Limits...
Topological data analysis of Escherichia coli O157:H7 and non-O157 survival in soils.

Ibekwe AM, Ma J, Crowley DE, Yang CH, Johnson AM, Petrossian TC, Lum PY - Front Cell Infect Microbiol (2014)

Bottom Line: Network analysis showed that Shiga toxin negative strain E. coli O157:H7 4554 survived significantly longer in comparison to E. coli O157:H7 EDL 933, while the survival time of E. coli O157:NM was comparable to that of E. coli O157:H7 EDL 933 in all of the tested soils.Two non-O157 strains, E. coli O26:H11 and E. coli O103:H2 survived much longer than E. coli O91:H21 and the three strains of E. coli O157.We show that there are complex interactions between E. coli strain survival, microbial community structures, and soil parameters.

View Article: PubMed Central - PubMed

Affiliation: Agricultural Research Service-US Salinity Laboratory, United States Department of Agriculture Riverside, CA, USA.

ABSTRACT
Shiga toxin-producing E. coli O157:H7 and non-O157 have been implicated in many foodborne illnesses caused by the consumption of contaminated fresh produce. However, data on their persistence in soils are limited due to the complexity in datasets generated from different environmental variables and bacterial taxa. There is a continuing need to distinguish the various environmental variables and different bacterial groups to understand the relationships among these factors and the pathogen survival. Using an approach called Topological Data Analysis (TDA); we reconstructed the relationship structure of E. coli O157 and non-O157 survival in 32 soils (16 organic and 16 conventionally managed soils) from California (CA) and Arizona (AZ) with a multi-resolution output. In our study, we took a community approach based on total soil microbiome to study community level survival and examining the network of the community as a whole and the relationship between its topology and biological processes. TDA produces a geometric representation of complex data sets. Network analysis showed that Shiga toxin negative strain E. coli O157:H7 4554 survived significantly longer in comparison to E. coli O157:H7 EDL 933, while the survival time of E. coli O157:NM was comparable to that of E. coli O157:H7 EDL 933 in all of the tested soils. Two non-O157 strains, E. coli O26:H11 and E. coli O103:H2 survived much longer than E. coli O91:H21 and the three strains of E. coli O157. We show that there are complex interactions between E. coli strain survival, microbial community structures, and soil parameters.

Show MeSH

Related in: MedlinePlus

(A) Sample-sample relationships in a topological network. Using physical, chemical, and biological characteristics of the samples, we obtained a network that comprised of 4 sub-networks (A–D) and a singleton (single node comprising of 2 samples). The coloring here is by location, where each location is given a color (Salinas is red, Imperial Valley is green and Yuma is blue). Nodes that have a mixture of soils have colors in between as depicted in the color bar. Sub-networks structure indicates that physical, chemical, and biological characteristics primarily segregate the samples according to location, with Salinas being most different from soils from Yuma and Imperial Valley, (B). The same network is colored by soilmanagement type (organic (represented by 0) vs. conventional (represented by 1). The red nodes represent samples with conventional soil management while the blue nodes represented the organic soil management. The green and orange colored nodes represented mixed organic and conventional soil management with varying percent of mixture of the two types of management. (C) Another network is built using the same parameters except for resolution. The soil samples are analyzed at a lower resolution to ask if structure (D) and the singleton will merge with any part of the sub-networks. Sub-network (D), which comprised of samples from Yuma and Imperial, became part of sub-network (B) (colored nodes). Samples from sub-network (D) are not part of the gray nodes. The singleton however remained a singleton. The size of each node reflects the number of data points contained in the node. For (A,B), the distance metric and filters were Person correlation and Principal Metric SVD and secondary metric SVD. Metric, Norm Correlation; Lens, Principal Metric SVD value (Resolution 30, Gain 4.0x, Equalized) Secondary Metric SVD Value (Resolution 30, Gain 4.0x, Equalized). For (C), all analysis parameters remained the same except for resolution (20 instead of 30).
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4155871&req=5

Figure 1: (A) Sample-sample relationships in a topological network. Using physical, chemical, and biological characteristics of the samples, we obtained a network that comprised of 4 sub-networks (A–D) and a singleton (single node comprising of 2 samples). The coloring here is by location, where each location is given a color (Salinas is red, Imperial Valley is green and Yuma is blue). Nodes that have a mixture of soils have colors in between as depicted in the color bar. Sub-networks structure indicates that physical, chemical, and biological characteristics primarily segregate the samples according to location, with Salinas being most different from soils from Yuma and Imperial Valley, (B). The same network is colored by soilmanagement type (organic (represented by 0) vs. conventional (represented by 1). The red nodes represent samples with conventional soil management while the blue nodes represented the organic soil management. The green and orange colored nodes represented mixed organic and conventional soil management with varying percent of mixture of the two types of management. (C) Another network is built using the same parameters except for resolution. The soil samples are analyzed at a lower resolution to ask if structure (D) and the singleton will merge with any part of the sub-networks. Sub-network (D), which comprised of samples from Yuma and Imperial, became part of sub-network (B) (colored nodes). Samples from sub-network (D) are not part of the gray nodes. The singleton however remained a singleton. The size of each node reflects the number of data points contained in the node. For (A,B), the distance metric and filters were Person correlation and Principal Metric SVD and secondary metric SVD. Metric, Norm Correlation; Lens, Principal Metric SVD value (Resolution 30, Gain 4.0x, Equalized) Secondary Metric SVD Value (Resolution 30, Gain 4.0x, Equalized). For (C), all analysis parameters remained the same except for resolution (20 instead of 30).

Mentions: Using the properties of physical, chemical, and biological characteristic of these soil samples as variables, we clustered the soil samples using TDA. The resulting network represents the soil samples clustering into sub-networks. Figure 1A shows 4 sub-networks A, B, C and D with B and C connecting to form a larger sub- network. There is also a singleton (1 node comprising of 2 soil samples from Salinas that stood apart from everything else). The network can also be colored by various factors and characteristics such as location and soil management type for visualization (Figure 1). In addition, we can also apply statistics to probe what factors distinguished our soils into sub-networks. We found that “location” was one of the key differences between the sub-networks (Kolmogorov-Smirnov test PV < 0.0003). In order to visualize the effect of “location” on the soil samples, Figure 1 is colored by “location.” We show that soil samples from the Salinas areas (A) completely formed a separate sub-network from soil samples from the Imperial and Yuma areas (B, C, and D) as indicated by the color. This indicates that physical, chemical, and biological characteristic of these soil samples collectively are quite different from location to location, especially the soil samples from Salinas, which formed a distinct sub-network (A). Soil samples from Yuma and Imperial are closer to each, forming a sub-network that looks like a dumb bell, with some samples from Imperial clustering at left side of dumb bell (B) and the rest of the network comprised of a mixture between samples from Yuma and Imperial. Interestingly, physical, chemical, and biological properties measured of these soil samples did not differentiate between conventional and organic soil management as seen from the non-enrichment of any one type of soil management in the network (also see Table 2, where the P-value for soil management as a differentiating factor between those sub-networks was 0.4126). To further investigate sub-network D and the singleton, another network analysis was performed using the same distance metric and mathematical lenses but at a lower resolution (20 instead of 30). Sub-network D, which comprised of samples from Yuma and Imperial, became part of sub-network C (Figure 1C). The singleton however remained a singleton, indicating that these samples are fundamentally different from the rest of the samples due to unknown reasons including quality of the samples.


Topological data analysis of Escherichia coli O157:H7 and non-O157 survival in soils.

Ibekwe AM, Ma J, Crowley DE, Yang CH, Johnson AM, Petrossian TC, Lum PY - Front Cell Infect Microbiol (2014)

(A) Sample-sample relationships in a topological network. Using physical, chemical, and biological characteristics of the samples, we obtained a network that comprised of 4 sub-networks (A–D) and a singleton (single node comprising of 2 samples). The coloring here is by location, where each location is given a color (Salinas is red, Imperial Valley is green and Yuma is blue). Nodes that have a mixture of soils have colors in between as depicted in the color bar. Sub-networks structure indicates that physical, chemical, and biological characteristics primarily segregate the samples according to location, with Salinas being most different from soils from Yuma and Imperial Valley, (B). The same network is colored by soilmanagement type (organic (represented by 0) vs. conventional (represented by 1). The red nodes represent samples with conventional soil management while the blue nodes represented the organic soil management. The green and orange colored nodes represented mixed organic and conventional soil management with varying percent of mixture of the two types of management. (C) Another network is built using the same parameters except for resolution. The soil samples are analyzed at a lower resolution to ask if structure (D) and the singleton will merge with any part of the sub-networks. Sub-network (D), which comprised of samples from Yuma and Imperial, became part of sub-network (B) (colored nodes). Samples from sub-network (D) are not part of the gray nodes. The singleton however remained a singleton. The size of each node reflects the number of data points contained in the node. For (A,B), the distance metric and filters were Person correlation and Principal Metric SVD and secondary metric SVD. Metric, Norm Correlation; Lens, Principal Metric SVD value (Resolution 30, Gain 4.0x, Equalized) Secondary Metric SVD Value (Resolution 30, Gain 4.0x, Equalized). For (C), all analysis parameters remained the same except for resolution (20 instead of 30).
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4155871&req=5

Figure 1: (A) Sample-sample relationships in a topological network. Using physical, chemical, and biological characteristics of the samples, we obtained a network that comprised of 4 sub-networks (A–D) and a singleton (single node comprising of 2 samples). The coloring here is by location, where each location is given a color (Salinas is red, Imperial Valley is green and Yuma is blue). Nodes that have a mixture of soils have colors in between as depicted in the color bar. Sub-networks structure indicates that physical, chemical, and biological characteristics primarily segregate the samples according to location, with Salinas being most different from soils from Yuma and Imperial Valley, (B). The same network is colored by soilmanagement type (organic (represented by 0) vs. conventional (represented by 1). The red nodes represent samples with conventional soil management while the blue nodes represented the organic soil management. The green and orange colored nodes represented mixed organic and conventional soil management with varying percent of mixture of the two types of management. (C) Another network is built using the same parameters except for resolution. The soil samples are analyzed at a lower resolution to ask if structure (D) and the singleton will merge with any part of the sub-networks. Sub-network (D), which comprised of samples from Yuma and Imperial, became part of sub-network (B) (colored nodes). Samples from sub-network (D) are not part of the gray nodes. The singleton however remained a singleton. The size of each node reflects the number of data points contained in the node. For (A,B), the distance metric and filters were Person correlation and Principal Metric SVD and secondary metric SVD. Metric, Norm Correlation; Lens, Principal Metric SVD value (Resolution 30, Gain 4.0x, Equalized) Secondary Metric SVD Value (Resolution 30, Gain 4.0x, Equalized). For (C), all analysis parameters remained the same except for resolution (20 instead of 30).
Mentions: Using the properties of physical, chemical, and biological characteristic of these soil samples as variables, we clustered the soil samples using TDA. The resulting network represents the soil samples clustering into sub-networks. Figure 1A shows 4 sub-networks A, B, C and D with B and C connecting to form a larger sub- network. There is also a singleton (1 node comprising of 2 soil samples from Salinas that stood apart from everything else). The network can also be colored by various factors and characteristics such as location and soil management type for visualization (Figure 1). In addition, we can also apply statistics to probe what factors distinguished our soils into sub-networks. We found that “location” was one of the key differences between the sub-networks (Kolmogorov-Smirnov test PV < 0.0003). In order to visualize the effect of “location” on the soil samples, Figure 1 is colored by “location.” We show that soil samples from the Salinas areas (A) completely formed a separate sub-network from soil samples from the Imperial and Yuma areas (B, C, and D) as indicated by the color. This indicates that physical, chemical, and biological characteristic of these soil samples collectively are quite different from location to location, especially the soil samples from Salinas, which formed a distinct sub-network (A). Soil samples from Yuma and Imperial are closer to each, forming a sub-network that looks like a dumb bell, with some samples from Imperial clustering at left side of dumb bell (B) and the rest of the network comprised of a mixture between samples from Yuma and Imperial. Interestingly, physical, chemical, and biological properties measured of these soil samples did not differentiate between conventional and organic soil management as seen from the non-enrichment of any one type of soil management in the network (also see Table 2, where the P-value for soil management as a differentiating factor between those sub-networks was 0.4126). To further investigate sub-network D and the singleton, another network analysis was performed using the same distance metric and mathematical lenses but at a lower resolution (20 instead of 30). Sub-network D, which comprised of samples from Yuma and Imperial, became part of sub-network C (Figure 1C). The singleton however remained a singleton, indicating that these samples are fundamentally different from the rest of the samples due to unknown reasons including quality of the samples.

Bottom Line: Network analysis showed that Shiga toxin negative strain E. coli O157:H7 4554 survived significantly longer in comparison to E. coli O157:H7 EDL 933, while the survival time of E. coli O157:NM was comparable to that of E. coli O157:H7 EDL 933 in all of the tested soils.Two non-O157 strains, E. coli O26:H11 and E. coli O103:H2 survived much longer than E. coli O91:H21 and the three strains of E. coli O157.We show that there are complex interactions between E. coli strain survival, microbial community structures, and soil parameters.

View Article: PubMed Central - PubMed

Affiliation: Agricultural Research Service-US Salinity Laboratory, United States Department of Agriculture Riverside, CA, USA.

ABSTRACT
Shiga toxin-producing E. coli O157:H7 and non-O157 have been implicated in many foodborne illnesses caused by the consumption of contaminated fresh produce. However, data on their persistence in soils are limited due to the complexity in datasets generated from different environmental variables and bacterial taxa. There is a continuing need to distinguish the various environmental variables and different bacterial groups to understand the relationships among these factors and the pathogen survival. Using an approach called Topological Data Analysis (TDA); we reconstructed the relationship structure of E. coli O157 and non-O157 survival in 32 soils (16 organic and 16 conventionally managed soils) from California (CA) and Arizona (AZ) with a multi-resolution output. In our study, we took a community approach based on total soil microbiome to study community level survival and examining the network of the community as a whole and the relationship between its topology and biological processes. TDA produces a geometric representation of complex data sets. Network analysis showed that Shiga toxin negative strain E. coli O157:H7 4554 survived significantly longer in comparison to E. coli O157:H7 EDL 933, while the survival time of E. coli O157:NM was comparable to that of E. coli O157:H7 EDL 933 in all of the tested soils. Two non-O157 strains, E. coli O26:H11 and E. coli O103:H2 survived much longer than E. coli O91:H21 and the three strains of E. coli O157. We show that there are complex interactions between E. coli strain survival, microbial community structures, and soil parameters.

Show MeSH
Related in: MedlinePlus