There is significant interest in using flow cytometry for the multiparameter analysis of small biological particles such as cell-derived microvesicles (including ectosomes and exosomes), as well as bacteria, viruses, and intracellular organelles. However, developing validated protocols to do this has proven to be a challenge, as results are greatly affected by pre-analytical sample preparation variables (1–3), by variation in staining, measurement and analysis procedures (4, 5), and by differences in instrument performance between different makes and models of flow cytometers (6), and even between the same instrument model. All of these factors are the subject of active investigation and discussion, which has raised awareness of all of the experimental details that can affect results. We have come to recognize a particularly subtle, yet pervasive artifact that we believe deserves attention.
In flow cytometry of cells, light scatter, generally forward angle, is often used to trigger the data acquisition system to make a measurement. Light scatter from cells is usually well resolved from background, and there is usually little ambiguity as to the nature of the events being detected. As particle size (diameter) decreases, light scatter intensity decreases in a nonlinear and measurement angle-dependent manner (7), so that for very small particles the light scatter intensity may be unresolved from that of incidental small particles in the sheath or sample solution or from optical or electronic noise from the instrument. The light scatter response of a flow cytometer has often been characterized with uniform polymer beads, which has led to the appreciation that different instruments with different angles of light collection perform differently and that side scatter generally exhibits better resolution than forward scatter, owing to the lower levels of background light scatter orthogonal to the excitation beam. Such studies have also led to the recognition that polymer beads are unsuitable as size calibration particles for application to membrane vesicles, as the latter have a lower refractive index, and thus lower scatter intensities compared to a polymer bead of the same diameter (8, 9). Thus, the current best practice appears to be to use polymer beads as reference particles to standardize instrument set up (6), rather than as size standards for calibration.
However, even when these factors affecting the measurement are recognized, there is a measurement artifact that occurs when light scatter is used to trigger detection of particles whose light scatter is below the noise threshold of the instrument, and whose origin lies in coincidence. We illustrate this artifact and propose that fluorescence triggering may be more appropriate for the measurement of small, dim particles.
Nile Red microspheres (0.53 μm diameter and 0.11 μm diameter) were from Spherotech (Lake Forest, IL). Working solutions of approximately 1 × 108 particles/mL were prepared in 0.1 μm filtered nanopure water using the vendor-provided mass concentration and the diameters and density of the particles according to the relationship N = (6W/3.14 × P × D3) × 1012 particles, where W is the weight of the polymer in grams, P is the polymer density (1.05 for polystyrene), and D is the particle diameter.
Samples were analyzed on a FACSCalibur flow cytometer (BD Biosciences, San Jose, CA) equipped with standard laser and filters using 0.1-μm filtered nanopure water as the sheath fluid. Each sample was run for 60 s at low sample flow, which was determined to be 0.124 μL/s as measured using Accucount counting beads (Spherotech). Samples were measured using either light scatter (FALS plus SSC) or fluorescence (FL2 channel: 488 nm excitation, 585/42 nm bandpass)) triggering, and data analyzed using FCS Express 4 (De Novo Software, Los Angeles, CA). The fluorescence intensity axis was calibrated using 8 peak Rainbow beads (Spherotech) that had been cross calibrated with Quantabrite PE beads (BD Biosciences) on that instrument. Data files have been deposited in the Flow Cytometry Data Repository (flowrepository.org).
The measurement of small, dim particles often requires working near the noise floor of the instrument. When detectors are operated at high gain a sample of filtered buffer will produce spurious events resulting from electronic or optical noise or signals from particles smaller than the pore size of the filter used. In typical use, trigger threshold levels are set to minimize the frequency of such background events to some minimal level, after which the sample is measured and the data analyzed and reported. However, if care is not taken to perform some necessary control experiments, the reported results can be wrong. It is instructive to illustrate this issue using a well-understood sample such as fluorescent microspheres.
We used two commercially available fluorescent polymer microspheres with diameters of 530 nm and 110 nm, and analyzed these at different particle concentrations using either light scatter or fluorescence channels to trigger detection (Figs. 1 and 2). A sample of filtered buffer was used to set the trigger threshold at a level such that the “noise” event frequency was ∼10/s.
As presented in Figure 1, the 530 nm particle is readily detected in both light scatter and fluorescence channels, and using a dual channel (FSC and SSC) trigger threshold or a single channel (fluorescence) trigger threshold (gray dashed lines). Single beads are clearly identified on both the side scatter and orange fluorescence channel, and give an absolute fluorescence intensity value of 4,900 mean equivalent soluble fluorophores (MESF)-PE using either triggering approach. Coincident events, which are approximately twice as bright, are also readily identified in both the SSC and fluorescence channels. Neither the light scatter nor fluorescence distributions change as the sample is diluted, though it is apparent that the fluorescence signals are more clearly resolved from background than the light scatter signals. Event rates dropped proportionally as the sample was diluted (Fig. 3), and the particle stock concentration, as estimated from the calibrated instrument flow rate and dilution of the bead working solution, is similar for both trigger channels (Table 1).
Table 1. Estimates of particle concentration and brightness using different trigger channels
Median FL intensity (MESF-PE)
[Bead] per mL
FSC + SSC
1.07 × 108
1.16 × 108
FSC + SSC
1.30 × 107
1.28 × 108
Presented in Figure 2 are histograms resulting from analysis of the 110 nm particles. Light scatter triggering produced scatter histograms that are indistinguishable from background, and a bimodal fluorescence intensity histogram consisting of a dim population indistinguishable from background and a brighter population of events with an apparent absolute intensity of ∼450 MESF-PE. As the sample was diluted, the event rate decreased and the bright fluorescent population disappeared, leaving only the dim background events. Since the dimmer population of events is clearly originating from the sample buffer and sheath, it is tempting and intuitive to set a marker on the “positive” events and report on these. Doing so gives a particle concentration ∼8-fold lower than expected based on dilution of the stock solution (Table 1).
However, analysis of this same sample using fluorescence triggering produces significantly different results. Fluorescence triggering reveals a single dim population barely resolved above the trigger threshold level with a median absolute intensity of ∼150 MESF, and a particle concentration consistent with that expected from dilution of the stock solution (Table 1). Serial dilution of the sample gives the expected decrease in event rate (Fig. 3), but even at the lowest dilution, the same fluorescence population distribution can be discerned above the background from a sample of buffer (Fig. 2B, 1:32 dilution). Note also that the side scatter distribution obtained using fluorescence triggering is below the trigger level required to minimize background counts in light scatter-based triggering. In other words, the signals are in the noise of the light scatter channel.
What is the source of the discordance between fluorescence and light scatter-based triggering of the 110 nm Nile Red beads? The “particles” detected and measured with light scatter-based triggering do not exist. The events detected are the result of the coincident occurrence of multiple particles in the probe volume at the same time. As seen in Figure 2B, the light scatter distributions determined using fluorescence triggering are not resolved from background noise and are, in fact, below the threshold level needed to perform light scatter-based triggering. If the signal level from a single particle in the trigger channel is below the threshold level then, by definition, the events detected are not single particles. Rather, it is the coincident occurrence of many particles in the probe volume that causes the trigger level to be exceeded and an event detected. Thus, the consequences of attempting to trigger detection on particles whose signal is below the noise threshold are not merely the failure to detect the particle of interest. In fact it is worse: detecting events that represent multiple particles, resulting in particle counts and intensity measurements that are wrong.
Here, we used a simple system of uniform fluorescent beads to illustrate this artifact. However, for the analysis of heterogeneous samples of small biological particles whose concentration is unknown, this artifact can easily go unrecognized and is likely responsible, at least in part, for the widely varying results reported on, for example, cell derived microvesicles. Indeed, it is now widely (but not universally) appreciated that, owing to their lower refractive index, such microvesicles scatter much less light than a polymer bead of comparable diameter. Thus, for a light scatter trigger threshold level set with reference to polymer bead, light scatter from single microvesicles of a similar diameter will not trigger detection. Very large vesicles (1 micron in diameter or above) may trigger detection, and measurements of such large vesicles may have value. However, the coincident occurrence of many small vesicles may also trigger detection, producing data that will be very difficult to interpret and reproduce. Coincidence is a general phenomenon in flow cytometry, and can strongly impact specific applications including cell sorting (10), analysis of platelet-granulocyte complexes (11) and, as we illustrate here, nanoparticle analysis (12). This phenomenon was recently documented in the analysis of microvesicles enriched from urine (9), for which it was given the colorful moniker of “swarm” detection. As shown here, coincidence in a single color measurement can produce incorrect particle intensity and concentration estimates. In a multicolor measurement, where different fluorescent labels are used to identify sub-populations of particles, coincidence can make it appear as if two labels are coincident on the same particle, when in fact they were each exclusive to different particles that happen to pass through the probe volume together. Thus, any claims for colocalization of two markers on a particle subset should be accompanied by demonstration that coincidence is not affecting the result, as well as appropriate calibration and reporting fluorescence intensity measurements.
This artifact is not specific to light scatter, but will occur whenever the trigger channel threshold is greater than the signal level of individual particles, and where the particle concentration is high enough such that a sufficient number of single particles occur coincidentally in the probe volume to trigger detection. However, we think it is most pervasive in light scatter-based triggering and suggest that, where possible, fluorescence triggering be used. In our experience, the signal to background ratio is generally higher in the fluorescence channel than in the light scatter channel. This is because optical filters reject scattered light, reducing optical noise, and most background particles in the sheath and sample fluids are not highly fluorescent. Fluorescence has the added advantage of well established protocols for calibration so that any two labs, using the same commercially available intensity reference and calibration beads, can confidently set their threshold levels in absolute units of MESF. Although ultrafiltration using increasingly small pore sizes can reduce the number and size background particles, these particles plus optical and electronic noise make the detection of the extremely dim scatter from small, low refractive index particles a very difficult challenge, in our experience. Moreover, the dependence of light scatter on refractive index and collection angle make standardization very difficult for light scatter-based triggering.
In summary, we illustrate an artifact that can occur when using flow cytometry to analyze samples consisting of particles whose signal levels are near the noise floor of the instrument. We show that for very small particles, fluorescence can provide improved resolution from background noise compared to light scatter. We recommend performing serial dilution and measurement of such samples to ensure that the population intensity distributions are not concentration dependent, which would be evidence of coincidence. Finally, we recommend that, where possible, fluorescence-based triggering be used so that trigger threshold levels can be reported in quantitative terms that can be shared between labs.