Immunophenotyping reveals longitudinal changes in circulating immune cells during radium-223 therapy in patients with metastatic castration-resistant prostate cancer

Patient recruitment

This prospective, single-arm, exploratory study included mCRPC patients treated with radium-223 at the Radboudumc and Canisius-Wilhelmina hospital, Nijmegen, The Netherlands between May 2016 and January 2018. Patients were eligible if they had histologically proven adenocarcinoma of the prostate, mCRPC with symptomatic bone metastases, and no (history of) visceral metastases. Patients with a history of a secondary malignancy or auto-immune disease were excluded. Prior radionuclide treatment and concomitant other anticancer treatments were not allowed except for luteinizing hormone-releasing hormone agonists or antagonists. The use of low-dose corticosteroids (maximally 10 mg prednisone daily) was permitted.

Written informed consent was obtained from all patients before enrolment. The study was conducted in accordance with the principles of Good Clinical Practice, the Declaration of Helsinki, and other applicable local regulations and was approved by the institutional review boards of both participating centers.

Radium-223 therapy and study-specific procedures

Patients were treated with radium-223 according to daily practice, with radium-223 being administered intravenously every four weeks at a dose of 55 kBq/kg, with a maximum of six injections. Laboratory evaluations, including lymphocyte and monocyte counts, were carried out within four weeks before radium-223 initiation, one week before each subsequent injection, and four weeks after the end of treatment. Within three months before the start of radium-223 therapy, bone scintigraphy, and CT or ⁶⁸Ga-PSMA-11 PET/CT of the thorax, abdomen, and pelvis were performed. At baseline and four weeks after the second, fourth, and sixth radium-223 injection, 30ml of blood was collected in heparin blood collection tubes for immunophenotyping.

Blood processing and storage

PBMCs were isolated using Ficoll-Paque gradient centrifugation. After adding Ficoll (Lymphoprep, Axis-Shield, Dundee, UK), samples were centrifuged at 2100 x g for 20 minutes at room temperature. The PBMC layer was transferred to a new tube and washed with phosphate-buffered saline (PBS). Viable cells were counted manually with a Bürker-Türk counting chamber. Cells were resuspended in culture medium (X-VIVO + 2% human serum) and mixed 1:1 with freezing medium (20% DMSO, 80% human serum albumin) at a concentration of 2-20 x10⁶ viable cells per ml. Cells were aliquoted into cryogenic vials and stored in liquid nitrogen.

Flow cytometry

PBMCs were thawed rapidly in a 37 °C water bath and diluted in RPMI 1640 media. Cell number and viability were determined with a hemocytometer using trypan blue. Immunophenotyping of PBMCs was performed using four antibody panels. We used a T cell panel consisting of antibodies against CD3, CD4, CD8, CD25, CD45RO, CTLA-4 (CD152), CCR7 (CD197), and FoxP3 (Supplementary Table S1). In addition, we developed two checkpoint panels. Checkpoint panel 1 included antibodies targeting CD3, CD4, CD8, LAG-3 (CD223), PD-L1 (CD274), ICOS (CD278), PD-1 (CD279), and TIM-3 (CD366). Checkpoint panel 2 consisted of a CD3, CD4, CD8, OX-40 (CD134), CTLA-4, PD-L1, PD-1, and TIM-1 (CD365) antibody mix. Lastly, the MDSC panel consisted of antibodies against lineage markers (CD3, CD19, CD20, CD56), CD11b, CD14, CD15, CD33, PD-L2 (CD273), PD-L1, and HLA-DR. Antibody specifications are tabulated in Supplementary Table S1. In all panels, fixable viability dye efluor 780 (eBioscience, San Diego, CA, USA) was used to exclude dead cells.

Cells were incubated with fixable viability dye diluted in PBS at 4 °C for 30 minutes. Subsequently, cells were incubated with an antibody mix, consisting of the cell surface markers diluted in brilliant staining buffer (BD Biosciences, San Jose, CA, USA). The cells were incubated at 4 °C in the dark for 30 minutes. For intracellular staining, cells stained with the T cell panel were fixed with Fix/Perm (eBioscience) and incubated for 2 hours at 4 °C. After washing, the cells were resuspended in permeabilization buffer containing antibodies against the intracellular markers anti-FoxP3 and anti-CTLA-4 and incubated for 30 minutes at 4 °C.

Staining intensity was measured with the FACSLyric (BD Biosciences). Instrument settings were verified and adjusted before each acquisition using single stainings. Data were analyzed with FlowJo Software (Tree Star Inc., Ashland, OR, USA). Positive and negative cell populations for each marker were determined using fluorescence minus one (FMO) controls. The gating strategy is shown in Supplementary Figure S1. Regarding checkpoint expression, both the percentage of positive cells and the median fluorescence intensity (MFI) were studied. To minimize the effects of intra-batch differences, we used the ΔMFI, which is defined as the MFI of the test sample minus the MFI of the negative FMO control. Checkpoint molecules expressed on <1 % of the cell of interest were excluded from further analyses.

Clinical response evaluation

Clinical response was defined as a decline in alkaline phosphatase (ALP) levels of ≥30% from baseline during therapy, according to the ALSYMPCA study criteria [1]. In addition, PSA responses, radiological responses, and OS were analyzed. PSA response was defined as a decline of ≥30% during treatment according to the ALSYMPCA study criteria. Within three months after completion or discontinuation of therapy, patients underwent radiological evaluation by bone scintigraphy and CT of the thorax, abdomen, and pelvis. Radiological evaluation of soft tissues was performed according to Response Evaluation Criteria In Solid Tumors version 1.1 [10], and bone scans were evaluated according to Prostate Cancer Working Group 3 criteria [11]. OS was defined as the time between the first radium-223 injection and either death from any cause or the last follow-up. All patients were followed until death or August 1, 2020.

Data analysis

Exploratory data analyses were performed using R version 3.6.2. The following packages were used for analyses and visualization: the Tidyverse collection of packages [12], scales [13], cowplot [14], and ggbeeswarm [15].

Lymphocyte and monocyte counts were normalized to adjust for interindividual variation. Cell counts were normalized per patient as follows: the patients’ mean count per cell type was subtracted from the measurement, and, subsequently, the overall mean of the count of that particular cell type was added. Linear regression was used to determine the average linear change of the cell counts across radium-223 injections (Figure 1).

• Download figure
• Open in new tab

Figure 1: Normalized absolute mononuclear cell counts during Radium-223 treatment.

(A) Lymphocyte and (B) monocyte counts at baseline (BL) and after each Radium-223 injection. Red dots indicate the group means. The blue line represents the fitted linear regression model, including 95% CI.

Circulating immune cell populations were analyzed for their abundance in peripheral blood during treatment with radium-223. PBMC-derived percentages of cell populations were pre-processed by a logit-transformation to enable analysis using linear regression models (except for ΔMFI; Figure 2, Supplementary Figures 3 and 6). Before analysis, the percentage/ΔMFI data were normalized per patient, as described above. (Figure 2A). Immune cell subsets were visualized in grouped scatterplots per injection number. A linear model was fitted to each immune cell subset to determine the average change per injection number. Furthermore, checkpoint expression on these cells was also analyzed using the ΔMFI.

• Download figure
• Open in new tab

Figure 2: Overview of the analyzed immune cell subsets throughout radium-223 treatment.

(A) Data pre-processing steps, (B) CD3⁺, CD4⁺, and CD8⁺ T cells, (C) Immune checkpoint-expressing T cells, (D) Memory and effector T cell subsets, (E) Immunosuppressive cell subsets. Red dots indicate the group means. The blue line represents the fitted linear regression model, including 95% CI.

Next, we quantified longitudinal trends in immune cell subsets as the 6-month change (i.e., change between the subset’s baseline value and the value after the sixth radium-223 injection). We used percentile bootstrapping to determine the 95% confidence interval (CI) of the 6-month change of an immune cell subset over time (unit: percentage change relative to baseline; Figure 3, Supplementary Figures 4 and 7). Specifically, we used 2000 replicates in each bootstrap. Per replicate, we re-sampled the 30 patients with replacement, normalized the data as described above, and fitted a linear model to the sample. Next, the 6-month change between these measures was calculated, including a 95% CI to determine the uncertainty of the bootstrap estimate. Checkpoint expression was analyzed similarly.

• Download figure
• Open in new tab

Figure 3: 6-Month change estimate of immune cell subsets during radium-223 therapy.

The percentage change relative to baseline is calculated using a bootstrap method. Per bootstrap, a linear model is fitted on the logit-transformed and normalized bootstrap sample. Subsequently, the model predictions at baseline and after six months are used to calculate the change in (A) CD3⁺, CD4⁺, and CD8⁺ T cells, (B) memory/effector T cells, (C) immunosuppressive cells, and (D) checkpoint-expressing T cells and monocytes.

In the end, we performed an exploratory subgroup analysis in patients with an ALP response in the same manner.

All R code is available on GitHub: https://github.com/jeroencreemers/Immunophenotyping-radium223-mCRPC.

Patient characteristics

A total of 35 mCRPC patients were screened, of which 33 patients met the in- and exclusion criteria (two patients received concomitant enzalutamide). In the final analysis, three patients were excluded because ≤2 injections of radium-223 were administered, and no follow-up blood withdrawal had occurred after discontinuation of treatment. The median age of the study population was 71 years. The median number of prior systemic prostate cancer therapies since diagnosis of metastatic hormone-sensitive prostate cancer was one (range 0-4). Seventeen patients (57%) had received prior docetaxel chemotherapy, and twenty patients (67%) received prior enzalutamide and/or abiraterone. Median baseline ALP was 148 IU/l, and the majority of patients (77%) had high volume (>20) bone metastatic disease. Baseline patient demographics and clinical characteristics are shown in Table 1.

Twenty-two (73%) patients received all six radium-223 injections. An ALP decline of ≥30% during therapy was achieved in twenty patients (67%). Four patients (13%) had a PSA decline of ≥30% during therapy. At the time of analysis, 27 patients (90%) were deceased. Median OS was 13.2 months (95% CI 10.2-16.2 months). The majority of patients (60%) showed no new bone metastases on bone scintigraphy after radium-223. Radiological evidence of new pathological lymph node metastases or development of visceral metastases was found in seven patients (26%) and four patients (15%), respectively. Clinical outcomes are shown in Supplementary Table 2.

Availability of PBMCs for immunophenotyping

PBMCs for immunophenotyping were available of 30, 29, 26, and 20 patients at baseline and after two, four, and six injections of radium-223, respectively (Supplementary Figure S2). Missing data was mostly a consequence of the preterm discontinuation of radium-223 therapy. In one patient (ID-25), no PBMCs were collected after the second injection of radium-223.

Absolute lymphocyte and monocyte count

We first analyzed the absolute peripheral lymphocyte and monocyte counts to account for potential longitudinal variation in mononuclear cells in the population. We fitted linear regression models to the normalized absolute lymphocyte and monocyte counts per injection number (as a surrogate marker for time). Lymphocyte counts declined by approximately a factor two during treatment with radium-223 (Figure 1A; slope: -0.12 * 10⁹ cells/L per injection number, 95% CI: -0.143 - -0.102), while monocyte counts remained relatively stable over the course of treatment (Figure 1B; slope: -0.003 *10⁹ cells/L per injection number, 95% CI: -0.012 - 0.005).

Immunophenotype of lymphocyte and monocyte subsets during radium-223 therapy

To investigate longitudinal changes in circulating immune cell subsets in mCRPC patients during treatment with radium-223, we applied two pre-processing steps to the PBMC data: a logit transformation and a normalization step (see Methods; Figure 2A).

We first analyzed the PBMCs for the presence of T cells (CD3⁺ cells; Figure 2B). Despite normalization, substantial variation remained present in the data. Overall, the percentage of T cells within the population of PBMCs decreased during radium-223 treatment. Within the CD3⁺ cells, we distinguished two T cell subsets: CD4⁺ T cells and CD8⁺ T cells (Figure 2B). No major changes were seen in these subsets. Next, we determined the fraction of CD4⁺ and CD8⁺ T cells that expressed costimulatory (OX-40, ICOS, TIM-1) and inhibitory (CTLA-4, LAG-3, PD-L1, PD-1, TIM-3) checkpoint molecules, respectively (Figure 2C). The expression of OX-40 and LAG-3 on both subsets was practically absent; hence these checkpoint molecules were omitted from further analyses. Expression of CTLA-4, TIM-1, and TIM-3 was observed on a small subset (<5%) of CD4⁺ and CD8⁺ T cells, while expression of PD-1 (>30%) and PD-L1 (>70%) was more pronounced. A larger proportion of CD4⁺ cells than of CD8⁺ cells expressed ICOS (6.6% and <1% at baseline, respectively). The fraction of CD4⁺ and CD8⁺ T cells expressing PD-1 and PD-L1, as well as the fraction of CD4⁺ T cells expressing ICOS, increased slightly during treatment, while other checkpoint-expressing subsets remained stable over time. The relative expression of checkpoint molecules on CD4⁺ and CD8⁺ cells (i.e., ΔMFI) is shown in Supplementary Figure S3.

Based on the expression of memory marker CD45RO and lymphoid tissue homing chemokine receptor CCR7, we identified central memory (CD45RO⁺CCR7⁺), effector memory (CD45RO⁺CCR7^-), and effector (CD45RO^-CCR7^-) cells within the CD4⁺ and CD8⁺ T cell subsets (Figure 2D). All memory and/or effector phenotypes within the CD4⁺ and the CD8⁺ subset appeared stable throughout radium-223 therapy.

Subsequently, we studied four immunosuppressive cell types: regulatory T cells (Tregs; CD3⁺CD4⁺CD25⁺FoxP3⁺), monocytic MDSCs (M-MDSCs; CD11b⁺CD14⁺CD15^-HLA-DR^low/-), polymorphonuclear MDSCs (PMN-MDSCs; CD11b⁺CD14^-CD15⁺HLA-DR^low/-), and early MDSCs (eMDSC; lin^-CD14^-CD15^-CD33⁺HLA-DR^low/-). PMN-MDSCs were not detected, likely because we used cryopreserved samples. While the portion of M-MDSCs and eMDSCs, as well as the subsets expressing PD-L1, remained stable during treatment, we observed a small increase in the Treg fraction (Figure 2E).

The uncertainty associated with the longitudinal changes in immune cell subsets

To measure the degree of sampling variation in the described longitudinal changes of immune cell subsets during radium-223 treatment, we used a bootstrapping approach.

We applied bootstrapping to the main T cell subsets (Figure 3A), the memory/effector T cell subsets (Figure 3B), and the immunosuppressive subsets (Figure 3C). With regard to the main T cell subsets, the bootstrap approach estimated an overall 6-month decrease of 20.3% (CI -31.8% - -8.8%) of the CD3⁺ subset (Figure 3A).

The changes in the CD4⁺ and CD8⁺ fractions of the CD3⁺ subset were uncertain (Figure 3A). In the memory/effector T cell phenotypes, the largest change estimate was observed in the CD4⁺CD45RO^- CCR7^- subset (bootstrap estimate +15.1%, CI -4% – 42.9%). All other memory/effector phenotypes had less pronounced change estimates with CIs spanning across both positive and negative changes (Figure 3B). As indicated in Figure 3C, the proportion of Tregs in peripheral blood seemed to increase over time (25.1%; CI 14.3% - 38.2%). While the bootstrap estimate indicated an increase in the fraction of M-MDSCs during therapy, the CI was wide (113.3%; CI 33.6% - 239.6%). In contrast to the other two immunosuppressive subsets, the eMDSC change estimate was inconclusive (9.9%; -36.9% - 89.7%).

Concerning the checkpoint-expressing subsets, we observed a pronounced increase in the ICOS- and TIM-3-expressing proportion of CD4⁺ and CD8⁺ T cells, albeit with wide CIs (Figure 3D; numerical values are added as Supplementary Table S3). Less pronounced increases were seen in the fractions of PD-1- and PD-L1-expressing CD4⁺ and CD8⁺ T cells and PD-L1-expressing M-MDSCs. For the CTLA-4- and TIM-1-expressing subsets, changes were small (i.e., CD4⁺CTLA-4⁺ and CD8⁺TIM-1⁺), and/or the sampling variation was large (Figure 3D). Interestingly, while the fraction of Tregs increased over time, the proportion of CTLA-4-expressing Tregs hardly changed.

In addition, the 6-month change in the relative expression of checkpoint molecules on lymphocytes and monocytes, as analyzed with the ΔMFI, followed the same trend as the immune cell counts. The 6-month change estimates of the ΔMFI ranged from 0-10% (Supplementary Figure S4; Supplementary Table 4).

Subgroup analysis in patients with an ALP response

Since our data indicated longitudinal changes in fractions of peripheral lymphocyte or monocyte subsets in the entire study population, and ALP changes are considered a surrogate marker for response to radium-223, we hypothesized that potential changes might be more pronounced in patients with an ALP response to radium-223 therapy (n=20; 67%). PBMCs for immunophenotyping were available on all time points for thirteen patients in this subset (Supplementary Figure S5). Visually, there is a strong resemblance between the subgroup with ALP responders and the entire population in terms of the occurrence of cells, their distribution, and time trends (Supplementary Figure S6). The bootstrapping approach supports this resemblance: we observed similar trends compared to the entire population, with higher uncertainty due to the smaller sample size (Supplementary Figure S7; Supplementary Table S5). Therefore, our data do not support a more pronounced induction of fractions of immune cell subsets over time in patients with an ALP response, compared to those without a biochemical response to radium-223 therapy.