2.1.

Cell Lines

MDA-MB-231-GFP, mouse LM2, and mouse 4T1(-tdTomato, -dsRed, -mCherry) cells were maintained in DMEM with 7% FCS and 1% Penicillin streptomycin at 37°C under a humidified 5% CO2 atmosphere. For in vitro experiments, cells were harvested by trypsin, diluted in culture medium followed by two cycles of centrifugation and resuspension in phosphate buffered saline (PBS). Cell dilutions were then transferred to clear, nonfluorescent PCR tubes (Thermo scientific S98752, Waltham, Massachusetts) followed by centrifugation and imaging.

2.2.

Animals

All procedures were performed in accordance with the Massachusetts General Hospital animal welfare guidelines. Female Nu/Nu mice were obtained from the Cox-7 defined-flora animal facility at MGH. Prior to experimentation mice were either anesthetized or euthanized by inhaled Isoflurane (2% or 5% respectively). For subcutaneous injections, MDA-MB-231-GFP were diluted in PBS, and 30 μL were injected in anesthetized mice via 30 gauge needle followed immediately by imaging. For direct injection of the superficial cervical lymph nodes, a skin flap was performed by cutting below the sternum and along both midaxillary lines followed by careful, blunt dissection to expose the cervical lymph nodes. Approximately 10 μL of $1\times {10}^4$ MDA-MB-231-GFP $\mathrm{cells}/\mu \text{L}$ were injected via 30 gauge needle into one of the superficial cervical lymph nodes, followed by replacement of the skin flap and imaging.

2.3.

Imaging System

Time resolved images were acquired with a custom built imaging system described in detail previously.²⁰ Briefly, fluorescence was excited with the pulsed broadband (480 to 850 nm) output of a titanium sapphire driven (800 nm @ 80-MHz repletion rate, Spectraphysics, Santa Clara, California) photonic crystal fiber (Thorlabs NL-PM-750, Newton, New Jersey) filtered through either $495\pm 25\mathrm{nm}$ or $543\pm 5\mathrm{nm}$ filter. The resulting fluorescence emission was detected with either a $529\pm 25\mathrm{nm}$ filter, or a 560-nm long pass filter coupled to an intensified CCD camera (PicostarHR, LAVision, Gmbh; 500-ps gate width, 600-V gain, 150-ps steps, $4\times 4$ hardware binning). Excitation powers were $\sim 50\mu \text{W}/{\mathrm{cm}}^2$ at both 495 and 543 nm. Camera integration times ranged from 100 ms to 3 s.

2.4.

Statistics

Variance between groups was first assessed via one way analysis of variation (ANOVA). Differences in means between two groups were assessed via a two tailed $T$-test, significance was determined at a $p$ value of 0.05.

3.1.

TD FL Multiplexing of FP Fluorescence and Tissue AF

We first validated the ability of TD FL multiplexing to detect small numbers of GFP expressing cells in the presence of high AF background and excitation leakage with in vitro measurements. 1 to 12 K MDA-MB-231-GFP cells were mixed and pelleted with $2.5\times {10}^6$ LM2 cells that serve as the background AF component. Fluorescence was excited at $495\pm 25\mathrm{nm}$ and detected using a $529\pm 25\mathrm{nm}$ filter. In order to minimize the influence of excitation leakage through the emission filter, the fluorescence temporal point spread function (TPSF) was analyzed between 2.3 and 11 ns after the excitation peak. The TPSF from a tube containing only GFP expressing cells can be fit with a single exponential decay corresponding to the GFP FL of 2.6 ns while the AF from LM2 cells is best fit with a biexponential decay with lifetimes of 0.6 and 2.1 ns.

In Fig. 1(a), we show that the recorded TPSF from the mixed cell pellets approaches the TPSF from pellets containing only GFP expressing cells as the number of GFP cells in the mixture is increased. Figure 1(b) shows the CW fluorescence intensity (grayscale), calculated by integrating the TPSF for all time points and the FL maps (color) obtained from single exponential fitting of the TPSF at each pixel. The presence of GFP fluorescence is evident from increases in both CW signal and FL. However, the FLs in the images do not accurately represent the known GFP lifetime of 2.6 ns since the observed TPSF is a mixture of the lifetimes of the tissue AF and the FP. In order to assess the contribution of each lifetime component, we fit the decay of the TPSF at each image pixel according to

Fig. 1

Time domain (TD) fluorescence lifetime (FL) multiplexing enables in vitro detection of GFP expressing cells in the presence of high autofluorescence and excitation leakage. MDA-MB 231-GFP cells were mixed with 2.6 million LM2 cells, pelleted and imaged. (a) TD measurements were made from 2.3 to 11 ns after the excitation peak to avoid excitation light leakage. Inset shows the instrument response function (black) obtained using a white paper placed on the imaging stage. Red dashed line shows the start of data acquisition (2.3 ns), avoiding the excitation signal. The autofluorescence temporal point spread function (TPSF) from LM2 cells is best fit with a biexponential decay with lifetimes 0.6 and 2.1 ns (dotted black line), while the shape of the normalized fluorescence TPSF from measured cell pellets approached the single exponential decay of GFP (lifetime 2.6 ns, green) with increasing number of GFP expressing cells, indicated by direction of the arrow. (b) Similarly, continuous wave (CW) fluorescence intensity and FL increase with the addition of GFP cells. (c) A dual basis function analysis [Eq. (1) in the text] of each TPSF reveals a constant AF amplitude ($a_{\mathrm{AF}}$, red) and an increasing GFP amplitude ($a_{2.6}$, green) with increasing GFP cell number. (d) The CW fluorescence intensity (red square, right axis) shows a large background due to the AF from LM2 cells and increases further with the addition of GFP cells. The GFP/AF amplitude ratio, $a_{2.6}/a_{\mathrm{AF}}$ (blue diamond, left axis) eliminates systemic variations between measurements (such as fluctuations in illumination power or integration time) and shows a linear dependence on the true GFP concentration, approaching zero for the case of pure AF cells.

We reiterate that in the above measurements, we have exploited an important advantage of TD imaging, namely the ability to exploit the distinct time scales of the fluorescence excitation and emission processes. This is illustrated in the inset of Fig. 1(a): the excitation laser pulse (black line) is rapidly diminished by 2.3 ns after the peak whereas the GFP fluorescence decay extends to an additional 8.7 ns. Thus, TD techniques allow the use of filter sets with small spectral band separations under conditions where excitation light bleed-through becomes a significant factor, such as low-probe concentrations or cells seated deeply within tissue.

Next, we performed in vivo experiments to ascertain the suitability of TD FL multiplexing for detecting small numbers of GFP expressing cells under the skin in nude mice. We injected 1 K to 15 k GFP cells subcutaneously in 30-μL saline solution followed immediately by imaging. As in our in vitro model, skin AF (excited at 495 nm) exhibits a biexponential decay, and changes in the measured TPSF are observed following the addition of GFP expressing cells [Fig. 2(a)]. From the relatively minor impact on the shape of the TPSF from 15 K subcutaneous GFP cells, it is apparent that the mouse skin AF contributes significantly more to the total fluorescence than did the AF cells in the in vitro measurements (Fig. 1). Indeed, the location of the injected cells is not clear in the CW images below 10-K GFP cells [In Fig. 2(b), black spots indicate boundary of injected area]. However, the subcutaneous location of the injected GFP cells can be observed in lifetime maps as areas of increased lifetime, compared with the background skin AF [Fig. 2(b)], well before these are visible in the CW images. As shown in Fig. 2(c), we once again observe a statistically significant linear relationship between the GFP:AF amplitude ratio and the number of implanted GFP cells. In addition, we can conclude that we have an approximate detection limit of 1.5-K subcutaneous GFP cells [Fig. 2(c)].

Fig. 2

TD FL multiplexing enables in vivo detection of subcutaneous GFP expressing cells at lower levels than CW imaging. MDA-MB-231-GFP cells in 30 μL of phosphate buffered saline (PBS) were injected subcutaneously in a nude mouse. (a) The fluorescence TPSF measured from 10 and 15 K subcutaneous GFP cells (gray lines) is a combination of the biexponential decay of mouse skin (blue solid line) and the single exponential decay of a tube of GFP expressing cells (green dashed line). (b) Subcutaneous GFP cells in CW intensity images (grayscale) are only visible for $>10\mathrm{K}$ cells (black spots indicate injection site boundaries) due to the strong background AF. In FL images (color), the location of GFP cells is revealed as a localized increase in FL, down to 1.5 K injected cells. (c) Both CW fluorescence intensity (red squares) and the amplitude ratio $a_{2.6}/a_{\mathrm{AF}}$ increase with the number of subcutaneous GFP cells. Note that the CW intensity is dominated by the AF background.

To demonstrate the applicability of TD FL multiplexing of FP and AF in a realistic animal disease model, we employed a model of experimental lymph node metastasis. Approximately 100 K MDA-MB-231-GFP cells were directly injected into the right-central superficial cervical lymph node of a freshly sacrificed mouse. Comparison of the CW images from before and after injection [Figs. 3(a) and 3(d), respectively] reveal a minor increase in intensity directly over the injected lymph node, with the predominant signal arising from tissue AF. In the single-exponential lifetime maps, the area directly above the injected lymph node shows a longer lifetime than the surrounding tissue [Fig. 3(e)]. Figures 3(c) and 3(f) show the amplitude maps generated using the dual basis function approach outlined above [Eq. (1)]; the basis functions consisted of a single exponential for the GFP fluorescence (2.6-ns lifetime) and a biexponential basis function for AF (lifetimes of 0.6 and 2.5 ns), determined from the TPSF of the mouse skin from the contralateral side. The contrast-to-background ratio (CBR) was estimated¹² for the CW and TD approaches as ${\mathrm{CBR}}_{\mathrm{CW}}=U^t/U^0$ and ${\mathrm{CBR}}_{\mathrm{TD}}=a_{\mathrm{FP}}^t/a_{\mathrm{FP}}^0$, where $U^t$ and $a_{\mathrm{FP}}^t$ are the average CW intensity and FP decay amplitude within the tumor region and $U^0$ and $a_{\mathrm{FP}}^0$ are the corresponding values outside the tumor region. The injected node is easily detectable in the FL amplitude map [Fig. 3(f)], with a much higher CBR (37.5) than the CW image (1.2). Images acquired after removing the skin [Figs. 3(g)–3(i)] and dissection of the superficial cervical lymph nodes [Figs. 3(j)–3(l)] confirm that the GFP cells were indeed located only in one superficial lymph node. These results demonstrate that lifetime multiplexing dramatically improves the contrast for detecting FPs in intact mice, potentially enabling early stage detection of metastasis in shallow organs such the mammary fat pad or lymph nodes. However, for imaging deeper organs, red shifted FPs¹⁸ are more appropriate.

Fig. 3

TD imaging enables detection of experimental lymph node metastasis with a high contrast-to-background ratio (CBR). The superficial cervical lymph nodes were exposed by performing a skin flap procedure in a freshly sacrificed nude mouse. Images were taken before (a–c) and after (d–f) injection of the right cervical node with $\sim 10\mu \text{L}$ of a $10\mathrm{K}/\mu \text{L}$ MDA-MB-231-GFP cell solution in PBS. The corresponding FL amplitudes, $a_{\mathrm{AF}}$, $a_{2.6}$ resulting from a multiexponential fit of the fluorescence TPSF with basis functions [Eq. (1)] for mouse skin AF and GFP, are shown in (c) and (f) as red and green components, respectively, of a composite RGB image (Thus, yellow indicates complete overlap of the two signals). Prior to injection, there were no detectable regions of increase in (a) CW intensity, (b) FL, or (c) FL amplitude corresponding to GFP ($a_{2.6}$, green). After injection of the GFP cells, the location of the injected node can be determined from a localized increase FL directly over the injected node (E), but is barely visible in the CW fluorescence intensity (d). The decay amplitude images (f) clearly reveals the location of the injected node under the mouse skin with a (m) $30\times$ increase in CBR compared with the CW image. Imaging performed in situ (g–i) and of dissected cervical lymph nodes (j–l) confirm that only the right superficial cervical lymph node was injected.

3.2.

TD FL Multiplexing Enables Simultaneous Imaging of Multiple Spectrally Overlapping FPs

In addition to the separation of FP and tissue AF signals, TD measurements enable multiplexed imaging of spectrally similar FPs based on their distinct lifetimes.^12,21 This is important, since the large overlap in the excitation and emission spectra of FPs poses a limit on the number of FPs that can efficiently be used simultaneously using spectral unmixing alone. Multiplexed imaging of several FPs can allow the simultaneous in vivo visualization of multiple genetically encoded tags with high specificity in a living animal. Figure 4(a) shows the distinct FLs of three red shifted FPs and mouse skin AF excited at 543 nm. As we observed with GFP, the FPs exhibit single exponential decay behavior, while the mouse skin AF for 543-nm excitation is best fit with a biexponential decay (lifetimes of 0.27 and 2.1 ns). In measurements of tubes containing cells expressing dsRed, mCherry, or tdTomato, we find FLs of $2.4\pm 0.0015\mathrm{ns}$, $1.52\pm 0.0016\mathrm{ns}$, or $3.08\pm 0.002\mathrm{ns}$, respectively [Fig. 4(c)], while the CW images do not distinguish the three FPs [Fig. 4(b)].

Fig. 4

TD FL multiplexing enables the separation of three red shifted fluorescent proteins (RFPs) from mouse skin autofluorescence. Measurements of pelleted RFPs were made in vitro before and after placement under the skin of a sacrificed nude mouse. (a) tdTomato, dsRed, and mCherry have distinct, single exponential fluorescence decays with corresponding lifetimes of 3, 2.4, and 1.5 ns, respectively, while mouse skin AF (543-nm excitation) is best fit by a biexponential decay (lifetimes 0.27 and 2.1 ns). (b) CW intensity provides no identifying contrast between the tubes containing the three RFPs. (c) A single exponential fit of the fluorescence TPSF correctly reveals the lifetime of the FP expressed in each tube. (d) CW image cannot distinguish tubes containing the three RFPs placed under the skin of a nude mouse. (e) Lifetime map of the tubes under the mouse skin from a single exponential fit to each pixel. The effect of tissue AF leads to incorrect lifetimes for the tubes. (f) The contribution of each source of fluorescence can be determined by performing a multiexponential fit of the total fluorescence TPSF with four basis functions: one corresponding to the biexponential decay of the skin AF, and three single exponential decays corresponding the known FL of each RFP. For visualization, the resulting FL amplitudes: $a_3$, $a_{2.4}$, and $a_{1.5}$ are shown in red, green, and blue, respectively, of a single RGB image. Here, the FL amplitude corresponding to each FP correctly highlights and identifies the contents of each tube. The tissue AF amplitude $a_{\mathrm{AF}}$ distribution is shown in grayscale.

Figure 4(d) shows the CW image of the three FPs placed under the skin of a sacrificed nude mouse. Despite the relatively high concentration of the FPs, tissue AF is $\sim 18\%$ as bright as the tdTomato CW intensity. In the TD, the tissue AF contribution is also manifest in the relatively muted FL image of these subcutaneous tubes [Fig. 4(e)]. It should be noted especially that the longer lifetime tdTomato and dsRed are more strongly affected by the tissue AF, effectively reducing their observed lifetimes. Thus, single exponential fits lead to erroneous lifetimes and quantitation when analyzing in vivo data.⁶ However, the basis function approach [Eq. (1)] allows the correct identification and quantitation of the different FPs based on the corresponding decay amplitudes $a_{\mathrm{FP}}$ and $a_{\mathrm{AF}}$ [Fig. 4(f)]. The ratio of the recovered amplitudes $a_3:a_{2.4}:a_{1.5}$ was $0.8:1:0.89$ for the FPs under the skin and was similar to the in vitro amplitude ratio of $0.9:1:0.8$, indicating the ability of the method to quantitate multiple FPs simultaneously present in an animal.