A core of kinase-regulated interactomes defines the neoplastic MDSC lineage.

Myeloid-derived suppressor cells (MDSCs) differentiate from bone marrow precursors, expand in cancer-bearing hosts and accelerate tumor progression. MDSCs have become attractive therapeutic targets, as their elimination strongly enhances anti-neoplastic treatments. Here, immature myeloid dendritic cells (DCs), MDSCs modeling tumor-infiltrating subsets or modeling non-cancerous (NC)-MDSCs were compared by in-depth quantitative proteomics. We found that neoplastic MDSCs differentially expressed a core of kinases which controlled lineage-specific (PI3K-AKT and SRC kinases) and cancer-induced (ERK and PKC kinases) protein interaction networks (interactomes). These kinases contributed to some extent to myeloid differentiation. However, only AKT and ERK specifically drove MDSC differentiation from myeloid precursors. Interfering with AKT and ERK with selective small molecule inhibitors or shRNAs selectively hampered MDSC differentiation and viability. Thus, we provide compelling evidence that MDSCs constitute a distinct myeloid lineage distinguished by a “kinase signature” and well-defined interactomes. Our results define new opportunities for the development of anti-cancer treatments targeting these tumor-promoting immune cells.


INTRODUCTION
Anti-cancer treatments are primarily aimed at causing arrest of tumor cell growth or tumor cell death. In recent years, immunotherapy has resurfaced as an attractive therapeutic alternative [1]. However, the expansion of immunosuppressive cell types in cancer patients strongly interferes with anti-tumor immune responses. These immunosuppressive cells enhance tumor progression/metastasis and counteract classical anti-neoplastic treatments. Amongst these, myeloid-derived suppressor cells (MDSCs) are major contributors to tumor progression. MDSCs differentiate from precursors within the bone marrow (BM) in tumor-bearing hosts. MDSCs distribute systemically and infiltrate tumors, where they contribute to tumor progression through a variety of mechanisms [2,3]. However, MDSC differentiation and functions are still poorly understood. This is due to the difficulty of isolating them from tumor-bearing subjects, or differentiating them in vitro so that they faithfully model in vivo cell subsets [4]. Nonetheless, counteracting their activities strongly enhances anti-cancer treatments [5]. Thus, finding treatments that would specifically eliminate MDSCs could improve the efficacy of anticancer therapies.
The use of high-throughput analytical techniques for the identification of cellular regulatory pathways and novel molecular targets is on the increase. Two independent studies on the proteome of blood and spleen MDSCs have been published using LC-MS/MS mass spectrometry and label-free quantification [12,13]. Although relevant data was obtained, none of these studies included control cell types such as myeloid DCs and NC-MDSCs. Therefore, studies that have been published so far have not discriminated pathways associated to cell lineage or the tumor environment.
To overcome these issues, we carried out in-depth proteomic analyses comparing myeloid DCs, MDSCs modeling tumor-infiltrating subsets or modeling NC-MDSCs. We found a kinase signature that defined neoplastic MDSCs which could be specifically targeted to interfere with MDSC differentiation from myeloid precursors.
Confidence-based protein networks were reconstructed using STRING software [14], with up-regulated or down-regulated proteins. Both high and medium confidence links were considered (score >0.4), as the number of networks was limited to allow careful confirmation. About 10 distinct protein networks were organized around a central group of kinases that included SRC family members ( Supplementary Fig. 1). These networks were associated to production of reactive oxygen species (ROS), protection against oxidative damage, intracellular vesicle trafficking and aminoacid metabolism. Decreases in spliceseosomal proteins, carbohydrate metabolism, lysosomal function and MHC II antigen presentation were also evident.
KEGG pathway mapping was applied to up-and down-regulated proteins. KEGG analyses showed strong inhibition of cellular processes associated to inflammatory disorders and a decrease in metabolism of aminoacids (Fig. 5).
Similar results were obtained with STRING software (Supplementary Fig. 2). Most notably, up-regulation of the pentose phosphate pathway, changes in cytoskeletal proteins and down-modulation of oxidative phosphorylation. Results from KEGG pathway mapping highlighted increased glutathione metabolism, activation of the pentose phosphate pathway and a decrease in spliceosomal proteins (Fig. 7).

A kinase signature defines the neoplastic MDSC lineage
Systems biology analyses delineated a kinase signature of the MDSC lineage (AKT and the SRC family, which included SRC, HCK and FYN) and neoplastic MDSCs (ERK and PKC kinases). Overall, the expression of FYN, HCK and total and phosphorylated SRC agreed with proteomic data, as assessed by flow cytometry and immunoblotting. The predicted participation of AKT was also confirmed (Fig. 8a). AKT expression was particularly high in MDSCs modeling tumor-infiltrating subsets as detected by immunoblotting. ERK1 and PKC isoforms were predicted to be differentially expressed in tumor-infiltrating MDSCs. While total ERK expression was equivalent between B16-MDSCs and NC-MDSCs, phosphorylated (active) ERK1 was increased in B16-MDSCs (Fig. 8a). The expression of phosphorylated PKC isoforms (phosphorylated pan-PKCs) was tested by immunoblot. In agreement with proteomic data and Ingenuity analyses, phosphorylated PKCs were present at higher levels in MDSCs modeling neoplastic subsets (Fig. 8a).  differentially expressed proteins, which shows detailed interaction relationships between the input nodes (differentially expressed proteins between MDSCs and DCs), and regulatory kinases encircled in blue. This interactome links AKT/SRC kinases with mitochondrial respiration and dysfunction, protection against oxidative stress and extracellular matrix remodeling. Nodes in red, up-regulated proteins. Nodes in green, down-modulated proteins. In white, predicted protein nodes. A, activation; B, binding; C, causes/leads to; CC, chemicalchemical interaction; CP, chemical-protein interaction; E, expression; EC, enzyme catalysis; I, inhibition; L, proteolysis; LO, localization; M, biochemical modification; MB, group/complex; P, phosphorylation/dephosphorylation; PD Protein-DNA binding; PP Protein-Protein binding; PR Protein-RNA binding; RB Regulation of Binding; RE Reaction; RR RNA-RNA Binding; T Transcription; TR Translocation. Dash arrows, indirect interactions.

AKT and ERK1 specifically contribute to MDSC differentiation from myeloid precursors
To assess the contribution of MDSC-associated kinases to myeloid differentiation, a collection of kinase inhibitors were added to myeloid precursors committed towards DC or B16-MDSC differentiation. Inhibitors were added at concentrations reported to interfere with cancer cell growth. High resolution impedance-based real-time cell monitoring (RTCA) was used to continuously monitor myeloid differentiation, viability and to calculate IC50s ( Fig. 8b and Table 1) [15]. Overall, all tested inhibitors affected equally to myeloid precursors differentiating towards DCs and MDSCs (Table 1). Treatments with the specific AKT inhibitor X or the MEK inhibitor PD0325901 were an exception. AKT inhibitor X was highly toxic to precursors differentiating towards MDSCs, while differentiating DCs remained unaltered ( Fig. 8b and Table 1). Treatment with the MEK inhibitor PD0325901 selectively inhibited MDSC proliferation. Overall, comparing the IC50s for differentiating DCs and MDSCs, AKT and MEK-ERK pathways specifically contributed to MDSC differentiation (Fig. 8c). Moreover, myeloid precursors committed to MDSC differentiation died within hours of adding the AKT inhibitor, strongly suggesting that AKT was involved in survival but not differentiation (Fig. 9a).
The results with kinase inhibitors were also confirmed with silencing shRNAs. Thus, myeloid precursors committed towards MDSC differentiation were expanded for three days from BM and transduced with a lentivector delivering immunoblot-validated shRNAs against AKT1 or ERK1 as described [8,16] (Fig. 9b). Transduced myeloid precursors died 48 hours after delivery of the AKT1-specific shRNA. Likewise, ERK1 silencing significantly inhibited cell growth.

Inhibition of the ERK pathway interferes with MDSC differentiation and accelerates DC maturation
Inhibition of the ERK pathway interfered with MDSC growth. As the MEK inhibitor PD0325901 is currently used for the treatment of several human cancers in clinical trials, its effects on differentiation of myeloid cell lineages was further tested. Thus, the three main myeloid cell populations differentiated from bone marrow precursors was quantified by flow cytometry after a week of PD0325901 treatment; Namely, CD11b + monocytic myeloid cells (Ly6C high Ly6G neg ), granulocytic myeloid cells (Ly6C + Ly6G high ) and conventional DCs (Ly6C + Ly6G neg/low CD11c + ) (Fig. 9c). Interestingly, PD0325901 treatment accelerated conventional CD11c + DC differentiation. At the highest tested concentration, the MEK inhibitor was strongly cytotoxic to myeloid cells committed to  MDSC differentiation, but not to those differentiating towards DCs which strongly up-regulated CD11c expression.
Overall, these results also confirmed that interference with the ERK pathway is inhibitory over MDSCs and promotes conventional DC differentiation.

DISCUSSION
High-throughput analyses of biological systems provide a unified view of biological processes, and can uncover novel molecular targets. However, sorting out meaningful information from large datasets is challenging and relies on choosing the right controls. In addition, some biological systems such as MDSCs are difficult to work with. Here we characterized the neoplastic B16 melanoma MDSC proteome by differentiating MDSCs modeling tumor-infiltrating subsets, and quantitatively comparing it with the proteomes of myeloid DCs and MDSCs modeling non-cancerous MDSCs. Myeloid DCs and NC-MDSCs provide the appropriate controls to discriminate pathways regulated by cell lineage or the tumor environment. We performed in-depth quantitative proteomics to construct highly detailed MDSC interactome maps. Regulatory networks were for the first time unambiguously ascribed to cell lineage or to a neoplastic environment.
Cell lineage differences were highlighted by comparing B16 melanoma-MDSCs with myeloid DCs. Mitochondrial dysfunction was a key characteristic of MDSCs, reflecting a shut-down of oxidative phosphorylation. MDSCs modeling non-cancerous subsets provided a convenient standard to discriminate cancer-specific pathways. Of these, the pentose phosphate pathway was one of the most prominent, probably used to produce NADPH for biosynthesis in the absence of oxidative phosphorylation. Decreased expression of mitochondrial NADPH dehydrogenase, up-regulation of free radical scavenging proteins, and cellular stress were hallmarks of neoplastic MDSCs compared to NC-MDSCs. As we found that NC-MDSCs differed from neoplastic MDSCs in cell stress pathways and inflammatory pathways, it is likely that NC-MDSCs are different to other subsets differentiated in non-neoplastic conditions such as cell stress and infection. Overall, published data agreed with our proteome maps [17][18][19][20][21][22][23][24].
Importantly, neoplastic MDSCs presented a specific kinase signature which controlled MDSC-related interactomes and clearly separated them from the myeloid DC lineage. While SRC, FYN, HCK, PI3K and AKT kinases differentiated MDSCs from DCs, ERK and PKC discriminated neoplastic MDSCs from non-cancerous subsets. The proteomic and systems biology data was confirmed by immunoblot and flow cytometry. Ingenuity analyses also predicted the PKC isoforms as a differential feature of neoplastic MDSCs. DCs and NC-MDSCs presented lower but detectable levels of phospho-PKC. As we used a pan-phospho-PCK antibody, we cannot rule out that some specific PKC isoforms discriminate neoplastic MDSCs from non-neoplastic counterparts. In fact, this is the case of 4T1 breast cancer MDSCs, for which there is evidence that isoforms beta and theta are specifically activated [22].
AKT was required for the survival of myeloid precursors differentiating into MDSCs, but was dispensable in precursors committed to DC differentiation. This is in agreement with the requirement of AKT for in vivo MDSC expansion [25], and with its anti-apoptotic role in hematopoietic cells [26]. Additionally, the MEK1 inhibitor PD032591 selectively affected differentiating MDSCs, while DC maturation was enhanced according to CD11c up-regulation. The ERK pathway is known to keep  Blots on the right, detection of the indicated kinases by immunoblotting. Two preparations from the indicated cell populations (top of the immunoblots) were loaded and lanes were labelled as 1 and 2. An immunoblot for GADPH detection is shown as a reference control, on the same membrane used for AKT detection above. b. Representative real-time cell monitoring (RTCA) results for myeloid precursors treated with the indicated concentrations of AKT inhibitor X, and grown either in DC-differentiation medium or in B16-MDSC conditioning medium as indicated on top. Data is plotted as means of cell index with error bars (standard deviations) from duplicate cultures, shown as a function of time. Relevant statistical comparisons are shown and indicated with *, **, and *** for significant (P < 0.05), very significant (P < 0.01) and highly significant (P < 0.001) differences, respectively. c. DC/MDSC IC50 ratios calculated for the indicated treatments. Ratios close to 1 (horizontal dotted line) indicate that treatments are equally inhibitory over MDSCs and DCs. Ratios higher than 1 indicate that MDSCs are more sensitive to the specific treatments than DCs.
DCs immature and favor tumor progression [27][28][29]. Here we also demonstrated that ERK contributes to MDSC differentiation. Our results show that inhibition of ERK and AKT pathways could enhance anti-tumor immune responses by depleting MDSCs and activating DCs. Nevertheless, the other differentially expressed kinases may still participate in MDSC functions apart from differentiation and survival, which could be susceptible of therapeutic intervention.

Cells and mice
293T, B16F0 cells and BM-DCs were grown as described [8,27]. Approval for animal studies was obtained from the Animal Ethics Committee of the University of Navarra, and from the Government of Navarra. Non-cancerous MDSCs (NC-MDSCs, 293T-MDSCs) and B16-MDSCs were obtained from C57BL/6 murine BM cells as described [8].

Drug treatments of myeloid cell cultures and impedance-based real-time living cell monitoring (RTCA)
Myeloid hematopoietic precursors were expanded from BM cells using granulocyte-monocyte-colony stimulating factor (GM-CSF), stem cell factor (SCF) and leukaemia inhibitory factor (LIF) for 2 to 3 days, following published conditions [8,30]. Then, myeloid precursors were seeded on two L8 cell culture chambers for the xCELLingence RTCA monitoring system (ACEA biosciences), at a density of 200000 cells per well. DC or B16-MDSC differentiation medium was added to myeloid precursors, and treatments were carried out simultaneously in duplicates. After 30 min, the indicated chemical inhibitors were added at concentrations reported to be cytotoxic to cancer cells. Control wells were treated with carrier solution (either water or DMSO). The following inhibitors were used: AKT inhibitor X (Calbiochem), tyrosine kinase inhibitor TX-1123 (Calbiochem), MEK inhibitor PD0325901 (SIGMA), ERK inhibitors SCH772984 and VTX-11e [31], broad PKC inhibitor GÖ 6983 (Santa Cruz Biotechnology), PKC inhibitor NPC-15437 dihydrochloride (Santa Cruz Biotechnology), selective LCK and FYN inhibitor PP2 (Santa Cruz Biotechnology), and the SRC and FYN inhibitor Saracatinib (MedChem Express). IC50s for each inhibitor were calculated using the RTCA data and analysis software, using duplicates for each drug treatment.

Lentivector production and cell transduction
Lentivectors were produced and titrated by flow cytometry or Q-PCR as described [32]. The pHIV-SIREN system developed by our group [16] was used as a backbone to clone the following validated shRNAs against ERK1 (GCATGCTTAATTCCAAGGGCTATTCAAGAGATAG CCCTTGGAATTAAGCATGTTTTTTACGCGT) and AKT1 (GTCTGAGACTGACACCAGGTATTTCAAGAG AATACCTGGTGTCAGTCTCAGATTTTTTACGCGT). A control shRNA-encoding lentivector targeting the human PD1 transcript (SIREN-shPD1a) was used [33]. The same shRNAs were cloned into the pSIRACT-GFP shRNA-cloning lentivectors, which were derived from pHIV-SIREN constructs by replacing the PGK promoter by the Actinin 4 promoter. The lentivector backbone was changed because PGK was strongly down-modulated in MDSCs, while actinin4 was strongly expressed. Data is plotted as means and standard deviations as error bars. The same is shown in the graph on the right, but delivering an ERK1-specific shRNA (shERK1). Relevant statistical comparisons are shown within the graph. c. Phenotype effects of sustained MEK inhibition with PD0325901 on myeloid precursors committed to MDSC differentiation or to DC differentiation. Ly6C+ cells were gated and the Ly6G-CD11c expression profiles are shown in flow cytometry density plots. Percentages of CD11c+ myeloid cells after 7 days of culture are highlighted within the graphs. ***, very highly significant differences.

Immunoblot
Immunoblots were performed as described [27]. Anti-GADPH was purchased from Calbiochem. Rabbit anti-human HCK was purchased from Millipore. From BD bioscience, mouse anti-pan ERK, mouse anti-AKT and mouse anti-AKT pT308. From Cell Signaling, rabbit anti-mouse T202/Y204 p-P44/42 MAPK, phospho-pan-PKC rabbit mAb. Peroxidase-conjugated anti-mouse and anti-rabbit antibodies were purchased from DAKO. Membranes were stripped and re-probed with antibodies for total and phosphorylated proteins, when required.

Cell staining and flow cytometry
Surface and intracellular staining were performed as described previously [27] using the indicated antibodies.

Mass spectrometry-based quantitative proteomics and bioinformatic analyses
A global experiment was carried out with three biological replicates in each experimental condition using B16-MDSC, NC-MDSC and DC cell pellets. The specific procedures for sample preparation, proteomic analyses, iTRAQ-based proteomic workflows and mass spectrometry using triple-TOF 5600 system (AB Sciex) have been published [8]. The mass spectrometry proteomics data were deposited to the ProteomeXchange Consortium (http://proteomecentral.proteomexchange. org) via the PRIDE partner repository with the data set identifiers PXD001103 and PXD001106.
Analyses of raw data (.wiff, AB Sciex) were performed with MaxQuant software [34]. For peak list generation, default AB Sciex Q-TOF instrument parameters were used except the main search peptide tolerance, which was set to 0.01 Da, and MS/MS match tolerance, which was increased up to 50 ppm. Minimum peptide length was set to 6. Two databases were used. A contaminant database (.fasta) was firstly used for filtering out contaminants. Peak lists were searched against UniProt murine database, and Andromeda was used as a search engine [35]. Methionine oxidation was set as variable modification, and the carbamidomethylation of cysteine residues was set as fixed modification. Maximum false discovery rates (FDR) were set to 0.01 at protein and peptide levels. Analyses were limited to peptides of six or more amino acids in length, and considering a maximum of two missed cleavages. Relative protein abundance output data files were managed using R scripts for subsequent statistical analyses and representation. Proteins identified by site (identification based only on a modification), reverse proteins (identified by decoy database) and potential contaminants were filtered out. Only proteins with more than one identified peptide were used for quantification. For possible quantification data rescue, up to one missing value for each group was rescued replacing it by the mean of the rest in-group samples. Data was normalized and transformed for later comparison using quantiles normalization and log2 transform respectively. The Limma Bioconductor software package in R was used for ANOVA analyses. Significant and differential data were selected by a p-value lower than 0.01, fold changes of <0.77 (down-regulation) and >1.3 (up-regulation) in linear scale. These parameters were used for differential expression threshold with volcano and profile plots.
The proteomic information was analyzed using bioinformatic tools. Studies with the Kyoto Encyclopedia of Genes and Genomes (KEGGS) Pathway mapping tool were performed as described (http://www.genome .jp/kegg/tool/map_pathway1. html). The identification of specifically up-or dysregulated regulatory/metabolic networks in MDSCs was analyzed with the open access STRING (Search Tool for the Retrieval of Interacting Genes) analysis tool (v.9.1) [14] and with the Ingenuity Pathway Analysis Tool (Qiagen).

Statistical analyses
GraphPad Prism and SPSS software packages were used for plotting data and statistical analyses. No data was considered an outlier. Real time cell monitoring data (RTCA, ACEA biosystems) was analyzed by exporting the Cell Index data as a function of time. It was confirmed that Cell Index in a growing population of cells was highly homogeneous and normally distributed at any given time-point. Therefore, the data was analyzed by one-way ANOVA and Tukey's pair-wise comparisons. IC50s were estimated for each treatment (using three published active concentrations per compound) in duplicates by RTCA, and means with standard deviations were obtained. IC50s were also highly homogeneous and normally distributed. The relative IC50 ratios for DCs vs MDSCs were also calculated, and compared by one-way ANOVA and Tukey's pair wise comparisons. Cell viability was quantified by trypan blue staining and data analyzed by one-way ANOVA and Tukey's pair wise comparisons. Triplicates per treatment were used for the analyses. Growth of myeloid cells transduced with lentivectors encoding shRNAs was www.impactjournals.com/oncotarget compared by a two-way ANOVA with "time of growth" as a random factor with data from four independent transductions, as described previously [8].