Plasma inflammatory cytokines and survival of pancreatic cancer patients

Abstract

Objectives

Inflammation and inflammatory conditions have been associated with pancreatic cancer risk and progression in a number of clinical, epidemiological, and animal model studies. The goal of the present study is to identify plasma markers of inflammation associated with survival of pancreatic cancer patients, and assess their joint contribution to patient outcome.

Methods

We measured circulating levels of four established markers of inflammation (C-reactive protein (CRP), interleukin-6 (IL-6), soluble tumor necrosis factor receptor type II (sTNF-RII), and macrophage inhibitory cytokine-1 (MIC-1)) in 446 patients enrolled in an ongoing prospective clinic-based study. Hazard ratios (HRs) and 95% confidence intervals (CI) for death were estimated using multivariate Cox proportional hazards models.

Results

Overall mortality was significantly increased in patients in the top quartile of CRP (HR = 2.52, 95% CI: 1.82–3.49), IL-6 (HR = 2.78, 95% CI: 2.03–3.81), sTNF-RII (HR = 2.00, 95% CI: 1.46–2.72), and MIC-1 (HR = 2.53, 95% CI: 1.83–3.50), compared to those in the bottom quartile (P-trend <0.0001 for all four comparisons). Furthermore, patients with higher circulating concentrations of all four cytokines had a median survival of 3.7 months; whereas, those with lower levels had a median survival of 19.2 months (HR = 4.55, 95% CI: 2.87–7.20, P-trend <0.0001).

Conclusion

Individual elevated plasma inflammatory cytokines are associated with significant and dramatic reductions in pancreatic cancer patient survival. Furthermore, we observed an independent combined effect of those cytokines on patient survival, suggesting that multiple inflammatory pathways are likely involved in PDAC progression. Future research efforts to target the inflammatory state using combination strategies in pancreatic cancer patients are warranted.

Introduction

Pancreatic ductal adenocarcinoma (PDAC) is the fourth leading cause of cancer death in the US, with a 5-year survival rate of only 8%¹. Several pathologic characteristics obtained at surgery, such as tumor size, resection margin, and number of involved lymph nodes have been associated with reduced survival². However, only 15–20% of PDAC patients are able to undergo resection³. Consequently, identification of additional biomarkers that can be easily assayed in all patients is urgently needed for elucidation of underlying mechanisms of disease progression and development of new therapeutic strategies.

Numerous studies have suggested a role of inflammation in pancreatic cancer development and progression⁴. Cytokines synthesized by the host, tumor cells, and stromal cells play a role in cellular proliferation, angiogenesis, and metastasis^5,6. Furthermore, epidemiological studies reveal an association between several inflammation-associated conditions, such as diabetes⁷ and obesity⁸, with PDAC survival, further implicating inflammation in PDAC pathogenesis.

Elevated levels of individual inflammatory markers, including C-reactive protein (CRP), interleukine-6 (IL-6), macrophage inhibitory cytokine-1 (MIC-1), and tumor necrosis factor-α (TNF-α), as well as combined scores, such as the Glasgow Prognostic Score (GPS), or CRP to albumin ratio, have been associated with decreased survival in PDAC patients^9–24. However, these studies were small and often did not account for potential confounding variables. Moreover, it is not known to what extent these inflammatory pathways, which have distinct roles in tumor development, such as cell proliferation, invasion, angiogenesis, and immune evasion^5,6 might be involved in disease progression and patient survival. We therefore evaluated the individual and combined association of these four markers of inflammation with patient survival in a large prospective study of well-characterized PDAC patients.

Materials and methods

Study population

Patients were drawn from an ongoing clinic-based study at the Dana-Farber Cancer Institute (DFCI; Boston, MA). All new patients with a diagnosis of PDAC who were seen in the DFCI Gastrointestinal Cancer Center outpatient clinic were prospectively identified and enrolled to this cohort study between December 22, 2004 and June 16, 2014. Patients were eligible for the study if they had pathologically confirmed PDAC and were age 21 or older. The study was approved by the Dana-Farber/Harvard Cancer Center Institutional Review Board. All participants provided informed consent for their biological specimens and clinical data to be used for research.

Pancreatic cancer cases

A total of 1,038 DFCI patients were approached for consent between December 22, 2004 and June 18, 2014, and 743 (72%) agreed to participate (Supplementary Figure 1). Of patients that provided consent, 485 completed a questionnaire on medical history, medication use, lifestyle, and family history. There were no differences between patients who did or did not complete the questionnaire in regards to gender, age at diagnosis, race/ethnicity, or stage at diagnosis. Of the 485 patients who completed the questionnaire, 450 patients provided blood samples at an average of 1.4 months after diagnosis (Table 1). We excluded patients with missing information about height (n = 1), gender (n = 1), unknown stage at the time of blood draw (n = 1), and missing inflammatory cytokine data (n = 1).

Table 1.

Baseline characteristics of pancreatic cancer patients included in the study

Characteristics	Pancreatic cancer cases(N = 446)
Mean (SD)
Age at blood draw, years	64.1 (10.4)
BMI, kg/m2	29.6 (74.2)
Physical activity, MET-hr/wk	14.3 (25.1)
Median (SD)
Time between diagnosis and blood draw, months	1.4 (7.6)
Time between blood draw and survey, months	0 (2.6)
Time between surgery and blood draw, monthsa	1.6 (5.3)
Number of metastatic sitesb	1.0 (0.7)
CA19-9 at blood draw, U/ml	644 (274,892)
Gender, No. (%)
Female	211 (47)
Male	235 (53)
Diabetes, No. (%)
No	279 (63)
Yes	141 (32)
Unknown	26 (6)
Cancer stage, No. (%)
Localized	12 (3)
Locally advanced	100 (22)
Metastatic	271 (61)
No. of evidence of disease	63 (14)
Grade, No. (%)
Well/moderately differentiated	84 (19)
Poorly differentiated/undifferentiated	96 (22)
Unknown	266 (60)
Smoking status at blood draw, No. (%)
Never	194 (44)
Former	216 (48)
Current	32 (7)
Unknown	4 (1)
Regular aspirin use at blood draw c , No. (%)
No	145 (33)
Yes	142 (32)
Unknown	159 (36)
Treatment d status at time of blood draw, No. (%)
Treatment naive	234 (53)
On treatment	164 (37)
Post treatment	48 (11)

SD standard deviation, BMI body mass index, MET-hr metabolic equivalent of task-hour

^aAmong patients who underwent surgical resection

^bAmong patients with metastatic disease

^cRegular use is defined as intake frequency of≥3 days/week

^dIncludes chemotherapy and radiation

Exposure assessment

CRP, MIC-1, and sTNF-RII were assayed in the laboratory of Dr. Nader Rifai (Boston Children’s Hospital, Boston, MA). The sTNF-RII is an established surrogate measurement for TNF-α due to its role in TNF-α signaling, lower diurnal variation, and increased stability in frozen plasma^25,26. Furthermore, unlike TNF-α levels of which tend to fluctuate, levels of sTNF-RII are stable over long periods of time²⁷. CRP was measured using an immunoturbidimetric assay (Roche Diagnostics, Indianapolis, IN), with a limit of detection of 0.03 mg/L. MIC-1 and sTNF-RII were measured by an ELISA assay (R&D Systems, Minneapolis, MN), with a sensitivity of 4.36 pg/mL for MIC-1 and 0.6 pg/mL for sTNF-RII. IL-6 was measured as part of the 16-plex pro- and anti-inflammatory cytokine panel (Human Cytokine A Premixed Magnetic Luminex Performance Assay, R&D Systems, Minneapolis, MN). The sensitivity of the assay is 1.11 pg/mL. Coefficients of variation for each assay were calculated using 10% blinded duplicate samples, and ranged from 2.5% for CRP to 6.1% for sTNF-RII. Laboratory personnel was blinded to patient status.

Covariate assessment

Data on patient and disease characteristics, such as age at time of blood draw, albumin levels, gender, body mass index (BMI) at time of blood draw, date of diagnosis, stage, treatment history, and date of death were extracted from the medical record. Information on race, smoking status, physical activity, and aspirin and non-steroidal anti-inflammatory drug (NSAID) use at time of blood draw were extracted from the self-administered questionnaire. GPS was calculated as previously described^9,24.

Statistical analysis

All inflammatory cytokines were log-transformed to improve normality. Correlation between cytokines was analyzed using Spearman correlation. We used the Wilcoxon rank-sum or Kruskal–Wallis test to evaluate differences in cytokine levels between two or more groups of interest.

We used the Cox proportional hazards model to evaluate the hazards ratios (HRs) and 95% confidence intervals (CIs) for mortality. Person-time was calculated as time between blood collection and death or last follow-up (November 16, 2016). To test the proportionality of hazards assumption, we evaluated the cross product of time and inflammatory cytokines. This test revealed a violation of the proportionality of hazards assumption which was addressed by including an interaction term between time and cytokine levels in the models, allowing calculation of HRs for different time points. Inflammatory cytokines were modeled as quartiles. To evaluate the trend of the association between inflammatory cytokines and survival across quartiles, we used the median of each quartile as a continuous variable in the model. In multivariate models, we adjusted a priori for age at blood collection, gender, grade (well differentiated, moderately differentiated, poorly differentiated, undifferentiated, unknown), cancer stage (localized/no evidence of disease, locally advanced, metastatic), treatment status (treatment naive, on treatment, post treatment), number of metastatic sites, BMI (continuous), and physical activity (continuous) at time of blood collection. We additionally examined potential confounding by smoking, aspirin use, NSAID use, and diabetes status at time of blood collection.

To investigate whether the combination of cytokine concentrations is more strongly related to mortality than each individual cytokine alone, we also evaluated the association between a combined inflammatory cytokine score and mortality. We calculated this combined score by summing the number of cytokines with levels above the population median. The score therefore ranged from 0 (no cytokines above the median) to 4 (all four cytokine levels above the median).

To compare discrimination between different survival models, we calculated the overall C-index. This metric is an extension of the receiver operating characteristic for the Cox proportional hazard model²⁸.

We performed subgroup analyses to examine potential effect modification by age (<median of 64.5 years, ≥64.5 years), gender, BMI (<25 kg/m², ≥25 kg/m²), diabetes (yes vs. no), smoking status (never vs. ever smoker), regular aspirin use (yes vs. no), grade of differentiation (well/moderately vs. poorly differentiated/undifferentiated), and metastatic status (non-metastatic vs. metastatic). We also performed a stratified analysis by treatment status (treatment naive vs. on/post treatment) for metastatic patients only. Statistical interaction was evaluated by including a cross-product term containing the stratification variable and inflammatory cytokine level (as quartiles) into the model and performing the likelihood ratio test.

Kaplan–Maier method and log-rank tests were used to illustrate and analyze survival curves. Statistical analyses were performed using SAS software (version 9.4; SAS Institute, Cary, NC). All P-values were two-sided.

Results

Characteristics of the study population are shown in Table 1. The study included 211 female (47%) and 235 (53%) male participants. The mean age at diagnosis was 64.1 years. At the time of blood draw, 3% of patients had localized disease, 22% had locally advanced disease, 61% had metastatic disease, and 14% of patients had no evidence of disease following surgical resection. Most patients (53%) were treatment naive at the time of blood draw, 37% were on active treatment with chemotherapy and/or radiation, and 11% were post treatment.

Median survival time in the entire patient cohort was 9.7 months (30.1 months for patients with no evidence of disease following surgical resection, 19.3 months with patients with localized disease, 12.4 months for those with locally advanced tumors, and 6.5 months for those with metastatic disease). Median follow-up was 9.3 months; at last follow-up, 413 patients (92.6%) had died.

We observed significant, but weak to moderate, correlations between cytokines, with correlation coefficients ranging from 0.12 (IL-6 and sTNF-RII) to 0.41 (MIC-1 and sTNF-RII) (Supplementary table 1). CRP levels differed significantly by time between diagnosis and blood draw, BMI, stage, and grade (Table 2). IL-6 levels were higher in regular aspirin users and patients with higher BMI. sTNF-RII levels were higher among older patients and patients with poorly differentiated and undifferentiated tumors. MIC-1 levels differed significantly by age, gender, time between diagnosis and blood draw, cancer stage, presence of diabetes, aspirin use, and treatment status.

Table 2.

Plasma inflammatory biomarkers according to selected patient and tumor characteristics

Characteristics	CRP (mg/L) (n = 437) median (IQ range)	P-valueb	IL-6 (pg/mL) (n = 427) median (IQ range)	P-valueb	sTNF-RII (pg/mL) (n = 423) median (IQ range)	P-valueb	MIC-1 (pg/mL) (n = 434) median (IQ range)	P- Valueb
Age a
<64.5 years	8.1 (24.9)		2.6 (2.4)		3443.9 (2060.2)		1698.2 (1819.5)
≥64.5 years	7.3 (20.6)	0.47	3.0 (2.2)	0.19	3726.6 (2390.5)	0.01	2064.8 (1931.4)	0.005
Time between diagnosis and blood draw a
<1.4 months	10.8 (24.2)		2.8 (2.4)		3508.0 (2309.6)		1719.5 (1689.2)
≥1.4 months	5.4 (17.7)	0.002	2.8 (2.2)	0.29	3730.7 (1890.7)	0.07	1995.0 (1938.5)	0.01
BMI a
<25.3 kg/m2	5.4 (18.7)		2.6 (1.8)		3531.7 (2619.3)		1837.2 (1824.2)
≥25.3 kg/m2	12.1 (24.0)	<0.0001	3.0 (2.4)	0.001	3641.2 (2619.3)	0.30	1938.2 (2002.4)	0.41
Physical activity a
<3.9 MET-hr/wk	7.4 (24.4)		2.9 (2.4)		3512.2 (2213.7)		2055.1 (2039.1)
≥3.9 MET-hr/wk	6.7 (25.5)	0.87	2.5 (1.7)	0.71	3606.8 (1827.2)	0.29	1937.0 (1413.4)	0.40
Gender
Male	9.5 (24.5)		2.8 (2.6)		3660.3 (2015.4)		2022.5 (2462.1)
Female	7.2 (18.7)	0.13	2.8 (1.8)	0.87	3493.4 (2395.8)	0.59	1773.4 (1656.8)	0.03
Diabetes
No	7.4 (20.6)		2.8 (2.2)		3547.0 (2174.2)		1819.2 (1744.2)
Yes	9.0 (26.9)	0.44	2.8 (2.3)	0.92	3644.6 (2129.6)	0.41	2221.1 (2811.6)	0.01
Cancer stage
Localized	4.5 (10.3)		1.8 (2.1)		3490.5 (2103.0)		1443.1 (1397.0)
Locally advanced	4.9 (16.0)		2.8 (1.9)		3290.3 (1827.9)		1727.1 (1924.3)
Metastatic	12.6 (30.5)		2.8 (2.5)		3803.6 (2182.2)		2165.3 (1991.8)
NED	2.9 (8.5)	<0.0001	2.7 (2.3)	0.14	3497.0 (1973.2)	0.05	1390.7 (901.5)	<0.0001
Grade
Well/moderately differentiated	4.1 (15.3)		2.7 (2.4)		3112.0 (1454.8)		1575.3 (1529.7)
Poorly differentiated/undifferentiated	11.1 (31.3)	0.01	3.0 (2.7)	0.16	3944.5 (2586.0)	0.005	1975.2 (1846.0)	0.09
Smoking
Never	8.0 (21.8)		2.8 (2.2)		3570.1 (2040.3)		1877.2 (1943.7)
Past	7.8 (23.0)		2.8 (2.5)		3714.3 (2070.0)		1949.8 (2034.3)
Current	11.7 (22.4)	0.55	2.8 (2.9)	0.39	3569.5 (1722.8)	0.30	1734.6 (2031.2)	0.85
Regular aspirin use c
Non-users	5.5 (24.0)		2.5 (1.6)		3509.2 (1826.6)		1869.3 (1451.4)
Users	10.6 (22.9)	0.09	3.0 (2.5)	0.01	3768.1 (2708.8)	0.03	2180.8 (2586.0)	0.04
Regular NSAID use c
Non-users	6.6 (25.0)		2.8 (2.2)		3661.9 (2090.5)		2064.8 (1746.5)
Users	8.9 (34.2)	0.21	2.8 (1.5)	0.96	3522.4 (1950.8)	0.59	1769.7 (1344.3)	0.44
Treatment status d
Treatment naive
On/post treatment	8.8 (21.1)		2.8 (2.1)		3529.3 (2182.0)		1672.5 (1611.1)
	6.9 (23.7)	0.74	2.8 (2.2)	0.11	3700.3 (2082.3)	0.53	2145.8 (2580.3)	<0.0001

CRP, C-reactive protein; IL-6, interleukin-6; MIC-1, macrophage inhibitory cytokine-1; sTNF-RII, tumor necrosis factor receptor 2; IQ, interquartile range; BMI, body mass index; MET-hr, metabolic equivalent of task-hour; NED, no evidence of disease, NSAID, nonsteroidal anti-inflammatory drug

^aCut point determined by the median value

^bCalculated using Wilcoxon rank-sum, or Kruskal–Wallis test

^cRegular use is defined as intake frequency of  ≥3 days/week

^dIncludes chemotherapy and radiation

Bold value denote significance P-value ≤ 0.05

For each measured cytokine, patients with levels in the highest quartile had significantly worse survival (log-rank P-value <0.0001) (Fig. 1a–d). In the multivariate model, compared to patients in the lowest quartile, the mortality hazard of patients in the highest quartiles at the median survival time of 9.5 months was 2.52 (95% CI: 1.82–3.49) for CRP, 2.78 (95% CI: 2.03–3.81) for IL-6, 2.00 (95% CI: 1.46–2.72) for sTNF-RII, and 2.53 (95% CI: 1.83–3.50) for MIC-1 (Table 3). Moreover, there was a significant linear trend across quartiles (P-trend <0.0001 for all four cytokines). Further adjustment for diabetes, smoking, aspirin use, and non-aspirin NSAID use did not alter the associations (data not shown). We observed similar HRs for death across cytokine quartiles at 6 and 12 months (Supplementary Tables 2 and 3).

Fig. 1. Patient survival by quartiles of inflammatory cytokines.

A combined inflammatory score was created by adding number of inflammatory markers with the value above the population median

Table 3.

Hazard ratios^a for death by inflammatory biomarker levels

	HR (95% CI)
	Quartile 1	Quartile 2	Quartile 3	Quartile 4	P-trendc
CRP (N = 437)
Range (mg/L)	0.05–2.34	2.35–7.72	7.81–24.26	24.42–183.74
Person-months	2356	1616	1376	865
Cases/deaths	109/97	109/101	110/100	109/106
Median OS, months	16.6	10.7	8.4	4.3
Age-adjusted	1.00 (Ref)	1.66 (1.22–2.27)	1.98 (1.45–2.71)	3.28 (2.40–4.49)	<.0001
Multivariate-adjustedb	1.00 (Ref)	1.48 (1.08–2.04)	2.08 (1.52–2.86)	2.52 (1.82–3.49)	<.0001
IL-6 (N = 427)
Range (pg/ml)	0.17–1.86	1.89–2.76	2.78–4.15	4.17–165.73
Person-months	1905	1791	1299	1030
Cases/deaths	105/95	104/95	110/102	108/103
Median OS, months	14.8	12.1	8.4	4.9
Age-adjusted	1.00 (Ref)	1.18 (0.86–1.62)	1.83 (1.35–2.48)	2.41 (1.78–3.26)	<.0001
Multivariate-adjustedb	1.00 (Ref)	1.11 (0.80–1.54)	2.22 (1.62–3.03)	2.78 (2.03–3.81)	<.0001
sTNF-RII (N = 423)
Range (pg/ml)	989.36–2655.95	2657.81–3591.33	3606.79–4771.55	4825.31–9918.17
Person-months	1811	1938	1300	1107
Cases/deaths	105/90	106/97	106/104	106/100
Median OS, months	12.5	12.8	8.1	7.0
Age-adjusted	1.00 (Ref)	1.05 (0.77–1.43)	1.59 (1.18–2.14)	1.90 (1.41–2.56)	<.0001
Multivariate-adjustedb	1.00 (Ref)	1.05 (0.76–1.45)	1.44 (1.06–1.94)	2.00 (1.46–2.72)	<.0001
MIC-1 (N = 434)
Range (pg/ml)	344.40–1215.34	1216.57–1896.73	1897.19–3172.78	3183.01–31829.82
Person-months	2021	2054	1119	999
Cases/deaths	108/95	109/98	109/104	108/104
Median OS, months	14.1	13.2	7.6	4.9
Age-adjusted	1.00 (Ref)	1.20 (0.87–1.64)	2.30 (1.69–3.12)	2.61 (1.91–3.56)	<.0001
Multivariate-adjustedb	1.00 (Ref)	1.34 (0.97–1.86)	2.06 (1.50–2.84)	2.53 (1.83–3.50)	<.0001

CRP, C-reactive protein; IL-6, interleukin-6; MIC-1, macrophage inhibitory cytokine-1; sTNF-RII, tumor necrosis factor receptor 2. OS, overall survival. HR, hazard ratio. CI, confidence interval

^aHazard ratios at the median survival time of our cohort (9.5 months)

^bAdjusted for age at blood draw, gender, grade (well differentiated, moderately differentiated, poorly differentiated, undifferentiated, and unknown), stage (localized/no evidence of disease, locally advanced, and metastatic), treatment status (naive, on, and post), number of metastatic sites, and BMI (continuous) at blood draw and physical activity (continuous)

^cP-trend values calculated by entering quartile medians as continuous variables in Cox proportional hazards model

We also examined the association between CRP levels and mortality using the conventional CRP cutoff of 10 mg/L. Compared to patients with CRP ≤ 10 mg/L, those with CRP >10 mg/L had a twofold increase in the risk of death (multivariate HR: 2.00, 95% CI: 1.62–2.47). When simultaneously adjusting for all four cytokines in the multivariate model, we observed a continued significant mortality hazard for patients in the highest quartile of CRP (HR: 1.59, 95% CI: 1.07–2.37), IL-6 (HR: 1.80, 95% CI: 1.23–2.65), and MIC-1 (HR: 1.80, 95% CI: 1.22–2.67), while the association between sTNF-RII and survival was no longer significant (HR: 1.10, 95% CI: 0.75–1.59). Compared to patients with an inflammatory score of 0 and median survival 19.2 months, those with a score of 4 had a median survival of only 3.7 months and adjusted HR of 4.55 (95% CI: 2.87–7.20; P-trend <0.0001) (Table 4, Fig. 1e).

Table 4.

Hazard ratios^a for death by combined inflammatory marker score

	Inflammatory scoreb
	HR (95% CI)
	0	1	2	3	4	P-trendc
Person-months	1765	1790	1240	959	460
Cases/deaths	74/65	103/90	94/88	94/92	71/68
Median OS, months	19.2	13.8	7.6	6.5	3.7
Age-adjusted	1.00 (Ref)	1.73 (1.16–2.57)	2.39 (1.60–3.56)	3.36 (2.25–5.03)	4.53 (2.90–7.09)	<.0001
Multivariate-adjustedd	1.00 (Ref)	1.82 (1.21–2.73)	2.60 (1.72–3.95)	3.52 (2.32–5.35)	4.55 (2.87–7.20)	<.0001

HR, hazard ratio; CI, confidence interval; OS, overall survival

^aHazard ratios at 9.5 months (median survival time)

^bCreated by adding number of inflammatory markers with a value above the study population median

^cP-trend calculated by entering inflammatory score as continuous variables in Cox proportional hazards model

^dAdjusted for age at blood draw, gender, grade (well differentiated, moderately differentiated, poorly differentiated, undifferentiated, and unknown), stage (localized/no evidence of disease, locally advanced, and metastatic), treatment status (naive, on, and post), number of metastatic sites, BMI (continuous) at blood draw and physical activity (continuous)

The model including CRP had a discriminatory index of 0.80 (95% CI: 0.71–0.88), comparable to those of GPS (C-index = 0.76, 95% CI: 0.66–0.85) and CRP to albumin ratio (C-index = 0.74, 95% CI: 0.64–0.83) (Supplementary Tables 4 and 5).

The significant association between increasing cytokine levels and worse mortality was consistent across most subgroups of known prognostic characteristics (Table 5). We observed a significant interaction between CRP and BMI (P-interaction = 0.03), and between CRP and aspirin use (P-interaction = 0.001), with association with worse mortality being stronger among patients with BMI ≤25 kg/m² and among aspirin users.

Table 5.

Hazard ratios^a for death among selected patient subgroups

	CRP	IL-6	sTNF-RII	MIC-1
	>10 mg/L vs. ≤ 10 mg/L	Q4 vs. Q1	Q4 vs. Q1	Q4 vs. Q1
	HR (95% CI)b	HR (95% CI)b	HR (95% CI)b	HR (95% CI)b
Age c
≤64.5 years	2.09 (1.53–2.86)	3.96 (2.48–6.33)	2.21 (1.41–3.45)	3.10 (1.96–4.90)
>64.5 years	2.00 (1.46–2.72)	1.78 (1.12–2.82)	2.18 (1.36–3.49)	2.41 (1.44–4.04)
P-interactiond	0.96	0.05	0.93	0.78
BMI c
≤25 kg/m2	2.83 (1.98–4.05)	4.70 (2.75–8.03)	2.52 (1.52–4.19)	2.56 (1.57–4.17)
>25 kg/m2	1.73 (1.23–2.31)	2.53 (1.60–4.00)	1.65 (1.10–2.49)	2.52 (1.60–3.96)
P-interactiond	0.03	0.12	0.12	0.49
Gender
Male	1.89 (1.40–2.55)	2.66 (1.74–4.07)	2.93 (1.90–4.51)	3.31 (2.10–5.24)
Female	2.44 (1.77–3.37)	3.22 (1.90–5.43)	1.41 (0.88–2.23)	2.40 (1.45–3.99)
P-interactiond	0.10	0.84	0.17	0.49
History of diabetes
No	1.91 (1.44–2.54)	2.93 (1.90–4.52)	1.67 (1.12–2.48)	2.06 (1.33–3.19)
Yes	2.65 (1.74–4.02)	4.25 (2.31–7.82)	3.27 (1.70–6.29)	5.20 (2.67–10.13)
P-interactiond	0.64	0.90	0.83	0.37
Metastatic disease
No	1.47 (0.95–2.27)	2.29 (1.18–4.47)	1.33 (0.74–2.39)	1.65 (0.93–2.93)
Yes	2.10 (1.59–2.76)	2.40 (1.58–3.66)	1.72 (1.13–2.62)	2.92 (1.88–4.53)
P-interactiond	0.15	0.29	0.96	0.11
Number of metastatic sites e
1	1.99 (1.42–2.79)	2.32 (1.37–3.92)	1.59 (0.93–2.72)	2.91 (1.75–4.84)
>1	2.05 (1.16–3.63)	2.44 (1.05–5.72)	1.70 (0.72–4.01)	6.41 (1.96–21.10)
P-interactiond	0.60	0.43	0.25	0.30
Grade
Well/moderately differentiated	2.42 (1.23–4.76)	2.41 (1.07–5.42)	3.77 (1.55–9.14)	2.82 (1.11–7.21)
Poorly differentiated/undifferentiated	2.56 (1.49–4.41)	5.20 (2.16–12.52)	2.30 (1.05–5.05)	3.23 (1.45–7.18)
P-interactiond	0.81	0.05	0.29	0.63
Smoking status
Never smokers	2.15 (1.52–3.03)	2.40 (1.47–3.93)	2.40 (1.51–3.82)	2.54 (1.52–4.23)
Ever smokers	1.99 (1.48–2.66)	3.43 (2.16–5.47)	1.73 (1.10–2.71)	2.93 (1.86–4.63)
P-interactiond	0.70	0.31	0.19	0.33
Regular aspirin use f
No	1.59 (1.05–2.41)	4.97 (2.55–9.70)	2.03 (1.04–3.96)	4.25 (2.03–8.91)
Yes	3.01 (1.91–4.75)	2.42 (1.32–4.45)	3.16 (1.68–5.94)	3.23 (1.62–6.43)
P-interactiond	0.001	0.34	0.35	0.62
Treatment status g
Treatment naive	2.08 (1.52–2.84)	2.88 (1.84–4.51)	1.70 (1.09–2.64)	2.02 (1.28–3.18)
On/post treatment	2.14 (1.54–2.95)	2.97 (1.82–4.83)	2.29 (1.44–3.64)	2.81 (1.63–4.82)
P-interactiond	0.43	0.72	0.60	0.54

CRP, C-reactive protein; IL-6, interleukin-6; MIC-1, macrophage inhibitory cytokine-1; sTNF-RII, tumor necrosis factor receptor 2; HR, hazard ratio; CI, confidence interval; BMI, body mass index

^aHazard ratios at 9.5 months (median survival time)

^bAdjusted for age at blood draw, gender, grade, (well differentiated, moderately differentiated, poorly differentiated, and unknown), stage (localized/no evidence of disease, locally advanced, and metastatic), treatment status (naive, on treatment, and post treatment), number of metastatic sites, BMI (continuous) at blood draw, physical activity (continuous), and excluding the stratifying variable

^cMedian population values

^dP-interaction was calculated by entering a cross-product term of stratifying variable and inflammatory cytokine (quartiles) into Cox proportional hazards model

^eAmong patients with metastatic disease

^fRegular use is defined as intake frequency of ≥3 days/week

^gRestricted to metastatic patients

To evaluate whether higher levels of inflammatory cytokines simply reflect a greater burden of disease, we adjusted for disease stage, number of metastatic sites and CA19-9 levels in the multivariate model and observed no significant change of the estimate. In addition, associations were similar by metastatic disease and number of metastatic sites (P-interaction >0.25) (Table 5).

Discussion

In this large prospective clinic-based study of 446 PDAC patients, subjects with higher levels of each of CRP, IL-6, and MIC-1 experienced a significant increase in overall mortality. Moreover, patients with elevations of all four markers combined had the largest comparative increase in mortality, suggesting involvement of multiple activated inflammatory pathways in PDAC progression. Compared to patients with all cytokines below the population median, those with all four cytokines above the population median had an almost fivefold increased hazard of death, with median survival of 4 vs. 19 months. These associations were independent of known prognostic factors, such as tumor stage, grade, number of metastatic sites, and CA19-9 levels. Furthermore, except for sTNF-RII, the effects of these cytokines on survival are mutually independent.

The inverse associations of CRP and IL-6 with patient survival in our study are consistent with previously reported findings^{10,12,13,19,20,22,23}. We further showed that the associations of these two cytokines are independent from each other, as well as from MIC-1 and sTNF-RII. While we observed a significant association between survival and sTNF-RII, a surrogate for TNF-α, the association was attenuated after adjusting for CRP, IL-6 and MIC-1. This therefore argues against an independent effect of TNF-α on survival, as has been previously suggested¹¹. Our finding of decreased survival among patients in the top quartile of MIC-1 compared to those in the lowest quartile is also consistent with previous studies^29,30; however, distinct from those prior studies, we were able to comprehensively adjust for potential confounders. We also found that the influence of high-MIC-1 levels was independent of CRP, IL-6, and sTNF-RII. To our knowledge, no previous studies have assessed the combined contribution of those four inflammatory cytokines on PDAC patient survival.

Several explanations have been proposed to address the relationship between inflammatory cytokines and survival. High levels of circulating cytokines may result from the systemic response of host to tumor, reflecting the tumor burden. CRP is produced by the liver as part of the acute phase response³¹, and its levels correlate with cancer progression^13,20,22,23. However, adjusting for clinical markers of aggressive disease did not change our observation. Moreover, in stratified analyses the associations between inflammatory cytokines and patient mortality were similar across subgroups of tumor burden.

Extensive experimental evidence supports the importance of inflammatory cytokines in tumor growth and progression, both by acting directly on tumor cells and by modifying the tumor microenvironment⁵. IL-6 exerts its protumorigenic effects by activating several signaling pathways involved in PDAC, such as JAK-STAT3, Ras-MAPK, and PI3K-Akt, leading to increased cellular proliferation, angiogenesis, and metastatic potential³². IL-6 is one of the key factors in the development of muscle wasting, or cachexia³³, which is responsible for about one third of PDAC-associated deaths and decreased response to treatment³⁴. Furthermore, it was shown that IL-6 leads to formation of desmoplastic stroma³⁵, a dense extracellular matrix which acts as a physical barrier for effective drug delivery³⁶. MIC-1, a member of the human transforming growth factor (TGF)-β superfamily, has both anti- and protumorigenic roles in colon, breast, prostate, and melanoma cancers³⁷. While differences in MIC-1 levels between PDAC patients and healthy controls have previously been reported^16,38,39, little is known about the molecular pathways underlying this association. Furthermore, it was shown that both IL-6^40,41 and MIC-1⁴² attenuate T-cell-mediated anti-tumor immune response. In PDAC⁴³, as well as other cancers, MIC-1 overexpression is associated with increased resistance to chemotherapy drugs³⁷. However, it is unlikely that resistance to treatment explains the effect of MIC-1 in our study, since there was no difference in survival in treatment naive or on/post treatment group by MIC-1 levels (Table 5).

Our results carry multiple potential clinical and translational implications. First, CRP, IL-6, and MIC-1 assays are inexpensive and non-invasive. Therefore, their prognostic potential should be further investigated in future studies. Furthermore, they could be used to identify patients who may benefit most from anti-inflammatory strategies. Indeed, interest in the use of circulating inflammatory cytokines as predictors of treatment efficacy was first suggested by the RECAP trial, a randomized phase II study of capecitabine with or without ruxolitinib (JAK/STAT inhibitor) in patients with refractory PDAC that showed a survival benefit with ruxolitinib in patients with high CRP or modified GPS¹⁵. Unfortunately, subsequent phase III trials did not confirm this finding, but have led to research efforts in other novel inflammation-mediated pathways, including inhibitors of TBK1, which regulates a KRAS-driven autocrine cytokine circuit⁴⁴, immunomodulatory agents such as CCR2 antagonists (ClinicalTrials.gov identifier NCT02732938), and vitamin D receptor VDR analogs, which have been shown to reprogram the tumor microenvironment^45. It is conceivable that levels of inflammatory cytokines may play a future role in helping to select the patients who are most likely to benefit from these novel agents.

Advantages of our study include the prospective study design, large number of patients, as well as detailed information on clinical, pathological, treatment, and lifestyle factors. We were also able to evaluate a variety of inflammatory cytokines singly and in combination. However, several limitations exist. We were not able to evaluate pancreatic cancer-specific mortality, but 95% of pancreatic cancer patients present with incurable disease at diagnosis, therefore it is highly unlikely that our patients died of other causes. Furthermore, our study consisted of predominantly white participants treated at a tertiary academic center, and therefore our results may not be generalizable to the overall population of pancreatic cancer patients. Reassuringly, though, the median survival of our study population is reflective of PDAC patients overall. Finally, while circulating inflammatory cytokines have the advantage of being easily measurable and accessible in patients, plasma levels may not adequately reflect inflammatory activity within the tumor or its microenvironment. Fortunately, recent technological advances that allow tumor RNA sequencing in bulk or on single-cell populations will pave the way toward elucidating the exact origin and mechanism of the high-inflammatory state of PDAC patients.

In conclusion, increasing levels of circulating inflammatory cytokines were associated with significantly decreased survival of patients with pancreatic cancer in this large prospective clinic-based study. The potential prognostic value of these markers, as well as their utility for patient selection for novel anti-inflammatory and immunomodulatory agents, should be evaluated in future studies. Moreover, efforts to understand the pathogenesis of the high-inflammatory state and development of novel agents to decrease inflammation in PDAC patients are warranted.

Study Highlights

What is current knowledge

• Inflammation and inflammatory conditions have been associated with increased risk of pancreatic cancer

• Individual inflammatory cytokines have been associated with survival of pancreatic cancer patients

What is new here

• The effect of inflammatory cytokines on pancreatic cancer patient survival is mutually independent and additive, suggesting that multiple inflammatory pathways are involved in PDAC progression

Conflict of interest

Guarantor of the article: Kimmie Ng.

Specific author contributions: Study conception and design: A.B., N.S., D.A.B., K.W., C.S.F., K.N. Data collection and/or assembly: A.B., N.S., N.P.N., M.M.Z., N.R., M.W.W., L.K.B., D.A.R., V.M.-O., B.M.W., M.H.K., C.S.F., K.N. Data analysis and/or interpretation: A.B., N.S., D.A.R., V.M.-O., Y.C., S.Z., E.M.P., B.M.W., M.K., D.A.B., K.W., C.S.F., K.N. Manuscript writing: A.B., N.P.N., M.M.Z., N.R., Y.C., S.Z., E.M.P., K.N. Approval of the final manuscript version: all authors.

Financial support:This work was supported by the National Institutes of Health research grants DF/HCC SPORE in Gastrointestinal Cancer-P50CA127003 (A.B., C.S.F.), Department of Defense CA130288, Celgene, Lustgarten Foundation for Pancreatic Cancer Research (C.S.F., B.M.W.), Perry S. Levy Fund for Gastrointestinal Cancer Research (B.M.W., C.S.F.), Broman Fund for Pancreatic Cancer Research (K.N.), Pappas Family Research Fund for Pancreatic Cancer (C.S.F.), the Noble Effort Fund, the Peter R. Leavitt Family Fund, and Promises for Purple (B.M.W.); and Robert T. and Judith B. Hale Fund for Pancreatic Cancer Research (A.B., B.M.W., C.S.F.).

Potential competing interests: The authors declare that they have no conflict of interest.