The B7-CD28 Family and Friends: Unraveling Coinhibitory Interactions

Summary:

Immune responses must be tightly regulated to ensure both optimal protective immunity and tolerance. Costimulatory pathways within the B7-CD28 family provide essential signals for optimal T cell activation and clonal expansion. They provide crucial inhibitory signals that maintain immune homeostasis, control resolution of inflammation, regulate host defense and promote tolerance to prevent autoimmunity. Tumors and chronic pathogens can exploit these pathways to evade eradication by the immune system. Advances in understanding B7-CD28 pathways have ushered in a new era of immunotherapy with effective drugs to treat cancer, autoimmune diseases, infectious diseases, and transplant rejection. Here we discuss current understanding of the mechanisms underlying the coinhibitory functions of CTLA-4, PD-1, PD-L1—B7-1 and PD-L2—RGMb interactions, and less studied B7 family members, including HHLA2, VISTA, BTNL2, and BTN3A1, as well as their overlapping and unique roles in regulating immune responses, and the therapeutic potential of these insights.

ETOC blurb:

Tight regulation of immune responses is critical. In this review, Sharpe and colleagues review inhibitory members of the B7-CD28 family and their roles in regulating immune responses with an emphasis on CTLA-4, PD-1 and the increasingly complex network of cis and trans interactions with their ligands in lymphoid organs and the periphery with implications for infection, autoimmunity and anti-tumor immunity.

Introduction

The two-signal model for T cell activation^1,2 catalyzed research that has defined mechanisms regulating T cell responses. This model posits activation of naïve T cells requires two distinct signals: signal 1, delivered by interaction of the T cell receptor (TCR) with antigen-MHC complex on antigen presenting cells (APC), conferring specificity to the response, and signal 2 (co-signal or costimulatory signal) provided by interactions of molecules on APC with their receptors on T cells.

While the two-signal model was proposed for the activation of naïve T cells, we now appreciate that second signals can also promote (costimulate) or attenuate (coinhibit) signals through the TCR of antigen-experienced cells such as effector, memory, and regulatory T cells. The balance between costimulatory and coinhibitory pathways is critical for regulating T cell activation, T cell tolerance and T cell exhaustion, and maintaining immune homeostasis. Dysregulation of these pathways can contribute to autoimmune disorders^3,4, cancer^5,6, and infectious diseases^7,8. This mechanistic understanding has provided a basis for developing immunotherapies for treating cancer, autoimmune and infectious diseases, and transplant rejection.

The complementary roles of the costimulatory receptor, CD28, and coinhibitory receptor, CTLA-4, which both bind B7-1 (CD80) or B7-2 (CD86)^9,10 provided the initial paradigm for what we now see as a broader role of costimulation and coinhibition within the B7:CD28 family. CD28 family members share common structural motifs, including an extracellular Immunoglobulin V-set (IgV), stalk, transmembrane domain, and a cytoplasmic domain with tyrosine signaling motifs. B7-2 provides the major signal for CD28 costimulation of naïve T cells¹¹. B7 binding to CD28 leads to phosphorylation of tyrosine motifs, attracting adaptor proteins with Src Homology 2 (SH2) or Src homology 3 (SH3) domains to initiate downstream signaling and recruitment of the transcription factors NFAT, AP-1, and NFκb. This drives T cell proliferation, differentiation, and cytokine production⁹ (Table 1). CD28 and other costimulatory receptors in this family, such as ICOS, have been extensively reviewed elsewhere^9,12–14.

Table 1:

Characteristics of CD28/CD28-like and B7 family members

CD28 family or B7 family	% Identity to CD28 or B7-1 EC domain	Regulation		Cytoplasmic signaling motifs/modifications	Expression
		Stimuli to drive expression	Transcriptional regulators		Hematopoietic	Non-hematopoietic	Receptor/Ligand
CD28	100	Constitutive, TCR	Alpha and beta motifs in promoter region (negative)	PI3K motif, PP2A	T cell, plasma cell, NK, NKT	Unknown	B7-1, B7-2, ICOSL (In humans)
CTLA-4 (CD152)	26	TCR/CD28, BCR	NFAT, FOXP3	PI3K motif, PP2A, SHP2, PKC-n ubiquitination	T cell, B cell, NK, NKT, DC	Unknown	B7-1, B7-2, RGMb (soluble CTLA-4 only)
ICOS (CD278)	23	TCR/CD28, IL-6/12/23	NF-κB, AP-1, FOXO1, STAT3	PI3K motif	T cell, NK, NKT	Unknown	ICOSL
TMIGD2 (CD28-H, IGPR1) *Human only	23	TCR (negative)	NF-κB, AP-1, FOXO1, RFX1	Y192, Y222	T cells, NK cells, DC, ILC	Endothelial, epithelial, Constitutive (naive T cells), downregulated (activated T cells), NK, monocytes	HHLA2
PD-1 (CD279)	16	TCR, BCR, Type I IFN, IL-1/6/7/12/21/33, Notch signaling, TLR2, TLR4, TLR9, NOD signaling (in DC)	NFAT, AP-1, STAT3, STAT4, NF-κB, Blimp1, T-bet, Eomes, TOX, TOX2, FOXO1	ITIM, ITSM, SHP1, SHP2, ubiquitination	T cell, B cell, NK cell, DC, myeloid cells, ILC	Rare: Tumor (melanoma)	PD-L1, PD-L2
B7-1 (CD80)	100	BCR, type I and type II IFN, GM-CSF, cAMP, MHC II, CD40, LPS, TLR,	IRF7, NF-κB, Jun N-terminal kinase, PU.1	Unknown	B cells, NK cells, DC, macrophages	Rare: Tumor (follicular lymphoma)	CD28, CTLA-4, PD-L1
B7-2 (CD86)	26	BCR, CD40, LPS, type I and type II IFN, GM-CSF	NF-κB, STAT1, IRF1, AP-1, PU.1	Unknown	B cells, NK cells, DC, macrophages	Tumor (hematopoietic)	CD28, CTLA-4
B7-H3 (CD276)	25	IFN-γ, IL-4 (negative)	NF-κB, STAT3, mTORC1, PI3K-AKT	Unknown	B cells, DC, macrophages (mice), T cell, NK, monocyte (human)	Tumor, Rare: Epithelia, fibroblast, osteoblast,	Unknown
ICOSL (CD275, B7-H2, B7h, B7RP-1)	24	Type II interferon, TNF-α, LPS, IL-10 (negative)	NF-κB, STAT3, IRF1, AP-1, BAFF	Unknown	T cell, B cell, monocytes, macrophages mast cells	Endothelial, epithelial, fibroblast, osteoblast	ICOS, CD28 (human)
BTNL2 (BTL-II)	23	TNF-α, IL-1β	NF-κB, IRF1, AP-1	Unknown	T, B cells, macrophages	Stomach, small intestine, lung, tumor	Unknown but expressed on activated T and B cells
PD-L2 (CD273, B7-DC)	22	IL-4, Type I and II IFN, common γ chain cytokines, TNF, GM-CSF, complement C5a, hypoxia	NF-κB, STAT3, AP-1, IRF1, PI3K-AKT, mTOR, TFEB, GATA2	Unknown	T, B cells, macrophages, DC, macrophages, BM-mast cells	Endothelial cells, nasal epithelial cells, tumor	PD-1, RGMb
PD-L1 (CD274, B7-H1)	21	Type I and type II IFN, common γ chain cytokines, TNF-α, IL-6/10/12/17/25/27, TGF-β, EGFR, Hypoxia, complement C5a, GM-CSF	STAT1, STAT3, IRF1, FOXOA1, NF-κB, AP-1, RAS/RAF/MEK, MYC, HIF-1α, HIF-2α, MAPK, PI3K-AKT, TFEB, GATA2, HSF1, ATF3, FOXP3	Ubiquitination	B cells, NK cells, DC, macrophages	Endothelial cells, epithelial cells, keratinocytes, mesenchymal stem cells, fibroblast, pancreas islet, placenta, tumor	PD-1, B7-1
B7-H4 (B7x, B7s1, Vtcn1)	21	IL-6, IL-10, TGF-β, hypoxia, GM-CSF (negative), IL-4 (negative)	PKC, ATM, STAT3, AP-1, NF-κB, HIF-1α, PI3K-AKT	Unknown	DC, monocytes, macrophages	Mesenchymal stem cells, stroma, tumor	Unknown
HHLA2 (B7-H5, B7-H7) *Human only	21	Type II IFN, IL-10, LPS, BMP-4	METLL3, IRF1, NF-κB, STAT3	Unknown	B cells, DC, monocytes	Endothelial cells, epithelial cells, kidney, colon, small intestine, lung, placenta, tumor	KIR3DL3, TMIGD2
BTN3A1 (CD277, BTN3.1, BTF5)	19	Type I IFN, mevalonate pathway, OXPHOS (negative)	AMPK, RUNXI1 (negative), CtBP1 (negative)	B30.2 (PRYSPRY)	T cells, B cells, NK cells, DC, macrophages, neutrophils	Endothelial cells, tumor	Mevalonate pathway phospho-antigens (intracellular)
BTN2A1 (BTN2.1, BTF1)	19	Mevalonate pathway, OXPHOS (negative)	AMPK, RUNX1 (negative)	B30.2 (PRYSPRY)	T cells, B cells, NK cells, DC, macrophages, neutrophils	Epithelial cells, tumor	Mevalonate pathway phospho-antigens (intracellular)
VISTA (VSIR, PD1-H)	16	Hypoxia, TGF-β, VEGF	FOXD3, NF-κB, STAT3, VEGF, HIF-1α, PI3K-AKT	Casein kinase 2, phospho-kinase C sites	T cells, DC, monocytes	Lung, tumor	VSIG-3, VSIG8, PSGL-1, LRIG, Synecdan-2
RGMb (DRAGON)	Not related	Curcumin (negative)	Unknown	Unknown (GPI anchor)	Macrophages	Neural tissues, epithelial cells, organs	PD-L2, Neogenin, BMP2, BMP4

Abbreviations are: CD, cluster of differentiation; CTLA-4, cytotoxic T-lymphocyte associated protein-4; ICOS, inducible T cell costimulator; TMIGD2, transmembrane and immunoglobulin domain containing 2; PD-1, programmed death-1; B7-H3, B7 homolog 3; B7-H4, B7 homolog 4; ICOSL, inducible T cell costimulator ligand; PD-L1, programmed death ligand-1; VISTA, V-domain immunoglobulin suppressor of T cell activation; BTNL2, butyrophilin like 2; PD-L2, programmed death ligand-2; BTN3A1, butyrophilin subfamily 3; BTN2A1, butyrophilin subfamily 2; HHLA2, HERV–HLTR-associating protein 2; RGMb, repulsive guidance molecule BMP co-receptor B; LRIG, leucine Rich Repeats and Immunoglobulin Like Domains 1; PSGL-1, P-selectin glycoprotein ligand-1; VSIG3, V-set and Immunoglobulin containing 3; VSIG-8, V-set and Immunoglobulin containing 8; KIR3DL3, Killer cell immunoglobulin-like receptor 3DL3; BMP, bone morphogenetic protein; TCR, T cell receptor; BCR, B cell receptor; IL, interleukin; TLR, Toll like receptor; NOD, Nucleotide oligomerization domain-like receptor; IFN, interferon; GM-CSF, granulocyte-macrophage colony stimulating factor; cAMP, cyclic adenosine monophosphate; MHC, major histocompatibility complex II, LPS, lipopolysaccharide; TNF, tumor necrosis factor; TGF, tumor growth factor; EGFR, epidermal growth factor receptor; VEGF, vascular endothelial growth factor; OXPHOS, oxidative phosphorylation; NK, natural killer cell; NKT, natural killer T cell; DC, dendritic cell; ILC, innate lymphoid cell; BM, bone marrow

This review builds on prior reviews^10,15–17 and focuses on recent advances in our understanding of coinhibitory pathways in the B7-CD28 family with an emphasis on Cytotoxic T-lymphocyte associated protein 4 (CTLA-4), Programmed Cell Death Protein 1 (PD-1) and its ligands Programmed death-ligand 1 (PD-L1) and Programmed cell death 1 ligand 2 (PD-L2) (Figure 1). We also discuss select B7 family members: V-domain Ig suppressor of T cell activation (VISTA), Human endogenous retrovirus-H long terminal repeat-associating 2 (HHLA2), and butyrophilins, including Butyrophilin subfamily 3 member A1 (BTN3A1) and Butyrophilin like 2 (BTNL2). We review how these pathways regulate differentiation and function of naïve, effector, memory and regulatory T cells as well as activity of other cell types in different anatomical locations and disease context. In addition, we examine interactions between these pathways, in particular how interaction of PD-L1 with B7-1 creates a bridge between the PD-1 pathway and the B7-CD28/CTLA-4 pathway and how interaction of PD-L2 with Repulsive Guidance Molecule BMP Co-Receptor B (RGMb) creates a bridge between the PD-1 pathway and the RGMb/BMP2/BMP4 and the RGMb/neogenin complexes. Finally, we consider how this knowledge is refining our view of T cell costimulation, strategies for therapeutic modulation, and key areas for future investigation.

Figure 1. Costimulatory and coinhibitory receptor-ligand interactions in the B7-CD28 family.

The TCR is activated by recognition of peptide antigen presented by MHC and this gives specificity to the T cell response. Optimal T cell activation requires a second, costimulatory signal. B7-1 or B7-2 on APC bind to CD28 on T cells and provide a costimulatory signal. T cell activation is modulated by further stimulatory or inhibitory pathways between APC and T cells. These pathways regulate the fate of T cells, their differentiation into effector, memory, and regulatory T cells and their function. Some of the receptors/ligands are expressed on both APC and T cells as described in Table 1 and the text. Tumor cells and other cells can also express some of these receptors/ligands as described in Table 1 and the text.

CTLA-4

The gene for cytotoxic T-lymphocyte associated protein 4 (CTLA-4) encompasses a signal sequence, IgV-like extracellular domain containing the MYPPPY motif that binds its B7 ligands, stalk, transmembrane domain, and cytoplasmic domain. A juxtamembrane cysteine enforces homodimerization. A splice variant lacking the transmembrane domain encodes a soluble form, while a splice variant lacking the IgV domain encodes a ligand-independent form. Both have biological activity and are reviewed elsewhere¹⁰.

CTLA-4 expression is controlled at transcriptional and post-transcriptional levels (Table 1). Within T cells, both TCR activation¹⁸ through NFAT¹⁹ and forkhead box P3 (Foxp3) drive Ctla4 expression. CTLA-4 is upregulated on CD4⁺ Foxp3^- and CD8⁺ T cells following initial activation. In contrast, CTLA-4 is constitutively expressed on regulatory T cells (Treg) due to Foxp3. CTLA-4 mRNA stability in the cytoplasm is regulated through the 3’ UTR²⁰. MicroRNAs can directly (e.g. miR-145 and miR-155), or indirectly, through modulation of Foxp3²¹, regulate CTLA-4 translation.

Cell surface expression of CTLA-4 protein is controlled at multiple levels. Following export from the Golgi apparatus, CTLA-4 is contained within intracellular vesicles that continuously cycle between the plasma membrane, cytoplasm, and/or lysosome. This process is at least partially mediated by the adaptor proteins AP-1 and AP-2 associating with the YVKM motif in the CTLA-4 cytoplasmic domain. Within endosomes, unphosphorylated CTLA-4 bound to AP-1 trafficks from the trans-Golgi network to the lysosome for subsequent degradation. Unphosphorylated CTLA-4 bound to AP-2 favors rapid internalization from the plasma membrane into endosomes in the cytoplasm. In contrast, when the YVKM motif is phosphorylated by the tyrosine kinases Lck or Fyn, CTLA-4, in conjunction with LPS responsive beige-like anchor protein (LRBA) and Rab GTPases, exhibits delayed internalization and may be recycled to the cell surface depending on the ligand bound, as discussed below.

The critical role of CTLA-4 in regulating tolerance was first demonstrated in mice, where germline deletion of Ctla4 led to systemic autoimmunity characterized by hyperproliferation of lymphocytes, multi-organ tissue destruction, elevated immunoglobulin levels, and death within 2–4 weeks of age^22,23. Similarly, in humans, haploinsufficiency of CTLA4 causes a syndrome marked by immune dysregulation, including lymphoproliferation, autoimmunity and inflammation manifesting as multi-organ lymphocytic infiltrates and cytopenia^24,25. Mutations in the cell machinery involved with trafficking CTLA-4-containing endosomes to the cell surface, such as LRBA^3,26 or associated scaffold-binding proteins, such as DEF, phenocopy germline CTLA4 haploinsufficiency, causing hypogammaglobulinemia, B cell lymphopenia, enteropathies, and often autoimmune-related cytopenias (e.g. autoimmune hemolytic anemia) ^26,27. In addition, many polymorphisms in CTLA4 in the regulatory or promoter regions, particularly the A49G polymorphism, have been associated with numerous autoimmune conditions, including Hashimoto’s and Graves’ Disease²⁸. These polymorphisms suggest subtle modulation of CTLA-4 may predispose to autoimmunity.

CTLA-4 inhibits T cell responses through both cell-intrinsic and extrinsic mechanisms. A major mechanism by which CTLA-4 modulates T cell responses is by regulating surface availability of its ligands, B7-1 (CD80) and B7-2 (CD86). CTLA-4 has a higher affinity for these ligands than CD28 and acts as a competitive inhibitor for CD28. Disease in CTLA-4 deficient mice can be abrogated by genetic deletion of both B7-1 and B7-2²⁹ or genetic modification of CD28 to disrupt the Lck phosphorylation site³⁰, suggesting that disease arises from unopposed costimulation through CD28.

One mechanism by which CTLA-4 can exert cell-extrinsic effects is through transendocytosis, a process by which it binds to and removes B7-1 and B7-2 from the surface of APCs, including dendritic cells (DC) and B cells. Because B7-1 is a non-covalent homodimer and CTLA-4 is a disulfide linked homodimer, CTLA-4—B7-1 interactions can be multimeric with low and high avidity. In contrast, B7-2 is a monomer and engages with CTLA-4 as a low avidity interaction^31,32. These different affinity receptor-ligand interactions further regulate CTLA-4. The high affinity CTLA-4—B7-1 interaction leads to CTLA-4 ubiquitination and shunting of the complex toward lysosomes for subsequent degradation³³. In contrast, CTLA-4 molecules bound to B7-2 do not trigger ubiquitination, enabling CTLA-4 to disengage from B7-2 and for CTLA-4 to recycle to the cell surface. Through these mechanisms, CTLA-4 engagement with B7-1 or B7-2 decreases their availability for CD28, thereby diminishing costimulation and T cell activation. In this way, Treg, which express CTLA-4 constitutively and at high levels, control CD28-dependent activation of naïve T cells by limiting their access to B7-1 and B7-2. Migratory DC, both conventional dendritic cell (cDC) cDC1 and cDC2 subsets, are a major target of CTLA-4 - mediated transendocytosis in the lymph node³⁴. Within the tumor microenvironment, CTLA-4 - expressing Treg also interact with B7-1 or B7-2 - expressing DC³⁵. Transendocytosis of B7-1 by CTLA-4 on Treg also can increase the availability of B7-1’s other binding partner, programmed death-ligand 1 (PD-L1), on APCs³⁶, as will be discussed further in the PD-L1 section of this review. This provides evidence that modulation of CTLA-4 expression can indirectly influence the PD-1/PD-L1 pathway.

CTLA-4 also exerts cell-extrinsic effects by inducing expression of indoleamine 2, 3-dioxygenase (IDO) in DC, which in turn contributes to the inhibitory effects of CTLA-4 on T cell responses³⁷. CTLA-4 may also inhibit T cell responses through cell-intrinsic mechanisms. Following ligand binding, CTLA-4 is phosphorylated, which leads to recruitment of the phosphatases SHP-2 and protein phosphatase 2A (PP2A) ³⁸. These phosphatases, in turn, directly inhibit intrinsic signaling through the TCR and CD28, attenuating T cell proliferation and interleukin-2 (IL-2) production.

CTLA-4 is a key mediator of Treg cell suppressive function and when co-expressed with other co-inhibitory molecules such as programmed cell death protein 1 (PD-1), lymphocyte-activation gene 3 (LAG-3), or T cell immunoreceptor with immunoglobulin and ITM domain (TIGIT), these Treg cells display enhanced suppressive function^39–42. Studies of antigen-specific Ctla4 deficient Treg cells, generated to examine the consequences of CTLA-4 deficiency without lymphoproliferation, revealed that Ctla4 deletion abrogates Treg function in vivo⁴³. Constitutive conditional loss of Ctla4 on Treg cells led to a lethal autoimmune syndrome akin to germline knockout, albeit with delayed kinetics⁴⁴. However, when Ctla4 is deleted inducibly on Treg cells in adult mice, these mice do not develop systemic autoinflammation, suggesting an important function for CTLA-4 during the perinatal period in thymic selection and central tolerance⁴⁵. Inducible loss of Ctla4 in adulthood on all cells or only on Treg cells protected mice from experimental autoimmune encephalomyelitis (EAE), and these Treg cells suppressed proliferation of T conventional cells. Massive expansion of Treg cells in the draining lymph node (LN) and central nervous system (CNS), together with upregulation of IL-10 and other coinhibitory receptors, including PD-1 and LAG-3, may explain the protection from EAE. Effector T cells from adult mice in which Ctla4 was deleted can become pathogenic in some settings but do not appear to contribute to EAE resistance of these mice^45,46. These studies demonstrate the role of CTLA-4 in limiting peripheral Treg cell expansion and activation, and capacity to control T conventional cells.

CTLA-4 also regulates specialized populations of Foxp3⁺ CD4⁺ T cells including T follicular regulatory (Tfr) cells^47,48 that suppress germinal center (GC) reactions and antibody responses, as well as T follicular helper (Tfh) cells, a CD4⁺ T cell subset that provides T cell help to B cells and is essential for GC formation and high affinity antibody production. Inducible Ctla4 gene deletion studies in adult mice revealed multifaceted roles for CTLA-4 in Tfr and Tfh cells. These studies show CTLA-4 suppresses Tfr cell differentiation and/or maintenance but are essential for the inhibitory function of differentiated Tfr cells. CTLA-4 also inhibits Tfh cell differentiation and capacity of differentiated Tfh cells to stimulate B cell responses. The transcription factors Tcf1 and Lef1 are critical to restrain Ctla4 expression in Tfh cells, an effect mediated by epigenetic modification through histone deacetylation of the Ctla4 locus in a model of vaccination⁴⁹. Further research is warranted to elucidate the potential of modulating CTLA-4 in Tfh, Tfr or Treg cells to augment vaccination efficacy and pathogen eradication.

While initial rationale for anti-CTLA-4 therapy focused on unfettering CD28 costimulation, it was later recognized that anti-CTLA-4 could also act through Fc-mediated depletion of high CTLA-4 - expressing cells, namely Treg cells^50–52. Subsequent studies demonstrated that Treg cells remained present in the tumor microenvironment of patients treated with anti-CTLA-4⁵³, suggesting that its clinical efficacy likely stems from a combination of Fc-mediated antibody-dependent cellular cytotoxicity (ADCC) and blockade of the CTLA-4—B7-1/2 axis in both the lymph node and the tumor^35,54. The balance of expression of activating and inhibitory Fc receptors differs between human and mouse, which may explain this difference^55,56. Recent evidence further suggests that anti-CTLA-4 may destabilize Treg cell function by altering glycolysis⁵⁷, at least in part due to the high linkage between CTLA-4 and FoxP3 function.

Within conventional CD4⁺ T cell populations, costimulation and coinhibition can modulate the strength of TCR signaling, favoring specific lineages of CD4⁺ T cell subsets. Germline loss of Ctla4 or antibody blockade promotes the generation of Th2 cells⁵⁸ in a STAT6 - dependent manner and the generation/induction of Th17 cells. The generation of Th2 cells provides a compelling explanation for the pronounced autoantibody production observed in CTLA-4 haploinsufficiency syndromes which display profound autoantibody generation, possibly by modulating the strength of TCR signal and interleukin-4 (IL-4) production.

In addition to regulating T cells, CTLA-4 also has functions in B cells. B cells from CTLA-4 germline deficient mice exhibit a more activated phenotype with elevated antibody titers compared to their WT counterparts^22,23. Initially, this was attributed to CTLA-4 expression on T cells, particularly Tfh and Tfr cells. However, CTLA-4 is present on both canonical B cells in the follicle, termed B2 cells, and a non-canonical, more innate-like, B1 cell population, namely B-1a cells⁵⁹.

On canonical B cells CTLA-4 expression is induced by T cell help⁶⁰. Blockade of CTLA-4 during in vitro stimulation of purified B cells with IL-4 in combination with CD40 or LPS led to decreased class switching to IgG1 and IgE-expressing B cells and IL-8, IL-10, and TNFα production via an NF-κb or STAT6-dependent pathway⁶¹. Although not tested in vivo, this suggests that CTLA-4 has a cell-intrinsic effect on B cells to modulate class switching and cytokine production.

CTLA-4 plays a critical role in B-1a cells, a distinct lineage of mature CD5⁺ B cells derived during fetal development and differing from follicular B cells⁶². B-1a cells persist as a small minority in the spleen but are the predominant B cell population in the peritoneal cavity. B-1a cells reside outside the GC and uniquely produce natural antibodies (often of low affinity) secreted through T cell independent mechanisms. CTLA-4 expression is constitutive on B-1a cells in both the spleen and peritoneal cavity. B-1a cells possess self-renewal capacity, and loss of CTLA-4 in this population enhances their proliferation and interaction with follicular dendritic cells (FDC). Intriguingly, B-1a cells express high levels of both B7-1 and B7-2⁵⁹. This raises intriguing questions about whether CTLA-4 in this cell population functions either in cis or trans with other B-1a cells or FDC to establish an additional layer of negative feedback. Furthermore, B-1a cells from mice lacking Ctla4 specifically in B cells exhibit enhanced activation and ability to differentiate into APCs and induce Tfh cells and GC formation when transferred to congenically distinct hosts⁵⁹.

Although a direct parallel between murine B-1a cells and their human counterparts has yet to be established, CTLA-4 expression has been observed in multiple human hematologic malignancies, including acute myeloid leukemia⁶³, non-Hodgkin’s lymphoma, and chronic lymphocytic leukemia (CLL). In CLL, a generally indolent disease characterized by clonal expansion of a mature CD5⁺CD23⁺ B cell population, the presence of CTLA-4 may mark an additional immunosuppressive mechanism.

PD-1

PD-1 has signal, IgV-like, stalk, transmembrane, and a cytoplasmic domain containing ITIM and ITSM motifs. A number of splice variants have been identified, as well as numerous polymorphisms linked to increased susceptibility to autoimmune diseases including lupus and multiple sclerosis and alterations in viral setpoint for chronic hepatitis B infection¹⁰.

Expression of Pdcd1 transcripts reflects a dynamic interplay between epigenetic modifiers (e.g. DNA methylation, histone modification) for chromatin availability and transcription factor binding to regulate gene expression, which varies according to cell state. PD-1 is not expressed on naïve CD4⁺ or CD8⁺ T cells, but rapidly upregulated in a TCR-dependent manner. TCR signaling is the primary driver of Pdcd1 expression⁶⁴ (Table 1) and is accompanied by demethylation of the Pdcd1 locus⁶⁵ and binding of the transcription factors NFATc1 and AP-1⁶⁶. Cell surface protein expression emerges within 24 hours or less and even prior to the initial cell division⁶⁴. Other pathways that can induce Pdcd1 expression include histone methyltransferases such as H3K4m3 to mediate NF-κB p50/p50 homodimer binding to the Pdcd1 promoter⁶⁷; cytokine signaling through IFNα⁶⁸, IL-6/STAT3⁶⁹ or IL-12/STAT-4; and Notch signaling⁷⁰. Following resolution of infection, CD8⁺ memory T cells demonstrate signs of Pdcd1 locus remethylation⁶⁵, partly through the transcription factors BLIMP-1, Tbet, and Eomes and recruitment of lysine-specific demethylase LSD1 via BLIMP-1⁷¹, to reduce Pdcd1 expression.

PD-1 levels may remain high on select populations, including CD8⁺ exhausted T cells, Tfh cells, and tissue resident memory cells (Trm) cells. Chronic TCR engagement, as occurs during chronic viral infection, drives a state of T cell dysfunction, known as T cell exhaustion, in which T cells exhibit limited proliferative potential and effector function. During T cell exhaustion, the Pdcd1 promoter undergoes remodeling and demethylation⁶⁵, marked by distinct epigenetic changes in a 23kb region upstream of the transcriptional start site^72,73. Under these conditions, NFAT signaling is repressed⁷⁴, while other transcription factors are expressed, including FOXO1⁷⁵, contributing to Pdcd1 transcription. Interestingly, in chronic LCMV infection, BLIMP-1 remains bound to the Pdcd1 locus. Emerging evidence suggests that IL-6 and subsequent STAT3 binding do not displace BLIMP-1 binding but rather overcome its inhibitory effects through increased histone acetylation⁶⁹. This implies that prolonged cytokine signaling can induce additional changes in chromatin availability, potentially contributing to sustained PD-1 expression.

Epigenetic remodeling of the Pdcd1 locus during chronic infection or tumor growth is at least partly due to the transcription factor TOX, as genetic deletion studies of TOX have demonstrated reduction or abolished PD-1 expression^76–78. However, even after antigen is cleared, as has been elegantly shown following antiviral treatment or spontaneous clearance of hepatitis C viral (HCV) infection^79,80, PD-1 surface levels decrease but its locus retains persistent alterations in chromatin accessibility. Similarly, both TOX and PD-1 expression and epigenetic remodeling persist in mouse CD8⁺ T cells exposed to iterative stimulation and rest cycles between stimulations⁸¹ or in humans at higher levels in effector memory CD8⁺ T cells recognizing epitopes from latent EBV and CMV viral infections than in influenza-specific CD8⁺ T cells⁸². These EBV and CMV-specific cells retain both proliferative capacity and cytokine production, suggesting a potential protective role for TOX and PD-1 expression following repeated stimulation. These findings suggest that both the persistent presence of antigen and recurrent, limited stimulation over time may modulate chromatin accessibility and PD-1 expression in overlapping but distinct ways.

Several studies have demonstrated that PD-1 expression can be further regulated at the post-translational level. Fucosylation at four N-glycan sites mediated by FUT8 is required for proper PD-1 localization at the plasma membrane⁸³. Moreover, cell surface expression of PD-1 is additionally regulated at the level of translation through polyubiquitination by the E3 ligase FBXO38⁸⁴. Finally, in a model of hepatocellular carcinoma, TOX stabilizes PD-1 in the cytoplasm, promoting its translocation to the cell surface⁸⁵. These findings warrant further investigation of posttranslational regulation of PD-1 expression.

PD-1 has two ligands, PD-L1 and programmed cell death protein 1 ligand 2 (PD-L2) (Figure 1, Table 1)^86,87. PD-1 engagement with its ligands opposes TCR and CD28 signaling^88,89, resulting in inhibition of T cell proliferation, differentiation and effector functions including IFNγ, TNFα, and IL-2 production. Upon binding either ligand, PD-1 undergoes phosphorylation of its immunoreceptor tyrosine-based switch motif (ITSM) and immunoreceptor tyrosine-based inhibitory motif (ITIM) by Fyn and Lck in T cells⁹⁰. This phosphorylation recruits phosphatases, particularly src homology 2-domain containing phosphatase 2 (SHP-2) but also SHP-1, which dephosphorylate key signaling molecules including the zeta chain of CD3 (CD3z), ZAP-70, and PKCθ⁹¹ and antagonizes the effects of CD28⁹² through dephosphorylation CD28 signaling components⁸⁸. Consequently, AKT, the PI3K pathway, PLCγ1, Ras-MEK-ERK signaling cascade are inhibited, leading to suppression of T cell responses. Ligand binding by PD-1 on B cells also triggers phosphorylation of both tyrosines of PD-1, followed by recruitment of SHP2 and downstream dephosphorylation of BCR - associated signaling proteins, including Ig beta, Syk, phospholipase C-γ2 (PLC-γ2) and ERK1/2⁹³. However, SHP-2 may not be the only mediator of PD-1 signaling. Mice with T cell selective deletion of Ptpn11 (gene encoding SHP-2) do not completely recapitulate the phenotype of Pdcd1 deficient mice during chronic viral infection⁹⁴, suggesting there may be other mechanisms downstream of PD-1, redundant to SHP-2. Alternatively, the effects of PD-1 may differ by cell type. For example, cell-specific effects of SHP-2 or PD-1-SHP2 signaling in T cells versus myeloid cells may provide one possible explanation⁹⁵.

PD-1 signaling also alters cell metabolism. T cells undergo a profound metabolic shift toward glycolysis following activation. PD-1 in T cells inhibits this switch by suppressing glucose and glutamine metabolism and promoting lipolysis and fatty acid oxidation^16,17,96. PD-1 deletion in myeloid progenitor cells can shift cells toward intermediates of glycolysis and the TCA cycle and elevated cholesterol⁹⁷. This increased reliance on alternate carbon sources may provide therapeutic opportunities.

Like CTLA-4, PD-1 may exert a push-pull relationship, influencing initial activation of naïve T cells, differentiation into effector or memory cells, and the balance between effector and suppressive cell populations in both the lymphoid organs and the periphery. Loss of PD-1, through genetic deletion or antibody blockade, leads to enhanced T cell activation and effector function^16,98, as has been shown in numerous disease contexts, including acute infection^99,100, autoimmunity^101,102, chronic viral infection^7,8,103,104 and malignancy^5,105. In acute infection models such as intranasal vaccina and influenza, the absence of PD-1 leads to more rapid and robust expansion of effector T cells¹⁰⁰. However, this effector T cells expansion during acute infection may come at the cost of diminished number and function of memory CD8⁺ T cells⁹⁹. Similarly, in mouse models of autoimmune disease, loss of PD-1 signaling, accelerates disease onset with increased numbers and function of pathogenic self-reactive T cells.

Early seminal work in the LCMV Clone 13 chronic infection model⁸ and HIV⁷ demonstrated a key role for PD-1 in T cell exhaustion. Virus-specific, exhausted CD8⁺ T cells highly expressed PD-1 and antibody blockade of PD-1 or PD-L1 led to improved proliferative capacity, increased production of cytokines such as IFNγ, as well as cytolytic function^7,8,106. This strategy of PD-1 or PD-L1 blockade has revolutionized cancer treatment, with FDA approval of PD-1 inhibitors for over 25 types of cancer and two tumor-agnostic indications: tumor mutational burden >10 mutations/megabase and microsatellite instable (MSI^high)¹⁰⁷.

Exhausted CD8⁺ T cells are heterogenous: stem-like progenitor exhausted cells retain the capacity to differentiate and give rise to effector progeny, while a more terminally differentiated population expressing higher levels of PD-1 in concert with other co-inhibitory molecules has the most impaired proliferative capacity and the most limited cytokine production¹⁰⁸. These populations of cells have been identified in lymph organs^104,109, and target organs during chronic viral infections, and tumors^5,105,110. PD-1 blockade increases proliferation and differentiation of the stem-like exhausted CD8⁺ T cells (defined by Tcf1⁺PD-1⁺) into effector-like cells. Using either LCMV clone 13 or a tumor model, two groups suggest that this occurs in part due to asymmetric cell division^111,112. CD28 is necessary for response of these CD8+ T cells^92,113 to anti-PD-1 blockade. Future work will likely delineate how PD-1 may coordinate these effects. Stem-like T cells in the lymph node respond to anti-PD-1. Recent work in tumor models shows that stem-like T cells further differentiate into effector cells in the tumor microenvironment in the presence of additional costimulation¹¹⁴. These data suggest that strategies to augment costimulatory signals in the tumor microenvironment may be needed to promote effector T cell differentiation and function.

PD-1 also is expressed on tissue resident memory cells (Trm) in mice following resolution of viral infection¹¹⁵ and on Trm-like cells in inflammatory lesions in brain parenchyma in mice and multiple sclerosis patients and tumors^116,117. Multiple studies link the presence of Trm-signatures or Trm-like cells to response to anti-PD-1 cancer immunotherapy^118,119 and in anti-PD-1 related colitis¹²⁰. One challenge with this nomenclature is that canonical Trm cells survive in the absence of antigen with a gene expression program related to retention in the tissues. Chronically stimulated cells may have gene programs overlapping with Trm due to their localization in tissue while simultaneously encountering antigen. Although PD-1 is expressed on both Trm cells and chronically stimulated exhausted cells, it remains to be discovered whether PD-1 drives a similar program in each of these situations. However, given the wide distribution of Trm cells across the human body, an intriguing question is how PD-1 pathway blockade interacts with signaling cues, such as the cytokine milieu, to reactivate select Trm-like populations. Further open questions include understanding how modulation of the PD-1 pathway influences the differentiation of different types of memory cells (e.g. central memory where PD-1 is not expressed vs Trm cells where PD-1 is expressed), the maintenance of these memory populations, and characteristics of memory T cells in the context of chronic inflammation. Interestingly, PD-1⁺ CD8⁺ T cells bearing a tissue resident signature have been identified in proximity to PD-L1⁺ macrophages in healthy human pancreatic donor tissue¹²¹, suggesting an important role for ongoing receptor-ligand signaling. Understanding the role of PD-1 in memory T cell biology could lead to more effective strategies for treating chronic infections and cancer.

Selective loss of PD-1 signaling on Treg cells enhances their suppressive capacity in autoimmune diseases¹²², infection¹²³, and tumors^124,125. In murine models of EAE and type I diabetes¹²², absence of PD-1 on Treg cells leads to increased suppression and protection against disease. Mechanistically, loss of PD-1 signaling reduced signaling through the PI3K-AKT pathway, contributing to enhanced Treg cell suppressive capacity¹²² as enhanced AKT signaling has been associated with decreased Treg cell suppressive capacity¹²⁶. Similarly, in a model of toxoplasmosis, PD-1 deficient Treg cells play a critical role in suppressing cDC1 function, thereby limiting pathogenesis. Loss of PD-1 on Treg cells in this model results in increased effector Treg, elevated IL-10, and increased immunopathology¹²³.

In cancer, PD-1 blockade has been linked to cases of tumor hyperprogression, characterized by rapid or explosive tumor growth. One hypothesis suggests hyperprogression arises from the enhanced suppressive capacity and abundance of Treg cells, enabling them to counteract anti-tumor effects of PD-1 blockade^124,125. In addition, PD-1 expression on Treg cells has clinical implications. In metastatic clear cell renal cell carcinoma (ccRCC), patients with a high PD-1⁺ Treg/CD8⁺ T cell ratio experience shorter progression-free survival, overall survival and lower overall response rate, suggesting the PD-1⁺ Treg/CD8⁺ T cell ratio may have predictive value¹²⁷. These findings further suggest that the ratio of PD-1⁺ Treg cells to PD-1⁺ CD8⁺ T cells at the beginning of therapy may influence treatment response. Therefore, strategies to deplete Treg cells or impair Treg cell function in combination with PD-1 blockade may improve treatment outcomes.

Like CTLA-4, PD-1 can be expressed on Tfh cells, Tfr cells¹²⁸ and B cells within the GC and outside the follicle to regulate both T cell dependent and T cell independent (TI) processes. PD-1 and PD-L1 inhibit the differentiation and suppressive function of Tfr cells. Vaccination of Pdcd1^-/- mice led to four-fold greater expansion of Tfr cells relative to wildtype controls without a change in the relative percent of Tfh cells, and these PD-1 deficient Tfr cells have more potential inhibitory function¹²⁸. PD-1 also can inhibit Tfh cell expansion, differentiation and help for IgG class switching¹²⁹. PD-1—PDL1/L2 interactions play an important role, as the PD-L1/L2 expressed on B cells can bind to and signal into PD-1 on Tfh cells, leading to reduced plasma cell numbers¹³⁰. However, as PD-1 may also be expressed on GC B cells, the PD-1 pathway may have context-dependent effects in regulating humoral immune responses. In one study, a subset of PD-1^high B cells impaired anti-tumor CD8+ T cell responses through an IL-10 dependent mechanism in a mouse hepatoma model, suggesting a way in which PD-1 signaling in B cells may inhibit anti-tumor immunity and prevent tumor growth control¹³¹.

High expression of PD-1 characterizes a CD4⁺ CXCR5^- T cell population found in peripheral blood and synovial fluid of patients with rheumatoid arthritis. This population, termed T peripheral helper cells (Tph), promotes plasma cell expansion via CXCL13 and IL-21¹³². Similar Tph cells have been identified in other autoimmune conditions including juvenile rheumatoid arthritis. These findings raise the question of whether PD-1 agonism may be a viable strategy to target T-B interactions in select autoimmune conditions. Indeed, recent findings show that PD-1 agonistic antibodies that target the lower juxtamembrane region of PD-1 and have enhanced Fc receptor binding activity can ameliorate disease in two autoimmune mouse models¹³³. A phase II clinical trial suggests PD-1 agonists have therapeutic potential in rheumatoid arthritis¹³⁴.

It is increasingly clear that PD-1 also functions on non-T cell populations. This is particularly evident in B1 cells residing in the peritoneum, NK cells, macrophages, and innate lymphoid cells (ILC) inhabiting the respiratory and mucosal surfaces. Initial studies of Pdcd1 deficient mice revealed heightened T cell-independent B cell responses, marked by elevated IgG3 production in response to DNP-Ficoll, a TI antigen¹³⁵. This was accompanied by enhanced expansion of CD5^- B-1b cells, suggesting a cell-intrinsic role for PD-1 signaling on B cells. Subsequent mechanistic studies demonstrated that PD-1 inhibited formation and reactivation of TI B cell memory cells, an interaction mediated, at least in part, by PD-L1 or PD-L2 on non-antigen specific B cells¹³⁶. Immunization with 2,4,6 trinitrophenol (TNP)-Ficoll selectively activated B-1b cells, inducing their expansion, isotype switching, and plasmablast/plasma cell differentiation. Likewise, exposure of Pdcd1 deficient mice or wild-type mice treated with anti-PD-1 to sublethal pneumococcal infection or vaccination with the polysaccharide S. pneumoniae-capsule enhanced isotype switching and IgG production by pneumococcal specific B cells, processes that are T cell independent¹³⁷. Further research is warranted to elucidate the mechanisms of these PD-1 -mediated B cell intrinsic pathways. Given that cancers may display recurrent and aberrant glycosylation patterns, exploring how the PD-1 pathway can be harnessed to engineer T cell independent anti-tumor responses holds promise for novel therapeutic approaches.

PD-1 expression has been observed on NK cells both intrinsically and through transendocytosis^138–140. ILCs also express PD-1. Particular attention has focused on IL-33 signaling in group 2 innate lymphoid cells (ILC2), which regulate immunity, including nosocomial and commensal pathogens, through antigen-independent mechanisms. Induction of IL-33 upregulates PD-1 to limit ILC2 responses in models of airway hyperreactivity¹⁴¹, cancer¹⁴², and within adipose tissue of mice fed a high fat diet ¹⁴³. PD-1 blockade increased ILC2 expansion and production of cytokines such as GM-CSF¹⁴².

Likewise, PD-1 also can be expressed on myeloid cells within the tumor microenvironment or during infection. PD-1 expression was higher on tumor-infiltrating macrophages with an immunosuppressive “M2-like” phenotype, exhibiting a foamy morphology with impaired phagocytosis¹⁴⁴. Within the tumor microenvironment, PD-L1 on tumor cells can induce negative signals in PD-1-expressing myeloid cells through SHP2, blocking type I IFN/STAT1/CXCL9 signaling to suppress cytotoxic T cell recruitment in lung metastases¹⁴⁵. In a model of listeria monocytogenes infection, PD-1 was selectively induced on DCs through engagement of TLR2, TLR3, TLR4, and nucleotide-binding oligomerization domain (NOD) proteins, inhibiting IL-12 or TNFα production¹⁴⁶. Collectively, these studies identify non-T cell mediated mechanisms by which PD-1 tempers inflammatory responses.

Finally, PD-1 has also been shown to be expressed on neurons. Both Pdcd1 deficient and selective deletion of Pdcd1 on CA1 hippocampal neurons increased CA1 neuron excitability through increased long-term potentiation, improved learning and memory behavioral test. Interestingly, PD-L1, which is expressed on neurons and microglia, was upregulated. PD-L2 was not examined¹⁴⁷. This study identifies a previously unappreciated role for PD-1 pathway in a non-hematopoietic tissue (neuron) and points to the need to better characterize the effects of coinhibitory pathways on neurological function.

There is crosstalk between PD-1 and CTLA-4 on immune cell populations. Activation through the TCR will result in upregulation of both CTLA-4 and PD-1. Moreover, loss of expression of CTLA-4 is accompanied by upregulation of PD-1 and vice-versa, as well as the increased expression of other costimulatory and co-inhibitory receptors¹⁴⁸. Blockade of either anti-CTLA-4 or anti-PD-1 can lead to inflammatory syndromes that mirror autoimmunity and are known as immune-related adverse events (irAE). For example, one study in metastatic melanoma reported rates of severe irAE (clinically grade 3 or higher) as 27% for anti-CTLA-4, 16% for anti-PD-1 and 55% for combination anti-CTLA-4/anti-PD-1¹⁴⁹, with other studies showing a dose-dependent effect of anti-CTLA-4 therapy¹⁵⁰. Although an active area of investigation, mechanisms to explain this higher rate are limited. A mouse model of myocarditis using Pdcd1^-/-Ctla4^+/- mice¹⁵¹ suggests that the myocardial pathology requires CD8⁺ T cells¹⁵². Given the distinct expression levels of these co-inhibitory receptors, with higher expression of CTLA-4 on Treg, enhanced CTLA-4 and PD-1 on effector Treg and elevated PD-1 on exhausted CD8⁺ T cells, further work is warranted to understand how PD-1 and CTLA-4 interact on different cell types in autoimmunity, infection and cancer.

PD-L1 and PD-L2

The PD-1 ligands, PD-L1 and PD-L2, have 36% sequence identity, with PD-L2 having ~3-fold greater affinity for PD-1 than PD-L1^87,153. Both are expressed in lymphoid and non-lymphoid tissues, including tumors (Table 1). Pro-inflammatory cytokines induce PD-L1 and PD-L2 expression, potentially as a physiologic negative feedback mechanism to attenuate T cell responses, restore homeostasis and limit immune-mediated tissue damage, but enabling tumor cell evasion of the immune system. PD-L1 and PD-L2 each have second unique binding partners: PD-L1 with B7-1¹⁵³ and PD-L2 with RGMb¹⁵⁴. In this section, we discuss current understanding of these interactions.

The PD-L1–B7-1 interaction serves as a point of intersection between the CTLA-4–B7-1 and PD-L1–PD-1 inhibitory pathways (Figure 2). The orientation of the PD-L1 molecule and presence of a stalk domain allows it to establish an interaction with B7-1 that can only occur in cis on the same cell surface but not in trans across two different cells^155–158. Co-expression of B7-1 and PD-L1 within DC or tumor infiltrating myeloid cells supports a functional relevance of this interaction^155,156. PD-1 and B7-1 share a common binding region on the IgV domain of PD-L1 and can competitively inhibit each other’s binding to PD-L1^153,155. Using DC from knock-in mice with mutations in the IgV domains of PD-L1 or B7-1 that abrogated the cis-PD-L1–B7-1 interaction, Sugiura et al. showed these mutant DC promoted sustained PD-1 signaling. In turn, this led to diminished anti-tumor and self-reactive T cell responses. These findings indicate that the cis interaction of PD-L1 and B7-1 on APC blocks PD-L1 binding to PD-1 and thereby impairs inhibition of T cell responses¹⁵⁶. The similar affinities of PD-L1 for the trans interaction with PD-1 (Kd 0.77 mM) and the cis interaction of PD-L1 with B7-1 (Kd 1.4 mM) suggest a competition which may be dominated by the high local concentrations of PD-L1 and B7-1 on a cell surface¹⁵⁵.

Figure 2. Novel interactions among B7-1, PD-L1, CD28, CTLA-4, and PD-1.

B7-1 can form a back-to-back homodimer (interaction surface denoted as dark green on B7-1 IgV domain). A single B7-1 can interact with CD28 to deliver a co-stimulatory signal for T cell activation (interaction surface denoted as orange on B7-1 IgV domain). The B7-1 homodimer can interact with CTLA-4 in a multimeric fashion and CTLA-4 can remove and internalize the B7-1 by trogocytosis (interaction surface denoted as orange on B7-1 IgV domain), resulting in loss of B7-1 costimulatory activity. PD-L1 can interact with PD-1 (interaction surface denoted as light green on PD-L1 IgV domain), leading to phosphorylation of the PD-1 ITIM and ITSM tyrosines, recruitment and activation of SHP-2, and inhibition of T cell activation. B7-1 can interact with PD-L1 to form a heterodimer in which the B7-1 can still engage CD28 or CTLA-4 but the PD-L1 cannot engage PD-1. The CD28 signal remains active but the PD-1 signal is diminished. The CTLA-4 interaction is unimolecular since the B7-1 is monomeric but the CTLA-4 can still trogocytose the B7-1, resulting in loss of B7-1 costimulatory activity and free PD-L1 that can engage PD-1 to deliver an inhibitory signal.

B7-1 is a back-to-back, non-covalent homodimer and surprisingly, PD-L1 interacts with this dimer interface on B7-1, reducing B7-1 to a monomer^156–158. This still allows the interaction of B7-1 with CD28 and CTLA-4. Co-culture of T cells with APCs that co-express B7-1 and PD-L1 showed that the B7-1—CD28 trans interaction efficiently engaged and resulted in formation of CD28 microclusters on T cells¹⁵⁷. However, the functional consequences of PD-L1—B7-1 interactions on CTLA-4—B7-1 interactions remain an area of active investigation. Zhao¹⁵⁷ proposed that the PD-L1—B7-1 interaction, by reducing B7-1 to a monomer, would prevent the formation of a CTLA-4—B7-1 multimeric structure. Nevertheless, Tegkuc et al³⁶ show that CTLA-4 transendocytoses the B7-1 from a PD-L1—B7-1 heterodimer, thereby freeing PD-L1 to engage PD-1, increasing coinhibitory activity and also decreasing B7-1—CD28 signaling. Thus, reduced B7-1 on APCs has dual suppressive effects on T cells responses by limiting B7-1 co-stimulation and increasing free PD-L1 to interact with PD-1.

Whether PD-L1 interacts with B7-1 in cis on other cell types may also matter. A recent study suggests that membrane invagination can enable either B7-1 and B7-2 molecules expressed on CD8⁺ T cells to interact with CD28 in cis and in turn, this cis interaction can drive migration and cytokine production in vitro and mediate anti-tumor immunity¹⁵⁹. The authors posit that this may enable further cell survival in peripheral tissues where APCs are scarce. However, the frequency of this occurrence relative to a competing mechanism for cis-interactions of B7-1 with PD-L1 expression on a CD8+ T cell remains to be determined.

Anti-B7-1 antibodies that target the B7-1 interface and selectively block the PD-L1—B7-1 interaction but not the B7-1—CD28 or CTLA-4 interaction can reduce disease severity in the autoimmune mouses models of experimental autoimmune encephalitis (akin to multiple sclerosis), arthritis, and Sjogren’s¹⁶⁰. Anti-B7-1 antibodies that target the PD-L1 interface, would liberate PD-L1 to more effectively deliver PD-1-mediated inhibitory signals. In comparison, anti-PD-L1 antibodies that selectively block the PD-L1–B7-1 interaction but still allow the interaction with PD-1 accelerated type 1 diabetes in NOD mice, particularly self-reactive effector T cells^161,162, and impaired the induction and maintenance of oral tolerance¹⁶³. Thus, selective blocking of the B7-1 interface in the PD-L1—B7-1 interaction offers exciting therapeutic directions.

Of note, almost all PD-L1 therapeutic and phenotyping antibodies target the PD-1 binding region of PD-L1. Since this region is masked by the PD-L1—B7-1 interaction, this can result in the underestimation of PD-L1 levels in flow cytometry and immunohistochemistry.

Extracellular PD-L1 exists in both secreted and exosomal forms, with elevated circulating levels identified in malignancy, pregnancy, and infection¹⁶⁴. Secreted PD-L1 can result from splice variants or cleavage from the cell surface by metalloproteinases in tumor cells^165,166 or activated DC in an inflammatory milieu^166–168. Some splice variants of PD-L1 contain a stop codon before the transmembrane domain, leading to a secreted form with both monomer and covalent homodimers identified^165,166. The homodimeric form can inhibit T lymphocytes, as PD-1 inhibition requires multimeric crosslinking of PD-1 with the TCR. Another secreted form of PD-L1 shed through proteolytic cleavage by metalloproteinase has unknown function¹⁶⁸.

Exosomes are extracellular vesicles shed from eukaryotic cells that can communicate with the extracellular environment via their target protein, DNA, or RNA^169,170. Both tumor and myeloid cells may produce exosomal PD-L1^171,172. Exosomal PD-L1 may not always correlate with cell surface PD-L1^169,173. High IFN-γ can increase both secreted and exosomal PD-L1^164,166, both of which have been proposed to act as a decoy for anti-PD-L1 antibody by sequestering the anti-PD-L1 antibody or attenuating anti-tumor immunity through inhibition of T cell function^170,174,175.

High level of exosomal PD-L1 in head and neck cancer¹⁷⁶, gastric cancer¹⁷⁷, renal cell carcinoma¹⁶⁷ and soluble PD-L1 in DLBCL¹⁷⁸ was associated with worse prognosis, at least in part due to its association with larger tumors or more advanced disease. Notably, data on soluble PD-L1 as a biomarker for response to immunotherapy are mixed^{166,173,179,180}. The specific type of extracellular PD-L1 may matter, as exosomal PD-L1 stratifies patients with metastatic melanoma on pembrolizumab (anti-PD-1) therapy into responders or non-responders while the other secreted forms of PD-L1 do not¹⁷⁰. More work is needed to stratify extracellular PD-L1 forms as a therapeutic biomarker and whether in select cases, targeting exosomal secretion of PD-L1^174,176 is a viable therapeutic approach.

Nuclear expression of PD-L1 has been reported in multiple cancers including renal cell, hepatocellular, esophageal, lung, colorectal and prostate carcinoma^181,182 and following doxorubicin chemotherapy in breast cancer¹⁸³. PD-L1 may translocate from the plasma membrane to the nucleus, where it can bind DNA and regulate gene expression, and sister chromatid cohesion and segregation in cancer cells¹⁸⁴. Three mechanisms for this translocation from the cell surface to the nucleus have been proposed, including via the nuclear import protein KPNB1 in non-small lung carcinoma cells¹⁸⁵, interaction with vimentin and importin proteins to enter the nucleus following deacetylation of its C-terminus by HDAC2 and promotion of clathrin-dependent endocytosis¹⁸⁶, or hypoxia and gasdermin C through pSTAT3 interaction with the PD-L1 cytoplasmic domain¹⁸⁷. Du et al. showed that once in the nucleus, PD-L1 can interact with the transcription factor Sp1 to stimulate Gas6 transcription and cell proliferation¹⁸⁵. Gao et al. showed that nuclear PD-L1 can stimulate expression of NF-κB pathway and immunoregulatory genes including VISTA and B7-H3¹⁸⁶. Together, nuclear PD-L1 has the potential to regulate tumor cell proliferation and immune evasion in multiple ways.

The PD-L2–RGMb interaction serves as a point of intersection between the PD-L2–PD-1 pathway and RGMb–BMP2–BMP4–BMPR and RGMb–neogenin–netrin complexes^188,189 (Figure 3). RGMb is expressed on both hematopoietic and non-hematopoietic cells (Table 1). RGMb is a glycosylphosphatidylinositol (GPI)-anchored protein that can be expressed on the cell surface but is mainly localized intracellularly¹⁹⁰. In addition to PD-L2, RGMb binds to bone morphogenic proteins 2 and 4 (BMP) and neogenin. RGMb does not directly signal but serves as a co-receptor with BMP for signaling through type 1 and type II BMP receptors or alternatively a RGMb dimer can dimerize neogenin¹⁹¹. RGMb and PD-1 bind to PD-L2 with a similar binding affinity (Kd 48.5 for PD-1-PD-L2 and Kd 58.8 for RGMb-PD-L2) and on distinct sites of PD-L2¹⁵⁴. RGMb can simultaneously bind to PD-L2 and BMP, as they recognize distinct binding sites on RGMb¹⁵⁴. Thus, PD-L2 may be able to engage in a tripartite complex binding both PD-1 and RGMb, as well as the RGMb supercomplex in trans to regulate signaling downstream. When on the same cell, RGMb can bring BMP and neogenin together wherein BMP acts as a bridge between two neogenin and RGMb dimers¹⁹¹ and activates signals downstream. RGMb also can be cleaved from the cell surface by phospholipases, resulting in a soluble form. RGMb, when shed from the cell surface, can bind to neogenin and facilitate neogenin dimerization. In addition, RGMb can bind to the soluble monomeric form of CTLA-4, but not the dimeric membrane form of CTLA-4. RGMb increases binding of soluble CTLA-4 to B7-1, enabling it to block B7-1 interaction with CD28 more effectively¹⁹². How this sequestration of B7-1 alters known interactions with PD-L1 remains to be investigated.

Figure 3: Novel interactions of PD-L2 with PD-1 or the BMPR–BMP2/4–RGMb–neogenin complex.

PD-L2 can interact with PD-1, leading to phosphorylation of the PD-1 ITIM and ITSM tyrosines, recruitment and activation of SHP-2, and inhibition of T cell activation. RGMb serves as a co-receptor for BMPs for the type I and type II BMP receptors. The BMPR–BMP2/4–RGMb interaction results in phosphorylation of the BMP receptors and activation of the MAPK, ERK, and SMAD1/5/8 pathways. An RGMb dimer can bridge two Neogenin molecules, forming a RGMb-neogenin complex. RGMb bridging of neogenin engages Rho pathway signaling and cytoskeletal rearrangement. PD-L2 can also interact with RGMb in the BMPR-BMP2-RGMb-neogenin supercomplex and this interaction results in respiratory and mucosal tolerance.

Although RGMb was initially identified in the context of neuronal cell differentiation during brain development, its immunoregulatory roles are now appreciated. RGMb-BMP interactions inhibit IL-6 secretion by macrophages via p38 and MAPK signaling¹⁹³. In models of airway hyperreactivity, RGMb interactions with either neogenin or PD-L2 regulate respiratory tolerance. Blockade of PD-L2 interactions on lung DC with RGMb⁺ macrophages and eosinophils using an anti-PD-L2 mAb that selectively inhibits the PD-L2-RGMb interaction (but not PD-1—PD-L2) abrogated respiratory tolerance induced by exposure to ovalbumin prior to T cell priming¹⁵⁴. Likewise, a soluble mutated form of PD-L2 capable of only binding to RGMb but not PD-1 reduced airway hyperreactivity and lung inflammation¹⁹⁴. However, a separate role for RGMb-neogenin interactions independent of PD-L2 has also been shown. Anti-RGMb mAb inhibited the expansion of RGMb⁺ interstitial macrophages and eosinophils and reduced IL-25 production from RGMb⁺ macrophages and eosinophils in a neogenin-dependent and PD-L2 - independent manner¹⁹⁰. These findings identify RGMb—neogenin interaction as an attractive therapeutic target for IL-25 driven diseases, including allergen or rhinovirus-induced airway hyperreactivity. Furthermore, the expression of neogenin on RGMb⁺ macrophages and surrounding epithelium suggests that interactions between RGMb and neogenin may occur in both cis and trans.

RGMb is also expressed in gut epithelium. An anti-RGMb antibody that blocked RGMb interaction with BMP2/4 and neogenin reduced GVHD and colitis¹⁹⁵. In contrast, blockade of the RGMb binding site on PD-L2 worsened GVHD. RGMb deficient mice showed increased severity of colitis¹⁹⁶, perhaps due to the elimination of RGMb interaction with its binding partners. In addition, the gut microbiota influences PD-L2 and RGMb expression. Blockade of PD-L2—RGMb interactions can overcome microbiota-dependent resistance to anti-PD-L1 therapy in tumor-bearing germ-free, antibiotic treated mice, or mice reconstituted with microbiota from patients who did not respond to anti-PD-1 treatment. Intriguingly, this study identified a novel function for RGMb on T cells and showed that interaction in trans between PD-L2 on APC and RGMb on T cells inhibited CD8⁺ T function¹⁹⁷. Altogether, these findings define mechanisms by which RGMb regulates gut inflammation or anti-tumor immunity through interaction with ligands on hematopoietic and non-hematopoietic cells. Further work is needed to understand the functional consequences of RGMb interactions with its respective binding partners and how to optimize strategies for designing RGMb therapeutics.

We next describe other B7 molecules that have a co-stimulatory or co-inhibitory function (Table 1). Here we focus on HHLA2, VISTA, BTNL2 and BTN3A1 and refer the reader to recent reviews on ICOSL¹⁹⁸, B7-H3¹⁹⁹ , and B7-H4²⁰⁰.

HHLA2

HHLA2 (Human endogenous retrovirus-H long terminal repeat-associating 2) is longer than other B7 family members having the usual IgV-IgC but followed by an extra IgC and transmembrane and cytoplasmic domains²⁰¹. HHLA2 and its ligand orthologs are present in humans, monkeys, and many other mammals, but not in rodents^202,203. HHLA2 is found in various tissues, cancer cells and immune cells (Table 1)^201–205. PD-L1 and HHLA2 are differentially regulated, with interferons and hypoxia being important for PD-L1 upregulation, but not HHLA2. A moderate percentage of lung, renal, cholangiocarcinoma, and prostate tumors express PD-L1 but a higher percentage express HHLA2. Few tumors co-express both, suggesting non-redundant pathways of immune evasion^205–210. HHLA2 deficiency inhibits tumorigenic pathways such as EGFR/MAPK/ERK pathway in NSCLC²¹¹ and JAK/STAT in hepatocellular cancer²¹². HHLA2 has been defined as a prognostic marker in multiple cancers^213–216.

HHLA2 can deliver a dominant inhibitory signal that impedes proliferation and cytokine secretion in activated T cells^201,217 through KIR3DL3 (killer cell immunoglobulin-like receptor, three immunoglobulin domains and long cytoplasmic tail 3)^209,218. KIR3DL3 contains three IgV domains, a transmembrane domain, and a cytoplasmic tail with a single ITIM motif. The KIR3DL3-HHLA2 pathway delivers a co-inhibitory signal to NK cells and T cells. KIR3DL3⁺ CD8⁺ T cells and KIR3DL3⁺ NK cells had reduced cytotoxicity for HHLA2⁺ targets^209,218. Recruitment of SHP1 and SHP2 to the KIR3DL3 ITIM motif mediated the inhibitory effect by reducing ERK1/2 protein kinase and NF-κB signaling²¹⁸. Antibody blockade of KIR3DL3 on NK cells reduced tumor burden in a lung metastasis model²¹⁸.

TMIGD2 (Transmembrane and Immunoglobulin Domain Containing 2), a member of the CD28 family, is the co-stimulatory receptor for HHLA2^202,203. TMIGD2 has a single IgV, transmembrane, and cytoplasmic domain but lacks the juxta-membrane disulfide that dimerizes CD28 and CTLA-4²⁰³. TMIGD2 is expressed on naive T cells and many NK cells but expression declines on activated and memory T cells²⁰². Engagement of TMIGD2 by HHLA2 transmits a costimulatory signal to T²⁰² and NK cells²¹⁹ via AKT-dependent signaling. Further work is needed to understand contexts when TMIGD2 costimulation and KIR3DL3 coinhibition are most important.

VISTA

VISTA contains a single large IgV domain with homology to PD-L1 but no IgC domain. VISTA has several unique features including an extended C-C’ loop and an extra Beta-strand, H, with a disulfide bond that clamps the structure so the histidine-rich CDR-like loops are positioned facing sideways rather than the more usual upper position^220,221. The cytoplasmic domain of VISTA contains multiple casein kinase 2 and phospho-kinase C sites but no tyrosine signaling motifs, suggesting that it can signal into VISTA expressing cells²²². VISTA is most abundantly expressed on myeloid and DC, but relatively less on T cells (Table 1).

The abundance of histidine residues in its extracellular domain allows VISTA to have pH-dependent interactions with its ligands, since the histidines become charged as the extracellular pH declines from pH 7.4 to pH 6. Such a low pH is found in the tumor microenvironment, areas of the lymph node, and in healing wounds²²¹. VISTA has five reported ligands, PSGL-1, Syndecan-2, LRIG-1 , VSIG8, and VSIG3, which have a ratio of binding at pH 6.0 to pH 7.4 of 125, 78, 9, 9, and 3, respectively^223–227. PSGL-1 is expressed on T and B cells, myeloid cells, and dendritic cells²²⁸. PSGL-1 signaling can inhibit T cell proliferation and push them towards an exhausted phenotype²²⁹, although the interaction between VISTA and PSGL-1 was not explicitly shown to mediate this effect. VSIG3-VISTA and VSIG8-VISTA interactions inhibit T cell activation and effector function^223,225.

VISTA can function as both ligand or receptor. VISTA is expressed at high levels in naive T cells to enforce quiescence and tolerance²³⁰. Expression of VISTA declines on activated T cell populations at later time points²³¹. Higher expression of VISTA is found on tumor infiltrating macrophages, MDSC^232–236 and tumor cells²³⁰. In synovial sarcoma, secretory factors from cancer cells such as VEGF can upregulate VISTA on endothelial cells²³⁷.

VISTA fusion protein inhibits proliferation and cytokine production by activated T cells during early stages of activation. Mice lacking VISTA develop arthritis and lupus-like autoimmune disease, due its function in tolerance^{231,235,238,239}. VISTA induces Treg formation from human CD4⁺ T cells²⁴⁰. VISTA promotes suppression in myeloid cells and tolerogenic DC through inhibition of MAPK and NF-κB pathways upon TLR signaling²⁴¹. In an imiquimod (IMQ)-induced mouse model of psoriasis, VISTA deficiency exacerbated disease due to increased IL-17 and IL-23 from DC²⁴². In addition, VISTA agonist antibodies ameliorated autoimmunity, asthma, and GVHD in mouse models^{222,239,243,244}.

Early studies with rat anti-mouse VISTA antibodies combined with anti PD-1 or anti-PD-L1 showed efficacy in multiple mouse tumor models^233–236. Combined blockade resulted in increased inflammatory signatures in myeloid derived suppressor cells²⁴¹, reduced differentiation to Tregs and increased T cell effector function^234–236. Newer VISTA antibodies that block the interaction with PSGL-1 in a pH-selective fashion at pH 6.0 but not pH 7.4 had an added benefit with anti-PD-1 in mouse tumor models^226,240,245. Such antibodies are concentrated in the acidic tumor microenvironment and not in major VISTA-expressing sites like the spleen. These pH-selective VISTA antibodies had an improved safety profile relative to non-pH selective antibodies in non-human primates²²⁷. Clinical development of pH-selective VISTA and other antibodies offers a novel strategy in cancer.

BTN3A1 and BTN2A1

BTN3A1 and BTN2A1 are members of the B7-like family of butyrophilins. Like other members of the butyrophilin family, they contain one IgV domain, one IgC domain, a transmembrane domain, a juxta transmembrane domain and a cytoplasmic domain containing a B30.2 (PRYSPRY) domain. BTN3A1 has three isoforms and is expressed on immune cells, tumor cells and endothelial cells^246,247 (Table 1). BTN3A1 and BTN2A1 are a ligand for the γ9Vδ2 TCR in human γδ T cells. This interaction triggers activation of these T cells in response to transformed or stressed target cells²⁴⁸. Activated γ9Vδ2 T cells are cytotoxic and critical in immune responses in cancer, tuberculosis, or other diseases in an MHC-independent manner^249,250. The B30.2 domains of BTN3A1 and BTN2A1 comprise an intracellular binding pocket for phosphorylated intermediates of the mevalonate pathway such as isopentyl pyrophosphate^251,252. The mevalonate pathway feeds into cholesterol and other biosynthetic pathways and is overactive in cancer cells. This interaction is mediated through positively-charged amino acids in the B30.2 domain and negatively-charged amino acids in these phosphoantigens²⁵³. A recent study showed that these phosphoantigens “glue” B30.2 domains of BTN3A1 and BTN2A1 together, resulting in a conformational change that transmits an “inside-out” signal to the extracellular domains of the butyrophilins allowing them to engage the γ9Vδ2 TCR^254,255 . Activation of the AMPK pathway can upregulate BTN3A1 and BTN2A1 cell surface expression in tumor cells, increasing γ9Vδ2 T cell cytotoxicity²⁵⁶. ICT101, a humanized agonistic antibody to BTN3A1/2/3 is in clinical trials in combination with anti-PD-1 for treatment of solid and hematologic malignancies²⁵⁷. Small molecule analogs of these phosphoantigens also have been developed for therapeutic use^254,256.

BTNL2

BTNL2 belongs to the butyrophilin family and contains 4 Ig domains in an IgV-IgC-IgV-IgC arrangement followed by transmembrane and cytoplasmic domains^258,259. Unlike other butyrophilin family members, BTNL2 does not contain a B30.2 domain in its cytoplasmic domain²⁶⁰. BTNL2 is expressed in the gut particularly on small intestinal epithelial cells as well as immune cells (Figure 1)^258,259,261. The receptor of BTNL2 is not known, but studies with BTNL2-Fc inferred its expression on activated T cells and B cells²⁵⁹ (Table 1). BTNL2-Fc inhibited T cell proliferation, IL-2 production^258,259,261 and promoted induced Treg development²⁶². BTNL2 polymorphisms have been reported and confirmed to be associated with multiple autoimmune diseases including inflammatory bowel disease^258,259, sarcoidosis^263,264, rheumatoid arthritis²⁶⁵ and colon or prostate cancer^266,267. Mice reconstituted with BTNL2 deficient bone marrow have lower survival in a mouse model of cerebral malaria²⁶⁸. BTNL2 has an important role in maintaining intestinal homeostasis. BTNL2 deficient mice had severe colitis as BTNL2 regulates the response of γδ T cells in intestinal epithelial lymphocytes²⁶⁹. BTNL2-Fc promotes secretion of IL-17 from γδ T cells or IL-22 from γδ T cells, CD4 and ILCs in the colon²⁷⁰. BTNL2 monoclonal antibody reduced tumor growth in mouse models of lung cancer, colon cancer and lymphoma²⁶¹. In contrast, BTNL2-Fc imparted a protective effect in colitis while BTNL2 antibody reduced tumor growth in colitis induced colon tumorigenesis²⁷⁰. Higher levels of BTNL2 mRNA expression were associated with worse survival in patients with lung and colon adenocarcinoma²⁶¹. While human BTNL2 has been detected with polyclonal antibodies in western blots and immunohistochemistry, no human mAbs have been reported, perhaps due to the difficulty of expressing human BTNL2, whose amino terminal sequence differs greatly from mouse.

Outstanding questions and concluding remarks

In summary, coinhibitory pathways in the B7—CD28 family play a crucial role in maintaining immune balance, protecting against infection, and preventing autoimmunity. These pathways not only regulate the activation of naïve T cells, but also control T cell fate decisions and function of effector, memory, and regulatory T cells. Coinhibitory signals are essential for tolerance and resolving inflammation, but their overactivity can contribute to the immunosuppressive environment found in tumors and chronic viral infections. Although our understanding of the immunoregulatory roles of coinhibitory pathways in the B7—CD28 family has advanced, these insights raise questions of fundamental and translational importance:

First, what are the ramifications of PD-L1—B7-1 and PD-L2—RGMb interactions? PD-1 and its ligands comprise a network rather than a linear pathway. PD-L1 bridges to the B7-1/B7-2/CD28 pathway and PD-L2 bridges to the RGMb/BMP2/BMP4/neogenin pathway. This knowledge opens new mechanistic insights into the efficacy of current immunotherapies as well as opportunities for designing new treatment strategies for cancer, autoimmunity, allergies and transplant rejection.

Second, how can the multifaceted roles of coinhibitory receptors in regulating different types of T cells be exploited for therapy? For example, PD-1 reinvigorates exhausted CD8⁺ T cells but can increase numbers and suppressive function of regulatory T cells. PD-1 bispecific antibodies that selectively target effector T cells may be more effective. Combination therapies that block PD-1 and deplete or attenuate Tregs also may enhance efficacy of PD-1 pathway blockade.

Third, what are the shared and unique molecular pathways triggered by coinhibitory receptors? Does co-expression of two or more receptors on T cells induce distinct signaling pathways compared to cells expressing one receptor? Transcriptional, proteomic and lipidomic approaches provide means to identify overlapping and distinct signaling nodes. Spatial technologies may localize receptor ligand interactions in specific tissues during cancer, autoimmunity and infections. Such studies could identify therapeutic synergy resulting from co-blockade on the same cell or different cells and inform synthetic biology strategies for targeting specific cellular interactions.

Fourth, what are the functions of coinhibitory receptors on non-T cells? The expression of these receptors on innate as well as adaptive immune cells points to a broader immunoregulatory role.

Fifth, the butyrophilin subfamily of the B7s has been perplexing but the recent finding that the B30.2 intracellular domain of the BTN3A1 butyrophilin binds intracellular ligands and transmits an “inside-out” signal to engage γδ cells reveals a new mechanism of immunoregulation. Do other intracellular ligands exist for butyrophilins and is this part of a broader form of regulation in other disease contexts?

The therapeutic potential of deeper understanding of coinhibitory pathways is underscored by FDA approval of CTLA-4, PD-1 and PD-L1 antibodies for cancer immunotherapy and recent development of PD-1 agonistic antibodies to treat autoimmune diseases. The answers to these questions will continue to illuminate our understanding of coinhibitory pathways in health and disease, paving the way for novel therapeutic strategies to effectively and safely treat a variety of immune-mediated disorders.