This study was a single institution study of 10 patients who were diagnosed with NSCLC with metastatic or unresectable disease that was confirmed with pathology. Eligible patients were starting a first or new line of systemic treatment at the time of enrollment. Patients did not have any known severe anemia. The study was conducted according to the guidelines of the Declaration of Helsinki and approved by the Institutional Review Board at the University of Southern California Norris Comprehensive Cancer Center (protocol HS-17-00854 approved on 13 February 2018) and all patients provided written informed consent. Patients were able to leave the study at any time at their own request or were able to withdraw at the discretion of the investigator for safety, behavioral, or administrative reasons. The reasons for discontinuation were documented.

Patient samples were collected from 1/26/2018 to 5/3/2021. Samples were taken prior to initiation of a new or first line therapy, and at follow-up visits coinciding with their treatment schedule to avoid unnecessary blood draws for up to 70 weeks, with a maximum of 7 LBx samples each taken approximately 7–12 weeks apart. Patients were monitored from the time of enrollment to the date of last follow-up and spanned an average of 314.3 (range 27–548) days. Patients were followed for survival analysis in which progression events were confirmed by clinical imaging. A total of 50 normal donor (ND) samples from individuals with no known pathology were used for comparative analysis.

Peripheral blood samples (average 7 mL) were collected in Cell-Free DNA blood collection tubes (Streck, Omaha, NE) and placed in a temperature stabilization box for transport. All samples were processed by the Convergent Science Institute in Cancer at University of Southern California within 48 h of collection as described previously [43]. In short, blood samples underwent erythrocyte lysis and all the nucleated cells adhered to custom glass slides (Marienfeld, Lauda-Königshofen, Germany) with approximately 3 million cells per slide. Cells were then incubated in 7% BSA, dried, and stored at – 80 °C for subsequent analysis. WBC counts of the samples were determined automatically prior to processing (Medonic M-series hematology Analyzer, Clinical Diagnostic Solutions INC., Fort Lauderdale, FL) allowing for the calculation of cells/mL.

For sample analysis, 2 slides per test were thawed for immunofluorescent staining as previously described [43, 44]. Slides were processed at room temperature using the IntelliPATH FLX™ autostainer (Biocare Medical LLC, Irvine, CA, USA). Briefly, cells were fixed with paraformaldehyde prior to incubation with 2.5 ug/ml of a mouse IgG1 anti-human CD31:Alexa Fluor® 647 mAb (clone: WM59, MCA1738A647, BioRad, Hercules, CA) and 100 ug/ml of a goat anti-mouse IgG monoclonal Fab fragments (115-007-003, Jackson ImmunoResearch, West Grove, PA), permeabilized using 100% cold methanol, followed by an antibody cocktail consisting of mouse IgG1/IgG2a anti-human cytokeratin (CK) 1, 4, 5, 6, 8, 10, 13, 18, and 19 (clones: C-11, PCK-26, CY-90, KS-1A3, M20, A53-B/A2; C2562, Sigma, St. Louis, MO), mouse IgG1 anti-human CK 19 (clone: RCK108, GA61561-2, Dako, Carpinteria, CA), mouse anti-human CD45:Alexa Fluor® 647 (clone: F10-89-4, MCA87A647, AbD Serotec, Raleigh, NC), and rabbit IgG anti-human vimentin (Vim) (clone: D21H3, 9854BC, Cell Signaling, Danvers, MA). Lastly, slides were incubated with Alexa Fluor® 555 goat anti-mouse IgG1 antibody (A21127, Invitrogen, Carlsbad, CA) and 4′,6-diamidino-2-phenylindole (DAPI; D1306, ThermoFisher) prior to mounted with a glycerol-based aqueous mounting media. The HDSCA3.0 workflow includes technical controls throughout the pipeline as previously described [29, 32, 43]. Controls consisted of ND samples spiked with known cell line cells (SK-BR-3 ATCC: HTB-30 and HPAEC ATCC: PCS-100-022) that were processed and analyzed according to standard protocol.

Samples were imaged using automated scanning microscopy at 100 × magnification. Image data sets were analyzed using OCULAR (Outlier Clustering Unsupervised Learning Automated Report) to identify rare event candidates using 761 morphometric parameters [4, 29]. Images of CTC candidates were presented to a hematopathologist-trained technical analyst for manual data reduction and phenotype classification. Rare events were classified into 12 categories (8 cellular and 4 oncosome categories) based on marker expression in the 4 channels. There were 2 types of circulating tumor cell: epi.CTCs and mes.CTCs. Epi.CTCs were classified as containing a nucleus by DAPI morphology, and presenting as CK positive, Vim negative, CD45/CD31 negative. Mes.CTCs were classified as Epi.CTCs with Vim expression. Other rare cells were described using the positive immunofluorescence marker expression in each of the four channels (for example: DAPI|CD45/CD31 = DAPI positive, CD45/CD31 positive, CK negative, Vim negative). Oncosomes were classified as round DAPI negative CK positive events with variable Vim and CD45/CD31 expression, and were observed both free floating and in close proximity to cells (for example: Onc CK|Vim = Oncosome, CK positive, Vim positive, DAPI negative, CD45/CD31 negative).

Cohort level comparisons and longitudinal analysis were performed using python (version 3.8.5) and the Scipy library (version 1.5.0). Statistical comparisons of analyte enumerations at the cohort level were done using the Wilcoxon rank sum test, also known as the Mann–Whitney U test [45, 46]. The Wilcoxon rank sum test was chosen due its non-parametric nature and robustness to outliers. Statistical significance was set at a p-value of 0.05.

PFS was set to the length of time from date enrolled to last follow-up with no documented progression events. OS was set to the length of time from date enrolled to date of death, or end of study date if there was no date of death. Statistical analysis and data visualizations for PFS and OS were created using R software (version 3.6.3) and the survival library (version 3.2-7). Kaplan–Meier curves were used to estimate the survival functions [47]. To compare two survival functions statistically, the log-rank test was used [48,49,50]. For kinetic PFS analysis, changes in LBx analyte counts were determined using the change between the two blood draws prior to progression, or the last two draws if there was no patient progression. For PFS and OS, we analyzed 16 LBx analytes and groups: total events, total CK expressing cells, total rare cells, total oncosomes, and each individual channel-type classification for cells and oncosomes. For PFS-Kinetics, we analyzed the change in these 16 factors over time. For each of these factors, analyses were performed at each of the three quartiles. Statistical significance was set at a p-value of 0.05. When median survival could not be calculated because the cohort did not reach 50% survival during the study, median survival is reported as N/A.