Proteome sequence features carry signatures of the environmental niche of prokaryotes

Abstract

Background: Prokaryotic environmental adaptations occur at different levels within cells to ensure the preservation of genome integrity, proper protein folding and function as well as membrane fluidity. Although specific composition and structure of cellular components suitable for the variety of extreme conditions has already been postulated, a systematic study describing such adaptations has not yet been performed. We therefore explored whether the environmental niche of a prokaryote could be deduced from the sequence of its proteome. Finally, we aimed at finding the precise differences between proteome sequences of prokaryotes from different environments.

Results: We analyzed the proteomes of 192 prokaryotes from different habitats. We collected detailed information about the optimal growth conditions of each microorganism. Furthermore, we selected 42 physico-chemical properties of amino acids and computed their values for each proteome. Further, on the same set of features we applied two fundamentally different machine learning methods, Support Vector Machines and Random Forests, to successfully classify between bacteria and archaea, halophiles and non-halophiles, as well as mesophiles, thermophiles and mesothermophiles. Finally, we performed feature selection by using Random Forests.

Conclusions: To our knowledge, this is the first time that three different classification cases (domain of life, halophilicity and thermophilicity) of proteome adaptation are successfully performed with the same set of 42 features. The characteristic features of a specific adaptation constitute a signature that may help understanding the mechanisms of adaptation to extreme environments.

Figure 1

Four unique features used for classifications regarding domain of life revealed by the feature selection algorithm of RF. Pairs of box-and-whisker plots are shown for each feature: Leu content, average protein size in a proteome, His content, and 10-Cys content. Box-and-whisker plots represent bacteria and archaea from top to bottom. The feature values are normalized from 0 to 1 from left to right. (+) signs represent outliers.

Figure 2

Three unique features used for classifications regarding halophilicity revealed by the feature selection algorithm of RF. Pairs of box-and-whisker plots are shown for each feature: positive charge, normalized frequency of beta turn, and Phe content. Box-and-whisker plots represent non-halophiles and halophiles from top to bottom. The feature values are normalized from 0 to 1 from left to right. (+) signs represent outliers.

Figure 3

Four unique features used for classifications regarding thermophilicity revealed by the feature selection algorithm of RF. Triplets of box-and-whisker plots are shown for each feature: information measure for loop, Val content, Tyr content, and Chou-Fasman parameter of the coil conformation. Box-and-whisker plots represent mesophiles, mesothermophiles and thermophiles from top to bottom. The feature values are normalized from 0 to 1 from left to right. (+) signs represent outliers.