FIELD OF THE INVENTION The invention is in the field of molecular diagnostics of cancer, in particular the molecular typing of the tumor micro-environment (TME) as a spatial immune phenotype of a tumor involving measurement of nucleic acid biomarkers and use of a multi-gene classifier. The invention provides methods for typing the TME of a solid tumor as being T cell -inflamed or non-T cell-inflamed, preferably as being either T cell-inflamed, T cell- ignored or T cell-excluded, involving a multi-gene classifier, and methods of predicting disease prognosis or clinical outcome of ICI therapy of patients suffering from cancer based on the predictive nucleic acid biomarkers disclosed herein.

BACKGROUND OF THE INVENTION

Triple negative breast cancer (TNBC) is an aggressive form of breast cancer (BC) (accounting for 10-20% of all BCs) that is characterized by absence of hormone receptors and has limited therapeutic options. TNBC is considered the most immunogenic BC subtype based on relatively high numbers of tumor-infiltrating lymphocytes (TILs), which is reflected by a higher likelihood of response to immune checkpoint inhibition (ICI) when compared to other BC subtypes (Hammerl et al., Semin Cancer Biol, 52:178-188 (2018)). Nevertheless, objective response rates (ORR) to ICI in metastatic

TNBC are variable, and do not exceed 24% when administered as mono- therapy (Kwa et al., Cancer, 124(10):2086-2103 (2018)). Clinical benefit has been observed for first-line treatment with programmed cell death -ligand 1 (PD-L1) inhibitors (e.g. the PD-L1 blocking antibody atezolizumab) in combination with nanoparticle albumin-bound (nab)-paclitaxel, which has been approved by the EMA and FDA for PD-Ll-positive metastatic TNBC. Although this combination therapy induces survival benefit in PD-Ll- positive TNBC (Schmid et al., Lancet Oncol., 21(l):44-59 (2020)), still a significant proportion of TNBC patients does not benefit from ICI (e.g. anti- PD1 or anti-PDLl therapy). Moreover, preliminary data of primary TNBC treated with an anti-PDl antibody plus chemotherapy in the neoadjuvant setting suggest that PD-L1 expression is not associated with benefit for ICI (Schmid et al., N Engl J Med., 382(9):810-821 (2020); Savas et al., Cancer Cell, 37(5):623-624 (2020)).

Collectively, these findings point towards the need for better predictive markers and understanding of the underlying immune cell contextures to select ICI-responsive TNBC patients.

Several studies have examined the predictive value of tumor mutational burden (TMB) and TIL abundance in TNBC. While a high TMB has been associated with response to ICI-based therapies in melanoma, lung cancer, and colorectal cancer (Marra et al., BMC Med., 17(1): 1-9 (2019)), no significant association between TMB and ICI response has been found for TNBC (Samstein et al., Nat Genet., 51(2):202-206 (2019); Voorwerk et al., Nat Med., doi: 10.1038/s41591-019-0432-4 (2019); Molinero et al., J Immunother Cancer., 7(1): 1-9 (2019)). TILs are frequently present in primary TNBC and correlate with good prognosis as well as response to neoadjuvant chemotherapy and ICI in the metastatic setting (Hammerl et al., Semin Cancer Biol. 52:178-188 (2018); Voorwerk et al., Nat Med., doi: 10.1038/s41591-019-0432-4 (2019); Hammerl et al., Clin Cancer Res., doi: 10.1158/1078-0432. CCR- 19-0285 (2019); Loi et al., Ann Oncol.,

25(8): 1544-1550 (2014); Loi et al., J Clin Oncol, 37(7):559-569 (2019); Emens et al., JAMA Oncol., 5(l):74-82 (2019); Denkert et al., Lancet Oncol., 19(l):40-50 (2018)).

Furthermore, TILs predict overall survival (OS) to anti-PDl as a monotherapy independent of PD-L1 expression (Relationship Between Tumor-Infiltrating Lymphocytes and Outcomes in the KEYNOTE- 119 Study of Pembrolizumab Versus Chemotherapy for Previously Treated, Metastatic Triple-Negative Breast Cancer. 2019). Emerging evidence now suggests that next to abundance of TILs, also the cellular composition and activation state of TILs contribute to clinical outcome. For example, the presence of tissue- resident memory CD8+ T (Trm) cells provides more prognostic information when compared to CD8+ T cells (Savas et al., Nat Med., doi:10.1038/s41591- 018-0078-7 (2018)), and hallmarks of an ongoing immune response, such as clonal T cell expansion, correlate to anti-PDl response (Voorwerk et al., Nat Med.., doi:10.1038/s41591-019-0432-4 (2019)). In addition, the spatial localization of TILs has prognostic value in TNBC (Keren et al., Cell, 174(6):1373-1387.el9 (2018); Gruosso et al., J Clin Invest., 129(4):1785-1800 (2019)).

In this regard, three main spatial phenotypes have been identified and recognized for their association with clinical outcome in TNBC as well as other sohd tumor types: inflamed (also referred to as “hot”; characterized by the presence of intratumoral lymphocytes), excluded (also referred to as “altered”; lymphocytes are restricted to the invasive margin) and ignored (also referred to as “cold” or “desert”; characterized by lack of lymphocytes) (Gruosso et al., J Clin Invest., 129(4): 1785-1800 (2019); Galon et al.,

Science, 336:61-64 (2006); Galon et al., Nat Rev Drug Discov, doi: 10.1038/s41573-018-0007-y (2019)).

Although hterature reports on the existence of spatial phenotypes in solid tumors, so far it has not been studied whether these phenotypes are predictive of response to ICI in TNBC. Hitherto, it has also not been reported which immune evasive processes underpin these phenotypes in sohd tumors, such as TNBC.

There is a need for biomarkers that predict spatial localization of CD8+ T cells in primary and secondary tumors, and which are prognostic. There is also a need for biomarkers that allow for the stratification of patients with solid tumors such as TNBC into groups that are likely to respond to ICI therapy or other immune therapies and groups that will likely not respond to ICI therapy or other immune therapies and which consequently may benefit from combinatorial approaches or a different therapy.

SUMMARY OF THE INVENTION

The inventors have surprisingly established that the well-known spatial immune phenotypes (i) ‘ inflamed’ (ii) ‘excluded’ and (iii) ‘ignored’ can be accurately assigned by using a gene classifier. They have furthermore found that the gene classifier can directly be used to assess disease prognosis and provides a prognostic factor for improved patient survival in TNBC and other solid tumor types, and that the gene classifier can directly predict response to ICI therapy such as anti-PDl or anti-PD-Ll treatment in metastatic TNBC and other solid tumors such as melanoma. Furthermore, the inventors have identified genes, the expression of which are as such associated with prognosis of solid tumors such as TNBC.

It was also established that the three spatial immune phenotypes in primary TNBC are characterized by distinct immune determinants as well as TME and immune response-mediated paths of T cell evasion. These findings now provide a rationale for therapies on the basis of spatial immune phenotypes to enhance response to ICI or other anti-cancer agents in solid tumors, thereby creating a new clinical situation. More specifically, it was established that anti-cancer immunotherapy, such as immune checkpoint inhibitor therapy, preferably a PD-1 or PD-L1 inhibitor, can be employed in relation to all three immune phenotypes with the proviso that: (i) for treatment of subjects having a tumor with the ignored phenotype the immunotherapeutic agent is to administered in combination with a WNT inhibitor and/or an inhibitor of M2 macrophages; (ii) for treatment of subjects having a tumor with the excluded phenotype the immunotherapeutic agent is to be administered in combination with a TGFβ inhibitor and/or a VEGF inhibitor; and (iii) for treatment of subjects having a tumor with the inflamed phenotype the immunotherapeutic agent is to be administered as a single agent or in combination with another immunotherapeutic agent, an epigenetic drug and/or an inhibitor of M2 macrophages. See Example 1, Figures 10 and 13.

Therefore, the invention provides in a first aspect a method for typing a tumor micro-environment (TME) of a solid tumor as being of the immune phenotype T cell-inflamed, T cell-excluded or T cell-ignored, comprising the steps of: - providing a test sample of a solid tumor comprising a TME from a subject; measuring in said test sample the gene expression level for (i) at least one gene selected from group 1 consisting of IGHG1, NKG7, IL2RG, IL7R, CCL18, PVRIG, PLAC8, CCL5, SIRPG,

CORO1A, LCK, TRBC1, GZMB, CXCL13, and WARS; and (ii) at least one gene selected from group 2 consisting of COL5A1, SPON1, CAMK2N1, FAP, SPOCK1, COL1A1, SCGB2A1, AKR1C2, CPE, SCGB2A2, TCN1, TPSABl, and MMP2; and (iii) at least one gene selected from group 3 consisting of PERP, THBS2, ASPN, COLlOAl, TUFT1, GREM1, CEACAM6, ENTPD3, IGFBP5, PBX1, CXADR, GPRC5A, SDCl and CALML5; comparing the measured test sample gene expression levels to a reference gene expression level, and typing the inflam oefd said solid tumor of said subject as being T cell- inflamed, T cell-excluded or T cell-ignored on the basis of the comparison of said measured gene expression level and said reference gene expression level.

In a preferred embodiment, the method of the invention is a method for typing a TME of a solid tumor as being of the immune phenotype T cell-inflamed, T cell-excluded or T cell-ignored; wherein said reference gene expression level comprises the gene expression level of said at least one gene in at least one reference sample of each of said three immune phenotypes, and wherein said inflam oefd said solid tumor is typed as being T cell-inflamed, T cell-excluded or T cell-ignored on the basis of the comparison of said measured gene expression level and said reference.

In another preferred embodiment of a method of the invention, said gene expression level is measured for at least 2 genes selected from each of group 1, group 2, and group 3.

In another preferred embodiment of a method of the invention, said gene expression level is measured for at least 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14 or 15 genes selected from group 1, and at least 3, 4, 5, 6, 7, 8, 9,

10, 11, 12, or 13, genes selected from group 2, and at least 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, or 14 genes selected from group 3.

In another preferred embodiment of a method of the invention, said gene expression level is measured for IGHGl, NKG7, IL2RG, IL7R, CCL18, PVRIG, PLAC8, SIRPG, COROIA, LCK, TRBC1, GZMB, CCL5, CXCL13, WARS, COL5A1, SPON1, CAMK2N1, FAP, SPOCK1, COL1A1, SCGB2A1, AKR1C2, CPE, SCGB2A2, TCN1, TPSAB1, MMP2, PERP, THBS2, ASPN, COLlOAl, TUFT1, GREM1, CEACAM6, ENTPD3, IGFBP5, PBX1, CXADR, GPRC5A, SDCl and CALML5.

In another preferred embodiment of a method of the invention, the sohd tumor is selected from the group formed by BRCA: breast carcinoma such as triple negative breast cancer (TNBC); CESC: cervical squamous cell carcinoma and endocervical adenocarcinoma; HNSCC: head and neck squamous cell carcinoma; KICH: chromophobe renal cell carcinoma; BLCA: bladder urothelial carcinoma, and SKCM: skin cutaneous melanoma.

In another preferred embodiment of a method of the invention, the sohd tumor is a triple negative breast cancer (TNBC).

In another preferred embodiment of a method of the invention, the sohd tumor is a primary tumor, recurrent tumor or a secondary (metastasized) tumor. In another preferred embodiment of a method of the invention, measuring the gene expression level is performed by qPCR, microarray analysis or next- generation sequencing (NGS).

In another preferred embodiment of a method of the invention, said method for typing is a method for predicting the prognosis of a subject with a solid tumor, and wherein, when said TME of a solid tumor from said subject is typed as T cell-inflamed said subject has a favorable prognosis, and wherein, when said TME of a solid tumor from said subject is typed as being T cell-excluded or T cell-ignored, said subject has an unfavorable prognosis.

In another aspect, the present invention provides a method for assigning a subject having a solid tumor comprising a TME to an immunotherapy-responsive or immunotherapy-unresponsive group, such as an ICI-responsive or ICI-unresponsive group, said method comprising the steps of: performing a method for typing according to the invention as described above; and assigning said subject to the immunotherapy- responsive group when said inflam oefd a solid tumor from said subject is typed as T cell - inflamed or assigning said subject to the immunotherapy- unresponsive group when said inflam oefd a solid tumor from said subject is typed as T cell-excluded or T cell-ignored.

In another aspect, the present invention provides an immunotherapeutic agent for use in the treatment of a subject having a sohd tumor comprising a TME, wherein said subject is assigned to the immunotherapy-responsive group according to a method for assigning a subject according to the invention; optionally wherein said immunotherapeutic agent is for administration in combination with an epigenetic drug and/or an inhibitor of M2 macrophages.

In another aspect, the invention provides an immunotherapeutic agent for use in the treatment of a subject having a solid tumor comprising a TME, wherein said subject is assigned to the immunotherapy- unresponsive group according to a method for assigning according to the invention; and wherein said TME of said solid tumor is typed as being T cell-ignored; and wherein said immunotherapeutic agent is for administration in combination with a WNT inhibitor or an inhibitor of M2 macrophages such as colony stimulating factor 1 receptor (CSF1) inhibitor.

In another aspect, the present invention provides an immunotherapeutic agent for use in the treatment of a subject having a sohd tumor comprising a TME, wherein said subject is assigned to the immunotherapy-unresponsive group according to a method for assigning a subject according to the invention; and wherein said TME of said solid tumor is typed as being T cell-excluded; and wherein said immunotherapeutic agent is for administration in combination with a TGFβ inhibitor and/or a VEGF inhibitor.

In a preferred embodiment of these aspects, the immunotherapeutic agent is an immune checkpoint inhibitor (I Cl), preferably a PD-1 or PD-L1 inhibitor.

In another aspect, the invention provides a method for treating a subject having a solid tumor comprising a TME, comprising the steps of: - performing a method for assigning according to the invention; - administering a therapeutically effective amount of an immunotherapeutic agent, optionally in combination with an epigenetic drug and/or an inhibitor of M2 macrophages, to said subject if the subject is assigned to the immunotherapy-responsive group.

In another aspect, the invention provides a method for treating a subject having a solid tumor comprising a TME, comprising the steps of: - performing a method for assigning according to the invention; - administering a therapeutically effective amount of an immunotherapeutic agent in combination with a WNT inhibitor or an inhibitor of M2 macrophages if said subject is assigned to the immunotherapy-unresponsive group and if said TME of said solid tumor is typed as being T cell-ignored. In another aspect, the invention provides a method for treating a subject having a solid tumor comprising a TME, comprising the steps of: - performing a method for assigning according to the invention; - administering a therapeutically effective amount of an immunotherapeutic agent in combination with a TGFβ inhibitor and/or a VEGF inhibitor if said subject is assigned to the immunotherapy-unresponsive group and if said TME of said solid tumor is typed as being T cell-excluded.

In preferred embodiments of said methods for treatment, the immunotherapeutic agent is an immune checkpoint inhibitor (I Cl), preferably a PD-1 or PD-L1 inhibitor.

In another aspect, the present invention provides a method for predicting a prognosis for a subject with a solid tumor comprising a TME, said method comprising the steps of: - measuring in a sample of a solid tumor comprising a TME from a subject a gene expression level for (i) at least one gene selected from group 1 consisting of IGHGl, NKG7, IL2RG, IL7R, CCL18, PVRIG, PLAC8, CCL5, SIRPG, COROlA, LCK, TRBCl, GZMB, CXCL13, and WARS; and (ii) at least one gene selected from group 2 consisting of COL5A1, SPON1, CAMK2N1, FAP, SPOCK1, COL1A1, SCGB2A1, AKR1C2, CPE, SCGB2A2, TCN1, TPSAB1, and MMP2; and/or (iii) at least one gene selected from group 3 consisting of PERP, THBS2, ASPN, COL10A1, TUFT1, GREM1, CEACAM6, ENTPD3, IGFBP5, PBX1, CXADR, GPRC5A, SDC1 and CALML5; comparing said measured gene expression level to a reference gene expression level; and typing said subject as having a favorable or unfavorable prognosis on the basis of the comparison of said measured gene expression level and said reference gene expression level.

The invention also provides a method for measuring or determining a gene expression level in a test sample of the TME of a solid tumor of a subject, comprising the steps of providing a test sample of the TME of a solid tumor of a subject; and measuring in said test sample the gene expression level for (i) at least one gene selected from group 1 consisting of IGHG1, NKG7, IL2RG, IL7R, CCL18, PVRIG, PLAC8, CCL5, SIRPG, CORO 1 A, LCK, TRBC1, GZMB, CXCL13, and WARS; and (ii) at least one gene selected from group 2 consisting of COL5A1, SPON1, CAMK2N1, FAP, SPOCK1, COL1A1, SCGB2A1, AKR1C2, CPE, SCGB2A2, TCN1, TPSABl, andMMP2; and (iii) at least one gene selected from group 3 consisting of PERP, THBS2, ASPN, COLlOAl, TUFT1, GREM1, CEACAM6, ENTPD3, IGFBP5, PBX1, CXADR, GPRC5A, SDCl and CALML5.

Preferably, in said method of measuring or determining a gene expression level, said step of providing a test sample and/or said step of measuring is as disclosed in relation to a method for typing of the invention.

The invention also provides a method of treating a subject having a solid tumor comprising a classified TME, comprising the step of: - administering a therapeutically effective amount of an immunotherapeutic agent to a subject having a solid tumor comprising a classified TME; wherein said subject is assigned to the immunotherapy-responsive group according to a method for assigning according to the invention and wherein said TME of said solid tumor is classified as being T cell-inflamed; and wherein said immunotherapeutic agent is an ICI or other type of immune therapy, preferably as indicated herein.

A classified inflam aesd defined in embodiments of the present invention refers to a inflam cleadssified by methods of this invention.

The present invention, in providing methods for the stratification of patients with solid tumors, such as TNBC, into groups that are likely to respond to ICI therapy or other immune therapies and groups that will likely not respond to ICI therapy or other immune therapies, now provides for companion diagnostic methods to aid in selecting or excluding patients or patient groups for treatment with immunotherapeutic agents and which patients or patient groups may benefit from combinatorial approaches or an altogether different therapy. The invention thereto provides a method of treating a subject having a sohd tumor comprising a classified TME, comprising the step of: - administering a therapeutically effective amount of an immunotherapeutic agent to a subject having a solid tumor comprising a classified TME; wherein said subject is assigned to the immunotherapy-unresponsive group according to a method for assigning according to the invention and wherein said classified TME of said solid tumor (preferably TNBC) is typed as being T cell -ignored; and wherein said immunotherapeutic agent is for administration in combination with an WNT inhibitor or with a colony stimulating factor 1 receptor (CSF 1) inhibitor.

The invention also provides a method of treating a subject having a solid tumor comprising a classified TME, comprising the step of administering a therapeutically effective amount of an immunotherapeutic agent to a subject having a solid tumor comprising a classified TME; wherein said subject is assigned to the immunotherapy-unresponsive group according to a method for assigning according to the invention and wherein said classified inflam oefd said solid tumor (preferably TNBC) is typed as being T cell-excluded; and wherein said immunotherapeutic agent is for administration in combination with a TGFβ inhibitor or a VEGF inhibitor.

The invention also provides a use of an immunotherapeutic agent in the manufacture of a medicament for treatment of a subject having a sohd tumor comprising a classified TME; wherein said subject is assigned to the immunotherapy-responsive group according to a method for assigning according to the invention and wherein said classified TME of said solid tumor is typed as being T cell-inflamed; and wherein said immunotherapeutic agent is an ICI or other type of immune therapy.

The invention also provides a use of an immunotherapeutic agent in the manufacture of a medicament for treatment of a subject having a sohd tumor comprising a classified TME; wherein said subject is assigned to the immunotherapy-unresponsive group according to a method for assigning according to the invention and wherein said classified TME of said solid tumor is typed as being T cell-ignored; and wherein said immunotherapeutic agent is for administration in combination with an WNT inhibitor or with a CSF 1 inhibitor.

The invention also provides the use of an immunotherapeutic agent (e.g. an ICI or other type of immune therapy) in the manufacture of a medicament for treatment of a subject having a solid tumor comprising a classified TME; wherein said subject is assigned to the immunotherapy- unresponsive group according to a method for assigning according to the invention and wherein said classified TME of said solid tumor is typed as being T cell-excluded; and wherein said immunotherapeutic agent is for administration in combination with a TGFβ inhibitor or a VEGF inhibitor.

DESCRIPTION OF THE DRAWINGS

Figure 1. Study Design. Different steps and types of analyses regarding spatial immune phenotypes. Colors of boxes reflect the cohorts used for each step (for details and clinical characteristics of cohorts see M&M section and Table 2). For cohort A and F spatial phenotypes were identified using IHC of CD8+ T cells on whole slides and for cohort B-E spatial phenotypes were assigned using the gene-classifier.

Figure 2. Workflow for digital image analysis of immune stainings. A. Whole slide images of CD8+ T cell IHC with border and center stamps (regions of interest, red and yellow, respectively) with close-up (20x magnification) of one border stamp (top), separation of tissue (red) and empty space (blue) (middle) and identification of CD8-positive (red) and negative (blue) cells (bottom). Yellow fine indicates outer tumor margin. B. Image analysis for multiplex IF of immune effector panel at border and center of an inflamed TNBC; from left to right: multicolour IF image, tissue segmentation (red: tumor; green: stroma; orange: empty space, yellow line: outer tumor margin); cell segmentation; and individually phenotyped cells (Inform software). Stamp size: 670x502 pm ² ; resolution: 2 pixels/pm ² ; pixel size: 0.5x0.5 pm ² .

Figure 3. Immune effector cells according to spatial phenotypes in TNBC. Boxplots show cell densities (in cell number/mm ² ) following staining and analysis using immune effector panel. A. Tumor border. B. Tumor center. C. Stroma border. D. Stroma Center. E. Boxplots show mean distances between CD8+ T cells and other cell types. F. Table with Hazard Ratios (HR) and 95% confidence intervals (Cl, between brackets) for MFS of immune cell densities (significant HR values are shown in bold, analyses not corrected for multiple testing). Significant differences are: ***, p<0.001; **, p<0.01; *, p<0.05; NS, p>0.5.

Figure 4. Performance and clinical validation of gene classifier. Gene classifier (as described in Results section and Figure 7 A) was tested for: A. Correct assignment of spatial immune phenotypes in primary TNBC; B. Correct assignment of spatial immune phenotypes in TNBC lymph node metastases; C. Prediction of response to anti-PDl treatment. Abbreviations: PPV: positive predictive value, NPV: negative predictive value.

Figure 5. Spatial phenotypes in non-TNBC cancers. A. Forest plots show HRs and CIs of individual classifier genes (Cohort B, not corrected for multiple testing). B. Stacked bar-graphs show frequencies of spatial phenotypes in different breast cancer subtypes (left, Cohort B) and various cancer types (right, Cohort E); below right panel ORRs are listed for ICI treatments of respective cancer types. C. Kaplan-Meier curves for OS stratified per spatial phenotype in various cancer types (Cohort E). D. Boxplots display average expression of gene-sets for the excluded as well as inflamed phenotype in responding and non-responding melanoma patients following ICI treatment (left, Hugo Cohort; right, Riaz Cohort). Significant differences are: **, p<0.01; *, p<0.05; NS, n>0.5. Abbreviations: PAAD: pancreatic adenocarcinoma; BRCA: breast carcinoma; PRAD: prostate adenocarcinoma; BLCA: bladder urothelial carcinoma; HNSC: head and neck squamous cell carcinoma; CO AD: colon adenocarcinoma; LUAD: lung adenocarcinoma; CESC: cervical squamous cell carcinoma and endocervical adenocarcinoma; KICH: chromophobe renal cell carcinoma; SKCM: skin cutaneous melanoma. References for ORR: 1: Henriksen et al., Can Treat Review, 2019; 2: Kwa et al., Cancer, 2018; 3: Fay et al., Ann Transl Med, 2019; 4: Olivia et al., Annals Oncol, 2019; 5: Kamatham et al., Cur Col Can Rep, 2019; 6: Kim et al., Invest Clin Urol, 2018; 7: Regzedmaa et al., Oncotargets Ther, 2019; 8: Liu et al., Frontiers Pharmacol, 2019; 9: Flynn et al., Ther Adv Med Oncol, 2019.

Figure 6. Spatial immune contexture is prognostic in TNBC. A, B. Representative whole slide images of CD8+ T cell spatial phenotypes with percentage of patients per phenotype (A) and corresponding Kaplan-Meier curves for metastasis-free survival (MFS), disease-free survival (DFS) and overall survival (OS) (B, p-values show log-rank test for trend; time is displayed in months). C. Representative multiplex IF images of immune effector cells at the tumor border and center of each spatial phenotype. D. Circle plots show mean and SD of immune cell densities (cells/mm ² ) at border and center. E. Histograms show mean distances in pm between CD8+ T cells and other cell types (x-axis) versus their respective densities (cells/mm ² , y-axis). F. Boxplots show total number of tertiary lymphoid structures (TLS, identified by consecutive stainings of CD20+ B cells (top) and CD4+ T cells (bottom), see black squares in images). Significant differences are: ***, p<0.001; **, p<0.01; *, p<0.05, NS, p>0.5. Figure 7. Gene classifier assigns spatial phenotypes of CD8+ T cells and stratifies metastasized TNBC patients according to ICI response. A. Heatmap shows median expression of classifier genes per spatial phenotype in the discovery set (red: high expression, blue: low expression; Cohort Al). B. Forest plots show HRs and CIs of classifier gene-sets (Cohort B). C. Kaplan Meier curves of assigned spatial phenotypes in primary TNBC patients (Cohort E). D. Forest plots show Odds Ratios (OR) for response to anti-PD-1 treatment of classifier gene-sets (Cohort D, TONIC trial). E. Boxplots display average expression of classifier gene-sets in responding (CR+PR+SD > 24 weeks) and non-responding (PD) patients (Cohort D). F. ROC curves predict clinical response (PR+CR+SD) with areas under the curve (AUC) and CIs for gene sets of excluded-, inflamed- or a combination of the two phenotypes (average expression of respective gene sets was used) (first 3 panels), or for standardly used predictive markers, such as frequency of stromal TILs and PDL1 positivity of immune cells (Cohort D) (last 2 panels). G. Proportions of assigned spatial phenotypes in patients with metastatic TNBC responding or not responding to anti-PD-1 treatment (pre- treatment biopsies) and H. Corresponding survival curves (Cohort D).

Figure 8. Predictive value of spatial immune phenotype gene classifier versus public classifiers. A. Box-plots display signature scores in responder (CR+PR+SD) and non-responder patients from TONIC trial (cohort E) (PD) according to a short (6-gene) and extended (18-gene) interferon gamma signature (IFN- _Υ -related mRNA profile) from Ayers et al., 2017 (J Clin Invest. 127(8):2930-2940). ROC displays area under the curve for predicting anti-PDl response (CR+PR+SD) using the extended signature. B. Multivariable analysis including spatial immune phenotype gene-classifier and the extended IFNy signature. C. Box plots and ROC according to a T cell-exclusion program signature from Jerby-Arnon et al., 2018 (Cell 175(4):984-997). D. Box plots and ROC according to a tertiary lymphoid structure signature from Cabrita et al., 2020 (Nature 577:561-565).

Figure 9. Genomic features of spatial phenotypes. The following parameters were tested for differential presence in spatial phenotypes (determined by the gene-classifier) in TNBC: A. BRCA status (proportion). B. Loss of bet a2 -micro globin (copy number). C. Total number of different types of mutations. D. Total number of predicted neo-antigens. E. Proportions of most abundant mutational signatures. F, G. Frequencies of mutational signatures-3 and 5. H. TCR repertoire skewness (based on the Gini-Simpson index). I. Total number of different TCR-Vbeta reads. For all above parameters Cohort B was used, spatial phenotypes were assigned according to classifier. Significant differences are: ***, p0.001; **, p<0.01;

*, p<0.05, NS, p>0.5.

Figure 10. Spatial phenotypes interrogated for immune determinants and evasive pathways. A. Heatmap shows scaled average frequencies of immune cell populations based on Cibersort deconvolution (red: high, blue: low, immune cell populations with significant differences among spatial phenotypes are indicated in bold); corresponding boxplots show immune cell populations with differential abundances among spatial phenotypes. B. Heatmap shows scaled average expression of gene-sets related to T cell evasion (differential gene-sets are indicated in bold). C. Volcano plot of differential gene expressions between excluded and inflamed (upper), and ignored and inflamed phenotypes (lower); top DE genes related to T cell evasion are shown in bold. D. IPA analyses of cells, molecules and pathways associated with spatial phenotypes; and lists of major characteristics per spatial phenotype (bottom). E. Correlations between expressions of COLlOAl and TGFB- or VEGF-signaling in the excluded phenotype. F. Correlations between expressions of CD 163 and WNT targets or negative regulators of PPAR genes in the ignored phenotype. G. Correlations between presence of activated dendritic cells (according to BATF3 expression) and expressions of chemokines or type-I IFN genes in the inflamed phenotype. H. Correlations between expressions of CD8A and various T cell evasive genes/gene-sets (all phenotypes) (all correlations show regression coefficients and p-values).

Figure 11. Spatial immune phenotypes are characterized by distinct T cell evasive mechanisms. A. Representative images of cells and molecules related to spatial phenotypes (spatial phenotype panel) at the tumor border and center. B. Circle plots show mean and SD of cell densities at border and centre regions per mm ² ; Collagen- 10 was scored as positive tissue area. C. Histograms show mean distances in pm between CD8+ T cells and other cell types (x-axis) versus respective densities (y-axis). D. Boxplots show numbers of high endothelial venules (HEV, identified via MECA-79 staining, black arrow) and MHC-II expression of tumor cells (no distinction between border and centre, pink arrow: tumor cells; yellow arrow: adjacent normal breast lobules; green arrow: immune cells). E. Neutrophil densities at border and centre and representative image is shown. F. Boxplots show numbers of different T cell markers stained on consecutive slides, and representative images are shown. Significant differences are: ***, p0.001; **, p<0.01; *, p<0.05, NS, p>0.05.

Figure 12. Spatial phenotypes in metastasized TNBC according to distant sites and induction treatment. A. Stacked bar graphs show frequencies of spatial phenotypes assigned via gene classifier in different metastatic lesions (number indicates total number of lesions). B. Frequencies of spatial phenotypes in Cohort Al (CD8 stainings, primary tumors), Cohort C (gene- classifier, primary tumors) and Cohort D (gene-classifier, metastatic lesions). C. Frequencies of spatial phenotypes from paired pre- and post- induction treatments (number indicates a change to inflamed phenotype). Significant differences are: ***, p<0.001; **, p<0.01; *, p<0.05; NS, p>0.05. Abbreviations: cis: cisplatin; cycl: cyclophosphamide; dox: doxorubicin; irr: irradiation; none: no induction.

Figure 13. Illustration of immune contextures per spatial phenotype in relation to paths of T cell evasion as well as response to ICI. Distinctive and dominant pathways (in bold) per phenotype. When phenotypes are targeted in an immune phenotype-specific manner (in boxes), this would sensitize TNBC to ICI (see Discussion section for details).

Figure 14. Standardly used predictive markers of ICI response in patients with metastatic TNBC. A. Boxplots show fraction of PDL1 -positive immune cells (upper plot) and fraction of stromal TIL (sTIL) per response group (all Cohort D). B. Kaplan-Meier curves for OS and PFS with different cutoffs (>1%, left) for PDL1 and sTIL (>5%). C. Stacked bar graphs show frequencies of spatial phenotypes stratified by immune cell PD-L1>1%; table shows spatial phenotypes and immune cell PD-L1 in multivariable models according to prognostic value (HR, 95% Cl between brackets and p-value) as well as predictive value (OR, 95% Cl and p-value). D. Stacked bar graphs show frequencies of spatial phenotypes stratified by sTIL>5%; table shows spatial phenotypes and sTIL in multivariable models according to prognostic value (HR, Cl and p-value) as well as predictive value (OR, Cl and p value).

Figure 15. Accuracy of the gene classifier. This Figure shows on the y-axis the accuracy of the gene classifier in predicting the spatial immune phenotypes when randomly chosen genes from each group mentioned in Table 1 are included in the classifier. The Figure shows that, starting with one random gene per group and increasing, accuracy improves. Already at 5 genes per group, an accuracy of >60% is attained; at 10 genes per group, an accuracy of >70% is attained.

DETAILED DESCRIPTION OF THE INVENTION Definitions

The term “cancer” or “tumor”, as used herein, refers to a disease characterized by dysregulated cell proliferation and/or growth. The term comprises benign and mahgnant cancerous diseases, such as tumors, and may refer to an invasive or non-invasive cancer. The term comprises all types of cancers, including carcinomas, sarcomas, lymphomas, germ cell tumors, and blastomas. In one preferred embodiment, the term cancer relates to solid tumor. Examples of solid tumor include stomach cancer, breast cancer, lung cancer, colorectal cancer, liver cancer, gallbladder cancer, pancreatic cancer, thyroid cancer, prostate cancer, ovarian cancer, uterine cervical cancer, bladder cancer, sarcoma, glioma, mesothelioma, colorectal tumors, hepatic tumors, and head and neck tumors, with preference for breast cancer, lung cancer, colorectal cancer, stomach cancer, prostate cancer, and liver cancer. In another preferred embodiment, the term cancer relates to breast cancer. In yet another preferred embodiment, cancer relates to invasive breast cancer. The term “invasive breast cancer”, as used herein, refers to a breast cancer that spreads beyond the layer of tissue in which it developed into surrounding healthy, normal tissue. Invasive breast cancer may spread from the breast through the blood and lymph system to other parts of the body. In one highly preferred embodiment, cancer relates to triple-negative breast cancer (TNBC), which does not express the genes for HER2neu (ERBB2), estrogen receptor (ER), and progesterone receptor (PR).

The terms “spatial phenotype” and “spatial immune phenotype” refer to the three-type model for the immune phenotype based on the distribution of immune cells in the TME including (a) T cell -inflamed, (b) T cell-excluded, and (c) T cell-ignored also known as T cell desert (Gruosso et al. 2019 J Clin Invest. 129(4):1785-1800; Galon et al. 2006 Science 336:61- 64; Galon et al. 2019 Nature Reviews Drug Discovery 18: 197-218; Chen & Mellman 2017 Nature 541(7637):321-330). It has been recognized that the three main spatial phenotypes are associated with different clinical outcome in TNBC as well as other cancer types. The three immune phenotypes can essentially be identified based on spatial distribution of tumor-infiltrating CD8 ⁺ T cells as histologically (IHC) discernable in whole tissue sections (e.g. at the border and center of a solid tumor as displayed in Figures 2 A and 6 A and schematically shown in Figure 13) from tumors of different individuals. T cell -inflamed (shortly: inflamed) is characterized by CD8 ⁺ T cells being evenly distributed across border and center (e.g. almost equal frequencies of CD8 ⁺ T cells at the border and center); T cell-excluded (shortly: excluded) is characterized by CD8 ⁺ T cells being predominantly located at the tumor border, not the center (e.g. >10 times more CD8+ T cells at the border compared to center); T cell-ignored (shortly: ignored) is characterized by negligible presence of CD8 ⁺ T cells neither at border nor center (hardly any CD8 ⁺ T cells present at the border and center). The above quantitative distribution of CD8 ⁺ T cells may be assessed by IHC, preferably by digital image analysis of IHC-stained tissue sections. As a criterion, the number of CD8 ⁺ T cells per mm ² in different compartments may be applied as follows: inflamed, >200 cells/mm ² at border and ratio between border and center <10; excluded, >200 cells/mm ² at border and ratio between border and center >10; ignored <150 cells/mm ² at border and center.

The terms “gene classifier”, “gene expression classifier”, and “multi-gene biomarker” are used interchangeably herein to refer to a gene signature or molecular indicator that can discriminate between different spatial immune phenotypes of a solid tumor type based on differential gene expression between these phenotypes. The gene classifier discriminates between inflamed, excluded and ignored phenotypes based on differentially expressed genes in a test sample compared to a reference, the reference preferably being the averaged expression of all classifier genes in a set comprising representative tumor samples of each of the three spatial phenotypes of the same cancer. Preferably, the ranking indicated in Table 1 is used. The phenotype of an unknown tumor (e.g. test sample) is assigned to one of the three classes based on highest Spearman rank-correlations between the test sample and (ranked) expressions of classifier genes in the three spatial phenotypes. The reference set may include tumor samples of which the spatial immune phenotype is determined by classical immune histochemical (IHC) methods, preferably based on differential presence and spatial distribution of CD8 ⁺ T cells in the reference samples, which may either be scored manually or through digital image analysis as explained in the Experimental section. Differential gene expression analysis in these reference samples for one or more of the genes of the classifier indicated herein may provide suitable reference data with which test data of unknown samples of the same cancer can be correlated. In the gene classifier of the present invention, the genes IGHGl, NKG7, IL2RG, IL7R, CCL18, PVRIG, PLAC8, CCL5, SIRPG, COROIA, LCK, TRBC1, GZMB, CXCL13, and WARS are overexpressed (upregulated) in the phenotype “inflamed” relative to the other two phenotypes; the genes COL5A1, SPON1, CAMK2N1, FAP, SPOCK1, COL1A1, SCGB2A1, AKR1C2, CPE, SCGB2A2, TCN1, TPSABl, and MMP2 are overexpressed (upregulated) in the phenotype “ignored” relative to the other two phenotypes, and the genes PERP, THBS2, ASPN, COL10A1, TUFT1, GREM1, CEACAM6, ENTPD3, IGFBP5, PBX1, CXADR, GPRC5A, SDCl and CALML5 are overexpressed (upregulated) in the phenotype “excluded” relative to the other two phenotypes. The relative gene expression level of a set of minimally 3 genes comprising at least 1 gene that is characteristic for each phenotype is preferably compared or correlated to the relative gene expression levels of the same set of genes from reference samples (Table 1). The above is applicable for typing test samples and distinguishing between the three immune phenotypes. One of skill will understand that when performing a typing method to distinguish between inflamed and non-inflamed, the relative gene expression of a set of minimally 2 genes comprising at least 1 gene that is characteristic for each phenotype is preferably compared or correlated to the relative gene expression levels of the same set of genes from reference samples. Hence, the reference sample, preferably comprises a set of reference samples, and, depending on the level of distinction required, said set preferably includes at least 1 sample of each of the 2 different immune phenotypes inflamed and non-inflamed, or said set preferably includes at least one sample of each of the 3 different immune phenotypes inflamed, excluded, and ignored.

The gene classifier, in one embodiment, allows discrimination between the spatial immune phenotypes “inflamed” and “non-inflamed” (wherein “non-inflamed” is either excluded or ignored). In addition, the gene classifier allows further discrimination of the spatial immune phenotype “non-inflamed” into the spatial immune phenotypes “excluded” and “ignored”. The gene classifier, in one preferred embodiment, allows discrimination between the spatial immune phenotypes “inflamed” and “excluded” and “ignored” of a specified cancer. The analytical steps to distinguish between the various spatial immune phenotypes is herein also referred to as “typing”, which comprises differential gene expression analysis between a test tumor sample and a reference sample (preferably a set of reference samples as indicated above) and ranking the gene expression data on the basis of the gene classifier to thereby identify the spatial immune phenotype that matches the reference phenotype, preferably using the gene classifier as displayed in Table 1, and preferably using microarray gene expression analysis or RNAseq. The term "typing" as used herein includes any method of analysing the gene expression level of one or more nucleic acid molecules to be analysed (e.g. the "test" or target nucleic acid). Preferably, typing in aspects of the invention includes methods for analysing gene expression in a test tissue relative to reference tissues or a reference dataset. Methods of the invention thus include methods of determining the gene expression of a test sample and comparing the expression data with reference data. Analysing the gene expression level in aspects of the present invention involves determining the expression level of multiple genes from the multi-gene classifier as described herein, and in particular as indicated in Table 1. The term "multiple" as used herein means 2 or more (such as 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42 or more). In preferred embodiments of aspects of the present invention, typing may include the analysis of 1 or more (such as 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, or 15) genes upregulated in the phenotype “inflamed”, together with the analysis of 1 or more (such as 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27) genes upregulated in the phenotypes “excluded” and “ignored” (together also indicated as “non-inflamed”) as described herein. In highly preferred embodiments of aspects of the present invention, typing may comprise the analysis of 1 or more (such as 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, or 15) genes upregulated in the phenotype “inflamed”, further comprising the analysis of 1 or more (such as 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, or 14) genes upregulated in the phenotype “excluded”, and further comprising the analysis of 1 or more (such as 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, or 13) genes upregulated in the phenotype “ignored”.

Typing in methods of the present invention is based on determining the presence of differential gene expression in a solid tumor sample by measuring the quantity of a gene product (RNA or protein, preferably mRNA) for at least two, preferably at least three genes of the multi-gene classifier described herein, preferably at least one gene that is upregulated in each phenotype. The terms “differentially expressed gene”, “differential gene expression” and their synonyms, which are used interchangeably, refer to a gene whose expression is activated to a higher or lower level in a tumor sample of a subject suffering from cancer, such as triple negative breast cancer, relative to a reference or control sample or a set of reference or control values, preferably the average gene expression of each individual classifier gene for a set of reference samples. For the purpose of this invention, “differential gene expression” is considered to be present when there is at least an about 1.1, 1.2, 1.3, 1.4, 1.5, 1.6, 1.7, 1.8, 1.9 or 2.0-fold, preferably at least about four-fold, more preferably at least about six-fold, most preferably at least about ten-fold difference between the expression of a given gene in a test sample and the expression in the reference.

Table 1 includes both gene names and/or reference accession numbers, such as ENTREZ ID (GenelDs), and Affymetrix Probeset IDs. These identifiers may be used to retrieve publicly- available annotated DNA, mRNA or protein sequences from sources such as the NCBI website, which may be found at the following uniform resource locator (URL): http://www.ncbi.nlm.nih.gov. The identifiers given in the tables herein pertain to NCBI-GenBank Flat File Release 244.0 incorporating data processed by the INSDC databases as of Tuesday June 22, 2021. Gene identifiers indicated herein include reference to those gene sequences in their entirety. The information associated with these identifiers, including reference sequences and their associated annotations, are all incorporated by reference.

The test sample used in methods of this invention is a tumor sample, preferably a whole tissue section or biopsy of a solid tumor. As used herein, the term “biopsy” refers to a sample of tissue (e.g. tumor tissue) that is removed from a subject for the purpose of determining, for example, if the sample contains cancerous tissue or for use in analysis using methods of the present invention. The test sample used in methods of this invention is preferably a tumor sample comprising the TME, preferably wherein the sample includes tumor cells, resident stromal cells, such as fibroblasts, myofibroblasts, and neuroendocrine cells, and immune cells. The sampling procedure therefore preferably allows for the provision of, or provides, a mixed cell sample or heterogeneous cell sample of the tumor tissue. In one preferred embodiment, a sohd tumor sample is a needle biopsy sample.

In aspects of this invention, the test nucleic acid is preferably RNA (e.g. mRNA) derived or isolated from the test sample. If desired, cDNA may be generated using the RNA as template of a reverse transcriptase reaction. The quantity of RNA may inter alia be determined by RNAseq or microarray analysis.

The term “tumor micro-environment”, as used herein, abbreviated as TME, refers to the status of interaction between tumor cells, resident stromal cells, and immune cells, in particular the T cell, preferably CD8+ T cell, presence and spatial distribution in the tumor and/or its surroundings. The spatial immune phenotype of a solid tumor is a feature of the TME and represents the distinct T cell evasive mechanisms active in that environment.

The term “immunotherapeutic agent”, as used herein, refers to any compound, biologic or cell used for the treatment of disease by activating or suppressing the immune system. Immunotherapeutic agents include immunomodulators (such as thalidomide, lenalidomide, pomalidomide and/or imiquimod; cytokines like IFN-a; and other peptides) and immune checkpoint inhibitors (ICIs, such as anti-PDl antibodies: e.g. nivolumab, pembrolizumab, and cemiplimab; anti-PD-Ll antibodies: e.g. atezolizumab, darmatizumab; anti-CTLA4 antibodies: e.g. ipilumimab; anti-LAG3 antibodies: e.g. relatlimab; as well as stimulatory antibodies directed against CD40 or other costimulatory receptors, such as APX005111), vaccines, anti-cancer monoclonal antibodies for targeted therapy (such as Herceptin and bevacizumab), chimeric antigen receptor (CAR) T cells and TCR-T cells, as well as oncolytic viruses. Chemotherapeutic agents include proteasome inhibitors such as carfilzomib, oprozomib, bortezomib, and ixazomib; tyrosine kinase inhibitors such as imatinib, lapatinib, acalabrutinib, afatinib, alectinib, avapritinib, axitinib, bosutinib, cabozantinib, crizotinib, dacomitinib, dasatinib, entrectinib, erlotinib, gilteritinib, ibrutinib, midostaurin, neratinib, nilotinib, pacritinib, pazopanib, pexidartinib, ponatinib, quizartinib, regorafenib, midostaurine, sorafenib, dasatinib, sunitinib, vandetanib, afhbercept, zanubrutinib and ziv-aflibercept; anthracyclines such as a doxorubicin, daunorubicin, idarubicin, mitoxantrone, valrubicin, epirubicin, pirarubicin, rubidomycin, carcinomycin and N-acetyl adriamycin; alkylating agents such as busulfan, cyclophosphamide, bendamustine, carboplatin, chlorambucil, cyclophosphamide, cisplatin, temozolomide, melphalan, bendamustine, carmustine, lomustine, lomustine, dacarbazine, oxaliplatin, melphalan, lomustine, ifosfamide, mechlorethamine, thiotepa, trabectedin and streptozocin; camptothecin such as topotecan, irinotecan, silatecan, cositecan, exatecan, lurtotecan, gimatecan, belotecan and rubitecan; an anti-metabobte such as gemcitabine; taxanes such as paclitaxel and docetaxel; anti-cancer agents (mitomycin C; plant-derived alkaloids including vincristine, vinblastine, vinorelbine, vinflunine, vinpocetine, vindesine, ellipticine and 6-3-aminopropyl-ellipticine; 2- diethylaminoethyl-ellipticinium; datelliptium; and orretelliptine).

The term “therapeutically effective amount”, as used herein, includes reference to a quantity of a specified agent sufficient to achieve a desired effect in a subject being treated with that agent. Ideally, a therapeutically effective amount of an agent is an amount sufficient to inhibit or treat the disease or condition without causing a substantial cytotoxic effect in the subject. The therapeutically effective amount of an agent will be dependent on the subject being treated, the severity of the affliction, and the manner of administration of the therapeutic agent. It is within the knowledge and capabilities of the skilled practitioner to determine therapeutically effective dosing regimens.

The term “administering”, as used herein, refers to the physical introduction of an agent or therapeutic compound to subject as disclosed herein, using any of the various methods and delivery systems known to those skilled in the art. The skilled person is aware of suitable methods for administration and dosage forms. Preferred route of administration for protein-based agents such as antibodies is by parenteral administration, including intravenous, intramuscular, subcutaneous, intraperitoneal, spinal or other parenteral routes of administration, executed inter aha by injection or infusion in the form of a solution. Administering can be performed, for example, once, a plurality of times, and/or over one or more extended periods of time.

The term “level”, when used in the context of measuring and comparing a gene expression level in a sample can refer to absolute or to relative quantification. Absolute quantification may be accomplished by inclusion of known concentration(s) of one or more control analytes and referencing the detected level of the target nucleic acid with the known control analytes (e.g., through generation of a standard curve). Alternatively, relative quantification can be accomplished by comparison of detected levels or amounts between two or more different target nucleic acids to provide a relative quantification of each of the two or more different nucleic acids, e.g., relative to each other. In addition, a relative quantitation may be ascertained using a control, or reference, value (or profile) of gene expression levels obtained from a control sample or in the form of a gene classifier.

The term “typing”, as used herein, refers to differentiating between, or stratification of, subjects according to their TME status, more specifically spatial immune phenotype status. The typing is based on a comparison of (i) the measured gene expression level of at least one gene selected from each group as listed in Table 1 with (ii) a reference gene expression level for said at least one gene selected from each group as listed in Table 1. In particular, the typing can be based on comparison of the measured gene expression levels with reference gene expression levels provided in the form of a classifier such as a trained algorithm designed to distinguish said spatial immune phenotypes on the basis of gene expression levels of said at least one gene of each group as listed in Table 1. The reference, as disclosed herein, is preferably composed of the expression level value in sohd tumor tissue samples with known spatial organization of CD8+ T cells. More preferably, the reference is the average gene expression of at least a subset of the genes in the classifier set forth in Table 1, such as the average expression value for 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15,

16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, or 42 genes of the gene classifier set forth herein. The reference can be provided in the form of a gene classifier that is a trained algorithm.

The term “subject”, as used herein, can be used interchangeably with the term “patient” or “individual”, and includes reference to a mammal, preferably a human, who is suffering from a solid tumor that comprises a TME. Where reference is made to subjects having a sohd tumor that comprises a TME, also included is reference to subjects in which said tumor is completely or partially resected and wherein said solid tumor may recur or metastasize or wherein said solid tumor has already recurred or metastasized. In other words, “having a sohd tumor that comprises a TME” does not exclude situations in which the solid tumor has already been resected, for instance in order to perform a method for typing of the invention.

The term “combination”, as used herein in relation to combination therapies, includes reference to administration of at least two therapeutic agents together at the same time (either in the same pharmaceutical composition or in separate compositions), separately of each other at the same time or separately of each other staggered in time. Simultaneous, separate and sequential administration of the therapeutic agents disclosed herein is for instance envisaged.

The term “WNT inhibitor”, as used herein, includes reference to agents that inhibit the Wnt/6-catenin signaling pathway in cancer. WNT inhibitors include, but are not limited to, IWR-1, IWP-2, Pyrvinium pamoate, Salinomycin (Procoxacin), CWP232228, LP-922056, Teplinovivint, Wnt-C59 (C59), NCB-0846, Salinomycin sodium salt, Adavivint, FH535, Wogonin, CCT251545, iCRT3, PNU-74654, Echinacoside, ETC- 159, KYA1797K, EMT inhibitor- 1, Prinaberel (ERB-041), MSAB, FIDAS-5, Triptonide (NSC 165677), IWP-4, KY-05009, SKI II, IQ 1, Gigantol, Ginkgetin, iCRT 14 , SSTC3, Prodigiosin (Prodigiosine), Specnuezhenide, Specnuezhenide ((8E)-Nuezhenide), KY02111, Hematein, DK419, SGC- AAKl-1, Prodigiosin hydrochloride, FIDAS-3.

The term “CSF1R inhibitor”, as used herein, includes reference to agents that inhibit colony stimulating factor 1 receptor signaling. CSF 1R inhibitors include, but are not limited to, CSFlR-IN-1, CSFlR-IN-2 (compound 5), AZD7507, ARRY-382, c-Fms-IN-8 (compound 4a), BLZ945, c- Fms-IN-10, PRN1371, OSI-930, Pexidartinib hydrochloride (PLX-3397 hydrochloride), Chiauranib (CS2164), Pexidartinib (PLX-3397), AC710, AC710 Mesylate, Sulfatinib (HMPL-012), AZ304, and CHMFL-ABL/KIT-155 (compound 34).

The term “TGFβ inhibitor”, as used herein, includes reference to agents that inhibit transforming growth factor-beta signaling. TGFβ inhibitors include, but are not limited to, Galunisertib (LY2157299), Asiaticoside, BIO-013077-01, BIBF0775, EMT inhibitor-1, SJ000291942, LY2 109761, BMS453 (BMS- 189453), SB-431542, LSKL Inhibitor of Thrombospondin (TSP-1), A 77-01, GW788388, EW-7195, Disitertide (P144), 10,11-Dehydrocurvularin, A 83-01 sodium salt, LDN-212854, LY3200882, Oxymatrine, Isoviolanthin, SB-505124, Pirfenidone (AMR69), Halofuginone (RU-19110), SB-505124 hydrochloride, A 83-01, Halofuginone (RU-19110), TAK1/MAP4K2 inhibitor 1, Demethylzeylasteral, Pirfenidone D5 (AMR69 D5), NCGC00378430, KY-05009, PD-161570 and bintrafusp alfa (M7824).

The term “VEGF inhibitor”, as used herein, includes reference to agents that inhibit vascular endothelial growth factor signahng. VEGF inhibitors include, but are not limited to, AEE 788, AG 879, AP 24534, Axitinib, BMS 605541, DMH4, GSK 1363089, Ki 8751, Nintedanib, RAF 265, Sorafenib, SU 4312, SU 5402, SU 5416, SU 6668, Sunitinib malate, Toceranib, Vatalanib succinate, XL 184, ZM 306416 hydrochloride, and ZM 323881 hydrochloride.

The term “inhibitor of M2 macrophages” includes such therapeutic compounds as Cyclooxygenase-2 (COX-2) inhibitor, including etodolac, as well as all-trans retinoic acid (ATRA) and MEL-dKLA (melittin- d(KLAKLAK)2 hybrid peptide).

The term “epigenetic drug”, as used herein, includes reference to agents that release epigenetic brakes of expression of genes and consequently enhance innate immune pathways, such as type I IFN pathway and chemo-attractant pathways. Epigenetic drugs (or modifiers) include, but are not limited to, inhibitors of DNA methyl transferases (DNMTi), such as 5-azacytidine and 5-aza-2-deoxycytidine; inhibitors of histone deacetylase (HDACi), such as, valproate, FK-228, SAHA, PDX-101; as well as inhibitors of histone methyl transferases (HMTi), e.g. inhibitors of EZH2, such as Tazemetostat.

Description of the preferred embodiments

The present invention provides methods for typing the TME of a solid tumor. Methods of the present invention can be used to identify three main spatial immune phenotypes in TNBC as well as other cancer types through molecular analysis of gene expression. The present invention provides methods for distinguishing or detecting the spatial immune phenotypes inflamed, ignored, and excluded, of a solid tumor. Due to the association of these phenotypes with clinical outcome in cancer, the present invention provides methods for predicting the prognosis of a cancer patient, as well as methods for predicting the response of a cancer patient to ICI therapy. Compositions and kits useful in carrying out such methods are also provided.

Methods for typing the tumor micro-environment (TME) of a solid tumor The present disclosure provides methods for patient diagnostics and stratification based on diagnosis of tumor-immune interactions.

The present invention provides methods for typing, which allow for determining the spatial immune phenotype of a solid tumor tissue, where the methods generally involve the steps of determining an expression level of a product of a gene set forth in Table 1 in a sohd tumor tissue sample obtained from a patient, generating an expression level value, classifying the solid tumor tissue sample as a solid tumor with (i) an inflamed, (ii) an ignored, or (iii) an excluded spatial immune phenotype by comparing the expression level values to a reference. The reference can be composed of the expression level values of the different phenotypes either according to samples with known spatial organization of CD8+ T cells or according to an established training set of samples. Preferably, the reference is the average gene expression of at least a subset of the genes in the classifier set forth in Table 1, such as the average expression values for at least 2, 3, 4, 5, 6, 7, 8,

9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, or 42 genes of the gene classifier set forth herein. As also indicated herein above, the reference may be composed of the expression level value of at least one gene of the phenotype inflamed and at least one gene of the phenotype non-inflamed when typing a test sample as belonging to either one of these phenotypes, or the reference may be composed of the expression level value of at least one gene of the phenotype inflamed, at least one gene of the phenotype ignored, and least one gene of the phenotype excluded when typing a test sample as belonging to either one of these three phenotypes. The expression values of an unknown sample are preferably correlated/compared to the expression values of a reference samples of each corresponding phenotype.

Methods of the present invention now enable to predict the prognosis of a subject with a solid tumor or to predict a patient’s response to therapeutic intervention such as ICI therapy.

The present methods for typing contemplate determining the gene expression value of at least one gene from each spatial immune phenotype group as listed in Table 1, of which the expression is upregulated.

In a preferred embodiment, genes of which the differential expression correlate to the phenotype “inflamed” are, in order of descending rank, selected from IGHG1, PBX1, WARS, IL7R, CCL5, COROlA, CXCL13, LCK, CCL18, TRBC1, PLAC8, GZMB, NKG7, IL2RG, SIRPG, and PVRIG.

In preferred embodiments of this invention, genes of which the differential expression correlate to the phenotype “ignored” are, in order of descending rank, selected from COL1A1, COL5A1, MMP2, FAP, SPON1, CPE, SPOCK1, CAMK2N1, AKR1C2, SCGB2A2, TPSABl, TCN1 and SCGB2A1.

In preferred embodiments of this invention, genes of which the differential expressions correlate to the phenotype “excluded” are, in order of descending rank, selected from IGFBP5, THBS2, SDCl, PBX1, PERP, CXADR, COL10A1, GPRC5A, CALML5, ASPN, TUFT1, CEACAM6, and ENTPD3.

One of skill will understand that the accuracy of assigning the correct spatial immune phenotype based on the gene expression classifier as identified herein will increase with the number of genes expression products assayed. Based on the teaching herein above, one of skill in the art will be able to decide if addition of more genes from the 42 classifier genes as indicated herein is needed in order to improve the accuracy of the assay for determining the spatial immune phenotype of a solid tumor.

Preferably, in a method for typing of the invention, the gene expression level is measured for at least 5, at least 6, at least 7, or at least 8 genes of each of spatial immune phenotype groups 1, 2 and 3. Figure 15 shows that an accuracy of more than 60% is achieved when the gene expression level is measured for at least 5 random genes from each of spatial immune phenotype groups 1, 2 and 3 as depicted in Table 1.

In one preferred embodiment, the methods of the present invention comprise the step of measuring the expression level of at least one gene upregulated in each of the phenotypes according to the sub-listing in Table 1 (e.g. at least 1 gene upregulated in inflamed, at least 1 gene upregulated in ignored, and at least 1 gene upregulated in excluded). In preferred embodiments of methods of the invention at least 3 genes are therefore measured, including at least 1 gene upregulated in each phenotype. The term “upregulated in each phenotype” preferably means upregulated relative to the gene expression in the other 2 phenotypes. Preferably, the expression level of at least 6 genes is measured, including at least 2 genes upregulated in each phenotype. More preferably, the expression level of at least 9 genes is measured, including at least 3 genes upregulated in each phenotype. Even more preferably, the expression level of at least 12 genes is measured, including at least 4 genes upregulated in each phenotype. Still more preferably, the expression level of at least 15, 18, 21, 24, 27, 30, 33, 36, 39, 42 genes is measured, including at least 5, 6, 7, 8, 9, 10, 11, 12, 13, 14 genes upregulated in the classifier.

In a highly preferred embodiment, the methods of the present invention comprise the step of measuring the expression level of (i) at least a first gene selected from the group consisting of IGHGl, NKG7, IL2RG, IL7R, CCL18, PVRIG, PLAC8, CCL5, SIRPG, CORO 1 A, LCK, TRBC1, GZMB, CXCL13, and WARS; and

(ii) at least a second gene selected from the group consisting of COL5A1, SPON1, CAMK2N1, FAP, SPOCK1, COL1A1, SCGB2A1, AKR1C2, CPE, SCGB2A2, TCN1, TPSABl, andMMP2; and

(iii) at least a third gene selected from the group consisting of PERP,

THBS2, ASPN, COLlOAl, TUFT1, GREM1, CEACAM6, ENTPD3, IGFBP5, PBX1, CXADR, GPRC5A, SDCl and CALML5.

A difference (e.g., an increase, a decrease) in gene expression can be determined by comparison of the level of expression of one or more genes in a sample from a subject to that of a suitable control or reference. Suitable controls include, for instance, a non-neoplastic tissue sample (e.g. a nonneoplastic tissue sample from the same subject from which the cancer sample has been obtained), a sample of non-cancerous cells, non-metastatic cancer cells, non- mahgnant (benign) cells or the like, or a suitable known or determined reference standard. The reference can be a typical, normal or normalized range of levels, or a particular level, of expression of a protein or RNA (e.g. an expression standard). The standards can comprise, for example, a zero gene expression level, the gene expression level in a standard cell line, the average level of gene expression previously obtained for a population of normal human controls, or the average level of gene expression over the set of expression products measured, such as the average level of all classifier genes measured.

In a most preferred embodiment, assignment of a test sample to one of three immune phenotypes is based on rank-correlations between expressions of classifier genes in the test sample and ranked expressions of classifier genes per spatial phenotype in the classifier, more preferably highest Spearman rank-correlations between the test sample and classifier. One of skill will understand that other methods of establishing rank correlation between expression data of classifier genes in a test sample and the spatial immune phenotype classifier described herein may be accomplished by different statistical methods, including Spearman rank correlation, Weighted Rank Correlation, Kendall rank correlation, Hoeffding’s D measure, Theil-Sen, Rank Theil-Sen, Distance Covariance, Pearson., etc. One of skill will understand that other correlation methods may also be used to assign a test sample to one of three immune phenotypes based on correlation with the classifier as proposed herein, including the use of Distance Covariance; Wilcoxon's, hnear regression or non-linear regression models.

The spatial-phenotype-classifier was not only predictive and apphcable in TNBC, but also in other solid tumor types. Hence, the classifier of the present invention may be used to assess the prognostic and predictive value in a pan-cancer setting. Spatial phenotypes were significantly prognostic not only in invasive breast cancer BRCA (all subtypes, including ER+), but also in bladder cancer (BLCA), skin cutaneous melanoma (SKCM), cervical squamous cell carcinoma and endocervical adenocarcinoma (CESC), head and neck squamous cell carcinoma (HNSC) and kidney cancer (KICH), but despite similar trends not in prostate (PRAD), pancreatic (PAAD), lung (LUAD) or colon cancer (COAD).

Based on the findings disclosed herein, a gene-expression classifier was developed that allows assignment of spatial immune phenotypes based on RNA expressions, thereby enabling assessment of prognostic and predictive values of the spatial immune phenotypes without the need for CD8 ⁺ T cell stainings.

Moreover, the spatial-phenotype-classifier significantly outperformed other, publicly available gene-classifiers that are recognized for capturing lymphocyte activity and location, and for predicting anti-PD 1 response in melanoma, such as IFNg-response, T cell-exclusion and TLS signatures (Figure 8).

Methods for predicting the prognosis of a subject with a solid tumor The present invention provides methods for predicting the prognosis of a subject with a solid tumor, using either a method for typing of the invention or a method for predicting a prognosis of the invention.

The methods generally involve determining an expression level of a gene product of a gene set forth in Table 1 in a solid tumor tissue sample obtained from a patient, generating an expression level value, classifying the solid tumor tissue sample as a solid tumor with (i) an inflamed, (ii) an ignored, or (iii) an excluded spatial immune phenotype by comparing the expression level value to a reference, and predicting the prognosis of said patient on the basis of the spatial immune phenotype of said solid tumor.

These spatial immune phenotypes (assigned on the basis of methods of this invention) were significantly associated with survival (distant metastasis-free survival (MFS), disease-free survival (DFS), and overall survival (OS)). Both the excluded and ignored phenotypes (assigned on the basis of methods of this invention) are significantly associated with poor metastasis-free survival (MFS). The inflamed phenotype is significantly associated with better MFS. The excluded and ignored phenotypes were associated with poor prognosis, whereas the inflamed phenotype is associated with good prognosis (e.g. Figures 5 and 7). Tumors with an inflamed phenotype had the best prognosis (10-year OS: 80%), excluded phenotypes intermediate (10-year OS), and ignored phenotypes the worst prognosis (10-year OS: 40%). Prolonged survival of excluded versus ignored phenotypes was statistically significant for OS.

Methods for predicting a patient’s response to ICI therapy The present disclosure provides methods for predicting a patient’s response to ICI therapy using a method for typing or method for assigning of the invention.

The methods generally involve determining an expression level of a product of a gene set forth in Table 1 in a solid tumor tissue sample obtained from a patient, generating an expression level value, classifying the solid tumor tissue sample as a solid tumor with (i) an inflamed, (ii) an ignored, or (iii) an excluded spatial immune phenotype by comparing the expression level value to a reference, and predicting a patient’s response to ICI therapy on the basis of the spatial immune phenotype of said solid tumor.

The present inventors have found that excluded or ignored phenotypes classified as such by using the methods of the present invention respond poorly to ICI, while the inflamed phenotype responds well to ICI.

Methods for treating a subject with a solid tumor

The present disclosure provides new and combined methods for treating a subject with a solid tumor, which may optionally be resected, recurrent or metastasized.

The methods generally involve determining an expression level of a product of a gene set forth in Table 1 in a solid tumor tissue sample obtained from a patient, generating an expression level value, classifying the solid tumor tissue sample as a solid tumor with (i) an inflamed, (ii) an ignored, or (iii) an excluded spatial immune phenotype by comparing the expression level value to a reference, and administering to said patient a therapeutically effective amount of an immune checkpoint inhibitor and/or other therapies on the basis of the spatial immune phenotype of said solid tumor.

ICI useful in aspects of this invention include, but are not limited to PD-1 inhibitors, PD-L1 inhibitors, CTLA-4 inhibitors, LAG3 inhibitors, as well as stimulatory antibodies directed against CD40 or other costimulatory receptors, or other ICI, and combinations thereof. PD-1 inhibitors useful in aspects of this invention include, but are not limited to, Pembrolizumab, Nivolumab, Cemiplimab, JTX-4014, Spartalizumab, Camrelizumab, Sintilimab, Tislelizumab, Toripalimab, Dostarlimab, INCMGA00012 (MGA012), AMP-224 and AMP-514. PD-L1 inhibitors useful in aspects of this invention include, but are not limited to Atezolizumab, Avelumab, Durvalumab, KN035, CK-301, AUNP12, CA-170, and BMS-986189. CTLA-4 inhibitors useful in aspects of this invention include, but are not limited to Ipilimumab. LAG3 inhibitors useful in aspects of this invention include, but are not limited to, relatlimab. Stimulatory antibodies directed against CD40 or other costimulatory receptors useful in aspects of this invention include, but are not limited to APX005111). Combinations of any of the above are also foreseen.

In particular, ICI therapy (including ICI monotherapy) will be most effective in subjects having a solid tumor with an inflamed spatial immune phenotype. Subjects having a solid tumor with an excluded or ignored spatial immune phenotype are less likely to respond to ICI monotherapy, with TME-mediated T cell evasion (e.g. determined by resident stromal components) being strongest in the excluded phenotype, intermediate in the ignored phenotype, and weakest in the inflamed phenotype. In contrast, the immune-response-mediated T cell evasion (e.g. determined by immune cells, most likely as a consequence of adaptive immune responses) is strongest in the inflamed phenotype, intermediate in the ignored phenotype, and weakest in the excluded phenotype. One of skill in the art will understand that knowledge of the spatial immune phenotype of a solid tumor allows for more tailored therapeutic intervention. For a schematic overview of proposed co-treatments see Figure 13. The findings presented in Example 1, amongst others Figure 10D, E, F and G, list immune determinants per phenotype as well as correlative analyses between immune determinants and actionable targets that provide a rationale for below-mentioned co-treatments.

In the case of an inflamed phenotype, ICI monotherapy is proposed. In case ICI monotherapy is not effective in patients having a tumor with an inflamed phenotype, a combination of multiple ICIs may provide a suitable treatment. Alternatively, or in combination therewith, therapeutic priming prior to ICI treatment using epigenetic drugs (including, but not limited to 5-azacytidine and valproate) and/or inhibitors of M2 macrophages (including, but not limited to CSF1R inhibitors, e.g. pexidartinib) may enhance the type I IFN and chemo-attractant pathways, and counteract adaptive immune responses that have occurred in these types of tumors, respectively, and as such further boost the numbers of intra-tumoral T cells that are prone to ICI-directed re-activation. In the case of an excluded phenotype, therapeutic priming prior to ICI treatment is proposed using inhibitors of TGFβ (including, but not limited to the bifunctional anti-PDL-1 mAb/TGFh trap M7824), and inhibitors of VEGF receptor kinases (including, but not limited to cediranib), which are to enhance migration of CD8+ T cells from the tumor margin towards the tumor center and to enhance the activation of intra-tumoral CD8+ T cells.

In case of the ignored phenotype, therapeutic priming prior to ICI treatment is proposed using blockers of the WNT pathway (including, but not limited to WNT974) and/or drugs that target M2 macrophages (including, but not limited to CSF1R inhibitors, e.g. pexidartinib), which are to enhance infiltration of CD8+ T cells into the tumor, and the activation of intra- tumoral CD8+ T cells. The above mentioned co-treatments are described in more detail, including references for clinical precedents for such cotreatments in Example 1.

It will be understood that treatment regimens in aspects of this invention may include treatment with immunotherapeutic agents in combination with chemotherapeutic agents and/or radiation therapy (including external beam radiation therapy, high dose rate brachytherapy, (targeted-) radionuclide therapy, and hyperthermia), and/or surgery including surgical resection of a tumor. For the purpose of clarity and a concise description, features are described herein as part of the same or separate embodiments, however, it will be appreciated that the disclosure includes embodiments having combinations of all or some of the features described.

The content of the documents referred to herein is incorporated by reference.

Numbered embodiments

Embodiment 1. A method for typing a tumor micro-environment (TME) of a solid tumor, comprising the steps of:

- providing a test sample of the TME of a solid tumor of a subject;

- measuring in said test sample the gene expression level for:

(i) at least one gene selected from group 1 consisting of IGHGl, NKG7, IL2RG, IL7R, CCL18, PVRIG, PLAC8, CCL5, SIRPG, CORO1 A, LCK, TRBC1, GZMB, CXCL13, and WARS; and

(ii) at least one gene selected from group 2 consisting of COL5A1, SPON1, CAMK2N1, FAP, SPOCK1, COL1A1, SCGB2A1, AKR1C2, CPE, SCGB2A2, TCN1, TPSAB1, andMMP2; and/or

(iii) at least one gene selected from group 3 consisting of PERP, THBS2, ASPN, COL10A1, TUFT1, GREM1, CEACAM6, ENTPD3, IGFBP5,

PBX1, CXADR, GPRC5A, SDCl and CALML5;

- comparing the measured test sample gene expression levels to a reference, and

- typing the TME of said solid tumor of said subject as being T cell -inflamed or non-T cell-inflamed on the basis of the comparison of said measured gene expression level and said reference.

Embodiment 2. The method according to embodiment 1, wherein said method comprises measuring the gene expression level for:

(i) at least one gene selected from group 1 consisting of IGHGl, NKG7, IL2RG, IL7R, CCL18, PVRIG, PLAC8, CCL5, SIRPG, CORO 1 A, LCK, TRBC1, GZMB, CXCL13, and WARS; and

(ii) at least one gene selected from group 2 consisting of COL5A1, SPON1, CAMK2N1, FAP, SPOCK1, COL1A1, SCGB2A1, AKR1C2, CPE, SCGB2A2, TCN1, TPSAB1, andMMP2; and (iii) at least one gene selected from group 3 consisting of PERP, THBS2, ASPN, COL10A1, TUFT1, GREM1, CEACAM6, ENTPD3, IGFBP5,

PBX1, CXADR, GPRC5A, SDCl and CALML5.

Embodiment 3. The method according to embodiment 2, wherein said method is a method for typing a TME of a solid tumor as being of the immune phenotype T cell-inflamed, T cell-excluded or T cell-ignored; wherein said reference comprises the gene expression level of said at least one gene in at least one reference sample of each of said three immune phenotypes, and wherein said TME of said solid tumor is typed as being T cell-inflamed, T cell-excluded or T cell-ignored on the basis of the comparison of said measured gene expression level and said reference.

Embodiment 4. The method according to any one of the preceding embodiments, wherein said gene expression level is measured for at least 2 genes selected from each of group 1, group 2, and/or group 3.

Embodiment 5. The method according to any one of the preceding embodiments, wherein said gene expression level is measured for at least 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14 or 15 genes selected from group 1, and at least 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, or 13, genes selected from group 2, and at least 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, or 14 genes selected from group 3.

Embodiment 6. The method according to any one of the preceding embodiments, wherein said gene expression level is measured for IGHGl, NKG7, IL2RG, IL7R, CCL18, PVRIG, PLAC8, SIRPG, CORO 1 A, LCK, TRBC1, GZMB, CCL5, CXCL13, WARS, COL5A1, SPON1, CAMK2N1, FAP, SPOCK1, COL1A1, SCGB2A1, AKR1C2, CPE, SCGB2A2, TCN1, TPSABl, MMP2, PERP, THBS2, ASPN, COL10A1, TUFT1, GREM1, CEACAM6, ENTPD3, IGFBP5, PBX1, CXADR, GPRC5A, SDCl and CALML5.

Embodiment 7. The method according to any one of the preceding embodiments, wherein the solid tumor is selected from the group formed by BRCA: breast carcinoma such as triple negative breast cancer (TNBC); CESC: cervical squamous cell carcinoma and endocervical adenocarcinoma; HNSCC: head and neck squamous cell carcinoma; KICH: chromophobe renal cell carcinoma; BLCA: bladder urothelial carcinoma, and SKCM: skin cutaneous melanoma.

Embodiment 8. The method according to embodiment 7, wherein the solid tumor is a triple negative breast cancer (TNBC).

Embodiment 9. The method according to embodiment 7 or embodiment 8, wherein the solid tumor is a primary tumor, recurrent tumor or a secondary (metastasized) tumor.

Embodiment 10. The method according to any one of the preceding embodiments, wherein measuring the gene expression level is performed by qPCR, microarray analysis or next- generation sequencing (NGS).

Embodiment 11. The method according to any one of the previous embodiments, wherein said method for typing is a method for predicting the prognosis of a subject with a solid tumor, and wherein, when said TME of a sohd tumor from said subject is typed as T cell-inflamed said subject has a favorable prognosis, and wherein, when said TME of a solid tumor from said subject is typed as non-T cell-inflamed, preferably typed as being T cell- excluded or T cell-ignored, said subject has an unfavorable prognosis. Embodiment 12. A method for assigning a subject having a solid tumor comprising a TME to an immunotherapy-responsive or immunotherapy- unresponsive group, such as ICI, said method comprising the steps of:

- performing a method for typing according to any one of embodiments 1-11;

- assigning said subject to the immunotherapy-responsive group when said TME of a solid tumor from said subject is typed as T cell-inflamed or assigning said subject to the immunotherapy-unresponsive group when said TME of a solid tumor from said subject is typed as non-T cell-inflamed, preferably either T cell-excluded or ignored.

Embodiment 13. An immunotherapeutic agent, such as ICI, for use in the treatment of a subject having a solid tumor comprising a TME, wherein said subject is assigned to the immunotherapy-responsive group according to a method for assigning according to embodiment 12.

Embodiment 14. An immunotherapeutic agent for use in the treatment of a subject having a solid tumor comprising a TME, wherein said subject is assigned to the immunotherapy-unresponsive group according to a method for assigning according to embodiment 12 and wherein said inflam oefd said solid tumor is typed as being T cell-ignored; and wherein said immunotherapeutic agent is for administration in combination with a WNT inhibitor or with an inhibitor of M2 macrophages, such as colony stimulating factor 1 receptor (CSF 1) inhibitor.

Embodiment 15. An immunotherapeutic agent for use in the treatment of a subject having a solid tumor comprising a TME, wherein said subject is assigned to the immunotherapy-unresponsive group according to a method for assigning according to embodiment 12 and wherein said inflam oefd said solid tumor is typed as being T cell-excluded; and wherein said immunotherapeutic agent is for administration in combination with a TGFβ inhibitor and/or VEGF inhibitor.

Embodiment 16. The immunotherapeutic agent for use according to any one of embodiments 13-15, wherein said immunotherapeutic agent is an immune checkpoint inhibitor (ICI).

Embodiment 17. A method for predicting a prognosis for a subject with a solid tumor comprising a TME, said method comprising the steps of: - measuring in a sample of a solid tumor comprising a TME from a subject a gene expression level for:

(i) at least one gene selected from group 1 consisting of IGHGl, NKG7, IL2RG, IL7R, CCL18, PVRIG, PLAC8, CCL5, SIRPG, CORO 1 A, LCK, TRBC1, GZMB, CXCL13, and WARS; and (ii) at least one gene selected from group 2 consisting of COL5A1,

SPON1, CAMK2N1, FAP, SPOCK1, COL1A1, SCGB2A1, AKR1C2, CPE, SCGB2A2, TCN1, TPSABl, andMMP2; and (iii) at least one gene selected from group 3 consisting of PERP, THBS2, ASPN, COLlOAl, TUFT1, GREM1, CEACAM6, ENTPD3, IGFBP5, PBX1, CXADR, GPRC5A, SDCl and CALML5;

- comparing said measured gene expression level to a reference; and

- typing said subject as having a favorable or unfavorable prognosis on the basis of the comparison of said measured gene expression level and said reference. Table 1. Gene classifier

Spatial

Immune ENTREZ

Phenotype SYMBOL ID ^a Affy_ID GENE NAME infl_rank ign_rank excl_rank ^b Inflamed IGHG1 3500 204719_at immunoglobulin heavy constant gamma 1 (Glm marker) 2 8 2 (group 1) NKG7 4818 209555_s_at natural killer cell granule protein 7 32 37 33

IL2RG 3561 209686_at interleukin 2 receptor subunit gamma 33 39 36

IL7R 3575 210839_s_at interleukin 7 receptor 11 17 20

CCL18 6362 209160_at C-C motif chemokine ligand 18 26 31 29

PVRIG 79037 214091_s_at PVR related immunoglobulin domain containing 41 42 41

PLAC8 51316 219564_at placenta specific 8 29 33 32

CCL5 6352 201525_at C-C motif chemokine ligand 5 13 25 24

SIRPG 55423 205883_at signal regulatory protein gamma 34 40 39

CORO 1A 11151 213831_at coronin 1A 14 23 23

LCK 3932 205382_s_at LCK proto-oncogene, Src family tyrosine kinase 25 29 27

TRBC1 28639 202436_s_at T cell receptor beta constant 1 27 35 30

GZMB 3002 209541_at granzyme B 31 38 35

CXCL13 10563 20388 l_s_at C-X-C motif chemokine ligand 13 16 27 26

WARS 7453 205828_at tryptophanyl-tRNA synthetase 9 13 12

Ignored COL5A1 1289 203355_s_at collagen type V alpha 1 chain 8 2 5 (group 2) SPON1 10418 208937_s_at spondin 1 23 18 22 CAMK2N1 55450 220133_at calcium/calmodulin dependent protein kinase II inhibitor 1 30 24 25 FAP 2191 209047_at fibroblast activation protein alpha 22 16 17

SPARC (osteonectin), cwcv and kazal like domains

SPOCK1 6695 205959_at proteoglycan 1 28 22 21 COL1A1 1277 211653_x_at collagen type I alpha 1 chain 1 1 1

SCGB2A1 4246 214218_s_at secretoglobin famdy 2A member 1 42 41 42

AKR1C2 1646 216594_x_at aldo-keto reductase famdy 1 member C2 35 26 34

CPE 1363 206953_s_at carboxypeptidase E 21 19 18

SCGB2A2 4250 201348_at secretoglobin family 2A member 2 38 28 38

TCN1 6947 218730_s_at transcobalamin 1 40 36 40

TPSABl 7177 205916_at tryptase alpha/beta 1 37 30 37

MMP2 4313 209392_at matrix metallopeptidase 2 10 7 9

Excluded PERP 64065 58916_at PERP, TP53 apoptosis effector 6 6 8 (group 3) THBS2 7058 58916_at thrombospondin 2 3 3 4 ASPN 54829 221728_x_at asporin 20 15 16 COL10A1 1300 202437_s_at collagen type X alpha 1 chain 15 12 11 TUFT1 7286 204469_at tuftelin 1 19 20 19 GREM1 26585 218736_s_at gremlin 1, DAN family BMP antagonist 17 14 15 ^ CEACAM6 4680 209613_s_at carcinoembryonic antigen related cell adhesion molecule 6 39 32 28 ENTPD3 956 205979_at ectonucleoside triphosphate diphosphohydrolase 3 36 34 31 IGFBP5 3488 209612_s_at insulin like growth factor binding protein 5 7 5 3 PBX1 5087 207803_s_at PBX homeobox 1 4 9 7 CXADR 1525 212865_s_at CXADR, Igdike cell adhesion molecule 12 10 10 GPRC5A 9052 206898_at G protein-coupled receptor class C group 5 member A 18 11 13 SDC1 6382 20415 l_x_at syndecan 1 5 4 6 CALML5 51806 203980_at calmodulin like 5 24 21 14 ^a Probeset id of the Affymetrix Human Genome U133 Plus 2.0 Array (Thermo Fisher Scientific Inc., Waltham, USA), used as described in the Examples. ^b Inflamed genes are upregulated (overexpressed) in inflamed phenotype relative to their level of expression in the other phenotypes;

ignored genes are upregulated (overexpressed) in ignored phenotype relative to their level of expression in the other phenotypes; excluded genes are upregulated (overexpressed) in excluded phenotype relative to their level of expression in the other phenotypes.

00

EXAMPLES

Example 1. Spatial immune phenotypes predict response to anti- PD1 treatment and capture distinct paths of T cell evasion in solid tumors such as triple negative breast cancer.

Materials and Methods

Cohorts of patients

Cohort A: Node-negative, primary TNBC from patients who did not receive adjuvant treatment. FFPE resection materials were used for: whole tissue stainings for CD8 stainings (n=228); stainings for multiple immune cells/molecules on consecutive sections (n=30); multiplexed stainings for immune effector cells (n=64) and cells/molecules related to spatial phenotypes (n=69); microarray gene expression analysis (n=101, Cohort Al); as well as RNAseq data analysis (n=43, Cohort A2); Complete clinicopathological records were available with >10 year follow up (n=122). Cohort B: Node-negative, primary breast cancer (BC) from patients who did not receive adjuvant treatment (n=867 of which n=196 basal -like BC), with microarray data retrieved from gene expression omnibus GSE2034, GSE5327, GSE11121, GSE2990 and GSE7390. Details of combined cohort have been described previously (Hammerl et al., Clin Cancer Res., doi: 10.1158/1078-0432. CCR- 19-0285 (2019)).

Cohort C: Node-negative, primary BC from patients who did not receive treatment, with RNAseq and WGS data (n=347 of which n=66 TNBC) (Nik- Zainal et al., Nature, 534(7605):47-54 (2016)).

Cohort D: Metastatic TNBC from patients treated with anti-PDl antibody in the TONIC-trial (n=53, of which n=44 paired samples) (Voorwerk et al., Nat Med., doi: 10.1038/s41591-019-0432-4 (2019)), with processed transcriptome data of pre- and post-induction treatment biopsies retrieved via controlled access (available through The European Genome-phenome Archive (EGA) EGAS00001003535). Stromal TILs were scored independently, according to an accepted international standard from the International Immuno-Oncology Biomarker Working Group (available via the World Wide Web: http://www.tilsinbreastcancer.org for all guidelines on TIL assessment in solid tumors). PD-L1 stainings (PD-L1 IHC 22C3 pharmDx assay (Agilent Dako)) were assessed and the percentage of positive tumor-infiltrating immune cells was scored.

Cohort E: TCGA data (Chang et al., Nat Genet., 45(10): 1113- 1120 (2013)) as well as sample annotation data of TNBC were retrieved from the USCS xena browser (n=5194 of which 1284 BC of which in turn 137 TNBC). Transcrip tome data of anti-PDl pre-treatment biopsies from melanoma patients (n=28) or treated with anti-PDl antibody (n=65) were retrieved from GSE78220 (Hugo et al., Cell, 165(l):35-44 (2016)) and GSE91061 (Riaz et al., Cell, 171(4):934-949.el5 (2017)).

Cohort F: Metastatic TNBC with whole tissue CD8 stainings and RNAseq (n=12 lymph node macro metastases).

Table 2 and Figure 1 show clinical details and application of these cohorts.

Table 2. Overview of study cohorts. Clinical features and access of the different cohorts. Characteristics for TNBC patients. Abbreviations: LNO: lymph-node negative; LN+: lymph-node positive; NA: not available; BRCA: breast cancer;

5 BLCA: bladder cancer; SKCM: skin-cutaneous melanoma; LUAD: lung adenocarcinoma; HNSC: head and neck squamous-cell carcinoma; PRAD: prostate adenocarcinoma; PAAD: pancreatic adenocarcinoma; COAD: colorectal adenocarcinoma; CESC: cervical squamous cell carcinoma and endocervical adenocarcinoma; KICH: chromophobe renal cell carcinoma.

10 Tissue stainings and image analysis Immunohistochemistrv (IHC)

IHC was performed on TNBC whole tissue sections (FFPE) comprising different histological subtypes, which were assigned by pathologists. IHC stainings were performed following heat-induced antigen retrieval for 20 min at 95°C. After cooling to RT, staining was visualized by the anti-mouse EnVision+® System-HRP (DAB) (DakoCytomation). The following primary antibodies were used: CD8 (C8/C144B, Sanio, 1:100, pH 9); CD3 (PSl, Sigma, 1:25, pH 6); CD4 (4B12, DAKO, 1:80, pH 9), CD 137 (BBK-2, Santa Cruz, 1:80, pH 6), CD278 (SP98, Thermo Fisher, 1:50, pH 9), CD66b (80H3, BIO-RAD, 1:100, pH 9), MECA-79 (Cl 11-6, Santa Cruz, 1:50, pH 9), and MHC-II (LN3, Thermo Fisher, 1:50, pH 9).

Multiplexed immunofluorescence (IF)

Multiplexed IF was performed using OPAL reagents (Akoya Biosciences) on whole shdes (using a randomly selected subset of cohort A with comparable fractions of all spatial phenotypes). In brief, stainings included multiple cycles of: antigen retrieval (15 min boiling in antigen retrieval buffer, pH 6 or pH 9 depending on primary antibodies) followed by cooling, blocking, and consecutive staining with primary antibodies, HRP -polymer and Opal fluorophores; cycles were repeated until all markers were stained. Finally, nuclei were stained with DAPI.

Immune effector panel (number indicates position of primary antibody):

1. CD56 (MRQ-42, Sanbio, 1:500) - OPAL620; 2. CD3 (SP7, Sigma, 1:350) - OPAL520; 3. CD20 (L26, Sanbio, 1:1000) - OPAL650; 4. CD8 (C8/144b, Sanbio, 1:250) - OPAL570; 5. CD68 (KP-1, Sanbio, 1:250) - OPAL540; 6. Cytokeratin-Pan (AE1/AE3, Thermofisher, 1:200) - OPAL690; 7. DAPI. Spatial phenotype panel (number indicates position of primary antibody):

1. CLEC9A (sheep polyclonal*, R&D Systems, 1:600) - OPAL570; 2. S100A7 (47C1068, Biotechne, 1:1000) - OPAL650; 3. CD lib (EP1345Y, Abeam, 1:200) - OPAL690; 4. CD8 (C8/144b, Sanbio, 1:250) - OPAL540; 5. CD 163

(MRQ26, Cell Marque, 1:50) - OPAL520, 6. COLlOAl (X53, Life Technologies, 1:50) - OPAL620; 7. Cytokeratin-Pan (AE1/AE3, Thermofisher, 1:200) - Coumarin; 8. DAPI.

* Sheep IgG VisUCyte HRP polymer (R&D Systems) was used as secondary antibody.

Manual scoring

IHC was scored for the frequency of CD8+ T cells at the border and in the center (illustrated in Figure 2A). The border region includes the invasive margin, and covers ~50% tumoral area (tumor cells and stroma) and ~50% peritumoral area (no or only isolated tumor cells, particularly in case of ILC subtypes), whereas the center region includes non-necrotic regions, and covers tumor and stroma. In case of lymph node (LN) metastases, only border regions that were not surrounded by lymphoid tissue were evaluated. Spatial phenotype of CD8+ T cells was determined using whole slide scans (Hamamatsu slide scanner) at lx magnification and using at least 8 regions of interest at 20x magnification in border and center. Scoring criteria were as follows: inflamed: almost equal frequencies of CD8+ T cells at the border and center; excluded: >10 times more CD8+ T cells at the border compared to center; and ignored: hardly any CD8+ T cells present at the border and center. All immune markers stained on consecutive slides were scored at 20x magnification (at border and center) and reported as percentage of positive cells (of total nuclei). Tertiary lymphoid structures (TLS) were identified as dense clusters of CD4+ T cells and CD20+ B cells on consecutive slides, whereas High endothelial venules (HEV) were identified as vessels that were MECA-79 positive (frequently found in TLS), and both TLS and HEV were reported as total number per tumor.

Digital image analysis

Following whole shde scans using VECTRA 3.0 (Akoya Biosciences), at least 8 stamps (regions of interest; stamp size: 670x502 mm ² ; resolution: 2 pixels/mm ² ; pixel size: 0.5x0.5 mm ² ) were set in non-necrotic areas at the tumor border (containing 50% peritumoral region) and center (both comprising tumor as well as stroma compartments, illustrated in Figure 3). In case parts of the tissue were disrupted or lost due to repeated staining cycles, fewer stamps were set or tissues were excluded from analysis (in case of <3 stamps at either border or center regions). Tissue-segmentation was performed according to cytokeratin and DAPI staining; cell-segmentation and phenotyping of individual cells was performed according to individual markers and presence of DAPI using Inform software; and enumerations at border (tumor and stroma) and center regions (tumor and stroma) were summarized for all stamps per sample. Spatial phenotypes were determined according to median CD8 ⁺ T cell density at border and center as follows: inflamed, >200 cells/mm ² at border and ratio between border and center <10; excluded, >200 cells/mm ² at border and ratio between border and center >10; ignored <150 cells/mm ² at border and center. All scans fulfilled either of these 3 spatial phenotypes. Collagen- 10 was identified through tissue segmentation and quantified as collagen- 10-positive tissue area. Nearest-neighbor analysis was performed in R using the PhenoptR package, to which end, the number of non-CD8 ⁺ T cells within a 10 mm radius of CD8 ⁺ T cells were calculated from the Inform -derived cell segmentation files in Phyton.

Gene expression and DNA mutational analysis RNA sequencing

RNAseq data was collected from fresh frozen TNBC using 150bp paired-end with LncRNA library (Ribo-zero RNA) on Illumina HiSeq. RNA was isolated from FFPE using the RecoverAll Total Nucleic Acid Isolation Kit for FFPE (Thermofisher). RNA was sequenced using the FFPE sample Eukaryotic RNA-seq Library (250~300 bp insert strand specific library with rRNA removal) on the Illumina Novoseq6000 platform at Novogene. Although FFPE starting material yielded poorer quality of RNA when compared to FF samples, we still captured sequencing data of n=12 out of n=15 samples with sufficiently high quality: i.e., these samples contained <50% duplicated reads (ranging from 20-45%); >50% mapped reads (ranging from 55-95%); and expressed >75% of classifier genes. Data normalization

Microarray data were normalized using fRMA (McCall et al., Biostatistics, ll(2):242-253 (2010)) and corrected for batch effects using ComBat. RNAseq data (cohorts A2, C, D E (TNBC)) were aligned with GRCh38 using the STAR algorithm (version 2.4.2a) and geTMM normalized (Smid et al., BMC Bioinformatics, 19(1): 1-13 (2018)) for differential expression (DE) analyses. For pan-cancer analyses, pre-processed data was used (i.e., The Cancer Genome Atlas (TCGA) for cancers other than breast cancer (BRCA): EB++Adjusted; and ICI-treated melanoma patients: FPKM normalized). TCR repertoire, neo-antigen and mutational signature analysis

TCR clonality was estimated using the MIXCR algorithm as described previously (Hammerl et al., Clin Cancer Res.,doi: 10.1158/1078-0432. CCR- 19-0285 (2019)); output was processed with tcR package in R and reported as TCR diversity (total number of TCR-Vb reads per sample) and TCR repertoire skewness (Gini-Simpson index of TCR-Vb reads per sample). Prediction of neo-antigens was performed with netCTLpan as described previously (Hammerl et al., Clin Cancer Res.,doi: 10.1158/1078-0432. CCR- 19-0285 (2019); Smid et al., Nat Commun., 7:12910 (2016)). Identification of mutational signatures was described elsewhere (Nik-Zainal et al., Nature, 534(7605):47-54 (2016)). Differential gene-, pathway- and immune cell subset analyses Differential gene expression (DE) analysis was performed in R using the limma package and voom function (Ritchie et al., 2015 Nucleic Acids Research, 43(7), e47). Differentially expressed genes (p<0.05, logFC>1) were used for ingenuity pathway analysis (IPA software, core analysis). Spatial phenotypes were also interrogated for DE of gene-sets related to T cell evasion (Hammerl et al., Clin Cancer Res. ,doi:10.1158/1078-0432. CCR-19- 0285 (2019)). Expression of a gene-set was determined as average expression of all genes in the respective set. Immune cell frequencies were estimated using the CIBERSORT algorithm in absolute mode. Gene-set enrichment analysis for Hallmark and Kegg data sets (v7.2) was performed using GSEA 4.1.0 software (Subramanian et al., Proc Natl Acad Sci., 102(43): 15545- 15550 (2005)) using weighted signal to noise ranking with 1000 permutations.

Gene classifier to assign snatial nhenotvnes

In a discovery set (Cohort Al, n=101 primary TNBC), we selected the top differentially expressed genes among inflamed, excluded and ignored phenotypes (>llogFC among all 3 phenotypes; p _adj <0.05) of samples with microarray data and corresponding CD8+ T cell staining data (Figure 7A). Expressions for each classifier gene were averaged for each of the three spatial phenotypes, ranks of gene expressions were calculated per spatial phenotype (Table 1), and assignments were based on highest Spearman rank-correlations between unknown samples and ranked expressions of classifier genes per spatial phenotypes of the discovery set. In a validation set (Cohort A2, n=43 primary TNBC), RNAseq data of independent samples with corresponding CD8+ T cell staining data were used to assign phenotypes based on highest rank-correlations with the discovery set (Al), and yielded 81% accuracy (Figure 4A). Correct assignment of unknown samples from Cohort B (RNAseq data) was verified by comparison of T cell characteristics, such as TCR-Vb repertoire diversity and numbers of intra- tumoral T cells, with those of Cohort A2 (RNAseq and CD8+ T cell stainings), and the classifier-assigned samples were found non -different compared to those from the validation set (Figure 5A, B). An additional validation set (Cohort F, n=12 metastasized TNBC) with RNAseq data and corresponding CD8+ T cell staining data showed 83% accurate assignment of spatial immune phenotypes in TN lymph node metastases (Figure 4B). Clinical vahdation was done using metastasized lesions from TNBC patients treated with anti-PDl antibody (Cohort D, pre-treatment), from which 3 out of 53 samples were excluded because of equally high rank-correlations. Assignment of spatial phenotypes in metastatic lesions did not depend on lesion site. We did observe significantly different proportions (decreased frequency of inflamed, as expected) in the metastasized (cohort D) versus primary setting (Cohorts Al and C) (Figure 12B). Predictive value of classifier gene-sets was determined by fitting ROC curves for anti-PDl response. Responders (CR, PR, SD > 24 weeks) and non-responders (PD) were separated using the pROC package in R. Excluded and inflamed gene- sets were calculated as average scores of all respective genes, and PD-L1 and sTIL scores were scored as described before ⁸ .

Statistical analysis

Statistical analysis was performed in R version 3.5.1 or GraphPad Prism 6. Log-rank test for trend was used to compare Kaplan-Meier curves; Cox- regression analysis was used to assess HR of immune phenotypes, clinical parameters (age, grade and size which were used as continuous variables), cell types or gene-sets; and Logistic regression was used to determine OR of gene-sets (glm.OR function). Multiple testing correction was performed for differential gene-expression analysis using the Benjamini-Hochberg method. Kruskal- Wallis test was used to assess differences in gene expression and immune cell densities among spatial phenotypes; Pearson-correlation was used to assess linear relationships between continuous variables; and Chi- Square test or Fishers’ exact test (in case of small sample sizes) were used to assess relationships among factorial variables. The following significance levels were used: *, p<0.05; **, p<0.01; ***, p0.001; ****, po.ooo1; NS, p>0.5. Results

Spatial contexture of lymphocytes but not myeloid cells is prognostic in TNBC

In order to assess tumor-immune interactions in TNBC, CD8+ T cell presence and spatial organization were studied in 236 untreated, primary TNBC using immunohistochemical staining (IHC) of whole slides (Cohort A; for study design see Figure 1 and for clinical details of cohorts see Table 2). These patients did not receive adjuvant chemotherapy enabling unbiased testing of the prognostic value of immune markers. We defined three spatial immune phenotypes: excluded (26%; predominant location of CD8+ T cells at tumor border, not center); ignored (28%; negligible presence of CD8+ T cells neither at border nor center) and inflamed (46%; CD8+ T cells evenly distributed across border and center) (see M&M section for detailed criteria of spatial phenotypes) (Figure 6A, Figure 2A). These spatial phenotypes were significantly associated with survival (distant metastasis-free survival (MFS), disease-free survival (DFS), and overall survival (OS): p<0.009; n=122 lymph node-negative TNBC). Tumors with an inflamed phenotype had the best prognosis (10-year OS: 80%), excluded phenotypes intermediate (10-year OS: 60%, hazard ratio (HR): 1.45, 95% Cl: 0.84-3.3), and ignored phenotypes the worst prognosis (10-year OS: 40%; HR:3, 95% Cl: 1.5-5.9) (see Figure 6B for univariate analysis). Prolonged survival of excluded versus ignored phenotypes was statistically significant for OS, but not MFS nor DFS. Notably, the prognostic value of spatial phenotypes was independent of nodal status, tumor size or age (data not shown).

In addition to CD8+ T cells, we assessed the presence of other immune effector cells using multiplexed immunofluorescence (IF) imaging of 64 tumors (Figure 6C, Figure 2B). CD4+ T cells and CD20+ B cells generally co-occurred with CD8+ T cells at the tumor border and center, whereas CD56+ NK cells were hardly present in TNBC (Figure 6D, Figure 3A-D). For instance, at the tumor border numbers of stromal CD20+ B cells, CD4+ and CD8+ T cells did not differ between excluded and inflamed phenotypes, yet the excluded phenotype had significantly fewer intratumoral B and T cells (Figure 3A, C). Moreover, distances between CD8+ T cells and tumor cells (CK-positive cells) were significantly larger in excluded versus inflamed phenotypes (Figure 6E, Figure 3E). Interestingly, despite a lack of lymphocytes in the ignored phenotype, we did observe stromal and intra tumoral CD68+ macrophages (Figure 6D, Figure 3A-D). Notably, densities of stromal CD8+ T cells and intratumoral CD4+, CD8+ T and CD20+ B cells, but not CD68+ macrophages, demonstrated significant correlations with OS or MFS (Figure 3F). Next, we evaluated the presence of tertiary lymphoid structures (TLS), defined as focal areas that are positive for CD4+ T and CD20+ B cells, which are considered important sites for T cell priming and initiation of an anti-tumor immune response. Interestingly, we observed a high number of TLS at the border of tumors of both the inflamed and excluded phenotype, but not in the ignored phenotype (Figure 6F). Of note, neither the presence nor abundance of TLS were significantly associated with survival (tested for OS, MFS and DFS in univariate and multivariable setting), nor when stratified per spatial immune phenotype (data not shown).

A gene classifier of spatial phenotypes predicts outcome to anti-PDl treatment in TNBC patients

We developed a gene-expression classifier to be able to assess prognostic and predictive values of the spatial immune phenotypes without the need for CD8+ T cell stainings. Briefly, we selected most discriminative genes (according to differential expression, DE) for the excluded, ignored and inflamed phenotypes in a discovery set for which both gene expression data and CD8+ T cell stainings were available (Cohort Al, n=101 primary TNBC, Figure 7A). Using DE and rank-correlations with phenotypes from the discovery set, we assigned spatial phenotypes in an independent validation set (Cohort A2, n=43 primary TNBC; gene expression data and CD8+ stainings), which resulted in correct assignment of spatial phenotypes in 81% of primary TNBC (Figure 4A; see M&M section for details on classification). Using a second validation set (Cohort F, n=12 metastatic TNBC; gene expression data and CD8+ stainings), we showed correct assignment of spatial phenotypes in 83% of TN lymph node metastases (Figure 4B). Subsequently, the prognostic value of this spatial -phenotype- classifier was tested in an independent cohort of primary, lymph-node negative, systemically untreated BC (Cohort B, n=196 basal -like tumors (Hammerl et al, Chn Cancer Res., doi: 10.1158/1078-0432. CCR- 19-0285 (2019)); only gene expression data available). Genes highly expressed in the excluded or ignored phenotypes included: THBS2, ASPN, COLlOAl, COL5A1 GREM1, SPON1, FAP and SPOCK1, which were all significantly associated with poor MFS (HR>1, p<0.05). On the other hand, genes highly expressed in the inflamed phenotype included: WARS, CXCL13, CCL5, GZMB, TRBC1, COROlA, CCL5, CCL18, IL2RG, NKG7, IGHGl, which were all significantly associated with better MFS (HR<1, p<0.05) (Figure 5A). Assessment of the entire gene-sets of the excluded and ignored phenotypes were associated with poor prognosis (excluded: HR=1.8, Cl: 1.2- 2.7; ignored: HR=1.6, Cl: 1.1-2.4), whereas the gene-set of the inflamed phenotype was associated with good prognosis (HR=0.62, CI:0.45-0.86) (Figure 7B). Upon testing the performance of the gene classifier in a third cohort of primary TNBC patients (Figure 7C, Cohort E, n=137), we validated the prognostic value of the spatial-phenotype-classifier (logrank, p=0.001). It is noteworthy that among all BC, basal-like BC had the highest proportion of the inflamed phenotype followed by her2 and luminal-B subtypes (Figure 5B). To test the capacity of the spatial-phenotype-classifier to predict outcome after anti-PDl treatment in TNBC, we applied the classifier to a dataset of metastatic patients from the TONIC trial (Voorwerk et al., Nat Med. 2019. doi: 10.1038/s41591-019-0432-4). In this phase II trial, all patients received anti-PD 1 after a short (2 week) immune induction treatment with low dose chemotherapy or irradiation (cohort D, n=53, biopsies from pre- and post induction treatment metastatic lesions, see Table 2 for details). We observed significantly higher frequencies of the excluded (41%) and ignored phenotypes (37%), and decreased frequencies of the inflamed (21%) phenotypes when metastasized TNBC was compared to primary TNBC (data not shown). This was not dependent on biopsy sites, which supports the prognostic nature of the classifier (data not shown). Expression of the exclusion gene-set was significantly higher in non-responding (progressive disease (PD) patients) patients (odds-ratio (OR): 3.5; Cl: 1.2-11.9), whereas expression of the inflamed gene-set was significantly higher in responding patients (complete response (CR) + partial response (PR) + stable disease (SD) for > 24 weeks according to iRECIST criteria (Seymour et al., Lancet Oncol., 18(3):el43-el52 (2017)) (OR: 0.4; Cl: 0.18-0.92) (Figure 7D, E). No association with therapy response was found for the ignored gene-set (OR=0.9; Cl: 0.5-1.85). When assessing receiver operating characteristic

(ROC) as a measure of predictive value of the excluded, inflamed or combined gene-sets, we observed areas under the curve (AUC) of 0.72 (CL0.52-0.89), 0.73 (Cl: 0.54-0.94) and 0.75 (Cl: 0.55-0.95), respectively. In comparison, PD-L1 expression on immune cells, a biomarker that is currently used in the clinical setting had an AUC of 0.66 (Cl: 0.51-0.82) (Figure 7F). The AUC for sTIL, another marker considered to stratify patients was 0.67 (Cl: 0.48-0.82) (Figure 7F). In addition, non-responder patients showed enrichment for the excluded and ignored phenotypes (90% of cases), whereas in anti-PDl -responders the inflamed phenotype was enriched (60% of cases, Chi-square, p=0.007, Figure 7G). In fact, using the spatial-phenotype-classifier we were able to predict outcome after anti-PDl treatment, i.e., the negative predictive value (NPV) of the inflamed phenotype to therapy response is 0.9, while the positive predictive value (PPV) is 0.6 (Figure 4C). In line, patients with the excluded and ignored phenotypes had shortened OS when compared to the inflamed phenotype (logrank, p=0.05, Figure 7H). Notably, spatial phenotypes predict clinical outcome independent of immune cell PD-L1 but not sTIL (Figure 14).

Prognostic and predictive value of snatial nhenotvnes in multiple cancers We applied the spatial-phenotype-classifier to other solid tumor (sub-)types to assess the prognostic and predictive value in a pan-cancer setting. Spatial phenotypes were significantly prognostic not only in invasive breast cancer BRCA (all subtypes, including ER+), but also in bladder cancer (BLCA), skin cutaneous melanoma (SKCM), cervical squamous cell carcinoma and endocervical adenocarcinoma (CESC), head and neck squamous cell carcinoma (HNSC) and kidney cancer (RICH) (Cohort E, Figure 5C). Similar trends were seen in prostate (PRAD), pancreatic (PAAD), lung (LUAD) and colon cancer (COAD) (Cohort E, Figure 5C). Although spatial immune phenotypes have been described for various cancers, they have not directly been related to response to ICI treatments such as anti-PD 1 treatments. Here we demonstrate that tumors generally responding poorly to ICI, such as PRAD and PAAD, had the highest proportions of the excluded or ignored phenotypes, while tumors generally responding well to ICI, such as SKCM and LUAD, had the highest proportions of the inflamed phenotype (Figure 5B). In line with TNBC, in advanced and metastatic melanoma, where RNAseq data of ICI-treated patients is publicly available

(Riaz et al„ Cell, 171(4):934-949.el5 (2017); Hugo et al„ Cell,165(l):35-44 (2016)), we observed that expression of the gene-set of the excluded phenotype was significantly increased in tumors of patients not-responding to ICI, and the gene-set of the inflamed phenotype was significantly increased in tumors of patients responding to ICI (Figure 5D). Moreover, the spatial-phenotype-classifier outperformed other, publicly available gene- classifiers that are recognized for capturing lymphocyte activity and location, and for predicting anti-PDl response in melanoma, such as IFNg- response, T cell-exclusion and TLS signatures (Figure 8).

Spatial phenotypes differ in TCR repertoire skewness and mutational signatures but not mutational burden

In order to test for potential drivers of spatial phenotypes, we first studied clinicopathological and genomic features in lymph node negative, systemically untreated, primary TNBC (Cohorts A and C). Spatial phenotypes were not associated with mitotic activity index (MAI), tumor grade, tumor stage or histological subtypes, except for tumors with medullary features that were (as expected) solely comprised of the inflamed phenotype (data not shown). In addition, following assignment of spatial phenotypes to Cohort C (n=66; RNAseq and WGS data), we observed that spatial phenotypes did neither differ with respect to frequency of BRCAl or BRCA2 germline mutations (Figure 9A), frequency of f>2 Microglobulin loss (Figure 9B) nor TMB or types of genomic alterations, including non- synonymous SNV (passenger and driver mutations combined), exonic frameshifts, indels (Figure 9C) or predicted neo-antigens (Figure 9D). In contrast, spatial phenotypes did differ with respect to mutational signatures and TCR clonality (Figure 9E-I). For instance, mutational signature-3 (related to homologous recombination deficiency) was enriched in the inflamed phenotype and signature-5 (related to age) was significantly enriched in the ignored and excluded phenotypes (Figure 9F, G). The highest TCR-Vb diversity as well as the most skewed TCR-Vb repertoire (harboring clonally expanded reads) were observed in the inflamed phenotype, and both these parameters were equally low in the excluded and ignored phenotypes (Figure 9H, I).

Spatial phenotypes are characterized by distinct immune evasive pathways Next, we studied whether spatial phenotypes capture different modes of immune-evasion (Cohort A, Figures 10, 11). Immune cell deconvolution by Cibersort (Gentles et al., Nat Med., 21(8):938-945 (2015)) confirmed above observations (Figure 6D) with respect to the abundance of immune effector cells and particularly revealed differential frequencies of plasma cells, activated memory T cells, follicular helper T cells, activated dendritic cells (DC), and Ml- and M2 macrophages (Figure 10A). Subsequently, we evaluated expression of gene-sets related to various mechanisms of T cell evasion (Hammerl et al., Clin Cancer Res, doi:10.1158/1078-0432. CCR-19- 0285 (2019)), complemented with Ingenuity Pathway Analysis (IPA®) and verified with gene-set enrichment analysis. Using this approach, we observed that the excluded phenotype was characterized by enhanced expression of genes associated with endothelial barrier, glycolysis, serine protease inhibition (SPI), and extracellular matrix (ECM) remodeling (Figure 10B-D, see Figure IOC for examples of individual genes); notably all these pathways were inter-linked with the TGFb pathway (Figures 10D, E). One of the most up -regulated genes in the excluded phenotype (compared to the inflamed phenotype) was COLlOAl (Figure IOC), the expression of which was strongly correlated to the TGFb- and VEGF- signahng pathways while being inversely correlated to the expression of CD8A (Figure 10E, H). The ignored phenotype was characterized by increased expression of genes associated with b-oxidation (Figure 10B) as well as the WNT, PPAR, LXR/RXR and MAPK pathways (Figures IOC, D). Moreover, the ignored phenotype showed enhanced gene expression of S100A7 (Figure IOC), a molecule that has been reported to promote oncogenesis and act as a chemo-attractant for M2 macrophages and other suppressive myeloid cell (Nasser et al., Cancer Res, 75(6):974-985 (2015)). Of the above oncogenic pathways in particular WNT was inversely correlated with the expression of CD8A as well as CD 163 (Figures 10F, H). Lastly, the inflamed phenotype showed enhanced expression of genes associated with necrosis, TNF-signaling, type-I and type-II IFN, antigen processing and presentation, T cell co-stimulation, but also co-inhibition (Figures 10B, C), which were all inter-related (data not shown). Importantly, the inflamed phenotype showed high gene expression of the T cell chemo-attractants CXCL9 and CXCL10 (Figure IOC), which according to our immune cell deconvolution and pathway analyses are derived from activated (BATF3/CLEC9A-positive) conventional DC (cDCl, Figure 10G). CD 163 and T cell co-inhibition (Spranger et al., Sci Transl Med., 5(200):200rall6. doi:10.1126/scitranslmed.3006504 (2013)), generally downstream of an immune response, were correlated with the expression of CD8A (Figure 10H).

Multiplex IF demonstrated that collagen- 10 was deposited into stromal areas between tumor and immune cells at the tumor center in the excluded phenotype, (Figure 11A, B). To assess how entry of T cells may be affected by such a physical barrier, we evaluated the presence of high endothelial venules (HEV, identified via MECA-79 stainings), and observed that these were present at high numbers at the border as well as center of excluded phenotypes (Figure 11D). In the ignored phenotype, IF showed (albeit only in subset of ignored tumors) that very high S100A7 expression by tumor cells (highest of all spatial phenotypes) was accompanied by high frequencies of CD 163+ macrophages (Figure 11A, middle panel, Figure 11B). Even though tumor-associated macrophages were not unique for the ignored phenotype, and were present at particularly high densities in the center of inflamed and to a lesser extent in excluded phenotypes, nearest- neighbor analysis revealed that macrophages and myeloid cehs showed relatively low distances to CD8+ T cells, regardless of frequencies and spatial phenotypes (Figure 11C). CD66b+ neutrophils (another immune cell type that has been reported for its immune-suppressive effects in the TME) co-occurred with macrophages and myeloid cehs and were found to be present at high numbers in the same subset of the ignored phenotype (Figure HE). Notably, the ignored phenotypes that did not show high M2 and neutrophil densities were characterized by enhanced expression of WNT targets. In the inflamed phenotype, IF revealed significantly enhanced numbers of stromal as well as intra tumoral CLEC9A+ DC (Figure 11 A, B). Interestingly, and despite overall low abundances of these cells (regardless of spatial phenotype), CLEC9A+ DC were found in relatively close proximity to CD8+ T cells (Figure 11C) and their cell densities significantly correlated with those of CD8+ T cells (data not shown), pointing to the recognized immune-enhancing action governed by cDCl cells. Nevertheless, and despite high densities of T cells and TLS, only a small fraction of CD4+ and CD8+ T cells in the inflamed phenotype expressed the co-stimulatory receptors ICOS or 4 IBB, which co-ocurred with a significantly decreased MHC-II expression by tumor cells (Figure 11D, F).

Discussion In this study, using cohorts of in total 681 patients with TNBC, 2706 with other types of BC, and 4003 with other cancers, we have analyzed spatial immune phenotypes in relation to prognosis and response to anti-PDl treatment as well as genomic features and T cell evasion. Our results present for the first time that spatial immune phenotypes predict response to ICI in TNBC, and are characterized by distinct T cell evasive pathways that provide a rationale to develop spatial phenotype-specific therapies for ICI-refractory TNBC.

We have developed and validated a novel spatial -phenotype-classifier that accurately predicts spatial locahzation of CD8+ T cells in primary as well as metastasized TNBC. Next to its prognostic value in TNBC, this classifier has prognostic value in various tumor types (BC, CESC, HNSC, KICH, BLCA, SKCM), and suggests that the classifier may be apphed to different histologies. Strikingly, we found that this classifier predicts resistance to anti-PDl treatment in metastatic TNBC as well as melanoma. In case of TNBC, we report an NPV as high as 0.9, which is not achieved with the currently used predictor PD-L1. In fact, the spatial immuophenotype classifier acts independently of PD-L1 (data not shown) and outperfoms alternative classifiers that relate to lymphocyte activity and location (Figure 8). Notably, TLS, whether captured by staining (as performed in this study) or a gene signature, was neither significantly associated with survival nor anti-PDl response, irrespective of stratification per spatial immune phenotype (data not shown), indicating that further research is needed to determine the exact role of TLS in shaping anti-tumor immune responses in TNBC. Although not excluding CD8 stainings, our observations imply that gene- based classification of spatial immune phenotypes would enable early identification of non-responders and facilitate decision-making by clinical oncologists with respect to treatment of TNBC patients with ICI. Improved decision-making would prevent non-responding patients to receive ineffective and expensive treatment, and potentially challenge the diagnostic need for tissue stainings, which require whole tissue sections that are often not available as well as uniform staining protocols and training of pathologists. Regarding diagnostic implementation of the gene classifier, expressions from those genes that are part of our gene classifier can be developed into a routine tool. Alternatively, NGS-techniques, expected to become part of systemic evaluations of tumor tissues for targetable alterations in the near future at departments of Pathology of Medical Centers, can be used towards the application of the gene classifier. In addition to the predictive value of spatial phenotypes, in the TONIC trial, we observed that proportions of inflamed phenotype increase following induction treatment with cisplatin and doxorubicin (Figure 12), suggesting that spatial phenotypes show plasticity and non-inflamed phenotypes can be primed for treatment with ICI. In this regard, the immune determinants that characterize these phenotypes (Figure 10D) and their correlative relationships with actionable targets (Figure 10E, F, G, H) provide a rationale for below-mentioned co-treatments (illustrated in Figure 13, and discussed below). In case of the excluded phenotype, inhibitors of TGFb, such as the bifunctional anti-PDL-1 mAb/TGFb trap M7824, and inhibitors of VEGF receptor kinases, such as cediranib, both being in clinical development for TNBC and the latter being FDA-approved for other malignancies, can potentially prime for ICI. In case of the ignored phenotype, blockers of the WNT pathway, such as WNT974 and/or drugs that target M2 macrophages, such as pexidartinib, a CSF1R inhibitor that depletes M2 macrophages, are of interest, and are currently being tested in TNBC. The inflamed phenotype, being enriched in patients responding to anti-PDl treatment, would be the phenotype of choice to start combination ICI treatment. In case ICIs are not effective, this phenotype could potentially benefit from combining multiple ICIs or priming with CSF1R inhibitors that target M2 macrophages. Another mode of priming the inflamed phenotype could be reactivation of type I/II IFN and/or chemoattractant pathways, thereby re-boosting antigen presentation as well as recruitment and function of intra-tumoral CD8 T cells; to this end, an option could be the epigenetic drug decitabine that is approved for other indications and has shown promising results in preclinical studies of TNBC.

The above-mentioned targets are part of larger immune networks that were revealed upon integrative analyses of TNBC samples using NGS and multiplexed IF. The charting of these larger networks enabled the identification of TME- and immune response-mediated paths of T cell evasion and their relationship to ICI response. Following this approach, we observed that the excluded phenotype was characterized by CD4+, CD8+, CD20+ and CD56+ lymphocytes that were preferentially located at the tumor border at large distances from tumor cells. This phenotype had high expression of collagen- 10, which is not present in normal tissues, is associated with epithelial-to-mesenchymal transition, as weU as poor survival in TNBC and various other tumor types. Recently, it has been suggested, based on collagen fiber density (not further specified) and in silico modeling of T cell -influx, that T cell-exclusion in TNBC is regulated by chemorepellents rather than barriers of extracellular matrix. In contrast, our gene expression and in situ stainings (Figures IOC, H, 11A, B) strongly suggest that T cell-exclusion is due to collagen- 10 deposition, possibly hinting towards a unique role of collagen- 10 imposing a physical barrier to T cell-influx (Figure 10H, 1st plot). Next to the collagen barrier, our data point to enhanced tumor cell glycolysis, which has been reported to suppress T cell-mediated apoptosis of TNBC in vitro, and which may further promote T cell-exclusion (Figure 10H, 3rd plot). In addition, several serpins and other protease inhibitors, such as SERPINE1, SPINK1 and SLPI, demonstrated high gene expression. These enzyme inhibitors limit the activity of matrix metalloproteases or granzymes, thereby again potentially inhibiting T cell-influx (Figure 10H, 2nd plot) or T cell- mediated apoptosis of tumor cells. All immune evasive pathways associated with the excluded phenotype are inter-related (data not shown) and strongly correlate to expression of TGFb and VEGF (Figure 10E), which most likely represent upstream regulators that contribute to TME-mediated T cell evasion.

The ignored phenotype was characterized by no or very low densities of CD8+ T cells and either showed high expression of target genes of the WNT and PPARg/RXR pathways or contained CD 163+ macrophages and CD66b+ neutrophils. We found inverse correlations between WNT pathway activity and presence of CLEC9A+ DC and CD8+ T cells (Figure 10H, 4th plot) as well as TCR repertoire skewness. It is suggested that the occurrence of both WNT and PPAR pathways are representative of pan-cancer mechanisms of TME-mediated T cell evasion. Notably, we observed strong inverse correlations with either pathway and the abundance of CD 163+ cells (Figure 10F), and argue that the presence of M2 macrophages represents a second immune escape mechanism of the ignored phenotype (Figure 10H, 5th plot). It has previously been reported that murine models of BC revealed that S100A7 expression induced M2 macrophage recruitment and promoted metastasis. In the current study with patient materials, however, we found that numbers of S100A7+ tumor cells as well as CD 163+ cells, located at the border, were positively correlated with MFS (data not shown) and (low) frequencies of CD8+ T cells, arguing that recruitment of these myeloid cells is part of a negative feedback loop that follows an initial immune response.

Finally, the inflamed phenotype was characterized by high numbers of intra-tumoral CLEC9A+ DC and lymphocytes. The prognostic value of TILs was mainly attributed to T and B cells located in tumor regions, a finding that is in line with earlier observations showing that proximity to tumor cells is a pre-requisite for effective anti-tumor activity of lymphocytes. The inflamed phenotype had a high TCR clonality independent of the level of neo-antigens and showed highest expression of genes associated with immunogenic cell death, type I/II IFNs and chemo-attractants (Figure 10G). The combination of an inhibitor of DNA methyl transferase (5- azacytidine), which targets the latter two pathways, and anti-PDl antibody (pembrolizumab) has shown precedent in phase I/II studies to treat myelodysplastic syndromes (Chien et al., Br J Haematol, 195(3):378-387 (2021)). Interestingly, we observed that gene sets associated with necrosis, but not any other form of cell death, strongly correlated with densities of CD8+ T cells (data not shown), suggesting that immunogenic cell death may be a trigger of the cDCl-initiated adaptive immune response. Despite high numbers of DCs, TILs in the inflamed phenotype over-expressed genes encoding for various immune checkpoints and only a minority of TILs expressed ICOS or 41BB (Figure 11F). In fact, a large fraction of the inflamed phenotype showed genetic alterations in MHC-I (Figure 9B) and down-regulated expression of MHC-II by tumor cells (Figure HE). All the above changes are again inter-related (data not shown) and considered part of an immune response-mediated negative-feedback loop (Figure 10H, 5th and 6th plots), and may contribute to the relatively low frequency of sustained clinical responses to ICI even in the inflamed phenotype.

In conclusion, our study has resulted in the development and validation of a gene-classifier that accurately assigns spatial immune phenotypes in TNBC and metastasized TNBC, and is associated with prognosis in TNBC and various other cancers. This spatial-phenotype-classifier predicts patient response to anti-PDl independently of currently used clinical markers and outperforms other gene-signatures, thereby addressing an urgent clinical need. Finally, in-depth analysis of NGS, immunologic and clinical sets of patient data points towards actionable targets that may proof beneficial for phenotype-stratified ICI therapy in TNBC.