Mechanisms of PD-1/PD-L1 expression and prognostic relevance in non-Hodgkin lymphoma: a summary of immunohistochemical studies

Immune checkpoint blockade therapeutics, notably antibodies targeting the programmed death 1 (PD-1) receptor and its PD-L1 and PD-L2 ligands, are currently revolutionizing the treatment of cancer. For a sizeable fraction of patients with melanoma, lung, kidney and several other solid cancers, monoclonal antibodies that neutralize the interactions of the PD-1/PD-L1 complex allow the reconstitution of long-lasting antitumor immunity. In hematological malignancies this novel therapeutic strategy is far less documented, although promising clinical responses have been seen in refractory and relapsed Hodgkin lymphoma patients. This review describes our current knowledge of PD-1 and PD-L1 expression, as reported by immunohistochemical staining in both non-Hodgkin lymphoma cells and their surrounding immune cells. Here, we discuss the multiple intrinsic and extrinsic mechanisms by which both T and B cell lymphomas up-regulate the PD-1/PD-L1 axis, and review current knowledge about the prognostic significance of its immunohistochemical detection. This body of literature establishes the cell surface expression of PD-1/PD-L1 as a critical determinant for the identification of non-Hodgkin lymphoma patients eligible for immune checkpoint blockade therapies.


INTRODUCTION
In order to develop within immunocompetent hosts, it is imperative that tumors evolve several immune escape strategies, such as mutations causing antigen loss or alteration of the antigen processing and presentation machinery [1,2]. Other mechanisms that lead to immune evasion have also been identified in lymphoma; they include the impairment of immune cell infiltration through endothelial defects, the inhibition of immune activation by the secretion of suppressive mediators such as TGF-β and IL-10 [3], the local recruitment of immunosuppressive Review Oncotarget 44961 www.impactjournals.com/oncotarget cells such as regulatory T cells (Tregs) [4], tumorassociated macrophages (TAMs) [5] and myeloid-derived suppressor cells (MDSC) [6,7], and the impairment of functional anti-tumor responses through the up-regulation of immune checkpoint gene expression [8][9][10][11]. One of the most commonly dysregulated checkpoints involves the interaction of the programmed death-1 receptor (PD-1, CD279) at the surface of T lymphocytes with its ligand programmed death-ligand-1 (PD-L1) or PD-L2, which are found at the surface of macrophages and some tumor cells. This interaction delivers inhibitory signals that ultimately cause apoptosis, anergy or functional exhaustion of the T cells involved. Recently, we and others reported the overexpression of PD-L1 in non-Hodgkin lymphoma tumor cells and the increased expression of PD-1 in tumorinfiltrating lymphocytes (TILs) [9,10,12,13]. Thus, treatment with immune checkpoint inhibitors targeting PD-1/PD-L1, either alone or in combination with other immune checkpoint inhibitors, can restore T cell effector function [14] and has emerged as a promising strategy for hematological malignancy therapy, particularly in patients with refractory Hodgkin lymphoma [15,16], relapsed follicular lymphoma (FL) [17], and other aggressive non-Hodgkin lymphoma (NHL) [18,19].
Here, we review the literature on PD-1 and PD-L1 protein expression levels in NHL and the mechanisms of their up-regulation, as well as the prognostic relevance of these proteins in NHL patients.

THE BIOLOGY OF THE PD-1/PD-L1/2 AXIS
The main physiological role of PD-1 is in limiting autoimmunity in an inflammatory context (e.g. in response to infection) by restricting the activity of T cells in peripheral tissues [20,21]. PD-1/PD-L1 interactions usually occur predominantly in peripheral tissues, however in some cancers, for example lymphoma developing in lymphoid organs, PD-1 engagement can reduce the anti-tumor response of effector T cells. At the intracellular level, signaling downstream of the PD-1/ PD-L1/2 interaction reduces the duration of the synaptic contact [22], by down-regulating TCR signaling through a pathway thought to involve the SHP2 phosphatase [23] and thereby impairing the immunological synapse formed  [35,43]. PD-L1 expression can also be induced by activation of the JAK/STAT pathway via inflammatory cytokines such as IL10 [13,39]. This is through activation of JAK2 via either its molecular alteration, the inhibition of SOCS-1 [36] or by microRNA miR-135a [37]. EBV infection directly activates the PD-L1 promoter via the AP-1/cJUN/JUN-B pathway and indirectly activates it via the activation of JAK3-STAT5 by inflammatory cytokines (IFN) [13,43]. Other indirect processes that may result in molecular anomalies that induce the activation of the JAK/STAT pathway typically include the nucleophosmin-anaplastic lymphoma kinase (NPM-ALK) translocation in NPM-ALK-positive anaplastic large cell lymphoma (ALCL) [40,41] or the MYD88 L265P mutation in diffuse large B cell lymphoma [42] .
Oncotarget 44962 www.impactjournals.com/oncotarget between effector T cells and antigen presenting cells. PD-1 is expressed by T cells, B cells and natural killer (NK) cell effectors, and has been described as an exhaustion marker in cancer and chronic viral infections [24][25][26][27]. PD-L1 (CD274) is physiologically expressed at the surface of B cells, T cells and macrophages, whereas PD-L2 (CD273) is mainly expressed by antigen-presenting cells and epithelial tissues [20]. In many solid cancers, PD-1 is upregulated by a large proportion of TILs, whereas its ligands PD-L1 and PD-L2 are expressed by a variety of tumor cells [28][29][30] and cause a reduction in anti-tumor immunity.

THE PD-1/PD-L1/2 AXIS IN DLBCL
Diffuse large B cell lymphoma (DLBCL) are the most common type of lymphoma in adults. The prognosis for DLBCL patients is heterogeneous and remains poor in 40% of cases despite the introduction of therapy combining rituximab with cyclophosphamidedoxorubicin-oncovin-prednisone (CHOP) [44][45][46]. The  [47]. These subtypes arise from different stages of B-cell differentiation and acquire distinct oncogenic abnormalities which promote tumor proliferation and survival [47,48]. The GCB subtype more frequently presents with genetic lesions such as BCL2 translocations, PTEN or ING1 deletions, MDM2 gains or amplifications and p53 mutations. In contrast, DLBCL ABC exhibit chronic BCR activation (e.g. through CD79A/B

Figure 4: PD-1/PD-L1 expression and their prognostic value in follicular lymphoma.
Oncotarget 44967 www.impactjournals.com/oncotarget mutations) and present with other genetic alterations such as BCL2 amplifications or INK4-ARF deletions. The chronic BCR signaling pathway is a well-known target for therapeutic interventions (e.g ibrutinib, PKC inhibitors, lenalidomide) but activating mutations (e.g of CARD11, Bcl10 translocations, A20 deletions) occasionally hamper drug efficacy [48]. However, the physiopathology of DLBCL is not limited to tumor cells since the DLBCL microenvironment (ME) has also proven to be mandatory for its carcinogenesis. Within the ME, the tumor stromal cells and the composition of the immune infiltrate influence the progression of the DLBCL disease [49][50][51][52]. In addition, the strength of the immune response can be functionally impaired by several tumor immune escape mechanisms, most notably those upregulating immune checkpoint molecules such as PD-1/PD-L1 [53].

PD-1/PD-L1/2 expression in DLBCL
PD-L1 is expressed by both DLBCL tumor B cells and by non-malignant cells from their immune microenvironment, such as macrophages [10,54]. In DLBCL, PD-L1 expression has been reported in around 20-30% of DLBCL cases but this figure varies greatly depending on the cut-off applied (which ranges from 5 to 30%) and the cell compartment analyzed (tumor/nontumor cells) [10,12,13,54] (Figures 2A and 2B) ( Table  1). All of the studies that have investigated PD-L1 levels in DLBCL have reported higher expression rates in the non-GCB DLBCL subtypes [10,12,13,54]. In contrast, the expression of PD-L2 has been less well documented, as most NHL cell lines do not express it [12]. One report found low PD-L2 expression in DLBCL cells without a significant difference between subtypes [10]. Recently, a retrospective study conducted a double staining of PD-L1 and PAX5 in DLBCL samples in order to precisely quantify the rate of PD-L1 + cells in both the tumor and non-tumor compartments [54]. They found that 10.5% of DLBCL samples expressed PD-L1 in tumor cells (n=132/1253 DLBCL samples; using a cut-off of 30%), while it was expressed in 15.3% of ME cells, which were essentially composed of macrophages (n=172/1121 DLBCL samples; using a cut-off of 20%). This study also confirmed the predominant expression of PD-L1 in non-GCB subtypes of DLBCL NOS.
In contrast to PD-L1, PD-1 expression has almost exclusively been detected in the ME cells of DLBCL, with a varying number of cells per mm² examined [10,55,56]. DLBCL tumor cells have been found to express a low level of cell surface PD-1 [10,[55][56][57], and sometimes coexpress both PD-1 and PD-L1 [10,58]. Kiyasu et al. [54] also reported that the number of PD-1 + TILs was higher in GCB DLBCL and was inversely correlated with the number of PD-L1 + tumor and ME cells, although these conclusions remain controversial [59,60]. PD-L1 expression is also considered to be a hallmark of EBV-associated lymphoproliferative disorders. These include EBV + plasmablastic lymphoma (PL) (where 20% of tumor cells are PD-L1 + ) [58], EBV + post-transplant lymphoproliferative disorders (PTLD) (where 60% of tumor cells are PD-L1 + ) [13], EBV + DLBCL of the elderly (where 100% of tumor cells are PD-L1 + ) [13] and the recently described EBV + DLBCL subtype [31] found in young patients (where 76% of patients display an expression of PD-L1 by more than 5% of their tumor B cells) [61]. This frequent up-regulation of PD-L1 by EBV + lymphoma cells, and the inhibition of EBV-induced lymphomagenesis following PD-1/ PD-L1 blockade in a mouse model [62], suggest a link between EBV infection and PD-1/PD-L1 upregulation. Furthermore, PD-1 and PD-L1 have also been shown to be expressed by infiltrating immune cells of EBV + lymphoma patients, suggesting that their expression underlies the tolerogenic immune response induced by EBV [58,61]. However, PD-L1 expression does not always correlate with EBV infection since it has also been reported in EBV -PTLD [13] and EBV -PL [13,58].

Mechanisms of PD-L1 and PD-L2 overexpression in DLBCL tumor cells
Nearly 20% of DLBCL NOS are reported to carry genetic anomalies and chromosomal alterations that lead to PD-L1/2 overexpression [35]. Specifically, the structural anomalies on chromosome 9p24.1 have been significantly correlated with PD-L1 expression in DLBCL [54,63]. Other translocations involving IGH genes that lead to PD-L1 overexpression have also been reported [35]. Recently, Georgiou et al. [35] reported 12% of gains, 3% of amplifications and 4% of translocations of the PD-L1/PD-L2 locus in the non GCB subtype, and lower rates in the GCB subtype. Around 30% of non-GCB DLBCL are also reported to carry MYD88 mutations that cause the chronic activation of the JAK/STAT pathway and in turn stimulate the expression of PD-L1 [35,39,42,43]. Likewise, EBV infection could account for the overexpression of PD-L1 in EBV + DLBCL since antiviral and inflammatory cytokine responses also activate the JAK/STAT pathway. In addition, the EBV-encoded latent membrane protein (LMP)-1 activates AP-1 (via cJUN/JUN-B components) and the JAK/STAT signaling pathways which, respectively, activate the PD-L1 enhancer and promoter [38].

The prognostic impact of PD-1/PD-L1 expression in DLBCL
As depicted in Figure 3, PD-1 and PD-L1 expression in DLBCL samples have a prognostic value in DLBCL [54,59,70]. Using a large series of 1200 DLBCL samples, Kiyasu et al. [54] demonstrated that patients with PD-L1 + DLBCL had inferior overall survival rates than PD-L1 -DLBCL patients. Moreover, patients with PD-L1 + tumor cells but low PD-1 + TIL counts had poorer prognoses than patients with PD-L1 -DLBCL and high PD-1 + TIL counts [54]. Elevated soluble plasma PD-L1 (sPD-L1) levels in DLBCL patients has also been shown to correlate with the lowest three-year overall survival rates (3-year OS of 76% versus 89%, P<0.001) [71,72]. Moreover, reports of tumor cell PD-L1 expression have been almost completely assigned to the non-GC subtype of DLBCL which have the worst prognosis among the DLBCL subtypes [10,12,54]. Other studies have correlated better DLBCL patient survival with their TILs having a higher PD-1 expression [55,59,70]. Conversely, however, DLBCL patients with a high PD-1 expression on their circulating CD4 + T cells showed an aggressive clinical course [73].

Expression of the PD-1/PD-L1/2 axis in FL
In contrast to DLBCL, most follicular lymphoma (FL) tumor cells do not express PD-L1 or PD-L2 [10,12,68,74], however PD-1 + cells are abundant in the ME of FL [10]. In FL, these PD-1 + cells include not only TILs but also follicular helper T cells (T FH ) from lymphomatous follicles or residual germinal centers [75,76] (Figure 2C). FL ME macrophage cells have an increased expression of PD-L1 ( Figure 2D), but these levels remain lower than those in the ME of DLBCL [10].

The prognostic impact of PD-1/PD-L1 expression in FL
Conflicting data have been reported regarding the prognostic impact of PD-1/PD-L1 expression in FL. As depicted in Figure 4, some studies have reported a correlation between low PD-1 + cell counts and either a high histological grade of FL or a higher risk of transformation to DLBCL [77]. Carreras et al. showed an association between high PD-1 + cell counts and better overall survival (OS) and progression-free survival (PFS), although these results remain controversial [78]. These divergent results could possibly be accounted for by the various subtypes of PD-1 + lymphocytes tested (T FH , exhausted T cells, or Tregs). The prognostic value of PD-1 staining in immune cells may also rely on the pattern of infiltration in FL samples, since high rates of intra-follicular PD-1 + cells have been correlated with a good prognosis while PD-1 + inter-follicular and diffuse infiltrates have poorer outcomes [75,[79][80][81]. In addition, flow cytometry measures of PD-1 expression levels evidenced two different CD4 + PD-1 bright and CD4 + PD-1 dim T cell subtypes that were associated with different outcomes [82]. In contrast to PD-1, the prognostic relevance of PD-L1 staining in FL remains poorly studied so far. Only one study reported a significant correlation between PD-L1 expression by tumor cells and lower survival rates of FL patients [83].

THE PD-1/PD-L1/2 AXIS IN OTHER SMALL B CELL LYMPHOMA
The PD-1/PD-L1 axis also remains poorly documented in other small B cell lymphoma. From the few available studies it appears that LPL, MCL and MZL tumor cells are usually negative for PD-L1 IHC staining [12,56,68]. In CLL, PD-1 is expressed by both reactive T cells and some paraimmunoblasts and prolymphocytes in the proliferative centers [56,84] (Figure 2E). Few studies have reported PD-L1 + cells among the circulating blood cells of CLL patients [83,84], and the expression of PD-L1/L2 by CLL tumor cells showed conflicting results [12,56,68,84].

THE PD-1/PD-L1/2 AXIS IN T CELL LYMPHOMA
The PD-1/PD-L1 axis is involved in both the development and the immune escape of some PTCL malignancies. From a diagnostic point of view, PD-1 expression is usually observed at the surface of angioimmunoblastic T cell lymphoma (AITL) tumor cells derived from normal T FH . This also applies to other lymphoma with a TFH phenotype according to the new WHO 2016 classification [31] including follicular T cell lymphoma (FTL) and other PTCL with a TFH phenotype www.impactjournals.com/oncotarget (PTCL-TFH) ( Figure 2F). PD-1/PD-L1 expression has also been observed on the surface of tumor ME immune cells in these PTCL subtypes, however PD-L1 expression has rarely been observed in tumor cells from these T FH subtypes, with no PD-L1 + tumors from 4 cases reported in one study [56] and 1/21 elsewhere [85]. PD-L1 expression appears to vary widely amongst tumor cells, with low rates ranging from 0% [56] to 17% of cases [85], however one study has suggested that it is higher in refractory PTCL NOS, with 28% of rPTCL NOS cases being PD-L1 + [86].
In contrast, Marzec et al. [41] reported a consistent overexpression of PD-L1 in anaplastic lymphoma kinasepositive anaplastic large cell lymphoma (ALK + ALCL) cell lines. This report was confirmed in biopsies from ALK + ALCL tumors, with frequencies of PD-L1 + cases varying from 34 to 100% of the analyzed cases [12,85,87]. Marzec et al. [41] showed that activation of the transcription factor STAT3 by the nucleophosmin-ALK (NPM-ALK) fusion protein could be responsible for the increased expression of PD-L1 at the cell surface of ALK + tumor cells. Moreover, the same group showed that NPM-ALK induces the activation of the IL-10 and TGF-β cytokines. Since IL-10 can also activate JAK/STAT signaling via STAT3, and thus induce the up-regulation of PD-L1 [39], one may reasonably speculate that NPM-ALK directly up-regulates PD-L1 via STAT3 or IL-10.
The expression of PD-1 and/or PD-L1 has also been reported in some cutaneous lymphoproliferative disorders. For instance, some mycosis fungoïdes (MF) and Sezary syndrome (SS) tumor cells have been found to express PD-1 at their surface. Similarly, some rare cutaneous disorders characterized by small-and average-sized CD4 + T cells have been found to be PD-1 + [88][89][90]. In addition, PD-L1 expression has been reported on SS/MF tumor cells and on SS/MF tumor-infiltrating dendritic cells in 27% and 73% of cases, respectively [85]. Likewise, cases of HTLV1-associated leukemia and lymphoma of adults (ATLL) have been shown to overexpress PD-L1 at the surface of their tumor or ME stromal cells [54,91,92], and the presence of PD-L1 + tumor cells, together with the lack of PD-L1 + ME cells, has been correlated with a poor prognosis in ATLL patients [92].
Finally, studies on extranodal NK/T cell lymphomanasal type have reported PD-L1 + tumor cells in 67% of cases. This was presumably secondary to EBV infection via LMP1-mediated upregulation of PD-L1 or via interferon (IFN) signaling [13,43].

CLINICAL SIGNIFICANCE OF STUDYING PD-1/PD-L1/2 EXPRESSION IN NHL PATIENTS
Preliminary data from clinical trials in solid tumors and HL indicate that patients with PD-L1 + tumor cells were those who benefitted most from PD-1/PD-L1 immune checkpoint blockade treatments [15,[93][94]. Nevertheless, significant responses were also observed in some of the PD-L1patients [95]. This observation reflects the limits of using PD-L1 protein expression alone as a single predictive biomarker since its expression is heterogeneous among tumor cells and can increase spontaneously or upon treatment. The cell-to-cell heterogeneity of PD-1, PD-L1 and PD-L2 expression among cells from NHL biopsies can now be assessed by single cell RNA-sequencing technologies, but these are not widely available in diagnostic laboratories so the current clinical need is for an IHC-based test to detect cell surface protein markers. This has its own problems, however, since there is a lack of standardized IHC procedures [96] with different antibody clones and detection thresholds [97]. In fact, the levels of PD-1 and PD-L1 protein expression and their impact on clinical response in NHL patients do differ according to lymphoma subtype and the staining assessment methodology (Table 1). In solid tumors recent evidence has suggested that both mutational load, preexisting CD8 + T cell infiltration and PD-L1 expression (with a 1% cutoff) represent predictive factors for immunotherapeutic response [94,98]. So far, none of these biomarkers have been validated in NHL, although we and others have reported significant levels of PD-L1 expression in these malignancies. However, preclinical trials with immune checkpoint blockade (ICB) therapies have shown a promising efficacy, especially in relapsed/refractory (r/r) NHL with an acceptable safety profile of toxicity [99]. For example, 36% of r/r DLBCL patients and 40% of r/r FL displayed a complete response (CR) or partial response (PR) with nivolumab [100]. Pembrolizumab had an overall response rate (ORR) of 37.5% in PMBL patients and 21% in CLL patients [101]. In contrast though, TCL patients treated with nivolumab responded at a lower rate, and an overall response was achieved in only 17% of patients [99,100]. In addition, EBV + lymphoma such as EBV + DLBCL or NKTCL [13,54,58,102] and EBV + or EBV -PTLD frequently show PD-L1-expressing tumor and ME cells. Thus, in these subtypes, PD-L1 may be used as a biomarker to identify patients who may benefit from anti-PD-1/PD-L1 inhibitors. Thus, the clinical impact of PD-L1 status in EBV + lymphoma and PTLD has not yet been firmly established. Nevertheless, a significant association between PD-L1 expression and poor outcomes has been detected in early-diagnosed NKTCL patients treated with chemotherapy containing asparaginase [54,102]. ALK + ALCL patients may also be eligible for treatment by PD-1/ PD-L1 inhibitors since their NPM-ALK rearrangement up-regulates PD-L1 expression. Nevertheless, correlating this expression with clinical outcome needs to be formally validated in ALK + ALCL patients. Ongoing preclinical trials and translational research into the PD-1/PD-L1 axis in lymphoma will assess the effectiveness of PD-1/ PD-L1 inhibitors in PD-1/PD-L1-positive r/r B cell and T cell lymphoma. Thus, the validation of standardized procedures for assessing PD-1/PD-L1 expression is www.impactjournals.com/oncotarget paramount for establishing a reliable predictive marker of response, which is currently missing from our therapeutic armament for NHL patients.
Encouraged by the promising results in melanoma [103] and multiple myeloma patients [104], we anticipate that r/r NHL patients will also benefit from ICB therapies combined with other immunotherapeutic agents (such as anti-CD20 or anti-CD30), or from a combination of multiple ICB therapies (such as those targeting PD-1, CD137 and LAG3). Indeed, although the combination of anti-PD-1 with anti-CTLA-4 did not show a greater clinical response than anti-PD-1 alone for the treatment of hematological malignancies [105], many other combinations of ICB therapy are currently under investigation in NHL patients [99].
In conclusion, tumors from most NHLs carry PD-1 + and PD-L1/L2 + cells. This encompasses both PD-1-expressing TILs as well as PD-L1/2-expressing macrophages and tumor cells in some NHL subtypes. Although PD-1 expression by TILs is the mere hallmark of their physiological activation, PD-L1/2 expression by malignant cells results from several tumor-intrinsic (genetic or oncogenic) and -extrinsic immune escape mechanisms selected by pressure from antitumor immunity. Hence PD-L1 overexpression is associated with the poorest prognosis in several types of aggressive NHLs. Thus, monoclonal antibodies selectively blocking the PD-1/PD-L1 axis could preserve TILs from exhaustion and promote antitumor immunity as an effective therapeutic strategy for NHL.