2.1.

In Vitro Modeling Fluid System

The in vitro modeling fluid system consisted of a square quartz capillary (Polymicro Technologies Inc.) and a fluid pump system. The square capillary (ID/OD: $100∕300\mathrm{\mu }\mathrm{m}$ ) was prepared by burning away a $2\text{-}\mathrm{cm}$ patch of the polyimide coating in the center of a $30\text{-}\mathrm{cm}$ -long piece. The square capillary was held perpendicular to the excitation source with its surface adjusted flat to the source. The capillary was connected to a homemade oscillatory pump and a syringe pump (Fisher Scientific International, Inc., Hampton, New Hampshire) with plastic tubing. Cell samples were loaded into a syringe and pushed by the syringe pump through the capillary at constant flow rate (typically $5\mathrm{\mu }\mathrm{L}∕\min$ ). Pulsatile flow can also be superposed with a homemade oscillatory pump, which was described in detail previously.²¹ The frequency and oscillation amplitude of the pulsatile flow pattern was obtained by analyzing sequencial images of a fluorescent microsphere flowing in the capillary. To model in vivo blood flow in the blood vessel of a mouse, the oscillation frequency was set to $11\mathrm{Hz}$ (typical mouse heart rate) and the oscillation amplitude was set to half of the average flow rate.

2.2.

Optical Setup

The excitation source was a $\mathrm{Ti}:\mathrm{sapphire}$ laser (Mira-900 Coherent Inc., Santa Clara, California), which generates $50\text{-}\mathrm{fs}$ pulses at $800\mathrm{nm}$ with a $76\text{-}\mathrm{MHz}$ repetition rate. For in vitro measurements, the in vitro modeling fluid system was used. The femtosecond laser pulses were focused by a long working distance Olympus $40\times 0.55$ numerical aperture (NA)-microscope objective (Olympus America Inc., Melville, New York) into the square capillary tube (see the preceding fluid system). A telescope system was used to expand the laser beam to overfill the back lens of the objective, which produced a $2\text{-}\mathrm{\mu }\mathrm{m}$ spot in the focal plane. The point spread function (PSF) spreading has attracted much interest in multiphoton microscopy literature.^{29, 30} The PSF of the two-photon excitation system was measured by scanning a fluorescent microsphere ( $100\mathrm{nm}$ in diameter) at the focal plane of the objective and matched with theoretical calculation. The average excitation power used for cell measurements was between 10 and $20\mathrm{mW}$ at the focus. Group velocity dispersion from glass and mirrors was precompensated by a prism pair in the laser path. The femtosecond pulse width was adjusted to maximize the SHG (second-harmonic generated) signal from a BBO crystal, which was placed at the focal plane of the objective. The pulse width was further adjusted to maximize the fluorescence signal from dye solutions flowing in the capillary placed at the focal plane of the objective lens.

For single-channel or two-channel ratiometric measurement, the fluorescence from the excitation volume was collected back through the same microscope objective and separated from the excitation beam by a dichroic mirror. While we did observe very small, comparable to background scattering, SHG signals, the filter set in the system was chosen to prevent SHG and third-harmonic generated (THG) signals entering the detection channels. A long-pass filter in the excitation laser beam path cuts pump laser light leaking into the detection path, and a short-pass filter in the collection path further ensures collection of two-photon fluorescence only. Beamsplitters and bandpass filters were used to separate the fluorescence from different dyes into two channels, which were detected with two photomultiplier tubes (PMT HC7421-40, Hamamatsu Photonics, K.K., Hamamatsu City, Japan) and recorded with two multichannel photon-counting scalers (MCS SR430, Stanford Research Systems, Inc., Sunnyvale, California). The temporal bin width of the MCSs was set to correspond to the transit time of a cell through the excitation volume (typically $1.31\mathrm{ms}$ for a flow rate of $5\mathrm{\mu }\mathrm{L}∕\min$ ).

For two-beam scanning cell size measurement, the femtosecond laser pulses were split into two identical beams focused by the microscope objective into the capillary, with two distinct localized excitation region lined up along the fluid flow direction. The fluorescence signals from the two excitation regions were collected back through the same objective and sent to two identical channels, each detected with a PMT and recorded with a MCS. The temporal bin width of the MCSs was set to a much shorter time (typically $92\mathrm{\mu }\mathrm{s})$ , so that the fluorescence signal from a single cell goes into several consecutive bins, which form a peak.

2.3.

Setup Modification for In Vivo Measurements

For in vivo measurements, the in vitro modeling fluid system was replaced by a custom-made stage to hold anesthetized mice. An $8\text{-}\mathrm{mm}\text{-}\mathrm{diam}$ hole was drilled in the center of the stage, on top of which a glass slide was placed. The position of the mouse was adjusted to flatten one of its ears against the glass slide. The vasculature and blood flow of the mouse ear was visualized with a CCD camera by the Olympus $40\times$ objective from transmitted light of a fiber optic illuminator (EW-41500-50, Cole-Parmer Instrument Company, Vernon Hills, Illinois). The femtosecond NIR laser beam was focused into the mouse ear with the objective from below. By switching off the short-pass filter in the fluorescence collection optical path, the back-scattered light from the femtosecond NIR laser beam was used to align the excitation beam to the blood vessel. The long focal length of the objective $\left(5\mathrm{mm}\right)$ enabled the depth of the laser focus to be adjusted throughout the whole cross section of the mouse ear (less than $1\mathrm{mm}$ ). To obtain a higher detection rate (probe a larger volume of blood) and reduce the autofluorescence background, the femtosecond NIR laser beam was adjusted to focus at the center of the blood vessel. The laser beam created an excitation region at the location where the blood flow speed is highest, while not generating autofluorescence from surrounding tissues. The aberration introduced by the highly scattering skin tissue made the size of the laser focus larger than $2\mathrm{\mu }\mathrm{m}$ . The laser power after the objective was adjusted below $20\mathrm{mW}$ to avoid photodamage to either cells or tissue. The tissue at the focus of the femtosecond laser beam was periodically monitored with the same objective lens transilluminated by visible light. No obvious physical damage of the tissue was observed for the laser power we used.

4.1.

Computer Simulation

A Monte Carlo simulation program of the two-photon excited flow cytomety system was written in MATLAB, which incorporated an accurate model of the two-photon excitation volume generated by a femtosecond NIR laser within a microcapillary. This program enables the user to set a wide variety of experimental parameters, including the laser beam characteristics, the flow rate and spatial distribution of simulated fluorescent particle within the capillary, the size and fluorescent dye distribution within particles, and white-noise levels.

To reveal the difference between one-photon and two-photon excitation, the axial direction of the laser beams ( $z$ axis) was set perpendicular to the fluid flow direction ( $x$ axis) and the coordinate was centered at the laser focal point. The program calculated the one-photon and two-photon excited fluorescence signal for grid points within the excitation volume $(100\times 100\times 100\mathrm{\mu }\mathrm{m})$ at resolution of $20\mathrm{nm}$ and saved the results as a lookup table. By integrating the grid points along the $x$ axis, the fluorescence intensity distributions in the flow cross section of the fluid channel ( $y\text{-}z$ plan) were obtained.

To investigate the wide distribution induced by the inhomogeneous spatial distribution of dyes, two models representing different fluorescent dye distributions were used: (1) a uniformly distributed fluorescent sphere ( $10\mathrm{\mu }\mathrm{m}$ in diameter) and (2) a fluorescent shell ( $10\mathrm{\mu }\mathrm{m}$ in diameter). The fluorescent sphere was used as a model for fluorescent dye that evenly distributed in a cell, such as CFSE; while the fluorescent shell was used as the model for dye that only distributed with the cytoplasm of a cell, such as DeepRed. Using the two-photon excited fluorescence look-up table already described, the fluorescence signal from a $10\text{-}\mathrm{\mu }\mathrm{m}$ fluorescent sphere or shell was calculated for grid points within the excitation volume at a $100\text{-}\mathrm{nm}$ resolution. Given the low Reynolds number (less than 10) of our flow cytometry system, it was assumed that the fluid flow was laminar and the fluorescent spheres or shells followed the fluid flow ( $x$ axis). The fluorescence intensity distribution of the fluorescent sphere or shell in the cross section of the fluid channel was determined by integrating the fluorescence along the fluid flow direction. By correlating the distribution of the $10\text{-}\mathrm{\mu }\mathrm{m}$ sphere and shell models, a two-channel dot plot was generated after normalization of the signal to unity in both channels.

4.2.

Fluorescent Microspheres

Fluorescence microspheres with different size and spectral properties were obtained from Molecular Probes, Inc. (Eugene, Oregon). To calibrate the ${\mathrm{T}}^3\mathrm{FC}$ system for cell size measurement, three types of fluorescence microspheres were used: A7302, AlignFlow flow cytometry alignment beads, $2.5\mathrm{\mu }\mathrm{m}$ ; A7303, AlignFlow Plus flow cytometry alignment beads, $6\mathrm{\mu }\mathrm{m}$ ; and F8836, FluoSpheres polystyrene microspheres, $10\mathrm{\mu }\mathrm{m}$ . All three types of fluorescence microspheres have the same emission maximum at $515\mathrm{nm}$ . The diameter of each type of microspheres is uniform with a cell volume less than 4%. For sample preparation, all types of fluorescent microspheres were diluted in distilled water to a concentration of $3\times {10}^5∕\mathrm{mL}$ . For in vivo flow cytomety measurements, F-8827 ${\mathrm{FluoSpheres}}^{\circledR }$ carboxylate-modified microspheres, $2.0\mathrm{\mu }\mathrm{m}$ , yellow-green fluorescent (Ex505/Em515) was used.

4.3.

Cell Culture

4.4.

Confocal Imaging of Cells

Confocal images of PMBC stained with both CFSE and DeepRed, KB cells stained with both G5-FI-FA and G5-DR-FA were acquired using an Olympus Confocal Microscope (Olympus America Inc., Melville, New York). The cells were rinsed, fixed with paraformaldehyde, and mounted on a slide and confocal images were taken in an Olympus Confocal Microscope. An argon laser was used to excite CFSE or FITC and the fluorescent signal was recorded in the green channel; while a $\mathrm{He}∕\mathrm{Ne}$ laser was used to excite DeepRed and the signal was recorded in the red channel. The green and red channel images of the PMBC stained with CFSE and DeepRed were combined to form a single image [Fig. 5a in Sec. 5]; while KB cells stained with both G5-FI-FA and G5-DR-FA images were separately shown in green [Fig. 5b in Sec. 5] and red channel [Fig. 5c in Sec. 5].

Fig. 5

Confocal microscopy images of (a) PMBC stained with CFSE (green) and DeepRed (red), (b) green channel of KB cells stained with G5-FI-FA and G5-DR-FA, and (c) Red channel of KB cells stained with G5-FI-FA and G5-DR-FA. The details are given in Sec. 4.

4.5.

Conventional Flow Cytometry Analysis

4.6.

Animals

Five- to six-week-old, specific-pathogen-free female NU/NU CD-1 and CD-1 mice were purchased from Charles River Laboratories (Portage, Michigan) and housed in a specific pathogen-free animal facility at the University of Michigan Medical Center in accordance with the regulations of the University’s Committee on the Use and Care of Animals as well as with federal guidelines, including the principles of Laboratory Animal Care. CD-1 mice were used for the fluorescent microsphere-circulation in vivo experiments. Immmunodeficient, NU/NU CD-1 mice were used for DeepRed-labeled cell circulation and free DeepRed dye injection in vivo experiments. All injections were performed via the tail vein.

4.7.

In Vivo Two-Photon Flow Cytometry Measurements

To conduct in vivo two-photon flow cytometry (IVTPFC) measurements, mice were anesthetized by inhalation of isoflurane. The mice were placed on the ${\mathrm{T}}^3\mathrm{FC}$ stage with one ear taped to the glass slide window for optical detection. A blood vessel roughly $50\mathrm{\mu }\mathrm{m}$ in diameter with ample blood flow was selected for cytometry measurement. Depending on the anesthesia condition, the mice may breathe heavily during the data acquisition, which induces movement of its ear. This tremble can be easily seen with the transilluminated microscopy image. The alignment of the laser focus and the blood vessel within the mouse ear was checked before and after each data acquisition to be within a few micrometers. Before injection of the solution of fluorescent-dye-labeled microspheres or cells, the background signals in both the S channel and the L channel were recorded as a control. The laser power used was below $20\mathrm{mW}$ and no photodamage to the ear of the mice was observed after the experiments.

To monitor circulating fluorescent microspheres in vivo, $200\mathrm{\mu }\mathrm{L}$ sodium chloride solution containing $2.3\times 109$ yellow-green fluorescent (Ex505/Em515) $2.0\text{-}\mathrm{\mu }\mathrm{m}$ microspheres was injected into CD-1 mice through tail vein. Immediately following the injection, the two-channel fluorescence signals from a blood vessel were recorded. The same amount of microspheres was injected through the tail vein $11\min$ after the initial injection.

To monitor circulating DeepRed-labeled fluorescent cells in vivo, $300\mathrm{\mu }\mathrm{L}$ PBS containing $3.6\times {10}^6$ DeepRed-labeled splenocytes was injected into NU/NU CD-1 mouse initially with additional injection of the same dose $25\min$ after. The two-channel fluorescence signal was recorded immediately after each injection and repeated at the same vessel location roughly $2\mathrm{h}$ and 1 day after the initial injection.

To monitor the fluorescence from free dye in vivo, $300\mathrm{\mu }\mathrm{L}$ DeepRed dye solution $\left(5\mathrm{\mu }\mathrm{M}\right)$ was injected into NU/NU CD-1 mouse through tail vein. The fluorescence was acquired in two channels immediately following the injection and repeated at the same vessel location roughly one and half hours after.

5.1.

Single-Cell Detection Using a Femtosecond Laser

Using the flow cytometry simulation program (details in Sec. 4), we ran two simulations of the fluorescence excitation distribution in the cross section of a fluid channel for one-photon [Fig. 2a ] and two-photon excitation [Fig. 2b]. For comparison, 400- and $800\text{-}\mathrm{nm}$ laser beams with a $2\text{-}\mathrm{\mu }\mathrm{m}$ waist at the focal plane were used. As shown in Fig. 2a, in the case of one-photon excitation, the laser beam excites all the fluorescent dyes along its optical path, creating a broad excitation region within a fluid stream. Under nonuniform flow conditions, such as in vivo capillary blood flow, multiple cells may enter the excitation region simultaneously. Thus, single-cell detection may not be achieved, unless a confocal pinhole is used to restrict the detection region to the focus.³⁶ In the case of two-photon excitation, however, the probe volume is confined to the laser beam focus, where the photon flux is highest. One can simply focus a femtosecond NIR laser beam into a fluid stream to create an excitation volume (a 0.5 NA objective creates an excitation region about $1\mathrm{\mu }\mathrm{m}$ laterally and $5\mathrm{\mu }\mathrm{m}$ longitudinally) equal to or smaller than a cell, so that only one cell enters the localized excitation volume to be selectively detected.

Fig. 2

Fluorescence intensity distribution of (a) one-photon and (b) two-photon excitation in the cross section of a flow channel. For comparison, both simulation results are based on a $800\text{-}\mathrm{nm}$ laser beam with a $2\text{-}\mathrm{\mu }\mathrm{m}$ waist. The fluid flow direction is along the $X$ axis, the laser beam propagates along the $Z$ axis and centers at zero of the coordinates.

For proof of principle, Jurkat cells labeled with CFSE and unstained Jurkat cells as a control were tested in our ${\mathrm{T}}^3\mathrm{FC}$ system with a single detection channel. A short segment of the raw data is presented in Fig. 3a . The horizontal axis denotes time and the vertical axis denotes fluorescence signal. Sharp peaks in the raw data correspond to bursts of photons emitted by single fluorescence-stained cells as they transit through the excitation beam. With identical excitation power, almost no fluorescence peak was observed above the background for the unstained Jurkat cells, while thousands of events were detected from CFSE-labeled Jurkat cells within a few minutes. Although the cells were uniformly stained (confirmed by flow cytometry data not shown here), the intensity distribution of the fluorescent signal from each cell spreads out from just above the background to the maximum possible signal level [Fig. 3b). This is due to the fact that each cell enters the excitation region at a random position and speed. Depending on the amount of excitation received, the fluorescence signal from each cell peaks at a different intensity level. Due to the uneven excitation under two-photon excitation, the fluorescence intensity in a single channel itself cannot be used as a parameter to quantify cellular properties as in conventional flow cytometry. Nevertheless, our data clearly demonstrated single-cell detection in nonuniform flow condition can be achieved under two-photon excitation.

Fig. 3

Single-channel signal from ${\mathrm{T}}^3\mathrm{FC}$ : (a) raw data, where each peak corresponds to a labeled cell passing through the laser focus, and (b) histogram of the intensity distribution of peaks in a single channel.

5.2.

Quantitative Two-Channel Ratiometric Measurement

To understand the condition for signal quantification using trigger and reporter dyes, mouse PMBCs stained with CFSE and DeepRed (details in Sec. 4) were investigated using the ${\mathrm{T}}^3\mathrm{FC}$ system. The two-photon excited fluorescence signal has a wide distribution in the two-channel dot plot [Fig. 4a ]. The fluorescence was filtered such that the short-wavelength detection channel (S channel) corresponds only to CFSE, and the long-wavelength channel (L channel) only to DeepRed. The wide distribution indicates the S channel signal was not correlated to the L channel signal, although saturation of the dye uptake (confirmed by conventional flow cytometry result) should yield such correlation. To examine the lack of correlation between the two channel signals, the dual-dye-stained PMBC was studied under a confocal microscope. The image shows that CFSE and DeepRed had different distributions within each cell [Fig. 5a ]. The green nuclei and yellow cytoplasm of the cell indicate that CFSE was uniformly distributed through out the whole cell and DeepRed is located only in the outer cytoplasm surrounding the nucleus (DeepRed is a cell-permeating probe that binds to the outer membrane of mitochondria). To model the staining condition using the computer simulation program, $10\text{-}\mathrm{\mu }\mathrm{m}$ fluorescent spheres and shells were used to represent CFSE and DeepRed, respectively. The computer simulated cross sectional fluorescence distribution of the sphere [Fig. 4c] and the shell [Fig. 4d] do not match, which is due to the inhomogeneous localization of these two dyes within each bead. The unmatched distribution in the cross section of the fluid stream produces a wide distribution in a two-channel dot plot [Fig. 4b], which matches the result of the dual-dye-stained PMBC [Fig. 4a]. Thus, we can conclude that the different spatial localization of the two dyes within each cell leads to the lack of correlation between two channel signals under uneven excitation generated by the femtosecond NIR laser beam.

Fig. 4

Computer simulation result matches with experimental data: (a) two-channel dot plot of PMBCs stained with CFSE and DeepRed, where the two-photon excited fluorescent signals from CFSE and DeepRed go to S channel (green) and L channel (red), respectively; (b) computer simulated two-channel dot plot of dual-dye-stained microspheres; (c) fluorescence intensity distribution of a green-dye-stained $10\text{-}\mathrm{\mu }\mathrm{m}$ sphere in the cross section of the flow channel; and (d) Fluorescence intensity distribution of red-dye-stained $10\text{-}\mathrm{\mu }\mathrm{m}$ shell model in the cross section of the flow channel. The fluid flow direction is along the $X$ axis, and the laser beam propagates along the $Z$ axis and centers at zero of the coordinates. The $Y\text{-}Z$ plane represents the cross section of the flow channel. The two-photon fluorescence signal from a sphere or shell is calculated as it centers at each point in the $Y\text{-}Z$ plane. The dye in the sphere model is evenly distributed in the whole volume of a $10\text{-}\mathrm{\mu }\mathrm{m}$ sphere, while the dye in the shell model is evenly distributed on the surface of the same $10\text{-}\mathrm{\mu }\mathrm{m}$ sphere. The computer simulated dual-dye-stained $10\text{-}\mathrm{\mu }\mathrm{m}$ microspheres are the overlap of the sphere and shell model. The fluorescent signals from the sphere and shell go to the S and L channels, respectively. (Color online only).

To make quantitative ratiometric measurements, it is essential that the fluorescence amplitudes of signal and reporter be correlated (i.e., spatially correlated fluorescent labels must be used to enable a narrow angular distribution). One solution to establish the required correlation between the trigger and reporter dyes is to use targeted fluorochromes. For example, two dyes could be conjugated onto a nanoparticle, which is taken up into the cells via specific receptors. Alternatively, the dye molecules may be linked separately onto nanoparticles that have the same targeting agent. By competing for the same receptors, the dye concentrations in the cell become correlated. Specifically, the KB cell, which is an epidermoid carcinoma cell line known to express the receptor for folic acid (FA), was selected as a model for proof of concept. The nanoparticles used to carry the FA-targeting agent and fluorescent molecules are PAMAM dendrimers. Our previous studies have shown that KB cells bind and internalize FA-conjugated dendrimers in a receptor-mediated fashion.³¹ Regardless of the dye molecules conjugated on the dendrimers, the FA on each of the dendrimers guides the PAMAM dendrimers to the FA receptor within the KB cells. This is confirmed with the identical structure in green [Fig. 5b] and red [Fig. 5c] channel confocal microscopy images. KB cells treated with the G5-FI-FA and G5-DR-FA at different concentrations (Table 1) were analyzed by the ${\mathrm{T}}^3\mathrm{FC}$ . The FITC (FI) and DeepRed (DR) dyes correspond to the S and L channels, respectively. On a two-parameter dot plot (S channel versus L channel); the KB cells fall into well-separated clusters of narrow angular distribution [Fig. 6a to 6h] according to the variation on the concentrations of G5-FI-FA and G5-DR-FA. This enables one to differentiate a single-cell event in a mixed sample. With a mixture of KB cell samples under different incubation conditions through our capillary system, signals from different samples remain in the corresponding angular distributions [Figs. 6i and 6j]. Due to the different fluorescence quantum yield and two-photon excitation cross section of FITC and DeepRed, the number of photons detected in the L channel (sample 8) is roughly 10 times of the number detected in the S channel (sample 1). Taking the intensity difference into account, the average angular distributions of samples were normalized accordingly. As shown in Fig. 6k, the average angular distribution increases with decreasing concentration ratio of G5-FI-FA to G5-DR-FA. The ideal angular distributions of different samples, assuming the uptake load of the nanoparticle is proportional to its incubation concentration, shows a linear fit with the experimental results [Fig. 6l]. Thus, the uptake of the dendrimer-dye-FA conjugates is proportional to the concentration of the dye-conjugated nanoparticles during cell incubation. This result is confirmed by conventional flow cytometry (Fig. 7 ). Conventional flow cytometry results [Fig. 7a] showed that the detected fluorescent signals (mainlyautofluorescence) for unstained KB cells (sample 9) and KB cells stained with nontargeted dendrimer-dye conjugates (samples 10 and 11) in both channels were much lower than those for KB cells stained with dendrimer-dye-FA conjugates. This was also observed by the ${\mathrm{T}}^3\mathrm{FC}$ , where the number of cells detected was three orders of magnitude lower with the same background threshold.

Fig. 6

Two-channel dot plot and angular distribution of KB cell samples stained with G5-FI-FA and G5-DR-FA at different concentrations. (a) and (b) Two-channel dot plot and angular distribution histogram from sample 1; (c) and (d) from sample 2; (e) and (f) from sample 3; (g) and (h) from sample 8; (i) and (j) from the mixture of sample 1 and sample 8 at a volume ratio of 1:1; (k) the average angular distribution of cell samples from 1 to 8; and (l) the average angle of KB cells stained with G5-FI-FA and G5-DR-FA at certain concentration ratio fits linearly to the ideal average angle of free G5-FI-FA and G5-DR-FA solutions used to incubate the KB cells. The concentration of dye conjugates used to incubate the KB cell samples are listed in Table 1.

Fig. 7

Conventional flow cytometry of KB cells treated with different concentrations of G5-FI-FA and G5-DR-FA (the respective concentrations used are indicated), as described in Sec. 4: (a) unstained control in PBS; (b) to (i) correspond to samples 1 to 8 shown in Table 1, with the $X$ and $Y$ axes respectively representing the FITC and DeepRed fluorescence on a log scale. In the FITC channel, the mean channel fluorescence from sample 1 is set as 100%; in the DeepRed channel, sample 8 is set as 100%.

The two-channel ratiometric method can be used to quantitatively measure the uptake of fluorescence-labeled-nanoparticles under nonuniform flow condition. If the fluorescence signal from one of the dye-conjugated nanoparticles is associated with cellular function, such as apoptosis, the ratiometric method can be used to quantitatively measure the cellular function. Furthermore, the ratiometric method is not limited to the dual dye staining scheme; the ${\mathrm{T}}^3\mathrm{FC}$ can detect spectral shift or change from a single cell. One could envision further developing the system to perform fluorescence resonance energy transfer (FRET) measurements.

5.3.

Two-Beam Scanning Cell Size Measurement

Light scattering is the standard method for determining cell size in flow cytometry. However, under nonuniform flow, this does not work (especially in vivo). Here, we conduct cell size measurement based on the fluorescence peak width normalized to flow velocity. By setting the temporal bin width of the MCSs to a much shorter time, a time-resolved fluorescence peak from 2.5-, 6-, and $10\text{-}\mathrm{\mu }\mathrm{m}$ fluorescent microspheres, CFSE-stained KB cells, and CFSE-stained Jurkat cells were obtained using the ${\mathrm{T}}^3\mathrm{FC}$ . The fluorescence peak width was normalized to the flow velocity, inversely proportional to the transition time between two foci. The average peak width of each sample is shown in Fig. 8a . As the average peak width is a convolution measure of the particle and the laser beam, it is not directly proportional to the size of the particle. A deconvolution function, which converts average peak width to average particle size, was determined using the fluorescent microspheres as calibration. The deconvoluted size of the samples is shown in Fig. 8b. The size information obtained this way is a measure of the distribution volume of the fluorescent dye within a cell or particle, rather than the actual physical size of the cell or particle. With partial distribution of CFSE within cells, the two-beam scanning results may not reflect the actual size of KB cells or Jurkat cells. Further, cell clipping, where only a portion of the cell passes through the laser focus, is another potential source of artifacts not overcome by the normalization.

Fig. 8

Size measurements made using ${\mathrm{T}}^3\mathrm{FC}$ : (a) average peak width (normalized duration time of cell passing through one excitation volume) of different samples, where the 2.5-, 6-, and $10\text{-}\mathrm{\mu }\mathrm{m}$ bars are the nominal diameter of fluorescent microspheres; KB stands for CFSE-stained KB cells and Jurkat stands for CFSE-stained Jurkat cells; and (b) actual particle size of the samples after deconvolution.

In addition to the cell size measurement, the ${\mathrm{T}}^3\mathrm{FC}$ is also capable of cell aggregation detection. Time-resolved fluorescent peaks from CFSE-stained single Jurkat cell suspension and Jurkat cell aggregates were obtained in the ${\mathrm{T}}^3\mathrm{FC}$ . Typical fluorescence peak profiles from the single-channel raw data are shown in Fig. 9 . The peak width (number of bins above the background) of an aggregated Jurkat cell (103 bins) is much wider than that of a single cell (18 bins). The pulsatile flow generation device was used to further demonstrate the two-beam method in highly nonuniform flow. Suspensions of $6\text{-}\mathrm{\mu }\mathrm{m}$ fluorescent microspheres were analyzed using single-beam and two-beam scanning. Under the two-beam scanning scheme, the normalized peak width distribution of the florescent microspheres [Fig. 10b ] is much narrower compared with the distribution using a single beam [Fig. 10a]. Considering the small standard deviation of the fluorescent microspheres (less than 4%), the normalized peak width should have a narrower distribution than what is shown in Fig. 10b. The broadened peak width distribution is due to the random of each fluorescent microsphere in the excitation region ( $2\mathrm{\mu }\mathrm{m}$ lateral); resulting different portion of each microsphere is scanned by the laser beam. We have shown previously that identical fluorescent microspheres produce different peak widths if they enter the excitation region at identical speed but different position relative to the laser beam.²¹

Fig. 9

Representative single-cell and aggregated Jurkat cell fluorescent profiles. The peak from a single Jurkat cell (dashed line) is much smaller than that of a cell aggregate (solid line). Each bin corresponds to $92\mathrm{\mu }\mathrm{s}$ .

Fig. 10

Peak width distribution of $6\text{-}\mathrm{\mu }\mathrm{m}$ fluorescent microspheres (a) using the single beam setup and (b) using two-beam normalization.

Although the two-beam scanning method does not offer absolute size measurement of single particles or cells, it grants sufficient accuracy to detect cell aggregation. To demonstrate the ability to detect cell aggregation in nonuniform flow, suspensions of 6 and $10\mathrm{\mu }\mathrm{m}$ fluorescence microspheres were flown in the ${\mathrm{T}}^3\mathrm{FC}$ . With two-beam scanning, the normalized peak width distribution of 6- and $10\text{-}\mathrm{\mu }\mathrm{m}$ fluorescence microspheres is shown in Figs. 11a and 11b , respectively. Setting the differential threshold as 21 bins (dashed line), 2.9 and 95% of 6- and $10\text{-}\mathrm{\mu }\mathrm{m}$ fluorescence microspheres were above this level, respectively. This clearly demonstrates that the two-beam scanning method is capable of identifying size differences under nonuniform flow conditions. For the aggregated (induced by ConA) Jurkat cells, 12.3% were above the threshold level [Fig. 11d], compared to 5.6% for the control cells [Fig. 11c]. As stated, the fluorescence peak width depends on the relative position of the particle in the excitation region. Similarly, the fluorescence peak width also relies on the orientation of aggregated cell cluster in the laser beam. Thus, the percentage difference between the aggregated and control sample may not reflect the actual difference in these samples. In spite of this, the results show that aggregation of Jurkat cells is reflected in the measurements under nonuniform flow condition.

Fig. 11

Aggregation measurements made using ${\mathrm{T}}^3\mathrm{FC}$ with the peak width distribution of (a) $6\text{-}\mathrm{\mu }\mathrm{m}$ fluorescent microspheres, (b) $10\text{-}\mathrm{\mu }\mathrm{m}$ fluorescent microspheres, (c) Jurkat cells stained with CFSE, and (d) aggregated (induced by ConA) Jurkat cells stained with CFSE. Thresholds for differentiating aggregates or large particles from individual small particles are marked as orange dashed lines. (Color online only.)

A limitation of applying the two-beam method to in vivo measurements is that it can not differentiate events from a single cell passing sequentially through the two foci from two different cells passing through each of the focus. This technique is more suitable for cell size characterization under controlled flow, in which the flow speed is maintained constant and cell trajectory is limited by a small capillary. For in vivo blood flow conditions, such as stoppage or backward flow, this method may not work. For in vivo application under slower velocities, the cell concentration must be kept low to ensure that a cell passes through two foci before the next one arrives.

5.4.

Monitoring Circulating Fluorescent Microspheres and Fluorescence-Labeled Cells In Vivo

To demonstrate the ability to perform multicolor detection of circulating fluorescent particles in vivo, fluorescent microspheres or fluorescence-labeled cells were injected through the tail vein of a mouse. Two-photon flow cytometry measurements were performed at a blood vessel in the mouse ear. Before the injection of the fluorescent microspheres or the DeepRed-labeled splenocytes, the background signals in both S channel and L channels were recorded with the femtosecond NIR laser beam focused in various locations within the mouse ear. The measured background control traces before each injection are shown to the left of the orange line in Figs. 12a and 12d . The thresholds were adjusted so that no peaks exist in the control traces. The amount of autofluorescence noise, which accounts for a large portion of the background, comes from the tissue at the focus of the laser beam. With translocation of the laser focus, a much higher autofluorescence signal is observed from blood vessel wall than from the blood itself. The sudden increase of background in Fig. 12d toward the end of the trace can thus be explained by a shift in the mouse ear. The background in the S channel of the mouse injected with DeepRed-labeled cells [Fig. 12e] was higher than that injected with fluorescence microspheres [Fig. 12b]. This can also be explained by the different site of laser focus in the mouse ear for the two experiments. In addition, the S channel background in Fig. 12d changed dramatically for the data taken $2\mathrm{h}$ and 1 day after the initial injection. This can be attributed to the tremble of the anesthetized mouse during data acquisition. In spite of this, the background fluorescence in the L channel remains at a low level throughout the whole experiments, allowing fluorescent peaks to be selected.

Fig. 12

(a) S channel (green) and L channel (red) traces from an ear blood vessel of a CD-1 mouse injected with $2.3\times {10}^9$ yellow-green fluorescent (Ex505/Em515) $2.0\text{-}\mathrm{\mu }\mathrm{m}$ microspheres at time zero and $11\min$ after. The control traces are shown to the left of the orange dashed line. The orange and blue dashed lines corresponds to the first and second injection, respectively. Each spike in the S channel corresponds to the fluorescent burst from microspheres passing through the excitation region. (b) Blow up of the dual-channel raw data showing detail of individual fluorescent peaks. (c) Number of detected events at different time points before and after injection. Each time point is represented by number of peaks above the background threshold during a period of $107\mathrm{s}$ . (d) S channel (green) and L channel (red) traces from an ear blood vessel of a NU/NU CD-1 mouse injected with $3.6\times {10}^6$ DeepRed-labeled splenocytes at time zero and $25\min$ after. The control traces are shown to the left of the orange dashed line. The orange, blue, black, and magenta dashed lines correspond to the first injection, the second injection, $2\mathrm{h}$ after initial injection, and 1 day after, respectively. Each spike in the L channel corresponds to the fluorescent burst from DeepRed-labeled splenocytes passing through the excitation region. (e) Magnified dual-channel raw data. (f) Number of detected events at different time points before and after injection. Each time point is represented by the number of peaks above the background threshold during a period of $214\mathrm{s}$ .

Figure 12a shows the S and L channel traces of a mouse injected with the $2.0\text{-}\mathrm{\mu }\mathrm{m}$ yellow-green fluorescent microspheres, in which a peak in the S channel corresponds to a microsphere passing through the excitation region [Fig. 12b]. Figure 12d shows the two-channel traces of a mouse injected with DeepRed-labeled splenocytes, in which a peak in the L channel corresponds to a cell [Fig. 12e]. Thus, circulating $2.0\text{-}\mathrm{\mu }\mathrm{m}$ yellow-green fluorescent microspheres and DeepRed-labeled splenocytes were detected in short and long spectral detection channels, respectively. The two-channel detection method enables differentiation of different cell or particles based on the fluorescence wavelength. It is also possible to conduct two-channel ratiometric measurements in vivo, as demonstrated previously in vitro. In this particular experiment, however, direct ratiometric measurements would not have been possible, since different individuals were used for DeepRed-labeled splenocytes or fluorescent microspheres, and thus the dissimilar spatial distributions of the the two fluorescent species is of no consequence.

The double-peak event in Fig. 12b is probably the fluorescent trace of clogged fluorescent microspheres. For the fluorescent microspheres, about 20% of the detected events contained double peaks. For DeepRed-labeled splenocytes, almost no large peaks were observed. A double-peak event is treated as a single event if the fluorescent signal between the two peaks does not fall below the background threshold value.

5.5.

The Dynamics of Circulating Fluorescent Microspheres and Fluorescence-Labeled Cells

The number of detected fluorescent microspheres and DeepRed-labeled splenocytes at different time points is shown in Figs. 12c and 12f, respectively. For both of experiments, fluorescent microspheres or DeepRed-labeled cells were detected in the ear blood vessel immediately after tail vein injection (less than $30\mathrm{s}$ delay), which was indicated by the sudden increases in frequency of events. The frequency dropped to less than 10% within the first 10 to $20\min$ . The frequency went up again with reinjection of the same amount of fluorescent microspheres or DeepRed-labeled splenocytes. Similar depletion dynamics were observed. The experiments were repeated at the same location in the mouse ear $2\mathrm{h}$ and 1 day after the initial injection. No fluorescent microspheres were observed (not show here), while a few circulating DeepRed-labeled splenocytes were detected.

The depletion dynamics of circulating fluorescent microspheres in CD-1 mice and DeepRed-labeled splenocytes in NU/NU CD-1 mice using the in vivo two-photon flow cytometer and conventional flow cytometer is shown in Fig. 13 . For the CD-1 mouse injected with fluorescent microspheres, the S channel traces after the first and second injection were used to calculate the frequency of detected events [Fig. 13a]. For the NU/NU CD-1 mouse injected with DeepRed-labeled splenocytes, the L channel trace after the first injection was used [Fig. 13c]. The frequency was calculated as the number of events within a certain time period. The time points beyond 10 or $20\min$ were not used to generate the fitting, as the number of detected cells fell below $1\mathrm{event}∕\min$ . The number of fluorescent microspheres [Fig. 13b] or CFSE-labeled splenocytes [Fig. 13d] in $10\mathrm{\mu }\mathrm{L}$ of blood drawn at different time points after injection was detected ex vivo using a conventional flow cytometer. Blood samples are not able to be drawn rapidly enough from the mouse for conventional flow cytometry to follow the depletion dynamics over the course of a few minutes as accurately as the in vivo flow cytometer enables.

Fig. 13

Depletion dynamics of circulating fluorescent microspheres or cells. (a) The $10\text{-}\min$ frequency dynamics in the blood vessel of a CD-1 mouse after the first (square marker) and the second (round marker) injection of $2.3\times {10}^9$ yellow-green fluorescent $2.0\text{-}\mathrm{\mu }\mathrm{m}$ microspheres. The frequency was calculated as the number of peaks in the S channel within $107\mathrm{s}$ (red), $71\mathrm{s}$ (green), or $53\mathrm{s}$ (blue). Both axes are in log scale. The coefficient of the linear fit (black dashed line) is the average of the coefficients from all six decay curves. (b) Frequency dynamics obtained by ex vivo analysis using a conventional flow cytometer. Blood was drawn at different time points from the mouse after injection of the same dose of microsphere solution. A linear fit is obtained (red dashed line) in the log-log plot. (c) The $20\text{-}\min$ frequency dynamics in the blood vessel of a NU/NU CD-1 mouse after the first injection of $3.6\times {10}^6$ DeepRed-labeled splenocytes. The frequency was calculated as the number of peaks in the L channel within 214 (red) or $107\mathrm{s}$ (blue). Both axes are on a log scale. The coefficient of the linear fit (black dashed line) is the average of the coefficients from both of the decay curves. (d) Frequency dynamics obtained by ex vivo analysis using a conventional flow cytometer. Blood was drawn at different time points from the mouse after injection of the same dose of DeepRed-labeled splenocytes. A linear fit is obtained (red dashed line) on the log-log plot. (Color online only).

The in vivo flow cytometry results show that the fluorescent microspheres deplete faster in the circulation of a CD-1 mouse than fluorescence-labeled splenocytes in NU/NU CD-1 mouse. The average slope coefficients are $-1.94$ and $-1.16$ for circulating fluorescent microspheres and DeepRed-labeled splenocytes, repectively. This difference in depletion rate was also observed from the ex vivo flow cytometry results, where the slope coefficients are $-1.46$ and $-1.06$ for circulating fluorescent microspheres and CFSE-labeled splenocytes, repectively. The reasons for the frequency drop could be that the microbeads or splenocytes clogged in the blood stream, stuck to the blood vessel surfaces, were removed by lysis or phagocytosis, or were filtered by organs in the circulation. The different depletion rates indicate the circulation time of splenocytes in an immmunodeficient NU/NU CD-1 mouse is longer than that of fluorescent microspheres in a CD-1 mouse.

For both circulating fluorescent microspheres and fluorescence-labeled splenocytes, the depletion rate in the first $10\min$ after injection obtained by in vivo flow cytometry is larger than the rate in the first $1.5\mathrm{h}$ obtained by ex vivo flow cytometry analysis. The difference of the depletion rates from in vivo and ex vivo measurements could be attributed to the different time scales in the depletion process. It could also be the systematic error in the ex vivo method introduced by the significant delay between blood withdrawal and flow cytometry analysis.

According to previous results,^{24, 32} one challenge to doing in vivo cytometry is the small number of labeled cells at the detection region relative to the number injected into the animal. To have a statistically significant result, a larger number of cells must be detected. One solution would be to access larger blood vessels with higher blood flow rates. A femtosecond NIR laser source has the advantage of being able to penetrate much deeper through biological tissue than a one-photon cw laser,³³ though it is in fact not always the case that larger vessels are deeper, a notable counterexample being the brain. In addition, the nonlinear excitation property of a femtosecond NIR laser confines the excitation region to appear only at the focus without using a confocal pinhole. This reduces photodamage to the surrounding biological tissue, which may extend the monitoring time in live animals.

5.6.

Monitoring Fluorescent Dye Staining Process In Vivo

The two-channel fluorescence traces from an ear blood vessel were monitored as free DeepRed dye solution was injected through the tail vein [Fig. 14a ]. As shown in [Fig. 14b], the background signal in the long channel rose immediately after the injection, due to the presence of the DeepRed fluorophores in the mouse blood. After the injection, the background signal gradually decreased, as the dye molecules permeated through the blood vessel walls or were filtered out of the circulation. The average background signal at different time points is ploted against time in a log scale [Fig. 14c], and a linear fit of the signal decay reveals a slope coefficient of $-0.56$ . Compared with the depletion rate of fluorescent microspheres or fluorescence-labeled splenocytes, the background decay rate is much slower.

Fig. 14

(a) S channel (green) and L channel (red) traces from a blood vessel in the ear of a NU/NU CD-1 mouse injected with $300\mathrm{\mu }\mathrm{L}\left(5\mathrm{\mu }\mathrm{M}\right)$ DeepRed solution at time zero. The control traces are shown to the left of the orange dashed line. The orange and black dashed lines correspond to the initial injection and $1.5\mathrm{h}$ after initial injection, respectively. Each spike in the L channel corresponds to the fluorescent burst from DeepRed fluorophores in cells passing through the excitation region. (b) Magnified dual-channel raw data at the time of injection showing the background rise in the L channel after injection of free DeepRed dye solution. Individual fluorescent peaks above the background noise can also be observed in the L channel after the injection of the free DeepRed dye solution. (c) The background two-photon excited fluorescent signal in the L channel at different time points after the injection of DeepRed dye solution. The fluorescent peaks above threshold were not counted in the background signal. The red dashed line is the linear fit of the log value of two-photon fluorescence against time. (d) Number of detected events at different time points before and after injection. Each time point is represented by the number of peaks above the background threshold during a period of 214 seconds. (e) The $20\text{-}\min$ peak frequency dynamics in the blood vessel of a NU/NU CD-1 mouse after the injection of $300\mathrm{\mu }\mathrm{L}$ of DeepRed solution. The frequency was calculated as the number of peaks in the L channel within 214 (red) or $107\mathrm{s}$ (blue) time duration. Both axes are on a log scale. The coefficient of the linear fit (black dashed line) is the average of the coefficients from both of the decay curves.

The peaks above the background were also observed in the L channel trace signal. DeepRed is a cell-permeating probe that binds to the outer membrane of mitochondria in living cells and measures their viability. As the DeepRed dye solution circulates in the vasculature, it labels the circulating lymphocytes. Therefore, the peaks should correspond to the in vivo labeled cells passing through the two-photon excitation region. The number of peaks above threshold at different time points was acquired [Fig. 14d]. The threshold was adjusted according to the gradually changing background fluorescence signal. The frequency of in vivo labeled cells decreases after the injection, with the exception of the $15\text{-}\min$ data point, which could be accounted for, for example, by an increase in blood flow with change in mouse temperature. An average depletion rate of $-0.37$ was obtained by fitting the number of detected events in the first $20\min$ after injection [Fig. 14e]. This rate is much slower than the those of fluorescent microspheres or fluorescence-labeled splenocytes shown previously. One and half hours after the initial injection of free DeepRed dye, only one event was observed in a $6\text{-}\min$ interval.