The Interplay between Circulating Tumor Cells and the Immune System: From Immune Escape to Cancer Immunotherapy

Circulating tumor cells (CTCs) have aroused increasing interest not only in mechanistic studies of metastasis, but also for translational applications, such as patient monitoring, treatment choice, and treatment change due to tumor resistance. In this review, we will assess the state of the art about the study of the interactions between CTCs and the immune system. We intend to analyze the impact that the cells of the immune system have in limiting or promoting the metastatic capability of CTCs. To this purpose, we will examine studies that correlate CTCs, immune cells, and patient prognosis, and we will also discuss relevant animal models that have contributed to the understanding of the mechanisms of immune-mediated metastasis. We will then consider some studies in which CTCs seem to play a promising role in monitoring cancer patients during immunotherapy regimens. We believe that, from an accurate and profound knowledge of the interactions between CTCs and the immune system, new immunotherapeutic strategies against cancer might emerge in the future.


Introduction
In malignant evolution, cancer cells acquire the capability to invade healthy tissues and colonize distant organs. This process, indicated as metastatic cascade, includes an ordinate series of events involving tumor cells, namely their invasion into surrounding normal tissues, intravasation in the bloodstream, arrest, and extravasation through vascular walls into secondary sites, formation of microscopic colonies, and final proliferation into overt, clinically detectable metastases [1].
Although metastasis accounts for a remarkable 90% of cancer-associated deaths, it remains a poorly understood process because of the complex interplay that the primary tumor establishes with stromal cells, often based on redundant and still unclear signaling pathways [1]. Among stromal cells, those of the immune system especially affect the outcome of tumor progression and metastasis.
Furthermore, it is not yet clear when the metastatic capability of a tumor appears, because direct and indirect evidence have contrasted the view that the spreading of tumor cells to secondary sites represents a late event in tumorigenesis. For example, 30.6% of breast cancer (BC) patients at diagnosis show micrometastases in the bone marrow, independently of disease stage [2], and prostate cancer cells disseminate early [3]. Moreover, karyotypic abnormalities of micrometastases in the bone marrow from BC patients, as well as from animal models, indicate that tumor cell dissemination occurs in the pre-invasive stage of the disease [4]. Consequently, characterization of the primary tumor may not be enough to assess the risk for disease recurrence and to choose the best treatment, while by investigating metastases, which represent the final result of the process, we might lose relevant information about the characteristics necessary to overcome one or more steps of the metastatic cascade.
On these bases, the study of circulating tumor cells (CTCs) has elicited increasing interest, especially after the advent of reliable technologies that permit isolation, quantification, and characterization of tumor cells after intravasation in the peripheral blood and before extravasation at secondary sites. Although this is only an intermediate step of the full metastatic process, it is a crucial step as demonstrated by the prognostic value of CTC numbers [5] and it is thought to contribute to the selection of phenotypic and biological properties that are necessary for tumor cells to complete the whole process [6]. Furthermore, since we can obtain tumor cells from a simple blood draw of a cancer patient as often as necessary, CTCs are considered of utmost importance not only in mechanistic studies of metastasis, but also for translational applications such as patient monitoring, treatment choice, and, in the case of ongoing resistance, treatment change.
For this reason, being aware of the novelty of immunotherapy of tumors in the precision medicine era, we thought to review what we know concerning the relationship between CTCs and the different cell populations of the immune system, with the aim to understand how the latter can prevent or even support the metastatic dissemination of the former.
Our review does not aim to debate different scenarios about the interactions between CTCs and the immune system, with the final goal of supporting one or the other, since the scientific knowledge on this topic is not yet complete. Rather, we feel the need to describe the state of the art regarding what we know on the role and mechanisms of the different immune cell populations in limiting or promoting the dissemination and persistence of CTCs.
To this purpose, in our literature searches, we used the keyword 'CTC' (or the full form 'circulating tumor cell') in conjunction with 'macrophages', 'dendritic cells', and so on, according to the paragraph titles. For the keyword 'immunotherapy', we combined it with 'CTC' and, in some cases, with a third keyword, e.g., 'biomarkers', 'targets', and 'PD-L1'.
We reviewed not only studies performed in cancer patients that correlate CTCs, immune system cells, and prognosis, but also studies presenting relevant animal models that, by studying experimental tumors induced by cancer cell lines, or experimental metastasis induced by injection of cancer cell lines in the peripheral blood, which contributed to the understanding of immune-mediated mechanisms of metastasis. Based on our selection criteria, we used the acronym 'CTC' only when talking about studies on human samples or on animal models in which CTCs were shed from primary tumors. In all other cases (in vitro assays and in vivo experimental metastasis), we talked about 'tumor cells' since in these conditions neoplastic cells were not actually circulating or part of a real metastatic process.
We then reviewed papers reporting results on immunotherapies, with particular regard to what we can directly monitor in cancer patients by studying CTCs.
Finally, we briefly looked at future opportunities, i.e., new immunotherapeutic strategies that we could implement using the knowledge on metastatic mechanisms gained from studies on CTCs.

Immune-Surveillance and Immune-Support of CTCs
The success rate of metastasis is low since only a few of the thousands of CTCs daily released in the bloodstream survive and form secondary lesions. For instance, in murine experimental metastasis models, Fidler found that only 1% of tumor cells survived in premetastatic lungs 24 h after intravenous injection [7], and Luzzi et al., at day 13 post-injection, observed that only 1% of liver micrometastases progressed to form macrometastases and that 36% of tumor cells remained isolated; 95% of which were dormant (2% proliferative and 3% apoptotic) [8]. Several reasons can explain this inefficiency, including the mechanical stress in blood vessels, the complexity of the translocation from one site to another, the presence of a final microenvironment unsuitable for proliferation, and the intervention of host immune surveillance [7][8][9].
In principle, we cannot exclude that the relationships between the immune system and tumor cells differ between primary tumor and peripheral blood. Indeed, in peripheral blood, CTCs migrate as single cells or tumor microemboli and undergo changes that are expected to modify their phenotype [10], so that CTCs might continue or interrupt their crosstalk with the immune system. Furthermore, after intravasation, CTCs lose all the interactions with both the extracellular matrix (ECM) and stromal cells; in this situation, cytokines and other soluble factors produced by immune system cells, which are usually only locally active, might no longer be effective on CTCs. In any case, it is conceivable (and generally accepted) that CTCs, leaving the protected microenvironment of the primary tumor, encounter further immune surveillance in nontumor tissues.
Immune cells ambiguously show both anti-and protumor effects. They can promote tumor progression and metastasis by creating an immunosuppressive, tolerogenic context or mediating ECM degradation and angiogenesis [11][12][13]. Tumor-derived soluble factors (TDSFs) mediate the recruitment of myeloid cells and hematopoietic progenitors in the target organ to adapt local microenvironment for CTC homing [14]. For example, tumor necrosis factor (TNF)-α supports survival and proliferation of tumor cells and increase vascular permeability [15,16]. Also, immunohistochemistry on biopsies from BC patients revealed that the chemokine receptor CXCR4 favors tumor cell homing in the bone marrow [17].
Interestingly, when Hensler and colleagues compared BC patients and healthy donors for gene expression profiles in both CTCs (obtained by immuno-magnetic sorting) and peripheral blood mononuclear cells (PBMCs), they found a higher expression of metastasis-related genes, such as those for ECM degradation and epithelial-mesenchymal transition (EMT), in patient PBMCs [18].
In the subsections below, we will consider the most important infiltrating and circulating immune cells that hinder or favor the dissemination of CTCs, in some cases directly interacting with them. The main mechanisms are depicted in Figure 1.

Natural Killer (NK) Cells
The number of circulating NK cells has been observed to increase in metastatic breast, colorectal, and prostate cancer patients as compared to healthy donors [19]. NK cells can intercept CTCs in the bloodstream and destroy them before extravasation, thus preventing metastasis [20]. A study performed on mice injected with human colon cancer cells tried to elucidate the mechanisms used by NK cells to kill cancer cells. Direct perforin-dependent killing resulted more effective than indirect killing with apoptosis-inducing factors, since the former significantly delayed primary tumor growth, reduced the number of CTCs by 80% as assessed by real-time PCR, and hindered the formation of metastases, with respect to the latter [21].
Accordingly, low NK cell activity is associated with advanced disease and metastasis [22,23]. Metastatic BC patients with >5 CTCs per 7.5 mL of blood had circulating NK cells showing deficient lytic ability in chromium-51 release assays as compared to those from patients with ≤5 CTCs, and an inverse correlation between CTCs and progression-free survival (PFS) was found [24]. Also in metastatic breast, colorectal, and prostate cancer patients, the cytotoxic activity of NK cells inversely correlated with the level of CTCs detected by the CellSearch system (CS, Menarini-Silicon Biosystems, Castel Maggiore (BO), Italy) [19].

CD8 + and CD4 + T Cells
Our current knowledge concerning the function of CD8 + cytotoxic T lymphocytes (CTLs) and CD4 + T helper (Th) cells in immune surveillance of CTCs is very limited. Their infiltration in primary tumors, together with that of NK cells, correlates with an increase of disease-free survival (DFS) and overall survival (OS), and a reduction of metastasis and relapse [25,26]. Accordingly, in patients with metastatic BC, low circulating lymphocyte levels and high CTC levels, as assessed by CS, were found to be independent poor predictive factors for PFS and OS [27]. In inflammatory BC patients, the presence of CS-detected CTCs correlated with a reduction of CD3 + T cells, CD4 + T cells, and CD8 + T cells synthesizing TNF-α and IFN-γ [28]. In stage IV non-small-cell lung cancer (NSCLC) patients, Ye et al. showed that the number of CTCs as determined by SET-iFISH correlated negatively with that of NK cells and CD3 + , CD4 + , and CD4 + /CD8 + lymphocytes, while positively with metastasis [29]. Another study performed on late stage NSCLC patients gave comparable results, since CTCs showing both epithelial and mesenchymal markers in RNA-ISH assays negatively correlated with CD3 + and CD8 + T cells; moreover, CTC levels positively correlated with metastasis and a worse clinical outcome [30].
In a different study, BC patients positive for CTCs after AdnaTest were characterized by a significant increase of peripheral first apoptosis signal receptor (FAS)-positive Th cells [31]. The induction of apoptosis in T cells could represent an escape mechanism used by CTCs (see Section 3) and potentially explain the observed CTC-related lymphocytopenia. However, it is not clear if CTCs control the number of lymphocytes or if the increase in CTCs is the consequence of a lymphocyte dysfunction due to other factors, in particular, immunosuppression exerted by the primary tumor.
Accordingly, in metastatic carcinoma patients, circulating Tregs were more abundant as compared to healthy donors, although the numbers of Tregs did not correlate with CTCs; probably due to the low-sensitivity test (immunocytochemistry) used to identify CTCs, as concluded by the authors themselves [40]. Conversely, in inflammatory BC patients, CS-enumerated CTCs positively correlated with circulating Tregs [28]. In the already-mentioned work by Ye et al. performed on advanced NSCLC patients, high numbers of CTCs correlated, not only with lymphocytopenia and metastasis, but also with high numbers of Tregs [29]. In addition, in a study performed with PCR and FACS, high pre-resection levels of both epithelial cell adhesion molecule (EpCAM)-positive CTCs and circulating Tregs correlated with a higher risk of post-resection recurrence and metastasis in hepatocellular carcinoma patients [41]. Thus, the combination of CTCs and circulating Tregs might provide a novel prognostic predictor of cancer progression.
Interestingly, in a transplantable BC mouse model, Tregs infiltrating the primary tumor were found to produce receptor activator of nuclear factor kappa-B ligand (RANKL), a tumor-necrosis-factor family member involved in bone regeneration and remodeling, as well as in mammary gland hyperplasia during pregnancy. Released RANKL stimulated lung metastatic dissemination of RANK + tumor cells through activation of IKK-α and downregulation of the metastasis inhibitor maspin [42]. Noteworthy, RANK has been detected on CTCs from melanoma patients [43].

Other Lymphocytic Subsets
In an invasive BC model, the tumor-released chemokine CCL2 and the subsequent IL-1β release from TAMs (see Section 2.8) were shown to induce IL-17 production in γδ T cells. In turn, IL-17 caused a systemic granulocyte colony-stimulating factor (G-CSF)-dependent expansion of neutrophils, that suppressed CTLs and led to increased metastasis [44,45]. CD4 + T helper 17 (Th17) cells also produce IL-17 and have been associated with MDSC-mediated immunosuppression (see Section 2.7) and metastasis [46]. Accordingly, in a study on a transplantable murine colorectal cancer (CRC) model, the authors found corresponding changes between serum IL-17A levels and the FACS-determined prevalence of CTCs at different stages of disease; moreover, IL-17A promoted angiogenesis and metastasis in vivo, as well as matrix metalloproteinase 9 (MMP9)-dependent invasiveness of tumor cells in vitro [47]. Mego et al. also showed a correlation between CS-derived CTC counts and the percentage of activated IL-17-producing CD8 + T cells [28]. Other papers proved that systemic IL-17, through activation of the IL-6/Stat3 pathway and release of MMP2/9, TNF-α, and vascular endothelial factor (VEGF), induces tumor cell migration and metastatic growth in lung cancer-bearing mice [48][49][50]. IL-22 is another cytokine produced by γδ and Th17 cells that induces Stat3 and MMP9 in cancer cells to increase their motility and, in human pancreatic ductal adenocarcinoma, their metastatic capability [51].
Mammary tumor antigen-activated CD4 + T cells can prepare the bone premetastatic niche by releasing RANKL and inducing osteolytic bone disease, with subsequent consumption and metastatic colonization of the bone cavity [52]. In another study, allergen-induced pulmonary inflammation was linked to a higher risk of lung experimental metastasis due to a CD4 + T cell-mediated activation of the vascular endothelium, which, in microfluidic in vitro assays, was required for an enhanced transendothelial migration of cancer cells [53].
In a recent study on spontaneous and transplantable tumor-bearing mice, the tumor-evoked regulatory B cells (tBregs), a lymphocyte subpopulation that derives from TDSF-conditioned B cells, induced the differentiation of CD4 + T cells into Tregs and fully activated the prometastatic functions of cancer-primed MDSCs through the TGF-β signaling, eventually increasing metastasis [36,54].
Finally, type 2 innate lymphoid cells (ILC2s) are gaining increasing interest as they can activate myeloid cells by secreting cytokines and directly induce T cells by expressing MHCII molecules [55]. ILC2s participate in immune surveillance against CTCs, since in transplantable murine models they are recruited to the primary tumor in an IL-33-reliant manner and mediate CTL activation together with DCs, hence producing a decrease in tumor growth, CTCs (evaluated by FACS), and metastasis [56].

Neutrophils
The primary tumor can secrete G-CSF, which induces granulopoiesis in the bone marrow and recruits neutrophils-cells actively involved in tumor progression and metastasis [57,58]. In spontaneous and transplantable BC mouse models, neutrophils were shown to accumulate in premetastatic lungs and produce H 2 O 2 , with subsequent killing of incoming CTCs (as demonstrated by histology) and inhibition of metastasis; the authors also stated (without presenting relative data) that the number of CTCs was not affected by neutrophils [59]. Nonetheless, the surveillance exerted by neutrophils over CTCs is so far largely unexplored.
Conversely, more data are available on neutrophil support to CTCs. As already discussed, systemically raised G-CSF upon γδ T cell activation polarizes neutrophils towards an immunosuppressive behavior that favors metastatic progression [44,45]. Moreover, G-CSF also induces homing of Ly6G + Ly6C + granulocytes in the premetastatic lungs of tumor-bearing mice, where they can release the Bv8 protein that induces angiogenesis, further mobilization of myeloid and tumor cells, and final metastasis [58].
Neutrophils have been extensively associated to tumor angiogenesis and, thus, to an easier dissemination of tumor cells [60]. For example, UV irradiation-induced metastasis of primary melanoma was shown to be mediated by activated neutrophils that prompt angiogenesis and TNF-dependent migration of melanoma cells towards vascular endothelial cells, both in vitro and in vivo [61].
Several in vivo imaging studies showed that CTCs colocalize with endothelium-bound neutrophils in the premetastatic vascular network, suggesting that neutrophils can retain cancer cells and facilitate their extravasation. For example, adherent neutrophils within the inflamed liver sinusoids interact with intravenously-injected lung cancer cells and increase their tethering ability to CD62 (selectin) on the vascular endothelium [62,63]. In vitro assays and in vivo experimental metastasis demonstrated that melanoma cells entrapped in premetastatic lung capillaries release IL-8 and recruit neutrophils that, in turn, upregulate β2 integrins and bind intercellular adhesion molecule (ICAM)-1 on tumor cells, favoring their anchoring to blood vessels [64]. Activated neutrophils also adhere to CTCs through interaction of Mac-1 on the former and ICAM-1 on the latter [63]. ICAM-1 is then able to trigger migration-related signaling pathways inside tumor cells, thus promoting extravasation and colonization of adjacent tissues [65].
Furthermore, once stimulated by G-CSF, IL-8, or other pro-inflammatory factors, neutrophils undergo a process called NETosis and release neutrophil extracellular traps (NETs), i.e., networks of DNA and antimicrobial proteins, which, in physiological conditions, catch and contain pathogens. During systemic inflammation in mice, NETs capture systemically-delivered cancer cells and improve their adherence to liver blood vessels, extravasation, and metastatic spreading [66]. NET-released high mobility group box 1 (HMGB1) augments cancer cell adhesion, proliferation, and migratory capabilities in vitro in a toll-like receptor 9 (TLR9)-dependent manner [67]. Remarkably, in a transplantable tumor model, CTCs themselves induced the formation of metastasis-supporting NETs in the absence of infection [68], and this could depend on neutrophil priming upon tumor release of G-CSF [69]. In a recent study performed on a murine intra-abdominal sepsis model that mimics postoperative inflammation, Najmeh and colleagues showed the central role of β1 integrin in mediating the interaction between CTCs and NETs [70].
In metastatic CRC patients who had undergone liver resection, postsurgical inflammation caused an increase in NETs, which correlated with a >4-fold metastasis-dependent reduction in DFS [67]. NETs have also been observed in patients with pancreatic ductal adenocarcinoma [71] and BC [68], while in gastric cancer patients the number of NETs resulted higher than in healthy donors and increased with disease progression [72]. However, to date, no findings are available for NETs and CTCs in human cancer.

Monocytes
Circulating monocytes comprise two subpopulations, classical and nonclassical monocytes. As we will discuss later, classical 'inflammatory' monocytes can extravasate and differentiate into macrophages with protumor and prometastatic functions. Conversely, in response to the CX3CL1 chemokine, nonclassical 'patrolling' monocytes (PMos) accumulate in capillaries, where they clear circulating cellular debris, and intervene in inflammation with a protective role [73]. Hanna et al. showed in different murine metastatic tumor models that, up to 24 h after intravenous cancer cell injection, PMos are recruited to premetastatic lung capillaries through the CX3CL1/CX3CR1 axis and engulf tumor material, while as early as 4 h after injection they interact with tumor cells in circulation and hamper their attachment to the lung microvasculature. Moreover, the PMo-secreted CCL3, CCL4, and CCL5 chemokines summon and activate NK cells, thus leading to further elimination of metastasizing tumor cells and prevention of lung metastasis [74].
Finally, the expression of TLR2 and TLR4 on whole, peripheral monocytes was found to inversely correlate with the level of CS-detected CTCs in metastatic breast, colorectal, and prostate cancer patients [19].

Myeloid-Derived Suppressor Cells (MDSCs)
MDSCs are heterogeneous, immature myeloid cells comprising a polymorphonuclear subset (PMN-MDSCs) and a monocytic subset (M-MDSCs), which can be respectively distinguished from granulocytes and monocytes due to their high immunosuppressive activity [75]. Infiltrating and circulating MDSCs can favor metastasis by creating a tolerogenic microenvironment in both the primary tumor and metastatic sites [54,[76][77][78]. Interestingly, in a study on portal vein blood samples from pancreatic cancer patients, the authors found a correlation between numbers of circulating M-MDSCs and active FACS-isolated K-RASmut mRNA+ CTCs, suggesting that the establishment of liver metastases in these subjects may be supported by immunosuppression-dependent CTC survival in the bloodstream [79].
Furthermore, MDSCs can directly stimulate aggressiveness of tumor cells. By releasing IL-6, MDSCs elicit STAT3 activation and invasive capabilities of BC cells, with subsequent increase in tumor and metastasis burden [80]. MDSCs can also facilitate neoplastic cell dissemination by releasing MMP9 and degrading the ECM [81], as well as by upregulating MMP2, MMP13, and MMP14 in BC cells [82]. MMP9 from PMN-MDSCs has also been implicated in the generation of an aberrant and leaky vasculature in the premetastatic lung [83]. Additionally, PMN-MDSCs can promote EMT in melanoma cells through activation of the TGF-β, epidermal growth factor (EGF) or hepatocyte growth factor (HGF) signaling pathways, leading to enhanced metastasis [84]. Ouzounova et al. recently showed in a transplantable murine BC model that tumor-infiltrated M-MDSCs promote inducible nitric oxide synthase (iNOS)-mediated EMT and cancer stem cell properties in tumor cells at the invasion frontline; then, PMN-MDSCs in metastatic lungs induce mesenchymal-epithelial transition (MET) in CTCs and restore their original phenotype to foster settlement and proliferation [85]. Thus, MDSCs seem to be involved in EMT, although the role of their two subsets needs to be clarified.

Macrophages
Kupffer cells are liver resident macrophages able to detect and arrest CTCs while passing in the bloodstream and remove metastasizing tumor cells from the hepatic parenchyma [86]. Indeed, CTC counts obtained by CS and Epispot in paired peripheral and mesenteric blood samples from CRC patients suggested that the liver entraps a fraction of CTCs [87]. Kupffer cells may act both by activating adjacent T cells against CTCs and by recognizing opsonized tumor cells and directly killing them [86,88].
Beyond tissue-resident macrophages, those derived from circulating monocytes and M-MDSCs can actively participate in cancer, since tumor-associated macrophages (TAMs) infiltrate advanced tumors [89,90] and their detection correlates with a poor prognosis [91][92][93]. TAMs can directly prompt the migration of tumor cells by secreting paracrine factors. As shown by Wyckoff et al., one of these factors is EGF, which also activates tumor cells to release colony-stimulating factor (CSF-1), thus promoting the motility of TAMs themselves [94]. Loop mechanisms like this explain why TAMs and tumor cells often move together in tumor stroma toward blood vessels [94,95]. Accordingly, the inhibition of CSF-1 or EGF signaling in tumor-bearing mice undergoing intravital imaging experiments reduced metastasis by decreasing the number of macrophages and tumor cells able to leave the primary tumor [94]. Another example is represented by TAM-released secreted protein acidic and rich in cysteine (SPARC), which results to be necessary for metastasis, since it favors migration of cancer cells by aiding their integrin-mediated interaction with nearby stroma [96].
Motility of cancer cells can also be enhanced upon physical interaction with TAMs, which induce both RhoA activity and Notch1 signaling in tumor cells. In turn, Notch1 regulates Mena INV expression, which is required for the formation of invadopodia, matrix degrading protrusions used by cancer cells for invasion, and transendothelial migration [107,108]. Consistently, when mice with BC xenografts were treated with a Notch1-blocking antibody, CTCs (counted as plated, colony-forming cells from blood) diminished when compared to the control group [108].
Another study investigated the interplay between macrophages and CTCs. Hamilton and coworkers first established two permanent CTC lines from blood samples of advanced stage small cell lung cancer patients and, then, cultured in CTC-conditioned media healthy donor-derived PBMCs, which afterwards differentiated into monocytes/macrophages expressing the TAM markers CD14, CD163, and CD68. In addition, macrophage supernatants contained several soluble factors linked to tumor cell invasiveness, angiogenesis, and immune protection, thus suggesting that CTCs might educate TAMs to support their dissemination in vivo [109].
TAMs are not the only macrophages with a metastasis-promoting activity. Circulating monocytes and myeloid progenitors can be chemoattracted by CTC-derived CCL2 into blood vessels of the metastatic organ and, then, can anchor to the endothelium and transmigrate [110]. Once extravasated, these cells can bind CSF-1 released by cancer cells and differentiate into metastasis-associated macrophages (MAMs) [11,111,112].
MAMs and other infiltrating myeloid cells can promote vasodilatation in the premetastatic niche by directly secreting VEGF or mediating protease-dependent release of ECM-bound VEGF. Thus, the more abundant blood flow allows further accumulation of MAMs and CTCs [110,112,113]. Another synergistic mechanism of metastatic seeding is based on the tissue factor (TF) expressed on systemically-injected cancer cells, which can recruit platelets and activate coagulation in blood vessels near the target organ; this allows the arrest of other cancer cells and monocytes/macrophages in circulation [114]. Many other recruitment mechanisms are still not fully elucidated, such as that recently reported after intravital imaging in mice by Headley et al. Here, the authors showed that metastasis-promoting CTC-shed 5 µm microparticles and were able to enter neutrophils, monocytes, and macrophages in metastatic lungs within 24 h after arrival of CTCs [115].
In the context of the premetastatic niche, MAMs acquire metastasis-promoting functions. Once activated by CCL2, MAMs produce the autocrine chemokine CCL3, which enhances the interaction between vascular cell adhesion molecule-1 (VCAM-1) on tumor cells and the α4 integrin on MAMs, resulting in a reciprocal, efficient retention and extravasation of macrophages and CTCs at the metastatic site, as demonstrated in various spontaneous and transplantable murine cancer models [116]. In addition, through both VEGF-mediated vasodilatation and physical interaction, MAMs support extravasation, survival, and proliferation of CTCs. Indeed, the elimination of MAMs in metastasis-bearing mice reduces the number of extravasating cells (as determined by real-time PCR) and further metastatic growth [112]. A possible explanation of this survival advantage for cancer cells was proposed by Chen et al., who showed that the binding between α4 integrins on MAMs and VCAM-1 on tumor cells can trigger the Ezrin-PI3K/Akt anti-apoptotic pathway in the latter [117].
Like TAMs in the primary tumor, MAMs in metastatic sites also produce proteases to aid tumor cell invasion, such as cathepsin S, which has been shown to degrade the JAM-B junctional adhesion molecule in the blood-brain barrier and enhance breast-brain metastasis in both murine models and patients [118].

Dendritic Cells (DCs)
DCs comprise the conventional (myeloid) and plasmacytoid (lymphoid) subsets [119]. In both cases, mature DCs are considered to be immunostimulatory and, accordingly, mature DCs that infiltrate the tumor have been associated with a better patient outcome in terms of tumor progression and metastasis [120][121][122]. In BC patients positive for CTCs according to CS, circulating DCs showed increased expression of TLR2, TLR4, and TLR8 and a decreased expression of TLR3 as compared to negative patients. However, it is not clear if this represents an effort of the immune system to respond to tumor-derived ligands or if it is associated to an immune dysfunction potentially having protumor effects [123]. Nonetheless, intravital imaging on metastatic lungs in mice injected with cancer cells showed that resident CD103 + DCs-a subset able to cross-present antigens directly to CTLs-can internalize CTC-derived microparticles and migrate to lymph nodes, where they enhance antitumor CTL responses and restrain the metastatic burden [115].
Tumor-infiltrating DCs often do not efficiently stimulate immune responses, since TDSFs can cause accumulation of immature DCs and decreased production of mature DCs inside the tumor, impairing their APC functions [120,124]. Mego et al. recently related the phenotype of circulating conventional and plasmacytoid DCs to both CS CTC counts and clinical outcome in inflammatory BC patients. The authors demonstrated that patients with ≥5 CTCs, as compared to patients with <5 CTCs, had a significantly more advanced disease stage, a worse OS, a reduced percentage of conventional DCs producing TNF-α, IFN-α, and IL-12 and a higher expression of CCR7 and CD86 on conventional and plasmacytoid DCs, respectively. These observations suggest that patients with ≥5 CTCs had defects in DC number and function, despite an enhanced activation and maturation, and a potentially compromised Th1-like immune response [125].
Both DC subsets have also been directly linked to metastasis. CCL2 and LCN2 from tumor cells induce EMT and generation of regulatory dendritic cells (DCregs), DCs that show an immunosuppressive behavior, low expression of the stimulatory molecules HLA-DR, and CD86, and high expression of the immunosuppressive molecule programmed death-ligand 1 (PD-L1). In turn, DCregs activate Tregs and inhibit tumor-specific CTLs, finally enhancing tumor growth and metastasis [126].
Another immunosuppressive DC subpopulation has been newly described in a murine pancreatic ductal adenocarcinoma model, where, in response to tumor GM-CSF, monocyte-derived CD11b + CD11c + MHCII + CD24 + CD64 low F4/80 low DCs infiltrated the premetastatic liver. These cells activated Tregs, inhibited CTLs and reduced metastasis through a mechanism involving the MGL2 lectin and PD-L2 [127]. Interestingly, besides impairing DC maturation as mentioned above, tumor-derived exosomes are also able to activate DC prometastatic activity. Indeed, exosomes expressing HSP72 and HSP105 can trigger a TLR2-and TLR4-dependent IL-6 release from conventional DCs and a consequent MMP9 expression in tumor cells with enhanced invasion and metastasis [128]. Furthermore, the number of CD83 + mature DCs in the primary tumor of CRC patients correlated with the presence of tumor cells in blood and lymph vessels as assessed by histology, and local lymph node metastases [129].
Likewise, the involvement of plasmacytoid DCs in metastasis has been investigated, although to date few references are available. This subset expands in bone lesions derived from BC and leads to a Th2-like response, as well as to an accumulation of Tregs and MDSCs. Subsequent immunosuppression and release of osteolytic cytokines elicit bone destruction, thus enhancing tumor growth and metastasis [130]. An accumulation of plasmacytoid DCs has also been observed in metastasis from other malignancies, such as BC [131] and melanoma [132]. Moreover, in the peripheral blood from gastric cancer patients, the number of these cells increased in the case of advanced disease and presence of lymph node metastasis [133].

Other Circulating Immune Cells Interacting with CTCs
CTCs can be found as single cells or clusters, named 'circulating tumor microemboli' (CTMs), that also comprise leukocytes, endothelial cells, fibroblasts, and other cells held together by cell adhesion proteins [134][135][136]. CTCs inside CTMs are protected from both immune recognition and therapeutics; this 'stealthiness' represents a tumor escape mechanism, which can also be provided by platelets alone (see Section 3). In this regard, Jiang at al. recently proposed a microfluidic method to isolate platelet-coated CTCs and CTMs that was not based on the detection of CTC surface epitopes, in this way overcoming the problems associated with masking by platelets and other cells. An intriguing hypothesis suggested by the authors is that platelets may enable CTC-immune cell interactions inside CTMs [137].
A particular a CD14 + CD11c + CD45 + myeloid subpopulation has been observed inside CTMs. Adams and colleagues named these cells 'cancer-associated macrophage-like cells' (CAMLs) and described them as giant cells (30-300 µm in length) with large multiple or polylobated nuclei (14-64 µm in diameter) [138]. The expression of CD14 and CD45 ranges from intense to absent, and the morphology of these cells is very variable, since it can be amorphous, round, oblong, tadpole-, or spindle-shaped. This phenotypic variability, similar to the plasticity of macrophages, suggests that CAMLs may also have different stages of maturation and differentiation [138]. CAMLs have been detected in the peripheral blood of patients with breast, prostate, pancreas and lung cancer in percentages ranging from 81 to 97% of total patients, whilst totally absent in healthy individuals [138,139].
Interestingly, CAMLs can be EpCAM-positive and/or CK8/18/19-positive, although it is still unclear if they directly express these epithelial markers (at different levels depending on the differentiation stage) or if they engulf material of epithelial origin. A possible explanation is that CAMLs internalize tumor cells/CTCs or their debris, as suggested by the presence of tumor-specific markers and mutations inside CAMLs [138]. Accordingly, other researchers working on macrophage-tumor cell fusions (MTFs) could cultivate resulting fused cells present in the blood from melanoma patients and observe primary tumor-specific mutations [140]. The internalization/fusion hypothesis is supported by the fact that CAMLs seem to originate in the primary tumor and increase in blood samples from patients responding to radiotherapy, chemotherapy, or other treatments, generally when dead tumor cells and debris accumulate [138,141]. CAMLs have also been shown to actively interact with CTCs or express CD146 and TIE2, markers that can suggest a pro-angiogenic activity [138]. In support of a protumor role of these cells, in metastatic BC patients, EpCAM + CAMLs correlate with shorter OS and PFS [142].
Therefore, both the origin and function of CAMLs in the tumor context are still under investigation, but they appear to be interesting liquid biopsy-based predictors of tumor activity and response to therapy. It cannot be excluded that CAMLs might also have an active role in helping CTC intravasation, extravasation, or survival in the bloodstream, thus participating in the metastatic process.

CTC Evasion from Immune Surveillance
Since the early phases of tumor progression, a delicate equilibrium arises between the antitumoral immunity and tumor escape mechanisms. The selection of less immunogenic tumor cell clones (cancer immunoediting) and the promotion of an immunosuppressive microenvironment able to limit immune responses and favor neoplastic progression are paramount in this process [12].
Several mechanisms have been hypothesized through which CTCs could escape or survive from encounters with immune cells [143]. Some mechanisms involve modifications in MHCI molecules: 1. The downregulation or loss of surface MHCI expression to escape the action of CTLs (an event that, in fact, makes them susceptible to the action of NK cells) [144]; 2. the acquisition of a 'pseudonormal' phenotype by the transfer of MHCI molecules from the surface of platelets to escape NK cell-mediated cytotoxicity [145]; 3. the expression of cytokeratins (CK8, CK18, CK19) that interfere with the recognition of MHCI complexes by T cell receptors (TCRs) on CTLs [146].
Other mechanisms not involving MHCI modifications are: 1. The expression of PD-L1, which prevents T cell-mediated destruction [147][148][149][150]; 2. the expression of CD47, which provides a 'don not eat me' signal [151][152][153]; 3. an altered expression of the apoptotic FAS and/or FASL proteins that may induce the apoptosis of T cells [31] or protect tumor cells from FAS-mediated apoptosis [154]; 4. the association of CTCs inside CTMs, where they are hidden and protected from immune attacks [155]; 5. the interaction with platelets, which induce EMT-like features in CTCs [156], promote their arrest and extravasation [157], and, as already mentioned, form a coating shield that provides them with 'stealth' properties and helps their survival in the circulation [158].

CTC Surface Markers Involved in Immune Escape and Metastatic Dissemination
Several CTC surface markers, potentially involved in CTC escape from the immune system, could represent therapeutic targets to prevent metastasis.
CD44 is a ubiquitous multistructural and multifunctional cell surface glycoprotein. It participates in a wide variety of cellular functions including cellular adhesion, hyaluronate degradation, lymphocyte activation, lymph node homing, myelopoiesis, lymphopoiesis, angiogenesis, and cytokine release. CD44 is overexpressed in several tumors [159]. The binding of CD44 on migrating tumor cells to CD62 on endothelial cells is responsible for the initial steps of extravasation. Moreover, CD62 is also expressed on platelets and its binding to CD44 creates a coat that protects tumor cells from cytotoxic effector cells in in vivo models [160]. Alternative splicing determines structural and functional diversity of this protein and may be related to tumor metastasis [161].
CTCs from patients with metastatic BC, capable of metastasizing in immunocompromised mice, express CD44 [152]. Katoh et al. collected CTCs from 150 patients affected by sporadic CRC and examined the relationship between expression of the CD44v9 mRNA and prognosis through reverse transcription PCR. They showed that the survival rate was significantly lower in stage III and unresectable stage IV CRC patients with CTCs positive for CD44v9 mRNA expression and speculated that CD44v9 mRNA in CTCs could be a useful marker to predict recurrence, prognosis, and treatment efficacy in CRC patients [162].

CD47
CD47 is a cell surface ubiquitous glycoprotein, belonging to the immunoglobulin (Ig) superfamily. Upon binding to its ligand signal-regulatory protein α (SIRPα), expressed on macrophages and DCs, CD47 inhibits phagocytosis by these cell types [163]. For this reason, CD47 is known as a 'don't eat me signal' [164] and its upregulation on CTCs might confer a nonimmunogenic profile, enabling them to escape from phagocytosis. An overexpression of the CD47 gene was found in CTCs from CRC patients as compared to corresponding primary tumor tissue, suggesting a potential survival advantage [153]. CD47 was also expressed in CD44 + CTCs from a progressive metastatic BC patient. The same patient had a primary tumor negative for CD47 but developed a bone metastasis with a high expression of CD47 after seven years; this suggests that CD47 expression was probably acquired during the initiation of metastatic dissemination [152].

PD-L1
PD-1 is a member of the B7/CD28 family of co-stimulatory receptors. It regulates T cell activation through binding to its ligands, PD-L1 and PD-L2, both of which are expressed on many other cell types. When PD-L1 binds to PD-1, a strong inhibitory signal is transmitted into the T cell, leading to a reduction of cytokine production and suppression of T cell proliferation. Under physiological conditions, the PD-1/PD-L1 (or PD-1/PD-L2) interaction is necessary to mediate the natural immune tolerance [165]. In some tumors this protective mechanism is led to perversion through the overexpression of PD-L1 with the consequent prevention of an immune response against cancer [166]. The expression of PD-L1 has been demonstrated on CTCs of several malignancies [147][148][149][150][167][168][169][170][171][172] and associated with a poor prognosis [148,149,167,170].

EpCAM
EpCAM is a cell surface glycoprotein that has gained considerable interest in the diagnosis and treatment of cancer because it is frequently overexpressed in epithelial tumors [173]. To date, EpCAM is the antigen of choice for CTC enrichment from patient blood samples and this principle is at the basis of the CS system, the only clinically-validated, FDA-cleared system for identification, isolation, and enumeration of CTCs from blood samples.
Although the available technologies are mostly EpCAM-dependent, the detractors of the use of this protein complain that a fraction of CTCs cannot be quantified yet through the CS (more aggressive, undifferentiated, or EMT cells?). The limits of current technologies for isolation of CTCs represent a hot topic, but their discussion is beyond the scope of our review. Other authors in this special issue, as well as ourselves elsewhere, have already extensively addressed these arguments [174]. For our purpose, it is enough to remember herein that, before arriving to patient's bedside, a tumor marker must demonstrate its analytical and clinical validity and, finally, its clinical utility. The quality of available data determines the level of evidence, the strongest being Level 1, i.e., the definitive demonstration of clinical utility that can be obtained through a single, high-powered, prospective, randomized, controlled trial or from a meta-analysis or overview of multiple, well-designed studies. The European Pooled Analysis Consortium (EPAC) demonstrated the clinical validity of the CS assay, with Level 1 evidence in 2014 [5].
We should also remember that the meaning of any malignant feature of CTCs should be judged according to the degree of clinical validation of a certain phenotypical/molecular characteristic that we are using to identify CTCs in peripheral blood [175]. For this reason, after looking for potential targets in immunotherapy and their association with patient survival, we found and reported here, for the most, studies exploiting EpCAM-based technologies.
EpCAM overexpression has been associated with both decreased and increased survival of patients [176]. Dynamic changes in EpCAM expression frequently occur during tumor progression and its downregulation was observed during EMT [177]. Evidence suggests that epithelial plasticity could also be implicated in tumor immune escape [178]. In particular, the acquisition of an EMT phenotype has been associated with an inhibition of CTL-mediated tumor cell lysis in the human MCF-7 cell line [179]. Since EMT CTCs have also been correlated with disease progression and chemotherapy resistance [10,180]; enrichment systems that allow simultaneous investigation of both EpCAM + and EpCAM − CTCs are being employed in order to obtain more complete information about the role of EpCAM in tumor progression [181].

CTCs as Biomarkers in Cancer Immunotherapy
The progress achieved in recent years in understanding the molecular mechanisms underlying cancer has allowed the development of targeted therapies, with the hope to select the most appropriate treatment for individual patients. Among these strategies, immunotherapy has brought enormous progress to cancer treatment.
The main goal of cancer immunotherapy is to reinforce the patient's suppressed immune system, ideally restoring its capability to eradicate cancer. The mechanisms of tumor escape from immune surveillance represent a 'druggable' Achilles' heel for restoring immune control.
Cancer immunotherapy approaches can be passive or active. Passive immunotherapy is mainly used in case of weak or negative immune response and consists of ex vivo-activated cells or molecules that, once re-injected into the body, compensate for missing or deficient immune functions. This approach includes infusion of tumor-specific monoclonal antibodies (mAbs) directed against several targets (i.e., oncogenic pathways and osteoclast functions), infusion of cytokines, and adoptive cell transfer (ACT). Active immunotherapy strategies aim to stimulate in vivo a pre-existing immune response. To apply active immune-therapeutics, the patient's immune system should be able to be competently stimulated and to mediate effector functions. This group includes vaccines, immune checkpoint inhibitors, and oncolytic viruses [182].
Since the detailed description of immunotherapeutic strategies goes beyond the aim of this review, we will focus our attention on the most successful treatments of solid tumors, in which CTCs have been investigated as biomarkers for patient monitoring (Table 1).
It is important to note here that most of the primary tumor targets listed below are not specific for CTCs alone. Indeed, some of them have already been detected as soluble markers or circulating tumor DNA (ctDNA) for a selection of tailored treatments, with lower costs and simpler execution as compared to CTC detection. For example, ctDNA was detected with a higher frequency than CTCs in metastatic BC [183]. However, while CTC numbers correlated with prognosis, baseline ctDNA levels did not. This suggests that ctDNA might be more useful in identifying mutations for therapeutic targets, rather than as prognostic biomarker [183].

Blocking of Oncogenic Pathways
Proteins involved in oncogenic pathways are often overexpressed during carcinogenesis and can be targeted to avoid the proliferation of tumor cells [184]. Tumor-specific mAbs induce tumor cell death by directly binding to tumor targets or stimulating antibody-dependent cellular cytotoxicity (ADCC) [185], complement-dependent cytotoxicity (CDC) [186], or antibody-dependent cellular phagocytosis (ADCP) [88]. Examples of mAbs directed against oncogenic pathways are anti-EGFR and anti-HER2.

EGFR
The epidermal growth factor receptor (EGFR) is a transmembrane receptor with tyrosine kinase activity that, upon activation by EGF or other ligands, initiates mitogenic signaling across several pathways [187]. Overexpression of EGFR is associated with a more advanced disease and a more unfavorable prognosis; since it occurs in several malignancies, this pathway represents an ideal therapeutic target [188]. Cetuximab (Erbitux, Eli Lilly) was the first mAb prescribed to treat patients with advanced CRC expressing EGFR and was approved by the FDA in 2004. Thereafter, the FDA approved other anti-EGFR mAbs, either as single agents or in combination with other drugs. EGFR expression on CTCs has been demonstrated in patients with advanced breast, prostate, lung, and colorectal cancer by using the CS platform [189][190][191][192]. In CTCs from CRC patients, wide intra/interpatient variability in expression and gene amplification levels of EGFR was observed, which might explain differences in treatment response [193]. The persistence of CTCs, enriched and detected by AdnaTest from 38 advanced RAS-BRAF-wild-type CRC patients during treatment with cetuximab-irinotecan or panitumumab, was related to a decrease in OS and PFS [194]. Kuboki et al. demonstrated that a high CTC count assessed by CS predicted a decrease in OS in 63 patients with advanced CRC receiving cetuximab in combination with chemotherapy as third-line treatment, but EGFR expression in CTCs did not predict response to cetuximab [195].

HER2
The human epidermal growth factor receptor 2 (HER2, also called HER2/neu or ERB-B2) is a transmembrane glycoprotein receptor with intracellular tyrosine kinase activity. Upon ligand binding, HER2 activates the PI3K/Akt signaling pathway, leading to cell proliferation and survival. HER2 gene amplification was initially reported in almost 30% of primary BC patients [196]. Trastuzumab (Herceptin, Genentech) is the first humanized IgG1-class mAb indicated for the treatment of HER2-amplified BC and was approved by the FDA in 1998.
HER2 expression on CTCs has been extensively tested in BC patients [197][198][199][200][201][202]. Notably, Meng and coworkers demonstrated in 24 patients with HER2-negative primary tumors that nearly 40% acquired a HER2 gene amplification in CTCs during cancer progression; they first obtained proof of concept that patients treated with a Herceptin-containing therapy had a partial or complete response [197]. Zhang et al. observed in 101 metastatic BC patients that HER2 status was different between CTCs and tumor tissues and that, in CTCs, it predicted the outcome of patients receiving anti-HER2 therapy. Indeed, about 62% of histologically HER2-positive patients had HER2-negative CTCs (detected through the CS system) at the time of sampling. Moreover, although all histologically HER2-positive patients received anti-HER2 therapy, the median PFS of HER2-positive CTC patients was significantly longer than that of HER2-negative CTC patients (8.5 vs. 3.5 months, p < 0.001) [202].

Osteoclast Regulation
Bone metastasis is common in solid tumors, particularly in breast and prostate cancer. It is the result of a complex process, in which tumor and immune cells participate by releasing cytokines and growth factors. The RANK/RANKL/osteoprotegerin (OPG) axis plays a key role in bone turn-over and is deregulated in many tumors. The differentiation and maturation of osteoclasts are mediated by the binding of RANKL to RANK. RANKL is produced by osteoblasts and stromal cells, while RANK is expressed on pre-osteoclasts. The RANK-RANKL interactions are tightly regulated by OPG, which acts as a soluble decoy receptor by preventing the binding of RANKL to RANK and blocking its activation, thereby inhibiting osteoclast genesis [203]. In a mouse model, tumor cells expressing RANK were shown to migrate to the bone, perhaps attracted by RANKL, which is abundantly expressed in this tissue [204]. Santini et al. demonstrated that RANK expression in primary BC is a predictive marker of bone metastasis occurrence and shorter skeletal DFS [205].
The RANK-RANKL interaction is a promising target for mAb immunotherapy in advanced cancer disease. Denosumab (Xgeva, made by Amgen Inc., and Prolia, made by Amgen, Inc.) is a human IgG2 mAb with a high affinity and specificity for human RANKL. By binding to RANKL, it prevents RANKL interaction with RANK (in a similar way to OPG), thus reducing the differentiation, activity, and survival of osteoclasts [206]. Xgeva and Prolia were approved by the FDA, respectively in 2010 and 2011, the former being indicated for the prevention of skeletal-related events in patients with bone metastases from solid tumors, and the latter for the treatment of bone loss in patients with prostate or breast cancer undergoing hormone ablation therapy.
Concerning the expression of RANK on CTCs, Gray et al. analyzed circulating melanoma cells (CMCs) from 56 melanoma patients (40 late-stage and 16 early-stage) for RANK expression through multiparametric flow cytometry. RANK + CMCs were detected in 22/40 late-stage and 4/16 early-stage patients. Interestingly, in two patients with 100% and 75% RANK + CMCs, immunofluorescence staining of metastases, which had been removed one month prior to blood collection for CTC analysis, showed only a small fraction (2%) of RANK + cells within the tumor. Furthermore, a change in CMC numbers after treatment start was not found, but the percentage of RANK + CMCs increased after therapy with BRAF inhibitors and this was associated with a shorter PFS, whereas in patients treated with immune checkpoint inhibitors the increase in RANK + CMCs was not apparent [43].

Immune Checkpoint Inhibitors
T cell activation is regulated at different levels during immune responses to prevent autoimmunity. The cytotoxic T-lymphocyte associated protein 4 (CTLA-4) and PD-1 immune checkpoint pathways play a key role in peripheral tolerance by operating at different stages of immune responses. CTLA-4 stops potential autoreactive T cells in the initial phase of naïve T cell activation, typically inside lymph nodes. Conversely, the PD-1 pathway regulates previously-activated T cells in the later stages of an immune response, primarily in peripheral tissues. As discussed above, cancer cells, which should be recognized and killed by T cells, have developed methods to evade the host's immune system by exploiting peripheral tolerance [207].

CTLA-4
CTLA-4 is a key inhibitor receptor that influences T cell function. In resting T cells, CTLA-4 is located in the intracellular compartment and is transported and expressed on the cell surface only after activation upon CD28 binding to B7-1 (CD80) and B7-2 (CD86) on APCs [208]. Once on the cell surface, the CTLA-4 inhibitory signal is transmitted through the binding of B7-1 and B7-2 on B cells and activated monocytes. Compared to CD28, CTLA-4 binds B7 molecules with a higher affinity and blocks further costimulation [209]. CTLA-4, therefore, downregulates T cell responses and APC function, resulting in immune tolerance [210].
Studies that use CTCs as biomarkers during anti-CTLA-4 mAb treatments are few. Khoja et al. and Klinac et al. employed CMCs as treatment response biomarkers in patients with melanoma receiving ipilimumab and/or other drugs, respectively using the CS and a manual multimarker immunomagnetic enrichment assay followed by microscopy quantification; however, in both cases the number of considered patients was insufficient to draw reliable conclusions [216,217]. Recently, Hong and collaborators isolated CMCs, by means of microfluidic enrichment, from 16 metastatic melanoma patients undergoing therapy with ipilimumab, and developed a CTC scoring assay to evaluate a 19-gene digital RNA signature. They showed that the use of this quantitative CTC score applied to the serial monitoring of patients was predictive of long-term response to immunotherapy, thus offering an alternative to the analysis of repeated tumor biopsies, which are invasive and insufficiently precise to guide new or ongoing treatments [218]. However interesting as these approaches are, to clarify the role of CMCs/CTCs in the monitoring of patients treated with anti-CTLA-4 mAbs, further studies are needed in large, well-defined subgroups of patients undergoing the same treatment regimen.

The PD-1/PD-L1 Axis
As discussed above, a tumor overexpressing PD-L1 protects itself from T cells, leading to exhaustion and neutralization. The PD-1/PD-L1 inhibitors prevent the PD-1/PD-L1 interaction, thus facilitating an efficacious immune response against the tumor. Clinical studies indicated that antibodies that block PD-1 and PD-L1 have a reliable effect on many advanced malignancies [219]. The PD-1 blockers pembrolizumab (Keytruda, Merck & Co., Inc., Kenilworth, NJ, USA) and nivolumab (Opdivo, Bristol-Myers Squibb Company, New York, NY, USA) were approved by the FDA in 2014 for patients with advanced melanoma and have subsequently been approved for other cancer types.
PD-L1 can be detected by immunohistochemistry on tumor or immune cells; however, the utility of this marker in predicting which patients might benefit from immune checkpoint inhibitors is controversial [220]. Indeed, since most PD-L1 positive tumors are not affected by anti-PD-1/PD-L1 therapy, the predictive value of PD-L1 in tumor biopsies is so low that it is unacceptable to use as a biomarker in treatment selection [221]. Furthermore, a discordant PD-L1 expression between primary tumors and metastases was observed [222].
Mazel and colleagues first reported the expression of PD-L1 in CTCs from 16 patients with HR + , HER2metastatic BC, as assessed by the CS platform. CTCs expressing PD-L1 ranged from 0.2 to 100% and PD-L1 intensity varied between different patients and between CTCs within the same sample. The authors supposed that PD-L1 + CTCs might be able to escape the immune system control and, therefore, represent a target for anti-PD-L1 therapies [147].
To understand whether CTCs that express PD-L1 could represent a predictive biomarker during anti-PD-1 therapy, Nicolazzo and colleagues investigated the expression of PD-L1 in CTCs from 24 patients with metastatic NSCLC treated with nivolumab and deduced that it assumes a predictive significance after several months from the beginning of the therapy [148]. Recently, many papers have been published in which the expression of PD-L1 on CTCs has been studied in patients undergoing immunotherapy [148,149,168,169]. Despite the different CTC enrichment and enumeration platforms and PD-L1 expression assays, the majority of these studies agree in affirming that a high PD-L1 + CTC burden is associated with worse OS and PFS [148,149,167,170]. However, it seems premature to draw definitive conclusions because of the low number of analyzed patients.
In Section 4.3.1, we cited the paper by Hong et al. describing a novel method based on a digital RNA signature and a CTC scoring assay to predict patient response to ipilimumab. Noteworthy, the authors also successfully tested their CTC score on samples from 33 melanoma patients receiving pembrolizumab, hence showing the potential clinical validity of this use of CMCs/CTCs as a noninvasive biomarker in cancer immunotherapy [218].

Adoptive Cell Transfer (ACT)
ACT has been proven to be an effective immunotherapeutic method for the treatment of cancer and has achieved promising results in anticancer clinical trials [223]. One of the first anticancer ACT therapies was based on infusion of tumor infiltrating lymphocytes (TILs) for the treatment of melanoma [224] and obtained an objective response rate of 49-72% when the preparative chemotherapy-induced lymphocytic depletion was performed before TIL infusion [225]. Other ACT-based approaches use CTLs, NK cells, genetically-engineered lymphocytes expressing highly active TCRs, and chimeric antigen receptors (CARs) [226]. In NK cell-based immunotherapy, some studies have shown a decrease in the number of CTCs, which appears to be associated with treatment efficacy and positive patient outcome [227][228][229]. In particular, Liang et al. compared two groups of patients with recurrent BC, the first (n = 18) treated with autogeneic NK cells and the second (n = 18) with allogeneic NK cells. In the group treated with allogeneic NK cells, they found that the number of CTCs, enriched by immunomagnetic sorting and counted by FACS, decreased significantly (from 13.13 ± 5.83 before treatment to 6.88 ± 4.95 one month after the final transfusion, p = 0.01). In contrast, changes in CTC levels in the group treated with autogeneic NK cells were not significant (p > 0.05) [227]. Using the same enrichment and counting systems, CTCs have also been studied to evaluate allogeneic NK cell-based ACT in stage IV NSCLC patients (n = 31). Lin et al. observed a decrease in CTC numbers, from 27.12 ± 9.286 one day before treatment to 14.02 ± 5.872 at 30 days after the treatment [228]. Qin et al. also observed a decrease in CTC counts in stage II-IV hepatic carcinoma patients (n = 16) one month after immunotherapy with allogeneic NK cells [229].

Cancer Vaccines
A large number of therapeutic vaccines against cancer have shown only little effect, perhaps due to the use of non-mutated self antigens (unable to mediate effective antitumor responses), monovalent antigen-targeting strategies that may select resistant tumor variants, or suboptimal delivery systems that resulted in weak and short-lived antigen-specific T cell responses. A further obstacle is represented by a highly immunosuppressive tumor microenvironment [230].
The identification and validation of predictive biomarkers that accurately reflect immune responses in tissues, including tumors, will be important future tasks of the Human Vaccines Project [231]. Some studies have already been published on the use of CTCs for this purpose. Stojadinovich et al. detected CTCs (by CS) in patients without clinical evidence of BC but at high risk of recurrence. Patients were immunized with the E75-peptide vaccine and a significant reduction in both total CTCs and HER2/neu + CTCs was demonstrated over the course of vaccination. These preliminary data suggest that enumeration of CTCs might serve as a surrogate marker to monitor the immunologic response to BC-targeting vaccines [232].

Immunotherapies Targeting CTCs
We have seen that CTCs were proven to be useful prognostic markers and that researchers are working assiduously to define their predictive utility. In this last paragraph, we report some preliminary studies on immunotherapies against CTCs. These studies, performed in vitro or in murine models, give an idea of the variety of targets and strategies that can be considered in cancer therapy.
The anti-CD44 antibody RG7356 prevented tumor cell adhesion to hyaluronic acid in vitro and caused tumor growth inhibition in vivo in xenograft models using cell lines expressing the CD44s isoform [233]. It has also been demonstrated that CD44-targeting in xenografts results in tumor cell phagocytosis mediated by macrophages. It has been supposed that the RG7356-based treatment induces an upregulation of cytokines and chemoattractants that recruit and activate TAMs [234]. Although TAMs mainly promote tumor progression, it has recently been shown that they are able to phagocytize tumor cells in the presence of mAbs targeting specific tumor antigens [235].
In primary human bladder cancer, a marked expression of CD47 was seen in CD44 + tumor initiating cells (TICs) that escape phagocytosis. In the same study, an anti-CD47 mAb restored the ability of macrophages to phagocytize TICs in vitro [236].
Monoclonal and recombinant antibodies have been used to directly target EpCAM + cells [237][238][239][240]. A recombinant vaccinia virus expressing full-length EpCAM (VV GA733-2) was able to promote macrophage-mediated ADCC of antigen-positive CRC targets in a murine model [241]. Although promising results were obtained in preclinical studies, the efficacy of EpCAM-directed mAbs and of VV GA733-2 still remains to be demonstrated through large clinical trials [242].
Other preliminary studies could have interesting implications in CTC research. Gül et al. showed in mice inoculated with B16F10 melanoma cells that unstimulated Kupffer cells could bind, but not eliminate tumor cells circulating into the liver; interestingly, when stimulated with a tumor-specific, anti-gp75 mAb, Kupffer cells efficiently removed tumor cells in an ADCP-mediated manner. Phagocytosis of tumor cells was dependent on both the high-affinity IgG-binding Fcγ receptor I (FcγRI) and on the low-affinity FcγRIV, and resulted in inhibition of liver metastasis [88]. Mitchell et al. developed peripheral blood leukocytes coated with liposomes previously conjugated with E-selectins and TNF-related apoptosis inducing ligand (TRAIL). When mice intravenously inoculated with tumor cells were treated with the functionalized leukocytes, intercepted cancer cells in the circulation underwent TRAIL-mediated apoptosis. Since the cytotoxic activity of these leukocytes resembled that of NK cells, they were named 'unnatural killer cells' [243]. In conclusion, these two examples may represent promising therapeutic strategies to improve the immune antitumor intervention in CTC-positive cancer patients.

Future Opportunities
Compared with the scenario delineated at the end of the last century, modern oncology has made an epochal breakthrough in patient treatment: the findings of molecular oncology now allow selection of the best treatment for individual patients at any time during disease evolution, maximizing the results and minimizing the side effects. Immunotherapy represents a further improvement, because it permits manipulation of the immune system in order to stimulate recognition and elimination of cancer cells, leaving intact normal surrounding tissues.
However, as in all the best stories, evil is lurking and promises to complicate matters. In our case, the challenge is represented by heterogeneity and plasticity of the tumor that, under pressure from treatment with a targeted drug, is able to evade towards alternative, yet-unexplored pathways, or to favor the growth of a minority of tumor cells insensitive to the ongoing treatment. The only option, in this not-so-unusual case, is to recognize as early as possible the incoming resistance and to change treatment.
Alternatively, we should combine different drugs to target, at the same time, different tumor cell subsets at the lowest dose sufficient to reduce tumor burden without inducing resistance and limit side effects of the drug combination. The cooperative support of immunotherapy, focused on eliminating damaged cancer cells, also promises, in principle, a successful eradication of microscopic groups of tumor cells. The utility of this kind of therapeutic strategy needs to undergo the proof of facts, in ad hoc designed clinical studies based on a robust group of biomarkers.
For the reason explained in the introduction, we believe that CTCs are a good point of observation of tumor evolution and research on the interplay between CTCs and the immune system might be useful to control tumor growth and, hopefully, block it.
By reviewing the literature, we showed that most cell populations of the immune system interact with CTCs, conditioning their shedding from the primary tumor, survival, homing, or even their further growth. In principle, the ambiguous relationship between CTCs and the immune system, with its antior protumor functions depending on the context, might be a resource rather than a problem if we are able to manipulate the immune system to our scope. The main lesson is that any interaction with the immune system as well as any characteristic of CTCs might be transformed into an Achilles' heel of the tumor.
In the near future, the knowledge on the metastatic mechanisms that we are gaining from studies on CTCs will allow the implementation of a growing number of immunotherapeutic strategies-some of these options are already real. For example, in 2004, Mosolits et al. identified a promising anti-EpCAM vaccine: they treated 13 CRC patients with an anti-EpCAM vaccine (n = 7) or an anti-idiotypic antibody mimicking EpCAM (n = 6) in combination with GM-CSF. They detected a long-lasting EpCAM-specific proliferative T cell response in the first group compared to the second [244]. In a more recent work, Choi and coworkers showed that vaccines based on DCs pulsed with EpCAM peptides elicit a strong antigen-specific CTL response and result in a significant suppression of tumor growth in a mouse model. This study, performed with the human hepatoma cell line HepG2, suggests that peptides from cancer stem cells (CSCs) might be a source of antigens for vaccination-based immunotherapy, with the aim to eliminate the CSCs responsible for tumor relapse [245].
On the other hand, since the EMT process is thought to play a role in tumor progression and therapy resistance, immunotherapies directed against tumor cells undergoing EMT are being investigated. The vaccines directed against brachyury, a transcription factor that is associated with tumor EMT, represent an example ( Table 1). As reviewed in detail in the work by Hamilton and colleagues, these vaccines are an attractive therapeutic strategy to be used in combination with other treatments [246].
Finally, research on CTCs and immunotherapy of tumors will soon benefit from the extraordinary advances recently achieved in the fields of genomics and transcriptomics, at both a molecular and a computational level. High-throughput technologies, such as next-generation sequencing (NGS), and computational tools for data analysis are now available to study tumor escape mechanisms and to unveil still unknown interactions between tumor and immune cells [247]. Strikingly, these technologies can also be applied to CTCs. For example, single-cell RNA-Seq is already being used to investigate origin, phenotype, and drug resistance pathways of CTCs [248,249].
It has recently been reported that a new procedure, based on the collection of a large volume of peripheral blood (the Diagnostic Leucapheresis), increases the total number of collected CTCs up to 30 fold, thus promising to make high-throughput technologies usable tools for the great majority of patients [250].
Future possible applications of these tools could allow, for instance, the discovery of new interactions between CTCs and the immune system; potentially-immunogenic neoantigens on CTCs in order to target them and/or the primary tumor by means of immunotherapy; neoantigens on CTCs that might reduce the success rate of immunotherapies recognizing only the original antigen (but not the mutated form); other escape mechanisms on CTCs that are shared with the primary tumor; and specific molecular targets on CTCs (rather than on biopsies) to identify and stratify patients who would benefit from immunotherapy. As demonstrated with the work by Hong et al. (which we discussed above), such innovative approaches can also allow us to follow CTC dynamics and reinforce the predictive value of CTCs as a biomarker for serial monitoring of patients during immunotherapies and other treatments [218].

Conclusions
By reviewing the main studies that directly or indirectly address the relationships between the immune system and CTCs we could observe that this is a new frontier of cancer research with a promising impact in the clinics.
However, concerning the feasibility of a prompt transfer from lab benchtop to patient bedside of the concepts discussed in these pages, it seems opportune to introduce two cautionary notes. First, the number of patients analyzed in the majority of the studies discussed here is very low; second, often, the experimental evidence obtained in in vitro or murine models has never been tested in clinical trials. It is clear that further studies in homogeneous groups of patients, who are undergoing the same treatment regimen and for whom a comprehensive characterization of their disease is available (including molecular markers and considering primary tumor, CTCs, and possible metastases), are necessary before we will be able to fully exploit the translational potential of CTC research in the oncoming era of cancer immunotherapy.
Finally, the recent molecular and computational genomic tools are gaining increasing attention thanks to the huge quantity of novel information they can provide to researchers. Over the next few years, we will probably witness a revolutionary progress in oncology and CTC-aided immunotherapy of tumors.
Author Contributions: R.Z. drew down the outline of topics; K.L. treated immune surveillance and escape; C.P. discussed biomarkers and immunotherapies; K.L., C.P., and R.Z. critically revised the selected scientific papers. Acknowledgments: The authors thank Elisabetta Rossi and Antonella Facchinetti (IOV-IRCCS, Padua, Italy) for critical reading of the manuscript and advice, as well as Christina Drace (IOV-IRCCS, Padua, Italy) for English editing. The concepts developed in this review were derived from the work performed at the CTC laboratory of IOV-IRCCS, Padua, Italy.

Conflicts of Interest:
The authors declare no conflicts of interest.