CD28 (alias: Tp44) is a co-stimulatory molecule that was first described in 1987 (31, 33). Generally, CD28 is expressed constitutively on CD4⁺ and CD8⁺ T cells, including those in the TIME. Expression has been demonstrated among bone marrow stromal cells and other immune subsets such as plasma cells, neutrophils, and eosinophils (34). CD28-directed ligands belong to the B7 family of which CD80 (alias: B7-1) and CD86 (alias: B7-2) demonstrate the highest binding affinity (Kd: 4uM and 15-40uM, resp.) (25, 35, 36). CD80 and 86 are upregulated on APCs upon activation and maturation following immune stress responses. CD28 signaling induces activation of NFAT, mTOR, ERK and NFkB lowering the threshold for TCR signaling and subsequent T cell activation and proliferation, survival and effector function ( Figure 2 ). However, in the TIME, CD80/86-mediated CD28 signaling may be hampered due to competitive binding of PD-L-1 and CTLA-4 to their respect receptors (37). Both PD-1 and CTLA-4 signaling impair CD28 signaling as intracellular domains of activated or PD-L-1 bound PD-1 bind to the CD28 cytoplasmic tail with high affinity. Moreover, CD28-mediated T cell co-stimulation has been shown to be crucial for PD-1 therapy in cancer patients as loss of CD28 via T cell exhaustion was shown to correlate to clinical irresponsiveness towards PD-1 ICI (37).

In HCC, CD28 has been described to be expressed on TILs albeit at lower levels compared to PBMC-derived T cell (38). This might be explained by the fact that TIL fractions are enriched for central and effector memory T cells (Tcm and TEMRA, resp.) rather than naïve CD8 T cells (Tn). Alternatively, diminished CD28 may be a reflection of T cell exhaustion (37). Among CD3⁺ TILs, Hsu and colleagues have demonstrated high levels of PD1/CD28-coexpression (39). Furthermore, CD28 is upregulated on CD8⁺ PD1^hi HCC TILs compared to the PD1^- and PD1^int compartments thereby potentially delineating tumour-reactive TILs (40). In the TIME multiple cell types can deliver CD28 stimulatory signals as CD80 and CD86 are expressed on intra-tumoural B cell, BDCA1⁺ myeloid DC (mDC), and CD14⁺ monocytes in both HCC and CCA (18). HCC tumour cells themselves, however, have demonstrated relatively low B7 family member expression (41, 42).

As CD28 has been described to be expressed constitutively on naive T cells, agonistic targeting of CD28 is hard to restrict to the TIME or to circulating tumour-specific T cells. This was illustrated by a first-in-human clinical trial of a CD28 agonistic antibody where all healthy volunteers developed life-threatening immune-related adverse events (irAE) as a direct result of a cytokine release syndrome (43). Direct CD28 stimulation of TILs does, however, did show potential in vitro, reinvigorating anti-tumour CD8 TIL function and metabolic activity (44). Alternatively, CD28 stimulation might rather be established indirectly through antibody-mediated blockade of anti-PD-L-1 and CTLA-4 allowing increased binding of CD80/86 to CD28 and/or by lifting of inhibitory signaling via PD-1 and CTLA-4. Especially in HCC the former might contribute to the effect of ICI therapies targeting PD-L-1 and CTLA-4, since HCC tumour cells of early-relapsed disease demonstrated enhanced interaction of PD-L-1 and CTLA-4 with CD80/86 compared to primary HCC tumour cells (45). In these cases anti-PD-1/CTLA-4 blockade might enhance CD28 signaling and thereby could support T cell activation by local antigen presentation, thus improving tumour immune surveillance.

Engagement of CD28 will trigger most T cells. To specifically target only already activated cells using ICS, one might focus on activation-induced co-stimulatory receptors such as inducible T-cell co-stimulator (ICOS; aliases: CD278, CVID1) (46) ( Figure 2 ). Only a small fraction of resting memory T cells shows expression of ICOS at low levels (25). ICOS expression has been demonstrated to be expressed at intermediate levels in immune-active tumour-infiltrating Th1 cells (47). Furthermore, both in colorectal cancer (CRC) and non-small cell lung cancer (NSCLC), Duhen and colleagues demonstrated tumour-reactive CD4⁺ Th or follicular T helper (Tfh) TILs co-express PD1 and ICOS (48). Important to consider when targeting ICOS is that in the TIME ICOS may also be highly expressed on FoxP3⁺ Tregs, identifying highly immune suppressive activated Tregs among ICOS^hi fractions (47, 49). Hence, stimulating ICOS may also augment immune suppression via Tregs. The ligand for ICOS is B7H2 (aliases: ICOSLG, B7RP1, CD275) (50). ICOS and CD28 share the B7H2 ligand, although ICOS binds to B7H2 with significantly higher affinity. B7H2 is expressed constitutively on mature APCs (e.g., B cells, macrophages, and DCs) (50). Moreover, B7H2 is largely expressed on somatic cells such as tumour cells including HCC, under local control of TNFα (51, 52). Similar to CD28, agonistic ICOS engagement triggers the activation of multiple pathways that support antigen specific T cell activation ( Figure 2 ).

Upregulation of ICOSL on HCC tumour-derived plasmacytoid DCs (pDCs) was hypothesised to activate type 1 regulatory T (Tr1) cells only (53). However, in HCC TILs, ICOS is upregulated both on Tregs as well as on CD4⁺ TILs that demonstrated features of recent activation and displayed increased proliferative capacity as well (54–57). Therefore, ICOS signaling in HCC might both enhance and hamper local tumour control by liver-resident immune cells. In contrast, CCA-derived intra-tumoural Tregs demonstrated high expression of ICOS but not effector Th or CTL (19).

Consistent with these expression patterns, in preclinical studies, ICOS signaling either promoted pro-tumour responses (via Tregs) or anti-tumour responses (via Th1, Tfh, or CTL) (47). Hence, agonistic ICOS mAbs alone will most likely not achieve any anti-tumour activity due to predominant action on Treg subsets. Interestingly, in patients treated with anti-CTLA-4 mAb, ICOS appears to be upregulated on effector T cells (58). Moreover, ICOS knock-out mice do not respond well to anti-CLTA-4, suggesting a significant role for ICOS-signaling on effector T cell-mediated anti-tumour activity particularly when Tregs are blocked (59). Concordantly, in a murine tumour model, concomitant stimulation of ICOS and blockade of CTLA-4 has proven to elicit potent synergistic anti-tumour responses, pleading for clinical exploration of combination regimen of ICOS stimulation with ICI, especially anti-CTLA-4 antibodies (60, 61). Future research may focus on antibody engineering to enhance the potential of ICOS^hi Treg depletion via antibody-dependent cytotoxicity and also the timing of Treg depletion with respect to ICOS stimulation. On another note, ICOS stimulation has recently been combined with PD-1 blockade. Yap and colleagues reported on a phase 1/2 trial on vopratelimab (humanised IgG1 agonistic ICOS antibody; alias: JTX-2011) combined or not with nivolumab in refractory advanced-stage solid tumours, including 2 CCA patients in the monotherapy arm (62). Though vopratelimab alone or in combination with nivolumab was tolerated well, objective response rates (ORRs) were only 1.4% and 2.3% respectively.

CD226 (alias: DNAM1, TLiSA1) is a constitutively expressed co-stimulatory receptor that was discovered first in 1985 (63). Being expressed mostly by effector T cells, Tregs, and NK cells, CD226 regulates immune activity via interplay with its ligands CD155 and CD112.

At the tumour site CD226 expression tends to be diminished due to PD-1 and TIGIT signaling, local regulation through TGF-ß, and proteasomal cleavage (64). Given the significant immunostimulatory role of CD226 via VAV1, agonistic CD226-targeting mAbs could potentially serve as promising anti-tumour regimen (65, 66). However, CD226 plays a significant role in blood platelet adhesion and activation as well, potentially complicating its clinical application. T cell Ig and ITIM domain (TIGIT) competes with CD226 for binding CD155 (Kd 114-119nM vs. 1-3nM) and CD112 (Kd 0.31-8.97uM vs. not measureable) (67). In HCC-derived CTL, Th, and Treg TILs the TIGIT/CD226 ratio appeared to be upregulated (68). Therefore, enhancing CD226 signaling specifically on TILs might be achieved indirectly using selective blockade of TIGIT. Indeed, low PD1 expressing CD8⁺ T cells reacted to anti-PD-1 and anti-TIGIT combination regimen in a CD226-dependent manner in vitro whereas these CTLs did not respond to anti-PD-1 alone. Concordantly, in the setting of NSCLC, Banta, et al. demonstrated the importance of CD226 expression for optimal anti-tumour CD8⁺ T cell responses in the context of therapeutic ICI via PD-1 or TIGIT (69).

TIM-1 is a co-stimulatory receptor that is part of the IgSF TIM subfamily. TIM-1 ligands are TIM-4 and phophatidylserine. TIM-1 expression is induced upon activation of T cells, NK cells, B cells, macrophages, DCs, and mast cells (70). In a mouse transplant model, agonistic TIM-1 mAb have shown to stimulate effector T cell function and deprogram Tregs (71). In HCC, TIM-1 expressing B cells have been demonstrated to delineate immune suppressive subsets. TIM-1 targeting might therefore have a stimulatory effect by deprogramming Bregs as well (72). Stimulating effector T cells and hampering regulatory subsets, TIM-1-mediated ICS might be a potential candidate for enhancing anti-tumour activity (73). However, recently TIM-1 expression by cervical cancer cells was associated with cancer proliferation and migration indicating harmful effects may also arise. The expression of TIM-1 on liver tumour cells is unknown, hence more studies are required before agonistic targeting of TIM-1 can be applied in the clinic (74).

SLAMF6 (alias: NTB-A) is expressed on B, NK, and T cells as well. Remarkably, SLAMF6 expression is strongly correlated to the expression of T cell factor 1 (TCF-1) (75). Both are described to delineate progenitor exhausted CD8⁺ T cells from terminally-exhausted CTLs, making SLAMF6-mediated ICS of great interest in rescuing tumour-specific CTLs (76). SLAMF6 has neither been characterised in HCC nor in CCA. However, in HCC progenitor exhausted CD8⁺ T cells have been demonstrated ex vivo by PD1^int, TCF-1⁺, TOX^lo expression, providing a potential reservoir for SLAMF6-mediated ICS to reinvigorate anti-tumour immunity (68).

The activation-induced co-stimulatory molecule 4-1BB (aliases: TNFRSF9, CD137, ILA) was first described in 1989 (77). 4-1BB is transiently expressed upon TCR engagement on TILs, including activated CD8⁺ T cells (78–80), memory and regulatory CD4⁺ T cells (81), follicular CD4⁺ T (Tfh) cells (82), but also on NK cells (83). Additionally, 4-1BB expression by TILs is enhanced partly under hypoxic conditions through hypoxia-inducible factor 1-α (79). 4-1BB binds uniquely to its 4-1BB ligand (4-1BBL; aliases: TNFSF9, CD137L). 4-1BBL is expressed on antigen presenting cells such as dendritic cells (DCs), B lymphocytes, and macrophages (84, 85). Furthermore, upon inflammation, non-immunological human cells (e.g., smooth muscle cells, endothelial cells, hematopoietic stem cells) demonstrate expression of 4-1BBLas well, suggesting a role for these cells in effector T cell enhancement (86, 87). Besides a membrane-bound form (m4-1BB), 4-1BB also exists in a soluble form (s4-1BB) that results from alternative splicing (88). Although initially s4-1BB was observed in patients with autoimmune disease, increased levels have been demonstrated in haematological malignancies as well (89). Moreover, s4-1BB is hypothesised to function as a cancer immune escape mechanism by competing with m4-1BB for binding to 4-1BBL, thereby hampering intratumoural 4-1BB T cell costimulation (90).

In HCC and intrahepatic CCA, 4-1BB has been described to be exclusively expressed by CD4⁺ and CD8⁺ TILs (91, 92). Interestingly, Kim and colleagues have shown that 4-1BB delineates a distinct activation status among exhausted PD1^hi CD8⁺ HCC-derived TILs (92). Compared to 4-1BB^- TILs, 4-1BB⁺ PD1^hi CD8⁺ T cells expressed higher levels of TCF-1, CD28, and T-bet/Eomes, indicating a greater potential for TIL reinvigoration. Accordingly, co-stimulation of CD8⁺ TILs using a humanised IgG4 4-1BB mAb further reinvigorated anti-PD-1-mediated CD8 function in vitro. As 4-1BB⁺ PD1^hi CD8⁺ TILs are hypothesised to be tumour-reactive T cells, 4-1BB costimulation using agonistic antibodies may be promising anti-tumour strategy in HCC patients (92). Care should be taken, however, as 4-1BB has also been demonstrated to be expressed highly on Tregs in colorectal cancer-derived liver metastasis (CRLM), as well as on TIL-derived Tregs in HCC (93, 94). Nevertheless, pre-clinical mouse studies have revealed a potential of dual anti-tumour activities of 4-1BB mAb by which both Treg depletion and CD8 T cell promotion can be achieved by smartly exploiting antibody isotypes and FcγR-availability (95).

Care should be taken though as in 2008, high doses of urelumab (alias: BMS-663513), a fully humanised IgG4 mAb targeting 4-1BB, led to two hepatotoxicity-related deaths. Lower dosage regimens, however, appeared to be safe in later studies in hematological cancer patients and were accompanied by CR/PR of 0%/6%, 6%/6%, and 17%/0% in DLBCL, FL, and other B cell lymphomas, respectively (26) (96) In addition, Utomilumab (alias: PF-05082566), a fully humanised IgG2 mAb, has been studied as a single agent, engaging 4-1BB to mediate T cell ICS. It was well-tolerated safety profile among 55 patients (MCL, CRC, GC, PDAC, (N)SCLC, CCA, BC, Lymphoma, Sarcoma, etc.) (97). When combined with pembrolizumab, no dose-limiting toxicities were observed and 6 out of 23 patients demonstrated CR or PR (2 CR: RCC, SCLC; 4 PR: TC, RCC, NSCLC, and HNSCC) (98). Interestingly, clinical activity correlated with increased levels of peripheral activated memory/effector CD8⁺ T cells. Also, other combination regimen of utomilumab with mogamulizumab (anti-CCR4) or avelumab (anti-PD-L1) appeared to be safe, but anti-tumour activity remained relatively small (99, 100). Based on these experiences a very diverse landscape of second-generation 4-1BB agonists has recently been developed, with many entering clinical Phase 1 and 2 trials (101).

OX40 (aliases: TNFRSF4, CD134) is a co-stimulatory molecule that was discovered in 1987 by Paterson and colleagues (102). OX40 expression can be induced following TCR cross-linking on activated CD4⁺ and CD8⁺ T cells, and Tregs. Moreover, it is induced upon the activation of NK cells, NKT cells, and neutrophils (103–105). OX40 is overexpressed on T cells upon sustained TCR stimulation in the presence of CD28-mediated costimulation as well as IL2 (106). In vitro, OX40 gets upregulated on CD4⁺ and CD8⁺ TILs when exposed to autologous tumour cells (107). Accordingly, OX40 appears to be expressed at higher levels in TILs or tumour-draining lymph nodes when compared to PBMC-derived immune cells in melanoma, ductal mamma carcinoma, and head and neck cancer (108–112). OX40 ligand (OX40L; alias: TNFSF4, gp34) functions as the unique ligand for the OX40 receptor (113). OX40L is expressed transiently on antigen presenting cells such as DCs, B lymphocytes, and macrophages upon ligation of certain pattern recognition or cytokine receptors (e.g., TLR2, TLR4, TLR9, CD40, TSLPR, IL18R) (114–117). Moreover, T cells have been demonstrated to upregulate OX40L themselves as a result of TCR crosslinking upon T-T cell interactions, giving rise to sustained CD4⁺ T cell longevity (118). Innate-derived immune cells such as NK cells and type 2/3 innate lymphoid cells (ILC) express OX40L when triggered by NKG2D or alarmin molecules, respectively (119, 120). As reported for 4-1BBL, non-immunological cells (endothelial cells, smooth muscle cells) can express OX40L under inflammatory conditions as well (121, 122). Though, the soluble variant of OX40L (sOX40L) is increased in some types of cancer, it cannot oligomerise and hence sOX40L does not properly stimulate OX40 (123).

In HCC, OX40 was found enriched in the TIME (124–127). Expression was enhanced specifically among Treg and CD4⁺ activated helper T cell (aTh) TIL subsets (124, 126). Ligation of OX40 markedly increased CD4⁺ T cell expansion in vitro (126). CD8⁺ TIL fractions, in contrast, showed only modest expression levels of OX40 (126). However, OX40 was largely co-expressed with exhaustion marker PD1 on CD8⁺ T cells from HCV-related HCC patients, suggesting prior antigen specific activation and thus hinting to pre-existing in situ anti-tumour reactivity (124). Moreover, OX40 correlated to higher expression of other immune-activation markers such as: CD68, TIM-3, and LAG3 (125). Though, OX40 expression is overall associated with anti-tumour immunity, in HCC it has also been correlated to more aggressive disease (i.e., displaying increased alpha feto-protein (AFP) and vascular invasion) and to impaired survival (125). In contrast, in CCA, increased expression of OX40 on PBMC-derived Th and CD8⁺ T cell was rather correlated to improved recurrence-free survival, hinting to a potential anti-tumour effect (128). Despite the association of OX40 to more aggressive disease in HCC, Treg, Th, and CTL in HCC-derived TILs can be skewed to the pro-inflammatory state upon multimerisation of OX40 using a hexameric OX40 ligand or bead-bound or Fc-engineered OX40 antibody in vitro. From these experiments it was suggested that FcγR(IIB)-mediated antibody multimerisation, to allow for OX40 trimerisation, is critical to effectively induce OX40-mediated anti-tumour immunity (126). Clinical trials on OX40 stimulation so far, however, have all applied traditional agonistic mAb with insufficient FcγR(IIB) affinities.

In a first attempt, a therapeutic agonistic mouse mAb to OX40 demonstrated tumour regression in 12 out of 30 late-stage cancer patients (129). Interestingly, in this study regression and SD were observed among patients with CCA. However, induction of human anti-mouse antibodies made re-administration of the agent impossible. More recently, Tavolimab (alias: MEDI0562), a humanised agonistic OX40 IgG1 antibody, was deemed safe in advanced-stage solid tumours but PR was observed in only 2 out of 50 patients, not comprising the single HCC patient included (130). Nonetheless, a marked increase in proliferation of peripheral CD4⁺ and CD8⁺ memory T cells was observed. Moreover, intra-tumoural FoxP3⁺ T cells decreased. Also, another fully humanised agonistic OX40 IgG1 antibody (BMS-986178) was shown to be safe in metastatic solid tumours (131). Clinically meaningful anti-tumour efficacy was not observed among any of the HCC patients included in this trial. Similar safety profiles on OX40 monotherapy have been observed in other recent clinical trials studying ICAGN01949 (fully humanised agonistic OX40 IgG1 mAb), GSK3174998 (fully humanised agonistic IgG1 mAb), or ivuxolimab (alias: PF-04518600; fully humanised agonistic IgG2 mAb) in advanced solid tumours and ivuxolimab in AML, respectively (132–135). Interestingly, one HCC patient, a non-responder to prior sorafenib treatment, demonstrated long-term tumour regression on ivuxolimab (132). A 30mg flat dose of ivuxolimab is currently evaluated in an expansion trial for efficacy, safety, and pharmacodynamics in HCC patients specifically. In patients treated with ICAGN01949 also one patient with metastatic CCA receiving 700mg demonstrated PR as best response with a largest decrease in tumour size of 41.9% from baseline (133). OX40 ligation may also hold promise in the neoadjuvant setting as was suggested by a recent study applying tavolimab pre-operably to HNSCC patients. In most patients enhanced immune activation in peripheral CD4⁺ and CD8⁺ T cells was observed and 4 out of 17 patients displayed expansion of putative tumour reactive CD103⁺CD39⁺CD8⁺ TILs. Importantly, in contrast to immune non-responsive patients, none of these patients developed recurrent disease (29).

Because of the modest efficacy of OX-40-mediated ICS, combination regimens have been clinically tested as well. BMS-986178 alone did not show any objective response in advanced-stage solid tumours. In combination with nivolumab and/or ipilimumab (anti-CTLA-4), ORR ranged from 0 to 13%, but did not surpass expected ORRs for anti-PD-1/CTLA-4 monotherapies (28). A second trial studying the combination of anti-OX40 antibodies with ICI (tavolimab vs. tavolimab with durvalumab (anti-PD-L1) or tremelimumab (anti-CTLA-4)) again demonstrated no improved clinical activity compared to monotherapy ICI (136). Combination therapies of multiple ICS strategies involving OX40-engagement (i.e., ivuxolimab with utolimumab and GSK3174998 with GSK1795091) do show anti-tumour immune reactivity. These trials, however, lacked a monotherapy arm (137, 138). Taken together, a clear clinical benefit using agonistic mAb to OX40 has not yet been demonstrated but there is remaining clinical potential for future FcγR(IIB)-binding antibodies.

Glucocorticoid-induced TNFR-related protein (GITR) (aliases: TNFRSF18, CD357, AITR) is a co-stimulatory molecule that was firstly described in 1997 (139). Tregs demonstrate constitutive high expression levels of GITR, whereas naive and memory TILs have lower expression levels (140, 141). Upon T cell activation via CD28 signalling, GITR expression can be enhanced rapidly in both Treg and effector TILs (140). GITR is expressed transiently at low to intermediate levels in B cells, and innate lymphocyte subsets such as macrophages, NK cells, and NKT cells as well (142–144). GITR binds uniquely to GITR ligand (GITRL; aliases: TNFSF18, CD357L) (145). GITRL is expressed on activated APCs such as macrophages, DCs, and B cells (143, 146, 147). GITRL has been demonstrated to be transiently upregulated upon TLR4 activation (148). Moreover, GITRL is expressed by endothelial cells (145, 149). It is therefore hypothesised to play a role in mediating leukocyte adhesion and migration.

In HCC, TIL-derived (CD4⁺CD25⁺ or CD4+FoxP3+) Tregs have enhanced GITR expression compared to Tregs in adjacent tissues or PBMC (150) (55). In vitro, co-culture of HCC-derived CD4⁺CD25^- Th cells and CD4⁺CD25⁺ Tregs showed that GITR ligation prevented hypo-responsiveness of effector CD4⁺ T cells, suggesting GITR engagement either hampered Tregs or stimulated Th cells directly (55). Even though GITR is most abundant on activated CD4+ Tregs, it is also detected on CD4⁺ Th and CD8⁺ effector TILs (151–153). Interestingly, these TILs demonstrated co-expression of GITR with other activation induced checkpoint inhibitors and stimulators such as CTLA-4, PD1, and 4-1BB offering opportunity for combination therapies (151, 152). When GITR ligation was combined with CTLA-4-mediated ICI in vitro on HCC-derived TILs, immunosuppression by tumour-derived Tregs was abrogated completely (151). in addition, HCC-derived CD8⁺ TILs were functionally enhanced when GITR ligation was combined with PD-1 blockade, further paving the way for combination therapies (92, 152). Similar results were obtained for intra-hepatic cholangiocarcinoma (iCCA) where CD4⁺ Tregs also demonstrated higher GITR expression compared to CD4⁺ Th, and CD8⁺ effector cells (19). Additionally, in vitro GITRL enhanced proliferation of both pre-stimulated CD4⁺ and CD8⁺ TILs compared to anti-PD-1- or -CTLA-4-mediated ICI. Interestingly, in CCA GITRL is downregulated on mDCs and monocytes in TILs when compared to adjacent liver tissues (19). Thereby, potentially hampering stimulation of anti-tumour TIL activity. Similar to CD4⁺ T cells, CD8⁺ TILs largely co-expressed GITR with other checkpoint molecules underscoring the potential of combination-therapies in both CCA and HCC.

Clinical activity of GITR-mediated ICS using TRX-518, a humanised agonistic GITR glycosylated IgG1 mAb, was evaluated in 43 patients with refractory solid tumours including 1 HCC, 1 fibrolamellar, and 1 CCA patient (30). TRX-518 was safe and depleted peripheral Tregs. Nevertheless, patients developed neither PR nor CR. CD8⁺ T cell exhaustion was hypothesised to cause clinical inactivity, pleading for combinatorial approaches using PD-1-mediated ICI. Similarly, MK-1248, a humanised agonistic GITR IgG4 antibody had no clinical effect as monotherapy for patients with solid cancers, including 1 HCC patient (154). However, when combined with pembrolizumab 1 and 2 out of 17 patients developed CR and PR, respectively (HNSCC, melanoma and cancer e.c.i.). In contrast, in an interim-analysis of a phase 2 trial on advanced solid tumours, no additional clinical activity of GITR engagement using BMS-986156 (humanised agonistic GITR IgG1 mAb) to nivolumab was demonstrated (2 and 19 out of 252 CR and PR, resp.) (27) Still, 5 out of the 12 HCC patients in the combination arm experienced radiological disease control of which four had SD and one had PR. In ICI-naïve but not ICI experienced melanoma patients, MK-4166 (humanised agonistic GITR IgG1 mAb), significantly increased the ORR (5 and 3 out of 13 CR and PR, resp.). Moreover, combination of MK-4166 with pembrolizumab did not result in enhanced clinical activity as was observed in the single HCC patient that got doublet therapy (155). In accordance, GWN323, a humanised agonistic GITR IgG1 mAb, appeared to be safe, but showed minimal clinical activity as monotherapy and modest clinical benefit in combination with spartalizumab (anti-PD-1 mAb) in both advanced solid tumours as well as lymphomas (156). In conclusion, GITR appears to be an attractive target for ICS in vitro. In light of the importance of multimerisation of TNFRSF like OX40 the focus should maybe be shifted towards novel strategies to deliver adequate GITR-mediated ICS such as multimerisation approaches, to potentially reproduce advantageous results in vivo as well.

In contrast to other TNFRSF members, CD27 (alias: TNFRSF7) is expressed constitutively on naive and effector T cells. This suggests that CD27 may act earlier upon activation in priming of T cells. Though CD27 gets downregulated on effector phenotypes, it is still expressed at moderate levels by central- and effector-memory T cells. NK cells also express CD27, albeit at lower levels in activated subsets. Moreover, CD27 is expressed by germinal centre and memory B cells (157, 158). It ligates specifically to CD70 (alias: TNFSF7, CD27L) that is expressed upon activation of DCs, B cells, and T cells.

In HCC tumours CD27 was found expressed on tumour-infiltrating B and T cell but the majority of CD27⁺ TILs were CD3⁺ T cells (159, 160). Compared to healthy controls, circulating CD27⁺CD19⁺ B cells in HCC appeared to be decreased especially with disease progression (161). Accordingly, expression of CD27 on T and B cells has been shown to be associated positively with patient survival (159, 162). CD27 has been demonstrated to be co-expressed with CD38 in HCC tumours, potentially delineating NK cells (159). Concordantly, a fraction of HCC-derived liver-resident NK cells expressed CD27. Additionally, NK cell CD27 was downregulated upon increased tumour burden which corresponded with impaired cytotoxic capacity in vitro, suggesting CD27 expression may associate with prognosis (163).

Besides T cell priming and effector differentiation, continuous ligation of CD27 is considered to play a significant role in the induction of T cell exhaustion as well as Treg survival. Nevertheless, CD27 agonistic antibodies exhibit anti-tumour functionality in both pre-clinical in vitro and in vivo models by enhancing CD27-mediated ICS and depletion of Tregs via ADCC (164, 165). Agonistic targeting of CD27 to stimulate anti-tumour T cell reactivity has been studied on a smaller scale when compared to other ICS-mediated T cell engagement. Treatment of 25 (Melanoma, CRC, OC, PC, RCC, NSCLC) and 31 patients (melanoma, RCC), with varlilumab (alias: CDX-1127), a fully humanised agonistic CD27 IgG1 mAb, in a phase 1 dose-escalation and -expansion trial, respectively was well tolerated (166). Only 1 patient experienced dose-limiting toxicity (grade 3 asymptomatic hyponatremia). Generally, pro-inflammatory immune activation was observed as characterised by increased terminally differentiated effector memory CD8⁺ T cells, HLA-DR expression on CD4+ T cells, and INFγ responses to recall antigens. One out of 15 RCC patients in the expansion cohort demonstrated a PR. Similarly, in hematologic cancers, varlilumab was tolerated well up to the maximum tested dose (167). 30 and 4 patients were tested in a dose-expansion and -escalation design, respectively of which 1 out of 4 Non-Hodgkin’s Lymphoma (NHL) patients developed CR. Both trials have shown modest clinical activity, pleading for the importance of combination therapies. Already assessed was the combination of varlilumab with nivolumab that could be applied safely in patients with several advanced solid tumours, not involving PLC (168). Though the combination regimen was not compared to nivolumab only, the ORR was not greater than expected for anti-PD-1 monotherapy. Results from a trial (NCT03396445) testing clinical efficacy of another fully humanised agonistic CD27 antibody (MK-5890) in a cross-over design comparing MK-5890 monotherapy to the combination with pembrolizumab (anti-PD-L1) are pending. To our knowledge, no data on treatment of PLC with CD27 targeting Abs are available yet.

CD40 (aliases: TNFRSF5, Bp50) is a co-stimulatory receptor that was discovered in 1986 (169). CD40 is constitutively expressed on APCs (e.g., macrophages, DCs, B cells) (170, 171). Furthermore, CD40 expression can be enhanced on fibroblasts as well as epithelial cells upon exposure to interferons (IFN) and tumour necrosis factor (TNF) (171, 172). In contrast to all other TFNRSF members, the orientation of CD40/CD40L on the DC:T cell synapse is inverted, with the receptor and ligand being expressed on the APC and T cell respectively, indicating a role in T cell priming rather than effector functionalities. CD40 ligand (CD40L; aliases: TNFSF5, CD154) serves as the sole ligand of the CD40 receptor (173). CD40L is primarily expressed by activated T cells and is upregulated upon TCR signalling (174). Under pro-inflammatory conditions, other immune cells that express CD40L are activated B cells, NK cells, mast cells, and basophils (175–178). CD40L expressing CD4⁺ T cells mainly interact with B cells in germinal centres and are therefore defined as Tfh cells (179). However, in immune oncology, CD40-mediated ICS is largely focused on the “licensing” of DCs allowing them to promote anti-tumour T cell activation and the skewing of macrophages (180). Generally, upon receptor crosslinking by CD40L, APCs upregulate major histocompatibility complex (MHC) molecules and co-stimulatory IgSF or TNFRSF ligands (e.g., CD70) (181). Moreover, CD40-activation induces secretion of cytokines that are crucial for CD8⁺ T cell activation as well as Th1 polarisation (e.g., IL12). By accomplishing enhanced antigen presentation and pro-inflammatory T cell support, CD40 activation potentiates anti-tumour immunity (181).

In HCC, CD40 is expressed on tumour infiltrating B cells and DCs. Total CD40 tumour expression correlated positively to improved survival, highlighting a significant role in anti-tumour immunity (159). Also in CCA, intra-tumoural CD40 expression was demonstrated to be an independent predictor for improved survival (182). However, in situ, the priming of anti-cancer immunity by APCs may be hampered in HCC, since intra-tumoural activated DCs showed less expression of CD40 when compared to adjacent tissues in the majority of patients, especially in those with tumour suppressor gene mutations (183, 184). Therefore, impaired DC maturation in HCC and CCA may prove to be a critical feature of tumour escape that offers therapeutic opportunity. Encouragingly, HCC PBMC-derived B cells transfected with HCC total RNA were able to induce cytotoxic T cell responses ex vivo upon activation with CD40L (185). Furthermore, CD40L-actived B cells of HCC patients were able to induce in vitro CD4⁺ and CD8⁺ responses in autologous TIL fractions to tumour-associated antigens (glypican-3, MAGE-C2) (18). These in vitro studies indicate that in HCC anti-tumour T cells can be induced when tumour antigens are effectively presented by well matured APC. As such there is potential for CD40 agonistic drugs in PLC. In four different CCA mouse models, treatment with an agonistic CD40 antibody alone achieved moderate anti-tumour immunity only (182). However, when combined with anti-PD-1 ICI, anti-tumour effects enhanced significantly by the induction of CD8⁺ T cell responses via activation of DCs and macrophages. Interestingly, anti-tumour activities were abrogated upon macrophage depletion, thus pointing out a critical role of myeloid cells for both CD40-mediated immunotherapy as well as effective anti-PD-1 therapy in CCA. These data highlight the possibility of targeting CD40 to facilitate T cell priming of non-inflammed tumours and support the combination of agonistic CD40-antibodies with T-cell targeting immunotherapy.

In 2007, selicrelumab (aliases: CP-870,893; RO7009789), a fully humanised agonistic CD40 IgG2 mAb, elicited a partial response in 4 out of 29 patients with advanced solid tumours. PR was observed in metastatic melanoma patients, and 1 out of 2 CCA patients experienced regression of a large hepatic metastasis (186). Even though, 55% of all study subjects developed grade 1-2 cytokine release syndrome (CRS) the safety profile was deemed acceptable. Similarly, in hematological cancer (B cell NHL), CD40 engagement via dacetuzumab (humanised agonistic CD40 IgG1 mAb; alias: SGN-40) was tolerated well amongst patients. Objective anti-tumour responses were reported in only a subset of patients (6 out of 50; 1 CR, 5 PR) (187). Other phase I and II trials on dacetuzumab in patients with multiple myeloma (MM) and diffuse large B-cell lymphoma (DLBCL), or on ChiLob7/4 (chimeric agonistic CD40 IgG1 mAb) in DLBCL patients have shown a similar safety profile as well as modest clinical activity (188–190). Recently, mitazalimumab, a fully humanised agonistic CD40 IgG1 (alias: ADC1013) has been shown to encompass a manageable safety profile among 95 patients with advanced solid tumours, unfortunately not including any PLC patients. Only 1 out of 95 patients (RCC) experienced partial response (191). These data have pleaded for identifying patients that are sensitive to CD40 engagement and again argue for combining CD40 agonistic antibodies with other regimen to enhance clinical activity. To enhance polarisation of macrophages to the pro-inflammatory M1 phenotype, sotigalimab (humised IgG1 agonistic CD40 antibody; alias: APX005M) was combined with inhibition of CSFR1 via cabiralizumab and co-administered with nivolumab (anti-PD-1) in patients with anti-PD-1/PD-L1-resistant melanoma, RCC, and NSCLC (192). Though the triplet therapy was tolerated reasonably and patient’s pharmacodynamics analysis suggested enhanced pro-inflammatory state, PR was reached in 1 out of 26 patients only. Lastly, in the phase 2 PRINCE trial studying nivolumab/chemotherapy, sotigalimab/chemotherapy, and nivolumab/sotigalimab/chemotherapy in metastatic PDAC patients, modest increase in OS has been observed in the nivolumab and sotigagalimab arms versus historical controls (193). However, only the nivolumab/chemotherapy arm met the primary study endpoint. Interestingly, analysis on PR-patients in the sotigalimab arm revealed that mainly genetic signatures related to CD4⁺ T cells, B cells, and DC subsets tend to predict for improved OS, in line with the envisioned mechanism of CD40 stimulation. Therefore, since CD40-mediated ICS induces T cell priming, agonistic CD40 antibodies should rather be used in combination regimen that offer tumour specific antigens as T cell targets or alleviate T cell suppression, like vaccination approaches or ICI respectively. Recent studies also suggests that we may not be looking at the right compartment to observe an immunomodulatory effect of the treatment. We may need to focus on tumour-draining lymph nodes for markers that predict response to therapy (194).

Up to now, much clinical experience on anti-cancer immunotherapy has been obtained from large cohort studies. However, these studies have primarily focused on ICI using antagonising mAb. In contrast, high affinity towards cognate epitopes as well as ligand competition are of less importance to agonistic ICS approaches and this needs to be considered when developing such strategies.

By nature, CD28-, ICOS-, and TNFRSF-mediated ICS signaling requires clustering or higher-order oligomerisation of costimulatory receptors: receptor super clustering (200). For example, soluble TNFRSF ligands demonstrate reduced agonistic capacities compared to membrane bound forms. A key factor for optimal ICS following agonistic binding is the potential of the mAb for multivalent binding and induction of receptor super clustering (201). Natural IgSF ligands (e.g., B7H1, ICOSL) configure as homodimers that account for bivalent engagement of separate singular IgSF receptors. Moreover, natural TNF ligands mostly present as tertiary confirmations that engage with cognate receptors in receptor-trimer complexes (i.e., 3:3 configuration). Receptor complex multimerisation has been demonstrated to greatly enhance co-stimulatory receptor activation. Preclinical and clinical trials have proven the synergistic effect of receptor clustering through the application of receptor ligands as well as multivalent ligands compared to monomeric signaling (202–204). Multivalent agonistic aptamers directed to OX40 and 4-1BB have shown to induce greater TNFRSF activation (205, 206). In vitro activation of HCC-derived TILs was enhanced upon exposure to multimeric OX40L and GITRL, whereas monomeric Abs did not or hardly stimulate CD4⁺ and CD8⁺ T cell proliferation, respectively (126, 152). Concordantly, whenever OX40 mAbs were coupled to beads, creating multimeric anti-OX40 IgG2, TIL expansion increased. These results highlight the significant role of TNFRSF multimerisation for activating TILs in HCC. Consequently, current agonistic ICS mAb may establish suboptimal receptor super clustering and this should be optimised to enhance ICS-mediated T cell activation.

Natural IgG is capable of bridging multiple receptors via the two connected antigen-binding fragments (Fab) that induce monomer-monomer or oligomer interactions albeit insufficient to support proper ICS (207). Agonistic mAb should either actively drive receptor oligomerisation, stabilise self-assembled receptor oligomers, or bridge between pre-existing receptor-trimers ( Figure 3 ). Receptor-trimer bridging can be achieved using FcγR-mediated receptor crosslinking. FcγRs bind the constant crystallizable fragment (Fc) domain of IgGs, and depending of the subtypes this binding induces activating or inhibitory signals regulating the function of various immune subsets (208). In human, six different FcγRs have been described of which 5 immune-activating (high affinity immunoglobulin-γ FcRI, FcγRI; low affinity immunoglobulin- γ FcRIIa, FcγRIIA; low affinity immunoglobulin-γ FcRIIc, FcγRIIC; low affinity immunoglobulin-γ FcRIIIa, FcγIIIA; low affinity immunoglobulin-γ FcRIIIb, FcγIIIB) and 1 inhibitory variant (FcγRIIB) (208). The inhibitory FcγRIIB is expressed on APC and myeloid subsets (DC, macrophage, B cell, NK cell, etc.) and uniquely functions as a scaffold crosslinking IgGs which can be exploited to crosslink also agonistic mAb (208).

In HCC-derived TILs FcγRIIB is widely expressed, facilitating the FcγRIIB-dependent effect of any agonistic ICS mAb. The majority of intra-tumoural B cells and cDCs express FcγRIIB and to a smaller extent it is expressed on monocytes and NK cells (126). Binding to inhibitory FcγRIIB has been widely demonstrated to be of utmost importance for agonistic mAbs to drive potent CD28-, CD40-, OX40-, and 4-1BB-mediated ICS via receptor trimerisation (209–211). After mAb engage with their cognate TNFRSF epitope their Fc domains are captured by FcγRIIB-expressing APCs or myeloid cells forming a scaffold facilitating TNFRSF clustering and activation in trans. Accordingly, Campos-Carrascosa and colleagues treated HCC-derived TILs with an Fc-engineered αOX40 human IgG1 antibody, termed αOX40_v12, that contained six mutations leading to increased affinity to FcγRIIB (E223D, G237D, H268D, P271G, Y296D, A330R). When compared to the native agonistic αOX40 human IgG1 antibody αOX40_v12 appeared to improve in vitro TIL expansion and functionality. These novel generation Fc-engineered agonistic mAbs are currently widely studied (126, 212, 213).

In contrast to current ICI that relies mostly on direct T cell activation, agonistic ICS mAbs may also stimulate CD8⁺ effector T cells indirectly by the depletion of immunosuppressive cells (e.g., Treg). Antibodies binding to activating FcγRs expressed by NK cells, macrophages, or granulocytes, can trigger cell-mediated cytotoxic effector functions such as antibody-dependent cellular cytotoxicity (ADCC) and phagocytosis (ADCP) (214, 215). Upon mAb-antigen recognition, target cells are opsonised by antibodies, recognised by activating FcγR expressing cells and subsequently directly killed or phagocytosed by these cells. In human, FcγR3, expressed on NK cells and monocytes, is considered the primary activating receptor driving ADCC (216).

Whereas inhibitory FcγR2B engagement stimulates downstream signaling through receptor multimerisation, activating FcγR binding facilitates ADCC and ADCP activities of agonist ICS mAbs (157). Notably, engagement by FcγR2B reduces antibody availability for activating FcγRs (217). Therefore, IgG isotype selection is critical for the design of ICS mAb therapies. IgG1 has highest affinity to all activating FcγRs, whereas IgG2 and IgG4 only bind moderately to FcγRII and FcγR1 (IgG4 only) (216). mAbs with so called high activating: inhibitory (A:I) ratios tend to have greater cell-mediated cytotoxic effector functions, but lower agonistic activity (218). To establish effective anti-tumour immunity through ICS mAbs (esp. of the IgG1 isotype), the relative effect of immune cell activation versus immune cell depletion is likely determined by the tumour’s immune context.

As the TIME of PLC is highly enriched by immunosuppressive Tregs that hamper cytotoxic T cell-mediated anti-tumour immunity, greater ADCC effects might be desired rather than direct immune cell activation. Both HCC- and CCA-derived tumour infiltrating Tregs express high levels of co-stimulatory IgSF (ICOS) and TNFRSF (4-1BB, OX40, GITR) members (18, 55, 91, 92, 124, 126). As such, agonistic ICS mAbs of the IgG1 isotype might be of particular interest in depleting immune suppressive Tregs through ADCC mediated by liver resident Kupffer cells and NK cells that express relative high levels of FcγRIII (219, 220). Similar to discussed Fc-engineering strategies to increase binding to FcγRIIB, approaches to specifically enhance binding of the Fc domain to activating FcγRs could be employed to potentiate ADCC/ADCP-mediated anti-tumour activity of ICS agonistic Abs. Full anti-tumour immunity is however likely not to be expected from depletion of immunosuppressive subsets alone. Among various solid tumours in human, including HCC and CCA, Treg depletion by GITR agonism did increase Teff: Treg ratios, but was not sufficient to activate cytolytic cells due to persistent in situ TIL exhaustion (30). Accordingly, cytotoxic T cell re-invigoration in PLC might be supported by depletion of Tregs or any other immunosuppressing subsets but should likely be supported by ICS-mediated T cell activation and ICI in the appropriate dosing-regimen as well.

In contrast to conventional ICI mAbs, ICS demonstrate a variable dose-response relationship. Classical mAbs primarily achieve clinical activity through receptor antagonism or ADCC, ADCP, and complement-dependent cytotoxicity (CDC). These modes of action establish optimal functionality at binding saturation of the cognate receptors. At peak receptor occupancy, increased dosage concentration will not induce any additional effect thereby reaching a plateau (201). In contrast, preclinical in vitro studies on ICS have shown clinical activity to decrease after a peak at specific concentrations has been reached (221, 222). Thus, agonistic ICS mAbs act in a bell-shaped dose-response rather than the classical sigmoidal dose-response functionality.

This mechanism might be partly attributed to the mAbs’ stoichiometric binding properties to the cognate receptors (201, 222). Theoretically, formation of maximal receptor superclustering should be achieved upon bisected molar concentrations of antibody to receptor thereby providing optimal bridging between the both. If antibody or receptor abundance exceed each other, inadequate antibody-receptor bridging would be established leading to isolated complexes with a 2:1 or 1:2 stoichiometry, respectively. At the dose-optimum, effective ICS-mediated T cell activation can be solely established directly via maximal receptor multimerisation. Therefore, optimal dosing strategies may require adaptations based on biomarkers (e.g., T cell proliferation or activation) to carefully monitor these dynamics.

Secondly, optimal activity at a certain dose-optimum might be explained by dynamical T cell functionality as well. Prolonged T cell activation through chronic antigen exposure via ICS-supported TCR signaling drives immune cell exhaustion leading to downregulation and activation-induced cell death. Whereas ICI mAbs disinhibit T cells that have been primed already, high dose agonistic ICS mAb concentrations could facilitate substantial T cell priming and overstimulation favouring subsequent T cell exhaustion. Combination regimen incorporating ICI might be a logical consideration to revert T cell exhaustion. However, albeit administered concomitantly rather than sequentially, various clinical trials failed to show any additional effect of ICS/ICI combination therapy over ICS monotherapies (27, 28, 135, 136, 154, 168). Therefore, better understanding of the process of T cell differentiation; in particular on the relation between priming, activation and exhaustion, is crucial. Effective combination regimen should preferentially first apply ICS agonists to enable tumour-specific T cell activation possibly even towards exhaustion. As activation and exhaustion are characterised by enhanced co-inhibitory receptor expression, this approach should then be followed by ICI mAbs to prevent suppression via co-inhibitory receptor ligands in the TIME (215, 223).

Other than ICI that bind to broadly expressed co-inhibitory receptors, co-stimulatory receptor expression might be relatively low on effector TILs, particularly on the cytotoxic immune compartments. As mentioned previously, co-stimulatory receptors are expressed fairly transiently on activated CD4⁺ and CD8⁺ T cells. However, HCC- and CCA-derived TILs demonstrate relatively lower expression of ICOS, OX40, and GITR on CD8⁺ T and (a)Th cells when compared to (a)Tregs (18, 126, 152). and thus it might be required to re-prime TILs prior to administration of any agonistic ICS Ab (combination) regimen. Strikingly, the pre-activation status of TILs correlated positively to ex vivo response rates upon OX40-mediated ICS (126). Several strategies could be of use here. i.e., toll-like receptor (TLR) agonists, vaccination, radiation or low dose metronomic chemotherapy to achieve immunogenic cell death. In HCC, previous locoregional therapies demonstrate to enhance Ki67 expression in HCC-derived TILs (126, 224). Thus, these data support the concept of priming the TIME prior to ICS-mediated anti-tumour immunity in PLC.