Investigation of white blood cell characteristics in cerebrospinal fluid samples at pediatric brain tumor diagnosis

Abstract

Purpose

The role of white blood cells (WBC) in the pediatric central nervous system (CNS) tumor microenvironment is incompletely defined. We hypothesized that the WBC profile in cerebrospinal fluid (CSF) correlates with the presence of tumor cells and prognosis in pediatric CNS tumors, as well as other patient and disease characteristics, and differs by tumor type, thus giving insight into the tumor immune response.

Methods

We conducted a retrospective analysis of CSF WBC profiles at CNS tumor diagnosis in 269 patients at our institution. We examined total nucleated cell count, absolute counts, and percentages by WBC subtype. We compared CSF WBC values by tumor cell presence, patient vital status, tumor location, and the most common tumor types.

Results

Patients who died of their tumor had a lower CSF lymphocyte percentage and a higher absolute monocyte count in CSF at diagnosis. The presence of tumor cells in CSF was associated with fewer lymphocytes and monocytes. Ventricular tumors had higher CSF lymphocyte, monocyte, macrophage, and total nucleated cell counts than extraventricular tumors. Germ cell tumors, low-grade glioma, high-grade glioma, and ependymoma had lower macrophage counts or percentages compared to other tumor types.

Conclusions

WBC profile in CSF at pediatric CNS tumor diagnosis correlates with patient prognosis and presence of metastatic cells, along with tumor type and other tumor characteristics like relationship to the ventricles. Prospective CSF profiling and study may be useful to future immunotherapy and other pediatric CNS tumor clinical trials.

Introduction

Brain tumors are one of the leading causes of death in children and adolescents, accounting for 30% of total pediatric cancer deaths in the United States and recently replacing leukemia as the most common cause of cancer mortality in children [1]. Whereas blood sampling can be helpful for testing biomarkers in other solid tumors like prostate and colon cancer, CSF sampling is critical for staging in most pediatric brain cancers. Cytological analysis of CSF is considered one of the best methods for diagnosing metastatic brain cancer [2, 3]; the presence or absence of tumor cells has important implications for treatment and prognosis. Examination of CSF is also useful for evaluating chemotherapy responses in CNS cancer patients [4]. It has easy accessibility through lumbar puncture, has continuous production and recycling similar to blood and lymph, as well as contact with brain tissue and, often, tumor tissue itself [3]. This makes it the optimal body fluid reservoir to find biomarkers relevant to brain tumors [4]. Most clinical use of CSF sampling has focused on the presence or absence of tumor cells, while information on immune cells in CSF is usually ignored. This clinically available information, however, could be a key window into the tumor immune response.

Investigation of immunity and immunological response for diagnosis, prognosis, and therapeutic intervention in pediatric cancer patients is evolving as a major aspect of oncology. For example, immunotherapy such as monoclonal antibody treatment results in significantly improved outcomes and is the standard therapy for neuroblastoma [5]. Additionally, clinical trials have shown that genome-edited chimeric antigen receptor (CAR) T-cells are a useful treatment for patients with aggressive leukemia [6]. CAR T-cells have also demonstrated antitumor efficacy preclinically in pediatric CNS tumors [7]. Patient-specific CAR T-cells have now been shown to activate a localized immune response clinically in pediatric CNS tumors, leading to hopeful early results even in aggressive tumors like diffuse midline gliomas [8]. However, there is much to learn when considering how a patient’s tumor and immune system respond to immunotherapy.

Few studies have investigated pediatric CNS tumor immunophenotyping and responsiveness to immunotherapy [9, 10], and no studies have compared WBC counts in CSF between tumor types in pediatric brain tumor patients. We hypothesized that pre-diagnostic factors, tumor type, and post-diagnostic factors would be associated with the WBC populations present in CSF at diagnosis in pediatric CNS tumors. We conducted quantitative analysis of WBC subtypes in CSF of pediatric brain tumor patients. We then analyzed the correlations of WBC makeup with tumor characteristics, components of the CSF, and patient prognosis.

Methods

Study design

We extracted data from the EHR on 534 pediatric patients that were diagnosed with CNS tumors at Children’s Hospital of Colorado from 1998 to 2018. Of these, 115 were removed because they did not have a CSF draw within 60 days of diagnosis and 111 were removed because they did not have a diagnosis of primary interest. Finally, 39 were excluded for analysis after removing those with missing covariate data (Fig. 1—Flowchart of patients according to the applied inclusion and exclusion criteria).

Fig. 1.

Flowchart of study subjects with applied inclusion and exclusion criteria

Data collection

The CSF samples were collected by lumbar puncture (LP) within 60 days of initial tumor diagnosis. The sample closest to diagnosis within this timeframe was used for each patient, such that each patient only had one sample included in the study. Total nucleated cell counts and percentages/absolute counts of white blood cell (WBC) subtypes were examined. Patients with CSF red blood cell (RBC) counts of 10,000 or higher were excluded from the analysis, because WBC counts were felt to be unreliable in these cases. The following variables were collected: age at diagnosis, steroid use at the time of CSF draw, presence of tumor cells in the CSF, spatial relationship of the tumor to the ventricles on MRI (ventricular vs. peri/non-ventricular), location of the tumor (supratentorial vs. infratentorial/spinal cord), and tumor subtype. Tumor subtypes included atypical teratoid rhabdoid tumors (ATRT), ependymoma, germ cell tumors, high grade glioma (HGG), low grade glioma (LGG), medulloblastoma, and other embryonal tumors (OET).

Statistical analysis

Summary statistics were reported for demographic and CSF variables as frequency and percentage (n (%)) or mean and standard deviation (mean ± SD). CSF variables were not normally distributed, so median and interquartile range (IQR) were reported. Box and whisker plots were produced for CSF variables, stratified by tumor type.

Zero-inflated negative binomial (ZINB) regression was used to model lymphocytes, monocytes, and macrophages as outcomes. Each outcome was analyzed as a percentage of total CSF and as an absolute count. Univariate models were fit for each outcome, and multivariable models were chosen using stepwise selection with the following candidate predictors: age at diagnosis, on steroids at CSF draw, presence of tumor cells in the CSF, spatial relationship of tumor to the ventricles (Ventricular vs. Peri/Non-ventricular), location of the tumor (Supratentorial vs. Infratentorial/Spinal Cord), diagnosis (ATRT, Ependymoma, Germinoma, HGG, LGG, OET, Medulloblastoma), and death. The models assumed the same set of predictors for the zero portion and non-zero portion of the model. For the outcomes of neutrophils (both percent and absolute count) and total nucleated cells, the zero-inflated models did not converge. Instead, we were only able to model whether there were zero neutrophils or zero nucleated cells (binary outcome). The same modeling approach as above was employed, but logistic regression was used instead of ZINB. Logistic regression was also used to model the presence of tumor cells in CSF (Y/N). Univariate model results are excluded for conciseness.

Time-to-death was defined as time from diagnosis to death, with patients censored at the last follow-up date. Univariate cox proportional hazards (Cox PH) models for time-to-death were fit for patient and CSF characteristics. Each of three WBC subtypes (lymphocytes, monocytes, and neutrophils) was forced in as an explanatory variable in a separate multivariable Cox PH model for time-to-death. Both absolute non-zeroes and percentages were considered. The models were all adjusted for clinical characteristics, selected by stepwise selection. R version 4.0.2 was used for all analysis, and the significance level was set to 0.05.

Results

Demographics

Data for patients with documented tumor location were retrieved from 269 pediatric patients (Table 1). The mean age at diagnosis was 8 years (SD 5.8). The three most common tumors were medulloblastoma in 36% of patients, ependymoma in 17% of patients, and germinoma in 16% of patients. Approximately 59% of tumors were infratentorial, 39% were supratentorial, and 2% were in the spinal cord. At the time of CSF sampling, 27% were receiving steroids.

Table 1.

Demographic information for patient population used in study

Characteristic	All (n = 287)	Complete cases* (N = 269)
Age at Dx, years	7.9 ± 5.7	8 ± 5.8
Diagnosis category
Medulloblastoma	105 (37%)	96 (36%)
Ependymoma	46 (16%)	45 (17%)
Low-grade glioma	33 (11%)	32 (12%)
Germinoma	42 (15%)	42 (16%)
ATRT	20 (7%)	18 (7%)
Other embryonal tumor	21 (7%)	17 (6%)
High-grade glioma	20 (7%)	19 (7%)
Grade category
High	254 (89%)	237 (88%)
Low	33 (11%)	32 (12%)
Tumor location
Infratentorial	167 (60%)	158 (59%)
Spinal cord	6 (2%)	5 (2%)
Supratentorial	106 (38%)	106 (39%)
Relationship to ventricles
Non-ventricular	24 (9%)	24 (9%)
Periventricular	113 (42%)	112 (42%)
Ventricular	135 (50%)	133 (49%)
On dexamethasone at CSF draw
No	212 (74%)	197 (73%)
Yes	75 (26%)	72 (27%)
Death
No	211 (74%)	202 (75%)
Yes	76 (26%)	67 (25%)

^*Complete is defined as patients with known tumor location. This cohort was used for the multivariable models

Pre-testing factors

The effect of clinical determinants from before CSF testing (pre-diagnostic factors of age, tumor location, and steroid use) on CSF WBC results were analyzed, and the results from multivariable regression are presented in Tables 2 and 3. After controlling for the presence of tumor cells, death, ATRT diagnosis, and LGG diagnosis, older age was associated with increased percentage of lymphocytes (Table 2). In addition, older age was associated with decreased percentages of monocytes and macrophages. The location of tumors was also associated with WBC characteristics. Ventricular tumors were less likely to have zero absolute lymphocytes, monocytes, macrophages, and total nucleated cells compared with extraventricular tumors (Table 3). Tumors located in the supratentorial region had significantly lower absolute counts of lymphocytes and monocytes. The patients on steroids at the time of the CSF draw were found to have lower absolute counts of monocytes and macrophages.

Table 2.

Multivariable outcomes of WBC percentages for pre-diagnostic factors, tumor types, and post-diagnostic factors

Table 3.

Multivariable outcomes of WBC absolute counts for pre-diagnostic factors, tumor types, and post-diagnostic factors

Tumor subtypes

The seven most common pediatric tumor subtypes were included in our study. Lymphocytes were the most common WBC type across all tumor types, and monocytes were the second most common (Fig. 2). Germinoma had the highest percentage of lymphocytes of any tumor type, followed by medulloblastoma (Fig. 2).

Fig. 2.

Percentages of WBC types in each tumor subtype at CSF collection (box represents median and interquartile range; whiskers represent 1.5× the interquartile range, unless no values exceed this distance, in which case they represent the minimum and maximum values)

Both germinoma and LGG had a higher likelihood of having an absolute macrophage count of zero compared to other tumors (Table 3). Ependymoma showed a higher likelihood of the presence of tumor cells (Table 3) and a lower percentage of macrophages compared to other types of tumors (Table 2). The CSF in patients with HGG also showed a higher percentage of monocytes and a lower percentage of macrophages compared to other tumor types (Table 2). Patients with HGG also had lower absolute macrophage counts compared to other tumor types (Table 3). Patients with LGG had a higher risk of having zero absolute lymphocyte counts and zero total nucleated cell counts compared to other tumor types (Table 3).

There were no statistically significant differences in WBC counts and percentages on multivariable analysis for OET. Those with medulloblastoma had lower risk of having a zero percentage of neutrophils and zero absolute counts compared to other tumor types (Table 2). Finally, after controlling for age, tumor location, and presence of tumor cells, patients with ATRT had higher absolute lymphocyte counts compared to other tumor types.

Post-testing outcomes

Presence of tumor cells in the CSF and mortality were the two outcome measures studied following CSF testing. We discovered that patients who eventually died of their tumor had a lower lymphocyte percentage, a higher absolute monocyte count, and a higher likelihood of tumor cells in CSF. Multivariable analysis also showed that the presence of tumor cells in CSF was associated with a higher likelihood of having zero percent lymphocytes and monocytes (Table 2).

CSF WBC results in multivariable time-to-death model

Supplemental Tables 1–5 shows CSF WBC results as predictors of time to death in our patient population. On multi-variable analysis, total nucleated cells and absolute counts/percentages of WBC types were not associated with time to death. The hazard of death, however, increased with tumor cell presence in CSF and with diagnoses of HGG, ATRT or OET, and decreased with supratentorial location.

Discussion

The application of CSF as a source to characterize the immune system in brain diseases such as multiple sclerosis is already well established [11]. We wanted to further investigate its utility in CNS tumors and analyzed the CSF to assess differences in WBC profiles between tumor types, as well as based on pre- and post-testing characteristics. Our study showed the macrophage make-up of CSF varied by tumor types. Compared to all other tumor types, Germinoma, LGG, HGG, and ependymoma had lower macrophage counts or percentages. We saw a higher lymphocyte percentage in the CSF of ATRT patients compared to the other tumor types. On the other hand, LGG was more likely to have fewer lymphocytes in the CSF.

Immune cells in CSF are helpful in discerning the cellular environment of the tumor. It has been shown that WBC infiltration of the CSF, specifically CD8+T cell infiltration, can highlight the exchange between brain tumors and CSF, providing non-invasive ways to predict the response to immunotherapy drugs and study the tumor microenvironment [11]. Understanding the tumor microenvironment is a gateway to advancing immunotherapies. The influence of serum WBCs in certain cancer treatments has been established; for example, in patients with non-small cell lung cancer, those with higher WBC levels following adjuvant chemotherapy benefited from more chemotherapy compared to those with decreased WBC [12].

Although immune cells play a variety of roles in cancer, the role of CSF WBCs in pediatric brain neoplasms is not entirely understood, but they may offer insight into the immune system’s relationship to malignancy. Therefore, we conducted this retrospective analysis of WBC values in CSF to better understand the role of immune reaction in different types of pediatric brain tumors.

Many studies have analyzed the tumor microenvironment in medulloblastoma subgroups. Patel et al. suggested that patients with medulloblastoma may experience tumor-induced systemic immune suppression, resulting in lower counts of WBC [13]. In contrast, other studies have found high macrophage numbers in this tumor subtype. In the sonic hedgehog (SHH) subgroup of medulloblastoma, there was an increased expression of inflammatory genes corresponding to monocytes and macrophages in addition to a significantly increased number of macrophages. The study also reported an association between the location of macrophages and proliferating tumor cells [14]. Furthermore, Teo et al. reported many macrophages in SHH medulloblastomas, suggesting a possible active immune response with these tumors [15]. Our study showed higher neutrophil percentages and absolute counts in patients with medulloblastoma. Neutrophils have been shown to have the ability to lyse tumor cells, but additional research also suggests tumor-associated neutrophils (TANs) may help tumor progression and lead to a worse prognosis [16].

In contrast to medulloblastoma, our analysis showed lower macrophage counts and percentages in HGG compared to other tumors. HGG tissue has been described as having hypoxic and immunosuppressive microenvironment [17]. Our CSF analysis found that both LGG and HGG had lower CSF WBC counts overall and were more likely to have zero counts of macrophages compared to non-glioma tumor types, suggesting a muted immune response in the tumor microenvironment. However, our results showed that HGG was associated with a high percentage of monocytes. One study investigated the role of the immune response in pediatric HGG and found the amplification of immune response-related genes and genes associated with monocytes in these patients [18]. This study also found an increase in the expression of immune response-related genes and macrophages in the mesenchymal subtype of pediatric glioblastoma (GBM). Chen et al. reported that the most abundant non-neoplastic cells in the GBM microenvironment are infiltrating monocytes and macrophages [19]. Circulating monocytes in HGG may have a role in immune suppression and tumor expansion. Fries et al. stated that subtypes of infiltrating monocytes might stimulate immunosuppression and promote the expansion of glioma by releasing abnormally high levels of epidermal growth factor (EGF) [20]. The elevated monocyte percentage in our study may also contribute to possible HGG expansion.

Based on immunological profiling, pediatric glioma can be classified into three categories: immune hot (IS-I), immune altered (IS-II), and immune cold (IS-III) [21]. In this study, Wang et al. found that HGGs exhibit low immune activity and are more likely than LGGs to be classified as IS-II or IS-III tumors [21]. In contrast to these studies, we found fewer WBC in the CSF of LGG patients compared to other tumor subtypes, including HGG. This may suggest a less robust immune response with less aggressive tumors. The contradictions in these findings may be due in part to the variations in data types. Our study used quantitative data derived from WBC absolute counts and percentages, compared to transcriptomic data in the Wang et al. study.

Studies have found that increased numbers of lymphocytes were associated with a favorable prognosis in patients with ependymoma, suggesting an antitumor role for lymphocytes. Studies of macrophages have been mixed: In one study, increased tumor-associated macrophages were associated with a dismal prognosis in ependymoma patients [22], while Yamada et al. found that several genes associated explicitly with the activity of macrophages correlated with good outcomes in ependymoma [23]. In our analysis, ependymoma CSF showed a lower percentage of macrophages compared to other tumor types.

Pediatric patients with ATRT had relatively higher absolute lymphocyte counts in CSF compared to other tumor types in our study. Lu and colleagues examined tumor tissues from four patients with ATRT and compared those with four patients with GBM. The results showed that infiltration of lymphocytes was significantly higher in ATRTs than in GBM [24], suggesting a more robust immune response that may make these tumors amenable to immunotherapy.

Zapka and colleagues examined tumor tissues from 28 patients with germinomas and found massive immune cell infiltration, including macrophages [25]. Our CSF data contrast with these findings: Germinoma was associated with lower counts of macrophages in our analysis compared to the other tumor types. This contradiction may be attributed to CSF versus tumor tissue differences, as well as to our sample size of 42 patients with germinomas compared to 28 in the Zapka et al. analysis.

In our study, malignant tumor cells were associated with lower lymphocyte and monocyte percentages after adjusting for tumor type, suggesting that metastatic tumor cells may have an immune suppressive role in the CSF. When considering location, tumors contacting the ventricles may be more likely to stimulate an immune response in the CSF, based on the higher lymphocyte counts we observed for tumors located ventricularly. Patients on steroids at the time of the CSF draw were found to have lower absolute counts of monocytes and macrophages, consistent with the anti-inflammatory role of steroids known to hinder tumor immunotherapy [26].

To our knowledge, this is the first study to investigate the relationship between tumor-specific WBC profiles in CSF and various pediatric brain tumor subtypes. Further, we assessed the association between the presence of tumor cells, WBC characteristics, tumor proximity to ventricles, and disease prognosis. Another advantage of the study is the single-institution nature, since a single lab favors consistency and uniformity of data. However, we acknowledge some limitations. Information bias or selection bias could not be controlled due to the study’s retrospective nature. A small sample size may influence the quality of evidence (i.e., low statistical power, inflated effect size estimation, and low reproducibility). Adjustment for multiple comparisons was not performed due to the exploratory nature of the study, and we acknowledge a larger scale study is needed to confirm and refine these findings.

In summary, CSF analysis is a diagnostic tool that can help guide prognostic and therapeutic management in CNS tumors, and our study raises inclusion of the WBC profile as a component of that analysis. Different types of pediatric brain tumors varied in their CSF WBC profiles, offering possible insight into the role and type of immunotherapy that may be useful in various pediatric CNS tumors. By giving an easily accessible window into a tumor’s microenvironment and local immune response, CSF WBCs may offer useful data on prognosis and treatment options. Future studies studying immune-related gene expression, more detailed WBC subtyping, and direct correlation of CSF WBCs with the local tumor microenvironment should be considered as part of immunotherapy and other clinical trials. This will allow a better understand the specific relationships between tumor subtypes and the complex immune responses that they elicit, creating a basis for understanding how tumor immunotherapy will interact with the body’s own tumor immune response.

Data availability

The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.