Standardized measurements for monitoring and comparing multiphoton microscope systems

Abstract

The goal of this protocol is to improve the characterization and performance standardization of multiphoton microscopy hardware across a large user base. We purposefully focus on hardware and only briefly touch on software and data analysis routines where relevant. Here we cover the measurement and quantification of laser power, pulse width optimization, field of view, resolution and photomultiplier tube performance. The intended audience is scientists with little expertise in optics who either build or use multiphoton microscopes in their laboratories. They can use our procedures to test whether their multiphoton microscope performs well and produces consistent data over the lifetime of their system. Individual procedures are designed to take 1-2 h to complete without the use of expensive equipment. The procedures listed here help standardize the microscopes and facilitate the reproducibility of data across setups.

Fig. 1. Basic schematic of a resonant-scanning multiphoton microscope.

An ultrafast, pulsed laser (red) has its power modulated by the Pockels cell, after which the laser is relayed through the scanning system (lenses and mirrors) to the sample, through the objective. The sample undergoes fluorescence excitation and emission, and the emitted light (green) is separated from the original beam using a dichroic mirror to be directed at the detection system (in this case a photomultiplier tube; PMT).

Fig. 2. Object distortion caused by RI mismatching and the correction factor.

a, Scanning the objective lens axially (indicated by the black arrow, Δobjective) through a spherical object (labeled as the ‘actual object’ in purple, Δfocus) submerged in water produces a three-dimensional image. Despite this, the image (labeled as the ‘apparent object’ in green) appears compressed along the axial direction. This compression arises due to the air-to-water interface, where the RI transitions from air (RI_IMMERSION of 1) to water (RI_SAMPLE of 1.33), causing light rays (purple rays) to bend. Consequently, the movement of the objective (Δobjective) is smaller than that of the actual focal plane (Δfocus). The resulting rendering of the sphere (‘apparent object’, green) exhibits axial compression, as the acquisition software assigns the objective’s travel distance (Δobjective) rather than the focal plane’s travel distance (Δfocus) to the object’s z axis. A correction factor can be calculated and applied to correct this distortion. b, A plot shows the correction factor converting the movement of the objective (Δobjective) or the z-stage (Δstage) to that of the actual imaging plane (Δfocus) as a function of NA of the objective. RI_IMMERSION = n₁ = 1 and RI_SAMPLE = n₂ = 1.33. The figure is adapted from ref. 34, Springer Nature Ltd.

Fig. 3. Laser power measurement.

a, The path of a resonant-scanned focused laser beam over a sample. The beam moves sinusoidally along the fast axis (X in this case) whilst being scanned in orthogonal Y direction with a linear galvo. The area over which the beam moves is known as the scan field. On the left and right edges, the beam slows as the scanner changes direction and turns around. In these areas the potential for photodamage is greatest, as the beam is traveling more slowly over the sample. Thus, the beam is typically ‘blanked’ or disabled during these epochs. In a resonant scanning microscope, the beam is usually blanked ~30% of the time. The image field (red lines and gray region) is the area over which the beam is on and capable of exciting fluorescence. The dotted lines indicate the blanked turn-around regions. A power meter cannot distinguish these states and so returns a time-averaged power value over the whole scan field if the microscope is scanning during a measurement. b, Position of the power meter head with respect to the laser beam exiting the objective. The sensor surface should be close but must not be at the working distance of the objective lens, as the focused beam may damage the sensor surface leading to unreliable measurements and permanent damage to the sensor. c, An example laser power calibration curve. Output power can be represented as a percentage of total available laser power or in direct power units (mW), depending on microscope configuration. The purpose is to create a lookup table that allows linear adjustment of power on the edges of the modulation range for modulation devices (such as Pockels cells) that have nonlinear response.

Fig. 4. FOV size measurement.

a, Two-photon image of a 1 mm grid with 100 µm divisions for the maximum scan angle of the microscope system (that is, the minimum zoom). Here, the FOV size is ~800 µm. b, An image of a 25 µm copper EM grid. This image shows the ~1,200 µm FOV and displays minor pincushion distortion at the left and right edges (note how the vertical line on the right and left sides bows in from the vertical line defined by the edge of the image). c, Grid lines are detected and overlaid on top of the EM grid image using the ‘MicsPerPixel’ software tool.

Fig. 5. FOV size comparison for two-photon and epifluorescent modes on a large FOV microscope (Mesoscope).

a, Tiled two-photon image of the Mesoscope FOV (~5,000 µm). b, Epifluorescent image of the Mesoscope FOV showing a ~45° rotation and vertical reflection, compared with the two-photon image.

Fig. 6. FOV homogeneity.

a, Homogeneity calibration image of the uniform fluorescent slide acquired using a 16× objective at maximum scan angle on a system that allows for large scan angles. Homogeneity dropoff profiles and different zoom factor overlays are shown (different systems may have different scan areas for the corresponding ‘zoom’ factor). The area of peak brightness is offset downward slightly in the y axis. This indicates a possible misalignment in the optics. The dark spots arise from imperfections in the slide surface. b, Intensity profiles along the diagonals (cyan and red) demonstrate nonuniformity of excitation at the maximum zoom factor.

Fig. 7. Two-photon FWHM as a function of NA and wavelengths.

The formula is adapted from Zipfel et al.. NA is the numerical aperture of the objective lens. In FWHMz, n is the RI of the medium where the sample is embedded and set as water in this plot. The RI of the water, n, is wavelength dependent and is [1.328, 1.3255, 1.3225] at [0.9, 1.1, 1.3] µm, respectively.

Fig. 8. Example measurement of PSFs.

a,b, Fluorescent beads (0.2 µm) were embedded in 0.75% (wt/vol) agarose gel; 40 µm z-stacks were acquired at the depth of 500 µm, and beads at the center (a) and the edge (b) of the FOV were measured. The example images are shown from the XY, XZ and YZ cross sections, respectively. The intensity profiles of the beads (red dashed lines) in the X direction on the XY plane and in the Z direction are plotted, which are fitted to a Gaussian curve (orange dashed line) to extract the radial-X and axial FWHM of the PSF (note the PSF degradation for the lateral position of the bead).

Fig. 9. Example image pixel histogram.

The pixel intensity value distribution is shown for one image frame while live scanning. The spread of this distribution is used to optimize laser pulse width.

Fig. 10. Pulse width optimization measurement.

The top panel shows a plot of mean image intensity versus GDD compensation value for a fixed average laser power. For this system, ~-12,000 fs² of compensation is necessary for the microscope to achieve highest excitation efficiency. The bottom panels show example images acquired of the different settings of GDD compensator and projected intensity plots that can be used for analysis described in the procedure.

Fig. 11. A hand-made tritium light source.

a, A 3 mm × 11 mm tritium vial next to a 5 cent coin. b, The assembled tritium light source. The pinhole is at the top and will be placed immediately beneath the objective to test the entire collection system or PMT window to directly test the PMT. c, The multicolor assortment of tritium capsules, each with a different color phosphor (shown without a pinhole for clarity). SM1A6 Thorlabs parts were used here for a specific setup, which is not part of this procedure.

Fig. 12. First-day performance for three multialkali PMT units of the same model.

a–d, Mean pixel value (a), SNR (b), ROC–AUC (c) and mean anode current (d) data shown for three Hamamatsu R10699 PMTs on the first day of installation. PMTs were tested within the full collection optics system (green channel) of the same microscope. Gain setting is represented by the control signal voltage.

Fig. 13. Change in PMT performance over time.

a–d, Mean pixel value (a), SNR (b), ROC–AUC (c) and mean anode current (d) collected for a Hamamatsu R10699 PMT unit when it was first installed and after 1.5 years of routine use. The dashed line shows pixel values expected based upon tritium decay alone. The gain setting is represented by the control signal voltage.

Fig. 14. Pixel grayscale value distributions and ROC–AUC at a series of gain settings for an example multialkali PMT.

For each gain setting, distributions of pixel (PX) values from dark response (black) and light response (red) images are shown. Corresponding ROC–AUC values are indicated at the top of each panel. The gain expressed as control signal voltage (mV).

Fig. 15. Comparison of two different GaAsP PMTs of the same model tritium light sources (red and green) and control (no light source).

The PMTs (Hamamatsu H10770PB-40). Mean (row 1), standard deviation (row 2), were measured with bandpass filters in place (PMT1: 570–620 nm bandpass; ROC–AUC (row 3) and SNR (row 4) for two example GaAsP PMTs for two different PMT2: 500–550 nm bandpass).

Fig. 16. Photon transfer analysis.

a, The average image of a 500-frame two-photon calcium imaging sequence in mouse visual cortex recorded at 8 frames per second. b, The PTC computed from the same sequence. It features a long linear portion corresponding to Poissonian noise dominating the frame-to-frame variance in all but the brightest regions. The slope of the robust linear fit (black line) reveals the photon sensitivity of 96.9 grayscale levels per photon. Note that the density of intensity values follows a long-tail distribution. The variance in bright regions of the image grows faster than predicted by the linear fit, reflecting the presence of physiological signals. The static images lack such deviations. c, The coefficient of variation (CoV) image reveals areas of higher variance than predicted from quantum noise alone. The calcium activity in cells produces a higher CoV, shading them green. d, Cell segmentation based on the maximum projection image; eight cells are delineated. e, The maximum photon flux per pixel expressed in the units of photons per pixel per frame. f, The fluorescence traces from the labeled cells expressed as photons per second. Scale bar, 10⁴ photons per second per cell.

Fig. 17. A different image sequence from another source.

a, The PTC indicates that most pixels do not see a photon in each frame. The PTC has a non-Poissonian segment where no photons are detected. b, The coefficient of variation image reveals no deviations from predicted variance. c, The maximum photon flux is substantially lower than in our first dataset, due to the finer pixel pitch. However, the number of pixels per cell is about four times larger. d, After ROI averaging, fluorescence signals produced comparable peak amplitudes with the first data set.

Fig. 18. Pockels cell resonance effect.

a, The image of a homogeneous fluorescent medium with no ringing effect. b, Same image as a but with Pockels cell ringing visible on the left side. c, A line profile along the yellow line in a that only shows a drop at the dark spot along the line. d, A line profile along yellow line in c that shows intensity oscillations on the left side of the image where Pockels cell ringing is present. This effect does not change with the zoom factor. The line profile is chosen to extend over a darker spot to highlight the magnitude of the ringing.

Fig. 19. The effect of dielectric coatings on GDD.

Optimal GDD was estimated using a fluorescent slide and the built-in GDD compensation on a Coherent Chameleon Vision II laser. The compensation curve was first measured with three dielectric coated mirrors (ThorLabs EO3) in the path (black line). Obvious sharp peakiness is seen in the required compensation value as a function of wavelength. After these three mirrors were swapped with metallic coated mirrors, the peakiness completely disappears (red line). The three EO3 mirrors did not contribute equally to the above effect (data not shown).