Renal cell carcinoma (RCC) is the third most lethal urologic malignancy, with a high mortality rate equivalent to the death of 100,000 individuals per year all over the world.1 According to its morphologic features, RCC can be classified into various subtypes, and clear cell RCC (ccRCC) represents the most common type.2 After radical nephrectomy of RCC, the principal type of treatment, more than 25 % patients still experience recurrence with metastatic disease.3 Several clinical prognostic systems have been put into application to stratify patients with RCC into different risk group of disease progression. Besides the conventional TNM staging system, that of the Mayo Clinic, the stage, size, grade, and necrosis (SSIGN) scoring system has also been widely used to predict prognosis of RCC patients, which integrates TNM stage, tumor size, Fuhrman grade, and histologic tumor necrosis for predicting cancer-specific survival (CSS) originally. Low-risk (LR) (SSIGN 0–3), intermediate-risk (IR) (SSIGN 4–7), and high-risk (HR) (SSIGN ≥ 8) groups have been established according to the SSIGN score in previous studies.4,5 The University of Los Angeles Integrated Staging System (UISS) is another well-established prognostic model in application. However, these models still have their limitations when it comes to classifying patients with different prognoses accurately because of the heterogeneous nature of RCC. Hence, new biomarkers that can further enrich the established prognostic system and new prognostic models are desperately needed. Such biomarkers could optimize risk stratification and provide guidance so that individual therapeutic strategies can be undertaken earlier.

Colony-stimulating factor 1 receptor (CSF-1R) is a single-pass type III transmembrane tyrosine–protein kinase encoded by c-fms proto-oncogene, which acts as a cell-surface receptor for colony-stimulating factor 1 (CSF-1) and interleukin 34. CSF-1R is essential in the survival, proliferation, differentiation, and recruitment of mononuclear phagocytes cell lineages. Moreover, CSF-1R could promote the release of a series of proinflammatory chemokines when combined with its ligands and thereby plays a vital role in inflammatory processes.6,7 In addition, both RAS-activated protein kinase-dependent and phosphatidylinositol 3-kinase (PI3K)-dependent pathways are activated through CSF-1R activation.8,9 Moreover, it has been determined that autocrine regulation involving the tumor cells themselves and paracrine regulation with host macrophages of CSF-1R facilitate the development and progression of solid tumors.7,10

The activation of CSF-1R on the surface of macrophages results in the alternatively activated (M2) polarization of tumor-associated macrophages (TAM).11,12 M2 macrophages exert the functions of an anti-inflammatory; favor angiogenesis; provide immunosuppression; and promote tumor growth—contrary to classically activated (M1) macrophages, which can mediate the defense of the host from a variety of viruses and bacteria and which play roles in anticancer immunity.11 Previous studies have demonstrated that inhibition of CSF-1R activity could decrease the infiltration of TAM and result in a delay in tumor progression. Overactivation of CSF-1R significantly increases the recruitment of TAM and accelerates the tumor growth and progression.13–15 Menke et al. found autocrine CSF-1 and CSF-1R coexpressed in RCC cells, and CSF-1 engagement of CSF-1R promoted the survival and proliferation of RCC as well as reduced apoptosis.16 Evidence is mounting that high expression of CSF-1R is closely associated with poor prognosis in a variety of malignancies.13,14,17,18 To date, however, to our knowledge, no study has evaluated the clinical significance of CSF-1R expression in ccRCC.

In this study, we analyzed the expression level of CSF-1R in clinical specimens obtained from ccRCC patients. The association of CSF-1R with clinicopathologic characteristics and patient outcomes including CSS was determined. In addition, the prognostic value of CSF-1R and its improvement to prognostic accuracy of conventional prognostic models were evaluated. A prognostic nomogram was constructed on the basis of the identified independent prognosticators in this study.

Methods

Patients

We retrospectively recruited 268 patients receiving either partial or radical nephrectomy for clear cell renal cell carcinoma (ccRCC) at Zhongshan Hospital, Fudan University (Shanghai, China), from January 2001 to December 2004. The study was approved by the ethics committee of Zhongshan Hospital, Fudan University. Each patient had signed the informed consent agreement. Clinicopathologic variables including age, gender, tumor size, pT stage, pN stage, metastasis, TNM stage, Fuhrman grade, necrosis, sarcomatoid, and ECOG PS of each patient were collected. The disease of patients was staged through radiographic reports combined with postoperative pathologic data and was reassigned in the light of 2010 American Joint Committee on Cancer TNM classification. Survival status was updated in October 2013. Median follow-up was 92 months (range 11–120 months). CSS was the end point in this study. We calculated CSS from the day of curative nephrectomy to the day of death or last follow-up.

Western Blot Analysis

Western blot analysis was performed as described previously.19 Anti-CSF-1R antibody (dilution, 1:1000; sc-13949; Santa Cruz, Santa Cruz, CA) and anti-GAPDH antibody (dilution, 1:1000; sc-365062; Santa Cruz) were used for Western blot analysis.

Tissue Microarray and Immunohistochemistry

Hematoxylin and eosin staining was used to review surgical specimens histologically; specimens were fixed in formalin and embedded in paraffin. Away from necrotic and hemorrhagic materials, representative areas were marked more centrally on the paraffin blocks. We used duplicate 1.0 mm tissue cores taken from two different areas to construct the tissue microarrays. Primary anti-CSF-1R antibody (dilution, 1:200; sc-13949; Santa Cruz) was used for immunohistochemistry. Two pathologists (Q.F. and L.C.) masked the clinicopathologic data evaluated the staining score of each specimen using the H score, which was on a scale of 0–300, multiplying the percentage of positive cell distribution by the staining intensity (where 3, 2, 1, and 0 indicate strong, moderate, weak, and negative staining, respectively).20 The mean score of duplicate tissue cores was used for statistical analysis. X-tile Plots v3.6.1 (Yale University, New Haven, CT) was used to dichotomize patients into the CSF-1R-low (n = 139) and CSF-1R-high (n = 129) groups on the basis of the optimal cutoff value of the staining score (i.e., 135).

Statistical Analysis

SPSS Statistics 17.0 (IBM, Armonk, NY) was used to perform Pearson’s χ 2 test for the comparison of categorical variables, and Student’s t test was used to compare continuous variables. Kaplan–Meier analysis combined with the log rank test was performed to determine CSS curves by MedCalc 12.7.0.0 (https://www.medcalc.org/). Univariate and multivariate analyses were performed using Cox proportional hazard models by SPSS Statistics 17.0. The prognostic accuracy of Cox regression models was quantified by Harrell’s concordance index (C index), ranging from 0.5 (no prognostic power) to 1 (perfect prediction), by Stata 12.0 (StataCorp, College Station, TX). The prognostic nomogram was constructed by R 3.0.2 (https://www.r-project.org/). All significance tests were two sided and were performed at a statistical significance level of 0.05.

Results

CSF-1R Expression Pattern and Its Association with Clinicopathologic Features of ccRCC Patients

We compared CSF-1R expression level in seven paired peritumor and tumor tissues from patients with ccRCC by Western blot analysis. In 71.4 % (5 of 7) patients, the CSF-1R protein expression in tumor tissues was higher than in peritumor tissues (Fig. 1a). Then we further focused on the CSF-1R expression in tumor tissues. We evaluated the expression of CSF-1R in tumor specimens from 268 ccRCC patients by immunohistochemical analysis. CSF-1R positive staining was widely localized in the cytoplasm of tumor cells. Representative CSF-1R immunohistochemical images of high expression (score = 240) and low expression (score = 10) are presented in Fig. 1. As shown in Supplementary Fig. S1, 135 was selected as the cutoff of the immunohistochemistry score derived from the “minimum P value” approach, which separates the patients into low CSF-1R (n = 139) and high CSF-1R (n = 129) expression groups with the best discriminatory power. The descriptive statistics of immunohistochemistry score in all patients and low/high-CSF-1R expression subgroups are provided in Fig. 1c, d. Associations between CSF-1R expression and clinicopathologic features of ccRCC patients are listed in Table 1. CSF-1R expression was significantly positively associated with pT stage (P = 0.047), tumor metastasis (P = 0.005), TNM stage (P = 0.003), and ECOG PS (P = 0.011).

Fig. 1
figure 1

CSF-1R expression in ccRCC tissues. Results from “minimum P value” approach and descriptive statistics of immunohistochemistry score data. a CSF-1R protein level in 7 paired peritumor and tumor tissues is detected by Western blot analysis. P stands for peritumor tissue, T stands for tumor tissue. b Representative CSF-1R immunohistochemical images of high expression (score = 240) and low expression (score = 10), respectively. Scale bar = 50 μm (original magnification, ×200). c, d Descriptive statistics of immunohistochemistry score data of patients

Full size image
Table 1 Correlation between CSF-1R expression and patient characteristics
Full size table

High CSF-1R Expression was Associated with Adverse Prognosis

We next analyzed the associations between CSF-1R expression and CSS by the Kaplan–Meier method. As shown in Fig. 2a, patients with low CSF-1R expression had better CSS compared to those with high CSF-1R expression (P < 0.001). We further determined the effect of CSF-1R expression in different risk groups according to the TNM, UISS, and SSIGN scoring systems. As presented in Fig. 2b–d, in the LR group of SSIGN scoring system, the patients could not be significantly stratified by CSF-1R expression for CSS (P = 0.058), whereas patients in the IR and HR groups could be significantly stratified by CSF-1R expression (P < 0.001 and P = 0.037, respectively). Furthermore, all of the risk groups based on the TNM and UISS systems could be stratified by CSF-1R expression (Supplementary Fig. S2).

Fig. 2
figure 2

Kaplan–Meier analysis for CSS of patients with ccRCC. a Kaplan–Meier analysis for CSS of all ccRCC patients according to CSF-1R expression (n = 268). Kaplan–Meier analysis in b SSIGN LR (n = 182), c IR (n = 71), and d HR (n = 15) groups. P value calculated by log rank test

Full size image

High Expression of CSF-1R was Identified as an Independent Prognostic Factor of CSS in ccRCC Patients

In order to assess the prognostic value of CSF-1R expression for CSS, we first performed univariate Cox analysis. As presented in Supplementary Table S1, tumor size (P < 0.001), pT stage (P < 0.001), pN stage (P < 0.001), metastasis (P < 0.001), Fuhrman grade (P < 0.001), necrosis (P < 0.001), sarcomatoid (P < 0.001), ECOG PS (P < 0.001), and CSF-1R (P < 0.001) were found to be negative prognostic factors for CSS. Then the prognostic factors that showed statistical significance in univariate analysis were analyzed by Cox multivariate regression. Besides pT stage, metastasis, Fuhrman grade, and sarcomatoid, CSF-1R expression (hazard ratio 2.209; 95 % confidence interval 1.363–3.579; P = 0.001) was also identified as an independent adverse prognostic factor for CSS in ccRCC patients after surgery (Fig. 3).

Fig. 3
figure 3

Multivariate Cox regression analysis for CSS in patients with ccRCC. Forest plot presented results of multivariable Cox regression analysis for CSS in patients with ccRCC

Full size image

Extension of Well-established Prognostic Models with CSF-1R Expression

To investigate whether incorporation of CSF-1R expression into well-established models—TNM, UISS, and SSIGN—would improve their prognostic accuracy, we calculated the C index of these models combined with or without CSF-1R expression for CSS, respectively. The C indexes of the TNM, UISS, and SSIGN scoring systems were 0.681, 0.712, and 0.692, which were significantly improved from 0.732 (P = 0.001), 0.753 (P = 0.001), and 0.735 (P = 0.001), respectively (Supplementary Table S2).

Prognostic Nomogram for CSS in Patients with ccRCC

According to the results of multivariate regression analysis, a prognostic nomogram was constructed integrating all the independent prognostic indicators for CSS (Fig. 4a). The calibration plots for the probability for CSS at 3, 5, and 10 years after surgery showed optimal agreements between the actual observed survival probability and the predicted survival probability by nomogram with prognostic accuracies of 85.66, 83.08, and 81.33 %, respectively (Fig. 4b–d).

Fig. 4
figure 4

Prognostic nomogram and calibration plots. a Nomogram was constructed by integrating all independent prognostic indicators. Calibration plots for probability for CSS at 3 years (b), 5 years (c), and 10 years (d) after surgery

Full size image

Discussion

In the current study, we demonstrated that high CSF-1R expression was significantly associated with poor prognosis of surgically treated patients with ccRCC. CSF-1R expression was determined to be an independent adverse prognostic factor for CSS. The incorporation of CSF-1R expression improved the prognostic accuracy of the conventional prognostic models. CSF-1R expression with other independent prognosticators could be integrated into a newly constructed and efficient prognostic nomogram. To our knowledge, our study is the first to report the clinical significance of CSF-1R in patients with ccRCC.

CSF-1R, a member of the family of growth factor receptors, is of great significance for the malignant transformation of tumors.14 Once activated by its ligands, several intracellular tyrosine residues of CSF-1R were phosphorylated, leading to the initiation of a cascade of biochemical events providing sustained signals for cell growth, transformation, and tumorigenesis.8 Caescu et al. uncovered nuclear factor κB and extracellular signal-regulated kinase 1/2 (ERK1/2) as CSF-1R pTyr-721 regulated signaling nodes.21 Digiacomo et al. demonstrated that macrophage migration was promoted by CSF-1R that was transactivated and synergized with prostaglandin E2 at the ERK1/2 level.22 Overexpression of CSF-1R is common in many types of carcinomas and is related to poor prognosis as well as associated with high TNM stage.13,17 Having recently determined that CSF-1R expression was up-regulated in tumor tissues compared to corresponding peritumor tissues, we next explored the characteristics of CSF-1R in ccRCC using The Cancer Genome Atlas ccRCC database. Analysis conducted by cBioPortal indicated that CSF-1R gene amplification was quite frequent in ccRCC. Further survival analysis demonstrated that both the difference in overall survival and disease-free survival between affected and unaffected patients failed to reach level of statistical significance. Although the amplification of genomic DNA could affect the expression level of the involved genes, the protein level in cells is not solely determined by this issue. Besides modulation of consumption, the protein level is extensively determined by complicated production processes, including amplification of genomic DNA, transcriptional activation, posttranscriptional processing, translational regulation, and posttranslational modification. Our subsequent work was thus focused on the clinical significance of the CSF-1R protein level in ccRCC. However, the exact molecular mechanism involved in the up-regulation of CSF-1R protein level in ccRCC merits further investigation.

The tumor microenvironment is instrumental in promoting invasion and metastasis of tumor cells by means of either direct interactions between stromal cells and tumor cells or by the secretion of chemotactic factors targeting the tumor cells. A wealth of evidence has suggested that invasion and metastasis of malignancies require both the autocrine loop and the paracrine loop of CSF-1R signaling. Our previous work as well as our current findings have demonstrated that either CSF-1 or CSF-1R alone is competent to be an independent adverse prognostic factor in ccRCC.6 Tumor cells could express both CSF-1 and CSF-1R, and the autocrine loop between them could promote tumorigenesis, which has been determined in several malignancies.8,16 CSF-1 also recruits TAM to the site in which they accelerate the progression of tumor through combining with its sole receptor, CSF-1R.23–26 The paracrine interaction between tumor cells and macrophages, which involves CSF-1, CSF-1R, and epidermal growth factor, acts as the driving force for the spread of the tumor through accelerating migration, invasion, and metastasis of tumor cells. Patsialou et al. have shown that abrogation of autocrine CSF-1R signaling could significantly reduce invasion, dissemination, and metastasis in a claudin-low breast cell line.27 Such work has indicated that the combination of CSF-1 and CSF-1R mainly affect the process of tumor acceleration to advanced stage and progression of metastasis. Our data demonstrated that patients in the IR and HR groups could be significantly stratified by CSF-1R expression whereas patients in the LR group could not, which also indicated its pivotal role during tumor progression. Besides this classical ligand-receptor activation mode, other vigorous mechanisms might also contribute to the prosperity of the CSF-1R signaling pathway, which merits further exploration.

The tumor microenvironment has far-researching significance for the diagnosis and treatment of tumors. The intercellular communication of tumor cells with their surrounding stroma needs the cooperation of a series of factors and receptors. However, current prognostic models ignore the irreplaceable role of cancer environment; they predict patient death or recurrence mainly on the basis of tumor histology and tumor morphology. CSF-1R, although expressed on ccRCC cells, as we demonstrated, may also be an important member of the tumor microenvironment. Moreover, our results revealed that CSF-1R expression could be identified as an independent prognostic factor for patients with ccRCC. The incorporation of CSF-1R expression into conventional prognostic models, as we did, might improve their prognostic efficacy.

The components and cytokines of the tumor microenvironment may also present some potential therapeutic targets that could be applied in individual-level treatment. Xu et al. highlighted the significance of CSF-1/CSF-1R signaling in recruiting TAMs, which limit the efficacy of radiotherapy.28 Escamilla et al. demonstrated that blockade of TAM influx through inhibiting CSF-1R disrupts tumor promotion and provides a more durable therapeutic response in prostate cancer.29 Some studies have also raised the translational potential of the inhibition of CSF-1R for malignancies.30 Taken together, these data indicate that targeting CSF-1R might have a potential application value in the treatment of ccRCC, which merits further investigation. The well-known programmed death-1 (PD-1), also known as CD279, is a coinhibitory receptor and is mostly expressed on the surface of T cells and B cells. When activated by its two ligands, PD-L1 and PD-L2, the PD-1 pathway could mediate local immunosuppression around the tumor microenvironment to protect tumor cells from immune surveillance.31 The associations between PD-1 expression by lymphocytes and RCC TNM stage, grade, and prognosis, as well as the PD-L1 expression by RCC cancer cells and its potential associations with clinical outcomes, have led to the development of new anti-PD-1/PD-L1 agents for the treatment of RCC.32 CSF-1R might be closely related to PD-1 under the immunosuppressive microenvironment during tumor development and progression of ccRCC. Whether CSF-1R and PD-1 follow a close concurrent expression pattern or represent independent pathways needs further investigation.

We acknowledge several limitations of our study. First, other populations and larger external cohorts should be enrolled to validate our findings. Second, although two tissue cores in the same tumor were used to make the tissue microarray blocks, only small pieces of information of the original tumor can be displayed using the tissue microarray technique. Finally, functional studies are needed to further elucidate the biologic mechanisms under this association.