Quantitative cerebrospinal fluid circulating tumor cells are a potential biomarker of response for proton craniospinal irradiation for leptomeningeal metastasis

Abstract Background Leptomeningeal metastasis (LM) involves cerebrospinal fluid (CSF) seeding of tumor cells. Proton craniospinal irradiation (pCSI) is potentially effective for solid tumor LM. We evaluated whether circulating tumor cells (CTCs) in the CSF (CTCCSF), blood (CTCblood), and neuroimaging correlate with outcomes after pCSI for LM. Methods We describe a single-institution consecutive case series of 58 patients treated with pCSI for LM. Pre-pCSI CTCs, the change in CTC post-pCSI (Δ CTC), and MRIs were examined. Central nervous system progression-free survival (CNS-PFS) and overall survival (OS) from pCSI were determined using Kaplan Meier analysis, Cox proportional-hazards regression, time-dependent ROC analysis, and joint modeling of time-varying effects and survival outcomes. Results The median CNS-PFS and OS were 6 months (IQR: 4–9) and 8 months (IQR: 5–13), respectively. Pre-pCSI CTCCSF < 53/3mL was associated with improved CNS-PFS (12.0 vs 6.0 months, P < .01). Parenchymal brain metastases (n = 34, 59%) on pre-pCSI MRI showed worse OS (7.0 vs 13 months, P = .01). Through joint modeling, CTCCSF was significantly prognostic of CNS-PFS (P < .01) and OS (P < .01). A Δ CTC-CSF≥37 cells/3mL, the median Δ CTC-CSF at nadir, showed improved CNS-PFS (8.0 vs 5.0 months, P = .02) and further stratified patients into favorable and unfavorable subgroups (CNS-PFS 8.0 vs 4.0 months, P < .01). No associations with CTCblood were found. Conclusion We found the best survival observed in patients with low pre-pCSI CTCCSF and intermediate outcomes for patients with high pre-pCSI CTCCSF but large Δ CTC-CSF. These results favor additional studies incorporating pCSI and CTCCSF measurement earlier in the LM treatment paradigm.


Quantitative cerebrospinal fluid circulating tumor cells are a potential biomarker of response for proton craniospinal irradiation for leptomeningeal metastasis
Leptomeningeal metastasis (LM), or seeding of tumor cells to the pia and arachnoid mater, or leptomeninges, is a devastating complication of solid tumor malignancies, and it is associated with a poor prognosis. While it is clinically detected in 5-8% of patients with solid tumors, 1-3 it is frequently underdiagnosed. Without treatment, survival is roughly 4-6 weeks with death usually resulting from neurologic deterioration. 4 With treatment, patients tend to have a median survival of 3-4 months, depending on the histology, functional status, and response to systemic therapy. 5 Though there are no universally accepted guidelines for diagnosis or treatment response of LM, it is recognized that neurological physical examination, cerebrospinal fluid (CSF) analysis, and central nervous system (CNS) imaging should be used in combination. 6 Though CSF cytology aids in LM diagnosis, up to 50% of LM patients will have a negative cytology. 7 Therefore, CSF cytology has a low sensitivity alone, and it is not useful as a biomarker since it is not quantitative. In metastatic patients, circulating tumor cells (CTCs) in the peripheral blood (CTC blood ) correlate with increased burden of metastasis 8 and can predict disease progression and survival, [9][10][11] though the utility of CTC blood in LM is unclear.
Recently, CSF CTCs (CTC CSF ) have been shown to have improved diagnostic performance for LM compared to magnetic resonance imaging (MRI) and CSF cytology. [12][13][14][15] The presence of at least one CTC CSF per 1ml of CSF was shown to improve diagnosis of LM and should be considered to be added to CSF cytology and MRI evaluation. 14 In a large retrospective cohort of patients with CNS metastases, including patients with LM, high CTC CSF was associated with worse overall survival. 16 CTC CSF was assayed prospectively in a clinical trial evaluating intrathecal Trastuzumab in patients with HER2+ epithelial cancer LM, and dynamic changes in CTC CSF were noted, with an increase in CTC CSF seen prior to MRI changes in patients with disease progression. 17 It is still unknown whether CTC CSF can be used to determine treatment response to LM-targeted therapies.
Chemotherapy agents for treatment of LM are limited in therapeutic efficacy, but craniospinal irradiation (CSI) can be used to treat the entire CSF compartment. Nevertheless, X-ray-base photon CSI can cause significant toxicity compared to proton-based CSI. 18 We have therefore evaluated and demonstrated that proton craniospinal irradiation (pCSI) can achieve therapeutic radiation doses with relatively few side effects. 19 Currently, there are no established biomarkers of response of LM-directed therapies. We aimed to understand whether neuroimaging, CTCs in the blood, or CTCs in the CSF correlate with clinical outcomes in patients with LM treated with pCSI.

Methods
We evaluated a prospectively collected consecutive case series of 58 patients who were treated with pCSI between January 2018 and December 2020; 21 of the patients were enrolled in a phase I trial evaluating the safety of pCSI (NCT03520504). 19 Patients were diagnosed with LM using radiographic criteria on MRI and/or positive CSF cytology (45 patients). 6 For each patient, we collected the patient demographics, treatment history, radiation dose and fractionation, and pre-and post-pCSI MRI studies. Patient age and Karnofsky performance status (KPS) at the time of pCSI were recorded. Median values with range were reported unless otherwise specified.
CTC CSF and CTC blood were analyzed by the CellSearch® platform, an immunomagnetic isolation technique using antibodies to EpCAM conjugated to magnetic beads for the capture of EpCAM-expressing epithelial cells that are CD45 negative and cytokeratin positive, prior to and after pCSI with a standard cutoff of 200/3 mL when higher numbers of CTCs were seen. CTC was analyzed within 48 h from CSF or blood collection (7.5 mL of whole blood was used for analysis). CTC CSF and CTC blood were measured within 8 weeks prior to pCSI and/or at multiple points after completing pCSI, approximately every 2-3 months. For patients with pre-and post-pCSI CTC CSF measurement, the change in CTC CSF (Δ CTC-CSF ) was calculated and dichotomized by median Δ CTC-CSF , and for patients with multiple post-pCSI CTC CSF measurements, the nadir was used for analysis. An identical procedure was performed to obtain the change in CTC blood pre-and post-pCSI (Δ CTC-blood ). We assessed correlation between CTC CSF and CTC blood pre-pCSI using a non-parametric Spearman's rank correlation rho.
MRIs of the brain and spine were examined within 60 days prior to pCSI. All patients had pre-pCSI MRI assessment of LM and were examined for the presence of brain and/or spine LM, parenchymal brain metastases, and hydrocephalus. Post-pCSI MRI assessment was examined for LM that either improved from prior scans, remained stable, or showed disease progression using RANO-LM criteria 6 by an experienced neuroradiologist blinded to the CTC results and clinical status. The association between MRI findings and survival outcomes was landmarked at 60 days post-pCSI.

Importance of Study
This study describes the clinical outcomes of the largest cohort of leptomeningeal patients receiving proton craniospinal irradiation (pCSI) to date. We observe the best survival in patients with low pre-pCSI circulating tumor cells in the cerebrospinal fluid (CTC CSF ) and intermediate outcomes for patients with high pre-pCSI CTC CSF but large change CTC CSF between preand post-pCSI measurements. Therefore, we establish that CTC CSF may have potential utility as a biomarker to prognosticate response to pCSI in LM patients.

Neuro-Oncology Advances
We report median follow-up separately for those who died and those who were lost to follow-up or had not died by December 31, 2020. We examined central nervous system progression-free survival (CNS-PFS), defined as the time from start of pCSI to either CNS progression or death, and overall survival (OS), defined as the time from start of pCSI to death from any cause. Survival times are reported with interquartile range (IQR). Univariate associations with CNS-PFS and OS were visualized using Kaplan Meier plots and significance was determined using a log-rank test. Multivariate Cox proportional-hazards regression modeling was used to test independence of univariate associations. Time-dependent receiver operating characteristic (ROC) curve analysis was performed to report 6-month and 1-year sensitivity and specificity accounting for data censoring. 20 The association between survival outcomes and pre-pCSI CTC count was tested using Cox proportional-hazards regression after verifying the assumption of proportional hazards. The CTC count over time, a continuous time-varying covariate, was simultaneously analyzed in a joint model of longitudinal and survival outcomes with shared random effects and an unspecified baseline risk function. Results from models treating CTC count as a continuous covariate are reported per 10 cells. To aid in clinical interpretability, we determined optimal cutoffs through maximally selected rank statistics using CNS-PFS as the associated outcome of interest and requiring no fewer than 10% of patients with non-zero CTC to be separated by a chosen cutoff, with the goal of partitioning CTC measurements into high and low groups. We defined statistical significance at alpha = 0.05.

CNS Progression-Free Survival and Overall Survival
The survival outcome summaries for CNS-PFS and OS are shown in Table 2. The median follow-up time for patients who were censored or still alive by the end of the study period (n = 15, 26%) was 15 months (IQR: 9-18 months). The median CNS-PFS was 6 months (IQR: 4-9 months) ( Figure 1A). The median overall survival was 8 months (IQR: 5-13 months) ( Figure 1B). Tumor histology, cytology positivity, patient age, and KPS were not found to be associated with OS or CNS-PFS.

Plasma circulating tumor cells (CTC blood ).-CTC blood
were measured in 17 patients pre-pCSI, and in all 17 patients they were also measured post-pCSI up to a median of 1.6 months (range: 0.6-6.1 months) before initiating a new systemic therapy. There was no correlation between pre-pCSI levels of CTC CSF and CTC blood (rho = 0.10, P = .72) (Supplementary Figure 3). An optimal cutoff in CNS-PFS survival analysis of pre-pCSI CTC blood was determined to be 1 cell/3mL, and 8 (47%) patients had CTCs ≥1 cell/3mL. No association was seen between pre-pCSI CTC blood ≥ 1 cell/3mL and CNS-PFS (P = .20) or OS (P = .80). After pCSI, 22 patients had CTC blood measured, including all 17 with pre-pCSI measurements. Joint modeling of CTC blood over time and survival outcomes showed that CTC blood was not significantly associated with CNS-PFS (HR: 1. and using the median Δ CTC-blood of 0 as a cutoff , neither OS (P = .70) or CNS-PFS (P = .90) were significantly associated using a log-rank test.

MRI features as a biomarker of response.-LM abnor-
malities were visible in the majority (n = 54, 93%) of pre-pCSI MRs. The presence of hydrocephalus (n = 13, 22%) on MRI prior to CSI was not associated with CNS-PFS or OS (Table 3). Parenchymal brain metastases (n = 34, 59%) were associated with a worse CNS-PFS (median 5 vs. 8 months for those without parenchymal brain metastases, P = .04) and worse OS (median 7 vs. 13 months for those without parenchymal brain metastases, P = .01) ( Table 3). The timedependent sensitivity and specificity for CNS-PFS and OS using the presence of brain parenchymal metastases are shown in Table 3. Patients with spine only LM in pre-pCSI MRI (n = 11, 19%) were found to have improved OS compared to patients with no visible LM (n = 4, 7%), brain only LM (n = 9, 15%), or both brain and spine LM (n = 34, 59%) (22 vs. 3 vs. 8 vs. 8 months, respectively) ( Table 3). Because the relationship between site of LM and OS involved overstratification and did not correlate with bulk of disease (i.e., no visible LM had a worse OS compared to spine only LM), we did not estimate sensitivities or specificities for LM site, and we did not include it in further combined models. There was no association between site of LM on MRI and CNS-PFS post-pCSI (P = .40).

Neuro-Oncology Advances
The time-dependent sensitivity and specificity for CNS-PFS and OS combining pre-pCSI CTC CSF and parenchymal brain metastases are shown in Table 3, resulting in a lower sensitivity of CNS-PFS and OS at 6 and 12 months. In a multivariate model, pre-pCSI CTC CSF was significantly associated with CNS-PFS (P = .02) independently of CNS parenchymal brain metastases on MRI (P = .05). In a multivariate model of OS, only the presence of CNS parenchymal brain metastases on MRI was associated with OS (P = .05) independently of pre-pCSI CTC CSF (P = .21).
Most patients (90%, n = 52) had at least one post-pCSI MRI assessment within 60 days of completing pCSI. In post-pCSI MRI assessment 44 of 52 patients (76%) showed stable or improved disease. Using a landmark of 60 days post-pCSI, patients with a worse MRI after pCSI (n = 8) did not have significantly different overall survival compared to those with an improved or stable MRI (n = 44) (12 vs 10 months, P = .40). The dynamic changes in CTC CSF over time are shown by post-pCSI MRI in Supplementary Figure 4.

Discussion
Our recent phase 1 clinical study of pCSI for solid tumor LM had shown a median CNS-PFS of 7 months (95% CI: 5-13) and a median OS of 8 months (95% CI: 6 to not reached) with 20% of patients living more than a year without CNS disease progression. 19 In our current study of the largest patient cohort to date, we found that pCSI is associated with similar median CNS-PFS of 6 months and OS of 8 months for patients with solid tumor LM despite many patients having been heavily pre-treated. Compared to the survival of other therapies such as systemic therapy alone (2.2 months), 21 systemic therapy with intrathecal cytarabine (3.8 months), 21 triple intrathecal chemotherapy (2 months), 22 and immune checkpoint inhibitors (2 months) 23 pCSI patients had favorable outcomes. In addition, 10% of patients were alive without CNS disease progression and 26% of patients were still alive and censored at the end of the study period, indicating that pCSI may prolong survival in select patients.
Our study also demonstrated that CTC CSF quantitative assessment in the CNS compartment has potential as a prognostic biomarker of response. Using maximally selected rank statistics, we determined that patients with pre-pCSI CTC CSF <53 cells/3 mL at baseline and/or Δ CTC-CSF ≥37 cells/3mL post-pCSI had the most favorable CNS-PFS survival outcomes; however, there were a limited number of patients with both pre-and post-pCSI CTC CSF measurements. Nevertheless, for patients with pre-CSI CTC CSF ≥53 cells/3mL at baseline, the quantitative post-CSI Δ CTC-CSF allowed for further stratification into 3 groups using CTC CSF as a biomarker of response: the most favorable group, the favorable group and the unfavorable group. Our results showed that pCSI may be most effective in improving survival outcomes in patients with lower CSF disease burden (the most favorable group), arguing for clinical trial evaluation of early integration of pCSI into the LM treatment paradigm and routine CTC CSF measurement. For patients with higher disease burden at baseline and suboptimal response to pCSI (the unfavorable group), more aggressive therapeutic strategies can be considered if consistent with patient's goals of care, including concurrent systemic therapy with pCSI, or early LM-directed therapy following pCSI including additional radiotherapy.
We did not find that CTC blood to be prognostic of CNS-PFS or OS. We also did not find a significant correlation between CTC CSF and CTC blood . Our results indicate that in this patient cohort with LM, survival outcomes are likely driven by CSF disease burden and response to treatment. As all of our patients received pCSI, they were less likely to have known poor risk factors including KPS <60, multiple and major neurologic deficits, encephalopathy, or extensive systemic disease with few treatment options, 24 but because LM patients with favorable prognostic factors still have a median OS of roughly 4 months, 25 our findings demonstrated that LM-directed treatments for patients with limited poor-prognostic factors may lead to beneficial outcomes.
On MRI, the presence of parenchymal brain metastases is thought to be associated with increased disease bulk and worse prognosis, 6 and we found that the presence of parenchymal brain metastases was associated with shorter CNS-PFS and OS. 26 Nevertheless, we did not see that a higher visible burden of LM on MRI was associated with significantly worse outcomes, and the patients with no visible LM although a small group (n = 4) actually had the worst overall survival. It is possible that this is driven by the biological differences between LM cells suspended in the CSF compared to those that form radiographically evident disease. 27 Furthermore, of the patients with no visible LM, 3 had parenchymal brain metastases, which we demonstrated to be associated with poor prognosis. While the parenchymal brain metastases on MRI pre-pCSI is prognostic of OS, the MRI post-pCSI is not prognostic, likely because of patients receiving additional therapies with varied efficacy and the limited number of patients with worse MRI after pCSI. Additionally, we did not detect a relationship between prior hydrocephalus, a poor prognostic factor, and survival after pCSI. All in all, when compared to CTC CSF, we concluded that MRI features provide excellent anatomic detail, but MRI may be less reliable in associating with treatment outcomes for LM patients undergoing pCSI.
This study has important limitations. First, given that all patients were selected to receive pCSI and a subset of patients had pre-pCSI and/or post-pCSI CTC measurements, there is selection bias, and we were not able to account for all potential clinical confounding factors. Furthermore, the actual pre-pCSI CTC measurements could be higher than reported due to the time lapse between CTC measurements and initiation of pCSI. Second, the biomarkers derived by the CellSearch® assay can only be applied in patients with solid tumor of epithelial origin (CD45-, EpCAM+). Third, this study was not designed to define nominal cutoffs for CTCs given its heterogeneity in patient clinical history and timing of measurements. Because the CTC assay data is truncated at 200 cells/3 mL and the optimal cutoffs were chosen to maximize the discrimination of CNS PFS, this study cut offs should serve as heuristics corresponding to large and small CTC values. Though we chose to minimize the number of analyses performed, and there is biologic plausibility to support the use of CTC CSF , these results should be interpreted carefully and with the understanding that more validation is needed. With this in mind, a prospective randomized clinical trial is currently underway to address these concerns (NCT04343573).
In conclusion, this is the largest study to date demonstrating CTC CSF could be a biomarker prognostic of survival outcomes in patients with LM who received pCSI, a potentially effective treatment for patients with solid tumor LM. We identified 3 groups of patients based on CTC CSF , with significant, long-term survival observed in patients with pre-CSI CTC CSF <53 cells/3mL, arguing for evaluation of the utility of early pCSI intervention in the LM treatment paradigm. Early treatment escalation after pCSI may be considered for patients with high pre-pCSI CTC CSF and a smaller nadir post-pCSI in order to prolong survival.

Supplementary material
Supplemental material is available at Neuro-Oncology Advances online.

Funding
This work was supported by grants from Cycle for Survival Equinox Innovation Initiative Award and the National Institutes of Health/National Cancer Institute (P30-CA008748).