Introduction

Immune checkpoint inhibitors (ICPIs) have dramatically changed the treatment of non–small cell lung cancer (NSCLC) in recent years by demonstrating an overall survival benefit.1 At present, ICPIs are being used as the first-line treatment with or without chemotherapy. Programmed cell death ligand 1 (PD-L1) is the only established biomarker for predicting the efficacy of ICPI therapy.1 Currently, PD-L1 expression is used as a companion diagnostic parameter to clinical markers and a predictive indicator of the therapeutic efficacy to anti-PD-1 / PD-L1 antibodies.2 It is well known that using the immunostaining method for PD-L1 expression evaluation has several limitations.3 Namely, it does not show how the expression changes over time, eg, in response to interferon γ induction in the tumor microenvironment.3 Therefore, even in the same patient, the expression of PD-L1 may differ depending on the time of sample collection.3 In addition, PD-L1 expression in tumors is not uniform and it tends to be localized around interferon γ–producing T cells, resulting in differences between measurements at various biopsy sites.3 Considering the challenges related to the methodology and interpretation of PD-L1 immunostaining, other predictive factors have been sought. In recent years, it has been reported that peripheral blood leukocytes may be a promising predictor in ICPI therapy.4-19 A study by Tanizaki et al11 demonstrated that a low absolute neutrophil count, high absolute lymphocyte count, and high absolute eosinophil count were significantly and independently associated with better survival. In that study, blood samples were obtained only within 7 days before ICPI therapy initiation. We also encountered a patient with a high peripheral blood eosinophil count and a successful response to ICPI therapy.20 In clinical practice, determination of complete blood count (CBC) is essential for chemotherapy of advanced lung cancer. If investigating the number of eosinophils and their percentage could help us predict the efficacy of ICPI therapy, it would be a cheap and convenient indicator. Eosinophils are known to play an important role in parasitic and allergic diseases, and the number of peripheral blood eosinophils is known to be increased in those patients.21 On the other hand, eosinophil infiltration into tumor lesions and peripheral eosinophilia have been reported in patients with cancer.22-25 The exact mechanism and clinical importance of eosinophilia in peripheral blood and tumor tissue remain unclear. However, attempts have been made to elucidate it, albeit little by little, eg, the involvement of cytokines such as interleukin 5 and interleukin 16 was examined.10,24 As mentioned above, ICPIs have become common antitumor drugs in the treatment of cancer in various organs, and more attention is being paid to the relationship between eosinophils and cancer immunity.10,17,26,27 The role of eosinophils in cancer immunity remains unknown; however, the relationship among peripheral eosinophils and antitumor effects, response duration, and survival has recently been of interest.6-8,10,11,16-18,26 As currently there is no valid biomarker to replace PD-L1, expectations associated with the role of peripheral eosinophils as biomarkers for ICPI efficacy have been rising.6-8,10,11,16-18,26 Yet, there are some unclear issues pertaining to the significance of peripheral blood eosinophils in ICPI therapy. We do not know what is of greater prognostic value: the eosinophil count before ICPI therapy initiation11,12 or a change in the eosinophil count after starting treatment.6,8,10,15,17,18 Even if changes in the eosinophil count after ICPI therapy initiation are more relevant, it is unclear when the measured eosinophil count is meaningful. Furthermore, there has been uncertainty as to whether the relative increase in eosinophils or an increase in absolute numbers is crucial.

In the present study, we focused on peripheral eosinophils and aimed to clarify the relationship between eosinophil counts and percentages and the efficacy of ICPI therapy. This study was performed to determine whether eosinophils before or during ICPI therapy had a greater impact on patient response and whether eosinophil counts or percentages were a better predictor of therapeutic response.

Patients and methods

Patients

We analyzed the medical records of all patients diagnosed with NSCLC in 3 tertiary hospitals in Japan (Mito Medical Center, University of Tsukuba–Mito Kyodo General Hospital, Ryugasaki Saiseikai Hospital, and Tsukuba University Hospital) between February 2016 and December 2019. Patients with NSCLC treated with ICPI monotherapy, or combination therapy with ICPI and chemotherapy, during this period were included in this study. Lung cancer was diagnosed based on the World Health Organization classification. Tumor-node-metastasis staging (TNM Classification, 8th edition) using head computed tomography or magnetic resonance imaging, bone scans, and ultrasonography and / or computed tomography of the abdomen was performed in all patients prior to ICPI therapy initiation. Patients with the following comorbidities and with a history of treatment for these conditions were excluded: parasitic infestations, allergic diseases, autoimmune diseases, and hematologic malignancies. Patients with chronic obstructive pulmonary disease and those with bronchial asthma and chronic obstructive pulmonary disease overlap requiring systemic steroid use were also excluded. Particular attention was paid to adrenal insufficiency as an immune-related adverse event (irAE). Patients who developed eosinophilia associated with adrenal insufficiency as an irAE were excluded from this study. Patient demographic data including age, sex, Eastern Cooperative Oncology Group score for performance status, histopathology, disease stage, PD-L1 expression, objective tumor response, and survival were obtained from the patients’ medical charts. Tumor response was evaluated as complete response, partial response, stable disease, or progressive disease, according to the Response Evaluation Criteria in Solid Tumors (Version 1.1).

Peripheral eosinophil count and percentage measurement

Eosinophil counts and percentages were measured at the same time as CBC measurements before and during ICPI therapy. Results were obtained from the medical records of each patient. Counts for leukocyte subpopulations were measured by routine clinical laboratory analysis using the Sysmex XN-3000 analyzer (Sysmex Co., Ltd. Kobe, Japan).

Statistical analysis

The χ2 test was used to compare nominal variables. We used the nonparametric Mann–Whitney test to compare values with unknown population variance. Comparisons of 3 or more independent groups were performed using the Kruskal–Wallis test. To determine the cutoff value, a receiver operating characteristics (ROC) curve analysis was performed. The optimal cutoff was determined as the value for which the point on the ROC curve had the minimum distance to the upper left corner. Time to treatment failure (TTF) was used to evaluate prolonged therapeutic efficacy. We adopted the definition of TTF that is commonly used in cancer treatment: the interval from initiation of therapy with ICPIs to treatment discontinuation or the last follow-up visit. Time to treatment failure was estimated by the Kaplan–Meier method and compared using the log-rank test. We used the Cox proportional hazards model and forward-backward stepwise method to determine the independent variables used in the final model. Time to treatment failure was the dependent variable in that model. All statistical analyses were conducted using the BellCurve software for Excel, version 3.0 (SSRI Co, Ltd, Tokyo, Japan). A P value less than 0.05 was considered significant.

Ethics

This study conformed to the Ethical Guidelines for Clinical Studies issued by the Ministry of Health, Labor, and Welfare of Japan. Written informed consent for a noninterventional retrospective study was obtained from each patient. The analysis of the medical records of patients with lung cancer was approved by the ethics committee of Mito Medical Center–University of Tsukuba Hospital (NO 16–66).

Results

Patient characteristics

We analyzed the clinical characteristics of 190 patients who met all inclusion criteria within the study period. Detailed data of the study patients are shown in Tables 1 and 2. Of those 190 patients enrolled, 149 (78.4%) were male. The median age was 69 (range, 29–87) years. Lung cancers included 113 cases of adenocarcinoma, 59 of squamous cell carcinoma, and 18 of other types (8 cases of poorly differentiated NSCLC, 3 of large-cell neuroendocrine carcinoma, 3 of pleomorphic carcinoma, 2 of NSCLC not otherwise specified, and 2 of adenosquamous cell carcinoma). A total of 136 patients had distant metastases. The ICPI monotherapy was started in 157 patients, and 33 patients received ICPI combined with carboplatin-containing chemotherapy. The following ICPIs were used: nivolumab in 84 patients, pembrolizumab in 77, and atezolizumab in 29. As shown in Tables 1 and 2, there was no difference in the patients’ baseline data in terms of treatment regimens used.

Table 1. Characteristics of 190 study patients with non–small cell lung cancer treated with immune checkpoint inhibitor therapy

Characteristic

ICPI monotherapy (n = 157)

ICPI + CBDCA-containing chemotherapy (n = 33)

P value

Age, y, median (IQR)

69 (63–76)

69 (64–74)

0.55

Sex

Male

124

25

0.65

Female

33

8

PS

0–1

133

30

0.43

2–3

24

3

Pathology

AD

95

18

0.47

SQ

49

10

Other typesa

13

5

Stage

IIIA–C

46

7

0.39

IVA

32

10

IVB

79

16

Treatment regimens were as follows: 84 patients on nivolumab monotherapy, 77 on pembrolizumab (48 on pembrolizumab monotherapy, 15 on pembrolizumab + carboplatin + pemetrexed, and 14 on pembrolizumab + carboplatin + paclitaxel), and 29 on atezolizumab (25 on atezolizumab monotherapy, 2 on atezolizumab + carboplatin + pemetrexed, and 2 on atezolizumab + carboplatin + paclitaxel). Comparisons of means between the 2 independent groups were performed using the χ2 or Mann–Whitney tests. Comparisons of 3 or more independent groups were performed using the Kruskal–Wallis test.

a Other types: 8 cases of poorly differentiated non–small cell lung cancer, 3 of large-cell neuroendocrine carcinoma, 3 of pleomorphic carcinoma, 2 of non–small cell lung cancer not otherwise specified, and 2 of adenosquamous cell carcinoma

Abbreviations: AD, adenocarcinoma; CBDCA, carboplatin; ICPI, immune checkpoint inhibitors; IQR, interquartile range; PS, performance status; SQ, squamous cell cancer

Table 2. Characteristics of the study patients by the drugs used

Characteristic

ICPI monotherapy

(n = 157)

ICPI + CBDCA-containing chemotherapy (n = 33)

P value

P

(n = 48)

N

(n = 84)

A

(n = 25)

PEM + A (n = 2)

PEM + P (n = 15)

Taxol + A (n = 2)

Taxol + P (n = 14)

Age, y, median (IQR)

72 (68–78)

68 (60–76)

69 (60–73)

68 (59–76)

67 (65–72)

69 (64–75)

71 (65–76)

0.12

Sex

Male

41

66

17

2

10

2

11

0.5

Female

7

18

8

0

5

0

3

PS

0–1

38

71

24

2

13

2

13

0.52

2–3

10

13

1

0

2

0

1

Pathology

AD

30

46

19

2

13

2

1

0.29

SQ

16

29

4

0

0

0

10

Othera

2

9

2

0

2

0

3

Stage

IIIA-C

9

25

12

0

2

0

5

0.24

IVA

14

18

0

0

7

1

2

IVB

25

41

13

2

6

1

7

a Other types: 8 cases of poorly differentiated non–small cell lung cancer, 3 of large-cell neuroendocrine carcinoma, 3 of pleomorphic carcinoma, 2 of non–small cell lung cancer not otherwise specified, and 2 of adenosquamous cell carcinoma

Abbreviations: A, atezolizumab; N, nivolumab; P, pembrolizumab; PEM, pemetrexed; others, see Table 1

Distribution of peripheral eosinophils

Peripheral eosinophil counts at the time of ICPI therapy initiation and the maximum eosinophil counts during ICPI therapy are presented as box plots in Figure 1A and 1B, respectively. The distribution of the peripheral eosinophil percentage at the time of ICPI therapy initiation and the maximum eosinophil percentage during ICPI therapy are shown in Figure 1C and 1D, respectively. When eosinophil counts increased in a patient, their eosinophil percentage increased as well.

Figure 1. A, B – box plots of pretreatment peripheral eosinophil counts (A) and maximal eosinophil counts measured during the course of immune checkpoint inhibitor therapy (B); C, D – distribution of the percentage of pretreatment peripheral eosinophils (C) and the maximum eosinophil percentage measured during the course of therapy (D)

Eosinophil measurement in response to immune checkpoint inhibitor therapy

Eosinophil counts and percentages at the time of ICPI therapy initiation as well as the maximum eosinophil counts and percentages during ICPI therapy in the study patients who did or did not respond to ICPI therapy are shown in Table 3. There was a significant difference between patients with controlled disease and those with progressive disease in terms of eosinophil counts at the time of ICPI therapy initiation and the maximum eosinophil counts and percentages during ICPI therapy. In Table 4, we analyzed patients’ data on peripheral eosinophils and response to ICPI therapy. Patients with complete response, partial response, stable disease, and progressive disease differed with regard to eosinophil counts at the time of ICPI therapy initiation and the maximum eosinophil counts and percentages during ICPI therapy. Changes in the percentage of eosinophils at the time of ICPI therapy initiation and the maximum percentage of eosinophils during ICPI therapy in each patient with NSCLC are shown in Supplementary material, Figure S1.

Table 3. Peripheral eosinophils in the study patients with controlled and progressive disease

Eosinophils

Controlled diseasea (n = 135)

Progressive disease (n = 55)

P value

At the time of ICPI therapy initiation

Count, n/μl

151 (97–297)

92 (62–245)

0.02

Percentage

2.4 (1.3–4)

1.9 (0.8–3.3)

0.07

During the ICPI therapy

Maximum count, n/μl

372 (187–741)

218 (133–381)

0.002

Maximum percentage

5.9 (2.9–8.3)

3.4 (1.6–5.3)

<⁠0.001

Data are presented as median (interquartile range). The Mann–Whitney test was used for statistical analysis.

a Patients with controlled disease included those with complete response, partial response, and stable disease.

Abbreviations: see Table 1

Table 4. Peripheral eosinophils and response to immune checkpoint inhibitor therapy in the study patients with non–small cell lung cancer

Eosinophils

CR (n = 4)

PR (n = 64)

SD (n = 67)

PD (n = 55)

P value

At the time of ICPI therapy initiation

Count, n/μl

512 (347–635)

152 (88–282)

148 (104–265)

92 (62–245)

0.02

Percentage

7.5 (5.8–7.9)

2.4 (1.3–3.8)

2.2 (1.4–3.9)

1.9 (0.8–3.3)

0.09

During the ICPI therapy

Maximum count, n/μl

777 (457–1309)

437 (228–841)

244 (153–537)

218 (133–381)

<⁠0.001

Maximum percentage

8.4 (7.2–11)

7.1 (4.1–10.3)

4.2 (2.5–6.8)

3.4 (1.6–5.3)

<⁠0.001

Data are presented as median (interquartile range). The Kruskal–Wallis test was used for statistical analysis.

Abbreviations: CR, complete response; PD, progressive disease; PR, partial response; SD, stable disease; others, see Table 1

Determination of peripheral eosinophil cutoff values

The ROC curve analysis was performed to determine the cutoff value that distinguished patients with controlled disease from those with progressive disease. The ROC curves for eosinophil counts and percentages at the time of ICPI therapy initiation and the maximum eosinophil counts and percentages during ICPI therapy are presented in Figure 2.

Figure 2. Receiver operating characteristics curves for the cutoff levels of eosinophil counts and percentages at the time of immune checkpoint inhibitor (ICPI) therapy initiation and the maximum eosinophil counts and maximum eosinophil percentages during the clinical course of immune checkpoint inhibitor therapy

The area under the ROC curve for eosinophil counts and eosinophil percentages at the time of ICPI therapy initiation was 0.6106 (95% CI, 0.5198–0.7014) and 0.5849 (95% CI, 0.4934–0.6764), respectively. The area under the ROC curve for the maximum eosinophil counts and the maximum eosinophil percentage during the ICPI therapy was 0.6442 (95% CI, 0.5615–0.7269) and 0.6814 (95% CI, 0.6007–0.7621), respectively. The most appropriate cutoff values for patients with controlled disease and those with progressive disease were an eosinophil count of 330/μl and an eosinophil percentage of 5% during the ICPI therapy. When the cutoff value was set to 5%, the sensitivity of the model reached 60.7%, the specificity, 27.3%, and the odds ratio, 4.13.

Association of eosinophils at the time treatment failure with immune checkpoint inhibitors therapy

Based on the cutoff values determined through the ROC curve analysis and those reported in previous studies,11,28 the maximum eosinophil count was set to 300/μl and the cutoff value, to 5%. Patients were divided into 2 groups to examine TTF. Regarding eosinophil counts and percentages at the time of ICPI therapy initiation, a significant difference in TTF between the 2 groups was observed, when the cutoff values were set to 150/μl (P = 0.046) and 3% (P = 0.03), respectively. As for the maximum eosinophil counts and percentages during ICPI therapy, a significant difference in TTF between the 2 groups was noted, when the cutoff values were set to 150/μl (P <⁠0.001), 300/μl (P <⁠0.001), 3% (P <⁠0.001), and 5% (P <⁠0.001), respectively (Figure 3). The multivariable analysis of TTF showed that a maximum eosinophil percentage greater than 5% during ICPI therapy was a significant predictive indicator for prolonged therapeutic efficacy, as were “good performance status” and “PD-L1 expression level of 50% or more” (Table 5).

Figure 3. Time to treatment failure curves with a cutoff value for eosinophil (Eo) percentage of 3% (A) and 5% (B) and for eosinophil count of 150/μl (C) and 300/μl (D) during immune checkpoint inhibitor therapy

Table 5. Results of multivariable analysis using the Cox proportional hazards model

Factor

Exp

95% CI

P value

PS (ECOG)

1.5112

1.1806–1.9344

0.001

PD-L1 ≥50%

0.6549

0.4326–0.9914

0.0454

Maximum eosinophil percentage ≥5%

0.3978

0.2630–0.6016

<⁠0.001

The forward-backward stepwise method was used for the choice of independent variables used in the Cox proportional hazards model.

Abbreviations: ECOG, Eastern Cooperative Oncology Group; Exp, exponential function; PD-L1, programmed cell death ligand 1; others, see Table 1

Discussion

The level of PD-L1 expression has been used as a biomarker of the efficacy of ICPI therapy. However, PD-L1 commonly exhibits a nonuniform immunostaining pattern across cancer tissue, which leads to inaccuracies in interpreting the overall expression. Numerous studies have sought to identify other biomarkers in ICPI therapy,10,17,26,27 but there is no established biomarker other than PD-L1.9 The identification of a biomarker that is also a routinely used clinical marker would be of value, as it may represent a convenient and low-cost option that does not require special equipment.

In our clinical practice, we observed a patient with NSCLC who responded to ICPI therapy during eosinophilia and progressed following eosinophilia normalization.20 Based on that experience, we focused on the change in the number of peripheral eosinophils before and during ICPI therapy by investigating the relationship between peripheral eosinophil counts and percentages and the efficacy of ICPI therapy. Eosinophilia may develop in adrenal deficiency due to ICPI use.29,30 However, according to a review by Domagała-Kulawik et al,31 hematologic irAEs do occur very rarely. The authors noted that hematologic irAEs should be distinguished from transient changes in laboratory blood test results at the time of ICPI therapy initiation. Eosinophils play the key role in allergic and parasitic diseases. Although their role in cancer immunity was not mainstream, there have been reports on cancer patients with peripheral eosinophilia and those with eosinophil infiltration into tumors.22-25 The mechanism and clinical significance of eosinophilia in peripheral blood and tumor tissues remains unclear, but the involvement of some cytokines has been suggested.10,24 In recent years, ICPIs have become common drugs for the treatment of cancer in various organs and more attention is being paid to the relationship between eosinophils and cancer immunity.10,17,26,27 In particular, peripheral eosinophils as a biomarker of ICPI therapy efficacy have been a matter of interest.6-8,10,11,16-18,26 Simon et al,10 who studied both basic and clinical aspects of peripheral blood eosinophils in ICPI treatment, demonstrated that response to ICPI treatment was associated with eosinophil accumulation in the peripheral blood of patients with melanoma.10 Interestingly, they showed an association between elevated serum levels of interleukin 16 and an increase in peripheral eosinophils during ICPI therapy. Using immunohistochemistry, they also observed an enhanced eosinophil degranulation and a positive correlation between eosinophils and CD8+ T cell infiltration of tumor tissues in patients with melanoma treated with ICPIs.10 Additionally, Umansky et al7 reported on the involvement of myeloid-derived suppressor cells and eosinophils in the therapeutic effect of ipilimumab.7 In patients with melanoma, for example, an elevated peripheral eosinophil count at the time of diagnosis,8 during ipilimumab treatment,6 and after initial ICPI therapy10 were associated with a favorable prognosis. The relationship between peripheral eosinophils and the therapeutic effect of ICPI use was investigated not only in patients with melanoma but also in those with NSCLC11,13,16 and Hodgkin lymphoma.19 Regarding patients with NSCLC, there have been reports focused on eosinophilia before ICPI use.11,12 Tanizaki et al11 used CBC measured within 7 days before ICPI therapy initiation.11 On the other hand, some studies have concentrated on the increase following ICPI therapy initiation.10,16 Simon et al10 evaluated peripheral blood drawn 12 to 32 days after the first administration of ICPIs.10 However, it is not clear at which timepoint, before or after ICPI therapy, peripheral eosinophilia should be deemed relevant. If the change in eosinophils after ICPI therapy initiation is of clinical significance, there also appears the question as to when eosinophilia should be considered important. In addition, it remains unknown which parameter is of greater importance—eosinophil counts or percentages—and what is the optimal cutoff value for each of them.

In the present study, we observed an increase in both peripheral eosinophil counts and eosinophil percentages after ICPI therapy initiation, regardless of whether the patients had controlled or progressive disease. The median time to note the maximum eosinophil percentage was 5 weeks in patients with controlled disease. The ROC curve analysis showed that the cutoff values for patients with controlled disease and those with progressive disease were an eosinophil count of 330/μl and an eosinophil percentage of 5%, where an eosinophil count of 300/μl may be of use in clinical practice. We observed a significant difference in TTF between patients stratified by these cutoff values. Finally, we identified eosinophil percentages greater than 5% during ICPI therapy as a significant predictive factor for prolonged therapeutic efficacy in a multivariable analysis. The results of this study suggest that it may be possible to predict response to ICPI therapy from peripheral eosinophils. Based on the obtained findings, we observed that changes in eosinophils after treatment were more significant than those at the pretreatment stage. Furthermore, we determined that changes in eosinophils noted within 5 weeks of ICPI therapy initiation were of greatest importance.

Although clinically significant changes in eosinophil levels has been considered a potential biomarker for eosinophilia, there have been different views on the significance of peripheral eosinophila. In other words, its importance as a biomarker depended on whether it exhibited therapeutic responsiveness, long-term progression-free survival, or good prognosis. Its significance was generally regarded as a feature of a favorable marker. On the other hand, there have been reports on patients who developed eosinophilia syndrome with unfavorable complications.32,33 Considering those complications, we hypothesized that eosinophilia after starting treatment is an indicator of cancer immunity initiation.

Admittedly, this study had several limitations. First, it was a retrospective study that included patients with various baseline characteristics. Second, it involved a limited number of patients with a short follow-up, and the number of patients required was not preset on a statistical basis. Third, we did not elucidate the relationship between changes in eosinophils following ICPI therapy and the biological role of eosinophils. Fourth, in this study, there was no significant difference in the patients’ baseline characteristics stratified by treatment regimens, so all patients were analyzed together. However, it was considered important to investigate the change of eosinophils according to treatment methods used. In particular, cytotoxic anticancer drugs can affect peripheral blood cells, so it may be necessary to analyze patients receiving them separately. We showed that ICPI therapy was effective in patients who had increased eosinophil counts and elevated eosinophil percentages. Yet, it remains unclear whether this therapeutic effect weakened consistently with the decreasing eosinophil count and percentage. This issue could not be solved. It also should be clarified whether a temporary or continuous increase in eosinophils is of importance. Our team is currently examining this issue. The latter seems to have a better prognosis so far (data not shown).

It would not be difficult or expensive to include the measurement of peripheral eosinophils when measuring CBC and investigate the association of eosinophil counts and percentages with the efficacy of ICPI therapy. Along these lines, we are now planning a sufficiently powered large-scale study with a higher number of patients to confirm the findings from this study.