Cytokines and angiogenic factors in patients with metastatic renal cell carcinoma treated with interferon-α: association of pretreatment serum levels with survival

Abstract

Background: To correlate serum cytokine and angiogenic factor (CAF) levels with overall survival (OS) in metastatic renal cell carcinoma (mRCC) treated with interferon-α (IFN-α).

Patients and methods: Serum CAF levels were measured in 103 patients treated on a randomized trial with IFN-α 0.5 million units (MU) twice daily or 5 MU daily. Concentrations of 17 analytes were determined by multiplex bead immunoassays [vascular endothelial growth factor A (VEGF_A) and several cytokines] or enzyme-linked immunosorbent assay (basic fibroblast growth factor). We used proportional hazards models to evaluate the effect of CAF levels and clinical factors on OS.

Results: Pretreatment serum interleukin (IL) 5, IL-12 p40, VEGF_A, and IL-6 levels and Memorial Sloan-Kettering Cancer Center risk grouping independently correlated with OS, with hazard ratios of 2.33, 2.00, 2.07, 1.82, and 0.39, respectively (concordance index = 0.69 for the combined model versus 0.60 for the CAF model versus 0.52 for the clinical model). Based on an index derived from these five risk factors (RFs), patients with 0–2 RF had a median OS time of 32 months versus 9 months for patients with 3–5 RF (P < 0.0001).

Conclusions: Serum CAF profiling contributes to prognostic evaluation in mRCC and helps to identify a subset of patients with 20% 5-year OS.

introduction

Metastatic renal cell carcinoma (mRCC) is well recognized to have a broad spectrum of clinical expression [1]. Pretreatment clinical features correlating with survival in the context of interferon-α (IFN-α) therapy have been identified by investigators at Memorial Sloan-Kettering Cancer Center (MSKCC) [2]. A similar clinical model, based on IFN-α or interleukin (IL) 2 therapies, validated the MSKCC factors and added ‘number of metastatic sites’ as an independent risk factor for survival [3]. One approach to refine prognostic models relies on biologically based factors.

Numerous reports have indicated that cancer exerts direct and profound immune suppression via a shift toward a humoral T-helper type 2 (Th2) response, leading to accumulation of increased numbers of immature myeloid-derived suppressor cells [4, 5] or regulatory T cells [6], decreasing numbers of functional dendritic cells (DCs) [7], and increasing apoptosis of T cells by tumor-derived gangliosides [8]. CD4+ T-helper (Th) cells are considered a double-edged sword because of their ability to promote or inhibit tumor growth [9, 10]. A Th1 response is associated with cytotoxic T-lymphocyte activation, is considered beneficial for antitumor immunity, and plays an important role in tumor rejection [11, 12]. By contrast, a Th2 response is associated with the metastasis microenvironment [13–15]. A subset of CD4+ cells known as regulatory T cells can be found in tumors where they inhibit effector T-cell proliferation, i.e. antagonize a Th1 response [6]. IFN-α exerts an antitumor effect indirectly by activating DCs, increasing DC cross-priming, costimulating lymphocytes, upregulating major histocompatibility complex molecules on the surface of tumor cells, or shifting the immune response to a Th1 response [12, 16].

The antitumor efficacy of twice-daily low-dose IFN-α was demonstrated in preclinical studies showing potent antiangiogenic effects [17]. We previously published the results of a randomized clinical trial in mRCC using IFN-α 0.5 million units (MU) twice daily or 5 MU daily and observed no significant differences in progression-free survival, response rate, and overall survival (OS) between the two arms [18]. However, we noted long-term survival in a few patients. In a companion study of the same patient population, we measured pretreatment serum levels of cytokine and angiogenic factors (CAFs) that are plausibly related to the mechanisms of action of IFN-α and report here the results.

patients and methods

Between 27 March 2002 and 19 December 2003, 118 patients with mRCC were randomly assigned to receive subcutaneous IFN-α2b 0.5 MU twice daily or 5 MU daily. The trial was approved by the University of Texas, MD Anderson Cancer Center Institutional Review Board. Before being randomly assigned, patients were stratified according to MSKCC risk model [2]. Key eligibility criteria were previously described [18].

CAF quantitation

Before initiation of therapy, ∼10 ml of blood was collected in a serum separator tube with clot activator. Within 1 h after collection, blood samples were centrifuged at 1200 g for 20 min at 4°C. Serum was divided into aliquots and stored at −80°C until batchwise analysis. Levels of vascular endothelial growth factor A (VEGF_A); tumor necrosis factor-α; IFN-α; IFN-γ; granulocyte–macrophage colony-stimulating factor; and several interleukins (ILs), including IL-1, IL-2, IL-4, IL-5, IL-10, IL-12 p40, IL-6, IL-8, IL-13, IL-15, and IL-17, were quantified from sera using the Multiplex Bead Immunoassay™ (BioSource International, Inc., Camarillo, CA). The assay is a solid-phase sandwich quantitative enzyme-linked immunosorbent assay (ELISA) that utilizes beads of defined spectral properties conjugated to analyte-specific capture antibodies. Briefly, antibody-coated beads were mixed with 50 μl of serum in 96-well plates. Plates were incubated for 2 h to allow cytokines in the serum to bind to their cognate antibody-coated bead. After incubation, plates were washed and a mixture of biotinylated detector antibody was added to each well. Plates were incubated for 1 h and washed. After washing, streptavidin conjugated to R-phycoerythrin was added and plates were incubated for 30 min and washed. Both fluorescence and spectral properties of beads were monitored using a Luminex100™. CAF concentration was determined by grouping beads of equal spectral properties into bead regions and quantifying the fluorescence emitted by each region. Fluorescence values were used to calculate CAF concentration using a standard curve derived from a mixture of analytes of known concentration. Each serum sample was analyzed in triplicate and serum CAF concentrations were reported in picograms per milliliter.

Serum levels of basic fibroblast growth factor (bFGF) were measured using the commercially available ELISA kit Quantikine bFGF HS ELISA (R&D Systems, Minneapolis, MN). ELISA plates were read using the Fluostar Optima Microplate Reader (BMG Lab Technologies Inc., Durham, NC).

statistical analysis

The primary objective of this study was to determine whether pretreatment serum CAF levels correlate with OS and add to the information provided by clinical factors such as those used in the MSKCC prognostic model. We defined OS time as the interval from the first IFN-α dose to death from any cause or the date of last follow-up.

Pearson's chi-squared tests were used to test the association between baseline categorical variables and treatment groups [19]. Wilcoxon rank sum tests were applied to compare the difference of continuous variables between the two treatment groups [19]. Unadjusted probabilities of OS were estimated using the Kaplan–Meier method [19]. Unadjusted between-group comparisons of OS were made using the log rank test [19]. Because the CAF concentrations were all highly skewed, these were log transformed in analyses. We used recursive partitioning method and martingale residual plots to determine the optimal cut points and dichotomized the baseline CAF values [20]. To avoid potentially unstable correlations, we imposed a constraint that no group has <20 patients. Cox proportional hazards (PHs) models were used to estimate the effect of clinical factors and baseline CAF levels [21]. Stepwise selection methods were employed to carry out model selection and to construct the most parsimonious models. The significant level was set at 0.05. The Harrell's concordance index (c-index) was calculated to assess predictive accuracy [22]. All computations were carried out in SAS 9.1.3 (SAS Institute, Cary, NC) and Splus 7.0 (Insightful Corporation, Seattle, WA) [21].

results

Patient characteristics for all 103 patients that form the basis of this report are shown in Table 1. There were no significant differences between the two arms with regard to continuous variables such as age, serum lactate dehydrogenase (LDH), calcium and alkaline phosphatase levels, hemoglobin, white blood count, platelet count, and erythrocyte sedimentation rate (ESR). The median follow-up time for the 103 analyzable observations was 57 months (range 3–69). For all 103 patients, median OS time was 20 months [95% confidence intervals (CIs) 16–30]. Consistent with the MSKCC model, there were significant differences in median OS times among patients according to risk groups (Figure 1A), with median OS times of 33 months [95% CI 24 to not estimatable (NE)], 16 months (95% CI 14–28), 13 months (95% CI 6 to NE) for patients with favorable-risk, intermediate-risk, and poor-risk disease, respectively (P = 0.006). There were also significant differences in median OS times between patients with clear-cell and nonclear-cell histology (Figure 1B), with median OS times of 24 months (95% CI 17–33) and 9 months (95% CI 6–22), respectively (P = 0.01). None of the baseline routine laboratory parameters (hemoglobin, leukocyte count, platelet count, absolute neutrophil count, absolute lymphocyte count, absolute monocyte count, absolute eosinophil count, ESR, calcium, albumin, alkaline phosphatase, and LDH) correlated with OS.

Table 1.

Patient characteristics (N = 103)

	Low-dose IFN-α (N = 54)		Intermediate-dose IFN-α (N = 49)		P valuea
	N	%	N	%
Gender					0.02
Female	15	75	5	25
Male	39	47	44	53
Race					0.41
Black	3	75	1	25
Hispanic	6	40	9	60
White	45	54	39	46
MSKCC risk groups					0.52
Favorable	16	53	14	47
Intermediate	34	55	28	45
Poor	4	36	7	64
Liver					0.58
Yes	16	48	17	52
No	38	54	32	46
Lung					0.60
Yes	40	54	34	46
No	14	48	15	52
Bone					0.46
Yes	8	44	10	56
No	46	54	39	46
Renal fossa/adrenal					0.75
Yes	17	55	14	45
No	37	51	35	49
Mediastinum					0.25
Yes	20	61	13	39
No	34	49	36	51
Retroperitoneum					0.08
Yes	7	35	13	65
No	47	57	36	43
Prior nephrectomy					0.26
Yes	50	54	42	46
No	4	36	7	64
Prior radiation					0.90
Yes	3	50	3	50
No	51	53	46	47
Histology					0.12
Conventional	48	56	38	44
Others	6	50	11	50

^a P value for chi-square test.

IFN, interferon; MSKCC, Memorial Sloan-Kettering Cancer Center.

Figure 1.

Overall survival from registration (N = 103). (A) Stratified by Memorial Sloan-Kettering prognostic category. (B) Stratified by histology (clear cell versus non-clear cell).

Pretreatment serum CAF levels ranged from undetectable (the lower limit of reliable detection varies with analyte, but is ∼0.1 pg/ml) to ∼6,000 pg/ml, a dynamic range of nearly five orders of magnitude. The median value of each triplicate set was used in the correlative analysis of OS.

By univariable analysis, we identified several clinical factors that were significantly associated with OS, including MSKCC risk grouping, number of metastatic sites (used as a continuous variable), histology, and prior nephrectomy. A Cox PH model which initially included these four clinical factors identified all but prior nephrectomy as independently significant (Table 2). The final model based on the three clinical factors discussed above had a modest c-index (analogous to the area under a receiver-operator curve) of 0.52.

Table 2.

Multivariable Cox proportional hazards model for clinical factors and OS (N = 103)

Variable	HR	95% CI	P value
MSKCC risk group
Poor	Reference	–
Intermediate	0.51	0.26–1.00	0.05
Favorable	0.34	0.17–0.72	0.005
Number of disease sites	1.25	1.02–1.54	0.03
Histology
Other	Reference	–
Clear cell	0.53	0.30–0.99	0.03

CI, confidence interval; HR, hazard ratio; MSKCC, Memorial Sloan-Kettering Cancer Center; OS, overall survival.

Pretreatment serum CAF levels were evaluated for correlation with OS. For 13 out of 17 analytes, a significant correlation was found by univariable analysis. In every case, higher values were associated with inferior survival, with hazard ratios (HRs) ranging from 1.74 to 2.9 (Table 3). These 13 analytes were then taken as a starting point in a Cox PH model, which indicated independent significance for IL-5, IL-12 p40, IL-6, and VEGF_A, with HR ranging from 1.87 to 2.33.The c-index for this model was 0.6 (Table 4).

Table 3.

Univariable Cox proportional hazards model for pretreatment serum CAF levels and OS (N = 103)

Variable	Cut point	Median (range)	HR (95% CI)	P value
IL-1β	8.97	15.7 (0.1–1452)	1.37 (0.85–2.23)	0.201
IL-5	4.19	5.9 (2.3–104)	2.02 (1.25–3.26)	0.004
TNF-α	100.6	47.7 (0.1–6100)	2.04 (1.18–3.52)	0.010
GM-CSF	55.1	38.5 (22.2–400)	2.64 (1.55–4.47)	<0.001
IFN-α	78.8	55.2 (8.4–1674)	2.24 (1.32–3.82)	0.003
IFN-γ	23.8	19.5 (6.7–225)	2.41 (1.42–4.11)	0.001
IL-10	5.36	6.8 (0.1–53)	1.74 (1.06–2.87)	0.030
IL-12 p40	222.1	150.3 (10.2–1577)	2.25 (1.41–3.59)	0.001
IL-2	44.3	17.3 (1.8–2703)	1.44 (0.83–2.50)	0.196
IL-4	20.1	13.7 (4.6–5230)	2.90 (1.69–4.98)	<0.001
IL-6	29.66	21.6 (2.2–2764)	2.39 (1.51–3.77)	<0.001
IL-8	20.49	13.4 (1.05–97)	1.99 (1.16–3.42)	0.012
IL-13	22.2	16.9 (8.4–203)	2.02 (1.20–3.38)	0.008
IL-15	8.19	9.2 (0.4–4010)	0.74 (0.48–1.15)	0.186
IL-17	35.1	24.8 (0.1–3285)	2.69 (1.57–4.61)	<0.001
bFGF	0.96	0.7 (0.1–14)	0.73 (0.46–1.18)	0.205
VEGFA	98.0	0.3 (0.1–798)	1.95 (1.15–3.30)	0.013

bFGF, basic fibroblast growth factor; CAF, cytokine and angiogenic factor; CI, confidence interval; GM-CSF, granulocyte–macrophage colony-stimulating factor; HR, hazard ratio; IFN, interferon; IL, interleukin; MSKCC, Memorial Sloan-Kettering Cancer Center; OS, overall survival; TNF, tumor necrosis factor; VEGF_A, vascular endothelial growth factor A.

Table 4.

Comparison of CAF model to CAF + clinical factors model in all patients (N = 103)

CAF model (c-index = 0.60)			CAF + clinical factors model (c-index = 0.69)
Variable	HR (95% CI)	P value	Variable	HR (95% CI)	P value
IL-5	2.33 (1.42–3.83)	<0.001	IL-5	2.33 (1.43–3.82)	0.001
IL-12 p40	1.87 (1.14–3.07)	0.013	IL-12 p40	2.00 (1.22–3.28)	0.006
IL-6	2.11 (1.30–3.46)	0.003	IL-6	1.82 (1.11–2.97)	0.017
VEGFA	2.12 (1.24–3.64)	0.006	VEGFA	2.07 (1.19–3.60)	0.010
			MSKCC risk group: intermediate versus poor	0.59 (0.30–1.17)	0.129
			Favorable versus poor	0.39 (0.18–0.83)	0.015

c-index, Harrell's concordance index; CAF, cytokine and angiogenic factor; CI, confidence interval; HR, hazard ratio; IL, interleukin; MSKCC, Memorial Sloan-Kettering Cancer Center; VEGF_A, vascular endothelial growth factor A.

We then considered a Cox PH model combining the clinical factors from our previous model with pretreatment serum CAF levels starting with the four independently significant clinical factors (Table 2) and the four independently significant pretreatment serum CAF values (Table 4). All four of the serum CAF analytes remained significant, but only MSKCC risk grouping (favorable versus poor/intermediate) remained in the model from among the clinical factors. The c-index of this model was 0.69 (Table 4). Then, we constructed a simple index, in which we gave a score of one point for each factor that was unfavorable, such that each patient could be assigned a combined score from 0 to 5. Kaplan–Meier curves for OS according to this index appeared to cluster, with patients having zero to two adverse factors having similar OS times, and those with three to five adverse factors also having similar OS times (Figure 2A). Dichotomization of OS according to number of adverse factors in our final model revealed a significant separation of outcome (Figure 2B). For the favorable group, median OS time was 32 months (95% CI 26–56) versus 9 months (95% CI 7–15) for the unfavorable group (P < 0.0001).

Figure 2.

Overall survival from registration (N = 103). (A) Stratified by number of adverse factors in our final proportional hazard model. (B) Stratified by 0-2 versus 3-5 adverse factors.

Our population consisted of 86 patients with conventional-type (i.e. clear-cell) renal cell carcinoma (RCC) and 17 with nonclear-cell RCC. Because of differences in the biology, clinical behavior, and response to various therapies between these two categories of RCC, we analyzed our results in the 86-patient subset with conventional histology. Surprisingly, serum IL-6 and VEGF_A levels dropped out as independent factors in the multivariable model of pretreatment serum CAF levels and in the model combining pretreatment serum CAF levels and clinical factors (Table 5).

Table 5.

Comparison of CAF model to CAF + clinical factors model in patients with clear-cell RCC (N = 86)

CAF model			CAF + clinical factors model
Variable	HR (95% CI)	P value	Variable	HR (95% CI)	P value
IL-5	2.11 (1.24–3.59)	0.006	IL-5	2.16 (1.27–3.67)	0.005
IL-12 p40	2.63 (1.57–4.41)	<0.001	IL-12 p40	2.85 (1.67–4.85)	<0.001
			MSKCC risk group: favorable versus poor versus intermediate	0.51 (0.30–0.90)	0.019

CAF, cytokine and angiogenic factor; CI, confidence interval; HR, hazard ratio; IL, interleukin; MSKCC, Memorial Sloan-Kettering Cancer Center; RCC, renal cell carcinoma.

Seventy-five of the 103 patients in this study received second-line therapy after disease progression on IFN-α (Table 6); the majority of them received chemotherapy with gemcitabine plus capecitabine. Eight patients received IL-2-based therapy or adoptive immunotherapy. Only nine patients received VEGF-directed therapy, as the currently approved targeted agents were not available during the conduct of this trial. Fourteen patients received other second-line therapeutics. OS times for these four groups of patients, as measured from the date of initiation of IFN-α therapy to the date of death, were not significantly different (data not shown). In our view, because of the small numbers in each of these groups and the incomplete nature of the data on second-line therapy, it would not be appropriate to add this variable to the Cox model analysis.

Table 6.

Second-line systemic therapy received by patients after interferon-α

Therapy	No. of patients	%
Any second-line therapy known to investigators	75 (out of 103)	72.8
VEGF-targeted therapeuticsa	9	8.7
Capecitabine + gemcitabine chemotherapy	44	42.7
Immunotherapyb	8	7.7
Other therapeuticsc	14	13.6

^aSorafenib (three patients); bevacizumab + erlotinib (five patients); SU-5416 (one patient).

^bInterleukin 2 based (seven patients); stem cell transplantation (one patient).

^cThalidomide (12 patients); recombinant lactoferrin (two patients).

VEGF, vascular endothelial growth factor.

discussion

This report presents a pretreatment serum CAF profile for patients with mRCC who received first-line therapy with single-agent IFN-α. When we combined the CAF model (based on analysis of baseline serum CAF levels) with the clinical model (based on MSKCC risk grouping, histology, number of metastatic sites, prior nephrectomy) into one model, we were able to improve c-index from 0.52 with the clinical factors to 0.69 in the model that included baseline CAF levels.

While it may be useful to have more refined prognostic models, the biologic implications of our findings are perhaps of greater importance. It is interesting but not surprising that pretreatment serum levels of IL-5, IL-12, IL-6, and VEGF_A were all strongly prognostic, with elevated levels negatively correlating with OS. Elevated levels of VEGF_A and IL-6 have been previously demonstrated to have an adverse impact on prognosis in mRCC [23]. IL-5 plays an important role in recruiting eosinophils, and eosinophilia has been reported to have an adverse impact on prognosis in patients with mRCC. In our analysis, we did not find a correlation between elevated pretreatment serum IL-5 levels and eosinophilia, but only five patients in our study had an elevated baseline eosinophil count.

Clinical studies and experimental mouse models have expanded our understanding of the complex relationship between the immune system, angiogenesis, and tumor development [10, 24]. The presence of elevated systemic levels of proinflammatory cytokines may portend unfavorable clinical outcomes in patients with advanced solid tumors [25]. Although the complexity of cytokine networks makes it difficult to precisely determine the relationship between clinical outcome and systemic cytokine levels, our data provide evidence that pretreatment serum levels of proinflammatory cytokines such as IL-12 p40 negatively correlate with OS in patients with mRCC.

Although immune activation can lead to tumor destruction, it is now becoming clear that immune mediators can also contribute to tumor progression [24]. Cytokines generated during strong innate immune responses can enhance adaptive immune responses and tumor destruction. In some contexts, these cytokines can also contribute to tumor growth, through stimulation of proangiogenic cytokines [10, 24].

In this study, we identified a subset of patients with mRCC who had a median OS time of ∼3 years and a 20% 5-year OS (Figure 2B) in the context of IFN-α treatment. Whether our prognostic index identifies a generally favorable group or, alternatively, the baseline serum CAF profile identifies a subset of patients for whom IFN-α is an effective therapy is not yet settled. The hypothesis that patients with a favorable serum CAF profile do well with IFN-α therapy is certainly testable and should be done, especially that low-dose IFN-α therapy is very well tolerated.

In future studies, we plan to address the challenges of assay reproducibility and expand the use of baseline serum CAF levels to the treatment contexts of bevacizumab, tyrosine kinase inhibitors, mTOR inhibitors, and the regimen of capecitabine plus gemcitabine. We will also evaluate the kinetics of serum CAF level changes after IFN-α treatment, as this may provide additional prognostic and biologic insight.