T cell specific dispersion cancer biomarker

Abstract

The invention relates to a method for determining the prognosis of a patient diagnosed with cancer, the method comprising: - obtaining from the patient a histology section containing a tumour area; - in a measurement step, contacting the histology section with a labelled probe capable of specifically labelling T cells, and obtaining an image of the tumour area, - in an image analysis step, quantifying the density of T cells present per mm2 in at least 2 separate, non-overlapping regions of the image of the tumour area; - calculating a dispersion value (alpha); and - assigning the patient a high probability of survival if the dispersion value is below a threshold value; or a low probability of survival if the dispersion value is above the threshold value.

Description

[0001] T Cell Specific Dispersion Cancer Biomarker

[0002] Field

[0003] The present invention relates to a method to predict the survival outcome of a patient diagnosed with a solid cancer, based on calculation of a dispersion value reflecting both the density, and variability of dispersion of T cells in a tumour histology section.

[0004] Background

[0005] Various studies have shown the importance of CD8 T cells for cancer patient prognosis, with the general trend that higher abundances of CD8 T cells within tumour tissues are favourable for the patient (Bruni et al 2020, Donnem et al 2016). Similarly, studies have shown that certain RNA expression profiles, associated with increased tumour lymphocyte infiltration are also associated with better patient prognosis (Thorsson et al 2018).

[0006] There are numerous approaches to profile the densities of CD8 T cells which use different measures as a prognostic and predictive score. To date, clinical biomarkers using imaging data (e.g. IHC, immunofluorescence) do take spatial information into account but with large limitations. The Immunoscore for T cells uses the information of tumour area and tumour border to stratify the spatial location of T cells (Galon et al 2020; Bruni et al 2020; Pages et al 2009). Other studies have used tumour area based CD8 T cell quantifications, stromal quantifications or intratumoural quantifications or a mix of all (Donnem et al 2016). The “Immunoscore” summarizes CD8 T cell densities and was originally introduced in colorectal cancer. This score was included in the 2020 European Society of Medical Oncology (ESMO) Clinical Practice Guide- lines for Diagnosis, Treatment and Follow-up for Localized Colon Cancer. An IHC measurement of T cells is performed, and the corresponding image data is stratified into tumour and tumour margin regions. The estimated density of the measured T cells in the two regions is combined into a biomarker value, and is then compared to a pre-defined threshold which classifies a patient as high, or low risk.

[0007] Based on the above-mentioned state of the art, the objective of the present invention is to provide improved means and methods to predict cancer outcomes. This objective is attained by the subjectmatter of the independent claims of the present specification, with further advantageous embodiments described in the dependent claims, examples, figures and general description of this specification.

[0008] Summary of the Invention

[0009] The dispersion parameter developed by the inventors is a mixture of the variability and the average cell count of T cells, per mm2. It captures the observed variability across tumours and the density of cytotoxic T cells which seem to be highly relevant to clinical outcomes. The established Immunoscore method was shown to be relevant in colorectal cancer and other cancers, but does not have prognostic power in the NSCLC cohort investigated in this study.

[0010] The invention described herein relates to a method of predicting the prognosis of a cancer patient by means of calculating a T cell count dispersion score. This is an interpretable score based on count data of T cells across individual sections, or ‘tiles’ of an image taken of a patient tissue section, fitted with a negative binomial model to derive the expected cell count of T cells and the dispersion value per patient from the fitted model. The inventors show that in a cohort of stage l-lll NSCLC patients, stratification based on the T cell dispersion achieves striking differentiation in overall survival (OS). Patients with high dispersion are characterized by more variation in T cell densities across tiles, given the observed average density of T cells in a representative cohort of individuals. The score provides powerful predictive value on sections of resected tumour tissues, but can also be applied to biopsy samples, opening an avenue for simple testing in the clinics at low cost, using existing CD8 immunohistochemistry (IHC) assays.

[0011] A first aspect of the invention provides a method for determining the prognosis of a patient diagnosed with a solid cancer. The method is performed on a patient’s histology section containing a piece of tumour area, such as a histology slide of a of resected tumour tissue obtained during an operation, or an exploratory tumour biopsy. The histology section will have been obtained from the patient prior to the measurement step set forth below.

[0012] The method comprises the following steps. in a measurement step, contacting the histology section with a labelled probe capable of specifically labelling T cells, and obtaining an image of the tumour area, in an image analysis step, quantifying the number, or density of T cells present in at least (>) 2 separate, non-overlapping regions of the image of the tumour area to provide a plurality of tile values; in a calculation step, calculating a T cell count dispersion value (alpha, a) from the plurality of tile values; and assigning the patient:

[0013] • a good prognosis if the a value is equal to or below a threshold, or

[0014] • a poor prognosis of the cancer patient if the a value is above a threshold.

[0015] In particular embodiments, alpha is calculated by fitting a negative binomial distribution (NBD) to the tile value data and extracting a first expected T cell count from the plurality of tile values in the data set (p); and also a T cell count dispersion value (a), reflecting the observed variability of T cell counts across all the plurality of tile values in the data set. The invention further relates to a computer program configured to carry out the image analysis step and the calculation and assignment steps above, and also a system of classifying patients by analysis of a tumour histology sample as described above.

[0016] Terms and definitions

[0017] For purposes of interpreting this specification, the following definitions will apply and whenever appropriate, terms used in the singular will also include the plural and vice versa. In the event that any definition set forth below conflicts with any document incorporated herein by reference, the definition set forth shall control.

[0018] The terms “comprising”, “having”, “containing”, and “including”, and other similar forms, and grammatical equivalents thereof, as used herein, are intended to be equivalent in meaning and to be open-ended in that an item or items following any one of these words is not meant to be an exhaustive listing of such item or items, or meant to be limited to only the listed item or items. For example, an article “comprising” components A, B, and C can consist of (i.e., contain only) components A, B, and C, or can contain not only components A, B, and C but also one or more other components. As such, it is intended and understood that “comprises” and similar forms thereof, and grammatical equivalents thereof, include disclosure of embodiments of “consisting essentially of or “consisting of.”

[0019] Where a range of values is provided, it is understood that each intervening value, to the tenth of the unit of the lower limit, unless the context clearly dictates otherwise, between the upper and lower limit of that range and any other stated or intervening value in that stated range, is encompassed within the disclosure, subject to any specifically excluded limit in the stated range. Where the stated range includes one or both of the limits, ranges excluding either or both of those included limits are also included in the disclosure.

[0020] Reference to “about” a value or parameter herein includes (and describes) variations that are directed to that value or parameter per se. For example, description referring to “about X” includes description of “X.”

[0021] As used herein, including in the appended claims, the singular forms “a”, “or” and “the” include plural referents unless the context clearly dictates otherwise.

[0022] "And / or" where used herein is to be taken as specific recitation of each of the two specified features or components with or without the other. Thus, the term "and / or" as used in a phrase such as "A and / or B" herein is intended to include "A and B," "A or B," "A" (alone), and "B" (alone). Likewise, the term "and / or" as used in a phrase such as "A, B, and / or C" is intended to encompass each of the following aspects: A, B, and C; A, B, or C; A or C; A or B; B or C; A and C; A and B; B and C; A (alone); B (alone); and C (alone).

[0023] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art (e.g., in cell culture, molecular genetics, nucleic acid chemistry, hybridization techniques and biochemistry, organic synthesis). Standard techniques are used for molecular, genetic, and biochemical methods (see generally, Sambrook et al., Molecular Cloning: A Laboratory Manual, 4th ed. (2012) Cold Spring Harbor Laboratory Press, Cold Spring Harbor, N.Y. and Ausubel et al., Short Protocols in Molecular Biology (2002) 5th Ed, John Wiley & Sons, Inc.) and chemical methods.

[0024] Dispersion value

[0025] The term T cell count dispersion value, or dispersion value in the context of the present specification relates to a statistical value that helps in understanding the variability and average cell count per mm2in an image of the tumour environment obtained from a histological sample. It captures the observed variability across different regions of the tumour area and the number, or density of T cells. In particular embodiments, the T cell count dispersion value is calculated using the formula (Var- mu) I mu2, wherein (mu) represents the mean density of T cells present per mm2of the at least two regions of the tumour area; and (Var) represents a measure of the variability in the number of T cells present at least two regions of the tumour. The dispersion value is not equal equivalent to the variance or standard deviation. The variance in normal distribution is simply the spread of the data around the mean independent of the value of the mean (Var = StdDev2). However, the dispersion differ in that this measure is inherently coupled to the mean (Var = p + p2*D).

[0026] Cell Biology, diagnostic method inventions: Markers, ligands

[0027] The expression of a T cell marker such as CD3, CD4, or CD8 may be assayed via techniques suited to labelling a histology slide such as fluorescence microscopy, light microscopy, or multiplex analyses of using a set of molecule probes followed by imaging mass cytometry.

[0028] The term molecular probe or labelled probe in the context of the present specification relates to a specific ligand, particularly an antibody, antibody fragment, an antibody-like molecule or aptamer, more particularly an antibody or antibody fragment, that can bind to a target T cell surface antigen molecule, with a dissociation constant of <10'7mol / l, particularly <10'9mol / l. The molecular probe comprises a detectable marker such as a particle, bead, dye, or an enzyme which catalyses a chromogenic substrate such as 3,3',5,5'-tetramethylbenzidine (TMB))

[0029] The term set of molecular probes relates to a panel of molecular probes for positive and / or negative selection of marker expression.

[0030] The term fluorescent dye in the context of the present specification relates to a small molecule capable of fluorescence in the visible or near infrared spectrum. Examples for fluorescent labels or labels presenting a visible colour include, without being restricted to, fluorescein isothiocyanate (FITC), rhodamine, allophycocyanin (APC), peridinin chlorophyll (PerCP), phycoerithrin (PE), alexa Fluors (Life Technologies, Carlsbad, CA, USA), dylight fluors (Thermo Fisher Scientific, Waltham, MA, USA) ATTO Dyes (ATTO-TEC GmbH, Siegen, Germany), BODIPY Dyes (4,4-difluoro-4-bora- 3a,4a-diaza-s-indacene based dyes) and the like. In the context of the present specification, the term labelled antibody is used for an antibody being covalently bound to a detectable label. Such detectable labels include for example, without limitation, octadecyl rhodamine B, 7-nitro-2-1 ,3-benzoxadiazol-4-yl, 4-acetamido-4'- isothiocyanatostilbene-2,2' disulfonic acid, acridine and derivatives, 5-(2'-aminoethyl)amino- naphthalene-1 -sulfonic acid (EDANS), 4-amino-N-(3-[vinylsulfonyl]phenyl)naphthalimide-3,6- disulfonate dilithium salt, N-(4-anilino-1-naphthyl)maleimide, anthranilamide, BODIPY, Brilliant Yellow, coumarin and derivatives, cyanine dyes, cyanosine, 4',6-diaminidino-2-phenylindole (DAPI), bromopyrogallol red, 7-diethylamino-3-(4'-isothiocyanatophenyl)-4-methylcoumarin, diethylenetriamine pentaacetate, 4,4'-diisothiocyanatodihydro-stilbene-2,2'-disulfonic acid, 4,4 - diisothiocyanatostilbene-2,2'-disulfonic acid, dansylchloride, 4-dimethylaminophenylazophenyl-4'- isoth iocyanate (DABITC), eosin and derivatives, erythrosin and derivatives, ethidium, fluorescein,

[0031] 5-carboxyfluorescein (FAM), 5-(4,6-dichlorotriazin-2-yl)aminofluorescein (DTAF), 2',7'-dimethoxy-

[0032] 4'5'-dichloro-6-carboxyfluorescein, fluorescein isothiocyanate, X-rhodamine-5-(and 6)- isoth iocyanate (QFITC or XRITC), fluorescamine, IR-144 (2-[2-[3-[[1 ,3-dihydro-1 ,1-dimethyl-3-(3- sulfopropyl)-2H-benz[e]indol2-ylidene]ethylidene]-2-[4-(ethoxycarbonyl)- 1 -piperazinyl]-1 - cyclopenten-1-yl]ethenyl]-1 ,1-dimethyl-3-(3-sulforpropyl)-1 H-benz[e]indolium hydroxide, inner salt, compound with n,n-diethylethanamine(1 :1 ), CAS No.: 54849-69-3), 5-chloro-2-[2-[3-[(5-chloro-3- ethyl-2(3H)-benzothiazol-ylidene)ethylidene]-2-(diphenylamino)-1-cyclopenten-1-yl]ethenyl]-3- ethyl benzothiazolium perchlorate (IR O), malachite green isothiocyanate, 4-methylumbelliferone, ortho cresolphthalein, nitrotyrosine, pararosaniline, phenol red, B-phycoerythrin, o- phthaldialdehyde, pyrene, pyrene butyrate, succinimidyl 1 -pyrene, butyrate quantum dots, Reactive Red 4 (Cibacron Brilliant Red 3B-A), rhodamine and derivatives, 6-carboxy-X-rhodamine (ROX),

[0033] 6-carboxyrhodamine (R6G), lissamine rhodamine B sulfonyl chloride rhodamine (Rhod), rhodamine B, rhodamine 123, rhodamine X isothiocyanate, sulforhodamine B, sulforhodamine 101 , sulfonyl chloride derivative of sulforhodamine 101 (Texas Red), N,N,N',N'tetramethyl-6- carboxyrhodamine (TAMRA) tetramethyl rhodamine, tetramethyl rhodamine isothiocyanate (TRITC), riboflavin, rosolic acid, terbium chelate derivatives, Cyanine-3 (Cy3), Cyanine-5 (Cy5), Cyanine-5.5 (Cy5.5), Cyanine-7 (Cy7), IRD 700, IRD 800, Alexa 647, La Jolta Blue, phthalo cyanine, and naphthalo cyanine.

[0034] Detailed Description of the Invention

[0035] A first aspect of the invention relates to a method for determining the prognosis of a patient diagnosed with a type of solid tissue cancer.

[0036] The method is performed on a patient histology section containing a tumour area. Such samples can be obtained from resections of tumour, or a smaller biopsy sample, and processed to provide histology sections according to standard pathology protocols prior to application of the method according to the invention. The method comprises the following steps. In a measurement step, the histology section is contacted with a labelled probe capable of specifically labelling T cells, and an image of the labelled tumour area is obtained. The means for this measurement step are not particularly limited and may, for example, be achieved by performing T cell immunohistological staining using a probe for a T cell, conjugated to a detectable label such as an enzyme, or a florescent dye, followed by acquisition of microscopy image of the labelled tumour area. Alternatively, an imaging mass cytometry image of the tumour area may be obtained where T cell markers are included in a panel of many multiplexed markers. CD8 biomarker measurements as a basis for the following steps of the method are routinely done with certified tests (e.g. Ventana CD8 IHC) known to the skilled person.

[0037] In an image analysis step, the number, or density of T cells present in at least (>) 2 separate, nonoverlapping regions (or tiles) of the image of the tumour area is quantified to provide a plurality of tile values as a basis for sample assessment. The shape of these tiles is not particularly limited, but their non-overlapping nature serves to reflect the T cell count in various areas of the tumour sample.

[0038] The T cell counts obtained from the image analysis are then used in a next a calculation step, where a T cell count dispersion value (a) of the plurality of tile values is calculated. The inventors show that a dispersion value captures the observed variability across tumours and the number, or density of cytotoxic T cells, and has a strong correlation with clinical parameters. The dispersion parameter ‘alpha’ is sometimes referred to as a dispersion parameter, or a dispersion coefficient, and can be obtained, for example by extracting the parameters from a negative binomial distribution (NBD) fitted to the tile data obtained in the image analysis step.

[0039] Finally, the method comprises assigning the patient either a good prognosis if the a value is equal to or below a threshold, or a poor prognosis of the cancer patient if the a value is above a threshold.

[0040] The inventors developed an algorithm that tiles the data into smaller rectangles and calculates cellular densities of CD8 T cells for each tile. Depending on the heterogeneity of the CD8 infiltration in the tumour, the values for those CD8 densities across the tiles vary strongly. The inventors were able to show that such a process requires a negative binomial model. They fit the negative binomial model to the data from all tiles and extracted model parameters (expected value: mean of CD8 densities across all tiles; dispersion: extra variation observed in the data) for a cohort of 192 NSCLC patients. Splitting the patient groups based on the average or the median value of the dispersion results in a strong prognostic signature for overall survival. Samples with high dispersion (i.e. the observed counts across the tiles are more dispersed than the average across patients, meaning T cells are more non-homogenously spread across various regions) according to the assessment were correlated with poor outcomes.

[0041] In some embodiments of the method according to the invention, the calculation step further comprises submitting a data set comprising the plurality of tile values to a computer-implemented classification method to provide this T cell count dispersion value (a). The computer-implemented classification method comprises the steps:

[0042] Firstly, determining a negative binomial distribution (NBD) to fit the data set.

[0043] Then, determining a first NBD parameter from the NBD corresponding to the expected T cell number or density across all the plurality of tile values in the data set (p).

[0044] Next, the method determines a second NBD parameter from the NBD corresponding to the T cell count dispersion value (a), wherein a characterizes the observed variability of T cell counts across all the plurality of tile values in the data set.

[0045] Finally, the computer-implemented classification method generates an output indicative of at least two classes: a first class indicative of good prognosis of the cancer patient if the alpha value is equal to or below an a threshold, or a second class indicative of poor prognosis of the cancer patient if the alpha value is above the a threshold.

[0046] In certain embodiments of the method according to the invention, the first NBD parameter and the second NBD parameter are determined by an optimisation algorithm selected from the list consisting of maximum likelihood estimation, Bayesian modelling, and expectation maximization. In particular embodiments, a maximum likelihood estimation approach is used.

[0047] In particular embodiments of the method according to predict the prognosis of a cancer patient, the NBD parameters p and a are determined by the probability mass function (1 ).

[0048] / r(fc + |)

[0049] (1) f (k; p, a) = Pr(X = k) = I

[0050] \

[0051] The inventors observed that the probability of observing the tile values given specific p and alpha values is maximized by use of this objective function.

[0052] In certain embodiments, the labelled probe capable of specifically labelling T cells is a ligand capable of specifically binding a T cell antigen selected from the group consisting of CD3, CD4, and CD8. It thus is capable of labelling and making visible T cells positive for at least one of the classical T cell markers which can be used to extra a value for alpha corresponding closely to clinical parameters (Fig. 3).

[0053] In certain embodiments, the ligand capable of specifically binding a T cell antigen is an antibody. The antibody is labelled to allow its detection by light, or as part of a mass spectrometric histology detection system. Antibody fragments capable of specifically binding their antigen, or antibody-like ligands might be used instead a full antibody molecule.

[0054] In certain embodiments, the labelled probe capable of specifically labelling T cells is an antibody specific for CD8. In particular embodiments, the labelled probe capable of specifically labelling T cells is an antibody specific for CD8.

[0055] In certain embodiments, the labelled probe is labelled by a fluorescent dye. Examples of fluorescent dyes that may be employed to practice the invention are given above. In particular embodiments, the probe capable of specifically labelling is an antibody conjugated to a fluorescent dye.

[0056] The inventors show that analysis of even small biopsy sample tumour area can provide a reasonable association with clinical parameters. In certain embodiments, in the image analysis step the number of T cells present least 5, 10, 20, 30, 40, or at least 50 separate, non-overlapping regions of the tumour area of the tumour area are quantified and subsequently used to calculate the T cell count dispersion value calculation.

[0057] In certain embodiments, the area of each of the at least two tiles is equivalent to the area of a 300 x 300 pm2square. In certain embodiments, the area of each of the at least two tiles is equivalent to a 700 x 700 pm2square. In certain embodiments, the area of each of the at least two tiles is equivalent to the area of a maximally 1200 x 1200 pm2. In certain embodiments, the dispersion value alpha is calculated is based on tile values of at least 7 tiles, wherein each tile has an area equivalent to the area of a 300 x 300 pm2square.

[0058] In certain embodiments, the histology section used to obtain an image for analysis is a tumour resection sample comprising more than (>) 4.5 mm2area of labelled tumour area.

[0059] In certain embodiments, the histology section used to obtain an image for analysis is a tumour biopsy sample comprising between 3 to 4.5 mm2of labelled tumour area. In certain embodiments, the tile areas of at least two such biopsy samples are used to calculate the alpha parameter according to the method.

[0060] In certain embodiments, the threshold applied by the patient classification method generates an output indicative of a first class indicative of good prognosis, based on a threshold for alpha that is between 0.5-0.9. In particular embodiments the threshold for a is about 0.8.

[0061] In certain embodiments of the method according to the invention, a good prognosis is defined as a more than 70% survival probability. Conversely, a poor prognosis is defined as less than or equal to 30% survival probability at 1500 days after diagnosis.

[0062] In certain embodiments, the T cell count dispersion value is calculated using the formula (Var - mu) I mu2, wherein (mu) represents the mean density of T cells present per mm2of the at least two regions of the tumour area; and ( Var) represents a measure of the variability in the number of T cells present at least two regions of the tumour.

[0063] In certain embodiments, the threshold value is in the range of 0.5 to 1 , particularly around 0.8. In certain embodiments, the threshold is the mean or the median T cell count dispersion value as previously determined in samples obtained from a cohort of at least 50, 100, or 150 patients diagnosed with the same cancer that the cancer patient providing the patient tumour sample has been diagnosed with.

[0064] In certain embodiments, the measurement step uses immunofluorescence and microscopy, or mass cytometry to acquire an image of the histology section.

[0065] In certain embodiments, the cancer from which the tissue section is taken, is lung cancer. In certain embodiments, the cancer is a non-small cell lung cancer.

[0066] Another aspect of the invention relates to a method to predict the prognosis of a cancer, executed on an image of a histology section having been obtained from the patient’s tumour, where the T cells in the image have been labelled. The method comprises in an image analysis step, quantifying the number, or density of T cells present in at least (>) 2 separate, non-overlapping regions of the image of the tumour area to provide a plurality of tile values; and in a calculation step, calculating a T cell count dispersion value (a) from the plurality of tile values (particularly by fitting a NBD model and extracting the a value from the NBD); and assigning the patient:

[0067] • a (high likelihood of) good prognosis if the a value is equal to or below an a threshold, or

[0068] • a (high likelihood of) poor prognosis of the cancer patient if the a value is above an a threshold.

[0069] The invention further encompasses the use of probes and image acquisition material as identified herein for use in the manufacture of a kit for the detection of dispersion of T cells in a tumour sample.

[0070] Likewise, the invention provides a biomarker composition, and a method of providing information for diagnosis as outlined in the present specification.

[0071] Another aspect of the invention relates to a computer program comprising instructions which, when the program is executed on by a computer, cause the computer to carry out the image analysis step, and the calculation step as described above.

[0072] The computer program receives an image as input, the image containing a section of tumour area labelled with a T cell marker, such that T cells can be counted by the software.

[0073] The computer program executes an image analysis step, whereby the image is divided into a plurality of non-overlapping regions (or tiles), and the number, or density of T cells present in at least 2 separate regions, to provide a plurality of tile values as a basis for a calculation step. In particular embodiments, the number or density of T cells in 5 or more regions of the image of the tumour area are counted.

[0074] The computer program performs a calculation step. T cell counts or densities obtained from the image analysis step, are used to calculate a T cell count dispersion value from the plurality of tile values. In particular embodiments, this value is obtained by extracting the parameters from an NBD fitted to the plurality of T cell counts for each tile obtained in the image analysis step.

[0075] Finally, the computer program assigns one of at least two classes to the patient sample, either the patient is classified as likely to have a good prognosis if the T cell count dispersion value is equal to or below a threshold, or as likely to have poor prognosis if the T cell count dispersion value is above a threshold. In certain embodiments, the threshold value is in the range of 0.5 to 1 , particularly around 0.8.

[0076] The invention further encompasses the following items:

[0077] A. A method to determine the prognosis of a patient diagnosed with cancer, the method comprising: in a measurement step, contacting a histology section containing a tumour area, the histology section having been obtained from the patient, with a labelled probe capable of specifically labelling T cells, and obtaining an image of the tumour area, in an image analysis step, quantifying the number, or density of T cells present in at least (>) 2 separate, non-overlapping regions of the image of the tumour area to provide a plurality of tile values; in a calculation step, calculating a T cell count dispersion value (a) of the plurality of tile values; and assigning the patient: a good prognosis if the a value is equal to or below a threshold, or a poor prognosis of the cancer patient if the a value is above a threshold.

[0078] B. The method according to item 1 , wherein a is calculated using the formula (Var- mu) I mu2, wherein (mu) represents the mean density of T cells present per mm2of the at least two regions of the tumour area; and ( Var) represents a measure of the variability in the number of T cells present at least two regions of the tumour.

[0079] C. The method according to item 1 or 2, wherein the calculation step further comprises submitting a data set comprising the plurality of tile values to a computer-implemented classification method to provide the T cell count dispersion value (a); wherein the computer-implemented classification method comprises the steps: determining a negative binomial distribution (NBD) to fit the data set; and determining a first NBD parameter from the NBD corresponding to the expected T cell number or density across all the plurality of tile values in the data set (p); and determining a second NBD parameter from the NBD corresponding to the T cell count dispersion value (a), wherein a characterizes the observed variability of T cell counts across all the plurality of tile values in the data set; and wherein the computer-implemented classification method generates an output indicative of at least two classes: a first class indicative of good prognosis of the cancer patient if the alpha value is equal to or below an a threshold, or a second class indicative of poor prognosis of the cancer patient if the alpha value is above the a threshold.

[0080] D. The method according to item C, wherein the NBD parameters p and a are determined by a probability density function (1 ).

[0081] E. The method according any one of the preceding items, wherein the probe capable of specifically binding T cells is a ligand capable of specifically binding a T cell antigen selected from the group consisting of CD3, CD4, and CD8.

[0082] F. The method according to any one of the preceding items, wherein the probe capable of specifically binding T cells is an antibody.

[0083] G. The method according to any one of the preceding items, wherein the probe capable of specifically binding T cells is an antibody specific for CD8.

[0084] H. The method according to any one of the preceding items, wherein the data set comprises at least 5, 7, 10, 20, 30, 40, or at least 50 tile values.

[0085] I. The method according to any one of the preceding items, wherein the area of each of the at least two tiles is equivalent to the area of a 300 x 300 pm2square, particularly is equivalent to the area of a square at least 700 x 700 pm2and maximally 1200 x 1200 pm2.

[0086] J. The method according to any one of the items, wherein the dispersion value a is calculated 3 based on at least 7 tile values.

[0087] K. The method according to any one of the preceding items, wherein the histology section is a tumour resection sample comprising more than (>) 4.5 mm2area of labelled tumour area.

[0088] L. The method according to anyone of the items A to K, wherein the histology section is a tumour biopsy sample comprising between 3 and 4.5 mm2of labelled tumour area.

[0089] M. The method according to any one of the preceding claims, wherein the threshold is the mean or the median dispersion value as previously determined for samples obtained from a cohort of at least 50, 100, or 150 patients diagnosed with a cancer originating in the same tissue as that of the cancer patient’s tumour.

[0090] N. The method according to any one of the preceding items wherein the threshold for a is between 0.5 to 0.9, particularly wherein the threshold for a is about 0.8. O. The method according to anyone of the items wherein the first NBD parameter and the second NBD parameter are determined by an optimisation algorithm selected from the list consisting of maximisation likelihood estimation, Bayesian modelling, and expectation maximization.

[0091] P. The method according to any one of the preceding items, wherein the image of a histology section obtained from the patient’s tumour has been obtained by immunofluorescence and microscopy, or imaging mass cytometry.

[0092] Q. The method according to any of the preceding items, wherein the cancer patient has been diagnosed with lung cancer, particularly non-small cell lung cancer.

[0093] R. The method according to any one of the preceding items, wherein a good prognosis is defined as a more than 70% survival probability, and a poor prognosis is defined as less than or equal to 30% survival probability at 1500 days after diagnosis.

[0094] S. A computer program comprising instructions which, when the program is executed on by a computer, cause the computer to carry out the image analysis step, and the calculation step as specified in any one of the preceding items.

[0095] Wherever alternatives for single separable features are laid out herein as “embodiments”, it is to be understood that such alternatives may be combined freely to form discrete embodiments of the invention disclosed herein.

[0096] The invention is further illustrated by the following examples and figures, from which further embodiments and advantages can be drawn. These examples are meant to illustrate the invention but not to limit its scope.

[0097] Description of the Figures

[0098] Fig. 1 shows A. workflow of biomarker analysis and calculating thresholds in a cancer cohort according to the invention. B. Whole slide image of a NSCLC tumour tissue section. Each segmented cell is represented by its centre point and coloured according to its cell phenotypes (CD8 T cells = black). The tissue was tiled into 700 x 700 pm2tiles. C. Histogram of the density count of CD8 across all tiles for the data shown in A. After, a negative binomial distribution was fitted to the data using a maximum likelihood estimation of the parameters p and a (light grey dotted line bottom right plot). Additionally, a Poisson distribution was fitted to the data as comparison (dark grey dotted line), but gave a poorer fit to the data, as did normal distribution (no shown).

[0099] Fig. 2 shows that CD8 dispersion can be used to predict OS of patients. A. Scatter plot showing the relationship of the dispersion parameter (y-axis) and the expected value (x-axis). Each dot represents the fitted value from the negative binomial model of a patient. Scatterplot on the righthand side shows the same data coloured by the patient population that shows higher than average dispersion (grey) and lower than average (black). B. Kaplan Meier plots of patient OS for high and low dispersion groups. C. Kaplan Meier plot showing OS for patients with a high and a low mean density of CD8 across tiles. D. Kaplan Meier plot showing OS for patients with a high and a low expected value of CD8 T cell density from the negative binomial model across tiles. Fig. 3 shows Cox proportional hazards model for CD8, CD3 and CD4 T cell density dispersion and clinical parameters. A. Multivariate cox proportional hazards model using the following variables (Dispersion, Stage, Histology, Adjuvant treatment and Neo-adjuvant treatment). CD8, or CD3, or CD4 low and high dispersion groups were determined based on the mean estimates of the CD8, CD3 or CD4 dispersion of the whole patient population B. As in A but the dispersion for CD8 is modelled as a continuous value.

[0100] Fig. 4 shows example images from a high (left side) and low (right side) CD8 dispersion tissue section. CD8 T cells are coloured in black, tumour area is grey, excluded healthy tissue (not relevant to calculation) coloured is light grey.

[0101] Fig. 5 shows impact of tile size on dispersion. A. Hazard ratios (HR) and 95% confidence intervals for the continuous dispersion score are shown on the y-axis and the different tile sizes used to extract the data from the tissue sections are displayed on the x-axis. B. Mean dispersion value on a random subset of the entire cohort computed across multiple tile sizes.

[0102] Fig. 6 shows impact of CD8 T cell density in healthy surrounding tissue to predict patient OS. A. Kaplan Meier plot showing the survival of a patient group with high and low CD8 T cell dispersion in the surrounding healthy tissue. B. Boxplot showing HR for the high dispersion group when either expanding or decreasing the area used for dispersion calculation with 700 x 700 pm2tiles. C. Multivariate cox proportional hazards model showing the prognostic value of the dichotomized CD8 from the healthy tissue and clinical variables for OS. D. Heatmap showing how the border expansion / decrease influences HRs when calculating dispersion with different tile sizes. P-values are colored gray scale, indicating their significance level.

[0103] Fig. 7 shows in silico determination of CD8 dispersion on biopsy sized samples. A. A grid representing biopsy sized regions overlaid on the tissue is shown. B. Boxplot showing the difference of dispersion calculated on biopsies and whole slide for all patients. Black points show ten random biopsy sized tissue samples for each patient and their distribution is visualized as box plot and is overlaid with indicated points representing the mean value for each patient across the 10-fold biopsy sampling. C. Boxplot showing the mean difference in dispersion calculated on biopsy sampling and whole slide dispersion across 10-fold sampling, for varying numbers of biopsy sized samples (1 to 30 biopsies of 3-4.5 mm2). D. Confusion matrices for dispersion calculated on one, two, three and five biopsy sized samples (from left to right) compared to ground truth derived from whole slides. Patients were split into 4 groups based on high and low p and a for probability and accuracy calculations. Additionally, the overall classification probability for only dispersion is shown on the bottom. E. As in D but using corrected dispersion values based on sample size for the classification of individual biopsies.

[0104] Fig. 8 shows differential immunoscore percentiles are not predictive of survival in NSCLC. Examples

[0105] Example 1: CD8 Dispersion biomarker

[0106] The advent of personalized medicine and ever-increasing numbers of available cancer drugs necessitate better prognostic and predictive biomarkers. In turn, this requires a deeper profiling and understanding of tumours and the tumour microenvironment (TME). To address this challenge the Integrated iMMUnoprofiling of large adaptive CANcer patient cohorts (IMMUcan) consortium collects samples and longitudinal clinical data from up to 3000 patients. All samples undergo bulk RNA-seq and whole exome sequencing (WES) for molecular profiling, whole slide multiplexed immunofluorescence (mIF) for broad and imaging mass cytometry (IMC) for detailed cellular profiling.

[0107] The inventors analyzed a retrospective sub-cohort of non-small cell lung cancer (NSCLC) consisting of stage l-IIIA resected tumour samples from 192 patients collected between 2012 and 2018 under the EORTC-SPECTA protocol (NCT 02214134). Median follow up was 2.8 years and 42 patients died of the disease. Bulk RNA-seq and WES data were processed using standard tools. An image analysis workflow for mIF and IMC data paired with a tiling approached from ecology and spatial data analysis were used for feature extraction and subsequent multivariate cox proportional hazard modelling.

[0108] The project leading to this application has received funding from the Innovative Medicines Initiative 2 Joint Undertaking under Grant Agreement n° 821558. This Joint Undertaking receives the support from the European Union’s Horizon 2020 research and innovation program and EFPIA

[0109] CD8 dispersion workflow

[0110] Negative binomial distributions are characterized by two parameters: the mean (p) and the dispersion (a). To estimate those two parameters per patient, first mIF derived CD8 T cell counts after segmentation (Figure 1 A) (Methods). The inventors restricted analysis to the tumour area as defined by a pathologist and used tiles of 700 x 700 pm to derive CD8 T cell counts per mm2for each tile per patient (Figure 1 B). A negative binomial model was fit to the data across all tiles per patient and model parameters p and a extracted using a maximum likelihood approach (Figure 1 B). A clear relationship between p and a was visible across patients (Figure 2A) indicating that the two parameters are not independent as also apparent from the definition of the negative binomial function (equation 1 , 2 and 3). Splitting the patient population by the average dispersion value of all patients and investigating their overall survival (OS) resulted in a strong difference with patients belonging to the high dispersion group having worst survival (Figure 2B). Splitting the population based on either the mean count or the expected value of CD8 T cells did not result in a difference in OS (Figure 2C, D). In a multivariate cox proportional hazards model patients with low CD8 dispersion had a 4-fold lower hazard compared to patients with high CD8 dispersion (Figure 3A) (HR = 0.29, p = 0.002). In addition to CD8, performing cell counts, and density calculations with the markers CD3, and CD4 across tiles within the labelled tumour area provided equivalent relationship to clinical parameters (Fig. 3A). We also tested the continuous variable of CD8 dispersion in a cox proportional hazards model and found and increase of the hazard of 2.62 with 1 unit increase of the dispersion (HR = 2.62, p = 0.013) (Figure 3B). Comparing images with high vs low dispersion revealed that the dispersion parameter seemed to split images with spatially homogeneous CD8 densities (low dispersion) vs those with variation in CD8 densities across tumours (high dispersion) (Figure 4).

[0111] Tile size and dispersion threshold

[0112] Next, we evaluated whether the size of the tiles mattered to define patient populations. We varied the tile size from 100 x 100 pm2to 3000 x 3000 pm2and extracted p and a per patient. Tile sizes below 300 x 300 pm2could not be used to achieve a prognostic score for patients (Figure 5A). From this, we conclude that for prognostic purposes dispersion values should be estimated from tiles sizes larger than 300 x 300 pm2and smaller than 1200 x 1200 pm2. We next investigated the robustness of a across different tile sizes. Generally, we found that the average dispersion value was very stable when sub-sampling patient images at one tile size (Figure 5B). However, we observed that the average dispersion value observed across patients decreased with increasing tile size suggesting that dispersion values derived from different tile sizes are not comparable.

[0113] Tumour area vs tumour area + border

[0114] In this study we made use of whole slide tissue sections from tumours which contained healthy tissue surrounding the tumour tissue. So far, we utilized only the tumour area annotated by a pathologist. We investigated the influence of the tumour border and surrounding normal tissue for prognostic purposes and found that the CD8 density in the surrounding healthy tissue did not contain prognostic information (Figure 6A, C). Next, we investigated how the addition of different amounts of healthy tissue surrounding the tumour would influence the prognostic value of the dispersion calculated with 700 x 700 pm2tiles (Figure 6B). We found that the expansion of the tumour region by 1 mm did not influence the prognostic power of the CD8 dispersion. Similarly, for different tile sizes expansions of up to 1200 pm had no negative impact on the prognostic value of the dispersion (Figure 6D). Expanding further into the normal tissue resulted in a loss of prognostic power, likely because the normal surrounding tissue does not contain valuable CD8 T cells information as shown above.

[0115] Sample size

[0116] Our cohort consisted of whole tumour resections which resulted in tissue sections with the area of the tumour tissue ranging from 6 up to 730 mm2(mean 268 mm2). To estimate whether our model could also be used on samples with the size of a biopsy (estimated at 3-4.5 mm2) we generated a grid of biopsy-sized tiles for each tissue section (Figure 7A). Next, we investigated the difference between the observed dispersion on biopsy-sized samples and the dispersion derived on from the whole section (Figure 7B). This revealed that from one biopsy, the dispersion of 50% of patients was underestimated by 50% from whole slide derived dispersion values. We then asked how many biopsies would be required to recapitulate the dispersion values from whole slide (Figure 7C). We found that generally, from biopsies we underestimated the dispersion compared to the whole slide derived dispersion given that we always maximally used 80% of the available biopsy-sized tiles per sub-sampling. Compared to using one biopsy, with 2 biopsies, the difference to whole slide derived dispersion decreased to about 30% and stabilized at around 10% difference with 30 biopsy-sized sub-samples.

[0117] To evaluate, if in a clinical test certain scenarios would be more likely to be predicted correctly from biopsies, we investigated whether our prediction accuracy differed for biopsy-sized samples of low and high dispersed whole slides. For a more detailed investigation, we split the patient population into 4 groups (low p low a; low p high a; high p low a; high p high a). With one biopsy, the overall accuracy was 61.1 % (Figure 7D). However, there were differences in the detection accuracy and probability of correctly identifying patients based on biopsies for the four different groups. We observed the highest probability of correctly identifying patients belonging to the low p and high a group based on biopsies (Figure 7D - 1 biopsy) and when just taking a into account the probability increased to 85% and did not increase much more when taking more biopsies (91 % with 5 biopsies). However, many patients classified as low p and high a based on whole slides were predicted to belong to the low-low class based on biopsies (28% accuracy with 1 biopsy). Identifying patients of the low-high group based on biopsies remained challenging even when taking 5 biopsies (68%). Ignoring p revealed that generally high dispersion samples were correctly identified with probabilities of 84-89% from one to five biopsies.

[0118] From these experiments it became clear that biopsy-based dispersion estimates would often result in better prognosis as expected due to the apparent underestimation of the dispersion from biopsies (Figure 7C). We argued that a model that would correctly identify patients with good prognosis would be preferential since patients for which the prediction would be poor could undergo further testing. Adjusting the biopsy-based dispersion values for the underestimation (Figure 7E) revealed that we could indeed increase the accuracy for the detection of the low-high group from 29% to 62% using just one biopsy and to 88% using five biopsies. This came at the cost of falsely classifying 20% of low-low patients as low-high using 1 biopsy. Overall, using the corrected biopsybased estimate of a we could correctly identify 80% of patients belonging to the low dispersion group using one biopsy and 90% using 5 biopsies.

[0119] Summary

[0120] The TME did not differ across clinical stage, smoking status or driver mutations. The TME of squamous cell carcinoma (LUSC) subtypes was enriched for immune suppressed cell types compared to adenocarcinoma (LUAD) subtypes in multiplexed immunofluorescence (mIF) and imagine mass cytometry (IMC) image data. Most samples (78%) contained B cell aggregates or tertiary lymphoid structures but their amount, size and location were not prognostic for overall survival (OS). In this cohort, not the density of CD8+T cells but the variability of CD8+ T cell density across tissue sections, next to factors such as CD8+T cell exclusion and broad T cell activation / exhaustion were strong prognostic factors for OS. Finally, these results highlight the use of multi-modal data integration for the identification of the strongest prognostic features.

[0121] The inventors used single cell data derived from whole slide immunofluorescence data as input to develop an algorithm that tiles the data into smaller rectangles and calculates cellular densities of CD8 T cells for each tile (Fig. 1 , Fig. 2). Depending on the heterogeneity of the CD8 infiltration in the tumour, the values for those CD8 densities across the tiles vary strongly. Such a process cannot be modelled with a standard Gaussian model but requires a negative binomial model. The negative binomial model was fit to the data from all tiles and extract model parameters (expected value: mean of CD8 densities across all tiles; dispersion: extra variation observed in the data) for a cohort of 192 NSCLC patients. Splitting the patient groups based on the average or the median value of the dispersion results in a strong prognostic signature for overall survival (Fig. 8).

[0122] The analyses reveal the potential of multi-modal data analysis and integration, and form the basis for analyses of all molecular and cellular profiling data from IMMUcan patients for biomarker discovery.

[0123] Example 2 Immunoscore

[0124] Immunoscore is a CD8 based biomarker system coming into the clinics for colorectal cancer with the following workflow:

[0125] • CD3, CD45, or CD8 staining

[0126] • tile the image data

[0127] • measure CD8 density in (1 ) tumour and in (2) tumour margin

[0128] • take the mean of the CD8 distribution as the patient value (corresponding to mu)

[0129] • compare the mu to the percentile of the population to identify groups

[0130] This method was not prognostic in the NSCLC cohort (Fig 8).

[0131] Methods

[0132] Cohort

[0133] Clinico-pathological data and samples were collected according to the EORTC-SPECTA protocol (NCT02834884), open in 20 European countries and 126 hospitals. All patients provided written informed consent at the time of sample collection for molecular and cellular analysis. The cohort consisted of 192 patients of clinical stage l-lll. All patients underwent surgery for tumour removal and formalin-fixed paraffin-embedded samples were generated from surgical specimens and sectioned at 4 pm thickness for processing. mIF staining and image acquisition

[0134] Multiplexed staining was performed on 4 pm thick tissue sections using an automated Ventana Discovery Ultra system (ROCHE). First, samples were deparaffinized followed by epitope retrieval (64 minutes at 98°C) and endogenous peroxidase quenching (Discovery Inhibitor, 8 minutes, Ventana). For each round of staining non-specific site blocking (Discovery Goat IgG and Discovery Inhibitor, Ventana), primary antibody incubation and secondary HRP-labelled antibody incubation for 16 minutes with Discovery OmniMap anti-rabbit HRP (Ventana, # 760-4311 ) or anti-mouse HRP (Ventana, #760-4310) were performed. Covalent dye labelling was performed using the OPALTM reactive fluorophore detection (Akoya Biosciences, Marlborough, MS, USA) for 12 min followed by heat denaturation of the antibodies for a next round of staining. The mIF stained slides were scanned using the PhenoImager™ HT (AKOYA) with the MOTiFTM mode for whole slide multispectral image acquisition. The scanner outputs a multi-channel 8-bit image in QPTIFF format.

[0135] Tumour regions on the tissue sections were annotated by a pathologist on consecutive Hematoxylin and Eosin (H&E) stained slides and manually transferred onto the mIF slides.

[0136] Image processing and single cell data generation

[0137] Unmixing of fluorescence signals was performed using non-negative matrix factorization as described in Eling, Dorier et al. (in preparation). Single cell segmentation was performed using nuclei detection with adaptive thresholding using the EBlmage package in R. Due to absence of membrane markers cell segmentation was performed expanding the nuclear mask by up to 5 pm or until touching a neighbouring cell. Fluorescence signal intensity was quantified using the mean expression over all pixels for each segmented cell. A threshold for CD8 was defined per image to separate CD8 positive from CD8 negative cells. CSV files were exported per image containing the location of each cell (X, Y coordinates), the intensity of CD8 and the location of each cell (in tumour or in healthy tissue region).

[0138] Tiling the images

[0139] The CSV data from each image was processed using R. First all cells were transformed to points in space using the sf package from R. A grid of defined size (edge length) was overlaid on top of all cells. Then, to calculate the density of cells in each tile, first a hull was calculated for each tile to identify the area that contained cells. The area from the hull was extracted and the number of CD8 T cells (CD8 positive cells) within the area was computed, resulting in a CD8+ / mm2 density per tile. To reduce noise in the data, tiles with a density above a threshold were discarded. The threshold was computed by taking the maximum density value observed across tiles for which the tissue area (convex hull) covers at least 90% of the tile size.

[0140] Model fitting

[0141] Under the assumption that the density per tile (rounded to the nearest integer) is negative binomial distributed (given that density is count data for a fixed surface), a maximum likelihood approach was used to fit a negative binomial distribution to the grid data for each sample. Therefore, assuming individual tile densities across a given patient are independent and identically distributed. The probability mass function of the parametric distribution used for computing the likelihood can be seen in equation 1 . This specific parametrization of the negative binomial allows for real-valued estimates of the parameters p and a. Given the parametrization from equation 1 , p equals the expected value of the random variable X (density count of CD8 T cells) (equation 2). Finally, the variance of a negative binomial distribution can be expressed in terms of p and a as shown in equation 3 highlighting that the variance is a mix of p and the dispersion variable a.

[0142] Cited references:

[0143] 1. Bruni D, et al., J. Nat Rev Cancer. 2020;20(11 ):662-680. doi:10.1038 / s41568-020-0285-7

[0144] 2. Donnem T., et al., Ann Oncol. 2016;27(2):225-232. doi:10.1093 / annonc / mdv560

[0145] 3. Thorsson V, et al., 2019 Aug 20 ;51 (2):411-412], Immunity. 2018;48(4):812-830.e14. doi:10.1016 / j.immuni.2018.03.023

[0146] 4. Galon J, et al., Immunity. 2020;52(1 ):55-81 . doi:10.1016 / j.immuni.2019.12.018

[0147] 5. Pages F, et al., Oncogene. 2010;29(8):1093-1102. doi:10.1038 / onc.2009.416

[0148] 6. Galon J, et al., Q J Nucl Med Mol Imaging. 2020;64(2):152-161 . doi:10.23736 / S1824- 4785.20.03249-5 Any patent document or scientific publication cited herein shall be deemed incorporated by reference herein in its entirety.

Claims

Claims1. A method to determine the prognosis of a patient diagnosed with cancer, the method comprising: in a measurement step, contacting a histology section containing a tumour area, the histology section having been obtained from the patient, with a labelled probe capable of specifically labelling T cells, and obtaining an image of the tumour area, in an image analysis step, quantifying the number, or density of T cells present in at least (>) 2 separate, non-overlapping regions of the image of the tumour area to provide a plurality of tile values; in a calculation step, submitting a data set comprising the plurality of tile values to a computer-implemented classification method to provide a T cell count dispersion value (a); wherein the computer-implemented classification method comprises the steps:• determining a negative binomial distribution (NBD) to fit the data set; and• determining a first NBD parameter from the NBD corresponding to the expected T cell number or density across all the plurality of tile values in the data set (p); and• determining a second NBD parameter from the NBD corresponding to the T cell count dispersion value (a), wherein a characterizes the observed variability of T cell counts across all the plurality of tile values in the data set; and• generating an output indicative of at least two classes: i. a first class indicative of a good prognosis of the cancer patient if the alpha value is equal to or below an a threshold, wherein the good prognosis is defined as a more than 70% survival probability at 1500 days after diagnosis; orII. a second class indicative of a poor prognosis of the cancer patient if the alpha value is above the a threshold, wherein a poor prognosis is defined as less than or equal to 30% survival probability at 1500 days after diagnosis; wherein the a threshold is between 0.5 to 0.9.

2. The method according to claim 1 , wherein the dispersion value is calculated using the formula (Var- mu) I mu2, wherein (mu) represents the mean density of T cells present per mm2of the at least two regions of the tumour area; and (Var) represents a measure of the variability in the number of T cells present at least two regions of the tumour.

3. The method according to claim 1 or 2, wherein the NBD parameters p and a are determined by a probability density function (1 ).( r(fc + |)(1) f (k; / , a) = Pr(X = k) = I - z —4. The method according any one of the preceding claims, wherein the probe capable of specifically binding T cells is a ligand capable of specifically binding a T cell antigen selected from the group consisting of CD3, CD4, and CD8.

5. The method according to any one of the preceding claims, wherein the probe capable of specifically binding T cells is an antibody.

6. The method according to any one of the preceding claims, wherein the probe capable of specifically binding T cells is an antibody specific for CD8.

7. The method according to any one of the preceding claims, wherein the data set comprises at least 5, 7, 10, 20, 30, 40, or at least 50 tile values.

8. The method according to any one of the preceding claims, wherein the area of each of the at least two tiles is equivalent to the area of a 300 x 300 pm2square, particularly is equivalent to the area of a square at least 700 x 700 pm2and maximally 1200 x 1200 pm2.

9. The method according to any one of the claims 3 to 8, wherein the dispersion value a is calculated according to the formula as specified in claim 3 based on at least 7 tile values.

10. The method according to any one of the preceding claims, wherein the histology section is a tumour resection sample comprising more than (>) 4.5 mm2area of labelled tumour area.11 . The method according to any one of the claims 1 to 9, wherein the histology section is a tumour biopsy sample comprising between 3 and 4.5 mm2of labelled tumour area.

12. The method according to any one of the preceding claims, wherein the threshold is the mean or the median dispersion value as previously determined for samples obtained from a cohort of at least 50, 100, or 150 patients diagnosed with a cancer originating in the same tissue as that of the cancer patient’s tumour.

13. The method according to any one of the preceding claims wherein the threshold for a is about 0.8.

14. The method according to any one of the preceding claims wherein the first NBD parameter and the second NBD parameter are determined by an optimisation algorithm selected from the list consisting of maximisation likelihood estimation, Bayesian modelling, and expectation maximization.

15. The method according to anyone of the preceding claims, wherein the image of a histology section obtained from the patient’s tumour has been obtained by immunofluorescence and microscopy, or imaging mass cytometry.

16. The method according to any of the preceding claims, wherein the cancer patient has been diagnosed with lung cancer, particularly non-small cell lung cancer.

7. A computer program comprising instructions which, when the program is executed on by a computer, cause the computer to carry out the image analysis step, and the calculation step of any one of the preceding claims.

Images