2.1.

Instrument Design

A diagram of the fiber-DFFC instrument is shown in Fig. 1(a). The specially designed optical bundle (EM-Vision LLC, Florida) consists of a central “excitation” and eight “collection” $300\text{-}\mu \mathrm{m}$ core diameter multimode optical fibers. The center-to-center separation for all fibers was $580\mu \mathrm{m}$. We noted previously that even small amounts of autofluorescence generated in optical fibers could obscure the weak signals from individual cells in vivo.⁷ Therefore, two micromachined filters were deposited directly on the fiber tip as shown: a 635-, 20-nm full-width at half-maximum (FWHM) bandpass for the excitation fiber, and an outer ring-shaped 650-nm long-pass filter for the collection fibers. The probe tip also had a 3-mm diameter ball lens to couple light from the sample into the fibers. The fiber probe was placed in firm contact with the surface of the sample, in this case either a phantom [Fig. 1(b)] or a mouse tail [Fig. 1(c)], so that measurements were taken in diffuse reflectance configuration.

Fig. 1

(a) Fiber-DFFC system schematic. The probe uses a central excitation ($x$) and eight collection fibers ($m$), half of which are coupled to the PMT. The probe is placed in contact with the (b) flow phantom or (c) tail of a nude mouse.

The excitation source was a broadband super-continuum (SC) fiber laser (Koheras SuperK Power, NKT Photonics, Birkerod, Denmark) fitted with a beam splitter unit (SpectraK Split, NKT Photonics) and a 640-nm bandpass filter (10-nm FWHM, Chroma Technologies, Bellows Falls, Vermont). We note that we used the SC only as a broadband light source and did not use the pulsed properties of the laser output. The laser was coupled into the excitation fiber using an SMA-coupled achromatic lens package (F230 SMA-B; Thorlabs Inc., Newton, New Jersey) and the power at the sample was 9 mW. The eight collection fibers were divided into two bundles of four fibers (alternating around the probe so that the detected signal was approximately symmetrical), which were also terminated on SMA-coupled lens packages (F260 SMA-B; Thorlabs) to collimate the output light. One output arm was passed through a 690-nm emission filter (50-nm FWHM, Chroma) and focused onto a photomultiplier tube (PMT; H6780-20, Hamamatsu Photonics, Bridgewater, New Jersey) with a 25-mm focal length lens (Edmund Optics, Barrington, New Jersey). The second output arm was not used for these experiments but could be used in the future, for example, for multispectral experiments. The output of the PMT was amplified with a current preamplifier (SR570, Stanford Research Systems, Sunnyvale, California) configured with 100-Hz low-pass filter, and acquired with a data acquisition card (DAQ; NI-USB-6251, National Instrument, Austin, Texas). The DAQ was controlled using a custom script written in MATLAB (The Mathworks Inc., Natick, Massachusetts).

2.2.

Validation in Optical Phantoms In Vitro

As an initial test of the fiber-DFFC system, we used flow cytometry microspheres (red laser cell sorting beads, Catalog No. C16507, Life Technologies Co., Carlsbad, California) suspended in phosphate buffered saline (PBS) at low concentrations of either 100, 500, or $1000\text{spheres}/\mathrm{mL}$ ($N=12$ samples for each). The microspheres mimic the size ($6\text{-}\mu \mathrm{m}$ diameter) and fluorescence intensity of cells that are brightly labeled with a near-infrared dye. The microsphere solution was pumped through a 3-D printed optical phantom printed from VeroWhitePlus plastic (similar to ABS) on an Object30 printer (Stratasys Ltd., Billerica, Massachusetts). The exact optical and autofluorescence properties of the printed material are not known, but we and others have observed that these approximate that of biological tissue at near-infrared wavelengths.¹² The phantom was 6 mm in diameter, 20 mm in length, with a 0.8-mm clear cylindrical channel centered 0.75 mm from the surface. We threaded a strand of $250\text{-}\mu \mathrm{m}$ internal diameter Tygon tubing (TGY-010-C, Small Parts, Inc., Seattle, Washington) through the channel and connected it to a syringe on a microsyringe pump (70-2209, Harvard Apparatus, Holliston, Massachusetts) configured to produce a flow speed of $5\mathrm{mm}/\mathrm{s}$ ($15\mu \mathrm{L}/\min$). Microspheres were, therefore, flowing between 625 and $875\mu \mathrm{m}$ from the surface, mimicking a superficial blood vessel. The fiber probe was placed in contact with the phantom, approximately above the tubing as shown in Fig. 1(b).

To test the fiber-DFFC quantitative accuracy in phantoms, we also used a high-sensitivity fluorescence imaging system similar to what we have described previously.⁹ Briefly, the camera was an electron multiplied charge coupled device (EMCCD; iXonEM 885 Andor Technology, Belfast, Northern Ireland) camera fitted with a $2\times$ magnifying objective, $1\times$ lens tube (Mitutoyo Objective Edmund Optics, Barrington, New Jersey) and a 710-nm bandpass filter (Chroma). In combination, this yielded a $\sim 5\times 5{\mathrm{mm}}^2$ image with a frame rate of $15\mathrm{frames}/\mathrm{s}$. The bare tube was illuminated with a 660-nm diode laser (DPSS-660; Crystalaser Inc., Reno, Nevada).

2.3.

Validation in Mice In Vivo

We performed preliminary tests of fiber-DFFC in nude mice (Charles River Labs) with fluorescently labeled MM cells (Northwestern University, Chicago, Illinois). All mice were handled in accordance with Northeastern University’s Division of Laboratory Animal Medicine policies on animal care. MM cells were harvested using a cell scraper, spun down at 400 g, and then resuspended in Roswell Park Memorial Institute medium with 0.1% bovine serum albumin at a concentration of $2\times {10}^6\text{cells}/\mathrm{mL}$. Cells were dyed using a final concentration of $1\mu \mathrm{mol}/\mathrm{L}$ of Vybrant-DiD (Thermofisher Scientific, Waltham, Massachusetts) and incubated for 20 min at 37°C. At the end of the incubation process, FBS was added (2% of total volume) to prevent cell clumping during centrifuging. Cells were centrifuged as before and washed once with PBS with FBS to remove any free DiD in suspension.

Nude mice were retro-orbitally (r.o.) injected with $50\mu \mathrm{L}$ of MM cells in suspension, at concentrations of either $2\times {10}^7$ labeled $\text{cells}/\mathrm{mL}$, or $2\times {10}^6$ labeled cells plus $2\times {10}^7$ unlabeled $\text{cells}/\mathrm{mL}$. In both cases, ${10}^6$ cells were injected, but in the second case only 1 in 10 cells was labeled. The rationale was to produce similar clearance kinetics in both cases, since this is known to be dependent on the concentration of the injected cells. Moreover, we note that in practice most injected cells are cleared rapidly from circulation or lost at the site of injection, so that only a small fraction is in quasistable circulation.¹³ Approximately 5 to 10 min after injection, mice were scanned with the fiber-DFFC probe, which was placed in firm contact with the ventral surface of the mouse tail, approximately over the tail artery 3 cm from the tip, where the tail diameter was 3 mm. Mice were held under inhaled isofluorane and were kept on a warming pad throughout the experiments.

2.4.

Data Processing

As we show, data generated by the instrument required only minimal processing. We first subtracted the background offset from the signal. In the in vivo case, it is generally not possible to perform a matched control (“blank”) measurement for subtraction, so we estimated the background directly from the measured signals as follows: we applied a 2.5-s median filter to the raw data, and then applied a moving average filter (“smooth” function in MATLAB) to smooth the result. We subtracted this estimated background from the raw data. We then applied a 0.1-s moving average filter to the background-subtracted signal. We note that this step was not necessary (particularly in phantoms) but decreased the standard deviation of the background and improved the overall SNR. We used MATLAB’s “findpeaks” function on the result to automatically identify fluorescent spikes from moving cells.

3.1.

Optical Phantoms In Vitro

Example data acquired from the phantom is shown in Fig. 2. When PBS only was pumped through the tubing [Fig. 2(a)], approximately $-4.5\mu \mathrm{A}$ of background signal was detected, indicating that the phantom had a substantial autofluorescence background similar to that observed in our in vivo experiments (note that the PMT produces a negative output, so that the current values presented here were inverted for visualization). When spheres were pumped through the phantom, “spikes” were observed on top of this background, as shown in Fig. 2(b).

Fig. 2

Example data traces collected from the surface of an optical phantom (a) without and (b) with fluorescent microspheres. We performed signal processing as described in the text (c) and (d) and identified peaks (red circles). We also (e) imaged bare tubing with an EMCCD camera to obtain independent measurements of microsphere numbers. Dotted lines represent the approximate location of the tubing. Frames shown were separated by 250 ms. (f) The number of spheres counted with the fiber-DFFC and EMCCD systems in 5-min scans is shown, indicating excellent agreement between the methods.

We processed these data as described above, and the result is shown in Figs. 2(c) and 2(d). The “spikes” observed were similar to those in our previous work,^7,8 and we interpret these to be the fluorescence signal generated as microspheres pass through the instrument field of view (FOV). Background subtraction and smoothing yielded an overall SNR improvement from $\sim 35$ to 40 dB. Here, SNR was defined as $20\log (I_{\text{peak}}/\sigma )$, where $I_{\text{peak}}$ was the peak amplitude and $\sigma$ is the background standard deviation. We also calculated the temporal width of the spikes, which on average had FWHM of 107 ms. Given that the flow speed was $5\mathrm{mm}/\mathrm{s}$, we estimate that the fiber-DFFC instrument FOV was 0.5 mm.

To test the quantitative accuracy of the system, we used the “findpeaks” function in MATLAB with a 30-nA threshold and determined the number of microspheres in 5-min traces for each sample [indicated with red circles in Fig. 2(d)]. We also used a fluorescence imager to directly visualize and manually count the microspheres in the Tygon tubing during the same intervals. An example image sequence is shown in Fig. 2(e), where each image is separated by 250 ms. The resulting microsphere count data obtained with the two methods are shown in Fig. 2(f), where each data point represents a separate 5-min scan. Based on the estimated concentrations of the microspheres and the flow rate, we expect 75, 37.5, and 7.5 spheres in 5 min for the three dilutions, respectively. Although we note that it is difficult to dilute spheres accurately at such low concentrations, these are in good general agreement with the average measured counts ($94\pm 19$, $19\pm 14$, and $5.5\pm 3.6$, respectively). Overall, these data demonstrate the count accuracy of the fiber-DFFC in flow phantoms.

3.2.

Testing in Mice In Vivo

An example 5-min processed data trace for an uninjected (control) mouse is shown in Fig. 3(a). The background autofluorescence signal varied between approximately $-5$ and $-10\mu \mathrm{A}$ over the mice tested here, but after processing was very flat with a standard deviation of approximately only 5 nA. As noted above, motion artifacts due to the breathing motion of mice was a major problem in our previous work,⁸ but the fiber-DFFC design allowed us to completely remove this issue, simply by securing the sample (tail) to the platform and placing the fiber probe in firm contact with the surface [Fig. 1(c)]. This is a major advantage of the fiber DFFC probe, since it eliminated the need to correct for this artifact using a second background measurement and a weighted subtraction.⁸

Fig. 3

Example 5-min data traces measured from (a) uninjected control mice, and mice injected with (b) ${10}^5$ or (c) ${10}^6$ fluorescently labeled MM cells. Red circles indicate detected peaks. An example spike is shown inset in (b). The distribution of peak (d) widths and (e) amplitudes is shown. The (f) normalized cell count as a function of time after injection is shown.

Data traces for mice injected with either ${10}^5$ or ${10}^6$ labeled MM cells are shown in Figs. 3(b) and 3(c). “Spikes” were again evident, which we interpret to be fluorescently labeled MM cells passing through the fiber-DFFC FOV as they moved along the artery. The distribution of the widths of the peaks is shown in the histogram in Fig. 3(d). Most were between 150 and 200 ms; given the 0.5-mm FOV, this implies an average linear flow speed of cells in the tail of 2.5 to $3.3\mathrm{mm}/\mathrm{s}$, although individual cells moving at speeds between 1 and $10\mathrm{mm}/\mathrm{s}$ were observed. A literature search failed to provide a specific value for the blood flow rate in mice tail blood vessels, but the range of flow rates here is generally consistent with previously reported blood flow speeds in mice and our own work.^7,14 Based on the background noise levels observed in the control animals, we set a peak detection threshold of 30 nA; this yielded a count rate of 24.5, 1.9, and $0\text{cells}/\min$ in 15-min data traces for the ${10}^6$, ${10}^5$, and control mice, respectively, [peaks are indicated with red circles in Figs. 3(b) and 3(c)].

The histogram of the peak amplitudes (heights) of the data shown in Figs. 3(b) and 3(c) is shown in Fig. 3(e). The mean amplitude was approximately $-110\mathrm{nA}$, which is equivalent to an SNR of 26.8 dB, where $\sigma$ in mice was 5 nA. However, individual spikes with SNRs as high as 40 dB were observed, indicating the high quality of the data here relative to our previous DFFC designs in which SNRs of peaks in vivo were on average only 14 dB.^7,8 While the mean and distribution of spike heights in Figs. 3(b) and 3(c) were similar, occasionally very large ($>500\mathrm{nA}$) spikes were observed at the higher concentration, which may have been due to more than one cell passing through the instrument FOV at the same time or simply a brightly labeled cell. Finally, the MM clearance kinetics of mice injected with ${10}^6$ labeled MM cells is shown in Fig. 3(f) in a 15-min period, starting 10 min after injection. Here, the count data are normalized to the 10-min time point, and the error bars represent the standard error observed in $N=4$ mice. As expected, MM cells cleared from circulation with approximately the same kinetics observed previously, with a half-life of $\sim 15\min$.¹³