Pembrolizumab in patients with thymic carcinoma: a single-arm, single-centre, phase 2 study

Summary

Background

Treatment options are limited for patients with thymic carcinoma. These aggressive tumours are not typically associated with paraneoplastic autoimmune disorders, and strong PD-L1 expression has been reported in thymic epithelial tumours. We aimed to assess the activity of pembrolizumab, a monoclonal antibody that targets PD-1, in patients with advanced thymic carcinoma.

Methods

We completed a single-arm phase 2 study of pembrolizumab in patients with recurrent thymic carcinoma who had progressed after at least one line of chemotherapy. This was a single-centre study performed at Lombardi Comprehensive Cancer Center, Georgetown University, Washington, DC, USA. Key inclusion criteria were an Eastern Cooperative Oncology Group performance status of 0–2, no history of autoimmune disease or other malignancy requiring treatment or laboratory abnormality, and adequate organ function. Patients received 200 mg of pembrolizumab every 3 weeks for up to 2 years. The primary objective of the study was the proportion of patients who had achieved a response assessed with Response Evaluation Criteria in Solid Tumors version 1.1. Analysis was per protocol, in all eligible patients. The study is registered with ClinicalTrials.gov, number NCT02364076, and is closed to accrual; we report the final analysis.

Findings

41 patients were enrolled from March 12, 2015, to Dec 16, 2016, of whom 40 were eligible and evaluable and one was excluded because of elevated liver enzymes at screening. The median follow-up was 20 months (IQR 14–26). The proportion of patients who achieved a response was 22·5% (95% CI 10·8–38·5); one (3%) patient achieved a complete response, eight (20%) patients achieved partial responses, and 21 (53%) patients achieved stable disease. The most common grade 3 or 4 adverse events were increased aspartate aminotransferase and alanine aminotransferase (five [13%] patients each). Six (15%) patients developed severe autoimmune toxicity, including two (5%) patients with myocarditis. There were 17 deaths at the time of analysis, but no deaths due to toxicity.

Interpretation

Pembrolizumab is a promising treatment option in patients with thymic carcinoma. Because severe autoimmune disorders are more frequent in thymic carcinoma than in other tumour types, careful monitoring is essential.

Funding

Merck & Co.

Introduction

Thymic epithelial tumours are rare malignancies of thymic epithelial cells, and approximately 500 new cases are diagnosed each year in the USA.¹ Thymic carcinomas are the most aggressive subtype of thymic epithelial tumours and constitute just over 10% of thymic epithelial tumours.² They are often not resectable, tend to metastasise to many locations, and are associated with a shorter overall survival than are thymomas. The recommended treatment of localised disease is surgery, whenever feasible, and radiation with or without chemotherapy. For unresectable and metastatic thymic carcinomas, the standard treatment is chemotherapy with cisplatin, doxorubicin, and cyclophosphamide, or carboplatin and paclitaxel, although the proportion of patients who achieve a response is less than 50%.³ Targeted therapies against thymic carcinoma have been unsuccessful, except for sunitinib, a multikinase inhibitor, which led to a response in 25% of patients.⁴ Novel, effective treatments are needed for this rare disease.

Immunotherapy has been crucial in the treatment of several malignancies in the past few years and has included the use of immune checkpoint inhibitors, such as antibodies against CTLA-4, PD-1, and PD-L1.^5,6 Thymic carcinomas have a high expression of PD-L1,^7,8 which is correlated with a better response to PD-1 and PD-L1 antibodies in several studies.^5,9,10 When PD-1, which is predominantly present on the surface of activated T cells, binds to PD-L1 on tumour cells, the cytotoxic T-cell response is downregulated.¹¹ The thymus is involved in T-cell development, and although thymomas are associated with myasthenia gravis and other autoimmune diseases, thymic carcinomas are infrequently associated with autoimmune disorders.³ On the basis of these considerations, we conducted an investigator-initiated, single-arm, phase 2 study of pembrolizumab, an anti-PD-1 antibody, in patients with advanced refractory or recurrent thymic carcinomas.

Methods

Study design and participants

This study was a single-arm, phase 2 study, done at Lombardi Comprehensive Cancer Center, Georgetown University, Washington DC, USA, in patients with thymic carcinoma who had progressed after at least one line of chemotherapy. Patients with a histological diagnosis of thymic carcinoma, which was determined in accordance with 2004 WHO classification, were eligible for inclusion in the study.¹² Histopathology slides were reviewed by the study pathologists (BK and JJC) to confirm the diagnosis and to assess PD-L1 expression. The inclusion criteria included: progressive disease upon study entry, advanced stage disease (based on the study¹³ by Masaoka and colleagues), an Eastern Cooperative Oncology Group (ECOG) performance status of 0–2, an age of at least 18 years, no history of autoimmune disease, adequate organ and bone marrow function, measurable disease in accordance with Response Evaluation Criteria in Solid Tumors (RECIST) version 1.1,¹⁴ at least one previous chemotherapy regimen, and no systemic therapy for at least 4 weeks before study enrolment. Patients with treated brain metastases were eligible if they were asymptomatic in the absence of steroid treatment and not progressing after local treatment. Exclusion criteria included active HIV or hepatitis infections, immunodeficiency, interstitial pneumonitis, or previous treatment with an immune checkpoint inhibitor. Written informed consent was obtained from all patients. The study was approved by the institutional review board of the Lombardi Comprehensive Cancer Center, Georgetown University (Washington, DC, USA). The protocol is in the appendix. As of July 10, 2017, the protocol was amended to include an additional cohort of patients treated with pembrolizumab and epacadostat (an IDO1 inhibitor) on the basis of promising data in patients with melanomas and non-small-cell lung cancer (NSCLC). Results will be reported separately.

Procedures

Patients were treated with 200 mg of pembrolizumab intravenously every 3 weeks for up to 2 years. There were no dose reductions allowed. Treatment was withheld in patients with grade 4 haematological toxicities and with non-haematological toxicities of grade 3 and above, including laboratory abnormalities and severe or life-threatening adverse events. Supportive care and management of immune-related adverse events (irAEs) were in accordance with standard guidelines.^6,9 Toxicity was graded with Common Terminology Criteria for Adverse Events version 4.0.¹⁵ Treatment was discontinued in cases of confirmed progression, unacceptable toxicity, and on the patients’ requests.

Pregnancy tests and coagulation tests were done before entry into the study. Physical examination, blood cell counts, and urine analysis were done before every treatment cycle, and thyroid function tests for total triiodothyronine, free thyroxine, and thyroid-stimulating hormone were done every two cycles. ¹⁸F-fluorodeoxyglucose-PET imaging was recommended before entry and was repeated if positive at baseline.

We assessed tumour response with CT scans every two cycles for the first year, every four cycles for the second year, and every 3–4 months thereafter. Response was independently assessed by one radiologist (MM) in accordance with RECIST version 1.1.¹⁴ Pleural lesions were measured in accordance with RECIST version 1.1.

To determine PD-L1 expression, we used immunohistochemistry (PD-L1 IHC 22C3 pharmDx kit; Agilent Technologies, Santa Clara, CA, USA) of archival formalin-fixed paraffin-embedded tissues, and we also extracted DNA and RNA from these tissues for sequencing (appendix p 1). We encouraged use of fresh biopsy samples (taken when insufficient archival material was available and upon progression) in addition to archival samples for DNA and RNA extraction. PD-L1 expression was classified as “high” if at least 50% of the tumour cells stained positive.⁹ PD-L1 expression in 1–49% of cells was classified as “low” expression, and expression was classified as “none” if no tumour cells in the sample expressed PD-L1. We used next-generation sequencing with a targeted exome approach with a 206-gene custom cancer-related gene panel (appendix pp 7–8) on the Illumina MiSeq platform (Illumina, San Diego, CA, USA), as previously described,¹⁶ with whole blood acting as a normal germline control (appendix pp 1–2). We used the NanoString nCounter MAX Analysis System with nSolver version 3.0 software (Seattle, WA, USA) to assess an 18-gene T-cell-inflamed interferon-γ gene expression profile signature, as previously described,¹⁷ and the cutoff value was set at −0·318, as previously described (appendix p 2).¹⁷

Because sequencing of the T-cell receptors of different tissues might identify shared epitopes,¹⁸ we sequenced DNA that we extracted from the tumour, blood, and muscle in one patient who developed severe myositis, in accordance with a described method.¹⁸ The blood was taken before treatment and after the autoimmune disorder was diagnosed and treated, the tumour tissue was archival, and the muscle biopsy was obtained when the severe myositis occurred. This measurement enabled us to determine T-cell receptor clones that could be involved in this adverse event or clones that expanded because of treatment.

Outcomes

The primary endpoint of the study was the proportion of patients who achieved an overall response (ie, a radiologically confirmed complete or partial response). Duration of response was assessed ad hoc (from the first assessment of response to progression). Secondary endpoints were progression-free survival, overall survival, and safety. Progression-free survival was calculated from the date of treatment start to progression or death, whichever came first. Overall survival was calculated from treatment start to death or the date that the patient was last known to be alive.

Statistical analysis

We tested the null hypothesis that 5% or fewer patients would have a response against the one-sided alternative hypothesis that more than 5% of patients would respond to pembrolizumab, assuming the true response to be 20%. We used a Simon’s two-stage optimal phase 2 design: in the first stage, 21 patients were to be accrued and, if more than one response was seen, the study would accrue a total of 41 patients. The null hypothesis would be rejected if five or more responses were observed in 41 patients. This design yields a type I error of 5% and power of 90%. We calculated the Clopper-Pearson exact 95% CI for the proportion of patients with a response using an exact binomial calculation. All analyses were done per protocol on all eligible patients.

The Kaplan-Meier method was used to determine progression-free survival, overall survival (with 95% confidence bands), and duration of response. The computation was performed in R version 2.15.2 and SPSS version 24.

We analysed the association between response and baseline characteristics and between response and the results of PD-L1 expression and mutation type and frequency by Fisher’s exact test as a post hoc analysis. The correlation between PD-L1 expression and response to therapy was assessed. The correlations between PD-L1 expression and progression-free survival and overall survival were analysed post hoc. The results of next-generation sequencing were analysed by assessing the association between response and presence of any mutation and between response and number of mutations per sample. Post-hoc analyses were done to test the associations between the most common mutations and progression-free survival and overall survival.

Post-hoc analyses of the association between irAEs and response and the 18-gene interferon-γ expression were done by Fisher's exact test. Post-hoc analysis of the associations between gene expression profiles and clinical outcome were assessed with logistic regression (best overall response) and Cox regression (progression-free survival and overall survival). Standard receiver operating characteristic (ROC) analyses were also performed for best overall response. We evaluated the appropriateness of the Cox model using the binary variable of the gene expression profile with graphical methods that assessed the parallelism of the negative log of the negative log of estimated survival functions and the observed survival versus expected survival. Models were fitted to the continuous and dichotomised gene expression profile variable (≤−0·318 vs >−0·318, cutoff as previously described; appendix p 3).¹⁷ For the binary gene expression profile variable, median progression-free survival and overall survival with corresponding 95% confidence intervals were derived by the Kaplan-Meier product limit estimator. The trial is registered with ClinicalTrials.gov, number NCT02364076.

Role of the funding source

Representatives of the funder, including RM and WMB, contributed to the study design, data analysis and interpretation, and writing of the report. The study database for data collection was maintained by the principal investigator (GG), but the funder had no role in data collection. All authors had access to the data and had responsibility for the decision to submit the Article for publication. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

Results

From March 12, 2015, to Dec 16, 2016, 41 patients were enrolled. One patient was not eligible because of elevated liver enzymes at screening and was excluded from further analysis; thus 40 patients were evaluable for safety and activity analysis. Baseline characteristics of the eligible patients are shown in table 1. Most patients had more than one previous line of chemotherapy and had extensive metastatic disease (appendix pp 4–5). The median follow-up was 20 months (IQR 14–26) at the data cutoff date of July 28, 2017.

Table 1:

Baseline characteristics

	Eligible patients (n=40)
Median age, years	57 (25–80)
Sex
Male	28 (70%)
Female	12 (30%)
ECOG performance status
0	19 (48%)
1	19 (48%)
2	2 (5%)
Race
White	33 (82%)
Asian	4 (10%)
African-American	2 (5%)
Latino	1 (3%)
Histology
Squamous	19 (48%)
Poorly differentiated	13 (32%)
Neuroendocrine	6 (15%)
Not specified	2 (5%)
Disease stage (Masaoka et al13)
III	1 (3%)
IVA	6 (15%)
IVB	33 (82%)
Metastatic sites
Median number of sites	2 (1–6)
Liver	15 (38%)
Brain	6 (15%)
Bone	9 (22%)
Previous therapy
Median number of previous therapies	2 (1–6)
Previous thymectomy	21 (52%)
Previous chest radiation	23 (58%)

Data are median (range) or n (%). ECOG=Eastern Cooperative Oncology Group.

One (3%) of 40 eligible patients had a complete response and eight (20%) had partial responses, giving an overall response of 22·5% (95% CI 10·8–38·5; figure 1A). 21 (53%) patients had stable disease (including one patient with an unconfirmed partial response), and in ten (25%) patients progression was the best response. Pleural lesions were present in 18 (45%) patients and were used as measurable lesions in ten (25%) patients. Disease control (complete or partial response plus stable disease) was achieved in 30 (75%, 95% CI 59–87) patients.

Figure 1: Overall response.

(A) Waterfall plot of the best response in all eligible patients. *Patients have an ongoing response or stable disease. (B) Measurements of target lesions at each timepoint in all eligible patients (where CT scans were taken every two cycles, which represents 6 weeks). (C) Duration of treatment. With a few exceptions, treatment cycles were repeated every 3 weeks. The arrows indicate that patients are either still on treatment, are still in follow-up after completing 2 years of therapy (35 cycles), or, in the case of patient number 28, were taken off the study but are still responding.

Median time to response was 6 weeks (range 6–24, IQR 6–12). There were no cases of pseudoprogression (figure 1B).

The median duration of response for nine patients with a complete or partial response was 22·4 months (95% CI 12·3–34·7) from first measurement of response, and the median duration of stable disease in 21 patients with stable disease from the start of treatment was 6·8 months (1·8–11·7; figure 1C). Eight (20%) patients had stable disease that was maintained for at least 6 months and three patients had stable disease that was maintained for more than a year.

At the date of the analysis (the cutoff date), 31 (78%) patients had disease progression and 17 (43%) patients had died: 14 (35%) from disease progression, one from heart failure, one from the toxic effects of subsequent treatment, and one from development of acute myeloid leukaemia. Median progression-free survival was 4·2 months (95% CI 2·9–10·3) and median overall survival was 24·9 months (15·5–not reached; figure 2). 1-year progression-free survival was 29% (17·6–48·5) and 1-year overall survival was 71% (57·6–87·1).

Figure 2: Progression-free survival (A) and overall survival (B).

Grey highlighting represents 95% CIs.

Patients received a median of six cycles (range 1–35) of pembrolizumab treatment. The major reason for treatment discontinuation was disease progression (28 [70%] patients). Treatment was also discontinued in three (8%) patients upon the patients’ requests. As of data cutoff, three patients had completed 2 years of treatment. Most patients only had mild (grade 1–2) adverse events that were similar to those reported with pembrolizumab treatment in other malignancies (table 2). The most common grade 3 or 4 adverse events were increased aspartate aminotransferase and alanine aminotransferase (five [13%] patients each). However, six (15%) patients developed one or more new-onset severe irAEs (appendix p 6); four of these patients were treated with intravenous steroids (125 mg methylprednisolone) followed by high-dose oral steroids (1–2 mg/kg prednisone once a day or 4 mg dexamethasone every 4 hours). No patients with severe irAEs died because of toxicity, but in four cases hospital admission was necessary. We interrupted treatment because of severe irAEs in three patients and because of concurrent severe irAEs and progression of disease in three patients. Two patients were diagnosed with polymyositis and myocarditis after two cycles of therapy, which responded to high-dose steroids but required placement of pacemakers. All patients with severe irAEs recovered. Three of six patients had a partial response to therapy (p=0·12).

Table 2:

All adverse events associated with pembrolizumab treatment, by worst grade in each patient

	Grade 1–2	Grade 3	Grade 4
Fatigue	16 (40%)	3 (8%)	0
Aspartate aminotransferase increased	11 (28%)	3 (8%)	2 (5%)
Alanine aminotransferase increased	5 (13%)	4 (10%)	1 (3%)
Alkaline phosphatase increased	10 (25%)	0	0
Diarrhoea	9 (23%)	0	0
Arthralgia	4 (10%)	1 (3%)	0
Fever	5 (13%)	0	0
Hypothyroidism	5 (13%)	0	0
Rhinitis	4 (10%)	0	0
Anaemia	2 (5%)	2 (5%)	0
Nausea	4 (10%)	0	0
Rash	4 (10%)	0	0
Dyspnoea	0	3 (8%)	0
Myalgia or myositis	0	3 (8%)	0
Creatine phosphokinase increased	0	1 (3%)	2 (5%)
Bilirubin increased	2 (5%)	0	0
Blurred vision	1 (3%)	1 (3%)	0
Dry mouth	2 (5%)	0	0
Flu-like symptoms	2 (5%)	0	0
Hyperthyroidism	2 (5%)	0	0
Hyperuricaemia	2 (5%)	0	0
Myocarditis	0	0	2 (5%)
Neutropenia	2 (5%)	0	0
Amylase increased	1 (3%)	0	0
Dehydration	1 (3%)	0	0
Watering eyes	1 (3%)	0	0
Hyperglycaemia	0	0	1 (3%)
Hypokalaemia	1 (3%)	0	0
Leucocytopenia	1 (3%)	0	0
Lipase increased	0	1 (3%)	0
Localised oedema (facial swelling)	1 (3%)	0	0
Palpitations	1 (3%)	0	0
Skin and subcutaneous tissue disorders	1 (3%)	0	0
Thrombocytopenia	0	1 (3%)	0

Data are n (%). Grades are in accordance with Common Terminology Criteria for Adverse Events version 4.0. There were no deaths due to adverse events

PD-L1 immunohistochemistry data were available for 37 patients. Positive staining for PD-L1 in at least 50% of tumour cells, indicating high PD-L1 expression, was found in ten (25%) patients, six of whom had a partial or complete response (p=0·005; figure 3). Of the ten patients with high PD-L1 expression, five had disease progression and one died. Of the 27 patients with low PD-L1 expression, 23 had disease progression and 14 died (appendix p 10). In a post-hoc analysis, progression-free survival was longer in patients with high PD-L1 expression than those with low or no expression (median 24 months, 95% CI 5·8–42·3 vs 2·9 months, 1·7–4·1; appendix p 10). Overall survival was also longer in patients with high expression than those with low or no expression (median not reached vs 15·5 months, 4·7–26·3; appendix p 10).

Figure 3: PD-L1 expression and response to pembrolizumab.

Error bars are SD. Horizontal lines are medians.

NanoString gene expression profiling was performed in 33 patients. Expression of the 18-gene T-cell-inflamed interferon-γ gene expression profile was higher in patients with a response than in non-responders, with an area under the ROC curve of 0·71 (95% CI 0·53–0·89; appendix p 11). There was a significant correlation between presence of the 18-gene interferon-γ signature and response (p=0·044; appendix p 12); ie, the higher the expression of the interferon-γ signature, the greater the response.¹⁷ Cox models of progression-free survival and overall survival using the continuous gene expression profile score did not stratify patients (data not shown); however, using the binary gene expression profile category (at least −0·318 vs less than −0·318) in these models showed significantly longer progression-free survival and overall survival in patients with a gene expression profile of at least −0·318 compared with those with lower expression of the gene expression profile (appendix p 13). For progression-free survival, 11 (85%) of 13 patients with a gene expression profile less than −0·318 and 14 (70%) of 20 patients with a gene expression profile at least −0·318 had a progression event (median progression-free survival 82 days, 95% CI 39–not reached vs 235 days, 87–not reached; hazard ratio 0·41, 95% CI 0·18–0·95). For overall survival, eight (62%) of 13 patients with gene expression profile less than −0·318 died versus 7 (35%) of 20 patients with a gene expression profile at least −0·318 (median overall survival 346 days, 95% CI 268–not reached vs not reached, 639–not reached; hazard ratio 0·28, 95% CI 0·09–0·87).

Targeted exome sequencing was successfully done in samples from 36 patients. 130 somatic mutations were observed, with a median of three somatic mutations per patient (range 0–12). In four of 36 patient samples no somatic mutation was identified. The most common type of mutation was missense (80 [62%]; appendix p 14), 51 (64%) of which were identified to be deleterious or damaging (appendix). 42 (32%) of the mutations detected are catalogued in the COSMIC database and 36 (86%) of those are predicted to be pathogenic. The most commonly mutated gene was TP53 (13 [36%] of 36 patients; appendix p 15). There was no correlation between the type of mutation and response to treatment or autoimmune toxicity (data not shown). TP53 mutations (15 mutations in 13 patients) were found more frequently in patients with low or no PD-L1 expression than in patients with high PD-L1 expression (p=0·043). All five patients with a CYLD mutation had high expression of PD-L1 (p=0·00067). Three of these five patients had a complete or partial response. Furthermore, three of the five mutations were stop-gain mutations, indicating truncation of the protein (appendix p 16). A recurring stop-gain mutation in the CDKN2A gene (COSM12475) was detected in four different patients.

In a post-hoc analysis, there was a correlation between TP53 mutation and shorter overall survival but no correlation between TP53 mutation and progression-free survival (appendix p 17). Of 13 patients with TP53 mutations, ten (77%) had disease progression and nine (69%) died. Of 23 patients with wild-type TP53, 17 (74%) had disease progression and six (26%) died. Median overall survival in patients with a TP53 mutation was 10 months (95% CI 6·5–13·5), and median overall survival was not reached in patients with wild-type TP53. The median progression-free survival was 1·5 months (95% CI 0–3·3) in patients with a TP53 mutation versus 8·3 months (6·0–10·6) in patients with wild-type TP53.

There was a longer progression-free survival and overall survival in five patients with CYLD mutations compared with patients with wild-type CYLD, but this difference was not significant (appendix p 17).

Notably, one patient who developed a severe autoimmune disorder had a mutation in the GTF2I gene (7:74146970T→A, 1271T→A, Leu424His), a mutation that we previously showed to be common in type A and AB thymomas, the most indolent thymic epithelial tumours, but infrequent in thymic carcinomas.¹⁶ Two patients had a repeat biopsy of their tumour upon progression: PD-L1 expression was identical to the previous measurement in both patients and, in one patient, the mutation analysis was also identical, whereas the gene expression profile decreased from −0·034 to −0·240 (but the gene expression profile was not examined in the other patient). None of our patients had high levels of microsatellite instability.

A measurement of differential abundance was used to identify any T-cell receptor clones that expanded between samples of pre-treatment blood and samples of post-treatment blood in a patient who developed severe myositis and myocarditis. 26 T-cell receptor clones increased in frequency in the post-treatment blood sample; of these, seven were found in the muscle biopsy and one was found in both muscle and tumour (appendix p 18).

Discussion

The response to pembrolizumab in thymic carcinomas was similar to that achieved in other epithelial tumours, such as non-small-cell lung cancer, and the duration of response was 22·4 months, which is substantially longer than that observed with sunitinib (16·4 months) in a similar setting.⁴ Three patients have completed 2 years of treatment so far. The rapidity and depth of the responses, together with the duration of the responses in this heavily pre-treated and widely metastatic population are promising.

Although we selected for patients with only thymic carcinoma and without previous autoimmune disorders, six patients developed severe irAEs. Two patients developed myocarditis, which required high-dose steroids and placement of a pacemaker. Both patients also developed polymyositis and fully recovered. Another patient developed myositis, which responded to steroids. Involvement of skeletal striated muscle and heart muscle has been described in two patients with metastatic melanoma who received combination treatment with nivolumab, an anti-PD-1 antibody, and ipilimumab, an anti-CTLA-4 antibody.¹⁸ In these patients, a T-cell clone was shared between the tumours and the muscle. In our study, we also found a shared T-cell clone that was present in four samples (in pre-treatment and post-toxicity blood and tumour and muscle biopsies) from one patient. These data suggest the potential for a tumour to be sharing epitopes that are found in normal muscle, as previously reported;¹⁸ however, because of the low number of templates in the tissues, it is difficult to draw a strong conclusion.

Treatment with nivolumab plus ipilimumab in patients with melanoma was associated with myocarditis in less than 1% of 2974 patients treated in previous analyses; this incidence was still higher than with nivolumab alone.¹⁸ Additional cases of myocarditis in patients treated with immune checkpoint inhibitors have since been published,¹⁹ and myocarditis has now been recognised as a potentially serious side-effect of other immune checkpoint inhibitors, such as atezolizumab and durvalumab, which are both PD-L1 inhibitors. The incidence of irAEs is substantially lower following treatment with antibodies against PD-1 and PD-L1 compared with antibodies against CTLA-4, but the incidence of severe irAEs appears to be much higher in patients with thymic epithelial tumours than in patients with other tumour types, in which incidence is typically less than 1%. The toxicity of these checkpoint inhibitors could vary substantially by disease type and might reflect the different role of the immune system in the pathogenesis of these tumours. In some patients, the diagnosis of a thymic epithelial tumour was made on the basis of small volume biopsies. We cannot completely rule out the possibility of misdiagnosing thymoma, a distinct disease entity associated with an increased risk of developing autoimmune disorders, from this limited biopsy tissue because there have been sporadic case reports²⁰ of primary thymic epithelial tumours showing mixed histology of thymic carcinoma and thymoma. However, the evidence suggests that most thymic carcinomas arise de novo.²¹

Adverse events in muscles and myocardium appear to be particularly frequent in thymic epithelial tumours when treating with immune checkpoint inhibitors. In a phase 1 study²² of avelumab, a PD-L1 antibody, eight patients with thymic epithelial tumours (seven with thymomas and one with thymic carcinoma) were treated and five patients developed severe autoimmune disorders, three of which were myositis. Two patients with thymomas responded to the treatment. In another phase 2 study²³ of pembrolizumab in South Korea, seven patients with thymomas and 26 patients with thymic carcinoma were enrolled. In this study, patients with a history of autoimmune disorders were not excluded: five of seven patients with thymoma developed severe autoimmune disorders and one patient died of nephritis. Three of 33 patients developed grade 4 myocarditis. This study reported a similar treatment response to our study (24·2% vs 22·5% in our study, although cross-trial comparisons should be made with caution). Clearly, immune checkpoint inhibitors are active in thymic epithelial tumours, but they also present a higher risk of autoimmune toxicity than in other tumour types. Since these toxic effects can be severe (particularly myocarditis), special caution must be taken, and methods to prevent these effects or predict high-risk patients are needed. In our study, female and Asian patients appeared to develop autoimmune disorders more frequently than in male and non-Asian patients, although the associations were not significant. We were unable to identify any other factor that might have affected toxicity (such as mediastinal radiation or history of cardiac disease). In a previous study, we showed that most patients with thymic epithelial tumours have autoantibodies against several cytokines in their blood, even in the absence of overt autoimmune syndromes.^7,24 This finding suggests that these tumours might promote autoimmunity. Expression of AIRE is downregulated in thymic epithelial tumours of patients with autoimmune disorders.²⁵ AIRE plays an important role in immunity by regulating the expression of autoantigens and negative selection of autoreactive T cells in the thymus. Mutations in AIRE cause the rare autosomal-recessive systemic autoimmune disease autoimmune polyendocrinopathy with candidiasis and ectodermal dystrophy. In the TCGA database, there is an overexpression of genes associated with this disease that shows some sequence similarity (eg, NEFM) or extensive sequence similarity (eg, RYR3) with major autoimmune targets, such as muscle.²⁶ Notably, the mainly neuronal striational muscle antigen RYR3 shares homology with muscular RYR1 and cardiac RYR2. This finding could explain the simultaneous presence of several autoimmune disorders and myasthenia gravis as we have reported elsewhere.²⁷

We observed a positive correlation between response, progression-free survival, overall survival, PD-L1 expression, and an interferon-γ signature, similar to what has been described in other tumours.¹⁷ Several studies^7,8 have shown that in thymic epithelial tumours PD-L1 expression is present.

The mutational burden, which correlates with response to PD-1 antibodies in NSCLC,²⁸ could not be assessed in our study because of the small size of our sequencing platform and the relatively low mutation burden in this tumour type.^16,26 None of our patients had high levels of microsatellite instability.²⁹ The pattern of mutations found was similar to previous reports.^16,30 The association between TP53 mutations and low PD-L1 expression and between CYLD mutations and high PD-L1 expression are of interest. TP53 mutations and their association with overall survival have been previously described in thymic carcinoma.³⁰ The frequency of other mutations was low compared with other, more common, epithelial tumours.^16,26 CYLD is a deubiquitinating enzyme and mutations in CYLD are responsible for familial cylindromatosis and other rare hereditary diseases.³¹ CYLD is a tumour suppressor gene, and the frequency of its mutations in thymic carcinomas is greater than 10%. CYLD mutations were correlated with high PD-L1 expression.

An important consideration when choosing treatment is cost. The high cost of anticancer drugs, including immune checkpoint inhibitors, is an active area of research and discussion. For immunotherapy to have a broader impact, further research is needed to find ways to mitigate the costs associated with immune checkpoint inhibitor therapy (such as shorter treatment duration and development of predictive biomarkers).

In conclusion, pembrolizumab is an active treatment in patients with advanced thymic carcinoma. The response and response duration compares favourably with other treatments that are available for this disease. The development of autoimmune disorders, although severe in some cases, is typically manageable, but requires careful monitoring, development of predictive and preventive measures, such as the use of frequent measurements of troponin and creatine phosphokinase, and screening for the presence of high levels of autoantibodies. Use of novel technologies, such as T-cell receptor sequencing, has the potential to identify patients at higher risk of toxicity.

Research in context.

Evidence before this study

At the time of planning our study in 2014, we searched PubMed and ClinicalTrials.gov with the terms “thymic carcinoma”, “thymic epithelial malignancy”, “pembrolizumab”, and “anti-PD-1”. We searched articles published before Dec 31, 2013. The search was not limited to English language publications. We did not identify any studies of anti-PD-1 therapy for thymic carcinoma. We used the data on PD-L1 expression in thymic carcinomas and the correlation between PD-L1 expression and efficacy of anti-PD-1 therapy observed in other tumour types and the higher mutational burden of thymic carcinoma as compared with thymoma, as the basis for this trial.

Added value of the study

The anti-PD-1 antibody pembrolizumab was active (objective response 22·5%) and led to durable responses in heavily pretreated patients with metastatic thymic carcinoma.

Six (15%) patients developed serious autoimmune toxicity.

Implications of all the available evidence

To the best of our knowledge, these findings represent the first prospective data to demonstrate the antitumour activity of anti-PD-1 therapy in pretreated metastatic thymic carcinoma. Given the relative paucity of effective systemic therapies in the setting of metastatic disease, pembrolizumab might represent a new therapeutic option for these patients. Patients with thymic carcinoma have an increased risk of developing serious autoimmune toxicities after treatment with pembrolizumab and therefore close monitoring of patients is necessary.