Development of contextual intravital imaging of oxygen and cell dynamics
To relate dynamic cellular processes such as motility to the local availability of oxygen in vivo, we needed to develop an imaging method that combines two-photon phosphorescence lifetime microscopy with time-lapse imaging of fluorescence. In general, measurement of phosphorescence lifetime in time domain involves signal accumulation from multiple acquisition cycles whereby each unitary acquisition consists of a short excitation gate followed by recording of phosphorescence photon arrival times. In a typical, spot-wise measurement regimen, all the required unitary acquisition cycles are executed one after another for each measurement position. When applied to imaging, this traditional PLIM regimen results in very long pixel dwell times, which situation is incompatible with dynamic imaging. To overcome this limitation, we designed a ‘fast’ scanning two-photon (FaST) PLIM regimen that differs from the conventional one in the manner of signal accumulation and in the unitary acquisition design. For signal accumulation, we used only one excitation/acquisition cycle per each pixel visit in repeated scanning of the entire field of view, as opposed to the multiplicity of excitations for each pixel dwell time in a singular frame scan in the conventional regimen for time domain PLIM (Fig. 1a). Compared to the conventional regimen, the use of frame priority scanning did not shorten the overall duration of lifetime image acquisition, but it attained a much faster fundamental scanning rate i.e., in proportion to the required number of unitary decay acquisitions per pixel (about ten to hundred times faster in our application). Thus, in conjunction with recording fluorescence during each excitation gate, whose duration was appropriately optimized, the frame priority mode of phosphorescence signal accumulation enabled much higher frame rate for concurrent cell dynamics recording than it would be possible conventionally. We implemented this scanning regimen using a commercially available galvanometer scanner with photon counting detectors and dynamic range-extended electro-optical laser gating (Additional file 1: Figure S1).
Nonetheless, although greatly improved, the attained fluorescence scanning rate of about one frame per minute was still too slow for tracking of fast moving cells such as T cells. In the next step of method development, we considered the unitary acquisition cycle duration, which is determined by the temporal spacing of consecutive excitation gates. In the conventional PLIM regimen, the consecutive excitation gates must be spaced by at least ~ 10 decay times to prevent triplet state pile-up [16, 24]. However, because our regimen separates the consecutive excitation gates in space, i.e., between consecutive pixels, the time between individual excitation gates in each pixel is as long as one scan duration (e.g., tens of seconds). Therefore, no longer constrained by triplet state pile up, we could shorten each pixel dwell time by measuring only a fraction of each decay, known as ‘incomplete decays’ (Fig. 1a and b). However, in silico simulations revealed large standard deviations in lifetime determination from incomplete decays when using the conventional three-parametric fitting, i.e. amplitude, lifetime, and offset (background) (Fig. 1c). Therefore, to improve the fit accuracy, we added a short “Pre-Pulse” period prior to each excitation gate, which provided an additional independent measurement that in principle corresponds to the offset (Fig. 1b). With the offset measured by the pre-pulse, the decay could be fitted using a single-exponential model with only two parameters: amplitude and lifetime. In silico simulations showed that standard deviations in lifetime estimations from incomplete decays with pre-pulse were significantly reduced compared to those without pre-pulse, and were similar to those determined using ‘complete’ decays (Fig. 1c and Additional file 1: Figure S2A and B). This way, not only was the scanning rate improved by additional ~ 2.5-fold, but the speed of lifetime image acquisition was improved by ~ 2.5-fold for a chosen level of precision (e.g., 5% error in lifetime estimation), compared to that using the complete decays regimen. Taken together, by combining frame-priority scanning with pre-pulse-enabled incomplete decay measurements in the FaST-PLIM regimen, we could obtain one high resolution lifetime image for PtP-C343 in as little as one to several minutes while imaging fluorescence at the desired rate of 2 images per minute. A more detailed description of the animal models and imaging and in silico modeling procedures is provided in the subsequent sections.
Mouse acute lymphocytic leukemia imaging model
CD11c-EYFP mice [28] were obtained from Dr. Michel Nussenzweig, The Rockefeller University, New York, NY; hCD2-DsRed mice [29] were from Dr. Dimitris Kioussis, The National Institute for Medical Research, Mill Hill, London, U.K. The mouse strains were interbred to yield a double reporter strain. Male and female mice 6 to 12 weeks of age were used for experiments, and they were euthanized by CO2 inhalation followed by cervical dislocation. The B-ALL mCer cell line was generated by transducing fetal liver cells with a BCR-Abl p190 construct, following by transduction with a plasmid encoding the mCerulean fluorescent protein [30]. These and other cells were cultured in IMDM medium containing 10–20% fetal bovine serum, 1% β-mercaptoethanol, and 1% penicillin/streptomycin. Mice were injected with 1.25 × 105 B-ALL cells i.v. in 200 μL Hank’s Balanced Salt Solution (HBSS). At specified time points, mice were anesthetized using a 10 mg/ml ketamine and 1 mg/ml xylazine cocktail i.p. (at a dose of 10 μl/g of body weight), followed by an injection of 50 μl every 20–30 min to maintain anesthesia. After determining complete anesthesia by toe pinch, the scalp skin and membrane beneath were removed surgically to access the skull bone marrow for imaging. Following this procedure, mice were head-immobilized with a custom-made stereotactic holder on a heated microscope stage in enclosure maintained at 37 °C throughout the entire imaging session.
Lung metastases imaging model
Cyan fluorescent MCA-205 fibrosarcoma cells were generated by transducing the parental cell line with mCerulean vector. The reporter mice, as described for the leukemia model, were injected with 2.5 × 105 MCA-205-mCer cells in 200 μL HBSS through the tail vein. At specified time points, mice were anesthetized as for the leukemia model, and were tracheotomized and intubated for mechanical ventilation (Inspira, Warner Instruments). The large lung lobe was exposed by partial ribcage excision and immobilized for imaging using a heated suction holder system (STH-2, VueBio.com).
Intravital oxygen and dynamic cell imaging by FaST-PLIM
In order to visualize oxygen tensions in vivo, we injected mice with 40 μl of 1.7 mM solution of PtP-C343 oxygen probe [19]. Imaging was performed using a two-photon microscope system consisting of two titanium sapphire femtosecond lasers (Mai Tai, Spectra Physic), two electro-optical modulators (EOM) (Linos), polarization-based merge optics, SP5 laser scanner, DMI6000 microscope chassis (Leica Microsystems), and 25x NA 1.1 water immersion objective (Nikon) (Additional file 1: Figure S1). The emitted photons were collected by a four-channel non-descanned detector (Leica) consisting of two hybrid photodiode photomultiplier detectors (HyD) and two photomultiplier detectors optically arranged with appropriate dichroic mirrors and bandpass filters (Semrock). Laser gating and time correlated single photon counting were managed by the SPC-150/DP-120 subsystem (Becker & Hickl). For the concurrent mode of FaST-PLIM imaging, one laser was used at 875 nm and ~ 1.4 W infrared power at laser output, and the beam intensity was decreased down to 12.5 or 25% by a neutral density filter and passed through one EOM and dual stacked broadband polarizers. At these settings, infrared power measured at the objective back aperture was 25–55 mW. In concurrent mode, images were captured in 256 × 256 format (2.42 μm pixel size) with bi-directional scanning. Pixel dwell time of ~ 200 μs was obtained at 5 Hz line scan frequency, or 100 μs at 10 Hz, for example. During each pixel dwell time, the EOM/SPC150-modulated laser beam was off for 10 μs, then pulsed on for 10–30 (typically 20) μs, and again off for the remaining time. The pulse duration was adjusted for the best balance of fluorescence and phosphorescence brightness. Signals from one HyD detector were routed by B&H multi-channel scaler to the “phosphorescence” channel during the laser-off times, and to “fluorescence” channel during laser-on times. Simultaneously, entire signals from all four detectors were collected by the Leica SP5 system.
The rate of oxygen imaging in FaST-PLIM, i.e., the number of accumulated phosphorescence frames needed for given precision of lifetime estimations, was determined largely by the rate of phosphorescence photons and the noise. To minimize the number of frame accumulations, we considered the trade-off between temporal vs. spatial resolution. Accepting spatial resolution in the order of a single lymphocyte’s diameter (~ 10–20 μm), we were not concerned with limiting laser power to remain within the quadratic range (which would be required for a diffraction-limited excitation spot size) [24], and we applied short-radius circular pixel binning prior to automated decay curve fitting for each pixel position. This way, and using PtP-C343 at biocompatible concentrations, enough photons were accumulated to generate one lifetime image in 1–5 min, i.e. in just 3–12 scans. We used concurrent phosphorescence and fluorescence acquisition mode for specimens whose fluorescence was sufficiently bright when excited at wavelengths appropriate for excitation of PtP-C343. If one of the fluorescent labels was dim or required excitation at a different wavelength, we performed acquisition in sequential mode, in which cellular motility was imaged using dual laser excitation at the full duty cycle immediately before or after (or between) recording the phosphorescence. For sequential FaST-PLIM imaging, the fluorescence component data were captured for 5–30 min in 512 × 512 format (1.41 μm pixel size) at 600 Hz bi-directional line scan frequency, in up to eight channels (two lasers x four detectors), followed (or preceded) by the phosphorescence component for 1–5 min, as in the concurrent mode.
Lifetime data analysis
Post-acquisition image processing pipeline included conversion of lifetimes into oxygen partial pressures (pO2) and co-registering the phosphorescence channel with fluorescence channels for oxygen and cell movement cross-examination. Phosphorescence lifetimes were determined using single-exponent model in SPCI 6 software (B&H):
$$ {y}_n={A}_n{e}^{-\left(t-{t}_n\right)/{\tau}_n}+{b}_n $$
Where y is the number of photons, assuming analog response and no shot noise; n is the pixel index, A is the amplitude; t is time; tn is the decay start time; τ is the lifetime; and b is the background (offset). The offset setting was for the pre-pulse time bins. Up to ~ 5 circular binning was used to reach the average amplitude of at least 100 photons per binned pixel, with less binning for stronger signals. At 5 binning, the effective spatial resolution of phosphorescence was in the order of ~ 25 μm, i.e., comparable to the diameter of a T cell. In incomplete decays with non-descanned detection, decay carryover will cause overestimation of the offset, which would propagate to lifetime. At the current conditions, there is maximum 0.3% offset error (for τ = 50 μs, pulse-to-pulse time ≥ 290 μs), lifetime error similar. The phosphorescence lifetime images were converted into pO2 (mmHg) images based on a Stern-Volmer calibration curve [19] (Microsoft Excel 2017).
Contextual cell tracking
The fluorescence lif-format files and pO2 image sequences were combined in the Imaris 8.4.2 analysis software (Bitplane AG, Saint Paul, MN). Voxel dimensions were specified according to the objective used for image acquisition. If drift was present, it was corrected based on tracked landmark features. T cell motility was analyzed using the automated spot detection followed by autoregressive spot tracking and manual error correction. Quantitative analyses were done on all tracks with duration > 5 min, typically. Oxygen tension channel was Gaussian kernel 3 smoothed and the individual T cell-centered local oxygen tensions were the averages in cell-sized circular areas (~ 8 μm diameter).
Software for acquisition and analysis
Image acquisition and initial processing was performed using LAS 2.7 (Leica Microsystems) for intensity data, and SPCM 9.77 (Becker & Hickl) for time-resolved data. Further image processing and analysis was conducted using SPCImage 6 and 7 (B&H), IMARIS 8.4.2 (Bitplane, Inc.), ImageJ 1.51w (NIH), Anaconda Python Distribution 5.1 and Python 3.5.5.
In silico simulations
Simulations were performed using Microsoft Excel 2016 with Analysis Pack, Virtual Basic for Applications (VBA) and Realstats Resource Pack (real-statistics.com). Simulation conditions were: 150 photons per decay, 3:1 phosphorescence to background ratio, Monte Carlo Poisson (shot) photon noise, 256 time bins, 10 μs pre-pulse or no pre-pulse, 20 μs excitation gate. For the “conventional” complete decay method, the dwell time was 500 μs, and for the partial decay with or without pre-pulse, the dwell time was 100 μs.
Statistical analysis
The two-sided student’s t-test was used to evaluate the null hypothesis that there was no difference between two groups, and a p-value of < 0.05 was considered statistically significant a priori. ANOVA with Tukey multiple comparison test was used where the means of more than two groups were to be compared simultaneously. Pearson’s correlation was used to evaluate correlations between two variables. r values > 0.1 with corresponding p values < 0.05 were considered to be significant a priori.