2.1.

Two-Color DiFC Instrument DiFC

The two-color DiFC system uses a similar design to our previously reported GFP-compatible b-DiFC system.^11,21,25 The system uses a 488 nm laser coupled into two specially designed optical fiber probes. Each probe consists of a single source fiber surrounded by a ring of 8 detection fibers [Fig. 1(a)]. The probe tips have internal mounted filters to reduce fiber autofluorescence including a central 488 nm band-pass filter (BP-f) and a ring-shaped 503 nm long-pass filter (LP-f). The eight detection fibers are grouped into two bundles of four which each terminate on an output fiber coupler, emission band-pass filters, a second focusing lens, and photomultiplier tube (PMT). The two sets of detection fibers are interleaved in the probe tip as shown in Fig. 1(b). The tip can then be aligned on the skin surface above a major blood vessel, for instance the ventral tail artery of a mouse [Fig. 1(c)], to excite and detect FP-expressing circulating cells.

Fig. 1

(a) DiFC probe design showing source and detection optical fibers, excitation BP and fluorescence collection LP filters, and aspheric lens tip. Diagram reprinted with permission from Patil et al.²¹ (b) Each two-color DiFC probe contains two sets of four fibers that are interleaved and coupled to two separate detector arms. (c) Two probes are aligned on the mouse tail, approximately above the ventral caudal vascular bundle. (d) The excitation spectra of GFP and tdTomato FP are shown, with a black line indicating the 488 nm laser wavelength. (e) The emission spectra of GFP and tdTomato are shown, along with the two-color DiFC emission filters used for each fluorophore, indicating modest fluorophore “bleed” between channels. (f) FP-expressing cells moving through a blood vessel are excited by laser light. Fluorescent light is collected by the detection fibers and split between two detector arms (filters and PMTs) for each fluorophore. (g) The detected light appears as transient peaks (“spikes”) in light intensity with a time delay between the two probes. In principle, this allows for detection of GFP+ CTCs and clusters, tdTomato+ CTCs and clusters, and two fluorophore multicellular CTC clusters. (h) Example fluorescence microscopy images of each are shown.

GFP and tdTomato were chosen as target FPs. Both tdTomato and GFP can be excited by 488 nm light [tdTomato with lower efficiency than GFP, as shown in Fig. 1(d)], which simplified the instrument construction. To achieve this, one of the two fiber probe outputs is fitted with a 535/50 filter (ET535/50m; Chroma Technology Corporation, Bellows Falls, Vermont, United States) and 536/40 nm filter (FL-004682; IDEX Health and Science, LLC, Rochester, New York, United States) for GFP detection, and the other output with a 610/75 nm filter (ET610/75m; Chroma Technology Corporation) for tdTomato. As we show, the tdTomato and GFP emission spectra are sufficiently separated that the two emission filter sets allow detection of both FPs with minimal inter-channel “bleed” [Fig. 1(e)]. The brightness of tdTomato—up to 2 to 3 times brighter than GFP—allows for detectable emission despite the lower excitation efficiency with the 488 nm laser.²⁶

The instrument uses two fiber probes that are placed on the skin surface, approximately above the blood vessel of interest [Fig. 1(c)]. The data from the four PMTs, resulting from the two fiber probes, can be visualized as four channels: probe 1 – green (P1-G), probe 1 – orange (P1-O), P2-G, and P2-O [Fig. 1(f)]. The use of two fiber probes allows us to determine the direction and speed of circulating GFP- and tdTomato-expressing cells. For example, a GFP+ cell moving in a blood vessel beneath probe 1 followed by probe 2 will be detected as peaks in light intensity on channels P1-G and P2-G with a time delay between peaks [Fig. 1(g)]. Similarly, tdTomato+ cells will be detected in channels P1-O and P2-O. This allows us to specifically identify cells moving in the blood vessel of interest as described further in the next section.

2.2.

Signal Processing and CTC Detection

The signal processing algorithm used for two-color DiFC follows the following steps, which is a modified version of our previously published algorithm:^20,21,25

2.3.

Cell Lines and CTC Clusters In Vitro

2.4.

Optical Flow Phantom Experiments In Vitro

To first validate the two-color DiFC system, we used a tissue-mimicking optical flow phantom as we have described previously.²¹ The phantom is a block of scattering plastic that approximately mimics the optical properties and autofluorescence of biological tissue in the visible range. We threaded strands of Tygon tubing (TGY-010-C; Small Parts, Inc., Seattle, Washington, United States) through a through-hole at 0.75 mm depth, mimicking the depth of a blood vessel in a mouse tail. A microsyringe pump is used to pass suspensions of GFP- and tdTomato-expressing cells at final suspension concentrations of approximately ${10}^3$ cells per mL for single cells and ${10}^4$ cells per mL for clusters (which corresponded to approximately ${10}^3$ CTCCs per mL) with a flow speed of $25\mu \mathrm{L}$ per minute through the tubing.

2.5.

Mouse Experiments In Vivo

All mice were handled in accordance with Northeastern University’s Institutional Animal Care and Use Committee (IACUC) policies on animal care. Animal experiments were carried out under Northeastern University IACUC protocol #21-0412R. Mice were caged in groups of five or less, and all animals were fed a diet of low fluorescence animal chow (AIN 93M Mature Rodent Diet, Ziegler Feed, East Berlin, Pennsylvania, United States).

We used an MM disseminated xenograft model (MM DXM) (which we have used previously²¹) with an equal mixture of GFP- and tdTomato-expressing MM.1S cells. 8-week-old male severe combined immunodeficient (SCID/Bg) mice (Strain code 250; Charles River Laboratories, Cambridge, Massachusetts, United States) were injected via tail vein with $200\mu \mathrm{L}$ of PBS containing $2.5\times {10}^6$ GFP-MM.1S cells and $2.5\times {10}^6$ tdTomato-MM.1S cells ($N=4$). DiFC scanning was performed on each mouse held under inhaled isoflurane when CTCs were expected to enter circulation $-28$ days.

Additionally, three NOD SCID Gamma (NSG) mice (Strain code 005557; The Jackson Laboratory, Bar Harbor, Maine, United States) were scanned with DiFC for control data.

2.6.

Whole-Body Hyperspectral Fluorescence Cryo-imaging

To visualize the spatial distribution of GFP and tdTomato expressing cells in the MM DXM model, animals were euthanized and submerged in optimal cutting temperature compound in preparation for hyperspectral fluorescence cryo-imaging. This approach, as described elsewhere,^25,27 produces high-resolution 3-D white light and fluorescence images of the entire animal by imaging the frozen specimen during automated serial sectioning. For this study, the specimen was imaged using a white light emitting diode (LED), a 530 nm LED (Mightex, Toronto, Ontario, Canada) with a 550 nm shortpass filter for tdTomato excitation, and a 470 nm LED (Mightex) with a 475 nm shortpass filter for GFP excitation. The resulting RGB and fluorescence image stacks were assembled and rendered using 3D slicer.²⁸ In 3D slicer, the GFP and tdTomato tumors were first segmented from the fluorescence image stacks while removing regions of autofluorescence in the stomach and intestines. Next, the overlapping tumor regions expressing both GFP and tdTomato were created by determining the intersection between the GFP and tdTomato segmentations. Finally, the intersecting region was subtracted from both segmentations to generate distinct GFP, tdTomato, and overlapping regions. These segmentations were then rendered into 3-D models for visualization.

3.1.

Two-Color DiFC Performance in Phantom

We first performed two-color DiFC in our flow phantom model in vitro [Fig. 2(a)] using suspensions of GFP+ and tdTomato+ MM or BC cells. The signals measured from cells expressing either of the FPs are readily distinguishable as they are primarily detected as peaks in either the green [GFP, Fig. 2(b)] or orange [tdTomato, Fig. 2(c)] channels. Our selected combination of FPs, filters, and phantom optical properties yielded very similar SNRs for both GFP+ and tdTomato+ cells [Figs. 2(d) and 2(e)]—MM cells of both fluorophores had a mean SNR of 28.3 dB while the BC GFP+ and tdTomato+ cells had mean SNRs of 34.1 and 34.0 dB, respectively.

Fig. 2

(a) We used an optical flow phantom model for initial in vitro validation of two-color DiFC. (b) and (c) Representative DiFC data recorded when suspensions of (b) GFP+ MM cells and (c) tdTomato+ MM cells were passed through the phantom. As shown, GFP+ cells were detected in the green channels, and tdTomato+ cells in the orange channels. (d) and (e) The combination of fluorophore expression and DiFC filters resulted in similar SNRs for both GFP and tdTomato cells. This was the case for both (d) MM and (e) 4T1 cells. (f) Particularly bright cells or 1FP clusters resulted in slight bleed between the two fluorescence channels of either probe. These were distinguished from 2FP clusters by analysis of the relative peak amplitudes as described in Sec. 2.2.

We also observed that some particularly bright individual cells and 1FP clusters result in $2\lambda$ peaks—one peak in the primary channel (green for GFP, orange for tdTomato) and one peak in the opposite channel due to fluorophore emission bleed [Fig. 2(f)]. However, we were able to distinguish these 1FP detections apart from detections of 2FP clusters as described in Sec. 2.2. To show this, we separately performed DiFC on suspensions of GFP+ and tdTomato+ MM cells (MM) and 1FP clusters (BC) in a phantom. We note that because MM cells are cultured in suspension they typically grow as individual cells and small clusters, whereas adherent 4T1 BC cells readily form large multicellular groupings in culture.²⁹

For all detected $2\lambda$ peaks, the green and orange peak amplitudes are plotted in Fig. 3. The dashed curves show the maximum expected amplitude of fluorophore spectral bleed between green and orange detection channels [Eq. (1)]. As such, detections plotted between the dashed lines are determined by our algorithm to be 2FP (blue circles) multi-cellular clusters and all detections outside the lines are 1FP (small green or pink circles) from single cells or 1FP clusters. The TR value in Eq. (1) was estimated as 0.057 for GFP+ cells and 0.089 for tdTomato+ cells by calculating the highest TR values of the single-fluorophore phantom MM and BC data with Eq. (2) [Figs. 3(a) and 3(b)]. We note that we selected a ‘conservative’ threshold that accounted for the worst-case coincidence of signal noise and peak detection. As shown in Figs. 3(a) and 3(b), this resulted in no false positive identification of 2FP clusters when running 1FP suspensions of cells through the phantom.

Fig. 3

Green channel and orange channel peak amplitudes of $2\lambda$ peaks can be compared to identify 1FP and 2FP detections. The dotted diagonal lines indicate hypothetical 1:1 GFP and tdTomato detections, and the dashed curves represent Eq. (1), which is the threshold for $2\lambda$ peaks that distinguishes 1FP or 2FP detections. Therefore, points between the dashed lines indicate 2FP detections, and outside the lines indicate 1FP detections. (a,b) Bright MM single cells and small clusters detected with two-color DiFC were identified as GFP-only (a) or tdTomato-only (b). (c,d) GFP and tdTomato 1FP BC clusters were also detected as 1FP. (e) GFP and tdTomato 1FP BC clusters were mixed and scanned simultaneously which resulted in few false 2FP detections. (f) Co-cultured GFP and tdTomato BC cells and clusters resulted in many 2FP detections.

We also collected DiFC data of GFP+ clusters and tdTomato+ BC clusters through the phantom, either one FP at a time [Figs. 3(c) and 3(d)] or mixed in a combined suspension [Fig. 3(e)]. As shown, all detections were correctly identified as 1FP for the single FP cluster cases [Figs. 3(c) and 3(d)]. In the case of mixed 1FP cluster suspension [Fig. 3(e)], most detections were correctly identified as 1FP, although a small number were labeled as 2FP. This occurred for only a small proportion of detections (1.5%) during scans with high flow rates of 15.5 detections per minute. As such, we surmise that the errors were likely the result of coincident detections of both 1FP GFP+ and tdTomato+ clusters in the phantom (i.e., both passed through the $\sim 1\mathrm{mm}$ diameter DiFC field of view at the same time), as opposed to incorrect classification of a single detection.

Finally, we created 2FP clusters of BC CTCs by co-culturing GFP+ and tdTomato+ cells. Suspensions were passed through a phantom and scanned with two-color DiFC [Fig. 3(f)]. The resulting DiFC data showed both 1FP and 2FP detections, which were 53.9% and 46.1% of all $2\lambda$ peaks, respectively. There was a large variety of 2FP clusters, some with 50% GFP and 50% tdTomato and others with more GFP or tdTomato cells.

3.2.

Two-Color DiFC in Multiple Myeloma Xenograft Model Mice In Vivo

We next performed two-color DiFC on MM tumor bearing mice [Fig. 4]. Mice were injected with a suspension of GFP+ and tdTomato+ MM cells (1:1 ratio). MM cells are known to initially rapidly home to the bone marrow niche after injection, steadily proliferate and then circulate in the peripheral blood in increasing numbers over time.²¹ Since cells were otherwise identical (aside from FP expression), we expected this proliferation would occur at approximately the same rate for both GFP+ and tdTomato+ cells. Representative two-color DiFC data measured periodically during tumor growth are shown in Figs. 4(a)–4(d). Here, Figs. 4(a) and 4(b) are raster plots, where each vertical line represents a DiFC detection of a GFP+ and tdTomato+ MM cell, respectively. When plotted together [Figs. 4(c) and 4(d)] we found that the number of both CTC types were, as expected, observed at approximately the same frequency and rate of increase.

Fig. 4

(a) GFP+ and (b) tdTomato+ MM CTC detections in a mouse over time in a disseminated xenograft model. Each vertical line represents a single CTC detection in time. (c) Both GFP and tdTomato detections are shown together. (d) The estimated concentration of CTCs in the peripheral blood (PB) increased over time and was similar for both cell populations.

3.3.

Correlation of GFP and tdTomato CTCs

We also visualized DiFC CTC detections as moving averages, where CTCs are counted in 2-min moving intervals through the scan. Figure 5 shows a representative DiFC scan with individual cell detections [Fig. 5(a)] and moving averages through the scan for GFP [Fig. 5(b)] and tdTomato [Fig. 5(c)] detections. These data are typical of DiFC measurements,¹¹ with transient periods of higher and lower detection rates observed during the scan for both GFP+ and tdTomato+ cells.

Fig. 5

(a) Representative raster plots of GFP+ (first row) and tdTomato+ (second row) MM CTC detections from a DiFC scan. The numbers of (b) GFP+ and (c) tdTomato+ CTCs detected in moving 2-min intervals fluctuated over the length of the scan. (d) The Pearson correlation coefficient (PCC) of paired GFP and tdTomato detection rates was 0.287. (e) PCCs for all DiFC scans with at least 0.5 CTCs per minute are shown. Overall, these data indicate only weak correlation between GFP and tdTomato detection rates suggesting differences of CTC numbers in circulations as opposed to fluctuations in blood flow rates.

An open question was whether this variation in count rates is due to short-term variations in CTC shedding from tumor into circulation, or transient changes in cardiovascular output (and therefore DiFC blood volume sampling) while mice are under inhaled anesthesia. We investigated the correlation between the paired moving averages of GFP and tdTomato detections using the Pearson correlation coefficient (PCC) as shown in Fig. 5(d). For this scan, a PCC of 0.345 was obtained, suggesting only a weak correlation between the GFP and tdTomato detection rates ($p$-value of 0.018).

Likewise, we calculated the PCC for all DiFC scans in this study with overall detection rates of at least 0.5 CTCs detected per minute, as shown in Fig. 5(e). Overall, the PCC data showed weak positive correlation between the channels (median PCC of 0.186 and mean of 0.199). Since GFP and tdTomato detection rates are poorly correlated, this suggests that temporal fluctuations are due to transient changes in rates of CTC shedding into the bloodstream, as opposed to changes in vascular output.

3.4.

CTC Clusters in Multiple Myeloma Mice

A total of 260 $2\lambda$ peaks were detected in the MM mice over a total 18 h of DiFC data. These are plotted in Fig. 6(a). Our analysis determined that 13.5% were 2FP detections [representative example is shown in Fig. 6(b)] and 86.5% were 1FP detections [representative example is shown in Fig. 6(c)].

Fig. 6

(a) Amplitudes of all $2\lambda$ peaks detected in MM mice. As in Fig. 3, the dashed curves indicate the threshold for which $2\lambda$ peaks are classified as 2FP or 1FP. The solid lines indicate the estimated amplitude of 3 GFP and 3 tdTomato CTCs. 1FP peaks with amplitudes of at least three CTCs are considered clusters. Representative (b) 2FP and (c) 1FP detections are shown. Histograms show the estimated sizes of (d) 2FP and (e) 1FP CTCCs (in number of cells).

We further estimated the size of the CTC clusters based on the combined peak amplitudes as summarized in Figs. 6(d) and 6(e). CTCCs are defined as two or more cells traveling together in circulation, or three or more nuclei travelling together in circulation. The latter definition excludes the possibility of a single CTC undergoing mitosis being identified as a CTCC.⁵ Therefore, in our calculations, 1FP detections that had amplitudes consistent with three or more GFP+ or tdTomato+ nuclei were considered 1FP clusters (89.3% of 1FP detections). 2FP detections contained at least one GFP- and tdTomato-expressing CTC, so that all 2FP detections were considered 2FP CTCCs. In combination, CTC clusters of either type (1FP or 2FP) represented 14.6% of all CTC detections. This is similar to the rate that we previously reported in a GFP-only MM tumor model.²¹

Figure 6(d) shows a histogram of cluster sizes for measured 2FP clusters (mean size = 9.3 cells; median size = 3 cells), and Fig. 6(e) shows a histogram of clusters sizes for measured 1FP clusters (mean size = 23.4 cells; median size = 16). The difference in estimated mean sizes here is due to the two (disparate) definitions used. However, in general this distribution of sizes is also similar to our previous work.²¹

We further note that we manually curated our CTCC detections after automated raw data processing and removed suspected artifacts, e.g., 2FP detections that appeared to arise from a motion artifact, or high amplitude (greater than 1500 mV) $2\lambda$ detections that saturated one of the detector PMTs and gave inaccurate estimation of the relative peak intensities.

3.5.

Whole-Body Hyperspectral Fluorescence Cryo-Imaging

After completion of DiFC scanning we performed whole-body hyperspectral fluorescence cryo-imaging of tumor bearing mice. Compared to whole body planar fluorescence or bioluminescence imaging techniques,³⁰ cryo-imaging provides significantly superior spatial resolution and quantification of fluorescence signals.²⁷ Based on prior characterization of the MM DXM model, we expected tumor to proliferate in the skeleton of the mouse.³¹ Hyperspectral fluorescence cryo-imaging confirmed these expectations, with GFP+ and tdTomato+ tumor appearing along and inside the spine, and within the brain, skull, scapula, humerus, femur, ribs and muscles of the legs, and arms. As shown in the whole-body rendering of a representative mouse [Fig. 7(a)] nearly all bulk tumor appeared to be single-FP, as opposed to homogenously mixed 2FP tumors. These data are consistent with our observation that most detected CTCCs were 1FP with relatively few 2FP CTCCs.

Fig. 7

Hyperspectral fluorescence cryo-imaging showed that most tumor masses in the mice were single FP. This is consistent with our observation that $>85\%$ of CTCC detections were 1FP and not 2FP. (a) The full body image of a representative mouse is shown as well as (b) a top-down view of a section of the spine displaying two examples of 2FP tumor masses. (c) and (d) Additional 2-D cross-sectional sagittal and axial views of one 2FP tumor mass are also shown along the spine. GFP is shown in green, tdTomato in red, and overlap in blue.

Upon close investigation of the spine in Fig. 7(b), two examples of mixed tumors can be seen in blue. One of the 2FP tumors is further visualized in sagittal and axial cross-sectional views along the spine in Figs. 7(c) and 7(d). Overall, though, mixed tumors represented only a small fraction of the total observed tumor volume.