Dazzo, F.B., J. Liu, A. Jain, G. Tang, C. Gross, C. Reddy, C. Monosmith, A. Prabhu, R. Peretz, G. Zhu, J. Wang, M. Li, N. Philips, A. Baruti, R. Longueuil, C. Meyers, D. Nasr, I. Leader, S. Zamani, C. Passmore, L. Doherty, S. Dixon, P. Smith, D. McGarrell, S. Pierce, S. Gantner, S. Nakano, A. Smucker, E. Polone, A. Tondello, A. Squartini, Y. Yanni, and R. Hollingsworth
Presented at the All Scientist and GLBRC Sustainability Meeting (2009-05-05 to 2009-05-07 )
A major challenge in microbial ecology is to develop reliable methods of computer-assisted microscopy that can process and analyze complex digital images of microbial populations and communities at single cell resolution, and compute useful ecological characteristics of their organization and structure in situ without cultivation. To address this challenge, our team of microbiologists, mathematicians and computer scientists has been developing image analysis software that can extract the full information content in digital images of actively growing microbial communities, thereby strengthening microscopy-based methods for understanding microbial ecology. Our software package, called CMEIAS (Center for Microbial Ecology Image Analysis System) consists of a suite of application tools for various stages of image bioinformatics, including image processing and segmentation, object analysis and object classification, data processing, statistical analysis and exploratory data mining. Free downloads of these computing tools are supported with well-illustrated user manuals, interactive training tutorials with images, and audio-visual tutorials at our website.
The first release version (v. 1.27) of CMEIAS operates within UTHSCSA ImageTool v. 1.27 in a PC running Windows 2000/XP Pro/Vista (32-bit). It analyzes foreground objects in images, reports their numerical abundance and 24 attributes of size, shape and luminosity, provides a 1-dimensional object classifier using a single, user-specified measurement attribute and ≤ 15 upper bin limits (≤ 16 classes), and also features a unique semi-automatic supervised hierarchical tree classifier of all major plus several minor microbial morphotypes (equivalent to 98% of the genera in Bergey’s Manual 9th ed.) operating in 14-dimensional space. The CMEIAS morphotype classifier can monitor spatial-temporal changes in abundance of each microbial morphotype present within dynamically growing communities at frequencies down to 0.1% with 97% accuracy, and has flexibility to add up to 5 additional (rare) morphotypes if present. A CMEIAS ver. 1.28 update is scheduled for release in spring ’09 and includes new user support files in a fully-featured chm Help file, an updated user manual, and expanded training tutorials following major revisions based on feedback from an international team of beta-testers. Also, installation of CMEIAS ver. 1.28 has been simplified by development of an intelligent setup.exe.
Major upgrades in CMEIAS software applications are at various stages of development. The suite will include a CMEIAS ver. 3.1 upgrade of the ImageTool host program with numerous revisions in displayed text that improve the user-friendly operation, three object segmentation tools, upgrades of all the object analysis and classification features provided in CMEIAS ver. 1.28, plus numerous new features that expand its utility and awesome computing power.
Three image-editing tools are included in the major CMEIAS upgrade to facilitate the segmentation of objects in images prior to their analysis. First, a script of CMEIAS Actions operating in Adobe Photoshop / Image Processing Tool Kit has been developed to semi-automate the editing of foreground and background pixels in brightfield, phase contrast, fluorescence and SEM images of microbial communities in various habitats, and operates with 97.1% accuracy. Second, a stand-alone CMEIAS Color Segmentation program has been developed to facilitate the segmentation of foreground microbial objects within complex RGB digital images where color differentiation really matters most, e.g., immunofluorescence, FISH, activity/viability stains, reporter gene expression, Gram stains, etc. This application interactively samples a training set of local color pixels representing the microbes of interest, then finds each cell’s boundary, and finally creates a new RGB segmented output image containing these colored microbes in a noise-free background. Its performance has been extensively tested on many complex color images analyzed pixel-by-pixel one at a time, and operates with ~99% accuracy. Third, a CMEIAS Object Separation plugin has been developed to operate within the ImageTool host program. It automatically splits touching objects (microbial cells) in thresholded images with 95.8% accuracy, and has a unique design that avoids the deletion of pixels resulting in erroneous size reduction of the segmented objects as occurs with other automated object separation tools (e.g., watershed) commonly featured in other image processing software.
The image analysis core of the CMEIAS 3.1 upgrade currently includes 92+ different measurements of object analysis and object classification, designed to extract five major ecologically relevant types of quantitative information in digital images of microbial communities. These include: (i) classification of morphological diversity using pattern recognition algorithms combined with up to 57+ additional measurement attributes of cell size, shape, luminance, and spatial distribution; (ii) microbial abundance, using cell counts, spatial density, dilution adjusted concentration, filamentous length, biovolume, biomass C, and biosurface area of each individual cell, each individual population of morphotypes or the entire community; (iii) metabolic activity / viability differentiated by color staining or microdensitometry, (iv) autecology / phylogeny using fluorescent molecular probes (e.g., FISH, immunofluorescence) or autofluorescent cell components (e.g., GFP, F420) and color image segmentation, and (v) in situ spatial pattern analysis of microbial colonization on surfaces.
Notable among the measurement attributes of CMEIAS Object Analysis are 8 biovolume formulas to measuring this cell abundance parameter accurately for each microbial morphotype classified by CMEIAS, attributes that employ user-specified variables in a universal allometric scaling formula to compute biomass carbon from biovolume data, attributes that discriminate biofilm architectures (e.g., % Substratum Coverage, Areal Porosity, Mean Diffusion Radius, Maximum Diffusion Radius), various attributes of luminosity for in situ microdensitometry studies of gene expression and metabolic activity, and various attributes for spatial distribution analysis (detailed below).
The CMEIAS 3.1 upgrade is designed to extract three different types of spatial distribution information from microbes that defines their colonization behavior in situ_. These include plot-less point pattern analysis of spatial randomness, plot-based quadrat-lattice analysis of local spatial density, and geostatistical analysis to mathematically model spatially autocorrelated attributes with kriging map interpolation over the entire georeferenced domain. This CMEIAS spatial analysis module features several measurement attributes that report the location of each individual microbial object (Cartesian x_, y coordinates relative to a landmark origin), linear distances to their 1st and 2nd nearest neighbors, Ĝhat nearest neighbor empirical distribution, and a cluster index that measures the degree to which each cell is spatially aggregated with its neighbors. A Cumulative Object Analysis plugin has been developed to extract data on spatial attributes per unit of sampling area (whole image, quadrat, or inscribed polygon area of interest). These attributes include local spatial density using area-weighted abundance measurements of cell concentration, cumulative length / biovolume / biomass C / and biosurface area, and spatial clustering based on descriptive statistics of the 1st and 2nd nearest neighbor distances. This plugin also reports the Cartesian coordinates for postings at the quadrat’s geometric center, the mean center of objects weighted by their local density, or at randomized location within the quadrat for all 3 types of in situ spatial analysis. This cumulative object analysis plugin is supported by an AreaPolygonAOI script for plot-based spatial analysis when the foreground objects of interest are enclosed within a user-defined polygon area instead of the entire image. The spatial analysis module is also supported by a stand-alone CMEIAS Quadrat Maker executable application designed to facilitate the optimization of grid lattice dimensions that divide an image into smaller, constant size quadrats for high-resolution spatial sampling of local microbial density, and then prepares a small image of each size-optimized quadrat ready for plot-based and geostatistical spatial distribution analyses.
The Object Classification plugin of the CMEIAS 3.1 upgrade features 5 different supervised object classifiers. Other than an expanded list of measurement attributes from which to choose, the 1-dimensional object size / shape classifier and the 14-dimensional object morphotype classifier retain their original design of operation featured in the earlier ImageTool-CMEIAS release version 1.27. A third CMEIAS object classifier has been developed to subclassify each morphotype group into Operational Morphological Units (OMU), based on their morphological signature defined by a multilinear matrix of upper bin limits optimized as the least-overlapping borders that separate clusters for each measurement attribute used in the classification scheme. These OMU classification assignments are defined by 2 size border files; one being a default file of size classifications derived from cluster analyses of microbial size data in Bergey’s knowledgebase of taxonomically described microorganisms; the other being an infinitely adaptable, user-defined size border file optimized for classification of all OMU’s present in the community under investigation. Supporting the OMU classifier are an improved set of easily recognized pseudocolor assignments for each morphotype in the classified output image, an edit tool to reassign OMU classes when located at the continuum of borders in the 14-dimensional space used to classify morphotypes, and a feature to reconstruct new images of morphotype-specific populations derived from classified community images containing multiple morphotypes. Also supporting the OMU classifier is a stand-alone executable CMEIAS Size Border Analysis data exploration and mining program that performs a Monte Carlo cluster analysis to optimize the decision boundaries of upper class limits for building the user-defined size border file. Each cluster scheme with a high Student t value and associated +>+95% confidence limit is tested and validated using the 1-D object classifier, and the resultant optimized scheme of “best cut” upper bin limits is then written into the user-defined size border code to operate the OMU classifier. The flexible multilinear design of this size border file expands the ability of the OMU classifier to recognize and classify an almost unlimited range of OMU diversity optimized for the specific community under investigation. The high-low size border tool in this program analyzes size range data extracted from populations of known microbial taxa to optimize the cluster scheme of upper bin limits to incorporate into the generic default size border file for classification of OMUs in a variety of communities and habitats, rather than the user-defined size border file that is optimized for a specific community.
Finally, the 2 remaining object classifiers in the CMEIAS 3.1 upgrade are of the semi-unsupervised type, associating the 14-D morphotype classifier and the OMU classifier with a user-selected set of additional measurement attributes available in the Object Analysis plugin. These classifiers produce an output matrix table of mixed object classification and object analysis data reported for each individual microbial object. These outputs of mixed data matrices are then evaluated by various exploratory morphotype-weighted cluster analysis and data mining techniques to compute their unsupervised classification based on the (dis)similarities between microbial community structures.
Two software applications are being built to process and analyze CMEIAS data within MS Excel. The CMEIAS Data Preparation macro compiles and concatenates CMEIAS object analysis and classification data extracted from multiple images within the same community dataset, and automatically builds the properly formatted input tables of data for further analysis. The CMEIAS Data Analysis addin computes descriptive statistics, 2-d scatterplots and frequency distribution analyses with user-selected bin parameters on CMEIAS object analysis data, analyzes the tolerance envelope of confidence intervals surrounding CMEIAS object classification data derived from multiple image datasets to determine if the sample size (number of images and cell objects found within them) is sufficient to estimate diversity, constructs ranked abundance plots of community structure, computes / tabulates and plots 19 diversity indices and 12 similarity coefficients of classification data to compare community structures, and analyzes plot-less, plot-based and georeferenced data for spatial distribution analysis.
Recent collaborative projects using CMEIAS include in situ analysis of the autecological biogeography of candidate biofertilizer inoculant strains of rhizobia that promote rice growth, microbial community analyses of corn leaf surfaces (normal vs. genetically engineered BT-corn), distribution and abundance of active communities within soil aggregates, seasonal succession and food-web dynamics of epilithic microbial biofilm communities on streambed rocks in Japanese freshwater streams, discrimination of microbial biofilm architectures developing on various surface polymers coated on glass and submerged in the Red Cedar River (MSU campus), whole community analysis of microbial biofilms developing in pristine Italian Alpine lakes, in situ spatial scale of bacterial cell-to-cell communication (quorum sensing) at single-cell resolution, and shifts in community structure of human vaginal microflora in health and bacterial vaginosis disease.
In summary, CMEIAS-based applications can help to fill major gaps in our understanding of microbial ecology by providing accurate, robust and user-friendly computing tools that extract ecologically important, quantitative information from digital images of microbes, at spatial scales relevant to their diversity, abundance, physiology and in situ distribution. The awesome computational power of CMEIAS used in conjunction with computer-assisted microscopy at single-cell resolution adds an exciting new dimension to microbial community analysis, and it is especially valuable when combined with molecular-based and other methods of polyphasic analysis. CMEIAS tools are admirably suited for microbial community analyses currently under investigation in this KBS-LTER program focused on row crop agriculture and in other studies of microbial community analysis conducted in other LTER programs. For instance, early studies using computer-assisted microscopy and CMEIAS image analysis are underway to examine the epiphytic microbial communities that develop on dominant aquatic plants (lilly pad, sawgrass and peiphyton) in the Florida everglades ecosystem as a link to the Florida Coastal Everglades LTER program hosted at Florida International University.
Finally, the CMEIAS website includes a page entitled “Publications using CMEIAS” which contain 68 entries (most are hyperlinked) at time of writing (4-27/09) that describe various aspects of CMEIAS research and its applications as background information for this extended summary.
Back to meeting | Show |