Plasma secretome analyses identify IL-8 and Nitrites as predictors of poor prognosis in nasopharyngeal carcinoma patients

Abstract

Predicting tumor recurrence and death in patients with nasopharyngeal carcinoma (NPC) remains to date challenging. We here analyzed the plasmatic secretomes of NPC untreated and relapsing patients, and explored possible correlations with the clinical and pathological features and survival characteristics of the corresponding patient cohorts, with the aim of identifying novel prognostic biomarkers. This study included 27 controls, 45 untreated NPC and 11 relapsed patients. A set of 14 plasma cytokines were analyzed using Millipore multiplex assay. Nitrites were assessed by Griess method. A comparative analysis of each groups’ secretome showed upregulation of IL-8, IL-12p70, IL-10 and IP-10 in untreated patients, and of IL-6, IL-10, MCP-1 and IP-10 in relapsing patients. Nitrites significantly correlated with IL-8 during relapse. Secretomes’ network analyses revealed prevalence of high correlations between IL8/IL-17A and IFN-γ/IL12p70 in the control group, between TNF-α/IL-8/IL-6, TNF-α/VEGF/IFN-γ and IL-10/MCP-1 in the untreated group, and between IL-8/IL-6/IL-10, TNF-α/IL-8/IL-6, IL12-p70/VEGF/IL-10/IFN-γ, IL-6/IL-10/IFN-γ and IL-8/IP-10 in the relapse group. IL-12p70, IP-10 and MCP-1 levels respectively associated with gender, age and node metastasis respectively. Recurrence-free survival (RFS) analysis showed that patients presenting High IL-8/Low NO immunological scores presented a combined 80% probability of relapse/death after 53 months (combined log-rank test p=0.0034; individual p=0.012 and p=0.016). Multivariate Cox hazard regression analysis revealed that IL-8 (HR=7.451; 95% CI [2.398–23.152]; p=0.001) and treatment type (HR=0.232; 95% CI 0.072–0.749; p=0.015) were independent prognostic factors. C&RT decision tree analysis showed that High IL-8/Low NO immunological scores predicted treatment failure in 50% cases starting the 36th month of follow-up (AUC= 1) for all of the studied cases and in 57% cases for patients receiving chemotherapy alone (AUC=1). Altogether, our results showed that NPC development is accompanied with cytokines deregulation to form specific interaction networks at time of diagnosis and relapse, and demonstrate that High IL-8/Low NO signature may constitute a predictor of poor prognosis which may be useful to improve risk stratification and therapy failure management.

Introduction

Undifferentiated NPC remains to date a major health problem for the southeastern Asian and North African populations [1], as despite the radio-chemotherapeutic sensitivity of the tumor, patients remain confronted to high rates of therapeutic failures, reflected by loco-regional/distant metastasis and death [2, 3]. The identification of patients at risk of therapeutic failure requires finding prognostic predictive biomarkers.

Recent studies have shown that tumor microenvironment (TME) enrichment in immune cells is crucial to tumor immune evasion and disease’ outcome [4–6], and that various soluble factors released by the TME, also found in the plasma secretome, were associated with tumor instability, growth and metastasis [7–9]. Many studies conducted on head and neck squamous cell carcinoma (HNSCC) highlighted the importance of cytokines as prognostic biomarkers. Among them, IL-6 and IL-8 were the most prominent cytokines linked to poor outcome and tumor progression [10–13]. Given these elements, the characterization of the TME immunological secretome have gained increased interest for the identification of predictive biomarkers of therapeutic failure [14–19]. In NPC, low NOS2 expression in the TME showed to be associated with intermediary plasmatic NO levels and risk of metastasis. Similarly, increased tissular expression of TNF-α and IL-6 were shown to be linked to enhanced tumor growth and metastatic capacities [20–22].

An increasing number of studies have been able to associate prognostic values to patients’ cytokine profiles combined to their clinical characteristics using data mining analysis [10, 23]. These emerging techniques proved to be crucial to decipher factors of risk and to bring clinical insights[24]

Considering that TME’s enrichment in immune cells at each step of tumor evolution and adaptation to immune selection pressures is translated at the level of the secreted secretome as a result of an evolution of tumor immunogenicity and growth of resistant tumor variants [25], we hypothesized that analyzing the plasma secretomes found in newly diagnosed and relapsing NPC patients with machine learning methods, would help in identifying predictive immunological signatures of disease relapse and death. Therefore, we sought compared the secretomes found in newly diagnosed and relapsing NPC patients using cytokine multiplex analysis and data mining algorithms in order to identify new biomarkers of disease evolution and response to therapy.

Patients and methods

Patients and controls

A total of 56 patients and 27 healthy donors were enrolled in this study. All patients were diagnosed with undifferentiated type of non-keratinized nasopharyngeal carcinoma (UCNT) by the otolaryngology and anatomopathological service of Mustapha Pacha Hospital. All subjects gave their consent to treating physicians. This study was approved by the ethics committee of the National Agency for Research Development in Health (ATRSS). Clinical characteristics of participants are showed in Table 1. Following recruitment, the “untreated patients” either received chemotherapy alone (Chemo, n=27) or chemotherapy followed by radiotherapy (Chemo + RT, n=15). The duration of the chemotherapy protocol consisted of three cycles (Docetaxel at 75 mg/m2 (Day1); Cisplatin at 75 mg/m² (Day1) and 5FU at 1250 mg/m² (Day1 to Day 3)) administered every 3 weeks. The radiotherapy protocol consisted of a total of 60–70 Gy administered over 6 to 7 week (2 Gy per day, 5 days a week).

Table 1.

Clinical characteristics of NPC patients and healthy subjects

Characteristic		Healthy Donors	Untreated NPC	Relapsed NPC
Gender	Male	22 (81.48%)	33 (73.33%)	8 (72.72%)
	Female	5 (18.51%)	12 (26.66%)	3 (27.27%)
Age (Mean±SD)		38.89±11.81	46.07±14.75	43.36±5.37
T stage	T1	-	8	-
	T2	-	15	-
	T3	-	11	-
	T4	-	11	-
N stage	N0	-	7	-
	N1	-	11	-
	N2	-	13	-
	N3	-	14	-
Distant Metastasis	Liver	-	-	2
	Bone	-	-	8
	Liver+Stomach	-	-	1
Stage disease	I	-	0	-
	II	-	6	-
	III	-	14	-
	IVa	-	11	-
	IVb	-	14	-
Followed treatment after blood sampling	Chemotherapy		30
	Chemotherapy + Radiotherapy		15

Methods

Plasma collection

Blood was collected in heparinized tubes from subjects and centrifuged at 1.500 rpm for 15 minutes at +4°C. Plasma was stored at −20°C until analysis.

Cytokines measurements

Plasmatic levels of cytokines, chemokines and growth factors (IFN-γ, IL-10, IL-12p40, IL-12p70, IL-1β, IL-4, IL-13, IL-17A, IL-6, IL-8 (CXCL8), IP-10 (CXCL10), MCP-1 (CCL2), TNF-α et VEGF) were measured by multiplexed immunobead based assay method using MILLIPLEX kit (Cat.# HCYTOMAG-60K, EMD Millipore, USA). The minimum detectable concentrations (Min. DC) for each cytokine and their percentage of detection inside the cohort is showed in Table 2. Cytokines with values below the Min. DC. were replaced by zero for statistical analysis.

Table 2.

Analysis of plasmatic nitrites and cytokines levels according to clinical parameters in untreated NPC patients (n = 45).

	Gender		Age (years)		T classification		N classification		Clinical staging
	Male (n=33)	Female (n=12)	<48 (n=22)	≥48 (n=23)	T1+T2 (n=23)	T3+T4 (n=22)	NO (n=7)	N1+N2+N3 (n=38)	I+II (n=6)	III+IV (n=39)
IFN-γ	50.96 ± 60.03	29 ± 36.47	49.13 ± 57.64	41.25 ± 53.81	57.52 ± 65.43	32.13 ± 39.48	34.51 ± 53.14	47.06 ± 56.06	40.17 ± 68	45.86 ± 54.02
11–10	1.61 ± 2.90	0.3915 ± 1.356	0.705 ± 1.44	1.833 ± 3.341	1.89 ± 3.28	0.65 ± 1.525	1.32 ± 3.5	1.275 ± 2.494	2.32 ± 3.87	1.121 ± 2.410
IL-12p40	130.3 ± 729.7	2.52 ± 8.74	193.3 ± 893.8	3.315 ± 7.30	187.3 ± 873.6	0.925 ± 3.081	1.78 ± 4.70	113.6 ± 680	7.12 ± 12.38	109.9 ± 671.3
IL-12p70	14.83 ± 28.24	1.33 ± 1.73*	5.76 ± 8.79	16.46 ± 33.19	16.71 ± 31.40	5.504 ± 13.89	4.28 ± 3.52	12.51 ± 26.85	14.68 ± 30.05	10.70 ± 24.36
IL-13	2.68 ± 7.86	90.41 ± 281.3	49.09 ± 208.8	4.061 ± 9.196	3.84 ± 9.23	49.32 ± 208.7	140.4 ± 369.8	5.011 ± 17.38	6.40 ± 13.90	29.10 ± 157
IL-17A	39.68 ± 58.07	50.03 ± 84.75	46.16 ± 65.2	38.89 ± 66.74	34.25 ± 33.52	51.01 ± 87.29	36.06 ± 48.37	43.62 ± 68.48	27.86 ± 41.19	44.69 ± 68.43
IL-1ϐ	0.23 ± 0.828	0.63 ± 1.47	0.23 ± 0.92	0.4177 ± 1.147	0.318 ± 0.982	0.34 ± 1.11	0	0.3888 ± 1.116	0.708 ± 1.73	0.27 ± 0.903
IL-4	9.18 ± 27.59	0	4.462 ± 16.34	8.9 ± 29.6	12.50 ± 32.55	0.70 ± 3.276	2.19 ± 5.81	7.566 ± 25.85	2.70 ± 6.63	7.35 ± 25.53
11–6	7.28 ± 15.46	3.21 ± 6.61	3.60 ± 3.47	8.681 ± 18.75	7.38 ± 15.73	4.96 ± 11.48	1.68 ± 1.99	7.028 ± 14.78	3.99 ± 5.01	6.53 ± 14.62
11–8	88.5 ± 259.4	161.5 ± 333.5	117.4 ± 320.4	98.88 ± 239.6	109.9 ± 313	105.9 ± 245.7	45.53 ± 75.52	119.5 ± 301.2	267 ± 603.5	83.49 ± 193.8
IP-10	138 ± 139	877.8 ± 361.4	868.9 ± 363.7	1608 ± 1603*	1031 ± 478.6	1473 ± 1665	887.5 ± 405.9	1313 ± 1309	1059 ± 529.6	1276 ± 1296
MCP-1	573.7 ± 329.1	437.8 ± 231.4	474.8 ± 148.9	597.4 ± 403.5	478.3 ± 186.2	599.3 ± 395.8	328.1 ± 93.34	576 ± 320.5*	374.1 ± 111.5	562.6 ± 323.2
TNF-α	24.76 ± 32.46	32.39 ± 52.32	29.44 ± 46.58	24.27 ± 28.98	30.07 ± 45.25	23.38 ± 29.91	16.63 ± 17.04	28.67 ± 40.85	40.90 ± 69.78	24.63 ± 31.87
VEGF	415.4 ± 516.1	319.1 ± 378	396.8 ± 481.7	383 ± 491	486.1 ± 591.7	288.9 ± 311.6	220.8 ± 147.4	420.8 ± 515	261 ± 217.5	409.5 ± 508.5
NO	31.16 ± 14.66	25.03 ± 9.861	30.67 ± 16.39	28.43 ± 10.83	30.01 ± 15.58	29.02 ± 11.79	32.54 ± 12.42	28.97 ± 14.02	20.92 ± 11.37	30.8 ± 13.68

Abbreviations: NPC, nasopharyngeal carcinoma; IL, interleukin; MCP-1, monocyte chemoattractant protein-1; IP-10, interferon protein-10, TNF-α, tumor necrosis factor- α; IFN-γ, interferon- γ; VEGF, vascular endothelial growth factor; NO, nitric oxide. Asterisks indicate a p value of c 0.05. The statistical analysis between groups was conducted using Mann-Whitney (Mean ± SD).

Nitrites measurement

Nitric oxide levels in the plasma was measured indirectly through its stable metabolite, nitrites by modified Griess reaction.

Statistical study

The obtained results are shown as Mean±SD, median or percentage. The statistical tests were applied according to Agostino and Pearson omnibus normality test. The statistical analysis between different groups was conducted using Kruskal-Wallis with Dunn’s test for multiple comparisons or ordinary one-way ANOVA with Tukey’s test for multiple comparisons. The comparison between two groups was conducted using Mann-Whitney. Spearman or Pearson coefficients were calculated to compare correlations between the analyzed cytokines and nitrites. The Kaplan-Meier test was used to estimate recurrence-free survival (RFS) up to 53 months of follow up for nitrites and cytokines with medians above 0. The log-rank test was used to calculate p values. The RFS was defined as the time from diagnosis to locoregional relapse, distant metastasis or death from any cause. The Cox proportional hazard regression model was used for univariate and multivariate analysis of cytokines, nitrites and clinical parameters regarding the RFS of untreated NPC patients. The backward selection was applied to keep only significant variables in the model. All variables were considered as binary and categorized as “Low” and “High” depending on the median value when appropriate. The CART (Classification and Regression Tree) decision tree was calculated using C&RT method to define the inflammatory mediators’ signatures predicting the worst and the best disease outcomes. All statistical analyses were performed with the Prism GraphPad version 6.01 or XLSTAT version 2016. For all analysis, a two-tailed p value < 0.05 was considered significant.

RESULTS

The profiles of the cytokine secretome at the time of tumor diagnosis and relapse are distinct

We and other described the importance of several inflammatory cytokines which support NPC development like TNF-α or IL-6 [20, 22, 26]. To identify new biomarkers in NPC pathogenesis, we firstly proceeded with a comparative analysis of the cytokine secretomes found in 27 healthy donors, 45 NPC patients prior to any treatment, and 11 patients at time of relapse. The multiplex comparison of 14 plasmatic cytokines was performed and compared to NO variations (Fig. 1).

Fig. 1. Secretome analysis.

Plasmatic nitrites and cytokines levels among controls (n=27), untreated (n=45) and relapsed (n=11) NPC patients were compared. The Horizontal lines show the mean value of the groups. The statistical analysis between the different groups was conducted using Kruskal-Wallis with Dunn’s test for multiple comparisons or ordinary one-way ANOVA with Tukey’s test for multiple comparisons according to the normality test analysis. (*=p<.05), (**=p<.01), (***=p<.001) and (****=p<.0001). Cytokines concentration unit (pg/ml) and nitrites (µM)

Means levels comparisons identified significant variations among 4 (IL-12p70, IL-8, IL-10 and IP-10) out of 14 analyzed cytokines in NPC patients compared to controls. As shown in Figure 1, the untreated NPC group compared to healthy donors group was characterized with increased levels of IL-10 (1.28±2.62 vs 0 pg/ml, p=0.02), IL-12p70 (11.23±24.84 vs 3.30±6.54 pg/ml, p=0.04), IL-8 (108.0±278.9 vs 11.76±8.65 pg/ml, p=0.0005) and IP-10 (1247.00±1220.00 vs 437.6±209.5 pg/ml, p<0.0001). Remarkably, IL-6 levels variations did not reach statistical significance compared to the control, despite being increased in the untreated group (p=0.06) (Fig. 1).

At the exception of IL-17A for which the plasmatic levels were decreased in the “relapsed NPC” group compared to controls (7.77±12.04 vs 25.06±25.21 pg/ml, p=0.04), the levels of IL-10 (4.55±7.94 vs 0 pg/ml, p=0.01), IL-6 (18.45±40.19 vs 2.60±6.05 pg/ml, p=0.02) and IP-10 (1694.00±1082.00 vs 437.6±209.5 pg/ml, p<0.0001) were significantly higher (Fig 1). We noted with interest presence of a major diminution of plasmatic IL-17A levels (7.77±12.04 vs 42.44±65.35, p=0.006) in relapsing compared to untreated patients. An opposite trend was found for MCP-1 (749.5±299.2 vs 537.5±309.6, p=0.04) (Fig. 1). No statistical differences were observed between the analyzed groups for TNF-α, VEGF, IL-12p40, IL-13, IL-1β and IL-4.

Further analysis of patients’ plasma at the time of tumor detection showed an overall increase in the percentage of individuals expressing detectable levels of the majority of the analyzed set of cytokines, in particular: IFN-γ, IL-10, IL12p70, IL17A, IL4 and IL-6. Interestingly, and despite sharing some similarities with the untreated group, a loss of IL-12p40, IL13, IL-1β and IL-4 expression was observed in the relapsing patients (Fig 2A).

Fig. 2.

Compared rates of individuals with detectable plasmatic cytokine levels in healthy, untreated and relapsed NPC patients (2A). Abbreviations: NPC, nasopharyngeal carcinoma; HD, healthy donors;; IL, interleukin; MCP-1, monocyte chemoattractant protein-1; IP-10, interferon protein-10, TNF-α, tumor necrosis factor- α; IFN-γ, interferon- γ; VEGF, vascular endothelial growth factor. MinDC: minimal detectable concentration (cutoff); (Red : Increase; pink : intermediary increase; Green : Decrease). (2B) Evaluation of the immune secretome associating with untreated NPC and NPC relapse. Population structure analysis was performed by PCA in untreated n=45 and relapsed NPC n=11 compared to controls (n=27).

A principal component analysis of the whole dataset recapitulated 40.27% of the total variance and showed presence of differences among groups’ distribution. We found that whereas the untreated patients cluster tended to be distributed along the F1 dimension (27.78% of total variance), with a major contribution of IL-8 and IL12-p70 to the variance, the relapsed patients cluster tended to be distributed vertically along the F2 dimension with a major contribution of MCP-1, IP-10 and nitrites to the variance (12.49% of total variance) (Fig 2B).

We concluded from these observations that the composition of the cytokine secretomes at the time of tumor diagnosis and relapse were distinct.

Specific cytokine network interactions support primary tumor development and tumor recurrence

As cytokines interactions forms a comprehensive network that is responsible for tumor progression, we next, explored for possible associations prevalent between the analyzed set of cytokines and chemokines at each step of disease evolution (i.e., at time of diagnosis and relapse); For this purpose, we performed a multiple correlation analysis for the prevalent cytokines found in each analyzed group of individuals.

Since the coefficient of correlation is influenced by the sample size, cytokines with less than 25% of values above the lower limit of detection were not taken into consideration for this analysis; thereby we excluded IL-12p40, IL-4, IL-13 and IL-1β from the matrix of analysis for the appropriate groups [27].

A correlogram representation of the cytokines associations for each analyzed secretome showed prevalence of important correlation variations with disease progression; for instance, group specific correlations were found between IL8/IL-17A and IFN-γ/IL12p70 in the control group, between TNF-α/IL-8/IL-6, IFN-γ/VEGF/TNF-α and IL-10/MCP-1 in the untreated group, and between IL-8/IL-6/IL-10, TNF-α/IL-8/IL-6, IL12-p70/VEGF/IL-10/IFN-γ, IL-6/IL-10/IFN-γ and IL-8/IP-10 in the relapse group (Fig. 3).

Fig. 3. Patients’ group specific secretome correlation analysis.

Correlograms of plasmatic cytokines in controls (3A), untreated patients (3B) and relapsed NPC (3C) groups. A moderate to strong positive correlations were observed between cytokines in NPC patient groups (untreated n=45 and relapsed n=11) compared to controls (n=27) with a distinct inflammatory, immunomodulatory and angiogenic profile. A Spearman or Pearson’s test coefficient were used according to the normality test analysis. • p<.05. (3D) Population variations comparison of cytokines’ interactions network.

We observed also that some correlations which were prevalent under physiological conditions, like those associating IL-8/IL-17A, IL-8/IFN-γ and IFN-γ/IL12-p70 were lost in the untreated group (Fig. 3). Interestingly some correlations like those associating IL-8/IFN-γ and IFN-γ/IL12-p70 were recovered under tumor recurrence.

A moderate positive correlation was observed between nitrites levels and IL-8 (r=0.63, p=0.035) in relapsed NPC (Supplementary Fig. 1).

We concluded from this analysis that specific cytokine network interactions prevailed under each step of NPC evolution to support primary specifically tumor development and tumor recurrence.

IL-12p70, IP-10 and MCP-1 expression associate respectively with gender, age and node metastasis in NPC

To determine whether the secretome observed in the enrolled patients might be linked to clinical parameters, the plasmatic concentrations of the analyzed cytokines as well as of nitrites were confronted to gender, age, clinical staging and TNM classification, (Table 2).

Plasmatic IL-12p70 levels were found to be significantly higher in male compared to female patients (14.83±28.24 vs 1.33±1.72 pg/ml, p=0.012) and IP-10 levels observed in patients older than 48 years were higher compared to those under 48 (1608±1603 vs 868.9±363.7 pg/ml, p=0.038). Interestingly, patients with loco-regional metastasis displayed significantly higher levels of plasmatic MCP-1 compared to metastasis free patients (576±320.5 vs 328.1±93.34 pg/ml, p=0.014). Nitrites, as well as the other tested cytokines did not show any significant association with the considered clinical and pathological features (Table 2).

Patients expressing High IL-8 / Low nitrites plasmatic signatures associated with poor prognosis

To evaluate the impact of the variations of cytokines and nitrites levels on the clinical outcome of the enrolled patients, we next performed a Kaplan-Meier survival analysis for recurrence-free survival. Survival was censored at 53 months of follow-up. The median values of the explored plasmatic biomarkers were taken as cutoff points to establish the “Low/High” stratification. Cytokines which displayed medians equal to zero (IL-10, IL-12p40, IL-1β, IL-4 and IL-13) were classified as either detectable (values above 0) or undetectable (values equal to 0).

A Kaplan-Meier curve analysis showed that among the nine plasmatic cytokines meeting the “Low/High” stratification, only IL-8 presented a prognostic value in which the High IL-8 levels were associated with poor prognosis. In Fig. 4A, we observed that after 53 months of follow-up, more than 66% of patients with High IL-8 levels met recurrence or death versus 28% of patients with Low IL-8 (p=0.012). No significant differences were observed for RFS in NPC patients for cytokines meeting the Detectable/Undetectable stratification (Data not shown).

Fig. 4.

Recurrence-free survival (RFS) estimation in untreated NPC patients (n=45) according to the median (low vs high) of plasmatic nitrites and the nine cytokines levels. (A) Only IL-8 and nitrites showed a statistically significant difference with high IL-8 associated to poor prognosis at the opposite of nitrites. (B) When combined; high IL-8 and low nitrites levels showed a statistically significant difference associated to poor prognosis. p values were calculated using the log-rank test.

In agreement with our previous results analyzing the impact of NO expression on RFS in a small cohort of patients (n=17, p=0.02; follow-up: 36 months) [20], we observed that when, 85% of enrolled patients presenting High nitrites levels survived after 23 months of first diagnosis, 50% of those with Low nitrites succumbed to death or showed signs of recurrence. This result confirmed that Low nitrites levels were significantly linked to worse prognosis (n=45; p=0.016; follow-up 53 months) (Fig. 4A).

Next, we considered studying the combining impact of IL-8 and nitrites values on RFS (Recurrence-free survival). We observed that patients presenting “High IL-8/Low nitrites” immunological scores presented the worst survival profile with more than 80% relapses or death cases after 53 months of follow-up compared to other groups (High IL-8/High NO, Low IL-8/Low NO, Low IL-8/High NO) which recapitulated 23% to 44% relapse/death cases (p=0.0034) (Fig. 4B).

We also observed that when only 27% of patients of the Chemo + RT arm relapsed/died after 53 months of follow-up, 59% of patients of the chemotherapy arm relapsed/died (p=0.0299). This result is in agreement with the literature and indicate that patients receiving chemotherapy and radiotherapy (Chemo + RT) have better RFS compared to those receiving chemotherapy alone (Supplementary Fig. 2)[28]. After stratification of the patients by type of treatment, the prognostic values (RFS) of High/Low IL-8 and High/Low NO were improved in the Chemotherapy arm (p=0.0008 and p=0.0102, respectively) but not in the Chemo + RT arm (p=0.5213 and p=0.6687) (Supplementary Fig. 3).

IL-8 and the treatment type are independent prognostic factors of RFS in NPC

In order to determine if IL-8 and nitrites were independent prognostic factors, we next run, after model validation (Supplementary Tab. 1), univariate and multivariate Cox proportional hazard regression analyses (Table 3). The univariate analysis showed that while High IL-8 associated with an increased risk of relapse/death (HR=3.13; 95% CI 1.20–8.16; p=0.02), High NO and the Chemo + RT protocol were associated with a reduced risk of relapse/death (HR=0.34; 95% CI 0.14–0.86; p=0.024) and (HR=0.32; 95% CI 0.106–0.959; p=0.042). A multivariate analysis showed that High IL-8 was an independent factor for poor prognosis (HR=7.451; 95% CI 2.398–23.152; p=0.001) while the Chemo + RT treatment appeared to constitute an independent prognosis factor for survival (HR=0.232; 95% CI 0.072–0.749; p=0.015). In contrast, High NO failed to reach significance (HR=0.44; 95% CI 0.156–1.242; p=0.121) (Table 3).

Table 3:

Univariate and multivariate Cox proportional hazards regression analysis for RFS of NPC patients (n=45).

Variables	Univariate analysis HR (95% CI)	P value	Multivariate analysis HR (95% CI)	P value
Age	1.703 (0.695–4.173)	0.244
Gender	1.203 (0.461–3.136)	0.706
T staging	1.195 (0.497–2.872)	0.691
N staging	1.659 (0.385–7.152)	0.497
Clinical staging	0.948 (0.277–3.237)	0.932
IFN-y	1.273 (0.527–3.075)	0.591
IL-10	1.033 (0.375–2.844)	0.95
IL-12p40	0.812 (0.238–2.775)	0.74
IL-12p70	1.848 (0.754–4.529)	0.18
IL-13	0.309 (0.072–1.334)	0.115
IL-17A	1.009 (0.418–2.437)	0.984
IL-1β	0.35 (0.047–2.615)	0.306
IL-4	0.247 (0.033–1.848)	0.173
IL-6	0.854 (0.355–2.053)	0.724
IL-8	3.127 (1.198–8.159)	0.02	7.451 (2.398–23.152)	0.001
IP-10	1.707 (0.696–4.188)	0.243
MCP-1	1.425 (0.582–3.487)	0.439
TNF-α	1.752 (0.725–4.235)	0.213
VEGF	1.296 (0.537–3.13)	0.564
NO	0.343 (0.136–0.866)	0.024	0.44 (0.156–1.242)	0.121
Chemo+RT*	0.320 (0.106–0.959)	0.042	0.232 (0.072–0.749)	0.015

Abbreviations: NPC, nasopharyngeal carcinoma; HR, hazard ratio; CI, confidence interval; RFS, recurrence-free survival; IL, interleukin; MCP-1, monocyte chemoattractant protein-1; IP-10, interferon protein-10, TNF-α, tumor necrosis factor- α; IFN- γ, interferon- γ; VEGF, vascular endothelial growth factor.

^*:Missing data for 3 patients in the Chemotherapy arm have been replaced by estimates of their values based on the mean/mode estimation test.

C&RT analysis identifies High IL-8/Low NO association as best predictor for therapy failure

Finally, we sought to determine the conditions of prediction of poor prognosis/treatment failure using a Classification and regression tree analysis (C&RT). Interestingly, the model was able to confirm that High IL-8/Low NO association was the best predictor for higher risk relapse/death (10/20, 50% cases). The model also identified Low IL-8/Low MCP1 association as the best determinant for response to therapy (13/25, 52%) at 53 months (86% purity) (Fig. 4A). The area below ROC curve (AUC) indicated an accuracy of 1 (Fig. 4B). A similar analysis at different time points, showed that High IL-8/Low NO association was the best predictor for higher risk of relapse as early as 36 months of follow-up (50% relapse, 75% purity) (Data not shown). The model applied to patients receiving chemotherapy alone as it associated High IL-8/Low NO association with higher risk relapse/death (8/14, 57% cases, 88.9% purity, AUC = 1) starting at 36 months of follow-up (Data not shown).

Discussion

Cytokines and nitric oxide play key functions in cancer immunity and tumor development; as such their role as cancer biomarkers has drawn extensive interest in recent years [29, 30]. In the current study we made a profile analysis of circulatory cytokines present in NPC patients in order to identify specific cytokines signatures which may be linked to a possible prognostic function. Our analysis included 45 untreated and 11 relapsed NPC patients presenting bone, liver or gastric metastasis.

A multiplex analysis of the evolution of the expression of cytokines in advanced NPC indicated a significant upregulation of plasmatic IL-12p70, IL-8, IL-10 and IP-10 levels at time of diagnosis. This first multiplex analysis performed on a North African population is in agreement with the findings reported for 132 Taiwanese, 37 Eastern Chinese and 39 Southern Chinese suffering UCNT [26, 31, 32]. Since IL-6 reached significance in the 3 studies performed on the Southeastern Chinese populations, we interpreted the close statistical significance found for this cytokine in our cohort (p=0.06) as determinant. We interpreted these similarities as a major indicator of a possible involvement of this group of cytokines in disease onset.

We suggest that the differences observed in our ability to detect significant upregulation of IL-1β, IL-4, IL-13, IL-17A, TNF-α, IFN-γ, MCP-1 and VEGF in our population as either as an effect due to the sensitivity of the used assays for the studied populations or a statistical effect due to intrinsic elements associated with the analyzed cohorts like the men: women ratio or the structure of NPC stage pyramid.

We determined following a PCA analysis, that the plasma secretomes found in newly diagnosed individuals was distinct from that found in relapsing NPC patients. This is indicated by a major elevation in IL-8 and IL12-p70 synthesis in untreated patients and of MCP-1, IP-10 and nitrites (NO) in individuals suffering tumor relapse. We interpreted these results as indicative of a major inflammatory and neoangiogenic cytokinic activity in untreated patients, and of an immunosuppressive and chemoattractive one in favor of metastatic extensions in patients with relapse. Interestingly, IL-17 expression associated with a significant reduction of its synthesis in the relapsing phase. This is evocative of a remodeling of the immunological secretome generated under the condition of the primary tumor and its relapsing variants development phases.

In agreement with our findings, several other studies, using single cytokine analysis, demonstrated increase in major cytokines expression (IL-6, IL-8, IL-10, VEGF, TNF-α and TGF-β) supporting tumor development in NPC [20, 22, 33–39]. Despite the divisive observations reported on IL-10 and TGF-β by Tan et al. [40], these results are in line with the trend of a development of a Th1/Th17 response among the analyzed patients.

Intriguingly, even though a similar observation on IL-12p70 upregulation was reported in Asian patients, no clear demonstration of a possible function of IL-12p70 in NPC control has been proposed to date [31, 32, 41].

Considering the role of IL-12p70 in Th1 polarization, IFN-γ synthesis and anti-tumor immunity, we observed that despite its high concentrations, it less correlated with IFN-γ synthesis in patients compared to controls. We hypothesize that, in presence of high concentrations of IL-17, IL-23 would antagonize IL-12 secretion to limit IFN-γ synthesis [42]. The finding showing that patients at relapse show a reduction in IL-17A synthesis and a greater association linking the IL12 to IFN-γ axis is in our opinion supportive of this hypothesis. As the reduction of IL-17A/IL-8 interaction may reflect a loss of influence of IL-17A in advanced/relapsing tumors, further investigations should explore the cellular origin(s) of the cytokine and the contribution of the Th17 response in NPC pathophysiology.

As most of our patients presented a stage III/IV disease at the time of diagnosis, we also hypothesize that the anti-tumoral activity of IL-12p70 in untreated and relapse patients might be compromised by dominant immune escape processes present in advanced stages of carcinogenesis like those involving PDL-1 blocking NK activity[43, 44]. A possible suppression of IL12p70 anti-angiogenic function in metastatic patients is also suggested by the enhanced IL12p70/VEGF correlation. Taken together, these arguments evoke a possible disruption of IL12p70 function in NPC by concurrent immune mechanisms.

Remarkably, we observed that most of IL12p70 upregulation occurred in men during tumor onset. This element confronted to the fact that men are 3 times at higher risk of developing NPC and the observation by Wilcoxen at al. of a gender-dependent IL-12 secretion by antigen presenting cells (APC), poses the challenging question of a possible influence of gonadal hormones on APC activity in NPC pathogenesis in presence of IL-10 [45]. Although not observable in our study, a trend towards a higher IL-10 synthesis in males would possibly be implicated in a gender associated susceptibility to a stronger immunosuppressive response.

While IL-10 mostly associated with IL-12p70 in relapsing patients, the cytokine was found to be increasingly associated with VEGF in metastatic patients; this result points toward a possible gradual implication of the molecule in angiogenesis. This hypothesis is in line with a potential cooperation of IL-10/VEGF to promote pro-tumoral tumor associated macrophages (TAM) under the proposed model of Dace D.S et al. of a proangiognencic function of IL-10 under hypoxia; a condition prevalent in UCNT[46, 47].

The finding that IL-10 upregulation also occurred in presence of elevated IFN-γ induced protein (IP-10) levels in untreated and relapsed NPC patients is also in favor of the idea of an installation of immunosuppressive environment favoring tumor promotion. Remarkably, this feature was markedly found to be associated with relapse by PCA analysis.

Besides the possibility that IP-10 upregulation might be also involved in Th1/CTL and Treg recruitment, and IL-8 inhibition [48, 49], the hypothesis of a contribution of IP-10 in NPC development is strengthened by the observation that its lower expression levels associated with the best survival profile particularly when in association with Low IL-8 and Low MCP-1 levels.

Considering that high IP-10 expression may contribute to metastasis in melanoma [50] and our data on the presence of a unique correlation associating IP-10 and IL-8 in relapsing tumors, suggests that IP-10 would possibly have an influence on tumor progression. Our Kaplan-Meier analysis using cytokine stratification is in agreement with that view. Therefore, future studies are necessary to bring more insight on 1) the influence of the chemokine on NPC development and recurrence, and 2) on NPC susceptibility in aging individuals which display increased IP-10 levels [51, 52].

In our cohort, significantly higher MCP-1 levels associated with nodal and mostly bone metastasis. This observation is in agreement with the function of this chemokine in tumor spreading and the findings showing that increased serum level of MCP-1 associate with bone metastasis [31, 33, 53]. Considering that MCP-1 is involved in monocyte attraction as well as M2-TAM polarization to promote tumor progression, it will be interesting to further clarify the impact of MCP-1 expression on M1/M2 polarization and relapse in NPC[54–56].

Previously, we reported that high NO amounts in plasma of untreated NPC patients were negatively associated with IL-6 levels [20]. In this study, we confirmed our previous findings regarding the implication of NO in NPC pathogenesis, but could not determine any significant associations with cytokines expression beside that linking NO to IL-8 in relapsed patients. Still, the increased correlations associating IL-6 and TNF-α to IL-8 indicated that this network of molecules would act as inflammatory components of NPC.

Interestingly, IL-6 showed to contribute more extensively to tumor metastasis and therapeutic failure via an augmented contribution to a distinct cytokine network formed by IL-8, IFN-γ and IL-10. Of note we previously reported that high IL-6 levels contributed to metalloprotease 9 (MMP-9) activation in association with low NO levels [20]. In agreement with these findings, we confirmed in this study the beneficial effect of the highest NO concentrations on patients’ survival by determining the hazard risk value associated with NO concentrations in NPC [20, 57]. The Kaplan Meier analysis confirmed that both NO and IL-8 could be considered as prognostic factors and showed that the association of these two biomarkers can enable for better prediction of NPC recurrence-free survival when the treatment type was not taken into account.

The predictive model using C&RT analysis consolidated our findings that the patients with High IL-8/Low NO immunological signature had the worst profile with 50% of relapse after 53 months of follow-up; conversely to those presenting Low IL-8/Low MCP-1. This finding draws our attention to the role of these chemokines in disease severity and death.

As our results confirmed the advantage of combining chemotherapy to radiotherapy to improve patients’ survival et avoid tumor relapse/death, we further confronted our model the treatment type. The analysis of the prediction model in association with the type of treatment showed by multivariate analysis that IL-8, but not NO, was an independent prognostic factor. IL-8 and NO prognostic values remained significant for the group of patients receiving Chemotherapy alone. This result indicates that radiotherapy activity may have potently eliminated the tumor and the influence of the inflammatory markers on tumor development. Of note, Merhi et al., showed that the circulatory levels of IL-8 and IL-6 dropped down after HNSCC treatment with an anti-PD1 but increased at progression [58]. Further analyses are therefore necessary to evaluate the expression of the cytokines and NO in surviving patients and those in relapse to assess the influence of the treatments.

Our findings on plasmatic IL-8 association with disease severity finds also echo in Cheng et al. study on 99 Chinese patients in which augmented IL-8 associated with overall survival [34] and in a recent demonstration by Yang et al., on epithelial ovarian cancer, where IL-8 and TAN levels were described as drivers of immune evasion and poor clinical outcome [59–61]. This points towards a possible contribution of tumor associated neutrophils (TAN) in NPC immunopathology which should be explored [62]. In this context, the direct contribution of EBV oncogene LMP-1 to IL-8 production in the TME should be clarified too [60, 61]. Consequently, pairing tissue and plasma secretomes should be performed in the future to improve our understanding of the complexity of the cytokines’ interactions prevailing in patients experiencing therapeutic failure and recurrence. In order to support our conclusions, extended cohorts of patients should be recruited to refine our predictive model and overcome the limitations met in cytokines’ detection levels. More sensitive tests should also help us in defining the contribution of Th2 markers (IL-4, IL-13) in the system.

In conclusion, this study provides demonstration that IL-8/NO combined signatures should be considered as predictive biomarkers of death/relapse in clinics. This study brings to the best of our knowledge, the first rational published arguments for the national health authorities to consider policies to enforce access to radiotherapy and decrease mortality related to NPC. Our results provide evidence that cytokines monitoring may help in improving disease management by early identification of individuals at highest risk of metastasis. The predictive value of the identified cytokines may help not only in determining which patients are at risk of therapeutic failure but also help in optimizing the selection of molecular targets to disrupt the pro-tumoral network and help in designing novel therapeutic interventions built on cytokine levels.

Supplementary Material

1

S. Fig. 1 Correlation between plasmatic nitrites and IL-8 levels among relapsed patients group (n=11). A statistically significant correlation was observed between nitrites and IL-8. Pearson’s test coefficient was used according to the normality test analysis.

S Tab. 2 Model validation by regression test with backward elimination. Only IL-13, IL-6, IL-8, type of treatment and nitrites were validated for multivariate COX analysis.

S. Fig. 2 Recurrence-free survival (RFS) estimation in untreated NPC patients (n=42) according to the type of treatment. Patients undergoing treatment with chemotherapy alone had significant shorter recurrence-free survival compared to those combining chemotherapy with radiotherapy. p value was calculated using the log-rank test.

S. Fig. 3 Treatment impact on RFS according to nitrites and cytokines values. Recurrence-free survival (RFS) estimation in untreated NPC patients (n=45) was calculated according to the median (High vs Low) of plasmatic nitrites and cytokines levels by treatment type. (A) Chemotherapy alone. Significant differences are observed for IL-8 and nitrites for patents in chemo arm (n=30). (B) no statistical significance was observed IL-8 or nitrites in the Chemo + RT arm (n=15). C. Table of p value was calculated using the log-rank test. * Values for patients (n=3) missing for the Chemo arm were calculated by applying the estimation test.

Fig. 5.

The CART (Classification and Regression Tree) decision tree of cytokines and nitrites in untreated NPC patients (n=45) according to patients’ outcome (RFS). (A) The diagram showed the corresponding profile of the worst and the best outcome with 100% of purity. The worst one corresponds to high IL-8, low nitrites. The best profile corresponds to low IL-8, low MCP-1. 0=Bleu color, without relapses; 1=pink color, with relapses. (B) The ROC curve analysis of cytokines and nitrites in untreated NPC patients according to RFS. The area under the ROC curve is 1.

Highlights.

• The profiles of the cytokine secretome at the time of tumor diagnosis and relapse are distinct

• Distinct cytokine network interactions support primary tumor development and tumor recurrence

• IL-12p70, IP-10 and MCP-1 expression associate respectively with gender, age and node metastasis in NPC

• Patients with High IL-8 / Low nitrites plasmatic signatures associate with poor prognosis

• IL-8 and the treatment type are independent prognostic factors of RFS in NPC patients

• C&RT analysis identifies High IL-8/Low NO association as best predictor for therapy failure