Article Text

Original research
Myeloid subsets impede the efficacy of anti-PD1 therapy in patients with advanced gastric cancer (WJOG10417GTR study)
  1. Hirokazu Shoji1,
  2. Chie Kudo-Saito2,
  3. Kengo Nagashima3,
  4. Hiroshi Imazeki1,2,
  5. Kai Tsugaru4,
  6. Naoki Takahashi5,
  7. Takeshi Kawakami6,
  8. Yusuke Amanuma7,
  9. Takeru Wakatsuki8,
  10. Naohiro Okano9,
  11. Yukiya Narita10,
  12. Yoshiyuki Yamamoto11,
  13. Rika Kizawa12,
  14. Kei Muro10,
  15. Kazunori Aoki2 and
  16. Narikazu Boku1,13
  1. 1Department of Gastrointestinal Medical Oncology, National Cancer Center Hospital, Tokyo, Japan
  2. 2Department of Immune Medicine, National Cancer Center Research Institute, Tokyo, Japan
  3. 3Biostatistics Unit, Clinical and Translational Research Center, Keio University Hospital, Tokyo, Japan, Tokyo, Japan
  4. 4Division of Gastroenterology and Hepatology, Keio University Hospital, Tokyo, Japan, Tokyo, Japan
  5. 5Department of Gastroenterology, Saitama Cancer Center, Saitama, Japan
  6. 6Division of Gastrointestinal Oncology, Shizuoka Cancer Center, Shizuoka, Japan
  7. 7Clinical Trial Promotion Department, Chiba Cancer Center, Chiba, Japan
  8. 8Department of Gastrointestinal Medical Oncology, Cancer Institute Hospital of JFCR, Tokyo, Japan
  9. 9Department of Medical Oncology, Kyorin University Faculty of Medicine, Tokyo, Japan
  10. 10Department of Clinical Oncology, Aichi Cancer Center Hospital, Nagoya, Japan
  11. 11Department of Gastroenterology, University of Tsukuba Hospital, Tsukuba, Japan
  12. 12Department of Medical Oncology, Toranomon Hospital, Tokyo, Japan
  13. 13Department of Medical Oncology and General Medicine, IMS Hospital, Institute of Medical Science, University of Tokyo, Tokyo, Japan
  1. Correspondence to Dr Chie Kudo-Saito; ckudo{at}ncc.go.jp

Abstract

Background Gastric cancer (GC) is one of the most common and deadly malignant diseases worldwide. Despite revolutionary advances, the therapeutic efficacy of anti-PD1/PDL1 monoclonal antibodies in advanced GC is still low due to the emergence of innate and acquired resistance to treatment. Myeloid cells represent the majority of human immune cells. Therefore, their increase, decrease, and abnormality could have a significant impact on the patient’s immune system and the progression of cancer, and reprogramming, inhibiting, and eliminating the tumor-supportive types may improve the immunological situation and efficacy of immunotherapy. However, the significance of myeloid cells in anti-PD1/PDL1 therapy remains unclear in GC. In the WJOG10417GTR study on GC, we sought to identify myeloid determinants that could predict anti-PD1 therapeutic efficacy and also serve as potential therapeutic targets.

Methods We collected tumor tissues and peripheral blood from 96 patients with advanced GC before and 1 month after anti-PD1 nivolumab monotherapy, and the isolated whole leucocytes were analyzed by flow cytometry for various immune cell populations, including many myeloid subsets. Then, the relationship between the cellular levels and progression-free survival (PFS) or overall survival (OS) was statistically analyzed.

Results We found that high levels of several myeloid subsets expressing molecules that have been targeted in drug discovery but not yet approved for clinical use were significantly associated with shorter PFS/OS as compared with low levels: PDL1+ and CTLA4+ myeloid subsets within tumors at baseline, PDL1+, B7H3+ and CD115+ myeloid subsets in peripheral blood at baseline, and LAG3+, CD155+ and CD115+ myeloid subsets in peripheral blood at post-treatment.

Conclusions This study revealed that these myeloid subsets are significant risk factors in nivolumab therapy for advanced GC. Targeting them may be useful as diagnostic biomarkers to predict potential anti-PD1 therapeutic efficacy, and also as therapeutic targets for accelerating the development of new drugs to improve clinical outcomes in immunotherapy for GC.

  • Gastric Cancer
  • Immune Checkpoint Inhibitor
  • Nivolumab
  • Biomarker
  • Myeloid

Data availability statement

All data relevant to the study are included in the article or uploaded as online supplemental information.

http://creativecommons.org/licenses/by-nc/4.0/

This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See http://creativecommons.org/licenses/by-nc/4.0/.

Statistics from Altmetric.com

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

WHAT IS ALREADY KNOWN ON THIS TOPIC

  • Anti-PD1/PDL1 therapy has attracted great attention as a promising strategy for treating a variety of cancers, including gastric cancer (GC). However, the clinical responses are limited to a small portion of patients, and despite intensive research, biomarkers to predict treatment response and effective combination regimens to enhance therapeutic efficacy have yet to be established. It has long been reported that immunosuppressive and proinflammatory myeloid cells promote cancer progression and metastasis directly and indirectly via inducing immunosuppression, exhaustion, and dysfunction. However, neither biomarkers nor cancer therapeutic drugs targeting myeloid cells have yet been established in clinical practice, and the significance of myeloid cells in anti-PD1/PDL1 therapy for GC remains unclear. One of the main reasons may be inappropriate sampling methods, such as freezing and Ficoll isolation, which result in the loss of a part of the myeloid population.

WHAT THIS STUDY ADDS

  • By using our own methods to isolate whole leucocytes from fresh tumor tissues and peripheral blood of patients with advanced GC before and 1 month after anti-PD1 nivolumab monotherapy, this study revealed that several myeloid subsets were significant risk factors for nivolumab therapy for GC. This study provides not only diagnostic biomarkers that may predict potential anti-PD1/PDL1 therapeutic efficacy but also therapeutic targets that may improve clinical outcomes in the treatment of GC.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

  • This study highlights the importance of the sampling quality, which significantly impacts any analytical results and suggests a new direction for future translational research, in which careful sampling should be widely practiced. The findings suggest that targeting the identified myeloid subsets is of great importance for improving the efficacy of PD1/PDL1 therapy in GC, and provides a new direction that this should be widely applied and evaluated in clinical trials for various other cancer types in the future.

Introduction

Gastric cancer (GC) is the fifth most common malignancy worldwide and the third-leading cause of cancer-related mortality, and many patients experience recurrence, even after curative resection, and disease progression during systemic chemotherapy for unresectable disease.1 In recent years, inhibitory monoclonal antibodies (mAbs) targeting immune checkpoint (IC) pathways mediated by specific molecules, such as CTLA4, PD1, PDL1, LAG3 and TIGIT, (ICIs), have attracted great attention in treatment of a variety of cancers, including GC.2 For example, anti-PD1 nivolumab monotherapy as a later-line treatment provided a significant survival benefit to patients with advanced GC and/or gastroesophageal junction cancer (collectively designated AGC).3 The therapeutic efficacy of nivolumab was significantly enhanced by combination with cytotoxic chemotherapy in AGC patients having a combined PDL1-positive score (CPS) ≥5 or ≥1, as well as in all randomly assigned patients.4 However, its initial and durable efficacy is limited to a small proportion of patients due to innate and acquired resistance to the treatment.2 Thus, it is urgently needed to increase the effectiveness of immunotherapy for AGC.

The significant success of the anti-CTLA4/PD1/PDL1 mAbs has certainly led to the development of various drugs in cancer immunotherapy: new drugs targeting different IC pathways, small molecule inhibitors targeting multiple signaling pathways, genetically modified cells, and vaccines.5 6 However, most of them are still in progress or have failed. On the other hand, identification of biomarkers to predict potential therapeutic efficacy has been widely pursued, and several ones, including the CPS-PDL1,7 microsatellite instability (MSI),8 or mutation burden9 in tumors have been successfully applied in clinical practice. However, these are not necessarily correlated with clinical outcomes, and thus more accurate biomarkers are also still needed in clinical settings.

As an unspectacular but extremely crucial strategy for translational research, we have been working on refining sampling methods for patient specimens because the sampling quality has a significant impact on any analytical results. Most clinical studies often use frozen samples due to the convenience and conventionally apply the Ficoll isolation method, which results in the loss of some leucocyte fractions, such as myeloid cells. However, unless the sample is analyzed in a state as close as possible to the patient’s body, it would be impossible to obtain results that can exactly contribute to clinical practice. We have been using a simple method to remove red blood cells with/without specially density-adjusted Ficoll to isolate as many leucocytes as possible. This approach led to identification of unique tumor-supportive myeloid subsets expressing IC molecules, including CTLA410 and LAG3,11 although these molecules are usually noticed only in T/NK cells. Myeloid cells account for the majority of the human immune system. Therefore, their increase, decrease, and abnormality could have a significant impact on the patient’s immune system and the progression of cancer, and reprogramming, inhibiting, and eliminating the tumor-supportive types may improve the immunological situation and efficacy of immunotherapy. However, the significance of myeloid cells in anti-PD1/PDL1 therapy remains unclear in GC. In this study, we isolated whole leucocytes from fresh tumor tissues and peripheral blood of AGC patients by our own methods before and 1 month after nivolumab monotherapy, and analyzed them by flow cytometry for various immune cell populations, especially myeloid subsets, in order to identify determinants that could predict anti-PD1 therapeutic efficacy and also serve as potential therapeutic targets.

Materials and methods

Patients and treatment

Eligibility and ineligibility for participation are shown in online supplemental table S1. Written informed consent was obtained from all patients before study enrolment. Finally, 100 patients with AGC were enrolled, and 96 patients (male×67, female×29; median age 69), excluding four patients due to the inability of treatment, received intravenous infusions of anti-PD1 mAb nivolumab at 3 mg/kg or 240 mg/body every 2 weeks until disease progression or unacceptable toxicity. Tumor response was evaluated according to the Response Evaluation Criteria in Solid Tumors version 1.1. Toxicities were graded based on the National Cancer Institute Common Terminology Criteria for Adverse Events 4.0.

Supplemental material

Sample collection and cell preparation

Endoscopic biopsy tumor tissues and EDTA-added peripheral blood were collected from the patients before and after nivolumab treatment (figure 1A). The post-treatment timing corresponded to approximately 1 month (after 21 days or more (22–43 days) from the first drug administration) when the immune status was considered to become stable after treatment initiation. Tumor tissues were dispersed into single-cell suspensions, and the cells were used for assays as tumor-infiltrating cells (TILs) after treatment with ACK Lysing Buffer (ThermoFisher) to remove red blood cells. Peripheral blood was centrifuged, followed by ACK treatment, and the cell pellet was used as peripheral blood cells (PBCs). After nivolumab treatment, it was impossible to collect peripheral blood from 12 patients, and tumor tissues from 25 patients due to poor condition.

Figure 1

Overview of the WJOG10417GTR study. (A) Patient enrolment and analyzed samples. Of 100 patients with advanced gastric cancer (AGC) enrolled in the WJOG10417GTR study, 4 patients who did not receive nivolumab therapy were excluded, and peripheral blood (PB) and tumor tissues from the remaining 96 patients were analyzed by ELISA and flow cytometry (FCM). In the ELISA analysis, 12 patients were excluded due to the inability to collect PB at post-treatment. In the FCM analysis, 5 patients were excluded due to less than 10% CD45+ cells (leucocytes) at pretreatment, and 16 patients (PB) or 29 patients (tumor) were excluded due to the inability to collect samples and less than 10% CD45+ cells at post-treatment. (B) progression-free survival (PFS) and overall survival (OS) of the patients (n=96).

Flow cytometric analysis

After Fc blocking, cells were stained with the following immunofluorescence-conjugated antibodies and other mAbs described in online supplemental figure S1: anti-CD45-APC-Cy7 (BioLegend), anti-CD11b-BV510 (BioLegend), anti-PDL1-BV785 (BioLegend), anti-CTLA4-BV785 (BD Biosciences), anti-LAG3-BV650 (BioLegend), anti-CD155-PerCP-Cy5.5 (BioLegend), anti-B7H3-PE-Cy7 (BioLegend), anti-CD115-BV711 (BioLegend), and the appropriate isotype control. Data were acquired using a BD LSR Fortessa X-20 cytometer (BD) and were analyzed by FlowJo software (BD). Before defining the specific molecular expressions, debris was first excluded by FSC/SSC followed by gating CD45+ leucocytes, and immunofluorescence intensity was compared with the isotype control (online supplemental figure S1). The TIL data were indicated as the percentage of the specific cells within the CD45+ leucocytes. The PBC data were indicated as the number of specific cells per ml of peripheral blood (×106/mL). Patients with CD45+ leucocytes <10% were considered ineligible for the integrated analysis and were excluded.

Statistical analysis

Significant differences (p<0.05) were evaluated using SAS software V.9.4 (Statistical Analysis System, RRID:SCR_008567) and R V.4.2.3 (R Foundation for Statistical Computing). Patients were divided into two groups, high and low levels of each marker, based on the cut-off values determined by change point of log HRs for progression-free survival (PFS) and overall survival (OS) using the Cox regression models with penalized splines. The HRs were obtained by univariable and multivariable models adjusting potential confounding factors (PS, prior gastrectomy, number of metastatic organs, and ALP), and 95% CIs were calculated. In this study, we used the cut-off values with the larger HRs calculated through careful analysis of each marker based on the method for modeling continuous-scale covariates as described elsewhere12 while avoiding obtaining outlier cut-off values (online supplemental methods). PFS/OS curves were estimated using the Kaplan-Meier method.

Results

High levels of myeloid subsets in tumor tissues at baseline are a significant poor prognostic factor in nivolumab treatment of AGC

96 patients received anti-PD1 mAb nivolumab (figure 1A, online supplemental table S2), and treatment-related adverse events, such as pruritus, rash, and hypothyroidism, were observed in around 5% of the patients, with grade 3 diarrhea occurring in only one case (online supplemental table S3). Median follow-up period was 21.0 months, and median number of nivolumab doses was four (range 0–42). Median PFS time (mPFS) for all patients was 1.8 months (95% CI 1.4 to 2.2), and median OS time (mOS) was 6.6 months (95% CI 4.9 to 9.3; figure 1B). Five cases (5.2%) did not show progressive disease (PD) at the end of the study period. In patients with target lesions, partial responses were achieved in six patients (13.0%) and stable disease in nine patients (19.6%), resulting in an objective response rate of 13.3% and a disease control rate of 33.3% (online supplemental table S4).

In TILs, high levels of antitumor effector CD3+CD8+ T cells at baseline, but not at post-treatment, were significantly associated with better PFS (HR=0.59, p=0.025; online supplemental figure S2A), and high levels of CD11b+ myeloid cells at baseline, but not at post-treatment, were significantly associated with worse OS (HR=1.82, p=0.016; online supplemental figure S2B). However, high levels of CD3CD56+ NK cells both before (HR=2.02, p=0.013) and after treatment (HR=2.16, p=0.016) were significantly associated with worse OS and high levels of FOXP3+CTLA4+ regulatory T cells (Tregs) at baseline were significantly associated with better OS (CD4+ Treg: HR=0.44, p=0.021; CD8+ Treg: HR=0.36, p=0.019; online supplemental figure S2). Speculating that the FOXP3 might be induced by immune activation and thus the FOXP3+CTLA4+ Tregs might not be of the immunosuppressive type, we also analyzed the relationship with another Treg subset expressing TIGIT, which is known to augment immunosuppressive functions of Tregs.13 However, high levels of CD8+FOXP3+TIGIT+ Tregs at baseline were still significantly associated with better PFS (online supplemental table S5). These results are inconsistent with the general concept that NK cells are potent tumor effector cells and that Tregs are immunosuppressive and impede immunotherapeutic efficacy, but suggest that anti-PD1 therapy is likely to be effective in patients with high levels of NK cell and Tregs.

We analyzed TILs in more detail for myeloid subsets. PDL1, which is expressed not only in cancer cells but also in various immune cells, plays a critical role as a PD1 ligand in putting a brake on antitumor immune responses,14 and its expression levels as the “CPS-PDL1” have been clinically used to select patients who receive anti-PD1/PDL1 therapy.7 However, patients with CPS-PDL1≥1 (39.5%) at baseline did not have a better prognosis than other cases (PFS: HR=1.04, p=0.849; OS: HR=1.10, p=0.701; online supplemental figure S3A). In contrast, high levels of the PDL1+ myeloid subset at baseline were significantly associated with worse PFS/OS (figure 2A,B), although the high levels at post-treatment tended to be associated with better prognosis (figure 2C,D). This suggests that baseline PDL1 expression in myeloid cells, rather than in tumor cells, is an important prognostic factor in anti-PD1 therapy.

Figure 2

High levels of myeloid subsets in tumor tissues at baseline are a significant risk factor in nivolumab treatment of advanced gastric cancer. Forest plots show the cut-off and median values (percentage in CD45+ cells) of each CD11b+ myeloid subset in tumor-infiltrating cells (TILs), HRs with 95% CI, and p values. (A) TILs at pretreatment for PFS (n=91). (B) TILs at pretreatment for OS (n=91). (C) TILs at post-treatment for PFS (n=67). (D) TILs at post-treatment for OS (n=67). OS, overall survival; PFS, progression-free survival.

Besides PDL1, there are many other molecules expressed in tumor-supportive myeloid subsets. In TILs, high levels of the CTLA4+ subset (PFS: HR=2.37, p=0.001; OS: HR=2.02, p=0.013) and the LAG3+ subset (PFS: HR=2.23, p=0.003; OS: HR=1.71, p=0.070) at baseline were significantly associated with poorer prognosis (figure 2A,B). High levels of the LAG3+ subset at both timings were significantly associated with worse PFS (pre: HR=2.23, p=0.003; post: HR=1.87, p=0.041; figure 2A,C). Kaplan-Meier curves clearly showed that patients with high levels of the CTLA4+ subset at baseline had the shortest PFS (mPFS: 0.9 months vs 2.0 months in low levels) and OS (mOS: 3.4 months vs 7.5 months in low levels) as compared with the other two subsets (figure 3). These suggest that the high infiltration of CTLA4+ myeloid subset within tumors at baseline is a significant risk factor in nivolumab therapy for AGC, as much or more than the long-established PDL1+ myeloid subset. After nivolumab treatment, the PDL1+ subset increased in 52% of patients, and the CTLA4+ and LAG3+ subsets increased in 44% of patients, but differences between pretreatment and post-treatment levels were not significantly associated with patient prognosis (online supplemental figure S4).

Figure 3

PDL1+ and CTLA4+ myeloid subsets in tumor tissues at baseline are significant poor prognostic factors in nivolumab treatment of AGC. Kaplan-Meier graphs show two groups, high (red lines) and low (blue lines), divided by the cut-off value of each myeloid subset in TILs at pretreatment (dotted lines) or post-treatment (solid lines) shown in figure 2, and median PFS (mPFS, months) or median OS (mOS, months) in the group. (A) CD11b+PDL1+ subset. (B) CD11b+CTLA4+ subset. (C) CD11b+LAG3+ subset. OS, overall survival; PFS, progression-free survival; TILs, tumor-infiltrating cell.

In TILs, myeloid subsets coexpressing three molecules, PDL1, CTLA4, and LAG3, were only present in a small proportion of both TILs and PBCs, but high levels of the PDL1+CTLA4+LAG3+ subset in TILs at post-treatment, but not pretreatment, were significantly associated with shorter PFS (mPFS: 1.7 months vs 3.2 months, HR=1.896, p=0.015; online supplemental figure S5A). This suggests that the PDL1+CTLA4+LAG3+ subset in TILs after treatment is a significant risk factor in nivolumab therapy for AGC, but that the subsets expressing single molecules, particularly the CTLA4+ subset, are more important high-risk factors. By immunohistochemical analysis (IHC) using pretreatment tissue sections that were available to us, we also analyzed the relationship between the levels of CD11b+CTLA4+ or CD11b+LAG3+ cell infiltration in tumors and patient prognosis. However, high levels of both subsets were significantly associated with longer PFS (online supplemental figure S6). The disparate results may have arisen from differences in the nature of the methods. Further studies are needed to draw a definitive conclusion.

High levels of myeloid cell subsets in peripheral blood are a significant poor prognostic factor in nivolumab treatment of AGC

In PBCs, post-treatment high levels of all cell populations, including antitumor effector T/NK/NKT cells as well as negative factors, such as Tregs, exhausted T cells (Texs) expressing IC molecules such as PD1, TIM3, LAG3 and TIGIT, and myeloid cells, were significant risk factors for PFS, although the reason is unclear (online supplemental figures S7–S9). Baseline high levels of CD4+ Tregs (HR=2.18, p=0.043) and Texs (HRs=1.55–2.28, p<0.05) were also significantly associated with worse PFS (online supplemental figure S9A). Post-treatment high levels of CD8+ Tregs were also significantly associated with worse OS (HR=2.55, p=0.014; online supplemental figure S9D). We also analyzed the relationship with another Treg subset expressing TIGIT. The HR decreased and no longer showed any significance for baseline Tregs, but the results for post-treatment CD8+ Tregs were nearly the same as those for FOXP3+CTLA4+ Tregs (online supplemental table S5). These suggest that systemic immune suppression and exhaustion greatly affect the therapeutic effectiveness of nivolumab in AGC.

Myeloid cell populations/subsets in the peripheral blood showed greater negative impacts than these cells, although neutrophil-lymphocyte ratio, which has long been the subject of widespread attention, obtained from serological data in routine blood tests showed no clinical significance (online supplemental figure S10). High levels of various myeloid cell populations, including CD68+ monocytes, CD11c+ DCs, and CD117+ mast cells, at both timings, were associated with worse PFS/OS (online supplemental figure S7). Particularly, high mast cell levels both before (PFS: HR=1.64, p=0.027; OS: HR=1.77, p=0.024) and after treatment (PFS: HR=1.96, p=0.008; OS: HR=2.10, p=0.019) had a significant impact on patient prognosis. Post-treatment high DC levels were also significantly associated with worse prognosis (PFS: HR=2.28, p=0.019; OS: HR=2.58, p=0.008). In PBCs as well as TILs, high levels of myeloid subsets expressing IC molecules at both timings were obviously poor prognostic factors: In the PDL1+ subset, HR=1.98 for PFS (p=0.009) and HR=2.21 for OS (p=0.015) at pretreatment, and HR=2.19 for PFS (p=0.016) and HR=2.45 for OS (p=0.010) at post-treatment; in the CTLA4+ subset, HR=1.76 for PFS (p=0.029) and HR=1.63 for OS (p=0.101) at pretreatment, and HR=3.32 for PFS (p=0.001) and HR=1.99 for OS (p=0.080) at post-treatment; and in the LAG3+ subset, HR=1.70 for PFS (p=0.017) and HR=1.19 for OS (p=0.483) at pretreatment, and HR=3.18 for PFS (p=0.004) and HR=2.49 for OS (p=0.028) at post-treatment (figure 4). After treatment, PDL1+ subset increased in 55% of patients, CTLA4+ subset in 58% of patients, and LAG3+ subset in 61% of patients (all subsets increased in 37% of patients). Patients with high increases in the CTLA4+ subset (PFS: HR=2.336, p=0.016; OS: HR=2.524, p=0.015) or LAG3+ subset (PFS: HR=1.874, p=0.038; OS: HR=2.368, p=0.013) had a significantly worse prognosis, whereas patients with high increases in the PDL1+ subset showed a significantly longer OS (online supplemental figure S11). These suggest that the differential increase in the CTLA4+ and LAG3+ subsets is also a significant risk factor for nivolumab therapy for AGC, even more than the long-attracted PDL1+ myeloid subset.

Figure 4

High levels of myeloid cell subsets in peripheral blood are a significant risk factor in nivolumab treatment of AGC. Forest plots show the cut-off and median values (number of cells (×106 /mL)) for each CD11b+ myeloid subset in peripheral blood cells, HRs with 95% CI, and p values. (A) Myeloid subsets at pretreatment for PFS (n=91). (B) Myeloid subsets at pretreatment for OS (n=91). (C) Myeloid subsets at post-treatment for PFS (n=80). (D) Myeloid subsets at post-treatment for OS (n=80). OS, overall survival; PFS, progression-free survival.

As compared with patients with low levels, patients with high PDL1+ subset levels at both before (mPFS: 1.6 months vs 3.3 months, and mOS: 5.6 months vs 12.4 months) and after treatment (mPFS: 1.5 months vs 2.3 months, and mOS: 5.7 months vs 10.0 months) had significant shorter PFS/OS (figure 5A). The CTLA4+ subset both before (mPFS: 1.6 months vs 3.3 months) and after treatment (mPFS: 1.4 months vs 2.5 months) had a large impact only on PFS, suggesting a critical risk factor for local tumor responsiveness to nivolumab treatment rather than patient survival (figure 5B). Patients with high LAG3+ subset levels at post-treatment had significantly shorter PFS (mPFS, 1.4 months vs 2.3 months) and OS (mOS, 4.7 months vs 9.0 months), while its impact was slightly smaller than that of the PDL1+ subset (figure 5C). These suggest that high levels of these myeloid subsets, not only locally in the tumor but also systemically in the peripheral blood, are significant risk factors for nivolumab therapy for AGC, although no correlation was seen in the values of the three myeloid subsets between TILs and PBCs before and after treatment (online supplemental figure S12). One possible reason may be that different data are used for PBCs (cell count) and TILs (percentage), although no correlation was also seen even when we analyzed using the TIL cell count.

Figure 5

PDL1+, CTLA4+, LAG3+ myeloid subsets in peripheral blood are significant poor prognostic factors, particularly for PFS, in nivolumab treatment of AGC. Kaplan-Meier graphs show two groups, high (red lines) and low (blue lines), divided by the cut-off value of each myeloid subset in PBCs at pretreatment (dotted lines) or post-treatment (solid lines) shown in figure 4, and mPFS (months) or mOS (months) in the group. (A) CD11b+PDL1+ subset in PBCs. (B) CD11b+CTLA4+ subset in PBCs. (C) CD11b+LAG3+ subset in PBCs. OS, overall survival; PBCs, peripheral blood cells; PFS, progression-free survival.

When the impact of the molecular expression level (mean fluorescence intensity, MFI) in these myeloid subsets on patient prognosis was statistically analyzed, only MFI-PDL1 in the CD11b+PDL1+ subset showed the same tendency in the results as the cell count and percentage results, although the MFI-CTLA4/LAG3 levels were extremely higher than MFI-PDL1 in both TILs and PBCs: Only high MFI-PDL1 levels were significantly associated with poor prognosis, while patients with high MFI-CTLA4/LAG3 levels had a rather favorable prognosis (online supplemental table S6). These suggest that PD1 blockade is ineffective even if the expression level of PDL1 is highly elevated. At least in immunotherapy for AGC, targeting PDL1 rather than PD1 may be preferable if patients are selected by the PDL1+ myeloid subset biomarker.

The myeloid subset coexpressing PDL1, CTLA4, and LAG3 was extremely rare in PBC as in TILs. However, high levels of the PDL1+CTLA4+LAG3+ subset both before (mPFS: 1.6 months vs 2.1 months in low levels, HR=1.971, p=0.010) and after treatment (mPFS: 2.9 months vs 4.5 months, HR=2.246, p=0.008) were also significantly associated with shorter PFS and showed a trend towards association with shorter OS (mOS: 9.0 months vs 15.6 months, HR=2.164, p=0.052; online supplemental figure S5B). These suggest that the PDL1+CTLA4+LAG3+ subset in PBCs as well as TILs is a poor prognostic factor, but that this subset in PBCs than in TILs is more important as a risk factor for nivolumab therapy for AGC.

Interestingly, patients with high levels of soluble PDL1 (sPDL1) in plasma at baseline also had significantly poorer prognosis (PFS: HR=2.09, p=0.015; OS: HR=2.68, p=0.002), although no significance at post-treatment (online supplemental figure S3B). These suggest that, at least in AGC, peripheral PDL1 levels, rather than local tumor levels, may be a better biomarker for predicting the nivolumab efficacy. The sPDL1 is known to be generated by proteolytic cleavage of the membrane-bound extracellular domain,15 and its high serum levels have been reported as a significant poor prognostic factor for AGC.16

Other myeloid subsets expressing molecules targeted for cancer therapy in peripheral blood

As a larger number of PBCs were harvested than TILs, PBCs were further analyzed by flow cytometry for myeloid subsets expressing molecules that have been targeted in drug discovery but not yet approved for clinical use, including one of the TIGIT ligands CD155,17 a member of the B7 family B7H3 (CD276),18 and CD115 (CSF1R).19 As compared with patients with low CD155+ subset levels, especially at post-treatment, had a large impact on both PFS (mPFS: 1.6 months vs 3.1 months) and OS (mOS: 6.5 months vs 12.9 months; figure 6A). Patients with high B7H3+ subset levels at baseline had shorter PFS/OS (mPFS: 1.6 months vs 3.3 months, and mOS: 5.6 months vs 12.4 months; figure 6B). Patients with high CD115+ subset levels both before (mPFS: 1.6 months vs 3.3 months, and mOS: 5.6 months vs 10.0 months) and after treatment (mPFS: 1.8 months vs 2.3 months, and mOS: 5.5 months vs 10.0 months) had shorter PFS/OS (figure 6C). Collectively, these results suggest that the myeloid subsets shown here are significant risk factors in nivolumab therapy for AGC. Targeting them may be useful as diagnostic biomarkers to predict potential anti-PD1 therapeutic efficacy and also as therapeutic targets for accelerating development of new drugs to improve clinical outcomes in immunotherapy for GC.

Figure 6

Other myeloid subsets expressing molecules targeted for cancer therapy in peripheral blood. Kaplan-Meier graphs show two groups, high (red lines) and low (blue lines), divided by the cut-off value of each myeloid subset in PBCs at pretreatment (dotted lines) or post-treatment (solid lines) shown in figure 4, and mPFS (months) or mOS (months) in the group. (A) CD11b+CD155+ subset in PBCs. (B) CD11b+B7H3+ subset in PBCs. (C) CD11b+CD115+ subset in PBCs. OS, overall survival; PBCs, peripheral blood cells; PFS, progression-free survival.

Discussion

In this study, the clinical outcomes, such as a response rate, PFS, OS and incidences of adverse events, were consistent with those reported in the ATTTRACTION-2 trial.3 Here, we identified myeloid subsets expressing PDL1, CTLA4, LAG3, CD155, B7H3, or CD115 as significant risk factors in nivolumab therapy for AGC. Targeting these subsets may be useful as diagnostic biomarkers to predict potential anti-PD1 therapeutic efficacy and also as therapeutic targets to improve clinical outcomes in the treatment of GC. These molecules have been actually targeted in drug discovery but not yet approved for clinical use for cancer, including GC. Therefore, at least in GC, it is expected that antibody drugs targeting these molecules expressed in the myeloid subsets may be useful and effective as an alternative or follow-up treatment to nivolumab therapy for AGC, if the right patients are selected by the myeloid subset levels as biomarkers.

In clinical settings, CPS-PDL17 has been widely recognized as biomarkers associated with anti-PD1/PDL1 therapeutic efficacy, whereas this is not necessarily correlated with clinical outcomes. In this study, most of the patients with CPS ≥1 had disease progression within 1 month after treatment, suggesting no relevant impact on patient prognosis. A non-correlation similar to our result and an association with a favorable prognosis has been reported,20 21 and the predictive value of the CPS for ICI therapy in GC is still controversial. In contrast, this study revealed that high levels of sPDL1 and PDL1+ myeloid subsets at baseline were significantly associated with poor prognosis. Several studies have already shown the correlation between sPDL1 levels and poor prognosis in the ICI therapy for GC.22 23 These suggest that peripheral PDL1, rather than local tumor PDL1, is the key negative factor impeding the therapeutic efficacy of nivolumab. This implies that immunotherapy targeting PDL1 rather than PD1 may be preferable, at least for AGC. The JAVELIN Gastric 100 study for AGC, however, demonstrated that OS of the group receiving anti-PDL1 mAb avelumab was not significantly longer than that of the control group continuing chemotherapy.24 Selecting patients using high levels of sPDL1 and/or a PDL1+ myeloid subset as biomarkers may lead to narrowing down the subpopulation of AGC patients who can obtain benefits from anti-PDL1 therapy.

It has long been reported that immunosuppressive and proinflammatory myeloid cells promote cancer progression and metastasis directly and indirectly via inducing immunosuppression, exhaustion, and dysfunction.25 Clinical studies also demonstrated that local and systemic increase of a variety of myeloid cell populations, such as neutrophils,26 M2-type macrophages,27 MDSCs,28 mast cells,29 and plasmacytoid dendritic cells,30 is a significant poor prognostic factor in GC. However, neither biomarkers nor cancer therapeutic drugs targeting myeloid cells have yet been established in clinical practice. One of the main reasons for this may be inappropriate sampling methods, such as freezing and Ficoll isolation, which result in the loss of a part of the myeloid population. In this study, which isolated and analyzed nearly all leucocytes, we were able to reveal that several myeloid subsets were significantly associated with poor prognosis of AGC patients receiving nivolumab.

High CTLA4+ subset levels in TILs at baseline and in PBCs at both timings were significantly associated with poorer prognosis. There are many studies on CTLA4+ myeloid subsets, including our previous study showing that CTLA4+ macrophages is a determinant of anti-PD1 resistance. The CTLA4+ macrophages are systemically expanded in the metastasis setting, and facilitate tumor progression and metastasis directly by generating lipid droplets in tumor cells, and indirectly by inducing immune exhaustion, leading to ineffectiveness of anti-PD1 therapy.10 Lipid droplets are known as an organelle that not only store fat but also control abnormal lipid metabolism in tumor cells, and an increase in lipid droplets is attracting attention as a feature of cancer stemness.31 We also showed that anti-CTLA4 therapy is effective in anti-PD1-resistant mouse metastasis models with increased CTLA4+ macrophages. Other studies have also showed the functional roles of CTLA4+ myeloid cells in cancer. For example, human monocytes and monocyte-derived DCs become to express CTLA4 on the maturation with inflammatory stimuli, and treatment with anti-CTLA4 mAb significantly enhances T-cell stimulatory activity as antigen-presenting cells.32–34 Such accumulating evidence suggests that CTLA4 blockade therapy may be effective in GC patients with increased CTLA4+ myeloid cells. Anti-CTLA4 ipilimumab monotherapy, however, showed no significant benefits in the phase II study for AGC as compared with the best support care,35 and even combination therapy with ipilimumab and nivolumab showed no significant improvement of OS as compared with chemotherapy alone in gastro-esophageal cancer.36 Probably, double blockade of the immune brakes might trigger intense immune responses, and increase Tregs for self-tolerance, Texs by overactivation, immature T cells that are rapidly generated without proper education and activation, and other immunosuppressive cells, leading to relapse and metastasis. Using CTLA4+ myeloid subset levels as a biomarker for patient selection may narrow down the patients who are likely to obtain benefits from anti-CTLA4 therapy in combination with/without anti-PD1 therapy, at least in GC.

The LAG3+ subset in both TILs and PBCs at both timings was specifically associated with PFS, suggesting a critical risk factor for local tumor responsiveness to treatment, although its impact on OS was slightly smaller compared with other subsets. We previously identified LAG3+ myeloid cells coexpressing the FSTL1 receptor DIP2A as another determinant of anti-PD1 resistance. These cells are abundantly expanded in the host with FSTL1+ tumor cells having EMT-like properties, and induces apoptosis in T cells partly via the LAG3 signaling, leading to immune dysfunction and the consequent ineffectiveness of anti-PD1 therapy.11 As a next-generation ICIs, anti-LAG3 therapy has been recently expected to overcome immune exhaustion and dysfunction by boosting antitumor effector functions. Actually, combination regimen with nivolumab and anti-LAG3 relatlimab was approved for treating melanoma, although only in the USA.37 If the right patients are appropriately selected using the LAG3+ myeloid subset levels as a biomarker, anti-LAG3 therapy may also be effective in the treatment of GC.

However, it should be noted that the results of IHC analysis, in which CD11b+CTLA4+ or CD11b+LAG3+ cell infiltration in tumors was significantly associated with favorable PFS in patients, were different from the results of these flow cytometric analyses. The disparate results may have arisen from differences in the nature of the methods, for example, the use of fresh samples for flow cytometric analysis, but formalin-fixed samples for IHC. It seems that caution is needed when comparing two different data sets at the same level. Further studies are needed to draw a definitive conclusion.

High CD155+ subset levels, especially at post-treatment, were significantly associated with PFS/OS. CD155 is an adhesion molecule belonging to the Nectin family that is expressed in myeloid cells and tumor cells, and the ligand for T-cell stimulatory receptor CD226, T-cell inhibitory receptor TIGIT, and CD96 on T cells and NK cells.17 It has been reported that CD155 overexpressed in tumor cells plays key roles in cell migration, invasion, and proliferation and are significantly associated with tumor progression and poor prognosis in clinical settings.17 Therefore, blocking CD155 may be useful not only for restoring antitumor immunity but also for weakening and eliminating cancer cells directly, especially in GC patients with elevated CD155+ myeloid levels after nivolumab therapy.

High B7H3+ subset levels at baseline were significantly associated with PFS/OS. B7H3 is a member of the B7 family expressed in myeloid cells and tumor cells and plays an inhibitory role in T-cell proliferation and activation.18 B7H3 overexpressed in tumor cells is known to contribute to tumorigenesis, metastasis and other malignant properties through different mechanisms, and its positivity in tumor tissues is associated with tumor progression and poor prognosis in clinical settings.38 It has been reported that blocking B7H3 not only restores antitumor immune responses but also attenuates malignant property of tumor cells in the in vitro and in vivo settings.18 B7H3-targeted therapies have been widely evaluated in many clinical trials in various drug forms, such as inhibitory mAbs, mAbs with antibody-dependent cellular cytotoxicity, antibody-drug conjugates, radioimmunotherapeutics, engineered chimeric antigen receptor T cells (CAR-T), and B7H3-CD3 bispecific mAbs.39 By using the B7H3+ myeloid subset levels at baseline as a biomarker to narrow down patients, these treatments may be effective for treating GC.

High levels of the CD115+ subset at both time points had a wide range of implications on patient prognosis. CD115/CSF1R, a receptor for M-CSF and IL-34, is highly expressed in myeloid cells and tumor cells, and its signaling is necessary for the survival, proliferation, differentiation, and activation of myeloid cells, especially macrophages that release tumor-promotive and proinflammatory cytokines.19 Targeting CD115 has attracted attention as a promising strategy for cancer therapy, and various specific drugs have been clinically developed. Pexidartinib is a kinase inhibitor small molecule that blocks the activity and has been already approved by the FDA for the treatment of tenosynovial giant cell tumor.40 LY3022855, an anti-CD115 mAb, has been now evaluated in patients with advanced solid tumors refractory to standard therapy in combination with anti-CTLA4 tremelimumab or anti-PDL1 durvalumab in early phases.41 In GC, CD115 positivity in tumor tissues is a significant poor prognostic factor associated with lymph node and peritoneal metastasis, although the evaluation was performed within tumor cells only, excluding immune cells.42 Considering our data together, these anti-CD115 inhibitors may be expectedly effective for treating GC, especially with proper selection of patients with elevated CD115+ myeloid subset levels.

There are some limitations in this study. The MSI testing was not performed in this study because the positivity rate for MSI-high is generally low, at only 6.3% in Japanese,43 and it was assumed that its coexistence with other factors was not so large. The cut-off value used here might overestimate the impact of each factor. We did not investigate the tumor expression of the molecules, such as CD155, B7H3, CD115, which might have affected the patient prognosis. The utility of each factor for selecting the optimal therapy could not investigated because of a single-arm treatment.

In conclusion, our study revealed that special myeloid subsets are significant risk factors in nivolumab treatment. Myeloid cells are the majority of cellular components in the human immune system, and its abnormality may widely and negatively affect on the entire host, including tumor cells, stroma, immunity, and therapeutic efficacy. Therefore, targeting these myeloid subsets and the associated molecules, including PDL1, CTLA4, LAG3, CD155, B7H3, and CD115, may be useful as biomarkers to predict therapeutic responses more accurately, and also as therapeutic targets for more effectively treating GC, possibly via reprogramming, inhibiting, and eliminating these myeloid subsets.

Data availability statement

All data relevant to the study are included in the article or uploaded as online supplemental information.

Ethics statements

Patient consent for publication

Ethics approval

The prospective observational WJOG10417GTR study (UMIN000032686) was conducted in collaboration with 10 hospitals according to the protocol (August 2018–November 2020) approved by the IRB of each participating hospital, including the National Cancer Center (No. 2017-473). All activities were conducted in accordance with the ethical principles of the Declaration of Helsinki and the Japanese Clinical Research Ethics Guidelines.

Acknowledgments

We would like to thank Dr. Takahiro Miyamoto, Ms. Ayako Murooka, Ms. Kana Uegaki, and Ms. Mika Takeda for their cooperation in this study. Also, we would like to express our sincere gratitude to all the patients and medical staff who provided great support to this study.

References

Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

Footnotes

  • Contributors HS directed and supervised this study, collected samples and clinicopathological data of patients and wrote the draft manuscript. CK-S conceptualized, designed, directed, and supervised this study, conducted assays, formally analyzed all data, and wrote the manuscript. KN conducted statistical analysis of the data. HI conducted assays and analyzed the data. KT, NT, TK, YA, TW, NO, YN, YY, RK, and KM collected samples and clinicopathological data of patients. KA supported with paperwork. NB conceptualized, designed, directed and supervised this study, collected samples and clinicopathological data of patients and is the guarantor of the study.

  • Funding This study was supported by Ono Pharmaceutical and Bristol Myers Squibb (grant number A2018-032).

  • Disclaimer The companies provided nivolumab and funded the study but were not involved in study design, data collection, or data analysis.

  • Competing interests HS received research grants from Ono Pharmaceutical and Takeda Pharmaceutical, and honoraria from Ono Pharmaceutical and Bristol-Myers Squibb. CK-S received a grant from Chiome Bioscience, and honoraria from Ono Pharmaceutical and Bristol-Myers Squibb. KN received honoraria from Pfizer, Fujimoto Pharmaceutical, Senju Pharmaceutical, and Toray. HI received honoraria from Ono Pharmaceutical. NT received honoraria from Ono Pharmaceutical, Bristol-Myers Squibb, and Taiho Pharmaceutical. TK received honoraria from Ono Pharmaceutical and Bristol-Myers Squibb. NO received honoraria from Taiho Pharmaceutical, Eli Lilly Japan, Eisai, Bayer, Chugai Pharmaceutical, Ono Pharmaceutical, Takeda Pharmaceutical, and GlaxoSmithKline. YN received honoraria from Eli Lilly, Daiichi Sankyo, Taiho, Ono Pharmaceutical, and Bristol-Myers Squibb. YY received honoraria from Chugai Pharmaceutical, Ono Pharmaceutical, Takeda Pharmaceutical, Taiho, Sanofi, Yakult, Nihon Servier, Lilly, Asahi Kasei Parma. KA received grants from Ono Pharmaceutical, Chugai Pharmaceutical, Eisai, Daiichi Sankyo, and Chiome Bioscience. KM received honoraria from Amgen, AstraZeneca, Chugai Pharmaceutical, Ono Pharmaceutical, Eli Lilly, Daiichi Sankyo, Taiho, and Bristol-Myers Squibb. NB received honoraria from Ono Pharmaceutical, Bristol-Myers Squibb, Taiho Pharmaceutical, Eli Lilly, and Daiichi Sankyo. Other authors have no competing financial interests.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.