Assessing Immune-Related Adverse Events of Efficacious Combination Immunotherapies in Preclinical Models of Cancer

The combination of ipilimumab [anti-cytotoxic T-lymphocyte antigen 4 (CTLA-4)] and nivolumab [anti-programmed cell death protein 1 (PD-1)] to target T-cell checkpoint receptors in the treatment of advanced melanoma has produced rapid and impressive anticancer effects (1, 2) and may result in significant efficacy against other cancers. These clinical data, together with results from preclinical mouse tumor models, demonstrate that multiple immunosuppressive pathways exist in tumors and that their cotargeting can increase the efficacy of host antitumor immunity. Thus, different combination immunotherapies are currently being tested to further increase clinical efficacy in a greater proportion of patients (3). However, a key feature and problem with targeting immune checkpoint receptors, particularly in combination, is the occurrence of immune-related adverse events (irAE; ref. 4).

IrAEs occur due to therapy-associated cytokine release and T-cell–mediated organ infiltration (4). In particular, grade 3/4 irAEs are a major cause for concern as a proportion of patients discontinue treatment and fatalities can occur if not promptly treated. In patients treated concurrently with ipilimumab and nivolumab, about 50% developed grade 3/4 irAEs (2) compared with patients treated with nivolumab (∼14%; ref. 5) or ipilimumab (∼20%–25%; ref. 6) alone. The most common grade 3/4 irAEs were colitis, diarrhea, transaminitis, and dermatitis (1, 2). One can predict that irAEs frequency and severity will increase with the myriad of combination immunotherapies currently being tested, thus potentially limiting their clinical utility. Indeed, fatal toxicities from single-agent therapy can hold back its development even if they display promising antitumor efficacy clinically. One example is urelumab, an agonistic antibody targeting CD137, which induced dose-dependent fatal hepatotoxicity. This led to the termination of a number of clinical trials including those that were assessing urelumab in combination (7). With the FDA approval of anti-CTLA-4 and anti-PD-1 mono- and combination therapy, irAEs will increase, including rare irAEs, as patient numbers increase from hundreds to the thousands (8). In addition, there is a pipeline of reagents targeting novel receptors that are currently being evaluated preclinically and those that demonstrate efficacy will make a compelling case to be tested clinically alone and in combination. However, there are currently no mouse models that develop a spectrum of clinical irAEs following immunotherapy. Cancer immunotherapy needs such translational vehicles to predict the antitumor efficacy and the severity of irAEs, that is, testing the therapeutic window of novel treatments alone and in combination in a timely manner. Indeed, budgetary, legal, and regulatory issues limit the number of different combination therapies that can be evaluated at any one time and the length of time required for the antitumor efficacy of that combination to become evident.

IrAEs mimicking clinical symptoms generally have not been observed in mouse models even when 3 different immune pathways were targeted (9). This may be due to the generally short time frame of preclinical mouse tumor models or the resistance of certain mouse strains to the induction of irAEs (4). Treated mice may have displayed biochemical changes consistent with autoimmunity but outwardly appeared healthy as has been observed in many patients treated with immune checkpoint inhibitors (10). While experimental mice are inbred and mostly immunologically-naïve, humans treated with immunomodulating antibodies are outbred and exposed to a large universe of antigens over many years before they are treated. Therefore, there is a need to lower immune tolerance in mice to be able to correctly detect irAEs induced by immunomodulatory antibodies. Indeed, the efficacy of immune checkpoint inhibitors and its associated irAEs may be partially attributed to depletion or attenuation of Tregs. In humans, increased Treg number/proportion has been associated with worse prognosis in some cancers (11, 12). Tregs can suppress antitumor immunity through inhibiting the effector function of many immune cell types (13). Preclinically, some of the antitumor efficacy of anti-CTLA-4 has been shown to be mediated through Treg depletion (14). In humans, the mechanism of action of ipilimumab is not fully understood, although a recent report in a patient with melanoma demonstrated anti-CTLA-4 mediated depletion of tumor-infiltrating Tregs via antibody-dependent cellular cytotoxicity through CD16-expressing monocytes (15). In addition, a number of studies have demonstrated that PD-1 blockade on Tregs decreased their suppressive function (16–18).

To better mimic the clinical symptoms and biochemical immunopathology observed in patients following immune checkpoint blockade, we utilized the Foxp3-GFP-DTR mouse to lower immune self-tolerance (19). Prolonged Treg depletion in these mice induces rapid fatal autoimmune disease similar to the most severe grade 3/4 clinical irAEs. We utilized 2 strains of Foxp3-DTR mice and 3 different tumor models to study the mechanism, efficacy, and safety of transient Treg depletion in combination with monoclonal antibodies (mAb) targeting immunomodulatory receptors PD-1, TIM-3, or CD137.

Inbred C57BL/6 Foxp3-DTR-GFP were kindly provided by Dr. Geoffrey Hill (19), whereas BALB/c Foxp3-DTR-GFP mice were generated by backcrossing C57BL/6 Foxp3-DTR-GFP mice 10 times to BALB/c WT mice. All mice were bred and maintained at the QIMR Berghofer Medical Research Institute. Six- to 12-week-old mice were used in all experiments and performed in accordance to QIMR Berghofer Medical Research Institute animal experimental ethics committee guidelines. In addition, mice were scored for clinical symptoms of illness with changes to weight, posture, activity, and fur texture monitored with mice euthanized when clinical symptoms reached the cumulative limit outlined by animal ethics. Blepharitis was also scored in some mice, with mice given a score of 0 to 3 based on disease severity as defined previously (20).

BALB/c-derived 4T1.2 mammary carcinoma, CT26L5 carcinogen-induced colon carcinoma, and C57BL/6 MC38 colon adenocarcinoma or E0771 mammary carcinoma cell lines were maintained in RPMI-1640 or DMEM supplemented with 10% FCS, penicillin/streptomycin, and l-glutamine as previously described (21–23). All cell lines were obtained between 2000 and 2008 and were routinely tested as negative for mycoplasma. Cell line authentication was not routinely performed.

Foxp3-DTR-GFP mice were injected subcutaneously with 4T1.2 (5 × 10⁴), CT26L5 (2 × 10⁵), MC38 (1 × 10⁶), or E0771 (5 × 10⁵) tumor cells. Tumor growth was measured by caliper square measurements and treatment started when the mean tumor size reached 40 to 50 mm². Mice were euthanized when tumor size reached 150 mm². To deplete Tregs, mice were injected intraperitoneally with 250 ng of diphtheria toxin (DT; Sigma-Aldrich), diluted in PBS. Some mice additionally received treatment at the indicated schedule with rat control IgG2α (2A3 or 1-117), anti-CD137 (3H3), anti-PD-1 (RMP1-14), anti-TIM-3 (RMT3-23; all from BioXCell or Leinco; each 250 μg/mouse, intraperitoneally), anti-CTLA-4 (9D9; Bristol Myers Squibb), anti-CD8β (53-5.8; BioXCell), anti-asGM1 (WAKO; each 100 μg/mouse, intraperitoneally). For anti-TNF treatment, groups of mice were intraperitoneally treated with 200 μg control IgG (2A3) or anti-TNF (MP6-XT3; BioXCell) as indicated.

Single-cell suspensions were generated from indicated organs as previously described (24). For surface staining, cells were stained with anti-CD45.2 APC-eFluor780 (104), anti-TCRβ PerCP-Cy5.5 (H57-597), anti-CD8α Brilliant Violet 711 (53-6.7), anti-CD4 Brilliant Violet 605 (RM4-5), biotin-CD137 (17B5), followed by Streptavidin-APC (BD Biosciences), anti-PD-1 PE-Cy7 (J43), anti-TIM-3 PE (RMT3-23 and 8B.2C12 for C57BL/6 and BALB/c mice, respectively) and Zombie Aqua (all from eBioscience or Biolegend) in the presence of 2.4G2 (anti-CD16/32, to block Fc receptors) on ice. To stain for Ki67 or Foxp3, samples were fixed and permeabilized with Foxp3 staining kit (eBioscience) before being stained with anti-Ki67 Alexa647 (B56) (BD Pharmingen) or anti-Foxp3 eFluor 450 (FLK-16s; eBioscience). To determine intracellular cytokine levels, single-cell suspensions were incubated for 4 hours in complete RPMI with monensin and brefeldin A (eBioscience) alone, or with PMA/ionomycin stimulation (eBioscience cell stimulation cocktail). Samples were then surface stained before being fixed/permeabilized (BD CytoFix/CytoPerm Kit) and stained with anti-IFNγ APC (XMG1.2) and anti-TNF PE (MP6-XT22) or their respective isotypes (Biolegend). To determine absolute counts in samples, liquid counting beads (BD Biosciences) were added directly before samples were ran on the flow cytometer. All data were collected on a Fortessa 4 (Becton Dickinson) flow cytometer and analyzed with FlowJo v10 software (Tree Star, Inc.).

Mice were bled either from the retro-orbital sinus or by cardiac puncture, allowed to clot, and centrifuged at 10,000 rpm for 10 minutes to separate sera. Cytokine levels were determined using mouse cytometric bead array (CBA) as per manufacturer's instructions (BD Biosciences).

Mouse tissues were perfusion fixed in 10% neutral-buffered formalin overnight, processed routinely, and embedded in paraffin. Four-micrometer thick sections were cut and stained with hematoxylin and eosin (H&E). H&E-stained tissue sections were imaged using Aperio Scanscope AT (Leica) and analyzed by Aperio ImageScope. The pathology of mouse colon and liver sections were scored by referring to McGuckin's (25) or Sparwasser's (20) standards respectively. The liver score is the sum of individual scores for inflammatory cell infiltration in portal tracts, parenchyma, and necrosis.

Sera alanine aminotransferase (ALT) levels were measured by Queensland Pathology using a Beckman Unicell DxC800 analyzer.

Statistical analysis was performed using GraphPad Prism software. Differences in tumor growth were determined by Mann–Whitney U test or a log-rank test (Mantel–Cox) for survival analysis. Differences between measurements in groups were determined by Student t test, a one-way ANOVA, or two-way ANOVA with Dunnett posttest analysis as indicated. P ≤ 0.05 was considered significant with P ≤ 0.05 indicated with (*), P ≤ 0.01 with (**), P ≤ 0.001 with (***), and P ≤ 0.0001 indicated with (****). Where data are presented on a log₁₀ scale, statistical comparison was also performed on this scale.

One mechanism by which anti-CTLA-4 and anti-PD-1 mediate their function is through depleting or attenuating Treg function. We therefore developed a model that can predict the therapeutic window to any additional therapy after the Treg:effector T-cell ratio has been lowered. Using the Foxp3-DTR mice, we asked how transient (1 dose DT) or prolonged (2 or 5 doses DT) Treg depletion impacted on tumor growth and development of autoimmune toxicity. Having optimized the dose of DT required to maximally deplete Tregs (Supplementary Fig. S1), we determined how transient or prolonged Treg depletion in BALB/c or C57BL/6 Foxp3-DTR mice affected tumor growth in 4 different tumor models (Fig. 1). Overall, both transient and prolonged Treg depletion suppressed 4T1.2 (Fig. 1A), CT26L5 (Fig. 1B), MC38 (Fig. 1C), E0771 (Fig. 1D) tumor growth similarly and sometimes completely, although the effect was tumor-dependent. However, prolonged Treg-depleted mice developed overt signs of illness, including blepharitis (chronic inflammation of the eyelid; Supplementary Fig. S2), weight loss, and reduced mobility, which necessitated their euthanasia. In contrast, transient Treg-depleted tumor-bearing mice displayed no overt signs of illness. Next, we performed comprehensive analysis on various organs and sera of 4T1.2 tumor–bearing mice and demonstrated that transient and prolonged Treg depletion in these mice mimicked low- and high-grade clinical irAEs, respectively (Fig. 2). This included increased lymphocytic and mononuclear immune infiltrates in the indicated organs (Fig. 2A), increased spleen and colon weights (Fig. 2B and C), an expansion of effector T cells (Fig. 2D), and an increase in IFNγ and TNF in the sera (Fig. 2E). In addition, we observed raised levels of anti-nuclear antibodies (ANA), total IgG, anti–double-stranded DNA (anti-dsDNA) antibodies of IgG and IgM isotypes, and prominent IgG depositions in the kidney glomeruli (Fig. 2F–J; methods described in Supplementary Text). Similar results were seen in MC38 tumor–bearing mice (Supplementary Fig. S3), although 5 doses of DT were required to induce severe irAEs. Finally, we demonstrated the relative equivalency between tumor-bearing mice receiving transient Treg depletion or multiple anti-CTLA-4/PD-1 therapy in lowering the immune tolerance threshold (Supplementary Fig. S4). Similar to transient Treg-depleted mice, anti-CTLA-4/PD-1–treated mice displayed no overt signs of illness while displaying varying immunologic changes. The equivalency between transient Treg depletion and anti-CTLA-4/PD-1 therapy in lowering the immune tolerance threshold in mice allowed us to evaluate the therapeutic window of treatment in a standard way.

Given that transient Treg depletion suppressed 4T1.2 and MC38 tumor growth for a significant period of time, we next evaluated in these models how the addition of anti-PD-1, -TIM-3, or -CD137 further regulated tumor growth and irAEs development (Fig. 3). Anti-PD-1 mAbs have proven utility in humans, whereas TIM-3 (T-cell immunoglobulin and mucin-domain containing-3) is another inhibitory receptor and preclinically, anti-TIM-3 alone has antitumor function and synergizes well with anti-PD-1 or Treg depletion (23, 26). CD137 is a costimulatory receptor expressed on many immune cells including activated T and NK cells (27). In clinical trials, anti-CD137 alone has demonstrated some efficacy although with unacceptable liver toxicities at higher doses (27).

First, we determined the level of PD-1, TIM-3, and CD137 expression on Tregs and effector T cells in the tumors and spleen of tumor-bearing mice 3 days after Treg depletion (when antibody treatment would commence; Supplementary Fig. S5). While expression of these receptors generally increased in the spleen following Treg depletion, it did not alter in the tumor microenvironment. In both 4T1.2 and CT26L5 tumor–bearing mice, DT plus anti-TIM-3 suppressed tumor growth most effectively (2 of 5 and 4 of 5 cures, respectively) compared with the other combinations (Fig. 3A, Supplementary Figs. S6A–S6C and S7). This improved efficacy appeared to be associated with a corresponding increase in splenic CD4⁺ and CD8⁺ IFNγ⁺ T cells (Fig. 3B). Overall, combination therapies were more effective in suppressing tumor growth compared with treatment with one or multiple DT or antibodies alone. Interestingly, DT plus anti-PD-1 or anti-TIM-3–treated mice showed minimal or no overt signs of illness, respectively. In both 4T1.2 and CT26L5 tumor models, we observed skin inflammation in the ear of mice treated with DT plus anti-PD-1 but not in those treated with DT plus anti-TIM-3. In contrast, DT plus anti-CD137–treated mice appeared physically ill and displayed inflammation in the ear and blepharitis. These symptoms were similar to the multiple DT-treated groups, although not as severe as euthanasia rarely occurred. Cessation of anti-CD137 therapy led to a decrease in skin inflammation, although the blepharitis remained.

In contrast, in MC38 tumor–bearing mice, both DT plus anti-PD-1 and anti-CD137 were equally effective (5 of 5 and 5 of 6 cures, respectively) compared with DT plus anti-TIM-3 (0 of 5 cures; Fig. 3C and Supplementary Fig. S6D–S6F). The improved efficacy of the DT plus anti-PD-1 or anti-CD137 combinations also appeared to be associated with an increase in CD8⁺ IFNγ⁺ T cells (Fig. 3D). Similar to the 4T1.2 tumor model, only MC38 tumor–bearing mice treated with DT plus anti-CD137 presented with overt signs of illness, most notably weight loss and mild blepharitis. However, this was less severe than multiple DT-treated mice, as these mice recovered following treatment cessation.

To determine whether DT plus anti-CD137–treated mice developed the same spectrum of irAEs as prolonged Treg-depleted mice, organs and sera were harvested from 4T1.2 or MC38 tumor–bearing mice after treatment with DT followed by 3 doses of immunomodulatory mAbs, as mice that received prolonged Treg depletion had developed visible irAEs by this stage (Figs. 4 and 5 and Supplementary Fig. S8). DT plus anti-CD137–treated mice had obvious organ immune infiltrates (Fig. 4A and E) as reflected by a significant increase in organ weights, histologic scores, and ALT levels (Fig. 4B–D and F–H). An increase in the colon and liver histologic score of DT plus anti-PD-1–treated 4T1.2 tumor–bearing mice was also observed, although other parameters were not changed (Fig. 4C and D). Examination of liver sections revealed a liver pathology of acute hepatitis, with portal and periportal inflammation without destruction of bile ducts or interface membrane. In contrast, DT plus anti-TIM-3–treated mice generally displayed similar levels of immune infiltrates as DT plus control IgG–treated mice (Fig. 4). In DT plus anti-CD137–treated mice, we also observed a significant increase in proliferating CD8⁺ T cells (Fig. 5 and Supplementary Fig. S9) and serum levels of IFNγ and TNF (Supplementary Fig. S10). Overall, transient Treg depletion plus anti-TIM-3 or anti-PD-1 provided the best therapeutic window (high antitumor efficacy and low toxicities) depending on the tumor model. In contrast, across 3 tumor models, transient Treg depletion plus anti-CD137 had a lower therapeutic window.

In the clinic, severe irAEs are commonly managed by cessation of therapy and/or treatment with corticosteroids or TNF-blocking antibodies (28). Given that DT plus anti-CD137 treatment increased TNF levels in both 4T1.2 and MC38 tumor–bearing mice, we asked whether neutralization of TNF in these mice could prevent severe irAEs (Fig. 6). During the treatment period, we observed no difference in tumor growth suppression between mice treated with DT plus anti-CD137 and anti-TNF or control IgG (Fig. 6A) or between groups that received prolonged Treg depletion and anti-TNF or control IgG (Fig. 6B). Importantly, DT plus anti-CD137–treated mice that received anti-TNF appeared physically less ill compared with their control IgG–treated counterparts. In general, these mice displayed lower inflammation in the skin of their ears and did not develop blepharitis (Fig. 6C). Surprisingly, TNF blockade did not reduce the severity of irAEs in prolonged Treg-depleted mice, as they became physically ill and the severity of their blepharitis was not changed (Fig. 6D). We next set up a similar experiment as above where one day after their last anti-CD137 treatment, organs and sera were harvested to confirm whether immunopathology was attenuated. We observed a significant decrease in spleen and colon weights (Fig. 6E and F). Importantly, the improved physical health of the anti-TNF–treated mice was reflected by lower levels of ALT compared with control IgG–treated groups and comparable to that of naïve mice (Fig. 6G). Similarly, histologic analysis revealed lower levels of immune infiltration in the liver, proximal colon, skin, and spleen (Fig. 6H).

To investigate which immune effector cells were mediating the antitumor effect and irAEs, we depleted CD8 or NK cells in DT plus anti-CD137–treated MC38 tumor–bearing mice (Fig. 7A–D). CD8⁺ T cells predominantly mediated the antitumor effect as all tumors grew out when they were depleted (Fig. 7C), whereas NK cell depletion did not impact on tumor rejection (Fig. 7D). Given that liver toxicity is the most clinically relevant irAE induced from anti-CD137 therapy, we examined serum ALT levels and liver pathology in these mice. CD8⁺ T cells were likely the main cause of liver pathology in this model, as both parameters were significantly reduced when they were depleted (Fig. 7E and F). In contrast, neither ALT levels nor histologic score were attenuated when natural killer (NK) cells were depleted. Overall, this model mimics what is observed in the clinic where TNF blockade can alleviate the severity of irAEs from anti-CTLA-4 and anti-PD-1 and suggests anti-TNF therapy may be effective to reduce irAEs in anti-CD137 therapy without impacting on antitumor efficacy.

In this study, we developed a preclinical mouse model that allowed the antitumor efficacy and irAEs of antibodies targeting various immunomodulatory receptors to be simultaneously assessed for the first time. Overall, transient or prolonged depletion of Tregs in Foxp3-DTR mice bearing different tumor types resulted in either growth suppression and/or cures while inducing mild or severe irAEs, respectively. These phenotypes mimic the range of responses and severity of irAEs that can occur in patients treated with ipilimumab/nivolumab alone. The equivalency between transient Treg depletion and anti-CTLA-4/PD-1 therapy in lowering the immune tolerance threshold in mice allowed irAEs to be more easily induced following treatment with additional immunomodulatory antibodies. Using transient Treg depletion with DT, we showed that anti-PD-1 or anti-TIM-3 mAbs have a higher therapeutic window than anti-CD137. While all 3 mAbs suppressed tumor growth, tumor rejection was more frequently observed in the DT plus anti-PD-1 or anti-TIM-3 combinations, depending on the tumor type, and these mice developed only mild irAEs. In contrast, DT plus anti-CD137–treated mice developed severe irAEs. These irAEs appeared because of an infiltration of activated proliferating effector T cells in tissues producing IFNγ and TNF. However, TNF blockade decreased the severity of irAEs in these mice without impacting on tumor growth.

Previous approaches to measure development of irAEs after immunotherapies have utilized strains of mice on an autoimmune background (29), aged mice (30), or the use of cyclophosphamide to deplete Tregs (31). However, our approach is simpler and cheaper in that young mice can be used and irAEs developed rapidly after commencement of treatment, similar to the clinical setting. Furthermore, a diverse range of experimental tumors derived from the C57BL/6 and BALB/c strains are available to model different cancers. Given that tissues in different anatomical sites sculpt and vary the tumor microenvironment to affect responses to immunotherapy (32), future work will now investigate whether the therapeutic window of different combination immunotherapies remains the same for the same tumor implanted orthotopically. We would also propose that the Foxp3-DTR mice can be used as an initial screen to assess the therapeutic window of novel immunotherapies followed by confirmation with anti-CTLA-4 plus anti-PD-1 given that complete Treg depletion is currently not feasible clinically. Interestingly, patients who discontinued treatment because of irAEs can demonstrate durable responses (2). We can now interrogate how changing the dose and scheduling frequency of the combination immunotherapies used in this study impacts on its therapeutic window, given these parameters have not been optimized for immune checkpoint inhibitors in the clinic.

Agonistic CD137 mAbs have previously been shown to enhance cytotoxic T-cell function and demonstrated antitumor efficacy in many preclinical tumor models (7). However, agonistic CD137 mAb has also been reported to suppress T-cell–dependent humoral immunity and reverse the course of established autoimmune disease such as rheumatoid arthritis, experimental autoimmune encephalomyelitis, and systemic lupus erythematosus (7). Thus, it was suggested that CD137 possesses dual immunoregulatory activity and therefore may enhance antitumor activity without autoimmune side effects associated with immunotherapy approaches (7). However, this contrasts with the toxicities that we and others have observed, preclinically and clinically. It was previously reported that injection of anti-CD137 into naïve mice induced liver toxicities that resolved upon cessation of therapy. However, it was not mentioned whether these mice displayed overt signs of illness during the course of treatment (33). In our studies, the development of visible irAEs including inflammation of the skin in the ear and eyelid occurred within one dose of anti-CD137 following transient Treg depletion. In contrast, no overt signs of illness were displayed by tumor-bearing mice treated with adoptive T-cell transfer in combination with one dose of anti-CD137 (34). Although there was no change in hepatic function, a single dose of anti-CD137 (clone 1D8) increased the content of liver CD8⁺ T, NKT, and NK cells. Similarly, another study reported that anti-CD137 led to mononuclear inflammation in the portal spaces of the liver and marked increase in CD8⁺ T-cell numbers (35). Our study confirmed these findings and further suggested that CD8⁺ T cells were the main contributor to the liver toxicity induced by anti-CD137 therapy in MC38 tumor–bearing mice. It was previously reported that anti-CD137 limited autoantibody production induced by CTLA-4 blockade in MC38 tumor–bearing mice (36). However, we did not observe a decrease in anti-dsDNA titers in our MC38 tumor–bearing mice treated with DT plus anti-CD137, although they were significantly decreased in anti-CD137–treated 4T1.2 tumor–bearing mice. Potentially, this may be due to the clone of anti-CD137 used, which may have different strength in agonistic activity or the period of antibody treatment. Importantly, this also suggests that pathways that control bona fide autoimmunity may be distinct to those that induce irAEs.

Currently, the efficacy and safety of 2 anti-CD137 mAbs [urelumab (BMS-663513); IgG4, PF-05082566; IgG2] are being evaluated in a number of clinical trials (7). While a phase II study of urelumab was terminated because of fatal hepatotoxicity, low-dose urelumab in combination with anti-PD-1 or anti-CD20 are now being assessed in advanced solid and hematologic malignancies (NCT02253992; NCT01775631; NCT01307267; NCT02179918). In contrast, phase I trials of another anti-CD137 mAb (PF-05082566) suggested it was well-tolerated, with evidence of disease stabilization in multiple patients (7). Nevertheless, given that DT plus anti-CD137–treated mice developed dermatitis, colitis, and liver toxicities, we predict severe irAEs may develop in patients who receive the dose-escalated anti-PD-1 and anti-CD137 combination and these symptoms should be closely monitored for. Given CD137 expression was much higher on Tregs within spleen and tumor compared with effector T cells in our tumor models, this may potentially explain the severity of irAEs observed in these mice, although the clone of anti-CD137 used in our studies is not known to crosslink FcR. While a number of contradictory preclinical studies have reported that anti-CD137 may deplete or attenuate Treg-suppressive function positively or negatively (7, 37, 38), a recent clinical data reported a decrease in regulatory CD4⁺ T cells after anti-CD137 therapy (7). It will be of interest to assess how anti-CD137 attenuates Treg function in future preclinical experiments and in the clinic. Moving forward, we can now determine whether the therapeutic window of DT plus anti-CD137 can be improved by co-administration of anti-TNF, using lower doses of anti-CD137, changing the route of administration from systemic to intratumor (39) or using alternative scheduling.

From our studies, DT plus anti-TIM-3 appeared the safest combination, as it induced the mildest irAEs. Importantly, this combination was also the most efficacious against 4T1.2 and CT26L5 tumors. The safety of anti-TIM-3 treatment may be due to their selectively higher expression on intratumor but not peripheral lymphocytes (26), and indeed, TIM-3 seems to be expressed primarily on intratumor T cells in cancer patients (40). Unlike CTLA-4- or PD-1–deficient mice (41, 42), TIM-3–deficient mice do not exhibit autoimmunity (43). This together with the antitumor efficacy and safety profile observed with anti-TIM-3 in our study provides a compelling case to target TIM-3 clinically. A phase I-b/II clinical trial has recently commenced to evaluate the safety and efficacy of anti-TIM-3 (MBG453, Novartis) alone or in combination with anti-PD-1 in advanced solid tumors (NCT02608268). Further studies are now required to decipher in which tumor type this combination will be most effective.

Currently, patients with active, known, or suspected autoimmune disease are generally excluded from immune checkpoint blockade as these therapies may exacerbate or trigger any underlying autoimmunity (44). Recently, a patient with melanoma was reported to have developed autoimmune diabetes following anti-PD-1 therapy and subsequent HLA typing found that this patient had an autoimmune diabetes high-risk genotype (45). In the autoimmune-prone NOD strain of mice, it has been shown that blocking PD-1-PD-L1 (programmed cell death ligand 1) signaling accelerates onset of type 1 diabetes (29). In contrast, in another report, patients with melanoma with ulcerative colitis responded to anti-CTLA-4 either with no worsening or with concomitant flaring of autoimmunity (46–48). Interestingly, we observed a significant increase in dsDNA titers in our MC38 tumor–bearing mice treated with DT plus anti-PD-1 and a trend for increased ANA, although not in the 4T1.2 tumor–bearing mice. This suggests that the genetics of the immune system may play a role in the type and location of pathologies that can be induced by antibodies targeting immunomodulatory receptors. In another study, a patient with melanoma treated with anti-PD-1 developed autoimmune myocarditis due to infiltration of CD8⁺ T cells, resulting in severe heart failure (49). BALB/c, but not C57BL/6, mice deficient for PD-1 have been reported to develop T-cell–mediated myocarditis (42, 50, 51). Potentially, this is an irAE that may be more prominent in patients with certain genetic profiles and may be exacerbated following anti-PD-1 and anti-CD137 treatment. Using the two different strains of Foxp3-DTR mice, we can now start to interrogate these issues.

In our study, 4T1.2 tumor–bearing mice that received prolonged Treg depletion and anti-TNF did not have a decrease in their irAEs compared with those that received DT plus anti-CD137 and anti-TNF. It has been suggested that TNF can promote or suppress autoimmune diseases under different circumstances (52). Therefore, the failure of anti-TNF to mitigate the severity of irAEs induced by long-term Treg depletion suggests other contributing factors, including autoantibodies produced by B cells. Our model now allows us to ask whether there are distinct or shared immune pathways that control development of irAEs and antitumor immunity and whether it is similarly or differentially activated depending on the combination immunotherapies used. For example, both IFNγ and TNF were significantly upregulated in the sera of DT plus anti-CD137–treated mice. Although, anti-TNF did not appear to impact on antitumor efficacy, one would postulate that neutralizing IFNγ might reduce irAEs severity but simultaneously reduce antitumor immune response. Future experiments to deplete different immune cells and cytokines will allow the dissection of these pathways to be made. Currently, combination therapies targeting T and myeloid pathways are being assessed in cancers that do not respond to co-blockade of checkpoint receptors and we can use our model to measure the therapeutic window of such combinations.

The next frontier in cancer immunotherapy lies in combination approaches and this can potentially benefit a greater proportion of patients with cancer. However, clinicians are currently faced with the dilemma of what combination immunotherapies to test in different cancers. Indeed the rationale for some combination immunotherapies currently in clinical trials may not be selected on the basis of scientific evidence but more because of their availability (3). Filling a need, we have developed a model that may be used to preclinically assess the therapeutic window of novel immunotherapy combinations in different tumor types to aid clinicians and pharmaceutical companies weigh up their risk/cost–benefit profile. Furthermore, our model offers an opportunity to dissect whether the molecular pathways governing the development of antitumor immunity and irAEs are related or distinct to allow more specific targeting.

M.J. Smyth reports receiving commercial research grant from Bristol Myers Squibb and Medimmune. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J. Liu, M.J. Smyth, M.W.L. Teng

Development of methodology: J. Liu, S.J. Blake, K.A. Fairfax

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Liu, S.J. Blake, M.C.R. Yong, M.J. Smyth

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Liu, S.J. Blake, K.A. Fairfax, M.J. Smyth, M.W.L. Teng

Writing, review, and/or revision of the manuscript: J. Liu, S.J. Blake, K.A. Fairfax, S. Allen, M.J. Smyth, M.W.L. Teng

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Liu, H. Harjunpaa, M.C.R. Yong, S. Allen, K. Takeda

Study supervision: M.J. Smyth, M.W.L. Teng

The authors wish to thank Liam Town, Kate Elder, and Joanne Sutton for breeding, genotyping, and maintenance and care of the mice used in this study and Ran Wang and Michael McGuckin for help and advice in scoring the pathology of mouse colon sections. We thank Dr. Mahendra Singh for help and advice in assessing the pathology of mouse liver sections and Dr. Antoni Ribas for helpful suggestions and comments.

M.W.L. Teng is supported by a CDF1 Fellowship and project grants from the National Health and Medical Research Council of Australia (NH&MRC) and grants from the Cancer Council of Queensland (CCQ) and QIMR Berghofer Walk to End Women's Cancer (WEWC) award. M.J. Smyth is supported by a NH&MRC Project Grant, NH&MRC Senior Principal Research Fellowship, the Susan Komen for the Cure, and the CCQ. K.A. Fairfax is supported by an NH&MRC Fellowship (516786). J Liu is supported by a University of Queensland International Postgraduate Research Scholarship, Centennial Scholarship, Advantage Top-Up Scholarship and a QIMR Berghofer Top Up award.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.