Platelet count, aspirin use, and characteristics of host inflammatory responses in colorectal cancer

This study is based on an earlier described prospectively recruited cohort of surgically treated stage I–IV CRC patients, operated in Oulu University Hospital in 2006–2014 (n = 356; Additional file 1: Table S1) [22‐24]. All the patients were required to sign a written informed consent to participate. The study was approved by the Ethics Committee of Oulu University Hospital (58/2005, 184/2009) and was performed in accordance with the Declaration of Helsinki. The clinical details, including the list of patients’ medication, were collected from the clinical records and follow-up data from the clinical records and from Statistics Finland [25, 26]. Time to recurrence (TTR) was defined as time from the operation to the recurrence of the same cancer, cancer specific survival (CSS) was defined as time from the operation to death from the same cancer, and overall survival (OS) was defined as time from the operation to death, irrespective of cause. Preoperative staging of rectal cancer was conducted with magnetic resonance imaging, and most rectal cancer patients with cT3 or cT4 tumors (n = 69) received preoperative radiotherapy or chemoradiotherapy (RT/CRT). Additionally, one colon cancer patient received neoadjuvant treatment. The study was designed in accordance with the REMARK criteria [27].

Blood and serum samples were collected preoperatively [22]. Blood platelet count, blood Hb, erythrocyte MCV, serum CRP and serum albumin were measured in the laboratory of Oulu University Hospital [22, 28]. Anemia was defined according to WHO criteria as blood Hb levels < 120 g/L in women and < 130 g/L in men [23]. It was classified according to erythrocyte MCV levels as microcytic (MCV < 80 fL), normocytic (MCV 80–100 fL), and macrocytic (MCV > 100 fL) [23]. CRP and albumin were used to calculate modified Glasgow Prognostic Score (mGPS): mGPS0, serum CRP ≤ 10 mg/L and serum albumin ≥ 35 g/L or < 35 g/L; mGPS1, serum CRP > 10 mg/L and serum albumin ≥ 35 g/L; mGPS2, serum CRP > 10 mg/L and serum albumin < 35 g/L) [18, 29]. In patients operated between 2006 and January 2010 (n = 148), serum analysis of 27 cytokines was performed with Bio-Plex Pro Human 27-Plex Cytokine Panel (Bio-Rad, Hercules, CA, USA), as described earlier [22]. As described earlier in more detail, 14 cytokines had many values outside the assay detection limits, and 13 cytokines (IL-1ra, IL-4, IL-6, IL-7, IL-8, IL-9, IL-12, IFN-γ, CXL10, CCL2, CCL4, CCL11, and PDGF-BB) with less than four values outside the assay working range were included in this study [22, 30].

Histopathological analysis was conducted using the hematoxylin and eosin stained sections. Tumor stage was reclassified according to TNM8, and the grading was conducted according to the WHO criteria. Lymphatic invasion was defined as tumor cells present in vessels with an endothelial lining but lacking a muscular wall, while blood vessel invasion was defined as tumor cells in vessels with a thick muscular wall or in vessels containing red blood cells [31].

Tissue microarrays were constructed for immunohistochemical analysis, consisting of 1–4 cores of 3.0 mm diameter for each tumor, depending on the size of the tumor, from the invasive margin (IM) and the tumor center (CT) [23, 32]. Immunohistochemistry for mismatch repair (MMR) enzymes MLH1, MSH2, MSH6, and PMS2 was conducted, as described earlier, to evaluate MMR enzyme status [23, 28, 33]. BRAF V600E specific VE1 immunohistochemistry was conducted with Ventana Bench-Mark XT immunostainer (Ventana Medical Systems, Tucson, AZ) [34], to evaluate BRAF mutation status. Our earlier study indicated that the method had a sensitivity of 100% and a specificity of 99.3% in detecting BRAF V600E mutation [34]. The densities of five types of tumor infiltrating immune cells at the invasive margin (IM) and center of the tumor (CT) were determined with an ImageJ based computer assisted analysis method [35], as described earlier [24, 36]. Intraepithelial (IEL) CD3⁺ and CD8⁺ lymphocytes were counted manually from the captured images, because the method could not segregate intraepithelial cells from those in tumor stroma [36]. The antibodies and protocols for immunohistochemistry are specified in Additional file 1: Table S2. All the histological and immunohistochemical analyses were performed blinded to other data.

The statistical analyses were conducted using R version 3.5.1 (R Foundation for Statistical Computing, Vienna, Austria), using the packages tidyverse (v.1.2.1), survival (v.2.42-6), survminer (v.0.4.3), plotROC (v.2.2.1), car (v.3.0-2), lm.beta (v.1.5-1), and powerSurvEpi (v.0.1.0), or IBM SPSS Statistics for Windows, version 24.0 (IBM Corp., Armonk, NY). The statistical significances of the associations between categorical and continuous variables were analyzed by independent samples t-test or Mann–Whitney test (comparing two classes), or one-way analysis of variances (ANOVA) or Kruskal–Wallis test (comparing three or more classes), as appropriate, while the statistical significances of the associations between two categorical variables were analyzed with χ² test. Pearson correlation coefficients (r) were used to determine the correlation between two continuous variables. To normalize their distribution, logarithmic transformation was applied to variables with positive skewness. Multiple linear regression models were used in multivariable analysis of the associations between blood platelet count, systemic inflammatory markers/tumor infiltrating immune cells, and selected clinicopathological factors. The normality of the residuals, homoscedasticity, and linearity assumptions were checked by a histogram and a normal probability plot of the residuals and by a scatterplot of residuals vs. predicted values. Cytoscape, an open source software platform for visualizing complex networks, was used in creating a 2D visualization of the relationships between blood platelet count, blood Hb levels, and serum levels of systemic inflammatory markers with the Prefuse force directed algorithm weighted by the statistical significances of the correlations between individual variables [37]. Receiver operating characteristics (ROC) analysis was used to evaluate the capacity of platelet count to discriminate the survivors from non-survivors. The survival outcomes of the patient subgroups were analyzed with Kaplan–Meier curves, log-rank test, and Cox proportional hazards regression analysis. In multivariable Cox models, the cases with missing data were assigned to the majority category of a given covariate to limit the degrees of freedom of the models. We confirmed that excluding cases with missing data on any of the covariates did not substantially alter the results (data not shown). An interaction between platelet count and aspirin use was assessed by including the product interaction term of these variables in the model. All p values are two-tailed. As our primary hypothesis testing, we evaluated the relationships between blood platelet count and TTR, CSS, and OS. For these primary, confirmatory survival analyses, we used α level of 0.05, while for the other exploratory analyses, we used α level of 0.005 [38], and regarded the results with p = 0.05–0.005 as of borderline statistical significance and interpreted them cautiously. Based on the most recent meta-analysis, we expected to observe a HR of 2.11 for the association between high platelet count and OS [11]. The statistical power (1−β) to detect this HR was estimated to be 0.98 at α level of 0.05, 0.92 at α level of 0.01, 0.88 at α level of 0.005, and 0.76 at α level of 0.001 (Additional file 1: Figure S1).

Among the 356 CRC patients, mean platelet count was 294.9 × 10⁹/L (SD 93.4 × 10⁹/L) (Table 1). High platelet count, studied as a continuous variable, was associated with female gender (p = 0.002), proximal tumor location (p < 0.001), high TNM stage (p < 0.001), especially high T-class (p = 0.003) and M-class (p = 0.001), and normocytic and microcytic anemia (p < 0.001). High platelet count also showed tendency towards an association (borderline statistical significance considering multiple hypothesis testing) with young patient age (p = 0.044) and MMR deficiency (p = 0.006). Platelet count had no association with aspirin use (p = 0.396).

We hypothesized that elevated platelet count would be associated with systemic inflammation in CRC. Supporting the hypothesis, increased platelet count, as a continuous variable, was associated with higher mGPS (p < 0.001; Table 1) and with higher serum C-reactive protein levels (univariable p < 0.001; multivariable adjusted p = 0.003; Additional file 1: Table S3).

For more detailed overview on the relationships between platelet count and systemic inflammatory mediators, serum analysis of 13 cytokines was conducted in 148 patients. This analysis indicated that platelet count positively correlated with most of the studied cytokines, including IL-1RA, IL-4, IL-6, IL-7, IL-8, IL-12. IFNγ, and PDGF-BB (univariable and multivariable adjusted p < 0.001 for all; Fig. 1; Table 2); the strongest correlations were with IL-7 (r = 0.531), IL-1RA (r = 0.509), and PDGF-BB (r = 0.474).

The strong associations between platelet count and serum cytokine levels lead us to hypothesize that platelet count would also contribute to tumor immune cell infiltration by the release of inflammatory mediators. Thus, correlations between platelet count and five types of tumor infiltrating immune cells (CD3⁺ T cells, CD8⁺ cytotoxic T cells, FoxP3⁺ regulatory T cells, Tryptase⁺ mast cells, Elastase⁺ neutrophils) in different tumor locations were analyzed (Fig. 2; Table 3). However, no statistically significant multivariable adjusted associations were detected between platelet count and the analyzed cell types.

We hypothesized that the effects of platelets in CRC could depend on whether patients receive aspirin medication inhibiting platelet activation. Therefore, we analyzed the relationships between aspirin use, clinicopathological characteristics (Additional file 1: Table S4), systemic inflammatory markers (Additional file 1: Table S5), and tumor infiltrating immune cells (Additional file 1: Table S5). There were 85 (23.9%) patients, who had aspirin in their list of daily medications in the clinical records at the time of surgery, with dosage of 100 mg/day (n = 75, 88.2%) or 50 mg/day (n = 10, 11.8%). Aspirin use was more prevalent in the older age group (p < 0.001), among the patients who did not receive preoperative RT/CRT (p = 0.010; borderline statistical significance considering multiple hypothesis testing), in lower N-class (p = 0.026; borderline statistical significance considering multiple hypothesis testing), and in patients with anemia (p = 0.039; borderline statistical significance considering multiple hypothesis testing). Aspirin use did not significantly associate with systemic inflammatory markers (Additional file 1: Table S6) but associated with a tendency towards higher numbers of FoxP3⁺ T cells at the IM (p = 0.005) and higher numbers of neutrophils in the CT (p = 0.031; borderline statistical significance considering multiple hypothesis testing).

Finally, survival analyses were conducted. ROC analysis indicated that blood platelet count did not show good ability to discriminate survivors from non-survivors (Fig. 3a–c). In the previous studies, 400 × 10⁹/L has been the most commonly used cut-off for high platelet count [11], and we decided to apply that to our primary analyses. However, subsequent Kaplan–Meier curves (Fig. 3d–f), and univariable Cox regression models (Additional file 1: Table S6) did not indicate statistically significant association between platelet count and survival. To evaluate the consistency of our data with different cut-off values, we also tested 300 × 10⁹/L and 200 × 10⁹/L as the cut-off points, and the results were parallel to those with 400 × 10⁹/L (Additional file 1: Figs. S2, S3, Table S6). We hypothesized that aspirin use could influence the prognostic effect of platelet count. However, combined evaluation of platelet count and aspirin did not show statistically significant associations with survival (Fig. 3g–i, Additional file 1: Figs. S2, S3, Table S6). Blood platelet count and aspirin use were not statistically significant in multivariable Cox models (Table 4). There was no statistically significant interaction between platelet count and aspirin use (TTR: p for interaction = 0.933, CSS: p for interaction = 0.536, OS: p for interaction = 0.500).