Blood serum amyloid A as potential biomarker of pembrolizumab efficacy for patients affected by advanced non-small cell lung cancer overexpressing PD-L1: results of the exploratory “FoRECATT” study

According to inclusion criteria, patients had undergone no previous systemic therapy for metastatic disease, and had an Eastern Cooperative Oncology Group (ECOG) performance-status (PS) score of 0 or 1 (on a 5-point scale, with 0 indicating no symptoms and higher scores indicating increasing disability). At least one measurable lesion according to Response Evaluation Criteria in Solid Tumors (RECIST) and adequate hematologic, hepatic, and renal functions were required, in addition to normal basal levels of TSH, fT3, fT4 and ACTH for the P cohort. Exclusion criteria for receiving pembrolizumab were symptomatic interstitial lung diseases, autoimmune diseases and systemic immunosuppression. Patients with brain metastases were eligible if asymptomatic or already treated with cranial radiotherapy. Progression was defined according to RECIST 1.1 criteria. CT scan evaluation was performed by a designated radiologist at baseline and every 9 weeks during the treatment. PD-L1 expression was assessed in formalin-fixed tumor samples at local laboratory by a designated pathologist with the use of the commercially available PD-L1 IHC 22C3 pharmDx assay (Dako North America). All adverse events were graded according to the National Cancer Institute Common Terminology Criteria for Adverse Events, version 4.0. Study protocol was reviewed and approved by the local institutional review board and Ethical Committee (Prot. 26,496/19, ID 2640, June 20^th, 2019) and conducted according to the principles of Declaration of Helsinki and EU General Data Protection Regulation (GDPR, 25.05.18). Patients enrolled in the study signed a written informed consent for biomarkers analysis and clinical data collection at the beginning of therapy. Patients’ data were anonymized before analysis.

Blood samples were collected at baseline (day of the first administration of pembrolizumab), before the start of the infusion for both cohorts; the same procedure was performed every 9 weeks in the only P cohort until withdrawal from treatment for either toxicity or disease progression. SAA plasma concentration was measured by a commercial enzyme-linked immunosorbent assay (ELISA) (Siemens Healthineers, Milano, Italy).

No power analysis was done to calculate the sample size given the explorative and original aim of the study. The primary end-point response rate (RR) was calculated as the proportion of partial and complete responses assessed by radiological imaging. The secondary end-points were progression-free survival (PFS), defined as the time from randomization until first evidence of objective tumor progression or death from any cause and overall survival (OS), as the time from starting treatment to death for any cause. The endpoints were evaluated according to the SAA value at baseline in both cohorts and during treatment only in the P cohort. SSA was analyzed both as continuous and discontinuous variable (low versus high), using a prefixed threshold value. ROC curve analysis was used to determine the value of SAA levels to adopt as cut-off, which more accurately predict response to pembrolizumab. The association of SAA level with clinico-pathological characteristics and RR was evaluated by Fisher’s exact test, χ2 test or Mann–Whitney test, as appropriate, or using Pearson correlation analysis. Estimates of survival times (PFS and OS) were calculated according to the Kaplan–Meier method and compared with log-rank test for the survival univariate analysis. All variables found to be associated (p < 0.20) with PFS, OS and RR in the univariate model were included in the multivariate analysis, using the Cox proportional-hazards regression model. Each analysis was performed with the use of a two-sided 5% significance level and a 95% CI. Data were analyzed using MedCalc (v.18.5).

In the present prospective study, we found that NSCLC patients with “low” SAA at baseline had a greater likelihood to respond to first-line pembrolizumab than those with higher pre-treatment SAA, while responses to chemotherapy was not affect by SAA level. Moreover, patients having “low” SAA reached longer PFS and OS irrespective of treatment received (pembrolizumab or chemotherapy) than those harboring “high” SAA.

We hypothesized that blood level of tumor-derived SAA could be useful to predict response to anti-PD-1 mABs in NSCLC patients, basing on the immune-modulating properties of SAA and its possible involvement in the process of tumor escape from immune-surveillance.

The acute-phase SAA proteins exhibit significant pro-inflammatory activity, by inducing the synthesis of several cytokines and promoting chemotaxis for monocytes and neutrophils, and act as apolipoprotein of high-density lipoprotein (HDL), involved in the reverse cholesterol transportation during inflammatory states [25]. SAA was also involved in the pathogenesis of the “secondary” amyloidosis and others chronic inflammatory diseases, such as rheumatoid arthritis, atherosclerosis, Alzheimer’s disease and cancer.

Elevated levels of SAA in blood were detected relatively early in cancer patients and, thus, several studies have evaluated SAA as possible serum biomarker for various subtype of cancer, including ovarian, lung, renal, endometrial, uterine and melanoma [26], without drawing any definitive conclusion. Afterwards, the correlation between SAA and cancer was reinvigorated by the finding of local SAA production within tumor tissues, which was significantly higher in lung cancer patients than in other pulmonary disease or other cancer types [27]. In melanoma patients, tumor-derived SAA regulated the plasticity of polymorphonuclear myeloid-derived suppressor cells (PMN-MDSCs) through the binding to the Formyl peptide receptor 2 (FPR-2) on the surface of these cells, resulting in the expansion of IL-10-secreting PMN-MDSCs, which downregulated T-cell anti-tumor activity [28]. At the same time, as negative feedback mechanism, SAA stimulated iNKT (invariant Natural Killer T) cells to interact with IL-10-secreting PMN-MDSCs, a process that inhibits their immune-suppressive activity by reducing IL-10 and enhancing IL-12 production. It is conceivable that tumors produce SAA as immune-escape mechanism, such as the PD-L1 expression, exploiting the capability of SAA to impair the function of T-cells, which represent the main players in anti-cancer immune-response elicited by PD-(L)-1 inhibitors.

The survival findings of our study suggest that SAA may have a prognostic rather than predictive significance when evaluated as biomarker for immunotherapy. However, given the immune-suppressive property on antitumor response described above and the observed strong correlation with response to pembrolizumab and not to chemotherapy, a potential effect of SAA on the activity of immunotherapy cannot be excluded. Remarkably, response rate was adopted as primary endpoint for defining the SAA cut-off to exploit its direct correlation with drug activity with less risk of confounding factors than survival endpoints.

However, this strong prognostic value of SAA could be clinically relevant because a simple blood sample makes possible to individuate a subset of patients (about two third of PD-L1 overexpressing NSCLC population) who may benefit the most from anti-PD-1 therapy (ORR 53%, mPFS 17.2 mo) compared to the remaining one-third of patients having “high” pre-treatment SAA, who contrarily reached a very poor OS (7 months) and whose ORR (7%) and mPFS (2.1 mo) potentially underlie a primary resistance to ICIs, probably including the limited cases of hyper-progression.

The discrimination power of SAA in the selection of patients was not influenced by PD-L1 expression, since the TPS score was not associated with neither SAA level nor response to treatment, and appeared independent from other known prognostic factors, excluding the absence of pleural effusion at baseline which also resulted an independent predictor for better PFS with pembrolizumab, supporting the well-known poor activity of ICIs on pleural metastases. The capability of SAA to predict outcomes of anti-PD-1 treatment was confirmed by the dynamic monitoring of blood protein level, which showed that patients who maintained “low” SAA during therapy reached better PFS and OS than those who maintained “high” SAA or changed the protein level. Furthermore, a subset of patients belonging to “intermediate profile” group change the SAA level from “high” to “low” during the treatment, raising the question whether is the tumor microenvironment favorably modified by immunotherapy increasing likelihood of response during treatment or the variation of the circulating immune-profile primarily influences the activity of ICIs. Early evidence with anti-PD-1 in NSCLC seems support the first hypothesis [29]. Interestingly, a small number of metastatic sites (≤ 2) and thus a low tumor-burden, was the unique clinical characteristic associated with “low” pre-treatment SAA, supporting the fact that blood SAA in cancer patients reflects the direct production by tumor rather than the liver synthesis in response to inflammatory condition. In fact, no relationship was found between SAA high concentration and increased BMI or presence of comorbidities, which were usually characterized by a chronic inflammatory status.

Other systemic inflammation-based prognostic scores have been proposed to improve the selection of patients to candidate to an appropriate therapy or to direct the treatment allocation and the design of randomized clinical trial. In particular, SAA is also included in the acute phase proteins analyzed by the VeriStrat test (Biodesix, USA), a blood-based proteomic assay which assigns patients to either “good” or “poor” status, and was tested as strong independent indicator of prognosis only for tyrosine kinases inhibitors treatment and chemotherapy, while its value for treatment with ICIs remains unknown [30].

Regarding immunotherapy-treated patients, neutrophil lymphocyte ratio (NLR) have been reported repeatedly to have prognostic value, regardless of treatment considered [31]; however, a combined score (the LIPI score) including LDH and derived-NLR, had showed potential predictive value in large retrospective real-life cohort of patients treated with IO [22]. Unfortunately, when applied to randomized patients in clinical trials, the LIPI score revealed as survival predictive biomarker irrespective of treatment administered [23, 32].

These unsuccessful but compelling data underline the great challenge to identify a reliable biomarker with a clear predictive value among the systemic inflammation components, since inflammatory response have a well-recognized prognostic significance in cancer patients across the histotype and treatments employed [33].

Besides the role of SAA, our exploratory study also gave a “real-life” description of efficacy of pembrolizumab prospectively assessed in PD-L1 over-expressing NSCLC patients. The RR (38%) and mPFS (8 mo) observed in our entire cohort were slightly lower compared with those reported in KEYNOTE-024 (ORR 44.8%, mPFS 8 mo) after a similar median time of follow-up [6]. These differences could be explained by the more accurately selected population enrolled in the trial, in which patients with ECOG PS 2, untreated brain metastases and receiving systemic corticosteroids were excluded. Furthermore, the median age of our patients was higher than the trial population one (70.5 versus 64.5 years).

Despite the relevant results and the prospective design which reduces the risk of bias, our study has some flaws. First, the small sample of patients included do not allow drawing any definitive conclusion about the role of this blood biomarker. Second, the value of SAA adopted as cut-off was derived from the analysis of a single-institution study population and thus need to be validated in larger and external series of patients. Third, radiological assessment was performed locally, lowering the strength of the RR and PFS results. Fourth, although our clinical experience suggests that the production of SAA by the tumor would represent a potential novel mechanism of immune-escape through the induction of an immunosuppressive microenvironment, pre-clinical data supporting evidences are not provided in the present study. To avoid the potential bias discussed, a further study (FoRECATT-2) is currently ongoing including an external multi-institutional validation set of patients and in vitro experiments to assess the effect of SAA on immune response. According to the novel approved indication of pembrolizumab, a cohort of patients treated with the combination of chemotherapy and immunotherapy will be also included.

In conclusion, our study provides to clinicians a strong and simple prognostic blood biomarker which may improve the selection of patients who are more likely to benefit from immunotherapy, avoiding the risk of immune-related toxicities for potentially not-responsive patients and also reducing the financial burden for Healthcare System through a more tailored therapeutic approach. Unlike the other available tissue biomarkers, such as PD-L1 and TMB, whose evaluation requires adequate sample achievable only with invasive procedures and is hampered by relevant spatial and temporal heterogeneity, the SAA dosing could represent a “non-invasive” dynamic biomarker which is assessable in every patient during the treatment, reflecting both disease and immune-environmental changes and allowing for an early detection of those cases with acquired resistance to immunotherapy. Finally, our findings if further confirmed, may support the development of novel therapeutic strategies employing anti-SAA e anti-FPR-2 monoclonal antibody which could improve the activity of ICIs by blocking the potential immunosuppressive action of SAA.