Non-contact laser microdissection and pressure catapulting: Automation via object-oriented image processing

Abstract

Laser Microdissection and Pressure Catapulting (LMPC) is a well-known non-contact method to isolate and collect specific cells from complex animal or plant tissues or living cultured cells. Thus, it is enabling pure and homogeneous sample preparation resulting in an eminent increase in the specificity of versatile downstream molecular analyses, like e.g., different PCR, nucleic acid and protein microarray analysis.

An important innovation is the integration of computer-assisted recognition methods by object-oriented image processing. Here we show the automated recognition of sample material by applying object-oriented image processing on parameters like, e.g., density, area, neighborhood and shape. Coupled with this kind of software modules, the LMPC system is able to detect, isolate and finally capture the specimen of interest in a fully automated manner.

Differential analysis of individual tissues and single cells will eliminate the averaging effect and allow the discovery of detailed differences between various cell types.

Zusammenfassung

Laser Microdissection und Pressure Catapulting (LMPC) ist eine etablierte und berührungsfreie Methode, Zellen aus komplexen tierischen oder pflanzlichen Geweben oder aus lebenden Kulturen spezifisch zu isolieren und zu sammeln. Dadurch wird eine saubere und homogene Probenpräparation gewährleistet, bei der die Aussagekraft einer Vielzahl nachgeschalteter molekularer Analysen steigt. Beispiele hierfür stellen verschiedene PCR-, Nukleinsäure- und Protein-Microarray-Untersuchungen dar.

Die Integration computergestützter Erkennungsmethoden ist eine wichtige Weiterentwicklung dieser Technologie. In dieser Arbeit zeigen wir die automatische Erkennung von Probenmaterial durch die Anwendung so genannter objektorientierter Bildverarbeitung. Hierbei werden Parameter wie Objektdichte, Fläche, Nachbarschaftsbeziehungen oder die Form zur Erkennung herangezogen. Die Verbindung derartiger Software mit einem LMPC-System erlaubt die Erkennung, Isolierung und das automatische Sammeln der Probe in nur einem Arbeitsgang.

Differenzierte Analysen einzelner Gewebe und Zellen führen zum Ausschluss von Mittelungseffekten. Sie erlauben die detaillierte Untersuchung verschiedener Zelltypen, wobei Unterschiede deutlich zu Tage treten.

Introduction

The ability of generating pure samples from heterogeneous specimen has become one of the most important necessity for differentiated analysis and high-content results in the field of genomic, transcriptomic and proteomic research. Thus, whenever high-resolution control of sample composition is crucial, laser microdissection and micromanipulation offers the possibility of working with homogeneous material by selecting or rejecting individual cells and tissue areas of interest, respectively.

The PALM MicroBeam (P.A.L.M. Microlaser Technologies AG, Bernried, Germany) enables such sample preparation by the combination of Laser Microdissection with Pressure Catapulting (LMPC). This technology utilizes a pulsed UV-A laser that is coupled through the epifluorescence path into an inverted microscope and focused on a minute spot via the objective lenses. Only within this spot the energy of the laser light is high enough to ablate material due to a photofragmentation process. By this means, selected specimen of different origins can first be laser microdissected and thereafter ejected against gravity (Laser Pressure Catapulting, LPC) into a collection device on top of the desired area.

As the PALM micromanipulation system has no physical contact to the heterogeneous specimen the risk of contamination or infection of the isolated probes is minimized. And additionally, as the extracted samples are derived from a clearly defined origin, this technology is a paramount prerequisite for homogeneous starting material.

Non-contact sample preparation by LMPC has become one of the most emerging techniques in functional genomics and proteomics and non-contact laser microdissection from subcellular compounds up to entire tissue areas is frequently performed in numerous research institutes or industrial laboratories throughout the world.

The pulsed UV-A laser beam used for LMPC is routinely focused to less than 1 μm when it impacts the sample at high energy density (Fig. 1). The energy transfer within this condensed focal spot is sufficient to break the molecular bonds of the irradiated matter by a so-called ‘ablative photo-decomposition’ process without any mechanical contact. Thus, material at the focal point is cleaved into atoms and small molecules. As this cutting is a photochemical and fast process devoid of heat transfer, adjacent material outside the focus area is not thermally affected [1], [2].

Additionally, the laser wavelength is below any statistical threshold of biological response per photon for DNA, RNA or proteins (Fig. 2), this light does not impair biological material [3], [4]. Therefore, these molecules can be isolated from the specimen for downstream analyses and applications and even living cells can be captured [5], [6].

As the effective laser energy is concentrated at the minute focal spot only (see below), it is even possible to perform laser microsurgery within living specimens. Numerous publications in the field of cell and developmental biology or from assisted human fertilization procedures have proven the safety of UV-A laser-based microsurgery and microdissection [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19].

The ablation process of the laser microdissection technology produces a clear gap between selected and unwanted material. Subsequently, the selected and circum-dissected area can be catapulted from the object plane with a single laser pulse for the distance of several millimeters against gravity directly into a capture device mounted within the laser beam. This patented “Laser Pressure Catapulting” technology marks a breakthrough in modern laser capture methods, and enables the entire non-contact preparation of pure and homogeneous samples in a fast and reliable manner [20], [21], [22], [23], [24]. The range of catapulted specimen spans from subcellular organelles up to entire small organisms, e.g., the nematode C. elegans, which easily survives this unique catapulting procedure [2], [5], [25].

Laser microdissection is feasible with objective magnifications from 5× to 100×. The cutting size of the implemented laser depends on the focal diameter of the beam. This focus results from the wavelength and beam quality of the laser, the magnification and numerical aperture (N.A.) of the applied objective, and finally from the specimen's absorbance behavior. For best focusing results, a laser of high beam quality and an objective with high N.A. is required. Higher aperture objectives, e.g., a 100× oil immersion objective (N.A.=1.3), are necessary for minimum cutting size of less than 600 to 700 nm [26], enabling microdissection and microsurgery even of single nuclei, filaments, chromosomes, or chromosomal parts [27]. The precise cutting of the PALM MicroBeam system is supported by an extremely precise (positioning better than 1 μm) motorized microscope stage.

For the capture of the catapulted samples different collection devices can be coupled to the MicroBeam. The most recent developed fully robotic unit called ‘RoboMover’ functions as a multi-purpose collection device, with adapters ranging from single caps of microfuge tubes up 96-well microtiter plates. Successful sample capture can easily be monitored with the so-called ‘cap-check’ function where isolated samples can be seen in the collection cap.

An intuitive graphical user interface for all functions is provided for facilitating the use of this laser microdissection system. This controlling software is the so-called ‘RoboSoftware’, managing the motion of the motorized microscope, the motorized microscope stage, the RoboMover capture device as well as the motor-driven laser settings and the optional fluorescence equipment. The RoboSoftware includes automated process routines as well as additional functions: A wide palette of drawing tools for marking the incision path in the preselection mode enables the outlining and color-coding up to thousands of independent target areas all over the entire slide or even from different slides. These selected target areas are listed in an ‘element list’ protocol, which allows target grouping and experimental scheduling. The list of elements (Fig. 3, upper right) is the main tool for summary and display of the outlined samples and corresponding area measures, the color dependent sorting of the outlined areas, and laser activation. Guided by the entries in the element list, the software will microdissect and/or catapult only elements showing the particular chosen color-code. This way, complex experimental setups can be planned and processed automatically. Different laser functions are available to process the preselected and outlined specimens and are described elsewhere [28], [29].

Non-contact laser microdissection and catapulting, as realized in PALM MicroBeam HT, allows largely automated and highly reliable capture of thousands of cells within short time; thus, enabling higher throughput specimen sampling, which is especially important for array techniques or proteomic studies.

Virtually any biological specimen can be used for sample capture with the non-contact LMPC technology and numerous protocols for sample preparation and downstream application techniques have been published during the past few years of laser micromanipulation and microdissection [5], [6], [24], [28], [30], [31], [32], [33].

Different sample preparation methods are usable with LMPC. Specimens can either be prepared directly onto a standard glass slide or on a membrane-spanned slide. Depending on the nature of the sample and the purpose of the experiment, catapulting may be performed from either of these two types of slides. Capturing from a membrane-spanned slide is more suitable for large tissue areas, tedious specimens or fragile samples such as cell smears, very small cells, cell nuclei, and chromosomes. The LPC membrane of these slides serves as a backbone that holds the selected tissue area close together and facilitates the capture. The laser first operates around the selected area following the preselected line and cuts the specimen together with the underlying membrane. With the catapult pulse of the laser, the entire isolated area is ejected out of the object plane and catapulted directly into the collection device. Due to the supporting membrane, the captured single cells or even large cell areas keep their morphology intact and can be visualized easily in the collection cap as cell-covered membrane-islets. Direct capturing from glass is even possible for archived material after removing the coverslip after removing the coverslip, but as the supporting membrane is missing, the harvested area is captured as small flakes into the collection vessel.

Specially developed membrane-mounted cell culture dishes can also be used for isolation of living cells for recultivation purposes. Thus, it is possible to catapult living cells grown on the surface of such membrane consumables and capture them alive [5], [6], [29], [30]. Common laser microdissection and capture techniques require a dry environment and are therefore restricted to fixed or frozen dry tissue sections or cell preparations [34]. In contrast, living cells need medium for growth and survival. Beyond the necessity of a humid environment, contamination is especially critical in living specimen, if the captured cells shall be recultivated after microdissection. Therefore, an entirely non-contact microdissection and capture method is desirable for isolation and ongoing cultivation of cultured cells. The LMPC process has no detrimental effects on the isolated cells, as they are shown to proliferate after isolation. These cells are still viable and unaltered after several LMPC rounds of laser microdissection, capture and recultivation as could be shown with CGH and M-FISH analyses [16], [17].

The high degree of automation realized in the latest generation of P.A.L.M. systems (PALM MicroBeam HT) can optionally be augmented by image analyzing software modules allowing automated functions for specimen identification and image processing. Coupled with any of these software modules, the MicroBeam system is able to detect, isolate and finally capture the specimen of interest, e.g., pre-specified tissue areas, fluorescent labeled rare cells, metaphase or FISH-treated cells in fully automated manner. Auto-marked areas can subsequently be extracted automatically by appropriate laser functions. These versatile automated software modules comprise also the advantage of fast and reliable detection and evaluation of particular cells or cell components, based on optimized classifiers or rule sets by means of morphological phenotypes. The efficient detection algorithms are adapted prior to LMPC to achieve integrated, interactive classifiers or rule sets for optimized recognition and accurate results.

An example of such a rule set generation is given here for the identification of lung cancer by applying object-oriented image processing for rapid sample recognition.

At present nucleic acids are extracted readily even from single cells and analyzed after subsequent amplification steps. In the case of proteins this is not within reach as amplification is not applicable to these biomolecules. Thus, the amount of homogeneous starting material for protein analysis has to be of much greater volume compared to RNA or DNA analyses.

However, modern laser microdissection and LPC instruments make great benefit from computer-assisted recognition methods. In the following case study material is used from lung cancer samples provided from pathological evidence (Fig. 4). The type of cancer that was detected was identified as squamous epithelial cancer and the goal was to research the differences between cancerous, stroma and inflammatory tissue.

In this study we show the automated recognition of the cancerous material by applying object-oriented image processing and contrast theoretically to two popular methods as are nearest neighborhood recognition and threshold recognition.

The image that is taken from the sample reflects a type of image, which consists of a variety of shapes and patterns to be categorized.

Distinction between the regions outlined in Fig. 5 cannot easily be undertaken by threshold conditions only, because a color threshold will solely point to nuclei, background or homogenous stained regions. As indicated by the red regions in Fig. 6 the threshold condition may lead to the same result in all three color channels, an increase of the threshold level results in blurring of the identified areas which in turn leads to decreased initial information. However, the threshold level will be placed: It will always remain difficult to render adjacent regions according to their morphological relationship, because any object classified by threshold condition will only be determined by its pixel inherent color.

Even if a fine-adjustment in the green threshold to 0.33 (non-relative to scene, i.e., the whole image of current processing) results in the green regions rendered in Fig. 7, it is difficult to extract the relevant geometries of cancer regions by concomitant boolean operations of the three color channels.

If nearest-neighborhood classification is applied, the pre-defined set of parameters of such a function is the main obstacle in pathology morphometry: Which parameter should be applied in which program sequence? If the sequence gets too sophisticated, there is great risk to overtrain the system by overloading the training set with too much and too detailed information. The result is quite often that on the training set excellent results in recognition are obtained; in test sets predominantly failure will occur.

Because of the drawbacks of the mentioned techniques of identification we decided for object-oriented pattern recognition as provided by DEFINIENS AG, München, in their Cellenger software suite. Applying this software, after an initial generation of seed objects, those objects are related to each other in a tree-like structure to gain more information from the scene as is revealed by plain color: the relationships between the objects. This information is crucial in the field of cancer, because, as in our example, squamous cancer exists in little islets that are dense of nuclei and those islets are also within a region of increased occurrence of those islets theirselves (Fig. 8).

Following those parameters - density, area, neighborhood and shape - one finds that, as shown in Fig. 9 for one set of objects, a parameter will be obtained which is no more purely discriminating but rather a complex combination describing a complex or fuzzy object.

As indicated by the pathologist, a factor influencing fuzzy logic classes of 0.3 reflects in detail his perception of which areas are to be marked as cancer and therefore extracted by LMPC.

If this parameter is applied, the result, which is conveyed to the LMPC instrument, is rendered in Fig. 10. It is obvious, that the identification and outlining by hand would be a time-consuming and tedious work, and as such subject to many errors caused by, e.g., non-standardized interpretation of the sample if manually done.

Therefore, the application of object-oriented pattern recognition is a very beneficial instrument to nurture the increase of precision provided by laser microdissection and non-contact extraction methods for life-science research.

Laser microdissection and pressure catapulting will supplement functional proteomic studies correlating protein profiles with morphological relevant features. Laser catapulting enables experiments with a higher throughput setup. From the respective capture device the specimens are dissolved and subsequently biomolecules of interest, i.e., DNA, RNA or proteins, are purified and committed to the corresponding downstream applications such as (RT-) PCR and microarray hybridization of microdissected DNA- and RNA samples, chromosomal analysis, MALDI/SELDI and nano-scale liquid chromatography (nLC) analysis of proteins [35], [36], [37], [38], [39], [40], [41], [42], [43], [44].

A very critical step for subsequent analyses is the homogeneity of the starting material. A selective procurement of specific cells via LMPC technology allows rapid and highly precise results. Furthermore, the starting amount of material can be lowered as only cells of interest are harvested.

For accurate and reliable isolation of starting material, it is necessary to comply with an optimized protocol [24], [33].

Typical RNA yields and concentrations from extraction of cells in compliance to these protocols were analyzed in an Agilent Bioanalyzer 2100 [45]. The clearly visible 18S and 28S rRNA-peaks indicate the excellent RNA quality with an integrity factor of RIN 10 (Fig. 11)!

Direct catapulting with no mechanical interference will save time, prevent the danger of loosing specimen during pipetting, and minimize contamination with unwanted material. In summary, this versatile laser micromanipulation and microdissection system is the state of the art tool, essential when pure sample generation is required throughout the entire field of modern molecular research and medical analyses.