Cerebrospinal Fluid Flow Cytometry Improves Patient Stratification in Central Nervous System Acute Lymphoblastic Leukaemia but Requires Careful Standardization: Results from the ALLTogether1 CSF-FLOW Sub-Study

The central nervous system (CNS) is an important sanctuary site for acute lymphoblastic leukaemia (ALL). Sensitive biomarkers are needed to guide CNS treatment intensity and identify patients at risk of CNS relapse. Currently, cerebrospinal fluid (CSF) cell counts and cytology are used to stage patients as CNS1 (no detectable CNS leukaemia), CNS2 (≤5 white blood cells (WBC)/µl, ≥2 visible blasts on cytospin), and CNS3 (>5 WBC/µl with blasts or overt clinical CNS involvement). However, most CNS relapses occur in CNS1 patients, and the prognostic value of CNS2 disease is disputed. Wide variations in CNS2 rates between and even within countries indicate a lack of diagnostic accuracy.

CSF-FLOW is a scientific sub-study within the ALLTogether1 trial for children and young adults (1-45 years of age) with newly diagnosed ALL (ClinicalTrials.gov ID: NCT04307576). CSF-FLOW builds on data indicating that CSF flow cytometry (FCM) identifies patients at higher risk of relapse even among CNS1 patients and can divide patients with a traumatic lumbar puncture (TLP: >10 red cells/µl of CSF) into TLP- (no blasts on FCM) and TLP+ (with blasts) groups with prognostic relevance. CSF-FLOW aims to: 1) evaluate the prognostic impact of CSF FCM in a large independent cohort with sufficient power to accurately stratify CNS relapse risk; 2) determine the utility of CSF FCM in improving conventional CNS classification; 3) identify key parameters influencing assay performance to develop a standardised clinical grade assay protocol. Aim 1 requires long follow-up. Here we report the results of aims 2 and 3.

As of April 22, 2024, CSF-FLOW recruited 1869 children from 12 European countries (57.1% males; 73.5% ≤10 years old). Using conventional staging criteria, 67% of patients were CNS1 (n = 1249), 10% CNS2 (n = 186), 2.1% CNS3 (n = 40), 11.4% TLP- (n = 213), 5.6% TLP+ (n = 104), and 3.8% inconclusive (n = 71). 1348 patients had a day 1 CSF (collected in Transfix/stabilisation buffer) FCM result, 20.5% of whom had >10 leukaemia associated immunophenotype (LAIP) events detectable (= CSF FCM+). CSF FCM+ was commoner in T-cell than B-cell precursor (BCP) ALL (41.6% vs 17.5%, p < 0.001) and in TLP patients compared to non-TLP (35.2% vs 17.6%, p < 0.001).

55.2% of BCP patients diagnosed as CNS2 by cytology were confirmed by FCM, whilst 44.8% had no detectable leukemic blasts by FCM. These could be false negatives due to technical issues such as sample delays (see below), or false positives due to misidentification of T lymphocytes as leukemic cells on cytology. We investigated CD3+ cell counts (by FCM) in BCP patients with discrepant results. The mean CD3+ count was 101 (range 0-782) indicated up to 90% of patients would have ≥2 CD3+ T cells on the cytology slide, indicating a high likelihood of false positive results.

FCM+ rates were significantly higher when larger CSF sample volumes were used (range = 45-5000 µl; p = 0.007). Despite use of Transfix, CSF FCM- patients had a significantly longer delay (days) between sample withdrawal and testing than CSF FCM+ (p = 0.04). Varying FCM rates between countries were mainly explainable by differences in sample volumes and analytical delays.

Patients underwent serial CSF sampling at days 1, 15 and 29 of treatment. Clearance kinetics varied by immunophenotype with 1.9% (21/1127) BCP and 13% (21/162) T-ALL patients still FCM+ at day 15 (p < 0.001). At day 29, 1% (11/1116) BCP and 3.1% (5/161) T-ALL patients were still FCM+ (p = n.s). Clearance at day 15 also varied by TLP status, with a higher rate of FCM+ in TLP patients compared to non-TLP (5.9% vs 2.8%; p = 0.04). Patients still FCM+ at day 15 had a slightly higher day 1 CSF blast count/µl (p < 0.001). Longer follow up is needed to assess the impact of flow positivity and clearance on outcome.

Overall, these results confirm that FCM is more sensitive and specific than cytospin-based methodologies and suggest that FCM can increase the accuracy of conventional CNS2 classification by reducing false positives due to T cells. With longer follow-up, CSF-FLOW is adequately powered to develop risk-prediction algorithms for patients at high and low risk of CNS-involving relapses. However, prior to universal adoption of FCM as a clinical-grade biomarker, our findings stress the importance of developing standardized protocols indicating required sample volumes, maximal sample transit times and identification of optimal blast thresholds for prognostic significance.