Immune mechanisms of toxicity from checkpoint inhibitors

Abstract

Immunotherapy has changed the treatment landscape for cancer over the past decade. Inhibitors of the immune checkpoint proteins CTLA-4, PD-1 and PD-L1 can induce durable remissions in a subset of patients with metastatic disease. However, these treatments can be limited by inflammatory toxicities that can affect any organ system in the body and in some cases can be life threatening. Considerable progress has been made in understanding the drivers of these toxicities as well as effective management strategies. Further research into understanding the molecular and cellular mechanisms that drive toxicity will enable better prediction of toxicity and development of optimized therapies for these toxicities that avoid interfering with antitumor immunity. In this review we discuss our current understanding of the inflammatory toxicities from immune checkpoint inhibitors and propose optimal treatment strategies for these toxicities.

Introduction

Immunotherapy has fundamentally changed the treatment landscape for both solid and hematologic malignancies over the past decade[1]. Monoclonal antibodies that block the immune regulatory checkpoint receptors cytotoxic T lymphocyte antigen (CTLA)-4, programmed death (PD)-1 and its ligand PD-L1 have shown broad activity against multiple tumor types, leading to prolonged survival for many patients and even durable remissions[1]. More recently, a third class of immune checkpoint inhibitor (ICI) that targets lymphocytes activation gene (LAG)-3 has demonstrated efficacy in combination with PD-1 blockade[2].

Despite the enormous potential of checkpoint inhibitors, their success has been somewhat limited by a diverse spectrum of inflammatory toxicities that are collectively referred to as immune-related adverse events (irAEs)[3]. These toxicities can lead to treatment delays and discontinuation, and in rare cases can be life-threatening. In contrast to the adverse events induced by chemotherapy and targeted therapy, irAEs have proven to be difficult to predict, both in their severity and their timing[4]. In addition, irAEs have been reported in nearly every organ system in the body, and while certain organs (e.g. skin) are more commonly affected than others (e.g. endocrine pancreas), why a particular organ becomes inflamed in a particular person remains entirely unclear[3].

Understanding the molecular and cellular mechanisms that drive irAEs is clearly of clinical importance, but these toxicities also offer a window into basic mechanisms of immune regulation in humans. IrAEs are the “phenotype” of loss of function by the targeted receptors, which provides insights into which cells and pathways are under control by CTLA-4, PD-1/PD-L1 and LAG-3 at “steady state,” and how loss of that control manifests as disease. Many irAEs resemble spontaneous autoimmune disease and have the potential to serve as more tractable models of autoimmunity, effectively serving as human models of human disease[1]. The timing and nature of the immune perturbation is known for irAEs, and so, at least in theory, the initial events in autoimmunity can be observed as they unfold in real time.

In this perspective, we discuss our current understanding of the pathogenesis of irAEs, treatment approaches, and major unanswered questions in the field. At this point, the information available on CTLA-4 and PD-1/PD-L1 related irAEs is considerably more advanced than what we understand about LAG-3; we will consequently focus our discussion on CTLA-4 and PD-/PD-L1 blockade.

How does the class of immune checkpoint inhibitor influence irAE development?

CTLA-4 and PD-1/PD-L1 have very different roles in immune regulation (Figure 1). CTLA-4 is a decoy receptor for the co-stimulatory proteins B7–1 and B7–2 expressed on antigen presenting cells (APCs), preventing the B7s from interacting with the costimulatory receptor CD28 on T cells. CTLA-4 thus reduces activation of naïve T cells and may also interfere with ongoing stimulation of T cells in inflamed tissues[1]. CTLA-4 is also highly expressed on regulatory T cells (Tregs) where it can bind to B7s on activated APCs with such high avidity that the B7s are physically removed from the APC membrane[5]. PD-1 is an inhibitory receptor expressed on T cells after activation, with the highest levels expressed after repeated stimulation. PD-1 directly inhibits TCR and CD28 signaling through activation of SHP-2 phosphatase after binding to one of its ligands, PD-L1 or PD-L2. PD-L2, which is not targeted pharmacologically, is expressed on activated immune cells[1]. PD-L1 is more broadly expressed and can be upregulated on both healthy and malignant tissue in response to a variety of inflammatory and stress signals, including interferon gamma (IFNγ) produced by activated T cells[1,6]. The phenotype of mice deficient in CTLA-4 is also quite distinct from that of mice deficient in PD-1 or PD-L1. CTLA-4 deficient animals die shortly after birth from overwhelming autoimmunity involving multiple organs, including the muscles of respiration[7]. PD-1 and PD-L1 deficient animals are largely healthy but are substantially more susceptible to autoimmunity or pathologic inflammation in the setting of other predisposing factors, such as when crossed to autoimmunity-prone mouse strains[1,8,9].

Figure 1: Mechanisms of action of CTLA-4 and PD-1/PD-L1 pathways.

A) CTLA-4 binds tightly to the costimulatory ligands CD80 and CD86, preventing them from interacting with the costimulatory receptor CD28 on T cells. On newly stimulated T cells in the lymph node, CTLA-4 can prevent full activation in a cell intrinsic manner. Tregs express high levels of CTLA-4, enabling them to bind to CD80 and CD86 on dendritic cells (DC) and physically remove these ligands from the DC membrane, preventing subsequent activation of T cells by this DC. B) PD-1 expressing T cells that have entered the tumor microenvironment can bind to their cognate peptide-MHC complex on the target tumor cell, but activation is inhibited when PD-1 binds to PD-L1 expressed on the tumor cell. PD-1 delivers a direct inhibitory signal to T cell through activation of SHP-2 phosphatase, which inhibits signaling downstream of the TCR and CD28. IFNg produced by activated T cells increases PD-L1 expression by target cells, acting as a negative feedback loop.

The distribution, severity, and frequency of irAEs is clearly related to the class of ICI used[3]. CTLA-4 blockade leads to more severe and more frequent toxicities than does PD-1 blockade in most cases. In particular, inflammation in the gut is much more common with CTLA-4 blockade than with either PD-1 or PD-L1 inhibitors[10]. Hypophysitis is also more common on CTLA-4 blockade[11]. In contrast, PD-1 blockade is more frequently associated with thyroiditis and may be a more common cause of pneumonitis and autoimmune diabetes[11]. Combination immunotherapy targeting both CTLA-4 and PD-1 is considerably more toxic than either treatment alone, though for the most part, the risk of toxicity appears to be additive rather than synergistic[12]. Although the variable distribution of affected organs, the inflammation severity, and inflammation frequency suggest differences in the underlying mechanisms of toxicity, we have minimal existing data to directly address this question.

Histopathologically, toxicities caused by CTLA-4 blockade are not readily distinguishable from toxicities caused by inhibition of PD-1 or PD-L1[13–15]. Similarly, in one detailed analysis of colitis caused by immunotherapy, no clear differences were observed between colitis caused by PD-1 blockade and colitis caused by CTLA-4 blockade, though this may simply reflect inadequate sample size given the limited number of cells analyzed[16]. From a clinical perspective, treatment of irAEs appears to be dependent on the organ involved and the severity of the inflammation, rather than on the class of checkpoint inhibitor that was used[17]. Whether treatment could be optimized by considering checkpoint inhibitor class, however, is unknown.

For many diseases, questions like these could be addressed through carefully controlled experiments in animal models. However, a major limitation in irAE research is the relative paucity of good animal models that replicate the human disease. CTLA-4 deficiency in mice is rapidly lethal as mentioned above, while CTLA-4 blockade with antibodies rarely induces any perceptible side effects[3]. PD-1 and PD-L1 blocking antibodies and combination immunotherapies are similarly well tolerated[3]. Combined PD-1 deficient, CTLA-4 heterozygous mice do develop some autoimmune disease like ICI toxicities, including myocarditis, but do not develop the more common inflammatory syndromes in the barrier organs[18,19]. Colitis has been reported in some treatment situations using checkpoint blocking antibodies, but the penetrance of the phenotype is variable and may be facility- or strain-dependent[20,21]. Mice xenografted with human bone marrow do develop worsening inflammation after checkpoint inhibitor treatment, but this likely represents an exacerbation of xenograft versus host disease rather than a true irAE[22].

Why are some organ systems more frequently targeted by irAEs?

The inflammatory toxicities of checkpoint inhibitors can affect any organ system in the body. However, most toxicities occur at barrier organs including the skin, gastrointestinal tract and liver, and lungs (Figure 2). Although many of these toxicities are mild, severe toxicities in any of these organs can be life-threatening[23]. Colitis is the most common severe toxicity from current ICIs. Approximately 20% of patients on PD-1 blockade develop some mild version of gastrointestinal inflammation, with 2–5% developing more severe inflammation. On CTLA-4 inhibitors, gastrointestinal inflammation occurs in closer to 40% of patients with 10–15% developing severe disease[10,12]. Dermatologic toxicities are even more common, affecting the great majority of patients on ICI therapy. Most of these toxicities are mild and many can be managed with topical therapies, though severe toxicities can occur[14,24]. The fact that barrier organs are so commonly involved suggests that the antigenic targets of the immune response may be the commensal microbiome, though this has not yet been demonstrated (Figure 3). The adaptive immune system develops through a complex process of selection that deletes most high affinity self-reactive T and B cells[25]. However, immune cells that recognize harmless microbial and environmental proteins are regulated exclusively through peripheral tolerance mechanisms such as the CTLA-4 and PD-1/PD-L1 pathway. Indeed, inflammation at barrier organs is a common feature of many genetic deficiencies in peripheral tolerance mechanisms, including CTLA-4 heterozygosity[26–28].

Figure 2:

Major sites of toxicity from immune checkpoint inhibitors.

Figure 3:

Potential Antigenic Targets in Immune Related Adverse Events. The barrier organs (gastrointestinal tract and liver, lungs, and skin) contain a large quantity of microbial and environmental antigens. Endocrine organs such as the thyroid and pancreas produce a variety of specialized proteins. The heart, skinand skeletal muscle express tissue-restricted proteins, which in the case of melanocyte antigens may be shared with melanoma.

New onset Celiac Disease, a disease of the small intestines caused by T cell recognition of the dietary protein gluten, has been described after ICI treatment[29]. These patients respond to a gluten free diet, consistent with an important role for CTLA-4 and PD-1/PD-L1 in regulating immune responses to gluten[29,30]. Intriguingly, patients who develop ICI enteritis have histopathology that is indistinguishable from those who develop ICI-associated Celiac Disease, suggesting a potentially broader role for these immune checkpoints in regulating responses to dietary proteins[29].

Endocrine toxicities are also common with ICI therapy. Clinically significant thyroiditis occurs in 8% of patients on PD-1 blockade, and hypophysitis occurs in 6% of patients on CTLA-4 inhibitor monotherapy[11]. Other endocrine toxicities, including autoimmune diabetes and adrenal insufficiency, are rare but are extremely important to recognize because they can be deadly[11]. Endocrine organs are frequent targets of many spontaneous autoimmune diseases as well[27,28]. These tissues express many specialized proteins that are not involved in any other processes in the body (e.g., the iodination machinery in the thyroid). Although central tolerance typically eliminates most self-reactive cells, these mechanisms are not perfect and analyses from animal models and humans with autoimmune disease indicate that peripheral tolerance plays a key role in protecting these organs even when central tolerance is functioning normally[25].

The cardiac toxicities may be different. Inflammation in other sterile organs such as the heart or brain are rare, though these toxicities are also among the deadliest when they occur[23]. The targets of these inflammatory responses are almost certainly normal self-proteins for which central tolerance has failed to fully eliminate self-reactive cells. For the barrier toxicities, and likely the endocrine toxicities as well, the immune targets probably are not shared with the tumor cells themselves, though no careful analyses have yet addressed this hypothesis. A recent analysis of myocarditis in both a murine model and in patients found that α-myosin was a direct target of cytotoxic CD8 T cells[19]. Three class I restricted T cell clones reactive to α-myosin were identified in the murine model, which was confirmed to be dependent on CD8 T cells[19]. Like the murine model, α-myosin reactive cells could be expanded from the peripheral blood of patients with ICI induced myocarditis, and these clones could be identified by single cell TCR sequencing of cardiac biopsies[19]. Shared TCRs between the inflamed heart and tumors have been previously identified, though whether these cells are both heart and tumor reactive remains unclear[31].

What is the relationship between ICI toxicities and spontaneous autoimmune disease?

Several ICI toxicities look similar to well-described autoimmune diseases, though whether these diseases are truly the same syndromes is not clear. This is most obvious in endocrine toxicities such as ICI thyroiditis, which resembles Hashimoto’s thyroiditis, and ICI-induced autoimmune pancreatitis, which has clinical features similar to Type 1 Diabetes. In many of these cases, however, the ICI associated clinical syndrome is more fulminate[32].

Understanding the relationship between ICI toxicities and spontaneous autoimmune disease is of both basic science interest and considerable clinical importance. From a basic science perspective, defining how ICI toxicities relate to other autoimmune diseases will provide insights into the mechanisms of autoimmunity with the potential to understand early events in autoimmunity, including the identification of new therapeutic targets and new methods for predicting autoimmune disease risk.

From a clinical perspective, understanding this relationship between autoimmunity and ICI toxicities may help identify potential therapeutic approaches for ICI toxicities based on those already in clinical use for autoimmunity. The resemblance between ICI colitis and inflammatory bowel disease (IBD) was the reason that TNFα blocking antibodies were first used to treat steroid-refractory ICI colitis during the early ICI clinical trials[33,34]. This resemblance was also the reason that integrin inhibitors and IL-12/23p40 antibodies have been investigated with apparent clinical success[35,36].

Patients with some, though not all, autoimmune diseases do appear to have an increased risk for developing ICI toxicities related to their autoimmune disease[37,38]. For rheumatologic diseases, this risk is relatively small[37,38]. Yet in a large retrospective study of patients with IBD, including ulcerative colitis and Crohn’s Disease, treated with ICIs, the risk of gastrointestinal toxicities was approximately 40% compared to 11% for a control cohort of patients on ICIs who did not have IBD[39]. Similarly, patients with microscopic colitis, a distinct inflammatory disease of the colon, appear to be at an increased risk of gastrointestinal toxicities from ICIs[40].

Detailed immune comparisons between ICI toxicity and spontaneous autoimmunity have only begun to be performed[3]. One of the major challenges of these analyses is determining how to select patients and tissues to make appropriate comparisons. Because autoimmune diseases are typically chronic and are often treated with immune suppressive therapy over prolonged periods of time, differences may emerge between autoimmune disease and ICI toxicity that relate to the length of time that the disease has been present or to changes induced by therapy rather than to fundamental differences in disease biology. For example, in a recent single cell RNA sequencing analysis, ICI colitis was compared to Ulcerative Colitis during a flare[16]. While differences in T cell populations were identified, the Ulcerative Colitis patients had all been previously treated and had had their disease for an extended period[16]. Similar comparisons are ongoing for other ICI toxicities, but these analyses will face similar limitations.

What factors determine who will develop an irAE?

Understanding the factors that determine who will develop an irAE is of critical clinical importance. While most patients who are treated with immunotherapy will not develop a treatment-limiting side effect, severe inflammatory toxicities do develop in a subset of patients and these toxicities can be life-threatening[23]. Inflammatory toxicities are a major limitation of combination immunotherapies and will likely limit other investigational combination therapies[12]. Being able to predict specific inflammatory toxicities is an important step in developing prophylactic therapies and may be useful in selecting cancer treatment options.

Detailed immune analyses are now being performed on a variety of ICI toxicities. For ICI colitis, single cell RNA sequencing on colon biopsies at the time of colitis diagnosis identified an expanded cytotoxic CD8+ T cell population producing IFNγ and granzyme B that was not present in immunotherapy treated controls or healthy controls who did not have colitis[16,41] (Figure 4). A smaller population of expanded CD4+ T cells making IFNγ was also seen, as was a population of myeloid cells that were producing and responding to TNFα[41]. B cell populations appeared to be unchanged[41]. TCR sequencing from these cells indicated that the expanded cytotoxic CD8+ T cells were clonally related to colonic resident memory T cells (Trms), a result that was supported by RNA velocity analysis[41]. These results were verified by two subsequent independent analyses, indicating that activation of Trms in the gut is a cardinal feature of ICI colitis[16,41,42]. Yet it remains unclear why this happens in some patients and not in others, as Trms are a part of a normal colonic immune system. Trms are thought to largely recognize antigens from the microbiome, and perhaps the interaction between Trms and specific microbial products determines whether the balance of immune homeostasis tips toward inflammation in some patients[43]. This interaction could be influenced by changes in diet, other medications, or concurrent infections. Intriguingly, limited data suggest that fecal microbiota transplant can resolve refractory ICI colitis further implicating the microbiome in this syndrome[44].

Figure 4:

Proposed and cellular mechanisms of immune checkpoint inhibitor (ICI) colitis. Resident memory CD8 T cells (Trm) express CTLA-4 and PD-1 and are in a quiescent state in the healthy colon. The colon also contains abundant regulatory T cells (Tregs) which also express CTLA-4. Upon treatment with CTLA-4 or PD-1 inhibitors, CD8+ Trms become activated, proliferate and produce granzyme B (GZMB) and IFNγ potentially in response to recognition of microbial antigens from the microbiome. This response damages the colonic epithelial cells causing changes in permeability, cell death, and ulceration of the mucosa. Tregs also proliferate though the CTLA-4 on their surface is blocked, reducing their suppressive capacity. IFNγ activates myeloid cells such as tissue macrophages that then secrete TNFα that acts in both a paracrine and autocrine fashion. These macrophages also produce additional inflammatory cytokines and chemokines which can recruit additional activated T cells from the blood, including gut homing CD4+ T cells. Potential treatments for ICI colitis include glucocorticoids which have multiple affects including induction of apoptosis in activated T cells, antibodies to TNFα, integrin inhibitors that prevent trafficking of additional T cells from the blood, JAKi that block IFNg signaling, and CTLA-4-Ig (e.g. abatacept) which can both inhibit priming of new T cells in the gut draining lymph nodes and potentially interfere with ongoing interactions between B71/2 and CD28 on activated T cells in the colonic mucosa.

The colitis-associated CD4+ T cells in ICI colitis are not clonally related to the CD4+ Trms, suggesting that they largely enter the colon from outside the gut[41]. This is consistent with the ability of gut homing integrin inhibitors to treat ICI colitis, which would not be expected if the inflammation was driven entirely by gut resident cells. Intriguingly, a recent analysis of immune correlates of ICI toxicity, which included a large population of patients with ICI colitis, found that elevated circulating CD4+ resident memory T cells significantly increased the risk of developing a toxicity from ICIs[45,46].

Whether the lessons we have learned from ICI colitis are generalizable to other toxicities is a major open question. Trms are present in each of the barrier organs but are not found in large numbers in the so called “sterile” organs such as the endocrine tissues or cardiac muscle Detailed analyses from a variety of toxicities are necessary to understand whether different toxicities have distinct immunologic origins. In addition, independently of ICI toxicity outcome, performing pre-treatment, organ-level immune analyses in patients will be critical to establish whether baseline immunologic differences are present that subsequently lead to irAEs.

How do we select optimal treatment strategies for irAEs?

The great majority of patients who develop an inflammatory toxicity from ICI therapy will respond to systemic glucocorticoids[17]. Early use of these drugs has almost certainly reduced the morbidity and mortality associated with irAEs, particularly those that occur in critical organs such as the heart and nervous system[47]. Nevertheless, glucocorticoids are not always effective and may also pose risks. In addition to the abundant side effects associated with systemic glucocorticoids, they are also broadly immunosuppressive and may interfere in antitumor responses[1].

Data on the influence of systemic glucocorticoids on antitumor immunity is currently mixed, and no high quality, prospective studies are available. Several studies have shown that patients who receive glucocorticoids for the management of irAEs have overall survival that is similar to patients who do not develop irAEs[48]. Although these studies indicate that systemic glucocorticoids do not erase the therapeutic benefit of ICI therapy if given after treatment onset, other studies have suggested that high dose glucocorticoids can prevent adequate ICI responses if they are in use at the onset of immunotherapy[49–51].

One of the major limitations of current evidence is that antitumor responses and the development of irAEs may not be mechanistically distinct, that is, patients who develop irAEs may be more likely to develop robust antitumor responses. The use of glucocorticoids in these patients may then be inhibiting some of this mechanistic “synergy” even if the glucocorticoids do not fully eliminate antitumor immunity. Although speculative, the best evidence supporting this conjecture comes from patients who developed hypophysitis after ipilimumab treatment for melanoma[52]. In a retrospective analysis looking at patients who were treated with low dose versus high dose glucocorticoids, patients treated with low dose glucocorticoids were found to have considerably extended overall survival and time to treatment failure compared to patients who received high dose glucocorticoids[52]. Similar results were found in a two-center retrospective analysis that looked at irAEs more generally and segregating patients based on whether they did or did not receive early high dose systemic glucocorticoids[53].

Ultimately, this question cannot be settled by observational data, and will require randomized trials where different irAE treatment approaches are compared to each other. For example, an ongoing study is comparing first line systemic glucocorticoids to first line infliximab for the treatment of ICI colitis and may provide the first prospective, controlled data on the influence of irAE treatment modality on antitumor outcomes (NCT04305145).

Considerable basic science data are available to address how immune suppressive treatments influence antitumor immunity. High dose glucocorticoids can induce apoptosis in activated T cells and inhibit important elements of T cell mediated immunity[1,3]. In animal models, blockade of TNFα is associated with improved antitumor responses, similar to what is seen with IL-6 blockade[54,55]. Neither cytokine is correlated with effective antitumor immunity in analyses of immune correlates of antitumor response in patients, and mutations in downstream signaling from these cytokines are not seen as resistance mutations in patients with acquired resistance to ICIs[56–63]. In contrast mutations in signaling downstream of IFNγ are frequently seen in animal models and in patients with resistance to ICI therapy[56–63]. IFNγ signaling is targeted by several agents, including the JAK inhibitor tofacitinib, which appears to be effective in ICI colitis, but also likely interferes directly with antitumor responses[16,64]. Similarly, CTLA-4 immunoglobulin (Ig) fusion proteins (e.g., abatacept) have been proposed as treatment of ICI toxicity, particular myocarditis[65,66]. While these medications have a high likelihood of being effective in managing ICI toxicity, they are also likely to directly reverse the activity of the ICIs themselves. Though this is certainly appropriate for life threatening toxicities such as myocarditis, more general use of CTLA-4-Ig for the management of irAEs is probably not warranted.

Concluding Remarks

Optimizing the management of irAEs from cancer immunotherapy is an urgent clinical problem and we continue to have multiple outstanding questions (see Outstanding Questions Box). While guidelines for management of irAEs have been developed, these are based almost entirely on expert opinion and small retrospective clinical studies. Prospective treatment trials are currently lacking, and the establishment of these trials will be an important next step in the field. As the mechanisms, severity, and response to treatment of each toxicity will likely differ, these trials should involve distinct toxicities rather than “bucket” multiple toxicities together. We have only begun to understand the mechanisms that drive these toxicities, which is a necessary first step in developing novel treatments. Understanding the molecular and cellular mechanisms of irAEs also promises to provide new insights into autoimmunity and immune homeostasis in people. The field of irAE research is rapidly growing and progress within the field will be essential to advancing the field of cancer immunotherapy more generally.

Outstanding Questions.

• How does the class of immune checkpoint inhibitor influence irAE development?

• Why are some organ systems more frequently targeted by irAEs?

• What is the relationship between ICI toxicities and spontaneous autoimmune diseases?

• What factors determine who will develop an irAE?

• How do we select optimal treatment strategies for irAEs?

Highlights.

• Inflammatory toxicities are an important limitation on all checkpoint inhibitor therapies

• The spectrum of toxicities differs depending on the specific checkpoint pathway inhibited (e.g. CTLA-4 versus PD-1)

• Checkpoint inhibitor toxicities at mucosal surfaces are associated with expansion and activation of resident memory cells

• The immune targets of checkpoint inhibitor toxicities are unknown and may be toxicity dependent (e.g. the microbiome for colitis and muscles proteins for myocarditis)

• Circulating effector memory cells may indicate an elevated risk of checkpoint inhibitor toxicity