Immune signatures of checkpoint inhibitor-induced autoimmunity—A focus on neurotoxicity

Abstract

Background

Neurologic immune-related adverse events (irAE-n) are rare but severe toxicities of immune checkpoint inhibitor (ICI) treatment. To overcome diagnostic and therapeutic challenges, a better mechanistic understanding of irAE-n is paramount.

Methods

In this observational cohort study, we collected serum and peripheral blood samples from 34 consecutive cancer patients with irAE-n (during acute illness) and 49 cancer control patients without irAE-n (pre- and on-ICI treatment, n = 44 without high-grade irAEs, n = 5 with high-grade nonneurologic irAEs). Patients received either anti-programmed cell death protein (PD)-1 or anti-PD ligand-1 monotherapy or anti-PD-1/anti-cytotoxic T-lymphocyte-associated protein-4 combination therapy. Most common cancers were melanoma, lung cancer, and hepatocellular carcinoma. Peripheral blood immune profiling was performed using 48-marker single-cell mass cytometry and a multiplex cytokine assay.

Results

During acute illness, patients with irAE-n presented higher frequencies of cluster of differentiation (CD)8⁺ effector memory type (EM-)1 and central memory (CM) T cells compared to controls without irAEs. Multiorgan immunotoxicities (neurologic + nonneurologic) were associated with higher CD8⁺ EM1 T cell counts. While there were no B cell changes in the overall cohort, we detected a marked decrease of IgD⁻ CD11c⁺ CD21^low and IgD⁻ CD24⁺ CD21^high B cells in a subgroup of patients with autoantibody-positive irAE-n. We further identified signatures indicative of enhanced chemotaxis and inflammation in irAE-n patients and discovered C-X-C motif chemokine ligand (CXCL)10 as a promising marker to diagnose high-grade immunotoxicities such as irAE-n.

Conclusions

We demonstrate profound and partly subgroup-specific immune cell dysregulation in irAE-n patients, which may guide future biomarker development and targeted treatment approaches.

Graphical Abstract

Open in new tabDownload slide

Over the past decade, immunotherapy with immune checkpoint inhibitors (ICI) has transformed clinical oncology. Today, approximately 40% of patients with advanced cancer are eligible for ICI treatment.¹ However, targeting immune checkpoints with monoclonal antibodies against cytotoxic T-lymphocyte-associated protein 4 (CTLA-4), programmed cell death protein 1 (PD-1), or programmed death-ligand 1 (PD-L1) not only promotes powerful antitumor response but also induces autoimmune phenomena, referred to as immune-related adverse events (irAEs).² Albeit rare with an incidence of 2%–12%, neurologic irAEs (irAE-n) are particularly severe immunotoxicities with case fatality rates of up to 35%.^3,4 Diagnosing irAE-n is challenging due to heterogenous presentation, missing diagnostic markers, and complex differential diagnoses such as tumor progression or central nervous system (CNS) infection, which require different treatment strategies than irAE-n.⁵ Combined with an insufficient awareness, these factors delay immunosuppressive treatment and thereby increase the risk of chronic toxicities. In fact, 40%–80% of patients with irAE-n develop long-term sequelae,^6,7 implying a dramatic reduction in quality of life because of immobility, sensory loss, or cognitive impairment. Currently, most cases of irAE-n are treated with ICI discontinuation and corticosteroids.^8,9 However, steroid-refractory irAEs exist¹⁰ and prolonged use may diminish antitumor response,¹¹ so developing targeted therapeutic approaches is crucial.

Despite the clinical relevance of ICI-induced neurotoxicity, the immunological mechanisms underlying irAE-n are poorly understood.¹² Animal models of irAE-n are missing, and immunological studies focus on more prevalent irAEs such as immune-related colitis.^13,14 Increasing evidence points toward a T cell-mediated pathology of irAEs,^13,15,16 but the recognition of autoantibody-positive neurologic and nonneurologic irAEs also suggests B cell involvement.^17,18 Others proposed proinflammatory cytokines as potential mediators of nonneurologic irAEs,¹⁹ but their role in ICI-induced neurotoxicity is unknown.

Identifying inflammatory pathways that drive irAE-n is paramount to overcome diagnostic and therapeutic challenges and thereby improve the safety of ICI treatment. To that end, the aim of this observational cohort study was to discover cellular and soluble immune signatures associated with irAE-n using in-depth peripheral blood immune profiling with single-cell mass cytometry (MC) and a multiplex cytokine assay. With this approach we (1) elucidated signatures indicative of enhanced cluster of differentiation (CD)8+ T cell activation, subgroup-specific B cell dysregulation, and immune cell trafficking in patients with irAE-n and (2) identified promising targets to diagnose and treat irAE-n.

Material and Methods

Ethics Approval and Patients’ Consent

This study was registered (DRKS00012668) and approved by the ethics committees of Charité Universitätsmedizin Berlin (EA1/099/17; EA4/219/21) and University Hospital Erlangen (no. 17_16_Bc; no. 2_20_B). Blood collection was further approved within the Melautim project in Erlangen (195_20 B). All patients provided written informed consent to participate in this study. In patients who lacked legal capacity, written informed consent was obtained from the legal guardian.

Patients

We consecutively enrolled adult cancer patients with ICI-induced neurotoxicity that were treated at our center (Charité Universitätsmedizin Berlin) between September 2017 and December 2021. Three additional irAE-n cases were enrolled at a cooperation cancer center (University Hospital Erlangen). We included patients receiving anti-PD-1, anti-PD-L1, and ICI combination treatment (= anti-PD1 + anti-CTLA-4). Diagnosis of irAE-n was given according to the consensus criteria for “definitive” or “probable” irAE-n.²⁰ Cases of “possible” irAE-n were only included if presentation was suggestive for an irAE-n and diagnostic workup could not be completed (eg, death, loss to follow-up). One additional patient with preexisting myasthenia gravis, which deteriorated to a myasthenic crisis after ICI treatment initiation, was included. Blood samples were collected during the acute disease stage. Fifteen of 34 patients with irAE-n received corticosteroids at the time of blood collection because rapid clinical worsening required immediate treatment. From 3 patients additional serum samples prior to ICI treatment were available.

As controls, we recruited consecutive cancer patients scheduled to be treated with ICIs. We matched irAE-n patients and controls in terms of cancer, ICI regimen, age, and sex. However, some irAE-n patients had rare malignancies, so perfect matching was not always possible (Table 1). To increase power, we included 10 additional controls (n = 34 patients with irAE-n; n = 44 controls). Control patients with previous ICI treatment in the last 6 months were excluded. Patients that developed high-grade (defined as Common Terminology Criteria for Adverse Events [CTCAE] grade ≥3) nonneurologic irAEs within 12 months of ICI treatment onset were analyzed separately (n = 5). We collected serum and peripheral blood samples before (V0) and 6 weeks after ICI treatment initiation (V1).

Mass Cytometry

Cell Preparation, Staining, and Barcoding.—Cryopreserved peripheral blood mononuclear cells (PBMCs, ∼5–10 × 10⁶ cells per sample) were thawed in RPMI (Sigma-Aldrich) supplemented with Benzonase HC (Millipore). Next, cells were washed, incubated for 5 min, resuspended in cytometry by time of flight staining medium, and filtered. Live-cell barcoding was performed as described previously.²¹ Veri-Cells Heavy Metal Ta PBMCs (Biolegend) were added as anchor samples for post-acquisition batch normalization. In total, 48 different immune cell markers were stained (Supplementary Table 1). Antibodies were obtained preconjugated to metal isotopes (Standard BioTools) or were conjugated using the MaxPar X8 Kit (Standard BioTools) or internal protocols.²¹ After cell surface staining and live/dead staining, cells were fixed and permeabilized (FoxP3 Staining Buffer Set, Miltenyi Biotec) for intracellular staining. Cells were then stored overnight in 4% paraformaldehyde solution (Electron Microscopy Sciences), washed, and resuspended in phosphate-buffered saline (PBS) supplemented with iridium-based DNA intercalator (Standard BioTools). After incubation, samples were washed, filtered, and adjusted to a concentration of 7.5 × 10⁵ cells/mL. MC acquisition was performed on a Helios mass cytometer (Standard BioTools). A detailed description of the staining and barcoding protocols is outlined in Supplementary Text 1.

Data Analysis.—Mass cytometry data were compensated using the CATALYST algorithm.²² Bead-normalized and debarcoded files were uploaded to OMIQ (https://omiq.ai, Santa Clara, CA). First, data were scaled and normalized, then cells were gated, subsampled, clustered using FlowSOM,²³ and metaclustered (Supplementary Text 2). Data were visualized using 2D t-distributed stochastic neighbor embedding (t-SNE) plots.²⁴ Between-group comparison of cell clusters was performed using significance analysis of microarrays (SAM).²⁵ Supervised MC analysis was performed using manual gating as outlined in Supplementary Text 3 and Supplementary Figure 1.

Multiplex Cytokine Assay

Serum samples were thawed, diluted 1:3 in dilution buffer provided with the kit (Cytokine/Chemokine/Growth Factor 45-Plex Human ProcartaPlex Panel 1, Thermo Fisher Scientific), and incubated with antibody-coated magnetic beads for 30 min at room temperature (RT) with shaking. Samples were rested overnight at 4°C, followed by a 1-h incubation period at RT on the next day. All consecutive steps were performed according to the manufacturer's instructions. Assay plates were read using the Luminex MAGPIX system and quantified using the xPONENT analysis software (Luminex Corporation). A list of all measured cytokines is depicted in Supplementary Text 4. For 13 cytokines, >50% of measurements were below the limit of detection and 2 assays (IL-9 and IL-12p70) showed batch effects. These cytokines were not considered for further analysis.

Statistical Analysis

Because of the exploratory nature of this study and the rarity of irAE-n, an à priori sample size calculation was not feasible. However, using our preliminary MC data from 6 irAE-n patients and 7 controls as a reference and the difference in CD8+ EM1 T cell frequency as the primary endpoint, we performed a power simulation with a fixed sample size of n = 20 irAE-n patients and n = 40 controls. Data were drawn from multivariate normal, exponential, and beta distributions with various correlation structures, and power of multivariate analyses of variance (parametric and nonparametric) was simulated. Simulations revealed that with the given sample size, between-group differences with an effect size of ≥0.02 could be detected with a power of 80% to a 5% significance level. To reduce detection bias, the investigators were blinded for the data analysis. Between-group differences were calculated using 2-sided Mann–Whitney U test, Wilcoxon test, or Kruskal–Wallis test for unpaired data, paired data, or multigroup data, respectively. Correlation analysis was performed using 2-sided Spearman’s correlation method. For the multivariate logistic regression model, we included markers where (1) q <0.0001 or P <.05, (2) fold change >1.5, (3) correlation coefficient ≤0.5, and (4) linear association between the independent variables and the log-odds of the outcome applied. P values of ≤.05 were considered significant. Because our analysis was exploratory, we did not perform alpha-level correction except for SAM analysis. For SAM, q-values (P values adjusted with a false discovery rate of 0.1) were calculated. We reported MC clusters where the following applied: (1) d-score >1 or <−1, (2) frequency of >1% of T cells or non-T cells, and (3) fold increase of >1.3 or fold decrease of <0.76. Statistical analyses were performed using omiq.ai (https://omiq.ai) and R Studio (version_2022.02.3). Graphical illustrations were performed using Graph Pad Prism (version 7.0), R Studio (version 2022.02.3), Microsoft PowerPoint (version 16.62), and biorender.com.

Results

Cohort Characteristics

Thirty-four cancer patients with irAE-n and 44 cancer patients without high-grade immunotoxicities (controls) were included (Supplementary Figure 2). Five patients that developed high-grade nonneurologic irAEs (CTCAE grade ≥3) were analyzed separately (Supplementary Table 2, Supplementary Figure 3). Clinical characteristics are summarized in Table 1 and Supplementary Figure 4. Data on neuronal autoantibodies from 29 of 34 irAE-n patients and all controls have been published previously.¹⁸

T Cell Profiles Associated With irAE-n

Considering that ICI targets are mainly expressed on T cells, we first focused our MC data exploration on T cell subsets. We evaluated approximately 20,000 peripheral blood T cells per participant (6 samples [10%] with <20,000 cells, in total 1,147,828 T cells) and analyzed the expression of 37 T cell markers (Figure 1A) in 50 different metaclusters (Figure 1B). Using SAM, we observed a distinct expansion of CD8+ C-C chemokine receptor (CCR)7⁻ CD45RA⁻ CD27⁺ CD28⁺ effector memory type 1 (EM1) T cells (cluster 21, d = −1.92; P < 0.0001; q < 0.0001; fold change = 1.58) in irAE-n patients compared to controls. In the CD4⁺ T cell compartment, we detected a decrease of 2 activated CD4⁺ T cell subsets: CD4⁺ CCR7⁻ CD45RA⁻ CD27⁺ CD28⁺ CD127⁺ CXC chemokine receptor (CXCR)3⁺ T cells (cluster 18, d = 1.28; P ≤ .05; q = 0.41; fold change = 0.55) and CD4⁺ CD127⁺ T cells (cluster 2, d = 1.31; P ≤ .05; q = 0.41; fold change = 0.52; Figure 1B–D). To corroborate our findings, we repeated the analysis using different numbers of metaclusters (n = 40, n = 60), which confirmed the increase of CD8⁺ EM1 T cells and the decrease of CD4⁺ CXCR3⁺ CD127⁺ EM1 T cells (Supplementary Figure 5). In addition, we aimed to validate our results using a supervised approach, enabling us to include 13 additional patients (n = 6 with irAE-n, n = 7 controls), for which a slightly modified MC panel was used. This analysis verified an increase of CD8⁺ EM1 T cells (P = .04; Supplementary Figure 6A) and detected an overall upregulation of CD8⁺ central memory (CM) T cells (P = .001; Supplementary Figure 6B). Interestingly, the increase of CD8⁺ EM1 T cells was mainly driven by patients receiving anti-PD-L1 (n = 3) and ICI combination treatment (= anti-PD-1 + anti-CTLA-4; n = 8), while CD8+ EM1 T cell frequencies were comparable between irAE-n patients and controls receiving anti-PD1 monotherapy (n = 17; Supplementary Figure 6A). In contrast, the type of cancer did not seem to affect the observed expansion of cytotoxic T cells: CD8⁺ EM1 T cells were increased in both patients with melanoma and patients with lung cancer, which were the most common cancers within the study cohort. However, due to small sample sizes, these differences did not reach statistical significance (Supplementary Figure 6A). CD8⁺ CM T cells were expanded in irAE-n patients compared to controls, independent of ICI regimen or cancer type (Supplementary Figure 6B).

Because 44% of patients with irAE-n received corticosteroids at the time of blood withdrawal, we investigated the effect of corticosteroid treatment on CD8+ T cell subset frequencies. Although no significant changes were observed, levels of CD8+ EM1 T cells tended to be higher in irAE-n patients with concomitant corticosteroid treatment compared to those without (Supplementary Figure 6C). Using a supervised approach, we did not find significant changes in the CD4+ T cell compartment (Supplementary Figure 6D).

Most patients with irAE-n had either multiple irAE-n or additional nonneurologic irAEs (Table 1). Therefore, we next evaluated whether CD8⁺ EM1 T cell frequencies were associated with the total number of irAEs. Intriguingly, patients with multiorgan immunotoxicity had higher levels of peripheral blood CD8⁺ EM1 T cells (P = .02; rho = 0.31; Figure 2A). Considering that 60% of patients with corticosteroid treatment at the time of blood withdrawal had multiorgan irAEs compared to only 41% in the group of patients without corticosteroid treatment, this may also explain the tendency toward higher CD8+ EM1 T cell levels in patients receiving corticosteroids. In contrast, higher frequencies of peripheral blood CD8⁺ EM1 T cells in irAE-n patients were not associated with better tumor response (Figure 2B).

In summary, these results demonstrate increased levels of CD8+ memory T cells in patients with irAE-n and indicate a positive correlation between the peripheral blood frequency of CD8⁺ EM1 T cells and the extent of autoimmunity in ICI-treated patients.

Innate Immune Cell Subsets and B Cell Profiles Associated With irAE-n

As early clinical studies report beneficial effects of B cell depleting therapies in patients with irAE-n²⁶ and data on myeloid immune cells are limited, we next investigated changes in non-T cell immune cell populations. To that end, we chose 21 leukocyte markers and performed SAM on 30 metacluster frequencies (Figure 3A and B). Approximately 50,000 non-T cells per participant were evaluated (n = 3 [5%] had <50,000 cells evaluable, in total 2,925,160 CD3⁻ leukocytes). We identified 5 differentially abundant clusters that illustrate 3 characteristic immune signatures in patients with irAE-n compared to controls: (1) a decrease of CD11c⁺ cDCs (cluster 14, d = 1.76; P < .0001; q < 0.0001; fold change = 0.75), (2) a decrease of natural killer cells (NK cells) (cluster 2, CD161⁺ NK cells; d = 1.28; P ≤ .05; q = 0.28; fold change = 0.71; cluster 8, CD161⁻ NK cells; d = 1.82; P < .0001; q < 0.0001; fold change = 0.56), and (3) a decrease of nonclassical CD16⁺ CD14⁻ monocytes (cluster 20, d = 2.23; P < .0001; q < 0.0001; fold change = 0.49) with a concomitant increase of CD14⁺ CD11b⁺ monocytes (cluster 4, d = −2.52; P ≤ .05; q = 0.08; fold change = 1.73; Figure 3C and D). Overall, changes in the innate immunity were reproducible independent of ICI regimen (Supplemental Figure 7A–E) and type of cancer (Supplemental Figure 8A–E), even though sample sizes were small.

Surprisingly, our analysis revealed no changes in the B cell compartment in patients with irAE-n. As we recently reported high sensitivity and specificity of neuromuscular autoantibodies in a subgroup of irAE-n patients with ICI-induced neuromuscular disease,¹⁸ we performed a B cell subgroup analysis in autoantibody-positive patients with ICI-induced myositis or myasthenia gravis (n = 9) and controls (n = 37). Intriguingly, we observed decreased levels of IgD⁻ CD11c⁺ CD21^low B cells (cluster 1, d = 1.19; P <.0001; q = <0.0001; fold change = 0.57) and IgD⁻ CD24⁺ CD21^high B cells (cluster 2, d = 1.63; P < .0001; q = <0.0001; fold change = 0.69) in patients with ICI-induced myositis or myasthenia gravis compared to controls (Supplementary Figure 9A–C). Collectively, these findings suggest dysregulation of innate immune cells in patients with irAE-n and indicate subgroup-specific B cell alterations in patients with autoantibody-positive irAE-n.

Cytokine Signatures Associated With irAE-n

Having identified cellular immune profiles of irAE-n, we next explored soluble immune signatures of ICI-induced neurotoxicity. Comparison of log2-transformed concentrations revealed an increase of chemokines mediating immune cell trafficking—namely C-X-C motif chemokine ligand (CXCL)10 (P < .001; fold change = 2.41), C-C motif ligand (CCL)3 (P = .02; fold change = 2.10), CCL4 (P = .02; fold change = 1.82), and CCL5 (P = .006; fold change = 2.41)— as well as an increase of tumor necrosis factor (TNF) (P = .006; fold change = 2.93) and an upregulation of IL-1 receptor antagonist (IL-1RA, P = .04; fold change = 2.47; Figure 4). Moreover, the growth factors vascular endothelial growth factor (VEGF)-α (P = .04; fold change = 1.76), placenta growth factor (PIGF)-1 (P = .01; fold change = 1.71), platelet-derived growth factor BB (PDGF-BB) (P < .001; fold change = 2.93), and hepatocyte growth factor (HGF) (P = .02; fold change = 1.45) showed higher concentrations in irAE-n patients. This elevation was independent of comedication with the anti-VEGF antibody bevacizumab (Supplementary Figure 10, Table 1). Noteworthy, most cytokine signatures could be observed across all 3 ICI regimens (Supplementary Figure 11) and different types of cancer (Supplementary Figure 12), although differences were not always statistically significant due to small sample sizes in the subgroups. Furthermore, apart from lower HGF levels in patients receiving corticosteroids, cytokine signatures were similar among patients with and without concomitant corticosteroid treatment (Supplementary Figure 13). In conclusion, these robust findings indicate inflammation and enhanced immune cell migration in ICI-treated patients with irAE-n.

Potential Peripheral Blood Markers of irAE-n

Attribution of neurologic symptoms to an irAE-n is often challenging because other etiologies (eg, tumor-related, infectious origin) may present with similar signs and symptoms. Therefore, the identification of markers that indicate ICI-induced autoimmunity is of great relevance. Using multivariate logistic regression, we investigated the relationship between different peripheral blood immune markers and irAE-n. We found that patients with higher CD8+ EM1 T cell frequencies (P = .05; OR 1.31; CI 1.03–1.82; threshold = 7.2% of T cells) and higher CXCL10 concentration (P = .02; OR 1.10; CI 1.04–1.24; threshold = 49.5 pg/mL) had a higher probability to suffer from an irAE-n (Figure 5A and B). A bivariate logistic regression model using CD8⁺ EM1 T cell frequency and CXCL10 concentration as independent variables and irAE-n (yes/no) as the dependent variable achieved an accuracy of 82.1% when split into a train- and test-set (split ratio = 0.5, Figure 5C). When only patients with irAE-n of the peripheral nervous system (PNS) were included (n = 16), the model achieved an accuracy of 84.0% (Figure 5D). Sample size of patients with irAE-n affecting the CNS only was too small to perform ROC analysis; however, CD8⁺ EM1 T cell frequencies and CXCL10 levels were comparable between patients with CNS- and PNS-irAE-n, respectively (Figure 5E). Low correlation of CD8⁺ EM1 T cell frequency and CXCL10 levels (Figure 5F) indicates that these parameters are independently associated with irAE-n. Together, our findings suggest that CD8+ EM1 T cells and CXCL10 may be feasible markers of ICI-induced autoimmunity.

Comparison of Patients With Neurologic and Nonneurologic irAEs

Next, we investigated whether the observed immune signatures were limited to patients with neurotoxicity or also occurred in patients with other irAEs. To that end, we divided our study cohort into 4 groups: patients with neurotoxicity only (IrAE-n; n = 19, median CTCAE grade 3), patients with neurologic and nonneurologic irAEs (IrAE-n + nonneuro irAEs; n = 15, median CTCAE grade 3), patients with nonneurologic irAEs only (nonneuro irAEs; n = 9 with CTCAE grade 1/2 irAEs plus n = 5 with CTCAE grade 3 irAEs, median CTCAE grade 2, Supplementary Table 2), and patients without irAEs (no irAEs; n = 35). Remarkably, supervised MC analysis revealed highest levels of CD8⁺ EM1 T cells and CD8⁺ CM T cells in patients with neurotoxicity or multiorgan immunotoxicity (neurologic plus nonneurologic), whereas patients without or with only nonneurologic irAEs had equally low frequencies (Supplementary Figure 14A). However, this observation reached statistical significance only for CD8⁺ CM T cells. Unsupervised multi-SAM of T cells and non-T cells did not identify statistically significantly abundant clusters, but again, levels of CD8⁺ EM1 T cells tended to be higher in patients with irAE-n compared to both patients without any or only nonneurologic irAEs (Supplementary Figure 14B). In contrast, levels of cDCs, NK cells, and monocytes were comparable between patients with neurologic and nonneurologic irAEs, whereas patients without irAEs had a tendency towards higher frequencies of these innate immune cell subsets (Supplementary Figure 14C). These findings suggest that T cell dysregulation is particularly pronounced in patients with severe (CTCAE ≥3) or multiorgan irAEs, while non-T cell signatures may be observed in all 3, patients with neurologic irAEs, patients with multiorgan irAEs, and patients with mild-to-moderate nonneurologic irAEs.

Regarding cytokine profiles, we noted that patients with multiorgan irAEs (neurologic plus nonneurologic) presented the highest cytokine concentrations, which was significant for CXCL10, TNF, PIGF-1, PDGF-BB, and IL-1RA. Patients without irAEs or with nonneurologic irAEs only, however, showed comparable cytokine levels (Supplementary Figure 15). Taken together, these observations suggest that patients with multiorgan irAEs or particularly severe toxicities such as neurologic irAEs have higher CXCL10, TNF, PDGF-BB, and IL-1RA levels than patients without any or only mild-to-moderate nonneurologic irAEs.

Lastly, we compared cytokine levels of paired pre- and on-treatment serum samples derived from 3 patients with neurologic and nonneurologic irAEs (one patient with myositis and myocarditis, one with myositis and gastritis, and one with cranial neuropathy and hepatitis, respectively) with those from patients without irAEs (n = 35) or with only nonneurologic irAEs (n = 11). Interestingly, patients that developed multiorgan immunotoxicity presented a 1.6-fold increase in CXCL10 serum levels after ICI treatment onset that was not present in patients without any or only mild-to-moderate nonneurologic irAEs (Supplementary Figure 16).

ICI-Induced Immune Signatures in Cancer Patients Without irAE-n

To investigate immune cell shifts and cytokine expression induced by immune checkpoint blockade itself (in the absence of immunotoxicities with CTCAE ≥3), we compared 43 paired serum samples and 32 paired PBMC samples of control patients at baseline (V0, pretreatment) and 6 weeks after ICI treatment onset (V1). Together, we observed an upregulation of CD8+ effector T cells, CD4⁺ CXCR3⁺ T cells, and CD161⁺ NK cells (Supplementary Figure 17A–D) as well as a decrease of proinflammatory cytokines and growth factors 6 weeks after ICI treatment onset compared to baseline (Supplementary Figure 18).

Discussion

To our knowledge, this is the first systematic investigation of cellular and soluble immune signatures in patients with ICI-induced neurotoxicity compared to ICI-treated cancer patients without irAE-n. Using deep immune profiling with MC and cytokine analyses, we identified (1) a prominent increase of CD8⁺ effector and CM T cells, (2) distinct cytokine profiles promoting inflammation and immune cell migration, (3) subgroup-specific B cell changes, and (4) alterations of innate immune cells in patients with irAE-n.

Immune checkpoint inhibitors act by disinhibiting T cells to fight cancer cells. However, excessive activation of cytotoxic T cells not only improves antitumor response but also enhances the risk of autoimmunity. Increasing evidence suggests that cross-reactive CD8+ T cells targeting tumor- and self-antigens play a crucial role in irAE development,^27,28 directly linking ICI efficacy with toxicity. It is therefore not surprising that CD8+ EM1 T cells were strongly associated with neurotoxicity in our cohort. Elevated levels of CD8+ memory T cells have also been reported in the blood and inflamed tissue of patients with ICI-induced colitis¹³ and arthritis,¹⁶ hence, cytotoxic T cell changes are most likely a marker of ICI-induced autoimmunity in general, rather than a specific marker for neurotoxicity. Considering that patients with particularly severe irAEs (eg, neurologic irAEs) and multiorgan irAEs (eg, neurologic plus nonneurologic irAEs) presented higher CD8+ EM1 T cell levels compared to patients with milder irAEs or irAEs affecting only one organ, these cells are likely to reflect the extent of ICI-induced immunotoxicity.

Moreover, recent studies demonstrated a strong relationship between ICI efficacy and irAE development²⁹ as well as between improved survival and elevated CD8+ effector T cells in the blood³⁰ and tumor tissue³¹ of ICI-treated patients. Even though we identified an expansion of cytotoxic T cells in controls after ICI treatment onset, we could not detect an association between CD8+ EM1 T cells and tumor outcome in irAE-n patients. Likewise, cancer outcome was comparable between irAE-n patients and controls. This may be partly explained by the fact that some irAE-n such as myositis have been associated with improved survival, while other neurotoxicities (eg, CNS-irAE-n) seem to have a harmful effect on overall outcome, possibly because of secondary complications (eg, immobility, long-term corticosteroid treatment).³²

Together, our results support the hypothesis that CD8+ T cells play a pivotal role in the pathogenesis of irAEs including irAE-n. From a clinical perspective, the next step is to determine the antigen-specificity of tumor- and tissue-infiltrating CD8+ T cells, as recently achieved for ICI-induced myocarditis.²⁷ This would not only allow testing for autoreactive T cells to diagnose specific irAEs, but also developing new treatment strategies such as tolerogenic vaccines that generate autoantigen-specific immune tolerance.³³

Cytokines also represent promising diagnostic and therapeutic targets of irAEs.³⁴ We identified increased levels of TNF and proinflammatory chemokines (CXCL10, CCL3, CCL4, and CCL5) in patients with acute irAE-n. Importantly, cytokine signatures did not appear to be altered by the ICI regimen, cancer type, or comedication. CXCL10, which is produced by monocytes and dendritic cells (DCs) upon interferon gamma (IFN)-γ and TNF activation and is known to promote autoimmune inflammation,³⁵ showed the strongest increase in irAE-n patients compared to controls. CXCL10 binds to the chemokine receptor CXCR3 and induces T cell migration to inflammatory sites.³⁵ Fittingly, we observed a reduction of peripheral blood CD4+ CXCR3+ T cells in irAE-n patients. In patients with multiorgan immunotoxicity, we observed a strong CXCL10 increase after ICI treatment onset and the highest CXCL10 levels during acute illness. These findings are in accordance with 2 recent studies that demonstrated an early and strong increase of CXCL10 in patients who later developed different types of irAEs.^19,36 Therefore, CXCL10 may be a promising marker to diagnose or predict high-grade immunotoxicity such as irAE-n. Moreover, eliminating CXCL10 or upstream cytokines from the peripheral blood using plasma exchange or monoclonal antibodies may be an effective strategy to treat irAEs. Early clinical and preclinical studies support the effectiveness of TNF blockade³⁷ and plasma exchange.^38,39 However, the CXCL10/CXCR3-pathway is—similar to CD8+ EM1 T cells—critical for effective tumor eradication,⁴⁰ hence, long-term use of antibodies targeting TNF and/or CXCL10 may attenuate antitumor efficacy.⁴¹

Apart from CXCL10 and TNF, we identified an elevation of CCL3, CCL4, and CCL5 in patients with irAE-n. This signature was also found in patients with ICI-induced myocarditis recently.⁴² As CCL3, CCL4, and CCL5 are known to possess both pro- and anti-cancer properties,⁴³ they have been discussed as targets for cancer immunotherapy. Especially CCL5, which can be antagonized by the antiretroviral drug maraviroc, may be a promising target to inhibit cancer progression while preventing irAEs.⁴⁴ One tumor-promoting mechanism of CCL5 is the induction of VEGF-α.⁴⁵ Interestingly, we discovered a distinct increase of multiple growth factors such as HGF, PIGF, PDGF-BB, and VEGF-α in patients with irAE-n. These factors not only promote cell growth and angiogenesis but also tumor progression and metastasis.⁴⁶ Combination treatment with ICIs and CCL5- or VEGF-inhibitors may therefore be an interesting strategy to amplify antitumor efficacy while reducing irAEs.⁴⁴

We detected a marked decrease of IgD⁻ CD11c⁺ CD21^low and IgD⁻ CD24⁺ CD21^high B cells in a subgroup of patients with autoantibody-positive neuromuscular irAE-n. While CD11c⁺ CD21^low B cells are known to expand in autoimmune diseases, decreased frequencies in the peripheral blood may be a sign of B cell homing to inflammatory sites.⁴⁷ Interestingly, early clinical studies demonstrated efficacy of the anti-CD20 antibody rituximab for ICI-induced myositis and myasthenia gravis, but not for ICI-induced multiple sclerosis.^26,48 As B cell depletion may not diminish antitumor activity,⁴⁹ future research should specifically explore therapeutic approaches targeting B cells in autoantibody-positive irAEs including irAE-n.

Regarding the innate immunity, we observed decreased levels of NK cells, cDCs, and nonclassical monocytes in the peripheral blood of patients with irAE-n compared to controls. This may reflect their enhanced recruitment to sites of inflammation and tumor tissue, especially because chemokines that promote immune cell migration were elevated. In fact, NK cells produce CCL5 to stimulate cDCs, which in turn secrete CXCL10 to recruit CD8+ and CD4+ T cells to the periphery.⁵⁰ This cascade may be one mechanism promoting irAE development, highlighting the interplay between T cells and myeloid cells as a feature of ICI-induced autoimmunity.

Our study has several limitations. First, we included patients with different cancers, different ICI regimens, and different types of irAE-n. Even though we could show that most immune signatures were detectable across different ICI regimens and cancers, we equally provide evidence that in certain irAEs—for example, autoantibody-positive irAE-n—distinct mechanisms may play a role. Studies exploring specific irAE-n separately will be useful in this regard. Second, the time from ICI treatment onset to blood withdrawal was not matched between irAE-n patients and controls (median of 13 vs 6 weeks, respectively). As immune cell profiles may be influenced by the number of ICI applications, this could have biased our findings. Third, 44% of irAE-n patients were treated with corticosteroids at the time of blood collection. Even though we did not observe significant differences between patients with and without corticosteroid treatment, individual alterations cannot be excluded. Fourth, the low number of pretreatment samples of patients with irAE-n is a weakness, which is explained by the rarity of neurotoxicities. To gain reliable conclusions regarding potential biomarkers, prospective studies with larger cohorts are needed to systematically compare immune profiles before and after ICI treatment. Fifth, we only investigated immune signatures in the peripheral blood of patients with irAE-n. It would be highly desirable to expand our analysis to sites of inflammation and link peripheral blood immune states with local immune signatures.

Lastly, our aim was to investigate immune profiles associated with ICI-induced neurotoxicity. However, we demonstrated that 50% of patients with irAE-n have coexisting nonneurologic irAEs. Therefore, a clear differentiation between neurologic and nonneurologic irAEs is challenging. Together with other recent data,^19,36 our results provide evidence that CD8+ effector T cells and CXCL10 may be severity markers of ICI-induced autoimmunity in general, which are particularly elevated in high-grade immunotoxicities such as neurologic irAEs.

In conclusion, our study identified immune signatures indicative of enhanced T cell activation, immune cell trafficking, and T cell-myeloid crosstalk in patients with irAE-n, providing new insights into the immunological mechanisms underlying ICI-induced autoimmunity. We are the first to discover distinct B cell changes in a subgroup of patients with autoantibody-positive irAE-n. The identified chemokines and immune cell subsets provide a rich source for the development of biomarkers and targeted therapies for ICI-induced immunotoxicities.

Conflict of interest statement

L.M.J., R.M., C.U., P.K., N.F., S.F., D.K., C.S., L.G.R., S.K.M., F.T.N., and M.E. have declared that no conflict of interest exists related to the submitted work. A.R.S. and H.E.M. are listed as inventors on patents relating to mass cytometry reagents; H.E.M. receives royalties from Standard BioTools. L.H. declares research support from Therakos; speakers and advisory board honoraria from 4SC, Amgen, BiomeDx, Bristol Myers Squibb, Curevac, Merck, Merck Sharp & Dohme, Myoncare, Novartis, Pierre Fabre, Sanofi, SUN, and Roche. L.H. further holds patents described in publications nos. (1) WO/2001/052874; PCT/EP2001/000363, (2) WO/2003/093419; PCT/US2003/013350, and (3) WO/2019/219705; PCT/EP2019/062378. W.B. and P.H. have received lecture fees from Bristol-Meyers-Squibb (W.B.) and/or NOGGO e.V. (W.B. and P.H.), all unrelated to this work. S.K. has received lecture fees from the Nationale Gesundheits-Akademie GmbH, outside of the submitted work.

Funding

Deutsche Forschungsgemeinschaft (DFG, German Research Foundation; Me3644/5-1, Me3644/8-1, and Me3644/11-1 to H.E.M. and EXC-2049 – 390688087 to M.E.), the Berlin Institute of Health SPARK program (to W.B.), the German Centre for Cardiovascular Research (DZHK), and the German Federal Ministry of Education and Research (BMBF) as part of the project MelAutim (01ZX1905A to L.H.), which is a systematic investigation of melanoma and autoimmunity in the context of immunotherapies. The funding sources had no influence on the design, conduct, analysis, and/or report of this research. P.H. receives funding from the Else-Kröner-Fresenius-Stiftung (2020_EKEA.80) and is recipient of a Rahel-Hirsch stipend granted by Charité Universitätsmedizin Berlin. L.M.J., C.S., S.K., and P.H. are participants of the BIH Charité Clinician Scientist Program funded by the Charité – Universitätsmedizin Berlin and the Berlin Institute of Health at Charité (BIH). C.S. receives funding from Charité 3R—Replace, Reduce, Refine, AnimalfreeResearch Switzerland, and SenUMVK Berlin.

Acknowledgments

The authors thank Petra Loge for excellent technical assistance in processing the blood samples.

Authorship statement

Designing of research study: L.M.J., A.R.S., H.E.M., L.H., M.E., W.B., P.H., S.K.; recruitment of patients: L.M.J., R.M., C.U., P.K., N.F., C.S., L.G.R., S.K.M., F.T.N., L.H., W.B., P.H., S.K.; sample preparation: L.M.J., C.S., L.G.R., S.K.M., F.T.N.; conduction of experiments: L.M.J., A.R.S., S.F., D.K.; data acquisition: L.M.J., A.R.S., S.F., D.K.; data analysis: L.M.J., A.R.S., H.E.M., S.F., S.K.; writing of manuscript: L.M.J.; critical revision of manuscript: A.R.S., H.E.M., R.M., C.U., P.K., N.F., S.F., D.K., C.S., L.G.R., S.K.M., F.T.N., L.H., M.E., W.B., P.H., S.K.; supply of material and reagents: H.E.M., L.H., M.E.; acquisition of study funding: L.M.J., M.E., W.B., P.H., S.K.

Consent for publication

The corresponding author has full access to all data, for which she has the right to publish apart from any sponsor. All coauthors have read the final version of the manuscript and approved submission and publication.

Data availability

Mass cytometry data files are available from FlowRepository [FR-FCM-Z62A and FR-FCM-Z62D]. Anonymized analyzed data will be made available by the corresponding author upon reasonable request.

References

Author notes