Profile of soluble factors in pleural effusions predict prognosis in mesothelioma

2.1Experimental subjects

Consecutive patients attending a specialist referral centre for mesothelioma were invited to participate in the NCARD Biobanking program. Patients with a confirmed pathological diagnosis of mesothelioma and a pleural effusion sample larger than 100 mL and with more than 5 × 106 total cells, that was available prior to treatment or pleurodesis were included in the discovery set. The validation cohort was chosen, independently, at a later time-point from the Biobank using the simplified selection criteria of confirmed pathological diagnosis of mesothelioma and a pleural effusion sample available prior to treatment or pleurodesis. Validation samples were selected randomly using the RAND function in excel. Demographic and clinical data were collected including age, gender, weight loss, performance status, blood biochemistry, tumour histology, smoking history, treatment and survival. This study was approved by Sir Charles Gairdner and Osborne Park Hospitals Human Research Ethics Committee and all participants provided written informed consent and conformed with the Code of Ethics of the World Medical Association (Declaration of Helsinki) [17].

2.2Pleural effusion samples

Pleural effusions were collected in sterile containers following pleural aspiration or drainage. Samples were transferred immediately at ambient temperature to the laboratory and centrifuged at 1000 g for 10 min. The supernatant was transferred to -80∘C for storage until later analysis and counts were performed on effusion cell pellets by haemocytometer. Cell composition was determined from a visual review of cytospin slides stained with Rapid Stain (Amber Scientific). Cells were classified as mesothelial, macrophage, lymphocyte, neutrophil, eosinophil or other by trained observers, with a total of 400 cells per sample counted.

2.3Cytokine, chemokine and cellular factors analysis

Concentrations of soluble factors were determined using a Quantibody® Human Immune Response Assay (RayBiotech, Norcross GA), which measured a panel of 40 cytokines, chemokines and cellular factors involved in inflammatory and immune signalling pathways. Pleural effusion supernatant was diluted 1:2 and assayed in quadruplicate, along with serial standards, buffer controls and in-house human control plasma samples. Concentrations were determined from standard curves run in parallel and values reported as pg/mL. Values for concentrations reported to be below the sensitivity limit of detection were imputed for statistical comparisons using a Missing at Random (MAR) pattern [18], implemented using a fully conditional specification Markov Chain Monte Carlo (MCMC) multiple imputation method in the SPSS statistical program (IBM Corp., Armonk, NY, ver 24).

2.4Enzyme-linked immunosorbent assays (ELISAs)

Mesothelin is the only clinically approved biomarker for mesothelioma. Mesothelin concentrations were determined using the MESOMARK kit (Fujirebio Diagnostics, Malvern PA) following the manufacturer’s instructions. A mesothelin value ⩾ 20 nM was considered positive for mesothelioma [19]. IL8 was measured using the human CXCL8/IL8 ELISA DuoSet® ELISA Development kit (R&D systems, Minneapolis MN), with a sensitivity of 31 pg/ml and PF4 was measured using the human PF-4 ELISA (Thermo Scientific, Frederick, MD), with a sensitivity of 20 pg/ml, following the manufacturers’ protocols.

2.5Statistical analysis

Unless otherwise stated data is presented as median and the interquartile (IQ) range. Patient survival was calculated from the time of sample collection. For hierarchical clustering and survival analysis soluble factor levels were log (10) transformed and normalised. Survival analysis was performed using the Kaplan Meier product limit estimator, where difference in survival of categorical variables were assessed using a log-rank test, and using the Cox proportional hazards regression methods where hazard ratios (HR) are presented with 95% confidence intervals of a 1 SD difference in the variable. The Cox regression results were also adjusted by sex, prognostic risk group [6] and age as a continuous variable. For multivariate analysis a forward stepwise multiple Cox regression analysis was used with a probability of F to enter (pIN) value of 0.05 unless otherwise stated. Due to the high dimensionality of the data set, a more stringent pIN value of 0.0003, (determined from the sample size and number of variables being tested [20]), was used in a separate multivariate analysis in order to identify a sparser set of the most strongly associated variables. Group comparisons were made using the Student’s t test, using an adjustment for non-normality if indicated by a significant Levene’s test for equality of variances. Receiver Operator Curve (ROC) analysis was performed with two survival groups determined based on the lower 95% confidence interval for median mesothelioma survival time in Australia [21] (i.e. (survival < 9 and survival ⩾ 9 months). A difference was considered to be statistically significant if the associated p-value was less than 0.05. Survival analysis, group comparisons and ROC analysis were performed using IBM SPSS statistics. Hierarchical clustering was performed using hclust and heatmaps were produced using ggplot2 [22] in the R statistical package [23].

3.1Patient characteristics

The 50 individuals with pleural mesothelioma in the discovery cohort were predominantly male, approximately 70 years old at diagnosis, with good performance status. The tumours were of predominantly epithelioid histology. In this cohort effusions were predominately managed with indwelling pleural catheters. This was an observational study and patients received a range of treatments, including novel clinical trial agents. Of the 50 individuals, 23 received chemotherapy of some form. Patients undergoing active treatment received standard pemetrexed/platinum combination in the first line, and generally vinorelbine as a second line chemotherapy (Table 1).

Figure 1.

Cellular composition of pleural effusions, determined from a cell count of approximately 400 cells.

Figure 2.

Heatmap of cytokine, chemokines and soluble factor results, clustered into groups using hierarchical clustering. Rows are standard deviation normalized cytokine values, columns are individual pleural effusion samples. Red (negative) and green (positive) row labels indicate cytokines that were identified as being independently associated with prognosis in multivariate Cox regression survival analysis.

A similar number of samples were studied from individuals in an independent validation cohort (Table 1). There was no difference between the discovery and validation groups in terms of age, gender distribution, tumour histology, performance status, effusion management, smoking status and treatment received. Despite random selection, as a group patients in the validation cohort had a significantly poorer prognosis than those in the discovery cohort (median survival 12.53 (7.37–17.70) versus 22.1 (16.98–27.22) months (p= 0.003), respectively (Table 1) highlighting the need for prognostic markers in this disease. Prognosis within the discovery cohort was significantly associated with sex, performance status and treatment in univariate analysis (Table 2). Patients who received chemotherapy, subsequent to the pleural fluid sample collection, had significantly better prognosis (HR = 0.433 (0.231–0.809); p= 0.009) compared to those who did not.

3.2Cell counts

In the discovery cohort there was considerable inter-patient variability in the cellular composition of the pleural effusion samples. As a percentage of total cells, lymphocytes on average comprised approximately a third of total cell counts (27% (34)), however there was a wide range in values from 1 to 94%. Effusions had on average 10.5 (22.5)% mesothelial and/or mesothelioma cells. Samples could be segregated by the presence of a predominant cell type with 24% of samples being predominately (i.e. > 50%) lymphocytic and 30% being comprised predominately of macrophages. There were two samples that had > 50% neutrophils; post hoc review of clinical records provided no indication of infection at this time for these patients (Fig. 1). No significant association was observed for groups dichotomised on percentage of lymphocytes, macrophages or neutrophils with survival in univariate analysis (Table 2; column 2), although the HR for the association of each with prognosis was positive for lymphocytes and macrophages and negative for neutrophils when analysed as a continuous variable.

Figure 3.

Kaplan-Meier plot of the association of cytokine cluster group with prognosis. Log-rank test χ2 = 11.41, p= 0.003. Vertical marks are censored time points.

3.3Cytokine, chemokine and soluble factor concentrations

The majority of the 40 molecules measured in the pleural effusions of the discovery cohort were measured in the linear range of the assay for all 50 samples; note, 0.25% (i.e. 5/2000) were below the limit of detection of the assay and 0.35% (7/2000) were above the upper limit of the assay (Table 3). Values below the limit of detection were imputed as described under Methods. Univariate analysis demonstrated that 14 of the 40 molecules (35%) were significantly associated with survival. For each of these 14 molecules, an increasing value was associated with a poorer prognosis (Table 3). Given the known interrelationships between cytokines and chemokines, multivariate analysis was performed. Five molecules had an independent association with survival; increasing concentrations of IL8, IL2Ra and PF4 were independently associated with a poorer prognosis and increasing concentrations of lipocalin-2 and IL4 were associated with a more favourable prognosis (Table 4). Of particular note, IL4 was significantly associated in univariate analysis in the opposite direction with a poorer prognosis, indicating that its relationship with prognosis is strongly confounded by its correlation with other factors. Using a more stringent analysis (pIN value of 0.0003), IL8 was the only pleural effusion cytokine that was significantly associated with prognosis.

3.4Cluster analysis

Hierarchical cluster analysis (of normalized log(10) transformed values for expression levels in pleural effusions) revealed three major clusters (Fig. 2), with patient cluster 1 (consisting of 11 patients) having an overall higher concentration of pleural effusion soluble factors than cluster 2 (17 patients) or cluster 3 (22 patients). In the second dimension soluble factors partitioned into two major cluster groups; soluble factor group 1 (top, Fig. 2) and soluble factor group 2 (bottom, Fig. 2). The five soluble factors independently associated with prognosis were distributed throughout the clusters (Fig. 2 – PF4, IL8 and IL2RA, indicated in red; lipocalin-2 and IL4, indicated in green).

Patients in cluster 1 had a strikingly poorer prognosis than either cluster 2 or cluster 3 patients (HR = 3.549 (1.557–8.089); p< 0.001 Fig. 3). Notably, only one of the 11 cluster 1 patients had a tumour with sarcomatoid histology, therefore sarcomatoid histology was not over represented in this cluster (1 sarcomatoid patient, χ2 = 0.654, p= 0.721). Histology (epithelioid or sarcomatoid/biphasic), which was not a significant prognostic indicator in univariate analysis, achieved statistical significance in a multivariate Cox regression analysis after adjustment for patient cluster group (Table 2, column 6), emphasising the importance of high soluble factor levels in determining survival in this cohort of patients.

Post-hoc analysis of soluble factor concentrations between individuals who survived less than 12 months (n= 18) versus those that survived greater than 24 months (n= 8) confirmed that IL8 (136 (137) vs 55 pg/ml (61), p= 0.007) and PF4 (35443 (56943) vs 15505 pg/ml (15623), p= 0.005) concentrations were higher in the short term survival group, as was TNFα (878 (305) vs 818 pg/ml (110), p= 0.01).

3.5Association between soluble factor levels and cell composition

When the association between soluble factor levels and the broad cellular composition of the effusion was examined IL4 was the strongest predictor of cellular composition, being positively associated with lymphocytes and negatively associated with neutrophils (Table 5). The neutrophil to lymphocyte ratio (NLR) in the effusion was also strongly negatively predicted by IL4.

Several other factors were associated with cellular composition. MCP1 and VCAM were positive predictors and uPAR, Fas and CD40 were significant negative predictors of pleural effusion lymphocyte frequencies. For pleural effusion neutrophils, in addition to IL4, IL6 and IL1a, were significant positive predictors and RAGE and IL10 were negative predictors. Pleural effusion macrophages were positively predicted by CRP and negatively predicted by PF4 and MIP1b. The NLR was negatively predicted by RAGE in addition to IL4 and positively predicted by thrombomodulin and procalcitonin. Notably, IL8 levels were not significantly associated with cell composition.

3.6Validation study

To determine if the observation in the discovery cohort that pleural effusion concentrations of IL8 and PF4 were independent prognostic predictors, an independent set of randomly selected pleural effusion samples was then studied and levels of these factors were measured using stand-alone ELISA kits. There were quantitative differences in concentrations of IL8 and PF4 for the samples when measured by the multiplex assay and by ELISA. Median IL8 levels were 85 (121) pg/ml in the multiplex assay and 399 (1445) pg/ml by ELISA (Table 3). However, there was a strong correlation between PF4 (r= 0.900, p= 0.037) and IL8 (r= 0.903, p< 0.001) for samples measured on both platforms.

Using the ELISA generated data, PF4 concentrations were more strongly associated with prognosis (HR (per SD increase) 1.850 (1.267–2.701), p= 0.001)) than IL8 concentrations (HR (per SD increase) 1.625 (1.066–2.477, p= 0.024)) in the discovery cohort (Table 3). The independent association of the two factors with prognosis was evident but less pronounced in the validation cohort when the two factors were adjusted for each other in the Cox regression analysis (IL8 HR = 1.501 (per SD increase) (0.988–2.279); p= 0.057: PF4 (per SD increase) HR = 1.687 (1.157–2.461); p= 0.007). Therefore, the validation study examining IL8 and PF4 concentrations determined using stand-alone ELISA assays confirmed the prognostic association of these factors that was originally observed in the discovery cohort using a Quantikine multiplex assay.

ROC analysis confirmed that IL8 and PF4 were predictive of survival with area under the curve (AUC) values of 0.712 (0.543–0.881, p= 0.019) and 0.756 (0.600–0.912, p= 0.005), respectively. Both IL8 and PF4 displayed similar specificity versus sensitivity characteristics (Fig. 4).

Figure 4.

ROC analysis for pleural effusion IL8 and PF4 concentrations in predicting prognosis in the validation cohort (survival < 9 months compared to survival ⩾ 9 months). AUC for IL8 = 0.712 (0.543–0.881, p= 0.019); AUC for PF4 = 0.756 (0.600–0.912, p= 0.005).