Background

Pancreatic cancer is a leading cause of cancer-related death and is predicted to be the second largest cause of death from cancer worldwide by the end of the current decade.1 Inoperable, non-metastatic, locally advanced pancreatic cancer (LAPC) has an intermediate prognosis (median overall survival (OS) 12–18 months) between metastatic and operable disease, and is treated with combination chemotherapy or chemoradiotherapy (CRT). The use of potent combination chemotherapy has delivered modest improvements in clinical outcomes, though its application in clinical practice is limited by toxicity.2,3 Recently, a greater understanding of tumour biology has driven the development of novel therapies.4,5,6 One example is the recently reported POLO trial of maintenance olaparib in patients with germline BRCA mutations, which is the first successful trial of biomarker-guided therapy in pancreatic cancer.7

Tissue and blood samples collected during clinical trials provide an unparalleled resource from which tumour biology can be further interrogated and correlated with clinical outcomes, thereby facilitating the generation of evidence-based hypotheses for biomarker-driven treatments. SCALOP was a multi-centre randomised Phase 2 trial (n = 114) which compared gemcitabine versus capecitabine as the radiosensitiser of choice for LAPC.8 Patients received three cycles of gemcitabine and capecitabine (GEMCAP) induction chemotherapy and eligible patients with responding or stable disease (n = 74) were randomised to a further cycle of GEMCAP followed by gemcitabine or capecitabine-based CRT to a total radiation dose of 50.4 Gy in 28 fractions. Survival analyses have previously been reported and demonstrate that capecitabine-based CRT was superior in terms of survival and toxicity profile.8,9 To facilitate later biomarker analysis, blood samples were collected from individual patients at baseline (prior to GEMCAP) and, for a subset, at week 17 (prior to consolidation CRT), as summarised in Supplementary Fig. 1.

In a prior systematic review, we identified a number of cytokines reported to be of diagnostic, prognostic or predictive significance in PDAC.10 Here, we report the results of a detailed prognostic biomarker analysis, including correlation with clinical outcomes, assessing a panel of 35 of these systematically identified cytokines in samples from the SCALOP trial cohort.

Methods

This study report aligns with the Reporting Recommendations for Tumour Marker prognostic studies (REMARK) criteria.11

Patients

The design and results of the SCALOP trial have been reported elsewhere.8,9 Briefly, patients with histologically/cytologically confirmed inoperable locally advanced pancreatic cancer with maximum diameter 7 cm or less, performance status (PS) 0–2, were eligible. Response was assessed following three cycles of GEMCAP chemotherapy, and those with responding or stable disease (according to RECIST criteria), PS 0–1 and tumour diameter 6 cm or less were eligible for randomisation to a further cycle of GEMCAP followed by gemcitabine (300 mg/m2 weekly) or capecitabine (830 mg/m2 twice daily on days of radiotherapy) concurrent with radiotherapy (50.4 Gy in 28 fractions), as summarised in Supplementary Fig. 1.

Sample collection and analysis

Peripheral venous blood was collected in ethylenediaminetetraacetic acid (EDTA) anti-coagulant vacutainers. Each was centrifuged at 3000×g for 10 minutes and plasma aliquoted into cryovials then snap frozen. Samples were stored at −80 °C at investigating centres until the end of the trial and subsequently shipped on dry ice to Wales Cancer Bank (WCB) for centralised storage. Centres that were unable to store frozen blood samples or which did not have a Human Tissue Authority (HTA) license were required to send samples to WCB for processing using pre-paid safe boxes. All samples were processed on the same day of collection and, where possible, within an hour of collection.

Biomarker analysis was undertaken at the molecular laboratory at Oxford Institute for Radiation Oncology, University of Oxford. Assays were performed while blinded to clinical outcome. Assayed cytokines are listed in Table 1. Insulin-like growth factor 1 (IGF-1), transforming growth factor-β (TGF-β) and β-nerve growth factor precursor (β-NGF) were quantified using DuoSet® enzyme-linked immunosorbent assay (ELISA) (R&D Systems, USA) and the signal detected using a POLARstar® Omega microplate reader. The remaining 32 cytokines were assessed by Human Multiplexed Magnetic Luminex assay (R&D Systems, USA) and measured using the Luminex MAGPIX® fluorescent detection system.

Table 1 Association of each biomarker with overall survival.
Full size table

Prior to assay, all plasma samples were thawed on ice and centrifuged at 10,000×g for 5 min at 4 °C to remove precipitate. Samples were limited to two freeze-thaw cycles. Supernatants were loaded onto ELISA or Luminex plates at the recommended dilutions (i.e. 50-fold for CCL-5 and PDGF, each at 1 µl per well, and two-fold for all other cytokines, which were loaded at 25 µl per well) with standard protein controls. The manufacturer’s protocol was followed in each case with no modifications. For ELISA assays, standard curves were constructed and used to quantify each marker. For Luminex assays, biomarker quantification was derived from analysis of raw data using xPONENT 4.2® software.

All biomarkers were measured in triplicate across all patients at each timepoint. If a run contained ≥50% out of range values for a given biomarker across all patients (>1 SD from the mean) then the entire run was removed for that biomarker. Any remaining values above the upper limit of detection or below the lower limit of detection were substituted with the highest or half the lowest value of that biomarker respectively for a given run. The mean value across the remaining runs was then used for analysis. This method for handling measurement errors has been used elsewhere.12

Statistical analysis

Statistical analyses were conducted using R v3.5.2 according to a pre-specified analysis plan. Univariate Cox proportional hazard models were used for each of the 35 biomarkers as continuous variables to determine the association with OS. OS was measured from randomisation until death from any cause. Multiple comparisons were accounted for by using the False Discovery Rate (FDR). Those found to be significant at the q value <0.2 were then further investigated for independence from existing prognostic clinical characteristics (i.e. cancer antigen 19–9 (CA19–9), treatment (capecitabine CRT vs gemcitabine CRT), PS (0 vs 1) and age (<65 vs ≥65 years)) using Spearman correlations (r ≤ 0.7 shows independence) and Wilcoxon rank sum tests (p ≥ 0.05 shows independence). PET-CT data were unfortunately unavailable for a majority of patients and could not therefore be included in these analyses. Those clinical characteristics found to be independent were then split into tertiles and associated with OS using univariate Cox regression to ensure that any associations with the biomarker as a continuous variable were linear. Optimal thresholds delineating low to high values were identified using the R “survivalROC” package based on time-dependent receiver operating characteristic (ROC) curves from censored survival data and their corresponding area under the curve (AUC). Both continuous and dichotomised biomarkers were then associated with OS using multivariable Cox proportional hazard models, along with existing prognostic clinical characteristics, to determine whether or not novel biomarkers maintained prognostic value. The proportional hazards assumption was tested by calculating Schoenfeld residuals, and the linearity assumption was assessed by plotting deviance residuals.

Results

Patient characteristics

Cytokine data were available from 63 patients in total. No significant differences in clinicopathological characteristics were identified comparing patients who did and did not have cytokine data available (Supplementary Table 1). Measurements of the full panel of cytokines and corresponding clinical outcome information were available for 60 patients. These data were assessed further in order to identify correlations between circulating cytokines and clinical outcomes.

Prognostic factor identification

Raw biomarker levels at both baseline and, where measured, at week 17 are shown in Table 1, as is an analysis of the association with OS of each of the 35 assayed cytokines. Of the biomarkers tested, two (C-C Motif Chemokine Ligand 5, CCL5 and interferon-ƴ, IFNy) measured at baseline (prior to commencement of GEMCAP chemotherapy) had significant associations with OS at the q < 0.2 FDR level. IFNy levels significantly correlated with age (p = 0.019). However, CCL5 was independent of existing clinical characteristics, including age (p = 0.859), PS (p = 0.660) and CA19–9 (r = 0.339) (Table 2). No biomarkers were associated with progression at the univariate level at the q < 0.2 FDR level, as illustrated in Supplementary Table 2. Consequently, no further analyses on progression were undertaken.

Table 2 Correlation between biomarkers significantly associated with overall survival by Cox univariable regression and clinical characteristics known to be associated with overall survival in pancreatic cancer.
Full size table

CCL5 is an independent prognostic biomarker in LAPC

CCL5 data was available at baseline for 60 patients, with no differences identified in patient characteristics between those with and without CCL5 data (Table 3).

Table 3 Patient characteristics for all patients randomised within the SCALOP trial subdivided by the availability of data relating to serum C-C chemokine ligand 5 (CCL5) quantification.
Full size table

We identified a linear association of CCL5 with OS. When associated with OS in a multivariable Cox proportional hazards model as a continuous variable, patients with high circulating CCL5 were found to have a HR of 1.01 for each ng/ml unit increase (95% CI 1.00–1.03; p = 0.013, n = 54) (Table 4). A time-dependent ROC curve was constructed to identify the optimal threshold for CCL5 (Supplementary Fig. 2). Dichotomisation of CCL5 at its optimal threshold of 25.4 ng/ml was significantly associated with OS, with a HR of 1.95 (95% CI: 1.04–3.65; p = 0.037) in the Cox multivariable model. Median OS was 18.5 months in patients with CCL5low (21/60) (95% CI: 11.76–21.32) and 11.3 months (95% CI: 9.86–15.51) in patients with CCL5high (39/60); as demonstrated in Fig. 1. Diagnostic tests showed that there was no evidence of departure from proportionality or violations of the linearity assumption, and that there were no extreme outliers or influential points.

Table 4 Univariable and multivariable Cox regression analysis by characteristic.
Full size table
Fig. 1
figure 1

Kaplan–Meier estimates of overall survival for dichotomised serum C-C chemokine ligand 5 (CCL5) concentration in patients randomised within the SCALOP trial.

Full size image

CCL5 circulating levels were additionally assessed in the context of other clinicopathological characteristics, including age, PS and CA19–9 (n = 54 for CA19–9 data). A signature utilising dichotomised CCL5 was determined (Supplementary Fig. 3) including age, PS and CA19–9. The CCL5 signature was created for each patient using the beta coefficients from the multivariable survival model and multiplying them by their corresponding covariate (CCL5 × 0.6678 − Age × 0.4072 + WHO PS × 0.7807 + CA199 × 0.0001663). With the optimal threshold determined at 0.398, 17 patients were classified as CCL5low and 37 as CCL5high. Using this approach, patients classified as CCL5low had a median OS of 19.68 months (95% CI 16.3–27.79) and CCL5high of 11.2 months (95% CI 8.25–13.24); hazard ratio 2.69 (95% CI 1.40–5.17; p = 0.003) (Supplementary Table 3).

Relative change in CCL5 levels between baseline (prior to start of GEMCAP) and week 17 (prior to start of CRT, data available for n = 47) did not demonstrate any association with OS (Supplementary Table 4). Independent measurement of CCL5 at week 17 did not associate with OS. High levels of circulating CCL5 were identified to be a poor prognostic factor independent of concurrent chemotherapy received during radiotherapy (capecitabine or gemcitabine). Taken together, these suggest that circulating CCL5 levels at baseline relate to intrinsic tumour properties. No benefit of CCL5 as a potential pharmacodynamic biomarker was identified within the context of the CRT treatment delivered in this randomised trial.

Discussion

Interpretation of results in the context of pre-specified hypotheses and other relevant studies

Pancreatic cancer responds poorly to chemotherapy and radiotherapy and the aim of this research was to determine the prognostic value of previously reported circulating biomarkers in patients with LAPC treated with chemotherapy and CRT. To the best of our knowledge, this is the first biomarker study from a prospective randomised clinical trial in LAPC. The cytokine panel tested in this study was based on a systematic review that identified cytokines of diagnostic, prognostic or predictive significance in PDAC.10 This included six cytokines (IL-1β; IL-6, IL-8, VEGF, TGFβ, IL-10) previously reported as consistently elevated in patients with PDAC compared to healthy controls, all of which have previously been found to have potential prognostic value (carrying higher risk of metastasis and lower OS).10,13,14,15,16,17,18,19 However, none of these six cytokines correlated with survival in the SCALOP trial.

We identified baseline levels of circulating CCL5 and IFNγ as being significantly associated with OS, of which CCL5 remained significant in multivariable analysis. CCL5 can be secreted by a variety of tumour cells (including pancreatic cancer) as well as non-malignant stromal cells including T regulatory cells (Tregs) and macrophages. Engagement of CCL5 with its receptor, C-C chemokine receptor type 5 (CCR5), can favour tumour growth by several mechanisms including induction of the mammalian target of rapamycin (mTOR) pathway,20,21 and by recruiting tumour associated macrophages (TAMs), leading to immunosuppression and release of pro-angiogenic cytokines.22,23 The CCL5-CCR5 axis has also been implicated in tumour migration and metastasis through modulating the activity of the PI3K/Akt, MAPK/ERK and NF-ĸB pathways, and via induction of matrix metalloproteinases.24

The interaction between pancreatic tumour cells, the immune system and the role of the CCL5-CCR5 axis has been investigated in preclinical studies. Tan et al. demonstrated that both human pancreatic cancer and murine pancreatic tumour (Pan02) secreted CCL5 and correspondingly, within the stroma, CCR5 was preferentially expressed by CD4+ Foxp3+ Tregs.23 The investigators also showed that Treg infiltration within the tumour could be reduced by systemic administration of CCR5 antagonists, resulting in restricted tumour growth. CCL5 secretion in pancreatic cancer was further investigated by Wang et al in a cohort of 120 resected PDAC tissues. While Foxp3 is typically utilised to identify Tregs, here the investigators demonstrated the presence of Foxp3 positive cancer cells. CCL5 was directly trans-activated by cancer-Foxp3, which in turn promoted Treg infiltration.24 Intriguingly, Jang et al. reported accumulation of Tregs around murine pancreatic tumours within one week of implantation, suggesting tumour cell-intrinsic secretion of CCL5 plays an important role in this rapid infiltration.25 This proposes a model of cell intrinsic CCL5 secretion resulting in autocrine and paracrine signalling promoting a pro-tumourigenic immunosuppressive TME. However, the prognostic implication of circulating CCL5 has not been previously reported in literature.

Implications for future research and clinical value of the work

PDAC has been described as an “immune desert” and conventional immunotherapy has failed to impact outcomes in this disease. Low mutational burden, induction of TAMs and myeloid-derived suppressor cells, and exclusion of CD8 + and other pro-inflammatory cells are some of the proposed mechanisms for failure of immunotherapy. Mitigating the factors that contribute to immune suppression may enhance response to immunotherapy in PDAC. Several strategies, including combinations of immunotherapy with gemcitabine, radiation, pancreatic cancer vaccine (GVAX) and CSF1R antibody, are being evaluated in clinical trials.26,27 The biomarker analysis of this clinical cohort is consistent with the pre-clinical hypothesis that the CCL5-CCR5 axis plays an adverse role in pancreatic cancer pathogenesis. Given its immunomodulatory effects, inhibition of the CCL5-CCR5 axis in combination with immunotherapy should be tested in pre-clinical models.

Discussion of limitations

We have been unable to validate a number of previously reported prognostic biomarkers in the prospective trial cohort reported here. It is unclear whether this relates to the trial cohort under study or to limitations in previous studies that identified these candidate biomarkers. In this study, we analysed biomarkers in patients randomised within SCALOP but not in 40/114 patients ineligible for randomisation due to disease progression, poor PS or because of patient or clinician choice. This may have biased the analysed cohort and validation of our findings in a larger cohort is required.

Although clear protocols for sample collection and processing were pre-specified, variations between centres is possible. For example, 18/27 centres did not have facilities or regulatory approvals for sample storage, thereby requiring unprocessed samples to be shipped centrally to WCB for processing. Given that this study predates the widespread adoption of preservative tubes that allow a 14-day window within which samples can be transported and processed, the requirement for many samples to be sent to WCB may have impacted on their quality. Nevertheless, all assays were performed under identical experimental conditions at a single centre using pre-defined protocols.

A final point of note is that the median biomarker levels we have reported here differ from their measurement in previous studies of pancreatic cancer.17,18,19 For example, the measured values for IL-6 and IL-8 are a factor of between 3 and 5 higher than in some previous reports. Levels of IL-1β were also higher here compared with previous reports (a median of 30 pg/ml compared with 0 pg/ml in the existing literature). The median CCL5 reading is also around 200-fold higher than for example reported previously in pancreatic cancer by Farren and colleagues,19 but at 30.1 ng/ml is similar in magnitude to values that have been reported for ovarian, breast and cervical cancer.28,29 These variations may reflect differences in the studied patient populations, in the storage and processing of samples, and in the assays used to process these samples. Nevertheless, they do not detract from the standardised conditions used to compare serum biomarker levels for patients within this study, from which we identified clinical support for pre-clinical studies that have previously postulated a role for the CCL5-CCR5 axis in the pathogenesis of pancreatic cancer.

Conclusion

Circulating CCL5 is an independent marker for poor prognosis for patients with LAPC treated with combination chemotherapy and consolidation CRT within the SCALOP trial. Further studies are required to validate CCL5 as a tumour marker in LAPC. Blockade of the CCL5-CCR5 axis may provide opportunities to modulate the efficacy of immunotherapy in pancreatic cancer.