A method for evaluating the use of fluorescent dyes to track proliferation in cell lines by dye dilution

Authors


Abstract

Labeling nonquiescent cells with carboxyfluorescein succinimidyl ester (CFSE)-like dyes gives rise to a population width exceeding the threshold for resolving division peaks by flow cytometry. Width is a function of biological heterogeneity plus extrinsic and intrinsic error sources associated with the measurement process. Optimal cytometer performance minimizes extrinsic error, but reducing intrinsic error to the point of facilitating peak resolution requires careful fluorochrome selection and fluorescent cell sorting. In this study, we labeled the Jurkat and A549 cell lines with CFSE, CellTraceViolet (CTV), and eFluor 670 proliferation dye (EPD) to test if we could resolve division peaks in culture after reducing the labeled input widths by cell sorting. Reanalysis of the sorted populations to ascertain the level of reduction achieved always led to widths exceeding the gated limits due to the contribution of errors. Measuring detector-specific extrinsic error by sorting uniform fluorescent particles with similar spectral properties to the tracking dyes allowed us to determine the intrinsic error for each dye and cell type using a simple mathematical approach. We found that cell intrinsic error ultimately dictated whether we could resolve division peaks, and that as this increased, the required sort gate width to resolve any division peaks decreased to the point whereby issues with yield made A549 unsuitable for this approach. Finally, attempts to improve yields by setting two concurrent sort gates on the fluorescence distribution enriched for cells in different stages of the cell cycle that had nonequivalent proliferative properties in culture and thus should be practiced with caution. © 2013 International Society for Advancement of Cytometry

Carboxyfluorescein succinimidyl ester (CFSE), CellTraceViolet (CTV), and eFluor 670 proliferation dye (EPD) are used to track cell proliferation in vitro and in vivo by flow cytometry [1-4]. These dyes work by entering live cells and fluorescently labeling cellular proteins through covalent interactions with amine groups [5]. When the cell divides, the fluorescence should be apportioned equally between daughter cells to generate subsequent populations with roughly half the parental intensity. Lipophilic dyes including PKH26/67 and CellVue Claret have also been successfully used [6-11]. Analysis of the division peak profiles can generate powerful metrics including the precursor frequency (% divided), mean division index, and the proliferation index (burst size) [12-14]. The measured width of the input population will dictate if generational peaks can be resolved [15]. The width or spread of a distribution will not only reflect the intrinsic variation of the particle or cell being measured, but also sources of error associated with the measurement process. As such, the true width of a population will be less than that observed due to the contribution of errors [16]. With regard to fluorescence measurements by flow cytometry, these errors can be classified as particle intrinsic and particle extrinsic. Intrinsic factors include particle size and uniformity, the spectral properties of the associated fluorescent label and the expression levels of binding targets [17-19]. Extrinsic factors derive mainly from the instrument and include the performance of the illumination source and detectors [18, 19]. It also includes free fluorochrome in the sample buffer. Extrinsic error can be minimized by optimal instrument performance whereas intrinsic errors are harder to limit due to the inherent biological heterogeneity and the choice of fluorescent reagent. For example, it is relatively easy to obtain excellent peak resolution with resting lymphocytes as they have minimal intrinsic variation/error. However, for more intrinsically variable cell types such as cell lines or stem cells even optimal labeling conditions and cytometer performance will not facilitate division peak resolution [9, 20]. One alternative to overcome poor peak resolution is to employ peak fitting algorithms to deconvolve meaningful data from the ever-broadening population of cells as they divide [12, 21, 22]. While this is a powerful approach, some approaches require high resolution of the divisional populations either for analyzing “proliferentiation” [14, 23-25] or for using elegant algorithms to assign generational numbers in the absence of nonproliferative controls [26]. An alternative approach employed by several groups including our own, uses cell sorting to reduce the spread of the input population and therefore limit the source of intrinsic variation associated with the cell [15, 20, 23, 26, 27]. When using the sort gate reduction method, measurement errors will mean that rerunning the sorted population through the system under the same conditions always leads to a measured width that is greater than the width of the initial sort gate [15, 23]. Without measuring and normalising for extrinsic error, it is impossible to know what degree of intrinsic reduction has been achieved by sorting the cells. It has been suggested that the contribution extrinsic factors make to this observed respreading can be determined by the analogous sorting and reanalysis of fluorescent beads with similar spectral properties to the dye under investigation [15]. As the intrinsic properties such as size and staining intensity of the beads should be largely uniform, any increase in width of the sorted population will derive mainly from the extrinsic component of the error sources [15]. We began by labeling Jurkat cells with CFSE, EPD, or CTV, sorting different populations using defined gate widths and then remeasuring the width of the resulting populations. We also performed the same sorting process using spectrally similar beads to determine channel-specific extrinsic error. In this way, we were able to deconvolve dye-specific intrinsic error and comment on which dye made the least contribution to measurement error. Furthermore, by culturing the different sorted populations, we could then comment on the threshold for division peak resolution based on input width. Second, we used Imaging Flow Cytometry (IFC) to assess the performance of each dye in the context of tracking cell division in vitro. Specifically, we measured the propensity for dye transfer as well as the degree of symmetrical inheritance during the division process, all of which will impact heavily on peak resolution independently of the input width. Third, we looked at the influence of increasing cell intrinsic error by labeling the more variable A549 cell line and performing an analogous sorting process. We found that while extrinsic errors were machine and configuration dependent, the cell-intrinsic error was relatively consistent for a given cell type. As the cell-intrinsic error increased the width of the sort gate required for peak resolution decreased and so did the yield to the point where using A549 cells in this assay became wholly inefficient. Lastly, we looked at the effects of setting two concurrent sort decisions on the labeled cell population to see if this could increase yield as well as generating equivalent populations for any downstream assay.

Materials and Methods

Fluorescent Dye Labeling of Live Cells

In all experiments, Jurkat E6.1 (6–12 µm) or A549 (12–20 µm) cells (both FHCRC derived clones, Cell Services, CRUK, passaged < 3 times) were labeled with either CTV, CFSE, or EPD as described previously [23] or in accordance with the manufacturer's guidelines. In all cases, dye levels were titrated to achieve the brightest (Median Fluorescence Intensity, MedFI), most uniform staining distribution (CV) and the best viability in culture for the cell lines used in this study (Supporting Information Figs. S1 and S4). Briefly, Jurkat cells (E6.1) or A549 cells, maintained in RPMI culture media (Invitrogen, Paisley, UK containing 10% FBS, Penicillin/Streptomycin, Glutamine, and 2-Mercaptoethanol) at 37°C/5% CO2 were harvested, counted and sized using a Vi-Cell (Beckman Coulter), washed once in serum-free media and resuspended in prewarmed PBS labeling solution (CTV, 4 µM Invitrogen, A10198, 0.1 µM CFSE, SIGMA, 21888, 4 µM EPD, 65–0840-90 eBiosciences) at a density of 4 × 106/ml at 37°C for 30 min with immediate mixing upon dye addition to cells [21]. FBS was added (10% v/v) for at least 10 min to absorb remaining free dye. Cells were washed into serum free media for immediate cell sorting. In certain experiments, Jurkats were also labeled with Dye Cycle Ruby (DCR, Invitrogen, V10273) at a concentration of 5 µM for 30 min at 37°C at a density of 0.5 × 106/ml to identify cells with G1, S and G2/M DNA content for cell sorting (Supporting Information Fig. S5D).

Cell Sorting

Cell sorting was performed using a CS&T calibrated BD FACS Aria III system (488 nm 40 mW, 633 nm 20 mW, 407 nm 30 mW, and 561 nm 50 mW) or BD Influx system (488 nm 100 mW, 640 nm 40 mW, 405 nm 50 mW, and 561 nm 150 mW) and a sorting strategy described previously [23]. Briefly, live dye-labeled cells were identified using PI (SIGMA 81845) dye exclusion in the yellow 585/15 channel (Influx 610/20), CTV in the Violet 450/40 channel (Influx 460/50), CFSE in the Blue 520/30 channel, and EPD in the Red 660/20 channel. Where appropriate, DCR fluorescence was collected in the 695/40 blue channel. Sort gate widths were set using the min and max co-ordinates reported through DIVA software (Aria, BD) or the min and max function within Sortware (Influx, BD) and binned to an 8-bit scale to correlate with the equation proposed by Nordon et al. [15]. Sorting was performed using either the 70 or 100 µm sort nozzles at 60 or 20 PSI respectively, a 16-16-0 sort mask (Influx, purity 1 drop mode) and a sample flow rate of ∼10,000 events per second at given sample density of 20 × 106/ml. Sorted droplets were collected into 15 ml Falcon tubes containing 200 µl of FCS. After sorting, a portion of sorted sample was analyzed with the same setting and flow rate to measure the width of the post-sort populations. Width was calculated using the Full Width at Tenth Maximum (FWTM) and an 8-bit scale. To assess machine-related measurement errors an analogous approach was employed using fluorescently labeled beads (Table 1) with matched spectral properties to the dyes under investigation [15]. Cyto-Cal Multifluour + Violet beads (Thermo Fisher, FC3MV) were used for machine QC (Supporting Information Fig. S2E).

Table 1. Details of the fluorescent microspheres used to determine the extrinsic error contribution for each detector channel.
Commercial nameSupplierCat NoSize (μm)Detector
FITC CalibriteBD340,4866Blue 530/30
Fluoresbrite YGPolysciences17,1553Blue 530/30
Fluoresbrite YGPolysciences18,14010Blue 530/30
FlowCheckBeckman Coulter6,605,35910Blue 530/30
FITC BeadsBangs Laboratories8917Blue 530/30
AlignFlow Plus UVInvitrogenA73056Violet 450/40
Violet Laser BeadsBangs Laboratory9155Violet 450/40
APC Calibrite BeadsBD340,4876Red 660/20
Accudrop BeadsBD345,2496Red 660/20
Drop Delay BeadSpherotecDDCP-70-27Red 660/20
AF647 BeadsBangs Laboratory8877Red 660/20

Cell Culture for Proliferation, Dye Transfer, or Telophase Analysis

Sorted or unsorted labeled Jurkat cells were resuspended in fresh complete growth medium (RPMI 1640 +10% FBS, 2-Me and Pen Strep) at a density of 0.5 × 106/ml. Typically, 200 µl of cells were cultured in 96 well round bottom plate. For dye transfer experiments by IFC, singly labeled Jurkat cells were cocultured in pair-wise conditions for 24–48 h. For telophasic analysis, labeled cells were cultured alone. At the defined harvest times, a portion of the cells were washed and stained (depending on the proliferation dye) with PI for dead exclusion or DAPI (SIGMA, D9542) and run on an LSR Fortessa (BD) to assess viability prior to fixation and staining the bulk culture population (if required).

Cell Staining for Cell Cycle Analysis by Conventional Flow Cytometry (CFC) and IFC

Sorted or unsorted cells were washed once in PBS then fixed in 70% ethanol (EtOH) for a minimum of 1 hour. Cell were then permeabilized with 0.1% Triton-X 100 for 5 min. Samples were stained with either mouse anti-pH3 Ser 10 (Cell signalling Inc, #9706S) or mouse anti-MPM2 (Millipore #05368) for 1 h at RT followed by 2 washes. Depending on the spectral properties of the tracking dye used, cells were incubated with antimouse AF488 or AF647 (Invitrogen, A-11001 and A-21236) for 45 min at RT in the dark. Samples were incubated with RNAse A and PI for 30 min at RT and acquired on a LSR Fortessa CFC system (using 50 µg/ml PI) or a FlowSight (Amnis corp) IFC system (using 1 µg/ml PI).

LSR Fortessa Acquisition and Analysis

The laser configuration of the LSR Fortessa was 488 nm (50 mW), 635 nm (40 mW), 406 nm (50 mW), and 561 nm (50 mW). CTV (Violet 440/40), AF488 (Blue 530/30), PI (Yellow 610/20), and AF647 (Red 670/14) fluorescence was collected and analyzed as described previously [23]. EPD (Red 670/14) and CFSE (Blue 530/30) emissions were also collected. Acquisition gates were set using FACS DIVA software (v 6.1.3) and postacquisition analysis performed using FlowJo software (Treestar). Briefly, if appropriate, live cells were identified by viability dye exclusion. Then single cells were identified based on the SSC-W parameter and intact cells based on the FSC-A v SSC-A parameters. For cell cycle analysis, the area and height of the PI signal was used to further eliminate doublet cells and the Watson pragmatic fitting algorithm was used to determine cell cycle phase statistics [28]. A minimum of 10,000 live, single intact cells was acquired for analysis.

FlowSight Acquisition and Analysis for Dye Transfer, Telophasic Identification, and Polarity Measurements

Cells were run on a 4-laser, ASSIST and Cyto-Cal bead calibrated FlowSight IFC system (Amnis Corp, Seattle) with Blue (2–60 mW) Violet (2 mW) and Red (2 mW) laser excitation set to avoid saturation in any of the spectral channels (CH2 CFSE or AF488, CH5 = PI, CH7 CTV and CH11 EPD, or AF647). A minimum of 20,000 single events were collected per sample using the INSPIRE software (Amnis) for coculture experiments and 100,000 total singlet and doublet cells for telophase analysis. For coculture samples, analysis was initially performed using IDEAS software (v5.0) to identify single objects using the Area and Aspect ratio of the default channel mask (M01, Supporting Information Fig. S3A). FCS files of total object channel fluorescence were then created from this population for analysis in FlowJo 10 (Tree star). Analysis of signal symmetry in telophasic cells was conducted as described previously [23-25, 29] with key modifications shown in Supporting Information Figure S3B. Briefly, single and doublet cells were identified using the Area and Aspect ratio of the default BF channel mask (M01) and mitotic cells from within these were identified based on PI intensity (4N) and MPM2 positivity based on the total fluorescence of the default channel masks (Supporting Information Fig. S3B, upper panels). Masking adaptations were made (Supporting Information Fig. S3C) in order to subdivide prophase, metaphase, and anaphase/telophase cells based on aspect ratio and nuclear spot counts as shown (Supporting Information Fig. S3B, lower panels). Anaphase cells and Telophase cells were then further subdivided using the aspect ratio and area of the MPM2 channel images (M11). The relative fluorescence of CFSE, CTV, EPD, and PI were measured by analyzing exported 16-bit TIFF images of validated telophasic cells into ImageJ (NIH) as shown in Supporting Information Figure S3D. Polarity was calculated as described previously [25].

Determining Cell-Intrinsic Contribution to Measurement Error (Uncertainty)

If the error or uncertainties within the measurement system can be approximated by normal or Gaussian distributions then the measurement uncertainty can be derived from the intrinsic and extrinsic uncertainty contributions [16]. The standard deviation of the measurement uncertainty, σmeasured is then given by the equation:

display math(1)

where σintrinsic and σextrinsic are the standard deviation of the intrinsic and extrinsic uncertainty or error contributions respectively. This simple assumption allows us to derive approximate contributions of the uncertainty in the measurement of the control beads and the uncertainty due to heterogeneity of the stained cells. Using this framework, we assume that the extrinsic standard deviation of the measurement process can be derived from the channel width of the postsort fluorescence intensity histogram of the beads, which includes effects such as instrument error (due to illumination, collection, and detection characteristics) and also bead variability (negligible). The total measurement uncertainty can be derived from the channel width of the post sort fluorescence intensity histogram of the cells that includes all the measurement uncertainties when using the beads but also the effect of the heterogeneous cell population. We do not attempt to deconvolve the effect of gating on the uncertainties on the postsort histogram width instead we have determined the uncertainty for each sort gate measurement. Clearly, it is impossible to deconvolve the uncertainty due to the bead size in a robust manner from this analysis, however, by comparing the intrinsic and extrinsic uncertainty of the same cell line using a variety of different control beads and different laser excitations we are able to comment on the limitations of this issue. Furthermore, we also performed a simpler, more accessible calculation:

display math(2)

where wi is the intrinsic width of the population derived from the total measured width (wt) minus the contribution of extrinsic errors (we).

Results

Intrinsic and Extrinsic Factors Contribute to Measurement Error and Population Width

First we determined the optimal conditions for labeling Jurkat cells with CFSE, CTV, or EPD. We titrated the three dyes and measured the MedFI, CV postlabeling, and culture viability at 24 h (Supporting Information Figs. S1A–C). The optimal labeling concentrations for CTV and EPD were 4 μM whereas even though 1 μM seemed optimal for CFSE, due to unexpected toxicity at 48 h, we had to use 0.1 μM of CFSE with Jurkat cells (Supporting Information Fig. S1D). Prior to sorting, the EPD labeled population width was 57 ch of an 8-bit scale whereas CFSE and CTV were 66 and 82 ch, respectively (Fig. 1A). In all cases, the intensity of the dye labeled cells exceeded the brightest bead in the 6 peak Cyto-Cal set (Supporting Information Fig. S1E) indicating the likely sources of measurement errors [18, 19]. To reduce particle intrinsic variation we set a series of sort gates across each fluorescent distribution and remeasured the sorted populations. Immediate reanalysis of the sorted populations revealed that EPD labeled Jurkats presented with the lowest increase in width, with CFSE exhibiting the largest increase (Figs. 1A and 1B). Furthermore, we also noted that as the width of the sort gate increased, there was an overall decrease in the amount of respreading. This was expected due to the fact that as the gating width increased the extrinsic effects will be included in the gate and will tend to 0. We did not observe any change in the overall MedFI values after sorting, showing that the dyes were stability retained during this period of analysis. Due to the decreased width of the EPD input population we were able to achieve higher sort yields from each gate set compared to CTV and CFSE (data not shown). When we cultured the sorted cells for 72 h. we found that the threshold for division peak resolution for all dyes was a sort gate width of 10 ch and a post sort width of <25 ch. Interestingly, however, the quality of peaks for CFSE and EPD were poor in comparison to CTV, particularly as the post sort width of each input population was roughly equivalent (Fig. 1C). It was impossible to ascertain the success of the sort gate reduction with respect to limiting intrinsic variability due to the contribution of extrinsic errors to the population width. To this end, we used a modification of the method proposed by Nordon et al. to measure the extrinsic error of each dye- specific detector channel [15]. Fluorescent beads were sorted using the same channel widths as the dye-labeled Jurkats and reanalyzed. We used multiple bead sets of different sizes containing varying dyes all with spectral properties similar to CFSE (Supporting Information Fig. S2A), CTV (Supporting Information Fig. S2B) or EPD (Supporting Information Fig. S2C). Averaging the spread of all the beads in each channel and plotting this out we could estimate what contribution each particular channel was making to the channel width increase and that the highest level of error derived from the violet 450/40 channel (Fig. 1C and Supporting Information Fig. S2D). This conclusion was also supported by the Cyto-Cal 6 peak bead data we obtained prior to each sort showing the violet channel performance was inferior to the other two detectors as judged by peak width (Supporting Information Fig. S2E). Using Eqs. (1; Fig. 1D) and (2; Supporting Information Fig. S2D) to derive the intrinsic error contributed by each dye we found CFSE contributed most to intrinsic error but both EPD and CTV performed equally well. The cut off for peak resolution was judged to be ∼19.5 ch by Eq. (1) and ∼9 ch by Eq. (2) based on Figure 1C.

Figure 1.

Reducing width of CFSE, CTV, and EPD labeled Jurkat cells by cell sorting reveals a respreading phenomenon that is influenced by error in the measurement of fluorescence. A: Jurkat cells were labeled with the indicated proliferation tracking dyes and sort gates of increasing channel widths were set through the peak channels using an 8-bit scale. The initial width of the un-sorted populations as well as for each reanalyzed post-sort population is shown in the top left of each plot as well as the MedFI (BD Aria). B: A graph showing the post sort channel for each sort gate set on labeled populations of the indicated dyes (C). The cells from A were placed into culture for 72 h and then analyzed for division peak resolution. The post sort width values are shown in the top left of each respective plot (LSRFortessa). D: A graph showing the intrinsic error associated with each dye using Eq. (1) and the values for measurement uncertainty and extrinsic uncertainty or channel width shown in Figures 1A and 1C. The dotted line denotes the threshold for division peak resolution base on data in C. In all cases, the mean +/− SEM of four independent experiments is shown.

EPD Division Peak Resolution Is Affected by Culture-Dependent Variability

While CFSE contributed heavily to the respreading of the post-sort population whereas EPD and CTV did not, we wanted to understand why EPD gave such poor peak resolution in culture. To this end, we colabeled Jurkat cells with CTV and EPD and sorted them using a 3 ch width in both fluorescence detectors to then follow the “quality” of the peak resolution in the same cells. Figure 2A shows that at 24 h both dyes gave well-resolved peaks, however by 72 h the EPD signal resolution reduced while the CTV remained unaffected. This suggested a culture-dependent source of biological variation with regard to EPD. It has been shown previously that EPD transfers to bystander cells in an in vitro setting and this randomised distribution of fluorescence could increase signal measurement spread and thus decrease peak resolution [3]. We wanted to directly test this hypothesis in our system by looking at the uptake of CFSE, EPD, and CTV by cells in co-culture conditions using IFC. Over a 48 h period, we could observe that all conditions lead to some evidence of dye transfer, but that EPD was by far the most heavily transferred, showing an increase of ∼1.5 logs. We used the IFC-derived multispectral imagery to confirm that the transfer was intracellular and not simply due to small cell-associated debris that would not be identified by pulse-derived doublet exclusion techniques (Fig. 2B). Interestingly, we found that the pattern of staining for EPD was very different from that of CFSE and CTV in that it was not diffuse throughout the cell, but rather restricted to a defined peri-nuclear cytoplasmic position. This led us to postulate that a further reason for the poor division peak resolution may be due to decreased symmetrical inheritance of the EPD labeled structures within the cells due to stochastic mechanisms [24]. The whole premise of proliferation dyes is that the signal must be apportioned to each daughter within symmetrical limits as the cell divides. We stained cells with CTV, CFSE, or EPD and cultured them for 24 h and used a modification of our previously described method for identifying late stage mitotic cells [23] to then measure signal distribution over the daughter poles (Supporting Information Fig. S3). Using our approach, we were able to identify cells in telophase and then measure the distribution of either CTV, CFSE, or EPD across the daughter poles using Image J as described previously [25]. Figure 2D shows that by calculating the polarity of each signal across the daughter poles, EPD was inherited in a significantly less symmetrical fashion than either CTV or CFSE. Interestingly, CFSE was significantly more symmetrically inherited than CTV [23-25].

Figure 2.

Division peak resolution is also affected by culture and proliferation dependent sources of variation. A: Jurkat cells were colabeled with CTV and EPD then sorted to an optimal width in each fluorescent channel. Cells were then cultured and the quality of peak resolution was determined for each dye on a LSRFortessa (B) Analysis of dye transfer to bystander cells under pair-wise coculture conditions using IFC (FlowSight). Bivariate plots show the level of fluorescence of labeled cells cultured alone (gray population) in each channel versus the mixed condition (black population) for each possible pair-wise combination. Selected multispectral single cell images are shown (20×) from the area of each bivariate plot denoted by the black arrows. C: Multispectral images (20×) of MPM2+ telophasic cells showing distribution of CFSE, CTV, and EPD across the daughter poles prior to completing cytokinesis. D: Quantification of proliferation tracking dye fluorescence signal across the daughter poles expressed as a polarity score with the PI fluorescence included as an internal control. One hundred telophasic cells were analyzed per group. Dotted line denotes the threshold for being considered polarized (>0.2). Significance was determined using a 1-way analysis of variance (CFSE v CTV = **, CFSE v EPD = ***, and CTV v EPD = ***, P < 0.05). The % of cells with a polarity score >0.2 is shown. [Color figure can be viewed in the online issue which is available at wileyonlinelibrary.com.]

Intrinsic Error Is Cell-Type Specific and Independent of Instrument Configuration

For the next set of experiments, we wanted to investigate what influence an increase in cell-intrinsic variability and/or instrument associated extrinsic error would have on the respreading of sorted cells. As a model for increased cell-intrinsic error, we chose the larger and more heterogeneous A549 cells line. First, we determined the A549 specific-labeling efficiency and toxicity for the three dyes (Supporting Information Fig. S4 A–C). CFSE was not well tolerated at concentrations labeling with an MFI and CV suitable for sorting. Moreover, EPD again presented with the punctate staining pattern seen with Jurkats (Supporting Information Fig. S4D) and transferred heavily to bystander cells (Supporting Information Fig. S4E). Collectively, CTV remained the clear choice for A549 cell analysis. Sorting A549 cells necessitated a change in instrument settings including increased nozzle size and drop in system pressure. As a control, we also sorted CTV labeled Jurkat cells and violet excited beads using the same instrument configuration ensuring that our only variable was the cell type, as machine configuration (extrinsic error) and dye-intrinsic error was constant. Finally, we also analyzed CTV labeled Jurkats and violet beads on an Influx cell sorter system. Figure 3A shows the CTV channel distributions for unsorted A549 cells and Jurkat cells on the Aria and the latter on the Influx system. When we plotted the channel width of the post sort populations, we found that the A549 cells presented with the greatest increase, followed by the Jurkat cells sorted with the 100 μm nozzle, then Jurkat cells sorted with the 70 μm nozzle. Interestingly Jurkats sorted on the Influx with the 100 μm nozzle gave the lowest increase in channel width. Measuring configuration and machine-dependent extrinsic error using beads revealed that at the time of our study, the 100 μm nozzle Aria configuration contributed more to the respreading compared to the 70 μm set up. The 100 μm influx and the 70 μm aria were almost identical (Fig. 3C). Finally, by deconvolving extrinsic error values using Eq. (1); Fig. 3D) and Eq. (2); Supporting Information Fig. S2F) we found that CTV labeled Jurkat cells had comparable intrinsic error. Interestingly, but perhaps unsurprisingly, A549 intrinsic error could only be reduced below the threshold for division peak resolution using a 3 ch sort gate, correlating with division peak resolution in culture that was equivalent to those obtained using a 10–15 ch sort gate on Jurkats cells (Supporting Information Fig. S4F).

Figure 3.

Intrinsic error is cell type dependent, independent of instrument type or configuration and ultimately determines peak resolution. A: Histograms showing the presorted and postsorted widths of CTV labeled A549 and Jurkat cells sorted using the indicated instrument and configurations. The eight bit channel widths are shown in the top left for each plot. B: A graph showing the increase in the post-sort channel width for each sort gate set on CTV labeled Jurkat or A549 cells under the indicated configurations. C: A graph showing the contribution extrinsic error makes under machine-dependent conditions to the respreading values shown in B using UV Align-plus microsphere beads. D: The cell intrinsic error derived using Eq. (1) from the values shown in B and C. The dotted line denotes the cut off for empirically determined division peak resolution.

Sorting Concurrently on Different CTV Intensities Selects Nonequivalent Populations

Having established that CTV was best suited for Jurkat and A549 cell lines, but that cell-type specific intrinsic error remains the overriding factor as to whether sort-gate reduction can facilitate division peak resolution in culture we looked at other important technical considerations surrounding this approach. As the maximal sort gate width for division peak resolution limited us to a projected yield for Jurkats between 30 and 16% (10 ch) and 6% for A549s (3 ch; see MIFloCyt checklist) we investigated the efficacy of sorting multiple populations across the CTV labeled distribution by setting two concurrent sort decision (Fig. 4A). It has been shown that G1 and G2/M cells differ in size and contain different levels of protein that may diversify the CTV-binding capacity [30] we also monitored forward scatter (FSC-A) values for the high (hCTV) and low (lCTV) sort populations to see if there was any bias within our sort populations [2, 26]. While we saw no significant difference in the FSC-A median of hCTV and lCTV cells (Supporting Information Fig. S5A) they had very different proliferative traces in culture (Fig. 4B) translating to a significant increase in the percentage of cells entering division at 24 in the hCTV population compared with the lCTV (Fig. 4C). To investigate the reason for this difference, we determined the cell cycle phase distributions of ethanol fixed hCTV and lCTV cells by staining with PI and pH3 (Fig. 4D) Figure 4E shows that within the sorted lCTV gate, significantly more cells were in G1 (∼70%), whereas in the hCTV sorted population, significantly more were in the S and G2/M phases (∼58%). Both EPD and CFSE also showed a higher staining intensity in cells from the G2/M phase (Supporting Information Fig. S5B). Moreover, culturing our hCTV and lCTV populations under serum limiting conditions showed that lCTV cells were dependent on serum for proliferation while hCTV sorted cells were not suggesting lCTV cells began the culture period predominantly in G1 and are dependent on serum for S-phase transition (Supporting Information Fig. S5C). We reasoned that the proliferative differences were due to hCTV-sorted cells being enriched for cells in later stages of the cell cycle compared with lCTV cells. To test this, we used the viable DNA stain, DCR, to exclude G2/M cells prior to setting hCTV and lCTV gates (Supporting Information Fig. S5D). Figure 4F shows that preselecting against G2/M cells ablated any proliferative advantage. As there was clear correlation between the nonquiescent nature of cell lines, protein content and fluorescent labeling width, we serum starved cells prior to labeling as a way to overcome the need for sorting (Supporting Information Fig. S6A). We also investigated if sort gate reduction using SSC-A, a better measure of object size [31], could be used to reduce cell heterogeneity prior to dye labeling (Supporting Information Fig. S6B). In both cases, we found that we could reduce intrinsic variation, but not enough to achieve peak resolution based on channel widths.

Figure 4.

CTV staining intensity dictates proliferative capacity of the sorted population. A: CTV-labeled cells were subjected to two concurrent sort decisions differing in intensity (Aria). Histograms showing CTV fluorescence of the total labeled population with the lCTV (low) and hCTV (high) sort gates shown as black lines (upper panel). The postsort CTV fluorescence distribution of each sort population is also shown (lower panels). B: CTV proliferation profiles for lCTV (upper panels) and hCTV (lower panels) sort populations after 24 and 48 h in culture (LSRFortessa). C: A graph of the % divided at 24 h for cultured hCTV and lCTV sort populations. D: Ethanol fixed cells from unsorted, lCTV sorted and hCTV-sorted populations were stained with PI and pH3. Bivariate plots are shown with gates positioned to report the frequencies of G1, S, G2, and M phase cells (LSRFortessa). E: A graph showing the % of cells in a given cell cycle phase in either unsorted, hCTV sorted or lCTV-sorted populations. Frequencies for G1, S, and G2 are scaled to the left y-axis and M-phase frequencies to the right y-axis. E: A graph of the delta in the % divided values at 24 and 48 h. between hCTV and lCTV sorted populations with and without the elimination of non-G1 cells based on DCR staining. In all cases, data are the Mean +/− SEM of at least three independent experiments. Significance was determined using students t-test (* = P < 0.05, and ** = P < 0.001). [Color figure can be viewed in the online issue which is available at wileyonlinelibrary.com.]

Discussion

While the sorting of dye-labeled cell lines to reduce intrinsic variation for the purpose of division peak resolution has been described previously [15, 20, 23, 26, 27], this study specifically provides an approach that can determine the suitability of a given dye and cell type for this method. The fluorescence width of CFSE, CTV, or EPD cell populations will be determined by intrinsic factors from the cell type in question as well as the extrinsic factors linked to the measurement process. For example, the levels of cellular protein to which these dyes bind will affect spread and if a population of cells has highly variable levels of protein content this will be reflected in the population width. In support of this statement, we saw that the width of EPD labeled Jurkat cells was reproducibly less than CTV and CFSE, correlating with the microscopic observation that EPD bound to a more uniform cellular niche than the other dyes. Considering that the sort gate width will influence yield, we wanted to determine the threshold for the reduction of intrinsic variation that would achieve peak resolution in culture. We found that determining this threshold was confounded as all forms of measurement are subject to errors, so in practical terms, when we reanalyzed the sorted populations to determine how much we had narrowed the distributions, we found that the population widths always exceeded the sort gates set. It was therefore important to limit these sources of error wherever possible and determine their contribution to the increase in population width. As tracking dyes are extremely bright and exceed the intensity levels of peak 6 in the Cyto-Cal bead set, the major sources of error will derive from the illumination sources of the cytometer as well as cell-intrinsic factors that include the photonic properties of the dye itself. [18, 19]. For example, a dye with a low quantum efficiency may not emit a photon for every one absorbed meaning the level of recorded fluorescence may not reflect the amount of dye actually bound to the cell [32]. To investigate this source of error, we selected three tracking dyes and tested how they performed in our assay. Our criteria for a “good dye” was that it should stain brightly with low toxicity, contribute little to the post-sort width increase, be retained well by cells in culture [3] and be inherited in a symmetric, nonpolarized fashion to deliver clear division peaks [23-25]. Determining the dye-specific contribution to measurement error was complicated by the fact that each dye was excited by a different laser and measured in different detectors. Therefore, the ability to normalize for channel specific extrinsic sources of error was paramount to any conclusions regarding the optimal dye for the cells tested. We did this by sorting channel-matched fluorescent beads [15], measuring the postsort increase in channel width and using these values to deconvolve the dye-specific contribution. One caveat to this approach was that in some cases, the bead-associated dyes were not identical to the tracking dyes, however, by analyzing multiple bead sets for each detector and averaging the values, we largely negated this issue. Although EPD showed promise due to tight initial labeling and increased sort gate yields combined with a low contribution to respreading, we found serious issues when the cells were placed in culture. First, as noted by others, EPD transferred to bystander cells in vitro [3], while in vivo this may be less of an issue, possibly due to proximity [33]. Worryingly, we also found that the chances of symmetrical EPD signal inheritance was reduced compared with CTV. Unfortunately, due to intellectual property covering CTV and EPD we can make no conclusions as to the chemical or biological basis behind the different retention and staining pattern of EPD. We speculate that EPD is still binding protein via amine groups, but only within certain cellular structures that are used for the uptake process that may include endosomes. While endosomes have been shown to be symmetrically inherited [23, 34], if not every endosome is labeled then a stochastic process of inheritance may explain why we see some polarized EPD signal inheritance [24, 35]. The issue of asymmetric EPD inheritance could have serious ramifications for its use as a cell proliferation tracking dye. Immunologically, if a population of EPD labeled T cells are exposed to a specific antigen, asymmetric, or polarized inheritance of signal could mean that the calculation of division parameters may be misleading. Interestingly, while CFSE was the most symmetrically inherited of all the three dyes, it also contributed to respreading and also gave poor peak resolution in culture. As fluorescein is considered to have good quantum efficiency [32] it is unlikely that this is the explanation. It has been shown that CFSE intensity will decay independently of division, likely due to efflux and catabolism, throughout the lifetime of the labeled population [26]. It is possible that differential rates of dye loss within the labeled population may explain the broadening of the division peaks as even supposedly homogenous cell lines contain genetic variations that may influence dye retention. We saw no evidence of either a decrease in the median of the sorted populations or a shoulder upon immediate reanalysis, suggesting that efflux/catabolism was not playing a role here. However, we cannot rule out an effect later during prolonged culture that may influence division peak resolution. Nonetheless, we would include this as a source of intrinsic variation/error. Ultimately, our study suggests that it is intrinsic variation/error that determines peak resolution. Nordon et al. also reported that once extrinsic error was measured and accounted for, a sort gate width of 10 ch on an 8-bit scale was the upper limit for peak resolution [15]. The equation proposed by Nordon et al. suggests that a total width of 19 ch is the upper limit for peak resolution, whereas we empirically determine it to be around 25 ch. The discrepancy may be explained by the fact we used FWTM to measure width whereas they presumably used Full Width at Half Maximum (FWHM). We noted that no matter the instrument or configuration used, once extrinsic error was measured and normalized for, the intrinsic error of CTV-labeled Jurkat cells was highly consistent. This all but eliminated any concerns we may have had with the use of beads with different dyes to the tracking dyes in question. The combination of low levels of toxicity, intrinsic error and transfer to bystander cells combined with symmetrical inheritance and well resolved division peaks made CTV the natural choice for labeling the more heterogeneous A549 cell line. However it is possible that different cell types may handle these dyes in a different manner therefore we still recommend empirical testing for a given cell type and dye combination. We simply offer a framework with which such a test can be done and strongly encourage that all aspects we report here are considered for a given cell type, including testing the labeling efficiency and toxicity of multiple dyes in a given cell type. For example, we are also looking at the performance of lipophilic dyes in cell lines [36] using our metrics as although they are used successfully to track division [6, 10], peak resolution is inferior compared with amine-reactive dyes even when sort reduction has been used [9]. In terms of the A549 line, the intrinsic error associated with these cells when using CTV meant that only a 3 ch sort gate could reduce intrinsic error enough to obtain some peak resolution. Even then, this was dependent on ensuring cells were used after no more than 2–3 passages (data not shown). Such a highly variable particle or cell will be more prone to measurement error as it passes through the cytometer compared with more uniform lymphocytes or beads. Particle size may influence orientation, affecting how consistently the laser excites the fluorochrome and how well the optics can capture the emitted photons. It was unfortunate that A549 cells were incompatible with this method as they are widely used in cancer-based research, often requiring additive or subtractive synchronization conditions to produce cohorts for biochemical analysis upon release back into cycle [37]. Synchronization can be a source of error [38, 39] and multiparameter, single cell flow cytometry can circumvent this [40]. However, asynchronously proliferating populations not only contain cells in different stages of a given cell cycle; they may also be temporally separated by a minimum of one division round. One particularly powerful temporal method is the BrdU-Hoechst quenching technique [41], the technical limitations of which can be largely overcome by substituting Edu for Brdu [42]. Tracking dye analysis provides key parameters such as the % divided, burst size and measuring “proliferentiation” [14, 20]. Ultimately, our data shows that for highly variable cell types such as A549 or in situations where cell numbers are limiting, sorting would be inappropriate. Instead, one would have to rely on mathematical deconvolution techniques to make largely qualitative measurements of the proliferative response [22]. As long as the algorithm and parameters used are constant, then any error should be normalised across all samples. Moreover, high resolution division tracking through the sorting method may actually aid mathematical fits [26]. However, for certain applications where the division history of a cell must be established by manual gating with a high degree of accuracy and precision such mathematical approaches may not be appropriate. For example, asymmetric cell division in the adaptive immune system has been suggested to be restricted to the first division [43], and when studying this mechanism it is imperative that we could accurately identify the division history of all cells [24, 25]. Attempting to overcome limitations of yield by setting two concurrent sort decisions on the fluorescent distribution selected for populations in different stages of the cell cycle with very different proliferative capacities. While this has been suggested previously as a possible caveat [15], it was not empirically shown. The same selective bias was noted for CFSE and EPD labeled cells, reflecting the fact that protein content varies during the cell cycle [30] and could also influence analogous sorts on transformed cells from patients as it has been shown previously that sorting on a fluorescent-distribution can yield cells with different functions and fates [44]. Attempts to synchronise cells through serum starvation were not successful as Jurkats, like many transformed lines, seemed resistant to this approach [45]. Even if full arrest in G1 had been achieved, our data suggest that the labeled width of the population alone would still exceed the limits of peak resolution. We have begun to use our system to assess the effects of anti-cancer drugs on asynchronously growing cell lines (Filby et al. manuscript in preparation) and are able to appreciate the temporal separation of cells within the same cell cycle stage but separated across subsequent division rounds down to the mitotic phase [23]. Therefore, we feel that the use of dye dilution assays on asynchronously growing cell lines is a potentially powerful method with many applications both from a conventional and imaging flow perspective as long as the intrinsic properties of the cell type allow. By applying this sort gate method and normalising the remeasured width using data from spectrally matched beads one can determine using simple mathematics whether or not high resolution division tracking possible in any cell type.

Acknowledgments

This study was conducted from March 2011 to June 2013. Data can be made available on request. The authors would like to thank Laura Bazley of the LRI FACS laboratory for helpful discussion and Nuria Martinez-Martin from the LRI Lymphocyte Interaction Laboratory for critical appraisal of the manuscript.

Ancillary