The overall survival of patients with pancreatic ductal adenocarcinoma is extremely low. Although gemcitabine is the standard used chemotherapy for this disease, clinical outcomes do not reflect significant improvements, not even when combined with adjuvant treatments. There is an urgent need for prognosis markers to be found. The aim of this study was to analyze the potential value of serum cytokines to find a profile that can predict the clinical outcome in patients with pancreatic cancer and to establish a practical prognosis index that significantly predicts patients’ outcomes. We have conducted an extensive analysis of serum prognosis biomarkers using an antibody array comprising 507 human cytokines. Overall survival was estimated using the Kaplan-Meier method. Univariate and multivariate Cox’s proportional hazard models were used to analyze prognosis factors. To determine the extent that survival could be predicted based on this index, we used the leave-one-out cross-validation model. The multivariate model showed a better performance and it could represent a novel panel of serum cytokines that correlates to poor prognosis in pancreatic cancer. B7-1/CD80, EG-VEGF/PK1, IL-29, NRG1-beta1/HRG1-beta1, and PD-ECGF expressions portend a poor prognosis for patients with pancreatic cancer and these cytokines could represent novel therapeutic targets for this disease.
Pancreatic ductal adenocarcinoma (PDAC) accounts for only 2.68% of all cancers, but it represents the fourth leading cancer-related death worldwide just remaining after lung and bronchus, prostate, and colorectum cancers in men and after lung and bronchus, breast, and colorectum cancers in women [
As defined by the NIH Biomarker Working Group, a biological marker (biomarker) is a characteristic that is objectively measured and evaluated as an indicator of normal biological processes, pathogenic processes, or pharmacologic responses to a therapeutic intervention [
Here we have focused on those inflammatory mediators that may constitute useful prognosis biomarkers for PDAC detection. Altered levels of circulating inflammatory cytokines have been found in cancer patients for nearly every cancer examined, even at early stages of the development, indicating that immune response has an important role during carcinogenesis and that circulating inflammatory markers may be useful cancer biomarkers [
The aim of this study was to investigate the prognosis significance of serum cytokines as a reflection of the host response to tumor in PDAC patients. A conditional stepwise algorithm based on likelihood rate analysis according to the Cox’s proportional hazard model was used to identify the best combination of significant prognosis factors. An equation was then derived for modeling the survival in our specific cohort. A leave-one-out cross validation was developed to assess the model.
All patients in the study were diagnosed with PDAC at Hospital Virgen de las Nieves (Granada, Spain) from 2008 to 2011 (
Clinicopathologic characteristics of the study population (
Age at diagnosis, years (mean ± StD) | 66 ± 10.5 |
|
|
Gender | Male: 50% |
|
|
Disease stage | III (28%) |
|
|
Type of chemotherapy | Gemcitabine + Erlotinib |
|
|
Clinical response | PR (14.29%) |
|
|
Survival time, months (mean ± StD) | 12.6 ± 12.6 |
|
|
Outcome: | |
Follow-up months (mean ± StD) | 12.6 ± 12.6 |
Death from pancreatic cancer | 100% |
Alive | 0% |
Lost to follow-up (censored cases) | 0% |
|
|
CEA level [ |
2219 ± 5017 |
Healthy: 0–37 | |
|
|
CA 19-9 level [U/L] (mean ± StD) | 899 ± 3185 |
Healthy: 0–5 |
PR: partial response; SD: stable disease; PD: progressive disease; StD: standard deviation.
Soluble proteins in the sera of PDAC patients were measured using a biotin label-based human antibody array (Human Antibody L-series 507 Array (RayBiotech, Norcross, GA, USA)), according to the recommended protocols. Briefly, all samples were biotinylated. Antibodies were immobilized in specific spot locations on glass slides. Incubation of array membranes with biological samples resulted in the binding of cytokines to corresponding antibodies. Signals were visualized using streptavidin-HRP conjugates and colorimetric. Final spot intensities were measured as the original intensities subtracting the background. Data were normalized to the positive controls in the individual slide. The antibody array data is provided in Supplementary Table 3 (see Supplementary Material available online at
All statistics and data analysis were performed using the
The overall model fit was considered significant based on chi-squared statistic test (
Leave-one-out cross validation (LOOCV) was applied to assess the performance of the prognosis model as the simplest and most widely used method for estimating prediction accuracy [
Clinical characteristics of the PDAC patients are summarized in Table
(a) shows Kaplan-Meier disease-specific survival curve for the whole population in the study. The Kaplan-Meier survival curve is defined as the probability of surviving in a given period of time. Each period of time is the interval between two nonsimultaneous terminal events. There were no survival data censored as no information about the survival time of any individual was lost. (b–h) Plots depict Kaplan-Meier survival curves of individual biomarkers tagged as significant prognosis markers: (b) clinical response; (c) age; (d) BDNF; (e) HVEM/TNFRSF14; (f) IL-24; (g) IL-29; (h) leptin-R; (i) LRP-6; and (j) ROBO4. The cut-off values were determined considering those points which maximized the dichotomization between poor and fair prognosis. The
First, a univariate approach was used in this study to identify relevant and independent measurable factors at prognosis that could be associated to a higher risk of PDAC death. Serum levels of cytokines before treatment and clinicopathologic parameters such as age, gender, stage, and clinical response were analyzed. Amongst the clinicopathologic parameters, age and the clinical response (progressive or nonprogressive disease, according to the RECIST criteria [
Prognosis factors in univariate analysis.
Variable | Overall survival | ||||
---|---|---|---|---|---|
HR | 95% CI |
|
|||
BDNF | 0.005 | 1.005 | 1.000 | 1.009 | 0.034 |
HVEM/TNFRSF14 | −0.079 | 0.924 | 0.858 | 0.996 | 0.038 |
IL-24 | 0.040 | 1.041 | 1.006 | 1.078 | 0.023 |
IL-29 | 0.012 | 1.012 | 1.002 | 1.023 | 0.021 |
Leptin R | 0.008 | 1.008 | 1.001 | 1.015 | 0.018 |
LRP-6 | 0.027 | 1.027 | 1.004 | 1.051 | 0.022 |
ROBO4 | 0.002 | 1.002 | 1.000 | 1.004 | 0.045 |
Age | 0.086 | 1.089 | 1.008 | 1.177 | 0.030 |
Clinical response | 2.064 | 8.706 | 1.057 | 71.692 | 0.013 |
|
|||||
Cytokines | Overall model fit ( |
||||
HR | 95% CI |
|
|||
|
|||||
IL-24 (1) | 0.042 | 1.042 | 1.003 | 1.023 | 0.026 |
IL-29 (2) | 0.014 | 1.014 | 1.005 | 1.081 | 0.017 |
Despite the fact that often only those statistically significant variables in univariate analysis are included in multivariate analysis, some variables not being significant in univariate analysis may appear jointly significant in a multivariate analysis. Thus, in addition to the statistically significant variables related to poor prognosis on the univariate analysis, those also selected by the features selection procedure were also included in the multivariate model. In proteomics studies, the number of samples is usually low compared to the number of variables, due to the limited availability or the cost of measurements. Taking this into account and in order not to introduce bias due to the small sample problem, a wrapper was used as a feature selection method using conditional forward stepwise algorithm based on likelihood rate to reduce the dimensionality of the data [
The best combination of cytokines selected by the multivariate Cox’s proportional hazard analysis is shown in Table
Prognosis factors in multivariate analysis.
Cytokines | Overall survival | Overall model fit | ||||
---|---|---|---|---|---|---|
HR | 95% CI |
|
|
|||
IL-29 | 0.081 | 1.084 | 1.010 | 1.164 | 0.026 | 0.004212 |
B7-1/CD80 | 4.351 | 77.574 | 1.138 | 5289.45 | 0.043 | 0.002494 |
PD-ECGF | 0.264 | 1.302 | 0.944 | 1.797 | 0.108 | 0.001350 |
EG-VEGF/PK1 | 0.003 | 1.003 | 1.000 | 1.005 | 0.049 | 0.000134 |
NRG1-beta1/HRG1-beta1 | 0.020 | 1.020 | 0.994 | 1.047 | 0.129 | 0.000286 |
|
||||||
Cytokines | Overall survival in the univariate analysis | |||||
HR | 95% CI |
|
||||
|
||||||
IL-29 | 0.012 | 1.012 | 1.002 | 1.023 | 0.021 | |
B7-1/CD80 | 0.373 | 1.452 | 0.876 | 2.407 | 0.148 | |
PD-ECGF | 0.044 | 1.045 | 0.997 | 1.096 | 0.068 | |
EG-VEGF/PK1 | −0.0001 | 1.000 | 0.999 | 1.000 | 0.640 | |
NRG1-beta1/HRG1-beta1 | −0.004 | 0.996 | 0.979 | 1.014 | 0.673 |
As combinations of biomarkers are likely to provide more accurate prognosis information, the most accurate subset of variables was sought using the conditional forward stepwise regression approach based on likelihood rate. To illustrate the interrelated effect on OS of the seven markers highlighted by the univariate analysis, the Cox’s proportional hazard analysis was employed to select those variables jointly correlated with the survival. As a result of this analysis, a model containing only IL-24 (
Regarding filtered cytokines obtained by multivariate analysis, a second statistically significant (
Whether these PI can contribute to accurately model survival for this patient cohort was assessed by regression analyses.
The Cox’s regression model. Observed (denoted by square, diamonds and triangles points) and predicted (denoted by solid line) prognosis curves for the PDAC patients according to (a) univariate o and (b) multivariate Cox’s proportional hazard model analysis. As explained in the text, the stepwise procedure based on the likelihood ratio was used to select a model containing a statistically significant subset of prognosis factors. The three predicted prognosis curves in (b) are derived from step 3 (where three cytokines are included), step 4 (four cytokines included), and step 5 (five cytokines included) of this stepwise procedure. The predicted survival curves are adjusted to a logarithmic distribution function, as expected. The coefficient of determination
Prognosis index for multivariate model with these five cytokines ranged from 0 to 40 in our cohort. Patients were categorized into two groups according to their prognosis index: poor prognosis (PI > 17) and fair prognosis (PI < 17). Survival curves were then compared among these two prognosis groups (Figure
Kaplan-Meier PI survival curves. (a) shows survival plot for PI derived from univariate model, embracing 2 cytokines. A cut-off of 1.5 was chosen to divide cohort of patients in short (<6 months) and long (>6 months) survival times. (b) shows survival plot for PI derived from multivariate model, embracing 5 cytokines. A cut-off of 17 was chosen to divide cohort of patients in short (<6 months) and long (>6 months) survival times. Both PI cut-off values were established considering the best discrimination between poor and fair prognosis. The
In this work, we have conducted an extensive analysis of serum prognosis biomarkers using an antibody array comprising 507 human proteins including cytokines, chemokines, adipokines, growth factors, angiogenic factors, proteases, soluble receptors, soluble adhesion molecules, and other proteins. The main objective of this analysis was to determine if a specific cytokine panel in patient before Gemcitabine and Erlotinib treatment could influence the survival time after this treatment. This is a powerful tool with great potential in applications for biomarker discovery [
In the course of our evaluation, we first identified 2 cytokines that correlated with patients’ prognosis in univariate analysis. Following, a panel of 5 cytokines clearly demonstrated a remarkably better overall performance for modeling OS. Therefore, the multivariate model consisting of B7-1/CD80, EG-VEGF/PK1, IL-29, NRG1-beta1/HRG1-beta1, and PD-ECGF was shown to be more accurate than the univariate model considering the most significant markers. The effectiveness of our model is clearly demonstrated with the evaluation performed by the LOOCV.
Notwithstanding proposed roles for B7-1/CD80, EG-VEGF/PK1, and NRG1-beta1/HRG1-beta1 in PDAC, to the best of our knowledge this is the first time that this combination of serum cytokines has been collectively described as prognosis factors for PDAC. An overview of these biomarkers is subsequently given.
It may not be possible for one single biomarker to provide the necessary prognosis information about the patient to base treatment options on. For this reason, panels of biomarkers are advisable to accurately predict the stage of the disease and how it will progress. Previous studies have indicated that tumor prognosis is closely associated with immune escape by tumor cells. A dynamic relationship between the host immune system and cancer is emerging [
We are aware of the limitation imposed by population size in this study. However, we have tried to apply a robust statistical analysis and validation. Although PDAC is amongst the less prevalent cancer and studies with large sample size are difficult to be carried out, its aggressiveness and the poor outcome urge to search novel prognosis biomarkers as the basis for rational treatment decisions, analysis of novel therapeutic interventions, and tailored treatment approaches [
In summary, we have identified for the first time a panel of five serum cytokines comprising B7-1/CD80, EG-VEGF/PK1, IL-29, NRG1-beta1/HRG1-beta1, and PD-ECGF with prognosis significance in PDAC. These molecules might not only allow a more accurate prediction of prognosis of patients with PDAC but also represent novel targets for therapeutic agents. Studies in prognosis biomarkers achieving true clinical impact and improving patient management and outcome are a matter of the utmost importance in PDAC. Besides, being able to foresee the prognosis of a PDAC patient may help to avoid futile therapy approaches and to improve quality of life of those whose long-term survival is unpromising.
Pancreatic ductal adenocarcinoma
Overall survival
Hazard ratio
Prognosis index
Confidence interval.
No potential conflict of interests was disclosed.
The study was fully supported by ROCHE FARMA S.A (ref. H/OH-TAR-10/131 and ref. H/OH-TRR-08/59) and Instituto de Salud Carlos III (ISCIII) (Clinical trial ref. EC08/00009), and the Government of Andalusia Project P12-TIC-2082.