A 67-year-old man with metastases of melanoma in the brain, adrenal glands, lung, subcutis, and lymph nodes started double immune checkpoint blockade; PD1-inhibitor nivolumab (1 mg/kg) and the CTLA-4 inhibitor ipilimumab (3 mg/kg) given four times at 3-week intervals (figure 1A). After two treatments, MRI scans (MRI) showed regression of the brain metastases (figure 1B). The day after the fourth treatment, he developed a fever (39.1°C), elevation of C reactive protein (CRP) (45 mg/L), and dizziness. He was admitted to the oncology ward at Sahlgrenska University Hospital and was given IV antibiotics. After 2 days, he developed paraparesis in his lower limbs, decreased consciousness, and respiratory failure. His CRP level increased to 130 mg/L.

He was admitted to the intensive care unit (ICU), intubated, and put on mechanical ventilation. On ICU day 1, MRI showed complete regression of his brain metastases but new diffuse lesions in the brain, brainstem, and cerebellum with a bilateral and asymmetric pattern (figure 1B), indicating acute disseminated encephalomyelitis.9 Analyses of cerebrospinal fluid (CSF) suggested autoimmune encephalitis (online supplemental table S1). Immunosuppression with methylprednisolone (1 g daily intravenously) was started. The next day he was alert during sedation stops but required respiratory support, reflecting brain stem damage. MRI on ICU day 3 revealed unchanged brain lesions and lesions in the spinal cord. Cyclophosphamide was administered (1.5 g intravenously as a single dose), but his condition did not improve. On ICU day 6, mycophenolate mofetil (1 g two times per day intravenously) was started to inhibit T and B cell proliferation.10 The next day, he had improved and could be extubated and transferred back to the oncology ward. He had flaccid paraparesis in his lower limbs with minor residual sensation and urine and fecal incontinence. High-dose cortisone and mycophenolate mofetil were continued, and intravenous immunoglobulins were given (35 g daily for 5 days). MRI 10 and 17 days after ICU admission revealed gradual decrease in lesions in the brain and spinal cord (figure 1B).

After 6 weeks, the patient was discharged from hospital care. Now, he had regained most sensory functions, and he could sit up and operate a wheelchair. After 10 weeks, the inflammatory lesions had resolved on MRI. The immunosuppressant treatment was tapered and permanently discontinued after 14 months. More than 2 years after the first dose of double immune checkpoint blockade, the patient no longer suffers from fecal incontinence and he can walk more than 100 steps with assistive devices. By becoming more independent, the patient’s quality of life has improved significantly and he remains tumor-free.

In addition to the encephalomyelitis patient, we also analyzed T cell characteristics in checkpoint inhibitor-treated patients with (n=9) or without other irAE (n=12) (table 1). The irAE were moderate to severe and occurred during double (n=2, including the encephalomyelitis patient) or single (PD-1) inhibition (n=8). Similarly, samples were obtained from patients without irAE during double (n=2) or single (PD-1) checkpoint inhibition (n=10). In addition, we performed repeated analysis of cytokines and soluble checkpoint proteins in CSF from the encephalomyelitis patient. As controls for this analysis, we used CSF from patients with autoimmune systemic lupus erythematosus without encephalitis (n=4).

The number of reported cases of encephalitis following both single and double immune checkpoint blockade is increasing and the fatality rate is high (WHO global database VigiBase, online supplemental figure S2).

Retrospective analysis of blood tests unexpectedly revealed a distinct pattern that was not evident during the patient’s stay. The brain damage marker S-100B and the inflammatory marker CRP covaried strikingly over time (figure 1C), and levels of both were elevated after two treatments with ipilimumab/nivolumab. At that point, the patient had no symptoms, and MRI showed regression of the brain metastases and no signs of encephalitis. After the fourth and final treatment, S-100B and CRP peaked, indicating combined inflammation and brain damage. The patient rapidly deteriorated, and MRI showed acute disseminated encephalomyelitis. These results suggest that brain damage markers in blood may indicate encephalitis before the appearance of typical signs on MRI or clinical neurological symptoms.

Extensive analysis of blood and CSF confirmed brain damage and inflammation (online supplemental figure S3, tables S1 and S4) but showed no signs of infection or autoantibodies (online supplemental tables S1 and S4). The brain damage markers neurofilament light polypeptide (NFL) and glial fibrillar acidic protein (GFAP) were extremely high in both CSF and plasma as symptoms peaked and gradually normalized during immunosuppression (online supplemental figure S3, tables S1 and S3). Tau and S-100B levels were moderately increased (online supplemental table S3). Collectively, these findings suggest severe axonal damage (NFL) and astrocyte injury (GFAP).11 Interferon-γ, tumor necrosis factor-α, and interleukin-6 levels were increased in CSF when symptoms peaked and normalized during recovery (online supplemental figure S3). The checkpoint proteins PD1, PD-L1, and Tim-3 showed a similar pattern (online supplemental figure S4). CTLA-4 and LAG-3 were not elevated.

Flow cytometry of peripheral T cells was done on ICU day 2 and repeated four times. The analysis included T cell subtypes, activation markers, costimulatory receptors, inhibitory immune checkpoints, transcription factors, and attack enzymes (online supplemental table S2). In addition to our encephalomyelitis patient, T cell phenotype was also analyzed in patients with other irAE as well as in patients without irAE (table 1). The most striking observation in the encephalomyelitis patient was high expression of inducible T cell costimulatory receptor (ICOS) on all subtypes of CD4 +T helper cells (figure 2A) (online supplemental figure S5) and on CD8+ cytotoxic T cells. ICOS expression on T cells gradually normalized during immunosuppression and in parallel with clinical improvement (figure 2B–D) (online supplemental figure S5B,C). Similarly, high ICOS on CD4+ and CD8+T cells was detected also in patients with other irAE (figure 2A) and decreased when irAEs had resolved (figure 2B). One patient developed grade 4 hepatitis during treatment with double checkpoint inhibition. In this patient, one sample was analyzed before irAE, when liver enzymes were normal. Interestingly ICOS on CD4+ and CD8+T cells increased in parallel with liver enzymes and decreased again after immunosuppression, and subsequent normalization of liver enzymes (figure 3). Collectively, our data indicate that ICOS may promote development of irAE. High ICOS expression could not be explained by longer duration of checkpoint inhibition because the difference was significant also if patients with late irAE (>6 months, 4 patients) were excluded from the analysis (irAE vs non-irAE; ICOS on CD8+ T cells, p<0.010; ICOS on CD4 +T cells, p<0.027). In addition, the difference in ICOS expression was significant also when only patients treated with single PD-1 inhibition was compared (irAE vs non-irAE; ICOS on CD8 +T cells, p<0.0085; ICOS on CD4+T cells, p<0.0062). This indicates that the difference in ICOS was not specific for anti-CTLA-4 treatment. Full data on T cell characteristics is shown in online supplemental figure S6. In addition to ICOS, TIM-3 (an inhibitory checkpoint protein and activation marker) was significantly higher on CD8 +T cells in patients with irAE and normalized when irAE were resolved. Also, PD-1 expression on CD4+ T cells was significantly higher in patients with irAE. The patients with irAE had similar proportion of immunosuppressive regulatory T cells as checkpoint inhibitor-treated controls (online supplemental figure S7). There was no difference in total number of CD8+ cytotoxic T cells, CD4+T helper cells, B cells, or natural killer cells, between our patient and checkpoint inhibitor-treated controls (online supplemental figure S8).