study to integrate digital pathology and machine learning and provide a systematic characterization of molecular features defining distinct tumor-immune phenotypes in human cancer. One conclusion is that tumor-immune phenotypes should be studied and interpreted in the context of disease biology. For example, the immune desert tumors in ovarian cancer are heterogeneous and comprise of two distinct molecular subtypes, the differentiated and the proliferative subtype, which are associated with different clinical outcomes in ICON7 study (
Using immunohistochemistry and sequence data also facilitated a discovery of two hallmark features characterizing the T cell excluded tumors, including 1) loss of antigen presentation on tumor cells and 2) upregulation of TGFβ and stromal activation. Further, this study further dissected the functional role of TGFβ in mediating T cell exclusion and immune suppression in ovarian cancers.
The data revealed that the downregulation of MHC class I in ovarian cancer cells may be regulated by epigenetic mechanisms. Supporting this finding, there was a strong anti-correlation between the HLA-A gene expression and promoter methylation levels. Further, IFNγ treatment as well as EZH2 or DNMT inhibition may overcome such epigenetic regulation and increase HLA-A expression in selected ovarian cancer cells. For example, a previous study has shown that a subset of cancers harbouring mutations in the SWI/SNF ATPase, SMARCA4, is sensitive to EZH2 inhibition. Indeed, as shown in
Further, loss of MHC-I expression regulated by epigenetic mechanisms as a result of immune pressure associated with an absence of CD8+T cell infiltration in relapsing tumors has been previously reported in two patients with metastatic Merkel cell carcinoma treated with antigen-specific CD8+T cells and immune checkpoint inhibitors. In vitro treatment of the primary tumor cells with 5-Aza may be used to restore the expression of the MHC-I haplotype lost. In addition, TGFβ may play a specific role in the downregulation of tumor MHC class I expression. TGFβ1 treatment decreased the surface expression of MHC class I of hypomethylated ovarian cancer cells, while TGFβ inhibition restored its normal expression level.
Secondly, the study identified another important role of TGFβ in mediating crosstalk with cancer stromal cells to promote T cell exclusion and immunosuppression. Using human primary fibroblasts as model systems, TGFβ treatment specifically activated fibroblasts and promoted the production of ECM, which may serve as a physical barrier hindering T cell infiltration. Furthermore, the data also suggests that TGFβ may contribute to an overall immunosuppressive tumor microenvironment in the T cell excluded tumors. TGFβ1 treatment specifically induced immune-modulatory molecules, such as IL6, IL11 and TNFAIP6 in human primary fibroblasts. Secreted in inflammatory conditions, TNFAIP6 has been reported to inhibit neutrophil migration via binding hyaluronan molecules expressed in the tumor microenvironment. Moreover, TNFAIP6 promotes the anti-inflammatory phenotype of macrophages (M2-like) thereby contributing to the immunosuppression.
Finally, TGFβ is associated with lack of response to anti-PD-L1 therapy in bladder cancer, especially within the T cell excluded tumors. To further assess the therapeutic potential of targeting TGFβ in ovarian cancer, tumor-bearing mice were treated with anti-PD-L1 and anti-TGFβ. Synergistic anti-tumor responses were confirmed in an immunocompetent mouse model of ovarian cancer. (To obtain the ovarian cancer mouse model, the BrKras (Brca1−/−; p53−/−; myc; Kras-G12D; Akt-myr) ovarian cancer cell line was obtained, and a tumor cell line was derived by one passage into FVB syngeneic immunocompetent mice. The subsequent BrKrasX1.3 cell line was subcutaneously implanted in FVB mice as an ovarian cancer immunocompetent mouse model.) Mechanistic studies also supported the hypothesis that TGFβ played an important role in promoting T cell exclusion and immune suppression. Both histological and flow cytometry analysis demonstrated a consistent trend of increased CD8+T cell presence in the mouse tumor tissues upon anti-TGFβ and anti-PD-L1 combination treatment. Blocking TGFβ signaling synergized with anti-PD-L1 and significantly remodelled the mouse tumor microenvironment to a more pro-inflammatory state, including increased M1-like and decreased M2-like macrophages, increased levels of T cell chemoattractant, CXCL9 and CXCL10, and increased density of cytotoxic T cells (GZMB+CD8+). Blocking TGFβ and PD-L1 signaling pathways triggered a strong T cell infiltration in the tumor core and enhanced tumor regressions and survival.
Disclosures herein may have important clinical significance in the field of cancer immunotherapies. Checkpoint blockades have demonstrated impressive efficacy in only subsets of patients with a pre-existing T cell immunity, with the response rate is even lower in ovarian cancer even lower. Therefore, there is a strong unmet need to further broaden and deepen the clinical efficacy of the immune checkpoint inhibitors, and TGFβ is an attractive target to overcome the immune escape mechanisms involved in the T cell excluded tumors.
In summary, the present embodiments comprise and provide the first systematic and in-depth characterization of the molecular features and mechanisms underlying the tumor-immune phenotypes in human cancer. Integrating digital pathology with machine learning and transcriptome analysis can identify mechanisms by which tumor cells and cancer-associated fibroblasts interact to shape the tumor-immune contexture in the tumor microenvironment. Further, methods for targeting the TGFβ pathway can be used as a novel therapeutic strategy to overcome T cell exclusion and immune suppression, and ultimately improve the response to cancer immunotherapy.
Three hundred seventy treatment naive patient samples with High Grade Serous Carcinoma (HGSC) were collected from the phase III ICON7 clinical trial. The tumor tissues were subjected to review by a pathologist to confirm diagnosis and tumor content. The cohort was divided into 2 sample sets for the present study: training set (n=155) and testing set (n=215). An independent validation collection (n=84 including 55 primary tumors and 29 paired metastases) was procured from Cureline, Inc (Brisbane, CA, US). All procured and clinical samples had an appropriate Institutional Review Board (IRB) approval. The ovarian cancer cell lines were obtained from the Genentech Cell Bank where they were authenticated by short tandem repeat profiling prior to banking and SNP fingerprinting after expansion. The human primary normal fibroblasts CCD-18-Co (colon, CRL-1459™; ATCC, Manassas, VA), HOF (ovary, #7336; ScienCell Research Laboratories, Carlsbad, CA) and Primary human bladder fibroblast (PHBF) (bladder, PCS-420-013™; ATCC) were procured from ATCC for in vitro TGFβ1 treatment.
Immunohistochemistry (IHC) and in situ hybridization (ISH) assays were performed on 4-μm FFPE tissue section. MHC-I IHC staining was performed as a single batch on the Ventana Discovery XT platform using the primary antibodies specific for HLA-A proteins (Abcam #ab52922, Clone EP1395Y, diluted at 0.05 μg/mL), the secondary anti-rabbit HRP antibodies and a haematoxylin counter-stain. CD8 IHC was performed at Histogenex on Ventana Benchmark using C8/clone 144B anti-CD8a monoclonal antibodies. Single-plex FAP RNAscope in situ hybridization (ISH) assay was performed. The RNAscope signal was scored on the basis of number of dots per cell as follow 0: 0 dot/cell, 1: 1-3 dots/cell, 2: 4-9 dots/cell, 3: 10-15 dots/cell, and 4:>15 dots/cell with>10% of dots in clusters. To evaluate heterogeneity in marker expression, H-score analysis was performed on FAP-ISH and MHC-I IHC. The H-score was calculated by adding up the percentage of cells in each scoring category multiplied by the corresponding score, resulting in scores are on a scale of 0-400.
The CD8-DAB IHC slides with a haematoxylin counter-stain were scanned at 20×magnification on a Panoramic 250 scanner (3DHistech) in MIRAX file format with 80% jpeg compression. Software was used to design an algorithm to distinguish cells of the tumor epithelium from those of the stroma, using cell nuclei shape and size based on the haematoxylin signal. Once the tumor cells were identified, the immediate region surrounding those cells was defined as ‘tumor compartment’ and the rest as ‘stroma compartment’. Within those areas, DAB+CD8 cells were counted, and the number of CD8+cells per region classified as ‘tumor compartment’, or ‘stromal compartment’ was reported as ‘tumor CD8 density’, or ‘stroma CD8 density’ respectively.
Macrodissection was performed on 370 formalin-fixed, paraffin-embedded (FFPE) tumor tissues from ICON7 as well as 84 FFPE tissues from Cureline, Inc. to enrich tumor percentage to greater than 70%. Total RNA was purified using High Pure FFPE RNA Micro Kit (Roche Diagnostics). RNA sequencing was performed using TruSeq RNA Access technology (Illumina®). RNA-seq reads were first aligned to ribosomal RNA sequences to remove ribosomal reads. The remaining reads were aligned to the human reference genome (NCBI Build 38) using GSNAP43,44 version 2013 Oct. 10. To quantify gene expression levels, the number of reads mapped to the exons of each RefSeq gene was calculated using the functionality provided by the R/Bioconductor package GenomicAlignments45. Raw counts were first converted to counts per million (cpm), filtered for lowly expressed genes (i.e. expressed in less than 10% of samples, and cpm<0.25), then normalized using TMM normalization in the edgeR package followed by voom transformation using the limma package. Principal component analysis (PCA) was used to assess and remove any sample outliers. These normalized log2 counts were used for downstream analysis.
Random Forest Regression. The scores for CD8+T cell density in tumor and stroma were found to strongly correlate (cor=0.74). To better capture and quantify the CD8 infiltration patterns, theses CD8 scores were converted into polar coordinates: CD8+T cell quantity=[squareroot ((CD8-tumor){circumflex over ( )}2+(CD8-stroma){circumflex over ( )}2)] and CD8+T cell spatial distribution=[atan (CD8-stroma/CD8-tumor)]. To identify the genes associated with these two metrics, a random forest regression model was built for each gene (gene˜Quantity+Distribution, randomForest package), with standard resampling of patients but no sampling of the variables (Quantity and Distribution). This revealed the specificity of these two metrics in predicting gene expression, for 16944 genes in the dataset. We did not consider the bottom 25% of genes whose expression was not associated with the variables (i.e., average MSE (mean squared error) below 1st quartile). Genes with expression was selected based on the quantity metric (i.e. percent increase in MSE for>3rd quartile, referred to genes associated with CD8 quantity) and/or by CD8 spatial distribution (i.e., percent increase in MSE for spatial distribution>3rd quartile). This resulted in 103 genes associated with CD8 quantity, 56 associated with CD8 spatial distribution and 193 genes common for these two metrics. Correlation analysis of these genes highlighted very similar transcriptional profiles for the 103+193 genes associated with CD8 quantity. Subsequent analyses, were focused to the genes specific for these two metrics: 56+103=159 CD8-associated genes.
Consensus clustering. Based on the 157 CD8-associated genes (excluding two genes without gene symbol), a consensus clustering was performed on the ICON7 training set (n=155) using the ConsensusClusterPlus R package with pearson distance metric and k-means clustering with 80% patient selection and 100% feature selection. Transcriptional heterogeneity was captured well with 4 clusters, yet those clusters were mostly differentiated by CD8 quantity. To additionally capture CD8 distribution, we set the optimal number of clusters to 6, which differentiated tumors by both CD8 quantity and distribution. The expression profile of the 6 clusters revealed that some clusters only differed in their cytotoxic activity, i.e., level of CD8 quantity (
PAM classification. The PAMR package in R was used to derive a classifier for the prediction of the three immune phenotypes. This classifier was built on the 157 CD8-associated genes, the number of necessary classifier genes ranging from 157 to 1 was evaluated, and the optimal number of genes i.e. 157 was selected corresponding to a minimal cross-validation error rate at a threshold value of 0.23. A tumor was assigned to an immune phenotype when the probability for that phenotype exceeded 0.7 and was below 0.5 for the other two immune phenotypes. A tumor was otherwise considered unclassifiable.
The multiGSEA function with the Camera enrichment method in the multiGSEA R package was used for gene set enrichment analysis comparing different immune phenotypes in the full ICON7 collection (n=370), with use of the Hallmark and KEGG gene set collections from the Molecular Signature Database. Immune subset and stromal fraction enrichment analysis for ICON7 samples were done using the online xCell cell types enrichment score tool (http://xcell.ucsfedu/).
Enrichment of deleterious mutations in 15 homologous recombinant deficiency (HRD) related genes and 4 dMMR genes were evaluated in TCGA-OV samples in different tumor-immune phenotypes. In addition, tumor mutation burden (TMB) and neoantigen loads were estimated in TCGA-OV samples. Enrichment analysis in each tumor-immune phenotype for above-mentioned genetic features in TCGA-OV was performed using Fisher's exact test corrected for multiplicity via Benjamini-Hochberg method in R.
The 100 genes that were reported in the CLOVAR signature were extracted to examine the molecular subtype of a tumor. Four major clusters were identified in the ICON7 cohort based on hierarchical clustering with Euclidean distance and Ward's linkage method. By checking the testing results and up/down pattern in the original report for each gene, the identified clusters were assigned to various molecular subtypes (e.g., Immunoreactive, Mesenchymal, Proliferative and Differentiated).
250 ng of genomic DNA from 48 ovarian cancer cell lines were assayed using the Illumina Human Methylation 450 BeadChip platform. The raw methylation data (.idat files) were read into the R software using illuminaio. Quality control was performed using the methylation R package minfi; all samples passed quality control. The methylation levels were normalized using the “noob” background correction and dye bias equalization methods as implemented in minfi. Both procedures have been shown to perform well and to be appropriate for cancer samples. Beta values, defined as ratios of the methylated allele intensity over the total intensity, were calculated for probes targeting CpG sites located between −1000 bp and +1000 bp from the transcription start site of the HLA-A gene.
SK-OV-3 and OVCA-420 (MHC-Ihigh), and OAW42 and PA-1 (MHC-Ilow) ovarian cancer lines were cultured in complete culture media (RPMI-1640+10% FBS). The cells were plated at 12,500-100,000 cells/well in 6-well tissue culture plate and complete culture media. After 24 hours, the cells were starved overnight in DMEM high glucose medium without FBS. Next, the starving media was replaced with culture media only (DMEM+2% FBS), 10 ng/mL rhTGFβ1 (Cat #PHG9204, Thermo Fisher, CA), 10 ng/mL rhTGFβ1+10 μM Galunisertib (Cat #S2230, SelleckChem, TX) or 5 ng/mL recombinant IFNγ (Cat #554617, BD Biosciences, CA) for 96 h at 37° C. Cells were then stained and analysed by flow cytometry. The “percentage of untreated” was calculated using this formula: [Geo Mean Fluorescence Intensity (IFNγ-treated cells)/Geo Mean Fluorescence Intensity (untreated cells)]×100. In order to see if MHC-I expression can be regulated by methylation, two MHC-Ilow lines OAW42 and PA-1 were plated at 250,000-500,000 cells/dish in 10-cm dish and serum starved as described above for TGFβ1 treatment. 10 μM and 1 μM 5-Aza-2′-deoxycytidine (5-Aza, Cat# A2385, Sigma-Aldrich) demethylating agent in culture media was used to treat OAW42 and PA-1, respectively, for 96 h prior to FACS analysis. Media was half-replenished with fresh 5-Aza 48 hours after treatment to keep concentration consistent.
The primary normal fibroblast PHBF (Bladder), CCD-18Co (Colon) and HOF (Ovary) were serum-starved overnight before treatment with media only (untreated), 10 ng/mL rhTGFβ1 or 10 ng/mL rhTGFβ1+10 μM Galunisertib for 24 hours and total RNA was extracted for RNA-seq analysis. To detect IL-6 protein in the supernatant, cells were treated for 48 hours with rhTGFβ1. After the 48 h, the supernatant was collected and analysed by Luminex using the Millipore kit. For the proliferation assay, PHBF, CCD-18Co, HOF were plated at 3,000 cells/well in a 96-well culture flat bottom plate for immunofluorescent assays (Corning, #3917) overnight. Cells were then cultured for 72 hours in DMEM high glucose+1% FBS with indicated concentration of TGFβ1 with or without Galunisertib. Next, CellTiter-Glo® reagents (Promega, G7570) were added to each well and luminescence signal was read with a microplate reader.
IV.I.10. p-SMAD2/3 Western Blot Assay
PHBF cells were plated at 100,000 cells/well in a 24-well cell culture plate overnight, serum starved for 24 h and then cultured in serum-free DMEM with indicated concentration of TGFβ1 with or without Galunisertib for 30 min. Cells were lysed in protein lysis buffer containing T-PER tissue protein extraction reagent (ThermoFisher, #78510), cOmplete™ Protease Inhibitor Cocktail (Sigma-Aldrich, #11697498001) and PhosSTOP™ phosphatase inhibitor cocktails (Sigma-Aldrich, #4906845001). Total protein was diluted and normalized to 0.5 μg/μL with 4×LDS Sample Buffer (ThermoFisher, #84788). 10 ug of total protein was loaded into each well of a NuPAGE 4-12% Bis-Tris Midi Gel (Invitrogen), followed by protein transfer from gel to the membrane using Trans-Blot Turbo (Bio-Rad). The Phospho-Smad2 was first revealed following the general protocol western blot from Bio-Rad. Briefly, the membrane was blocked for lh, incubated with Phospho-Smad2 antibodies overnight at 4° C. (Ser456/467, 1:200, Cell Signaling #3108, clone138D4), washed and incubated with secondary antibodies goat anti-rabbit. To analyse the total Smad2/3, the membrane was stripped and then incubated with Smad2/3 antibodies (1:1000, Cell Signaling # 8685).
Before staining, Fc receptors were blocked for 10 min at room temperature using FcR blocking reagent human (Cat # 130-059-901, Miltenyi Biotec, CA). Cells were stained during the blocking step with the LIVE/DEAD™ Fixable Near-IR Dead Cell (Cat #L10119, Invitrogen, CA). Then, cells were incubated at room temperature for 15 min with anti-human HLA-ABC-PE (Cat#560168, BD Biosciences, CA) or isotype control mouse IgG1κ-PE (Cat #556650, BD Biosciences) antibodies, washed and samples were acquired on BD LSRFortessa™ flow cytometer.
IV.I.12.a. In Vivo Mouse Tumor Experiments
The Genentech Institutional Animal Care and Use Committee (IACUC) approved all animal studies and experiments were conducted according to National Institutes of Health (NIH) guidelines, the Animal Welfare Act, and U.S. Federal law. Female FVB mice were obtained from Jackson Laboratories (stock 001800). All mice were housed at Genentech under specific pathogen-free (SPF) conditions and used at 8-12 weeks of age. Investigators performing mouse experiments were not blinded. The BrKras (Brca1−/−; p53−/−; myc; Kras-G12D; Akt-myr) ovarian cancer cell line was obtained from Sandra Orsulic's lab. The tumor cell line was derived by one passage into FVB syngeneic immunocompetent mice. The subsequent BrKrasX1.3 cell line was selected for this study. Two million of BrKrasX1.3 ovarian cancer cells in 100 μL sterile HBSS were subcutaneously injected in the right flank of FVB mice. When tumors reached a volume of ˜50-180 mm3 (about 12 days after inoculation), animals were distributed into treatment groups based on tumor volume to form homogeneous groups at baseline and treated the next day with anti-GP120 isotype control antibodies (mouse IgG1 clone 10E7, 20 mg/kg first dose followed by 15 mg/kg), anti-PD-L1 (mouse IgG1 clone 6E11, 10 mg/kg first dose followed by 5 mg/kg thereafter)+anti-GP120 (10 mg/kg), anti-TGFβ (mouse IgG1 clone 1D11, 10 mg/kg)+anti-GP120 (10 mg/kg first dose followed by 5 mg/kg thereafter), or a combination of anti-PD-L1 (10 mg/kg first dose followed by 5 mg/kg thereafter) with anti-TGFβ (10 mg/kg), 3 times a week for 3 weeks (intravenously for the first dose and intraperitoneally thereafter). Tumors were measured 2-3 times per week by calliper, and tumor volumes were calculated using the modified ellipsoid formula, ½×(length×width2). Complete response (CR) was defined as a complete regression (undetectable) of the tumor without any recurrence. Partial regression (PR) was defined as tumor regression after the last dose for at least two time points followed by uncontrolled tumor growth and stable disease (SD) was defined as at least two time points with stable tumor volumes after the last dose followed by uncontrolled tumor growth. Animals were euthanized immediately if tumor volume exceeded 2000 mm3, or if tumors or body condition ever fell outside the IACUC Guidelines for Tumors in Rodents.
IV.I.12.b. Ex Vivo Analysis on Mouse Tumor Samples
Tumors were collected 7 days after treatment initiation (Day 8). Tumors were weighed, minced in small pieces with a razor blade and transferred into GentleMACS C tube (Miltenyi Biotec) containing 5 mL of digestion media (cocktail of dispase, collagenase P and DNAse I in RPMI+2% FBS). Tumors were first mechanically dissociated by running the program m_imp_tumor02 on the GentleMACS followed by 20 min of incubation at 37° C. on a rotator. Then, the cell suspension is filtered with a 70 μm mesh on a 50 mL falcon containing MACS buffer+2% FBS. Fresh digestion media is added to the undissociated tissue and samples were incubated for another 20 min at 37° C. Next, tissues were mechanically dissociated by running the program m_imp_tumor03 two times. The cell suspension is filtered on the 70 μm mesh. Red blood cells were lysed with ACK buffer. Washed cell suspension were then counted using a Vi-CELL XR (Beckman Coulter, Brea, CA).
For the staining, 4 million of live cells were transferred into FACS tube and washed with FACS stain buffer (1X PBS pH 7.4, 0.2% BSA, 0.09% NaAzide). Cells were then incubated for 10 min at room temperature with FcR blocking reagent mouse (2 μL/tube, Miltenyi Biotec, #130-092-575) and Zombie UV (1 μL/tube, BioLegend, #423108). The cells were then stained with the following antibodies: CD3-APC-Cy7 (2 μg/mL, BD Biosciences, clone 145-2C11, #557596), CD4-Alexa Fluor700 (0.5 μg/mL, BD Biosciences, clone RM4-5, #557956), CD25-PE (1 μg/mL, BD Biosciences, clone PC61, #553866), CD45-BV510 (0.5 μg/mL, BD Biosciences, clone 30F11, #563891), CD8-BV421 (1 μg/mL, BioLegend, clone 53-6.7, #100738), Ly6G-PercP-Cy5,5 (1 μg/mL, BD Biosciences, clone 1A8, #560602), SiglecF-BB515 (1 μg/mL, BD Biosciences, clone E50-2440, #564514), CD11b-BV421 (0.5 μg/mL, BioLegend, clone M1/70, #101236) for 30 min at 4° C. Cells were fixed and permeabilized with BD Cytofix/Cytoperm™ (BD Biosciences, #554714) for 20 min at 4° C. to stain CD206-AlexaFluor647 (2.5 μg/mL, BioLegend, clone C068C2, #141712), iNOS-PE (0.3 μg/mL, Thermo Fisher Scientific, clone CXNFT, #12-5920-82) and GranzymeB-AlexaFluor647 (1 μg/mL, BD Biosciences, clone GB11, #560212). To stain Ki67-FITC (10 μL/test, BD Biosciences, clone B56, #556026) and FOXP3-APC (2 μg/mL, Thermo Fisher Scientific, clone FJK-16s, #17-5773-82), cells were fixed and permeabilized with eBioscience™ Foxp3/Transcription (Thermo Fisher Scientific, #00-5523-00) for 45 min at 4° C.
Flow Cytometry data were collected with a BD LSRFortessa X-20 cell analyser and analysed using FlowJo Software (Version 10.4.2, FlowJo, LLC, Ashland, OR).
Blood was harvested by terminal heart bleed 7 days after treatment initiation and collected on BD microtainer tubes with serum separator additive (BD biosciences). Tubes were centrifuged for 10 min at 1,000 g at 4° C. and the serum collected and stored at −80° C. until analysis. To profile the cytokines/chemokines present in the serum, the samples were diluted 1:2 in assay diluent (Millipore) and the Mouse Cytokine/Chemokine Immunology Multiplex Assay 32-plex (Millipore) was performed.
Immunohistochemistry (IHC) was performed on 4 μm thick formalin-fixed, paraffin-embedded tissue sections mounted on glass slides. Staining was performed on the Lab Vision Autostainer (ThermoFisher Scientific, Kalamazoo, Michigan). Sections were de-paraffinized and rehydrated to deionized water. Antigen Retrieval was performed with 1×DAKO Target Retrieval Solution (Agilent Technologies, Carpinteria, CA) for 20 min at 99° C. and cooled to 74° C. Subsequently, endogenous peroxidase was quenched by incubating in sections in 3% H2O2 for 4 minutes at room temperature. Phospho-SMAD2 was detected using a rabbit monoclonal anti-pSMAD2 (clone 138D4, Cell Signal Technologies, Danvers, MA), and a rabbit monoclonal anti-CD8a (clone 1.21E3.1.3, Genentech, Inc, South San Francisco, CA) incubated for 60 min at RT. The primary antibody was detected with PowerVision Poly-HRP anti-Rabbit (LeicaBioSystems, Buffalo Grove, IL) and visualized with a Metal Enhanced DAB chromogen (Thermo Scientific, Kalamazoo, Michigan). Sections were counterstained with Mayer's haematoxylin, dehydrated, mounted with permanent mounting medium, and cover slipped.
CD8 digital pathology analysis: Brightfield CD8-IHC slides were scanned at 20×magnification using the Nanozoomer slide scanner (Hamamatsu). Image analysis was performed on native .ndpi files using custom algorithms developed in Definiens Developer XD software (Munich, Germany). DAB (CD8) and Haematoxylin (nuclear counterstain) were isolated by HSD colour transformation (van Der Laak et al, 2000). Cells were segmented by thresholding on isolated haematoxylin stain then split using a watershed segmentation algorithm. DAB positivity was evaluated within individual cell boundaries to classify CD8+cells. An automated region classification algorithm was applied within pathologist-annotated tumor borders to classify viable, necrotic, and stromal regions. Very small, punctate nuclei with dark haematoxylin counter stain were defined as necrotic. Sparse regions with small or elongated nuclei were classified as stroma or surrounding tissue (
Whole slide digital images of each immunolabeled tissue section were obtained using a Nanozoomer digital slide scanner (Hamamatsu). Tumor areas were manually annotated by a pathologist to include tumor using the MATLAB (MathWorks) software package. MATLAB was subsequently used to identify all viable cell nuclei based on size, shape, and labelling characteristics and to calculate mean DAB intensity for each nucleus. Four immunoreactivity levels (negative, weak, moderate, and strong) in a training set of the control and tumor tissue images. Nuclei were binned as weak positive, moderate positive, or strong positive and images were reviewed for algorithm accuracy. Final quantification results were reported as the digital histoscore (1*percent of weak nuclei+2*percent of moderate nuclei+3*percent of strong nuclei, range 0-300).
Some embodiments of the present disclosure include a system including one or more data processors. In some embodiments, the system includes a non-transitory computer readable storage medium containing instructions which, when executed on the one or more data processors, cause the one or more data processors to perform part or all of one or more methods and/or part or all of one or more processes disclosed herein. Some embodiments of the present disclosure include a computer-program product tangibly embodied in a non-transitory machine-readable storage medium, including instructions configured to cause one or more data processors to perform part or all of one or more methods and/or part or all of one or more processes disclosed herein.
The terms and expressions which have been employed are used as terms of description and not of limitation, and there is no intention in the use of such terms and expressions of excluding any equivalents of the features shown and described or portions thereof, but it is recognized that various modifications are possible within the scope of the invention claimed. Thus, it should be understood that although the present invention as claimed has been specifically disclosed by embodiments and optional features, modification and variation of the concepts herein disclosed may be resorted to by those skilled in the art, and that such modifications and variations are considered to be within the scope of this invention as defined by the appended claims.
The present description provides preferred exemplary embodiments only, and is not intended to limit the scope, applicability or configuration of the disclosure. Rather, the present description of the preferred exemplary embodiments will provide those skilled in the art with an enabling description for implementing various embodiments. It is understood that various changes may be made in the function and arrangement of elements without departing from the spirit and scope as set forth in the appended claims.
Specific details are given in the present description to provide a thorough understanding of the embodiments. However, it will be understood that the embodiments may be practiced without these specific details. For example, circuits, systems, networks, processes, and other components may be shown as components in block diagram form in order not to obscure the embodiments in unnecessary detail. In other instances, well-known circuits, processes, algorithms, structures, and techniques may be shown without unnecessary detail in order to avoid obscuring the embodiments.
This application is a divisional application of U.S. application Ser. No. 17/033,161, filed on Sep. 25, 2020, which claims priority to and the benefit of U.S. Provisional Application No. 62/907,062, filed on Sep. 27, 2019, each of which is hereby incorporated by reference in their entireties for all purposes.
| Number | Date | Country | |
|---|---|---|---|
| 62907062 | Sep 2019 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 17033161 | Sep 2020 | US |
| Child | 18534549 | US |
