Automated workflow for the cell cycle analysis of (non-)adherent cells using a machine learning approach

Understanding the details of the cell cycle at the level of individual cells is critical for both cellular biology and cancer research. While existing methods using specific fluorescent markers have advanced our ability to study the cell cycle in cells that adhere to surfaces, there is a clear gap when it comes to non-adherent cells. In this study, we combine a specialized surface to improve cell attachment, the genetically-encoded FUCCI(CA)2 sensor, an automated image processing and analysis pipeline, and a custom machine-learning algorithm. This combined approach allowed us to precisely measure the duration of different cell cycle phases in non-adherent cells.

Our method provided detailed information from hundreds of cells under different experimental conditions in a fully automated manner. We validated this approach in two different Acute Myeloid Leukemia (AML) cell lines, NB4 and Kasumi-1, which have unique cell cycle characteristics. Additionally, we tested the impact of drugs affecting the cell cycle in NB4 cells. Importantly, our cell cycle analysis system is freely available and has also been validated for use with adherent cells.

In summary, this report introduces a comprehensive, automated method for studying the cell cycle in both adherent and non-adherent cells, offering a valuable tool for cancer research and drug development.

Methods
RNA extraction, RNA-seq protocol and data analysis

Total RNA was extracted from dry pallets of cells collected prior- and post-acquisitions and purified using the Zymo Research Quick-RNA Miniprep (W/O directzol). Reverse transcription was performed with the SuperScript II Kit (Invitrogen), according to the manufacturer’s protocol. RNA-seq was performed according to the True-seq Low sample protocol selecting only polyadenylated transcripts. In brief, before starting mRNA isolation and library preparations the integrity of the total RNA was evaluated by running samples on a Bioanalyzer instrument by picoRNA Chip (Agilent), then converted into libraries of double stranded cDNA appropriate for next generation sequencing on the Illumina platform. The Illumina TruSeq v.2 RNA Sample Preparation Kit was used following manufacturer’s recommendations. Briefly, 0.1-1 μg of total RNA were subjected to two rounds of mRNA purification by denaturing and letting the RNA bind to Poly‐T oligo-attached magnetic beads. Then fragmentation was performed exploiting divalent cations contained in the Illumina fragmentation buffer and high temperature. First and second strand cDNA is reverse transcribed from fragmented RNA using random hexamers. First strand cDNA was synthesized by SuperScript II (Invitrogen) reverse transcriptase and random primers and second strand cDNA synthesized by DNA polymerase I and Rnase H. The subsequent isolation of the cDNA was achieved by using AMPure XP beads (depending on the concentration used, these beads can efficiently recover PCR products of different sizes). The product recovered contained overhanging strands of various lengths due to the fragmentation procedure. The 5’ and 3’ ends of cDNA are repaired by the 3’-5’ exonuclease activity and the polymerase activity and adenylated at 3’ extremities before ligating specific Illumina oligonucleotides adapters followed by 15 cycles of PCR reaction using proprietary Illumina primers mix to enrich the DNA fragments. Prepared libraries were quality checked and quantified using Agilent high sensitivity DNA assay on a Bioanalizer 2100 instrument (Agilent Technologies). Raw reads 51bp PE for NB4 and Kasumi-1 cells were quality-filtered and aligned to the hg18 reference genome using nf-core/rnaseq v3.9 pipeline using STAR as aligner and Salmon for quantification with default parameters. Gene counts for each sample were log1p transformed, mean value among the two replicates was taken to compute Pearson correlation among gene expression pre- and post- time-lapse acquisition.

EdU incorporation and assessment by Flow Cytometry

A two-hour EdU pulse was performed by replacing half of the total media volume of the cells with 2X concentrated EdU in the corresponding growth medium, followed by subsequent fixation. Click-iT™ EdU Alexa Fluor™ 647 Flow Cytometry Assay Kit (CN: C10419, Thermo-Fisher Scientific, Waltham, MA, USA) and Click-iT™ EdU Cell Proliferation Kit for Imaging, Alexa Fluor™ 647 dye (CN: C10340, Thermo-Fisher Scientific, Waltham, MA, USA) were used for flow cytometry and imaging, respectively. The experiments were performed according to the manufacturer’s protocols for the mentioned kits. DNA staining with DAPI using 500 μL of 5 μg/mL DAPI in PBS for 106 cells, followed by overnight incubation at 4°C, was additionally performed for cell cycle profiling by flow cytometry. Alternatively, 5 μg/mL Hoechst® 33342 (Thermo-Fisher Scientific, Waltham, MA, USA) was used to stain DNA for imaging purposes.

Fluorescence Time-Lapse Imaging

Images were acquired with a Leica Thunder Imager (Leica Microsystems, Wetzlar, Germany), equipped with a Lumencor Spectra X Light Engine (Lumencor, Beaverton, USA) for fluorescence excitation, a motorized stage and a Leica DFC9000 GTC camera. For non-adherent cells, images were acquired with LAS X software (Leica Microsystems, Wetzlar, Germany, version 3.7.5.24914) using a 20X/0.75NA air objective and a binning 2x2 was applied to increase the SNR. The mCherry and mVenus signals were detected respectively with 540-580 nm and 460-500 nm excitation filters, 585 and 505 nm dichroic mirrors and 592-668 nm and 512-542 nm emission filters. The brightfield channel was also acquired for representation purposes. We imaged 20 to 25 fields of views per well and focal points were manually set in each position before starting the acquisition and kept constant during the whole time-lapse thanks to the Adaptive Focus Control (AFC, Leica Microsystems). The total duration of the time-lapse on non-adherent cells was 72 hours, and the time interval was set to 1 hour to prevent cell phototoxicity.

The total duration of the time-lapse on MDA-MB-231 adherent cells was 120 hours, and the time interval was set to 30 minutes.