The identification of blood-borne biomarkers correlating with melanoma patient survival remains elusive. Novel techniques such as mass cytometry could help to identify melanoma biomarkers, allowing simultaneous detection of up to 100 parameters. However, the evaluation of multiparametric data generated via time-of-flight mass cytometry requires novel analytical techniques because the application of conventional gating strategies currently used in polychromatic flow cytometry is not feasible. In this study, we have employed 38-channel time-of-flight mass cytometry analysis to generate comprehensive immune cell signatures using matrix boolean analysis in a cohort of 28 stage IV melanoma patients and 17 controls. Clusters of parameters were constructed from the abundance of cellular phenotypes significantly different between patients and controls. This approach identified patient-specific combinatorial immune signatures consisting of high-resolution subsets of the T cell, NK cell, B cell, and myeloid compartments. An association with superior survival was characterized by a balanced distribution of myeloid-derived suppressor cell-like and APC-like myeloid phenotypes and differentiated NK cells. The results of this study in a discovery cohort of melanoma patients suggest that multifactorial immune signatures have the potential to allow more accurate prediction of individual patient outcome. Further investigation of the identified immune signatures in a validation cohort is now warranted.

Metastatic melanoma represents the most aggressive type of skin cancer with high mortality rates and a short median survival. Nonetheless, some patients benefit from cancer therapy and survive long term. The identification of immunological biomarkers correlating with clinical outcome is crucial for predicting which patients will fall into this group. Peripheral blood biomarkers help to identify patients at high risk of relapse, as well as those most likely to benefit from a certain therapy, allowing their allocation to appropriate treatment. Studies thus far have mostly examined limited types of immune cells in isolation, rather than attempting to integrate multiple components. Thus, for example, mounting evidence indicates that myeloid-derived suppressor cells (MDSCs) (1) are important in tumor evasion and are negatively associated with the clinical course in several cancer entities (24). Many other cells are also involved in immune responses to tumors, mediating both anti- and procancer effects, including αβ T cells (2, 5, 6), γδ T cells (7), NK cells (8) and B cells (9). Here, we aimed to identify informative signatures by simultaneous analysis of all these immune cell types.

Thus far, immune monitoring studies have relied chiefly on ELISPOT or polychromatic flow cytometry (PFC). This latter technology, although well established, is limited in the number of independently assessable parameters. Time-of-flight mass cytometry (CyTOF), in contrast, currently enables up to 40 channels to be assessed by using metal isotope–tagged Abs (1012) and thereby the unique opportunity to assess a richly extended level of immune cell analysis. Nevertheless, the generation of large and complex CyTOF datasets poses challenges for data analysis, which will gain immensely in complexity as the number of usable channels increases. Here, we have taken an unbiased straightforward approach to this data analysis challenge using matrix boolean analysis (MBA) (13, 14). MBA does not require any modification of the raw data and is visually verifiable at each step, in contrast to algorithm-based approaches applied to such datasets (1517).

The aim of the current study was to use MBA to explore biomarker candidates among CyTOF-generated peripheral blood immune signatures of 28 melanoma patients with detailed clinical characterization and follow-up data. Hierarchical clustering of optimal melanoma-linked signatures identified associations with overall survival (OS) and an established biomarker, serum lactate dehydrogenase (LDH) (18).

Cryopreserved PBMCs from 28 stage IV melanoma patients (Dermatology Department, University Hospital, Tübingen) with unresectable distant metastases at the time of blood draw and with available follow-up data including LDH levels ± 2 wk of blood draw were included in this study. Nine patients had received no prior systemic treatment whereas 19 had a mixed treatment background. Seventeen matched individuals served as healthy controls. Cohort characteristics are shown in Table I. This study was approved by the local Ethics Committee (561/2014BO2).

Table I.
Patient characteristics
VariableCategoryPatients
Controls
n%n%
Age >58 y 14 50 53 
≤58 y 14 50 47 
Median age, y 58  60  
Gender Female 14 50 35 
Male 14 50 11 65 
LDH Elevated >250 12 43  
Normal ≤250 15 54 
Unknown 
Systemic treatments prior to blood draw Chemotherapy 18 64 
BRAF/MEK inhibitors 
Anti-CTLA-4 Ab 14 
Anti-PD-1 Ab 
Other 
No systemic treatments 32 
Systemic subsequent treatments Chemotherapy 32 
BRAF/MEK inhibitors 11 
Anti-CTLA-4 Ab 13 46 
Anti-PD-1 Ab 
Other 
No further treatments 21 
Tissue pattern Soft tissue 14 
Lung 21 
Other organs 18 64 
VariableCategoryPatients
Controls
n%n%
Age >58 y 14 50 53 
≤58 y 14 50 47 
Median age, y 58  60  
Gender Female 14 50 35 
Male 14 50 11 65 
LDH Elevated >250 12 43  
Normal ≤250 15 54 
Unknown 
Systemic treatments prior to blood draw Chemotherapy 18 64 
BRAF/MEK inhibitors 
Anti-CTLA-4 Ab 14 
Anti-PD-1 Ab 
Other 
No systemic treatments 32 
Systemic subsequent treatments Chemotherapy 32 
BRAF/MEK inhibitors 11 
Anti-CTLA-4 Ab 13 46 
Anti-PD-1 Ab 
Other 
No further treatments 21 
Tissue pattern Soft tissue 14 
Lung 21 
Other organs 18 64 

Established panels and protocols were used (Supplemental Table I) as described previously for a similar panel (19). Samples were acquired with an LSR-II (Becton Dickinson). FlowJo V9.3.2 (Tree Star) was used for data analysis (Supplemental Fig. 1A).

Panel-optimization led to an optimal Ab/isotope composition. Mass tag barcoding (20) was performed as described (15). Up to 15 barcoded samples were acquired as one multiplexed batch per day on a CyTOF I (DVS Sciences, Canada) as described earlier (11). Cryopreserved control donor PBMCs were included in each CyTOF run to control for potential interday changes (Supplemental Fig. 3).

Matrix boolean analysis.

CyTOF data were transformed into FCS files by CyTOF software and de-barcoded. FlowJo V9.3.2 was used for analysis. CD45+ single cells (cisplatin-neg191Ir-pos193Ir-pos) were gated for the lineage of interest followed by boolean-gating of the markers that defined the compartment-specific signatures. The output of this analysis is an unbiased matrix displaying all included parameters against each other.

A median threshold of 100 cells was defined for the matrices resulting from the boolean gating. Differences in the frequencies of cellular phenotypes were analyzed between patients and healthy controls (Mann–Whitney U testing with GraphPad V6.0d) to identify melanoma-associated patterns for the composition of immune signatures.

Clustering and correlation with clinical data.

Analysis was performed with Statistical Package for the Social Sciences V22.0.0.1 (IBM) separately for all five compartments (myeloid cells, αβ T cells, γδ T cells, NK cells, and B cells), applying three methods to identify correlations between cellular and clinical parameters. First, the complete linkage method (Euclidean distances) was used for hierarchical clustering of patients based on the pattern of immune signatures. Survival of patients assigned to each cluster (minimum n = 5) was compared with the remaining patients. Second, correlations of patient clusters with LDH levels, age, and gender were analyzed by Fisher’s exact test. Third, each phenotype with significant differences in frequencies between controls and patients was analyzed for associations with OS. Patients were allocated to two groups by dichotomization using median frequencies of each phenotype.

In addition to the separate analysis of all five compartments, hierarchical clustering was applied considering all phenotypes significantly different between patients and controls, independent of the cellular compartment, to generate a comprehensive immune signature (CIS). Correlations between CIS clusters and patient clinical data were investigated using the same methods as for the single compartments. A stepwise backward Cox regression model and Wald tests were used to estimate the relative impact of single OS-associated cluster models when analyzed together with LDH. Results are reported as hazard ratios (HRs).

All p values ≤0.05 were considered significant for all statistical analysis. Survival correlations were estimated by Kaplan–Meier and log-rank testing.

First, a panel of metal isotope–tagged monoclonal Abs was assembled to investigate a broad immune signature in melanoma patient PBMCs and healthy controls (Table I). Independent analysis of the basic cellular compartments of the same individuals was performed by PFC for CyTOF validation (for gating see Fig. 1A, Supplemental Fig. 1A), but due to the high complexity of markers that were included in this CyTOF approach, a full validation with PFC is currently not possible. Results of both methods were in agreement regarding the mean or individual frequencies of T cells, monocytes, and NK cells (Fig. 1B).

FIGURE 1.

Mass cytometry: validation of the method and the MBA approach. (A) Representative serial gating of mass cytometry data to assess the basic cellular compartments monocytes (CD14+), T cells (CD3+CD4+ and CD3+CD8+), and NK cells (CD3CD4CD8CD16+CD56dim). (B) Comparison of means of the subset frequencies displayed in (A), between polychromatic flow and mass cytometry. Data from both experiments are derived from the same lot of PBMC samples from the same patients. (C) Entire spectrum of immune signatures in the αβ T cell compartment (CD3+ γδ TCR), revealed by MBA. Phenotypes in both patients and controls reflect the expected differentiation phenotypes, varying from naive to different effector memory stages to TEMRA cells identified with common markers (CD27, CD28, CD45RA, CCR7, CD57). Additionally, CD95 and CD161 were used as markers of activation.

FIGURE 1.

Mass cytometry: validation of the method and the MBA approach. (A) Representative serial gating of mass cytometry data to assess the basic cellular compartments monocytes (CD14+), T cells (CD3+CD4+ and CD3+CD8+), and NK cells (CD3CD4CD8CD16+CD56dim). (B) Comparison of means of the subset frequencies displayed in (A), between polychromatic flow and mass cytometry. Data from both experiments are derived from the same lot of PBMC samples from the same patients. (C) Entire spectrum of immune signatures in the αβ T cell compartment (CD3+ γδ TCR), revealed by MBA. Phenotypes in both patients and controls reflect the expected differentiation phenotypes, varying from naive to different effector memory stages to TEMRA cells identified with common markers (CD27, CD28, CD45RA, CCR7, CD57). Additionally, CD95 and CD161 were used as markers of activation.

Close modal

The signatures of the T cell compartment for all individuals documents that the MBA approach is feasible and reflects the entire ensemble of T cell differentiation stages (Fig. 1C: naive, effector memory, central memory, TEMRA subsets) and potential activated subtypes, in accordance with expectations from the literature (21).

Heterogeneous signatures in both patients and healthy individuals were observed in all five separately analyzed compartments (Fig. 1C, Supplemental Fig. 2: myeloid cells, αβ T cells, γδ T cells, B cells, and NK cells). For correlation with clinical data, a stepwise analytical method was applied for all compartments. First, melanoma-associated signatures were defined, considering only phenotypes whose frequencies were significantly different between patients and controls. Next, hierarchical clustering was performed to identify subgroups of patients with distinct signatures (Figs. 24), serving as a basis for the following correlations with clinical data (Supplemental Table I, Table II).

FIGURE 2.

Immune signature of the myeloid compartment in stage IV melanoma patients. (A) Comparison of the frequencies of myeloid phenotypes (CD3CD19CD11b+CD33+CD14+) in patients and controls by the Mann–Whitney U test (p ≤ 0.05 taken as significant). (B) Correlation of OS with the phenotypes displayed, based on dichotomization by median frequencies via Kaplan–Meier analysis and log-rank testing (p ≤ 0.05 taken as significant). (C) Myeloid immune signature of patients based on the frequency of the phenotypes described in (A) and (B). The upper part of the heatmap displays the marker distribution of the phenotypes displayed in (A)–(C). Each line represents one patient. Hierarchical clustering was performed to identify groups of patients with similar phenotypic distribution. (D) Correlation of the clusters of patients identified in (C) with OS via Kaplan–Meier analysis and log-rank testing at p ≤ 0.05. Patients in the c1 branch (dominated by the MDSC-like phenotypes 3 and 6) have a poor OS.

FIGURE 2.

Immune signature of the myeloid compartment in stage IV melanoma patients. (A) Comparison of the frequencies of myeloid phenotypes (CD3CD19CD11b+CD33+CD14+) in patients and controls by the Mann–Whitney U test (p ≤ 0.05 taken as significant). (B) Correlation of OS with the phenotypes displayed, based on dichotomization by median frequencies via Kaplan–Meier analysis and log-rank testing (p ≤ 0.05 taken as significant). (C) Myeloid immune signature of patients based on the frequency of the phenotypes described in (A) and (B). The upper part of the heatmap displays the marker distribution of the phenotypes displayed in (A)–(C). Each line represents one patient. Hierarchical clustering was performed to identify groups of patients with similar phenotypic distribution. (D) Correlation of the clusters of patients identified in (C) with OS via Kaplan–Meier analysis and log-rank testing at p ≤ 0.05. Patients in the c1 branch (dominated by the MDSC-like phenotypes 3 and 6) have a poor OS.

Close modal
FIGURE 4.

Immune signatures of the NK cell compartment of stage IV melanoma patients. (A) Comparison of the frequencies of identified NK cell phenotypes (CD3CD14CD19CD16+CD56dim) in patients and controls by the Mann–Whitney U test (p ≤ 0.05). (B) Correlation of the phenotypes displayed in (A) with OS, based on dichotomization by median frequencies via Kaplan–Meier analysis and log-rank testing. (C) NK cell immune signature of patients based on the phenotypes described in (A) and (B); hierarchical clustering was performed to identify relationships between patients.

FIGURE 4.

Immune signatures of the NK cell compartment of stage IV melanoma patients. (A) Comparison of the frequencies of identified NK cell phenotypes (CD3CD14CD19CD16+CD56dim) in patients and controls by the Mann–Whitney U test (p ≤ 0.05). (B) Correlation of the phenotypes displayed in (A) with OS, based on dichotomization by median frequencies via Kaplan–Meier analysis and log-rank testing. (C) NK cell immune signature of patients based on the phenotypes described in (A) and (B); hierarchical clustering was performed to identify relationships between patients.

Close modal
Table II.
Correlation of clinical data with the identified clusters in the different compartments
Survivalap valueLDHbp value>250 Cluster (n)Rest (n)>250 Cluster, %Rest, %
Monocytes 0.187      
 c1 × c2 0.031 0.013 7/8 6/19 87.5 31.6 
 c1a × rest 0.009 0.006 6/6 7/21 100.0 33.3 
 c2a × rest 0.324 0.678 3/8 10/19 37.5 52.6 
 c2a1 × rest 0.092 0.648 2/6 11/21 33.3 52.4 
 c2b × rest 0.439 0.120 3/11 10/16 27.3 62.5 
 c2b1 × rest 0.270 0.209 2/8 11/19 25.0 57.9 
 c2b1a × rest 0.657 0.385 2/7 11/20 28.6 55.0 
 c2b1a1 × rest 0.733 1.000 2/5 11/22 40.0 50.0 
T cells 0.277      
 c1 × restc 0.405 0.041 9/23 4/4 39.1 100.0 
NK cells 0.414      
 c1 × c2 0.941 1.000 3/6 10/21 50.0 47.6 
 c2a × rest 0.659 0.420 3/9 10/18 33.3 55.6 
 c2a1 × rest 0.465 1.000 3/7 10/20 42.9 50.0 
 c2b × rest 0.930 0.252 7/11 6/16 63.6 37.5 
 c2b1 × restc 0.527 0.440 6/10 7/17 60.0 41.2 
 c2b1a1 × rest 0.208 0.098 0/9 13/23 0.0 56.5 
Comprehensive signature 
 c1 × c2 0.005 0.018 5/17 8/10 29.4 80.0 
 c1a × restc 0.495 0.120 3/11 10/16 27.3 62.5 
 c1b × rest 0.161 0.385 2/7 11/20 28.6 55.0 
 c1b1 × rest 0.078 0.648 2/6 11/21 33.3 52.4 
 c2a × rest 0.015 0.013 7/8 6/19 87.5 31.6 
 c2a1 × rest 0.012 0.033 6/7 7/20 85.7 35.0 
 c2a1a × rest 5.0 × 10−4 0.016 5/5 8/22 100.0 36.4 
Survivalap valueLDHbp value>250 Cluster (n)Rest (n)>250 Cluster, %Rest, %
Monocytes 0.187      
 c1 × c2 0.031 0.013 7/8 6/19 87.5 31.6 
 c1a × rest 0.009 0.006 6/6 7/21 100.0 33.3 
 c2a × rest 0.324 0.678 3/8 10/19 37.5 52.6 
 c2a1 × rest 0.092 0.648 2/6 11/21 33.3 52.4 
 c2b × rest 0.439 0.120 3/11 10/16 27.3 62.5 
 c2b1 × rest 0.270 0.209 2/8 11/19 25.0 57.9 
 c2b1a × rest 0.657 0.385 2/7 11/20 28.6 55.0 
 c2b1a1 × rest 0.733 1.000 2/5 11/22 40.0 50.0 
T cells 0.277      
 c1 × restc 0.405 0.041 9/23 4/4 39.1 100.0 
NK cells 0.414      
 c1 × c2 0.941 1.000 3/6 10/21 50.0 47.6 
 c2a × rest 0.659 0.420 3/9 10/18 33.3 55.6 
 c2a1 × rest 0.465 1.000 3/7 10/20 42.9 50.0 
 c2b × rest 0.930 0.252 7/11 6/16 63.6 37.5 
 c2b1 × restc 0.527 0.440 6/10 7/17 60.0 41.2 
 c2b1a1 × rest 0.208 0.098 0/9 13/23 0.0 56.5 
Comprehensive signature 
 c1 × c2 0.005 0.018 5/17 8/10 29.4 80.0 
 c1a × restc 0.495 0.120 3/11 10/16 27.3 62.5 
 c1b × rest 0.161 0.385 2/7 11/20 28.6 55.0 
 c1b1 × rest 0.078 0.648 2/6 11/21 33.3 52.4 
 c2a × rest 0.015 0.013 7/8 6/19 87.5 31.6 
 c2a1 × rest 0.012 0.033 6/7 7/20 85.7 35.0 
 c2a1a × rest 5.0 × 10−4 0.016 5/5 8/22 100.0 36.4 
a

Kaplan–Meier analysis.

b

Fisher’s exact test.

c

No associations identified when further dissected.

Myeloid compartment.

Lineage gating identified the myeloid compartment (CD3CD19CD14+CD11b+CD33+). CD4, CD15, CD16, and HLA-DR, representing established markers, were analyzed by MBA. Additionally, CD38, CD85j, CD95, CX3CR1, and C3aR were included. Seventeen phenotypes were identified (Supplemental Fig. 2). Eight of 17 phenotypes were significantly different between patients and controls (Fig. 2A). Frequencies of myeloid lineage cells per se did not correlate with patients’ OS (Table II, p = 0.187). Single phenotype analysis revealed a positive association only of phenotype 7 (CD4dim,CD15+,CD16,CD38+,CD85j+,CD95+,CX3CR1,C3aR,HLA-DR+) with OS (Fig. 2B: p = 0.005). This potentially APC-like phenotype had a lower abundance in patients.

Coexpression of CD14 and CD15 was found in 5 of 17 identified phenotypes within the myeloid compartment of both patients and healthy subjects. This finding is interesting in that MDSCs are usually defined as being either CD14+ or CD15+ (22). An additional PFC experiment (n = 18) verified the presence of this population (Supplemental Fig. 1B: identified via forward and side scatter) and identified ∼6% of all monocytic cells that share this characteristic. Thus we suggest CD15 as a potential discriminative marker for the phenotypic assessment not only of polymorphonuclear (PMN) MDSC-like phenotypes.

Hierarchical clustering of the myeloid signatures identified two clusters in the first instance (Fig. 2C: c1 and c2) of which the c1 branch correlated negatively with OS (Fig. 2D: c1: p = 0.031, c1a: p = 0.009). Cluster c1 is dominated by the phenotypes 3 and 6, reflecting a marker expression pattern widely associated with MDSCs (low/absent HLA-DR expression). In contrast, c2 is positively associated with OS and defined by a more even abundance of cells with an APC- (phenotype 7) and MDSC-like (phenotypes 3 and 6) phenotype.

The cluster c1 and its subcluster c1a additionally correlated with increased LDH levels (Table II, p = 0.013 and p = 0.006, respectively). Reciprocally, there was a trend toward a positive correlation with OS for c2a1 (Fig. 2D: p = 0.092).

We conclude that these identified signatures, consisting mainly of cells with MDSC- and APC-like phenotypes of various differentiation stages, confirm that the myeloid compartment is positively associated with a better course of disease in stage IV melanoma patients when it is abundant and composed of cells with a high phenotypic diversity, but has a strong negative prognostic potential when it is limited to dominantly suppressive phenotypes.

αβ T cells.

αβ T cells were defined as CD3+, γδ TCR-negative, and subgrouped via CD4, CD8, CD27, CD28, CD45RA, CCR7, CD57, CD95, and CD161. The heatmap in Fig. 1C displays 68 identified phenotypes for classical CD4+ or CD8+, as well as CD4/CD8 double-positive and double-negative T cells. Frequencies of 16 of the 68 identified phenotypes were significantly different in patients and controls (Fig. 3A) and were therefore included in the patients’ signature analysis (Fig. 3C). Phenotype 15 (CD27+CD28+CD45RACCR7CD57CD95+CD161+) represents an early-differentiated CD4 phenotype that was less frequent in melanoma patients (Fig. 3A, p = 0.047) and positively correlated with OS (Fig. 3B, p = 0.032).

FIGURE 3.

Immune signature of the αβ T cell compartment of stage IV melanoma patients. (A) Comparison of the frequencies of T cell phenotypes (CD3+ γδ TCR) in patients and controls by Mann–Whitney U testing at p ≤ 0.05. (B) Correlation of the phenotypes displayed in (A) with OS, based on dichotomization by median frequencies via Kaplan–Meier analysis and log-rank testing (p ≤ 0.05). (C) αβ T cell immune signatures of patients based on the phenotypes described in (A) and (B); hierarchical clustering was performed to identify relationships between patients.

FIGURE 3.

Immune signature of the αβ T cell compartment of stage IV melanoma patients. (A) Comparison of the frequencies of T cell phenotypes (CD3+ γδ TCR) in patients and controls by Mann–Whitney U testing at p ≤ 0.05. (B) Correlation of the phenotypes displayed in (A) with OS, based on dichotomization by median frequencies via Kaplan–Meier analysis and log-rank testing (p ≤ 0.05). (C) αβ T cell immune signatures of patients based on the phenotypes described in (A) and (B); hierarchical clustering was performed to identify relationships between patients.

Close modal

Clustering of the signatures revealed interpatient heterogeneity in the CD4 (Fig. 3C: phenotypes 30, 38 and 39) as well as in the CD8 compartment (phenotype 31), whereas the frequencies of the other phenotypes were homogeneous over the whole group of patients. There were no correlations of any of the identified signature clusters with OS, although the few patients that were not part of the c1 clusters (high abundance of early differentiated CD4 T cells that were CD95+) possessed higher LDH levels (Table II, p = 0.041). However, females were more common in the c1 subcluster c1b1 that is characterized by a moderate abundance of early and intermediate differentiated CD4 T cell phenotypes (p = 0.033). This was the only association with demographic factors throughout the entire study. Thus, by this analysis, the immune signatures of the αβ T cell compartment possess no prognostic potential for OS, despite the correlation of the single CD4+ phenotype 15.

γδ T cells.

MBA was performed on CD3+ γδ TCR+ T cells. They were subdivided into the main subsets by expression of Vδ1 or Vδ2 TCRs. CD27, CD28 and CD45RA were included for the identification of different memory phenotypes (Supplemental Fig. 2). No differences of the identified phenotypes were found between patients and controls, therefore justifying no further consideration in our systematic approach.

B cells.

B cells were predefined as CD19+ and further analyzed for CD27, CD38, and CCR7 expression by MBA. This combination allows a basic impression of the differentiation status—considering CD27CD38 as early-differentiated, CD27+CD38 as intermediate-memory, and CD27CD38+ as late-differentiated cells. CCR7 was used as a marker of Ag-experienced B cells, relocating to the T zone in lymphoid organs (23). Signatures were dominated by the presence of the early-differentiated phenotypes (Supplemental Fig. 2). No differences were observed between patients and controls and thus B cell phenotypes were also not considered further in this analysis.

NK cells.

The NK cell compartment was predefined as CD16+CD56dimCD3CD19CD14. Activation was examined via CD69 and CD38 for cytolytic activity. Perforin was assessed as a surrogate for basal lytic capacity, associated with a more differentiated stage, identified through the expression of CD57 (24). We also included CX3CR1 because there is evidence that its expression is associated with lytic capacity (25), as well as CD85j, an inhibitory receptor associated with tumor escape mechanisms (26). Finally, we included CD161, a negative regulatory component of the NK receptor complex (Supplemental Fig. 2).

Ten of the 16 identified phenotypes showed differences between patients and controls (Fig. 4A) and were therefore included in the signature analysis. On the single phenotype level, only phenotype 7, a cell subset with high cytotoxic potential (CD38+CD57+CD69CD85jCD161CX3CR1+Perforin+), which was significantly increased in melanoma patients (Fig. 4A: p = 0.0054), tended to correlate with OS (Fig. 4B: p = 0.069). None of the identified signature clusters was associated with OS (Fig. 4C). We found a trend for elevated LDH levels for the c2 subcluster c2b1a1 that is characterized by a moderate abundance of potentially activated NK cell phenotypes (Table II, p = 0.098).

All phenotypes of the respective compartments that were found to differ between patients and controls were selected for analysis of a CIS, to seek the closest association of individual patient immune status and survival time. Fig. 5A displays the signatures of myeloid cells, αβ T cells, and NK cells of all patients. Hierarchical clustering identified two main signatures in the first instance (c1 versus c2) correlating with OS (Fig. 5B: p = 0.005). Patients with the c1 CIS have a survival benefit compared with those with c2. These c1-CIS patients share the characteristic of having a very heterogeneous distribution of phenotypes in the myeloid compartment and no dominance of cells with MDSC-like phenotypes (as in c2), as well as high frequencies of effector CD4 T cells and differentiated NK cells (Fig. 5A, NK phenotypes 1 and 3). Further dissection of cluster c1 resulted in a tendency toward improved OS (Fig. 5B: p = 0.078) for patients in subcluster c1b1. Reciprocally, a poorer outcome was observed for patients with the c2 CIS, whereby c2a, c2a1, and c2a1a resulted in a significant survival disadvantage (p = 0.015, p = 0.012, p = 0.0005, respectively). Increased LDH levels were observed for c2, c2a, c2a1 and c2a1a (Table II, p = 0.018, p = 0.013, p = 0.033, p = 0.016, respectively).

FIGURE 5.

Association with OS. (A) Combination of the myeloid, αβ T cell and NK cell compartment to generate a comprehensive signature. Hierarchical clustering was used to identify similar distributions in the three compartments at the single patient level. (B) Kaplan–Meier analysis combined with log-rank testing showing a negative association of the c2 branch (i.e., in the myeloid section dominated by the previously reported MDSC-like phenotypes 3 and 6) with survival.

FIGURE 5.

Association with OS. (A) Combination of the myeloid, αβ T cell and NK cell compartment to generate a comprehensive signature. Hierarchical clustering was used to identify similar distributions in the three compartments at the single patient level. (B) Kaplan–Meier analysis combined with log-rank testing showing a negative association of the c2 branch (i.e., in the myeloid section dominated by the previously reported MDSC-like phenotypes 3 and 6) with survival.

Close modal

All OS-associated cluster models (myeloid compartment: c1 × c2 and c1a × rest; CIS: c1 × c2, c2a × rest, c2a1 × rest, and c2a1a × rest) were evaluated in combination with LDH using Cox regression analysis to identify the most informative immune signature. Of all tested variables, CIS cluster c2a1a had the highest prognostic capacity for identifying patients with the poorest outcomes (HR: 6.3; p = 0.004), even better than the established marker LDH (HR: 1.9; p = 0.150).

Hence, these detailed immune cell analyses derived by CyTOF from limited amounts of PBMCs allow the identification of comprehensive combinatorial immune signatures with potential prognostic relevance.

The PFC-based analysis of single cell subsets to investigate their prognostic or predictive potential in cancer is an established and successful methodology. Here we present a comprehensive consideration of multiple cellular phenotypes of different immune cell compartments, assembled into immune signatures, using the recently developed technique of mass cytometry. With advances in analytic technology, the combinatorial number of cellular markers used in minimal sample sizes is rapidly rising, allowing thus far unattainable insights into the complexity of immune cell distribution in the peripheral blood. This should assist in achieving a more detailed understanding of cancer surveillance and will add sophistication to existing models in clinical studies. For instance, in this discovery study the large number of markers and various cell subsets of different compartments included in the CIS was the best model for predicting patients OS, even better than the myeloid signature that was by itself predictive, according to multivariate analysis. Thus, high dimensional analytical approaches, as reported in this study, possess potentially greater power for predicting the course of disease than those that are based on any single cell subset phenotype. CyTOF results generated on leukemia (17) and multiple myeloma (27) samples have recently been used for phenotypic description. Here, we used a CyTOF approach with 28 markers, resulting in 228 possible combinations, requiring novel analytical methods. Unbiased analyses are indispensable but there is yet no agreement as to which approach is most appropriate. In contrast to clustering-based approaches such as spanning-tree progression analysis of density-normalized events or cluster identification, characterization, and regression (28), dimensionality reduction methods such as principal component analysis (11, 12) have the benefit of maintaining single cell resolution of resulting plots. One nonlinear version of dimensionality reduction, t-distributed stochastic neighbor embedding (29), might be particularly well suited to CyTOF data analysis (17). Combining this method with clustering approaches can aid in comparing the composition of cell populations (15, 16). However, despite their strengths in describing cellular composition, these approaches may be less powerful when it comes to searching for specific differences between sets of samples, which is why we have applied the MBA approach. This generates a large matrix, displaying each marker against the others. No conversion, no downsampling, or other modifications of the raw data are needed; execution is fairly simple and each identified phenotype is easily verifiable. Another advantage is that there is no stepwise reduction and evaluation of single markers in the consensus of the whole ensemble.

However, there are certain limitations to our approach. The phenotypic exploration of cancer-associated immune signatures ignores potential functional differences. Another critical issue in our study is the relatively high number of applied tests comparing phenotypes between patients and controls in a limited number of subjects for the composition of immune signatures. We did not perform correction for multiple testing as a methodology to reduce false-positive findings, as this procedure would increase the risk of rejecting potentially promising phenotypes. We do acknowledge that the resulting OS-associated immune signatures can only be considered as candidates, which need to be validated in future studies. Thus, in general, we consider the approach presented here as a novel tool for future biomarker discovery and immunomonitoring. Consistent with this, we were able to identify, in agreement with the literature, a broad range of T cell differentiation stages (21), and several cell subsets consistent with previously described MDSC subsets (22, 30). This suggests that this methodology yields results consistent with expectations. Moreover, the application of hierarchical clustering for identifying groups of patients with similar immune signatures followed by Kaplan–Meier survival analysis avoided the use of the commonly employed dichotomization method. This is important because the latter mostly uses median frequencies of the cellular phenotypes of interest, which could lead to artifactual cutoff points.

One focus was the analysis of the myeloid compartment due to the crucial importance of MDSCs for survival of patients with melanoma (2, 4, 31, 32). This was important because of the difficulty of unequivocally defining surface phenotypes of the most relevant subpopulations of such cells (22, 33, 34). Our analysis reflects the broad distribution of different phenotypes in this compartment. Interestingly, the myeloid immune signatures were the only ones observed that correlated with both OS and the single marker LDH. Consistent with previous reports, we found that patients had higher frequencies of commonly employed MDSC-associated subsets (phenotype 3 and 6, Fig. 2C) that are HLA-DRlow/neg (1, 30, 33). The identified signatures did reveal a negative association with OS of cluster branch c1 (Fig. 2C, 2D) that is dominated by these MDSC-associated subsets, confirming their prognostic role.

Interestingly, however, many HLA-DR+ phenotypes exist in the signatures (Fig. 2C compared with Supplemental Fig. 2), indicating that the Ag-presenting capacity could be influenced in patients, for example as described for dendritic cells (35). Patients with a more heterogeneous myeloid signature (c2 branch) may be less affected by MDSC-associated immune suppression, as suggested by their having a better course of disease and lack of correlation with elevated LDH levels. High frequencies of phenotype 7 (HLA-DR+), represented in the c2 cluster branch, were associated with prolonged OS in patients and were present at similar levels as in healthy controls. Correlation of the c1 branch with elevated LDH levels, itself known to be prognostic (18), supports the prognostic capacity of this relatively homogeneous cluster of signatures. These findings demonstrate that the myeloid compartment is not associated with a poor outcome per se, but seems to possess both positive and negative prognostic potential in combination with other components.

It is striking that despite the use of cryopreserved PBMCs, which contain only low amounts of PMN cells, coexpression of CD14 (general monocytic marker) and CD15 (usually a PMN MDSC-associated marker) was found and verified on ∼6% of all myeloid cells—a finding that requires detailed analysis in upcoming studies. Interestingly, this coexpression was observed in both patients and healthy individuals. A similar CD14+CD15+ MDSC-phenotype was found to be negatively associated with OS and progression-free survival in non-small cell lung cancer (36).

Cancer rejection is for the most part ascribed to αβ T cells (5, 37, 38). We analyzed frequencies of a broad spectrum of T cell subsets to investigate whether this compartment also possesses prognostic qualities. All phenotypes that were the basis of our T cell signature revealed a relatively homogeneous interpatient expression pattern. This is also reflected in the inconsistent clustering of the signatures. However, we did not observe T cell signatures that were associated with OS. Nonetheless, our analysis suggests that mainly the CD4+ memory T cell subsets are influenced by melanoma itself, as visualized in 16 phenotypes (but only five within the CD8+ cells) that differ between controls and patients (Fig. 3). Only the CD4 phenotype 15 that might reflect a Th17 regulatory phenotype (39) correlated positively with OS. The latter might act as a marker for the potency of the T cell compartment that is known to be impaired in many patients, for example via CTLA-4–dependent (40) or PD1-dependent (41) pathways.

No association of γδ T cell signatures with melanoma was observed in contrast to what we recently reported (42). We were unable to confirm the described OS associations of the Vδ1 compartment due to the low abundance of the latter and the limited amount of sample material available. Similar reasons might be applicable to the B cell compartment, albeit these cells play an important role when invading tumors (43) or in the production of tumor-specific Abs (44).

The involvement of NK cells in cancer is well described (8). NK-like cytotoxicity and its correlation with cancer risk was demonstrated in an 11 y follow-up study (45). Tumor-infiltrating NK cells were described as a positive prognostic marker in different carcinomas (4648). However, we found no significant correlations with OS. The lack of correlations of peripheral NK signatures with OS might be explained through alterations of the latter that are limited to the tumor itself, although the current study was restricted to the analysis of circulating immune cells. Future analysis and correlation of immune signatures with tumor-infiltrating immune cells is warranted.

In contrast to regular PFC, CyTOF allows simultaneous evaluation of all major immune cell compartments. Integrated CIS modeling, considering all melanoma-associated phenotypes (T cell, NK cell and myeloid compartment), resulted in the strongest association with patients’ OS (Fig. 5) using multivariate analysis. We find these associations particularly intriguing because they implicate networks of cell subsets rather than individual subsets that can be more readily identified by traditional approaches. They also pose a unique challenge in that they require a high-dimensional approach such as CyTOF for them to be identified. Patients with a CIS in the c1 branch (heterogeneous composition over all compartments) had a clear survival advantage over patients from the c2 branch (Fig. 5A). The c2 clusters additionally correlated with increased LDH levels. These c2 clusters are defined by a low diversity of phenotypes in the myeloid compartment with a prevalence of the MDSC-associated phenotypes and a more homogeneous distribution of phenotypes in the NK compartment.

The CIS pattern indicates that myeloid cells and particularly the abundance of MDSCs are crucial for the course of disease and most probably for the interaction with other major immune compartments that are known to be involved in cancer rejection. For instance, the differentiated NK phenotypes are relatively underrepresented in the signatures of patients with a poor course of disease, which might be explained through previously described potential negative regulatory interactions between MDSCs and NK cells (49). Interestingly, no major differences were found in the T cell compartment of the CIS (Fig. 5A). This could be explained by MDSC activity unaccompanied by a change in T cell frequency, or the induction of an anergic state of potentially tumor-reactive T cells through, for example, ligation of CTLA-4 (50). Data on interactions between T cell reactivity to tumor Ags in vitro and MDSC levels in vivo in these melanoma patients are consistent with this interpretation (2). Namely, patients clustered in c1 have a better OS than those in c2 (Fig. 5B, p = 0.005). Further dissection of the patients in the c2 branch allowed the identification of a well-defined subgroup with extraordinarily poor prognosis (c2a1a) (Fig. 5B).

Despite recent progress in analytical strategies, CyTOF data analysis remains challenging. We present a reproducible and comprehensive method circumventing the problems of algorithm-based approaches to define biomarker candidates. Our results further suggest that the disease course is reflected in patient-specific immune signatures. Combinatorial analysis as presented here might help to improve personalized immunotherapeutic approaches and could replace single phenotype biomarker approaches in future. Nonetheless, although our major findings are in overall agreement with the current literature, this is a discovery study requiring future validation in larger prospective cohorts composed of patients that received specific therapies, for instance checkpoint blockade therapy.

This work was supported by grants from the German Research Foundation (DFG Pa 361/22-1) (G.P.), the Baden-Württemberg Stipendium-Plus Program (K.W.-H.), and the Singapore Immunology Network (Agency for Science, Technology and Research) (A.L., E.W.N.), and by Singapore Immunology Network (Agency for Science, Technology and Research) immunomonitoring platform funding (E.W.N.). A.L. is a scholar of the International Society for Advancement of Cytometry.

The online version of this article contains supplemental material.

Abbreviations used in this article:

     
  • CIS

    comprehensive immune signature

  •  
  • CyTOF

    time-of-flight mass cytometry

  •  
  • HR

    hazard ratio

  •  
  • LDH

    lactate dehydrogenase

  •  
  • MBA

    matrix boolean analysis

  •  
  • MDSC

    myeloid-derived suppressor cell

  •  
  • OS

    overall survival

  •  
  • PFC

    polychromatic flow cytometry

  •  
  • PMN

    polymorphonuclear.

1
Gabrilovich
D. I.
,
Ostrand-Rosenberg
S.
,
Bronte
V.
.
2012
.
Coordinated regulation of myeloid cells by tumours.
Nat. Rev. Immunol.
12
:
253
268
.
2
Weide
B.
,
Martens
A.
,
Zelba
H.
,
Stutz
C.
,
Derhovanessian
E.
,
Di Giacomo
A. M.
,
Maio
M.
,
Sucker
A.
,
Schilling
B.
,
Schadendorf
D.
, et al
.
2014
.
Myeloid-derived suppressor cells predict survival of patients with advanced melanoma: comparison with regulatory T cells and NY-ESO-1- or melan-A-specific T cells.
Clin. Cancer Res.
20
:
1601
1609
.
3
Kalathil
S.
,
Lugade
A. A.
,
Miller
A.
,
Iyer
R.
,
Thanavala
Y.
.
2013
.
Higher frequencies of GARP(+)CTLA-4(+)Foxp3(+) T regulatory cells and myeloid-derived suppressor cells in hepatocellular carcinoma patients are associated with impaired T-cell functionality.
Cancer Res.
73
:
2435
2444
.
4
Walter
S.
,
Weinschenk
T.
,
Stenzl
A.
,
Zdrojowy
R.
,
Pluzanska
A.
,
Szczylik
C.
,
Staehler
M.
,
Brugger
W.
,
Dietrich
P. Y.
,
Mendrzyk
R.
, et al
.
2012
.
Multipeptide immune response to cancer vaccine IMA901 after single-dose cyclophosphamide associates with longer patient survival.
Nat. Med.
18
:
1254
1261
.
5
Weide
B.
,
Zelba
H.
,
Derhovanessian
E.
,
Pflugfelder
A.
,
Eigentler
T. K.
,
Di Giacomo
A. M.
,
Maio
M.
,
Aarntzen
E. H.
,
de Vries
I. J.
,
Sucker
A.
, et al
.
2012
.
Functional T cells targeting NY-ESO-1 or Melan-A are predictive for survival of patients with distant melanoma metastasis.
J. Clin. Oncol.
30
:
1835
1841
.
6
Zelba
H.
,
Weide
B.
,
Martens
A.
,
Derhovanessian
E.
,
Bailur
J. K.
,
Kyzirakos
C.
,
Pflugfelder
A.
,
Eigentler
T. K.
,
Di Giacomo
A. M.
,
Maio
M.
, et al
.
2014
.
Circulating CD4+ T cells that produce IL4 or IL17 when stimulated by melan-A but not by NY-ESO-1 have negative impacts on survival of patients with stage IV melanoma.
Clin. Cancer Res.
20
:
4390
4399
.
7
Déchanet-Merville
J.
2014
.
Promising cell-based immunotherapy using gamma delta T cells: together is better.
Clin. Cancer Res.
20
:
5573
5575
.
8
Waldhauer
I.
,
Steinle
A.
.
2008
.
NK cells and cancer immunosurveillance.
Oncogene
27
:
5932
5943
.
9
Yuan
J.
,
Adamow
M.
,
Ginsberg
B. A.
,
Rasalan
T. S.
,
Ritter
E.
,
Gallardo
H. F.
,
Xu
Y.
,
Pogoriler
E.
,
Terzulli
S. L.
,
Kuk
D.
, et al
.
2011
.
Integrated NY-ESO-1 antibody and CD8+ T-cell responses correlate with clinical benefit in advanced melanoma patients treated with ipilimumab.
Proc. Natl. Acad. Sci. USA
108
:
16723
16728
.
10
Ornatsky
O.
,
Baranov
V. I.
,
Bandura
D. R.
,
Tanner
S. D.
,
Dick
J.
.
2006
.
Multiple cellular antigen detection by ICP-MS.
J. Immunol. Methods
308
:
68
76
.
11
Newell
E. W.
,
Sigal
N.
,
Bendall
S. C.
,
Nolan
G. P.
,
Davis
M. M.
.
2012
.
Cytometry by time-of-flight shows combinatorial cytokine expression and virus-specific cell niches within a continuum of CD8+ T cell phenotypes. [Published erratum appears in 2013 Immunity 38: 198–199.]
Immunity
36
:
142
152
.
12
Bendall
S. C.
,
Simonds
E. F.
,
Qiu
P.
,
Amir
A. D.
,
Krutzik
P. O.
,
Finck
R.
,
Bruggner
R. V.
,
Melamed
R.
,
Trejo
A.
,
Ornatsky
O. I.
, et al
.
2011
.
Single-cell mass cytometry of differential immune and drug responses across a human hematopoietic continuum.
Science
332
:
687
696
.
13
Horowitz
A.
,
Strauss-Albee
D. M.
,
Leipold
M.
,
Kubo
J.
,
Nemat-Gorgani
N.
,
Dogan
O. C.
,
Dekker
C. L.
,
Mackey
S.
,
Maecker
H.
,
Swan
G. E.
, et al
.
2013
.
Genetic and environmental determinants of human NK cell diversity revealed by mass cytometry.
Sci. Transl. Med.
5
:
208ra145
.
14
Strauss-Albee
D. M.
,
Fukuyama
J.
,
Liang
E. C.
,
Yao
Y.
,
Jarrell
J. A.
,
Drake
A. L.
,
Kinuthia
J.
,
Montgomery
R. R.
,
John-Stewart
G.
,
Holmes
S.
,
Blish
C. A.
.
2015
.
Human NK cell repertoire diversity reflects immune experience and correlates with viral susceptibility.
Sci. Transl. Med.
7
:
297ra115
.
15
Becher
B.
,
Schlitzer
A.
,
Chen
J.
,
Mair
F.
,
Sumatoh
H. R.
,
Teng
K. W.
,
Low
D.
,
Ruedl
C.
,
Riccardi-Castagnoli
P.
,
Poidinger
M.
, et al
.
2014
.
High-dimensional analysis of the murine myeloid cell system.
Nat. Immunol.
15
:
1181
1189
.
16
Shekhar
K.
,
Brodin
P.
,
Davis
M. M.
,
Chakraborty
A. K.
.
2014
.
Automatic classification of cellular expression by nonlinear stochastic embedding (ACCENSE).
Proc. Natl. Acad. Sci. USA
111
:
202
207
.
17
Amir el
A. D.
,
Davis
K. L.
,
Tadmor
M. D.
,
Simonds
E. F.
,
Levine
J. H.
,
Bendall
S. C.
,
Shenfeld
D. K.
,
Krishnaswamy
S.
,
Nolan
G. P.
,
Pe’er
D.
.
2013
.
viSNE enables visualization of high dimensional single-cell data and reveals phenotypic heterogeneity of leukemia.
Nat. Biotechnol.
31
:
545
552
.
18
Balch
C. M.
,
Gershenwald
J. E.
,
Soong
S. J.
,
Thompson
J. F.
,
Atkins
M. B.
,
Byrd
D. R.
,
Buzaid
A. C.
,
Cochran
A. J.
,
Coit
D. G.
,
Ding
S.
, et al
.
2009
.
Final version of 2009 AJCC melanoma staging and classification.
J. Clin. Oncol.
27
:
6199
6206
.
19
Wistuba-Hamprecht
K.
,
Pawelec
G.
,
Derhovanessian
E.
.
2014
.
OMIP-020: phenotypic characterization of human γδ T-cells by multicolor flow cytometry.
Cytometry A
85
:
522
524
.
20
Bodenmiller
B.
,
Zunder
E. R.
,
Finck
R.
,
Chen
T. J.
,
Savig
E. S.
,
Bruggner
R. V.
,
Simonds
E. F.
,
Bendall
S. C.
,
Sachs
K.
,
Krutzik
P. O.
,
Nolan
G. P.
.
2012
.
Multiplexed mass cytometry profiling of cellular states perturbed by small-molecule regulators.
Nat. Biotechnol.
30
:
858
867
.
21
Farber
D. L.
,
Yudanin
N. A.
,
Restifo
N. P.
.
2014
.
Human memory T cells: generation, compartmentalization and homeostasis.
Nat. Rev. Immunol.
14
:
24
35
.
22
Poschke
I.
,
Kiessling
R.
.
2012
.
On the armament and appearances of human myeloid-derived suppressor cells.
Clin. Immunol.
144
:
250
268
.
23
Rasmussen
T.
,
Lodahl
M.
,
Hancke
S.
,
Johnsen
H. E.
.
2004
.
In multiple myeloma clonotypic CD38- /CD19+ / CD27+ memory B cells recirculate through bone marrow, peripheral blood and lymph nodes.
Leuk. Lymphoma
45
:
1413
1417
.
24
Lopez-Vergès
S.
,
Milush
J. M.
,
Pandey
S.
,
York
V. A.
,
Arakawa-Hoyt
J.
,
Pircher
H.
,
Norris
P. J.
,
Nixon
D. F.
,
Lanier
L. L.
.
2010
.
CD57 defines a functionally distinct population of mature NK cells in the human CD56dimCD16+ NK-cell subset.
Blood
116
:
3865
3874
.
25
Hamann
I.
,
Unterwalder
N.
,
Cardona
A. E.
,
Meisel
C.
,
Zipp
F.
,
Ransohoff
R. M.
,
Infante-Duarte
C.
.
2011
.
Analyses of phenotypic and functional characteristics of CX3CR1-expressing natural killer cells.
Immunology
133
:
62
73
.
26
Favier
B.
,
Lemaoult
J.
,
Lesport
E.
,
Carosella
E. D.
.
2010
.
ILT2/HLA-G interaction impairs NK-cell functions through the inhibition of the late but not the early events of the NK-cell activating synapse.
FASEB J.
24
:
689
699
.
27
Hansmann
L.
,
Blum
L.
,
Ju
C. H.
,
Liedtke
M.
,
Robinson
W. H.
,
Davis
M. M.
.
2015
.
Mass cytometry analysis shows that a novel memory phenotype B cell is expanded in multiple myeloma.
Cancer Immunol. Res.
3
:
650
660
.
28
Bruggner
R. V.
,
Bodenmiller
B.
,
Dill
D. L.
,
Tibshirani
R. J.
,
Nolan
G. P.
.
2014
.
Automated identification of stratifying signatures in cellular subpopulations.
Proc. Natl. Acad. Sci. USA
111
:
E2770
E2777
.
29
van der Maaten
L.
,
Hinton
G.
.
2008
.
Visualizing data using t-SNE.
J. Mach. Learn. Res.
9
:
2579
2605
.
30
Filipazzi
P.
,
Valenti
R.
,
Huber
V.
,
Pilla
L.
,
Canese
P.
,
Iero
M.
,
Castelli
C.
,
Mariani
L.
,
Parmiani
G.
,
Rivoltini
L.
.
2007
.
Identification of a new subset of myeloid suppressor cells in peripheral blood of melanoma patients with modulation by a granulocyte-macrophage colony-stimulation factor-based antitumor vaccine.
J. Clin. Oncol.
25
:
2546
2553
.
31
Solito
S.
,
Falisi
E.
,
Diaz-Montero
C. M.
,
Doni
A.
,
Pinton
L.
,
Rosato
A.
,
Francescato
S.
,
Basso
G.
,
Zanovello
P.
,
Onicescu
G.
, et al
.
2011
.
A human promyelocytic-like population is responsible for the immune suppression mediated by myeloid-derived suppressor cells.
Blood
118
:
2254
2265
.
32
Martens
A.
,
Wistuba-Hamprecht
K.
,
Geukes Foppen
M.
,
Yuan
J.
,
Postow
M. A.
,
Wong
P.
,
Romano
E.
,
Khammari
A.
,
Dreno
B.
,
Capone
M.
, et al
.
2016
.
Baseline peripheral blood biomarkers associated with clinical outcome of advanced melanoma patients treated with ipilimumab.
Clin. Cancer Res.
22
:
2908
2918
.
33
Montero
A. J.
,
Diaz-Montero
C. M.
,
Kyriakopoulos
C. E.
,
Bronte
V.
,
Mandruzzato
S.
.
2012
.
Myeloid-derived suppressor cells in cancer patients: a clinical perspective.
J. Immunother.
35
:
107
115
.
34
Bronte
V.
,
Brandau
S.
,
Chen
S. H.
,
Colombo
M. P.
,
Frey
A. B.
,
Greten
T. F.
,
Mandruzzato
S.
,
Murray
P. J.
,
Ochoa
A.
,
Ostrand-Rosenberg
S.
, et al
.
2016
.
Recommendations for myeloid-derived suppressor cell nomenclature and characterization standards.
Nat. Commun.
7
:
12150
.
35
Chevolet
I.
,
Speeckaert
R.
,
Schreuer
M.
,
Neyns
B.
,
Krysko
O.
,
Bachert
C.
,
Van Gele
M.
,
van Geel
N.
,
Brochez
L.
.
2015
.
Clinical significance of plasmacytoid dendritic cells and myeloid-derived suppressor cells in melanoma.
J. Transl. Med.
13
:
9
.
36
Vetsika
E. K.
,
Koinis
F.
,
Gioulbasani
M.
,
Aggouraki
D.
,
Koutoulaki
A.
,
Skalidaki
E.
,
Mavroudis
D.
,
Georgoulias
V.
,
Kotsakis
A.
.
2014
.
A circulating subpopulation of monocytic myeloid-derived suppressor cells as an independent prognostic/predictive factor in untreated non-small lung cancer patients.
J. Immunol. Res.
2014
:
659294
.
37
Kvistborg
P.
,
Shu
C. J.
,
Heemskerk
B.
,
Fankhauser
M.
,
Thrue
C. A.
,
Toebes
M.
,
van Rooij
N.
,
Linnemann
C.
,
van Buuren
M. M.
,
Urbanus
J. H.
, et al
.
2012
.
TIL therapy broadens the tumor-reactive CD8(+) T cell compartment in melanoma patients.
OncoImmunology
1
:
409
418
.
38
Braumüller
H.
,
Wieder
T.
,
Brenner
E.
,
Aßmann
S.
,
Hahn
M.
,
Alkhaled
M.
,
Schilbach
K.
,
Essmann
F.
,
Kneilling
M.
,
Griessinger
C.
, et al
.
2013
.
T-helper-1-cell cytokines drive cancer into senescence.
Nature
494
:
361
365
.
39
Weaver
C. T.
,
Harrington
L. E.
,
Mangan
P. R.
,
Gavrieli
M.
,
Murphy
K. M.
.
2006
.
Th17: an effector CD4 T cell lineage with regulatory T cell ties.
Immunity
24
:
677
688
.
40
Wolchok
J. D.
,
Saenger
Y.
.
2008
.
The mechanism of anti-CTLA-4 activity and the negative regulation of T-cell activation.
Oncologist
13
(
Suppl 4
):
2
9
.
41
Yang
Z. Z.
,
Grote
D. M.
,
Ziesmer
S. C.
,
Xiu
B.
,
Novak
A. J.
,
Ansell
S. M.
.
2015
.
PD-1 expression defines two distinct T-cell sub-populations in follicular lymphoma that differentially impact patient survival.
Blood Cancer J.
5
:
e281
.
42
Wistuba-Hamprecht
K.
,
Di Benedetto
S.
,
Schilling
B.
,
Sucker
A.
,
Schadendorf
D.
,
Garbe
C.
,
Weide
B.
,
Pawelec
G.
.
2016
.
Phenotypic characterization and prognostic impact of circulating γδ and αβ T-cells in metastatic malignant melanoma.
Int. J. Cancer
138
:
698
704
.
43
Ladányi
A.
,
Kiss
J.
,
Mohos
A.
,
Somlai
B.
,
Liszkay
G.
,
Gilde
K.
,
Fejös
Z.
,
Gaudi
I.
,
Dobos
J.
,
Tímár
J.
.
2011
.
Prognostic impact of B-cell density in cutaneous melanoma.
Cancer Immunol. Immunother.
60
:
1729
1738
.
44
Gilbert
A. E.
,
Karagiannis
P.
,
Dodev
T.
,
Koers
A.
,
Lacy
K.
,
Josephs
D. H.
,
Takhar
P.
,
Geh
J. L.
,
Healy
C.
,
Harries
M.
, et al
.
2011
.
Monitoring the systemic human memory B cell compartment of melanoma patients for anti-tumor IgG antibodies.
PLoS One
6
:
e19330
.
45
Imai
K.
,
Matsuyama
S.
,
Miyake
S.
,
Suga
K.
,
Nakachi
K.
.
2000
.
Natural cytotoxic activity of peripheral-blood lymphocytes and cancer incidence: an 11-year follow-up study of a general population.
Lancet
356
:
1795
1799
.
46
Villegas
F. R.
,
Coca
S.
,
Villarrubia
V. G.
,
Jiménez
R.
,
Chillón
M. J.
,
Jareño
J.
,
Zuil
M.
,
Callol
L.
.
2002
.
Prognostic significance of tumor infiltrating natural killer cells subset CD57 in patients with squamous cell lung cancer.
Lung Cancer
35
:
23
28
.
47
Coca
S.
,
Perez-Piqueras
J.
,
Martinez
D.
,
Colmenarejo
A.
,
Saez
M. A.
,
Vallejo
C.
,
Martos
J. A.
,
Moreno
M.
.
1997
.
The prognostic significance of intratumoral natural killer cells in patients with colorectal carcinoma.
Cancer
79
:
2320
2328
.
48
Ishigami
S.
,
Natsugoe
S.
,
Tokuda
K.
,
Nakajo
A.
,
Che
X.
,
Iwashige
H.
,
Aridome
K.
,
Hokita
S.
,
Aikou
T.
.
2000
.
Prognostic value of intratumoral natural killer cells in gastric carcinoma.
Cancer
88
:
577
583
.
49
Mao
Y.
,
Sarhan
D.
,
Steven
A.
,
Seliger
B.
,
Kiessling
R.
,
Lundqvist
A.
.
2014
.
Inhibition of tumor-derived prostaglandin-e2 blocks the induction of myeloid-derived suppressor cells and recovers natural killer cell activity.
Clin. Cancer Res.
20
:
4096
4106
.
50
Greenwald
R. J.
,
Boussiotis
V. A.
,
Lorsbach
R. B.
,
Abbas
A. K.
,
Sharpe
A. H.
.
2001
.
CTLA-4 regulates induction of anergy in vivo.
Immunity
14
:
145
155
.

The authors have no financial conflicts of interest.

Supplementary data