Isolating live cells after high-throughput, long-term, time-lapse microscopy

Abstract

Single-cell genetic screens can be incredibly powerful, but current high-throughput platforms do not track dynamic processes, and even for non-dynamic properties they struggle to separate mutants of interest from phenotypic outliers of the wild-type population. Here we introduce SIFT, single-cell isolation following time-lapse imaging, to address these limitations. After imaging and tracking individual bacteria for tens of consecutive generations under tightly controlled growth conditions, cells of interest are isolated and propagated for downstream analysis, free of contamination and without genetic or physiological perturbations. This platform can characterize tens of thousands of cell lineages per day, making it possible to accurately screen complex phenotypes without the need for barcoding or genetic modifications. We applied SIFT to identify a set of ultraprecise synthetic gene oscillators, with circuit variants spanning a 30-fold range of average periods. This revealed novel design principles in synthetic biology and demonstrated the power of SIFT to reliably screen diverse dynamic phenotypes.

Fig. 1 |. Overview of SIFT.

a, Schematic of the screening process. Any pooled library or complex culture (mixture of colored ovals) can be loaded into the microfluidic device and characterized by long-term time-lapse microscopy. After phenotyping, cells of interest (red cylinders) are transported by optical trapping to a clean area on the chip—previously sterilized with the support of a series of push-down valves—and individually collected from the device without cross-contamination. Isolated cells may be propagated for any downstream off-chip analysis. b, A hypothetical example of two different genotypes (represented by generic wild-type (W) and mutant (M) forms) that are frequently mischaracterized by individual single-cell snapshots (colored circles). Only by estimating the full distributions of the phenotypic property (right panel), by assaying the same genotypes over long periods of time (left panel), are the correct classifications reliably assigned.

Fig. 2 |. Cell retrieval principles of the SIFT platform.

a, A macroscopic image of the entire microfluidic chip, loaded with dyes for visualization, is shown in the left panel. Dotted boxes correspond to regions represented in other panels. Scale bar, 5 mm. A schematic overview of the single-cell isolation process is shown in the right panel, corresponding to a not-to-scale representation of the blue boxed region from the left panel. Cell(s) from a lineage of interest (marked by red star) are optically trapped and transported to a clean section of the chip—segregated by a series of collection valves (green box)—for further characterization and/or immediate collection. b, Kymograph of cell transportation via optical trapping. An Escherichia coli cell of interest is removed from a growth trench (leftmost two panels) and dragged through an open collection valve (rightmost two panels) for off-chip collection. White circles mark the optical trap position; dotted arrows indicate the movement direction of the trapped cell. Similar cell transfers were performed more than 200 times across 5 independent screening runs with similar results. Scale bar, 5 μm. c, On-chip inlet-cleaning valve infrastructure. A top-down representation of the entrances for one microfluidic lane (left panel; magnified representation of the orange dotted box from the left panel of a) shows deposited bacterial biofilms cleaned by circulated bleach (arrows), while downstream cells residing in growth cavities are protected by a closed push-down valve (red indented box). A cross-sectional view of the push-down valves (right panel; magnified depiction of dotted grey box in the left panel) shows valve actuation. Pressurization of a ceiling chamber causes deflection of a thin PDMS layer that pinches flow channels (yellow) closed. d, Images of a feeding lane inlet before (left panel) and after (right panel) biofilm removal by the chip-cleaning procedure outlined in c. Similar results were observed across five independent screening runs. Scale bar, 25 μm.

Fig. 3 |. Clean, reliable isolation of live individual cells by SIFT.

a, Sample fields-of-view from the mock color screen comprised of overlaid phase contrast and fluorescence images. E. coli were engineered to constitutively express either red, yellow, cyan or no fluorescent protein (rFP, yFP, CFP or no FP), and mixed to a ratio of 1:1:1:100, respectively. Scale bar, 5 μm. b, Collection of the chip flow-through prior to SIFT screening the mock library of mixed colored cells (left panel). Three out of three individual cells were isolated from each targeted lineage in the intended sequence. Each agar well is a plating of an independent collection bin; numbers denote the order of collection. Similar results were observed across three independent mock screening runs. Scale bars, 10 mm. c, number of genomic mutations specific to 15 isolated single cells transported via an optical trap for 15 s, 30 s and 75 s (the standard trapping time, two-times longer and five-times longer, respectively). All cells were continuously moved for the entire duration of the respective trapping times to mimic screening transportations, as shown in the schematic. Data were gathered from a single screening run. d, Growth-rate-derived generation times immediately following optical transport for varying times, ranging from just over the standard trapping time (15 s) to greater than 5 times longer. Mean generation times (red circles) were measured over a 12-h observation period, with error bars as s.e.m. Mean generation time of 50 non-trapped cells (dotted line; μ) and twice the standard deviation (shaded region; 2σ) represent natural growth heterogeneity within the same experiment. Data were gathered from a single screening run.

Fig. 4 |. Genetic screen of a dual-feedback oscillator library.

a, The dual-feedback oscillator with linked positive and negative feedback loops (top left panel). All circuit genes, including the GFP reporter, were encoded on two plasmids, transcriptionally regulated by identical P_lac/ara-1 hybrid promoters, and marked for protein degradation by SsrA (LAA) tags in E. coli (top right panel). Library mutations were targeted to promoter, operator and rBS sites of both lacI and araC^const, the araC gene with a point mutation (y13H) conferring constitutive activator functionality (bottom panel). b, Statistical estimates of period CV and mean period of all library variants with oscillatory-like signals over a 24-h period using SIFT (N_osc = 7,803). Variants selected to highlight phenotypic diversity (grey circles with roman numerals) have time-traces shown in c. red circles represent variants that were isolated, some of which were individually characterized in a follow-up mother machine run. Mean generation time was 25 min. Library phenotype data were gathered from a single screening run. c, Time-traces of selected variants, as described in b. Orange dots indicate called peaks. <T>, mean period; CV_T, period CV; a.u., arbitrary units. d, representative time-traces of the original dual-feedback circuit (SL126) and the best-performing (that is, with the lowest period CV) isolated mutant (SL278), characterized in the mother machine under conditions without any supplemented IPTG or arabinose.

Fig. 5 |. Genetic screen of a dominant-negative repressilator library.

a, An expression library encoding various levels of constitutively expressed (P_const) dominant-negative Tetr (Tetr^−d) was integrated onto the backbone of the repressilator circuit in an E. coli ΔclpXP background. b, The mutant Tetr^−d dimerizes with WT Tetr and prevents binding to cognate operator sites, effectively sequestering a pool of TetR proteins. ORF, open reading frame. c, Statistical estimates of period CV and mean period of all library variants with oscillatory-like signals over a 24-hour period using SIFT (N_osc = 1,277). Variants selected to highlight phenotypic diversity (grey circles with roman numerals) have time-traces shown in d. red circles represent variants that were isolated with SIFT, some of which were individually characterized in a follow-up mother machine run. Mean generation time was 24 min. Library phenotype data were gathered from a single screening run. d, Time-traces of selected variants, as described in c. Orange dots indicate called peaks. <T>, mean period; CV_T, period CV; a.u., arbitrary units. e, representative single-cell time-traces of the original repressilator with an integrated reporter (top panel; SL305) and the best-performing (that is, with the lowest characterized period CV) Tetr^−d repressilator library isolate (bottom panel; SL229). f, Period histograms of the strains from e. Histograms include 8,161 periods from the original circuit and 4,779 periods from the top isolate. g, Autocorrelation functions (ACF) of strains from e.