Biological Rationale and Clinical Evidence of Carbon Ion Radiation Therapy for Adenoid Cystic Carcinoma: A Narrative Review

Adenoid cystic carcinoma (ACC) is a rare, basaloid, epithelial tumor, arising mostly from salivary glands. Radiation therapy can be employed as a single modality for unresectable tumors, in an adjuvant setting after uncomplete resection, in case of high-risk pathological features, or for recurrent tumors. Due to ACC intrinsic radioresistance, high linear energy transfer (LET) radiotherapy techniques have been evaluated for ACC irradiation: while fast neutron therapy has now been abandoned due to toxicity concerns, charged particle beams such as protons and carbon ions are at present the beams used for hadron therapy. Carbon ion radiation therapy (CIRT) is currently increasingly used for ACC irradiation. The aim of this review is to describe the immunological, molecular and clinicopathological bases that support ACC treatment with CIRT, as well as to expose the current clinical evidence that reveal the advantages of using CIRT for treating ACC.


INTRODUCTION
Adenoid cystic carcinoma (ACC) is a rare tumor, which has a dual component of myoepithelial and ductal cells. Most of the ACCs arise from minor salivary glands, and account for about 10% of all malignant salivary tumors (1). Other involved head and neck (HN) sites include the external ear and the lacrimal glands. Less frequently, ACC might be diagnosed in non-HN sites, including the breasts, the lungs, the prostate gland, the esophagus, the cervix, the vulva or the skin (2)(3)(4). Overall, around 3000 cases have been identified between 1973 and 2007 in the US National Cancer Institute's Surveillance, Epidemiology and End Results (SEER) program (5) and around 2600 cases have been described between 1983 and 1994 in the European EUROCARE registry (6).
Photon beam radiotherapy (RT) technological developments gave a substantial contribution in HN patient prognosis improvement last decades especially in patients bearing squamous cell carcinoma (7). RT plays a key role in different phases of ACC management: in the adjuvant setting after surgery for potentially resectable cases or as a definitive modality for non-operable tumors; or in a reirradiation context for local recurrent disease (8). Due to the radioresistance of ACC, RT techniques using high linear energy transfer (LET) particles have been evaluated for more than 50 years, in particular fast neutron beams (9). However, the use of neutron beams, while efficient, has been abandoned for late toxicity due to the difficulty of obtaining an advantageous dose gradient between the target and the organs-at-risk. On the other hand, carbon ion radiation therapy (CIRT) harbors both the high LET of fast neutron beams and the specific spatial distribution of charged particle beams. Charged particle beams are characterized by a Bragg peak that deposits most of the dose in a very short path, the deepness of which depends on the start energy of the beam, allowing targeted irradiation. CIRT has consequently been increasingly evaluated for ACC irradiation during the last years.
To this date, no in vitro or preclinical study have specifically evaluated CIRT irradiation effects on ACC cell lines. Nevertheless, multiple immunological and biological properties justify ACC management with CIRT ( Table 1). We reviewed the literature on the biological bases and the clinical evidence that support the use of CIRT in ACC management.

ACC Is a Heterogeneous Group of Tumors From a Molecular Point of View
Based on whole-exome sequencing (WES) of 34 tumor samples from primary and metastatic ACC tumors isolated from eight patients, Liu et al. (10) demonstrated that there was an important spatial and temporal clonal diversity within and between primary and metastatic tumors. The average mutation rate was evaluated around 0.32 per million base pair and the incidence of shared mutations between primary and metastatic tumor samples was 21.9%; truncal genetic alterations included NOTCH pathway genes (such as NOTCH1 or SPEN) or the t (6,9) translocation (MYB-NFIB fusion). Nevertheless, this apparent diversity of tumor mutations allows a variety of potential systemic targeted treatments. For instance, Ho et al. (11) demonstrated that tumoral activation of the VEGF/KIT/PDGFR pathway could be effectively targeted by anti-angiogenic agents (such as Axitinib), while the NOTCH-mutated ACCs could be specifically blocked  (12,13). NOTCH-mutated ACC are over-represented in solid variant and are associated with a higher rate of liver and bone metastases, as well as shorter relapse-free survival (RFS) and overall survival (OS) (12). The molecular diversity of ACC tumors is striking in a relapsed or a metastatic context (14). Up to 26.3% of relapsed or metastatic ACC patients are NOTCHmutated, including 18.3% of activating NOTCH mutations, which have the poorest prognosis among all NOTCH mutations types. Ho et al. (14) further demonstrated that mutations in the KDM6A gene, which intervenes in chromatin remodeling, had a pejorative prognosis for relapsed or metastatic ACC, and that TERT pathway mutations were exclusive of NOTCH mutations and of MYB fusions. Consequently, it appeared that four distinct relapsed/metastatic ACC molecular subgroups could be proposed, based on the presence of NOTCH, TERT mutations and MYB fusion: MYB-mutated/NOTCHmutated; MYB-mutated/other mutations; MYB wild-type/ NOTCH-mutated and MYB-mutated/TERT-mutated.

ACC Is a Heterogeneous Group of Tumors From an Anatomopathological Point of View
ACC is characterized by a biphasic composition made of ductal cells (characterized by CK7 protein expression) and myoepithelial cells (characterized by CK5/6, P63, P40, D2-40, Calponin, a-SMA, S-100, and vimentin protein expression). The histological tumoral organization defines three ACC variants based on the predominant anatomopathological pattern: cribriform, tubular (both characterized by CK7 protein expression), or solid pattern (characterized by a glandular architecture and a loss of myoepithelial differentiation). Solid patterns usually have a higher Ki67 index. Multiple histological grading systems have been proposed based on the estimated proportion of the solid pattern component [Perzin grading using a 30% threshold (15), Spiro grading using a 50% threshold (16)]. It should be noted, however, that Van Weert et al. (17) recently demonstrated that the mere presence of solid components, independently of its proportion, was a negative prognostic factor. Bell et al. (18) evidenced that the c-Kit protein was systematically expressed, and EGFR was consistently negative, in the solid ACC subtype. C-Kit expression was limited to inner ductal epithelial cells, and EGFR expression was restrained to the outer myoepithelial cells, the latter being found in the majority of tubular and cribriform ACC patterns.

CIRT Anti-Tumor Efficiency Is Not Influenced by Tumor Heterogeneity
Based on three cellular sublines of resistant prostate tumors in murine models, Glowa et al. (19) demonstrated that the values of the tumor control dose 50 (TCD50) differed significantly less for CIRT than for photon RT. They concluded that response to CIRT was relatively independent on the molecular and histological tumoral heterogeneity. Additionally, Masunaga et al. (20) found that quiescent tumor cells were more sensitive to CIRT than to photon RT. They suggested that CIRT anti-tumor efficacy could be relatively independent from intra-tumoral cellular heterogeneity, resulting from the co-existence of quiescent and proliferative cell populations in various proportions.

ACC Radioresistance Properties Might Be Overcome by CIRT
ACC are radioresistant tumors. The main molecular actors of ACC EMT and radioresistance features are provided in Table 2.

Epithelial-Mesenchymal Transition and Cancer Stem Cell Properties Are Related to ACC Radioresistance
Wang et al. (21) demonstrated that hypoxic conditions promoted ACC epithelial-mesenchymal transition (EMT) and cancer stem cell (CSC) properties. Molecular actors of ACC EMT and stemness were characterized by Chen et al. (22) who found that HSP27 protein overexpression increased cell migration and invasion properties and induced an up-regulation of Snail 1 and Prrx1, which are potent EMT regulators. Expectedly, increased HSP27 levels in ACC correlated with radioresistance of ACC cell lines in vitro; it was also found that high HSP27 expression in ACC had a poor prognosis value. Finally, acquisition of CSC properties in ACC correlated with increased expression of CD133 and CD44. Shimoda et al. (23) further evidenced stemness was a generic characteristic of metastatic ACC cells, which express stemness-related transcription factors (TF) (such

CIRT Is Valuable for Radioresistant and Hypoxic Tumors
Peschke et al. (30) found on murine models of radioresistant prostate carcinomas that the TCD50 were 32.9 Gy for CIRT and 75.7 Gy for photon RT for single dose irradiation, and 38.0 Gy for CIRT and 90.6 Gy for photon radiotherapy for multiple-dose irradiation. This observation suggested that CIRT was more potent than photon radiotherapy for intrinsically radioresistant tumors, which is the case for ACC. In addition, Grimes et al. (31) evidenced that carbon ion beams had a lower oxygen enhancement ratio (OER) than proton beams, in particular towards the Bragg peak where the LET substantially increased for carbon beams, making CIRT valuable in case of hypoxic tumors. Nevertheless, Antonovic et al. (32) underlined that hypoxia could anyhow influence the outcome of CIRT because of the non-negligible OER of the low LET contributions in the spread-out Bragg peak (SOBP). Taking into account inter-fraction local oxygenation changes, occurring after tumor shrinkage (even for hypoxic tumors), CIRT OER was estimated around 1.2. Finally, using a glioma model, Liu et al. (33) demonstrated that CIRT superiority in tumorigenesis and angiogenesis inhibition compared with photon beams, resulted from modulation of VEGF level in the tumor micro-environment (TME).

Perineural Invasion Is a Characteristic Feature of ACC Local Malignancy
ACCs are characterized by an elevated propensity to locally invade surrounding tissues through perineural invasion (PNI).
Shan et al. (34) demonstrated that the BDNF/TrkB axis plays a causative role in ACC PNI. Gao et al. (35) found that the CCL5/ CCR5 axis increases salivary ACC PNI invasion and that blocking this chemokine axis inhibited perineural invasion in ACC cell lines. CCR5 chemokine receptor expression was elevated in salivary ACC tissue samples. Kobayashi et al. (36) observed that NGF and TrkA signaling contributed to PNI, and that both were expressed in around 65% of ACC patients.

ACC Cell Immunogenicity Is Limited to Few Tumor-Associated Antigens
ACC cells express few tumor-specific neo-antigens (TNA) against which specific anti-tumor immune responses could be directed, which is explained by their low tumoral mutation burden (TMB). Based on 60 ACC tumor samples, Ho et al. (40) estimated that ACC TMB was around 0.31 non-silent mutation per megabase. Nevertheless, TMB-high ACCs have already been described, either microsatellite-instable (MSI) or POLE-mutated, but these cases represent only a minority of ACC patients (41). Consequently, T-cell receptor (TCR) clonotype diversity in ACCs is usually lower than in most other solid tumor types and CD8 tumor-infiltrating lymphocytes (TIL) are rare (41). On the other hand, ACC cells express diverse types of tumor-associated antigens (TAA), in particular cancer-testis antigen (CTA). Based on 84 head and neck (HN) ACC tumor samples, Veit et al. (42) found that NY-ESO-1 and pan-MAGE CTA were significantly expressed in 57.1% and 31.2% of ACC patients, respectively. In addition, tumor expression of these two CTA has been found to be linked to a worse prognosis, since median overall survival (OS) was 282 months in the absence of NY-ESO-1 and pan-MAGE expression, 190.5 months when one of these two antigens was present, and only 90.5 months in case of simultaneous co-expression. Beppu et al. (43) estimated that MAGE-A CTA was detected in 60% of ACC tumors and represented an independent risk factor for locoregional recurrence. Finally, in addition to CTA, ACC also expresses less immunogenic tissue-differentiation TAA. Prostate-specific membrane antigen (PSMA) expression without predictive value was found in 94% of ACC patients (44).

ACC Cells Evade Immune Surveillance by Regulating the Expression of Membrane Receptors
Based on immunohistochemistry (IHC) analyses on 36 ACC tumor samples, Mosconi et al. (45) found that ACC expressed a high level of inhibitory immune membrane proteins, particularly PD-L2 and HLA-G. Notably, PD-L1 expression was systematically negative, and CTLA-4 expression was low, which are the targets of most current immune checkpoint inhibitors. PD-L2 expression has a prognosis value: Chang et al. (46) demonstrated that low PD-L2 expression was associated with a shorter RFS in a cohort of 70 patients with malignant salivary gland tumors (including 15 ACC). In addition, ACC cells have a reduced expression of ICAM-1 adhesion protein (47), and IHC analysis of tumor samples demonstrated reduced staining for surface antigens of T cells, NK cells, macrophage (TIA1 and CD68) (47), and antigenpresenting cells (APC) (CD1a and CD83) (45). It has consequently been proposed that the reduced ACC membrane expression of ICAM-1 might promote immune evasion by limiting the ICAM-1/LFA-1-mediated interaction between ACC cells and anti-tumoral immune cells. Conversely, high ICAM-1 expression was associated with a significantly better DFS for ACC (47).  (49) demonstrated that alterations in the PI3K and the WNT pathways (in particular, involving FGF17, BCL2, beta-catenin, and BAMBI genes) strongly correlated with a lack of immunecell infiltrates in ACC TME. In addition, it was found that cytokine landscape of ACC TME contributed to its protumoral properties: CCL2 chemokine produced by ACC cells recruit M2-polarized TAM, which, in response, increase tumor cell invasive and migrative properties by secreting glial cell linederived neurotrophic factor (GDNF).

CIRT Increases Tumoral Cell Immunogenicity
Imadome et al. (50) found that CIRT upregulated stressresponsive genes and immunity-related cell-communication genes (notably ICAM1) in tumor cells. At the same time, CIRT increased gene expression of cytokines and chemokines. Six to 36 hours after CIRT irradiation, Ohkubo et al. (51) demonstrated that an increased expression of ICAM-1 membrane receptor could be observed, which interacts with APC through their LFA1 receptors, but which is usually downregulated in ACC tumors (47). Based on murine lung tumor models, it was demonstrated that combination of CIRT with DC inhibited the development of lung metastases (51), while Ando et al. (52) evidenced that CIRT increased tumor cells immunogenicity and DC activation to a higher level than photon beam radiotherapy, as evidenced by greater CD40 and IL-12 level.

CIRT Activates Adaptative Immunity
Hartman et al. (53) found in vitro that CIRT and photon beam radiotherapy had common radiobiological properties such as induction of cell cycle arrest, surface expression of immunemodulating molecules, and activation of cytotoxic lymphocytes. Nevertheless, other authors have suggested that CIRT-induced immune activation might be more potent. Spina et al. (54) demonstrated that CIRT could induce a more pro-inflammatory cytokine landscape compared with photon beam radiotherapy on mammary tumor cell lines: high levels of IL-2, IL-1b, and IFNg were observed after CIRT (compared with an isolated IL-6 increase after photon radiotherapy). Simultaneously, CIRT induced an activated CD8 TIL phenotype in TME, as evidenced by high levels of granzyme B, IL-2, and TNFa expression, while the authors found that photon therapy decreased CD8 TILs. Takahashi et al. (55) demonstrated that, when combined with anti-PD-L1 and anti-CTLA-4 immunotherapies, CIRT substantially increased CD8 TIL infiltrates and HMGB-1 level, a potent danger-associated molecular pattern (DAMP), suggesting a rational therapeutic strategy combining immune checkpoint inhibitors with CIRT.

TME Acquires Anti-Tumoral Polarization After CIRT Irradiation
In a murine glioblastoma model, Chiblak

OAR Toxicity for Digestive and Pelvic ACC Treated With Photon RT
The largest cohort of Bartholin's gland ACC has been reported by Cardosi et al. (2) consisting of 12 patients treated with surgery; seven patients underwent adjuvant RT. One vulva radio-necrosis was observed, followed by a grade 5 sepsis, and one patient developed concomitant digestive and urinary fistulas. It should be recalled that fistulas have a significant impact on the quality of life of the patient. Zelga et al. (64)demonstrated that the simplest and safest treatment of radiation-induced rectovaginal fistulas was a fecal diversion with an ileostomy. While cardia (65) and esophageal (3) ACC exist, no report of RT-induced toxicity with photon RT has been published to this date; nevertheless, CIRT is expected to reduce radiation exposure to the unaffected digestive tract and to the heart.

Materials and Methods
A search was conducted on the PubMed, Medline, Google Scholar, Cochrane library and Web of Science databases using the following keywords: ["particle therapy" or "hadrontherapy" or "carbon ion" or "CIRT" or "heavy ion" or "ion beam" or "ion radiation"] and ["adenoid cystic carcinoma" OR "ACC"]. Search was independently conducted by PL and EO. Inclusion criteria, defined using the PICOS framework, were the following: clinical studies (trials, prospective or retrospectives studies, and case reports) evaluating CIRT for ACC, in any setting, and reporting toxicity and efficacy data. Exclusion criteria were pre-clinical or purely dosimetric studies. References from the selected studies were screened for potential additional articles.

Head and Neck (Primary Tumor)
The current clinical experience of CIRT for ACC is summarized in the Table 3. CIRT has been evaluated as a sole modality for definitive irradiation of head and neck ACC by Japanese centers. Mizoe   . With a median follow-up of 50 months, 3-year local control was 79% when treated in a definitive context and 82% when treated in an adjuvant setting. Acute toxicity was observed for 34.4% of the patients in a definitive setting and for 41.6% in an adjuvant setting. CIRT as a boost to IMRT has also been evaluated for laryngeal ACC (80), demonstrating an excellent local control on a cohort of eight ACC patients without any relapse at 24 months. Lacrimal ACC has also been treated with a CIRT boost (81), with a 2-year local control of 93%.   constraints, adapted to the RBE calculation model was therefore recommended in order to enhance the target control probability, such as for brain stem and optic pathway (90,91). For a full CIRT irradiation, ACC is generally irradiated with a sequential strategy consisting of a first phase of nine to ten fractions to a low-risk volume (including the surgical bed and zones at risk of perineural spread), followed by a second phase of six to seven fractions to a high risk volume (boost), with a unique nominal dose per fraction, according to the protocol adopted in Japan since 1997 (67). In this context, CIRT is usually delivered with a limited number of beams, typically two or three, achieving both a high conformation and an improved normal tissue sparing. However, in the absence of isocentric CIRT gantry, only fixed beam irradiations are available (which is for instance the case at CNAO) and it is difficult to change the beam arrangement between the two sequential phases: as a consequence, part of the low-risk volume receives unintended dose from the beam paths of the boost phase. A simultaneous integrated boost (SIB) approach is thus being evaluated at CNAO, in comparison to the sequential protocol, to improve the dose distribution of the two target volumes.

Treatment Planning System Considerations
Robust planning is of prime importance for CIRT, to take into account range and positioning uncertainties (92). Finally, the use  (97) demonstrated that in-beam positron emission tomography (PET) imageries allowed detection of morphological changes for head and neck proton therapy in clinical practice; while the primary fluence is lower in CIRT which introduces additional uncertainties compared to proton treatments, such devices have been successfully evaluated in phantoms for CIRT (98,99) with a 1mm to 2mm agreement on the range prediction.

DISCUSSION
We have detailed the rationale to use CIRT for ACC management, based on immunological, molecular, and pathological considerations, even though no in vitro or preclinical study have specifically evaluated CIRT irradiation on ACC cell lines to date. In addition, clinical data demonstrate that CIRT is associated with superior local control compared to conventional photon radiotherapy. In the future, further improvement of the outcomes of ACC treatment with CIRT may be possible by personalizing the treatment by taking into account ACC molecular and pathological features. Vered et al. (100) found that around 85% of ACC expressed EGFR receptors, which motivated the use of anti-EGFR systemic therapies in addition to radiation therapy, as radiosensitizers and to increase micrometastatic disease control. Adeberg et al. (101) recently evaluated adjunction of Cetuximab to photon radiation therapy (54 Gy) with a CIRT boost [24 Gy(RBE)], in 33 head and neck ACC patients. The toxicity was noticeable since 17% of the patients developed grade 3 rashes, 22% grade 3 radiation dermatitis and 48% grade 3 mucositis, but tumoral control was encouraging with a three-year DFS and OS of 67% and 90%, respectively. The ongoing NCT02942693 trial is a 2-arm study evaluating six weeks of Apatinib, an anti-VEGFR2 drug, followed by mixed irradiation with 56 Gy in photons and 15 Gy(RBE) in carbons. Immunotherapy combined with CIRT, although potentially attractive, considering the immunomodulatory effects of CIRT, is not currently evaluated in clinical trials. However, CIRT might face competition with low-LET RT techniques for ACC management in the future. Takagi et al. (102) evaluated the outcome of 40 ACC patients treated with proton beam therapy (PBT) and 40 patients treated with CIRT; no difference between PBT and CIRT were observed in terms of OS, PFS, or local control at 5 years. However, in addition to the inherent weakness of a retrospective non-randomized comparison, there was a significant difference in the equivalent dose prescription between treatment groups in favor of PBT. Patient selection for CIRT (instead of PBT) must consequently be defined; NTCP or dosimetric considerations could be used. In some specific cases, fear of adverse events in case of rapid tumor shrinkage second to CIRT irradiation might justify considering normofractionated PBT irradiation. Reirradiation with photon stereotactic radiotherapy has been evaluated for relapsed head and neck ACC patients (103), with a total dose of 30 Gy in 5 fractions: the 3-year LC of 49% was lower than that observed with CIRT (possibly due to the relatively low prescribed dose), but this technique might be easily implemented in radiotherapy centers where hadrontherapy is not available. Other ongoing trials comparing CIRT with low-LET RT techniques include the ETOILE trial (NCT02838602) comparing CIRT and PBT or photon radiotherapy for radioresistant tumors, including ACC, and the COSMIC trial (NCT04214366) comparing CIRT with a mix of CIRT and photon beam radiotherapy for ACC. In addition, financial sustainability is an issue that may weigh against CIRT compared with other low-LET techniques. Jensen et al. (104) estimated that IMRT with a carbon boost might have a mean survival benefit of 0.86 years for a single head and neck ACC, with an incremental cost-effectiveness ratio of 20.638€ per year.
Finally, other high-LET techniques are currently being investigated for ACC, such as boron neutron capture therapy (BNCT), currently under development in various European and Asian centers. Kato et al. (105) treated one ACC patient to a dose of 14Gy, which was well-tolerated without any grade ≥2 toxicity. Other reports include Kankaanranta et al. (106), who re-irradiated four inoperable recurrent ACC and Aihara et al. (107), who treated four inoperable head and neck ACC (including two recurrent diseases). This latter reported complete response for all patients within 6 months, a median OS of 32 months and no grade ≥3 toxicity.

CONCLUSION
The use of CIRT for ACC management is motivated by immunological, molecular and clinicopathological considerations.
Although no prospective randomized trials have been published to this date and might not be easily feasible, due to the rarity of ACCs and the scarce availability of particle beam RT facilities, clinical studies demonstrated that CIRT was well tolerated and associated with a substantial tumor control in diverse clinical situations, especially in advanced unresectable stages. Current technical developments ensure safe treatments.