Molecular Imaging in Immunotherapy: A Systematic Review

: Immunotherapy is an effective therapeutic option for several cancers. In the last years, the introduction of checkpoint inhibitors (ICIs) has shifted the therapeutic landscape in oncology and improved patient prognosis in a variety of neoplastic diseases. However, to date, the selection of the best patients eligible for these therapies, as well as the response assessment is still challenging. Patients are mainly stratified using an immunohistochemical analysis of the expression of antigens on biopsy specimens, such as PD-L1 and PD-1, on tumor cells, on peritumoral immune cells and/or in the tumor microenvironment (TME). Recently, the use and development of imaging biomarkers able to assess in-vivo cancer-related processes are becoming more important. Today, positron emission tomography (PET) with 2-deoxy-2-[18F]fluoro-D-glucose ([ 18 F]FDG) is used routinely to evaluate tumor metabolism, and also to predict and monitor response to immunotherapy. Although highly sensitive, FDG-PET in general is rather unspecific. Novel radiopharmaceuticals (immuno-PET radiotracers) able to identify specific immune system targets are under investigation in pre-clinical and clinical settings. In this review, we will provide an overview of the main new immuno-PET radiotracers in development. We will also review the main players (immune cells, tumor cells and molecular targets) involved in immunotherapy. Furthermore, we report current applications and the evidence of using [ 18 F]FDG PET in immunotherapy, including the use of artificial intelligence (AI).


Introduction
The origins of immunotherapy date back to the discovery of smallpox vaccination by Edward Jenner in 1796. Scientific interest in the mechanisms of immune surveillance of diseases has been high since then. In the last years, several findings on the involvement of both innate and adaptive immunity in the mechanisms of immune surveillance of cancer led to the emergence of immunotherapy, which is today considered one of the most promising anti-cancer approaches [1]. The knowledge of the complex and dynamic interactions between a tumor and the immune system has been fundamental for the development of increasingly specific, personalized and targeted therapies. This was facilitated by the progress of modern medicine through increasing multidisciplinary and translational approaches, including omics sciences (e.g., genomics, proteomics, metabolomics). The same knowledge is also exploited by the growing field of molecular imaging through the identification of diagnostic biomarkers to support clinical decisionmaking and patient selection by non-invasive, in-vivo assessment of specific therapy target expression, therapy response and toxicity. In recent years, also technological development assumed an important role in supporting clinical decision-making and selecting patients as immunotherapy candidates. First, new digital tomographs have greatly improved the detectability of small lesions, but also the use of artificial intelligence (AI), may improve the interpretation of so-called "big-data" extracted from images and allow for their integration with other medical data. Here, we will review the state of the art and the perspectives in the aforementioned aspects in the field of immunotherapy, with a special focus on molecular imaging.
No language restriction was applied to the search, but only articles in English were reviewed. The systematic literature search returned 822 articles. According to the PRISMA flow-chart, after duplicate removal, 86 articles have been considered, fully read, analyzed, and extensively described according to their title and abstract as previously described [2]. We also checked for further relevant articles in the references of the articles included in the retrieved literature.

Tumor Microenvironment
As recently observed, cancer should not be considered as a simple collection of transformed cells, but as an organic and evolving complex of transformed tumor cells and other cellular and molecular components that together constitute the tumor microenvironment (TME) [1]. On one side, tumor cells and stromal cells, such as cancerassociated fibroblasts (CAFs), adipocytes, and vascular endothelial cells, co-evolve and grow within a protective extracellular matrix (ECM) with the help of cytokines, glycoproteins, proteoglycans and growth factors. On the other side, tumor-infiltrating immune cells present in the TME try to identify and destroy growing tumor cells. This defensive behavior may also be associated with tissue inflammation and a protumorigenic effect responsible for the selection of more aggressive tumor clones, able to block and evade the same immune defensive mechanism [3]. The two major players of the anti-tumorigenic effect of the immune system are the natural killer (NK) cells and the tumor-infiltrating lymphocytes (TILs), which recognize and kill the tumor cells through different mechanisms. First NK cells target cells that do not present the major histocompatibility class I (MHC-I) on their surface. Once activated, NK cells release cytolytic granules to trigger programmed cell death. Differently, CD8 + T-lymphocytes and CD4 + T helper lymphocytes 1 (Th1) are activated by antigen-presenting cells (APCs), such as dendritic cells (DCs), but also by mononuclear phagocytes and B cells. The APCs can recognize and capture extracellular proteins from tumor cells and present them as antigens in the form of major histocompatibility complex molecules of class II (MHC-II) [4]. NK cells, CD8 + and CD4 + Th1 T-lymphocytes destroy tumor cells producing cytokine and chemokines, especially interleukin 2 (IL2) and interferon-gamma (IFNγ), which also prevent metastatic spread of the tumor [5].
Second tumor-infiltrating immune cells play a more ambivalent role and may as well have an anti-tumorigenic or a pro-tumorigenic effect based on a delicate balance of different mechanisms within the TME. Namely, other TILs can exhibit a pro-tumorigenic effect, such as CD4 + T helper lymphocytes 2 (Th2), producing interleukins with an inflammatory effect, and the regulatory CD4 + T-lymphocytes (Treg), which are indispensable for maintaining homeostasis, down-regulating the CD8 + T-lymphocyte activity [1,6]. Also, the presence of tumor-associated macrophages (TAMs) is related to a poor prognosis and reduced overall survival (OS). TAMs phenotype can be polarized by the TME characteristics, resulting in the development of pro-inflammatory macrophages (M1 type, driven by IFNγ and tumor necrosis factor-alpha (TNFα) or anti-inflammatory macrophages (M2 type, driven by IL4 and IL13). While the former ones have an antitumorigenic effect, the latter ones promote tumor progression by stimulating angiogenesis, remodeling the ECM, promoting metastasis and immunosuppression [1,3,5,7,8]. Tumor-associated neutrophils (TANs) may also show a similar behavior: they can be polarized, resulting in the development of an anti-tumorigenic type (N1) or a protumorigenic type (N2). Driven by the action of transforming growth factor-beta (TGF-β) secreted by the CAFs, N2 stimulate cancer proliferation and migration through the secretion of matrix metalloproteinases (MMPs) and IL1β [3,4,9]. Finally, the role of tumorinfiltrating B cells (TIBs) is even more controversial: CD20 + TIBs have been shown to behave as APCs with an anti-tumorigenic effect in non-small-cell lung cancer (NSCLC), ovarian cancer [10,11] and in melanoma, as part of the tertiary lymphoid structures (TLS) associated with CD8 + T-lymphocytes [12]. Moreover, CD20 + TIBs can produce granzyme B and perforin, which directly leads to the activation of apoptotic pathways in tumor cells. In contrast, however, Shalapour et al. demonstrated that TIBs seem to inhibit the antitumorigenic effect of CD8 + T-lymphocytes, promoting tumor cell growth and migration [13].

Immunotherapies targets
The understanding of the complex mechanisms regulating immune cell infiltration in the TME is still evolving. A comprehension of this intricate system is crucial in order to develop new immunotherapy strategies for cancer treatment, such as immune checkpoint inhibitors (ICIs), adoptive cell transfer (ACT), oncolytic virus therapies, cancer vaccines and cytokine therapies (Fig. 1).

Figure 1.
Simplified schematic representation of the tumor microenvironment and the major immunotherapy strategies for cancer treatment.

Immune checkpoint inhibitors
ICIs are nowadays the most commonly used immunotherapy strategy. ICIs represent a class of monoclonal antibodies (mAbs), able to counteract the immunosuppressive action of the immune checkpoint (IC) proteins and re-establishing the anti-tumorigenic function of the tumor-infiltrating immune cells present in the TME [1,3,5,14]. As the above-mentioned Treg, ICs are fundamental for maintaining homeostasis, downregulating CD8 + T-lymphocyte activity, and for preventing autoimmune reactions [15]. Cancer cells have learned to take advantage of this mechanism, producing and overexpressing ICs on their cell surface and down-regulating the immune system [16].
Cytotoxic T-lymphocyte associated protein 4 (CTLA-4), programmed cell death protein 1 (PD-1) and its ligands (PD-L1 and PD-L2) are the protagonists of this process. CTLA-4 is a glycoprotein expressed on the T cell surface, which allows tumor cells to escape the immune system by binding to CD80/86 on APCs. Anti-CTLA-4 therapies, such as ipilimumab, block this mechanism [17]. Furthermore, tumor cells can express PD-L1 and PD-L2 on their surface, thereby blocking the activity of T cells, B cells and NK cells, binding their surface receptor PD1. Anti-PD1 therapies, such as nivolumab and pembrolizumab, and anti-PD-L1 therapies, such as avelumab, atezolizumab and durvalumab, block this mechanism [18].

Adoptive cell transfer
Another widely used immunotherapy technique is the adoptive cell transfer (ACT), whose main representatives are chimeric antigen receptor (CAR)-T cells. CAR-T cells are autologous isolated T cells, genetically engineered through the use of antibody fragments to carry a receptor capable of recognizing specific antigens expressed by tumor cells, being reinfused to patients to stimulate cancer immune surveillance through secretion of perforin and granzyme granules and activation of programmed cell death. While the firstgeneration CAR-T cells should be reproduces ex-vivo, the second and third generation are capable to duplicate in-vivo and persistent in the patient's body [19][20][21]. CAR-T cells were developed for the treatment of hematologic disorders, such as acute lymphoblastic leukemia (ALL) and chronic lymphocytic leukemia (CLL) in adult patients [22][23][24]. Their use for solid cancer is still challenging due to the higher heterogeneity of tumoral antigens, the resulting increased difficulty in configuration recognition, higher difficulty to penetrate tumor tissue through the vascular endothelium and a low survival in the TME [21].

Oncolytic viruses, cancer vaccines and cytokine therapies
Another immunotherapy technique uses oncolytic viruses described as modified viruses aimed to target and kill tumor cells. In 2015, the Food and Drugs Administration (FDA) approved talimogene laherparepvec (T-Vec), also known as Imlygic, the first oncolytic modified herpes simplex virus, that infects and destroys melanoma cells in patients with unresectable metastatic disease [25,26].
Cancer vaccines are another approach in immunotherapy, relying on tumor-specific antigens to trigger T-cell activation in the TME. Its largest field of application is melanoma (MZ2-E, MZ2-D and gpl00 antigens [27][28][29]), but this technique has also been used in prostate cancer (sipuleucel-T, a dendritic cell based cancer vaccines [30]).
Despite the efficacy and initial enthusiasm for the use of cytokine therapies, especially in patients with chronic myeloid leukemia and melanoma, IL2 and TNFα have recently been sidelined in favor of the above-mentioned more promising ICIs and ACT approaches, owing to their poor tolerability and severe toxicity [1,3,5,14]. Immunotherapy is not exempt from autoimmune-like reactions and immune-related adverse events (irAEs), such as skin rash, pneumonitis, colitis, hepatitis and thyroiditis, that occur in approximately 50% of patients overall, and in 14% as grade 3-4 side effect, according to the Common Toxicity Criteria for Adverse Events (CTCAE) [18]. The exact reasons for response heterogeneity, tumor relapse and side effects are still not clear. A high immunohistochemical (IHC) staining of PD-L1 tumor expression [37,38] and/or PD-1 and CTLA-4 TILs expression [39] on biopsy specimens is the first fundamental step for treatment selection, being generally related to higher response rate to ICI therapy. However, evaluation of ICs expression is generally assessed by a single-tumor biopsy, leading to its potential underestimation [40], as tumor heterogeneity is both spatial (interand intra-tumoral heterogeneity of ICs expression) and time-related (more aggressive cell clones developing over time) [41,42]. Thus, while multiple-lesion biopsy sampling is not feasible most of the time, grading heterogeneity among primary and metastasis would be a valuable approach.

Molecular Imaging
To overcome the known limitations in the representativeness of biopsy specimens, non-invasive imaging techniques, such as computed tomography (CT), magnetic resonance imaging (MRI), positron emission tomography (PET) and single-photon emission computed tomography (SPECT) as single or hybrid modalities (i.e., PET/CT, PET/MRI, SPECT/CT) may provide valuable information for staging, patient selection, treatment response assessment, restaging and follow-up , even though optimal biomarkers to predict immunotherapy response and toxicity have not been identified yet [34].

Critical issues in the development of a good radiotracer in immunotherapy
The development of a new molecular imaging probe to answer a specific biological question is a complex process, that requires a multidisciplinary team, significant funds and several decisional stages from the laboratory bench to animal models, and from animal models to the clinic. A good molecular probe needs to overcome biological problems, such as rapid clearance, metabolism, degradation and too early excretion, nonspecific accumulation in non-target tissues, poor perfusion of the tumor and pharmacological delivery barriers [43]. Also, further potential issues related to radioactive decay and toxicity have to be considered. Ideally, the radiotracer: 1) must be composed of a radioactive isotope with an identically behavior as the stable isotopes of the same element, 2) must not change the chemical and physical properties of the biological system in which it will be introduced, 3) must not deviate the normal physiological state of the system under consideration, 4) must be highly specific and/or selective, with high plasma clearance and low plasma protein binding, 5) must have the same chemical and physical characteristics of the unlabeled compound, 6) must have good in-vivo stability, the labelling between the radionuclide and the compound must be strong enough to avoid the detection of free radionuclide, and 7) must have a blood half-time adequate to the specific biological question [44,45].
In the field of immunotherapy, new radiotracers have been developed through direct or indirect labelling methods.
Ex-vivo cell labeling techniques are currently increasing. Here, immune cells are isolated from a patient and incubated with the radionuclide ex-vivo, despite the high fragility (short viability) of the re-injected cells and only preclinical data available, till now. An in-vivo approach would allow the possibility of PET imaging at any time after cell infusion, especially this could be of particular interest in the case of CAR T cell infusion, allowing to verify their arrival and accumulation in the tumor and metastases rather than a simple assessment of their presence in the peripheral blood by serial sampling, as per current procedural lines [ [ 111 In]oxine complex, but the use of [ 89 Zr] is more promising, owing to the higher spatial resolution, higher sensitivity, and better signal-to-background ratio of PET compared with SPECT, and owing to its comparably long half-life of 3.27 days compared to 2.80 days for 111In , which is helpful to monitor cell distribution in murine models after administration [51]. However, the efflux of [ 89 Zr] is the major drawback of this radionuclide, owing to the high radiotoxicity exhibited by free [ 89 Zr]. This problem has been solved by Bansal et al., which covalently labeled [ 89 Zr] to cell surface proteins of mouse-derived melanoma cells, mouse dendritic cells and human mesenchymal cells, although only 30-50% of the cells were found to be efficiently bound [48].
In-vivo labeling techniques have been easier to implement and are in more advanced research phases with some clinical applications, already. In particular, radionuclide-labeled immunoglobulins G (IgG) antibodies (mAbs), antibody fragments or small proteins may be used to detect in-vivo the expression of ICs (PD-L1, PD-1 and CTLA-4), and/or other key molecules of immune checkpoint pathways and immune responses through SPECT or PET/CT images [15]. The choice of the right antigen expressed only on target cells with the right radionuclide is fundamental to create a good radiotracer: [ 111  Full-length antibodies (IgG, mAbs) are large proteins with a molecular weight > 150 kDa, resulting in low tumor infiltration and long blood half-life: indeed, they are eliminated in two or three weeks by either excretion (being filtered, reabsorbed and metabolized through proximal tubule of the nephron) or catabolism (through intracellular catabolism by lysosomal degradation to amino acids after uptake by pinocytosis or by a receptor-mediated endocytosis process). This demands the use of long physical half-life radionuclides increasing patient's radiation exposure and requiring image acquisition up to one week after injection. On the contrary, antibody fragments and small proteins are preferred over IgG/mAbs due to their shorter biological half-life. In fact, unbound molecules are rapidly cleared from the circulation through kidney excretion, resulting in high-contrast images with improved tumor-to-background signal ratio [53]. This phenomenon allows the use of shorter half-life radionuclides, such as [ 18 F] and [ 68 Ga], which are already widely used in clinical practice, and where images are acquired earlier (after 1-2 hours), which is a big advantage in clinical routine. Another significant advantage of using smaller imaging agents is their faster diffusion and better penetration, compared to the heavier full-length IgG/mAbs, into dense solid tumors, leading to better intra-tumoral distribution [54,55]. Another interesting new option is the use of bispecific antibodies, which are able to bind both endogenous T cells of patients and tumor-specific antigens [56]. Finally, the chelator is also critical to prevent dissociation of the radionuclide from the compound, as is often the case with [ 64 Cu], which tends to accumulate in the liver when it dissociates from the chelator [57].
Among all the new immune-cancer related radiotracers created in recent years, those associated especially with immune checkpoints and adoptive cell transfer techniques, and those targeting CD8+ T lymphocytes, which seem to have the most important anti-tumorigenic role in the TME, are the most promising ones. All the radiotracers tested in clinical trials are listed in Table 2.

PD-1/PD-L1 Pathways
As mentioned above, PD-1/PD-L1 pathway is the most widely used immunotherapeutic approach, being also the most studied pathway in pre-clinical and clinical studies. However, different targets, such as antibodies and small molecules, have been considered to identify a good radiotracer.

Labelled antibodies
Since the development of the first [ 64 Cu]DOTA labeled antibody (IgG) PD-1 radiotracer by Natarajan et al. [61] in 2015, being tested on mice bearing B16−F10 melanoma tumors, several immune checkpoint radiotracers have been developed both for SPECT and PET imaging. The development of these radiotracers required some changes in the mice-model used in the preclinical laboratory: immunodeficient mice commonly used to study xenograft tumor models do not provide important information on radiotracer distribution in healthy organs, such as liver, spleen, thymus, lung and brown fat. Therefore, immunocompetent mice have been developed to allow studying both tumor and healthy tissue distribution of these new radiotracers [62]. In 2016, Nedrow et al. studied the in-vivo biodistribution of [ 111 In]diethylenetriaminepentaacetic acid (DTPA)anti-PD-L1 antibody in an immunocompetent murine model of melanoma. Authors reported how splenic uptake of the radiotracer affected the tumor uptake and how the co-injection of labeled antibodies and 100-fold of unlabeled antibodies significantly reduced splenic uptake of the radiotracer at 24 hours, shifting the concentration of [ 111 In]DTPA-anti-PD-L1 in the blood and potentially increasing tumor uptake [63]. Owing to these results, the assessment of the influence of the cold antigen on radiotracer accumulation in the tumor has been introduced also in many other preclinical studies [64][65][66]. Several antigens have been studied for the development of PD-L1 radiotracers, including full-length IgG/mAbs, antibody fragments and small molecules, while only full-length IgG/mAbs have been used for the development of PD-1 radiotracers.
Interestingly, Kikuchi et al. [67] demonstrated that radiotherapy induced a PD-L1 upregulation in tumor, enhancing the efficacy of either treatment alone. They assessed a [ 89 Zr] labeled anti-mouse PD-L1 mAb and dynamic immuno-PET/CT imaging of two murine tumor models (head and neck squamous cell carcinoma and melanoma), alone and during anti-PD-1 mAb immunotherapy. PET/CT imaging demonstrated significantly increased radiotracer uptake in irradiated head and neck tumors compared with nonirradiated flank tumors, whilst anti-PD-1 therapy did not have the same effect.
Similar results have been proven also by Christensen et al. [68]. Authors studied the efficacy of the anti-PD-L1 (clone 6E11) conjugated with dibenzocyclooctyne-Desferrioxamine (DIBO-DFO) chelator and radiolabelled with [ 89 Zr] ([ 89 Zr]DFO-6E11) on NSCLC xenografts and syngeneic tumors models with different levels of PD-L1 in CT26 tumor-bearing mice subjected to external radiation therapy (XRT) in combination with PD-L1 blockade. [ 89 Zr]DFO-6E11 detected the differences in PD-L1 expression among tumor models and quantified the increase in PD-L1 expression in tumors of irradiated mice.

Radiolabeled small molecules
Several small molecules have been studied to overcome the aforementioned problems related to the use of full-length mAbs. Gaochao et al. [69]  In 2017, Chatterjee et al. [57] produced a PET radiotracer labeling the [ 64 Cu]DOTAGA with the WL12 peptide, able to bind PD-L1 with high affinity (half maximal inhibitory concentration (IC50) ≈ 23 nM). They evaluated the ability of [ 64 Cu]DOTAGA-WL12 to detect PD-L1 expression in-vivo by PET imaging in NSG (non-obese diabetic (NOD) scid gamma) mice, immunodeficient laboratory mice, harboring Chinese hamster ovary (CHO) tumors with constitutive human PD-L1 expression (hPD-L1) and isogenic negative control tumors. The radiotracer was able to early detect tumor PD-L1 expression, within 60 min after administration, however, with increased uptake in the liver due to the dissociation of Cu2+ from the chelator and its subsequent trans-chelation to plasma proteins, such as albumin and ceruloplasmin. Thereafter, Ravindra et al. [71] studied the WL12 peptide labeling with [ 68 Ga]DOTAGA. In immunocompetent mice, the percentage of injected dose per gram (%ID/g) at 60 min resulted much higher in hPD-L1 cells (11.56 ± 3.18), with better tumor-to-background contrast, compared to other cell lines (< 4.97). Moreover, the uptake of [ 68  , to monitor PD-L1 levels in nude mice bearing LN229 xenografts (brain cancer) with a positive expression for PD-L1. They found that LN229 xenografts were markedly visualized from 24h after injection of [ 89 Zr]Df-KN035, with high uptake enduring up to 120 h with a favorable tumor-to-muscle ratio. The same group also evaluated the impact of chemotherapy, radiotherapy or epidermal growth factor receptor (EGFR) tyrosine kinase inhibitor (TKIs) effect on the TME (PD-L1 levels) of NSCLC. PET imaging with [ 89 Zr]Df-KN035 was performed before and after EGFR-TKI gefitinib treatment to evaluate PD-L1 expression, observing that the high dose of gefitinib inhibited tumor growth and decreased the tumoral uptake of [ 89 Zr]Df-KN035 [78].

Radiolabeled Atezolizumab
Other anti-PD-L1 mAbs have been investigated in several clinical trials, such as [ 89 Zr]avelumab in a breast cancer mice-model [64] and the promising [ 89 Zr]DFOatezolizumab in a human tumorgraft model. In 2019, Vento et al. [80] showed for the first time the expression of PD-L1 in a renal cell carcinoma removed from a patient (with favorable nivolumab response) and implanted orthotopically into NOD/SCID mice. This was demonstrated using in-vivo PET images of atezolizumab (monoclonal anti-PD-L1 antibody with a mutant Fc) radiolabeled with [ 89 Zr]DFO. Also in 2019, Bensch at al. [59] presented the first-in-human study of [ 89 Zr]atezolizumab immune-PET in 22 patients with metastatic bladder cancer, NSCLC, or triple-negative breast cancer. They found a better correlation between progression-free survival (PFS) and OS with pre-treatment radiotracer uptake, compared to conventional IHC staining of PD-L1. These findings highlight the limitations of a single biopsy evaluation and show the benefit of in-vivo assessment by imaging. As in preclinical studies, 10 mg of unlabeled atezolizumab was administered together with the radiotracer to prevent rapid clearance during imaging, which was performed on days 4 and 7. Physiological biodistribution has been reported as following: low uptake in healthy brain, subcutaneous tissue, muscle, compact bone, and lung; higher uptake (increasing over time) in bone marrow, intestines, kidney, and liver. Overall, the observed [ 89 Zr]atezolizumab uptake in lymphoid tissue (lymph nodes and spleen mainly) might serve as a surrogate for the activation state of the body's immune system or as a measure for abundant PD-L1 expression. Tumor [ 89 Zr]atezolizumab uptake was generally high, with an overall geometric mean SUVmax of 10.4, but often with a heterogeneous intratumoral, intertumoral and interpatient radiotracer distribution.

Radiolabeled Pembrolizumab
In 2020, different targets have been individualized for the development of new immuno-PET radiotracers imaging PD-1 expression. In-vivo pharmacokinetics and wholebody distribution of [ 89 Zr]labeled PD-1 targeting pembrolizumab with PET in humanized mice have been recently studied by van der Veen et al. [81], with disappointing results: tumor uptake of [ 89 Zr]pembrolizumab was lower than uptake in normal lymphoid tissue, but higher compared to other organs. High uptake in lymphoid tissues (such as spleen, lymph nodes and bone marrow) was reduced by excess unlabeled pembrolizumab administration, which inversely did not affect tumor uptake. Instead, Wenping et al. [82] evaluated the biodistribution of [ 89 Zr]labeled pembrolizumab, a humanized IgG4 kappa monoclonal antibody targeting PD-1, in healthy cynomolgus monkeys as a translational model of tracking PD1-positive immune cells with better results: distribution in lymphoid tissue such as mesenteric lymph nodes, spleen, and palatine tonsils increased over time, although it was reduced when a large excess of cold pembrolizumab was co-administered with the radiotracer. Moreover, except for the liver, low radiotracer distribution was observed in all non-lymphoid tissues including lung, muscle, brain, heart, and kidney.

Radiolabeled Nivolumab
The early studies with nivolumab were fraught with problems. Similarly to [ 64 Cu]DOTAGA-WL12, Cole et al. [83] reported the presence of consistent high liver uptake concerning the in-vivo distribution of [ 89 Zr]labeled nivolumab in healthy nonhuman primates (NHP), but also the possibility to block splenic uptake by coadministration of excess cold nivolumab. England et al. [84] developed another [ 89 Zr]nivolumab radiotracer labeled with p-SCN-DFO chelator and a humanized murine model of lung cancer suitable for immuno-PET images. However, in their study, a proper accumulation of [ 89 Zr]Df-nivolumab was not reached before 48 hours, and uptake values were calculated at 168 h post-injection; moreover, salivary gland and splenic uptake could be another potential complication for proper image interpretation.
In 2018, Niemeijer et al. [58] reported the first-in-human study using whole-body PET imaging with [ 89 Zr]nivolumab (anti-PD-1) and [ 18 F]BMS-986192, an [ 18 F]labeled anti-PD-L1 fibronectin-based protein (adnectin), in 13 patients with advanced NSCLC before treatment with nivolumab. Acquisition time was 7 days after administration of [ 89 Zr]nivolumab, reflecting PD-1-expressing tumor-infiltrating immune cells, and 1 hour after injection of [ 18 F]BMS-986192, reflecting PD-L1 expression in tumor lesions. Both radiotracers showed favorable distribution in tumor, but with a high heterogeneity between patients and lesions, and a high accumulation in the spleen for the presence of immune cells, and in the liver due to catabolism of the radiotracers. Finally, a correlation between radiotracer uptake in tumors and response to nivolumab treatment has found a higher uptake of both radiotracers in responders: [ 18 F]BMS-986192 SUV peak (median 6.5 (responder) vs. 3.2 (non-responder), p = 0.03), and a similar association was found for [ 89 Zr]nivolumab (median SUV peak 6.4 vs. 3.9, p = 0.019). Moreover, due to low central nervous system (CNS) tracer penetration, both tracers showed lower accumulation in brain metastases compared to the extracerebral lesions. For non-CNS lesions, the median [ 18 F]BMS-986192 SUVpeak was higher for lesions with ≥50% tumor PD-L1 expression by IHC than for lesions with <50% expression. Similarly, [ 89 Zr]nivolumab uptake was higher in lesions with a higher aggregates of PD-1 positive tumor-infiltrating immune cells at tumor biopsy.

Cytotoxic T-lymphocyte-associated protein 4 (CTLA-4) pathway
The most important studies on CTLA-4 pathways have been conducted by Ehlerding et al. [85,86], which investigated the capacity to visualize CTLA-4 expression in a NSCLC humanized mouse model by in-vivo [ 64 Cu]DOTA-ipilimumab, [ 64 Cu]NOTA-ipilimumab and [ 64 Cu]NOTA-ipilimumab-F(ab')2 PET images. The visualization of CTLA-4 expression was feasible with all three radiotracers, although all of them demonstrated high absolute uptake in salivary glands, yielding higher salivary gland-to-blood ratios which were slightly lower for F(ab9)2 (1.78 ± 0.72 vs 1.19 ± 0.49, respectively at 48 h). [ 64 Cu] labelled ipilimumab (both antibody and fragment) showed a clear salivary gland uptake in humanized mice, and as expected, the fragment showed a faster clearance compared to the entire antibody [87]. Therefore, these radiotracers may help elucidate the response of CTLA-4-targeted checkpoint in immunotherapy.

Adoptive cell transfer (ACT) therapy
Different radiotracers for direct and indirect activated cell tracking were developed in the last years [88], such as [ 89 Zr]oxine-labelled dendritic cells that showed tumoral uptake at PET imaging of melanoma [89]. Namely, NK cells (designated to eliminate malignant cells [89]) were labelled with an [ 89 Zr]oxine complex to perform PET cell tracking in clinically relevant rhesus macaques model for the treatment of hematologic malignancies. Liver, lungs and spleen showed the higher %ID/cc, while a low signal was observed in bone marrow [90]. [ 64 Cu]-or [ 89 Zr]-labelled antibody against NKp30 were also tested in a renal carcinoma murine model, resulting in an [ 89 Zr] higher on-target contrast [91]. Mesenchymal stem cells (MSCs) are multipotent stromal cells that were shown to target tumors, however, their role in neoplastic tissue is still under investigation [92]. To analyze their biodistribution, [ 18 F]-labelled Fe3O4@Al(OH)3 nanoparticles were used to track MSC after intravenous injection with combined PET/MRI exam, showing intense liver uptake [93].

Chimeric Antigen Receptor T cell therapy (CAR-T)
CD8+ cytotoxic T lymphocytes engineered to express both chimeric antigen receptor (IL-13) and herpes simplex virus type-1 thymidine kinase (HSV1-TK) can entrap the PET radiotracer 9-[4-[18F]fluoro-3-(hydroxymethyl)butyl]guanine [ 18 F]FHBG within glioma cells. In 2017, Keu et al. [94] showed that non-invasive PET imaging with [ 18 F]FHBG can track HSV1-tk reporter gene expression present in CAR-engineered CTLs. In their experience on 7 high-grade gliomas (HGG), [ 18 F]FHBG imaging was safe and enabled the longitudinal imaging of T cells stably transfected with a PET reporter gene in patients. Further, the sodium iodide symporter (NIS) has been proposed as a non-immunogenic radionuclide reporter in ErbB T1E28z CAR therapy in two models of triple-negative breast cancer (MDA-MB-231 and MDA-MB-436). The two models are characterized by a different immune checkpoint inhibitor expression. In particular, MDA-MB-231 has a greater expression of PD-L1 compared to MDA-MB-436, and it is inversely correlated with CAR-T tumor retention [95].
A different strategy involves the direct labelling of IL13Rα2-CAR T cells (as described in the ongoing clinical trial NCT02208362) with [ 89 Zr]oxine in a glioma mouse model delivered intraventricularly or intravenously [96]. The prostate-specific membrane antigen (PSMA) is a human protein and has a low expression in normal tissue. In a study by of Minn et al., anti-CD19 CAR T cells have been transfected to express PSMA, in order to be tracked with [ 18 F]DCFPyL PET in a model of acute lymphoblastic leukemia [97]. UniCAR T cells are modular CAR T cells that can recognize a specific target on the surface of tumor cells. Fused with PSMA-11 epitope, they produce a theragnostic complex that might be useful for retargeting UniCAR T cells in prostate cancer tissue and for treatment response assessment by PET imaging [98].

CD8+ T lymphocytes
As mentioned above, non-invasive monitoring of active cytotoxic CD8+ T lymphocytes in TME is crucial to predict the immunotherapies' antitumor responses [99,100]. Their presence has been related to a favorable prognosis in different tumors [101,102].
CD8 is expressed on cytotoxic CD8+ T lymphocytes. It is an antigen highly suitable as a target for immune cell-specific radiotracers [103]. A great effort was made in the last years to optimize radiotracers for detection of CD8+ and CD4+ T cells in TME by immuno-PET imaging. One promising approach is a mini-body labeled with [ 89 Zr] that was tested in different immunotherapeutic models, such as colon carcinoma mouse model [103], melanoma, breast, colon, kidney tumor bearing mice [104], gastric adenocarcinoma and Hela cervical cancer [105]. Of note, the minibody [ 89 Zr]Df-IAB22M2C has been safely tested in human patients affected by different solid malignancies [60]. As expected, the minibody has a fast blood clearance due to its small size, whichallows for early image acquisition. A [ 64 Cu]labeled antibody was proposed for the in vivo quantification of CD8+ T cells as a prognostic biomarker in an immunocompetent mouse model of colorectal cancer [106] and breast [107] MuLV) is a selflimiting tumor whose rejection, complete by 4 weeks, includes infiltration, activation and proliferation of T cells. As expected, the [ 18 F]FDG signal reflected the glucose consumption showing the maximum uptake when the tumor reached the peak of growth, while [ 18 F]AraG was found to preferentially accumulate in activated CD8+ cells and with a continuous increase during tumor rejection. Of note, in the colon adenocarcinoma mouse model, responders to anti-PD-1 showed a stronger [ 18 F]AraG signal before treatment compared to non-responders [111]. The inducible T-cell costimulatory receptor (ICOS), [112]. This receptor has been investigated as a PET imaging target for predicting and monitoring T-cell-mediated immune response to cancer immunotherapy. Namely, an [ 89 Zr] labeled minibody against ICOS was successfully tested in immunotherapy models of Lewis lung cancer, both in the primary tumor and in tumor-draining lymph nodes. ICOS is an imaging target that allows for an early response detection, before changes in tumor volume occur [113]. In 2017, [ 18 F] has been directly introduced to the tryptophane backbone, and radioactive 5-18 F-L-α-methyl tryptophan was proposed as indoleamine-2,3-dioxygenase 1 (IDO1) imaging agent. IDO1, preliminary tested on melanoma mice [130], is responsible for the first step of the degradation of the essential amino acid tryptophan into immunosuppressive kynurenine and its inhibition restores the anti-tumour immune response [114]. In 2020, Pandit-Taskar N et al. [60] assessed the safety and utility of [ 89 Zr]IAB22M2C, a radiolabeled minibody against tumor-infiltrating CD8-positive (CD81) T lymphocytes, for targeted imaging of CD81 T cells in cancer patients. This first-in-human prospective study included 6 cancer patients (1 melanoma, 4 lung cancer, and 1 hepatocellular carcinoma). Patients received a single dose of [ 89 Zr]IAB22M2C (about 111 MBq), followed by PET/CT scans at different timepoints (until 144 h post administration). No side effects were registered, and serum clearance was biexponential. The highest uptake was observed in the spleen, followed by the bone marrow. The maximum uptake in normal lymph nodes was reached between 24-48h. Uptake in tumor lesions was seen already 2 h after injection, but the uptake in [ 89 Zr]IAB22M2C-positive lesions increased until 24 h (SUV from 5.85 to 22.8).

Other immuno-PET radiotracers Tumor associated macrophages (TAM)
Immunosuppressive macrophages in the tumor microenvironment are associated with poor prognosis. A [ 64 Cu]-labelled polyglucose nanoparticle (Macrin) was proposed as a radiotracer for quantitative PET imaging of macrophages in colon adenocarcinoma and in orthotopic lung murine tumor models. The diameter of Macrin nanoparticle is 20 nm, and specific macrophage uptake has been demonstrated by dorsal window chambers. The clinical relevance of this nanosystem is in predicting response, especially to TAMtargeted therapies [115]. Integrin CD11 is expressed on the cell surface of TAM that is associated with poor prognosis in high-grade glioma [116]. An [ 89 Zr]-labeled anti-CD11b antibody for non-invasive imaging of TAMs showed significant uptake in glioma compared to contralateral normal brain parenchyma [117]. [ 89 Zr]HDL has been proposed as tumor macrophagic nanotracer for PET imaging to evaluate the immunotherapeutic response in a mammary adenocarcinoma model [118]. Goggi

CD3
CD3 is a direct T-cell marker, and the antibody [ 89 Zr]DFO-CD3 was shown to predict xenograft tumor model anti-CTLA-4 therapeuty response. Three days after [ 89 Zr]DFO-CD3 administration, responders showed higher tumor-to-liver PET uptake compared to non-responders [120]. Also, CD3 has been exploited as a T cell marker in BiTE antibody. BiTE stands for "bispecific T-cell engager". This category represents special antibodies, which are specific for both a surface target antigen on cancer cells and CD3 on T cells. Their therapeutic efficacy was firstly published in 2009 [121]. Here, an [ 89 Zr]-labelled antibody against CEA and CD3 positive cells showed accumulation in colon and breast tumor in PET , and therapeutic property in CEA positive tumors [122]. The full-sized antibody employed in these applications showed long blood half-life and high?? liver uptake. However, smaller molecules such as minibodies may have more optimal pharmacokinetic properties. A different possible imaging target expressed on activated T cells is the IL-2 receptor, also called CD25 [123]. The PET radiotracer [ 18 F]FB-IL2 has been described for the first time in 2012 by Di Gialleonardo et al. [124]. It showed an enhanced lung tumor uptake after irradiation and/or combination with immunization based on the recombinant Semliki Forest (rSFV) viral-vector [125]. An important challenge on the road to [ 18 F]FB-IL2 clinical translation is its complex and time-consuming production [126]. To overcome this obstacle, the two variants [ 18 F]AlF-RESCA-IL2 and [ 68 Ga]NODAGA-IL2, which are characterized by an easier synthesis, were compared to [ 18 F]FB-IL2. They showed good in-vitro and in-vivo characteristics, with high uptake in lymphoid tissue and human peripheral blood mononuclear cell (hPBMC) xenografts [63].

[18F]FDG PET/CT
Despite these promising results, the above-mentioned novel radiopharmaceuticals are still all under investigation. In clinical routine, [ 18 F]FDG PET/CT has reached a pivotal role in the evaluation of response to immunotherapy, although it is far from a specific immuno-PET radiotracer. [ 18 F]FDG allows to visualize rapidly proliferating glucose-avid cancer cells (Warburg effect) and monitor immunotherapy response in different cancer types, such as melanoma, non-small-cell lung cancer, head & neck cancer, urothelial cancer and renal cell cancer [130][131][132][133]. However, during ICIs therapy, the presence of TILs and the inflammatory response may also lead to a transient increase in tumor volume and increased [ 18 F]FDG uptake in responding tumoral tissue. This phenomenon, known as pseudo-progression, may potentially lead to an underestimation of the benefit of ICIs therapy: Both morphological Response Evaluation Criteria in Solid Tumors (RECIST) 1.1 and functional PET Response Criteria in Solid Tumors version (PERCIST) 1.0 cannot address pseudo-progression adequately [134][135][136]. Therefore, new morphological and functional criteria have been developed in the last decade to overcome this problem, such as the immune-related response criteria (irRC), the immunotherapy-modified Response Evaluation Criteria in Solid Tumors (imRECIST), the immunotherapy-modified PET Response Criteria in Solid Tumors (imPERCIST), PET Response Evaluation Criteria for Immunotherapy (PERCIMT) and Lymphoma Response to Immunomodulatory Therapy Criteria (LYRIC) [137][138][139][140][141]. The major radiological and functional response criteria to immunotherapy are summarized in Table 3. Table 3. Summarized the radiological and functional response criteria to the immunotherapy in tumor.

Radiological response criteria irRC imRECIST
Complete response (CR) Disappearance of all target lesions. Determined by two observations not less than 4 weeks apart.
Disappearance of all target and non-target lesions, without any new lesions. Any pathological lymph nodes must have reduction in short axis to <10 mm. Determined by two observations not less than 4 weeks apart.

Partial response (PR)
Sum of product of all lesions decreased by >50% for at least 4 weeks; no new lesions; no progression of any lesions.
At least a 30% decrease of the sum of maximum diameters of target lesions; no new lesions; no progression of disease.

Stable disease (SD)
Sum of product of all lesions decreased by <50% or increased by <25% in the size of one or more lesions.
Does not meet the criteria for CR, PR or PD, taking as reference the smallest sum of maximum diameters of target lesions.

Progressive disease (PD)
A single lesion increased by >25% (over the smallest measurement achieved for Sum of the maximum diameter of lesions increased by >20% over the smallest achieved sum of maximum diameter. the single lesion) or the appearance of new lesions, that has to be confirmed in 2 consecutive observations at least 4 weeks apart.
The appearance of new lesions and/or progression of non-target lesions are considered iUPD and must be confirmed 4-8 weeks later as iCPD. Progression is not confirmed in case of shrinkage of these lesions at 4-8 weeks and evaluation must be reset. Complete resolution of [ 18 F]FDG uptake within all lesions, to a level of less than or equal to that of the mean liver activity and indistinguishable from the background (blood pool uptake).

Functional
PET Deauville score* = 1, 2, or 3, with or without a residual mass on CT, target nodes/nodal masses must regress to < 1.5 cm in longest diameter. Does not meet the criteria for CR, PR or PD. Does not meet the criteria for CR, PR or PD.

Progressive metabolic disease (PMD)
Increased of at least 30% in the sum of SULpeak of all target lesions detected at baseline or new FDG-avid lesions are considered UPMD and must be confirmed 4-8 weeks later as CPMD.
Progression is not confirmed in case of PMR or SMD at 4-8 weeks and evaluation must be reset. *Deauville score: 1 = no uptake above background; 2 = uptake < mediastinum; 3 = uptake > mediastinum but < liver; 4 = uptake greater than liver; 5 = uptake markedly higher than liver (2-3 times in normal liver) and/or new lesions; X = new areas of uptake unlikely to be related to lymphoma.
In the morphological criteria, both irRC and imRECIST have introduced the category of immune unconfirmed progressive disease (iUPD) for the appearance of new lesions at the first restaging, that must be confirmed at least after 4-8 weeks to be reported as immune confirmed progressive disease (iCPD) [142]. In the functional criteria, the indeterminate response, related to the immune-mediated flare effect, has been introduced also in imPERCIST, PERCIMT and LYRIC criteria. Moreover, the complete disappearance of [ 18 F]FDG uptake in all lesions means a complete metabolic response, regardless of a change in tumor size (in contrast to morphological criteria) [143]. Finally, with the intent of differentiating pseudo-progression from true progression, PERCIMT introduced also new criteria of progressive disease (PD), such as the appearance of four or more new lesions (< 1.0 cm in functional diameter), or three or more new lesions (> 1.0 cm in functional diameter) or two or more new lesions (> 1.5 cm in functional diameter) [139]. Compared with morphologic imaging, [ 18 F]FDG PET/CT has also the advantage of an early detection of autoimmune reactions and irAEs (such as hypophysitis, pneumonia, colitis, hepatitis, and thyroiditis) due to increased uptake of [ 18 F]FDG at these sites. This potentially allows for a rapid intervention in life-threatening cases, and aides in switching therapy in less severe cases [143].
In a recent meta-analysis on the incidence of pseudoprogression during ICIs for solid tumor, Park et al. [144] reported that the incidence according to tumor types were similar for the three more representative ones (6.4% in melanoma, 5.0% in NSCLC, and 7.0% in genitourinary cancer). The incidence of pseudoprogression according to the agent types were also analyzed: the pooled incidence in the studies of PD-1/PD-L1 inhibitor monotherapy was 5.7% (5.6% for PD-1 and 6.8% for PD-L1), whereas one study of melanoma that used a CTLA-4 inhibitor showed higher incidence of pseudoprogression (9.7%).
Several semiquantitative parameters, such as metabolic tumor volume (MTV) and total lesion glycolysis (TLG) rate, have been studied to evaluate the predictive and prognostic value of the metabolic tumor burden at baseline [ 18 F]FDG PET/CT. Other parameters, such as spleen to liver ratio (SLR) or bone marrow to liver (BLR) SUVmax ratio, were shown to represent surrogates (indirect index) of the hematopoietic tissue metabolism. Seban et al. [145] in a retrospective study on 55 melanoma patients undergoing 18F-FDG PET/CT before anti-PD1, reported that low tumor burden (TMTV=total metabolic tumor volume; <25cm 3 ) correlates with survival and objective response, while hematopoietic tissue metabolism (BLR >0.79; SLR >0.77) inversely correlates with survival.
These results are considered preliminary but may assume a more prominent l role in the evaluation of patients under immunotherapy in the future. Today, the use of [ 18 F]FDG PET/CT is not still standardized in clinical practice and cannot not be considered as a specific marker of immune response.

Artificial Intelligence (AI) in Immunotherapy
An improvement in the acquisition, reconstruction and interpretation of medical image datasets is attempted with the use of AI. Through several approaches (Machine learning, radiomics, and deep learning), the dealing with so-called "big-data" is enabled, with the goals to condense information, to enable better patient selection, to improve workflow and to obtain precise predictive models [146].

Radiomics assessing the response to immunotherapy and survival prediction
Radiomics is a machine-learning technique that can evaluate intratumor heterogeneity, which has been suggested to be an important resistance mechanism leading to treatment failure. Valentinuzzi et al. aimed to investigate whether immunotherapy [ 18 F]FDG radiomics signature (iRADIOMICS) was able to predict response of metastatic NSCLC (stage IV) to pembrolizumab compared to the clinical standards (iRECIST and PD-L1 immunohistochemistry). 30 patients underwent 18F-FDG PET/CT at baseline, and after one and four months. Multivariate iRADIOMICS was found superior to the current standards in terms of predictive power and survival, with an area under the curve (AUC) of 0.90 vs. an AUCmax of 0.6 for PD-L1 and 0.86 for iRECIST [147]. Polverari et al. evaluating evaluated [ 18 F]FDG PET/CT radiomics features extracted from the primary lesion. They were able to predict response to immunotherapy in 57 advanced NCSLC patients They observed that patients with high tumor volume, tumor lesion glycolysis (TLG), and heterogeneity features ("skewness" and "kurtosis") had a higher probability of immunotherapy failure [148]. Wei et al. assessed 194 patients with advanced NSCLC through a radiomics signature approach (mpRS) to predict response to immunotherapy using an improved least absolute shrinkage and selection operator (LASSO) method, reaching an AUC of 0.81, with similar results for survival prediction [149]. Park et al. tested radiomics features on pre-gefitinib or pre-erlotinib [ 18 F]FDG PET/CT images, in order to assess their predictive value in recurrence or advanced NSCLC patients with EGFR mutation. Radiomics features were found to have an incremental predictive value of early EGFR TKI failure, being also highly predictive for PFS assessment [150]. Parvez  They observed that an increase in spleen-to-liver ratio (SLR) greater than 25% at 3 months [ 18 F]FDG PET/CT was associated with poor outcome after immunotherapy [153]. Finally, in a retrospective study on 112 metastatic melanoma patients treated with immune checkpoint inhibition, Basler et al. assessed that a model combined of blood biomarkers (LDH+S100) and non-invasive 18F-FDG PET/CT-based radiomics represent a promising biomarker for early differentiation of pseudoprogression, potentially avoiding added toxicity or delayed treatment switch [

Deep-learning approach
Deep-learning (DL) is an AI application organized in multiple, progressive neural networks and subsequent related processes (layers). DL approaches simultaneously learn relevant features and prediction models from input images with no need of previous computation and extraction of "custom-tailored" imaging variables. Park et al. developed a DL model to predict cytolytic activity score (CytAct), using semi-automatically segmented tumors on [ 18 F]FDG PET/CT paired with tissue RNA sequencing. In the immune checkpoint blockade (ICB) cohort, a higher predicted CytAct of an individual lesion was associated with more tumor shrinkage after ICB treatment (p < 0.001). Furthermore, higher minimum predicted CytAct in each patient was associated with significantly prolonged PFS and OS (p = 0.001 and p= 0.004, respectively), while in patients with multiple lesions, ICB responders had significantly lower variance of predicted CytActs (p = 0.005) [160].

Conclusion
In recent decades, the impressive improvement in translational medicine has led to a better understanding of the TME and all the dynamic mechanisms behind tumor development and growth, such as the relationship between tumor cells and infiltrating immune cells in the TME. These discoveries have led to the development of immunotherapy, and subsequently of imaging biomarkers for the TME, which may provide new insights into TME and cancer behavior. While most of these new radiotracers are still in a preclinical phase, they will certainly provide complementary information to clinically used [ 18 F]FDG PET/CT for the prediction of immunotherapy response in the context of personalized medicine, also broadening the theragnostic horizon. Today, [ 18 F]FDG PET/CT represent a standard technique employed for tumor staging and response assessment, despite the known limitations due to pseudo progression and enhanced activity of inflammatory components. Therefore, [ 18 F]FDG PET/CT is still not validated as a specific biomarker in immunotherapy and not well established for response assessment, yet. Nevertheless, new developments in PET/CT technology and images analysis are becoming fundamental tools to guide therapeutic decision-making and prognosis stratification in clinical practice. AI-based approaches may help identify subpopulations with an increased risk of immunotherapy failure, assessing the immuneprofile non-invasively, and predicting survival. Hence, they may well evolve as a fundamental support for clinical decision-making. However, due to problems related to data standardization, privacy concerns, lack of extensive data, the development of AI approaches is still challenging. In our opinion and according to the presented data, new radiotracers combined with technical improvements and AI applications will further refine the selection of the most appropriate therapy for each patient, thus potentially avoiding high-cost strategies (not exempt from toxicity), towards individualized precision healthcare. Retrieved from https://app.biorender.com/biorender-templates/t-5fc930de9e43ce3283cc6197-pro-and-anti-tumour-immune-cells-in-the-tumor-microenvironme; (2) "Tumor Microenvironment 2", by BioRender.com (2021). Retrieved from https://app.biorender.com/biorender-templates/t-5f63a4bebecfd300b1f68c0c-tumor-microenvironment-2; (3) "Overview of Metastatic Cascade", by BioRender.com (2021). Retrieved from https://app.biorender.com/biorender-templates/t-5f46e7082c321c00b3f65e38-overview-of-metastatic-cascade; (4) "Immunotherapy Overview", by BioRender.com (2021). Retrieved from https://app.biorender.com/biorender-templates/t-5f2da5b6a2391a00a70941ba-immunotherapy-overview; (5) "T-cell Deactivation vs. Activation", by BioRender.com (2021). Retrieved from https://app.biorender.com/biorender-templates/t-5c082704e97e1400127428b8-t-cell-deactivation-vs-activation; (6) "CAR T Cell Therapy Overview", by BioRender.com (2021). Retrieved from https://app.biorender.com/biorender-templates/t-5e41ad5c0dd2690088b718bd-car-t-cell-therapy-overview.

Conflicts of Interest:
MH is a recipient of grants from GE Healthcare, grants for translational and clinical cardiac and oncological research from the Alfred and Annemarie von Sick Grant legacy, and grants from the Artificial Intelligence in oncological Imaging Network by the University of Zurich. IAB is a recipient of grants from GE Healthcare, grants from the Sick legacy and the "Jimmy Wirth Foundation". The other authors declare no conflict of interest.