WO2019053244A1

WO2019053244A1 - Biomarkers for responsiveness checkpoint inhibitor therapy

Info

Publication number: WO2019053244A1
Application number: PCT/EP2018/075017
Authority: WO
Inventors: Carsten KRIEG; Mitchell LEVESQUE; Burkhard Becher
Original assignee: Universität Zürich
Priority date: 2017-09-15
Filing date: 2018-09-17
Publication date: 2019-03-21
Also published as: WO2019053244A9

Abstract

The invention relates to a method of assigning to a patient suffering from cancer a likelihood of being responsive to checkpoint inhibitor therapy. The method comprises the steps of: 1) determining in a blood sample obtained from the patient the relative frequency of classical monocytes, wherein the relative frequency is the percentage of classical monocytes on the number of peripheral blood mononuclear cells; 2) comparing the relative frequency with a threshold and 3) attributing a high likelihood of being responsive to checkpoint inhibitor therapy to the patient if the relative frequency is above the threshold.

Description

Biomarkers for Responsiveness to Checkpoint Inhibitor Therapy

The present invention relates to the use of biomarkers in order to stratify patients before checkpoint inhibitor therapy.

Background of the invention

Checkpoint Inhibitor Therapy (CIT) has revolutionized the treatment of cancer as it shows effective treatment of otherwise untreatable neoplasia, such as melanoma.

The physiological function of immune checkpoints is to prevent autoimmunity by down- regulating the immune system and promoting self tolerance. Tumor cells however use this system to evade detection and attack by the immune system.

Programmed cell death protein 1 (PD-1 ) is a known immune checkpoint. Immunotherapy with anti-PD-1 aims to block the interaction of tumor-reactive T cells with PD-1 ligands (PD-L1 and PD-L2) expressed on various cells types including leukocytes and the tumor cells themselves. Clinical trials on PD-1 and PD-L1 blockade for patients with advanced melanoma have demonstrated consistent therapeutic responses, thus prompting their application to several other cancers. Nivolumab, an anti-PD-1 monoclonal antibody, has been approved by the US FDA for the treatment of patients with metastatic melanoma, non-small cell lung carcinoma (NSCLC), metastatic renal cell carcinoma, metastatic squamous NSCLC and refractory Hodgkin's lymphoma. Pembrolizumab has shown similar efficacy and is now FDA approved as a second line treatment drug for melanoma and for the treatment of patients with NSCLC, advanced gastric cancer, advanced bladder cancer, head and neck cancer, classical Hodgkin's lymphoma, and triple negative breast cancer. Further anti-PD-L1 mAbs have been developed for the treatment of advanced human cancers including metastatic urothelial bladder cancer.

Clinical outcomes of anti-PD-1 immunotherapy are however highly variable, with only a fraction of patients showing durable responses, some with early progression and others with late response, while the majority of treated patients show no clinical benefit. The discovery of reliable biomarkers to stratify patients prior to treatment is urgently needed in order to improve identification of responders to anti-PD-1 immunotherapy.

Description

Based on the above mentioned state of the art, the objective of the present invention is to provide means and methods to identify and discriminate responder and non-responder to anti- PD-1 immunotherapy and to establish new methods for patient stratification. This objective is attained by the claims of the present specification. Terms and definitions

In the context of the present specification, the term checkpoint inhibitory therapy relates to a therapy that overrides an immune checkpoint mechanism and enables the immune system to attack tumor cells. The agents used in checkpoint inhibitory therapy may be checkpoint inhibitory agents or checkpoint agonist agents.

In the context of the present specification, the term checkpoint inhibitory agent or checkpoint inhibitory antibody is meant to encompass an agent, particularly an antibody (or antibody-like molecule) capable of disrupting the signal cascade leading to T cell inhibition after T cell activation as part of what is known in the art the immune checkpoint mechanism. Non-limiting examples of a checkpoint inhibitory agent or checkpoint inhibitory antibody include antibodies to CTLA-4 (Uniprot P16410), PD-1 (Uniprot Q151 16), PD-L1 (Uniprot Q9NZQ7), B7H3 (CD276; Uniprot Q5ZPR3), Tim-3, Gal9, VISTA, Lag3.

In the context of the present specification, the term checkpoint agonist agent or checkpoint agonist antibody is meant to encompass an agent, particularly but not limited to an antibody (or antibody-like molecule) capable of engaging the signal cascade leading to T cell activation as part of what is known in the art the immune checkpoint mechanism. Non-limiting examples of receptors known to stimulate T cell activation include CD122 and CD137 (4-1 BB; Uniprot Q0701 1 ). The term checkpoint agonist agent or checkpoint agonist antibody encompasses agonist antibodies to CD137 (4-1 BB), CD134 (OX40), CD357 (GITR) CD278 (ICOS), CD27, CD28.

In the context of the present specification, the term antibody is used in its meaning known in the art of cell biology and immunology; it refers to whole antibodies including but not limited to immunoglobulin type G (IgG), type A (IgA), type D (IgD), type E (IgE) or type M (IgM), any antigen binding fragment or single chains thereof and related or derived constructs. A whole antibody is a glycoprotein comprising at least two heavy (H) chains and two light (L) chains inter-connected by disulfide bonds. Each heavy chain is comprised of a heavy chain variable region (VH) and a heavy chain constant region (CH). The heavy chain constant region is comprised of three domains, CH1 , CH2 and CH3. Each light chain is comprised of a light chain variable region (abbreviated herein as VL) and a light chain constant region (CL). The light chain constant region is comprised of one domain, CL. The variable regions of the heavy and light chains contain a binding domain that interacts with an antigen. The constant regions of the antibodies may mediate the binding of the immunoglobulin to host tissues or factors, including various cells of the immune system (e.g., effector cells) and the first component of the classical complement system.

The term antibody-like molecule in the context of the present specification refers to a molecule capable of specific binding to another molecule or target with high affinity / a Kd < 10E-8 mol/l. An antibody-like molecule binds to its target similarly to the specific binding of an antibody. The term antibody-like molecule encompasses a repeat protein, such as a designed ankyrin repeat protein (Molecular Partners, Zurich), a polypeptide derived from armadillo repeat proteins, a polypeptide derived from leucine-rich repeat proteins and a polypeptide derived from tetratricopeptide repeat proteins.

The term antibody-like molecule further encompasses a polypeptide derived from protein A domains, a polypeptide derived from fibronectin domain FN3, a polypeptide derived from consensus fibronectin domains, a polypeptide derived from lipocalins, a polypeptide derived from Zinc fingers, a polypeptide derived from Src homology domain 2 (SH2), a polypeptide derived from Src homology domain 3 (SH3), a polypeptide derived from PDZ domains, a polypeptide derived from gamma-crystallin, a polypeptide derived from ubiquitin, a polypeptide derived from a cysteine knot polypeptide and a polypeptide derived from a knottin.

In the context of the present specification, CD14 refers to a protein identified by UniProt ID P08571 .

In the context of the present specification, CD16 refers to CD16a, identified by UniProt ID P08637 and/or CD16b, identified by UniProt ID 075015.

In the context of the present specification, CD33 refers to a protein identified by UniProt ID P20138.

In the context of the present specification, HLA-DR refers to "Human Leukocyte Antigen - antigen D Related", encoded by the human leukocyte antigen complex on chromosome 6 region 6p21 .31 .

In the context of the present specification, the designation "positive" or "+" with regard to the expression of a certain marker molecule is used in its meaning known in the field of immunology. In particular, if a cell population is designated "positive" or "+" with regard to the expression of a certain marker molecule, this shall mean that the cell population can be stained by a common fluorescent-dye-labelled antibody against the marker molecule and will give a fluorescence signal of at least 30% higher intensity, particularly of at least double the intensity, more particularly of at least one, two or three log higher intensity compared to unlabelled cells or compared to cells labelled with the same antibody but commonly known as not expressing said marker molecule or compared to cells labelled with an isotype control antibody.

In the context of the present specification, the designation "negative" or "-" with regard to the expression of a certain marker molecule is used in its meaning known in the field of immunology. In particular, if a cell population is designated "negative" or "-" with regard to the expression of a certain marker molecule, this shall mean that a cell population cannot be stained by a fluorescent-dye labelled antibody as described above against the marker molecule.

In the context of the present specification, the designation "hi" with regard to the expression of a certain marker molecule is used in its meaning known in the field of immunology. In particular, if a cell population is designated "hi" with regard to the expression of a certain marker molecule, this shall mean that the cell population stained by a common fluorescent-dye-labelled antibody against the marker molecule will give a fluorescence signal significantly stronger, particularly at least 2 x stronger or 10 x stronger than that given by a cell population designated "+" with regard to the expression of said marker molecule.

In the context of the present specification, the term "classical monocytes" is used in its meaning known in the field of immunology. In particular, the term refers to monocytes that are characterized by expression of the CD14 cell surface receptor and absence of expression of the CD16 cell surface receptor (CD14⁺CD16^").

A first aspect of the invention relates to a method of prognosis, specifically of assigning to a patient suffering from cancer a likelihood of being responsive to checkpoint inhibitor therapy. The method comprises the steps of

a. determining in a blood sample obtained from said patient the relative frequency of classical monocytes, wherein the relative frequency is the percentage of classical monocytes on the number of peripheral blood mononuclear cells,

b. comparing the relative frequency with a threshold and

c. attributing a high likelihood of being responsive to checkpoint inhibitor therapy to the patient if the relative frequency is above the threshold.

A second aspect of the invention relates to a method of therapeutic scheduling, specifically of assigning a patient suffering from cancer to checkpoint inhibitor therapy. The method comprises the steps of

a. determining in a blood sample obtained from the patient the relative frequency of classical monocytes, wherein the relative frequency is the percentage of classical monocytes on the number of peripheral blood mononuclear cells,

b. comparing the relative frequency with a threshold and

c. assigning the patient to checkpoint inhibitor therapy if the relative frequency is above the threshold.

In certain embodiments of the first or second aspect of the invention, the classical monocytes are characterized by expression of CD14⁺CD16^"CD33^hiHLA-DR^hi. In certain embodiments of the first or second aspect of the invention, the threshold is 1 .6 times (x) the average relative frequency of classical monocytes in a blood sample obtained from a healthy subject. In certain embodiments, the threshold is 1 .7 x the average relative frequency of classical monocytes in a blood sample obtained from a healthy subject. In certain embodiments, the threshold is 1 .8 x the average relative frequency of classical monocytes in a blood sample obtained from a healthy subject. In certain embodiments, the threshold is 1 .9 x the average relative frequency of classical monocytes in a blood sample obtained from a healthy subject. In certain embodiments, the threshold is 2.0 x the average relative frequency of classical monocytes in a blood sample obtained from a healthy subject. In certain embodiments, the threshold is 2.1 x the average relative frequency of classical monocytes in a blood sample obtained from a healthy subject. In certain embodiments, the threshold is 12.2 x the average relative frequency of classical monocytes in a blood sample obtained from a healthy subject.

In certain embodiments of the first or second aspect of the invention, the relative frequency of classical monocytes is determined by flow cytometry.

In certain embodiments of the first or second aspect of the invention, the cancer is melanoma. In certain embodiments, the cancer is metastatic melanoma.

In certain embodiments of the first or second aspect of the invention, the checkpoint inhibitor therapy comprises treatment with a checkpoint inhibitory agent. In certain embodiments, the checkpoint inhibitor therapy is selected from the group comprising anti-PD-1 immunotherapy, anti-PD-L1 immunotherapy, anti-CTLA-4 immunotherapy, anti-TIM-3 immunotherapy, anti- Lag-3 immunotherapy, or a combination of said therapies.

In certain embodiments of the first or second aspect of the invention, the checkpoint inhibitor therapy is anti-PD-1 immunotherapy. In certain embodiments, the anti-PD-1 immunotherapy comprises administration of an anti-PD-1 immunotherapy agent selected from the group comprising Nivolumab and Pembrolizumab.

In certain embodiments of the first or second aspect of the invention, additionally, the level of a marker selected from expression of albumin, expression of c-reactive protein and relative frequency of immature granulocytes is determined in the blood sample and wherein an increase in the additional marker is predictive of the patient being responsive to checkpoint inhibitor therapy.

According to another aspect of the invention, a checkpoint inhibitory agent or a checkpoint agonist agent for use in the therapy of cancer is provided, wherein a patient receiving the checkpoint inhibitory agent or a checkpoint agonist agent is characterized by a relative frequency of CD14⁺CD16^"CD33^hlHLA-DR^hl monocytes determined as a percentage of classical monocytes in relation to the number of peripheral blood mononuclear cells determined in a blood sample obtained from the patient, and the relative frequency is above 1 .6 x, 1 .7 x, 1 .8 x, 1 .9 x, 2.0 x, 2.1 x, 2.2 x the average relative frequency of classical monocytes in a blood sample obtained from a healthy subject.

In certain embodiments of this aspect of the invention, the cancer is melanoma, in particular metastatic melanoma.

In certain embodiments of this aspect of the invention, the checkpoint inhibitory agent is selected from the group comprising an anti-PD-1 antibody, an anti-PD-L1 antibody, an anti- CTLA-4 antibody, an anti-TIM-3 antibody, an anti-Lag-3 antibody, or a combination of said antibodies.

According to another aspect of the invention, a method of monitoring the success of checkpoint inhibitor therapy is provided. The method comprises the steps of

a. determining in a blood sample obtained from a patient receiving the checkpoint inhibitor therapy the relative frequency of classical monocytes, wherein the relative frequency is the percentage of classical monocytes on the number of peripheral blood mononuclear cells,

b. comparing the relative frequency with a monitoring threshold and

c. classifying the checkpoint inhibitor therapy as successful if the relative frequency is above the monitoring threshold.

According to another aspect of the invention, a diagnostic kit is provided. The kit comprises ligands, in particular antibodies, binding to CD14, CD16, CD33 and HLA-DR.

In certain embodiments of this aspect of the invention, the kit further comprises ligands, in particular antibodies, binding to CD3, CD4, CD1 1 b, CD19, CD45RO and CD56.

In certain embodiments of this aspect of the invention, the kit further comprises at least one antibody binding to CD45, CD61 , CD66b, ICAM-1 , CD86, CD1 1 c, CD7 or PD-L1 . In certain embodiments, the kit further comprises at least two ligands, in particular antibodies, each binding to one of CD45, CD61 , CD66b, ICAM-1 , CD86, CD1 1 c, CD7 and PD-L1 . In certain embodiments, the kit further comprises at least 5 ligands, in particular antibodies, each binding to one of CD45, CD61 , CD66b, ICAM-1 , CD86, CD1 1 c, CD7 and PD-L1 .

In certain embodiments of this aspect of the invention, the ligands comprise means for detection by flow cytometry. In certain embodiments, the ligands are fluorescently labeled.

Wherever alternatives for single separable features are laid out herein as "embodiments", it is to be understood that such alternatives may be combined freely to form discrete embodiments of the invention disclosed herein. The invention is further illustrated by the following examples and figures, from which further embodiments and advantages can be drawn. These examples are meant to illustrate the invention but not to limit its scope.

Brief description of the figures

Fig. 1 shows the stratification of responders and non-responders and identification of differences in immune cell populations using mass cytometry. (A) Experimental setup for melanoma patient sample processing using metal-labeled antibodies and acquisition by mass cytometry. (B) Dendrogram tree built on hierarchical clustering using Ward linkage of the normalized median marker expression from single cells of patient PBMCs. Bars on top of the heatmap represent individual samples from responders (green) versus non-responders (red). (C) Heatmap of significantly differentially expressed markers between responders and non- responders before and 12 weeks after therapy initiation, in pre-processed live single cells. Heat scale shows normalized median marker expression ranging from under-expressed (blue) to over-expressed (orange) where changes in marker expression between responders and non-responders was significant (p=<0.1 ). Colored bars on top of the heatmap represent individual samples from responders (green) and non-responders (red). (D) Cells from healthy donors and patients were used as an input for the FlowSOM algorithm. Thirteen algorithm- chosen markers were used to form 7 machine-assisted clusters. Visualization of 15Ό00 events in non-responders (NR), responders (R), and healthy donors (HD) using the tSNE algorithm. The heatmap represents the expression of respective marker within a cellular cluster and was used to annotate clusters, which were overlaid in color code (panel on the right). (E) Direct comparison of cluster frequencies in healthy donors (HD, black), non-responders (NR, pink) and responders (R, green) from batch 1 (circles) and batch 2 (triangles) extracted from tSNE immune clusters in C. Numbers in brackets indicate p-values (^***p=<0.01 ).

Fig. 2 shows the differences in T cell activation status and in the frequency of the T cell subpopulations before and after 12 weeks of therapy in responders and non-responders. Heatmaps showing significantly different (p=<0.1 ) normalized median marker expression in responders (green bar on top) and non-responders (red bar on top) in CD4⁺ T cells (A) and CD8⁺ T cells (B) before and after therapy. (C and D) FlowSOM was used to generate indicated T cell subpopulations and resultant cluster frequencies from batch 1 (circles) and batch 2 (triangles) are plotted as in Fig. 1 E. Heat scale shows over-expression (orange) or under- expression (blue) of the respective marker. Numbers in brackets show p-values (^*p=<0.1 , ^***p=<0.01 ).

Fig. 3 shows the increased activation in CD4⁺ or CD8⁺ CD69⁺T^mem/eff cells after immunotherapy start in responders. CD4⁺ and CD8⁺ memory/effector T cells after therapy were extracted and activated polyclonally. Frequencies of PD-1 , CTLA-4, and cytokines in (A) CD4⁺ CD69⁺ T^mem/eff cells and (B) CD8⁺ CD69⁺ T^mem/eff cells in responders (green) and non-responders (pink) were compared. Healthy subjects (grey) served as controls. (C) A matrix using the afore-mentioned markers was created and cells were sorted into this matrix using FlowSOM. Shown are the significantly different combinations after comparing responders to non-responders in CD4⁺ T cells. (D) Bar graphs displaying differences in cluster frequencies derived from C. Samples from batch 1 are indicated by circles and from batch 2 by triangles (^*p=<0.1 , ^**p=<0.05, ^***p=<0.01 ).

Fig. 4 shows the patient stratification based on myeloid cell markers and expansion and enhanced activation of classical monocytes in responders. (A) Dendrogram tree built using hierarchical clustering and Ward linkage as in Fig 1 B. (B) Heatmap of significantly differentially expressed markers in the myeloid compartment (p=0.1 , CD3^"CD19^"). Heat represents median marker expression normalized to the mean of 0 and a standard deviation of 1. (C) Visualization of FlowSOM-generated myeloid clusters (CD3 negative CD19 negative) in non-responders (NR), responders (R) and health donors (HD) using tSNE. Per plot, 1 CT000 cells are displayed. CD7 and CD56 positive cells were excluded from analysis. The heatmap represents the expression of respective markers within a cellular cluster. (D) Frequencies of myeloid cells in healthy donors (HD, black, n=10), non-responders (NR, pink, n=9) and responders (R, green, n=1 1 ) extrapolated form B before and after therapy. (E) Heatmap comparing significantly differently expressed myeloid markers in CD14⁺ and CD33⁺ monocytes between responders and non-responders before and 12 weeks after therapy. Heat scale shows normalized median marker expression ranging from under expressed (blue) to over-expressed (orange). (F) Validation of results on a second independent cohort of 31 patients using flow cytometry and CD3, CD4, CD1 1 b, CD14, CD19, CD16, CD33, CD56, HLA-DR markers. Bars on top of the heatmaps represent individual samples from responders (green) versus non-responders (pink). Numbers in brackets show adjusted p-values (^**p=<0.01 ).

Fig. 5 shows the simultaneous detection of T cell differentiation and activation markers in blood. PBMCs from 5 healthy donors and 10 melanoma patients were stained with a panel of 31 antibodies and analyzed by mass cytometry. Biaxial mass cytometry plots show the staining quality by gating on a respective positive and negative cell population of the shown differentiation and activation marker. Each plot shows a representation of four independent experiments.

Fig. 6 shows the simultaneous detection of TH and CTL profiles in human blood. PBMC from melanoma patients were stimulated for 4 hours with PMA lonomycin in the presence of brefeldin A and monensin. Two-dimensional mass cytometry plots show a representative experiment of four independent experiments. Fig. 7 shows the characterization of the circulating myeloid compartment in the blood of melanoma patients. Shown are dot plots from mass cytometry staining panels on cells from human blood. Data is representative for one out of four independent experiments.

Fig. 8 shows the comparison of 30 clinical parameters including measure monocytes to progression free survival using a multivariate Cox-model on the patient samples used for discovery mass cytometry approach.

Fig. 9 shows the citrus analysis of the T cell panel. (A) Depiction of Model Error Rate; (B) significant different clusters extracted from the minimal cluster model; (C) spanning tree of most relevant clusters in the minimal cluster model; (D) heatmap of markers expressed in significant different clusters shown in B (R) responders, (NR) non-responders.

Fig. 10 shows the citrus analysis of the myeloid cell-enriched (CD3negCD19neg) panel. (A) Depiction of Model Error Rate; (B) significant different clusters extracted from the minimal cluster model; (C) spanning tree of most relevant clusters in the minimal cluster model; (D) heatmap of markers expressed in significant different clusters shown in B (R) resonders, (NR) non-responders.

Fig. 1 1 shows the FACS validation panel. PBMC from an independent, randomized, blinded patient cohort were stained for CD3, CD4, CD1 1 b, CD14, CD19, CD16, CD33, CD45RO, CD56, and HLA-DR, acquired and analyzed using the above gating strategy.

Fig. 12 shows the comparison of 30 clinical parameters plus monocyte frequencies from the validation experiment to progression free survival using a multivariate Cox-model on the FACS validation cohort.

Examples

Introduction

During chronic inflammation, such as that occurring in cancer, continuous antigen exposure increases expression of programmed cell death 1 (PD-1 ) on T effector cells, which is thought to blunt immune responses against cancer cells and to be a sign of exhaustion. Therefore, PD- 1 among other check-point inhibitors is being increasingly targeted to treat cancer. Immunotherapy with anti-PD-1 aims to block the interaction of tumor-reactive T cells with PD- 1 ligands (PD-L1 and PD-L2) expressed on various cells types including leukocytes and the tumor cells themselves. Clinical trials on PD-1 and PD-L1 blockade for patients with advanced melanoma have demonstrated consistent therapeutic responses, thus prompting their application to several other cancers. Recently, nivolumab, an anti-PD-1 monoclonal antibody, has been approved by the US FDA for the treatment of patients with metastatic melanoma, non-small cell lung carcinoma (NSCLC) and metastatic renal cell carcinoma. Nivolumab has additionally been approved to treat patients with metastatic squamous NSCLC and refractory Hodgkirf s lymphoma. Pembrolizumab has shown similar efficacy and is now FDA approved as a first line treatment drug for melanoma. Pembrolizumab is also effective in patients with NSCLC, advanced gastric cancer, advanced bladder cancer, head and neck cancer, classical Hodgkin^'s lymphoma and triple negative breast cancer. Additional anti-PD-L1 monoclonal antibodies (mAbs) have been developed for the treatment of advanced human cancers including metastatic urothelial bladder cancer.There does not appear to be a significant difference among PD-1/PD-L1 mAbs, however there are no current side-by-side comparison studies.

Despite these encouraging results, clinical outcomes still remain highly variable, with only a fraction of patients showing durable responses, some with early progression and others with late response, while the majority of treated patients show no clinical response. In addition, to date, few biomarkers are available for stratifying patients and identifying potential responders prior to therapy. Thus, predictive biomarkers are urgently needed to improve the identification of potential responders to anti-PD-1 immunotherapy, or patients who might benefit more from other treatment modalities. This would further reduce toxicity due to multiple non-beneficial treatments, optimize the therapeutic window, and diminish the cost of therapy. Immunohistochemistry for the evaluation of tumor-associated PD-L1 has been used in several clinical trials with the notion that positive PD-L1 expression would predict response to anti-PD1 therapy. However, positivity for PD-L1 did not always correlate with effective response, furthermore, responsiveness to anti-PD1 therapy has also been observed in patients with PD- L1 negative tumors. These observations together with the evident heterogeneity of PD-L1 expression, clearly indicate that the power of PD-L1 as a sole predictive biomarker is limited and alternative methods are critical to improve patient care.

Some recent reports used single-cell analysis to evaluate the expression of PD-1 and downstream signaling molecules on tumor infiltrating and circulating CD8⁺ T cells, with the aim of identifying predictive biomarkers. However, these approaches are hampered by the limited accessibility of patient material and low dimensionality and/or lack of systematic bioinformatics analysis pipelines. Single-cell analysis of human immune compartments has so far been limited by the parameters that can be visualized by flow cytometry. Due to spectral overlap, modern flow cytometers are restricted to <20 parameters that can be measured simultaneously; whereas, with no spectral overlap, mass cytometry can readily measure more than 34 markers and has been applied recently to characterize hematopoietic, myeloid and T cells in both humans and mice. Thus, mass cytometry is a powerful tool to uncover rare immune subsets or biomarker combinations that associate with a phenotype of interest, such as response to immunotherapy. The inventors used peripheral blood mononuclear cells (PBMC) from melanoma patients as a readily accessible and minimally invasive biopsy that has been shown to be more representative than tumor biopsies to probe immune signatures associated with responsiveness to anti-PD-1 immunotherapy. High dimensional, single cell mass cytometry was used along with live cell barcoding, an optimized immune marker panel and a customized, interactive bioinformatics pipeline to identify differences in baseline PBMCs from responders and non-responders to checkpoint inhibitor, anti-PD1 . This approach not only has the potential to identify informative predictive features but might also illuminate the biological mechanisms of differential therapeutic response. Examining the peripheral blood of cancer patients before anti-PD-1 treatment, the inventors found (i) significantly fewer effector memory CD4⁺ T cells and naive CD8⁺ T cells, (ii) elevated CD8⁺ central memory cells, and (iii) increased CD14⁺ activated classical monocytes in patients who later responded to anti-PD-1 immunotherapy. Moreover, by subsequently examining cytokine production in the T cell subsets, the inventors observed a significant increase in CTLA-4⁺, TNF- ⁺, PD-1 ⁺, granzyme-B⁺, and IL-2⁺ cells in responders after therapy.

Results

Stratification of responders versus non-responders using single cell mass cytometry

The aim of this study was to identify biomarkers that could be used to predict responsiveness to anti-PD-1 immunotherapy. The inventors took advantage of cryopreserved PBMCs isolated from the blood of a unique cohort of 20 melanoma patients before and 12 weeks after anti-PD- 1 immunotherapy (20 patients over two time points = total sample number 40), and 10 healthy controls.

Table 1. Characteristics of melanoma patients and healthy donors used for the biomarker discovery study. Numbers in parenthesis display the age range of subjects.

Healthy Donors

N 10

Age (years) 60.3 (46-71 )

Sex (male/female) 6/4

Melanoma patients

Responders Non-responders (NR)

(R)

N 1 1 9

Age (years) 62.0 (42-81 ) 57.8 (45-75)

Sex (male/female) 9/2 5/4

Pre-treatments Radiotherapy 6/11 5/9

Chemotherapy 3/11 3/9

Ipilimumab 9/11 7/9

Other 0/11 2/9

In order to discover biomarkers in the blood of these melanoma patients the inventors generated a snapshot of immune responses before and 12 weeks after anti-PD-1 immunotherapy on the cohort of 20 patients and 10 healthy donors shown in Table 1 . For the CyTOF analysis, PBMCs were thawed and stained with three mass cytometry panels (Fig. 1A and Table 3). For all three panels, one phenotypic and one functional T cell panel, as well as one myeloid panel, the same number of PBMCs was used and labeled with barcodes that combine 2 or 3 (out of 5) metal tags. The first staining panel contained 31 T cell markers to identify all major immune cell populations and cover all stages of T cell differentiation and activation (Table 3). After acquisition, each sample was debarcoded using Boolean gating. Staining quality was evaluated by defining a biological positive and negative control (Fig. 5).

After data pre-processing, the inventors performed a hierarchical clustering on normalized median marker expression values, in every patient before and after therapy (20 patients over two time points = 40 samples). As demonstrated in Fig. 1 B, two major clusters were observed. The first clade contained 100% responders (n=15/15), whereas the second clade consisted of 72% non-responders (n=18/25) and 28% responders (n=7/25). Thus, normalized median marker expression was sufficient to robustly separate most responders from non-responders. The unbiased clustering approach stratified the patients into responders and non-responders prior to therapy, which encouraged the inventors to analyze the dataset in depth.

Altered T cell memory compartment before therapy in responders

The inventors next tested the hypothesis that the changes in normalized median marker expression were driven by changes in the relative abundance of the various cell populations between responders and non-responders. Therefore, the inventors analyzed the differential median expression of the 29 markers, comparing responders and non-responders, before and after therapy initiation (Fig. 1 C). Significant increases in the expression of HLA-DR, CTLA-4, CD56, CD45RO, CD1 1 a, CD25, and CCR5 and down-regulation of CD3, CD27, CD28, CD127, and CD4 was observed in responders versus non-responders.

Next, the inventors sought to identify which cell populations, from afore mentioned markers, describes cellular differences in relative frequency between responders and non-responders. Markers were selected using the PCA informativeness score established by Levine et al. (Levine et al., Cell 162, 184-197, 2015), cells were clustered using the FlowSOM algorithm (Van Gassen et al., Cytometry Part A 87, 636-645, 2015; Weber, L. M. & Robinson, M. D., 2016, doi:10.1 101/047613) with consensus clustering and a two-dimensional t-stochastic neighbour embedding (tSNE) projection was used for visualization, as shown in Fig. 1 D (Maaten, L. V. D. & Hinton, G., Journal of Machine Learning Research 9, 2579-2605, 2008). Based on the marker intensities detected in the clusters, the inventors manually annotated the seven major cell populations (CD4 T cells, CD8 T cells, NK cells, NKT cells, B cells, γδΤ cells, and myeloid cells) and then separated them into the three groups. The inventors subsequently examined differences in frequencies in between groups of the identified clusters (Fig. 1 E) using a generalized mixed model (see Methods). In responders, the frequency of CD4⁺ T cells and CD8⁺ T cells was reduced, while the frequency of myeloid cells was significantly increased (p- values = 1 .55e-05, 1.74e-03 and 1.74e-03 respectively). The inventors also observed an increase in NKT cells and a decrease in γδΤ cells (p=3.07e-03 and 2.52e-03). These findings are consistent with previous reports showing that enhanced NKT cell frequency after immunotherapy correlated with positive clinical response in melanoma patients while increased γδΤ cells correlated with decreased clinical benefit.

Since T cells are the major targets of anti-PD-1 immunotherapy and given the altered T cell composition in responders before immunotherapy, the inventors next compared the normalized median marker expressions on T cells between non-responders and responders before and after therapy. CD4⁺ T cells in responders showed an up-regulation of CTLA-4, HLA- DR, CD69, BTLA, and CD1 1 a (Fig. 2A) already at baseline before therapy. CD8⁺ T cells in responders showed an increase in CD45RO, CTLA-4, CD62L, CD69, CD1 1 a, CCR4, BTLA, PD-1 , CCR6, HLA-DR and granzyme-B expression (Fig. 2B). Besides being a regulator during T cell expansion, CTLA-4 is also a marker of activated T cells. Further, the inventors found that T cell depletion in the peripheral blood of melanoma patients is more pronounced in responders compared to non-responders (Fig. 2C and D). This phenomenon may be due to their enhanced ability to migrate to the tumor site. Indeed, in the CD4⁺ T cell compartment of responder patients, the inventors also found an up-regulation of CD1 1 a, which has been shown to be essential for migration to lymph nodes and distal sites.

Since the inventors wanted to understand whether there are differences in T cell subpopulations, the inventors next extracted CD4⁺ T cells and CD8⁺ T cells from the FlowSOM- generated clusters in Fig. 1 C and subdivided them into CD45RO^"CD62L⁺ naive, CD45RO^" CD62L- effector cells (TE), CD45RO⁺CD62L^" effector memory (EM) cells, CD45RO⁺CD62L⁺ central memory (CM) cells or CD127^"CD25⁺ regulatory T cells (T_regs) using FlowSOM (Fig. 2C and D).

The inventors then compared the frequencies of resultant T cell sub-clusters between responders and non-responders before and 12 weeks after therapy. The patients who eventually responded to the therapy showed a significant reduction in the CD4⁺ EM T cells, as well as reduction in CD8⁺ naive T cells population at baseline and after treatment (p-values: 8.21 e-03, 6.95e-03). Additionally, the CD8⁺ T cell subpopulation of responders showed an increase in CM T cells before and after treatment (Fig. 2C and D). The inventors' findings are consistent with the current paradigm of peripheral T cell development in humans, supporting the notion that T cell development progresses linearly from naive via memory to effector cells along with the loss of some properties, such as the ability to self-renew, expand and persist but in turn gain effector function and tissue specificity in vivo. In this context, it has been shown that, following adoptive T cell transfer, differentiated memory T cells can differentiate into potent effectors in vivo following interaction with their cognate antigen.

Anti-PD-1 immunotherapy alters the properties within the T cell compartment

In order to compare the functional properties of T cells between non-responders and responders, the inventors designed a second panel to investigate cytokine production (Fig. 6) in polyclonally activated cells. PBMCs were processed as described above. Briefly, single cell suspensions were cultured for 4h in the presence of PMA/lonomycin, barcoded, stained, fixed and analyzed by mass cytometry. In order to get a functional profile from antigen-experienced T cells, activated CD69⁺ memory and effector T cells (T^mem/eff cells) were extracted and cytokine (IL-2, IL-4, IL-10, IL-13, IL-17A, GM-CSF, TNF-a, IFN-γ, Grz-B), PD-1 and CTLA-4 positive T cell subpopulations were identified. The inventors found no difference in cytokine production between responders and non-responders prior to therapy. However, after therapy the inventors found a significant up-regulation of PD-1 , IL-4, and granzyme-B in CD4⁺CD69^+mem/eff T cells in responders, while IL-17A-positive cells were less abundant (Fig. 3A). For CD8⁺CD69^+mem/eff T cells, an up-regulation of CTLA4 and granzyme-B was detected in responders (Fig. 3B). In order to link these signatures to a specific cell population, the inventors then created a matrix containing all possible marker combinations in CD4⁺CD69^+mem/eff T cells (Fig3C) or CD8⁺CD69^+mem/eff T cells (not shown). Using this approach, no differences in the CD8⁺ T cell subpopulations were found. Fig. 3D shows the different cell populations from this matrix when comparing CD4⁺ T cell subsets in responders to non-responders. For CD4⁺CD69^+mem/eff T cells, the inventors found 6 clusters to be increased and 2 clusters to be reduced in responders. Among the enlarged clusters, the most representative signature was CTLA-4⁺, granzyme-B⁺, TNF- a ⁺, PD-1 ⁺ and IL-2⁺, which was present in clusters 10, 16 and 19 (p-values= 2.37e-2, 4.22e-02, and 3.95e-02). The decrease of IL-2 in CD4⁺CD69^+mem/eff T cells from cluster 5 and the expansion of cluster 48 in responders reflect the higher activation status in the CD4⁺ T cell compartment that the inventors observed from panel 1. These findings indicate that anti-PD1 immunotherapy can efficiently shape the functional profile of CD4+ T cells and that this cell population might significantly contribute to the efficacy of anti-PD1 immunotherapy against melanoma. Importantly, Quezada and colleagues provided the first evidence for the anti-tumor activity of CD4⁺ T cells in a mouse model of melanoma showing that the transfer of a small numbers of homeostatically expanding naive CD4⁺ T cells can induce profound regression of established tumors. Given the shift of frequency from naive to central memory T cells in responders before therapy and the increase in CTLA-4, IFN-γ, granzyme-B and PD-1 after therapy, the inventors' findings indicate that anti-PD-1 immunotherapy supports functionally activated T cells.

Myeloid cells predict responsiveness to anti-PD-1 immunotherapy

Since the inventors found increased frequencies of myeloid cells in anti-PD-1 therapy responders before therapy (Fig. 1 D), the inventors generated a third panel (Fig. 7) and extracted live myeloid cells by manually gating out CD3⁺ and CD19⁺ cells and excluding CD7⁺ and CD56⁺ cells and marker expression from further analysis. Unsupervised clustering of normalized median marker expression values in myeloid cells again separated patients into two distinct clusters, with one clade being mostly composed of (86%) non-responders (n = 12/14) and the other clade consisting of 76% (n = 19/25) of responders (Fig. 4A).

The inventors next searched for changes in normalized median marker expression between non-responders and responders, before and 12 weeks after therapy, and the inventors found that 16 markers (i.e., CD86, HLA-DR, CD141 , ICAM-1 , CD1 1 c, PD-L1 , CD38, CD16, CD33, CD1 1 b, CD303, CD62L, CD1 c, CD64, CD14, and CD34) were significantly up-regulated in the myeloid compartment of responders (Fig. 4B).

Next, FlowSOM was used to subdivide the myeloid compartment into 4 major clusters, which were annotated as CD14⁺CD16^"HLA-DR^hi classical monocytes, CD14^"CD33^|0WCD1 1 b⁺HLA- DR'° myeloid cells, plasmacytoid dendritic cells (CD123⁺CD303⁺HLA-DR⁺CD1 1 c^" pDC) and classical CD1 c⁺CD1 1 c⁺HLA-DR⁺ dendritic cells (cDC, Fig. 4C). As shown in Fig. 4D, the frequency of CD14⁺ classical monocytes was significantly increased in responders before therapy, whereas the frequency of CD14^"CD33^|0WCD1 1 b⁺HLA-DR'° myeloid cells were decreased (both p-value=2.23e-02). Importantly, the expansion of CD14⁺ classical monocytes was maintained after anti-PD-1 therapy.

The higher frequency of CD14⁺ classical monocytes in responders before therapy is striking. In recent years, the role of myeloid cells in cancer has been extensively debated and numerous studies have addressed the role of the so called myeloid derived suppressor cells, which have been shown to arise during chronic inflammation. It has been proposed that high levels of myeloid cells with immunosuppressive features may lead to T cell dysfunction and failure to respond to immunotherapy. In line with this hypothesis, the lower frequency of CD14^" CD33^|0WCD1 1 b⁺HLA-DR'° myeloid cells in the inventors' melanoma patients agree with earlier reports showing an increase in the objective clinical responses and long-term survival. However, the phenotypic, morphological and functional heterogeneity of these cells generates confusion when investigating their roles in inflammatory responses.

In a final step, the inventors more specifically examined the marker expression of the myeloid cell clusters before and after therapy, as shown in Fig. 4C, and found an up-regulation of ICAM- 1 , CD1 1 b, CD1 1 c, HLA-DR, and PD-L1 in CD14⁺ classical monocytes of responders already before therapy (Fig. 4E). Interestingly, despite their reduced frequencies, CD14^"CD33⁺ myeloid cells also showed an activated phenotype by over-expressing HLA-DR, CD141 , CD33, CD1 1 c, CD1 1 b and CD86.

The inventors' finding that classical CD14⁺ monocytes are activated during anti-PD-1 immunotherapy was entirely unexpected, but there is evidence that monocytes correlate with sustained response to anti-CTLA-4 treatment. This report further described that the engagement of FcgRIIIA on CD16⁺ monocytes by CTLA-4 antibody resulted in Treg cell lysis. Furthermore, a study on untreated melanoma patients with high tumor burdens detected the presence of deregulated monocyte populations with a dramatic decrease of HLA-DR and inflammatory markers on intermediate (CD14⁺CD16⁺) and non-classical monocytes (CD14^" CD16⁺).

Finally, using a multivariate Cox-proportional hazards model, the inventors analyzed the association of 42 standard clinical variables plus the measure monocytes with progression- free survival (PFS) (Fig. 8) and used a logistic regression model for the association of these variables to best overall response (data not shown). Gender (hazard rate: 7.646; 95%CI: 3.037-12.225; P=0.01 ), hematocrit (hazard rate: 52.121 ; 95%CI: 10.991 -93.251 ; P=0.013) and albumin (hazard rate: -1.199; 95%CI: -2.239 - -0.159; P=0.024) levels were identified as independent factors associated with progression under anti-PD1 therapy. This finding is consistent with a large body of epidemiological data showing that albumin is an independent prognostic factor for survival in several cancers and that low serum albumin is associated with higher cancer-related mortality.

As an independent validation of the computational results, the inventors employed Citrus, which is a clustering-based supervised algorithm that identifies stratifying signatures, to compare the identified cell types and marker expression differences that could distinguish between non-responders and responders before therapy (Fig. 9 and 10). Citrus independently confirmed the reduction observed in the T cell compartment and the increase in the myeloid compartment before therapy, as shown in panels 1 and 3.

Validation of cellular immune signature by flow cytometry

To facilitate the translation of the inventors' observations into clinical practice, the inventors designed a flow cytometry-based validation panel using a reduced number of markers. The inventors selected a combination of markers that were significantly differentially expressed in Fig. 1 C and 4B and markers known from the literature to define the cellular composition in the blood (Fig. 1 1 ). A blinded validation was performed on PBMCs from a second independent cohort of 31 melanoma patients containing 15 responders and 16 non-responders before anti- PD-1 therapy (table 2). As for the previous set of data, the inventors assessed the correlation between commonly documented clinical factors and patients PFS under treatment, including the monocyte frequencies form the validation panel (Fig. 12). Only hemoglobin (hazard rate: - 0.78; 95%CI: -1 .47 - -0.1 ; P=0.025) and hematocrit (hazard rate: 358.28; 95%CI: 53.9-662.66; P=0.021 ) were independent factors associated with progression under anti-PD1 therapy. Table 2 - Characteristics of melanoma patients and healthy donors used for the validation study. Numbers in parenthesis display the age range of subjects.

Healthy Donors

N 14

Age (years) 63.4 (46-91 )

Sex (male/female) 7/7

Melanoma patients

Responders (R) Non-responders (NR)

N 15 16

Age (years) 58.9 (31-93) 61.9 (27-89)

Sex (male/female) 9/6 8/8

Pre-treatments

Radiotherapy 10/15 5/16

Chemotherapy 0/15 4/16

Ipilimumab 10/15 13/16

Melanoma Inhibitor 2/15 2/16

Other 0/15 5/16

The results obtained by this approach confirmed the decrease in T cells (CD3⁺ CD56^", p=1.67e- 02) and the increase of CD14⁺ monocytes (CD3^"CD19^"CD14⁺CD16^"HLA-DR⁺, p=1 .99e-06) before therapy in responders, as already shown by mass cytometry (Fig. 4F).

In the inventors' cohort of melanoma patients, the inventors not only discovered a previously unappreciated higher frequency of classical monocytes (CD14⁺CD16^") amongst responding patients before therapy, but the inventors also found higher levels of HLA-DR and ICAM-1 on these cells. Together with the finding that patients responding to the therapy have higher frequencies of central memory T cells in circulation and a more activated (CTLA-4⁺, TNF-a ⁺, PD-1 ⁺, granzyme-B⁺ and IL-2⁺) T cell compartment after therapy, the inventors' results suggest that the presence of highly activated classical monocytes may be a prerequisite for a successful response to anti-PD-1 immunotherapy. The inventors propose to investigate the signature of PD-1 responders further in larger cohorts, as well as across the different applications of checkpoint inhibitor therapy. Altogether, besides representing a potentially powerful clinical determinant of response, the inventors' findings may help to elucidate the mechanisms underlying anti-PD-1 activity.

Material and Methods

Patient Samples

Fifty-one cryopreserved peripheral blood mononuclear cells (PBMC) samples of melanoma patients before and about 12 weeks after (median: 84 days, range: 23-162 days, average: 87.3 days) anti-PD-1 immunotherapy initiation were supplied by the Department of Dermatology, University Hospital Zurich, Switzerland (see Tablel ). Response was defined as the patient's Best Overall Response in the course of treatment. That is, the responder group comprises every patient who showed signs of clinical benefit within the first 15 weeks of treatment. Clinical benefit was defined as non-progression (including stable disease). The non-responder group included every patient who discontinued treatment due to significant disease progression or showed signs of progression after the first 15 weeks of treatment. Progression was defined as either a significant increase in tumor size, new metastatic sites, or the need to treat the patient with a secondary treatment such as radiotherapy. As healthy controls, age- and sex-matched PBMCs were acquired from the Red Cross Blood Bank, Zurich, Switzerland. All human biological samples were collected after written informed consent of the patients and with approval of the Local Ethics Committee (Kantonale Ethikkommission Zurich, KEK-ZH authorization Nr. 2014-0425) in accordance to GCP guidelines and the Declaration of Helsinki. Patient data and analysis

Standard clinical parameters (30) were measured in responders and non-responders that were collected before and after anti-PD-1 treatment. To assess the potential correlation between PFS and any of the clinical variable collected the inventors performed a Cox proportional hazard regression model. Age was considered as a continuous variable, whereas all other variable were dichotomized using clinical limit of normal as cut-offs. Candidate prognostic factors with a significant p value (<0.05) were then included in the multivariate analysis. In order to confirm findings from the cox model the inventors tested the same clinical parameters for differences between responders and non-responders using the linear mixed models (LMM) described in Data Analysis below (data not shown). The function coxph in the R package survival was used to fit a Cox proportional hazards regression model to the data.

Stimulations, stainings, and mass cytometry acquisition Cryopreserved PBMCs were thawed, incubated for 10 minutes in pre-warmed complete RPMI (RPMI, 10% FBS, Glutamine, Penicillin and Streptomycin) containing 200 g/ml DNAse, spun down, washed in cRPMI and in some cases rested overnight at 37°C at high cell densities (10⁷/mL). Cells from each sample were washed, counted and seeded in 96-well plates. Cells were seeded into 96-well non-tissue culture treated round bottom plates (BD Falcon) and left untreated or stimulated for 4 hours with 150 ng/ml phorbol-12-myristate-13-acetate (PMA) and 1 mM ionomycin in the presence of 10 μg/mlBrefeldin A (Sigma) and monensin (BD). For live cell barcoding cells were transferred into V-bottom plates (Costar) washed in cold FACS buffer (PBS + 2% FCS + 2mM EDTA + 0.05% sodium azide) and incubated for 15 minutes at 37°C with a combination of sample unique metal-labeled anti-human CD45 antibodies. Cells were then washed twice with ice-cold FACS buffer and incubated with 200 μΜ Cisplatin-Pt-198 (Fluidigm) for 2 minutes at room temperature. Cells were washed and surface proteins were stained with antibody at 37°C for 15 minutes and for an additional 10 minutes at 4°C. Cells were washed with FACS buffer and for intracellular staining some samples were permeabilized using cytofix/cytoperm-buffer (BD) for 20 minutes on ice and stained with an intracellular antibody cocktail (table 2) for 40 min on ice. Finally, cells were incubated over night with 250 nM iridium intercalator (Fluidigm) to label cellular DNA. Subsequently, cells were washed with PBS and with distilled water. Mass cytometry acquisition was performed on a CyTOF2.1 (Helios) instrument.

Antibody Conjugation

Purified antibodies lacking carrier proteins were purchased from the companies listed in Supplementary Table 1 . Antibody conjugation was performed using a metal-labeling kit (Fluidigm).

Data Analysis

After mass cytometry acquisition, beads were removed and samples were normalized using the standalone MATLAB normalizer (Version 2013b) (Finck, R. et al., Cytometry Part A 83, 483-494, 2013). To ensure data quality, marker expression was controlled in FlowJo (Version 10.1 r5). Each patient sample containing a unique metal barcode and was de-barcoded using Boolean gating in FlowJo. All analyses on CyTOF data was performed after arcsinh (with base equal to 5) transformation of marker expression. In the following, the inventors developed a custom R workflow in order to discover different biomarkers when comparing marker expression between responders and non-responders

(https://github.com/gosianow/carsten_cytof_code). All markers were included in the analysis and samples with less than 50 cells were excluded. Differential marker expression analysis was performed by fitting linear mixed models (LMM) using the Ime4 R package (Bates, D. et al., Journal of Statistical Software 67, 1-48, 2015). The median marker expression is a response variable (y) and the experimental group response (non-responder, responder or healthy donor) is a fixed effect. To account for any batch effects among samples, the inventors use each individual experiment as an additional fixed variable (batch). The inventors account for the fact that samples are paired (the same sample measured before and after therapy) by introducing the patient ID as a random effect variable. To test for differences between responders and non-responders the inventors used the generalized linear hypothesis (glht) function from the multcomp R package (Hothorn, T. et al., Biometrical Journal 50, 346-363, 2008) to test for the four following contrasts: (1 ) the difference in marker expression between responders and non-responders before therapy, (2) differences after therapy, (3) overall differences in both combined and (4) an interaction that is comparing differences before and after therapy. Except for functional components (Fig. 3), the inventors noted that in almost all cases the therapy did not have an impact on the observed significant differences. Based on this observation and in order to gain power, the inventors report results of the overall differences between responders and non-responders. To account for multiple cluster comparison, the inventors adjusted the resulting p-values using the Benjamini-Hochberg procedure of multiple-testing correction.

Differential marker expression is visualized using heatmaps as the change between responders and non-responders for significant markers (adjusted p-value < 0.1 ). Colours represent normalized median marker expressions to mean of 0 and standard deviation of 1 . To rank markers according to their importance, the inventors used the feature-scoring algorithm based on principal component analysis (PCA) from Levine et al., which identifies the non-redundant markers in each patient, while capturing the overall diversity. Top scoring (Levine PCA score averaged across samples) markers were used for subsequent clustering and dimension reduction analysis.

In order to cluster single cell data, the inventors used the SOM function from the FlowSOM R package³⁹ and ConsensusClusterPlus function from ConsensusClusterPlus R package (Wilkerson, M. D. & Hayes, D. N., Bioinformatics 26, 1572-1573, 2010), a combination of methods that is one of the best performing clustering approaches (Weber & Robinson, 2016). In a first step, the inventors used Flow SOM to assign cells to a 10 times 10 grid according to their similarity using the self-organizing map (SOM) algorithm. In the second step, the resulting 100 codes, vectors of marker expression representing the 100 grid nodes, were clustered using ConsensusClusterPlus hierarchical clustering with average linkage. Since the inventors knew the mapping between cells and nodes, the inventors could reconstruct the final clustering for each individual cell. The inventors applied ConsensusClusterPlus to cluster the codes into a range of clusters from 2 to 20 and to calculate a score (delta area), which the inventors used to define the appropriate number of clusters present in the data based on the so called elbow criterion. For data visualization, the inventors used tSNE dimension reduction, to represent the annotated cell populations in a 2D map (Maaten & Hinton, 2008).

In order to analyze differences in relative cell population abundance (frequency) between responders and non-responders to anti-PD-1 therapy, the inventors performed analysis analogous to differential marker expression analysis described above. Here, the response variable (y) was the number of cells in a given cluster in each sample, and instead of a LMM, a generalized linear mixed model (GLMM) with the binomial family was applied.

Cytokine analysis based on a bimatrix

For the selected subpopulations, the inventors investigated changes in cytokine production between responders and non-responders. The strategy described above, based on the comparison of median marker expression or abundance of annotated clusters, was not sensitive enough. Thus, the inventors exploited a new type of analysis based on a so called bimatrix. Bimatrix is a binary matrix with rows representing cells and columns corresponding to the cytokines of interest where each entry encodes whether a cell is positive (1 ) or negative (0) for a given cytokine. Thresholds for defining the positive status of a cell were defined for each batch of data individually by investigating expression profiles in FlowJo. Subsequently, the inventors performed two types of comparisons. First, the differential frequency analysis based on GLMM which compare the abundance of positive cells in responders and non- responders for each individual cytokine (Fig. 3A and 3B). For the second analysis, the inventors considered an entire cytokine set profile of each cell. Cells described by the bimatrix were clustered by using the SOM method into 49 clusters (7 times 7 grid) to generate profiles of the cytokine production (Fig. 3C) and the relative abundance of these profiles was compared between responders and non-responders using the GLMM approach described above.

Validation by flow cytometry

For validation of the CyTOF data, the inventors generated a flow cytometry panel as described above. In brief, the panel was based on a combination of markers with significantly different expression taken from Fig. 1 C and 4B and markers known from literature to define the cellular composition in the blood. Validation samples were taken in a blinded fashion on PBMCs from a second independent cohort of 31 melanoma patients containing 15 responders and 16 non- responders before anti-PD-1 therapy. After thawing, cell suspensions were stained in staining buffer (PBS, 5mM EDTA, 0.5% BSA) containing Fc-block (Miltenyi) with the following antibody cocktail (clones in brackets, all from Biolegend until noted otherwise): CD1 1 b-BrilliantViolett (BV) 421 (ICRF44), CD14-PE (HCD14), HLA-DR-FITC (L243), CD4-BV71 1 (OKT4), CD33- BV605 (WM53), and Live/dead-stain-NearlnfraRed. CD56-Pe-Cy7 (NCAM1 ) and CD1 1 c- AlexaFluor700 (B-Ly6) were both from BD Biosciences, CD16-APC (3G8) from ThermoFischer and CD45RO-ECD (2H4LDH1 1 LD89(2H4), from Beckman Coulter. At least 100Ό00 live cells were acquired using Diva software on a Fortessa flow cytometer (BD) and analyzed using FlowJo software (TriStar). From FlowJo 3 the frequencies of three cell populations, which were CD3⁺CD4⁺ CD4⁺ T cells, CD3⁺CD4^" CD8⁺ T cells and CD14⁺CD16^"HLA-DR⁺ monocytes were extracted from the three treatment groups. For statistical testing, the inventors applied a generalized linear model (GLM) with beta family, using the glmmADMB R package (link: httpglmmadmb.r-forge.r-project.org, accessed: 12 December 2016), where the response y is an relative abundance (proportion) of a cell population in the sample. The contrast for the comparison between responders and non-responders was tested using the glht function and a Benjamini-Hochberg procedure was applied to correct the resulting p-values for multiple- testing.

Claims

1 . A method of assigning to a patient suffering from cancer a likelihood of being responsive to checkpoint inhibitor therapy, wherein said method comprises the steps of

a. determining in a blood sample obtained from said patient the relative frequency of classical monocytes, wherein said relative frequency is the percentage of classical monocytes on the number of peripheral blood mononuclear cells, b. comparing said relative frequency with a threshold and

c. attributing a high likelihood of being responsive to checkpoint inhibitor therapy to said patient if said relative frequency is above said threshold.

2. A method of assigning a patient suffering from cancer to checkpoint inhibitor therapy, wherein the method comprises the steps of

c. assigning said patient to checkpoint inhibitor therapy if said relative frequency is above said threshold.

3. The method according to claim 1 or 2, wherein said classical monocytes are characterized by expression of CD14⁺CD16^"CD33^hiHLA-DR^hi.

4. The method according to any one of the above claims, wherein said threshold is 1 .6 times ("x"), 1 .7 x, 1 .8 x, 1.9 x, 2.0 x, 2.1 x, or 2.2 x the average relative frequency of classical monocytes in a blood sample obtained from a healthy subject.

5. The method according to any one of the above claims, wherein said relative frequency of classical monocytes is determined by flow cytometry.

6. The method according to any one of the above claims, wherein said cancer is melanoma, in particular metastatic melanoma.

7. The method according to any one of the above claims, wherein said checkpoint inhibitor therapy comprises treatment with a checkpoint inhibitory agent, in particular said checkpoint inhibitor therapy is selected from the group comprising anti-PD-1 immunotherapy, anti-PD-L1 immunotherapy, anti-CTLA-4 immunotherapy, anti-TIM-3 immunotherapy, anti-Lag-3 immunotherapy, or a combination of said therapies.

8. The method according to any one of the above claims, wherein said checkpoint inhibitor therapy is anti-PD-1 immunotherapy, in particular anti-PD-1 immunotherapy comprising administration of an anti-PD-1 immunotherapy agent selected from the group comprising Nivolumab and Pembrolizumab.

9. The method according to any one of the above claims, wherein additionally, the level of a marker selected from expression of albumin, expression of c-reactive protein and relative frequency of immature granulocytes is determined in said blood sample and wherein an increase in said marker is predictive of said patient being responsive to checkpoint inhibitor therapy.

10. A checkpoint inhibitory agent or a checkpoint agonist agent for use in the therapy of cancer, wherein a patient receiving said checkpoint inhibitory agent or a checkpoint agonist agent is characterized by a relative frequency of CD14⁺CD16^"CD33^hiHLA-DR^hi monocytes determined as a percentage of classical monocytes in relation to the number of peripheral blood mononuclear cells determined in a blood sample obtained from said patient, and said relative frequency is above a threshold, wherein said threshold is 1 .6 x - 2.2 x the average relative frequency of classical monocytes in a blood sample obtained from a healthy subject.

1 1 . The checkpoint inhibitory agent or checkpoint agonist agent for use in the therapy of cancer according to claim 10, wherein said cancer is melanoma, in particular metastatic melanoma.

12. The checkpoint inhibitory agent for use in the therapy of cancer according to claim 10 or 1 1 , wherein said checkpoint inhibitory agent is selected from the group comprising an anti-PD-1 antibody, an anti-PD-L1 antibody, an anti-CTLA-4 antibody, an anti-TIM- 3 antibody, an anti-Lag-3 antibody, or a combination of said antibodies.

13. A method of monitoring the success of checkpoint inhibitor therapy, wherein said method comprises the steps of

a. determining in a blood sample obtained from a patient receiving said checkpoint inhibitor therapy the relative frequency of classical monocytes, wherein said relative frequency is the percentage of classical monocytes on the number of peripheral blood mononuclear cells,

b. comparing said relative frequency with a monitoring threshold and

c. classifying said checkpoint inhibitor therapy as successful if said relative frequency is above said monitoring threshold.

14. A diagnostic kit comprising ligands, in particular antibodies, binding to CD14, CD16, CD33 and HLA-DR.

15. The kit according to claim 14, wherein said kit further comprises ligands, in particular antibodies, binding to CD3, CD4, CD1 1 b, CD19, CD45RO and CD56.

16. The kit according to claim 14 or 15, wherein said kit further comprises at least one, particularly at least two, more particularly at least 5 ligands, in particular antibodies, binding to CD45, CD61 , CD66b, ICAM-1 , CD86, CD1 1 c, CD7 and PD-L1 .

17. The kit according to any one of claims 14 to 16, wherein said ligands comprise means for detection by flow cytometry.