Dynamic light sheet generation and fluorescence imaging behind turbid media

Scattering of light severely compromises the image quality when turbid media such as thick tissues are observed using conventional fluorescence microscopes. On the one hand, multiple scattering leads to a randomization of the illumination into a speckle pattern; on the other hand, the emitted fluorescence signal gets scrambled as well and cannot be traced back to its origin. Slicing, peeling, clearing etc. hence belong to common tasks of biologists who try to reduce the turbidity of their sample in order to unravel the happenings in tissues and developing animals.

Great technical developments in terms of multi-photon microscopy [1], adaptive optics [2–4], wave-front shaping [5–15], speckle(auto)correlation [16–18], time-reversal [19] and optical phase conjugation [20–28] have improved microscopy in and/or behind turbid media to a great extent. However, the image quality, imaging speed and modalities are still subject to possible improvements.

In this paper, we introduce direct and dynamic formation of variants of light sheet illumination behind scattering layers. Light sheet microscopy [29–31] combines the speed advantage of wide-field imaging with selective plane excitation to reduce out of focus fluorescence and has become a popular tool for biologists for fast three dimensional imaging. Light sheet microscopes illuminate only a thin slice of the sample and the emitted fluorescence from this plane is collected with a detection objective placed perpendicular to the excitation. Nevertheless, traveling through scattering media damages this type of illumination pattern as well leading to a progressive widening of the illumination slice.

We use optical feedback based wave-front shaping to transform the speckle pattern behind a scattering layer into a light sheet within the region of interest. Hence, the setup employs three objective lenses: one objective lens for illumination; a second one facing the first on the opposite side of the sample to provide feedback signal for the optimization; and a third one placed perpendicular to the other two in order to detect the fluorescence signal coming from the excited slice of a fluorescently labelled sample.

Figure 1 explains the setup as well as the problems faced for light sheet microscopy in turbid media in more detail. If one illuminates a sample (in our case a glass capillary filled with fluorescent beads) with a wide-field scheme, both detection objectives capture two perpendicular perspectives of the same scene at their respective focal planes. In this mode, the bright beads at the focal plane are heavily blurred with out of focus fluorescence. If the illumination is shaped into a light sheet, only a thin plane perpendicular to the optical axis of the detection objective 2 is excited. Hence, the detection objective 2 captures an image of clearly distinguishable beads from the sheet with significantly less out of focus signal. Detection objective 1 in that case faces a projection of the light sheet. If its focal plane is set in the middle of the sheet, a thin line of strongly illuminated beads is captured. The length of the line corresponds to the height of the light sheet and its width corresponds to the thickness of the light sheet at the focal position. However, if a scattering layer is placed within the illumination path, the illumination light forms a diverging speckle pattern. The images captured by the detection objectives again become blurry due to out of focus fluorescence and the intensity of single beads depends on their overlap with the illuminating speckles.

The aim of our work is to optimize the spatial phase distribution of the illumination to achieve constructive interference of speckle fields behind the scattering layer and make them form a light sheet within a certain region. With our technique, we recover homogeneous selective plane excitation in spite of the scattering layer.

Arbitrary intensity distributions rather than a single focus behind turbid media have already been demonstrated. Malavalli et al. created an array of multiple foci behind turbid media by first determining the phase distribution to create a single focus and then adding linearly another phase mask to it [11]. This strategy can be applied to create a light sheet as well. The second mask to be added must then act as a negative cylindrical lens to elongate the focus into a light sheet. However, several disadvantages would come along. First, the region of possible elongation is fundamentally limited by the memory effect (The memory effect describes the wave-front correlation valid within a certain angular range [35, 36]). Additionally, the focal length of the cylindrical lens requires fine-adjustments to match exactly to the position of the focus along the optical axis. Finally, the intensity within the focus will be spread over a larger region and thus necessitates dynamic and/or static intensity adjustments during and after the optimization. Also Vellekoop et al. showed in his pioneering work optimization on multiple spots [6]. However, in both works, no special effort was put to ensure a homogenous intensity distribution like we did as mentioned in Results section.

The enhancement factors of excitation intensity within ROI varied 7 to 15. Instead of creating a light sheet, one could create one single focus with higher enhancement [6] and scan it over the desired area as well [9, 13]. Our setup creates the light sheet directly and requires no scanning system and additional optical components to create conjugate planes. This makes the microscope simple and robust. As long as the signal to background ratio is enough, the lower enhancements compared to the single focus optimization do not hinder the advantages of light sheet imaging.

In our experiments, the excitation beam and the fluorescence signal could reach the camera directly without being scattered. However, in most biological settings, the detection path will not be clear and the signal coming from the sample plane will be aberrated and scattered as well. This would hinder the evaluation of the light sheet using its image on the camera. In this case, this type of optimization using the direct beam can only be performed by weak scattering on the detection side where significant amount of the direct beam reaches the camera as ballistic photons. In such a scenario, adaptive optics tailored for the detection arm of light sheet microscopes can be implemented as well [37]. If the detection side is subject to severe scattering, as we assume for many biological applications, alternative approaches using fluorescent probe as feedback signal can be applied. For this purpose, a bright and isolated fluorescent particle can be placed at the sample plane of interest. A fluorescently labelled structure of the specimen itself might act as a source as well. The scattered fluorescence signal from the source can be focused on a single point detector on the back side of the overall turbid media. The wave-front of the illumination can then be optimized to increase the signal at the detector which corresponds to an intensity increase of the excitation at the region of interest. Such methods using the image brightness as a metric for different optimization algorithms have already been used for implementation of adaptive optics in confocal and multiphoton microscopy as compared by Wright et al. [38]. To use a labelled structure in the specimen is surely the better choice than invasively bringing a feedback source into the sample which can also destroy the structures in the ROI. However, the labelling might be then too dense leading to a mixed signal from a bigger area. On the other side, sparse labeling can bleach quickly. In that case, quantum dots would offer practically non-bleaching signal. The proposed method would work for creating a single focus in the sample which can be then scanned within the range of memory effect. To collect signal from a line shaped source instead of a point source can lead to an increase of overall intensity of a light sheet. However, the homogeneity cannot be deduced from the signal increase at the detector since the scattered fluorescence from different parts of the line shaped source would be mixed up. To still ensure homogeneity and create directly a light sheet without scanning, a chain of spectrally different quantum dots can act as feedback source. Many quantum dots can be excited by the same wavelength but will emit fluorescence at a different wavelength. Optimizing on different colors from adjacent spots can lead to a more homogenous intensity increase of light sheet type illumination.

Our optimization speed in the range of tens of seconds is moderate. Cui demonstrated focusing through highly scattering media within 400 ms via spatial frequency modulation [39]. Tang et al. used nonlinearity assisted iterative optimization (IMPACT) and focused light through mouse brain tissue and skull within ~5 s [15]. In a later in vivo study on neurons, the optimization time was reduced down to ~2 s [40]. Liu et al. carried the time scale of time-reversed ultrasonically encoded (TRUE) optical focusing down to a few milliseconds [23]. Wang et al. achieved similar speeds via digital optical phase conjugation employing a digital micro-mirror device and a FPGA with custom firmware [27]. Even focusing behind turbid media in sub-millisecond time scale was shown by the process of field self-organization inside a multimode laser cavity [41]. All these works offer higher speeds than our system. However, they require elaborate optical paths and/or dedicated electronic hardware and software which makes the systems complex and expensive. The ones utilizing phase conjugate mirrors show also wavelength dependency. Once established, our setup is easy to adjust, can tolerate misalignments and is inexpensive. The calibration of the SLM can easily be scaled for different wavelengths. This makes our system suitable for usage not only in physics labs but also in biology labs. Another faster and simple approach was developed by Edrei et al. based on so called shower curtain effect [42]. But it requires coherent light which makes fluorescence light unsuitable as contrast mechanism for imaging. A moderate optimization speed was also achieved by N’Gom et al. [43]. Here, portions of the transmission matrix of the scattering medium are constructed by computationally intensive semi-definite programming. Single and multiple foci were created behind turbid samples with a speckle persistence time in minutes scale. Besides, imaging inside turbid media has been enabled by other computational methods utilizing compressive sampling. Duran and Tajahuerce et al. used transmitted light captured by a single pixel camera and could reconstruct sparse high contrast objects hidden in turbid media [44, 45]. However, imaging of fluorescence emitting low contrast objects still remained a challenge.

We have shown that our wave-front shaping microscope enables homogenous light sheet illumination of fluorescent samples behind turbid media. This type of illumination is particularly advantageous while exciting a big area at once with reduced out of focus light. Our setup can compensate sample dynamics on a time scale below minutes and suits for example applications for developmental biology such as organ growth and morphogenesis in embryo and larvae. Another possible application area is optical coherence tomography (OCT) in ophthalmology where the backscattered light from retina within rather “clean” eye can serve as feedback signal. Dedicated wave-front correction techniques improving the penetration depth and image quality in OCT have already been shown by Fiolka et al. [46] and Yu et al. [47]. These works can be extended to achieve light sheet type illumination as well.

This work of illumination optimization presents a necessary first step for imaging of turbid biological scenes, which eventually needs to be complemented with methods to recover the fluorescent image after scattering. These complementary methods need to be performed based on the signal from detection path 2 facing the light sheet plane, since phase aberrations would occur differently on our two different detection sides.