Iterative multi-photon adaptive compensation technique (IMPACT) has been developed for wavefront measurement and compensation in highly scattering tissues. Our previous report was largely based on the measurements of fixed tissue. Here we demonstrate the advantages of IMPACT for in vivo imaging and report the latest results. In particular, we show that IMPACT can be used for functional imaging of awake mice, and greatly improve the in vivo neuron imaging in mouse cortex at large depth (~660 microns). Moreover, IMPACT enables neuron imaging through the intact skull of adult mice, which promises noninvasive optical measurements in mouse brain.
© 2014 Optical Society of America
In vivo fluorescence microscopy [1–4] has been widely adopted for biomedical research. The molecular specificity and high sensitivity have enabled breakthroughs in a variety of research fields. One of the major limitations of optical measurement is the shallow imaging depth in biological tissue [5–7]. The inhomogeneous refractive index distribution leads to aberration and scattering, which can distort the optical wavefront. The deviation from the ideal spherical wavefront causes a deteriorated point spread function.
Initially developed for astronomical telescopes, adaptive optics (AO) has been adopted by more and more labs to achieve better microscopy imaging results [5, 8–13]. The first major success of applying AO to microscopy was in the field of ophthalmology [14–17]. One example is the laser scanning confocal reflectance imaging, in which the backscattering at the laser focus serves as the guide star. A wavefront sensor (typically Shack-Hartmann) directly measures the backscattered wavefront and feeds the correction profile to a continuous surface deformable mirror (DM). This scheme very much resembles the astronomical AO as it involves the laser guide star and the wavefront senor that directly measures the wavefront and feeds the information to the DM. Recently, such a scheme has been adopted for fluorescence microscopy application, in which the fluorescence generated by the laser focus serves as the guide star [8, 9, 13]. In scattering tissue, directly recording the backscattered light on a wavefront sensor is problematic because the backscattering can happen at any point along the beam path. A more practical solution is to perform low coherence interferometry as in optical coherence tomography and microscopy . In such systems, the wavefront information is obtained by phase-shifting interferometry. A recent study shows that low coherence gated backscattered light can work well in highly scattering mouse cortex .
Recently, significant research efforts are devoted to non-direct wavefront measurement or sensor-less methods. Generally, one changes the laser wavefront and observes the variation of the acquired images. This includes, for example, the phase retrieval methods , the image metric optimization methods [11, 20] and the pupil segmentation methods . The common feature is that the wavefront is not directly measured but inferred from the image variations, and a number of images need to be recorded during the measurements.
A closely related research field to AO is imaging in random scattering media [22, 23]. In the same way, the loss of resolution and signal strength is due to the wavefront distortion. Different from AO in astronomy, the wavefront distortion in random scattering media is highly complicated. For example, in conventional AO, one uses up to tens of low order Zernike modes to describe the wavefront distortion. Such a representation becomes insufficient for highly scattering samples [5, 7]. As a result, conventional AO methods for wavefront measurement and compensation do not work well for such samples. To make the matter even worse, the well-controlled guide star widely used in astronomical AO is in general not available in such systems.
Besides the wavefront complexity, the other challenge for practical in vivo applications is that we need to measure the wavefront at high speed. On the one hand, the cells are dynamic in their native environment [4, 24]. On the other hand, there could be motion induced artifact when imaging awake animals (for example, in the study of behavioral neuroscience).
To meet these challenges, we have developed iterative multi-photon adaptive compensation technique (IMPACT)  that utilizes 1) iterative feedback and 2) nonlinearity to rapidly measure wavefront in a highly scattering environment. In biological systems, the size and distribution of the fluorescence target is unpredictable. The iterative feedback can help focus light onto the stronger target and the nonlinearity (2nd, 3rd, and higher order ones) can help force the system to converge to a single focus. To date, IMPACT is perhaps the only scheme that has been applied to in vivo microscopy to handle the complicated wavefront distortion encountered in highly scattering biological tissue, such as the brain and skull of adult mice. In terms of operation speed, IMPACT benefits from the fast segmented DM and provides a measurement speed of 1.5 millisecond per spatial mode. In this work, we report our recent progress of applying IMPACT to practical in vivo animal imaging. We apply IMPACT to functional imaging of awake mice, imaging mouse cortex at large depth, and imaging through the intact skull of adult mice.
2. Principle and experimental setup
2.1 Principle of IMPACT
The principle and operation procedure of IMPACT have been discussed in the initial report . Basically, we use a segmented MEMS DM as the wavefront modulation and correction device. The MEMS mirror can in principle run at 30 kHz in a real-time control system. Using a Windows workstation, we achieved a consistent update speed of 8 kHz. The essence of IMPACT measurement is to split the DM’s pixels and run parallel phase modulation with each pixel at a unique frequency. This key idea was developed based on W.B. Bridges’ multidither coherent optical adaptive techniques (MCOAT) in the 1970s . However, there are a few major differences: 1) MCOAT modulates all the pixels in parallel and 2) the modulation is a small range phase dither (a phase swing over a small range). The feedback loops gradually shift the center value of the phase dither to achieve wavefront correction. In comparison, IMPACT only modulates a portion of the pixels while keeping the rest of the pixels stationary. Different from the small range dither, IMPACT uses a linear phase shift as a function of time over the entire 2π phase range. The unique modulation frequency becomes the unique phase slope value. At the end of the modulation, Fourier transform is used in IMPACT to determine the correction phase values. Another unique feature of IMPACT is that the pixel groups take turn to serve as the modulation group and the stationary (reference) group. In practice, we use three iterations to arrive at the converged wavefront, which takes ~2 seconds in total.
How large a fraction of the pixels shall we modulate? If we only modulate a small fraction such as 1/8, the majority of the light beam is not modulated. The modulation signal will contain a high DC value, which wastes the precious fluorescence signals and yields no useful information. On the contrary, if we modulate a large fraction such as 7/8, the stationary reference will be weak. As each pixel controls an equal amount of the numerical aperture (NA), the reference beam will have a low NA and hence a larger than diffraction limited focus. As a compromise of using the fluorescence signal efficiently and having a high quality reference focus, we modulate 50% of the pixels in practice.
How do we split the pixels to the two 50% groups? To have a near diffraction limited reference focus, we want the 50% of the pixels to spread uniformly over the available NA. If we spread them periodically, there will be some side peaks at the focal plane although the nonlinearity can help suppress it. A better way is to randomly and uniformly spread the 50% of the pixels over the available NA. An example of the splitting pattern is shown in Fig. 1.
To measure the phase values of N pixels, we will need at least 2N + 1 measurements to satisfy the Nyquist–Shannon sampling theorem. In practice, we use 4N measurements. In the Fourier domain, we uniformly spread the modulation frequencies between 0.25f and 0.5f (f is the MEMS update rate, 8 kHz in our current system). Besides the beating between the modulated pixels and the stationary pixels, there is the beating between the modulated pixels, which is between 0 and 0.25f and is therefore isolated from the correct signal frequencies. If we used 2N + 1 measurements, we would have to put the frequency between 0 and 0.5f and therefore there would be the crosstalk between the modulation-modulation beating and the modulation-reference beating.
To use it for imaging, we park the laser beam at the point of interest and start the parallel phase modulation, one half at a time. We typically repeat the modulation three times to ensure the convergence of the wavefront measurement. As the measurement progresses, the laser focus gets stronger. In the control program, we gradually reduce the laser power to maintain a near constant DC signal level to preserve the precious fluorescence signal from the sample. At the end of the measurement, we display the compensation wavefront on the DM and start laser scanning as in a conventional scope.
2.2 Experimental setup
A schematic drawing of our experimental setup is shown in Fig. 2.The laser source was a 80 MHz 140 fs Ti:Sapphire oscillator (Chameleon, Coherent, CA, USA). A two-axis Galvo system scanned the beam in the transverse direction. The Galvo was imaged by a pair of relay lenses onto the segmented MEMS DM that was subsequently imaged onto the rear pupil plane of a water-dipping objective (Nikon 16x NA 0.8, Japan). The fluorescence signal was directed by a dichroic beam splitter towards a 5 mm aperture GaAsP PMT (Hamamatsu, Japan). In all animal imaging experiments, the mice were head fixed according to established procedures .
3. Experimental results
Although optical recordings have traditionally been performed in anesthetized animals to minimize motion artifacts, anesthesia may potentially interfere with neuronal activity and prevent a direct observation of the behavior related neuronal activity . To use AO in such studies, we need to measure the optical wavefront quickly to avoid motion induced artifacts. Different from other methods, IMPACT does not need to record multiple images but only looks to maximize the nonlinear signal at the focus. Through experiments, we find that IMPACT works well even for awake animal imaging, which was a challenging task for AO microscopy. As an example, we performed function imaging on the GCaMP6f expressing transgenic GP5.17 mice , which have faster fluorescent dynamics but weaker intensity than the mice with GCaMP6f virus injection. We performed craniotomy on the S1 cortex and labelled astrocytes with SR101 . As the GCaMP signal from the spontaneous neuronal activity is dynamic in both time and intensity, we used the signal from labeled astrocytes for IMPACT measurements and compared the signal improvement on astrocytes. Figure 3(a) shows the astrocyte image at 220 µm under the dura with system correction [Fig. 3(c)]. The fine processes of the astrocytes are barely resolved. We parked the laser at the position marked with ‘ + ’ in Fig. 3(a) and ran IMPACT. Despite the unpredictable motion of the awake mice, IMPACT performed wavefront measurement and correction, and improved the image quality. The image with full compensation is shown in Fig. 3(b), and the full compensation profile is shown in Fig. 3(d). In Fig. 3(e), we compare the signal intensities at the location marked by the arrows in Fig. 3(a), which shows the improved signal intensity and contrast. With the full compensation, we imaged the spontaneous neuronal activity at 14 Hz with ~47 mW laser power. The fluorescence dynamics of the dendrites and spines are shown in Fig. 3(f)-3(h) and Media 1.
We also employed IMPACT to image the mouse S1 cortex in vivo at large depth. We used anesthetized Thy1-YFP (H line) mice with craniotomy for in vivo imaging, and we imaged the dendrites and spines of layer 5 pyramidal cells at 650~670 µm under the dura. With system correction only, the dendrites and spines are hardly resolvable, as shown in Fig. 4(a) and Media 2. In comparison, the dendrites and spines are clearly resolved [Fig. 4(b) and Media 3] with the full compensation [Fig. 4(c)] applied. The signals at the location marked by the arrows in Fig. 4(a) are compared in Fig. 4(d), which shows the signal intensity and contrast improvement with IMPACT. Moreover, the spines on the apical dendrites can also be resolved with IMPACT as marked by the yellow arrow in Fig. 4(b).
High spatial resolution imaging through highly scattering tissue, such as the skull of adult mice, is a challenging if not impossible task in the past. For imaging neurons in the cortex, the skull is usually either removed or thinned. However, skull removal may introduce substantial glial activation after the surgery, causing artifacts in the biological observation . Thinning can work for a short period of time, but the bone tissue can grow near the thinning sites after the surgery, which causes severe wavefront distortions and reduces the image quality for long-term studies. A highly non-invasive approach is greatly desired for the study of brain function. Demonstrated in the early work , IMPACT can handle the complex wavefront distortions and allow fluorescence imaging through fixed skulls. Here we show that IMPACT allows in vivo imaging of neurons through the intact skull of adult mice, which provides a unique solution to noninvasive brain imaging.
We performed in vivo two-photon fluorescence imaging through the intact skull of adult mice. An anesthetized P43 Thy1-YFP (H line) mouse was imaged after the scalp above the S1 cortex area was removed. The thickness of the skull was ~120 µm, determined by the second harmonic signal from the bone. In Fig. 5(a), we show the volume view of the dendrite and spines imaged with full correction at 110-126 µm under the bottom of the skull (Media 4). In Fig. 5(b) and 5(c), we show one cross-section of Fig. 5(a) acquired with full correction and system correction, respectively. Without full correction, the dendrite is barely visible (Media 5). With full correction [Fig. 5(d)], the fine spines can be resolved. Comparing the signals at the dendrite, IMPACT improved the fluorescence signal by a factor of ~20, much greater than that provided by other AO imaging methods. The improvement in image contrast and resolution could be even more significant, apparent from the comparison between Fig. 5(b) and 5(c).
All procedures involving mice were approved by the Animal Care and Use Committees of HHMI Janelia Research Campus.
These imaging experiments suggest that IMPACT works well for in vivo applications. Particularly it works for imaging awake, head-restrained animals, which is deemed a very challenging task for AO imaging. Such a capability comes from the high speed operation of IMPACT: the wavefront can be updated at ~8 kHz and the beating frequency is distributed between 2 kHz and 4 kHz, away from the slow scale variation. Of course, if the motion scale is large enough such that the fluorescence region shifts out of the laser focus, IMPACT may not be able to operate properly. Through experiments, we found that IMPACT worked well for awake, head-restrained animal imaging.
Another advantage of IMPACT is that it can compensate the wavefront distortion with a large number of spatial modes. Especially for the case of highly scattering tissue such as the skull of adult mice, the wavefront becomes highly complicated [Fig. 5(d)]. One cannot use just a few tens of Zernike modes to faithfully represent the wavefront. IMPACT takes advantage of the segmented MEMS DM of a large number of pixels (492 in the current system) and can handle complicated wavefront. As a result, IMPACT obtains huge signal improvement and makes imaging possible for difficult tasks. Employing MEMS DM with more pixels, the signal improvement can be even greater.
A main difference between IMPACT and other sensorless AO microscopy is that IMPACT does not require taking multiple images. Instead, IMPACT only looks to maximize nonlinear (2nd, 3rd, higher orders) signals. In the present implementation, the signal only comes from the laser focus. The laser power during IMPACT measurement is similar to the actual imaging power. Through the experiments on GFP and YFP expressing samples, we observed no photobleaching and phototoxicity effect on the sample. An alternative scheme would be to combine IMPACT with temporal focusing [31, 32] to collect fluorescence signal from a line or an area to use fluorescence signals from more fluorophores. This could also make the system more resistant to abrupt sample motions for imaging awake animals.
Examining the setup (Fig. 2), we find that the IMPACT scope is largely similar to a conventional two-photon scope except that a MEMS DM and one addition pair of relay lenses are inserted. As the cost of the segmented MEMS DM dropped dramatically over the past two years, we expect more adoption of IMPACT in bioimaging applications.
The major limitation of IMPACT is that the imaging field of view shrinks as we compensate for more and more complicated wavefront distortions. For compensating simple wavefront distortions such as the spherical aberration or other low order Zernike modes, the field of view can be rather large (100-200 microns) in the cortex, as shown in Fig. 3. However, as we push for imaging deeper, the field of view shrinks to tens of microns in the cortex and less than ten microns through the skull of adult mice. In the past, we tried running IMPACT at multiple locations and then stitched the images together after the measurement . Such a scheme is sufficient for studying slowly varying systems, such as the plasticity of the brain. For observing highly dynamic systems, such as the calcium activity in the cortex, better schemes need to be developed. New ideas are needed to achieve near simultaneous observation of a large volume to carry out such studies. Multi-conjugate correction achieves its success in both astronomical AO and ophthalmology AO [16, 17]. Similar or brand new methods are expected to revolutionize the large volume deep tissue fluorescence microscopy.
In summary, we report the application of IMPACT for in vivo imaging. Taking advantage of the iterative feedback and the inherent nonlinearity in the multiphoton signals, IMPACT can work in highly scattering system such as the cortex and the skull of adult mice. The improved signal strength, resolution, and contrast can facilitate in vivo calcium imaging of mouse cortex. We show for the first time that IMPACT enables in vivo imaging of neuron through the intact skull of adult mice, which provides a new solution to the noninvasive study of mouse brain. Although we mainly discuss the imaging applications in this work, IMPACT can surely be used for spectroscopy applications to achieve greater signal to noise ratio or shorten the measurement time.
We thank Hod Dana from GENIE team at Janelia for providing the GP5.17 mice. The research is supported by the Howard Hughes Medical Institute.
References and links
5. J. Tang, R. N. Germain, and M. Cui, “Superpenetration optical microscopy by iterative multiphoton adaptive compensation technique,” Proc. Natl. Acad. Sci. U.S.A. 109(22), 8434–8439 (2012). [CrossRef] [PubMed]
8. X. Tao, B. Fernandez, O. Azucena, M. Fu, D. Garcia, Y. Zuo, D. C. Chen, and J. Kubby, “Adaptive optics confocal microscopy using direct wavefront sensing,” Opt. Lett. 36(7), 1062–1064 (2011). [CrossRef] [PubMed]
9. X. Tao, J. Crest, S. Kotadia, O. Azucena, D. C. Chen, W. Sullivan, and J. Kubby, “Live imaging using adaptive optics with fluorescent protein guide-stars,” Opt. Express 20(14), 15969–15982 (2012). [CrossRef] [PubMed]
11. D. Débarre, E. J. Botcherby, T. Watanabe, S. Srinivas, M. J. Booth, and T. Wilson, “Image-based adaptive optics for two-photon microscopy,” Opt. Lett. 34(16), 2495–2497 (2009). [CrossRef] [PubMed]
12. P. Kner, J. Sedat, D. Agard, and Z. Kam, “Applying adaptive optics to three-dimensional wide-field microscopy,” in MOEMS-MEMS 2008 Micro and Nanofabrication, (International Society for Optics and Photonics, 2008), 688809–688809–688812.
13. K. Wang, D. E. Milkie, A. Saxena, P. Engerer, T. Misgeld, M. E. Bronner, J. Mumm, and E. Betzig, “Rapid adaptive optical recovery of optimal resolution over large volumes,” Nat. Methods 11(6), 625–628 (2014). [CrossRef] [PubMed]
16. Z. Popovic, P. Knutsson, J. Thaung, M. Owner-Petersen, and J. Sjöstrand, “Noninvasive imaging of human foveal capillary network using dual-conjugate adaptive optics,” Invest. Ophthalmol. Vis. Sci. 52(5), 2649–2655 (2011). [CrossRef] [PubMed]
17. J. Thaung, P. Knutsson, Z. Popovic, and M. Owner-Petersen, “Dual-conjugate adaptive optics for wide-field high-resolution retinal imaging,” Opt. Express 17(6), 4454–4467 (2009). [CrossRef] [PubMed]
18. M. Rueckel, J. A. Mack-Bucher, and W. Denk, “Adaptive wavefront correction in two-photon microscopy using coherence-gated wavefront sensing,” Proc. Natl. Acad. Sci. U.S.A. 103(46), 17137–17142 (2006). [CrossRef] [PubMed]
19. R. Fiolka, K. Si, and M. Cui, “Complex wavefront corrections for deep tissue focusing using low coherence backscattered light,” Opt. Express 20(15), 16532–16543 (2012). [CrossRef]
24. T. Lämmermann, P. V. Afonso, B. R. Angermann, J. M. Wang, W. Kastenmüller, C. A. Parent, and R. N. Germain, “Neutrophil swarms require LTB4 and integrins at sites of cell death in vivo,” Nature 498(7454), 371–375 (2013). [CrossRef] [PubMed]
25. W. B. Bridges, P. T. Brunner, S. P. Lazzara, T. A. Nussmeier, T. R. O’Meara, J. A. Sanguinet, and W. P. Brown Jr., “Coherent optical adaptive techniques,” Appl. Opt. 13(2), 291–300 (1974). [CrossRef] [PubMed]
26. Z. V. Guo, S. A. Hires, N. Li, D. H. O’Connor, T. Komiyama, E. Ophir, D. Huber, C. Bonardi, K. Morandell, D. Gutnisky, S. Peron, N. L. Xu, J. Cox, and K. Svoboda, “Procedures for Behavioral Experiments in Head-Fixed Mice,” PLoS ONE 9(2), e88678 (2014). [CrossRef] [PubMed]
28. H. Dana, T.-w. Chen, A. Hu, B. C. Shields, C. Guo, L. L. Looger, D. S. Kim, and K. Svoboda, “Thy1-GCaMP6 transgenic mice for neuronla population imaging in vivo,” PLoS ONE. in press.
30. H.-T. Xu, F. Pan, G. Yang, and W.-B. Gan, “Choice of cranial window type for in vivo imaging affects dendritic spine turnover in the cortex,” Nat. Neurosci. 10(5), 549–551 (2007). [CrossRef] [PubMed]
32. E. Tal, D. Oron, and Y. Silberberg, “Improved depth resolution in video-rate line-scanning multiphoton microscopy using temporal focusing,” Opt. Lett. 30(13), 1686–1688 (2005). [CrossRef] [PubMed]