Detection of Aneuploidy in Cerebrospinal Fluid from Patients with Breast Cancer Can Improve Diagnosis of Leptomeningeal Metastases

The incidence of leptomeningeal metastasis in patients with breast cancer is estimated to be around 5% (1, 2). Although the incidence is relatively low, once patients become symptomatic, the symptoms can be devastating and the prognosis deteriorates with a median overall survival (OS) of 4–8 weeks (3), which increases to 3–8 months when treated (4–7). The detection of leptomeningeal metastasis in patients with breast cancer is hampered by limited sensitivities of routinely used techniques. According to the EANO-ESMO guidelines, patients presenting with typical symptoms of leptomeningeal metastasis and characteristic abnormalities on gadolinium-enhanced (Gd)-MRI may be diagnosed with (probable) leptomeningeal metastasis without cytologic confirmation (8). However, Gd-MRI has a sensitivity of 53%–80% and specificity of 77%–93% (4, 6, 7, 9, 10). A lumbar puncture for cerebrospinal fluid (CSF) cytology is recommended, which has a sensitivity of 45%–75% at first examination and increases to 64%–84% when a second lumbar puncture is performed (11–14). These limited sensitivities inevitably lead to delayed or missed diagnoses, thus improvement of diagnostic methods is urgently needed to allow for more timely treatment that may result in improved survival and quality of life.

During the past decades, multiple biomarkers have been interrogated for their ability to improve the detection rate of leptomeningeal metastasis, specifically in the CSF (15). Promising sources for the future application of biomarkers which are directly derived from leptomeningeal metastasis are tumor cells in CSF, detected by EpCAM-based methods (16–22), and the tumor-derived cell-free DNA (ctDNA) fraction within the total cell-free DNA (cfDNA) pool present in the CSF (23). Solid tumors, such as breast cancer, release tumor DNA by apoptosis and necrosis in all bodily fluids (24). To date, only small studies have been performed focusing on detection of ctDNA in CSF of patients with leptomeningeal metastasis originating from breast cancer. For instance, in a patient with clinical suspicion of leptomeningeal metastasis of whom the CSF cytology result was three times negative mutations in ESR1, PTEN and MRPS33 in CSF-derived cfDNA could be detected (23). In this patient, leptomeningeal metastasis was confirmed at autopsy, suggesting that assessment of tumor-derived cfDNA in CSF could detect leptomeningeal metastasis more sensitively than CSF cytology (23). In patients with lung cancer, EGFR mutations have been detected in CSF of patients with leptomeningeal metastasis (25). Similarly, in patients with brain metastases derived from solid tumors, as well as in patients with primary brain tumors, somatic alterations in CSF-derived cfDNA have been detected in 63% and 50%, respectively (26). However, not all tumors carry hotspot mutations, and therefore the use of mutations for detection of disease in CSF requires prior knowledge on the tumor's genetic make-up. Genome-wide untargeted approaches have the great advantage that upfront knowledge about the genetic alterations to be detected is not required. For example, the modified fast aneuploidy screening test-sequencing system (mFAST-SeqS) method employs selective amplification of long interspaced nuclear elements (LINE-1 sequences), which are present throughout the genome (27). This method allows for the detection of somatic copy-number alterations (CNA) at a chromosome arm resolution, representing a fast and affordable assessment of tumor fractions requiring only low amounts of DNA input (∼1 ng; ref. 27). Belic and colleagues showed that CNA patterns observed with either the mFAST-SeqS method or genome-wide shallow sequencing of plasma-derived cfDNA from patients with metastatic breast cancer were highly correlated, whereas known chromosomal profiles from cell line DNA could be captured with mFAST-SeqS as well (27). Considering virtually all breast cancers harbor CNAs, analysis of CNA represents an attractive alternative in CSF-derived cfDNA as well (28).

In this retrospective proof-of-concept study, we assessed the value of mutational analyses by performing targeted next-generation sequencing (NGS) and performed aneuploidy analyses on archival CSF samples using the mFAST-SeqS method in a large cohort of patients with breast cancer who underwent a lumbar puncture for the clinical suspicion of leptomeningeal metastasis. Furthermore, the prognostic value of CSF-derived cfDNA analyses, together with other routinely collected clinical parameters, CSF cytology, and CSF chemistry, was associated with OS.

Adult patients (≥18 years old) with a history of breast cancer who underwent a lumbar puncture for clinical suspicion of leptomeningeal metastasis (clinical signs and symptoms, e.g., headache, nausea, mental changes, gait difficulties, meningeal rigidity, cranial nerve palsies, spinal symptoms, and abnormalities at neurologic examination) and from whom stored remnant CSF was available were included in this retrospective analysis. CSF samples used in this study were collected and stored as part of standard diagnostic work-up. Remaining CSF has been stored at −80°C at the Department of Neuro-Oncology at the Erasmus MC Cancer Institute (Rotterdam, the Netherlands). The following clinical data were collected: medical history, information of the lumbar puncture (date and volume of CSF used for cytology), routine CSF chemistry results (including leukocyte count and protein and glucose concentration), cytology result as reported by the pathologist [positive, equivocal (suspicious or atypical cells), or negative], MRI results as reported by the radiologist and neurologic signs and symptoms as derived from the medical record prior to or at time of CSF collection, and follow-up (final diagnosis, start of systemic therapy/radiotherapy after lumbar puncture, and date of death). Leptomeningeal metastasis was defined as either malignant cells at cytology (“cytology+”) and/or when characteristic MRI abnormalities were observed (enhancing leptomeninges, leptomeningeal nodules, or linear and/or radicular enhancement; “radiology+”; ref. 8). To evaluate whether patients with central nervous system (CNS) metastases at CSF collection developed additional CNS localizations over time or whether patients without CNS metastasis at CSF collection did develop these during follow-up, consecutive scans were evaluated for development of leptomeningeal metastasis and/or brain metastases (scored as “final CNS diagnosis”). For the comparison with mutation analysis and mFAST-SeqS, the cytology results were used as reference. This study has been performed according to the “Code of Conduct for Responsible Use (2011)” (29) and the study design was approved by the Medical Research Ethics Committee of the Erasmus MC Cancer Institute (Rotterdam, the Netherlands; MEC-2019-0504). Nononcologic female patients with a clinical indication for a diagnostic lumbar puncture for the following diagnoses were included as control group (n = 12): idiopathic intracranial hypertension (2×), impaired consciousness, headache with spontaneous recovery, acute headache (no subarachnoidal bleeding), Alzheimer disease, frontotemporal dementia, vertigo, suspected demyelinating disease (2×), neurosarcoidosis, and idiopathic facial nerve paresis.

CSF samples were centrifuged for 10 minutes at 2,000 × g at 4°C. After centrifugation, supernatant was stored at −80°C until further handling. For cfDNA isolation, CSF samples were thawed at room temperature and 0.5–4.1 mL was used. cfDNA was isolated and eluted in 20 μL buffer using the QIAamp Circulating Nucleic Acid Kit (Qiagen) as per the manufacturer's instructions and stored at −20°C. cfDNA concentrations were quantified using the Quant-iT dsDNA High-sensitivity Assay (Invitrogen, Life Technologies) according to the manufacturer's instructions, and the Qubit Fluorometer (Invitrogen) was used as read out.

A targeted NGS approach with molecular barcoding using Oncomine Breast cfDNA Assay v2 (Thermo Fisher Scientific) was applied for low limit somatic variant detection according to the manufacturer's instructions. This assay consists of 10 genes frequently affected in breast cancer, covering 157 hotspots in genes including AKT1, EGFR, ERBB2, ERBB3, ESR1, FBXW7, KRAS, PIK3CA, SF3B1, and TP53. The amount of cfDNA used for sequencing ranged from 3.3 to 22.3 ng. Analyses were done as reported previously, using Ion S5 XL sequencing system and 540 chips, and evaluated with a standard variant calling pipeline (30). First, raw Ion S5 sequencing results with the Oncomine cfDNA assays were loaded into the TorrentSuite variant caller 5.10. By applying additional filtering, hotspot variants were called when (i) at least 500 unique molecules for that particular position were sequenced resulting in a limit of detection of 0.2% and (ii) if the mutant sequence was covered by three unique molecules with at least three reads per unique molecule.

We used the recently described mFAST-SeqS method, which has initially been established as a minimally invasive screening method for fetal aneuploidy from maternal blood (31), but has been adapted by Belic and colleagues to estimate tumor fractions in cfDNA (27). LINE-1 (L1) amplicon libraries were prepared as described by Belic and colleagues. Briefly, using target-specific L1 primers and Phusion hot start II polymerase II, a primary PCR step was performed to amplify L1 sequences throughout the genome using a single primer pair. This was followed by a secondary PCR step that amplifies all molecules from the first PCR step and adds adaptors and sample-specific index sequences. After both PCR steps, the PCR products were purified using AMPure XP Beads (1.4 ×, Beckman Coulter). The resulting libraries were quantified using the NEBNext Library Kit for Illumina (New England Biolabs), pooled equimolarly for 20 samples (2 nmol/L), supplemented with 25% of a PhiX control library, and sequenced on a MiSeq System (Illumina) generating 150 bp single reads aiming for at least 100,000 reads (32).

We trimmed the primers of the first PCR of the sequenced reads using Trimmomatic (v0.38). The trimmed reads were mapped on human reference genome hg19 using Burrows–Wheeler alignment (v0.7.17) and the read counts per chromosome arm were determined. Reads with a mapping quality >15 were counted and read counts were normalized to the total read count per sample. Subsequent computational random downsampling in steps of 5,000 reads for 24 samples (12 cases and 12 controls) at 100 iterations showed that reliable results were obtained down to 90,000 reads per sample (Supplementary Fig. S1). In short, we determined the delta in genome-wide z-score for every iteration compared with the genome-wide z-score obtained at 100,000 reads and observed an increase in the average delta genome-wide z-score with lower total numbers of read counts for both cases and controls. The average delta z-scores became divergent between cases and controls, whereas the associated SD showed a clear increase below 90,000 reads. Importantly, no false-positive chromosome arms were observed in any of the control samples at 90,000 reads. On the basis of the foregoing, we included samples with at least 90,000 mapped reads in the analysis. To test for over- and underrepresentation of each chromosome arm, we calculated z-scores by subtracting the mean and dividing by the SD of normalized read counts for the respective chromosome arm from a panel of 12 CSF controls. Because little to no reads aligned to the short arms of the acrocentric chromosomes 13p, 14p, 15p, 21p, 22p, and Y, these were excluded from the analysis. To get a general overview of aneuploidy, we squared and summed z-scores per chromosome arm into a genome-wide z-score. Following the original threshold set by Belic and colleagues (27), we considered samples with a genome-wide z-score ≥ 5 as aneuploid.

Descriptive statistics were calculated for variables of interest. Mann–Whitney U test was performed for univariate analyses of continuous variables and a Fisher exact test was used for categorical variables. To calculate the correlation between DNA input and molecular coverage, we calculated the Spearman rho. OS was calculated from time of CSF collection until death (event) or last follow-up (censored). Models associating variables of interest and OS, time to developing leptomeningeal metastasis, or brain metastases, were constructed using Cox proportional hazards methodology (enter method). The Kaplan–Meier method was used to graphically represent OS. Two-sided P values below 0.05 were considered significant. All statistical analyses were performed using IBM SPSS version 25. Figures were constructed with GraphPad Prism.

From January 2002 to April 2016, 121 patients with breast cancer underwent a lumbar puncture for suspected leptomeningeal metastasis, of whom leftover CSF was available for cfDNA analyses (Fig. 1). At CSF sampling, the median age was 55 years [interquartile range (IQR), 45–63 years]. Thirteen patients (10.7%) had a positive cytology, whereas 2 (1.7%) and 106 (87.6%) patients had an equivocal or negative cytology result, respectively (Table 1). The median total amount of cfDNA, isolated from a median of 1.8 mL of CSF, was 8.72 ng. The median cfDNA concentration was 5.17 ng/mL CSF (IQR, 3.62–10.75 ng/mL CSF).

As CSF cytology is the gold standard to diagnose leptomeningeal metastasis, we compared clinicopathologic variables between patients with a positive and negative CSF cytology result (Supplementary Table S1A). For the comparison between positive versus negative cytology, the two equivocal samples were excluded. The only clinical sign that was more frequently observed in the positive cytology group was meningeal rigidity. All 13 patients with a positive cytology were already diagnosed with metastatic disease at the time of CSF collection and 6 of these patients were previously diagnosed with brain metastases. Regarding routinely performed CSF analyses, the CSF leukocyte concentration was significantly higher in patients with positive cytology. The glucose concentration was significantly lower in patients with positive cytology. No differences in CSF protein concentration and total cfDNA concentration were observed between cytology-positive and -negative samples. There was no difference in breast cancer subtype between the cytology-positive versus -negative group.

To study the concordance between CSF cytology and mutations in CSF, we performed targeted NGS on a subset of 30 patient samples, of which nine had a positive, one had an equivocal, and 20 had a negative cytology result. We found that the molecular coverage (i.e., the number of uniquely sequenced molecules) was significantly correlated with the amount of DNA available for sequencing (Spearman rho, 0.68; P < 0.001). When <10 ng was used, the molecular coverage was below 500 molecules in 73% of samples (Supplementary Fig. S2). As the median total CSF–derived cfDNA yield in our cohort was only 8.72 ng, the majority of samples (69/121) had too low DNA yield for reliable NGS analysis with our currently used method. Sufficient molecular coverage was obtained for eight of 30 samples, in four of which hotspot mutations were detected (Supplementary Table S2). All four samples in which mutations were detected were reported as cytology positive.

From the total cohort of 121 patients, the mFAST-SeqS method yielded sufficient number of reads (>90,000) for 114 samples from 112 patients, allowing for reliable determination of the aneuploidy status. For 2 patients, two sequential CSF samples were available, of which the first was included for the cohort-wide analyses (Supplementary Table S3). Aneuploidy (mFAST-SeqS z-score ≥ 5) was observed in 10 of 13 (76.9%) samples that were cytology positive, which was significantly more often than in the cytology-negative group (9%; P < 0.001; Fig. 2; Supplementary Table S1B). Three patients with a positive cytology did not show genome-wide aneuploidy according to the threshold of ≥5 that we employed, indicating a false-negative rate of 23.1%. Patients with aneuploidy more frequently had gait difficulties, cranial nerve palsies, lymph node metastasis, and a higher CSF protein concentration (Supplementary Table S1B). The total cfDNA concentration was not different between either patients with and without aneuploidy or patients with and without abnormal cytology (Supplementary Table S1A and S1B), suggesting that specific detection of a tumor-specific signal within the total pool of cfDNA is more informative.

Besides a genome-wide aneuploidy score, the mFAST-SeqS method also provides an aneuploidy score per chromosome arm. From all 88 patients without genome-wide aneuploidy, 6 patients did show alterations in two or more individual chromosome arms. In patients with aneuploidy of ≥2 chromosome arms, 2 patients were cytology positive, 2 patients had intracranial metastases (dural/brain metastasis), and the other 2 patients were not diagnosed with CNS metastases (Supplementary Table S4). Hence, using a threshold for aneuploidy when ≥2 chromosome arms are aneuploid decreases the false-negative rate to 7.7%, whereas none of the healthy control samples (n = 12) showed ≥2 aneuploid chromosome arms in a leave-one-out analysis.

Fourteen cytology-negative patients were scored as having aneuploidy. Of these 14 patients, 4 patients were diagnosed with leptomeningeal metastasis at time of CSF collection (n = 1) or immediately afterwards (n = 3) either on MRI or at second lumbar puncture. Three patients were not diagnosed with leptomeningeal metastasis on imaging, but had dural metastases at CSF collection, of which 1 patient developed cytology-proven leptomeningeal metastasis 619 days after initial lumbar puncture. Finally, 1 patient developed parenchymal brain metastases after 188 days (Table 2). The remaining 6 patients had a high aneuploidy score, but no final diagnosis of CNS metastases, and therefore represented potential false positives (5.4% of all patients; 25% of all patients with aneuploidy). However, 4 patients died soon after CSF collection (after 43, 48, 51, and 83 days), whereas the other 2 patients were still alive at time of analyses (4.8 and 14.5 years of follow-up, respectively). Unfortunately, no autopsies were performed to determine the cause of death and excluded CNS involvement in these patients, but based on the clinical data, case 119 had near certain leptomeningeal metastasis, but a negative cytology. An overview of clinical symptoms, CSF chemistry, imaging results, extracranial metastatic localizations, and OS of the 14 patients with genome-wide aneuploidy, but negative CSF cytology, is provided in Supplementary Table S5. If we were to use the cutoff of ≥2 aneuploid chromosome arms instead of using the cutoff of a genome-wide z-score ≥ 5, 2 additional patients without a final diagnosis of CNS metastases become positive, which would increase the potential false-positive rate to 7.1%.

With a median follow-up of 10.7 years and 30 patients still alive at the time of data analysis, the median OS of the entire cohort was 1.78 years (IQR, 0.42–11.7 years).

Next, we associated routine CSF chemistry results with OS. CSF leukocyte count was available for 119 patients and a leukocyte count above the upper limit of normal (>4 × 10⁶/L) was associated with a greater hazard of death [HR, 1.78; 95% confidence interval (CI), 1.07–2.98; P = 0.027]. CSF protein levels were available for 120 patients; a CSF protein level above the upper limit of normal (>0.58 g/L) was associated with a greater hazard of death (HR, 2.43; 95% CI, 1.34–4.40; P = 0.003). Glucose levels were available for all 121 patients; a glucose level below the lower limit of normal (<2.5 mmol/L) was associated with a greater hazard of death (HR, 10.34; 95% CI, 3.49–30.62; P < 0.001). Of note, only 4 patients had a glucose concentration <2.5 mmol/L. CSF cytology results were available for all patients at time of CSF collection. To test whether the cytology result was associated with OS, we used the definition of CSF cytology positive as those samples in which malignant cells were reported by the pathologist, and equivocal results were analyzed as cytology negative. A positive cytology at the time of CSF collection was associated with increased risk of death (HR, 5.38; 95% CI, 2.92–9.91; P < 0.001). In addition to the above described routine CSF chemistry results, the presence of aneuploidy (z-score ≥ 5) and the presence of any metastases (extracranial and/or brain metastases) at time of CSF collection were associated with a greater hazard of death (HR, 3.41; 95% CI, 2.07–5.61; P <0.001 and HR, 12.86; 95% CI, 6.63–24.95; P <0.001, respectively). Other clinical parameters, including age at CSF collection and estrogen receptor (ER) and HER2 status of the primary tumor, were not associated with OS.

In subsequent multivariable analysis, including all univariable significant variables, only CSF aneuploidy (Fig. 3A) and the presence of any metastatic localization at time of CSF collection (Fig. 3B) were significantly associated with a greater risk of death (HR, 2.24; 95% CI, 1.13–4.43; P = 0.021 and HR, 12.79; 95% CI, 6.29–26.02; P <0.001, respectively; Table 3A; Supplementary Fig. S3).

To investigate whether the presence of aneuploidy was associated with time to leptomeningeal metastasis development and time to brain metastases development, we also performed Cox regression only for those patients who were not diagnosed with leptomeningeal metastasis (n = 97) or brain metastases (n = 99) at the time of CSF collection, respectively. In univariable analysis for development of brain metastases, only HER2 status and metastatic disease at CSF collection were significantly associated with a higher likelihood for brain metastases development. In the multivariable analyses, both variables still showed a significantly greater hazard of developing brain metastases (Table 3B).

Similarly, in univariable Cox regression with time to leptomeningeal metastasis as dependent variable, metastatic disease at time of the lumbar puncture (no vs. yes) and the aneuploidy score (<5 vs. ≥5) were significantly associated with a higher likelihood of the development of leptomeningeal metastasis. In multivariable analysis, both variables still showed a significantly greater hazard of developing leptomeningeal metastasis (Table 3C).

Plasma cfDNA analyses are increasingly being implemented in routine diagnostics in patients with metastatic cancer, and more recently, cfDNA analyses of CSF have sparked interest to characterize molecular aberrations of primary brain tumors and brain metastases. Our study comprises the largest breast cancer cohort in which CSF-derived cfDNA analyses have been performed. Here, we show that aneuploidy, measured by the mFAST-SeqS method (i) identifies 77% of patients with cytologically proven leptomeningeal metastasis, (ii) identifies a subgroup of patients with CNS metastases prior to routine diagnostics, and (iii) has prognostic value.

Specifically, we established for the first time the relation between aneuploidy detection in CSF-derived cfDNA and OS and showed that the previously established genome-wide z-score of ≥5, developed to select plasma samples with high tumor fractions (>5%–10%; ref. 27), yields prognostic value for patients with breast cancer. Importantly, as tumor DNA can be derived from brain metastases and leptomeningeal metastasis, CSF-derived cfDNA analyses will not necessarily discriminate between these two conditions. Notably, an mFAST-SeqS z-score of ≥5 was associated with developing leptomeningeal metastasis, but not associated with developing brain metastases. Although the number of events in both Cox models was limited, it is likely that tumor DNA derived from leptomeningeal metastasis is more abundantly present in CSF than tumor DNA derived from parenchymal brain metastases. More data, especially from negative control samples and EANO-ESMO–confirmed leptomeningeal metastasis cases, are necessary to determine the optimal diagnostic cutoff for aneuploidy in this specific setting. Using the current threshold of genome-wide z-score ≥5 for the definition of “aberrant,” we missed three cases that were cytology positive. Two of the three missed cases had alterations on multiple chromosome arms, which in our cohort only occurred in a total of 6 patients, suggesting a lower threshold or other way of scoring aneuploidy could decrease the false-negativity rate without greatly affecting the false-positivity rate. The third patient with a false-negative mFAST-SeqS result had no alterations on any of the single chromosome arms. The initial cytology report of this CSF sample mentioned “no malignant cells,” but at second examination, the report mentioned “low cellular CSF with two atypical cells compatible with adenocarcinoma,” which might be below the limit of detection of the mFAST-SeqS method. An alternative explanation for this false-negative sample, might be a relatively copy-number neutral breast cancer, which will also be missed by this method.

On the other hand, 14 cytology-negative patients were positive for aneuploidy at the current cutoff. Four of these cytology-negative patients were actually diagnosed with leptomeningeal metastasis at time of CSF collection or immediately after the initial lumbar puncture by imaging or a second CSF assessment. This indicates that these initial cytology results should be considered as false negatives and demonstrates the potential additive value of mFAST-SeqS method to conventional cytology. We identified only 6 patients with CSF aneuploidy without a final diagnosis of leptomeningeal metastasis or brain metastases. It is possible that due to wide-spread metastatic disease in four of these patients, tumor DNA from the blood was diffused over the blood–brain barrier. Hence, comparative studies between plasma and CSF obtained at the same time from the same patient are needed to elucidate whether the same chromosomal alterations are detected in blood and CSF.

Although targeted UMI-based NGS approaches are known to enable much more sensitive ctDNA detection compared with mFAST-SeqS method (down to 0.2% vs. 5%–10%; ref. 27), due to low cfDNA amounts, the majority of our samples yielded a molecular coverage that was too low for reliable detection of mutations. On the basis of our findings, we believe that current panel-based sequencing with Oncomine Breast cfDNA Assay v2 (Thermo Fisher Scientific) only provides an option for those samples with sufficient DNA yield, which only can be determined after cfDNA isolation. For low cfDNA yielding CSF samples, singleplex digital PCR (dPCR) assays can be performed, which was recently shown by van Bussel and colleagues (16) for EGFR mutations in patients with non–small cell lung cancer. However, in contrast to melanoma and non–small cell lung cancer, breast cancer mutation analyses of the primary tumor or metastatic lesion are not routinely performed because until recently no targeted treatments were available requiring knowledge of the tumors' mutational profile. More importantly, the mutational profile of breast cancer is quite heterogeneous (33), requiring a broad targeted panel or ideally whole-exome or whole-genome sequencing of tumor tissue, followed by a patient-specific dPCR for CSF analyses. Hence, a more general approach aiming at detection of virtually universal cancerous alterations, such as provided by the mFAST-SeqS method, seems a more attractive option, requiring low amounts of DNA input and no upfront knowledge of the genetic make-up of the tumor. However, for samples with sufficient cfDNA yield, that is, above 10 ng, targeted NGS is feasible as shown by the four cytology-positive samples, in which we detected hotspot mutations, and is more sensitive than mFAST-SeqS method for the detection of tumor-derived cfDNA.

Although our study represents one of the largest cohorts of CSF-derived cfDNA analyses, due to the retrospective nature of the study, the clinical data collection was suboptimal and dependent on the extent of information that has been captured in the medical record by the treating physician. Moreover, in this proof-of-concept study, we used CSF that has been stored at −80°C for many years, so we cannot exclude that long-time storage might have influenced DNA quality and quantity compared with freshly obtained CSF.

Future studies employing the mFAST-SeqS method in patients suspected of leptomeningeal metastasis, should ideally be prospectively conducted and focus on determination of the optimal cutoff for aneuploidy in a sample based on a larger series of negative control samples and true-positive cases. Moreover, standardized clinical, pathologic, and radiological assessments, according to the EANO guidelines, should be performed to investigate to what extent the mFAST-SeqS method complements the current diagnostic armamentarium to diagnose leptomeningeal metastasis in a patient's CSF.

In conclusion, aneuploidy, as measured by the mFAST-SeqS method, provides a robust and affordable technique to detect tumor-derived DNA in the CSF of patients with CNS metastases from breast cancer. The detected aneuploidy is associated specifically with development of leptomeningeal metastasis and OS, but not with development of brain metastases. Future prospective trials investigating leptomeningeal metastasis should employ this method in combination with other promising techniques, such as EpCAM-based tumor cell detection assays, to improve detection of leptomeningeal metastasis in CSF of patients with advanced cancer. Ultimately, the combination of standardized clinical symptoms and neuro-imaging scoring together with sensitive detection of tumor-derived material in the CSF will improve leptomeningeal metastasis diagnosis.

L. Angus reports personal fees from Merck and Pfizer outside the submitted work. J.W.M. Martens reports personal fees from Novartis outside the submitted work. No disclosures were reported by the other authors.

L. Angus: Conceptualization, data curation, formal analysis, investigation, writing–original draft. T. Deger: Formal analysis, writing–review and editing. A. Jager: Conceptualization, writing–review and editing. J.W.M. Martens: Conceptualization, writing–review and editing. V. de Weerd: Writing–review and editing, carried out the experiments (FAST-SeqS). I. van Heuvel: Resources, writing–review and editing. M.J. van den Bent: Conceptualization, resources, writing–review and editing. P.A.E. Sillevis Smitt: Conceptualization, resources, writing–review and editing. J.M. Kros: Conceptualization, writing–review and editing. E.M.J. Bindels: Software, Writing–review and editing, processed the experimental data and provided technical support for the sequencing experiments. E. Heitzer: Writing–review and editing, designed the mFAST-SeqS method and contributed to the experimental set-up. S. Sleijfer: Conceptualization, writing–review and editing. J.L.M. Jongen: Conceptualization, resources, writing–review and editing. S.M. Wilting: Conceptualization, supervision, writing–original draft, writing–review and editing.

This research did not receive any specific grant from funding agencies in the public, commercial, or nonprofit sectors. We thank Charlotte C.J. Rimmelzwaan and Laura Pasquet for their help with the CSF cfDNA isolations and cfDNA mFAST-SeqS experiments.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.