Effective identification of bacterial type III secretion signals using joint element features

Abstract

Type III secretion system (T3SS) plays important roles in bacteria and host cell interactions by specifically translocating type III effectors into the cytoplasm of the host cells. The N-terminal amino acid sequences of the bacterial type III effectors determine their specific secretion via type III secretion conduits. It is still unclear as to how the N-terminal sequences guide this specificity. In this work, the amino acid composition, secondary structure, and solvent accessibility in the N-termini of type III and non-type III secreted proteins were compared and contrasted. A high-efficacy mathematical model based on these joint features was developed to distinguish the type III proteins from the non-type III ones. The results indicate that secondary structure and solvent accessibility may make important contribution to the specific recognition of type III secretion signals. Analysis also showed that the joint feature of the N-terminal 6(th)-10(th) amino acids are especially important for guiding specific type III secretion. Furthermore, a genome-wide screening was performed to predict Salmonella type III secreted proteins, and 8 new candidates were experimentally validated. Interestingly, type III secretion signals were also predicted in gram-positive bacteria and yeasts. Experimental validation showed that two candidates from yeast can indeed be secreted through Salmonella type III secretion conduit. This research provides the first line of direct evidence that secondary structure and solvent accessibility contain important features for guiding specific type III secretion. The new software based on these joint features ensures a high accuracy (general cross-validation sensitivity of ∼96% at a specificity of ∼98%) in silico identification of new type III secreted proteins, which may facilitate our understanding about the specificity of type III secretion and the evolution of type III secreted proteins.

Figure 1. Distinctive N-terminal position-specific Aac, Sse and Acc feature in T3S proteins.

Element positions are depicted on the horizontal axis. The heights of characters represent the preference or enrichment level. (A), (C) and (E): Aac, Sse and Acc preference for T3S proteins, respectively. (B), (D) and (F): Aac, Sse and Acc preference for non-T3S proteins, respectively.

Figure 2. Comparison of preference profile for Aac-Sse-Acc joint features between T3S and non-T3S sequences.

(A) and (C): Total number of non-zero distributed joint features at each position for T3S or non-T3S sequences. Full set of joint features include 120 different elements. The ratio of data size between T3S and non-T3S proteins is ∼1∶2 in (A) and 1∶1 in (C). (B) and (D): Cumulative frequency of the most enriched 10 (T3S-10 or non-T3S-10) or 20 (T3S-20 or non-T3S-20) joint features in T3S or non-T3S sequences. The ratio of data size between T3S and non-T3S proteins was about 1∶2 in (B) and 1∶1 in (D). Only the first 50 positions at the N-terminal end of T3S and non-T3S sequences were included for analysis.

Figure 3. Performance evaluation of T3SEpre.

ROC curves resulted from different T3S protein prediction software based on 5-fold cross validation using the same datasets. The parameters were optimized respectively (refer to Table 1).

Figure 4. Stableness and inter-species applicability of T3SEpre.

(A) ROC curves for T3SEpre models with different training to test data ratios. ‘Xx% vs. Yy%’ : ‘the percentage of training data versus that of testing data’. (B) Inter-species/group robustness of T3SEpre. Leave-One-Out strategy was adopted with the exception that, ‘One’ : data from ‘one species/group’. ‘Animal’ and ‘Plant’: ‘animal pathogens/symbionts’ and ‘plant pathogens/symbionts’, respectively. Sn and Sp represent sensitivity and specificity respectively. The recall rate of BPBAac and T3SEpre on each subgroup or species was indicated.

Figure 5. Performance of models with successively shortened N-terminal sequences.

(A) The first 100, 80, 60, 40 and 20 amino acid positions. (B) The first 60, 50, 40, 30, 20 and 10 amino acid positions. (C) The first 20, 15, 10 and 5 amino acid positions. (D) 1–10, 1–5, 6–10 and 11–15 amino acid positions.

Figure 6. Translocation of predicted T3S proteins.

(A) Cya translocation assays of Salmonella T3S protein candidates. Each construct was transformed into Salmonella SL1344 (Wild-type) and T3SS-deficient SL1344 strain (InvA-mutant). Duplicate was included for each test. Constructs pBADB-CyaA and pBADB-sipC-CyaA were used as negative control (NC) and positive control (PC), respectively. For each construct, Student’s t test was adopted to compare the cAMP level in the target wells co-incubated with the wild-type strain and InvA-mutant strain. Statistically significance was indicated by star (p<0.05). (B) Statistical analysis for Cya translocation assays of yeast T3S protein candidates. Constructs pBADB-CyaA and pBADB-sipC-CyaA were used as negative control (NC) and positive control (PC), respectively.