In this context, point-scanning confocal or multiphoton approaches are usually too slow, as the image is recoreded one pixel at a time (Fig. 1). Indeed, in these approaches signal level prescribes pixel accumulation times of typically 1–10 μs, corresponding to pixel rates of only 10⁵ to 10⁶ pixels.s⁻¹.

Several approaches have been explored during the last 15 years to improve the imaging speed of multiphoton microscopy up to ~10⁷ pixels.s⁻¹, including fast point-scanning and multifocal approaches (Fig. 1, Tables 1 and ​and2,2, and [21] for a review). However, besides hardware limitations (i.e. scanning speed, readout time, data transfer or storage) the pixel rate of any microscope is fundamentally limited by the signal level that can be obtained within the pixel accumulation time without causing fluorophore saturation or photodamage (including phototoxicity to the biological sample and photobleaching of the fluorophores). Hence, even though fast point-scanning can be implemented using resonant scanners, polygonal mirrors or acousto-optic deflectors [21], the useful pixel rate is still limited by fluorophore photophysics of the single-point excitation approach (third column in Table 1). The main strategy to circumvent this limitation is to parallelize the sample illumination and the signal detection. Using multifocal excitation (Fig. 1), overall pixel rate can be increased while maintaining the same illumination time per pixel (Table 1). However with this approach, an increase in imaging speed requires a proportional increase in laser average power (Table 1 and Table 2), similar to linear microscopy (Supp. Table 1). Available laser power therefore limits the achievable speed gain. Moreover, increasing the laser average power may eventually lead to linear absorption and photodamage, as it is the case in linear microscopy.

Among the strategies for improving the imaging speed of multiphoton microscopy, the recent implementation of scanned light-sheet microscopy using two-photon excitation (2p-SPIM in [16] and light-sheet 2p-microscopy in this review) introduces a new paradigm. In this technique, a sheet of light is generated by scanning a weakly focused Gaussian beam faster than the image acquisition time to illuminate an entire plane of the sample, which is then imaged with a camera oriented orthogonally to the sheet. Compared to a static light-sheet generated with a cylindrical lens [22], this scanned light-sheet approach generates typically 100 times stronger 2PEF signal [16], which is critical for live imaging. Light-sheet 2p-microscopy is the only technique improving overall pixel rate over point-scanning 2p-microscopy with longer pixel accumulation time and lower peak intensity (Table 1 and Table 2). This fundamental property results from the orthogonal geometry of the illumination and detection pathways (which are collinear in conventional microscopy), allowing the use of a low numerical aperture (NA) illumination focusing (resulting in a large illumination volume) without degrading the axial resolution and the overall signal rate [16]. The use of low-NA illumination has three important advantages for multiphoton live imaging. First, it results in lower peak intensity, and therefore less higher-order nonlinear photodamage to the tissue [23]. Second, parallelization of the illumination is done along the light propagation direction, reusing the same excitation energy, thus requiring less laser power than in multifocal approaches, in turn limiting linear absorption and photodamage. Finally, the weakly focused excitation beam is less sensitive to sample-induced optical aberrations and resolution loss with depth than in the case of high-NA focusing [16]. In addition, in the conditions presented in Table 3 [16], the laser is scanned ~15 times during the image acquisition with 1 ms between two passes. This temporal excitation pattern potentially results in lower photobleaching, as time is given for fluorophore dark state relaxation [24].

The depth performance of point-scanning 2p-microscopy relies on three phenomena: (i) superior penetration of illumination light, (ii) robust confinement of excitation volume, and (iii) efficient collection of the fluorescence. Let us compare this technique with confocal microscopy, light-sheet 2p-microscopy and light-sheet 1p-microscopy with respect to these three points (Fig. 3).

However, light-sheet microscopy does has one advantage other collinear techniques (confocal and point-scanning 2p-microscopy): the use of lower illumination NA leads to less sensitivity to optical aberrations and thus contributes to maintain better axial resolution in light-sheet 2p-microscopy imaging of inhomogeneous embryos [16].

In many cases such signals can be obtained from unstained tissues, with the additional benefit of not suffering from photobleaching. SHG is exclusively observed from dense organized non-centrosymmetric electronic structures. Some natural sources of SHG are fibrillar collagen, myofilaments, astroglial fibers, starch, and polarized tubulin assemblies such as mitotic spindles that can be observed in embryos (Fig. 2d-f). THG does not require molecular asymmetry but is observed only near optical heterogeneities. In practice, THG signals are obtained from dense non-aqueous objects such as lipid droplets [36], mineralized or absorbing structures, and generally from interfaces between media of different refractive indices. Coherent Raman processes such as CARS and SRS derive their contrast from molecular vibrational modes and can be used for micro-spectroscopy and chemically selective imaging. SRS provides increased contrast compared with CARS at the cost of increased experimental complexity. The most widespread use of coherent Raman microscopy for biological studies is currently the selective imaging of lipid distribution in tissues based on contrast from CH-bond vibration [37–40].

THG and SHG are efficiently produced using femtosecond excitation pulses and require a single laser. For that reason, combination with fluorescence-based point-scanning 2p-microscopy is straightforward. Several studies have reported harmonic and multimodal harmonic/fluorescence imaging of embryos in various models: fruit fly [5,41], zebrafish [30,42–43], mouse [44], or worm [45]. SHG imaging carried out in light-sheet mode and combined with light-sheet 2p-microscopy, has also been recently demonstrated [16]. CARS/SRS microscopy requires two synchronized and overlapped excitation beams, and contrast is optimized with picosecond rather than femtosecond pulses. For these reasons combination of coherent Raman scattering with 2PEF is more complex, but multimodal fluorescence/harmonic/Raman imaging is becoming a reality.

One particularly attractive aspect of these imaging approaches for embryo imaging is that they provide label-free structural or molecular vibrational imaging. In some experimental situations, it is challenging to obtain long-term, strong, specific, and non-invasive fluorescent labeling. Label-free nonlinear imaging has proven to be a particularly effective addition to fluorescence for studying early division patterns in the zebrafish embryo [30] (Fig. 2d), and lipid storage in worms [37–38,46]. Harmonic-generation contrast usually provides additional structural information on molecular or supra-molecular order that would not be easily detectable with fluorescent labeling strategies: for instance, collagen macromolecular organization into fibrils [47], myosin structural conformation in sarcomere [48], microtubule array polarization during brain maturation [49] (Fig. 2e), or detection of sub-micron-scale anisotropy in the cornea [50]. In addition, we note that artificial nanostructures usually produce strong nonlinear optical signals that, unlike fluorescence, do not bleach or saturate with high laser intensity. Thus, harmonic signals from nanoprobes may be detected with high sensitivity, as illustrated by SHG imaging of nanocrystals inside zebrafish embryos [51] (Fig. 2f).

01