2.1.

Optical Measurement System

A schematic of the instrument is shown in Fig. 1. A 488-nm laser (Innova 200, Coherent Inc., Santa Clara, California) is focused by two cylindrical lenses onto a BD FACSCalibur flow cell with attached light collection lens (Becton Dickinson, San Jose, California). Side-scattered light is transmitted through a dichroic beam splitter (FF505-SD, Semrock, Rochester, New York) onto a Hamamastu (Bridgewater, New Jersey) R636-10 PMT. Fluorescence is reflected by the same dichroic, passed through a dove prism, dispersed by a spectrometer (f1.8i Holospec using a HVG-590 grating, Kaiser Optical Systems Inc., Ann Arbor, Michigan), and detected with a $1600\times 200$ pixel electron multiplying charge coupled device (EMCCD) (Newton model DU970N-UVB, Andor Technology PLC., South Windsor, Connecticut). Forward-scattered light is quantified with an amplified detector (PDA100A, Thor Labs, Newton, New Jersey) equipped with a beam stop to block direct laser light.

Fig. 1

A schematic of the spectral flow cytometer. Light is represented by colored lines, analog signals by narrow black lines, TTL pulses by two narrowly separated black lines and USB cables by double thick black lines. Components are described in the text.

The output of the PMT (i.e., the side-scatter signal) is split and one part sent to a simple noninverting amplifier (OPA37GP) with its output going to a USB data acquisition (DAQ) card (9221, National Instruments, Austin Texas). The other part is sent to a Tektronix (Beaverton, OR) 2445B oscilloscope for real-time visualization of side scatter and generation of a transistor-transistor logic (TTL) trigger for the EMCCD camera from the side scatter signal. Since the PMT generates a current of electrons, the side-scatter voltage measured by the high impedance oscilloscope and amplifier is negative. The camera is controlled with Newton Andor Software and the spectra from the camera are sent directly to the computer via a USB connection. The camera sends out a TTL signal when it is collecting data and this goes directly to the DAQ card without amplification. The output of the forward-scatter detector is amplified with the same type of amplifier as the side-scatter signal and sent to the DAQ card. Collection of data from the DAQ card via a USB connection is performed with Igor Pro software (Wavemetrics, Tigard, Oregon) including NiDAQmx Tools.

A “home-built” pressure system is used to provide sheath flow and a Harvard Apparatus (Holliston, Massachusetts) PHD 2000 syringe pump is used to generate sample flow. Flowing particles are hydrodynamically focused in the BD FACSCalibur flow cell.

The spectral resolution was measured by placing a 532-nm laser in the beam path and examining the width of the scattered laser line. Spectral calibration was performed using spectral calibration lamps and a 532-nm laser.

2.2.

Samples

Fluorescent polystyrene microspheres were used as controls for instrument setup, as well as for demonstration of the instrument’s ability to use spectral properties to separate particles with similar spectra. Two types of microspheres were chosen for their different sizes and spectral characteristics: flow check fluorospheres (Beckman-Coulter) are 10 μm in diameter and have a bright, two-peak emission spectra. Align flow beads (Invitrogen) are 2.5 μm in diameter and have a simpler emission peak. Microspheres were analyzed individually so that pure populations could be characterized, the samples were then mixed and the populations separated in software.

Tumorigenic rat fibroblast, MR1, cells were grown in Dulbecco’s Modified Eagles Medium (DMEM) supplemented with 5% (v/v) of fetal calf serum (HyClone Thermo Fisher Scientific, Waltham, Massachusetts), $100\text{units}/\mathrm{ml}$ of penicillin, and $100\mu \text{g}/\mathrm{mL}$ of streptomycin. To obtain cell suspensions for staining and subsequent flow-cytometric analysis, cells were harvested from monolayer cultures in the exponential phase of growth by treatment with 0.25% trypsin in a phosphate-buffer (pH 7.4) containing 1 mM EDTA and 25 mM HEPES. Cells were then fixed in 70% EtOH/30% phosphate buffered saline (PBS) and stored at $-20°\mathrm{C}$ until use.

MR1 cells were stained at a concentration of 2 to $3\times {10}^6\text{cells}/\mathrm{ml}$ with $5\mu \text{g}/\mathrm{mL}$ (7.5 μM) propidium iodide (PrI) and/or $3\mu \text{g}/\mathrm{mL}$ (7.6 μM) ethidium bromide (EB) in PBS containing $50\mu \text{g}/\mathrm{mL}$ RNase A. These fluorophors were chosen to be a stringent test of the spectral deconvolution abilities of our measurement and analysis techniques. These concentrations were chosen so that the concentration of molecules would be similar for the two stains. Cells were incubated in staining solution for 30 min at 37°C and stored for a short period at 4°C if not used immediately. Single-stained samples were split in half; with one half being used for measurement while the other half was used for mixing samples stained with EB and PrI. Single-stained cell samples were washed before data collection, or before mixing, in order to reduce free dye in the solution. [Significant fluorescence from free dye was measured when unwashed samples were analyzed and these spectra are shown in Fig. 2(b).] Cell suspensions were centrifuged at 3000 rpm for 1 min in a microcentrifuge, the supernatant was removed and PBS was added. After a 5-min wait the wash step was repeated.

Fig. 2

(a) Spectra of ethidium bromide (dashed-blue) and propidium iodide (solid-red) bound to cells. These spectra are each averages of nearly 1000 cells and have been scaled to have the same maximum intensity in order to optimize visualization of spectral differences. (b) Spectra of free ethidium bromide (dashed-blue) and propidium iodide (solid-red). The spectrum of sheath, which is primarily the Raman spectrum of water is in black. Each spectrum is the average of hundreds of spectra.

2.3.

Data Collection

The sheath was $1\times$ PBS with no calcium/magnesium [Hyclone Dulbeccos PBS (Powder)] and was run at 5-6psi. The samples were loaded into syringes and injected into the system at a rate of 5 to $15\mu \text{l}/\min$. Laser intensity at the sample was 7 mW at 488 nm. The camera was binned to four bins vertically and 400 bins horizontally, the majority of fluorescence was seen in one vertical bin, and the spectral information was collected along the horizontal axis. The camera was set to a readout rate of 2.5 MHz and the sample integration time was 60 μs. Flow check fluorospheres were collected every day that cells were run to monitor instrument performance. For microspheres and cells, spectral collection was triggered on the initial falling voltage of a side-scatter event using the fast external trigger option of the camera. Background spectra, i.e., the spectra of the solutions the cells were in, were collected by setting the camera on internal trigger and then gating out all of the events that showed evidence of a cell being present. The Raman signal was collected with only sheath in the flow cell, using an internal trigger with no sample being injected. Spectral information was converted to ASCII for offline analysis.

2.4.

Data Analysis

Data analysis was performed with Igor Pro 6.1. The forward-scatter and side-scatter time traces were fit to Gaussians from which peak height and width were extracted. The spectra were analyzed by two different methods and a noise-reduction technique was demonstrated, all of which are described below.

2.4.2.

Fitting to basis/component spectra

The methods of fitting multicomponent spectra to basis are best explained using cells as an example. When cells are multiply stained, the spectrum of each cell is a linear combination of the stains bound to the cells, any intrinsic signal from the cells, the solutions the cells are in, and the sheath. Measurement of the sheath reveals that our instrument measures the Raman spectrum of water and this is one of the basis spectra used. Free PrI and free EB fluorescence have a different emission spectrum than when they are bound to DNA in cells. Therefore, fluorescent spectra were obtained of each solution the cells were in.

Example basis (component) spectra are shown in Fig. 2. Averaged spectra of EB and PrI bound to cells are show in Fig. 2(a). Averaged spectra of solutions of free PrI and EB, as well as of the PBS sheath, are shown in Fig. 2(b). The Raman/Sheath contribution was subtracted out of the PrI and EB spectra and all spectra have had a constant subtracted.

Equation (1) shows the equation used to fit the individual cell spectra. The fit coefficients are $c_0$, $c_1$, $c_2$, $c_3$, and $c_4$. $E(\lambda )$ is the spectrum of EB bound to cells. $P(\lambda )$ is the spectrum of PrI bound to cells. $S(\lambda )$ is the spectrum of the solution the cells are in. $R(\lambda )$) is the spectrum of the sheath that is primarily a Raman band of water. ${\chi }^2$ was optimized using the Levenberg-Marquardt algorithm and a non-negativity constraint was applied to all fit coefficients. Examples of cell spectra, the fit, and the contributions of each basis spectra to the fit are shown in Fig. 3. One cell shows a large EB contribution, while the other shows a large PrI contribution. The cells had been washed and there was very little dye in the solution surrounding the cells, consequently the solution background spectrum is very weak. A constant (determined using the data from 400 to 480 nm) was subtracted from the plotted cell spectra.

Eq. (1)

$$C(\lambda )=c_0+c_{1}E(\lambda )+c_{2}P(\lambda )+c_{3}S(\lambda )+c_{4}R(\lambda ).$$

Fig. 3

Cell spectra, their fits and the contributions of the basis spectra to the fits. (a) A cell with more EB than PrI. (b) A cell with more PrI than EB.

3.1.

Polystyrene Beads

Results from measurement of flow check beads are shown in Figs. 4Fig. 5–6. A side-scatter signal results in a negative voltage, therefore the three example side scatter traces shown in Fig. 4(a) are negative. The corresponding forward-scatter traces are shown in Fig. 4(b). There is some baseline variation for the forward-scatter signal. However, this variation does not affect the data quality, as each of the forward- and side-scatter traces was fit to a Gaussian plus a constant from which the peak height and width were extracted. Representative spectra from a sample of flow check and align flow beads are shown in Fig. 5(a). Two spectra of individual flow check beads are shown and two spectra of individual align flow beads are shown. Comparison of these spectra in red and black shows that there is strong spectral overlap. A spectrum of two flow checks beads simultaneously passing through the excitation laser is also shown. All spectra shown are processed only by subtraction of a constant.

Fig. 4

Raw side scatter (a) and forward scatter (b) time traces from measurements of flow check fluorospheres. Traces from a single event have matching colors and line types.

Fig. 5

(a) The two black spectra are align flow beads while the two red spectra are flow check beads. The yellow (top) one is a doublet of flow check beads. (b) Averaged spectra of flow check and align flow beads and Raman/Sheath.

Fig. 6

(a) Forward scatter height versus side scatter height for a mix of flow check and align flow beads. Side scatter is a negative signal, so stronger side scatter is more negative. The flow check beads are the population in the upper left corner and the align flow beads are the population in the lower right corner. (b) Fit coefficient for the flow check beads versus the fit coefficient for the align flow beads.

To facilitate comparison of this instrument with a conventional instrument, the coefficient of variation (CV) of the fluorescence intensity from flow check beads from 564 to 606 nm was calculated and found to be 5% for the data set presented. For some data sets, the CV was smaller by more than a factor of 2. This calculation simulates the use of a common $585\pm 21\mathrm{nm}$ filter on a conventional flow cytometer.

When a mix of flow check and align flow beads were run, the beads were easily separable based on forward-scatter height and side-scatter height, as shown in Fig. 6(a). The beads were also easily separable based on their spectra. Each spectrum was fit to Eq. (2), where $F(\lambda )$ is the average spectrum of flow check beads run earlier, $A(\lambda )$ is the average spectrum of align flow beads run earlier, and $R(\lambda )\mathrm{}$ is the average spectrum of the sheath. All three spectra are shown in Fig. 5(b). The coefficients $c_0$, $c_1$, $c_2$, and $c_3$ are determined from each fit. The fit coefficient for the flow check beads, $c_1$, is plotted versus the fit coefficient for the align flow beads, $c_2$, in red in Fig. 6(b). To evaluate the accuracy of this separation, the data from cells with large forward scatter [i.e., the top left group in Fig. 6(a)] are overlaid in green and data from cells with small forward scatter overlaid in blue [i.e., the bottom right group in Fig. 6(a)]. Out of 643 data points, there are five beads that were not categorized the same way by scatter and spectral fluorescence analysis, two at the origin and three at the top right of the graph. Of the two points at the origin, one is the point on Fig. 6(a) with the smallest side scatter and could have been gated out. The three at the top right of the graph are likely doublets of a flow check and an align flow bead. All three points have high forward scatter and intense side scatter, in fact one is the point in Fig. 6(a) with the highest forward scatter. Because align flow beads are so much smaller than flow check beads, an align flow bead stuck to a flow check bead will not change the scatter significantly.

3.2.

Cells

Results of a set of experiments using cells stained with PrI and EB are shown on a density plot in Fig. 7. The $x$-axis is the fit coefficient for the EB basis spectra multiplied by the area under the peak of the EB basis spectra (601.3 to 612.9 nm). The $y$-axis is the fit coefficient for the PrI basis spectra multiplied by the area under the peak of the PrI basis spectra (612.9 to 624.4 nm). The color shade represents the log of the number of cells. Cells stained only with EB are shown in shades of blue and have strong EB fluoresence. Cells only stained with PrI are shown in shades of green and have strong PrI fluorescence. Cells singly stained with EB and cells singly stained with PrI were mixed. Measurements taken $\sim 3\min$ after mixing are shown in red. Measurements $\sim 45\min$ after mixing are shown in cyan. At only 3 min after mixing, it is clear that both dyes are dissociating from and then rebinding to the cells. For example, the initially PrI stained cells have lost some PrI stain and gained some EB stain. At 45 min all cells have a similar proportion of EB and PrI bound.

Fig. 7

Density plots from four samples with the density of cells represented by the darkness of the colors. Cells stained with only EB are blue. Cells stained with only PrI are green. These two cell populations were washed and mixed. Measurements made $\sim 3\min$ after mixing are red. Measurements made $\sim 45\min$ after mixing are light blue.

3.3.

Noise Reduction with Principle Component Analysis

The first four principle components of a PCA analysis of the measurements made 3 min after mixing, are shown in Fig. 8. The first three principle components clearly contain spectral information. The fourth principle component is noise, as are all of the higher principle components. The spectral data were remade using only the first three principle components. An example is shown in Fig. 9(a), which demonstrates the resultant noise reduction. To assess whether significant artifacts were introduced by the reconstruction, the reconstructed spectra were fit to basis spectra and the results compared to fits of the original spectra. The results in Fig. 9(b) show that the spectral fits do not change significantly.

Fig. 8

Results of a PCA analysis of the spectra of the sample measured 3 min after mixing cells individually stained with PrI and EB. (a) The first three principle components (b) The fourth principle component.

Fig. 9

(a) Comparison of a measured spectra with a PCA reconstructed spectra. (b) Results of fitting the measured data set compared with results of fitting the reconstructed data set.

3.4.

Potential for Automated Analysis Without Running Compensation Samples

The starting and final spectra for the ALS analysis are shown in Fig. 10(a) where they can be compared with the basis spectra used earlier for fitting the data. The starting spectra are Gaussians and clearly have a different shape than the basis spectra. The final spectra are quite similar to the basis spectra. The PrI spectra is slightly red-shifted compared to the basis spectra. The shape of the EB spectra deviates slightly from the EB basis spectra, especially where it overlaps with the Raman spectra. The relative fluorescence intensities obtained from the ALS analysis are compared to earlier analyses in Fig. 10(b). The results are quite similar, however, the ALS results show decreased EB and increased PrI. The ALS analysis uses fewer basis (component) spectra to describe the measured spectra, because the spectrum of the solution the cells were in was not included in the ALS analysis. As Fig. 3 illustrates, the contribution of this spectrum was quite small. Nonetheless, to assure that not using the solution spectra was not the cause of the differences in Fig. 10(b), the spectra were fit without using the solution spectra. The lack of the solution spectra in the ALS analysis was not the cause of the differences in Fig. 10(b) (data not shown).

Fig. 10

(a) The Gaussian starting spectra for the ALS analysis are shown in orange and green. After 12,000 ALS iterations, the ALS spectra are in solid red and blue. The measured basis spectra are shown in dotted red and blue lines. The measured sheath spectrum is black. (b) Comparison of fluorescence contributions from EB and PrI to individual spectra obtained from fitting the data to component spectra and from ALS. The black dots and red circles are the same data shown in Fig. 9(b), however, the scales have been changed.