A Microfluidic, Multi-Antibody Cell Capture Method to Evaluate Tumor Cells in Cerebrospinal Fluid in Patients With Suspected Leptomeningeal Metastases Open Access

Thirty-four cell lines from various epithelial malignancies as well as germ cell tumors were selected for this study (Table 1). Cell lines were obtained from the American Type Culture Collection (Manassas, Virginia), cultured according to the supplier’s recommendations, evaluated for morphology, and confirmed for mycoplasma negativity prior to the start of experiments.

Due to challenges in obtaining fresh normal human CSF, artificial CSF (aCSF) was used to establish the analytic performance characteristics of the assay. aCSF consists of Hank balanced salt solution (Corning, Glendale, Arizona), bovine serum albumin (Millipore Sigma, St Louis, Missouri), and 10 μL of whole blood (obtained from 6 unique healthy donors). The composition of this matrix was designed with a higher glucose, protein, and leukocyte count than physiologic CSF to account for the possible iatrogenic contamination of clinical specimens with peripheral blood during lumbar puncture. A physiologic comparison between CSF, blood, and aCSF is shown in Table 2.^24–27 

Contrived samples consisted of 10 μL of a specified cell line in trace RPMI media (Corning) spiked into 3.9 mL aCSF. A “high-spike” sample contained approximately 500 tumor cells and a “low-spike” sample contained approximately 50 tumor cells. Negative controls (“null-spike” samples) contained no tumor cells. The number of tumor cells present in a sample was verified by positive nucleic acid stain from Syto11 (ThermoFisher Scientific, Carlsbad, California) and analyzed on a fluorescence imaging cytometer, Celigo (Nexcelom Biosciences, Lawrence, Massachusetts).

The critical processing steps of the cell capture assay are summarized below. All steps were performed at ambient temperature. Contrived samples were created and immediately stored in CEE-Sure CSF collection tubes (Biocept) containing 0.6 mL of diazolidinyl urea (Sigma) for no less than 48 hours to recreate clinical sample transportation time. Samples were centrifuged at 400g for 5 minutes. Supernatant was aspirated and cell pellets were resuspended and incubated with primary capture antibody cocktail PN1986 developed to optimally capture tumor cells based on prior experience in blood (Table 3; Biocept). Unbound capture antibodies were removed through multiple washes and centrifugations at 400g for 5 minutes followed by aspiration. Samples were then incubated with a biotinylated goat–anti-mouse secondary Fab (Jackson Immuno Research, West Grove, Pennsylvania). After additional wash cycles with phosphate-buffered saline (Corning), samples were drawn through a proprietary streptavidin-coated microfluidic device (Biocept). Cells labeled with the primary antibody capture cocktail PN1986 were immobilized via biotin-streptavidin affinity in the microfluidic device. Microfluidic devices loaded with samples underwent several rounds of washing with phosphate-buffered saline (Corning) followed by methanol to fix and permeabilize cells. Fluorescent immunocytochemistry of the immobilized cells was performed in parallel using the following antibody conjugates: streptavidin, Alexa Fluor 647 conjugate (ThermoFisher); CD45, Alexa Fluor 594 conjugates (Biolegend, San Diego, California); and cytokeratins, Alexa Fluor 488 conjugates (ThermoFisher). Cells underwent subsequent staining with 4′,6-diamidino-2-phenylindole (ThermoFisher). Microfluidic devices were scanned on an automated fluorescence imaging scanner (Bioview, Billerica, Massachusetts) and evaluated by clinical laboratory scientists. Cells that were streptavidin positive (SA⁺), cytokeratin positive (CK⁺), and CD45 negative (CD45⁻) were considered tumor cells, whereas cells that were streptavidin negative (SA⁻) or CD45 positive (CD45⁺) were considered not to be tumor cells. For accuracy studies, cytokeratin was evaluated for CD45⁻ and SA⁺ cells. Cells were classified as CK⁺ and CK⁻ tumor cells depending on the presence and absence of the cytokeratin.

CSF specimens were collected from 65 unique patients under Institutional Review Board (IRB)-approved protocols at Providence Saint John’s Health Center (Santa Monica, California) (protocol number BIOC-035, WCG IRB tracking number 20181339 and Providence Saint Joseph’s Health IRB protocol 2017000293). Patients consented to their samples being used for research and their data to be shared via signing the informed consent associated with the IRB-approved protocols. CSF was collected on site, transferred to CEE-Sure CSF collection tubes (Biocept), and transported to Biocept at ambient temperature. Cell capture was performed at the Biocept Clinical Laboratory Improvement Amendment (CLIA) laboratory in San Diego, California. Samples were stratified according to indication for obtaining CSF, cancer history and type, and overall LMD assessment. Pathologic diagnosis was required for primary malignant conditions. An overall LMD assessment was required for all cases indicated for LMD workup with an epithelial primary tumor. This assessment was made no less than 4 months after the time of sample collection and was determined by the treating neuro-oncologist and informed by standard of care diagnostic methods such as clinical signs and symptoms, MRI imaging, CSF analysis, and CSF cytology.

Recovery linearity was determined by spiking various concentrations of the BT474 cell line with from 0 to 625 target cells per 3.9 mL of aCSF and 10 μL of whole blood from a healthy donor. The experiment was performed in triplicates. After 48 hours storage at ambient temperature, the samples were processed through CNSide assay (Biocept) and analyzed. Linear regression analysis of capture antibody cocktail mix was performed for 15 samples and yielded an R² = 0.9999 for the carcinoma assay (Figure 1).

Fifty-four null-spike samples were evaluated. Cells were evaluated with respect to cytokeratin positivity/negativity. Mean values and standard deviations were calculated for both CK⁺ and CK⁻ cells captured. A detected result cutoff was established at the 95th percentile, and values were adjusted to volume. The detected cutoff values for CK⁺ and CK⁻ tumor cells were 0.15 and 0.20 cell per 1 mL aCSF, respectively.

Eighteen samples were prepared with either BT474 or T24 cell lines at high-, low-, and null-spiking concentrations. Results are shown in Table 4. The average high-spike TCCs were 72% (408 of 567) and 79% (419 of 528) in samples prepared with BT474 (n = 3) and T24 (n = 3), respectively. The average low-spike TCCs were 60% (32 of 53) and 79% (34 of 43) in samples prepared with BT474 (n = 3) and T24 (n = 3), respectively. No tumor cells were detected in the null-spiked samples (n = 6). The CV percentages for high-spike samples prepared with BT474 and T24 were 7% and 8%, respectively, whereas the CV for low-spike samples was 24%. Null-spiked samples did not show detection of tumor cells.

Customized and automated digital scanners (Bioview) were used to generate digital cell images from the microfluidic channel. To assess the reproducibility of image capture between different scanners, BT474 cell line triplicates were prepared at high-, low-, and null-spiking concentrations. Samples were digitally imaged on 3 scanning instruments designated “Y,” “W,” and “X.” Results are shown in Table 5. The average TCC percentages for high-spike samples (n = 3) and low-spike samples (n = 3) were 82% (436 of 531) and 63% (33 of 53), respectively. Null-spiked samples (n = 3) demonstrated no tumor cell capture. The CV percentages for high- and low-spike samples were 6% and 33%, respectively.

Automated liquid handlers (Hamilton, Reno, Nevada) were used to automate sample processing steps prior to loading the samples into the microfluidic chip for cell capture. To determine the reproducibility of the automated liquid handlers, a precision test was carried out using 18 samples consisting of BT474 cell lines at high-, low-, and null-spiking concentrations (6 individual samples per concentration). Samples were processed in duplicate on 3 automated liquid handlers designated “A,” “B,” and “C.” Results are shown in Table 6. The average TCC values for high-spike samples (n = 6) and low-spike samples (n = 6) were 71% (370 of 518) and 75% (38 of 51), respectively. Null-spiked samples (n = 6) demonstrated no tumor cell recovery. The CVs for high- and low-spike samples were 13% and 10%, respectively.

Twenty-four samples were prepared with either BT474 or T24 cell lines at high- and low-spiking concentrations. Following a 48-hour incubation period in CEE-Sure sample collection tubes, samples were evaluated in 24-hour intervals during the course of 7 consecutive days. Results are shown in Figure 2.

TCC was negatively correlated with the time of sample incubation. The negative correlation coefficients for high-spike samples prepared with BT474 (n = 1) and T24 (n = 1) were −0.3 and −0.8, respectively. The negative correlation coefficients for low-spike samples prepared with BT474 (n = 1) and T24 (n = 1) were −0.5 and −0.6, respectively. A decrease in cell recovery following longer incubation time was not unexpected, as cell viability is compromised when cells are kept outside of their normal tumor microenvironment.

In total, 122 contrived samples were used to assess analytic accuracy. Sixty-eight positive samples were prepared using 34 unique cell lines of different carcinoma types (see Table 1) at both high- and low-spiking concentrations. Fifty-four negative samples were prepared with a null spike. Samples were evaluated using the detection cutoffs established above. Sixty-eight positive control samples demonstrated detection. Fifty-two negative samples demonstrated no detection, and 2 negative samples demonstrated detection. Analytic sensitivity was 100% (68 of 68), and analytic specificity was 96% (52 of 54) (Table 7).

Additionally, average TCC and CV were assessed in positive control samples (Table 8). The average TCC in samples prepared with a low spike (n = 34) was 52% (23 of 48), with a CV of 67%. The average TCC in samples prepared with a high spike (n = 34) was 50% (261 of 522) and demonstrated a CV of 43%.

The ability of the cell capture assay to perform accurately in clinical samples was assessed in a total of 65 CSF samples collected from patients with and without indications for LMD workup, with and without epithelial primary tumors, and with and without a positive overall LMD assessment (Table 9). The detection cutoff values established above were used in this evaluation (see Materials and Methods). Overall LMD assessment was determined by the treating neuro-oncologist no less than 4 months after the time of sample collection and used all information from standard of care diagnostics such as clinical evaluation, brain and spine MRI, CSF analysis, and CSF cytology.

For samples that had indications for LMD workup and an epithelial primary tumor, CSF was divided at the time of procedure, and cell capture and CSF cytology were both performed. Cytopathologic evaluation was performed locally by the institution’s pathology service. Table 9 shows the cytology and cell capture results for matched samples.

Of this sample set, 14 cases had nonmalignant indications for CSF sampling (eg, autoimmune, infectious, neurodegenerative, etc.). Of the remaining 51 cases (those with indications for LMD workup and a history of cancer), 48 had an epithelial primary tumor (29 breast cancers, 12 non–small cell lung cancers, and 7 others) and 3 had a nonepithelial primary tumor (1 glioma and 2 lymphomas). Of the 48 cases with an epithelial primary and indications for LMD, overall LMD assessment was positive in 24 cases and negative in 24 cases.

Twenty-two of the 24 cases with an epithelial primary tumor, indications for LMD, and an overall positive assessment for LMD yielded a detected result on cell capture. The number of tumor cells captured among detected samples ranged from 2 to 28 197, with an average of 2022 (SD = 6161). Cytology results among these cases were 5 positive, 11 atypical, and 6 negative. No report could be obtained for 2 cases. The overall sensitivity for cell capture was 92% (22 of 24). The overall sensitivity for cytology was 22% (5 of 22) if only positive cases are included and 72% (16 of 22) if positive and atypical cases are included.

Thirty-nine of the remaining 41 cases that had nonmalignant indications, a nonepithelial primary tumor, or a negative overall LMD assessment yielded a not-detected result on cell capture. The overall specificity for cell capture was 95% (39 of 41). The cytology results among patients with indications for LMD, epithelial primaries, and a negative overall LMD assessment were all negative. The overall specificity for cytology was 100% (24 of 24).

The absence of a sensitive means to evaluate tumor cells in CSF has limited the ability of oncologists to promptly confirm suspected LMD in solid tumor patients and manage this disease after the initiation of treatment. Often, this results in a forced reliance on clinical symptoms and MRI imaging (with a reported sensitivity of 76% and specificity of 77%) to make critical management decisions.²⁸ 

The purpose of this study was to demonstrate the analytic performance of a cell capture assay for detecting tumor cells in CSF. The assay methodology includes the use of multiple primary capture antibodies, a secondary biotinylated antibody signal amplification step, digital image capture, and a CD45 exclusion criterion to rule out background leukocytes. Multiple parameters were examined in this study, including linearity, precision, sample stability, and accuracy.

Analytic linearity studies demonstrated an R² value of 0.9999 from 0 to 650 cells. Correlating higher cell counts was limited by the validated range of the orthogonal method. Additional studies are required to establish linearity at the upper end of the results obtained (data not shown).

Using contrived negative samples, we established an analytic cutoff at the 95th percentile for both CK⁺ and CK⁻ tumor cells at 0.15 and 0.20 cells/mL, respectively. We suspect that the cutoff for CK⁻ cells was higher due to nonspecific interactions that occurred from one of the antibodies in the capture antibody cocktails. These cutoff values yielded acceptable concordance, sensitivity, and specificity in contrived and clinical samples (see below) and could be adjusted in future studies using larger sample sizes of negative clinical samples.

Intra-assay precision studies demonstrated cell recovery (reported as TCC) ranging from 60% to 79% among 2 cell lines at high- and low-spiking concentrations. A variation (reported as CV) was seen among low-spike samples, being 24%. Minimal variation was seen among high-spike samples, with CV ranging from 7% to 8%. Interinstrument precision studies (performed separately for both digital scanning equipment and liquid handlers) demonstrated cell recovery ranging from 63% to 82% in a single cell line at high- and low-spiking concentrations. The variation among low-spike samples was 10% to 33%. Moderate variation was seen among high-spike samples, with CV ranging from 10% to 13%. Additionally, all null-spike (negative) samples tested in precision studies detected no tumor cells.

Sample stability studies demonstrated a negative correlation between the time to process and the tumor cells recovered, with negative correlation coefficients ranging from 0.3% to 0.8% among 2 cell lines at high- and low-spiking concentrations processed in 24-hour intervals during the course of 7 days. A variation of 23% to 25% CV was observed amongst these samples. Sample stability for this assay was established at 6 days, when all samples demonstrated >40% cell recovery. The 40% cutoff was selected in advance of the study based on historical BT474 cell capture data obtained from daily CLIA controls. Cell count variations can occur due to factors such as cell culture, cell spike accuracy, and automated cell count accuracy. Here we used Biocept’s validated cell spike protocol. Still, variations in cell count can occur based on the above factors. The primary purpose of spike controls is to verify the assay, such that antibodies were added, and confirm streptavidin coating of the microfluidic device. Spike cell counts obtained by the Celigo imaging device are used to save time and costs in the CLIA testing environment. So, this automated cell spike count protocol, optimized for BT474 cell count can, for example, be inflated in some instances due to artifacts, leading to higher spike counts and resulting in lower percentages of capture counts. Cells captured and identified by the BioView software are subsequently confirmed by trained clinical laboratory scientists. In conclusion, the cell line capture data should be viewed as an overall attempt, even if technically challenging, to demonstrate assay stability or performance relative to sample storage time.

Analytic accuracy studies performed across 34 different cell lines at high- and low-spiking concentrations demonstrated cell recovery ranging from 3% to 136% (note that these values were obtained with respect to the orthogonal methods through counting the cell spiking number by a cytometer [Celigo], the feasibility of which proved challenging at lower spiking concentrations, hence the value exceeding 100%). The average cell recovery ranged from 50% to 52% among high- and low-spiked samples respectively. Significant variation was observed for contrived samples between different cell lines for both low (CV = 67%) and high (CV = 43%) cellular concentrations. We attribute this high degree of variation to the heterogeneity of cell surface antigen expression among different tumor cell lines, since it was not observed during precision studies when 1 or 2 cell lines were studied for CV.²⁹  The analytic sensitivity and specificity of the assay was 100% and 96%, respectively.

Sixty-five clinical CSF specimens from unique patients were obtained from 2 sites and categorized based on indication, type of primary malignancy, and overall LMD assessment. Overall LMD assessment was made no less than 4 months after the time of sample collection and was determined by the treating neuro-oncologist using all available information generated by standard of care methods. Twenty-four samples were from patients with indications for LMD, primary epithelial malignancies, and a positive overall assessment for LMD, in which CTCs were expected to be detected and demonstrated 92% concordance with cell capture. Fourteen samples were from patients with indications for a nonmalignant condition, in which CTCs were expected to not be detected and demonstrated 86% concordance with cell capture. Three were from patients with indications for LMD and primary nonepithelial malignancies, which were expected to not be detected and demonstrated 100% concordance with cell capture. Twenty-four samples were from patients with indications for LMD, primary epithelial malignancies, and a negative overall assessment for LMD, which were expected to not be detected and demonstrated 100% concordance with cell capture. The overall sensitivity and specificity for cell capture in the tested population was 92% and 95%, respectively. These values were comparable to published accounts of other cell capture assay studies performed in CSF, with reported sensitivities ranging from 92% to 100% and specificities ranging from 73% to 95%.²²  The overall sensitivity and specificity for cytology in the tested population was 22% or 72% (depending on whether “atypical” results are included) and 100%, respectively.

We used a specific cocktail of antibodies (PN1986) based on our empirical studies using various antibody cocktails developed for a previous assay (Target Selector, Biocept) for identifying tumor cells in blood and found it to be best for CSF in our assay (details not shown). However, we cannot exclude that there may be other combinations of antibodies that could potentially be better or equivalent. The assay may also falsely detect very low numbers of nontumor cells (streptavidin⁺, CK⁺, CD45⁻ cells) such as choroid plexus cells and other epithelial cells that could be due to a disrupted blood-brain barrier or surgical procedures that patients undergo when accessing CSF and contribute to the false-positive rate. In future studies we will further define optimal conditions for sample stability, including effect of abnormal proteins, drugs, immunoglobulins, clinical conditions (cancer types, systemic tumor burden, CNS tumor burden, parenchymal lesions in brain versus leptomeninges on imaging, treatments, fever, sepsis, etc), on assay performance.

In summary, we detected epithelial cancerous cells with a microfluidic multiple-antibody cell capture assay in contrived and patient samples with various epithelial malignancies for LMD. The prognosis of patients with tumor cells detected in CSF requires further investigation using a larger sample group, especially in cases in which CSF cytology is negative. As intrathecal therapies and systemic therapies with CNS activity become more available, the use of highly sensitive CSF testing may be useful for the early diagnosis and longitudinal assessment of patients’ disease, particularly in assessing response to treatment prior to further radiologic evaluation, which is not practical to perform as frequently as cell capture derived from an intrathecal catheter (eg, Ommaya reservoir) or repeat lumbar puncture procedure. This cell capture assay is part of a larger suite of assays designed for use in CSF (CNSide, Biocept) that is currently undergoing a prospective clinical trial (NCT05414123) to evaluate its clinical performance and measure how the assay impacts clinical decisions for patients with LMD.

We would like to thank our team members from Biocept’s CLIA Laboratory team for their contributions (names are listed in alphabetic order of last name): Sean Adversalo, BS, CPT (California Department of Public Health); Icelee Balangcod, BS, MT(AMT); Abijail Guillen, MS, MLS(AMT); Michael Hadley, BS, CG(ASCP); Elliott Hansen, BS; Victoria Hathaway, BS; Lan Huynh, BS; John Peters, BS, CG(ASCP); Karina Rich, BS, CG(ASCP); Edgar Sales, BS, CG(ASCP); and Erica Schmidt, BS, MLS(ASCP).