No evidence for depletion of circulating lymphocyte populations in primary brain tumor patients receiving radiotherapy alone

Abstract

Purpose:

To investigate the impact of radiotherapy (RT) for brain tumors on circulating lymphocyte populations and understand this from a dosimetric perspective.

Methods:

This prospective study enrolled primary brain tumor patients treated with either photon or proton RT. To specifically study the effect of RT, only patients without concurrent chemotherapy were eligible. For each patient, blood samples were collected before, during and after RT, as well as at the first follow-up, for flow cytometry analysis of blood cells. We used a stochastic model to compute blood dose distributions from treatment plans and patient-specific blood flow simulations. Dose to the lymph nodes in the head and neck region was also assessed. Dose distributions and lymphocyte trends from patients treated with photon versus proton therapy were compared.

Results:

Thirty-five patients with a median age of 50 were included: 27 patients with benign tumors (mostly meningiomas) and 8 with malignant tumors (mostly gliomas). There were no significant RT-induced changes in either the total lymphocyte count or its subpopulations in this patient cohort. In line with this finding, dosimetric analyses showed that total blood doses were ~0.1-0.2 Gy and that most lymph nodes in the head-and-neck area are spared. Proton RT was associated to a greater dosimetric sparing of both blood and lymph nodes.

Conclusion:

In this exploratory analysis, we found no evidence that RT alone results in significant changes in peripheral lymphocyte counts in primary brain tumor patients treated to limited intracranial volumes. The lack of apparent RT-induced cytotoxicity is consistent with the limited dose received by either the blood or the regional lymph nodes.

Introduction

Radiation-induced lymphopenia (RIL) has been observed after irradiation of patients with solid tumors in a variety of treatment sites, including the brain, head and neck, thoracic, abdominal and pelvic regions, and is associated with inferior outcomes, though the strength of this association varies by tumor site ^1,2. The occurrence of RIL is often attributed to lymphocyte cell death due to irradiation of lymphocytes in the blood circulation³. Other studies have emphasized the importance of limiting dose to the secondary lymphoid organs (SLOs) such as the lymph nodes (LNs) or the spleen⁴ to avoid RIL, while still others point to dose to the bone marrow (BM) as a major contributor to RIL⁵. Most treatment sites result in substantial irradiation of more than one of these compartments, making it difficult to assess the relative contributions.

For external beam radiotherapy (EBRT) of the brain, some of the factors contributing to RIL can be ruled out. Skull irradiation may cause local suppression of functional activity of BM⁶, but this is expected to have negligible effect on the systemic lymphocyte counts due to the small amount of active BM present in the skull⁷. In addition, although the brain does not contain classical LNs and has only limited conventional lymphatic structures⁸, it does possess a glymphatic system that facilitates clearance of interstitial waste products from the central nervous system⁹. Importantly, this glymphatic pathway is not a conduit for circulating lymphocytes and therefore does not represent a major reservoir or trafficking route of these cells within the brain. Therefore, it has been argued that lymphopenia after cranial irradiation is caused by the direct cytotoxic effect of ionizing radiation on lymphocytes circulating in the blood^10,11. Studies have shown that treatment field volume matters, with larger fields and less conformal techniques increasing lymphopenia risk, while smaller target volumes and proton therapy reduce it^12-17. These findings have been interpreted in support of the hypothesis that RT induces lymphopenia in brain tumor patients by killing circulating lymphocytes since larger treatment fields result in larger volumes of blood irradiated¹⁸. However, blood dose does not equal lymphocyte dose as it ignores the recirculation of lymphocytes between the blood and lymphoid systems¹⁹. Considering the extent of this recirculation and the estimate that only a few percent of the recirculating lymphocyte pool is in the blood at any given time, the hypothesis that the cytotoxic effect of blood irradiation is the major mechanism causing RIL has been questioned²⁰. The impact of lymphocyte recirculation on the dose it receives during RT of glioblastomas has been evaluated by extended existing blood dose tools^21,22 to incorporate lymphocyte exchange between the blood and other lymphoid organs²³. However, calculated lymphocyte doses were found to be insufficient to explain the severity of RIL reported in clinical studies^24,25. Various mechanistic models have been proposed to better understand RIL mechanisms and identify patients at risk, but the dominant biological pathways remain unclear²⁶. Preclinical results in mice suggest that sparing cervical lymph nodes during brain irradiation can mitigate lymphopenia²⁷, highlighting the role of these lymphoid structures. Recently, we developed a mathematical model to simulate whole-body lymphocyte recirculation. Accounting for the complex interplay between lymphocyte migration and heterogeneous radiation dose distributions throughout the body, we were able to predict lymphocyte dose distribution and cell survival induced by preclinical irradiation studies in murine animals. Our analysis reaffirmed the importance of lymphoid structures and indicated that the severity of clinically observed lymphopenia cannot be attributed solely to the direct killing of circulating lymphocytes during irradiation²⁸.

In addition, lymphocyte subpopulations exhibit differential radiosensitivity, which may contribute to selective depletion patterns following irradiation. While B cells consistently demonstrate greater radiosensitivity than T cells^29-31, findings on other subsets are more variable: CD4+ and CD8+ T cells show similar radiosensitivity³², though regulatory T cells appear relatively radioresistant³⁰. Natural killer (NK) cells have been reported as both highly radiosensitive²⁹ and relatively resistant³⁰. These differences in radiosensitivity suggest that RT effects on lymphocytes are best captured through subset-level analysis rather than aggregate counts alone.

The clinical studies that demonstrated lymphopenia after cranial irradiation focused on high-grade glioma patients receiving concurrent chemotherapy^33-36. However, chemotherapy such as temozolomide independently causes severe lymphodepletion^37-39, potentially aggravated by concurrent corticosteroids⁴⁰. Finally, even untreated gliomas induce systemic T cell lymphopenia by disrupting lymphocyte trafficking and sequestration into the BM⁴¹. Nevertheless, a study comparing sequential versus concurrent RT and chemotherapy showed that concurrent partial-brain RT worsens chemotherapy-induced lymphopenia⁴². In contrast, low-grade brain tumor patients treated with RT alone did not experience noteworthy lymphopenia⁴³.

To clarify RT's independent role in lymphopenia, we analyzed lymphocyte subpopulations in a cohort of 35 patients with primary brain tumors receiving RT without concurrent chemotherapy. Lymphocyte subpopulation analysis may reveal 1) effects masked by total lymphocyte count, such as preferential depletion of CD4+ T cells as seen in other settings ^33,34,36,39; and 2) patterns of selective depletion that provide mechanistic insights into RT effects on different lymphocyte populations. Furthermore, we stratified analyses by modality – photon versus proton radiotherapy – to ensure that potential modality-specific effects were not masked in aggregate analysis, given that proton therapy’s superior conformality may reduce low-dose exposure to circulating lymphocytes⁴⁴. We coupled this with detailed dosimetry to blood and lymph nodes to examine potential correlations with changes in lymphocyte composition at multiple time points during RT.

Materials and methods

Eligibility criteria

Following IRB approval, primary brain tumor patients treated at our institution with either photon or proton therapy without concurrent chemotherapy were prospectively enrolled after providing written consent. Additional eligibility criteria were a Karnofsky Performance Status (KPS)>50 and age 18-80 years. Patients with uncontrolled intercurrent illness were excluded.

Blood Sample Collection and Processing

Peripheral blood samples were collected and analyzed at four timepoints: before the start of RT treatment (BT), mid-treatment (MT) (after one week), at the end of treatment (ET), and at follow up (FU). Peripheral blood was collected in EDTA tubes. Complete blood counts were performed at each timepoint using a Coulter counter. For plasma isolation, samples were centrifuged at 2000 rpm for 20 minutes at 4°C. Plasma was aliquoted and stored at −80°C. For mononuclear cell isolation, blood was transferred to Cell Preparation Tubes (CPT) and centrifuged at 1,600g for 25 minutes at room temperature. The mononuclear layer was collected and resuspended in media (RPMI-1640 with 20% DMSO) for cryopreservation.

Immune Phenotyping (Flow Cytometry)

Lymphocyte subsets in peripheral blood samples were analyzed by flow cytometry using a multi-color antibody panel. Residual erythrocytes were lysed using ACK lysing buffer, and samples were filtered through a 40μm cell strainer to maintain a single-cell suspension prior to acquisition. Cells were incubated with FcR blocking reagent to minimize non-specific binding, followed by surface marker staining with fluorochrome-conjugated antibodies against CD3, CD4, CD8, CD25, CD127, CD45RA, and PD-1 for 15–20 minutes in the dark. Data were acquired on a BD LSRII flow cytometer, and samples were fixed in 1% paraformaldehyde (PFA) before acquisition to ensure stability. Debris was excluded using forward (FSC-A) vs side scatter (SSC-A), followed by FSC-A vs FSC-H and SSC-A vs SSC-H gating to select single cells. Remaining cells were gated for CD3 and CD56 expression to identify CD3+ T cells, CD3-CD56+ NK cells, and CD3+CD56+ NKT cells. CD3+ T cells were further subdivided based on CD4 and CD8 expression, and percentages of CD25+ (activation/regulation), CD45RA+ (naïve marker), CD127+ (memory/survival), and PD1+ (activation/exhaustion) cells were quantified within each subset. The gating strategy is depicted in Figure S1.

Fitting time series data

Since blood counts represent sparse measurements of an underlying continuous process, we fitted a Gaussian Process (GP) model to each patient's time series. This approach is superior to using raw data directly since both biological fluctuations⁴⁵ and measurement noise⁴⁶ can be considerable, creating spurious extrema. GPs naturally resist overfitting to outliers when data is sparse by favoring smoother trajectories when evidence is limited, while denser sampling enables the model to capture more complex dynamics with higher confidence. Technicalities about the fitting process are summarized in the Supplementary Materials.

We assumed measurement noise σ with σ/β ≈ 12.5%⁴⁶ where β(t) denote the mean GP predictions (of absolute lymphocyte counts or subsets thereof). Meaningful deviations from baseline during RT were defined by the condition ∣μ(t) − μ(0)∣ > 2σ (i.e. a deviation of twice the noise level) within the first 50 days after the start of treatment.

Simulating radiation dose to blood.

Patient-specific blood doses were calculated using an extended version of HEDOS, a previously developed compartmental model of blood flow onto which the dynamics of the dose delivery is superimposed^21,22. Blood doses are represented by differential dose volume histograms (DVHs). As the blood dose may depend on whether one of the larger cranial blood vessels are being irradiated, a more fine-grained result was achieved by applying female and male phantoms of brain vasculature⁴⁷. Blood flow was explicitly tracked through these phantoms and seamlessly integrated in the larger compartmental model. The appropriate phantom vasculature was then mapped to each patient anatomy to construct patient-specific vasculatures. To this end, a diffeomorphic registration method was developed which matches the outer surfaces of the phantom and patient brain, thereby preserving the internal topology of the vasculature, figure 1. More details can be found in Beekman et al.²².

Figure 1.

Cranial blood flow simulations in the patient reference frame. A: Phantom vasculature inside the phantom mesh; B: Patient reference frame; C: Vasculature in the patient anatomy; D: Snapshot of blood flow simulation with blood particles colored by their current dwell time.

Radiation dose to lymph nodes.

While the brain itself does not have LNs, nodal groups in the superior cervical neck may receive significant collateral radiation dose. LN dose assessment requires segmentation. We only considered superior cervical LNs, as these are the only LNs which possibly receive a non-negligible dose, whereas other nodal groups would be too far removed from the high dose region. However, these LNs are typically not delineated in patients treated for brain tumors. As an efficient workaround, a template patient was constructed based on the guidelines outlined by Grégoire et al.⁴⁸. Subsequently, we registered the template subject to each of the patients in our cohort by matching bony anatomy in the H&N region. This yielded visually satisfactory segmentation propagation since the H&N LNs closely follow bony anatomy (Fig 2.). Using these LN segmentations, the dose distribution for each patient was calculated and combined into an overall H&N LN dose distribution.

Figure 2.

LN regions segmented through registration of a template. Results of two patients are shown with the radiation plan overlaid. This slice shows nodal levels VIII, IX and Xa.

Group statistics of dose volume histograms

Typically, dose distributions are collapsed into some point estimate – e.g. the mean – for statistical analysis, but the choice of point estimate lacks clear justification. Therefore, often a range of dosimetric quantities, such as V_x (volume receiving at least x Gy of dose) or D_y (dose to y-th volume percentile) is scanned over. However, this ignores the autocorrelation in both V_x and D_y, and their cross-correlation with each other, potentially leading to Type I errors. Since the fundamental issue stems from testing multiple derived metrics from the same underlying distribution, a more principled approach would be to consider the entire dose distribution thereby avoiding the multiple comparisons problem and preserving the full dosimetric information.

Statistical hypothesis testing requires a test statistic to quantify group differences. While this is straightforward for scalar data (e.g., difference in means), generalizing this to distributional data is non-trivial. The mean for a collection of distributions must be defined, as well as a notion of distance or difference between distributions. Furthermore, differences may manifest not only in the location of distributions, but also their shape, spread or other characteristics.

Here we use dose quantile averaging to define the mean dose quantile of a group. To quantify differences between dose distributions, we use the L₂ distance between their quantile functions, also known as the Wasserstein-2 metric⁴⁹:

$$d_{12}={\left({\int }_0^1{\mid Q_1(u)-Q_2(u)\mid }^{2}du\right)}^{1∕2}$$

Motivation for this choice is included in the Supplementary Materials.

With this metric choice established, we can perform group statistics based on the entire dose distributions. To this end, we used a non-parametric multivariate ANOVA test introduced by Anderson⁵⁰ in which the metric choice is crucial to calculate the inter and intra group variances. Since p-values are computed non-parametrically using permutations, this test is also referred to as PERMANOVA.

Besides overall hypothesis testing, the dose regions over which significant differences occur are of interest. We propose simultaneous confidence bands (SCB), a technique from functional data analysis, where bootstrap samples of the group means are used to create a confidence band around functional data differences. This solves the multiple testing problem by providing a single band that controls error rates across all points simultaneously rather than testing each point individually.

We center the confidence band around one group mean to visualize where and how much the mean functions differ, and when this is statistically significant⁵¹. Since we use a W₂ metric that corresponds to L₂ distance between quantile functions, we compute SCBs directly in quantile space.

Statistical analysis

Statistical comparisons of scalar valued distributions between groups – e.g. protons vs photons – were performed using Mann-Whitney U tests. This applied to dose means, PTV volumes, and lymphocyte count data at a single timepoint. The significance threshold was set to a p-value of 0.05. To statistically determine whether lymphocyte counts deviated from baseline at the population level, we computed the normalized difference between each timepoint and baseline (t=0) for each patient’s GP trajectory, i.e. the fold change (y_t − y₀)/y₀. For markers expressed as a percentage of the parent population, we computed the difference with respect to baseline in percentage points. We then used bootstrapping to construct 95% SCB (see above) around the population mean. Statistically significant deviations from baseline correspond to timepoints where the SCB excludes zero.

Results

Patient cohort

Thirty-five patients with a median age of 50 were enrolled in this study. The majority of patients had benign tumors (77%), predominantly meningiomas (81% of benign cases). Malignant tumors were predominantly gliomas (88%). Patients with malignant brain tumors received adjuvant chemotherapy after completion of RT. In almost all patients (91%), the primary tumor was surgically removed prior to RT. Proton therapy was used for 26 patients (23 pencil beam scanning, 3 passive scattering), and photon therapy (volumetric modulated arc therapy) was used for 9 patients. A complete set of blood draws (at all 4 timepoints) was obtained for 13 patients, with others missing one or multiple blood draws. Table 1 summarizes patient characteristics overall and by treatment modality; Table S1 stratifies by tumor malignancy.

Table 1.

Brain tumor patient cohort grouped by modality.

Characteristic	All brain tumors n=35	Photons n=9	Protons n=26
Patient Demographics
Sex
Male	11	2	9
Female	24	7	17
Median age (years)	50 (22-75)	63 (33 – 75)	41.5 (22 – 62)
Median BMI (kg/m2)	26.0 (17.5 – 47.3)	25.1 (21.5 – 31.1)	26.5 (17.5 – 47.3)
Tumor Characteristics
Malignancy
Benign	27	8	19
Malignant	8	1	7
Primary location
Skull base	15	6	9
Dural convexity	11	2	9
Parenchymal	9	1	8
Laterality
Right	18	3	15
Left	12	5	7
Central/bilateral	5	1	4
Treatment Characteristics
Prior surgical resection	32	8	24
Radiotherapy
Fractions	30 (28-33)	30 (30-33)	30 (28-33)
Fraction RBE dose (cGy)	180 (170-180)	180 (180-180)	180 (170-180)
Steroid use
During RT only	1	0	1
During FU only	2	1	1
During RT and FU	1	1	0
Anti-epileptic drugs
During RT only	2	0	2
During FU only	0	0	0
During RT and FU	10	2	8
Chemotherapy after RT	8	1	7

BMI=body-mass-index; RBE=relative biological effectiveness; Gy=Gray; RT=radiotherapy; FU=follow up. Values represent counts, except for continuous variables for which the median and range are presented.

Longitudinal lymphocyte counts and composition

Trends of the longitudinal lymphocyte population and its subsets, shown both in aggregate and stratified by treatment modality, are shown in Figure 3a. No clear trend is observable and none of the differences between photons and protons where statistically significant at any timepoint. Figure 3b shows a 2D tSNE embedding of the lymphocyte subpopulation data in different patients and at different time points during treatment. It appears that in general, intra-patient variation is much smaller than inter-patient variation. Consequently, subtle longitudinal deviations from stable lymphocyte population trajectories throughout treatment and follow up are obscured by the more dominant inherent variation among patients.

Figure 3.

a) Absolute lymphocyte count and relative counts of NK, CD4+ T and CD8+ T cells, shown in aggregate and separated by treatment modality. Percentages are defined relative to the total lymphocyte count. b) tSNE-plot of lymphocyte subpopulation data. Each patient is colored differently, and timing is indicated by marker style. Ellipsoids enclose the patient data but for outliers, which are indicated by the dashed lines. BT=before treatment; MT=mid treatment; ET=end treatment; FU=follow up.

Fold changes in absolute cell counts (cells/μL) from baseline over time are shown in Figure 4a. The absolute lymphocyte count (ALC) shows no significant deviation from baseline throughout the observation period. Considering individual patients, only nine patients experienced a nadir (a deviation exceeding the noise level), but no clear commonalities were found between these patients (Fig. S3). NK cells demonstrated an initial increase during RT; however, it was insignificant and the wide SCB suggests this was driven by a few patients rather than representing a consistent population effect. Both CD4+ and CD8+ T cell counts remained stable, with no evidence of radiation-induced changes.

Figure 4.

a) Fold change of lymphocyte counts from baseline values for all patients (left), and stratified by modality (middle) and tumor malignancy (right). b) Change from baseline in the expression levels of indicated markers (rows), expressed as percentage points relative to the parent population (columns). Statistically significant deviations occur when the SCB excludes zero and are indicated by stars in corresponding color. The vertical dashed line represents the mean RT treatment duration.

The subgroup trends should be interpreted with caution. The SCB were constructed using bootstrap resampling, which assumes the sample is representative of the population. This might not hold for the smaller subgroups in which the trends were observed (photon treatment with n=9; malignant tumors with n=8). Additionally, the SCB accounts for population-level variation only. A more conservative approach would be to also propagate the uncertainty arising from both measurement variability and data scarcity (Fig. S4).

Changes in T cell subset composition (expressed as a percentage of the parent population) are shown in Figure 4b. Neither CD4+ nor CD8+ T cells showed significant shifts in the proportion of CD45RA+, CD25+ or PD1+ subsets during or immediately following RT. While the fitted curves suggest potential trends at later timepoints, the limited number of measurements preclude definitive conclusions about long-term changes. Finally, CD127 expression was excluded from the subset analysis as the recorded frequencies (<2%) were inconsistent with established physiological ranges for peripheral blood T cells, suggesting a technical limitation.

Dosimetric results

The planning target volume (PTV) size tended to be larger for proton than for photon treatments (mean PTV size of 105.6 cc vs 62.5 cc, p=0.19). Nevertheless, brain doses tended to be higher for photons, which was significant in the low-dose regime below 5 Gy (Fig 5, left). The better low dose sparing of protons also translated in more sparing of the cervical LNs (Fig 5, middle) as well as lower accumulated blood doses (Fig 5, right). In Table S2 statistics on dose means are given for completeness.

Figure 5.

Proton vs photon dose analysis of the brain (left), the LNs (middle), and the blood (right). The 95% SCB centered around the photon group mean (green line) is shown in green shading. Gray regions indicate statistical significance (p<0.05) where the proton group mean (black line) falls outside the SCB. Distribution-wide p-values from PERMANOVA are shown in the top left corners.

Discussion

We found no evidence for substantial RT-induced lymphopenia in primary brain tumor patients without concurrent systemic therapy. We performed lymphocyte subset analysis in a group of 35 primary brain tumor patients with measurements at four timepoints: before-treatment, mid-treatment, end-treatment and follow-up. Neither the ALC nor any of the lymphocyte subsets showed evidence of a clear depletion trend during RT, suggesting that RT of brain tumors has, in the absence of concurrent chemotherapy, a very limited effect on lymphocyte populations. Stratification by treatment modality and tumor malignancy was performed to examine whether radiation-induced changes were masked when analyzing the cohort as a whole. However, the statistical uncertainty surrounding these observations due to small subgroup sizes and sparse temporal sampling, warrants caution for overinterpretation. To further investigate this finding that somewhat contrasts to the reported literature, we explored the dosimetry further. To this end, novel approaches to accurately calculate H&N lymph node and blood doses were established. Additionally, since our dataset consisted of both patients treated with photons and with protons, we asked ourselves whether these different treatment modalities result in appreciable dosimetric differences which may obscure the signal if analyzed together. We borrowed ideas from optimal transport and functional data analysis to allow for the comparison of the entire dose distributions between groups, rather than assessing an arbitrarily chosen point estimate. Further, modeling lymphocyte dynamics with Gaussian Process regression is a principled method to reconstruct continuous trajectories from sparse and noisy clinical timepoints. This addresses a common challenge in oncology where the invasiveness of repeated sampling limits the frequency of longitudinal measurements.

Even though proton treatment plans treated larger targets on average, they were associated with a ~25% lower blood dose than the photon treatments as well as lower LN dose, showcasing their superior low dose sparing to normal tissues. However, we found that RT of brain tumors is generally associated with low blood doses and typically does not lead to substantial LN irradiation, making the clinical relevance of this dosimetric sparing for avoiding lymphopenia questionable. Indeed, no significant lymphopenia was observed in either photon or proton treated patients.

We showed that clinical mean blood doses accumulated over 30 fractions of 2 Gy is of the order of 0.1-0.2 Gy. This is at least an order of magnitude lower than Yovino’s et al. estimation of a mean blood dose of approximately 2 Gy, with 99% of the volume receiving >0.5 Gy¹⁸. In a previous study by partially the same authors, a mean blood dose of about 0.36 Gy was derived for typical cranial RT plans, which is more in line with our results⁴². A quick calculation validates these more conservative blood dose calculations: According to the ICRP 89⁵², the brain contains only about 1.2% of the total blood volume. If the blood were standing still during irradiation and ¼ of the brain was irradiated, 0.3% of the blood would receive a dose of 2 Gy, amounting to a mean blood dose of 0.6 cGy per fraction. With 30 fractions, the mean blood dose will therefore only be 0.18 Gy. While the rapid blood flow through the brain means that a large fraction of the total blood volume will receive some dose, this circulation only affects the shape of the dose distribution, but not the mean.

Even these lower estimates might still lead to radiation-induced death of circulating lymphocytes, albeit mild: assuming a single exponential survival function³¹ SF = e^−0.4·D implies that about 0.24% of the circulating lymphocytes die each fraction. However, the blood pool is estimated to contain only ~2-5% of the mobilizable pool^53,54, with most lymphocytes in the lymphoid organs. Thoracic duct studies in humans show that the daily influx of lymphocytes into the blood pool is at least 3 times more⁵⁵, so that cell loss incurred in the blood is effectively diluted over the total recirculating lymphocyte pool. Consequently, the effective fractional lymphocyte depletion in the blood is 20-50-fold lower than predicted from blood dose. This amounts to a predicted depletion of about 0.01% per fraction, or 0.3% over the full RT regimen relative to baseline. This is practically undetectable and clinically irrelevant.

The above analysis tacitly assumes that all lymphocytes in the blood actively recirculate. In reality, only lymphocytes expressing CCR7 and CD62L (mostly naïve and central memory cells) enter lymph nodes effectively, while cells with a more effector phenotype (effector memory and terminally differentiated cells) primarily remain in blood and peripheral tissues⁵⁶. Consequently, effector populations may be more affected by blood dose, while naïve or central memory cells may be more affected by lymph node dose.

Taken together, it remains unclear why RT aggravates lymphopenia in glioma patients. Possibly, the locally delivered high-dose RT may cause significant biological changes such as altering the integrity of vessel endothelium, cytokine profiles, or the expression of trafficking molecules in the irradiated tissue. Wijerathne et al. showed that irradiation of rat brains leads to locally enhanced adhesion of lymphocytes to the irradiated vessel endothelium and subsequent extravasation⁵⁷. This would result in an increased net lymphocyte flux from the blood into tissue and effectively act as a lymphocyte sink. While this is only speculative, it is good to realize that a small change in lymphocyte trafficking patterns between the blood and other lymphoid tissues could substantially affect lymphocyte counts in the blood.

A limitation of this study is the relatively small patient cohort. This was a consequence of the prospective nature of the study as longitudinal flow cytometry of blood samples is not routinely performed in standard clinical practice. To contextualize this limitation: with a coefficient of variation (CV) of ~30% in baseline ALC (mean ~1500 cells/μL), our cohort of N=35 provided ~80% power to detect a clinically meaningful decline of 15% or more from baseline (Cohen's d≈0.5, α=0.05, two-sided). While smaller changes cannot be excluded, the observed stability of lymphocyte populations throughout RT is in concordance with our dosimetric analysis, which showed that the dose to blood and lymphoid organs resulting from typical RT treatments of brain tumors is very small.

Conclusion

Conformal RT of primary brain tumors result in very low doses to the blood and spares most neighboring LNs. While proton treatments are associated with lower blood and LN dose compared with photon treatments, neither modality results in clinically relevant lymphocyte irradiation. In line with this, blood lymphocyte populations in a prospective cohort of primary brain tumor patients who did not receive chemotherapy prior to or during RT showed no evidence of substantial change during RT treatment.

Data Sharing.

The datasets generated and analyzed during the current study are not publicly available due to patient privacy regulations and institutional data protection policies, but anonymized data are available on reasonable request and subject to appropriate data sharing agreements and ethical approval.