An adaptive optics module for deep tissue multiphoton imaging in vivo.

Animals. All animal experiments were conducted according to the National Institutes of Health guidelines for animal research. Procedures and protocols on mice and zebrafish were approved by the Institutional Animal Care and Use Committee at Janelia Research Campus, Howard Hughes Medical Institute; and the Animal Care and Use Committee at the University of California, Berkeley. Mice were housed in cages in groups of 1–5, under a normal light cycle, at a temperature of about 73 °F and humidity of 37%. Both males and females were used in this study, with ages ranging from 5 to > 30 weeks old, and included (Jackson Laboratories): Thy1-YFP-H, Thy1-GFP-M, Gad2-IRES-Cre X Ai14, C57BL/6J. Details on animal preparations are available below. Excitation Source. The 2P excitation source was a mode-locked titanium:sapphire laser (Chameleon Ultra II; Coherent) operating at 920 nm. The 3P excitation source consisted of a two-stage optical parametric amplifier (Opera-F; Coherent) pumped by a 40 W diode-pumped femtosecond laser (Monaco 1035–40-40; Coherent) operating at 1035 nm and 1 MHz, providing a broad tuning range (650–920 nm and 1200–2500 nm). Opera-F was operated at 1300 and 1700 nm for 3P excitation, for which the average output power was ~1.5 and ~0.9 W (1.5 and 0.9 μJ per pulse at 1 MHz repetition rate), respectively. For 1300 nm excitation, to reduce the group delay dispersion (GDD) at the sample plane, we used a homebuilt single-prism compressor 28 . After compensation, the pulse duration at the focal plane of the objective was measured to be ~54 fs using an autocorrelator (Carpe, APE GmbH). For 1700 nm excitation, since the GDD is anomalous for many of the glasses and crystals used in our microscope, the resulting negative GDD at the sample plane cannot be compensated for by using our prism-based compressor. Instead, the high normal dispersion of ZnSe (from a bulk compressor available inside Opera-F) and silicon (from a 3-mm thick window placed at Brewster’s angle 29 ) was used to obtain a pulse duration at the sample of ~70 fs after compensation. Adaptive optical microscope setup. Simplified diagrams of our homebuilt 2P microscope and 3P microscope are shown in Extended Data Fig. 2 . For the 2P microscope ( Extended Data Fig. 2a ), a pair of achromat doublets (AC254–200-B and AC254–300-B; Thorlabs) conjugated the segmented deformable mirror (DM) surface to a liquid-crystal spatial light modulator (SLM; Holoeye, PLUTO-NIR) used to introduce artificial aberrations. The SLM plane was then conjugated to a pair of galvanometers (6215H; Cambridge Technology) that were optically conjugated to each other and the back focal plane of a high-numerical aperture (NA) water-dipping objective (Olympus XLPLN25XWMP2, NA 1.05, 25× for mouse brain imaging; or Nikon CFI LWD, NA 0.8, 16× for zebrafish imaging), using three pairs of achromat doublets (AC254–150-B and AC254–60-B, AC508–080-B and AC508–080-B, AC508–75-C and SLB-50–600PIR1; Thorlabs and OptoSigma). For the 3P microscope ( Extended Data Fig. 2b ), femtosecond pulses at 1300 or 1700 nm were reflected off a segmented deformable mirror (DM). The DM was conjugated to a pair of galvanometers (6215H; Cambridge Technology) that were optically conjugate to each other and the back focal plane of a high- NA water-dipping objective (Olympus XLPLN25XWMP2, NA 1.05, 25×), using three pairs of achromat doublets (Original 3P system: AC254–100-C and 45–804, AC508–080-C and AC508–080-C, AC508–100-C and SLB-50–600PIR2; Thorlabs, Edmund Optics, and OptoSigma. New 3P system: AC254–400-C and AC254–300-C, SL50–3P and SL50–3P, SL50–3P and TTL200MP; Thorlabs). Here, the DM image underfilled the back aperture of the objective, leading to an effective NA of ~0.9, corresponding to a lateral and axial resolution of ~0.6 μm and ~2.3 μm, respectively, at 1300 nm. The deformable mirror in this project is a hexagonal tip-tilt-piston DM (Hex-111-X) manufactured by Boston Micromachines Corporation ( Supplementary Fig. 9 , Supplementary Note 1 ). For both the 2P and 3P microscopes, a field stop (iris diaphragm, Thorlabs) was located at the intermediate image plane between the DM and the X galvo to block unwanted diffraction orders and light reflected off mirror segments at large tilt angles. To translate the focus axially, the objective was mounted on a piezoelectric stage (P-725.4CD PIFOC; Physik Instrumente). The fluorescence signal was collected by the same objective and reflected from a dichroic beam splitter (FF665-Di02–25×36; Semrock), spectrally filtered (2P: FF01–680/SP, Semrock. 3P: FF01–680/SP, Semrock; together with BLP01–442R-25, Semrock, or ET575lp, Chroma, to block the third-harmonic generation signal at 1300 and 1700 nm, respectively), and detected by a photomultiplier tube (H7422–40 or H10770PA-40; Hamamatsu). A Pockels cell was used for controlling the excitation power (2P: M350–80; 3P: M360–40; Conoptics). For the experiments using 1700 nm excitation, we used D 2 O, instead of H 2 O, for the objective immersion media, because of its much lower absorption 29 . Custom-written software was used for image acquisition. Aberration measurement method. The general aberration measurement procedure was conceptually similar to that described in our previous work 10 . Experimentally, the laser focus was parked at one sample location and the fluorescence signal from this location was used for aberration measurement. The pupil was segmented to 37 regions, corresponding to the number of segments in the DM. The 37 segments were separated into two groups of alternating rows ( Extended Data Fig. 1 ). First, we fixed the tip, tilt, and piston of one group of 17 segments, but added to each of the other 20 pupil segments a specific tip angle Θ i and tilt angle Φ j ( i , j = 1,2,… n ) ( Extended Data Fig. 1 , Step 1), each of which were chosen randomly from an array of n angles evenly spaced between −Ψ/2 and Ψ/2. These applied tip and tilt angles caused the beamlet reflecting off the segment to displace along the X and Y axes in the objective focal plane by X i and Y j (X i = f* tan(2Θ i / M ) and Y j = f* tan(2Φ j / M ); f : focal length of the objective; M : magnification from the DM to objective back focal plane), respectively, which changed their interference with the reference focus formed by the other 17 beamlets. With the tip and tilt angles of all segments fixed, using the segments themselves we then modulated the phase or intensity of all 20 beamlets ( Supplementary Note 2 ), each at a distinct frequency ω s ( s = 1,2,…20), and recorded the fluorescence signal for a time duration T ( Extended Data Fig. 1 , Step 2). The recorded signal trace was then Fourier transformed (FT) and the Fourier magnitudes at each distinct modulation frequency ω s were measured ( Extended Data Fig. 1 , Step 3), whose values indicated how much individual beamlets interfered with the reference focus at the focal displacement of (X i , Y j ). The above procedure was repeated n × n times, so that all 20 segments sampled the full tip and tilt angles and their corresponding beamlets scanned over a 2D grid in the focal plane with the dimensions 2 f* tan(Ψ/ M ) by 2 f* tan(Ψ/ M ) ( Extended Data Fig. 1 , Step 4). For each beamlet, plotting the Fourier magnitudes versus the displacements (X i , Y j ), we constructed a 2D map of interference strength of this beamlet with the reference focus at different focal displacements ( Extended Data Fig. 1 , Step 5). Fitting the map with a 2D Gaussian function, we found the displacements leading to maximal interference between the beamlets and their reference focus, which gave us the tip and tilt angles (i.e., phase gradient) to be applied to this segment in the corrective wavefront. We then repeated Steps 1–5, but now with the group of 20 pupil segments fixed and forming the reference focus. Modulating the remaining 17 segments, we obtained their 2D maps of interference strength versus displacement ( Extended Data Fig. 1 , Step 6) and the phase gradients required to shift the corresponding beamlets to coincide with the reference focus. The total fluorescence signal acquisition time was 2 × n × n × T. With all the beamlets intersecting at the same location, the next step was to determine the phase offsets that would allow them to constructively interfere 30 . Consider the case of finding the phase offset that enables two beamlets to constructively interfere. With one beamlet assigned as the reference (with unknown phase θ r ), by incrementally adding a phase offset Δ θ to the other beamlet (with unknown phase θ 1 ) at step size ω (i.e., Δ θ = ω t), the intensity variation can be described by I = 2 + 2cos( ω t + θ 1 − θ r ). The phase offset that gives the maximal intensity (i.e., constructive interference) is thus the opposite of the phase of the function cos( ω t + θ 1 − θ r ). One approach is by Fourier-transforming the time-dependent signal and reading out the phase at frequency ω /2 π . To determine the phase offsets of 37 segments, we employed the concept of multidither coherent optical adaptive technique 11 , 31 – 33 . We modulated the phases of the first 20 rays by piston-displacing each corresponding mirror segment at a distinct frequency ω s while keeping the phases of the remaining rays, which formed a reference focus, constant ( Supplementary Note 3 ). We then Fourier transformed the recorded fluorescence trace ( Extended Data Fig. 1 , Step 7), and read out the phase offsets that would lead to constructive interference with the reference focus at the modulation frequencies ω s ( s = 1,2,…20) ( Extended Data Fig. 1 , Step 8). We next modulated the phases of the remaining 17 segments while keeping the phases of the first 20 segments unchanged and found the phase offsets for these 17 segments required to ensure constructive interference among beamlets. With all beamlets intersecting and constructively interfering at the focal plane, we obtained the final corrective wavefront and applied it to the DM for aberration correction. Because the reference foci used for both phase gradient measurements ( Extended Data Fig. 1 , Step 1–6) and phase offset measurements ( Extended Data Fig. 1 , Step 7–9) were aberrated to begin with, for larger aberrations, the whole procedure was iterated, as needed, to achieve optimal aberration correction ( Extended Data Fig. 1 , Step 10). Typical operation parameters and imaging considerations. An example set of operation parameters used for aberration measurement in the mouse brain in vivo ( Fig. 2 ) included an integration time of T = 90 ms for each of the 11 × 11 ( n × n ) tip and tilt angles, which scanned the modulated rays over 19 μm × 19 μm 2D grids in the focal plane. The overall fluorescence acquisition time was therefore 21.8 s (2 × n × n × T). Additionally, for the phase measurement portion of the algorithm, the total fluorescence acquisition time was 1.8 s (360 ms per iteration, with a total of 5 iterations). Additional hardware (e.g., DM settling time) and software overheads added to the fluorescence acquisition time and the overall time used for aberration measurement and correction was 3–4× the fluorescence acquisition time. For the data in Fig. 2 , 3 rounds of aberration measurements were performed to obtain the final corrective wavefront. However, our method does allow for considerably shorter fluorescence acquisition times for the gradient measurement step without affecting performance, because the process of determining the location of centroids in the gradient measurement step is robust against noise. For cranial window induced aberrations, reducing the integration time for each tip and tilt angle by 8.3× (from 90 ms to 10.8 ms) and the total sample illumination time by 5.3× from the typical parameters used in this manuscript yielded the same correction performance ( Supplementary Fig. 6 and Supplementary Table 3 ). “No AO” images were taken after system aberration correction ( Extended Data Fig. 2c – e ), as well as the adjustment of the objective correction collar to compensate for the spherical aberration introduced by the glass windows overlaying the mouse brain and spinal cord. Furthermore, to minimize additional aberration modes that arose from a tilted window and sample (e.g., coma) 34 , we used the third-harmonic generation signal from the window-tissue interface to determine the angle of the window ( Supplementary Fig. 10 ) and adjusted the mouse using a 2D goniometer stage till the window was perpendicular to the excitation beam. These procedures constitute the best practice and lead to the best performance that conventional optics could achieve. The deterioration in image signal and contrast observed in our “No AO” images, therefore, resulted from tissue-induced aberrations exclusively, and could only be compensated by AO. Digital image processing. Due to brain motion at depth, the StackReg 35 image registration plug-in in ImageJ was used for rigid registration in 2D. The “smooth” function from ImageJ that replaced the value of each pixel with the average of 3 × 3 pixels centering on this pixel was applied to all 3P images. 2P images were presented using the “Green hot” lookup table in ImageJ; 3P images were presented with the “Green hot” and “Magenta hot” lookup tables for 1300 and 1700 nm excitation, respectively. For images where the signal was too weak before AO correction, a linear scaling factor was applied to all pixel values to improve visibility with the scaling factor listed on the image and figure legend. The images presented did not undergo any other digital manipulation. For calculating the spatial frequency space representations of fluorescence images, a Gaussian Blur filter (ImageJ) with blur radius σ = 1 pixel was first applied to the corresponding fluorescence image to eliminate pixelation artifacts that would otherwise show up as high spatial frequency components of non-negligible amplitude (sometimes past the diffraction limit). The spatial frequency space representations of fluorescence images are shown in logarithmic scale. Bead samples. Carboxylate-modified fluorescent microspheres (Fluosphere™; Invitrogen) were immobilized on poly(l-lysine)-coated microscope slides (12–550-12, Fisher Scientific). Zebrafish preparation. Zebrafish procedures have been described previously 36 . Briefly, Tg(β-actin:HRAS-EGFP) zebrafish embryos ( Danio rerio ) were grown at 28 °C in E3 zebrafish embryo medium. To generate optically transparent embryos, melanin synthesis was inhibited by transferring the embryos into 1× phenylthiourea (PTU) solution in E3 medium, 10–16 hours post fertilization. Before imaging, the chorions were manually removed with forceps under a stereomicroscope. Four-day-old larvae were then anesthetized in E3 medium containing 1× tricaine and immobilized on a dish by embedding them in 0.5% low–melting point agarose with 1× PTU and 1× tricaine. During imaging, E3 medium containing 1× PTU and 1× tricaine was used as immersion medium. Mouse preparation (brain imaging). Cranial window implantation procedures were performed 37 , using aseptic technique, on mice that were anaesthetized with isoflurane (1–2% by volume in O2) and given the analgesic buprenorphine (SC, 0.3 mg per kg of body weight). For 2P imaging experiments, a 3.5-mm diameter craniotomy was made over V1 with dura left intact. A glass window made of two coverslips (Fisher Scientific, thickness no. 1.5) bonded with ultraviolet cured optical adhesives (Norland Optical Adhesives 61) was embedded in the craniotomy and sealed in place with dental acrylic. For 3P imaging experiments, a 5-mm diameter craniotomy was made over V1, with dura left intact. The glass window consisted of a donut-shaped coverslip (inner diameter 4.5 mm, outer diameter 5.5 mm; Potomac Photonics) bonded with ultraviolet cured optical adhesives (Norland Optical Adhesives 61) on top of a 5-mm diameter coverslip (Denville Scientific, thickness no. 1). The window was embedded in the craniotomy and sealed in place with dental acrylic. A titanium head-post was attached to the skull with cyanoacrylate glue and dental acrylic. Acute imaging was performed an hour after the surgery; chronic imaging happened at least one week after the surgery. Mice were head-fixed and anesthetized using isoflurane (1–2% by volume in O2) during imaging. Some mice also underwent virus injection procedures as described previously 37 . For the examples shown in Extended Data Fig. 9 , neurons in the hippocampus were infected with a mixture of AAV-Syn-Cre (10× dilution from 1.8 × 10 13 GC/mL) and AAV-CAG-FLEX-tdTomato (3.3 × 10 13 GC/mL) in a wild-type mouse (C57BL/6J). Three injection sites were chosen (AP: −1.7 mm, ML: +1.5 mm; AP: −2.0 mm, ML: +2.0 mm; AP: −2.3 mm, ML: +2.5 mm), at five different depths (0.6, 0.8, 1, 1.2, and 1.4 μm). 100 nL of viral solution were injected at each spot. A cranial window was implanted, as described above, 14 days after virus injection and acute imaging was then performed. Mouse preparation (spinal cord imaging). Acute spinal cord windows were prepared as described previously 12 . Briefly, mice were anesthetized with two successive intraperitoneal injections of 1mg/kg body weight urethane each, 30 mins apart. A tracheotomy was performed and mice were intubated to prevent asphyxiation. The T11–13 vertebrae were exposed and stabilized using spinal clamps (STS-A, Narishige). A dorsal laminectomy was performed at T12 to expose the spinal cord. After a wash with Ringer solution (135 mM NaCl, 5.4 mM KCl, 5 mM HEPES, 1.8 mM CaCl 2 , pH 7.2), the spinal cord was covered with a glass window made of a single coverslip (Fisher Scientific No. 1.5). The window and the surrounding custom imaging chamber were stabilized using 2% agarose in Ringer solution. 20 mL/kg body weight of 0.9% physiological saline is administered subcutaneously after the surgery, to ensure hydration during the imaging session. Blood flow through the central blood vessel was continuously monitored throughout the imaging experiment to ensure tissue health. For the functional imaging experiments, wild-type (C57BL/6J) mice had previously been intrathecally injected with ~5 × 10 10 GC of AAV8-Syn-jGCaMP7s. Animals were euthanized at the end of the imaging session. Temperature stimulation (spinal cord imaging): The left hind limb of the mouse was dehaired and gently fixed inside a custom-designed stimulation device as previously described 12 . The entire limb was continuously exposed to a homogeneous high flow-rate water flow of variable temperature. For each trial, the spinal cord was imaged for a total of 70 s. The first ~10 s were imaged at 31°C baseline water to obtain the baseline fluorescence and noise. For the following ~30 s the flow was switched to water that was pre-incubated at colder temperatures with the same flow rate. For the remaining ~30 s the flow was switched back to the 31°C baseline water. At least another 120 s passed before the next trial was performed. The actual temperature inside the stimulation chamber was monitored and recorded using a microprobe thermometer (BAT-12, Physitemp) with a Type-K thermocouple placed right next to the mouse’s limb, simultaneously with the fluorescence images. The electric valves controlling the water flow switch were triggered by and synchronized with the image acquisition system. Analysis of calcium imaging data. To account for motion in the spinal cord caused by respiration, we processed the image sequences using an iterative cross-correlation-based registration algorithm 37 . We averaged across trials, manually selected the region of interest, and calculated the mean fluorescence within this region. From this, we calculated the fractional change in 3P fluorescence ( ΔF/F 0 ) due to neural activity, with F 0 being the baseline fluorescence calculated as the mean fluorescence during exposure to the baseline temperature (initial ~10 s of the temperature stimulation). For the traces shown in Fig. 3j , we performed a 5-frame moving average.

Aberration measurement method. The general aberration measurement procedure was conceptually similar to that described in our previous work 10 . Experimentally, the laser focus was parked at one sample location and the fluorescence signal from this location was used for aberration measurement. The pupil was segmented to 37 regions, corresponding to the number of segments in the DM. The 37 segments were separated into two groups of alternating rows ( Extended Data Fig. 1 ). First, we fixed the tip, tilt, and piston of one group of 17 segments, but added to each of the other 20 pupil segments a specific tip angle Θ i and tilt angle Φ j ( i , j = 1,2,… n ) ( Extended Data Fig. 1 , Step 1), each of which were chosen randomly from an array of n angles evenly spaced between −Ψ/2 and Ψ/2. These applied tip and tilt angles caused the beamlet reflecting off the segment to displace along the X and Y axes in the objective focal plane by X i and Y j (X i = f* tan(2Θ i / M ) and Y j = f* tan(2Φ j / M ); f : focal length of the objective; M : magnification from the DM to objective back focal plane), respectively, which changed their interference with the reference focus formed by the other 17 beamlets. With the tip and tilt angles of all segments fixed, using the segments themselves we then modulated the phase or intensity of all 20 beamlets ( Supplementary Note 2 ), each at a distinct frequency ω s ( s = 1,2,…20), and recorded the fluorescence signal for a time duration T ( Extended Data Fig. 1 , Step 2). The recorded signal trace was then Fourier transformed (FT) and the Fourier magnitudes at each distinct modulation frequency ω s were measured ( Extended Data Fig. 1 , Step 3), whose values indicated how much individual beamlets interfered with the reference focus at the focal displacement of (X i , Y j ). The above procedure was repeated n × n times, so that all 20 segments sampled the full tip and tilt angles and their corresponding beamlets scanned over a 2D grid in the focal plane with the dimensions 2 f* tan(Ψ/ M ) by 2 f* tan(Ψ/ M ) ( Extended Data Fig. 1 , Step 4). For each beamlet, plotting the Fourier magnitudes versus the displacements (X i , Y j ), we constructed a 2D map of interference strength of this beamlet with the reference focus at different focal displacements ( Extended Data Fig. 1 , Step 5). Fitting the map with a 2D Gaussian function, we found the displacements leading to maximal interference between the beamlets and their reference focus, which gave us the tip and tilt angles (i.e., phase gradient) to be applied to this segment in the corrective wavefront. We then repeated Steps 1–5, but now with the group of 20 pupil segments fixed and forming the reference focus. Modulating the remaining 17 segments, we obtained their 2D maps of interference strength versus displacement ( Extended Data Fig. 1 , Step 6) and the phase gradients required to shift the corresponding beamlets to coincide with the reference focus. The total fluorescence signal acquisition time was 2 × n × n × T. With all the beamlets intersecting at the same location, the next step was to determine the phase offsets that would allow them to constructively interfere 30 . Consider the case of finding the phase offset that enables two beamlets to constructively interfere. With one beamlet assigned as the reference (with unknown phase θ r ), by incrementally adding a phase offset Δ θ to the other beamlet (with unknown phase θ 1 ) at step size ω (i.e., Δ θ = ω t), the intensity variation can be described by I = 2 + 2cos( ω t + θ 1 − θ r ). The phase offset that gives the maximal intensity (i.e., constructive interference) is thus the opposite of the phase of the function cos( ω t + θ 1 − θ r ). One approach is by Fourier-transforming the time-dependent signal and reading out the phase at frequency ω /2 π . To determine the phase offsets of 37 segments, we employed the concept of multidither coherent optical adaptive technique 11 , 31 – 33 . We modulated the phases of the first 20 rays by piston-displacing each corresponding mirror segment at a distinct frequency ω s while keeping the phases of the remaining rays, which formed a reference focus, constant ( Supplementary Note 3 ). We then Fourier transformed the recorded fluorescence trace ( Extended Data Fig. 1 , Step 7), and read out the phase offsets that would lead to constructive interference with the reference focus at the modulation frequencies ω s ( s = 1,2,…20) ( Extended Data Fig. 1 , Step 8). We next modulated the phases of the remaining 17 segments while keeping the phases of the first 20 segments unchanged and found the phase offsets for these 17 segments required to ensure constructive interference among beamlets. With all beamlets intersecting and constructively interfering at the focal plane, we obtained the final corrective wavefront and applied it to the DM for aberration correction. Because the reference foci used for both phase gradient measurements ( Extended Data Fig. 1 , Step 1–6) and phase offset measurements ( Extended Data Fig. 1 , Step 7–9) were aberrated to begin with, for larger aberrations, the whole procedure was iterated, as needed, to achieve optimal aberration correction ( Extended Data Fig. 1 , Step 10).

Supplementary Material Video 3 Video 1 Video 2 Video 4 Video 6 Video 8 Video 5 Video 7 Video 9 Supp Info