Strategy

The method used to construct the dynamic intracellular PPI networks was divided into three key steps: (i) data selection and preprocessing, (ii) selection of the target protein pool, and (iii) construction of the refined PPI networks for zebrafish. The overall framework for the proposed method is depicted in Figure 1.

Our strategy was to collect first all intracellular protein interactions into a candidate intracellular PPI network for zebrafish. The candidate intracellular protein interaction network was further validated and pruned since it could not exactly represent the actual intracellular protein interactions in C. albicans infection. A dynamic model was used to describe the candidate intracellular protein interactions. By using microarray data for C. albicans-infected zebrafish, the interaction abilities in the dynamic model could thus be determined. Significant PPIs based on these interaction abilities were thus identified to obtain the refined intracellular PPI. The same procedure was applied to construct the zebrafish intracellular PPI network for primary and secondary infection on the basis of time-course microarray data obtained from the corresponding experiments. Details of procedures used to construct the dynamic intracellular PPIs of zebrafish are described in the following sections.

Dataset selection and target protein pool determination

Three types of data were used in our proposed method: (i) time-course microarray profiles of gene expression of C. albicans-infected zebrafish, (ii) PPI data of H. sapiens, and (iii) orthologous gene data between zebrafish and H. sapiens. There were two sets of time points for primary and secondary C. albicans infection for obtaining microarray time-profile data for zebrafish gene expression (primary infection: 1, 2, 3, 6, and 14 dpi; secondary infection: 14.1, 14.25 14.5, 14.75, 15, 15.25, 15.5, and 16 dpi; see Figure 2). Manipulation of the animal model was approved by the Institutional Animal Care and Use Committee of National Tsing Hua University (IRB Approval No. 09808). The PPI data of H. sapiens were extracted from the Biological General Repository for Interaction Datasets (BioGRID) database (http://thebiogrid.org/) [12]. The gene ortholog data of zebrafish and humans were obtained from the InParanoid database (http://inparanoid.sbc.su.se/) [13]. For both primary and secondary infection, one-way analysis of variance (ANOVA) was applied to the microarray time-series profile of gene expression to select for differentially expressed proteins. The p-value was set at 0.01 with Student t-test for the protein pool selection. Here we treat proteins' expression as the corresponding genes' expression and view the gene pool as the protein pool. A total of 422 and 1284 proteins were thus identified as differentially expressed for primary and secondary infection, respectively. After these target protein pools for primary and secondary infection were determined, candidate PPI networks were constructed on basis of the protein pool and PPI information available from data mining. There were a total of 420 and 2312 PPI interactions included in our candidate network by integrating multiple databases (BioGRID and InParanoid7).

Construction of zebrafish intracellular PPI networks

These candidate PPI networks were obtained by including all available PPI interactions, which cannot truly represent the zebrafish intracellular PPI networks under our experimental settings. Hence, false-positive PPIs were removed on the basis of our experimental data with the help of dynamic modeling of PPI networks.

where p_k(t) represents the protein activity level for the k-th zebrafish target protein at time t. We denote M_k as the number of PPIs in zebrafish for the k-th target protein; K_k denotes the degradation effect for the k-th zebrafish target protein; p_m[t] denotes the protein activity level for the m-th zebrafish protein that can potentially interact with the k-th target protein, and b_km denotes the corresponding interaction ability between the two proteins. The basal level is denoted by q_k, and the stochastic noise due to model uncertainty and fluctuation of the microarray data is represented by n_k(t).

where parameters can be identified by solving a constrained least-squares problem. After the parameters were identified, Akaike's information criterion (AIC) was used to select significant PPI interactions [14]. The AIC includes both the estimated residual error and model complexity in one statistic. This value increases as the number of parameters increases and decreases as the variance of the residual error decreases. However, the variance of the residual error may decrease with increasing number of parameters. In other words, there exists a tradeoff between estimation accuracy and model complexity. Appropriate model order and significant interactions can be determined by ranking models by increasing AIC. By following these procedures, the intracellular PPI networks for zebrafish were thus constructed.

Inspection of the constructed zebrafish intracellular PPI networks for primary and secondary infection

We identified 57 proteins and 341 PPIs in the constructed zebrafish intracellular PPI networks during the primary infection, and 90 proteins and 385 PPIs during the secondary infection (see Additional file 1 for the complete networks). A comparison between the two constructed networks indicated that there were 20 proteins that were common to both PPI networks and 37 and 70 proteins that were specific to primary and secondary infection, respectively (see Additional file 1 for complete PPI network lists and figures).

The functions of the 37 proteins that were specific to primary infection were mapped to the following main biological processes: metabolic processes (30.1%), cellular processes (14%), cell communication (10.3%), the cell cycle (8.8%), and the immune response (6.6%). Similarly, the 70 proteins identified to be specific to secondary infection were involved in metabolic processes (26.1%), cellular processes (16.7%), cell communication (13%), developmental processes (9.1%), and transport (5.1%).

Centrality analysis of the constructed zebrafish intracellular PPI networks for primary and secondary infection

To uncover meaningful implications or insights of the constructed PPI networks, we conducted centrality analysis for both intracellular PPI networks. In particular, we considered three common network centralities, i.e., node degree and betweenness centrality.

Inspection of proteins common to PPI networks for primary and secondary infection

The TGF-β pathway is involved in the control of the primary and secondary immune response

Acvr1b (ALK4, activin receptor type 1B) was identified to be a hub protein in both primary and secondary infection (Tables 1 and 2). Activins are members of the TGF superfamily that act as local regulators of biological processes and are associated with cell growth and differentiation [19]. Correspondingly, activin is crucial to the control of innate and adaptive immune responses [10].

To regulate the immune response, TGF-β mediates its effects via SMAD proteins [9]. To clarify the role of TGF-β in the innate and adaptive immune response, we focused on the protein interactions of SMAD proteins and Acvr1b in primary and secondary infection, and compared the differences between initial and recurring infections. In primary infection, the SMAD protein Smad2 was found to interact with Acvr1b; whereas in secondary infection, Smad2, Smad3, and Smad7 were found to interacted with Acvr1b. The SMAD proteins can be divided into three major groups according to their function: receptor-regulated SMADs (R-SMAD), the common mediator SMAD (Co-SMAD), and inhibitory SMADs (I-SMAD) [9]. The R-SMADs Smad2 and Smad3, which, upon phosphorylation, interact with Co-SMAD and translocate to the nucleus [11], were identified as key proteins in primary and secondary infection. However, Smad7, an I-SMAD, was found to interact with Acvr1b during secondary, but not primary, infection (Figures 3 and 4).

Smad7 has been reported to play an essential role in the negative regulation of TGF-β signaling by interfering with the binding of TGF-β to type I receptors [25]. Furthermore, we compared the expression profile of Smad7 over time during primary and secondary infection to see if there was a significant change in between infections (Figure 5).

The time course of Smad7 expression showed a significant increasing trend of inhibitory Smad7 expression during secondary infection, suggesting that TGF-β signaling is suppressed in secondary infection relative to primary infection. This is in agreement with previous findings suggesting that Smad7 is involved in the reciprocal inhibition of TGF-β and IFN-γ [26].

In the reciprocal inhibition of TGF-β and IFN-γ, Smad7 is the key component responsible for polarizing responses toward either immunity or tolerance to infection. More specifically, although Smad7 can suppress the TGF-β signaling pathway to initiate infection tolerance, it can also promote immunity triggered through the IFN-γ signaling pathway. This is of interest, as the dual role of Smad7 in determining whether immunity or pathogen tolerance occurs is suggestive of a key mechanism that controls the immune response.

The gene expression time course in primary and secondary infection showed that Smad7 expression, which is at basal level in primary infection, increased rapidly during secondary infection (Figure 5). The difference in Smad7 expression between the primary and secondary infections may indicate that after initial low-dose infection, the zebrafish immune system was able to tolerate the invading pathogen, thereby shifting the immune response toward infection tolerance. However, in secondary infection with a lethal pathogen dose, increased Smad7 expression suggests that the immune response is triggered to defend against the invading pathogen; thus, the pathway responsible for infection tolerance is inhibited in this phase (Figure 6).

In addition, TGF-β has also been suggested to inhibit the function of inflammatory cells and immune responses [8],[27]. The regulation of TGF-β and its relationships with various immune cells are depicted in Figure 6. Under normal conditions, there is a feedback system that characterizes the relationship between the TGF-β pathway, the innate immune response, and the adaptive immune response (Figure 6). Even though immune cells can secrete cytokines that promote TGF-β signaling, TGF-β signaling can inhibit activation of these immune cells, thus acting as a feedback system. Smad7 during secondary infection was found to suppress TGF-β signaling, leading to attenuated inhibition of immune cells (Figure 6). Consequently, the increased proliferation of immune cells such as T and B cells in the adaptive immune response promotes defense against the invading pathogen.

In summary, the identification of Acvr1b in primary and secondary infection suggests that TGF-β signaling is indeed involved in the control of innate and adaptive immune responses. Furthermore, the discovery that Smad7 interacted with Acvr1b only during secondary infection suggests that TGF-β controls immune responses via a SMAD-dependent pathway. Therefore, the control mechanism can be described as a feedback system involving TGF-β signaling and the adaptive immune response (Figure 6).

The proteasome plays a role in controlling the adaptive immune response

Psmd1 and Psmd13, 26S proteasome regulatory subunits, were identified to be significant primarily during secondary infection. Proteasomal activity has been shown to be related to inflammatory and autoimmune diseases such as systemic lupus erythematosus and rheumatoid arthritis because of its role in activating an anti-apoptotic and pro-inflammatory regulator of cytokine expression [28]. Therefore, the identification of Psmd1 and Psmd13 in our constructed intracellular PPI networks for secondary infection indicates that the proteasome system plays a pivotal role in the zebrafish immune response. Furthermore, the numbers of linkages from primary to secondary infection for both Psmd1 and Psmd13 increased significantly, suggesting that the proteasome is more active during secondary infection, and is therefore more important in the adaptive immune response of zebrafish.

Regulation of apoptosis in primary and secondary infection

Many of the ten most significant hub proteins discussed above are related to the apoptotic process, as shown in Table 1. Further investigation revealed that apoptosis was activated during primary infection but was inhibited during secondary infection. In our constructed zebrafish intracellular PPI networks, we identified three proteins (Casp2, Acvr1b, Hsp90a.1) that were involved in apoptosis out of the ten hub proteins (Table 4).

For Casp2, the number of linkages increased significantly during secondary infection relative to the number during primary infection. The caspase family of proteins has a dominant role in activating apoptosis [29]. Analysis of Casp2 protein interactions revealed that it interacted with Bcl2 during secondary but not primary infection (Figures 7 and 8). Bcl2 is a member of the Bcl2 family that regulates cell death by inhibiting the apoptotic process [30]. Thus, the finding that Casp2 and Bcl2 interacted during secondary infection suggests that apoptosis is suppressed during secondary infection, in contrast to the induction of apoptosis during primary infection.

Acvr1b, a type 1B activin receptor, has been shown to be related to the apoptotic process in both primary and secondary infection. Activins are members of the TGF-β superfamily and are local regulators of biological processes that are associated with cell growth and differentiation [19]. The TGF-β pathway is also involved in inducing apoptosis and the SMAD family of molecules act as key signal transducers during this apoptotic process [31]. Smad7 protein was found to interact with Acvr1b during secondary but not primary infection (Figures 3 and 4). Smad7 is an inhibitory protein that interferes with the phosphorylation of pathway-restricted SMAD proteins such as Smad2 and SMAD3 by binding to type I receptors [11],[32]. Therefore, the interaction between Acvr1b and Smad7 supports our observation that apoptosis is inhibited during secondary C. albicans infection.

Hsp90a.1, a heat shock protein, was identified to be a key hub protein in the zebrafish intracellular PPI networks. The number of its linkages increased significantly from primary to secondary infection. Hsp70 and Hsp90 directly interact with proteins regulating the programmed cell death machinery and thus block the apoptotic process [23]. The identification of Hsp90a.1 as an important protein mainly during secondary infection in our constructed network again suggests that apoptosis is inhibited during secondary infection. Furthermore, Hsp90 stabilizes the 26S proteasome (Psmd1 and Psmd13, 26S proteasome proteins that are a part of the ten hub proteins, as listed in Table 4), and thereby enables the cell to remove unwanted or harmful proteins.

In summary, the finding that apoptotic proteins such as Casp2, Acrv1b, and Hsp90a.1 are more prominent during secondary rather than primary infection is intriguing. Increasing evidence supports that apoptosis has a crucial role in innate and adaptive immunity during infection [33]-[36]. Our results indicate that apoptosis was inhibited in secondary but not primary infection, suggesting that during infection, apoptosis can be adopted as an offensive or defensive strategy by the pathogen or zebrafish, respectively.

The identification of Ncstn implies a relationship between bacteria- and fungus-induced immune responses

Ncstn, a part of the γ-secretase protein complex, was found to play a significant role during both primary and secondary infection in our constructed PPI networks. Ncstn can generate a peptide epitope that facilitates immune recognition of intracellular mycobacteria with related components of γ-secretase through MHC II-dependent priming of T cells [18]. Such pathogen recognition mechanisms are crucial to adaptive immunity in the host. The identification of Ncstn during C. albicans infection of zebrafish suggests that Ncstn responds not only to bacterial infection but also to fungal infection.

Taken together, initial investigation of our constructed PPI networks for primary and secondary infection revealed that the immune responses activated after secondary infection are generally stronger. As shown in the in vivo experiment, zebrafish that had been infected with 1 × 10⁵ CFU C. albicans have a higher survival rate and survive longer after secondary infection with a more lethal C. albicans dose (1 × 10⁷ CFU) compared with zebrafish without prior infection (Figure 9). Identification of the aforementioned hub proteins in our constructed zebrafish intracellular PPI networks encouraged us to explore how the zebrafish immune system responds to infection and whether the response differs in primary and secondary infection.

Note that our dynamic modeling approach is not free from errors. False-positive and false-negative interactions in the initial putative PPI network can affect the accuracy of our constructed network. In order to minimize the effect of false-positive interactions, we applied AIC in the last step of network construction to eliminate the false-positive interactions based on model order selection. False-negative interactions are harder to avoid since if a PPI link is missing in the initial putative network, there is no effective method to recover the link. Therefore, we used BioGRID and InParanoid7 database, the most comprehensive PPI database available, to build our initial candidate network. We understand that PPI links may still be missing in BioGRID and InParanoid7. Such error can be improved when more PPI databases are available and can be integrated to form a comprehensive initial putative network.

Zebrafish strain and maintenance

Male, wild-type AB strain zebrafish were used in the study. Zebrafish were adults approximately 9 months old and weighed 0.33–0.37 g. Fish were maintained in 10 L tanks at 28.5°C under a 14/10 h day/night cycle.

C. albicansstrain and growth conditions

The SC5314 strain of C. albicans was used in this study. A single colony from fresh YPD agar plates (1% yeast extract, 2% peptone, 2% dextrose, 1.5% agar) was inoculated into 5 ml YPD broth and then incubated with shaking at 180 rpm at 30°C for 24 h. Cells were harvested by centrifugation, washed once with sterile PBS or Hank's balanced salt solution (HBSS), and then resuspended in sterile PBS or HBSS. Suspensions of C. albicans cells were diluted with PBS or HBSS and then injected into zebrafish.

Infection and survival assay

Zebrafish were anesthetized by immersion in water containing 0.17 g/ml tricaine (Sigma, USA) and then intraperitoneally injected with 1 × 10⁵ (primary infection) and 1 × 10⁷ (secondary infection) colony-forming units (CFU) of C. albicans at day 0 and 14, respectively, by using a 26.5 gauge syringe (Hamilton Syringe 701 N). After infection, fish were immediately transferred to the tanks to recover immediately and kept in separate 10 L tanks maintained with daily water changes. The tanks were housed in an incubator with a 14/10-h day/night cycle at 28.5°C. The fish were closely monitored and mortality was determined every hour. The phenotypes of infected zebrafish, including bleeding, ulcer/lesion, and dropsy/abnormal swelling, were shown in Figure 10.

Purification of C. albicansand zebrafish RNA

C. albicans-infected zebrafish were immersed in Trizol reagent (Invitrogen, USA) and then ground in liquid nitrogen using a small mortar and pestle. The ground sample was then disrupted by using a MagNALyser System (Roche) with glass beads (cat. no. G8772-100G, Sigma), and then shaken at 5,000 rpm for 15 s. After phase separation by addition of chloroform, the total RNA was purified using an RNeasy Mini Kit (Qiagen, Germany). The purified RNA was quantified at an OD of 260 nm wavelength by using an ND-1000 spectrophotometer (Nanodrop Technology, USA) and the RNA quality was analyzed using a Bioanalyzer 2100 (Agilent Technologies, USA) with a RNA 6000 Nanolabchip kit (Agilent Technologies).

Microarray experiments

Total RNA (1 μg) was amplified using a Quick-Amp labeling kit (Agilent Technologies), and then labeled with Cy3 (CyDye, PerkinElmer, USA) during the in vitro transcription process. For the C. albicans and zebrafish arrays, 0.625 and 1.65 μg of Cy3 cRNA, respectively, were fragmented to an average size of approximately 50 to 100 nucleotides by incubation with fragmentation buffer at 60°C for 30 min. The fragmented and labeled cRNA was then hybridized to an oligomicroarray at 60°C for 17 h. The microarrays were washed, then dried by using a nitrogen gun, and then scanned for Cy3 at 535 nm by using an Agilent microarray scanner (Agilent Technologies, USA). Scanned images were analyzed by Feature Extraction 9.5.3 software (Agilent Technologies), and image analysis and normalization software were employed to quantify the signal and background intensities for each feature. Raw data were uploaded onto the NCBI GEO Database (The array data has been uploaded onto NCBI GEO database with accession number: GSE51603).

The time points of the time-course microarray data for the primary infection were 1, 2, 3, 6, and 14 days post-infection (dpi), and those for secondary infection were 14.2, 14.6, 14.12, 14.18, 15, 15.6, 15.12, and 15.18 dpi (Figure 2). Each time point consisted of two replicates with 10 zebrafish in each replica as well as the control group with comparable conditions.

While the microarray dataset presented in this manuscript was newly generated and first reported in this study, it is also a most recent report from a series of pathogen-host interaction studies completed by our research group. Routine validation assays, including histological analysis, have already been performed and reported previously to ensure the quality and reproducibility of our data (ex.[37]-[39]). In comparison with our published reports that were focusing on the primary infection, the results are highly consistent between the current study and our past findings.