Whole-brain imaging and characterization of Drosophila brains based on one-, two-, and three-photon excitations

To study functional connectome, optical microscopy provides the advantages of in vivo observation, molecular specificity, high-speed acquisition, and sub-micrometer spatial resolution. Now, the most complete single-neuron-based anatomical connectome is built upon Drosophila; thus it will be a milestone to achieve whole-brain observation with sub-cellular resolution in living Drosophila. Surprisingly, two-photon microscopy cannot penetrate through the 200-μm-thick brain, due to the extraordinarily strong aberration/scattering from tracheae. Here we achieve whole-Drosophila-brain observation by degassing the brain or by using three-photon microscopy at 1300-nm, while only the latter provides in vivo feasibility, reduced aberration/scattering and exceptional optical sectioning capability. Furthermore, by comparing one-photon (488-nm), two-photon (920-nm), and three-photon (1300-nm) excitations in the brain, we not only demonstrate first quantitative reduction of both scattering and aberration in trachea-filled tissues, but unravel that the contribution of aberration exceeds scattering at long wavelengths. Our work paves the way toward constructing functional connectome in a living Drosophila.


Introduction 1
Drosophila is an important model animal to study connectomics since its brain 2 is complex with 10 5 neurons but still small enough to be completely mapped by 3 optical microscopy with single-cell resolution. Compared to other model animals, the 4 genetic tool box is more complete with Drosophila, and a connectome map based on 5 in vitro structural registration of more than 30,000 cells has been established (Chiang 6 et al., 2011), serving as an invaluable reference for functional connectome study. To 7 study the functional connectome, two-photon fluorescence (2PF) microscopy is now 8 the most popular tool because of its advantages on low photobleaching and 9 phototoxicity, subcellular spatial resolution, and deep penetration depth (Helmchen & 10 Denk, 2005). When observing living mouse or zebrafish brain with 2PF microscopy, 11 the penetration depth approaches 1 mm, which is typically limited to about five 12 scattering lengths (Helmchen & Denk, 2005;Horton et al., 2013). However, even 13 using the same fluorophore and excitation wavelength, the reported imaging depths in 14 a living Drosophila brain are much more limited. For example, when imaging 15 GCaMP with excitation wavelength around 920-nm, activities from mushroom bodies 16 (MB) had been recorded at only several tens of micrometers in depth (Honegger,17 Campbell, & Turner, 2011; Y. L. Wang et al., 2004). Using the same combination of Although the thickness of a Drosophila brain is only about 200 µm, which is much 22 smaller than the typical imaging depth of 2PF microscopy in other model animals like 23 mouse and zebrafish, to the best of our knowledge, no study has demonstrated in vivo 1 whole-brain imaging in Drosophila, nor has characterized the image attenuation of a 2 living Drosophila brain. The whole-brain observation capability is a major milestone 3 toward establishing functional connectome in this model animal. 4 The underlying difficulty of living Drosophila whole-brain imaging is that, 5 different from mouse and zebrafish, where blood vessels are responsible for oxygen 6 exchange, air vessels, i.e., tracheae, are in charge of oxygen exchange in Drosophila 7 brains. The micro-tracheae in the brain are a few micrometers in diameter (Beitel & 8 Krasnow, 2000), comparable to near infrared wavelengths, and thus induce 9 extraordinarily strong aberration/scattering from the air/tissue interface since the 10 refractive index (RI) difference between air and tissue is much larger than that 11 between blood and tissue. This tracheae-induced aberration/scattering impedes deep 12 tissue observations inside a living Drosophila brain. However, the optical properties 13 of trachea-filled tissues have not been well studied. Therefore, the aim of this work is 14 to unravel the optical effect of tracheae, and to design a suitable method to increase 15 imaging depth in trachea-filled tissues. 16 To increase imaging depth in living animals, there are several known 17 approaches. For example, photo-activatable fluorophores (PAFs) have been used to 18 suppress out-of-focus fluorescence (Wei, Chen, & Min, 2012), and high-energy lasers 19 were used to enhance excitation efficiency at deep tissue (Theer, Hasan, & Denk, 20 2003 Wang et al., 2014). Nevertheless, the best depth achieved by AO in a living 3 Drosophila brain to date is still less than 100 µm (Pedrazzani et al., 2016), since the 4 aberration inside the insect's brain is much larger than that of vertebrate's brain (RI 5 difference between air and tissue is at least ~ 1 order of magnitude larger than that 6 between blood and tissue). In addition, typically AO does not compensate the 7 scattering effect (Booth, 2014), which limits its impact in improving the imaging 8 depth in Drosophila. 9 On the other hand, long excitation wavelength is well known to greatly improve  Furthermore, high-order optical nonlinear excitations, such as three-photon absorption, 16 are often combined with long wavelength, thus providing better excitation 17 confinement, i.e. better optical sectioning capability, than 2PF, to improve image 18 contrast in deep tissue. Combining these factors together, long wavelength excitation 19 is promising for whole-brain imaging in Drosophila. 20 Here, two approaches are adopted to achieve whole-brain observation in a 21 Drosophila brain. The first one is to pump our air inside the tracheae, i.e. degassing. 22 Since the tracheae-induced aberration/scattering is largely removed in the degassed 23 brain, 2PF microscopy penetrates through the whole brain. However, the Drosophila 1 is not alive after degassing. To achieve in vivo whole-brain observation, the second 2 approach is to use three-photon imaging based on excitation wavelength at 1300-nm 3 in a GFP-labeled living Drosophila brain. The three-photon fluorescence (3PF) 4 method provides exceptional excitation confinement and simultaneously reduced 5 aberration/scattering, thus allows high-contrast and high-resolution image throughout 6 the whole brain. The accompanying third harmonic generation (THG) modality 7 provides detailed map of the densely distributed tracheae in the brain, useful for 8 structure-function studies. 3 (D) are 1PF and 2PF images of a degassed brain at the same depths. Although brain-edge structures 4 (arrowheads) are still visible at 60 µm depth in (A), the image loses contrast in brain center (arrow). In 5 the degassed brain, the penetration depth of 1PF is significantly improved in (B), but not approaching 6 the bottom of brain. In (C), 2PF penetrates deeper than 1PF, but becomes blurry at depth beyond 100 7 µm. Whole-brain imaging is achieved in (D) by 2PF in a degassed brain, which nevertheless has no  Within a living brain, the 1PF images lose the image contrast at around 40 µm 8 (Video 1), as no structures are visible in the brain center where the white arrow points 9 in the 60-µm panel of Figure 1A. The arrowheads indicate structures that located at 10 the edge of the brain. To verify the effect of trachea, a brain is degassed, i.e. air in 11 tracheae is pumped out. Figure 1B shows that 1PF in the degassed brain provides 12 much better contrast in the center of brain at the same 60 µm depth, but cannot exceed 13 130 µm. The 1PF imaging depth of the degassed brain is comparable with that in 14 mouse brains, which is mainly limited by scattering (Helmchen & Denk, 2005). 15 Comparing the results in Figures 1A and 1B, it is obvious that degassing removes the 16 additional attenuation contributed by tracheae. 17 On the other hand, it is well known that using long excitation wavelengths 18 with 2PF modality efficiently improves imaging depth, approaching 1 mm in mouse 19 brains (Helmchen & Denk, 2005). Using the same excitation wavelength (~ 920-nm) 20 and contrast agent (GFP families), Figure 1C shows the 2PF imaging depth in a living 21 Drosophila brain indeed increases compared to Figure 1A, but reaches only about 100 22 µm, which is not adequate to penetrate the whole brain, mainly due to the tracheae. 23 By combining 2PF modality with a degassed brain, Figure 1D presents the 24 first whole-brain imaging in a Drosophila. Reasonable contrast and resolution are 1 maintained throughout the nearly 200 µm depth, manifesting that the trachea-induced 2 aberration/scattering is the major restraint for deep-brain imaging in this model 3 animal. However, please note that the animal is no longer alive after degassing. 4 Moreover, the degassing process distorts the brain structures, making its association 5 with the Drosophila structural connectome difficult. 6 In order to enable the study of whole-brain functional connectome, 3PF at 7 excitation wavelength ~ 1300-nm was adopted. Figure 1E shows that this imaging 8 modality provides subcellular resolution throughout a living Drosophila brain. 9 Associating the 3PF images with an existing Drosophila brain database (FlyCircuit,  During the three-photon excitation process, a complementary contrast, i.e. 13 THG, is generated by the interfaces with RI differences, and thus is sensitive to the 14 tissue/air boundaries in tracheae. Figure 1F shows that THG signals reveals the 15 detailed distribution of tracheae, which are randomly distributed inside the brain and 16 causing strong aberration/scattering.   is provided by 2PF plus degassing. Comparing to mouse brain whose µatt of 2PF and 6 3PF are less than one-third and one-eighth of 1PF (Horton et al., 2013), the influence 7 of longer excitation wavelength is much smaller in Drosophila brains. To better 8 understand the results, the µatt has to be decomposed, as we explain in the following.

9
In Figure 2D, the attenuation coefficients of in vivo and degassed brains are 10 plotted in black and green, respectively. It is interesting to note that the attenuations of 11 degassed brains are comparable to that of a mouse brain (pink circles), which is 12 mostly contributed from scattering of neurons (Horton et al., 2013). The slight 13 difference may be due to the effect of residual wax in the tracheae. From the 14 comparison of in vivo Drosophila brain (total attenuation, µatt) and mouse brain 15 (attenuation from neurons, µneu), the contribution of trachea µtra = µattµneu is derived 16 as red triangles in Figure 2D. 17 The results show that both µneu and µtra reduce with increasing wavelengths, with 18 respectively l -2 and l -0.5 dependencies (see Figure supplement 2). Since the 19 attenuation of neuronal tissue µneu is dominated by scattering, which includes 20 Rayleigh scattering (l -4 dependence) and Mie scattering (l -1 dependence), the l -2 21 dependency should be a natural result of the mixture. On the other hand, the l -0.5 22 dependency suggests that µtra may not be governed by scattering, whose wavelength 1 dependency should be in between l -4 and l -1 , but by aberration from air-tissue 2

interfaces. 3
To further confirm that scattering does not play the key role in the attenuation of 4 the trachea-filled brain, we compare the 2PF µatt of Drosophila and mouse in Figure   5 2D, the latter (~ 5.55 mm -1 ) is less than half of the former (~ 13.5 ± 0.5 mm -1 ). 6 Nevertheless, 2PF imaging in mouse brains typically reaches almost 1 mm, i.e. about 7 five scattering lengths; while in living Drosophila brains, only about 100 µm, i.e. less 8 than two attenuation lengths, is achieved. Similar situation occurs with the 3PF 9 modality. The surprisingly limited penetration depth suggests that scattering does not 10 dominate the attenuation, but the trachea-induced aberration dictates, especially in 11 multiphoton imaging cases. The underlying reason of the limited depth is that 12 aberration not only induce excitation intensity loss, but also strongly deteriorates 13 point-spread-function (PSF), especially in the axial direction (Figure supplement 3). 14 As a result, capability of 2PF optical sectioning quickly loses at relatively shallow 15 region, leading to blurred images and considerably limited imaging depth, as 16 manifested in Figure 1C. On the other hand, 3PF imaging modality not only exhibits 17 less aberration ( Figure 2D), but also benefits from better optical sectioning ( Figure  18 supplement 3), thus provides deeper imaging depth than 2PF. In contrast, the single-19 photon fluorescence images in Figure 1A (see the arrow) demonstrate loss of image 20 contrast mainly due to scattering. 21 In Figure 2A and 2B, the trachea-contributed aberration is more efficiently 22 removed by degassing in the 2PF modality (2-fold reduction in µatt, Figure 2B) than 23 1PF (~ 30 % reduction, Figure 2A). In addition, from Figure 2D, µtra for 2PF and 3PF 1 are larger than µneu, especially in the 3PF case. These two observations further suggest 2 that tracheae-contributed aberration is the main factor to impede multiphoton deep-3 brain imaging in a living Drosophila.

12
As shown in Figure 1, to reaching the milestone of in vivo whole-brain imaging 13 in Drosophila, 3PF at 1300-nm should be the optimal choice. Here we quantify the 14 depth limit of 3PF by analyzing its signal-to-background ratio (SBR) in Figure 3A and 15 3b, along with comparison between 2PF and 3PF (see Materials and Methods for SBR 16 calculations). The former reaches unity at ~ 100 µm, since no structure is 1 distinguishable beyond this depth (Video 2), while the SBR of the latter significantly 2 outstands the former. From Figure 3B, the imaging depth of 3PF reaches at least 200 3 µm. The result once again supports that 3PF is capable to image through a whole 4 Drosophila brain. 5 The spatial resolutions of 3PF and THG in deep-brain region (186 µm depth) are 6 given in Figure 3C -3H. For 3PF, the minimal full-width-at-half-maximum (FWHM) 7 is 1.0 µm laterally ( Figure 3E) and 6.5 µm axially ( Figure 3G), which is sufficient to 8 resolve a single cell in the Drosophila brain (Tuthill, 2009). For THG, the FWHM is 9 slightly larger, 1.5 µm in lateral ( Figure 3F  bound of the THG spatial resolution. 12 In conclusion, we have, for the first time, characterized the optical properties of 13 the Drosophila brain, which is filled with air, with single-photon, two-photon, and 14 three-photon modalities. We found that the main limiting factor that impedes in vivo 15 whole-brain single-photon imaging is scattering, but for multiphoton imaging, 16 aberration from tracheae structures plays a more dominant role. The aberration affects 17 not only signal attenuation, but also image visibility. Although degassing enables 18 whole-Drosophila-brain imaging by reducing trachea aberration, the only way to 19 achieve in vivo whole-brain imaging with single cell resolution is 3PF at 1300-nm 20 excitation, which exhibits less scattering, aberration, and better optical sectioning. It is 21 possible to combine with AO to further reduce aberration (Tao et al., 2017), thus 22 allowing functional imaging on the scale extending from a single neuron, a complete 23 brain network, toward a whole-animal connectome (Lo & Chiang, 2016). 1 2

Microscope setup 4
One-photon microscope. 5 The one-photon imaging was done on a commercial microscope LSM 780 6 (Zeiss, Germany). The built-in laser (488-nm) and photomultiplier tube was used to 7 single-photon excitation and signal detection. A water immersion objective was used 8 (Olympus, XLPlan N, 25× NA 1.05) for its high transmission in both visible and IR 9 wavelength ranges. A pinhole with ~ 60 µm diameter was used to achieve optical 10 sectioning. The image formation was done by the controlling software Zen (Zeiss, 11 Germany). 12

Multiphoton microscope. 13
The setup was the same as that in Ouzounov et al (Ouzounov et al., 2017). A 14 home-built laser-scanning microscope that is compatible to long wavelength 15 excitation is constructed. A Ti: sapphire laser at 920-nm with an 80 MHz repetition 16 rate, and an optical parametric amplifier at 1300-nm with a 400 kHz repetition rate, 17 were used as excitation sources of 2PF and 3PF respectively. The same water 18 immersion objective as single-photon microscope was used. The power levels for 19 both lasers after the objective were limited to less than 20 mW for all imaging depths. 20 The fluorescence and THG signals were epi-collected with a dichroic beamsplitter 21  The signal current from the detectors was converted to voltage, amplified and low-9 pass filtered by a transimpedance amplifier (Hamamatsu, C9999) and another 1.9 10 MHz low-pass filter (BLP-1.9+, Minicircuits). Analog-to-digital conversion was 11 performed by a data acquisition card (PCI-6115, National Instruments).  The living Drosophila was immobilized in a pipette tip with volume 100 µL after 19 anesthetized by ice bathing. A window was cut into the head by using fine tweezers, 20 after placing a drop of Ca 2+ -free saline on the brain to prevent desiccation, and fat 21 bodies above the brain were removed, under a stereomicroscope. The dissection saline 22 was then replaced with a drop of Ca 2+ -containing saline (108 mM NaCl, 5 mM KCl, 2 23 mM CaCl2, 8.2 mM MgCl2, 4 mM NaHCO3, 1 mM NaH2PO4, 5 mM trehalose, 10 1 mM sucrose, and 5 mM HEPES [pH 7.5, 265 mOsm]). No cover glass was placed 2 between the brain and the objective. 3 To check the optical effect caused by the tracheae structure, degassing the 4 Drosophila brain was performed by pumping out the air inside tracheae. The 5 degassing protocol followed the previous publication of in situ Drosophila brain 6 imaging (C.-W. Lin et al., 2015). The degassing protocol started from immersing the 7 Drosophila in 4 % paraformaldehyde and 2 % triton, expelling air in tracheae by 8 using a vacuum chamber that was depressurized to -72-mmHg for 2.5 minutes, wait 9 for 1.5 minutes, and then releasing to normal pressure for 2 minutes. The degassing 10 process was completed by repeating the above procedure 4 times. After degassing, the 11 same microsurgery preparation was performed, and observed under the same 12 microscope. 13 14 Signal analyses. 15

Attenuation coefficient calculation. 16
This section explains how to derive the attenuation coefficients (µatt), which is 17 the inverse value of attenuation length (la), i.e., µatt = la -1 , in the results of Figure 2. intensity, F (n) (d), is related to the excitation intensity as: 2 (2). 3 For 1PF, 2PF and 3PF, n equals 1, 2 and 3, respectively. Combining eqs. 1 and 2, 4 where a is a proportional constant.  Therefore, µatt is determined by dividing the inverse value of the slopes over n.

13
To obtain the slopes of the curves in Figure 2A -2C, the data points for linear 14 regression fitting are selected by the same criteria. They are all selected with the 15 starting point of signal decay to the depth limit of the corresponding imaging modality. 16 For example, the depth range for fitting in Figure 2B starts ~ 20 µm and ends ~ 110 17 µm. However, the imaging depth limit of 3PF represent the bottom of the brain, 18 instead of the penetration depth limit. 19 20 Signal-to-background (SBR) calculation. 21 The SBR calculation was used in Figure 3 to determine the imaging depth 22 F ( n) (d) ∝ (I(d)) n F ( n) (d) = a×e −n×µ att ×d ln(F ( n) (d)) = ln(a)− n×µ att ×d limitation, and the method was similar to our previous publication (Ouzounov et al., 1 2017). Briefly, the signal value was averaged from the pixels with top 1 % intensities 2 from the white box in inset of Figure 2A, and the background value was the average 3 of all pixel intensities in the white box of Figure 3A, where no brain tissue was found 4 throughout the whole depth range. With this definition, the signals mostly came from 5 the in-focus regions, and the background arose from out-of-imaging-plane 6 contributions. Both SBR results are normalized to the 2PF value at 100 µm depth, 7 where no distinguishable features are observable. 8 9

10
The three-photon brain imaging cannot be done with the great help from Chris Xu and 11 Tianyu Wang from School of Applied and Engineering Physics in Cornell University. 12 In addition, we appreciate the generous support of transgenic Drosophila from Chun 13