This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).

This paper describes how the quantitative analytical tools of CMEIAS image analysis software can be used to investigate

Our team of microbiologists, mathematicians and computer scientists has been developing a suite of software applications for computer-assisted microscopy to enhance studies of microbial ecophysiology in natural and managed habitats. The long-range goal is to build a computing toolkit that strengthens microscopy-based approaches for understanding microbial ecology at spatial scales directly relevant to their ecological niches, without the need for cultivation [

A central goal in spatial ecology is to define what a measured characteristic at one location can reveal about that same variable at neighboring locations. Analyses of

At the core of these positive (cooperative) and negative (conflict) interactions that reduce spatial randomness are various types of microbial cell-to-cell communication events that regulate genes affecting their colonization behavior. The key connection between biofilm spatial ecology and cell-to-cell communication occurs when positive or negative interactions are found to be spatially autocorrelated,

This first experimental system involved many types of spatial pattern analyses to test for microbial interactions indicative of cell-to-cell communication among individual microbes in natural biofilms. Microbial assemblages were allowed to develop on clean glass microscope slides submerged for four summer days in the Red Cedar River on the campus of Michigan State University (East Lansing, MI, USA). Slides were retrieved and their underside wiped clean. The top surfaces of the slides were mounted in water with a cover slip and examined by phase-contrast light microscopy using a 100X Planapochromat Phase 3 objective lens to resolve individual bacterial cells. Digital 8-bit grayscale images of the biofilms were acquired using a monochrome digital camera, then segmented to binary, spatially calibrated and analyzed using CMEIAS image analysis software [

The accuracy of CMEIAS image analysis software to measure and compute the object centroid-to-centroid 1^{st} and 2^{nd} nearest neighbor distances was evaluated using a high-resolution 3-frequency grid distortion target (Edmund Optics, Barington, New Jersey, USA) as ground truth. The accuracy of CMEIAS color segmentation software used to process the color images for analysis was previously measured as 99+% [

The second experimental system was designed to further advance our understanding of bacterial cell-to-cell communication during their colonization of plant roots. CMEIAS image analysis was used to reevaluate the spatial scale of calling distances and the variations in intensity of gene expression activated by extracellular signal communication molecules produced by neighboring cells of

Axenically grown seedling roots were inoculated with both the AHL-producer and AHL-sensor strains of ^{9} cells per plant, grown gnotobiotically, harvested and examined by laser scanning microscopy in the epifluorescence confocal mode [

^{st} and 2^{nd} nearest neighboring cell, the cumulative Empirical Distribution Function of the 1^{st} nearest neighbor distance, and the CMEIAS Cluster Index that measures the clustering intensity of each cell in relation to its local environment [^{st} and 2^{nd} nearest neighbor distances used to compute these spatial attributes were 3.2% and 2.3%, respectively, with an overall combined accuracy of 97.2% (n = 38).

Statistical analyses indicated that the frequency distribution of the 1^{st} and 2^{nd} nearest neighbor distances and cluster indices of cells in these biofilms were significantly skewed and lacked normality (

These results justified subsequent analyses to test whether the spatial patterns of microbial distribution deviated from complete spatial randomness, and if so, the intensity of their clustered aggregation. Several methods of spatial statistics were performed on the CMEIAS data to test this hypothesis. The first evaluation, called the ^{st} nearest neighbor distances measured between individual cells in the sample compared to the distribution that would result if the pattern were completely random. In this latter case, the data points would distribute along the diagonal, dashed blue staircase. The resultant plot of the empirical distribution function indicated that both biofilm samples had aggregated patterns of distribution, and that this colonization behavior was more intense in biofilm CS25 (

The same data on 1^{st} and 2^{nd} nearest neighbor distances were evaluated by four other spatial point pattern statistical tests (Holgate Aggregation, Russ Randomness, Clark and Evans Dispersion/Spatial Density, and Hopkins and Skellam Aggregation) for complete randomness in spatial distribution of cells. These 4 tests indicate spatially aggregated patterns when the corresponding upper class limits of their indices computed from the 1^{st} and 2^{nd} nearest neighbor distances are >0.5, <1.0, <1.0 and >1.0, respectively. The results of all of these point pattern tests rejected the null hypothesis of spatial randomness in favor of the alternative hypothesis of aggregated distributions, and the intensity of this colonization behavior was significantly higher in the CS25 biofilm sample (

Concepts derived from fractal theory are fundamental to an understanding of the landscape complexity of scale-related phenomena in ecology [

The cluster index, computed as the inverse of 1st nearest neighbor distances, is a sensitive, local

The CMEIAS 1-dimensional classifier was used to sort individual cells in the two biofilm images into bins based on division of a scale defined by a single measurement feature [

The frequency distribution of the cluster indices for these microbes is presented in

Geostatistical analysis is the most powerful category of spatial pattern analysis that can be used to unravel the spatial uncertainty of interactions between individual microorganisms. This geospatial method examines the continuity or continuous variation of spatial patterns over the entire domain by testing whether a user-defined, continuously distributed regionalized variable (called the “Z-variate”) is spatial autocorrelated,

The mathematical model that statistically best fits the variogram's autocorrelation data (represented by the solid blue line in

After the autocorrelation model was optimally fit to the data, 1,000 simulations of multigrid refinements of the model were computed to produce the corresponding kriging maps. These high-resolution pseudocolored graphics elegantly map the spatial autocorrelation in the data. This knowledge is used to derive accurate, unbiased estimates of the spatial continuity of Z-variate values within the sampling unit, thereby precisely resolving detailed spatial patterns with known variance for each interpolated point [

Considered collectively, the results presented in

It is generally thought that high bacterial population densities are needed to exceed the threshold level of AHL signal concentrations required to activate genes and their physiological functions. Hence, this specific type of microbial communication has become known as ‘quorum sensing’ that functions primarily as sensors of high population density, thus optimizing the expression of functions that are most beneficial when simultaneously performed by dense populations. Despite its wide appeal, this quorum sensing paradigm has been challenged since the methods commonly used to detect it

An example of direct ^{st} nearest neighboring (red-fluorescent) AHL-source cell (

The use of sensor cells that report intracellular fluorescent proteins permits the measurement of expression levels of gene activation by AHL gradients

Albert Einstein once said: “Sometimes what is counted doesn't count, and what really counts cannot be counted!” Studies on cell communication within microbial biofilms can benefit by incorporating a spatial context, because in this habitat, “spatial relationships really matter”. This paper describes how CMEIAS-assisted microscopy can help measure what really counts in this field: quantitative data in spatial ecology that bestow new insights to help piece together the complicated story of biofilm development by gaining a clearer understanding of spatially-dependent interactions involving cell communication at real-world spatial scales important to the microbe's perspective. Characterizing the spatial scale of bacterial interactions resulting from cell-to-cell communication is important because it is a strong determinant of spatial patterns that reflect their colonization behavior, and certain ecological processes may operate at a particular scale but not at all at a different scale.

Eighteen different experimental tests were described using two types of sensors. Sixteen tests using the first sensor type (nearest neighbor-based cluster indices) differentiated the intensity of autocorrelated spatial structure in aggregated colonization behavior between two natural freshwater biofilms. Two other tests using the second type of sensor (genetically engineered strains to report signal molecule production and perception) indicated the range of calling distances between individual bacterial cells

Spatial ecology studies reported here confirm and expand upon previously acquired data [

Future work should continue to further test and validate these ecophysiological models of cell communication and colonization behavior by microbes at single-cell resolution during dynamic stages of biofilm development, including their spatiotemporal redistribution within various natural habitats.

This work was supported by the US-Egypt Science & Technology Joint fund project 58-3148-1-140, the Michigan State University Kellogg Biological Station Long-Term Ecological Research program, and the MSU Center for Microbial Ecology. I greatly appreciate the collaborations with Anton Hartmann, Stephan Gantner, Youssef Yanni, Andrea Squartini, Pedro Mateos, Rawle Hollingsworth and numerous current students participating in ongoing research developments of CMEIAS image analysis software described here. This work is dedicated in memory of my friend and colleague, Rawle Hollingsworth.

Cumulative empirical distribution function of the nearest neighbor distances between individual bacteria within images of biofilms CS4 (green diamonds) and CS25 (red squares). Differences in intensity of aggregated patterns are indicated by empirical distribution values that ascend above the diagonal dashed line of complete spatial randomness.

Frequency distribution of each cell's Cluster Index within CS4 and CS25 biofilms.

Isotropic semivariograms of the spatially autocorrelated cluster indices of cells in the CS4 (

2-dimensional Kriging maps of the spatially autocorrelated local cluster indices for the microbes within the CS4 (

Computer-assisted microscopy of AHL-mediated cell-to-cell communication between red-fluorescent source and green-fluorescent sensor reporter strains of

(

Statistical inference tests of the distributions of nearest neighbor distances and clustered indices between individual bacteria within images of the CS4 and CS25 freshwater biofilm assemblages.

^{st} Nearest Neighbor Distance (μm) |
^{nd} Nearest Neighbor Distance (μm) |
^{−1}) | ||||
---|---|---|---|---|---|---|

| ||||||

Normal Distribution? (Shapiro-Wilks W) | 0.906 ^{a} |
0.715 ^{a} |
0.937 ^{a} |
0.817 ^{a} |
0.924 ^{a} |
0.979 ^{a} |

Significant Skewness? | 1.028 ^{a} |
2.794 ^{a} |
1.143 ^{a} |
1.922 ^{a} |
0.848 ^{a} |
0.399 ^{a} |

Median | 1.591 | 1.257 | 2.711 | 1.905 | 0.629 | 0.796 |

Significantly Different Median? (Mann-Whitney U) | CS4 > CS25 (Yes, U = 115293 ^{a} |
CS4 > CS25 (Yes, U = 113652 ^{a} |
CS25 > CS4 (Yes, U = 112649 ^{a} |

Significance level is indicated when the P value was ≤0.05.

Ecological statistics to test if the microbial distribution in the CS4 and CS25 biofilms deviate from complete spatial randomness, and compare the intensity of spatial aggregation in their colonization behavior.

^{b} |
^{b} |
|||
---|---|---|---|---|

Holgate Aggregation | 1^{st} & 2^{nd} NND ^{a} |
0.556 (0.001) | 0.561 (0.001) | CS25 ≫ CS4 |

Russ Randomness | 1^{st} & 2^{nd} NND |
0.961 (0.01) | 0.727 (0.001) | CS25 > CS4 |

Clark & Evans Dispersion | 1^{st} NND, Spatial Density |
0.929 (0.01) | 0.764 (0.001) | CS25 > CS4 |

Spatial Density (cells/mm^{2}) |
Cell Count per Image^{2} |
54,735 | 63,458 | CS25 > CS4 |

Hopkins & Skellem Aggregation | Random Point to Nearest^{st} NND |
1.523 (0.000) | 12.582 (0.000) | CS25 ≫ CS4 |

Fractal Geometry | Cumulative Intersection | 1.892 | 2.140 | CS25 > CS4 |

Effective Range Separation Distance | Object Centroid X,Y |
4.4 μm | 46.5 μm | CS25 ≫ CS4 |

Moran's I Spatial Autocorrelation Index | Object Centroid X,Y |
4.244 | 10.579 | CS25 > CS4 |

NND = nearest neighbor distances;

P values are indicated in parentheses.