1. Introduction

With the development of systems biology and the modelling of cellular behavior there is an increasing interest in experimental data based on single cells [1, 2, 3]. Recently it was reported that even genetically identical cells under similar environmental conditions can show remarkably different phenotypes [4, 5, 6, 7]. Such experiments give new insights regarding cellular behavior and highlight the importance of technical advances for acquiring single cell data.

The development of fluorescence based microscopy techniques and the development of fluorescent proteins and other fluorescent probes have enabled imaging of live cells in real time, allowing intracellular processes to be visualized [3]. By using highly sensitive cameras it is now possible to follow intracellular events in live cells with both high spatial and temporal resolution. Such microscope images with high resolution are necessary to extract quantitative data on cellular processes [8]. This is particularly important when aiming to quantify events in small cells and organelles. The difficulty of acquiring high-resolution experimental data that allows intracellular processes to be quantified is for instance listed as one of the major obstacles for the advancement of the field of yeast systems biology [9]. However, following specific cells over time introduces new challenges compared to taking snapshots of different populations of cells over time. Photobleaching of fluorescent probes or proteins makes quantitative time-lapse studies more difficult. In microscope based single cell analysis there is also a need for powerful image processing programs. These programs must be able to handle the huge amounts of data resulting from multi-color stacks acquired over time, often from cameras with large bit depths [10].

Optical tweezers have proven to be an excellent tool for manipulating cells and bacteria, and have recently been highlighted for the use in single cell analysis [3, 11]. However, when performing single cell experiments it is still important to investigate a statistically relevant number of cells, in order to avoid generalizing the behavior of a few individual cells as the behavior of the entire population [12]. A possible solution to make optical tweezers experiments more high throughput is to perform experiments on several cells simultaneously, by implementing several optical traps. This can be realized by the use of a spatial light modulator (SLM) in the optical path forming the optical tweezers [13, 14]. By updating the hologram displayed on the SLM it is possible to dynamically reconfigure the positions of individual traps in three dimensions. Using such holographic optical tweezers (HOTs) opens up the possibility of performing single cell experiments in parallel. So far, the reports on using HOTs for biological applications have been limited, e.g. [15, 16, 17, 18].

When trapping cells using HOTs, problems arise that are not encountered when trapping, e.g., homogeneous plastic beads. Cells are naturally differing in size, shape and composition, depending on, e.g., age or stage in the cell cycle. In a 2D array of optical traps all cells will therefore not lie in the same image plane. This problem is even more cumbersome if the nucleus, or other organelle of interest, is not visible at all in the image plane, as illustrated in Fig. 1(a). We have found this to be a problem in our previous research on cellular signaling pathways, where we combined optical tweezers and microfluidics in order to stress single cells to monitor the nuclear shuttling of proteins [17, 19]. The aim of those experiments has been to quantify the degree of nuclear localization of proteins over time and during changing environmental conditions. In our previous experiments, the trapped cells were fixed relative to a single imaging plane. If the nucleus was not not visible in that single image plane that could be due to two reasons: either there was no nuclear localization of proteins or the nucleus was in a different image plane. Even if the nucleus was visible in the image, it could be imaged slightly out of focus resulting in a lower (blurred) signal. This would then erroneously be interpreted as a lower response to the changing environmental conditions.

One solution to these problems is to acquire an axial stack of images instead of a single image at each point in time. Thus, if the axial spacing between those images is fine enough, the nucleus will be in focus in one of the images in the stack. Such stacks can be obtained by translating the entire array of trapped cells through the image plane by either moving one of the external lenses in the optical setup along the axial direction, thus altering the divergence of the laser trapping beam, or by introducing a lens function on the hologram displayed on the SLM. Alternatively, the trapping plane can be fixed, while moving the imaging plane instead [20]. Since many biological probes are highly sensitive to photobleaching, this sets an upper limit to the number of images that can be practically acquired, thus restricting the total time a cell can be imaged and/or lowering the time resolution. Instead, it would be desirable if the features of interest could be brought into a single imaging plane. This principle was recently demonstrated by Oddos et al., who used dual optical tweezers to align immune synapses in live-cell conjugates into a single imaging plane [21]. This allowed both the spatial and temporal resolution to be increased in time-lapse experiments. In our case, if the positions of all nuclei always coincided with the imaging plane only one image would have to be acquired at each time point, thus allowing more images to be acquired, and if desired, with higher temporal resolution. Since HOTs allow 3D manipulation, it is possible to individually adjust the axial position of each trapped cell, as illustrated in Fig. 1(b).

 

Fig. 1. (a) Illustration of how budding yeast cells align in optical traps with nuclei in different planes in a 2D array of traps, depending on size, shape etc. All nuclei (or other organelles of interest) of the cells in the array will thus not be imaged correctly. (b) Using HOTs it is possible to individually adjust the axial position of each individual trap, such that the nuclei of the cells always coincides with the imaging plane.

Download Full Size | PDF

In this paper we describe how automated image acquisition and image analysis can be used together with hologram generation in order to move the nucleus of each cell trapped in a HOTs array into the same imaging plane. This is demonstrated for fluorescence images of baker’s yeast (Saccharomyces cerevisiae) with green fluorescent protein localized to their nuclei through activation of the high osmolarity glycerol (HOG) pathway, involved in the osmoregulation of yeast.

2. Experimental procedure

The HOTs setup is built around a motorized inverted epi-fluorescence microscope (DMI6000B, Leica Microsystems), see Fig. 2. The laser beam (1070 nm, IPG Photonics) was first magnified by a telescope (f ₁ = 50 mm, f ₂ = 200 mm) to match the beam diameter to the size of the SLM (X8267-15, Hamamatsu Photonics). A second telescope (f ₃ = 375 mm, f ₄ = 100 mm) was used to image the SLM plane onto the back focal plane of the microscope objective (100×, NA 1.3). Thus, an image of the Fourier plane of the SLM coincides with the trapping plane. Note that the second telescope also reduces the laser beam to fit the back aperture (which in this setup is positioned in the back focal plane) of the microscope objective. To introduce the trapping laser beam without disturbing the possibility of multicolor imaging, a dichroic mirror was inserted between the filter cube cassette and the objective turret. The images were acquired with a 14 bit electron multiplying CCD camera (C9100-12, Hamamatsu Photonics). For the acquisition of fluorescence images a GFP filter cube was used (Semrock).

The SLM and the image acquisition were controlled from LabView. The holograms displayed on the SLM were calculated with a modified version [22] of a freely available LabView software using a 3D version of the Gerchberg-Saxton algorithm [23].

In the experiments an array of yeast cells were first captured with the HOTs (typically around 12mW of laser power per trap). A stack of images was then acquired, by translating the trapping plane through the image plane using the SLM (13 images were acquired for each stack). These images were analyzed in Matlab via LabView to find out the correct focus plane for each cell trapped in the array. The x and y coordinates of the traps were used as inputs to the image analysis routine, allowing each cell to be analyzed separately in a region of interest (ROI) specified by the cell coordinates. The optimum axial coordinate for each trap was then fed back into the hologram calculation routine, and a new hologram was calculated and displayed on the SLM, adjusting the axial position of each trapped cell. A final image was then acquired of the cells in the optimized trapping pattern, referred to as the optimized image.

To determine in which axial plane the nucleus is in focus for each individual cell, we scored for images with the highest contrast. For fluorescence images, many measures of the degree of contrast in an image have a maximum at the focal position [24]. For our application, we searched for nuclear localization of proteins, by filtering the image with a disc-shaped averaging filter (8 pixels diameter, roughly corresponding to the expected size of a nucleus). For each ROI in the filtered images we chose the focal plane for each cell as the plane containing maximum pixel intensity, i.e., maximum average intensity within the nucleus. The corresponding axial position of the trap was then automatically fed back into the SLM control software, which calculated and displayed the new hologram.

 

Fig. 2. Schematic drawing of the holographic optical tweezers setup. The laser beam is expanded by a telescope comprising lenses L1 and L2, then reflected off the SLM. The SLM plane is imaged onto the back focal plane of the microscope objective by a second telescope (L3 and L4). The desired trapping pattern is formed in or close to the image plane, which is imaged onto an EM-CCD camera. The position of the dichroic mirror (reflecting the laser up through the microscope objective) above the fluorescence filter cassette allows for independent multicolor imaging without affecting the laser trapping.

Download Full Size | PDF

We also developed a technique for axial positioning of cells based on brightfield images of the trapped cells. To determine in which plane each cell was in focus, we made use of the fact that the image in which the cell was in focus had a local minimum in contrast (between two contrast maxima), as suggested by Gordon et al. [24]. The contrast was determined by taking the standard deviation of the image pixel intensities within the ROI corresponding to each cell. For each cell in the array the axial coordinate where the ROI had the lowest standard deviation was chosen as the in-focus plane.

Yeast cells (S. cerevisiae), strain BY4741, expressing Hog1-GFP were used in the experiments. The yeast cells were grown in yeast nitrogen base, YNB, until an exponential growth phase was reached. The nuclear localization of Hog1-GFP was induced by stressing the cells with 1M sorbitol before loading the cells into the array.

 

Fig. 3. (a) The distribution of 24 nuclei found from an axial scan using the SLM. (b) GFP image at z = 0 μm, the plane where most nuclei are in focus simultaneously. (c) Optimized GFP image captured after each trap had been individually adjusted in the axial direction, allowing all nuclei to be in focus in the same image. Note the natural cell-to-cell variation in intensity and degree of nuclear localization. Scale bar 5 μm.

Download Full Size | PDF

3. Results and discussion

The axial distribution of focus planes for the nuclei found in the fluorescence images when scanning the 2D array along the optical axis is shown in Fig. 3(a). In the experiment, 24 yeast cells were trapped in a rectangular array with 8 μm between neighboring trapping positions. As can be seen from the histogram (Fig. 3(a)), the nuclei are distributed over several image planes spanning over several micrometers. The single plane where most nuclei are in focus (z = 0 μm), which is the optimum plane to image if all traps are in a single plane, is shown in Fig. 3(b). After each trap had been individually optimized in the axial direction - to align all nuclei in the same plane - an optimized image was captured with the EM-CCD camera, see Fig. 3(c).

To analyze the optimized image in more detail, the GFP intensities of the nuclei measured in the stack were compared with the nuclear intensities in the optimized image for each cell, see Fig. 4. Even though all nuclei are visible in Fig. 3(c), the axial adjustment is not perfect, as can be noted in Fig. 4, since the green rings (corresponding to the optimized image) would then always be equal to the intensity found in the plane defined as the focus plane for each cell (blue plus signs). For comparison, the nuclear intensities from the single image plane from the stack where most nuclei are in focus, z = 0.0 μm, are shown as red triangles. We believe the seemingly non-optimized resulting image to be the result of several effects. To some extent, the cells move in the traps due to Brownian motion. The different organelles within the cell (in our case the nucleus) might also move within the cell over time [25]. To quantify the influence of such movements around the equilibrium trap position, 30 images of yeast cells in the optimized trapping pattern were analyzed (data not shown). From these images it was obvious that the measured intensities in the nuclei fluctuate over time. However, from the 30 acquired images it was also obvious that the GFP was photobleached considerably over time (the reduction in signal possibly also including some degree of adaptation to the stress conditions). Taking both the photobleaching into account (black squares) and the fluctuations (characterized by a standard deviation, σ, represented with error bars) into account, the optimized image is equally high or highest in intensity for 22 of the 24 cells in the array, and the optimized image is better than the image corresponding to the best single trapping plane for all cells. Allowing a 2σ variation of the nuclear signal, the signals in the optimized image are stronger than, or equally strong as, the signals found in any of the measured planes for 23 of the 24 cells.

 

Fig. 4. Analysis of the average intensities in the nuclei of the individual cells. The optimum plane to use with a 2D array is at z = 0.0μm (red triangles). The maximum average nuclear intensity found in the stack is shown as blue plus signs. The green circles corresponds to the optimized image. The fact that the nuclear intensities in this image are not always highest can to a large extent be explained by small movements of the nuclei around the equilibrium position (error bars representing the standard deviation) as well as photobleaching (black squares).

Download Full Size | PDF

However, there are also other explanations of the seemingly non-optimized resulting image. During the axial scan the cells might also move or rotate within the trap when a new hologram is displayed on the SLM, which might lead to a cell not finding the same equilibrium position in the trap in the optimized trapping pattern as at a specific axial coordinate during the axial scan. Since the yeast cells are trapped with a laser power low enough to make gravitational forces noticeable, another explanation could be that the axial equilibrium positions are affected by unintended deviations from the desired trap intensities. Such deviations could result from non-optimal holograms or from limitations of the SLM when it comes to realizing the desired phase patterns. It is known that symmetric trap patterns, such as the ones used for the results shown in Figs. 3–4, in contrast to asymmetric trap patterns, need to be carefully optimized in order to yield a high uniformity of the trap intensities [26]. This fact was supported by the use of less symmetric trapping patterns for which a good optimized image was easier to obtain, see Fig. 5. However, since an iterative algorithm indeed was used to optimize the phase holograms, ghost spots and undesired variations in trap intensities should be avoided, also for symmetric patterns [27]. Furthermore, simulations on the optical performance of typical holograms used in our experiments showed that they yielded uniform trap intensities. Thus, we believe that these intensity variations are, to a large degree, induced by SLM imperfections.

 

Fig. 5. The use of less symmetric trapping patterns gives better performance. Top: GFP images; (a) the image at the single trapping plane where most nuclei are in focus and (b) the image where the axial position of each cell has been adjusted individually, in order for all nuclei to be in focus in the same image. Bottom: brightfield images; (c) image at the best single plane, and (d) optimized image. Scale bar 5 μm.

Download Full Size | PDF

In these experiments we used the signal of interest, i.e. localized Hog1-GFP in stressed cells, to optimize the axial positions of the individual yeast cells. However, in protein shuttling experiments it is important to avoid unnecessary exposure in order to avoid photobleaching of the signal. Thus, it could be better to use a counter stain to label the organelle of interest. For the particular example of nuclear shuttling of GFP-tagged proteins in yeast, we recommend a nuclear protein or histone protein tagged with a red fluorescent protein such as mCherry. This allows the optimum trap positions to be found using this complementary fluorescent probe, rather than the GFP-tagged protein in the signaling pathway of interest. This additional tagging would also facilitate subsequent image analysis aiming to quantify the degree of nuclear localization under varying environmental conditions.

Using HOTs for adjusting the axial position of cells has the advantage of allowing more images to be acquired in the correct focal plane before photobleaching becomes an issue, allowing better time resolution or studies over longer periods of time, see Fig. 6. Without HOTs the analysis of parallel single cells must be performed on a surface, which either requires axial stacks to be acquired with the associated problem of photobleaching and limited time resolution, or forces one to use objectives with lower numerical aperture (with a larger depth of focus) in order to allow simultaneous imaging of cellular features at different axial positions, at the cost of lower spatial resolution. However, a lower numerical aperture reduces the amount of fluorescence light that can be collected from the sample, reducing the sensitivity of the measurements. Collecting as much fluorescence light as possible is particularly important when working at low light levels as is often the case when working with fluorescent proteins tagged to cellular proteins with normal expression levels. Also, using HOTs for manipulating cells facilitates fast switching of environment around cells when combined with a microfluidic device [17, 19]. Another advantage with the presented approach is that axial drift of the microscope becomes less of an issue, since the trapped cells follow the movement of the objective. In order to avoid axial drift in conventional microscopy, auto-focusing routines can be used to keep the distance between the microscope objective and the image plane constant. However, the problem of features of interest being located in different planes is not avoided. In addition, auto-focusing routines mainly relying on image analysis tend to score brighter cells more effectively than cells with lower fluorescence [24]. By analyzing each cell individually, as with the HOTs approach, the image of each cell, even less fluorescent ones, can be optimized.

 

Fig. 6. A comparison of the photobleaching using the HOTs approach presented in this paper and axial stacking. The curves show the average nuclear intensities for 24 cells trapped in an array, as in Fig. 3, for 40 consecutive images (red solid) or stacks (blue dashed). For the curve representing the stack, the nuclear intensity for each individual cell is taken from the image in the stack giving the highest signal, at each time point. The background is subtracted before normalizing the intensities to one for the first image/stack. Photobleaching makes the signal fade over time, with a faster rate when stacks are acquired at each time point. Thus, the optimization of the individual trap positions in order to bring the nucleus into one single imaging plane, allows more images containing useful information to be acquired.

Download Full Size | PDF

4. Conclusion

In this paper we have shown how holographic optical tweezers and simple image analysis routines can be used to automatically adjust the axial position of individual cells optically trapped in an array. As a result the nucleus of each trapped cell can be forced to coincide with the imaging plane of the microscope. Such optimization of images allows one single image to be acquired that includes all relevant information, rather than requiring entire axial stacks of images. In time lapse studies this allows more useful images to be acquired before photobleaching becomes an issue. Since only one image needs to be acquired at each time point rather than an axial stack, the time resolution can thus also be increased.