1.

Introduction

Two-photon excitation fluorescence (TPEF)¹ and second-harmonic generation (SHG) imaging² are forms of nonlinear microscopy being used in all major areas of biomedical research. Although these techniques allow intrinsic optical sectioning and three-dimensional (3-D) imaging, aberrations caused by the alignment and quality of the optical elements, the objective-sample refractive-index mismatch, and the inhomogeneous structure of the sample^3–7 limit their performance.³ These enlarge, both laterally and axially, the focal spot within the specimen and reduce the multiphoton excitation efficiency, which leads to a reduction in the quality of the images.

In the last decade adaptive optics (AO) yielded substantial improvements in the images obtained with multiphoton microscopy.^3–6,8–11 In an ideal situation, for each location within the sample, the corresponding wavefront should be accurately measured and corrected. However, since the plane-by-plane wavefront assessment may be complex in thick samples, most of the authors used sensorless techniques for aberration correction.^3–6 Strategies are based on different algorithms modifying the incoming wavefront to improve the recorded image according to different quality metrics, such as intensity, contrast, or sharpness.^6,12

These experiments provided image quality improvements by means of an indirect measurement of the actual aberrations within the sample. Only a few studies combined wavefront sensing and correction in multiphoton microscopy.^8–11

It is well known that the use of a high numerical aperture (NA) objective induces spherical aberration (SA) due to the refractive-index mismatch at the interface between materials.^13,14 When a microscope is focused into the specimen, SA is induced, leading to a degradation of the images of deeper layers. Then, SA plays an important role when imaging thick samples with a multiphoton microscope. Moreover, it has been reported that SA is often the dominant term^3,4,6,15,16 and its correction leads to image enhancement.

Methods to correct for (or minimize) this unwanted SA in multiphoton microscopy were mainly based on the use of an objective correction collar^15,16 or a sensorless AO device.^3–6 For these experiments high-NA immersion objectives and previously chosen cover glasses (to minimize the refractive-index mismatch and the SA produced at the corresponding interface) are often used. The correction collar is a manual method only valid for a well-defined set of cover thickness values. An optimal performance depends on the relationship between the refractive indices of the immersion medium and the sample, and how uniform the refractive index within the specimen is.¹⁵

Alternatively, sensorless schemes offer a different strategy to compensate for different aberration terms by indirect assessment.^3–6 AO multiphoton microscopes based on modal Zernike correction¹² have reported interesting results. However, the effects of partial correction or induction of pure aberrations terms have not been studied in detail. In particular, SA increases imaging depth (i.e., induces a larger depth of focus), a feature that can be used in 3-D microscopy. Therefore, it is conceivable that the manipulation of the wavefront aberrations of the incident laser beam (via AO) to induce predefined values of SA may help to optimize multiphoton signals from deeper layers. In this sense, the aim of this study is to further analyze the influence of the SA on 3-D multiphoton imaging and in particular the effect in imaging deeper sample layers. We will examine the possibility of using a nonimmersion objective and an AO module to optimize both TPEF and SHG imaging in nonbiological samples and ocular tissues.

2.

Experimental Setup and Procedure

2.1.

Adaptive Optics Multiphoton Microscope

The experimental setup (Fig. 1) combines an AO module and a research prototype multiphoton microscope.⁹ A 76-MHz, 110-fs Ti:sapphire mode-locked laser system (Mira900f, Coherent, St. Clara, CA) centered at 760 nm was used for the multiphoton excitation. This was coupled into an inverted microscope (Nikon TE2000-U, Tokyo, Japan). A pair of nonresonant galvanometric mirrors (GSI, VM1000, Billerica, MA) together with a $Z$-axis step motor (PI C-136, Karlsruhe, Germany) allowed the acquisition of $XY$ images at different depths within the sample under study. A Plan Fluor $20\times /0.5\mathrm{NA}$ air long-working distance microscope objective (WD 2.1, Nikon ELWD Series) was used for the experiments described herein. Nonlinear signals (both TPEF and SHG) were collected via the same objective (backscattering configuration). A dichroic mirror separated these nonlinear backscattered signals from the infrared light used for multiphoton excitation. These signals emitted by the sample passed a spectral filter and reached the photomultiplier tube. Spectral filters in the detection channel were used to isolate either TPEF (broadband fluorescence filter, FGL435, Thorlabs Inc., Newton, NJ) or SHG (narrow-band spectral filter, $380\pm 10\mathrm{nm}$; FB380-10, Thorlabs Inc.) signals. Image recording was completely controlled through a custom LabView™ (Austin, TX) interface. Image post-processing was performed with MATLAB™ (The MathWorks, Inc., Natick, MA). A continuously variable neutral density filter (not shown in Fig. 1) placed in front of the laser was used to set the appropriate average illumination power at the sample’s plane.

Fig. 1

Schematic diagram of the adaptive optics (AO) multiphoton microscope. The AO module is composed of a Hartmann–Shack wavefront sensor and a deformable mirror; PMT, photomultiplier tube.

3.

Results

3.1.

Effect of Aberration Correction on Single Plane Images

The influence of SA was first tested by recording individual planes of different samples as shown in Fig. 2. A direct comparison can be made for retinal and corneal tissues providing TPEF (left panels, photoreceptors mosaic) and SHG (right panels, stroma collagen fibers) signals, respectively. As expected, the addition of 0.1 μm of SA led to a reduction in the image quality for both specimens (bottom panels). All image quality metrics (total intensity, entropy, signal-to-noise ratio) reduced when inducing SA (between 8% and 40% depending on the metric).

Fig. 2

Effects of SA on multiphoton images. Top row images ($\mathrm{SA}=0\mu \text{m}$) must be compared to those of the bottom row ($\mathrm{SA}=+0.1\mu \text{m}$). Samples correspond to chicken retina (left) and porcine cornea (right). False color has been used for a better understanding: green for TPEF and blue for SHG. Imaged single planes correspond to the best focus within the sample. Bar length: 50 μm.

Another preliminary experiment on the dominance of SA over other aberration terms was also carried out. As an example Fig. 3 includes TPEF images of a plane located 100 μm inside a piece of cellulose. The impact of using just SA [Fig. 3(b)] or also other terms such as astigmatism and coma [Fig. 3(c)] can be directly compared to a control image [no AO module in use, Fig. 3(a)]. This illustrative example confirms the dominance of the SA and helps to understand our choice of SA as a unique term to improve multiphoton imaging in thick samples.

Fig. 3

Impact of SA and other aberration terms on TPEF image quality. (a) Image control; (b) SA; (c) SA, astigmatism, and coma. All images have the same intensity scale. Bar length: 50 μm.

3.2.

Effect of Spherical Aberration on Depth Imaging

Tomographic images of the samples for different amounts of SA were first acquired [see an example of these images in Fig. 4(a)]. As explained in Sec. 2, for each tomographic image (and SA value) the averaged intensity profile was computed. Figure 4(b) shows the impact of inducing different controlled amounts of SA as a function of the sample’s depth compared to the case of complete aberration correction.

Fig. 4

(a) Comparison of tomographic TPEF images of a single cellulose fiber for different values of induced SA (0 μm, left; $-0.1\mu \text{m}$, center; $-0.2\mu \text{m}$, right). (b) Intensity profiles as a function of depth, when inducing different controlled amounts of SA compared to the case of complete aberration correction (black line).

When SA was not induced, the TPEF signal decreased with depth because of the influence of the specimen aberrations themselves [black line in Fig. 4(b)]. The intensity values reduced at the best imaged plane when SA was induced: the higher the SA the more reduction. For  $\mathrm{SA}=-0.1\mu \text{m}$ the intensity was below the optimal one and provided similar values for deeper locations, as seen by comparing black and black dashed lines. However, this behavior clearly changed when adding larger values of SA. For this sample, this occurred when SA was larger than $-0.15\mu \text{m}$ (gray lines). The arrow in the plot indicates the location where the behavior was reversed, i.e., a better TPEF signal was obtained when imaging deeper into the sample. This implies an extended imaging depth (or alternatively an increase in the depth of focus) from deeper positions within the sample. It is also interesting to observe that for $\mathrm{SA}=-0.3\mu \text{m}$ (and larger values not shown here) the effects are more noticeable for deeper planes, although a shift in the optimal focal plane is also produced.

From the intensity profiles the optimum SA value was determined (see Sec. 2) and 3-D multiphoton imaging immediately performed (i.e., $XY$ images along the $Z$ direction). These stacks of images for different depth positions were recorded for two experimental conditions: with the optimum SA value and without SA (for comparison purposes). This 3-D imaging gives us information on the interaction between SA and the quality of images from deeper planes. Figure 5 presents an example of TPEF images at different depth locations for the two experimental conditions: $\mathrm{SA}=0\mu \text{m}$ (top row) and $\mathrm{SA}=-0.2\mu \text{m}$. The improvement of images at planes located 50 and 75 μm within the sample can be qualitatively observed.

Fig. 5

TPEF images for different depth positions acquired with all laser beam aberrations corrected (top row) and $-0.2\mu \text{m}$ SA induced (optimum value for this sample). The sample corresponds to a piece of cellulose. Bar length: 50 μm.

For a better understanding of this SA impact, Fig. 6(a) presents the corresponding total intensity values (across the entire images) as a function of depth. The tendency for the rest of the image quality metrics (not shown here) was also similar. Moreover, although different image quality metrics are expected to measure different image properties,¹⁸ the reversal point positions (see arrow) hardly differ among the different metrics (not more than 3 μm for the samples here used). Figure 6(b) presents the improvement in total TPEF signal versus depth position. This becomes positive when the effect the SA is reversed. Similar results are obtained for a human epiretinal membrane also providing TPEF signal. Results are depicted in Fig. 7.

Fig. 6

(a) Effect of inducing $-0.2\mu \text{m}$ SA (gray line) compared to the case of null SA (black line). (b) Improvement in the total TPEF intensity of the images as a function of depth.

Fig. 7

Improvement in TPEF signal as a function of depth for a human epiretinal membrane when inducing $-0.2\mu \text{m}$ SA compared to the case of complete aberration correction. The arrow indicates the depth location of the images within the sample. Bar length: 50 μm.

For completeness, the procedure was also applied to samples providing SHG signal. The increase in depth imaging for a sample of starch grains is also evident for two different metrics (total intensity and entropy) as depicted in Fig. 8. For this sample, a value of $-0.1\mu \text{m}$ SA was the optimal value to observe the improvement.

Fig. 8

SHG signal improvement measured with total intensity (black line) and entropy (gray line) as a function of depth when inducing $-0.1\mu \text{m}$ SA. The arrow indicates a depth location of 12 μm. The sample corresponds to starch grains. Bar length: 50 μm.

Finally, the procedure was also applied to another biological sample: porcine corneal tissue providing SHG signal originated at the stroma. It is known that within the cornea, the deeper the image plane is located, the lower the SHG signal recorded,²⁵ as upper panels of Fig. 9 show. However, when the illumination laser beam contains a certain amount of SA (bottom panels of Fig. 9), beyond a particular depth location the intensity levels do not follow that behavior. For this particular sample, $-0.1\mu \text{m}$ SA was the optimal value to observe the effect analyzed in this work. In Fig. 10, this improvement in SHG signal as a function of depth is depicted.

Fig. 9

SHG imaging of corneal stroma at different depth locations illustrating the effects of adding $-0.1\mu \text{m}$ SA to the illumination laser beam (bottom panels). Upper panels correspond to the case of all laser beam aberrations corrected. The sample is an ex vivo porcine cornea. Bar length: 50 μm.

Fig. 10

Improvement in the SHG intensity of corneal stroma images as a function of depth when inducing $-0.1\mu \text{m}$ SA. Vertical dotted lines correspond to the locations of images in Fig. 9.

4.

Discussion

During the last decade, there has been a remarkable interest in the development of novel AO capabilities for improved 3-D high-resolution multiphoton microscopy imaging. Most reported techniques are sensorless and improved the focal spot within the sample by pre-shaping the wavefront of the input beam to compensate for the specimen-induced aberration according to a sequential adjust of individual Zernike modes.⁶

In this sense, this study has reported that an accurate manipulation of the aberrations in the illuminating beam provides improved 3-D multiphoton imaging. The AO module has not been used for a modal plane-to-plane correction, but to induce controlled amounts of SA, working in closed loop. This manipulation of the incident laser beam aberrations leads to an improvement in the nonlinear signal from images (as measured with different metrics) acquired at deeper locations within the sample. This effect might be understood as increased imaging depth or extended depth of focus and has been demonstrated in nonbiological and biological (ocular tissues) samples providing both TPEF and SHG signals.

For each sample, the appropriate SA value was computed from fast tomographic nonlinear imaging (for this experiment the highest induced SA value was $-0.3\mu \text{m}$, in steps of 0.05 μm). This value corresponds to the image with the highest area under the intensity profile. With this optimum induced SA, each sample was imaged with depth in a regular manner ($XY+Z$ scanning). Although the required SA values were particular for each sample, they were never larger than $-0.2\mu \text{m}$ here. The percent of improvement at a certain depth location depended also on the used sample. The shallowest plane at which this improvement takes place was $\sim 7\mu \text{m}$ for starch. For the rest of the samples, it was placed between 25 and 30 μm.

The effects of SA in both linear (bright-field and confocal) and nonlinear microscopy have been a topic of interest.^{4,15,16,26–28} Implementations to eliminate or minimize the effect of SA in the former include an objective correction ring,^13,15,16 changes in the effective tube length of the microscope objective,^28,29 structured illumination techniques,^30,31 point-spread function engineering,³² and AO devices,^6,33–35 among others.

Although to minimize SA some objectives have a correction ring to adjust for the thickness of the microscope slide, additional SA is also present when focusing deeper within the specimen. Moreover, when imaging specimens in a backscattered configuration, the main drawback is the noticeable reduction of the signal for deeper planes. The anterior layers will produce aberrations that might interfere with the measurement from deeper layers as the light travels through. Lo and coworkers analyzed the effects of a collar in multiphoton microscopy of tendon and skin (maximum depth: 28 μm).¹⁵ Although their results showed some improvements, they ignored the nominal SA values required for each sample and depth. They also claimed that the collar’s ability to compensate for the SA decreased with depth and that the required correction collar settings were particular for each sample and distribution of refraction indices. Muriello and Dunn reported a similar experiment.¹⁶ The combination of a water immersion objective lens and a correction collar decreased the depth dependence of signal levels in TPEF images of kidney tissue. Débarre et al. combined an immersion objective collar and a DM to pre-compensate for different aberration terms (SA included) in a mouse embryo.⁶ This procedure is robust but highly time-consuming (minutes).

Despite that aberrations are specimen-dependent, these tend to be dominated by a small number of low-order Zernike aberration modes. The magnitude of the aberrations increases with NA.^36,37 Although the ideal compensation would require a plane-by-plane SA correction, in the present work, we have demonstrated that this is not always necessary. We report that an appropriate value of SA can experimentally be measured for each sample in order to improve the image quality of deeper planes because the depth of focus is extended. This procedure, although with some limitations, represents a more elegant and accurate method than an objective collar, where the effect is limited by the discrete values provided by the collar itself.

Unlike most experiments reported in the literature (using immersion high-NA objectives), the objective here used is a low-NA air long-working distance objective. These “air-immersion” objectives would be more suitable for future in vivo experiments. In this sense, the SA value produced in the illumination beam has a double function: (1) a basal correction compensating for the refractive-index mismatch air-microscope cover and (2) an additional correction to partially compensate for the specimen-induced SA. The reduction of image quality is affected by SA to a large extent.³⁶ Despite that the rest of aberration terms are not corrected and a unique value (experimentally computed) of SA is used, the image quality is improved.

Compared to the experimental conditions using a total laser beam aberration correction,⁹ the “side-effect” of the proposed technique is that the signal at the best imaged plane of the sample is reduced. However, for deeper planes the behavior is reversed, the signal is improved, and more details are visible. Although the enhancement depends on each particular sample, the extended imaging depth has been demonstrated for both TPEF and SHG signals. Moreover, the procedure here reported overpasses (to some extent) the limitation of a single aberration correction per depth (since a unique customized aberration is used) which significantly decreases scan times and enables more rapid imaging. The approach reported in this paper provides a fine balance on the effect of SA aberrations in stacks of nonlinear images. While SA reduces the image quality at best focus, it provides a better series of images at deeper layers. If the amount of incoming light can be easily selected, this can be a better compromise providing a range of improved images.

The present procedure is much faster than a possible plane-by-plane correction. The typical acquisition time for tomography imaging modality is about 5 s for a 100-μm-thick sample (at 3 μm $Z$-step). The sequential use of six SA values with the computation of the intensity profiles took approximately 40 s. Once the SA value has been determined, the regular $XY$ imaging (planes taken 5 μm apart) will take $\sim 1\min$. The images for a plane-by-plane correction provide better quality, but the time used is much longer. In particular, with our experimental configuration, $>5\min$ would necessary to complete the entire operation.

A plane-to-plane correction provides a more accurate SA correction, but the time required for it is excessively long when living samples are imaged. In this sense, the main advantage of our procedure is that it could be used to improve the imaging of living specimens such as embryos or brain neurons, among others.

Probably one of the most interesting consequences is provided by results in Fig. 10. The imaging of corneal tissue in living animals (human included) is a challenging application. Fast recording procedures are of special interest for in vivo applications. The use of the present technique with a nonimmersion objective when imaging the living cornea would produce improved images while decreasing acquisition times, minimizing photobleaching effects, reducing tissues damage, and keeping the safety limits (strongly necessary in ophthalmology).

5.

Conclusions

We have used an AO multiphoton microscope in a backscattered configuration and a nonimmersion low-NA objective to improve 3-D nonlinear imaging. The technique was based on an accurate aberration control of the illuminating beam. In particular, controlled amounts of SA led to an increase in imaging depth independently of the specimen-induced aberrations, which is essential in multiphoton microscopy. The benefits of this AO procedure have been tested in both nonbiological and biological samples, with particular interest in ex vivo ocular tissues. In this type of tissue, an extended imaging depth might be of huge interest to better visualize deep retinal and corneal layers through volume rendering reconstructions. The procedure here described partially overpasses the limits imposed by the aberrations, improves the multiphoton microscopy sectioning capabilities, and represents a further step to produce 3-D images with overall increased quality. This technique might be of great interest in future studies of living samples, especially clinical applications including corneal tissue imaging.