High throughput wide field second harmonic imaging of giant unilamellar vesicles

Imaging techniques based on nonlinear optical effects can probe specific intrinsic aspects related to the spatial arrangement of materials. Among them, second harmonic (SH) microscopy received wide recognition in the field of bioimaging, thanks to its unique selection rules, which prohibit coherent generation of photons from symmetric and isotropic structures. This makes the SH generation process a sensitive tool to image noncentrosymmetric structural arrangement of molecules such as collagen and fibrils.^1–4 The number of generated SH photons is determined by two main factors: (i) laser excitation conditions^5,6 and (ii) the degree of asymmetry of the sample.⁷ 

Excitation and collection in SH imaging are traditionally performed in a scanning confocal configuration. However, for samples with low inherent asymmetry, this approach results in either extremely long acquisition times or laser fluences that are incompatible with biological samples.⁸ For this reason, various approaches were taken to improve the throughput. Many approaches are based on parallelizing the imaging process, for example, through harmonic holography,^9,10 lensless imaging,¹¹ spatiotemporal wide-field illumination,^12–15 multifocal imaging,^16,17 or through wide-field counter propagating SHG geometries.^18,19 Other approaches work with higher laser repetition rates and, instead, increase the speed of the scanning process.^20,21

Previously, we have shown that utilizing a wide-field geometry with an appropriate repetition rate makes it possible to increase the throughput of an SH microscope by three orders of magnitude compared to scanning systems.^22–24 With this approach, it is possible to use laser fluences/peak powers typical for damage thresholds for confocal scanning imaging but with dwell times that are 10⁶ times longer, massively enhancing image quality (while cell proliferation is not hampered).²³ Also, interfaces between isotropic media can be SH imaged label-free using image acquisition times in the millisecond range.²⁴ This enhancement in throughput allowed us to perform nonresonant SH imaging of interfacial water on the membranes of model lipid bilayers^25–28 as well as in living, activated neurons.²⁹ These studies and others that included a wide plethora of interfaces³⁰ investigating interfacial water with a ∼100 ms acquisition time per image allowed us to extract important physico-chemical parameters as well as their dynamic behavior. Changes in the surface potential and the free energy³¹ could be mapped with reasonable spatiotemporal resolution. It also became clear that even shorter timescales would enable a better handle on the fundamental physics and chemistry that occurs at interfaces.

Here, we explore a further improvement in high-throughput wide-field SH microscopy by tuning the pulse duration of the incoming pulses. To perform this task, we built a single-stage optical parametric amplifier (OPA) pumped by a medium repetition rate femtosecond laser system, combined with group delay dispersion (GDD) compensation, leading to wavelength tunable femtosecond laser pulses with a pulse duration of 23 fs at the sample focus. We utilize the system to SH image interfacial water at the lipid membrane interface of unstained giant unilamellar vesicles (GUVs). We compare the nonresonant SH emission of interfacial water to resonant SH imaging from GUVs stained with a membrane-sensitive dye (FM4-64) and 2-photon fluorescence (2PF) obtained from the same GUVs labeled with Nile Red. The pulse compression achieves 78% of the theoretically possible throughput improvement compared to longer pulse durations, and the obtained nonresonant responses are a factor of 10⁷ smaller than the resonant response. Finally, we show how the nonresonant SH intensity images of interfacial water can be converted to a membrane potential map of GUVs.

Using considerations of this kind, we previously showed²² that a wide field imaging geometry with ∼190 fs microjoule pulses delivered at an ∼100 kHz repetition rate can improve the SH throughput by two to four orders of magnitude compared to scanning systems that use nanojoule pulse energies with ∼100 MHz repetition rates. Based on these results, further improvements in throughput are possible to some extent by increasing the laser fluence and repetition rate. However, this is only possible up to a sample-dependent limit [∼100 mJ/cm² for unstained CHO cells⁸, ∼20 mJ/cm² for human embryonic kidney (HEK) cells²³], after which the deposited energy will lead to sample degradation. In order to increase the throughput further, the parameter that has not been optimized yet is the laser pulse duration $(τ)$⁠. It is clear from Eq. (1) that decreasing $τ$ while keeping other parameters constant should lead to a higher imaging throughput.

To achieve shorter pulses, we built an OPA and combined it with a single beam wide field SH microscope. The microscope is sketched in Fig. 1(a). We used a 1030 nm pulsed laser with a 1 MHz repetition rate, 230 fs duration, and 3 μJ pulse energy (Ekspla, Femtolux 3) to pump a custom-built OPA described below. In this configuration, 900 nm 23 fs pulses were focused in the back focal plane [BFP in Fig. 1(a)] of the excitation objective lens (Olympus LUMPLFLN60XW, a 1.0 NA water immersion objective) to achieve wide field illumination of the sample plane. This objective together with a 150 mm achromatic lens (ACN 150, Thorlabs) creates a 90 μm circular illumination area in the sample plane, resulting in a fluence of 5 mJ/cm² and a peak power of 200 GW/cm². The generated SH photons were collected in a transmission geometry using a 60× Olympus LUMFLN60XW 1.1 NA objective with coverslip correction, which projected the image onto an electron multiplying intensified CCD (EM-ICCD) camera (PI-MAX 4, Princeton Instruments).

The custom-built OPA layout follows a common design scheme for wavelength tunable OPAs.³² As illustrated in Fig. 1(b), the pump laser output was split (P1) into two pulses, each used as a pump source for two mixing processes. The first one generates a supercontinuum (SC) seed beam for optical parametric amplification and the second one up-converts the rest of the near-infrared light (1030 nm) into visible light (515 nm) using SH generation. The SC beam was generated inside a 4 mm thick yttrium orthovanadate (YVO₄) crystal by focusing the laser light with a 50 mm lens (L1). This crystal was chosen due to its low SC generation threshold (200 nJ), a broad emission spectrum that covers the required spectral region, and good, long-term stability.³² Optimal SC generation was achieved by pumping the YVO₄ crystal slightly above the threshold using a 300 nJ pulse. The SC light was collimated and the remaining fundamental beam was removed by a short-pass dichroic mirror (DM1). The remaining fundamental beam with an energy of 2.7 μJ was focused into a 1 mm thick beta barium borate (BBO) crystal. The laser polarization, BBO orientation, and focusing conditions were adjusted to maximize the SH conversion efficiency (∼67%). The generated visible light (515 nm) was collimated and filtered from the remaining fundamental light with long pass dichroic mirrors (DM2, DM3). This beam was used as the OPA pump beam to amplify the SC radiation. SC amplification was performed in a 5 mm thick lithium triborate (LBO) crystal. Spatial and temporal overlap of the SC and the pump radiation was achieved inside the LBO crystal. The time-overlap was tuned by a delay line in the SC beam path (M4, M5). Wavelength tuning was performed by rotating the LBO crystal, and thus, changing the crystal optical axis orientation relative to the directions of the incoming beams and by correspondingly adjusting the delay line, in order to compensate for group delay dispersion (GDD) of SC radiation. The output of this single-stage OPA featured a broad-band spectrum (on the order of 100 nm), which is compressible down to several tens of femtoseconds. Figure 1(c) displays several output spectra obtained. The amplified signal beam was compressed in a prism compressor that was used to compensate for GDD by changing the spacing between the prisms. It is important to note that this compression step can be used not only to compensate for the GDD accumulated in the optical path of the OPA itself, but also to precompensate for the GDD introduced in the optical path of the microscope layout. Figure 1(d) shows the measured pulse duration as a function of wavelength. The wavelength tunable OPA output radiation consisted of 1 MHz, 250 mW pulses with a duration of <50 fs.

Next, we perform label-free imaging of giant unilamellar vesicles [GUVs, Figs. 2(a)–2(d)], which are widely used as a model system in membrane research and synthetic biology.^33–35 GUVs were formed from a 1:1 mixture of zwitterionic 1,2-diphytanoyl-sn-glycero-3-phosphocholine (DPhPC) and anionic 1,2-diphytanoyl-sn-glycero-3-phosphate (DPhPA) lipids by PVA-assisted swelling in a 45 mM sucrose solution, followed by transfer into an observation chamber with glucose and divalent salt solution of matching osmolarity³⁶ (see supplementary material for more details about sample preparation).⁴⁹ After formation, the molecular structure of the prepared lipid bilayers is symmetric, and thus, GUVs do not generate any SH photons. When 5 mM of CaCl₂ salt is added to the solution outside of the GUV, this symmetry is disrupted, which enables SH generation to take place at the interface, in the same manner as was observed earlier on free-standing lipid membranes in aqueous solution.³¹ The SH contrast originates from a net transmembrane asymmetry in the orientational distribution of water molecules. The asymmetry stems from water molecules that have a preferential orientation in the anionic head group generated electric field on the inside of the GUV, while at the outside, direct Ca²⁺ ion-head-group interactions reduce the electrostatic field from anionic groups, disorienting the interfacial water [Fig. 2(a)].

To assess the achieved improvement in the imaging system, we compared the imaging throughput with and without the reduced pulse duration. To do so, the microscope was reconfigured to operate with two laser illumination conditions: the compressed output of the OPA and an output of a laser system with a longer pulse duration (200 kHz, 190 fs, 20 μJ, Pharos, Light Conversion). Both laser sources were aligned to excite the same area at the sample plane with an equal fluence of 5 mJ/cm², a repetition rate of 200 kHz, and an acquisition time of 500 ms. Figures 2(b) and 2(c) show the comparison between SH images of unstained GUVs recorded with two different laser systems. All imaging parameters are summarized in Table I. Figure 2(b) shows an SH image obtained with the 190 fs laser system. This image is underexposed with a barely distinguishable GUV outline. Figure 2(c) shows an image taken using the compressed output of the OPA for excitation. Here, higher contrast and good image quality are obtained with the same acquisition time. Figure 2(d) shows a comparison of SH intensity obtained with the two configurations, where the shorter pulses of the OPA yield ∼6.4 times more SH photons. It is worth noting that this value is lower than the theoretically predicted factor of ∼8.2 given by Eq. (1). This difference originates from third-order dispersion that is accumulated in the microscope optics and leads to pulse stretching. This type of dispersion cannot be compensated with a two prism compressor and requires a more sophisticated compressor geometry. Nevertheless, this result shows that a further decrease in pulse duration would not bring a significant advantage in SH throughput since such short pulses suffer from significant material dispersion.

Having compared the improvement achieved by reducing the pulse duration in a nonresonant SH experiment, it is also interesting to consider the differences with resonant SH and two-photon fluorescence (2PF) imaging. Doing so will allow us to assess the difference in throughput between both processes. While it is widely known that the resonant enhancement should result in a throughput that is several orders of magnitude higher than nonresonant excitation,³⁷ we are not aware of a direct comparison in the literature. To do so, we stained GUVs with 10 μM FM4-64, a voltage-sensitive membrane selective probe.³⁸  Figure 2(e) shows the obtained image, recorded with the parameters provided in Table I. The intensity scale of the image was adjusted to the parameters of Figs. 2(b) and 2(c). Table I lists the gain in throughput corrected for all imaging parameters to be 1.3 × 10⁷, which is in agreement with expectations. Figure 2(f) shows a two-photon fluorescence image stained with 1 mol. % Nile Red (keeping the lipid composition the same) showing as well a drastic improvement over the nonresonant response. Figure 2(g) shows an intensity cross section following the dashed lines of Figs. 2(e) and 2(f). While the 2PF intensity is found from inside the GUV as well as on its surface, the SH intensity of FM4-64 originates only from the interface. This is partially due to the selection rules of both processes (2PF is allowed in isotropic media, while SHG is not), the optical response of the dye, and the difference in cross-sectioning (SHG has a much narrower depth of focus as it only occurs at the highest electromagnetic field intensities of the focus). Finally, due to its chemical structure, FM4-64 is known to distribute to membranes, enhancing the interfacial SH response. This is in agreement with previous studies comparing resonant 2PF and SHG on GUVs.^38–40 

Although the bilayer-induced interfacial asymmetry leads to much weaker nonresonant SH responses than can typically be achieved with a resonant SH/2PF process, it contains valuable information about the molecular biophysics of the system: The obtained SH intensity can be converted into an electrostatic potential value, which, in turn, can be used to derive spatiotemporal maps of electrostatic interfacial free energy and surface charge density.³¹ Being able to SH image interfacial water on GUVs with 0.5 s acquisition time allows for the investigation of a variety of biophysically relevant processes, such as ion-membrane transport,²⁷ and other spatiotemporal properties of the electric double layer. It also offers a useful alternative to similar experiments conducted on free-standing lipid membranes in aqueous solution.^25,31,41

Based on Eq. (3), SH intensity is converted to a surface potential difference $Δ Φ 0$⁠. To obtain the scaling factor, we recorded SH images of free-standing lipid membranes (FLMs) with same lipid compositions as a function of external electric bias (U) across the membrane, as was demonstrated previously.^31,42 In the case of a symmetric FLM, we have $Δ Φ 0∼U$⁠. The image averaged intensity took the form of a parabola whose fit function was used to convert the GUV intensities into corresponding $Δ Φ 0$ values. In this procedure, FLMs and GUVs were imaged with the same acquisition time and illumination conditions. Figure 3(c) shows the calibration curve where the average intensity is normalized to pixel-wise intensities. Using the corresponding $Δ Φ 0$ values for an image intensity value, the bottom panel of Fig. 3(c) is constructed. Figure 3(d) shows the spread in membrane potential values. The distribution is remarkably broad but in agreement with earlier studies.^27,31,42 Once the surface potential is determined, one can convert it into electrostatic free energy, binding constants, or surface charge density, whatever is relevant for the particular topic of study. Having spatiotemporal information about membrane potential and energetics is extremely valuable to understand a wide plethora of membrane interfacial processes from membrane biophysics to ion channel operation and neurological activity.^29,30,41 Valuable insights about membrane energetics can, in the future, be obtained by combining aspiration and patch clamping^47,48 or resonant with nonresonant responses. While a probe such as FM4-64 is widely recognized and used as a membrane potential marker, the membrane potential readout is largely empirical: without knowing the orientational distribution function and the interfacial concentration, one cannot achieve the same precision and certainty as with water imaging. Also, being a positively charged surfactant molecule, it is likely to structurally perturb and eventually break up the lipid membrane, which is probably the reason for its high toxicity. Therefore, future studies combining resonant and nonresonant approaches will present a useful approach to combine the best of both worlds: Label-free, quantified, relatively weak optical responses with labeled, indicative, strong optical responses that allow shorter acquisition times.

In this study, we performed high-throughput wide-field SH imaging of interfacial water at the interface of unstained GUVs with ultrashort laser pulses. We compared the nonresonant SH response of GUVs generated using 23 or 190 fs pulses. We also compared the nonresonant SH imaging process to 2PF and resonant SH imaging from GUVs stained with a membrane-selective fluorophore. The influence of pulse duration was tested using a home-built single-stage optical parametric amplifier (OPA) pumped by a medium repetition rate femtosecond laser system, combined with group delay dispersion compensation. The OPA produced 23 fs laser pulses at a maximum of 1 MHz repetition rate. We compared the nonresonant SH emission of interfacial water to resonant SH imaging from GUVs stained with a membrane-sensitive dye (FM4-64) and 2PF obtained from the same GUVs labeled with Nile Red. The pulse duration reduction manifested 78% of the theoretically possible throughput improvement compared to the longer pulses. The obtained nonresonant responses are a factor of 10⁷ smaller than the resonant response. Finally, we showed how the nonresonant SH intensity images of interfacial water can be converted to a transmembrane potential map of GUVs. Combining the low intensity, but quantitative SH response of water, with that of a high intensity but qualitative response probe represents a promising strategy for future research to understand the interfacial processes in, virtually, any aqueous system.

This work was supported by the Julia Jacobi Foundation, the Swiss National Science Foundation (Grant No. 200021-182606-1), the European Union’s Horizon 2020 research and innovation program under Marie Skłodowska-Curie Grant Agreement No. 721766 (No. H2020-MSCA-ITN), and European Research Council Grant Agreement No. 951324 (No. H2020, R2-tension).

The authors have no conflicts to disclose.

Ethics approval is not required.

M. Eremchev: Data curation (equal); Formal analysis (equal); Investigation (equal); Software (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). D. Roesel: Data curation (equal); Formal analysis (equal); Investigation (equal); Software (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). P.-M. Dansette: Investigation (equal); Writing – original draft (equal); Writing – review & editing (equal). A. Michailovas: Supervision (equal). S. Roke: Supervision (equal).

The data that support the findings of this study are available within the article and its supplementary material.