Protocol for in vivo fluorescence lifetime microendoscopic imaging of the murine femoral marrow

2.6.1 Laser systems

1. Near-infrared, femtosecond-pulsed laser (Ti:Sa, Chameleon Ultra II, Coherent, Duisburg, Germany) (excitation source, wavelength range 700 nm–1,080 nm; repetition rate: 80 MHz; pulse width at output: 140 fs)

2. Optical parametric oscillator (OPO) (APE, Berlin, Germany) (optional, excitation source for reporter fluorescence excitation in the red and far-red emission range, wavelength range 1,050 nm–1,300 nm; repetition rate: 80 MHz; pulse width at output: 200 fs)

2.6.2 Detection system

1. Dichroic mirror (775 nm, Chroma, Marlborough, MA, USA)

2. Interference filter (466 ± 30 nm) (optional, for NAD(P)H fluorescence detection)

3. GaAsP hybrid PMT (e.g. R10467U-40 Hamamatsu, Herrsching, Germany; quantum efficiency (QE) 45 % at 550 nm, QE > 30 % in the range 390 nm–680 nm) for TCSPC detection

4. TCSPC electronics (in this protocol, two-channel TCSPC, LaVision BioTec; two time-to-digital converters, each 20 MHz continuously max counting rate, 186 MHz photon burst, dead time 5.5 ns, time bin 27 ps, jitter <10 ps, temporal resolution 80 ps, limited by the ultrafast photodiode, dynamic 11 bit/event)

5. Ultrafast photodiode for generating the laser pulse trigger of the TCSPC electronics (e.g. DET08CL Thorlabs, response time 250 ps)

6. Photomultiplier tubes for imaging fluorescence emission of the reporter mice s.a. (e.g. H7422-40 or −50 (for detection >600 nm), Hamamatsu, Herrsching, Germany)

3.1.1 Surgical preparation

1. Conduct all procedures in a sterile environment.

2. Anesthetize mice with isoflurane using a vaporizer and a nose cone.

3. Apply ophthalmic ointment to prevent corneal drying.

4. Shave the surgical site using a sterile clipper.

5. Depilate the area with a depilatory cream.

6. Sterilize the skin with a suitable antiseptic solution, such as Braunoderm.

3.1.2 Surgical implantation of the microendoscope to the femur and post-operative care

1. Adhere to the surgical and implantation protocol detailed in Stefanowski and Fiedler et al. [2]. A schematic drawing of the GRIN lens and endoscope tubing with the glued in GRIN lens are shown in Figure 1 A and B, respectively. This protrusion of the GRIN lens was chosen to image the middle of the femoral bone marrow cavity (approx. 500 μm from the bone cortex). An illustration of the assembled endoscopic implant with the fixator plate, endoscope and reference plate and the fitting specimen holder that will be used to align the endoscope with the multiphoton-microscope objective are shown in Figure 1C and D, respectively.

2. NOTE: Animal experiment license in conformity with the regulations of your institution is required.

3. Administer immediate post-operative analgesia by injecting Buprenorphine: 0.03 mg/kg, subcutaneously

4. Provide sustained analgesia by adding Tramadol at a concentration of 0.05 mg/mL to drinking water (ad libitum).

5. Monitor mice daily for signs of pain, infection, or other complications.

6. Provide appropriate post-operative care, including adequate hydration, and environmental enrichment. The mice should show no impaired or abnormal movement from day 2 after surgery onwards. A schematic drawing of the mouse carrying the implanted micro-endoscope is shown in Figure 1E.

3.1.3 System calibration and performance assessment

1. We recommend to assess response of the time-resolved imaging system, i.e. instrument response function (IRF), as the measured fluorescence lifetime can be affected (mathematically convolved with the IRF) especially if the IRF is broad. To measure the IRF, we recommend using Potassium-dihydrogen-phosphate (KDP) crystals that provide a reliable second harmonics generation (SHG) signal, i.e. an instantaneous photophysical process, with no time delay. Since SHG doubles the frequency of the impinging light, we used 930 nm to detect the signal in the 466 ± 30 nm hPMT channel of our TCSPC. In our case, we determined a FWHM of (221 ± 7) ps from the Gaussian fit of our acquired IRF curve (Figure 2A and B). We didn’t find any dependence of the IRF on the location within the field of view of the GRIN system. Measuring the IRF in this way, accounts for the entire time-resolved imaging system including the laser, optical beam path, objective lens, GRIN lens, detector and detection and triggering electronics.

2. We also recommend to measure whether the position of noise is reflected by a quasi-infinite lifetime (at P(0|0)) within the phasor as a control (Figure 2A). Thereby, we assume that noise is an undampened oscillation. This can be measured by acquiring TCPSC stacks without any sample, thus only recording dark noise and possible ambient background.

3. As another system calibration step, we used NADH in solution (PBS) and acquired FLIM images to check whether our system accurately measures the fluorescence lifetime of unbound NADH known from literature (Figure 2C and D). Our results were in good agreement NADH of Δτ = 30 ps (τ _{NADH, exp} = 480 ps) τ 0.45 ns [28] state. With a measurement time bin of 55 ps, we found this difference to be negligible. This can also be done for other lifetime references, such as POPOP (in EtOH, expected lifetime of τ 1.35 ns [28]).

4. Lastly, we recommend ensure that no artefacts are induced by the system (we refer to this at the relevant step later in the protocol as well). Prominent artefacts are pile-up, which is the detection and consequent loss of a second photon within the time window of a single excitation pulse, and after-pulsing, the delayed generation of a spurious signal by the detector after detecting a true photon. While these effects also depend on the chosen hardware (technical properties of the detector and electronics), both these effects can be mitigated in most cases by temperature control of the detector and adapting the detector gain or laser power.

3.2.1 System setup

1. Configure the laser-scanning microscope with the appropriate objective lens.

2. Turn on and preheat the heating pad to maintain body temperature of the mouse during imaging.

3. Power on all necessary electronic devices and lasers, including, Ti:Sa laser, OPO laser, microscopy stages and electronics, TCSPC system, ultrafast photodiode and computer and launch the image acquisition software (e.g., ImSpector Pro 208).

4. Configure the software to simultaneously acquire the TCSPC imaging channel (for endogenous NAD(P)H- or exogenous fluorescent protein signals, e.g. FRET-based Ca²⁺-sensor such as TN-XXL (9)) and PMT imaging channels (for reporter fluorescence and second harmonics generation (SHG)), for co-registered TCSPC/PMT acquisition [10], [29].

5. Set the Ti:Sa laser wavelength to 760 nm for NAD(P)H-FLIM or the desired fluorophore excitation.

6. Set the OPO laser wavelength to the desired excitation wavelength to detect fluorescence of the reporter mouse.

7. NOTE: Ti:Sa and OPO beams are overlapping under the objective lens, allowing excitation of various fluorophores at the same time.

8. Configure save settings and the desired mode of operation (e.g., 3D image acquisition, time-lapse acquisition of 2D or 3D data).

9. NOTE: 2D TCSPC images are time-resolved and, thus, 3D stacks of e.g. 227 images (@ time bin = 55 ps, acquired over 12.5 ns imposed by the repetition rate of the excitation laser, 80 MHz).

10. Before in vivo imaging, check if stable system performance is granted by using a well-characterized reference slide (e.g., Convallaria majalis fluorescent slide).

11. Verify laser powers, detector gains, and TCSPC performance to match previous measurements and experiences.

12. Ensure that the TCSPC system accurately detects fluorescence decay curves without artifacts like after-pulsing or pile-up effects [30] (s. System calibration section). Adjust the gain accordingly.

3.2.2 Animal preparation

1. Anesthetize the mouse with isoflurane and maintain anesthesia throughout the imaging procedure.

2. Place the mouse on a heating pad to maintain a body temperature of 37 °C and apply ophthalmic ointment to keep the eyes from drying out throughout anesthesia.

3. Expose the GRIN endoscope top surface area by removing the plug screw of the reference plate.

4. Position the mouse on the heating pad and onto the specimen holder base plate.

5. Slide the reference plate into the funnel adapter and mount the adapter to the IVM table/custom specimen holder by carefully sliding the funnel into the opening of the specimen holder arm from below. The readily prepared mouse and adapter mounted into the specimen holder is shown in Figure 1 D.

6. Clean the imaging surface of the GRIN lens with a lint-free swab.

7. NOTE: When the reference plate is mounted and fixed to the specimen holder, the system is rigid, so that the top and bottom surface of the GRIN lens can be carefully wiped clean. For cleaning use ethanol or ideally an optical cleaning solution (n-hexane and isopropanol).

8. Position the specimen holder with the mouse under the microscope objective as schematically shown in Figure 1 F.

9. Focus on the top surface of the GRIN through the oculars.

3.2.3 Image acquisition

1. Adjust the initial focus position by elevating the focal plane to 200 µm above the GRIN lens surface, either in the acquisition software or at the microscope stage.

2. Initiate live view and gradually lower the focus using low laser power (a few percent, using e.g. 1,100 nm wavelength) until the GRIN surface becomes visible in the PMT detectors. Set this position as “zero”

3. NOTE: This step will be helpful to not become disoriented and accidentally focus on the GRIN lens surface at higher laser powers, potentially damaging the anti-reflex coating.

4. Center the scanning area based on the GRIN lens circular edge.

3.2.4 Acquisition settings

1. Set the following acquisition parameters to a field of view of 500 × 500 pixels (600 × 600 µm) to cover the full diameter of the GRIN lens (500 µm). This position/scan area will not be changed for the entire imaging session.

2. Set the line scanning rate (in the range 400 Hz–1,000 Hz or adapt this to the galvoscanner of your microscope). Sum-up several frames to increase collected photon numbers, e.g. 4× line average with unidirectional scanning mode.

3. Set the z-step size to 5 µm (slightly less than the axial resolution of the GRIN system used, NA = 0.5).

4. For time-lapse imaging, set the time interval and acquisition time window to observe cellular dynamics of interest.

5. NOTE: To accumulate sufficient number of photons of dim endogenous fluorescence signals, as it is the case in NAD(P)H-FLIM, set up the system so that 10–15 sequential images are acquired. This will result in pixel dwell times of 150 µs–300 µsÿ Adjust the time interval for the time-lapse acquisition larger than the total frame time (38 s–75 s).

3.2.5 Fluorescence imaging for cell-specific fluorescence through the GRIN endoscope

1. Excite reporter fluorescence, e.g. in LysM:tdRFP mice, tdRFP at 1,100 nm (OPO).

2. Detect emitted light with photomultiplier tubes e.g. in the green channel (525 ± 20 nm) and red channel (593 ± 20 nm).

3. Adjust PMT gain to optimize brightness and contrast for your scenario. Here, we showcase images for visualization of myeloid LysM⁺ cells (shown in Figure 3 A).

3.2.6 Fluorescence lifetime imaging through the GRIN endoscope

1. In parallel to OPO excitation, excite NAD(P)H at the optimal wavelength (e.g., 760 nm for NAD(P)H).

2. NOTE: Alternatively, you can choose another fluorophore to perform FLIM and adjust the excitation wavelength accordingly. To demonstrate this, we acquired fluorescence intensity images of flavin adenine dinucleotide (FAD) (Figure 3G). FAD similarly to NAD(P)H a metabolic cofactor necessary for numerous metabolic redox reactions and can provide additional information about cells metabolic redox state, calculated from the fractions I_FAD/(I_NAD(P)H + I_FAD) of the FAD and NAD(P)H emission signals [31], [32]. We show exemplary images of these images together with the optical redox ratio in Figure 3G.

3. Adjust the GaAsP hybrid PMT (or equivalent FLIM detector) gain to generate fluorescence decay curves over the entire image that show a clean exponential-like decay. An example of NADH fluorescence decay curves in NADH solution, with and without GRIN endoscope) is shown in Figure 2 A.

4. Check the fluorescence decay for a small area within the image or even single pixels. Gradually increase laser power in small increments (few percent at a time), to obtain sufficient signal per pixel (typically, maximum 200–300 photons/pixel integrated over the pixel decay curve or 10–20 photons/pixel at its peak are required for reliable FLIM data analysis). The indicated pixel dwell times (150–300 µs) ensure collecting this number of NAD(P)H fluorescence photons through the used GRIN lens system, both in solution as well as in the femoral marrow, however, higher laser excitation powers are needed in vivo.

5. Maintain an average maximum laser power of 25 % (in our case corresponding to 55 mW under the GRIN endoscope at 760 nm, Ti:Sa laser) to avoid tissue photo damage.

6. NOTE: Since the laser powers can deviate depending on microscope setup and laser model, it is helpful to determine the damage threshold experimentally in your system, i.e. (1) by measuring the transmitted laser power under the GRIN endoscope and (2) by performing immunofluorescence histology of heat-shock protein in the exposed tissue.

7. Acquire 15 sequential images per z-plane or time-point to ensure good photon statistics, especially for NAD(P)H-FLIM (as set up in step 24).

8. NOTE: This step is crucial to gather sufficient photons for reliable FLIM analysis. For NAD(P)H endogenous fluorescence in the bone marrow, we collect 4·10⁶–80·10⁶ photons over the entire circular FOV (125,665 pixel). In single cells of 10–30 µm diameter (area of 55–490 pixel), we collect at least 750 photons. For brighter (exogenous) fluorophores, less than 15 consecutive images may be needed to collect the specified number of photons.

9. Acquire 3D stacks (and/or time-lapse images) of cell-specific fluorescence and time-resolved fluorescence for FLIM (as set up in step 24) in a co-registered manner.

3.2.7 FLIM data pre-processing

1. Sort the acquired data, bundling the 15 sequential FLIM images for subsequent summing.

2. Sum these FLIM images (e.g. of the same z-plane or time-point). This can be automated, e.g. by using the FIJI macro recorder function, modified for your folder structure.

3. Crop the images to remove the GRIN lens edge and outside areas. An example of a NAD(P)H-FLIM intensity projection of the unsegmented bone marrow is shown in Figure 3 A.

4. NOTE: Acquired signals outside of the GRIN border might appear due to reflections at the tubing surface; however, as these are not originating from biological structures within the bone marrow, they should be excluded.

3.2.8 Cell type specific fluorescence image pre-processing

1. Pre-process the PMT (cell-specific fluorescence) image stacks using FIJI by applying the “Despeckle” filter to reduce noise and afterwards the “White Top-Hat” filter to homogenize the background.

3.2.9 Analysis of co-registered cell type specific fluorescence and NAD(P)H-FLIM data

1. For the cell segmentation of the specific reporter fluorescence, use the Labkit plugin in FIJI. Labkit allows you to train a random-forest algorithm to distinguish two or more classes. In this case, train the algorithm to distinguish between background and signal pixels in the cell-specific fluorescence images, e.g. tdRFP in LysM⁺ myeloid cells.

2. NOTE: Segmentation of summed-up NAD(P)H FLIM images is not recommended, as all cells in tissue contain NADH and NADPH. No differentiation between cells is thus possible.

3. Save the classification results as binary masks.

4. For single-cell analysis ensure that segmented cells are well separated (by visual inspection). If necessary, use the 3D Watershed function in FIJI to further separate overlapping cells. Employ the 3D Object Counter function in FIJI to generate masks for individual 3D cell objects, i.e. to segment individual cells. Exemplary fluorescence signal and segmentation outlines of a single cross section are shown in Figure 3 B.

5. For the FLIM image analysis we propose as standardized quality control by using the signal-to-noise ratio (SNR) metric to ensure that the results are based on sufficient photon counts and are therefore warranted by the data. (i) Calculate SNR maps using the formula: SNR = (µ_Signal–µ_BG)/σ_BG and applying this to your summed-up FLIM image by using the build-in FIJI “macro” function (“process → math → macro”). µ_Signal is the signal originating from the time-resolved fluorescence (here NAD(P)H fluorescence), µ_BG and σ_BG are the mean value and standard deviation of background histogram, e.g. dark image of the microscope.

6. NOTE: The background image, i.e. dark image, is acquired if the detector (TCSPC detector) is covered and therefore measures only the dark noise (of electronics).

7. Apply a threshold on the SNR maps of 10, create binary masks and apply them to your summed FLIM images to filter out low-SNR pixels. An example of a SNR-map and of corresponding map with SNR<10 pixels highlighted red is shown in Figure 2 B. To compare the influence of SNR-thresholding we show examples of the FLIM analysis results–lifetime maps, general metabolic activity and enzyme specific activity–with and without SNR-thresholding in Figure 2 D and E and the phasor plots and fluorescence decay curves in Figure 2 F, G and H, respectively.

8. NOTE: The accurate position and fluorescence lifetime calculation from experimental FLIM data using any evaluation method, including the phasor approach, is influenced by noise. If the SNR is too low (i.e. below 10) the length of the phase vector (phasor) will be shortened due to the theoretical infinite lifetime of background noise, i.e. an undampened overlap of oscillating waves. The vector length will be extended and reach a maximum as the SNR increases (an increase of SNR 10 does not significantly increase the vector length, as the vector length over SNR behaves asymptotic [33].

9. Optional: Compare the SNR metric with the collected photon counts to assess signal quality and time resolution [34].

10. NOTE: The absolute photon count is directly red-out from the TCSPC detector. The SNR, as any ratio, can be distorted, analogous to a small sample bias, if the absolute photon counts are low (for example a SNR of 10 could be 50 photons collected, which is too low to get any valid result, if the average background and noise amount to five photons).

11. NAD(P)H-FLIM images need to be masked by the segmentation of the cell-specific fluorescence, in order to analyze metabolic profiles at single cell level. For this the reporter fluorescence images and FLIM images need to match geometrically, i.e. co-registered.

12. NOTE: In rare cases, if the detection channels are misaligned, x and y positions of the cell-specific fluorescence and FLIM images need to be matched. Use the lens edges in both 3D images stacks to evaluate the xy-shift and find the correct position. The images can be translationally shifted by using the built-in FIJI “Transform → Translate” function. Geometric mismatches in x and y might happen, when co-registering images at different excitation and detection wavelengths. Even though this should affect only the z-dimension – i.e. chromatic aberration – xy-shifts might appear due to a tilted GRIN lens within the tubing, e.g. during lens gluing or drying process. This can be characterized using reference samples (e.g. fluorescent beads) and acquiring images through the GRIN endoscope at 1,100 nm and 760 nm, respectively, i.e. wavelengths used to excite (i) the cell-specific reporter fluorescence (tdRFP) and (ii) NAD(P)H fluorescence (for FLIM).

13. Determine the geometric mismatch in axially, i.e. in z direction. For GRIN microendoscope used in this work, we determined a chromatic aberration of 1.7 µm/10 nm excitation wavelength shift. With this formula the focal shift can be calculated substituting Δλ for the difference of the wavelength you used for exciting your cell-specific fluorescence in the reporter mouse and time-resolved fluorescence (FLIM). For tdRFP and NAD(P)H co-registration, this means 1,100–760 = 340 nm, leading to a focal shift of 58 µm.

14. NOTE: The wavelength-dependent geometric focal mismatch in z (chromatic aberration) can be determined by acquiring images of a reference sample (e.g. fluorescent beads in agarose gel) at different wavelengths over the spectral range of the lasers you plan to use.

15. Based on the result in 46, geometrically match cell-specific fluorescence and FLIM images along the axial (z) direction. Therefore, crop the 3D FLIM and fluorescence stacks. For tdRFP fluorescence and NAD(P)H-FLIM, the focal shift of 55 µm is corrected by cropping the first 11 z-slices of the NAD(P)H-FLIM stack.

16. Apply the binary masks of the segmentation to the summed-up, geometrically matched FLIM images. This step can be done with either the collective cell masks or the 3D object segmentation masks, in case of single-cell analysis.

3.2.10 Phasor analysis of NAD(P)H-FLIM data and statistical analysis

1. Apply the model-free phasor approach to analyze FLIM images processed as described. Use a phasor algorithm available on GitHub. For FLIM data acquired with the LaVision Biotec TCSPC, you can use the customized phasor algorithm developed in our lab (https://github.com/RLeben/phasorFLIM_enzymeAllocation.git).

2. NOTE: The algorithm transforms time-domain to phase-domain data using discrete Fourier transform and calculates the real and imaginary components of the normalized phase vector. Afterwards it displays the distribution of the phasors as a 3D heatmap histogram and reconstructs average fluorescence lifetime maps, general metabolic activity maps, and specific activity maps for 13 NAD(P)H-binding enzymes, we identified as most abundant from bulk mRNA data or particularly relevant in a subcellular compartment [33].

3. NOTE: You may use other algorithms as well, such as approximation algorithms based on gradient or genetic iterations, however, these typically require a model for the fluorescence decay.

4. NOTE: Statistical analysis of cell-specific metabolic profiles can be done by plotting the histograms of the metabolic enzyme activity maps. Alternatively, they can be visualized and interpreted by single-cell clustering algorithms, e.g. using machine-learning approaches.

Final NOTE: The implant-based microendoscope allows longitudinal imaging. Repeat section “Imaging Experiment” steps 1. – 49. on consecutive days, weeks or even months to investigate cells in the bone marrow longitudinally.