Sex Differences in Using Systemic Inflammatory Markers to Prognosticate Patients with Head and Neck Squamous Cell Carcinoma

In 2017, cancer was estimated to be the cause of nearly one out of every four deaths in the United States. Head and neck squamous cell carcinoma (HNSCC), comprising cancers of the oral cavity, pharynx, and larynx, accounts for 4% of all cancers, resulting in nearly 50,000 new cases and 10,000 deaths last year alone (1). Primary risk factors for HNSCC include smoking, alcohol use, and human papilloma virus (HPV) infection of the oropharynx. In fact, etiology of over 70% of oropharyngeal cancers diagnosed in the United States after the year 2000 could be attributed to HPV infection (2, 3).

Interestingly, men have historically been much more likely to develop both HPV-unassociated and -associated HNSCC, with their risk being approximately 3 times as high as it is for women (4–9). Indeed, HNSCC is the 7th most common cancer in men but only the 13th most common in women in the United States (10). This substantial male bias is insufficiently explained by sex differences in smoking and alcohol consumption (11). Furthermore, females present with lower rates of HPV infection despite not engaging with proportionally fewer oral sexual partners (8, 9). Sex bias likely reflects fundamental biological differences between males and females as mediated by sex chromosomes and hormones. Yet literature on their role in HNSCC is sparse (11, 12).

Males and females share robust differences in their susceptibility to various autoimmune and infectious diseases, emphasizing that sex is an important biological variable affecting the immune system (13). However, potential implication of sex-based immunologic differences on the pathogenesis of various malignancies has not been thoroughly investigated. Markers of subclinical inflammation [platelet, neutrophil, lymphocyte counts, and neutrophil-to-lymphocyte ratio (NLR)] have been proposed as prognostic indicators for both overall survival (OS) and recurrence-free-survival (RFS) in a variety of cancer types, including HNSCC. In general, both high platelet counts and high NLR are associated with decreased survival (14–24). In this study, we determined sex-specific prognostic value of peripheral immune/inflammatory responses in various malignancies with focus on HNSCC. In addition, CIBERSORT (Cell-type Identification By Estimating Relative Subsets Of Rna Transcripts) and ESTIMATE (Estimation of Stromal and Immune cells in Malignant Tumor tissues using Expression data) were used to analyze RNA-sequencing data from The Cancer Genome Atlas (TCGA) to compare male and female immune cell infiltration in HNSCC tumors and to validate our findings at the level of the tumor microenvironment (25, 26). This study offers an important insight that select cancers in males and females may have differential immune involvement and therefore provide support for stratification in the use of platelet count and NLR by patients’ sex in the era of precision oncology.

The Medical University of South Carolina (MUSC) Institutional Review Board approved all study activities, and the studies were conducted in accordance with the U.S. Common Rule. For each case, we abstracted data on demographic characteristics, clinical and pathologic variables at diagnosis, and outcome using two different data sources: the Hollings Cancer Center cancer registry and the MUSC Clinical Data Warehouse. The registry is part of a state-mandated data system that collects cancer incidence on all cases in South Carolina. The Clinical Data Warehouse is a single, secure, and integrated database extracted from the MUSC OACIS Clinical Data Repository, which includes patient demographics, International Classification of Disease-coded diagnoses, International Classification of Disease-coded procedures, medications, and laboratory test results. These databases were subsequently linked, in a blinded fashion, through an honest broker at the Clinical Data Warehouse, and entered into a secured study database.

Independent variables obtained from the Clinical Data Warehouse included pretreatment hematologic parameters (platelet count, red blood cell count, white blood cell count, absolute neutrophil count, absolute lymphocyte count, and hemoglobin level). Variables obtained from the registry included sociodemographic characteristics (age at diagnosis, sex, and race), lifestyle factors including smoking status (never, current, former) and alcohol use (never, current, former), location of the primary tumor, tumor–node–metastasis pathologic staging, and the p16 (HPV) status. The p16 status, assayed by immunohistochemistry, was used as a surrogate marker for HPV. All values were collected at the time of diagnosis.

The Cancer Registry at Hollings Cancer Center, MUSC, was used to identify 9,365 total patients with primary tumors from 9 different anatomical locations (bladder, colorectal, esophageal, head and neck, renal, liver, lung, skin melanoma, and pancreas). Cancer types were defined by ICD-10-CM codes, and histologic subtypes were all included in analysis. The study population consisted of histologically confirmed cases diagnosed between January 1, 2000, and June 30, 2012.

OS was the primary measured outcome. Statistical analyses were performed using Kaplan–Meier (KM) curves, log-rank test, and Cox proportional hazards regression. Variables considered in statistical analyses included platelet count (continuous variable), absolute neutrophil count (continuous variable), absolute lymphocyte count (continuous variable), age (continuous variable), race (white, black, other), tobacco history (never, current, former), alcohol history (never, current, former), and combined American Joint Committee on Cancer pathologic stage (I, II, III, IV). For the nine cancer type screen, a platelet cutoff factor of 300 K/mm³ was used to separate patients into two groups (high and low). Patients with platelet counts lower than 150 K/mm³ or higher than 450 K/mm³ were not included in this screen to avoid the possible effect of primary or secondary hematologic comorbidities on survival (18). Based on results from a previous meta-analysis, an NLR cutoff of 3 was used to separate patients into two groups (high and low; ref. 27). The χ² test was used to test homogeneity in distribution of high and low groups for males and females.

All hematologic parameters at time of diagnosis were not known for all patients. However, all patients with known platelet count or NLR values were included in analyses. The total number of male and female patients with known hematologic parameters for each cancer is shown in Supplementary Table S1.

RNA-sequencing data from TCGA (01-28-16 release) were downloaded from the Broad Institute GDAC Firehose (http://firebrowse.org). RNA-sequencing data for the HNSCC cohort were used (n = 520), and we downloaded RSEM-normalized values generated from the RNA-seqV2 platform.

To prepare the expression data for CIBERSORT analysis, genes with RSEM_avg < 10 were removed. The expression matrix was then normalized using voom (from R package limma) to transform RSEM to a distribution resembling microarray data. The transformed values were then uploaded to the CIBERSORT web portal (http://cibersort.stanford.edu/). CIBERSORT's LM22 signature matrix was used to analyze the expression matrix. Permutations were set to 1,000, and quantile normalization was turned off. Cell categories were simplified by summing component percentages into four groups: T lymphocyte, B lymphocyte, macrophages, and granulocytes.

ESTIMATE scores (Immune, Stromal, and ESTIMATE) for the TCGA HNSCC cohort used in this article were downloaded from the ESTIMATE portal website (http://bioinformatics.mdanderson.org/estimate/). Note that these scores were calculated based on the RNA-seqV2 platform, which we also used to calculate the CIBERSORT scores. Both ESTIMATE and CIBERSORT data were recombined with the clinical annotations (including patient gender and pathologic stage) extracted with the rtcga package. Pearson correlation and Welch t test were used to compare CIBERSORT- and ESTIMATE-derived values. All analyses were performed with R, using the packages rtcga, ggplot2, limma, and survival (28–31).

OS data for the TCGA cohort were retrieved with rtcga. Only patients with CIBERSORT P values less than 0.05 were included in this analysis. Optimal cutoff points for total T lymphocyte count and granulocyte count were determined by using an outcome-oriented method of maximizing log-rank statistics. This computation was done using the R package survminer. The cutoff point determined for T lymphocyte and granulocyte counts were 36% infiltration and 6.3% infiltration, respectively. Female and male patients were then separated into high and low groups prior to KM analysis.

To determine whether sex differences in circulating platelet, lymphocyte, or neutrophil counts exist, such hematologic parameters were compared in male and female patients for each of the nine cancer types (Fig. 1). All stages of cancers were included, and the samples were collected at the time of diagnosis. As a group, females presented with statistically significantly higher platelet counts for most cancers, higher neutrophil counts for only one cancer (liver), and higher lymphocyte counts for six cancers (colorectal, head and neck, kidney, lung, melanoma, and pancreas). It is notable that patients with liver cancers presented with particularly lower platelet, lymphocyte, and neutrophil counts compared with patients with other cancer types.

To evaluate the association between platelet counts and OS for each sex separately, KM curves were generated, stratified by platelet count and sex (Fig. 2). The platelet cutoff value used to sort patients was 300 K/mm³ in order to be consistent with previously published studies (16). In addition, the cutoff was numerically convenient for dividing the counts into quartiles within the range of physiologically normal platelet counts (150–450). Patients with platelet counts less than 150 K/mm³ or greater than 450 K/mm³ were not included in this preliminary screen in order to avoid confounding hematologic comorbidities (18). Log-rank test results indicated that colorectal, esophageal, head and neck, and liver cancers show the largest differences in prognoses between males and females. For these four cancers, high platelet count was associated with decreased OS in males but not in females. As for other cancer types, there was either no (renal, melanoma, pancreas) or non sex-specific (bladder, lung) relationship between platelet and OS.

Based on results from a previous meta-analysis, an NLR cutoff of 3 was used to separate patients into two groups (27). KM curves were generated, stratified by NLR category and sex (Fig. 3). Log-rank tests demonstrated significant difference in OS distribution between patients with high and low NLR in both sexes for most cancers. However, head and neck cancers were remarkable for exhibiting a sex discrepancy in prognostic value. For head and neck cancer, NLR was not significantly associated with OS for males [HR = 1.261; 95% confidence interval (CI), 0.998–1.593] as compared with females (HR = 2.627, 1.716–4.02). In addition, Cox regression results showed head and neck cancer was uniquely significant in that sex affects NLR's contribution toward OS (β = 0.702, P = 0.0044; Supplementary Table S2).

Due to a great sex discrepancy in the association between platelet count/NLR and OS and adequate sample size for HNSCC, we focused subsequent analyses on this patient population (Table 1). When comparing by sex, females were more likely to be negative for smoking history and alcohol use. However, this could be influenced by response bias. They also tended to present with an earlier staged tumor. In addition, primary lesion locations differed between males and females, where males presented with relatively fewer oral cavity tumors and more laryngeal tumors. In terms of hematologic parameters, females had higher mean platelet counts, but there was no difference in absolute neutrophil count, absolute lymphocyte count, WBC count, or NLR value. As expected, females had lower RBC and hemoglobin counts (32). In this cohort of patients, the number of patients with known HPV status was insufficient (85 males, 21 females, n = 106) for us to make a statistical conclusion on the contributions of HPV status.

To see if we could reproduce findings from previous reports, we assessed whether platelet count and NLR were associated with OS in our cohort with both sexes combined (18, 22, 27, 33). We also tested additional cutoff criteria, including separating patients into four subgroups (platelet count: 150–224, 225–299, 300–374, 375–450 K/mm³; NLR: quartiles) in order to gain more insight into their contributions towards survival. Indeed, when patients were not separated by sex, both platelet count (log-rank test: P = 0.0011) and NLR (P = 0.00014) were significantly associated with OS (Fig. 4; Supplementary Fig. S1). However, when males and females were considered separately, platelet count was no longer associative for females, and NLR is no longer associative for males (Fig. 4).

Furthermore, we postulated that both platelet count and NLR might be correlated with cancer staging. Platelet count did positively correlate with stage and female sex (ANOVA; Stage: P = 0.00162, Sex: P = 0.00033) but NLR (P = 0.096, P = 0.219), neutrophil counts (P = 0.099, P = 0.902), and lymphocyte counts (P = 0.335, P = 0.420) did not (Supplementary Fig. S2; Supplementary Table S3).

To probe into the nuances of NLR-predictive value, we tested whether neutrophils or lymphocytes were individually associated with OS. Log-rank tests revealed that neutrophil count was associated with worse prognosis for both males and females, but lymphocyte count was associated with better prognosis only for females (Supplementary Fig. S3). However, univariate analyses are unable to address the possibility of confounding interactions.

Therefore, multivariate analysis was performed with Cox proportional hazards regression. Regressions were modeled for each sex separately based on platelet, neutrophil, and lymphocytes count (all continuous variable), age, race, tobacco/alcohol history, and stage (Supplementary Table S4). In agreement with our findings above, platelet count was associated with worse survival for males exclusively. On the other hand, only higher lymphocyte count correlated with increased survival in females. Surprisingly, neutrophil count was no longer predictive of survival.

To test if the results hold for a homogenous anatomical subgroup of HNSCC, we performed the same Cox regression using only oropharyngeal cancers. The impact of neutrophil count and lymphocyte count on survival was consistent with that of the mixed group (data not shown). However, platelet count was now barely significant for females (P = 0.050) but not for males (P = 0.092). Therefore, we believe that platelet count as a continuous variable does not have a strong impact on survival, especially upon adjusting for stage.

Thus far, we have demonstrated remarkable sex-specific differences in the prognostic value of peripheral hematologic parameters. In order to validate our findings at the level of the tumor microenvironment, we performed unbiased analysis using RNA-sequencing data of bulk tumor samples from TCGA. We separated the patients by sex and attempted to ascertain if there were any differences in immune-signatures.

ESTIMATE and CIBERSORT methods were used in parallel to evaluate the immune profile of the TCGA HNSCC cohort. ESTIMATE uses single-sample gene set expression analysis to predict tumor purity and presence of infiltrating immune and stromal cells in bulk tumor data (26). Immune score measures the extent of immune cell infiltration, Stromal score measures stromal cell infiltration, and ESTIMATE score is a measurement of tumor purity. Females expressed significantly higher ESTIMATE scores and immune scores, signifying higher expression of immune infiltration and tumor purity (Fig. 5A). There was no significant difference in stromal infiltration.

CIBERSORT is a method for characterizing the cellular composition using expression data from a heterogeneous sample (25). Moreover, the goodness-of-fit of the CIBERSORT algorithm offers an independent measurement of immune infiltration (34). In samples with higher immune cell infiltration, CIBERSORT is better at estimating the distribution of immune cells. As a result, lower P values/RMSE or higher Pearson correlations are associated with greater immune infiltration. Analyzing the female samples with CIBERSORT resulted in lower P values, lower RMSE, and higher Pearson correlations (Fig. 5B).

To test agreement between the two distinctive algorithmic methods, the Pearson correlation produced by CIBERSORT was plotted as a function of the ESTIMATE Immune score for all samples. There was a strong positive correlation (R = 0.59; P = 6 × 10⁻⁵¹; unpublished observations), suggesting that both methods were consistent in predicting immune infiltration.

To determine which immune cell subsets were most responsible for increases in immune infiltration, we performed correlation analyses between CIBERSORT-relative cell percentages and ESTIMATE Immune score. T lymphocytes, as a group, and M1/M2 macrophages were the only cell types that were positively associated with Immune score (Fig. 5C and D; Supplementary Fig. S4F). On the other hand, M0 macrophages, B lymphocyte, and granulocytes—including mast cells, eosinophils, and neutrophils—were negatively associated with immune score (Supplementary Fig. S4).

KM curves were plotted to evaluate the association between CIBERSORT T lymphocyte/granulocyte counts and OS in patients with notable immune infiltration (CIBERSORT P value < 0.05; Fig. 5E and F). Log-rank tests showed that having a high T-lymphocyte count was significantly correlated with longer survival in both sexes. However, the difference in survival between female high and low count groups (HR = 0.201, 0.048–0.842, P = 0.0281) was more than double that of male high and low count groups (HR = 0.495; 95% CI, 0.281–0.871, P = 0.0147). On the other hand, having a high granulocyte count was associated with worse survival in males only (HR = 1.80, 1.07–3.03, P = 0.0268). The effect of granulocyte count on female survival was not significant (HR = 1.54, 0.73–3.27, P = 0.256). We further implemented the χ² test to check whether the difference in HR between males and females is affected by the differences in size of subgroups. The result indicates that there are no significant differences in distributions of the high and low groups between females and males (P = 0.021 and 0.44, for T lymphocytes and granulocytes, respectively, with α = 0.01).

Persistence of sexual dimorphism in select cancer incidence and survival even after statistical correction for occupational and behavioral factors implies the existence of cellular and molecular differences in disease progression between the two sexes. Indeed, sex chromosomes and hormones have been shown to regulate various steps in cancer pathogenesis via both tumor cell–intrinsic and –extrinsic mechanisms (35, 36). Here, we examined the latter with focus on sex-biased involvement of various hematopoietic compartments primarily in patients with HNSCC. In males, both univariate and multivariate analyses indicated that use of circulating platelet and neutrophil counts as prognostic markers was more effective. In females, circulating lymphocyte counts, rather than neutrophils, were more correlated with OS. Interestingly, gene expression analyses using TCGA database suggested that females have higher immune cell infiltration into the tumor microenvironment, and T lymphocyte signature can more strikingly predict their survival. To the best of our knowledge, this is the first study to comprehensively demonstrate sex bias in the clinical utility of platelet, granulocyte, and lymphocyte counts as biomarkers to prognosticate HNSCC patients via analyses at both the level of periphery and the tumor microenvironment.

Our results provide a valuable starting point to encourage intense investigations for molecular mechanisms underlying sex specificity in the predictive value of peripheral blood counts. Of note, studies have suggested the existence of sex bias in platelet activity, immunity, and hemostasis. For instance, a platelet proteome analysis detected higher levels of signaling and activation proteins in male platelets compared with female platelets (37). Platelet-derived TGFβ has been shown to not only promote tumor growth and metastasis (38), but also play a role in converting peripheral naïve T cells to regulatory T cells, which can suppress antitumor immune function (39). Greater activity of male platelets may lend to its increased significance in cancer prognosis. Also, although we did not detect a sex difference in total lymphocyte count, higher peripheral percentages of overall and IL2 producing T lymphocytes have been reported in women compared with men (40). Higher peripheral T-cell count in females may mediate more efficient immune surveillance and tumor infiltration. The latter has been associated with better prognosis in multiple cancers including—but not limited to—melanoma, colorectal, ovarian, and breast, many of which demonstrate sex bias in their incidence and survival (41–45). Furthermore, sex differences may exist in the function or subtype identities of immune cells that studies involving absolute peripheral counts fail to uncover. Indeed, both T regulatory and T helper cells would be considered as part of absolute lymphocyte count despite playing distinctive physiologic roles. Due to mostly comparable neutrophil and lymphocyte counts between the sexes in our study cohort, observed disparity in NLR's prognostic value could be due to differences in lymphocyte activity or cellular subtype ratios. Overall, our findings implicate that antitumor immunity may be largely different between males and females.

Our study corroborates other studies that highlight elevated circulating platelet count and NLR as biomarkers for poor prognosis in a variety of cancer types, association of which remains not universally accepted and yet established in the clinic (46, 47). One important barrier to their clinical implementation is their lack of consistency and reproducibility (48–50). Our study shows that failing to account for sex may lead to confounding of these variables with cancer progression.

Our study limitations included the followings. First, the patient population came from a single study center (MUSC). Multi-centric prospective studies are required to reduce selection bias. Second, the anatomical distribution of HNSCC cases differed between males and females. Third, although we independently validated our study results using TCGA database, CIBERSORT and ESTIMATE methods are—as their names suggest—estimates of immune infiltration based on gene signature. Quantitative analysis by flow cytometry or immunohistochemistry would be optimal for measuring infiltration of specific immune cell subtypes.

There currently exists a revolution in sex biology, in which researchers are required to consider sex in all appropriate experimental designs and analyses. In accordance with its importance, we argue here that sex is a critical variable in our understanding of the role of circulating platelet, granulocyte, and lymphocyte counts in cancer progression. Detailed knowledge will be highly relevant to clinical practice given that peripheral blood counts are easily obtainable and pharmacologic tools to modulate the function of sex hormones are readily available.

No potential conflicts of interest were disclosed.

Conception and design: C.Y. Lin, H. Kwon, X. Li, Z. Li

Development of methodology: C.Y. Lin, H. Kwon, Z. Li

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C.Y. Lin, H. Kwon

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C.Y. Lin, H. Kwon, G.O. Rangel Rivera, D. Chung

Writing, review, and/or revision of the manuscript: C.Y. Lin, H. Kwon, G.O. Rangel Rivera, X. Li, D. Chung, Z. Li

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C.Y. Lin, H. Kwon, D. Chung, Z. Li

Study supervision: X. Li, Z. Li

This work was supported by multiple NIH grants: P01 CA186866, R01 CA213290, R01 CA188419, R01 AI077283 (to Z. Li); and R01 GM122078 and R21 CA209848 (to D. Chung). This work was also supported by Hollings Cancer Center's Cancer Center Support Grant P30 CA138313. H. Kwon was supported by scholarships from the Abney Foundation and Hollings Cancer Center. Z. Li is the Abney Scholar for Stem Cell Biology and Therapy and is supported by the South Carolina SmartState Center of Excellence.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.