Historically, several in vitro/ex vivo microscopy imaging techniques have been used to study cellular interactions within the uterus and the placenta. As these experimental methods have revealed compelling facts about the biologic phenomena of cell–cell contacts in these organs, they cannot be used to study complex dynamic behavior of living cells inside their physiologic environment. For this, recent advances in intravital imaging techniques, together with two-photon microscopy, offer an exciting opportunity to study such dynamic immunologic processes at the cellular level in the complex uterine and placental tissues. In this article, we review experimental imaging techniques that have been used for studying the uterus and placenta. In particular, we describe the advantages of intravital techniques and discuss novel procedures that can be used in reproductive immunology. We also describe several technical details involved in image sequence post-processing required to extract useful data. Finally, we conclude by discussing how the reproductive immunology field may benefit from the broad use of these intravital techniques.
Imaging techniques are useful to better understand the complex cellular interactions that occur inside organs. Apart from the actual image acquisition step, however, successful visualization of living organs requires several preparatory procedures, working in concert, under constant development and improvement. These procedures include ex vivo or in vivo surgical techniques for explanting/exposing organs, fixation stages for preserving stabilized structures during imaging, cell type labeling with transgenic technology, in vivo technology, and techniques for maintaining the physiology as much possible as in a non-manipulated animal.
With a high quality preparation, both types of samples (in vitro/ex vivo or in vivo) can be placed in the proper microscope to perform the final image acquisition. Both sample types have their advantages and limitations. In vitro/ex vivo techniques are easier to handle, allowing images at different angles than in vivo techniques. In vivo techniques, however, better represent the normal physiology of the organ as compared with ex vivo/in vitro techniques. Thus, a good strategy is to validate the in vitro/ex vivo observations of cellular interactions with in vivo cellular behavior.
Intravital imaging techniques (IVT) for observing cell-cell contacts require ever-increasing advances in microscopes capable of acquiring images deeper within the organ. Two-photon microscopy (2pM) is presently the technology of choice for obtaining the depth demanded with IVT. One disadvantage of 2pM in IVT, however, is that these lasers are significantly more expensive than traditional microscopy equipment. Also, IVT involves surgery that alters the normal body physiology, which can add extra uncertainty into biologic experiments, despite any palliative methods. Nonetheless, the disadvantages of IVT are unavoidable for studies at the cellular level, since other non-invasive methods such as magnetic resonant imaging or ultrasound imaging cannot obtain these resolutions and magnifications.
There exists a large body of literature that describes imaging techniques that reveal information at the fetal-maternal interface. However, in this review we shall concentrate specifically on techniques that can uncover complex interactions that occur at the cellular level within the uterus and the placenta.
In vitro/ex vivo Imaging at the Fetal-Maternal Interface
Observation of fixed samples of the uterus and the placenta, as well as in vitro co-cultures of uterine/placental cells, can provide an initial possible scenario about cellular interactions before and during pregnancy. Even before conception, oviduct epithelial cells can retain sperm cells, demonstrating that cell-cell contacts may occur.[4-7] An increase in the leukocyte infiltrate (mostly composed by macrophages, dendritic cells, and granulocytes) in the endometrium during the estrus phase can catalyze cellular interactions inside this organ, which would have a profound impact upon implantation.[8-10]
After conception, close proximity and interactions between blastocyst trophoblasts, and uterine epithelial cells were observed by histologic analysis and by in vitro co-cultures, interactions that are important for future blastocyst implantation and leukocyte attraction.[11-16] By histologic examination, it was also shown that dendritic cells stay inside the uterus[17, 18] and placenta during pregnancy, perhaps becoming entrapped inside the placenta tissue. Natural killer cells, also studied with histologic and in vitro techniques, were identified at specific regions inside the uterus[21-25] and the placenta, in close proximity to spiral arteries, which in turn depend upon natural killer cell activity for their appropriate formation.
Although useful, in vitro microscopy images of fixed tissue only provide a static snapshot of cell distributions at a particular moment in the cell history. Thus, this imaging modality cannot provide any information about cell interactions, since non-interacting cells would appear to produce the same image as strongly interacting cells. Ex vivo adhesion tests are also not useful for this purpose since the method itself is artificial. Thus, cellular interaction events require observational methods that reveal the dynamic processes among cells within the live tissue.
Intravital Imaging at the Fetal-Maternal Interface
The field of intravital imaging seeks to reveals how and where cells interact and perform their function within live organisms. One can compare intravital imaging to ethology, where animal behavior is observed within their natural habitats and niches. Just as ethology explores the impact of environmental alterations on a population's behavior, intravital microscopy allows observation of cells acting inside their natural organs. Thus, we can see how cells respond, before and after, to a changing environment, starting from a normal physiology to one after immunization and/or infection.
Despite the important immunologic processes that occur within the uterus during the estrous cycle and during pregnancy, it is surprising that intravital image acquisition of this organ has not been fully explored. Recently, IVT studies of the uterus are uncovering a rich and biologically accurate picture of the complex phenomena taking place. The observation of cellular dynamics provides not only a deeper understanding of the particular immunologic processes at work in the organ, but also the role of tolerance and regulation for the entire organism.
One of the earliest experiments to perform intravital imaging of uterus was achieved with the development of a superfusion chamber and was limited to observations of microvascular flow of an open organ. Later, another experimental design permitted image acquisition of the skin-pouch transplanted uterus, where the influence of cytokines, anti-inflammatory agents, diets[31, 32] and drugs in angiogenesis and endometriosis were tested. In more recent experiments with a technique similar to observations of cremaster vascular flow, the uterine horns were exposed to observe the radial artery in decidual branches.
For in vivo placenta observations, much less has been done. Image acquisition in this organ has been restricted to explanted observation of tissues by confocal microscopy. As confocal microscopes provide better resolution when compared with two-photon microscopes, their uses have several drawbacks, including: a smaller depth field, higher photobleaching that affects image quality, and more photodamage to the tissue. Thus, high resolution two-photon microscopy is ideally suited for studying placentas. Through this imaging modality, we can ask new questions about the spatial-temporal active transport dynamics of macromolecules, such as immunoglobulins. Apart from studying immunologic processes in normal pregnancies, these observations may address situations where the animals are concomitantly infected during pregnancy.
Image post-processing techniques
Due to endogenous background motion caused by surrounding organs or external forces that are inherent in intravital imaging, such as intestinal peristaltic movement, quantitative extraction of cell dynamics requires image post-processing. This background motion between image acquisitions can result in video frame displacements in all directions with respect to anchor positions, determined by well-defined features such as blood vessels or other anatomic structures in the image sequence. To minimize these motions, software-based stabilization corrections are applied to the entire 3-dimensional intravital image stack sequences over time. To gain a simple understanding of how these algorithms work, we imagine two consecutive images (A and B) that are positioned in a certain position is space (x, y, and z positions). Assume that these images are originally coincident, pixel by pixel, and denoted as A = It (x,y,z) at time t and B = It+Δt (x,y,z) at time t+Δt, in the image sequence. Now imagine that the tissue in the image frame It+Δt moves x = 10 μm to the right (so that we write B = It+Δt (x+Δx, y, z)) with respect to the previous image frames position It(x,y,z). The effect of the stabilization correction is to apply an equal and opposite displacement (Δxc = −10 μm, to the left) that will return (or align) the image It+Δt to some ‘tissue anchor’ reference, such as a blood vessel, originally defined within the image frame It(x,y,z).
The overall result of the stabilization correction is twofold: (i) to improve the final video quality and (ii) to make it possible to accurately quantify cell dynamics by subtracting out endogenous motion from their trajectories. Another benefit is to be able to characterize blood flow hemodynamics in the microcirculation of the uterus. Simply stated, the image displacement or drift (Δx, Δy, Δz), in all three directions (x, y, and z planes), between all consecutive image pairs It and It+Δt is minimized so as to achieve a stable and better movie quality.
Linear Background Movement of Tissue Under Observation
If the background tissue movement between images in the sequence consists, predominantly, of displacements in the transverse (x, y) and vertical (z) direction with respect to the focal plane, then a linear spatial alignment will suffice for stabilization. As described, the displacement vectors ut,t−1 = (Δxt,t−1, Δyt,t−1, Δzt,t−1) are found between each z-image stack pair at time t and t−1 that maximize the alignment of pixel intensities (Fig. 1). The process of finding this maximum pixel alignment is referred to as image registration, and has a long history in the image processing community.[36-38] The image sequence is stabilized when the negative displacement (−ut,t−1) is applied to all images in the stack at time t with respect to the stack at t−1 (Fig. 1).
Fluorescent two-photon microscopy is not only three-dimensional but also multi-channel. Different wavelengths correspond to different reporter proteins, and hence can distinguish biologic cells and tissue of interest. In channels that tag specific cells, all other biologic tissue is transparent. As an example, GFP expression under the Foxp3 promoter is used for observing or tracking Tregs. Thus, the stabilization algorithm just described uses image stack data from a channel where the anatomic structures of the tissue are relatively stable. In the case of uterus and placenta tissues, for example, the stable features could be collagen fibers and blood vessels in the labyrinth zone. Once the displacement vector ut,t−1 = (Δxt,t−1, Δyt,t−1, Δzt,t−1) is known, it can be applied to all other channels, where no anatomic structures are visible, to subtract out the component background motion from the cell trajectories.
Nonlinear Alignment for Elastic Movements
Not all endogenous background motion results in linear displacements. Living tissue and organs are elastic and can expand, contract, pushing and pulling the tissue under microscopic study in a nonlinear way. Indeed, over several image frames in the sequence, the tissue appears to undergo breathing or peristaltic motion, that is impossible to remove by applying linear displacement corrections. For this type of motion, nonlinear image registration is used to align consecutive image stacks. Here, the alignment is accomplished by first warping the image pair, with a so-called diffeomorphic mapping,[41, 42] whose effect is to produce the appropriate nonlinear distortion necessary for maximal image overlap. In other words, let us consider two types of images in a mirror: one that exactly represents the image and the other that is a distorted version of the original due to imperfections or curves in the mirror, similar to the mirrors in a fun house. The result of a diffeomorphic mapping is to create a matrix that captures the way that an image must distort to return to the form of the original image. As in the linear case, the motivation of a stabilization algorithm is to subtract these nonlinear motions, albeit more complex, from the cell velocities to extract the true tracks and velocity of the cells without the effects of its surrounding environment.
Brief Details of Image Registration Algorithms
Recently, many image registration algorithms have been reviewed and there are now standard methods for approaching these complex problems. Registration algorithms can be based either on characteristic features in the images or pixel level statistical measures between images. In pixel level image registration, alignment is accomplished through a statistical measure of the pixel intensity overlap between two images, called the cross-correlation (CC). Intuitively, the CC is a two-dimensional function that compares two images by indicating the degree of overlap at every pixel location. Thus, if one image I1(x,y) is displaced by Δx with respect to another I2(x−Δx,y), then the CC function will have a maximum at (Δx, 0). In this way, the CC function is used to obtain the displacement vector ut,t−1 between pairs of images (at times t and t+Δt) that maximally match the pixel intensities. This alignment process just described is called a convolution, and is intuitively akin to moving one image over the other and asking at which point maximum alignment occurs. The CC function is a computationally intensive operation (requiring N2 operations) since one image is displaced relative to the other, in a loop for all points (x,y) in the image. Therefore, in practice, algorithms we have developed obtain the displacement vector of one image relative to the other using a more efficient and well-known method called the phase correlation (PC), which is based upon the Fourier transformation (for details the reader may consult Szeliski's recent book). Recently, we developed a software application, specific for IVT image sequences that uses these image registration algorithms for aligning images together with an optimization procedure for obtaining the best z-stack configurations at each time point.
Extracting Cell Tracks
An important quantitative parameter of biologic interest is the cell velocity along their trajectories or track history. Either linear or nonlinear image registration is used to obtain the displacement vectors ut = (Δxt, Δyt, Δzt) for each image from its predecessor with respect to a fixed tissue anchor. Recently, we developed and implemented a novel tracking algorithm, based upon the Sequential Monte Carlo method that is specifically designed to track cells from three-dimensional two-photon image data. By subtracting out the background movement (or ‘tissue velocity’) component of the entire organ, accurate track data from many cells can be obtained from IVT video sequence.
To obtain sufficient magnification levels, an individual microscopy image z-stack with cellular resolution has a small view field. To reconstruct the entire environment in three dimensions, spatially adjacent z-stacks must be stitched together in a large mosaic. Using the same image registration techniques described above, large mosaics can be constructed by finding the maximal alignment, or overlap between all adjacent images. The problem of constructing large mosaics from many images has been extensively studied and many algorithms exist.[43, 45] We have developed an algorithm to seamlessly blend adjacent image stacks, using sophisticated multiscale (pyramid) techniques described in these references, so that the resulting mosaic image appears as a single exposure. A sample of how this imaging stitching algorithms work is shown in Fig. 2.
Intravital imaging of uterus and placentas
Motivated by the promise that intravital imaging offers for understanding immunologic details not previously observed, we have develop new surgical procedures, a novel organ holder, and a set of custom bioimage algorithms that not only improve the quality of the resulting images, but aid quantitative extraction of data for in vivo two-photon microscopy within the uterus and placenta. Interested readers can consult these references for further details of the methods and experimental setups. Thus, given the quality of the results (Figs. 3 and 4) we have obtained, we expect that these novel techniques should be useful to investigators working in the field of reproductive immunology.
In vitro/ex vivo and intravital techniques have been used to address relevant questions inside the reproductive immunology. In this review we summarized the use of these techniques to observe cellular contacts at maternal-fetal interface, with special focus on intravital imaging techniques. Nonetheless, the advantage of intravital imaging techniques comes at a cost. Indeed, intravital image acquisition is more laborious and technically demanding with respect to the required equipment/procedures than alternative in vitro/ex vivo image acquisition techniques. Moreover, imaging the living animal results in endogenous background motion, both translational and vertical with respect to the microscope axis that must be subtracted from the final time-lapsed images, for the purpose of extracting quantitative information of cell movements and speeds. We (and others) have developed specialized software algorithms that greatly reduce the surrounding global tissue movement, by finding globally optimal alignments of the between temporally adjacent 3D image stacks in the xyz directions. Our stabilization algorithm is efficient and can stabilize motion in any direction, without prior knowledge or interaction from the user.
Finally, we believe that the outgrowing uses of such intravital imaging techniques dedicated to observe cellular interactions inside the virgin and pregnant uterus will strongly contribute to the characterization of immune cell behaviors inside this organ.
This work was supported by a FCT research grant to CET (PTDC/EBB-BIO/115514/2009).