The FSTL1-DIP2A axis is a significant biomarker for predicting anti-PD1 therapeutic efficacy in advanced gastric cancer.

Introduction
Gastric cancer (GC) is the fifth most common cancer and the fifth leading cause of death from malignant diseases worldwide [1]. Surgical resection of the tumor is the definitive treatment for GC, and the addition of chemotherapy before (neoadjuvant), after (adjuvant), or during surgery improves patient survival [2]. However, in the advanced and metastatic setting, such conventional treatments are insufficient to ensure long-term patient survival.
Targeting the immune brakes mediated by immune checkpoint (IC) molecules, such as CTLA4, PD1, PDL1, LAG3 and TIGIT, has recently attracted great attention as a promising strategy for treating various types of cancers including GC, and clinical development of monoclonal antibodies (mAbs) targeting the IC pathway (called “immune checkpoint inhibitors” [ICIs]) has provided great benefit to patients even in the advanced and metastatic setting [3, 4]. However, the treatment is effective only in a limited number of patients. Several biomarkers predictive of treatment responses have been identified, including CPS-PDL1 [5], microsatellite instability [6], and mutation burden [7] in tumors. However, these are not necessarily correlated with clinical outcomes. Many other attempts have also been undertaken around the world, including development of combination therapies with various other drugs, such as mAbs targeting different IC pathways, small molecule inhibitors targeting IC pathways, genetically modified CAR-T/NK cells, and neoantigen peptide vaccines [8]. However, most have failed or are still being evaluated. Thus, more breakthrough strategies are needed for the treatment of advanced and metastatic GC.
A member of the SPARC family follistatin-like 1 (FSTL1) has been demonstrated as a key driver of epithelial-to-mesenchymal transition and cancer stemness of various types of cancers, including gastrointestinal cancer such as GC [9, 10], colorectal cancer [11], and hepatocellular carcinoma [12]. For example, FSTL1 knockdown with the specific siRNAs can induce apoptosis in human FSTL1+ GC cells that highly express FSTL1 via the STAT6 signaling pathway, suggesting a key role for FSTL1 in GC cell survival [9]. FSTL1 knockdown with the specific shRNAs also suppress cell proliferation, sphere colony formation, migration, invasion, and metastasis in the in vitro and in vivo settings using human FSTL1+ GC cells, suggesting a crucial role of FSTL1 in the overall malignant properties of GC [10]. Furthermore, clinical studies have demonstrated that FSTL1 positivity in tumor tissues is a poor prognostic factor in various types of cancers, including gastrointestinal cancer such as GC [10, 13], colorectal cancer [11, 14], esophageal cancer [15], and hepatocellular carcinoma [16]. For example, FSTL1 expression in GC tissues from patients is aberrantly increased, especially in advanced stages rather than early stages, and the high FSTL1 levels are significantly correlated with infiltrating depth, lymph node metastasis, tumor stage, tumor size, tumor-node-metastasis (TNM) stage, and/or shorter survival in patients [10, 13]. We additionally revealed the immunological role of FSTL1, including suppression of anti-tumor immunity via expanding immunosuppressive DIP2A+ cells, such as mesenchymal stem/stromal cells [17], myeloid-derived suppressor cells [11], and regulatory T cells [18], and also via inducing apoptosis in CTLs [11] and NK cells [19]. Its receptor DIP2A is essential for the FSTL1 actions in both cancer cells and immune cells [11, 19]. Of note, blocking FSTL1 with the specific mAb successfully induces anti-tumor immunity even in the anti-PD1-resistant mouse tumor models [17]. Therefore, it is expected that targeting the FSTL1/DIP2A axis may be a promising strategy to overcome anti-PD1/PDL1 resistance in clinical settings.
However, the relationship between the FSTL1/DIP2A levels and anti-PD1 therapeutic efficacy remains to be elucidated in clinical practice. Furthermore, despite their ease of access, there is little clinical evidence regarding FSTL1 and DIP2A+ immune cells in the peripheral circulation. Therefore, we analyzed peripheral blood obtained from patients with unresectable advanced or recurrent GC including esophagogastric junction cancer (abbreviated as AGC) before and after nivolumab monotherapy for FSTL1 by ELISA, and for DIP2A+ immune cells by flow cytometry, and statistically analyzed the relationship with patient prognosis.

Materials and methods

Patients and sample preparation
The WJOG10417GTR study (UMIN000032686) was collaboratively conducted in 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). Written informed consent was obtained from all patients before study enrollment. Ninety-six AGC patients received intravenous infusions of nivolumab, an anti-PD1 mAb, at 3 mg/kg or 240 mg/body every 2 weeks until disease progression or unacceptable toxicity. The details of the WJOG10417GTR study, including eligibility, tumor responses, and toxicities, were previously reported [20]. All activities were conducted in accordance with the ethical principles of the Declaration of Helsinki and the Japanese Clinical Research Ethics Guidelines. EDTA-added peripheral blood was collected from the patients before and one month after treatment, but it was not possible to collect peripheral blood from 12 patients due to poor condition after nivolumab therapy. The timing of post-treatment was set to within approximately one month (22–43 days) after the first administration, when the immune status was estimated to have stabilized. After centrifugation of peripheral blood, whole cells in the pellet were used as peripheral blood cells (PBCs) for flow cytometric analysis following treatment with ACK Lysing Buffer (ThermoFisher), and plasma was used for ELISA.

ELISA
Plasma was assessed for human FSTL1 using an ELISA kit (Abcam # ab213782) according to the manufacturer’s instructions.

Flow cytometric analysis
After Fc blocking, PBCs were stained with the following immunofluorescence-conjugated antibodies: anti-CD45-APC-Cy7 (BioLegend #304,014), anti-CD3-BUV496 (BD Biosciences #564,809), anti-CD11b-BV510 (BioLegend #301,334), anti-CD56-BUV650 (BioLegend #362,532), anti-DIP2A-FITC (Bioss #bs-9312R-FITC), anti-LAG3-BV650 (BioLegend #369,316), 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 firstly excluded by FSC/SSC followed by gating CD45+ leukocytes, and immunofluorescence intensity was compared to the isotype control (Supplementary Figure S1). The data were presented as the number of specific cells per ml of peripheral blood. Patients with CD45+ leukocytes < 10% who were considered ineligible for integrated analysis were excluded as described before [20].

Statistical analysis
Significant differences (P value < 0.05) were statistically evaluated using GraphPad Prism 7 software (MDF), SAS software version 9.4, and R version 4.2.3. Data were analyzed by one-way ANOVA or two-way ANOVA with Bonferroni post-hoc test for pairwise comparison of multiple groups on the basis of the normal distributions. The normality of the dataset was analyzed using the Shapiro–Wilk test, and the dataset showing a non-normal distribution was analyzed using the Mann–Whitney U test. Progression-free survival (PFS) and overall survival (OS) were estimated using the Kaplan–Meier method. Hazard ratios (HRs) were obtained by univariable and multivariable models adjusting potential confounding factors (PS, prior gastrectomy, number of metastatic organs, and ALP), and 95% confidence intervals were calculated. Cutoff values were determined by change point of log HRs using the Cox regression models with penalized splines, and we used the cutoff values with the larger HRs calculated through careful analysis of each marker based on the method for modeling continuous-scale covariates according to a standard survival analysis textbook [21]. The detailed and specific methodology has been described in the Supplementary Methods.

Results

FSTL1 and DIP2A+ cells increase in peripheral blood of AGC patients with progressive disease after nivolumab treatment
To examine the relationship with treatment responses, we divided AGC patients into three groups, although there was a large imbalance in sample sizes due to the very small number of treatment responders: one showed progressive diseases (PD) within one month after treatment (designated “early PD” group), one showed PD more than 1 month after treatment (designated “late PD” group), and one did not show PD at least throughout the study period (designated “nonPD” group). At baseline, FSTL1 levels were higher in the early PD group (P = 0.006) and the late PD group (P < 0.001) as compared to those in the nonPD group (median 0 μg/ml), and these were further increased after treatment in 25% of early PD patients (P = 0.046 versus nonPD) and 45% of late PD patients (P = 0.002), although no significant differences were observed before and after treatment (Fig. 1A). In the early PD group, two patients with extremely high FSTL1 levels at both time points had an extremely poor prognosis, with progression-free survival (PFS) within 20 days and overall survival (OS) within 2 months.
At baseline, DIP2A+ subset levels in all cell populations tested (CD11b+ myeloid cells, CD3+ T cells, and CD56+ NK cells) were also higher in early PD group (T-cell and NK subsets, P < 0.001) and late PD group (T-cell subset, P < 0.001; and NK subset, P < 0.001) as compared to those in the nonPD group, albeit no significant difference in the myeloid subset (Fig. 1B–D). Although no significant differences were observed before and after treatment, the DIP2A+ myeloid subset further increased after treatment in 49% of early PD patients and 45% of late PD patients (Fig. 1B). Also, the DIP2A+ T-cell subset further increased after treatment in 57% of early PD patients (P < 0.001 versus nonPD) and 64% of late PD patients (P < 0.001; Fig. 1C), and the DIP2A+ NK subset further increased after treatment in 54% of early PD patients (P = 0.004) and 60% of late PD patients (P = 0.002; Fig. 1D). In the nonPD group, the number of DIP2A+ cells was small, especially in the T-cell and NK subsets, and changed only slightly after treatment. In the early PD group, two patients with extremely high DIP2A+ myeloid levels after treatment had a poor prognosis, with PFS within 2 weeks and OS within approximately 3 weeks, and two patients with extremely high DIP2A+ NK levels after treatment also had a poor prognosis, with PFS within 2 weeks and OS within approximately 2 months. These results suggest that increase in peripheral DIP2A+ cells as well as FSTL1 is markedly associated with worse responses to nivolumab therapy.

High baseline FSTL1 levels are significantly associated with poor prognosis of AGC patients after nivolumab treatment
We next divided patients into two groups based on the cutoff values, and differences in PFS and OS were statistically compared between them. Patients with high baseline FSTL1 levels showed significantly shorter PFS (median PFS [mPFS] = 1.71 months versus 2.00 months, hazard ratio [HR] = 1.775, P = 0.020) and OS (median OS [mOS] = 5.03 months versus 9.00 months, HR = 1.872, P = 0.015) as compared to the low group (Fig. 2), whereas no significant differences were observed in the comparison of post-treatment levels (Supplementary Figure S2A, Table 1). No significant association was seen between baseline FSTL1 levels and patient characteristics. These results suggest that high baseline FSTL1 levels are a significant poor prognostic factor for nivolumab therapy for AGC.

High baseline levels of DIP2A+ subsets are significantly associated with poor prognosis in AGC patients after nivolumab treatment
DIP2A+ cell levels both before and after treatment were strongly associated with patient prognosis. Patients with high DIP2A+ myeloid levels showed significantly shorter PFS (pre: mPFS = 0.84 months versus 1.87 months, HR = 2.088, P = 0.020; and post: mPFS = 1.41 months versus 2.25 months, HR = 2.245, P = 0.035) and OS (pre: mOS = 2.12 months versus 6.77 months, HR = 2.237, P = 0.020, and post: mOS = 4.47 months versus 9.00 months, HR = 1.956, P = 0.098) as compared to the low group (Fig. 3A, Supplementary Figure S2B, Table 1). Even when divided into two groups based on median values, significant differences in PFS were seen between high and low DIP2A+ myeloid levels at post-treatment (Supplementary Table S1).
Patients with high DIP2A+ T-cell levels both before (mPFS = 1.68 months versus 2.27 months, HR = 1.940, P = 0.007; Fig. 3B, Table 1) and after treatment (mPFS = 1.87 months versus 4.27 months, HR = 2.717, P = 0.001; Supplementary Figure S2C, Table 1) showed significantly shorter PFS as compared to patients with low levels. Also, patients with high DIP2A+ NK levels both before (mPFS = 1.58 months versus 2.37 months, HR = 1.997, P = 0.009; Fig. 3C, Table 1) and after treatment (mPFS = 1.97 months versus 3.01 months, HR = 1.978, P = 0.012; Supplementary Figure S2D, Table 1) showed significantly shorter PFS as compared to patients with low levels. Even when divided into two groups based on median values, significant differences in PFS were seen between high and low DIP2A+ myeloid levels at post-treatment (Supplementary Table S1). Patients with high post-treatment DIP2A+ T-cell levels (mOS = 7.49 months versus 23.52 months, HR = 2.106, P = 0.033; Supplementary Figure S2C, Table 1) and patients with high post-treatment DIP2A+ NK levels (mOS = 6.95 months versus 23.52 months, HR = 2.121, P = 0.025; Supplementary Figure S2D, Table 1) showed significantly shorter OS as compared to patients with low levels. Among patients with low post-treatment DIP2A+ T/NK levels, approximately 50% of patients survived for > 6 months. These suggest that high DIP2A+ cell levels are significant poor prognostic factors for nivolumab therapy for AGC, and patients with low DIP2A+ subset levels contain a subpopulation of anti-PD1 responders with long-term survival without tumor progression, known as “durable responders”. High levels of DIP2A+ T/NK subsets, but not the DIP2A+ myeloid subset, tended to be associated with several patient characteristics: high baseline DIP2A+ T-cell levels with low PDL1-CPS (CPS < 1, P = 0.058), high post-treatment DIP2A+ T-cell levels with HER2 positivity (P = 0.062).

AGC patients with low baseline FSTL1/DIP2A levels are treatment responders with durable responses
When baseline FSTL1 levels, which were significantly associated with poor prognosis, were combined with baseline levels of either DIP2A+ subset, patients with low levels of FSTL1 and the DIP2A+ myeloid subset showed significantly longer PFS (P < 0.004) and OS (P < 0.006) as compared to patients with high levels of either factor (Fig. 4A). Approximately 25% of the FSTL1-low/DIP2A-low patients survived for a long time during the study period (mOS = 9.26 months; Table 2). However, combination of FSTL1 with DIP2A+ T/NK subsets produced much greater synergy. Patients with low levels of both FSTL1 and the DIP2A+ T-cell subset showed significantly longer PFS (mPFS = 4.27 months versus 1.72 months, HR = 3.62, P = 0.001) and OS (mOS = 13.44 months versus 5.16 months, HR = 2.69, P = 0.012) as compared to patients with high levels of both (Fig. 4B, Table 2). Patients with low levels of both FSTL1 and the DIP2A+ NK subset showed significantly longer PFS (mPFS = 5.32 months versus 1.61 months, HR = 3.28, P = 0.003) and OS (mOS = 12.35 months versus 5.03 months, HR = 4.17, P = 0.001) as compared to patients with high levels of both (Fig. 4C, Table 2). These suggest that high baseline levels of both FSTL1 and DIP2A+ cells are significant risk factors that negatively affect anti-PD1 efficacy, and that patients with low levels of both, particularly FSTL1 and DIP2A+ T/NK subsets, are likely to be durable responders.

Discussion
In this study, we demonstrated that levels of FSTL1 and DIP2A+ cells were elevated in the peripheral circulation of AGC patients with PD, and that these levels were further increased in many of the patients after nivolumab treatment. When immune suppressive signals mediated by IC molecules such as PD1 are released, anti-tumor immune responses are induced and activated in the host. However, during this process, adverse effects such as "cytokine release syndrome", in which various pro-inflammatory cytokines and chemokines are released in large quantities from various cells, including tumor cells, mesenchymal stem cells, and immune cells, within the patient's body are often caused in clinical settings [22]. FSTL1 may be a representative molecule that is significantly increased among these numerous factors, having a significant impact on prognosis of GC patients, probably because FSTL1 is significantly associated with GC pathogenesis [10, 13]. The increased FSTL1 could subsequently induce and activate DIP2A+ cells upon receiving the stimulation. This may possibly be the reason why FSTL1 and DIP2A+ cells were further increased after nivolumab therapy. We also demonstrated that combination of high baseline levels of both FSTL1 and DIP2A+ cells, particularly DIP2A+ T/NK subsets, in peripheral blood are significant poor prognostic factors for nivolumab therapy for AGC. Patients with low levels of both are likely to be durable responders. These suggest that the FSTL1/DIP2A axis is a distinctive biomarker that can select the AGC patients who may obtain benefit of anti-PD1 therapy. As peripheral blood is relatively easier to collect than tumor tissues, it is expected that diagnostic testing for peripheral FSTL1/DIP2A may be easily applied into clinical practice.
Several studies have demonstrated that in GC, FSTL1 plays a key role in cancer metastasis and stemness [9, 10], and that FSTL1 expression in tumor tissues from patients is a poor prognostic factor in clinical settings [13]. However, the relationship between FSTL1/DIP2A levels, especially those in the peripheral circulation, and the clinical outcomes in the anti-PD1/PDL1 therapy remains to be determined in clinical practice. Our study revealed that high baseline levels of both FSTL1 and DIP2A+ cell subsets, particularly DIP2A+ T/NK subsets, are significant poor prognostic factors for nivolumab therapy for AGC. Patients with low levels of both are likely to be responders with long-term survival. These suggest that the FSTL1/DIP2A axis is a distinctive biomarker that can predict and select anti-PD1 responders. We previously demonstrated that FSTL1 is one of the determinants of resistance to anti-PD1 therapy using various mouse tumor models [11, 17, 19]. This suggests that strategies targeting the FSTL1/DIP2A axis may clearly hold promise in selecting potential responders and non-responders to ICI therapy for GC.
Several biomarkers have been identified to predict the efficacy of anti-PD1/PDL1 therapy, and have been actually used in clinical practice. However, their effectiveness and accuracy are questionable because they do not necessarily correlate with clinical efficacy. Indeed, the CPS-PDL1 analysis in the WJOG10417GTR study showed no significant association with patient prognosis as previously reported [20], although we did not conduct MSI testing, since the positivity rate of MSI-high in Japanese was generally low at 6.3% [23], and the coexistence with other factors was not considered to be significant. In the present study, however, the baseline levels of FSTL1 and DIP2A+ cells in peripheral blood were significantly associated with poor prognosis of AGC patients after nivolumab therapy. Therefore, at least in AGC, the FSTL1/DIP2A levels are expected to be able to select patients suitable for anti-PD1/PDL1 therapy more accurately than the CPS-PDL1.
We previously showed that anti-FSTL1 blocking therapy is significantly effective in inducing anti-tumor immunity even in mouse anti-PD1-resistant tumor models [17]. Therefore, development of anti-FSTL1 mAb may contribute to enhancing anti-PD1/PDL1 efficacy, at least in the treatment of AGC. Our previous study also showed that mouse and human DIP2A+ myeloid subsets becomes to express LAG3 in response to FSTL1 stimulation, and that both anti-FSTL1 and anti-LAG3 treatments synergistically induce more potent anti-tumor immunity in anti-PD1-resistant mouse tumor metastasis models [11]. Indeed, in the present study, the majority of DIP2A+ cells co-expressed LAG3 in PD patients (especially early PD patients), but at a much lower rate in nonPD patients (Supplementary Figure S1). Therefore, although this is only a hopeful expectation, combination of anti-FSTL1 with anti-LAG3 mAbs, rather than with anti-PD1/PDL1 mAbs, may be better for treating AGC, and may provide greater benefit as a follow-up treatment for anti-PD1/PDL1 failure. New drugs targeting other IC molecules, such as LAG3 and TIGIT, have been being developed as cancer immunotherapeutics [24]. However, there are concerns that releasing multiple immune brakes mediated by IC pathways could cause sudden immune activation, which could lead to adverse events [25]. As FSTL1 suppresses anti-tumor immunity through different molecular and cellular mechanisms from the IC pathways, it is unlikely that FSTL1 blocking therapy will cause similar risks, and it is expected that the main benefit may be enhancement of anti-tumor effect. Although this is just a hope for the future, development of anti-FSTL1 mAb with a different mechanism of action from ICI’s may be better for synergistically improving ICI efficacy, particularly in AGC patients with high levels of both FSTL1 and the DIP2A+ subsets at baseline.
However, some limitations of this study should be kept in mind. The usefulness of each factor for selecting the optimal treatment could not be evaluated due to the single-arm treatment setting. This study is an exploratory study, with no set sample size for hypothesis testing, and the relatively small sample size may affect the universality of the results, although many other exploratory studies have been widely conducted on small numbers of patients, sometimes using even fewer patients than those in our study [26, 27]. These studies differ in that they were designed as adjuncts to clinical trials with different statistical hypotheses, but are similar to our study in that no specific patient numbers were set as exploratory studies. In our study, we conducted a post-hoc power analysis to evaluate the power of the HR. Assuming a sample size of n = 96, a median progression-free survival of 1.8 months, a registration period of 2.7 years, and a follow-up period of 0.5 years, it was shown that HR = 1.81 could be detected with 80% power. This suggests that HR values above 1.81 could be considered reliable. It should be noted, however, that it is difficult to draw definitive conclusions based on these results alone, and it is necessary to carry out prospective validation studies to confirm the replicability of the results before universalizing them to other populations. The cutoff values determined by visual assessment of the continuous trends and change points of the log HR may have overestimated the effect of each factor. Also, in the comparative analysis of the three groups, it should be noted that the small sample size of the nonPD group limits the statistical reliability and generalizability of the results to clinical settings even if statistical significance was obtained. Interpreting the results requires extreme caution, taking into account potential biases and uncertainties. Lastly, we could not conduct immunohistochemical analysis for FSTL1/DIP2A expression using tumor tissues, which may have had some effects on patient prognosis as those in peripheral blood, because tumors were not available. Clarifying the relationship between tumor FSTL1/DIP2A expression levels and the anti-PD1/PDL1 therapeutic efficacy is another future challenge to strengthen the conclusions presented in this study.
Nevertheless, this study provided significant clinical evidence that the FSTL1/DIP2A axis may be useful as biomarkers to predict anti-PD1/PDL1 therapeutic efficacy in AGC. The FSTL1/DIP2A-targeting strategy may contribute to improving clinical outcomes in the treatment of GC.

Supplementary Information
Below is the link to the electronic supplementary material.