Deep-tissue focal fluorescence imaging with digitally time-reversed ultrasound-encoded light

Abstract

Fluorescence imaging is one of the most important research tools in biomedical sciences. However, scattering of light severely impedes imaging of thick biological samples beyond the ballistic regime. Here we directly show focusing and high-resolution fluorescence imaging deep inside biological tissues by digitally time-reversing ultrasound-tagged light with high optical gain (~5×10⁵ ). We confirm the presence of a time-reversed optical focus along with a diffuse background--a corollary of partial phase conjugation--and develop an approach for dynamic background cancellation. To illustrate the potential of our method, we image complex fluorescent objects and tumour microtissues at an unprecedented depth of 2.5 mm in biological tissues at a lateral resolution of 36 μm×52 μm and an axial resolution of 657 μm. Our results set the stage for a range of deep-tissue imaging applications in biomedical research and medical diagnostics.

ARTICLE

Received 21 Mar 2012 | Accepted 25 May 2012 | Published 26 Jun 2012 DOI: 10.1038/ncomms1925

Deep-tissue focal uorescence imaging with digitally time-reversed ultrasound-encoded light

Ying Min Wang1,*, Benjamin Judkewitz1,2,*, Charles A. DiMarzio3 & Changhuei Yang1

Fluorescence imaging is one of the most important research tools in biomedical sciences. However, scattering of light severely impedes imaging of thick biological samples beyond the ballistic regime. Here we directly show focusing and high-resolution uorescence imaging deep inside biological tissues by digitally time-reversing ultrasound-tagged light with high opticalgain(~5105 ). We conrm the presence of a time-reversed optical focus along with a diffuse background a corollary of partial phase conjugation and develop an approach for dynamic background cancellation. To illustrate the potential of our method, we image complex uorescent objects and tumour microtissues at an unprecedented depth of 2.5 mm in biological tissues at a lateral resolution of 36 m52m and an axial resolution of 657 m. Our results set the stage for a range of deep-tissue imaging applications in biomedical research and medical diagnostics.

1 Bioengineering and Electrical Engineering, California Institute of Technology , 1200 E California Boulevard, Pasadena , California 91125, USA.2 Department of Clinical Research, London School of Hygiene and Tropical Medicine , Keppel Street , London WC1E 7HT , UK . 3 Electrical and Computer Engineering, Mechanical and Industrial Engineering, Northeastern University , 360 Huntington Avenue, Boston , Massachusetts 02115 , USA. *These authors contributed equally to this work. Correspondence and requests for materials should be addressed to Y.M.W. (email: [email protected]) or to B.J. (email: [email protected]).

NATURE COMMUNICATIONS | 3:928 | DOI: 10.1038/ncomms1925 | www.nature.com/naturecommunications

2012 Macmillan Publishers Limited. All rights reserved.

1

ARTICLE

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms1925

Realizing high-resolution uorescence imaging within scattering biological tissues is a central goal in biomedical imaging. Considerable eorts have been made to extend the imaging

depth of optical methods 17, but focal excitation of uorescence has so far been fundamentally limited to a depth of one transport mean free path, or approximately 1 mm in most biological samples. This is because conventional focusing approaches treat scattered light as noise and select for the ballistic light component, which exponentially decreases with depth. However, scattered light contains important information about the sample, which can in fact be utilized. When light passes through scattering samples, its wavefront is seemingly randomized, but the randomization occurs in a deterministic and time-symmetric way. Th ese properties of elastic light scattering have recently been used to focus light through turbid samples by iterative wavefront optimization 814 and by time reversal using optical phase conjugation 10,1517. Th ese methods are, in many ways, analogous to adaptive optics methods used in astronomy to cancel out the eect of atmospheric scattering 18,19. However, in contrast to astronomy, where it is suffi cient to image through a turbid medium (the atmosphere), the goal of biomedical imaging is to image inside.

To achieve focusing inside tissues, Xu et al.20 proposed a scheme termed time reversal of ultrasound-encoded light (TRUE), which combines optical phase conjugation 21 with ultrasound encoding22. Th ey used focused ultrasound, which is much less scattered than light in biological tissues, to create a virtual source of light frequency shied by the acousto-optic eect. Scattered light emanating from this source was then time reversed by a photorefractive crystal acting as a phase conjugate mirror. Th e authors inferred the formation of a time-reversed optical focus from a line-scan across millimeter-scale absorbers embedded in tissue-mimicking phantoms. Although it promises improved absorption contrast 20,23,24, the use of this technique for high-resolution uorescence imaging in biological tissues remains fundamentally challenging. Because of the low ultrasound modulation effi ciency 25 , the phase conjugate mirror has to provide orders of magnitude higher than unity gain to excite detectable uorescence. Th is requirement cannot be met by traditional phase conjugate mirrors based on photorefractive crystals, whose gain is typically much less than one 26,27.

Moreover, the signicant challenge of undesired background illumination because of partial phase conjugation needs to be addressed. With complete time reversal, the TRUE focusing technique can be conceptually represented as photons retracing their paths back to the location of the virtual source. However, this view disregards the wave nature of light: complete time reversal requires full control over phase, amplitude and polarization of the entire scattered eld over the full solid angle which is fundamentally unfeasible (see below). As a result, even with perfectly aligned optics and noise-free recording of the scattered wavefront, the time-reversed focus is necessarily accompanied by a background 2830, which would obscure the uorescence signal originating at the desired optical focus.

Here we present a new strategy to overcome these challenges by combining digital phase conjugation 31 with dynamic wavefront manipulation. We directly visualize the formation of an optical focus, exciting uorescence between layers of highly scattering tissue. In doing so, we conrm the presence of the accompanying background predicted by theory that can be dynamically reproduced and subtracted. Th is digital background cancellation procedure, along with the high phase conjugate gain and resolution of our technique, enables the rst demonstration of focused uorescence imaging 2.5 mm deep inside biological tissue.

ResultsPrinciples. Our setup for uorescence imaging with time-reversed light is diagrammatically shown in Fig. 1 . As the performance of our approach critically depends on achievable resolution, phase

conjugate mirror gain and delity of phase conjugation, these parameters deserve further discussion .

Light scatters as it propagates through tissue, resulting in a speckled light eld at the ultrasound focus. Th e speckles within the ultrasound focus are frequency-shied by the acousto-optic eect, creating a source of frequency-shied light ( Fig. 1a ). As our technique selectively records and phase-conjugates the frequency-shied light, the size of the ultrasound-modulated volume determines the resolution of the phase-conjugated optical focus. We use a high-numerical aperture-focused ultrasound transducer with a calculated focal width of 34 m. To further conne the ultrasound-modulated volume along the axis of ultrasound propagation, we operate both the ultrasound source and the laser in pulsed mode 32 such that light only enters the sample when the ultrasound pulse has reached the target modulation volume (see Methods; calculated resolution along the axis of ultrasound propagation: 54 m).

With traditional phase conjugate mirrors, the power in the phase-conjugated beam ( POPC ) is proportional to the power in the signal beam ( PS) (ref. 27). Th is proportionality is referred to as the gain of the phase conjugate mirror:

G PP

= OPC S

Because of the low ultrasound modulation effi ciency and the small area ratio between ultrasound focus and scattered wavefront, the scattered light eld reaching the phase conjugate mirror consists mostly of light that is not frequency-shied ( f0) and a minute fraction (in our setup on the order of 10 4 of the total power) of frequency-shi ed, ultrasound-tagged light ( f0fUS). Th erefore, to excite detectable uorescence at the optical phase conjugate focus, a phase conjugate mirror with gain orders of magnitude larger than unity is required. Th is is currently not achievable by traditional phase conjugate mirrors, even with advanced phase conjugation schemes 3335.

To selectively phase-conjugate only the frequency-shied light with high gain, we implement an improved digital optical phase conjugation scheme (DOPC) that consists of a high dynamic range scientic CMOS (sCMOS) camera and a high-resolution phase-only spatial light modulator (SLM) 31. Th e scattered, ultrasound-tagged light eld interferes with an equally frequency-shi ed reference beam ( f0+fUS) and is imaged onto the sCMOS camera. Using digital phase-shiing holography 36 , the phase of the frequency-shied wavefront ( ( x,y)) with respect to the reference beam is measured. By reecting oa phase-only SLM displaying a phase conjugate map ( ( x,y)) at the image plane of the camera, the same reference beam is modulated to become a phase conjugate beam that is sent back into the sample (Fig. 1b). Th e phase conjugate beam traverses back through the tissue sample to converge at the location of the ultrasound focus resulting in an optical focus deep inside the tissue sample.

In our DOPC setup, the power in the phase conjugate light that leaves the DOPC setup is only dependent on the power in the reference beam that reects othe SLM displaying the phase conjugate map; thus, the DOPC is fundamentally not limited in terms of gain ( equation (1) ). In our experiments, we adjusted the intensity of the reference beam during playback to achieve a gain of approximately 5105, such that the phase conjugate focus contains suffi cient energy to excite uorescence that can be collected and detected outside of the tissue by a photomultiplier tube (PMT).

Th eoretically, with complete phase conjugation, the light eld within the ultrasound focus can be reconstructed without error (see Supplementary Methods ). However, the assumption of complete phase conjugation breaks down in practice real phase conjugate mirrors, whether based on photorefractive crystals or SLMs, have nite etendue and can only intercept a fraction of the output wavefront. As a result, a background always exists in the case of partial phase conjugation 2830. In a random scattering medium, the ratio of

(1)(1)

NATURE COMMUNICATIONS | 3:928 | DOI: 10.1038/ncomms1925 | www.nature.com/naturecommunications

2012 Macmillan Publishers Limited. All rights reserved.

2

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms1925

ARTICLE

a

Digital Phase Conjugation

Ultrasoundtransducer (fUS) sCMOS camera

Spatial light modulator

Reference beam (f0+fUS)

Photodetector

Lenses

Sample beam (f0)

Beamsplitters

Dichroic beamsplitter

Sample

b

Digital Phase Conjugation

Ultrasoundtransducer (fUS) sCMOS camera

Spatial light modulator

Photodetector

Reference beam (f0+fUS)

Lenses

Dichroic beamsplitter

Beamsplitters

Sample

Figure 1 | Schematic of the imaging principle. ( a) In the recording step, a 0.8-mm wide sample beam ( f0) scatters as it propagates through the tissue sample. A conned region of the scattered light in the tissue sample is frequency-shifted ( f0fUS) by a focused ultrasound pulse. The ultrasound focus thus becomes a virtual source within the tissue. Both the frequency-shifted light and the non-shifted light further scatter through the tissue and are collected. This output wavefront interferes with a reference beam ( f0+fUS) and the resulting interference pattern is imaged onto a scientic CMOS (sCMOS) camera in the digital phase conjugate mirror module. The digital phase conjugate mirror selectively measures the phase map ( (x,y)) of the frequency-shifted light through digital phase-shifting holography. The ultrasound is turned off after recording. ( b) In the playback step, the conjugate of the recorded phase map ( (x,y)) is displayed on a spatial light modulator (SLM) placed at the image plane of the sCMOS camera. The reference beam reects off the SLM and is transformed into the phase conjugate beam that is propagated back into the tissue, reconstructing an optical focus at the ultrasound modulation location. Any excited uorescence is collected and measured outside the tissue using a photodetector.

the peak intensity of the phase conjugate focus to the average intensity of the accompanying background, the peak-to-background ratio, can be analytically derived. Following the framework of Vellekoop et al.37,38 , we nd that the peak-to-background ratio is determined by the number of optical modes intercepted and time-reversed by the phase conjugate mirror, N , and the number of input modes in the ultrasound focus, M (see Supplementary Methods). When both phase and amplitude of the scattered eld are time-reversed, a case similar to the use of traditional phase conjugate mirrors, the peak-to-background ratio is:

PBRphase amplitude

& = +

As N is nite in a real-phase conjugation setup, a phase conjugate background is inevitable. In our experimental setup, N is limited by the number of optical modes imaged onto the SLM and the fact that only the horizontally-polarized component is measured and time-reversed. The diuse background that inevitably results can excite uorophores outside of the focus, contributing to noise in the detected uorescence signal. Because of its spatial extent, the total background excitation can drown out the desired focal uorescence signal detected by a single channel PMT outside the sample. We show in our experiments that this background, though indeed signicant, can be dynamically subtracted by digital manipulation of the measured phase conjugate map, allowing us to realize high-resolution focal uorescence imaging in biological tissues.

Direct visualization of optical focus. To directly visualize and characterize the focus formed by time-reversed light, we placed an optically transparent hydrogel slab containing a thin quantum dot layer between two pieces of ex vivo chicken breast tissue, each 2.5 mm thick ( Fig. 2a ). When we focused light into the tissue without any wavefront manipulation (at phase display on the SLM), the light

1

N M

(2)(2)

When only the phase of the scattered eld is time-reversed, a case similar to our technique using the DOPC, the peak-to-background ratio for large N is:

PBR PBR

phaseonly phase amplitude

= +

p p

4 1 14

( )

N

&

(3)(3)

M

3

NATURE COMMUNICATIONS | 3:928 | DOI: 10.1038/ncomms1925 | www.nature.com/naturecommunications

2012 Macmillan Publishers Limited. All rights reserved.

ARTICLE

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms1925

a b d e f

Ultrasound transducer

45 k

15 k

0

y

5 mm

500 m thin

0

500 m thin

x z

0

Input light

c

g h i

From SLM

15 k

2.5 k

5 mm

3 mm

2.5 mm

QDot sheet

0

Figure 2 | Demonstration of optical focusing between thick layers of biological tissue. ( a) Schematic of the sample arrangement, consisting of a thin sheet of quantum dots between two 2.5 mm thick sections of ex vivo chicken tissue. ( bd) Fluorescence emission camera images of the area (in the yz plane) indicated by the dashed blue square in a. (b) Diffuse illumination pattern obtained by focusing into the tissue without wavefront modulation (at phase display on the SLM). ( c) Illumination pattern resulting from optical phase conjugation of ultrasound-tagged light, showing a focus on top of a diffuse background. ( df) Background images and ( gi) corresponding background subtracted maps (positive values) obtained by the following techniques: ( d,g) mechanically shifting the sample by 5 m to disrupt phase conjugation; ( d,h) digitally shifting the phase map by 50 pixels; and ( f,i) modulating the original phase map by subdividing it into 8 16 areas and alternately adding 0 or phase shift to each area (see Methods). Scale bar, 500 m.

was highly scattered and failed to form a focus ( Fig. 2b ). In contrast, Fig. 2c shows the uorescence excited by phase conjugation of ultrasound frequency-shied light. A cone of light converging into the location of the ultrasound focus was clearly visible, albeit on a signicant background. Taking into account the thickness of the quantum dot sheet (500 m) and the expected size of the ultra-soundfocus(34 m, see Methods), the peak-to-background ratio was determined as ~5.5.

Background subtraction. As discussed above, the diuse background seen in Fig. 2c was expected because of the lack of complete control of the entire light eld in a phase conjugation experiment. We further observed that the diuse background was concentrated around the focus, an eect that was also reported by Vellekoop et al. when focusing light through a layer of highly scattering zinc oxide particles 28 . With a large number of speckles in the focus (see above), the presence of this background drastically reduced the contrast at the focus and poses a critical challenge to optical focusing using time reversal. With the DOPC system, however, the ability to digitally manipulate the phase conjugate eld allows for the possibility of playing back a light eld that closely mimics the background, thus enabling background subtraction.

Accurate background subtraction requires better understanding of the cause of its spatial localization. One possible explanation of this eect is that it is caused by correlations in the scattering transmission matrix 28.Th us, like the delity of the optical phase conjugate focus, the presence of a concentrated background would depend on the precise alignment of the sample with the phase-conjugated beam. Alternatively, the observed background could be caused by the macroscopic concentration of diuse light around the target area an eect that would be expected to be more prominent in highly forward-scattering samples such as biological tissues, and that would be invariant to microscopic misalignments of the sample. To identify the dominant eect contributing to the background in our forward-scattering sample, we displaced the sample by 5 m and displayed the conjugate of the phase map recorded before the displacement. As can be seen in

Fig. 2d , this shi entirely disrupted the focused beam, whereas the diuse background was unaected. A er subtraction of this background from the raw image, a focus was revealed at much higher contrast ( Fig. 2g ).

However, mechanical displacement is an impractical method for background subtraction for most applications. Instead, we can digitally alter the recorded phase maps to mimic the diuse background illumination. We achieved this by two methods: digitally shiing the phase map by 50 pixels ( Fig. 2e,h ) or dividing phase maps into large subregions and phase-shiing every other subregion by ( Fig. 2f,i ; see Methods) a strategy related to dierential background rejection techniques previously used in two-photon microscopy 39,40 . As digital shiing may introduce undesirable asymmetry to the phase map, the latter method for background subtraction was chosen for all subsequent experiments. We note that a suitable background image could not be obtained by simply displaying a at phase map on the SLM (as shown in Fig. 2b ). Such approach would fail to adapt to dierent locations of the ultra-sound focus and would be unable to compensate for geometrical aberrations in the tissue.

Performing time reversal and subtracting the background in this manner for each location of the focus, we scanned the position of the ultrasound transducer and conrmed that the optical focus followed the locations of the ultrasound focus ( Supplementary Movie 1).

Determining the point spread function. To measure the point spread function and to quantify the resolution of our imaging system, we placed a uorescent quantum dot lled polyacrylamide (PAA)bead(<20 m in diameter) between two pieces of ex vivo chicken breast tissue ( Fig. 3a ). Figure 3b shows an epiuorescence image of this sample. Because of the forward scattering nature of our biological sample ( g=0.965;ref.41),theapproximatelocation of the bead can be inferred. However, tissue scattering results in very strong blurring that would prohibit imaging at high resolution. In contrast, Fig. 3c shows a well-resolved image of the bead collected using time-reversed light. To obtain the image, the ultrasound focus was scanned in the xy plane and an optical focus obtained by phase conjugation was formed at each scan position indicated by the blue dots. Background subtraction by dynamic digital phase map manipulation was performed at every step. As the bead is smaller than the ultrasound focus, the imaged size of the bead eectively estimates the three-dimensional resolution of the imaging system. The proles in each dimension ( Fig. 3d f corresponding to the x,

y and z dimensions, respectively) were t by Gaussian point spread functions with

NATURE COMMUNICATIONS | 3:928 | DOI: 10.1038/ncomms1925 | www.nature.com/naturecommunications

2012 Macmillan Publishers Limited. All rights reserved.

4

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms1925

ARTICLE

b

a c

e f

Ultrasound transducer

y

657 169 m

Input light

52 5 m

d

2 2.5 mm thick ex vivo tissue

xz

36 3 m

Figure 3 | Determination of point spread function. ( a) Schematic of the setup used for the point spread function measurement. A uorescent bead is embedded between two 2.5 mm thick sections of ex vivo tissue. ( b) Epiuorescence image of the sample in the xy plane, showing very strong blurring dueto tissue scattering. ( c) Fluorescence image obtained by scanning the position of the ultrasound transducer in x and y, detecting the uorescence excited by time-reversed light and using adaptive background cancellation as described in the text. ( df) Prole of the uorescent bead in x (d), y (e) and z (f) direction. Blue dots indicate locations of collected data points. Pixels between data points are interpolated for display using bicubic interpolation. Scale bar, 50 m.

a e

b

Ultrasound transducer

2 2.5 mm thick ex vivo tissue

x z

y

Input light

c

f

d

g

Figure 4 | Fluorescence image of complex objects. ( a) Diagram of sample arrangement. ( b) Epiuorescence image of an abstract CIT feature. ( c) Epiuorescence image obtained when the sample is placed under 2.5 mm of biological tissue. The features are not resolved. ( d) Raster-scanned image of the embedded CIT feature obtained using our imaging system. ( e) Epiuorescence image of tumour microtissues. ( f) Epiuorescence image obtained when the sample is placed under 2.5 mm of biological tissue. ( g) Raster-scanned image of the embedded tumours obtained using our imaging system. Blue dots indicate locations of collected data points. Pixels between data points are interpolated for display using bicubic interpolation. Scale bars, 50 m.

widths of 36 3m and 565 m (full-width at half- maximum) in the plane perpendicular to the axis of light propagation, and 657169 m along the axis of light propagation (values 95% condence of t).

Fluorescence image of complex objects embedded in tissue. We demonstrate the deep-tissue imaging capability of our system by raster scanning a known complex feature. We patterned quantum dot features of an abstract CIT design in a 500- m thin patch of PAA gel (a hydrogel that is commonly used for ultrasound phantoms 42 ), embedded between two pieces of chicken tissue, each 2.5 mm thick ( Fig. 4a ). Figure 4b shows an epiuorescence image of the features before embedding. Because of the tissue scattering, the embedded features cannot be resolved with epiuorescence imaging ( Fig. 4c ). In comparison, the CIT features are clearly resolved using our method ( Fig. 4d ).

Fluorescence image of embedded tumour microtissues. We also obtained images of tumour microtissues embedded in tissues.

Th e microtissues are arranged in a 500- m thin patch of PAA gel, embedded between two pieces of chicken tissue, each 2.5 mm thick. Figure 4e shows an epiuorescence image of the tumours. The tumours embedded between ex vivo tissue are not resolved with epiuorescence imaging ( Fig. 4f ). In contrast, the tumours imaged with our method are well resolved and the dierential uorescence intensities of the tumours are also reected in the image (Fig. 4g).

Discussion

Fluorescence imaging in biological tissues beyond one transport mean free path is one of the most challenging goals in biomedical optics. In this study, we provide the rst demonstration of focal uorescence imaging in the diusive regime with time reversal of ultra-sound-tagged light. We implemented a DOPC system with high gain to directly observe the time-reversed optical focus and the accompanying phase conjugate background. We took advantage of the capabilities of the DOPC to digitally manipulate the phase-conjugate map to dynamically estimate and subtract the uorescence contribution of the phase conjugate background that would otherwise

5

NATURE COMMUNICATIONS | 3:928 | DOI: 10.1038/ncomms1925 | www.nature.com/naturecommunications

2012 Macmillan Publishers Limited. All rights reserved.

ARTICLE

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms1925

obscure the focal uorescence signal. Using this technique, we characterized the point spread function of the system as having an anisotropic lateral resolution of 34 m52 m and an axial resolution of 657 m. Furthermore, we illustrated the capabilities of our method by successfully imaging uorescent objects 2.5 mm deep in ex vivo tissue, equivalent to about 75 scattering mean free paths 30.

As conrmed by our results, the imaging resolution perpendicular to the axis of light propagation is determined by the ultrasound focal volume. Th e system s resolution can therefore be improved by using an ultrasound transducer with a higher central frequency and a higher numerical aperture. However, some expected trade-os should be noted. Higher frequency ultrasound is more strongly attenuated in biological tissues 43 , thus reducing the practical focusing depth of the ultrasound. In addition, a smaller modulation volume would further diminish the population of the frequency-shi ed light and increase the challenge of detecting a small signal on top of a large background during phase measurement. However, these issues can be addressed by the development of faster, higher dynamic-range cameras, and with advanced ltering methods 44,45.Th e point spread function along the axis of light propagation is limited by the angular spread of the focused light cone. As the angular spread is a function of illumination geometry and tissue scattering, the resolution could be improved by using high-numerical aperture illumination orcounter-intuitivelyby imaging thicker, more scattering samples. Finally, the resolution along the axis of light propagation could further be improved by taking advantage of multiphoton excitation.

We estimate that at the plane of the time-reversed focus, < 1 % of the energy is within the focus. Th is means that ~ 99 % of the remaining energy is spread over the diuse background, which if uncorrected can obscure the focal uorescence signal. In non-homogeneous uorescent samples, we showed that dynamic background subtraction eectively uncouples the focal uorescence signal from that excited by the background. In some applications like photodynamic therapy, where the goal is to deliver more light into the focus, a further increase in peak-to-background ratio may be desirable. There are two ways to achieve a higher peak-to-background ratio. First, the number of optical modes ( N ) controlled by the DOPC can be increased by increasing the number of pixels on the SLM. Second, the number of optical modes in the ultrasound focus ( M)canbe decreased by decreasing the size of the ultrasound focus.

Because our method is based on optical time reversal, it relies on mechanical stability of the sample. Th e acquisition cycle per pixel should therefore be faster or on the same order of magnitude as the speckle decorrelation of the tissue. Th is condition is easily met in ex vivo experiments: our current pixel acquisition time (6.7 s) was shorter than the decorrelation time of the sample (41 s). For in vivo applications, decorrelation times are typically much faster: published values range from the order of milliseconds 4648 to seconds49, depending on the tissue type and immobilization strategies. For such applications, the pixel acquisition time would have to be reduced accordingly. We anticipate that this will ultimately be possible with the use of faster SLMs 50 , and with the continuing development of faster, higher dynamic range cameras. In all our experiments, the irradiance of the laser beam at the sample was<10mWmm2. The laser power would have to be decreased, or the diameter of the beam increased, to meet clinical safety standards (2 mW mm 2). Taken together, such improvements would ultimately enable a wide range of in vivo applications, including molecular imaging, early cancer diagnosis, photodynamic therapy and targeted excitation of optogenetic tools in deep tissues.

Methods

Sample preparation. Frozen ex vivo chicken breast tissue was cut into 2.5 mm thick slices and embedded in 10 % PAA gel inside an open-top quartz glass sample cuvette with four polished sides ( Starna Cells , CA). Th e tissue slices

had a measured scattering coeffi cient of

s=30mm1 (see also ref. 30). Using the previously published anisotropy 41 of g=0.965, we calculated the reduced scattering coeffi cient to be

s=

s (1g)=30mm1 (10.965)=1.05mm1. This is in agreement with the widely referenced approximate value of ~ 1 mm 1 (see for example, refs 5,20).

Th e PAA gel was polymerized using 4 ml phosphate-buered saline, 1.5 ml acrylamide, 0.4 ml bis-acrylamide, 62.5 l ammonium persulfate and 25 l tetramethylethylenediamine. PAA beads containing quantum dots (Qtracker 655 Non-targeted Quantum Dots, Invitrogen ) were made using a reverse micelle protocol modied from Beningo and Wang 51 with a starting concentration of 200 nM quantum dots in the aqueous phase. Th e beads obtained varied in size and were strained through a 40 m cell strainer (Biologix, USA). Th e actual sizes of the beads used in all experiments were determined by observation under a uorescence microscope.

Th e CIT feature was made by polymerizing clear PAA gel (500 m thick) ona SU-8 mold (designed in-house, manufactured by the Stanford Microuidics Foundry). Th e patterned depression in the clear PAA gel was then lled with PAA containing quantum dots (Qtracker 705 Non-targeted Quantum Dots, Invitrogen) with a starting concentration of 1 M in the aqueous phase.

Cancer microtissues , obtained by the hanging-drop technique using HepG2 cells 52, were custom ordered from InSphero AG (Switzerland). The spheroids were xed with 2% paraformaldehyde (Sigma-Aldrich, USA), washed with 50mM borate buer saline ( Thermo Scientic , USA), permeabilized with 0.1 % Triton-X 100 ( Sigma-Aldrich , USA), and stained with DY-521XL long stokes shi NHS-ester dye ( Dyomics , Germany) that binds to the proteins in the cancer microtis-sues. Th e concentration of the staining solution was 14 nM. Based on calibration with known uorophore concentrations, we estimated the resulting stain concentration in the tumour to be ~ 5 M. Th e tumour microtissues were arranged and embedded in a 500- m thin PAA gel patch.

Measurement of sample scattering coefcient. Th e scattering coeffi cient of the chicken tissue was measured interferometrically with a Mach Zehnder interferometer. As only ballistic light signicantly interferes with the reference beam, the reduction in amplitude of the fringes with and without a sample in the sample beam path could be used to nd the scattering coeffi cient 30.

Setup . All data shown was recorded using a custom-built experimental setup diagrammatically shown in Supplementary Fig. S1 . A 2.7 W, 532 nm Q-switched laser (Navigator, SpectraPhysics, USA) pulsed at 20 kHz with a pulse width of 7 ns and a coherence length of 7 mm was used as a light source at optical frequency f0.

A er passing an optical isolator and a xed attenuator, it was split into a reference beam and a sample beam. Th e sample beam was attenuated by a neutral density lter wheel, spatially ltered by a single mode optical bre (Nufern 460HP, 20 cm length), collimated to a 0.8-mm waist beam and directed into the sample cuvette. Th e irradiance at the sample was ~ 10 mW mm 2. We note that this is above the ANSI standards for skin irradiance (2 mW mm 2). However, we do not foresee lowering the laser power as being a fundamental limitation.

Inside the sample, a fraction of the light was frequency-shi ed to f0+fUS by an

ultrasound transducer (element size: 6.35 mm, focal length: 6 mm; V3330, Olympus NDT, Olympus , USA) operated at fUS=45MHz. To achieve maximal resolution along the axis of ultrasound propagation, the transducer was driven with short pulses (pulse length: 1 cycle at 45 MHz, 55 V peak-to-peak) triggered by the laser Q-switch signal at a xed delay such that the ultrasound pulses coincided withthe laser pulses at the ultrasound focus 32 (the trigger delay was jittered by 5.5ns to minimize the detection of coherent eects between ultrasound-tagged and untagged light; see Supplementary Methods ).

To scan the ultrasound focus, the transducer was mounted on a three-axis computer-controlled micromanipulator ( Sutter Instruments , USA). Aer passing through the sample, the scattered beam was recombined with the reference beam (horizontally-polarized), which had also been frequency-shi ed by fUS by an acousto-optic modulator (AFM-502-A1, IntraAction , USA). A er passing a horizontally-aligned polarizer and another beamsplitter, the combined beams reached the surface of a phase-only SLM (vis-PLUTO, Holoeye , Germany), carefully aligned (1:1 pixel-to-pixel match) to the image plane of a high dynamic range sCMOS camera (pco. edge, PCO AG , Germany). Th e lens used to image the SLM onto the sCMOS camera was a commercial compound lens ( Micro-Nikkor 105 mm f / 2.8 , Nikon ).

Compared with the DOPC system rst described by Cui and Yang 31, our improved DOPC system directly imaged the SLM pixels onto the camera pixels, and thus, enabled reliable alignment and day-to-day quality assurance (see below). As the image had to be reected by a beamsplitter, we chose a plate beamsplitter (High-Energy Nd:YAG 50 / 50, Newport Corporation , CA) whose reective surface faced both the SLM and the camera, so as to avoid image aberrations and ensure precise alignment.

SLM curvature compensation. Th e reliability of DOPC critically depends not only on the precise alignment of the SLM and the camera, but also on the SLM curvature and the reference beam aberrations. Both of these eects can be compensated for digitally by nding a compensation phase map for the SLM that, when displayed, time reverses the reference beam. Because a time-reversed beam would trace its path back through the spatial lter (the SM bre), the compensation map

6

NATURE COMMUNICATIONS | 3:928 | DOI: 10.1038/ncomms1925 | www.nature.com/naturecommunications

2012 Macmillan Publishers Limited. All rights reserved.

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms1925

ARTICLE

was optimized by maximizing the power of the light that returned back through the single-mode bre (measured by a photodiode).

Phase recording . At each scan position, phase recording was achieved in a phase-shi ing digital holography setup 36: the ultrasound pulse was cycled through four phases (0, /2, , 3 / 2). A total of 10 30 frames were recorded (at 30 frames per second) for each phase and averaged, resulting in four intensity maps ( I0, I /2, I and I3 /2), which were used to reconstruct the complexeld according to E = ( I /2 I3 /2) + i(I0 I ). Th e phase map was calculated as = tan 1(Re(E)/Im(E)) (or = Arg( E)). To minimize artifacts introduced by slow phase uctuations of the reference beam, the acquisition of the intensity maps was interleaved by cycling through all four phases for each block of four sequential frames acquired by the sCMOS camera (exposure time: 28 ms, frame rate: 30 Hz). To achieve suffi ciently fast cycling and between-frame switching, an arbitrary function generator (AFG 3252, Tektronix , USA) generated two output signals (one 45 MHz sinusoidal cycle with phase shi of 0 or /2), which were each inverted by an RF 180 power splitter ( Mini-circuits , USA) to obtain the four phase-shi ed signals. A microcontroller board ( Arduino , Italy; obtained from SparkFun Electronics , USA ) connected to an RF switch (Mini-circuits, USA) was programmed to select the appropriate phase for each frame acquired by the sCMOS camera. Th roughout the phase recording, the SLM displayed a at (all 0) phase map. An acquisition cycle took 6.7 s per pixel. For complete timing information, see Supplementary Fig. S2 . For an analysis of whether the coherence length limits detection of scattered photons, see Supplementary Fig. S3 .

A typical ultrasound-tagged phase map recorded on the camera contained N=8104 modes (area of the sensor divided by speckle autocorrelation area). Together with an estimated M (number of speckles in the ultrasound focus) of 1 2 10 3 we obtained an upper bound for the peak-to-background ratio of ~ 60 ( equation (3) ).

Detection of uorescence excitation by time-reversed light. The time-reversed beam was obtained by reecting the blank reference beam othe SLM displaying the measured phase conjugate map. Th e backscattered uorescence excited by the time-reversed optical focus was reected o a dichroic mirror and detected by a single-channel PMT tted with the appropriate bandpass lters (Semrock 650-40, 710-40 or 675-67, for Qtracker 655, Qtracker 705 or DY-521XL long stokes shi NHS-ester dye, respectively).

Quality assurance of DOPC . Because of the dependence of our system on precise alignment, mechanical stability and low dri , we included a parallel sample beam path to assess and assure the performance of our setup on a day-to-day basis. Consisting of ground glass diusers and an additional observing camera, it was analogous to the setups previously used by our group to demonstrate turbidity suppression by phase conjugation 30.

Speckle decorrelation time . With the sample beam turned on, we acquired images of the speckle eld on the sCMOS camera at a rate of 1 frame per second for 180 s. We measured the correlation of the rst frame with each subsequent frame and dened the decorrelation time as the time a er which the correlation fell below 1 1

/e.

References

1. Denk, W., Strickler, J. & Webb, W. Two-photon laser scanning uorescence microscopy.Science 248, 7376 (1990).

2. Helmchen, F. & Denk, W. Deep tissue two-photon microscopy.Nat. Meth. 2, 932940 (2005).

3. Minsky, M. Microscopy Apparatus.US Patent 3,013,467 (1961).4. Huang, D. et al. Optical coherence tomography.Science 254, 11781181 (1991).

5. Ntziachristos, V. Going deeper than microscopy: the optical imaging frontier in biology.Nat. Methods 7, 603614 (2010).

6. Ntziachristos, V., Ripoll, J., Wang, L. V. & Weissleder, R. Looking and listening to light: the evolution of whole-body photonic imaging . Nat. Biotech. 23, 313320 (2005).

7. Lichtman, J. W. & Conchello, J.- A. Fluorescence microscopy.Nat. Methods 2, 910919 (2005).

8. Vellekoop, I. M., Lagendijk, A. & Mosk, A. P. Exploiting disorder for perfect focusing.Nat. Photonics 4, 320322 (2010).

9. van Putten, E. G.et al. Scattering lens resolves sub-100nm structures with visible light.Phys. Rev. Lett. 106, 193905 (2011).

10. Hsieh, C. L., Pu, Y., Grange, R. & Psaltis, D. Digital phase conjugation of second harmonic radiation emitted by nanoparticles in turbid media . Opt. Express 18, 1228312290.

11. Popo, S., Lerosey, G., Fink, M., Boccara, A. C. & Gigan, S. Imagetransmission through an opaque material . Nat. Commun. 1, 81 (2010).

12. Katz, O., Small, E., Bromberg, Y. & Silberberg, Y. Focusing and compression of ultrashort pulses through scattering media . Nat. Photonics 5, 372377 (2011).

13. Mudry, E., Le Moal, E., Ferrand, P., Chaumet, P. C. & Sentenac, A. Isotropic diraction-limited focusing using a single objective lens . Phys. Rev. Lett. 105, 203903 (2010).

14. McCabe, D. J. et al. Spatio-temporal focusing of an ultrafast pulse through a multiply scattering medium.Nat. Commun. 2, 447 (2011).

15. Yaqoob, Z., Psaltis, D., Feld, M. S. & Yang, C. H. Optical phase conjugation for turbidity suppression in biological samples . Nat. Photonics 2, 110115 (2008).

16.Leith, E. N. & Upatniek, J. Holographic imagery through diusing media.J. Opt. Soc. Am. 56, 523 (1966).17. Goodman, J. W., Huntley, W. H., Jackson, D. W. & Lehmann, M. Wavefront-reconstruction imaging through random media . Appl. Phys. Lett. 8, 311313 (1966).

18. Davies, R. & Kasper, M. Adaptive optics for astronomy. Annu. Rev. Astron. Astrophys. 50, 147 (2012).

19.Zawadzki, R. J.et al. Adaptive-optics optical coherence tomography for high-resolution and high-speed 3D retinal in vivo imaging.Opt. Express 13, 85328546 (2005).

20.Xu, X., Liu, H. & Wang, L. V. Time-reversed ultrasonically encoded optical focusing into scattering media . Nat. Photonics 5, 154157 (2011).

21.Yariv, A. Phase conjugate optics and real-time holography.IEEE J. Quantum Electron. 14, 650660 (1978).

22. Mahan, G. D., Engler, W. E., Tiemann, J. J. & Uzgiris, E. Ultrasonic tagging of light: theory.Proc. Natl Acad. Sci. USA 95, 1401514019 (1998).

23. Liu, H., Xu, X., Lai, P. & Wang, L. V. Time-reversed ultrasonically encoded optical focusing into tissue-mimicking media with thickness up to 70 mean free paths.J. Biomed. Opt. 16, 086009 (2011).

24. Lai, P., Xu, X., Liu, H., Suzuki, Y. & Wang, L. V. Reection-mode time-reversed ultrasonically encoded optical focusing into turbid media . J. Biomed. Opt. 16, 080505 (2011).

25.Kothapalli, S. R. & Wang, L. H. V. Ultrasound-modulated optical microscopy. J Biomed. Opt. 13, 054046 (2008).

26. Feinberg, J., Heiman, D., Tanguay, A. R. & Hellwarth, R. W. Photorefractive eects and light-induced charge migration in barium titanate . J. Appl. Phys. 51, 12971305 (1980).

27. Gunter, P. & Huignard, J. P. Photorefractive Materials and their Applications. 1 Basic Eects, (Springer, 2006).

28.Vellekoop, I. M. & Mosk, A. P. Universal optimal transmission of light through disordered materials.Phys. Rev. Lett. 101, 120601 (2008).

29.Gu, C. & Yeh, P. C. Partial phase conjugation, delity, and reciprocity . Opt. Commun. 107, 353357 (1994).

30. McDowell, E. J. et al. Turbidity suppression from the ballistic to the diusive regime in biological tissues using optical phase conjugation . J. Biomed. Opt. 15, 025004 (2010).

31.Cui, M. & Yang, C. H. Implementation of a digital optical phase conjugation system and its application to study the robustness of turbidity suppression by phase conjugation.Opt. Express 18, 34443455 (2010).

32. Atlan, M., Forget, B. C., Ramaz, F., Boccara, A. C. & Gross, M. Pulsed acousto-optic imaging in dynamic scattering media with heterodyne parallel speckle detection.Opt. Lett. 30, 13601362 (2005).

33.Feinberg, J. & Hellwarth, R. W. Phase-conjugating mirror with continuous-wave gain.Opt. Lett. 5, 519521 (1980).

34.Lanzerotti, M. Y., Schirmer, R. W. & Gaeta, A. L. High-reectivity, wide-bandwidth optical phase conjugation via four-wave mixing in potassium vapor . Appl. Phys. Lett. 69, 11991201 (1996).

35. Tschudi, T. et al. Image amplication by two- and four-wave mixing in BaTiO3 photorefractive crystals.IEEE J. Quantum Electron. 22, 14931502 (1986).

36. Yamaguchi, I. & Zhang, T. Phase-shi ing digital holography.Opt. Lett. 22, 12681270 (1997).

37. Vellekoop, I. M. Controlling the Propagation of Light in Disordered Scattering Media. PhD thesis, Univ. Twente (2008).

38.Vellekoop, I. M. & Mosk, A. P. Focusing coherent light through opaque strongly scattering media.Opt. Lett. 32, 23092311 (2007).

39. Leray, A. & Mertz, J. Rejection of two-photon uorescence background in thick tissue by dierential aberration imaging . Opt. Express 14, 1056510573 (2006).

40.Leray, A., Lillis, K. & Mertz, J. Enhanced background rejection in thick tissue with dierential-aberration two-photon microscopy . Biophys. J. 94, 14491458 (2008).

41.Cheong, W. F., Prahl, S. A. & Welch, A. J. A review of the opticalproperties of biological tissues.IEEE J. Quantum Electron. 26, 21662185 (1990).

42. Zell, K., Sperl, J. I., Vogel, M. W., Niessner, R. & Haisch, C. Acoustical properties of selected tissue phantom materials for ultrasound imaging . Phys. Med. Biol. 52, N475N484 (2007).

43.Goss, S. A., Frizzell, L. A. & Dunn, F. Ultrasonic absorption and attenuation in mammalian tissues.Ultrasound Med. Biol. 5, 181186 (1979).

44. Li, Y., Hemmer, P., Kim, C., Zhang, H. & Wang, L. V. Detection of ultrasound-modulated diuse photons using spectral-hole burning . Opt. Express 16, 1486214874 (2008).

45. Gross, M. et al. Detection of the tagged or untagged photons in acousto-optic imaging of thick highly scattering media by photorefractive adaptive holography.Eur. Phys. J. E 28, 173182 (2009).

7

NATURE COMMUNICATIONS | 3:928 | DOI: 10.1038/ncomms1925 | www.nature.com/naturecommunications

2012 Macmillan Publishers Limited. All rights reserved.

ARTICLE

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms1925

46. Draijer, M., Hondebrink, E., van Leeuwen, T. & Steenbergen, W. Review of laser speckle contrast techniques for visualizing tissue perfusion . Lasers Med. Sci. 24, 639651 (2009).

47. Hajjarian, Z., Xi, J., Jaer, F. A., Tearney, G. J. & Nadkarni, S. K. Intravascular laser speckle imaging catheter for the mechanical evaluation of the arterial wall . J. Biomed. Opt. 16, 026005 (2011).48. Lev, A. & Sfez, B. In vivo demonstration of the ultrasound-modulated light technique.J. Opt. Soc. Am.. A Opt. Image Sci. Vis. 20, 23472354 (2003).

49. Cui, M., McDowell, E. J. & Yang, C. An in vivo study of turbidity suppression by optical phase conjugation (TSOPC) on rabbit ear . Opt. Express 18, 2530 (2010).

50.Conkey, D. B., Caravaca-Aguirre, A. M. & Piestun, R. High-speed scattering medium characterization with application to focusing light through turbid media.Opt. Express 20, 17331740 (2012).

51.Beningo, K. A. & Wang, Y. L. Fc-receptor-mediated phagocytosis is regulated by mechanical properties of the target . J. Cell Sci. 115, 849856 (2002).

52.Kelm, J. M., Timmins, N. E., Brown, C. J., Fussenegger, M. & Nielsen, L.K. Method for generation of homogeneous multicellular tumor spheroids applicable to a wide variety of cell types . Biotechn. Bioeng. 83, 173180 (2003).

Acknowledgements

We thank Dr. Ivo M. Vellekoop, Professor Scott E. Fraser, Seung Ah Lee, Mooseok Jang for helpful discussions, and Guoan Zheng as well as Roarke Horstmeyer for comments on the manuscript. Th is work is supported by NIH 1DP2OD007307-01 and

R21 EB012255 as well as DARPA W31P4Q-11-1-0008. B.J. is a recipient of a Sir Henry Wellcome Postdoctoral Fellowship by the Wellcome Trust. Y.M.W. acknowledges support from the National Science Scholarship, awarded by the Agency for Science, Technology and Research, Singapore.

Author contributions

Th e idea was conceived by C.A.D. and C.Y. Th e imaging scheme was developed by Y.M.W., B.J. and C.Y. Th e experiments were designed and performed by Y.M.W. and B.J. Th e data analyses were performed by Y.M.W. and B.J. All authors contributed to the preparation of this manuscript.

Additional information

Supplementary Information accompanies this paper at http://www.nature.com/ naturecommunications

Competing nancial interests: Th e authors declare no competing nancial interests.

Reprints and permission information is available online at http://npg.nature.com/ reprintsandpermissions/

How to cite this article: Wang, Y. M. et al. Deep-tissue focal uorescence imaging with digitally time-reversed ultrasound-encoded light. Nat. Commun. 3:928doi: 10.1038 / ncomms1925 (2012).

License: Th is work is licensed under a Creative Commons Attribution-NonCommercial-Share Alike 3.0 Unported License. To view a copy of this license, visit http:// creativecommons.org/licenses/by-nc-sa/3.0/

8

NATURE COMMUNICATIONS | 3:928 | DOI: 10.1038/ncomms1925 | www.nature.com/naturecommunications

2012 Macmillan Publishers Limited. All rights reserved.