Quantitative mRNA imaging throughout the entire Drosophila brain

Improved fluorescence in situ hybridization enables smFISH in cleared whole-mount Drosophila brains with confocal microscopy; a custom Bessel beam structured illumination microscope allows single-mRNA detection across the entire brain. We describe a fluorescence in situ hybridization method that permits detection of the localization and abundance of single mRNAs (smFISH) in cleared whole-mount adult Drosophila brains. The approach is rapid and multiplexable and does not require molecular amplification; it allows facile quantification of mRNA expression with subcellular resolution on a standard confocal microscope. We further demonstrate single-mRNA detection across the entire brain using a custom Bessel beam structured illumination microscope (BB-SIM).

expressing neurons. PDF neurons are part of the circadian pacemaker network and consist of two groups, 4 large ventrolateral neurons (l-LNvs) and 4 small ventrolateral neurons (s-LNvs) in each hemisphere 10 . Brains labeled with FISH probes targeting GFP and PDF mRNAs in distinct colors and imaged on a confocal microscope revealed the characteristic two groups of l-LNvs and s-LNvs on each side of the brain. GFP-coding mRNA signal was not observed outside of the PDF neurons or in wild type flies, confirming specificity of the hybridization (Figure 1b).
To determine whether we could resolve changes in less abundant genes, we targeted Timeless (Tim), a transcription factor that cycles daily 11 and regulates the circadian cycle in Drosophila. We acquired confocal images of light-dark cycle entrained wild type files at zeitgeber times (ZT) 2 and 14, and we observed as expected a strong Tim signal in PDF neurons at ZT14, while the intensity was greatly diminished at ZT2 (Figure 1c, 1d and S2) 9,12 .
Because of its multiplexing potential, the technique lends itself to multiple applications. FISH can be a powerful characterization and validation tool for the rapidly expanding collections of genetic lines used in neuroscience and optogenetics. We selected three Gal4 lines (Mi1, Mi4, Mi9) 13 with known neurotransmitter expression and crossed them with UAS-myr::HaloTag to generate expression of a HaloTag reporter in the desired neurons 14 . We profiled those brains by labeling the HaloTag protein with its fluorescent ligand in a first color (Methods) and hybridizing FISH probes targeting genes involved in distinct neurotransmitter pathways in a second color (Gad1, vGlut and Chat, respectively associated with GABAergic, glutamatergic and cholinergic transmission). The overlap between the HaloTag and the FISH signals confirmed the expected neurotransmitter signatures ( Figure S3). We then interrogated the simultaneous expression of three genes associated with distinct neurotransmitter pathways. Resulting confocal stacks (Figure 1e, Mov S1) display non-overlapping spatial patterns of expression, suggesting minimal co-expression of these neurotransmitters. Such multiplexing experiments provide a rapid, powerful tool to link gene expression patterns to brain-wide architecture, function and connectomics patterns. mRNA localization has been proposed as a mechanism to regulate translation, specifically in neurons 15 . However, this model remains mostly untested in animals because single mRNA imaging in the large volumes relevant to neuronal connections in vivo is challenging. Hence, we sought to test the feasibility of our method to detect the location of single molecules in whole mount tissue. High-resolution imaging requires tissue transparency and matched refractive indices (RI) of the tissue and lens immersion media. To this end, we designed and built a specific Bessel Beam Selective Plane illumination microscope (SPIM) capable of structured illumination (SIM) ( Figure S4). The setup was engineered to image in media with a refractive index matched to the measured index of xylene-cleared Drosophila tissue (Supplementary Note 1). Excitation from two directions allows SIM resolution enhancement along both X&Y ( Figure  2a, Methods) 16 .
We first tested whether we could detect transcription sites in the whole mount brain. The multiple nascent pre-mRNAs at the locus of a transcribing gene typically generate a bright nuclear focus when imaged with FISH 4 . We imaged various lines differing in the location or number of genomic insertions of a GFP reporter gene (all under the control of a Pdf driver). In lines harboring a single insertion site (attp2 or attp18), we observed zero or one bright focus in each of the PDF neurons, but never observed two foci per nucleus (Figure 2b). In contrast, we frequently observed two foci per nucleus in a double insertion line (attp2; attp18). To further confirm our interpretation, we quantified the number of Tim and Pdf nuclear foci in s-LNvs, and observed a dramatic change (~5x) in the number of nuclear foci for Tim at ZT2 and ZT14 while the Pdf nuclear foci displayed little variation ( Figure S5). These results agree with previous findings that Tim transcription is reduced at ZT2, while Pdf transcription does not substantially change over the daily cycle 12 .
We next set out to detect single mRNA particles. Flies with the Pdf-Gal4 reporter expressing GFP in PDF neurons were hybridized with probes targeting the Pdf and the Gfp transcripts in separate colors, and high-resolution SPIM/SIM volumes containing the PDF neurons were acquired (Figure 2c). We detected diffraction-limited foci of the GFP-encoding mRNAs in the PDF neuron cell bodies. We could not resolve individual endogenous Pdf mRNAs within the cell bodies because they were too dense (left panel). Using an algorithm to characterize the individual spots 17 , the brightness of the mRNAs typically appeared as a unimodal intensity distributions above the fluorescent background ( Figure S6), consistent with individual mRNAs. A characteristic of single mRNA detection is that the intensity of detected mRNAs remains constant regardless of the expression level. To demonstrate this, we used a set of driver lines with increasing number of UAS elements upstream of the GFP reporter, generating increasing levels of expression in the PDF neurons 18 ( Figure S7). When imaging those brains at high resolution (Figure 2d), we counted as expected an increasing number of spots per cell with increasing number of UAS elements; importantly, the brightness of individual spots did not vary, confirming that we are detecting individual molecules ( Figure 2e).
Tissue thickness can decrease detection efficiency due to aberrations and scattering. To assess the impact of tissue depth on our detection performance, we measured the distribution of fluorescence intensities and spot sizes of Tim mRNAs imaged in different brain regions. Spot intensities and sizes were constant through the entire brain sample, suggesting that depth had a minimal impact on detection quality and that our setup achieves diffraction-limited performance throughout the brain volume (Figure 2f,g, S6g).
We demonstrated the potential of our technique by quantifying the absolute expression of Tim in PDF neurons at ZT2 and ZT14 (Figure 3a,b). These results are in good agreement with confocal measurements of total intensity (Figure 1d and S2) and provide the basis for absolute gene expression studies at the scale of the entire brain.
While most mRNAs localize to neuronal cell bodies, we also detected molecules in processes (Figure 3c-f). The mRNA levels decreased with increasing distance from the cell bodies (similar to mammalian neurons in culture 15 ) in a cell type specific manner suggesting distinct architectures or transport regulation (Figure 3f).
The whole-mount smFISH technology enables the interrogation of gene expression levels at the single-cell level using a standard confocal microscope. Because of its simplicity and versatility (multiplexable, compatible with HaloTag labeling and immunofluorescence), we anticipate it will constitute an important tool for addressing the cellular and molecular basis of brain function. Furthermore, when combined with high-resolution microscopy, it is able of detecting and counting individual molecules within an intact entire brain. This permits visualizing neuronal architecture and connections simultaneously with the levels and localization of mRNA expression. Combined with the powerful genetic tools inherent to Drosophila, this technique constitutes a unique tool to address questions such as the role of mRNA localization in memory formation. Expansion of the technique, e.g. multiplexing many more mRNAs 7,19 , will further exploit its potential.
FISH. Amino-labeled oligonucleotide probes 5 (Biosearch Technologies) were labeled to NHSester fluorophores. To perform FISH, rehydrated Drosophila brain tissues were exposed to 5% acetic acid at 4°C for 5 min. After being fixed in 2% paraformaldehyde for 55 min at 25 °C, the tissues were incubated in 1xPBS with 1% of NaBH 4 at 4°C for 30 min, followed by a 2 hr incubation in pre-hybridization buffer (15% formamide, 2x SSC, 0.1% triton X-100) at 50 °C. The brain tissues were transferred to 50uL of hybridization buffer (10% Formamide, 2xSSC, 5xDenhard's solution, 1mg/ml Yeast tRNA, 100ug/ml, salmon sperm DNA, 0.1% SDS) with FISH probes (50-100ng/uL per reaction, containing probe sets against one or multiple genes) and incubated at 50 °C for 10 hr, followed by an additional 10 hr incubation at 37 °C. After a series of wash steps, the brain tissues were dehydrated and introduced to xylene for tissue clearing. A detailed description of the FISH protocol is included in the supplementary information.
HaloTag staining and IHC. The brains of flies expressing the HaloTag reporter were stained with HaloTag-JF646 during fixation. After dissection, brain tissues were transferred to 2% paraformaldehyde containing 2uM of HaloTag-JF646 20 . They were then incubated with agitation for 55 min at 25 °C, followed by a series of 1xPBS washes. Antibody staining was performed after the FISH labeling. After a series of wash steps, the brain tissues were blocked with 10% normal goat serum in PBT at room temperature for 2 hr, followed by overnight primary antibody staining (Rabbit polyclonal anti-GFP Fraction, Life Technologies A11122, 1:1000 dilution with 5% NGS) at 4°C. After washing the primary antibody, brain tissues were incubated in secondary antibody (AF488 Goat anti-rabbit, Invitrogen A11034, 1:1000 dilution with 5% NGS) overnight at 4°C. Brain tissues were then fixed in 2% paraformaldehyde for 55 min at room temperature before dehydration and xylene clearing.
Confocal Imaging. For confocal imaging, all brain tissues were attached to poly-L-lysine coated cover glass and mounted in DPX (Janelia Adult Drosophila CNS DPX mounting protocol). We used a Zeiss LSM 880 confocal microscope (561 nm and 633 nm laser lines) imaged through a 25X DIC NA 0.8 oil objective to detect GFP transcripts in Pdf-Gal4;JFRC5 constructs and wild type flies (Figure 1). Images were acquired with sequential excitation as stacks with 1.5 um z-spacing. Detector gain and laser power were kept constant for all samples. To detect Gfp transcripts in the UAS titration experiments, we used a Zeiss LSM 800 confocal microscope (561 nm and 633 nm laser lines) imaged through a 63 X DIC NA 1.4 oil objective. Images were acquired with sequential excitation as stacks with 1.0um z-spacing. Detector gain and laser power were kept constant for all samples. To detect the overlapping neurotransmitter expression in Gal4 lines, we used a Zeiss LSM 880 confocal microscope (561 nm and 633 nm laser lines) imaged through a 63X DIC NA 1.4 oil objective. Images were acquired with sequential excitation as stacks with 1.00 um z-spacing. The multiplex neurotransmitter images were acquired on a Zeiss LSM 880 confocal microscope (488 nm, 561nm and 633 nm laser lines) imaged through a 25 X DIC NA 0.8 oil objective. Images were acquired with hyperspectral lambda detection scan as stacks with 1.00 um z-spacing. Summary of imaging parameters are in Supplementary Table 1.
SPIM/SIM Imaging. To generate the SPIM/SIM images, the brain tissue was mounted on a 1.5x3mm poly-L-lysine coated coverslip attached to the end of a 30mm glass rod. The objectives and sample were all immersed in the imaging medium. The images in this work were acquired with a NA Bessel max of 0.62, which creates a Bessel beam with a central peak width 0.35µm for 560 nm illumination. 11 phases were collected for each excitation direction. The exposure time for each image was 20 msec. The lateral field of view, which is determined by the length of the Bessel beam, was 35x35µm. Z stacks up to 24µm were acquired by moving the sample with a piezo stage with closed loop position control. Larger volumes were covered by tiling, moving the sample with piezo inertia drives in XY&Z. For more details see Supplementary Note 1, Figure S4 and

Figure S1
Figure S1 | Simultaneous imaging of Pdf mRNA (green) and GFP protein (purple) in wholemount tissue. Maximum intensity projection of confocal stack of a Pdf-Gal4; 5xUAS-IVS-mCD8::GFP fly brain. Pdf mRNA (labeled with FISH probes) accumulates mainly in the cell bodies of PDF neurons identified by the GFP protein signal (immunohistochemistry with antibodies against GFP, see Methods).      Beams from three lasers (488 nm, 561 nm, 640 nm) are expanded to a diameter of 3 mm, and vertical polarization set with a half wave plate. This combined beam is sent through an Acousto-optic Tunable Filter (AOTF, AA Optoelectronic) and beam expander. From the beam expander, the beam passes through an annular apodization mask (made in house using a laser mill to ablate the desired pattern on a Thorlabs Neutral Density filter) and a pair of galvo mirrors (Cambridge Technology) that allow lateral and axial positioning of the beam. The combination of a liquid crystal variable waveplate (Edmund Optics) and polarizing beamsplitter allows the beam to be directed to either excitation objective; this diagram shows the beam for just one path. The annular mask, X galvo, Z galvo, and back aperture of the excitation objectives are all at conjugate planes, so beam position is constant at those points. When imaging, the illumination plane and detection objective are fixed, and the sample is moved to image through Z to create a volume.   Background detections appear as a low intensity spots while mRNAs accumulate in a distinct high-intensity distribution. The fit to a 2-component lognormal distribution (purple) provides the respective contributions of background (red) and mRNAs (blue) to the overall histogram. Bottom: Jaccard Coefficient quantifying the similarity between the detected spots and the expected mRNA counts, as a function of an arbitrary intensity threshold (the Jaccard calculation is based on the 2-component lognormal fit, see Supplementary Note). The Jaccard Coefficient provides a metric of detection accuracy taking into account both sensitivity and selectivity; it reaches a maximum for the intensity threshold that separates optimally the background from mRNA signal. The maximum value indicates that the spots detected at the optimum are ~90% similar to the original mRNAs. g) Intensity and sizes of Tim mRNA spots as a function of depth. The fluorescence intensities and spot sizes of Tim mRNAs detected in panel c do not change as a function of depth. Each circle represents one detected spot; the line is a sliding average.

Figure S7
Figure S7 | Top: comparison of Gfp mRNA fluorescence intensity per cell in fly lines using confocal microscopy. The GFP reporter is driven by 3, 5, or 10 UAS repeats, resulting in a gradual increase in expression of the Gfp mRNA. In contrast, the endogenous Pdf mRNA levels do not increase with increasing UAS number (bottom). The colors indicate the two different PDF neuron types. Each circle corresponds to one cell; Mean and Standard Deviation are represented next to each bee swarm plot; 5 brains were imaged for each condition. Figure)

Supplementary Note 1: BB/SIM microscope
The SPIM/SIM microscope is closely related to the Bessel Beam Plane illumination microscope described by Planchon, et al. 1 and Gao,et al. 4 , which uses a Bessel beam, with a very narrow central peak (0.3 m) and long axial extent (30-50 m), swept or stepped to illuminate a thin sheet in the sample. As is generally true for plane illumination microscopy, the lateral resolution is limited by the lateral resolution of the detection objective, while the axial resolution is largely determined by the thickness of the light sheet. Stepping the beam creates a virtual grating: stripes perpendicular to the scan direction. A set of images collected with different positions of the grating can be analyzed using Structured Illumination (SI) analysis 3,4 to extract high frequency information and reconstruct the image with lateral resolution twice that of the detection objective. However, that resolution enhancement is limited to the direction perpendicular to the grating. Imaging whole mount Drosophila brains with nearly isotropic resolution requires two new developments: creation of optically clear samples and illumination patterning in two directions for SI resolution enhancement in both X&Y. As shown in Figure 1, the microscope therefore has three objectives: the imaging objective plus two orthogonally mounted excitation objectives. For aberration free imaging, the system must be designed for media with refractive index precisely matched to the cleared tissue.
The first step is development of an immersion fluid precisely matched to the refractive index of xylene cleared Drosophila tissue. Dehydration of tissue followed by treatment with xylene or other high index solvents to create optically clear samples is an old technique (see, for example 2 ). Such samples are usually mounted in a solid medium between a slide and coverslip. These samples are durable, but the geometry does not allow sufficient optical access to allow plane illumination, and the refractive index of the material surrounding the tissue is poorly controlled. A liquid imaging medium allows tuning of the refractive index to match the tissue and a dipping imaging geometry compatible with selective plane illumination.
The refractive index of the tissue was determined by imaging known targetsfluorescent beads-through cleared tissue in test media using a standard confocal microscope. Refractive index mismatch between the tissue index and the surrounding medium creates lensing effects which can be measured in the bead images. The composition of the test medium was varied systematically to minimize effects on the imaging. Best results were achieved with the refractive index of the medium at 1.5525. The other requirements for the imaging medium are: miscibility with xylene, low fluorescence, and relatively low vapor pressure. To meet these, the system is designed for imaging in a mixture of 90% 1,2-dichlorobenzene and 10% 1,2,4-trichlorobenzene.
The detection objective and two excitation objectives are custom designed (Special Optics/Navitar) with: working distance = 3.0 mm; design NA = 1.0 (measured NA 0.9-0.93); optical match to the index of the medium; small physical package allow 3 objectives to be mounted orthogonally with a common focal point; and solvent resistance. To allow adequate optical access to the sample, the tissue is mounted on a 1.5x3mm coverslip attached to the end of a 30 mm glass rod which holds the sample at the focus of the three objectives. The objectives and sample are all immersed in the imaging medium. A simplified optical layout of the system is given in supplementary figure M1.
The control electronics and software are identical to those described in Planchon, et al 1 . and Gao,et al. 4 , with the addition of control of the liquid crystal variable retarder to switch between beamlines. Images are collected by stepping the excitation beam along one axis with the camera exposing through all steps. The step size is (((N-1)/2)*/(2*NA Bessel max )) where N is the number of phases to be collected. That image is saved, and the imaging is repeated for all phases, with the pattern phase shifted by 2/N relative to the last. This process is repeated using excitation through the second beamline, which creates an image with patterning in the orthogonal direction. The sample is then moved in z, and the next plane imaged in the same way. For two color images, the full set of phases is collected for first color, then a full set of phases for the second color, before moving the sample and imaging the next plane. The two volumes created by imaging with X and Y modulation can show 100-200 nm displacement in z from each other, due imperfect alignment of the beams and beam deflection due to residual index variation in the tissue. Therefore, the "X image" is aligned in Z to the "Y image" before combining the two directions in the structured illumination analysis. The structured illumination analysis algorithm is detailed in Gustafsson,et al. 3 ; modifications specific to Bessel beam imaging are discussed in Gao, et al 4 . The reconstructed images in different colors are registered to each other by applying x, y and z offsets measured on beads labeled with both colors.
The images in this work were acquired with an excitation aperture NA Bessel max of 0.62, which creates a Bessel beam with a central peak width 0.35 m with 560 nm illumination. 11 phases were collected for each excitation direction. The exposure time for each phase is 20 msec; collection of one z slice in one color (22 images) takes 1.5 sec. PSF measurements with these conditions on 150 nm beads yield a lateral fwhm of 0.20 m, and axial fwhm of 0.29 m. The XY field of view, which is limited by the axial extent of the Bessel beam, was 35x35m. Z stacks up to 24 m are acquired by moving the sample with a piezo stage with closed loop position control (Physik Instrumente P-753). Larger volumes are covered by tiling, moving the sample with piezo inertia drives (Physik Instrumente, LPS-45) in XY&Z. Light collected by the detection objective is imaged through a 400 mm achromat (Edmund) onto the camera (Hammamatsu Orca 4.0).

Supplementary Note 2: Stepwise Protocol for whole-mount RNA FISH of Drosophila adult brain
Probe Library Design and Labeling FISH probe libraries were designed based on transcript sequences using the online Stellaris Designer and purchased from Biosearch Technologies. Libraries for each gene typically consist of ~50 probes, but we were able to detect mRNAs with as little as 20 probes (e.g. in the case of Pdf) when the mRNA length or sequence composition did not permit designing 50 probes; increasing the number of probes increases the signal to noise ratio. Each probe is 18-22nt long with a 3' end amine-modified nucleotide that we directly couple to an NHS-ester dye according to the manufacturer's instructions (Life Technologies). Excess dyes were added to the reaction to ensure all probes were coupled to dyes. To separate the free dyes from dye-coupled oligos, we used the Qiagen Nucleotide Removal Columns. This approach yields 85-100% of dyelabeling efficiency. We validated our approach using HPLC purification. Probe library sequences are listed in Supplementary  Mount tissue on glass coated with poly-L-lysine  Dehydrate with graded EtOH series: 30%, 50%, 75%, 100%, 100%, 100%, 10 min each  Clear in xylene, 3X, 5 min each  For confocal, mount in DPX (Electron microscopy sciences) For BB-SIM, hold in xylene until imaging.

Characterization of Single mRNA detection Sensitivity
In order to quantitatively assess the sensitivity and selectivity of our technique, we imaged zstacks of PDF neurons (wild type flies, ZT14) labeled against Tim mRNA (Supplementary Figure  S6a-e). We computed the fluorescence intensity distribution of detected spots in our z-stack using an intentionally low threshold to ensure that some of the brighter background pixels were included in the analysis. The resulting intensity distribution of detected particles is bimodal ( Supplementary Fig. S6f), consisting of low-intensity background spots, and high intensity spots corresponding to the individual mRNAs. The bimodal intensity distribution is well fit by a sum of two lognormal distributions ( Supplementary Fig. 6f, Equation 2); lognormal is the expected distribution resulting from the fluorescence intensity of individual molecules 10 .
Using these estimates, we can quantitatively assess the similarity between the detected spots ( ( ℎ ℎ ) + ( ℎ ℎ )) and the desired distribution (0). We use the Jaccard Coefficient as a metric 11 ; the Jaccard Coefficient quantifies the similarity between two ensembles and as = ∩ ∪ . In our case: = ( ℎ ℎ ) (0)+ ( ℎ ℎ ) As expected, the Jaccard Coefficient reaches a maximum when the chosen intensity threshold is positioned at the shoulder between the two components of the bimodal distribution. The maximum value is excellent ( = 0.9 ; ~3% false positive detections; ~7% missed mRNAs; Supplementary Fig6f), indicating that the imaging and analysis provide an accurate absolute count of the number of individual mRNA molecules. As observed across the whole brain, the size or intensity of individual spots does not substantially vary with depth within the collected zstack ( Supplementary Fig. 6g).

Quantification of Pdf mRNA localization along processes
In order to quantify the density of Pdf mRNAs along processes, we co-labeled brains with FISH probes against the Pdf gene and a HaloTag fluorescent ligand against a reporter myr::HaloTag fusion expressed in the PDF neurons. We acquired an initial low-resolution BB-SIM stack of the entire brain to provide the entire anatomical context (Fig. 3c), and then a second, highresolution stack, centered on the PDF neurons (Fig. 3d). We used the myr::HAloTag signal to trace processes that initiated from PDF neurons, using the Simple Neurite Tracer plugin in Fiji 5 .
Only processes which origin could be unambiguously traced to a PDF neuron were considered.