From biomarkers to therapeutic targets: the promise of PD-L1 in thyroid autoimmunity and cancer

The programmed cell death-1/programmed cell death ligand-1 (PD-1/PD-L1) immune checkpoint proteins hold promise as diagnostic, prognostic, and therapeutic targets for precision oncology. By restoring antitumor T cell surveillance, the high degree of effectiveness of the immune checkpoint inhibitors (ICIs) has revolutionized cancer treatment. However, the majority of patients (65-80 %) treated with ICIs experience significant side effects, called immune-related adverse events (irAEs), resulting in autoimmune damage to various organs. Therefore, broadening the clinical applicability of these treatments to all cancer types requires an improved understanding of the mechanisms linking cancer immune evasion and autoimmunity. The thyroid is the endocrine gland the most frequently involved in autoimmunity and cancer, the growing incidence of which is raising serious public health issues worldwide. In addition, the risk of developing thyroid cancer is increased in patients with autoimmune thyroid disease and thyroid dysfunction is one of the most common irAEs, especially with PD‑1/PD-L1 blockade. Therefore, we chose the thyroid as a model for the study of the link between autoimmunity, irAEs, and cancer. We provide an update into the current knowledge of the PD‑1/PD-L1 axis and discuss the growing interest of this axis in the diagnosis, prognosis, and management of thyroid diseases within the context of autoimmunity and cancer, while embracing personalized medicine.


Introduction
The thyroid is the endocrine gland the most frequently involved in autoimmunity and cancer [1,2]. The burden due to the rising incidence worldwide of these pathologies represents a major public health issue [1]. At a glance, thyroid cancer (TC) and autoimmune thyroid diseases (AITD) appear to involve diametrically opposite immune responses (Figure 1). Thyroid cancer progresses, at least in part, because it hijacks the mechanisms of tolerance to immune attack [3]. In contrast, AITDs result from the breakdown of the same mechanism of self-tolerance, and the overactive immune response damages the thyroid [4]. Of note, the risk of developing thyroid cancer is increased in patients with AITD [4]. When immune checkpoint inhibitors (ICIs) are used to treat cancer, the immune system is reinvigorated, resulting in the emergence of significant side-effects called immune-related adverse events (irAEs), which occur frequently in the thyroid gland [5]. Therefore, we overview the involvement of the key immune-

Ivyspring
International Publisher checkpoint programmed cell death-1/programmed cell death ligand-1 (PD-1/PD-L1) proteins in cancer and autoimmunity, focusing on the thyroid as a model. We provide an overview of the promise of these proteins in assisting the diagnosis, prognosis, and management of both cancer, and autoimmune diseases of the thyroid, while embracing personalized medicine (Figure 1). At the molecular level, the highly activated lymphocytes, which are characterized by the expression of PD-1 and IFNγ, are in close contact with thyrocytes and thyroid tumor cells. Either IFNγ or/and BRAF V600E induce the expression of PD-L1 by the epithelial cells, which binds to its receptor PD-1 on lymphocytes and thereby restrains autoimmune attacks in AITD (D) and promotes thyroid tumor immune tolerance (F). Note that both AITD and TC release the serum-soluble form of PD-L1, which may contribute with the tissue PD-L1 to the immune escape and be considered as biomarkers and therapeutic targets (sPD-L1, E). Abbreviations: AITD, Autoimmune thyroid disease; Anti-TSH-R-ab, Anti-receptor of the thyroid-stimulating hormone antibody; Anti-TPO-ab, Anti-thyroid peroxidase antibody; Auto-ab, Auto-antibodies; CLT, Chronic lymphocytic thyroiditis; GD, Grave's disease; IFNγ, Interferon-gamma; irAEs, Immune-related adverse events; PD-1, Programmed-cell death receptor 1; PD-L1, Programmed-cell death ligand 1; SNP, Single nucleotide polymorphism.
A number of breakthroughs have occurred in the last fifteen years that help explain how T cellmediated immunity is tightly controlled by inhibitory signals, also termed immune checkpoints, which protect healthy tissue from unnecessary and excessive immune damage [6]. Multiple immune checkpoints have been identified, including PD-1, which is mainly expressed on activated T cells, but also on regulatory T cells (Tregs), B cells, NK cells, and myeloid cells. Its ligands include PD-L1 and PD-L2. PD-L1 (also called B7-H1 or CD274) is induced at the cell surface of cells of healthy tissues and immune cells in response to various cytokines (notably IFNγ and TNFα), whereas PD-L2 is mainly expressed on activated dendritic cells and macrophages [6,7]. When PD-1 is bound to its ligands PD-L1 and PD-L2, PD-1 switches off T cell antigen receptor (TCR)-signaling, keeping in check the downstream CD8+ T cell survival and cytotoxic attack, while promoting differentiation of CD4+ T-cells into Tregs, to coordinately resolve inflammation [7,8].
Emerging evidence suggests that the immunosuppressive activity of PD-L1 is not restricted to acute inflammation. Indeed, the PD-1/PD-L1 pathway restrains chronic inflammation and downstream tissue injury in several autoimmune diseases, such as diabetes mellitus type I or systemic lupus erythematosus, among others [9,10]. Similarly, during tumor development, the immune system efficiently recognizes the tumor cells as 'foreign' by detecting the mutated or overexpressed proteins that can produce tumor-specific neoantigens at their cell surface [3,8]. Here again, PD-L1 is expressed by tumor cells to regulate immune attack [7]. Herein lies the rationale for blocking checkpoint molecules with immunotherapies (the so-called immune checkpoint inhibitors) to restore antitumor T cell surveillance, now approved by the Federal Drug Administration (FDA) for cancer treatment [7]. Nowadays, the major clinical impact of immunotherapies targeting PD-1/ PD-L1 has changed the landscape of cancer treatment. However, a substantial percentage of patients do not benefit from these treatments [11]. Based on the high response rates of patients with PD-L1-positive tumors, a high level of PD-L1 expression within a tumor was the first FDA approved predictive biomarker for the selection of patients who were likely to respond to anti-PD-1/PD-L1 immunotherapies [12,13].

The PD-1/PD-L1 axis in thyroid diseases
Notwithstanding the clinical interest of PD-1/ PD-L1 in autoimmunity and malignancies, their role in thyroid diseases remains poorly understood. The purpose of this literature review is to discuss the potential involvement of the PD-1/PD-L1 couple in thyroid pathogenesis and to clarify their role as a biomarker, and therapeutic target, particularly for the challenging management of aggressive thyroid lesions.

Method of the systematic review
One author (G.D'A) performed a comprehensive literature search of the articles published in English or French using the Medline/PubMed, EMBASE, ResearchGate, and Cochrane Library databases up to May 1, 2020. Manual search through the reference lists of the articles and related reviews was also conducted. The selected studies had to focus on i) thyroid dysfunction related to ICIs, ii) PD-1/PD-L1 expression in AITD, iii) all types of TC, and benign thyroid lesions or iv) pre-clinical and clinical use of ICIs for the treatment of TC. Original articles were included, but also the latest reviews and metaanalysis containing the newest or most relevant information. All the methods for measuring PD-1/ PD-L1 expression on tumor cells and/or tumorinfiltrating lymphocytes (TILs) in primary cancer tissues (on the cell membrane, cytoplasm, or both) were accepted. A total of 162 records were identified, screened, and 106 were excluded (1 Non-English Non-French, 79 irrelevant or not specific enough, 26 duplicate studies). Of the 56 studies included, there were five systematic reviews and meta-analyses. From each included study, the extracted data concerned the characteristics of PD-1/PD-L1 evaluation, the number of patients and their clinicopathological features and outcome (progression-free survival (PFS) and overall survival (OS)).
Thyroid autoimmune diseases are not a contraindication for immunotherapy Given the role of PD-1/PD-L1 in limiting autoimmunity [9,10,14], patients with preexisting autoimmune Hashimoto's thyroiditis (HT) and/or an asymptomatic elevated titer of thyroid autoantibodies are at-risk of disease worsening or de novo irAEs [16,17]. Paradoxically, the occurrence of irAEs and particularly of thyroid dysfunction emerges as a lesser evil than previously anticipated when treating with ICIs [18].
Indeed, cancer patients with solid tumors who experience irAEs consistently display a significant longer overall survival and overall response rate compared to those lacking toxicity, thus irAE correlates with better prognosis [16,[18][19][20][21][22][23][24][25][26]. Nevertheless, several flaws in the selection of the patients may preclude the establishment of this relationship [27]. These include: i) as large-sized prospective investigations are missing, the incidence of irAEs varies significantly across studies, as does the tumor type (melanoma being of better prognostic than NSCLC), the tumor staging (over-representation of stages III versus stages IV), and the type of ICI used (anti-CTLA4 being more toxic than anti-PD-1/PD-L1). Ii) In addition, patients who live longer ('long-term survivors') are more likely to receive multiple lines of therapy and to develop more irAE than patients who die earlier. However, these confounding factors for recruitment and the latter classical "immortal time bias" can erroneously link better prognosis with the occurrence of irAE. That said, it should be acknowledged that a significant association between irAE and the ICI response persists after adjusting for potential confounders in a multivariable analysis [19,20]. Of interest, regardless of the affected organs (thyroid, skin, lung…), the presence or the treatment of irAEs does not require interruption or compromise ICI treatment [14,[16][17][18]. In light of these observations, it can be assumed that patients with preexisting AITD, and more generally autoimmune or inflammatory diseases, might represent a subgroup of cancer patients who can benefit substantially from anti-PD-1/PD-L1 immunotherapies [18]. To move forward to personalized management, further large-scale validation is needed to determine whether a certain type or severity of an autoimmune or inflammatory disease could point to treatment efficacy and used as a clinical predictive biomarker of anti-PD-1/PD-L1 efficacy.

ICI-induced thyroid autoimmunity: a new entity distinct from thyroid autoimmunity
The common clinical presentation and management of irAEs and AITD fuel the hypothesis that they share an underlying mechanism (Figure 1). Both can be the consequence of the reactivation of the immune system by ICIs by misfiring healthy thyroid cells and killing the 'wanted' tumor cells [14]. Downstream of inflammatory attack, the pathogenesis of AITD and irAE involve the secretion of a cytokine storm, that can be integrated into a single cytokine score (the CYTOX), which correlated to immunotherapy toxicity in skin irAE [28]. In keeping with this scenario, the most severe forms of both irAEs and AITD are characterized by reduced iodine uptake and increased inflammatory metabolism (as evidenced by diffuse thyroid 18 fluorodeoxyglucose uptake on PET scan image) [17,29]. However, at odds with common pathogenesis, only 22% of patients with thyroid irAEs had elevated thyroid peroxidase antibodies (anti-TPO-ab) compared with >90% reported in Hashimoto's thyroiditis [17,29]. Likewise, the ICI-related thyroiditis resulted from a unique immune phenotype, drastically different from that of HT [16,17]. Indeed, while HT patients exhibited an increase in circulating CD3+, CD4+, and CD8+ T-cell populations, no such increase was observed for ICItreated patients with irAEs, which instead showed a specific decrease in immature NK cells, and HLA-DR lo/neg immunosuppressive cells [17].
Altogether, the latter differences support the current hypothesis that thyroid irAEs may represent, in part, a new autoimmune entity, even if further studies are needed to improve our knowledge of their specific pathogenesis (Figure 1). With the increasing use of cancer immunotherapy, and particularly of the ICI combinations, the incidence and severity of thyroid irAEs are expected to increase, emerging as a significant health concern. Therefore, this as yet unmet clinical need requires non-invasive prognostic approaches that identify patients at risk of thyroid irAEs. As discussed above, some characteristics of routine practice may be helpful in the detection of severe thyroid irAEs, ensuring patient management at an asymptomatic stage [17]. Yet, large scale collaborative efforts are critical to robustly define new prognostic biomarkers, as well as to evaluate the occurrence of thyroid irAEs as predictive biomarkers of the response of ICIs.

PD-1/PD-L1 expression in autoimmune thyroid disease
With an incidence reaching 50/100,000 per year in the female population, autoimmune thyroid diseases are the most prevalent organ-specific autoimmune diseases [2]. The two most common AITD are Hashimoto's thyroiditis and Graves' disease [30] that still poses management concerns. In both, an infiltration by immune cells can be observed but is only massive in HT (also known as chronic lymphocytic thyroiditis, CLT) [31]. The slow progression of AITD raised questions regarding the interaction of the immune system with thyroid cells [32]. Along this line, worsening AITD upon PD-1/PD-L1 blockade strongly suggests a critical underlying role for this pathway [9,14,33]. However, most of the studies into immune checkpoints carried out so far have been conducted in thyroid oncology, and only three studies on this subject in AITD have been published ( Table 1) [34][35][36][37]. Lubin et al. reported an increased expression of PD-L1 in HT glands, or papillary thyroid carcinoma (PTC) arising in such a background, while they observed little, if any, expression in healthy thyroid tissues [34]. It is noteworthy that the immune landscape of thyroid tumors displayed some features similar to AITD and more specifically to CLT. Both humoral-and cell-mediated immune responses are found to be enhanced in the microenvironment of TC and AITD (albeit to a lesser level) compared to benign thyroid nodules [38]. Lymphocyte infiltration is observed on histology within and surrounding the tumor, and also throughout the gland. Lymphocytic thyroiditis is characterized by small organized, rather than diffuse, lymphoid follicles, hereafter referred to as focal thyroiditis (FT).
Last year, Álvarez-Sierra et al. provided a specific and comprehensive analysis of the PD-1/PD-L1 axis in human AITD [39]. They demonstrated using a combined approach that the majority of infiltrating lymphocytes of GD and HT glands were effector and memory CD4+ and CD8+ T cells positive for PD-1. In addition, its ligand, PD-L1 was expressed by the neighboring thyroid follicular cells in 81% of GD glands and 100% of HT glands. Therein the infiltrating T cells produced IFNγ, which was sufficient to induce the expression of PD-L1 in vitro. Therefore, as all inflammatory actors were gathered together in the same area, this led to the notion that the PD-1/PD-L1 pathway could be active in AITD glands to safeguard against autoimmune attack. Consistently, polymorphisms in PD-1 or PD-L1 were associated with GD ( Figure 1) [40,41]. The challenge now is to elucidate the cellular mechanisms affecting PD-L1 expression in AITD. Harnessing such an immunosuppressive environment may hold promise as a new therapeutic target to fight autoimmune diseases (Figure 2).

PD-1/PD-L1 expression as a diagnostic biomarker of thyroid cancer over benign lesions
Thyroid carcinomas are the most common cancers of the endocrine gland [1]. The incidence has tripled worldwide over the last three decades and is expected to increase further. However, only a small proportion, 5 to 15% are malignant [42]. The current diagnostic strategy, mainly built around clinical evaluation, ultrasound evaluation, and fine-needle aspiration cytology (FNAC), can accurately differentiate benign from malignant thyroid nodules in 75% of patients [43,44]. However, a major challenge remains concerning the nodules that fall within the grey-zone of the 'indeterminate' category of the Bethesda classification, which provides the current cytonuclear diagnostic criteria of malignancy [44]. In these cases, a repeated FNAC or the use of surgery is recommended, and 60% of nodules are indeed benign as shown on histopathological examination [43,45]. Such diagnostic uncertainty hinders the management of these patients, leading to overtreatment, unnecessary surgery, and an increase in health care costs [46].
Due to its nonspecific cytologic aspects, the encapsulated follicular variant of papillary thyroid carcinoma (EFVPTC) has long been a challenge for diagnosis, as this cancerous lesion was likely to fall into the grey-zone. However, recently, EFVPTC was reclassified into the pre-cancerous 'non-invasive follicular thyroid neoplasm with papillary-like nuclear features' (NIFT-P) [53]. Interestingly, PD-L1 expression emerges to be higher in invasive EFVPTC compared to NIFT-P and/or benign nodules [52,54], in total agreement with their respective reclassification as cancerous and benign lesions [53]. Recently, Dell'Aquila et al. confirmed the use of PD-L1 as a biomarker of malignancy or aggressive EFVPTC disease, both in tissue biopsies and remarkably in FNAC, compared to NIFT-P [54]. Therefore, the authors proposed PD-L1 as a useful biomarker for the diagnosis of NIFT-P and the risk of EFVPTC invasiveness [54]. It will be of strong interest to compare the accuracy of PD-L1 to that of ancillary molecular techniques developed for the diagnosis of indeterminate nodules [55], and to determine whether PD-L1 can overcome their complex clinical interpretation [53].
Diagnostic value of PD-L1 in serum and FNAC as a non-invasive biomarker It is worth mentioning that the above PD-L1 IHC analyses of surgery biopsies remain invasive. Thus, exploiting the release by cancer cells into the blood of high amounts of PD-L1 (multiple myeloma, and hepatocellular carcinoma, among others [56,57], Aghajani et al. recently detected a soluble form of PD-L1 (sPD-L1) in the serum (but not in plasma) from PTC patients but not in healthy controls [58].
Altogether, PD-L1 appears to be a helpful diagnostic biomarker for indeterminate nodules, whether at the mRNA or protein level and in tissues or serum [48,50,52,54,57]. Nevertheless, no study has been designed specifically for this subgroup of patients. Therefore, further comparative clinical studies need to be performed with false-positive nodules (benign nodules, or inflammatory/ autoimmune thyroid diseases) and TC to ascertain its validity as a specific diagnostic biomarker of malignancy for indeterminate thyroid nodules (Figure 2).

PD-1/PD-L1 expression as a prognostic biomarker in thyroid carcinoma
Approximately 5% of the general population over the age of 60 years' harbors a TC. The majority of TC cases have an excellent prognosis, but a small proportion of patients will experience a more aggressive disease with a risk of recurrence [5,59]. Despite optimal care, distant metastatic spread occurs in 7-23% of cases, and represents the primary cause of thyroid cancer-related deaths with a ten-year OS rate inferior to 50% [5,59,60]. Among the clinicopathological features, the genomic alterations of TERT and BRAF have been associated with such a poor TC prognosis, the latter being targeted for the treatment of aggressive TC [5]. However, because of limited improvement in patient survival and acquired drug resistance, the usefulness of these genomic biomarkers requires new biomarkers, such as PD-1/PD-L1, which have the advantage of also being targeted by immunotherapies (as discussed later).

PD-L1 overexpression supports the progression and recurrence of well-differentiated thyroid carcinoma
Of clinical relevance, the overexpression of PD-L1 appears to predict worse prognosis, identifying patients at-risk of TC progression, representing real hope for the early management of advanced WDTC (18 clinical studies totaling 3107 patients, Table 3) [34-37, 48-50, 57, 61-70]. Indeed, the expression of PD-L1 independently correlated with all clinicopathological markers of poor TC prognosis [5]: such as male gender, an age > 45 years [49,50,69], tumor size (including extra-thyroid or multifocal lesions) [36,49,57,63,66,69], lymphovascular invasion and lymph node metastasis [34,36,49,62,64,65,69,70], or TNM stage [48,50,63]. Not surprisingly given its role in tumor immune escape, PD-L1 overexpression (protein and mRNA) is associated with reduced progression-free survival (PFS) in PTC patients [48,49,57,64,69]. Along this line, its expression tends to persist in up to 50% of the corresponding metastatic lymph nodes, being associated with a higher density of immunosuppressive Tregs, especially in cases of extranodal extension or cancer relapse [50,64,65,71]. Importantly, Zhou et al. have demonstrated that PD-L1 overexpression could confer a metastatic advantage in WDTC. Silencing PD-L1 with shRNA delayed follicular thyroid carcinoma (FTC) growth and metastasis in immunocompromised nude mice, which was suggested by the authors to be related to its role in resistance to cell death [63]. Of note, independently of its role in immune suppression, PD-L1 was suggested to contribute to glycolysis, stemness, resistance to cell death, and epithelial-tomesenchymal transition: a pre-requisite for metastasis [72][73][74]. Although critical for therapeutic intervention, the relevance of these potential tumor-intrinsic PD-L1 effects in human FTC deserves further study [63]. Indeed, aside from these encouraging data obtained with mice, we currently lack data evaluating long-term prognosis in patients with lymph node metastasis as a function of a PD-L1 cutoff. Before implementing PD-L1 as a prognostic biomarker and therapeutic target, this critical issue needs to be appreciated to overcome over-diagnosis and treatment of WDTC that would not harm patients.

Capitalizing on serum PD-L1 as a robust liquid thyroid cancer prognostic biomarker
Consistent with its prognostic value in other malignancies [56], Aghajani et al. showed that a high level of soluble PD-L1 in serum, over 0.44 ng/ml, and its detection in matched tumor tissues were associated with a significantly shorter PFS in PTC patients (uniand multivariate analyses) [57]. As a result, this study places PD-L1 in serum as a marker of bad prognosis for WDTC [57]. A longer follow-up time and a larger number of patients are needed to confirm the bad prognostic value of PD-L1 in serum in WDTC.
Anaplastic thyroid carcinoma is a rare malignancy that does not exhibit a long-term response to therapy. Histologically, it is derived from the dedifferentiation of a preexisting PDTC and WDTC. Of interest, a progressive increase in PD-L1 expression was consistently reported for WDTC (classical and insular PTC variants, PD-L1 positivity in 0-6.1% of positive cases), PDTC (25%) to ATC (22-90%) [61,62,[75][76][77][78][79][80][81]. Mechanistically, the loss of key tumor suppressors (TP53 and CDKN2A) that drive WDTC dedifferentiation correlated with PD-L1 overexpression [82]. This is in line with the work of Na et al., who demonstrated that both the expression of immunosuppressive markers (CTLA-4, PD-L1, and HLA-G) and thyroid infiltration by immune cells (ImmuneScore) were upregulated on TC dedifferentiation [75]. Given the low number of patients included -which reflects the rarity of these lethal malignancies -very little if any statistical association between PD-L1 and the other clinicopathological features of bad prognosis, such as the BRAF V600E mutational status was provided.
Unlike the other differentiated TC, medullary thyroid carcinoma is an aggressive cancer derived from parafollicular thyroid calcitonin-producing cells that do not respond to thyroid-stimulating hormone (TSH) suppression or radioactive iodine therapy (RAI), conferring a worse outcome. Of the three studies, PD-L1 positivity was reported in 6.3 to 21.8% of cases and correlated with the other MTC prognostic factors (TNM stage and postoperative calcitonin) [78][79][80]. For a cohort of 201 patients, Shi et al. showed that PD-L1 overexpression correlated with recurrent disease (40%), and with a reduced PFS in a multivariate analysis [80]. Unfortunately, the two other studies contain major flaws, including the small size of the cohort [78] and a short follow-up period [79], which preclude any statistical correlation of PD-L1 expression with PFS/OS. Similarly to BRAF V600E , the mutated RET protein kinases in MTC cells could regulate the expression of PD-L1 and the subsequent effectiveness of PD-1/PD-L1 blocking therapies, as suggested for melanoma and lung cancers [83,84]. Yet, it is regrettable that none of the studies evaluated the association of PD-L1 with these oncogenic MTC drivers. Here again, the upregulation of the PD-1/PD-L1 pathway in MTC emerged to impact patient prognosis, as it correlated with MTC size, TNM stage, lymph nodes, and distant metastasis [78][79][80].
However, in agreement with the conclusions of the three previous reviews [69,85,86], we acknowledge that alone PD-L1 remains an imperfect biomarker of TC prognosis, correlated [48,49,69,80,81] or not [36,61,62,77,78] to PFS/OS. Attention should be paid, however, to the correlation of the expressions of PD-L1/PD-1 on tumor cells and tumor-infiltrating lymphocytes (TILs) with an impaired PFS and OS in patients with ATC [81]. Therefore, we would like to nuance this claim by proposing a multi-marker PD-L1 panel by integrating the circulating and tissue levels of PD-L1, with the density of TILs or other clinicopathological features. This multi-marker panel centered around PD-L1 may offer the opportunity to improve the diagnosis of indeterminate lesions, the assessment of early recurrent disease, or the future prediction of treatment (see below) (Figure 2).

Future challenges and uncertainties of PD-L1 as a biomarker of thyroid pathologies
The above compelling studies suggest that the introduction of PD-L1 into clinical practice could improve the diagnosis and the evaluation of the prognosis of TC. However, its real value is not well established, as conflicting results in the percentage of PD-L1 positive tumors have been reported, ranging from 0 [51] to 80 % of WDTC [48][49][50], and up to 90% of ATC [35]. To explain these discrepancies, the limitations include: i) The overexpression of PD-L1 alone is not specific enough to distinguish the 'true positive' malignant from 'false-positive' benign nodules. Indeed, an inflamed microenvironment or an autoimmune pathology (such as HT or GD) can also increase the PD-L1 expression and lead to misdiagnosis.
ii) Across the studies, there is no standardization of PD-L1 staining, the primary antibody, and analysis (with a variable cutoff and scoring) [87]. Among 20 studies, nine different PD-L1 antibodies were used (the commercial E1L3N, E1J2J, Ab174838, MAB1561, MABC290 clones, the complementary SP142 and SP263, and the FDA-approved companion 22C3 antibody) and importantly only 11% of all cases were assessed with FDA-approved PD-L1 diagnostic tests.
iii) The interpretation is also complicated by the cell type analyzed, as PD-L1 expression was mainly measured on tumor cells, and rarely on tumor-infiltrating immune cells [61,65,79].
iv) A major technical bottleneck is the limited sensitivity of anti-PD-L1 towards the mature highly glycosylated protein [88], leading to 'false negative' PD-L1 staining at the plasma membrane of thyrocytes [48,51], but mainly in the intracellular location [35, 48-52, 89, 90]. Further harmonization is thus required with new antibodies that recognize the mature PD-L1 cell surface protein.
v) Moreover, the spatial heterogeneity and dynamism of PD-L1 expression within an AITD gland or a thyroid cancer make challenging the assessment of its real staining score, in a static picture, in one biopsy, at one specific moment [39]. We assume that the longitudinal monitoring of PD-L1 levels in liquid biopsies could help counteract this specific issue [57,58].
Solving these limitations will be particularly critical when the PD-L1 assay will be systematically tested to optimize the implementation of thyroid precision immunotherapy, as discussed below.

Clinical efficacy of immune checkpoint inhibitor immunotherapies for thyroid cancer
The PDTC and ATC subtypes, together with refractory WDTC and MTC are devastating diseases that will inevitably progress under standard therapy. In the field of precision thyroid oncology, the BRAF oncogene is frequently mutated in TC (25-80%), promoting aggressive tumor growth. Given this tumor oncogene addiction, several selective BRAF inhibitors (BRAFi, dabrafenib/vemurafenib) were successfully developed. Inevitably, however, the BRAF-mutated TCs relapse in only 6-7 months [91,92]. Reactivation of the MEK/MAPK pathway occurs in most drug-resistant patients [92,93,94]. To bypass this resistance, various combination therapies were developed with MEK (MEKi, trametinib) or pan-RAF inhibitors (LY3009120) [92,93,94]. Of note, the treatment with BRAFi and MEKi (dabrafenib + trametinib) inhibitors improved patient PFS and OS and was FDA approved for BRAF-mutated ATC in 2018 [95]. However, even with this combination, patients relapse and require new therapeutic options.  Male gender. Statistical significance: p < 0.05. A dash is given when the association was not specifically searched for. * When stage I is compared to stages > I, ** Median PFS was significantly lower in the PD-L1+/CD8+low subgroup, *** After excluding PTC cases with a background of CLT.  The recent advent of PD-1/PD-L1 inhibitors is changing dramatically the prognosis of many refractory cancers that were thought incurable. A high level of infiltration of T lymphocytes (CD8+, CD4+, FoxP3+) within the tumor microenvironment of WDTC positively correlated to PD-L1 expression [35,49,50]. Similarly, 21.4% of TCs coexist with an autoimmune microenvironment [35], rich in T lymphocytes that can similarly foster WDTC progression through PD-L1 overexpression [34-36, 48, 50]. The high level of infiltration of lymphocytes within their microenvironment, together with tumor overexpression of PD-L1, are two characteristics that correlate with the effectiveness of immunotherapy in many solid cancers [61,62,76,81]. This provides the molecular basis for immunotherapy for the subset of PTC patients who are refractory to RAI. However, in the beginning, there was some pessimism regarding the utility of immunotherapy in TC management [96,97]. First and foremost, because most thyroid carcinomas are WDTC of good prognosis, and current therapies (mainly surgery and RAI) are effective in controlling the disease. Compared to this strategy, scarce random clinical trials that exploit the PD-1/PD-L1 inhibitors to treat refractory WDTC were met with limited success [98], in line with their low neoantigen burden [99]. Likewise, this poor clinical benefit of ICIs should be interpreted with caution as the patients were either heavily pretreated, or not selectively enrolled after the failure of RAI and disease progression.
A combination of immune checkpoint inhibitors plus BRAF inhibitors arms the immune system against BRAF V600E+ thyroid cancer ATC is the deadliest TC cancer, and despite aggressive incorporating chemotherapy, radiotherapy, and targeted therapy with BRAFi, patients after diagnosis with ATC have a typical survival of only months [91]. The recent demonstration that ATC is characterized by a high tumor mutation burden, a high level of PD-L1 expression, a high level of TIL infiltration together with the safety of ICIs [98] all renew the interest in using immunotherapy against this deadly TC [100]. However, here again, three prospective clinical trials showed poor response to anti-PD-L1 immunotherapies before rapid progression and death [29,[101][102][103][104].
The ongoing challenge is thus to increase the efficacy of ICIs while limiting resistance and adverse effects. Since the initial failure of monotherapies, ICIs are now increasingly tested in combination therapies on patients with aggressive TC. However, while the combinations of ICIs with radiotherapy or chemotherapy are ineffective against ATC [103,104], it should be pointed out that the BRAF V600E oncogene controls the overexpression of the PD-1/PD-L1 axis in TC [35,62,67,69,70,75,105]. The positive correlation between BRAF V600E+ and PD-L1/PD-1 expression is of great clinical significance, as it might be used to select TC patients for a combination of ICIs with BRAFi. Preliminary results show a superior PFS and OS for the combination of BRAFi plus ICIs compared with single-agent treatment. Thirty percent of patients with refractory BRAF V600E+ ATC achieved complete and long-lasting remission with this combination, an impressive result that has never been observed with all current treatments [29,101,102]. In line with this initial success, evidence from animal models recapitulated the significant tumor shrinkage and a longer OS, with the BRAFi + ICIs combination, compared to the lack of efficacy of monotherapies [51,89,90].
From a molecular point of view, how can we explain the robust antitumor effect of the BRAFi + ICI combination? The biological rationale for the use of BRAFi capitalizes on the oncogene addiction of the tumor cells for growth and survival. BRAFi targets and kills tumor cells carrying the BRAF V600E oncogene. Likewise, the expression of the downstream target of the BRAF oncogene PD-L1 may decrease at the tumor cell surface. In this case, BRAFi therapy would be effective, the tumor would regress, and the combination would not lead to additional antitumor benefit.
However, the targeted BRAFi therapy alone is not very effective, and this was associated with the persistence of a high level of tumor PD-L1 expression [66,106,107]. These intriguing findings generate several hypotheses into the mechanisms of resistance, which, as discussed above, include: i) the rapid development of tumor resistance [108], ii) polyclonal tumor heterogeneity, or iii) poor tumor vascularization [109]. All would be associated with intrinsic activation of the MAPK pathway and the downstream expression of PD-L1.
While it is speculative, we suggest that when cancer cells die, they can release tumor antigens within the tumor microenvironment. Once recruited, the 'professional' phagocytes (i.e., dendritic cells and macrophages) ensure the uptake of these tumorantigens, of the antigen presentation and the priming of T cells [110]. Therefore, BRAFi can induce 'immunogenic cell death' that stimulates a cytotoxic immune response against residual tumor cells that survive the initial BRAFi treatment [111]. Thus, the recent evidence of the immunogenic cell death due to the tyrosine kinase inhibitor (KI) crizotinib is of particular interest [112]. In addition to tumor cells, the past decade has provided evidence that BRAFi can activate the MAPK pathway of host immune cells that express WT BRAF, thereby boosting their activity and tumor infiltration into the tumor. These extrinsic mechanisms fit the reported establishment of an inflamed 'hot' ATC microenvironment with massive T cell infiltration, a high level of production of type I IFN, and indirect tumor upregulation of PD-L1 [66,106,107].
Collectively, the specific antitumor activity of BRAFi combined with its effect on the immune system may support the complex antitumor effects of the targeted therapies and the additive efficacy of BRAFi + ICI combination [66,106,107]. In principle, this synergistic combination would also overcome innate or acquired resistance mechanisms for each drug. In addition to ATC, this combination should therefore be effective against advanced refractory WDTC, which shares with ATC the BRAF V600E+ oncogene. In light of these initial positive results obtained with a limited number of patients, and the mechanistic recapitulation in animal models, we strongly encourage further decisive testing of the ICIs + BRAFi combination on aggressive CT, including mechanistic studies and prospective randomized clinical trials from multi-centers.

Moving forward
More than ever, the field of immunotherapy is experiencing an increase in the use of ICI combinations with other KI such as lenvatinib in the treatment of ATC, which modulate the host immunity, the inflammatory response, and epithelialto-mesenchymal transition [113]. Another combination associates ICIs with RET inhibitors (Nivolumab + Cabozantinib) in treating hepatocellular carcinoma [114], likely by downregulating PD-L1 [83]. Hu-Lieskovan et al. provided a pre-clinical study that helps explain the efficacy of the triple-combination BRAFi + MEKi + anti-PD-1/PD-L1 in BRAF V600E+ melanoma [115]. In line with the effect on the immune system of BRAFi in BRAF V600E thyroid cancers, combining BRAFi with MEKi shapes the tumor immune landscape by increasing tumor antigen presentation, and by enhancing the migration of effector T cell to the tumor site, which contrasts with single-or dual-combinations. As these infiltrated tumors expressed PD-L1, PD-1 blockade induced sustained tumor regression in contrast to mono or dual anti-MAPK targeted therapies [115]. Before use in the clinic, the efficacy and side effects of this promising triple-combination BRAFi + MEKi + anti-PD-1/ PD-L1 are currently being explored in BRAF mutated ATC, PDTC or MTC in the NCT03181100 clinical trial ( Table 5). We guess that the ongoing development of combinations of ICIs with inhibitors of oncogenic drivers will provide a major therapeutic breakthrough in the treatment of refractory TC.
The hope for future immune checkpoint inhibitor combination strategies As of May 2020, a survey revealed that sixteen active clinical trials are evaluating immunotherapy targeting the PD-1/PD-L1 axis in unresectable, recurrent and/or metastatic TC after standard care (http://www.clinicaltrial.gov/) ( Table 5). To unleash massive immune attack within the tumor, ten trials with anti-PD-1, and six with PD-L1 inhibitors, are in the pipeline either as monotherapy or in combination with RAI, KI, radio, and/or chemotherapy, other immunotherapies or even oncolytic bacteria. Through these trials, we hope to be in a unique position to develop an armory of effective anti-PD-1/PD-L1 combinations for managing and defeating deadly TC.

Summary
Among the endocrine glands, the thyroid is the most frequently involved in autoimmune and cancerous pathologies, two conditions that are increasing worldwide. Compared to previous reviews and meta-analyses focusing on TC [69,85,86], we present a comprehensive overview of the shared involvement of the PD-1/PD-L1 pathway of thyroid carcinoma, and also of AITD, and thyroid irAEs (Figures 1 and 2). On the one hand, we discussed the implication of the PD-1/PD-L1 brake in restraining the activity of T cells and thereby the progression of AITD in patients who experience a higher risk of developing thyroid irAEs and a better outcome with PD-1/PD-L1 blockade. This may support the biological rationale of acting on immune checkpoints for AITD treatment. Similarly, PD-L1 is hijacked by the cancer cells to evade the immune system.
The BRAF V600E oncogene, or an inflammatory microenvironment (mostly caused by AITD), can induce PD-L1, which impacts thyroid tumor growth, aggressiveness, and dedifferentiation. As a result, PD-L1 (tissue or liquid) emerges as an attractive diagnostic/prognostic biomarker. However, alone PD-L1 is not sensitive or specific enough for patient stratification and should be integrated into a multi-marker panel with the other clinicopathological features, to improve the diagnostic and prognostic accuracy of TC, especially in identifying the malignancy of indeterminate nodules. Finally, immunotherapies targeting the PD-1/PD-L1 pathway may represent novel therapeutic promises for advanced, recurrent and/or metastatic thyroid carcinomas, primarily in combination with BRAFi (Figure 2).

Conclusion
We have witnessed in the last decade, the revolution of immunotherapy as a real 'life saving' option for cancers that were thought incurable. We guess that the next ten years are going to be as exciting with regard to the translation of the PD-1/PD-L1 pathway from bench to thyroid precision oncology. While much progress has been made, it is fair to say that more advances are needed before the introduction into routine clinical practice of PD-1/PD-L1 as biomarkers and therapeutic strategies. For this purpose, we encourage large scale collaborative efforts to gain further insight into the basic role of PD-L1 in thyroid pathogenesis as well as to overcome the present obstacles of clinically challenging detection, stratification, and resistance to current therapies.