In vivo imaging of cerebral energy metabolism with two-photon fluorescence lifetime microscopy of NADH.

2.1. Customized imaging system Figure 1a Fig. 1 (a) 2-photon imaging portion of our custom built in vivo imaging system, modified for FLIM measurements. SH: shutter, M: reflecting mirror, P: polarizer, EOM: electro-optic modulator, XY: galvanometer-based scanners, DM: Dichroic Mirror, PCM: photon counting module, PMT: photomultiplier tube. (a) Inset: reflectance image of the sealed cranial window model, approximately 3 mm in diameter at 570 nm. (b) In vivo image of NADH autofluorescence in a cortical astrocyte (Scale bar: 10 µm), with (c) pixel-wide distribution of 2-component fit performed with commercial SPCImage software, clearly showing 4 distinct peaks. (d) Example 4-component lifetime fit of in vivo cortical NADH fluorescence decay, computed with custom software. Blue profile: Collected, binned photon counts, Black-dashed lines: fitting boundaries, Green profile: computed IRF, Red profile: Computed decay profile (e) residuals for computed 2-, 3-, and 4-component lifetime fits to the example in vivo lifetime data, with reduced χ 2 errors of 195.16, 2.64, and 2.54, respectively. depicts the 2P FLIM portion of our multimodal microscope [ 24 , 25 ], custom-designed for in vivo investigations of cerebral metabolism in small animals. Excitation is provided by a tunable Ti:sapphire laser (Mai Tai, Spectra Physics), delivering ~360 fs pulses after the objective at 80 MHz and tuned to λ = 740 nm. The excitation intensity is modulated with an electro-optic modulator (ConOptics, Danbury CT), raster-scanned across a field of view (FOV) ranging between 75 and 300 µm using galvanometer-based scanner mirrors (Cambridge Technology, Inc), and focused on the sample by a water immersion objective (Olympus XLumPlan Fluor, 20X, 1.00 NA, 2 mm working distance). Using a series of dichroic and bandpass filters, emitted light is spectrally separated and detected by a custom-designed four-channel detector array with high collection efficiency using photomultiplier tubes (PMTs) (Channel 1: 680 ± 30 nm, Channel 2: 595 ± 25 nm, Channel 3: 525 ± 25 nm, Channel 4: 460 ± 30 nm). NADH fluorescence is collected in channel 4 by a novel hybrid PMT with high detection efficiency and minimal afterpulsing (HPM-100-40 Becker & Hickl GmbH). As detailed in Subsection 2.4, fluorescence signal of the exogenous dye sulfurhodamine 101 (SR101) was detected in channel 2 simultaneously with the NADH collection. The SR101 intensity was used to co-register the NADH image with cellular structures and to correct for absorption changes associated with altered cerebral blood flow during physiological manipulations. For FLIM investigations described herein, scanning and data collection were performed using commercial time-correlated single photon counting (TCSPC) hardware (SPC-150, GVD-120, DCC-100, Becker & Hickl GmbH) and control software (SPCM, Becker & Hickl GmbH). NADH autofluorescence is spectrally identical to that of nicotinamide adenine dinucleotide phosphate (NADPH). In principle, the detected autofluorescence could arise from a mix of both NADH and NADPH. In metabolically active brain tissue, however, NADPH is widely believed to contribute minimally to the fluorescence signal, due to its much lower concentration in brain tissue [ 26 , 27 ], low quantum yield [ 28 ], and insensitivity to metabolic perturbation [ 13 , 19 , 29 ]. 2.2.

Imaging protocol

High resolution (256 x 256 pixels) images were first acquired under baseline physiological conditions using Δ t ≈50 ps binning intervals and 256 temporal channels (12.8 ns-long decay profiles) over a field of view (FOV) ranging from 75 to 200 µm. Excitation intensity was adjusted to yield photon count rates of approximately 500,000 counts per second for these measurements, as indicated by the count rates continuously displayed in the SPCM software. Although incident power varied with cortical depth, it remained well below 50 mW for all measurements. For a given baseline measurement, such as the one provided in Fig. 1b , a repetitive raster scan was performed over the FOV at frame intervals of up to ~900 ms for 120 s. The FLIM system accumulates photons over all frames into time-resolved photon distributions, one distribution for each pixel, yielding a 256 x 256 x 256 data cube resolved in x , y , and t with high signal to noise ratio. We performed an additional 3x3 spatial pixel binning to obtain decay profiles with ~5000 photons. We regard this value as the approximate minimum number of photons required to perform our multiple-component lifetime fits, containing 2 constrained lifetimes and 2 free lifetimes and described in Subsections 2.3 and 3.2, at each pixel [ 30 ]. After collecting physiological baseline measurements, a time series of measurements over the same FOV was collected with 32 x 32 pixel resolution. Spatial resolution was sacrificed to decrease the time-point measurement interval for the time series to 2.5 s, enabling time-resolved FLIM measurements during metabolic manipulations. For each time-point measurement, the FOV was repeatedly raster scanned for 2.5 s. Each individual frame required ~50 ms. The software accumulated photons over all ~50 frames and temporally binned using the same Δ t ≈ 50 ps intervals. Each time series consisted of continuous time-point measurements acquired for 5 minutes. To minimize photobleaching, excitation intensity was reduced to yield photon count rates of approximately 50,000 counts per second for these time-resolved measurements. Fluorescence decay measurements were added over all pixels in the field of view. For a given time-point measurement, decay profiles were added with their two neighboring time points, yielding temporally smooth transient profiles. Each decay profile for each time point was comprised of 400,000 recorded photons. The total greatly exceeds the necessary amount for our fitting procedure, yielding more robust and temporally smooth profiles. These experiments constitute our global characterization of cerebral tissue’s response to anoxia. Subsequent investigations will feature modified experimental protocols with pointwise measurements, enabling both high spatial and temporal resolution for characterizing more localized, cell-specific changes in cerebral metabolism. 2.3. Data analysis: instrument response function and lifetime fitting As an initial step for each experiment, the lifetime calculation accuracy was verified, and simultaneously, the instrument response function (IRF) was computed, using FLIM measurements of dissolved NADH (0.5 mM in saline).

Early attempts to characterize the IRF using second harmonic generation

(SHG) of collagen and sucrose were confounded by mismatch in spectral properties [ 17 , 31 ]. The IRF in a two-photon microscope can be recorded by using SHG or Hyper-Rayleigh scattering [ 32 ]. Both methods deliver ultrafast signals at a wavelength detectable by the FLIM system. However, in contrast to fluorescence, most of the light is emitted in forward direction. The use of SHG may also require a change in laser wavelength. This can cause a shift in amplitude and phase of the timing reference pulses and thus a shift in the measured IRF. As NADH lifetime decay is well characterized [ 19 , 21 ] and its spectral features well match cytosolic and enzymatic bound NADH in vivo , NADH solution serves as a suitable reference emitter for evaluating these system characteristics under experimental conditions identical to our in vivo baseline measurements. The IRF was computed using an iterative procedure. No a priori assumptions were made for the NADH decay parameters. For the first iteration, the decay portion of the experimental NADH solution measurement (from the peak time to the final time point, t final ) was fit as the sum of two decaying exponentials and a constant offset. A preliminary IRF was then computed by the expression I R F ( t ) = F F T − 1 ( F F T ( I m e a s u r e d ( t ) ) F F T ( I t h e o r e t i c a l ( t ) ) ) , (1) where FFT and FFT −1 denote the Fast Fourier Transform and inverse Fast Fourier Transform, respectively. For each subsequent iteration the entire experimental decay curve of measured NADH solution (from t = 0 to t final ) and the computed IRF ( t ) from the previous iteration were used to compute coefficients for a new theoretical decay. The new theoretical profile was then used in Eq. (1) to re-compute IRF ( t ). The procedure continued for 5 iterations. The decay coefficients yielded consistent values after the second iteration. The coefficients and IRF ( t ) from the iteration yielding the smallest fitting error (reduced χ 2 ) were chosen. The fluorescence decays measured from in vivo brain tissue were fit with the expressions I t h e o r e t i c a l ( t ) = I o f f s e t + ∑ i = 1 N α i exp ( − t τ i ) , (2) I m o d e l ( t ) = I R F ( t ) ⊗ ( I 0 δ ( t ) + I t h e o r e t i c a l ( t ) ) , (3) where N = 4, as described in Subsection 3.2, α i and τ i represent the amplitude and lifetime of component I, and I offset represents the constant baseline intensity resultant from detector noise or ambient room light. I 0 Δ( t ) models the ultrafast signal component observed in the detected response. This component generally constitutes a miniscule amount of the detected response and its origin is not well understood. In our experiments, the ultrafast component likely arises from backscattered excitation light [ 19 ]. Generally, negligible amounts of excitation light penetrate through the emission filters and reach the detectors. However, in highly scattering tissue such as the brain, the contribution of the ultrashort excitation pulse should be accounted for and can be modeled as a Dirac delta function [ 19 , 33 ].

2.2.

Imaging protocol

High resolution (256 x 256 pixels) images were first acquired under baseline physiological conditions using Δ t ≈50 ps binning intervals and 256 temporal channels (12.8 ns-long decay profiles) over a field of view (FOV) ranging from 75 to 200 µm. Excitation intensity was adjusted to yield photon count rates of approximately 500,000 counts per second for these measurements, as indicated by the count rates continuously displayed in the SPCM software. Although incident power varied with cortical depth, it remained well below 50 mW for all measurements. For a given baseline measurement, such as the one provided in Fig. 1b , a repetitive raster scan was performed over the FOV at frame intervals of up to ~900 ms for 120 s. The FLIM system accumulates photons over all frames into time-resolved photon distributions, one distribution for each pixel, yielding a 256 x 256 x 256 data cube resolved in x , y , and t with high signal to noise ratio. We performed an additional 3x3 spatial pixel binning to obtain decay profiles with ~5000 photons. We regard this value as the approximate minimum number of photons required to perform our multiple-component lifetime fits, containing 2 constrained lifetimes and 2 free lifetimes and described in Subsections 2.3 and 3.2, at each pixel [ 30 ]. After collecting physiological baseline measurements, a time series of measurements over the same FOV was collected with 32 x 32 pixel resolution. Spatial resolution was sacrificed to decrease the time-point measurement interval for the time series to 2.5 s, enabling time-resolved FLIM measurements during metabolic manipulations. For each time-point measurement, the FOV was repeatedly raster scanned for 2.5 s. Each individual frame required ~50 ms. The software accumulated photons over all ~50 frames and temporally binned using the same Δ t ≈ 50 ps intervals. Each time series consisted of continuous time-point measurements acquired for 5 minutes. To minimize photobleaching, excitation intensity was reduced to yield photon count rates of approximately 50,000 counts per second for these time-resolved measurements. Fluorescence decay measurements were added over all pixels in the field of view. For a given time-point measurement, decay profiles were added with their two neighboring time points, yielding temporally smooth transient profiles. Each decay profile for each time point was comprised of 400,000 recorded photons. The total greatly exceeds the necessary amount for our fitting procedure, yielding more robust and temporally smooth profiles. These experiments constitute our global characterization of cerebral tissue’s response to anoxia. Subsequent investigations will feature modified experimental protocols with pointwise measurements, enabling both high spatial and temporal resolution for characterizing more localized, cell-specific changes in cerebral metabolism.