Introduction

B cells, along with T cells, form the core of the adaptive arm of the immune system. They are generated continually and throughout the life of the organism in the bone marrow, a primary lymphoid tissue, from haematopoietic stem cell progenitors that progress through sequential developmental steps. Each developing B cell expresses a unique B cell receptor (BCR), which is composed of two identical heavy chain proteins and two identical light chain proteins. In both humans and rodents, once a developing B cell expresses a correctly assembled BCR on its cell surface, and that BCR has been confirmed to not be autoreactive, the IgM+ immature B cell exits the bone marrow, enters the blood and migrates to the spleen.

Within the spleen, the architecture of which is somewhat different in mice versus humans1, immature B cells undergo additional developmental steps (Box 1; Fig. 1). The immature B cell is then considered a naive B cell (NBC), which has the capacity to be primed against proteins derived from foreign pathogens (that is, antigens). With a half-life of approximately 6 weeks, the NBCs circulate in the blood and enter lymph node organs through high endothelial venules. Unless they are activated by an antigen, B cells will spend approximately 24 hours in follicles, which are structures found in the lymph node cortex. They are drawn to these structures because NBCs express CXC-chemokine receptor 5 (CXCR5), which binds to its ligand, CXC-chemokine ligand 13 (CXCL13); CXCL13 is expressed in the follicle. Following this brief residence in the follicle, NBCs follow the gradient of a different molecule, sphingosine 1-phosphate, to the medullary cords and out through the efferent lymphatics. After transitioning through the thoracic duct, NBCs re-enter the blood and repeat this journey2.

Fig. 1: B cell biology.
figure 1

Following their development in the bone marrow, immature B cells enter the circulation and complete their development in the spleen. In the spleen, they first become naive B cells (NBCs). If activated by an antigen (Ag), NBCs may then enter an emerging germinal centre (GC), where affinity maturation occurs. The highest-affinity B cells emerge from the GC as memory B cells or plasma cells. Some NBCs can also be activated in extrafollicular environments, for example in the T cell area or in the medullary cords. NBCs activated in extrafollicular environments quickly become plasma cells. In the presence of IFNγ, B cells also can also acquire the expression of T-bet, and some T-bet+ cells will remain in the spleen as memory B cells. Some T-bet+ cells can adopt a double-negative 1 (DN1) or DN2 phenotype, with the latter being influenced by cytokines such as IL-21. CXCR, CXC-chemokine receptor; Ig, immunoglobulin; TFH cell, T follicular helper cell.

Because of the remarkable diversity of the BCR repertoire, B cells can respond to a seemingly infinite number of antigens. Immune responses to antigens occur within secondary lymphoid organs (lymph nodes, tonsils and spleen): the spleen supports immune responses to blood-borne antigens, whereas lymph nodes sample tissue-derived antigens3. Binding of an antigen to the BCR results in intracellular events such as calcium mobilization and protein phosphorylation4.

Depending on the molecular structure of the antigen, NBCs may or may not need additional help from T cells to fully respond to the antigen. To induce this help, B cells that have been activated by particular protein antigens form cognate interactions with T cells that are specific for a linear peptide fragment of the same antigen. These cognate B cell–T cell interactions occur in the germinal centres (GCs) in secondary lymphoid organs. Within the GCs, B cells with a range of affinities for the antigen compete for help from these cognate T cells, which express key costimulatory proteins, such as CD40 ligand (CD40L), and also make cytokines that direct the B cell response.

To increase their affinity for antigens, B cells also undergo secondary BCR diversification. The BCR-encoding genes accumulate mutations in their antigen-binding regions through the activity of activation-induced cytidine deaminase (AID). This process results in multiple variants of the original germ line BCR, and the B cells that emerge from this mutagenesis programme with the highest affinity for the antigen will ‘win’ the GC reaction competition and become dominant clones in the immune response5. Such B cells can also undergo class switch recombination to generate B cells that produce other classes of antibodies. A proportion of these B cells (switched and unswitched) become memory B cells (MBCs) or plasma cells (PCs), which secrete copious amounts of antibodies. MBCs and long-lived PCs (LLPCs) provide protection against reinfection by the same pathogen6.

Due to their vigorous proliferation and the accompanying mutagenesis programme, GC B cells can be vulnerable to cancerous transformation. For this reason, reagents that bind surface glycoproteins expressed by B cells such as CD20 have been successfully used to target B cell lymphomas for depletion, removing both cancerous and non-cancerous CD20+ B cells from the patient7. Given the excellent safety profile of many B cell depletion therapies (BCDTs) in cancer8, these drugs were reimagined for use in autoimmune diseases in which B cells play a direct or indirect role.

We begin this Review with select recent advances in B cell biology that are relevant to the mechanism of action of BCDT in autoimmune disease. Then we outline the clinical results from disease settings in which BCDTs have been used extensively — such as systemic lupus erythematosus (SLE), rheumatoid arthritis (RA) and multiple sclerosis (MS) — and discuss some emerging indications that may benefit from BCDT. Lastly, we discuss the potential mechanism of action of BCDT, possible reasons why BCDT has failed in some indications, and some new therapeutic directions for BCDT based on recent advances in B cell biology.

Recent advances in B cell biology

To understand how BCDT works in autoimmune disease, and to design superior B cell-targeted therapies, we must first understand some fundamental aspects of B cell biology. The topic of how BCDT impacts autoimmune disease was last reviewed in this journal in 2006 (ref.9); much has changed since then. Here, we note some highlights from the literature that are relevant to the use of BCDT in treating autoimmune disease.

Activation of B cells

The GC is an important microenvironment that supports the interaction of antigen-specific B and T cells. GCs are readily observed in histological sections of secondary lymphoid organs (tonsils, spleen and lymph nodes) in mice and humans. Somatic hypermutation mediated by AID may increase the affinity of the BCR, but may also reduce its affinity, or even generate an affinity for self-antigen. Due to its anatomical organization (Box 2), the GC environment imposes a framework of competition among GC-resident B cells such that those B cells with the highest affinity for an antigen persist and become MBCs and PCs. Advances in our understanding of this GC reaction may inform the design of specific therapies that interfere with this specialized microenvironment. For example, in the past 10 years, our understanding of how T follicular helper cells (TFH cells) and regulatory TFH cells engage with B cells has exploded10. In mouse models, entanglement of GC-resident TFH cells with GC B cells promotes sustained interactions that select for B cells with the highest affinity for an antigen through a process of feedforward loops involving CD40L, inducible T cell costimulator (ICOS) and BAFF (also known as TNFSF13B)11. This entanglement is reminiscent of the types of CD40–CD40L-driven B cell–T cell interactions that occur in human tonsils12. Although beyond the scope of this Review, removing or disabling TFH cells to prevent the GC reaction is a possible strategy for treating autoimmune disease.

The GC is not the only environment in which B cells can participate in the immune response. Extrafollicular B cell responses provide an early source of antibodies during infection13. In mouse models, antigen-specific B cells destined for the extrafollicular response maintain high levels of EBI2 (also known as GPR183) expression, which keeps these cells out of the GC microenvironment, and upregulate CXCR4, which allows them to localize to the medullary cords of the lymph nodes14,15. Although initially considered to be T cell-independent B cell responses, we now know that extrafollicular responses recruit TFH cells16. Moreover, the extrafollicular response can foster somatic hypermutation, as has been shown in infection models such as Salmonella enterica subsp. enterica serovar Typhimurium infection, in which GC responses are rare17. In addition to these mouse studies, extrafollicular class switching, driven by the cytokines BAFF and APRIL (also known as TNFSF13), has been documented in human tonsils and gut18,19. Extrafollicular B cell responses have also been observed in the synovium and salivary glands of patients with RA or Sjögren syndrome20,21,22,23. The extrafollicular pathway of B cell activation is also an early and potent source of PCs because BLIMP1 (also known as PRDM1) is rapidly upregulated in extrafollicular B cells. When expressed in B cells, BLIMP1 represses the expression of genes involved in BCR signalling and GC B cell function, so cells adopt a PC phenotype24.

Extrafollicular responses have implications for autoimmune diseases. Indeed, in MRL/FASlpr mice, a model for SLE, autoreactive plasmablasts (PBs) are generated not within the follicle or the GC but rather in T cell-rich areas of lymphoid tissues25,26. Moreover, in other lupus-prone mouse models, autoreactive antibody-producing B cells can be generated in extrafollicular regions independently of T cells; B cell survival factors such as BAFF and pattern recognition receptors such as Toll-like receptor 7 (TLR7) play important roles in the generation of these B cells27,28. Although extrafollicular responses clearly produce class-switched B cells, in contrast, and in keeping with older findings in the literature, GCs are dominated by non-class-switched IgM+ B cells in mice and humans29. Since class-switched antibodies are critical in the pathogenesis of diseases such as SLE, this observation calls into question the utility of targeting the GC itself as a therapeutic strategy.

In humans, it is difficult to study the extrafollicular response because tissue is not readily available. However, some inferences may be made from clinical observations. For example, so-called monocytoid B cells, which express CD19 but have a monocyte-like morphology, have been found in lymph node sinusoidal spaces, and the expression of AID in these cells suggests that an extrafollicular response may be occurring30. Moreover, B cells associated with the tonsillar epithelium can undergo class switch to IgG and IgA and undergo somatic hypermutation in response to locally produced cytokines, again suggesting an extrafollicular response18. For the most part, however, extrafollicular responses in humans are inferred because some circulating MBCs are CXCR5 and have relatively low levels of somatic hypermutation of the BCR, suggesting that they matured outside the GC31.

From these findings taken together, although we traditionally think of GCs as a hotbed for autoimmune B cell genesis, targeting extrafollicular responses may also have significant utility in treating B cell-mediated autoimmune diseases.

T-bet+ B cells

Multiple B cell subsets have been described in the past decade. Of these, B cells that express T-bet (also known as TBX21) may play a particularly important role in autoimmune diseases. T-bet is a transcription factor that historically has been considered a hallmark of IFNγ-producing T helper 1 cells. However, T-bet can also be expressed in B cells32, and even in dendritic cells33, in both mice and humans. T-bet+ B cells are commonly referred to as double-negative (DN) B cells, as they express neither CD27 (also known as TNFRSF7) nor IgD. Although DN B cells are CD27, and CD27 usually marks MBCs, DN B cells are considered to be antigen-experienced34 (Box 3).

DN B cells are further divided into DN1 and DN2 subsets. B cells that express high levels of T-bet (T-bethi B cells) can be directly generated from NBC precursors in response to IFNγ (which is locally produced by T cells)35, and consequently become ‘activated naive’ or DN1 B cells36. In addition, some T-bethi B cells have the capacity to quickly differentiate into PCs in an IL-21-dependent manner in patients with SLE37. These B cells are DN2 cells. Importantly, DN2 cells also lack CXCR5, and thus extrafollicular responses may have contributed to their generation38. As discussed in subsequent sections, T-bethi DN1 and DN2 cells are enriched in patients with SLE. Because of their propensity to rapidly differentiate into antibody-secreting cells in response to cytokines, T-bethi DN2 cells could be a source of surges of autoantibodies, whether directly pathogenic or a reflection of generalized B cell activation.

Levels of T-bet can also be used to distinguish MBCs that recirculate (T-betlow or T-bet) from those that are spleen resident and excluded from lymphatics (T-bethi) in mice and humans. Both subsets arise from antigen-activated NBCs, but subsequently diverge39. Of note, in mice, T-bethi MBCs accumulate with age40,41 and in autoimmune diseases such as MS42, and could therefore contribute to disease chronicity, particularly in aged patients. Although more research is required to understand whether T-bethi MBCs have a propensity to promote autoimmune disease pathology, they deserve some consideration in particular because many T-bethi MBCs do not circulate. Hence, we do not know whether they are depleted by anti-CD20 or anti-CD19 (ref.39), although the few that do circulate are sensitive to BAFF inhibition43. Thus, the very large population of MBCs that reside within the spleen (and possibly other tissues, as has been observed in chronic infections) may not be fully depleted, especially if they reside in vasculature-poor regions44. Such spleen-resident MBCs may be especially pathogenic in autoimmune conditions, particularly in aged individuals.

Antibody-independent B cell functions

Patients who positively respond to BCDT do not necessarily show a corresponding decrease in autoantibody levels. This effect has been observed in SLE45, MS46 and RA47,48. In such cases, BCDT must therefore alter autoimmune diseases in a manner that is independent of antibody production. In this section, we provide evidence that B cells can be a potent source of cytokines that may influence autoimmune pathology. However, B cells can also act as antigen-presenting cells, and this has been intensively studied in the context of MS.

Most of our evidence that B cells can make cytokines comes from animal models in which gain-of-function and loss-of-function approaches have been used to query whether B cells are necessary and/or sufficient for providing a particular cytokine. One antibody-independent function of B cells is to initiate and/or sustain tertiary lymphoid tissues (TLTs) in inflamed tissues. In mice, TLTs have been observed in a number of autoimmune disease settings and in some cases can be driven by the membrane-bound expression of lymphotoxin-αβ (LTαβ) by B cells49, although other cell types and other cytokines (IL-17 and IL-22) can also provide this signal.

Beyond their function in TLT formation, B cells produce numerous soluble cytokines. For example, B cells can be an important source of IL-2 (refs50,51), TNF and CCL3 (ref.52) in parasite infections. In addition, although IFNγ is typically associated with T helper 1 cell responses, this cytokine can also be produced by B cells during bacterial infection53,54,55,56. It is not entirely clear how B cells are induced to make IFNγ. It may be tied to downregulation of the expression of the MHC class I receptor LY49G2 on B cells57, or in response to cytokines such as IL-27, IL-12 and IL-18 (ref.58). In addition to IFNγ, the proinflammatory cytokines IL-6 and granulocyte–macrophage colony-stimulating factor (GM-CSF) are produced by B cells during neuroinflammation in mice, as well as in humans59,60.

PCs are also important cytokine-producing B cells. For instance, innate response activator (IRA) B cells have been described in mice: these cells have a phenotype that is similar to PCs expressing syndecan 1 (also known as CD138), and IRA B cells can produce GM-CSF61. Moreover IL-17, which is typically produced by T helper 17 cells, can also be produced by both mouse and human PBs and PCs62. In mice, PBs and PCs that produce GM-CSF are thought to be derived from peritoneal cavity B-1 cells (Box 1) that migrate to the spleen, where they produce IgM. In contrast, IL-17 is produced directly by PCs in the spleen upon infection with Trypanosoma cruzi61,62. IgA-producing PCs in the mouse intestine can express either TNF or inducible nitrogen oxide synthase (iNOS), and this expression is required for resolution of Citrobacter rodentium infection63. PC-derived iNOS, which is induced by the unfolded protein response and contributes to PC survival64,65, also plays a role in controlling Helicobacter pylori infection in humans66. Lastly, by isolation of intestinal PCs for ex vivo flow cytometry analysis and by immunofluorescence analysis on gut tissue from patients with inflammatory bowel disease, PCs in the intestine have also been shown to produce tissue-damaging molecules such as matrix metalloproteinases and granzyme B in response to inflammation67,68.

In summary, these examples of alternative B cell functions beyond antibody production, including cytokine production, provide a potential explanation for the surprising efficacy of BCDT in autoimmune disease.

Regulatory B cells

B cells also have the capacity to quiet inflammation, and IL-10, which is produced by subsets of myeloid cells, B cells and T cells, clearly mediates much (but likely not all) of this anti-inflammatory effect in mice69. Again we have learned about ‘unusual’ B cell behaviour by studying parasitic infections in mice — in this context, IL-10 production by B cells can limit allergic responses70,71, and B cells can perform this function independently of regulatory T cells9,70. In mice, regulatory B cells (which are CD1d+CD5+) have been shown to produce IL-10 (ref.72) in response to both CD40L and IL-21 provided by T cells73,74. IL-10-producing regulatory B cells have also been described in humans66. Depending on the sequence and/or nature of the activation signal, human B cells can produce IL-10 or, alternatively, secrete proinflammatory cytokines such as soluble LTα3, TNF and IL-6 (ref.75). TLR signalling on B cells may also elicit different cytokine responses, for example producing either the anti-inflammatory IL-10 or the proinflammatory IL-13, depending on the type of TLR and the maturation stage of the B cell (NBC vs MBC)76. Importantly, the ability to produce IL-10 by regulatory B cells in response to CD40 stimulation can be lost in regulatory B cells, as has been observed in patients with SLE77.

Anti-inflammatory cytokines can also be produced by PCs. This was clearly demonstrated in several mouse models: PCs producing IL-10 and IL-35 can attenuate neuroinflammation in experimental autoimmune encephalomyelitis (EAE)78. PC-dependent IL-10 production during EAE was further confirmed in a separate study in mice and humans in vitro79, and the PC-specific transcription factor interferon regulatory factor 4 (IRF4) was shown to induce IL-10 expression80. Our laboratory has shown that mucosally derived IgA-producing PCs are a potent source of IL-10, and that they can express IL-10 within the inflamed brain and spinal cord during EAE81. Because the microbiota has been implicated in the production of IL-10 by mucosal B cells82, this suggests that the dysbiosis which occurs during some chronic autoimmune diseases may impinge on the ability of B cell lineage cells to suppress inflammation. At the very least, the concept that ‘regulatory PCs’83 might exist must be carefully considered when one is interpreting BCDTs that differentially target PCs. Importantly, in humans, IL10-producing PCs can potentially be localized in the diseased tissue itself84.

Systemic autoimmune diseases

Our main BCDT tool is anti-CD20 antibodies — this therapy has been extensively tested in numerous autoimmune disease settings (Fig. 2). However anti-CD19 and anti-BAFF antibodies have also been tested in autoimmune disease indications. These agents, as well as new therapies at various stages of preclinical and clinical development, are described here in the context of systemic autoimmune diseases (Table 1).

Fig. 2: Roles of B cell lineage cells in autoimmune disorders.
figure 2

B cells and their effectors such as antibodies and cytokines differ in their contributions to pathobiology depending on the disease. Seven autoimmune disorders (multiple sclerosis, N-methyl-d-aspartate receptor (NMDAR) encephalitis, myasthenia gravis, systemic lupus erythematosus (SLE), rheumatoid arthritis, myelin–oligodendrocyte glycoprotein (MOG) spectrum disorder (MOGSD) and neuromyelitis optica spectrum disorder (NMOSD)), their targets, and the implicated B cell subtype or pathogenic autoantibodies are shown. Based on the vast options for how B cells can influence pathobiology, patients must be carefully assessed to ensure that B cell depletion therapy reduces pathogenic but not beneficial B cell subsets. ACh, acetylcholine; AChR, nicotinic acetylcholine receptor; ACP, anti-citrullinated protein; AQP4, aquaporin 4; dsDNA, double-stranded DNA; GM-CSF, granulocyte–macrophage colony-stimulating factor; MuSK, muscle-specific tyrosine kinase; RF, rheumatoid factor; TLT, tertiary lymphoid tissue; TNF, tumour necrosis factor.

Table 1 Selection of BCDTs being tested or currently approved for autoimmune diseases

BCDT in SLE

SLE has been a challenging disease to treat effectively, largely because it is heterogeneous, and this has become even more apparent in the bioinformatics era85. Indeed, there is only one approved biologic for SLE: belimumab, a fully humanized anti-BAFF monoclonal antibody (mAb) that blocks the binding of BAFF to its cognate BAFF receptors on B cells. Unlike diseases, such as neuromyelitis optica (NMO), that are driven by autoantibodies to surface antigens, SLE is characterized by class-switched antibodies to intracellular antigens, particularly nuclear antigens86. The pathogenicity of these antibodies is because of their propensity to form immune complexes that cause tissue damage in multiple target organs; for example, lupus nephritis, which occurs in a subset of patients with SLE, affects the kidneys.

BAFF is a key B cell survival factor that, when overexpressed in mice, results in an expanded B cell compartment and relaxed negative selection within the GC87,88,89. Moreover, some patients with SLE exhibit sustained elevations in BAFF levels90. BAFF binds to three receptors, BAFF receptor (BAFFR; also known as TNFRSF13C), TACI (also known as TNFRSF13B) and BCMA (also known as TNFRSF170), all of which play important roles in promoting the survival of B cells. Different B cell subsets have a differential reliance on each receptor91,92.

As B cells are the effectors of this disease, BCDT has been examined in multiple human trials. Although belimumab was successful in phase III trials and is approved for the treatment of SLE93,94, many patients do not respond to belimumab, and we do not know whether this agent will alleviate lupus nephritis because individuals with renal disease were excluded from the trial95. In contrast, ianalumab, a novel BAFF-targeting antibody that prevents BAFF signalling by blocking BAFFR on B cells96, is worth considering, and this drug has had promising effects in Sjögren disease96.

In spite of early successes in open-label trials, an anti-CD20 antibody (rituximab) failed to meet its primary end points in two randomized controlled trials97,98, and worryingly, some patients experienced flares when B cells naturally reconstituted themselves after the drug was ceased99. The patients who experienced flares were those whose B cell compartments were repopulated quickly and disproportionately with PBs, and these patients had high levels of antibodies to double-stranded DNA (anti-dsDNA)100. Indeed, a PB gene signature is associated with severe disease101. One compelling hypothesis to explain these flares could be that BAFF expression, which is elevated after rituximab treatment, promotes B cell expansion on drug use cessation102,103.

Another mechanism for depleting B cells is to target the CD19 co-receptor using obexelimab, a humanized antibody that is also engineered to engage the Fcγ receptor FcγRIIb as a means of inhibiting BCR and/or TLR9 signalling to increase efficacy104. Unlike CD20, CD19 is expressed on bone marrow-resident pro-B cells and pre-B cells as well as some PBs and PCs105 (although the most mature PCs, the LLPCs, many of which reside in the gut, are CD19 (ref.106)). Thus, antibodies that target CD19 may have broader effects on the B cell compartment than anti-CD20 antibodies. Although this may be an advantage for antibody-mediated diseases, there may also be safety concerns that accompany the loss of precursor B cells and many subsets of antibody-secreting cells. In a phase II trial of obexelimab, investigators tapered patients off immunosuppressive steroids before randomly assigning them to treatment groups. Although patients receiving obexelimab sustained their level of disease inactivity despite steroid withdrawal in initial studies95, phase II clinical trials, completed in 2018, failed to meet their primary end points107.

Removing B cells by inhibiting Bruton tyrosine kinase (BTK) is also being investigated. BTK is expressed in B cells and myeloid cells, and its kinase activity is induced by both BCR and Fc receptor engagement. On the basis of preclinical animal models with BTK inhibitors, a reversible orally administered BTK inhibitor, GDC-0853, is currently being tested in a phase II trial in SLE95. It is difficult to speculate on the safety and efficacy of GDC-0853 using the lessons learned from anti-CD20 antibodies because GDC-0853 will affect both B cells and Fc receptor-expressing cells. Nevertheless, because of safety concerns around B cell depletion (susceptibility to infection, for example), the rapid reversibility of this drug is certainly appealing.

One possible way to improve on these results is to combine therapies. On the basis of the possibility that ‘bad’ (that is, proinflammatory) B cells emerge after rituximab treatment in response to elevated BAFF levels, three separate trials are examining anti-CD20 treatment followed by anti-BAFF therapy. One of those trials failed to show a difference between rituximab versus rituximab plus belimumab, whereas the other trial did show an improvement in dual-treated participants. The third trial (BEAT-Lupus) will hopefully clarify whether this is a viable treatment option for patients with SLE95.

Another possibility is to target two cytokines with one drug. Atacicept (a TACI–Ig; a recombinant fusion protein of the extracellular domain of TACI and a modified Fc portion of human immunoglobulin) neutralizes both BAFF and the related cytokine APRIL. Thus, treatment with atacicept could inhibit signals downstream of all three BAFF receptors. Unfortunately, the atacicept trial in SLE was halted due to safety concerns: atacicept-treated patients experienced serious infections and proteinuria actually increased108.

BCDT in RA

RA is a chronic autoimmune disease that is mainly characterized by joint inflammation and bone and cartilage damage. RA can also affect a wide variety of body systems, including the skin, eyes, lungs and heart. Like SLE, autoantibodies have been associated with RA, notably rheumatoid factor (RF; antibodies that recognize the Fc region of IgG) and anti-citrullinated protein antibodies (ACPA; antibodies that recognize proteins that have been post-translationally modified through deamination of arginine residues). The presence of ACPA is more specific than the presence of RF for diagnosing RA. Patients with RA who are seropositive for autoantibodies typically experience more severe disease109. Because of the role of antibodies in RA, anti-CD20 agents have been tested, and the use of rituximab as a BCDT has shown great promise47,110. Of note, in a phase II trial of rituximab, circulating antibody levels were not altered by treatment47,48, even though patients who were seropositive had a more robust benefit from treatment than those who were seronegative111.

By contrast, treatment with atacicept failed to reduce the severity of RA (no significant change in the American College of Rheumatology criteria for 20% reduction in disease severity), despite reducing levels of RF112,113. Therefore, it remains unclear why anti-CD20 works in RA but atacicept does not, although one of the atacicept trials tested patients in whom treatment with TNF antagonists had failed and may therefore have represented a more treatment-resistant patient group. One possibility is that B cells in the synovium may propagate TLT formation that can support local production of antibody-secreting B cells114,115. Since the abundance of B cells in the synovium positively correlates with joint damage progression116, B cell-rich TLT could be a viable target for anti-CD20 therapy. Indeed, anti-CD20 therapy depletes synovial B cells to various degrees, and the depth of depletion correlates with clinical response117,118.

Changes in B cell-associated cytokine production may be another key part of the mechanism of action of BCDT in RA. Indeed, B cells from patients with RA produce less IL-10, and BCDT corrects this defect, perhaps because the cells that repopulate the periphery are biased towards IL-10 production119,120. B cells from patients with RA also produce proinflammatory cytokines, and PCs in the joint produce TNF, one of the key cytokines involved in RA121. Importantly, joint-resident B cells also make RANKL (also known as TNFSF11), which could actively promote bone destruction122. Thus, a ‘reset’ in cytokine production as a consequence of BCDT may result in clinical benefit for the patient.

Other indications for BCDT

Mouse studies that examine the effects of B cell deficiency on different disease states provide us with clues for how B cells affect virtually every tissue. For example, B cells promote insulin resistance123 and hypertension124,125. In insulin resistance, not all B cells are pathogenic: IgA-producing PCs can indeed limit obesity-induced insulin resistance126. The number of B cells within the myocardium increases in murine models of myocardial infarction, and B cell depletion in the context of myocardial injury is beneficial (reviewed elsewhere127). B cells have also been implicated in a number of vascular conditions, including aneurysm128 and atherosclerosis129. As in insulin resistance, B cells may exert beneficial and pathogenic effects in atherosclerosis; for example, BAFF overexpression accompanied by the expansion of some B cell subsets seems to be protective130. These preclinical experiments provide an impetus to explore BCDT in metabolic, myocardial and vascular disorders.

In humans, other indications that benefit from BCDT include pemphigus vulgaris, a severe and rare autoimmune disease that results in blisters on the skin and mucosal membranes, in which rituximab has remarkable effects131. Other disease states that are driven by autoreactive B cells, such as Sjögren syndrome, respond variably to anti-CD20 therapy, and trials targeting the BAFF pathway using ianalumab and belimumab are ongoing132.

Lastly, although BCDT obviously has a beneficial effect in B cell-derived malignancies, the role of B cells in other cancers has also been extensively studied. Although beyond the scope of this Review, not surprisingly, B cells can have both protumorigenic and antitumorigenic effects in non-B cell cancers, and this likely depends on the intratumour context and the nature of the tumour antigen(s)133.

Neurological autoimmune diseases

A number of neurological disorders are diagnosed by, or linked to, abnormal antibody levels in the cerebrospinal fluid (CSF)134. Therefore it was reasoned that with enough time, the PB and/or PC reservoir of central nervous system (CNS)-directed autoantibody-producing B cells would be eroded in patients treated with BCDT. In this section, we briefly discuss B cells and BCDT in MS. We also discuss the efficacy of BCDT in other neurological autoimmune disorders that are associated with autoantibody production (Fig. 2; Table 1).

B cells in MS

MS is a chronic demyelinating disease of the CNS with no cure or known cause. Genetic and environmental factors play roles. In MS, a complex interplay of neurodegenerative and immunological processes damages the myelin sheaths that surround neurons. There are three clinical variations of MS: relapsing–remitting MS (RRMS), primary progressive MS (PPMS) and secondary progressive MS (SPMS). RRMS is typified by episodes of clinical relapses — during which immune cells infiltrate the CNS and form focal lesions that can be detected by MRI — followed by clinical improvement and remyelination of lesion areas. Progressive forms of the disease (PPMS and SPMS) are considered to be more neurodegenerative in nature and have lower levels of inflammation and immune activation135. Without treatment, most patients with RRMS convert to SPMS, and reductions in conversion rates are often measured in clinical trials as a parameter for success. Not surprisingly, the efficacy of therapeutics differs between MS subtypes.

Historically, MS has been thought of as a T cell-mediated disease. This was fuelled by genetic observations (the MS risk is associated with the HLA-DRB1*15:01 allele)136, histopathology studies (T cells are much more abundant than B cells in MS lesions)84,137 and the key observation that adoptive transfer of activated myelin-specific T cells into naive mice is sufficient to cause EAE138. However, in other species, such as the marmoset, B cells are necessary for disease139,140. Ironically, the importance of B cells in MS has been appreciated for many years. One of the earliest diagnostics for MS was the presence of IgG oligoclonal bands in the CSF, implying the presence of intrathecal antibody-producing cells141,142. Evidence of B cell accumulation in CNS lesions143 and ectopic B cell follicles in the meninges of patients with SPMS (and, to a lesser extent, those with PPMS)137,144 further implicated B cells in the aetiopathology of the disease. Deposition of complexes containing antibodies and complement components in MS plaques also suggests that B cells are important in MS pathogenesis145,146, although in a separate study these complexes were not found to be a specific feature of lesions, nor a hallmark of MS in particular147. To date, no specific autoantibody has been demonstrated to be pathogenic and/or pathognomonic.

Studies have begun to characterize B cells in the blood of individuals with MS, and have examined the localization of B cells in the inflamed CNS. Specifically, NBCs in the peripheral blood of patients with MS have altered cytokine profiles: typically a decreased ability to produce IL-10, and increased production of IL-6, GM-CSF and LTα3 (refs60,75,148,149,150). Interestingly, these aberrant cytokine profiles are normalized in B cells in the blood of patients with MS who have reconstituted their B cell compartment after BCDT60. Therefore, these studies suggest that the proinflammatory cytokine profiles of B cells in the blood may alter the disease in individuals with MS and can be modulated by BCDT134,151.

In terms of CNS-resident B cells, patients with active disease can accumulate B cells in the CSF, including antigen-experienced B cell subsets such as MBCs and PCs152. Sequencing of the immunglobulin heavy chain variable region repertoire to ascertain clonal relationships between peripheral B cells and CNS-derived B cells revealed clonal relationships between B cells in the CNS and B cells in the cervical lymph nodes. Moreover, by building lineage trees based on the relative amount of affinity maturation compared with a germ line sequence, it was determined that ancestral B cell clones are enriched in the cervical lymph nodes compared with the CNS. This elegant study suggested that, although antigen-experienced B cells are clearly present in the CNS, most B cell maturation and activation occurs in the periphery153.

B cell infiltration is not equally distributed throughout the CNS. CD20+ B cells tend to accumulate in perivascular spaces and within the subarachnoid space of the leptomeninges rather than in the tissue parenchyma84. B cell-enriched TLTs within the subarachnoid space are particularly evident in patients with SPMS, and these B cell-rich and PC-rich structures are typically associated with grey matter pathology in layer I of the underlying cortex (subpial lesions)137,154,155. However, it is difficult to know whether TLT structures are impacted by BCDT as their small size makes them extremely difficult to capture by MRI.

From these findings taken together, B cells show significant alterations in accumulation, localization and activation in patients with RRMS versus healthy controls, and thus BCDT was posited as a treatment option.

BCDT in MS

Most MS treatments focus on depleting or altering the trafficking of peripheral immune cells. This is very effective for treating RRMS, in which waves of proinflammatory leukocytes from the periphery breach the blood–brain barrier and subsequently promote focal demyelinating lesions. Traditionally, clinical trials in patients with MS use clinical disability and/or radiological evidence of lesions (by MRI) as end points to measure success. Newer trials use a combination of clinical relapse rate, MRI-based evidence of lesions and progression of disability (which together form ‘no evidence of disease activity’, (NEDA-3)) as an end point that is measured 1, 2 or even 5 years after treatment initiation. Disease-modifying therapies, which abrogate specific immune processes involved in RRMS pathogenesis, are very good treatment options for patients with MS. These therapeutics include fingolimod156, ozanimod157 and natalizumab158, whose clinical efficacy may be because they reduce lymphocyte trafficking to the CNS. Three BCDTs that target CD20 have shown efficacy in MS: rituximab159, ocrelizumab160,161 and ofatumumab162,163.

HERMES, a phase II placebo-controlled clinical trial of rituximab in patients with RRMS, was remarkably successful and beat expectations. A priori criteria for the HERMES trial included a 60% reduction in active gadolinium-enhancing lesions over a 24-week period in response to rituximab (although this efficacy may well have occurred before 24 weeks)159. Treatment effectively depleted B cells in the peripheral blood after 4–8 weeks and in the CSF after 24 weeks46,164,165,166. Amazingly, patients who received rituximab treatment had a 91% reduction in lesion load and a clinical benefit after only 12 weeks of treatment. A persistent reduction in new lesion development was observed for up to 48 weeks. Importantly, IgG titres in the CSF were unchanged and oligoclonal bands persisted46,165. The benefit also persisted after the B cells repopulated following cessation of treatment46,165. In a more recent multicentre study167, patients treated with rituximab had greater reductions in annual relapse rates than those treated with IFNβ, a front-line therapy168. From these findings taken together, rituximab treatment considerably reduces clinical disease and reduces the lesions seen by brain imaging in people with MS, and these improvements correlate with a depletion of B cells in the CSF and blood but not with changes in IgG in the CSF.

Other CD20-targeting BCDTs are also beneficial in people with MS. The concurrent clinical trials OPERA I and OPERA II tested the efficacy of the fully humanized mAb ocrelizumab in MS with similar results. Patients with RRMS who were treated with ocrelizumab had significant clinical benefit (46–47% lower annualized relapse rate) and radiological improvement (94–95% reduction in active lesions)169. Ocrelizumab was also the first therapeutic that showed efficacy — albeit limited — in patients with PPMS in the ORATORIO phase III clinical trial. In this trial, 12-week clinical disease progression was the primary end point: 32.9% of ocrelizumab-treated patients and 39.3% of IFNβ-treated patients reached that end point160. Furthermore, MRI scans of the brain with T2-weighted images showed that ocrelizumab-treated patients had much smaller lesion volumes than patients who received IFNβ. Although ocrelizumab was not as efficacious in PPMS as it was in RRMS, the ORATORIO clinical trial nevertheless met its primary end point and four of the five secondary end points. Since the progressive form of MS has very few treatment options, ocrelizumab has been approved by the FDA for this patient population and has begun to meet the significant unmet need. Notably, ocrelizumab seems to work better in a subset of younger patients (younger than 51 years) with PPMS who also have active inflammation170.

Attempts to improve these anti-CD20 therapeutics are ongoing. A subcutaneously delivered form of ofatumumab had similar results to other BCDTs in phase II trials (such as MIRROR and APLIOS, NCT03560739)162, and outperformed teriflunomide, a front-line therapy for MS, in a phase III trial163. One major advantage of subcutaneous delivery, compared with intravenous delivery (which is used for most BCDTs), is that it renders a drug easier for patients to self-administer. The phase III trials ASCLEPIOS I and ASCLEPIOS II (NCT02792218 and NCT02792231) are currently testing the efficacy of ofatumumab delivered via an autoinjector pen. Ublituximab is a next-generation anti-CD20 IgG1 mAb that binds an epitope on CD20 (including residues 168–171 and 158 and 159) that is distinct from the epitope bound by rituximab and ocrelizumab171 or the epitope bound by ofatumumab172. Ublituximab has been glycoengineered to have a low fucose content in its Fc region, which should enhance its affinity for FcγRIIIa receptors and maximize antibody-dependent cellular cytotoxicity (ADCC) toward B cells173,174. In phase II clinical trials, ublituximab also led to B cell depletion and a reduction in gadolinium-enhancing T1 lesions, T2 lesion volume and annualized relapse rate171. Ublituximab is currently being evaluated in phase III clinical trials (ULTIMATE 1 (NCT03277261) and ULTIMATE 2 (NCT03277248)). Advancements in anti-CD20 therapies are key as they may both increase efficacy and improve access for patients.

Due to the success of anti-CD20 therapies, other approaches for targeting B cells have been tested in the clinic. Targeting CD19 instead of CD20 affects a wider array of B cells, and includes PBs, which may significantly contribute to inflammation in the CNS175,176. In a phase I clinical trial, an anti-CD19 antibody (inebilizumab) showed a trend towards fewer gadolinium-enhancing lesions as well as a reduction in the number of new or newly enlarged T2-weighted lesions177. Evobrutinib, a BTK inhibitor, has also been tested in a phase II clinical trial. Evobrutinib reduced brain lesions as assessed by MRI, but did not alter the annualized relapse rate or disability progression178. However, the trial enrolled patients with an unusually high baseline level of clinical disability, and patients were older than those in similar trials. Thus, it is possible that some of the enrolled patients may have been transitioning from RRMS to SPMS, a form of MS that is fairly resistant to BCDT178. The evolutionRMS 1 and evolutionRMS 2 (NCT04338022 and NCT004338061) trials are currently examining the potential of evobrutinib in phase III clinical trials. Tolebrutinib (formerly SAR442168 or PRN-2246), another promising BTK inhibitor, is slated for phase III clinical trials in RRMS (GEMINI 1 (NCT04410978) and GEMINI 2 (NCT04410991)), SPMS (HERCULES (NCT04411641)) and PPMS (PERSEUS (NCT04458051)). Two features that distinguish BTK inhibitors from those targeting CD20 or other surface markers are that BTK inhibition may also target myeloid cells, which express BTK179, and that the increased bioavailability of BTK inhibitors within the CNS relative to antibody-based therapies may increase their efficacy.

Therapies that inhibit the BAFF/APRIL pathway have also been investigated in MS as an alternative strategy for ‘starving’ B cells, PBs and PCs of essential growth factors. The ATAMS phase II clinical trial was designed to do this by treating patients with MS with atacicept, the TACI–Ig fusion protein that binds and sequesters both BAFF and APRIL180. The hope was that atacicept would have a greater impact than CD20-directed therapies on MS because it would also impact PC survival. However, to the surprise of MS researchers, patients treated with atacicept experienced a dose-dependent increase in annualized relapse rates, and the atacicept treatment arm contained a lower proportion of relapse-free patients than the placebo arm. Therefore, due to the increased frequency of relapses, the clinical trial was prematurely discontinued180. In ATON, a phase II randomized clinical trial, patients who presented with optic neuritis as a clinically isolated syndrome were treated with atacicept and monitored over time181. Unfortunately, enrolment in ATON was halted because of the results in the ATAMS trial, but preliminary results in ATON showed less degeneration of retinal nerve fibre layer thickness with atacicept treatment. Despite this encouraging reduction in retinal nerve fibre layer damage, a greater proportion of patients in the atacicept group (6/17) had a second clinical attack and converted to clinically definite MS than in the placebo group (3/17). From the findings taken together, atacicept is not an appropriate treatment for MS; we speculate on reasons why this may be the case in the section entitled “Potential mechanism of action of BCDT”.

Overall, some BCDTs are excellent treatment options for patients with MS (particularly RRMS, in which active inflammation is a prominent feature). However, the failure of another BCDT, atacicept, highlights our incomplete understanding of the nuances of B cells in MS. Data from these BCDT trials suggest that B cell subsets are heterogeneous in their capacity to regulate the immune response in MS. Moreover, the efficacy of anti-CD20 in MS is not related to a reduction in CSF antibody levels.

BCDT in anti-MOG spectrum disorders

The nearly universal presence of oligoclonal bands in the CSF of patients with MS spurred on efforts by many laboratories to measure autoantibodies to myelin–oligodendrocyte glycoprotein (MOG), the most abundant and accessible myelin antigen. Early studies indicated that some patients with MS were seropositive for anti-MOG IgG182,183. Moreover, immunization of mice with recombinant human MOG induces an EAE-like disease that is characterized by an anti-MOG IgG1 response that is necessary for clinical symptoms184,185,186. However, contrary to initial hypotheses, on closer examination, antibodies in the blood and CSF of patients with MS were not specific for the correctly folded form of MOG187,188. Soluble forms of the antigen are measured by standard enzyme-linked immunosorbent assay or western blot techniques189, but detection of antibodies to conformationally appropriate MOG, which is more biologically relevant, uses transfected human cells that express full-length MOG on the cell surface190. With these better diagnostic criteria, patients with MS had levels of anti-MOG that were very low, similar to those of healthy controls191,192,193,194, thus upending the concept that anti-MOG antibodies play a pathogenic role in MS. However, antibodies to other CNS antigens, particularly neuronal and glial targets, have been identified in patients with MS and may participate in the aetiopathology of the disease195,196,197,198. These antibodies are specific to diverse antigens, often to intracellular targets that reflect cell damage, and are possibly generated secondary to the initiating autoimmune attack188,199,200,201.

Anti-MOG antibodies can cause CNS autoimmune disorders other than MS, but these are currently not well stratified or classified194,202. These disorders have a range of phenotypes and are collectively referred to as MOG spectrum disorders (MOGSDs). Patients with MOGSDs largely present with an acute disseminated encephalomyelitis-like phenotype in younger patients and opticospinal phenotypes in older patients. Antibody-mediated demyelination (possibly via anti-MOG antibodies) plays a role in these syndromes, and these disorders are now being teased apart from a similar set of diseases, NMO spectrum disorders (NMOSDs)203. MOGSDs provide an opportunity to assess the impact of BCDT in an antibody-mediated CNS disease setting. However, the heterogeneity of patients makes it difficult to conclusively determine the success or failure of a therapy204,205. Current treatment options for patients with an MOGSD include steroids or plasma exchange206. In early reports, rituximab had mixed clinical efficacy in MOGSD205,207,208. More recently, a study in 121 patients with MOG antibody-associated disorder (MOGAD) showed that rituximab reduced relapse rates in patients with this disorder, but many patients continued to relapse despite B cell depletion209.

BCDT in other CNS diseases

MOGSD is not the only CNS disorder with a link to autoantibodies that may be altered by BCDT. NMOSD, a disease that primarily affects the spinal cord, optic nerves and brainstem, was previously considered a clinical variant of MS210. Examination of autoantibodies obtained from patients with NMOSD indicated that NMO-IgG, which are antibodies to aquaporin 4 (AQP4), are a unique feature not found in patients with MS211,212. In a report of patients with NMOSD at the national UK NMO clinic, 73% of patients were seropositive for NMO-IgG213. Of the NMO-IgG-seronegative patients, 42% were positive for anti-MOG IgG, further complicating the delineation between NMOSD and MOGSD. APQ4, which is targeted by NMO-IgG, is found on glial cells, and binding of NMO-IgG decreases aquaporin surface expression, resulting in increased permeability of the blood–brain barrier and complement activation, which promotes pathologies such as gliosis and axonal damage212,214. An anti-CD19 antibody, inebilizumab, was recently approved for NMO-IgG-seropositive patients with NMOSD following a promising phase II/III placebo-controlled clinical trial (N-Momentum (NCT02200770))215. Anti-CD20 treatment also shows good clinical efficacy in NMOSD, although the therapeutic benefit does not correlate with a decrease in NMO-IgG levels in the blood207,216. A TACI–Ig fusion protein, RC18, is also currently being examined in NMOSD in an interventional phase III clinical trial (NCT03330418). Therefore, the effect of BCDT in NMOSD is an active area of research, but observed reductions in disease are not entirely explained by a reduction in the autoantibody response. This observation suggests that NMOSD pathogenesis is not exclusively caused by NMO-IgG, or, alternatively, that measurements of these antibodies in the serum do not reflect ongoing pathogenesis in the CNS216.

Autoantibody-mediated encephalitis is a collection of neurological syndromes with numerous or unknown potential causes217. One of most common autoantibody-mediated encephalitis syndromes is characterized by a response to N-methyl-d-aspartate receptor (NMDAR). NMDAR is an ion channel and glutamate receptor that is critical for neurotransmitter signalling. Two primary causes of NMDAR encephalitis are currently understood. A small proportion of patients have anti-NMDAR encephalitis following infection with herpes simplex virus 1 (ref.218). More commonly, women with ovarian teratomas that express NMDARs can generate an NMDAR autoantibody response to the teratoma, and these antibodies eventually target NMDARs within the CNS219. This phenomenon was described in the popular Brain on Fire book and film, which were based on the experiences of a journalist at the New York Post who experienced anti-NMDAR autoimmunity217. Pathology from NMDAR autoantibodies is mediated by crosslinking NMDARs at the synapses on CNS neurons, leading to reduced levels of NMDAR, which likely alters memory, learning and other cognitive functions220. Despite the ovarian teratoma trigger, the anti-NMDAR immune response seems well compartmentalized to the CNS. Early reports have shown that rituximab leads to clinical improvement for anti-NMDAR encephalitis; however, it is not clear whether these improvements are because of changes in B cell numbers or because of changes in autoantibody levels221. Two BCDT trials are being conducted in anti-NMDAR encephalitis using rituximab (IANMDAR, a prospective phase II trial in paediatric patients with anti-NMDAR encephalitis (NCT03274375)) and inebilizumab (ExTINGUISH, a randomized phase IIb trial (NCT04372615)).

Myasthenia gravis (MG) is a neuromuscular disease that results in severe muscle weakness due to antibody-mediated blockade of nicotinic acetylcholine receptor (AChR) at neuromuscular junctions. This largely occurs due to an IgG1 or IgG3 autoantibody response to the AChR222. Targeting AChR leads to complement-mediated damage at the synapse and overt clinical symptoms. A subset of patients with MG test negative for antibodies to AChR and patients in those seronegative subgroups can be positive for other autoantibodies, such as anti-muscle-specific tyrosine kinase (anti-MuSK) IgG4 or anti-LRP4 IgG1 or IgG2. Standard treatment for MG includes the use of immunosuppressive steroids223, but BCDT has also been investigated as a more targeted approach. Of note, patients with anti-MuSK antibodies fared better with rituximab therapy than patients with other autoantibodies such as anti-AChR224,225. Efficacy in patients with anti-MuSK antibodies correlated with a reduction in anti-MuSK IgG4 in the serum; of note, the anti-AChR antibodies were predominantly IgG1 and IgG3 (ref.225). Taken together, BCDT provides therapeutic benefit in some but not all patients with MG and the serotype is a key indicator for clinical success225.

Lastly, a constellation of rare CNS diseases driven by IgG4 antibodies are receiving increased interest. IgG4, which has low affinity for the complement component C1q and Fc receptors, is an unusual IgG because it readily undergoes Fab arm exchange with other IgG4 molecules, leading to bispecific IgG4 molecules. Such molecules cannot form large immune complexes, and hence they cause little or no inflammation226. Although IgG4 was initially considered as an anti-inflammatory immunoglobulin, IgG4 is implicated in several autoimmune diseases, most notably MG227. Additionally, IgG4 has been linked to neuropathies, encephalitis (IgG4 antibodies targeting leucine-rich glioma-inactivated protein 1 (LGI1)), and pemphigus (pemphigus foliaceus and pemphigus vulgaris). As mentioned, rituximab is selectively efficacious in patients with MG who have anti-MuSK IgG4 (ref.225).

Because of the rapid efficacy of BCDT in patients with anti-MuSK IgG4, it is possible that the cells that make these IgG4 antibodies (PBs) are not long lived; thus, BCDT prevents replenishment of the pathogenic IgG4 antibody-producing pool. Indeed, O’Connor and colleagues have shown that the precursors to these PBs are derived from antigen-experienced CD27+ B cells that are impacted by BCDT, and when incompletely depleted, these CD27+ B cells can drive disease flares228. Alternatively, the cells making anti-MuSK IgG4 may express the CD20 antigen and are thus susceptible to outright depletion. If this is true, these IgG4-mediated diseases may be particularly sensitive to BCDT. However, the scarcity and heterogeneity of IgG4-mediated autoimmune disorders means that designing good clinical trials is challenging.

In summary, clinical trials in CNS diseases in which B cells and/or autoantibodies participate highlight that BCDT cannot be used indiscriminately for therapy. Accordingly, the choice of BCDT requires careful examination of B cell subsets and their antibody-dependent and antibody-independent functions.

Potential mechanism of action of BCDT

The successes of anti-CD20 treatment in autoimmune diseases such as RA and MS are some of the most encouraging examples of positive clinical outcomes for disease-modifying therapies in autoimmune diseases that have been traditionally difficult to treat. However, not all B cell-driven diseases respond to BCDT (SLE, for example, does not). The reason for such disparity is still unclear. Here, we discuss some working hypotheses for why BCDT may not always work and speculate on why it sometimes does work when we did not expect it to (Fig. 3).

Fig. 3: Effects of BCDT on B cell populations.
figure 3

Most of our knowledge of how B cell populations change after B cell depletion therapy (BCDT) is derived from treatment with anti-CD20 agents and is depicted here. We hypothesize that the efficacy of BCDT in some autoimmune diseases begins with a reset of the B cell pool in the periphery. By elimination of CD20+ memory B cells, some of which may be producing proinflammatory cytokines such as granulocyte–macrophage colony-stimulating factor (GM-CSF), the periphery is enriched with naive B cells and IL-10-producing regulatory B cells. Moreover, short-lived plasmablasts that are potentially autoreactive are not replenished by autoreactive memory B cells after BCDT. Long-lived plasma cells (LLPCs), however, which are present in the bone marrow (BM) and gut, are not eliminated. Surges in BAFF levels that occur after BCDT may promote the accumulation of immunoregulatory LLPCs, which could have a beneficial effect in some diseases, such as multiple sclerosis. In target tissues, B cells in tertiary lymphoid tissues (TLTs) may not be replaced. Consequently, the ability for other pathogenic immune cells to take up residence in the inflamed tissue may be compromised, and other immunoregulatory cells, such as T follicular helper cells (TFH cells), may take up residence, resulting in reduced disease. LTα3, lymphotoxin-α3; PC, plasma cell; TNF, tumour necrosis factor.

Antibody design

On the basis of the success of BCDT in some indications, second-generation anti-CD20 mAbs have rapidly emerged. Different anti-CD20 mAbs evoke distinct cytotoxic mechanisms: complement-dependent cellular cytotoxicity (CDC), FcγR-mediated depletion through cellular effector mechanisms including ADCC and antibody-dependent cellular phagocytosis, and directly inducing cell death229. Anti-CD20 mAbs can be divided into two types (type I and type II) on the basis of their ability to redistribute CD20 into plasma membrane lipid rafts, and their potency in assays measuring CDC, homotypic adhesion and programmed cell death induction230. More recently, with use of cryo-electron microscopy it has been shown that type I mAbs can act as molecular ‘seeds’ to increase the local mAb concentration, which promotes CDC231. The first generation of genetically engineered chimeric anti-CD20 mAbs, such as rituximab and ocrelizumab, are type I mAbs. Rituximab comprises a murine variable region (Fab) and a human constant region (Fc), whereas ocrelizumab is a humanized anti-CD20 IgG1. Ocrelizumab has superior ADCC activity in comparison with rituximab232,233. The main mechanism through which type I anti-CD20 mAbs deplete B cells is by redistribution of CD20 into lipid rafts: mAb-driven CD20 clustering results in CDC, although the process of cell death is independent of raft formation234. However, internalization of type I anti-CD20 mAbs from the surface via FcγRIIb results in resistance to therapy and decreased clinical efficacy235. Type II mAbs, such as obinutuzumab, are fully humanized mAbs. Compared with rituximab, this type II mAb binds to a different epitope, and does not induce efficient clustering of CD20 or antibody internalization, thus increasing efficacy and decreasing possible treatment resistance236. Hence these type II mAbs may have greater efficacy in diseases in which BCDT has shown only marginal efficacy.

Tissue and B cell lineage specificity

BCDT has a range of outcomes depending on the disease. One relatively straightforward reason for this variability is that pathogenic B cells in some scenarios are difficult to access (for example, because B cells localize into TLTs in the synovium or the leptomeninges). The depth of B cell depletion may therefore depend on the target tissue. For instance, the levels of synovial B cells were decreased but not fully eliminated in RA after BCDT117. Also, GC B cells in Peyer patches and peritoneal B-1 cells resist rituximab treatment in human CD20 transgenic mice, as these antibodies need to access the vasculature to deliver the antibody-bound B cells to effector cells of the reticuloendothelial system44. In MS, B cell depletion is effective in the blood and CSF164,165, but variable in the CNS237,238. Inflammation in MS can be CNS compartmentalized into TLTs (for example, in the leptomeninges), so B cells may persist in these locations, particularly because survival factors such as BAFF are locally produced by astrocytes239. Furthermore, a substantial reservoir of T-bet+ MBCs that express CD11c resides in the spleen and does not circulate. T-bet+ MBCs can be precursors to antibody-producing cells that could play a pathogenic role in autoimmune diseases such as SLE39. It is unclear whether anti-CD20 treatment effectively depletes these tissue-resident cells.

BCDT may also have variable outcomes in different autoimmune diseases if the pathogenic B cell lineage does not express CD20 and turnover is slow. For example, CD20 is largely absent on subsets of LLPCs in healthy human bone marrow240 and gut106. For this reason, anti-CD20 therapy does not substantially deplete LLPCs. New approaches that target CD19, such as using an anti-CD19 mAb or a CD19-specific chimeric antigen receptor (CAR)-based T cell therapy, have been explored for some B cell malignancies241. However, a population of LLPCs that do not express CD19 nevertheless persists after infusion242. Thus, when anti-CD20 therapy has robust efficacy in diseases that are known to be mediated by autoantibodies, such as ANCA-associated vasculitis, NMO and pemphigus vulgaris, this efficacy may be due, at least in part, to a loss of short lived PBs or PCs that are not replaced because precursor B cells are also depleted.

On the other end of the efficiency spectrum, anti-CD20 therapy has low efficacy in SLE, the canonical B cell-driven disease. We posit that this may be because pathogenic LLPCs that are CD20 and thus resistant to rituximab have a critical role in SLE. Not only does this potentially explain why anti-CD20 therapy failed in SLE, in which LLPCs are an important source of anti-DNA antibodies243, it may also explain why, after BCDT, autoantibodies persist in RA244 and NMO245, oligoclonal bands remain in MS46 and protective antibodies elicited by previous vaccinations can still be found246.

Impact on cytokine-producing cells

The efficacy of BCDT in the absence of altered autoantibody levels suggests that antibody-independent B cell functions, such as cytokine production, could be important. Cytokines produced by B cells may alter the function of other immune cells such as T cells and myeloid cells. For example, in MS, B cells produce very high quantities of cytokines such as TNF, IL-6 and GM-CSF75,247. IL-6 produced by B cells has been clearly linked with the generation of a T helper 17 cell inflammatory immune response in EAE, and a similar observation has been made in humans: circulating B cells from patients with MS overproduce IL-6 in vitro, and anti-CD20 treatment normalizes IL-6 production59. Indeed, in patients with MS, B cells that were reconstituted following BCDT had totally different cytokine profiles compared with B cells from untreated patients. Reconstituted B cells were characterized by limited proinflammatory cytokine production and higher production of anti-inflammatory IL-10 (ref.60). Thus, BCDT may produce a clean slate for peripheral B cells, providing a much-needed cytokine reset for the patient.

Antigen presentation and MBCs

In addition to producing antibodies and cytokines, B cells can internalize and process antigens, and then present them to CD8+ and CD4+ T cells via MHC class I and MHC class II, respectively. Interestingly, in EAE induced by full-length recombinant human MOG, the antigen presentation capacity of B cells was critical to disease induction, and passive transfer of pathogenic autoantibodies could only partially rescue disease severity if MHC class II was specifically deleted in the B cell compartment248,249. Although B cells are considered to be less efficient antigen-presenting cells than dendritic cells, MBCs can play an important role in antigen presentation, particularly if B cells present the same antigen as the responding T cell250.

B cells also mediate the proliferation of brain-homing autoreactive myelin-specific CD4+ T cells independently of any external T cell stimuli, and this function is lost after rituximab treatment251. These myelin-specific CD4+ T cells increase in number predominantly in patients with MS who express the HLA-DR15 haplotype and decrease in number after anti-CD20 treatment251, providing further evidence that interrupting B cell–T cell interactions by BCDT may be relevant to MS aetiopathology. In summary, it is possible that BCDT is efficacious in part because it abrogates proinflammatory antigen presentation by B cells.

Other hypotheses on mechanism

The effects on non-B cells may explain the success of anti-CD20 therapy in MS. The existence of a rare population of CD8+ T cells that concomitantly express CD20 was first described in 1993 by flow cytometry252. CD20+ T cells have also been described in patients with RA253 as well as MS254. Some patients with MS have an increased number of peripheral CD3+CD20low T cells, which can be efficiently depleted by rituximab255. Subsequently, with use of myelin–MHC class I tetramers to identify myelin-specific CD8+ T cells in peripheral blood, the number of myelin-specific CD20+CD8+ T cells was found to be increased in patients with MS, and the number of these cells was reduced after anti-CD20 treatment256. In addition, the presence of CSF-resident highly proliferative IFNγ-producing CD4+ and CD8+ T cells that express CD20 positively correlates with disease severity in MS254. Thus, anti-CD20 therapy may be effective partially because it removes CD20+ T cells.

In addition to this potential off-target-cell effect, removing B lymphocytes, an entire branch of the immune system, will change the availability of growth and survival factors that were previously consumed by these cells. Following anti-CD20 treatment in patients with MG, the relative proportion of PCs to B cells is increased, and IgA+ MBCs outnumber other MBCs257. It is unclear why this happens, but it is important to remember that BAFF levels increase after rituximab treatment102,103. This elevation of BAFF may have different effects in different disease settings. In some patients with MS, BAFF levels are elevated in both the serum and the CSF258,259. Moreover, Steri et al. showed that a variant BAFF (also known as TNFSF13B) allele that favours overproduction of BAFF has been associated with MS260. This observation suggests that BAFF may be deleterious in MS. However, others have found that high blood BAFF levels correlate with a more stable disease course of MS261, and, as mentioned, BAFF inhibition with atacicept worsened MS disease180. The consequences of high BAFF levels in MS are unclear because of these contradictory observations. We have found that high BAFF levels strongly protect against EAE. In our study, mice that overexpressed BAFF had an abundance of IgA-producing PCs in the gut. These PCs also produce IL-10, and, when transferred into mice that lack IL-10, can confer protection against EAE81. BAFF and the related cytokine APRIL are key survival cytokines that sustain LLPCs262. APRIL can also promote class switching to IgA18,19. Thus, one potential mechanism for the efficacy of anti-CD20 therapy in MS is that BAFF levels are increased after treatment and consequently promote the survival of immunoregulatory PCs, particularly gut-derived IgA+ PCs. Indeed, patients with RA who were treated with rituximab still have of IgA+ PCs in the lamina propria in gut biopsy samples263. This may also explain why inhibition of BAFF by atacicept had a negative outcome for patients with MS180.

This hypothesis may seem at odds with the findings of Steri et al.260. However, the study by Steri et al. did not examine whether patients with the variant allele had elevated levels of BAFF trimer or of higher-order structures of the cytokine. This is of critical importance because only highly oligomerized forms of BAFF (60-mer), or APRIL bound to heparin sulfate proteoglycan, stimulate TACI91,92. Thus, in contrast to the situation in SLE, in which high BAFF levels after rituximab treatment may be counterproductive102,103, increased BAFF levels capable of triggering TACI may be beneficial in MS.

New therapeutic directions

On the basis of what we have learned about B cells in the past 10 years, we might now be able to design therapies that provide superior specificity and sensitivity to thwart inflammatory B cells and foster the function and accumulation of immunoregulatory B cells (Fig. 4). In this section, we provide a few perspectives on what this could look like in the context of the findings discussed herein. One therapy type that has not been discussed, as it is beyond the scope of this Review, is drugs that target the CD40 pathway264,265. Although the first anti-CD40L drug (ruplizumab; BG9588) was not viable because it caused thrombotic side effects266, new and safer variants are being tested. Given the importance of CD40-dependent T cell–B cell interactions in the GC as provided by TFH cells, we optimistically await the results of testing new forms of anti-CD40 or anti-CD40L therapies in autoimmune disease settings267,268.

Fig. 4: Future approaches using B cell-targeted therapies in autoimmune disease.
figure 4

The next generation of B cell-targeted therapies, rather than B cell depletion therapies, will need to target specific proinflammatory and anti-inflammatory B cell functions. BAFF homotrimers versus higher-order aggregates (60-mers) of BAFF or APRIL impact different B cell subsets differently. Inhibiting BAFF homotrimers could have a beneficial effect because it may remove T-bet+ B cells and/or B cells that participate in the extrafollicular response. Because a subtype of T-bet+ B cells (double-negative 2 (DN2) cells, which are considered to be non-circulating memory B cells) gives rise to antibody-producing plasmablasts and plasma cells in an IL-21-dependent manner, IL-21 blockade may ameliorate disease by inhibiting these non-circulating memory B cells. However, supplementation with BAFF 60-mers or APRIL may promote the accumulation of IgA+ immunoregulatory plasma cells, which could be beneficial in multiple sclerosis (MS). Likewise, the levels of immunoregulatory plasma cells may be increased by manipulating the microbiota. The germinal centre (GC) itself is the source of somatically hypermutated B cells that play an important role in diseases such as systemic lupus erythematosus. Interrupting the entanglement between B cells and T follicular helper cells T (TFH cells) by blocking CD40 (or other candidates such as inducible T cell costimulator (ICOS)) could provide a means for staving off this source of potentially bad B cells. In tertiary lymphoid tissues (TLTs), removal of B cells may change the cellular environment of TLT, reducing its ability to foster autoimmune reactions.

Targeting T-bet+ B cells

Class-switched antibodies can be derived from T-bet+ B cells. In patients with SLE, these unusual cells are poised to make autoantibodies, dominate the blood of patients with active disease, and may be derived from an extrafollicular response rather than from the GC38. Recently, Reizis and colleagues showed that PBs that emerge from extrafollicular responses and make anti-dsDNA antibodies rely on type I interferon produced by plasmacytoid dendritic cells269. Targeting these T-bet+ DN2 cells by blocking IFNγ or type I interferon is an appealing potential way to treat SLE given the importance of these cells in the disease. Indeed, anifrolumab, which blocks the interaction between the interferon receptor (IFNAR) and its type I interferon ligands, has shown modest promise in a phase III trial in SLE270.

Harnessing the regulatory potential of PCs

IL-10 is a direct target of the transcription factor IRF4 in PCs80. The inhibitory molecules PDL1, PDL2 and LAG3 are likewise expressed by PCs, and PCs are primed to make IL-10 as they have an open chromatin configuration at the IL10 locus271. We have also shown that IgA-producing PCs can produce IL-10, and that the gut is a rich source of IL-10-producing PCs that can regulate neuroinflammation81. IgA-producing PCs in the gut can be extremely long-lived (measured in decades), and most LLPCs are negative for both CD19 and CD20 (ref.81). It is unknown whether the lifespan potential of a PC correlates with its suppressive function. Nevertheless, if we can learn what factors enhance the accumulation of these cells — possibly BAFF or, in the context of the gut, particular microorganisms which we know are important for dictating IgA PC numbers272 — then this becomes a potential therapeutic approach for treating some autoimmune diseases.

Reconsidering MBCs

As illuminated by single-cell sequencing, the BCRs that are recruited into an immune response are not always the ones that we might expect. For example, Tan et al. identified a startlingly common insertion between the V and DJ fragments of the BCR that encodes the 98 amino acid collagen-binding domain of leukocyte-associated immunoglobulin-like receptor 1 (LAIR1). The BCRs that contain LAIR1 can undergo somatic hypermutation, consequently losing their capacity to bind collagen but gaining the ability to bind to Plasmodium falciparum-infected erythrocytes273, thus repurposing previously self-reactive B cells. Moreover, Murugan et al. showed that the initial B cell response to P. falciparum in humans who had never been exposed to the parasite was driven by B cell clones that had already undergone somatic hypermutation, demonstrating that MBCs can be recruited into de novo immune responses to completely different antigens274.

This makes good evolutionary sense: setting up a GC is an energetically expensive process that requires chemotaxis, antigen trapping and a substantial remodelling of stromal cell networks275. In addition to its role as a platform for affinity maturation of NBCs, why not use this environment to repurpose and ‘redeem’276 self-reactive B cells to participate in immune responses? While this is an exciting frontier in the evolution of our understanding of B cells, undoubtedly there are collateral consequences for the host when dealing with a live-or-die infection. Therefore, we must determine whether such repurposed B cell subsets can mediate damage to host tissues and, if they do, whether they are targeted by current therapeutics. Our intense research into the dynamics of the human B cell response to the SARS-CoV-2 pathogen will undoubtedly yield further surprises.

Conclusions

The field of BCDT began with a rather crude instrument (antibodies that deplete all CD20+ B cells). The clinical experiments performed with anti-CD20 therapy, however, have been surprising and revealing, spurring on research into how B cells impact disease. Our rudimentary idea that PCs and antibodies are ‘bad’ has matured considerably into a more nuanced view of B cells in health and disease. Moving forward, it will be important to not merely ascribe increasingly specific subset designations on the basis of emerging RNA sequencing data in the absence of a holistic view of the environment in which B cells differentiate and exist. For example, the microbiome is a key environmental factor that influences B cell phenotype63. Taking cues from our co-evolution with pathogens, new perspectives on the roles of B cells in autoimmune disease will undoubtedly emerge. These perspectives, combined with advances in antibody technologies, should increase the efficacy and specificity of BCDTs in autoimmune diseases.