During some forms of apoptosis the mitochondrial outer membrane becomes depolarized and partially permeable to proteins. This results in a massive nonspecific release of hydrophilic proteins from the intermembrane space into the cytoplasm [1]. Among these proteins are apoptosis-inducing factor (AIF) and endonuclease G (endoG). The release of these proteins results in activation of the apoptotic caspases, degradation of nuclear DNA, and cell death [2, 3]. However, both AIF and endoG have been found to directly participate in DNA degradation in a caspase-independent way [4]. The protein AIF-homologous mitochondrion-associated inducer of death (AMID), which is probably not located in the mitochondrion, shares sequence homology with AIF and exerts similar apoptotic effects on nuclear chromatin [5]. Interestingly, endoG, AIF and AMID have all been found to influence chromatin changes during apoptosis [6].

EndoG is a mitochondrial nuclease with a molecular weight of 30 kDa. Its N-terminus contains a mitochondrial localization sequence (MLS), which is cleaved upon successful transport of the endoG precursor polypeptide across the outer mitochondrial membrane. EndoG migrates from mitochondria into the nucleus after apoptogenic stimuli [7, 8]. Addition of endoG to isolated cell nuclei resulted in cleavage of the chromatin into large fragments (~50 kbp) and subsequently into inter- and intra-nucleosomal-size fragments with periodically repeated single-stranded breaks. The first phase of endoG activity equates with the large-scale degradation of DNA during apoptosis, but the second phase would not seem to be able to generate the characteristic "laddered" fragmentation of chromatin observed in apoptotic nuclei. This may suggest that endoG normally interacts with other nucleases. Indeed, cooperation between endoG, DNase I and exonuclease III has been shown to occur only on isolated dsDNA [6]. Another proposed interaction partner for endoG was found by protein analytic in vitro methods to be flap endonuclease 1 [9], but it was not yet shown in living or fixed cells as many other possible interactions mentioned here.

AIF (also known as AIFM1 or PDCD8) is an evolutionary conserved flavoprotein. It shares a high degree of sequence homology with bacterial, plant, and fungal oxidoreductases. The human AIF is expressed as a precursor polypeptide of molecular weight ~67 kDa. This precursor contains an N-terminal MLS, which is cleaved, and the active AIF (~57 kDa) is created in the mitochondrial intermembrane space [10]. AIF is probably bound by its N-terminus to the surface of the inner mitochondrial membrane [11, 12]. The function of AIF in the mitochondrion under non-apoptotic conditions is not clear, but there is evidence that AIF may serve to sequester free radicals and that it can play important role in oxidative phosphorylation [13, 14]. However, human AIF is also able to induce apoptosis [10]. None of these effects could be inhibited by the pan-caspase inhibitor z-VAD-fmk, thus they are caspase-independent [15]. Translocation of AIF into the nucleus occurs during apoptosis [15, 16]. The C. elegans homologue of AIF, known as WAH-1, was shown to induce apoptosis and to migrate from mitochondria to the nucleus, where it interacts with a nuclease, CPS-6 (a homologue of the mammalian endoG), and together they mediate chromatin DNA degradation [17]. There is no clear evidence as to how mammalian AIF is involved in the process of chromatin degradation, but AIF can physically interact with DNA and RNA [18]. Interaction of human AIF and endoG, in analogy to what happens in C. elegans, has not yet been shown, although protein analysis in vitro results suggest its possibility [9]. However, other important proteins have been proposed to interact with AIF, namely cyclophilin A [19] and heat shock 70 kDa protein 1A (HSP70-1) [20]. Cyclophilin A is involved in transport of AIF into the nucleus and may be also involved in chromatin degradation [19, 21]. HSP70-1 binds AIF in the cytoplasm and blocks its transport into the nucleus [22]. Another protein, DNA topoisomerase II α, was found to be involved in chromatin degradation during caspase-independent apoptosis and is therefore one of our candidate proteins that may interact with AIF in nucleus [23].

AMID (also known as AIFM2 or PRG3) is a flavoprotein with amino acid sequence similarity to NADH oxidoreductases in all species, as well as to AIF (22% identity). In contrast to AIF, AMID does not contain a recognizable MLS [5]. AMID was found to associate with the outer mitochondrial membrane [5] or to be freely distributed in the cytoplasm [24]. Apoptosis induced by AMID is regulated by p53, is caspase-independent, and is not affected by overexpression of the Bcl-2 protein [5]. Artificially induced overexpression of AMID caused chromatin condensation at the nuclear periphery and formation of apoptotic bodies [24]. However, this was not confirmed by another study [5]. Overexpression of AMID also caused a loss of structurally differentiated mitochondria [5]. However, recent results did not show that AMID acts similarly to AIF and it's localization was also questioned [25, 26].

Therefore, in this work we use modern in silico methods for sequence analysis, predicting the subcellular localization of proteins, prediction of interactions, molecular modeling and docking and we combined these methods with fluorescence microscopic imaging of endoG, AIF, AMID, and other apoptotic proteins in living or fixed cells to analyze their localization and interactions.

We clearly show that sequence homology of AIF and AMID is restricted to Ndh conserved domain, that corresponds to oxidoreductase function and not to apoptotic function of AIF, which was shown to reside in large C-terminal part of AIF sequence, which is totally missing in AMID sequence (Fig. 1A) [54]. AMID sequence contains, aside from Ndh conserved domain and N-terminal part, only very small C-terminal part consisting of only several residues and thus it is highly improbable that AMID will act similarly as AIF during apoptosis. We correctly predicted that endoG and AIF would localize to the mitochondria, HSP70-1 and cyclophilin A to the cytoplasm, and DNA topoisomerase II α to the nucleus (Table 1). The cytoplasm was found as the most probable cellular location for AMID (Table 1). This result is in agreement with previous observations [25], although some of them described the distribution of AMID in the cytoplasm differently [5, 24]. No recognizable localization signal was found in the AMID amino acid sequence. However, the MLS was detected as the first 48 amino acids of the endoG sequence and as the first 61 amino acids of the AIF sequence [55]. This finding corroborates the many published observations that these proteins are translocated into the nucleus during apoptosis [7, 15, 56]. Interestingly, the predicted NLS of endoG is located within the MLS, so that once endoG enters the mitochondrion the NLS will be cleaved together with the MLS. NLS was found in sequences of DNA topoisomerase II α, which was expected, due to function of this protein. Interestingly, NLS was also found in the sequence of HSP70-1 suggesting its possible role in nucleus. HSP70-1 was found in nucleoli after the heat shock [57]. The algorithms and TMHMM 2.0 server also predicted that the sequence of endoG, cyclophilin A, HSP70-1 and DNA topoisomerase II α do not contain any transmembrane regions. However, PSORTII algorithms and TMHMM 2.0 revealed that AIF and AMID sequences contain one predicted transmembrane region thus suggesting that AIF and AMID could be membrane proteins [11, 58]. When apoptosis is induced the AIF protein is spliced probably by calpain I and looses its N-terminal part that contains the predicted transmembrane region and AIF is thus released into the intermembrane space and out of mitochondria [59]. Sequence of cyclophilin A showed no recognizable regions and is therefore interesting, that this protein is often found in nucleus, which may support the possibility that cyclophilin A interacts with AIF which contains NLS and together they translocates to nucleus [19]. The lipid anchoring is also possible for AMID, mediated by its identified putative N-myristoylation site. This site allows AMID to incorporate into various cellular membranes. Our findings thus predict that AMID is by N-terminal part incorporated into cellular membranes from cytoplasmic side.

Fluorescence imaging of living cells and microscopy of immunostained fixed cells (Fig. 2) confirmed the predicted localization of studied proteins (Table 1). AIF and endoG localized to mitochondria (Fig. 2A, E, G) and DNA topoisomerase II α to nucleus (Fig. 2J). DNA topoisomerase II α was also found to relocate during apoptosis similarly to chromatin condensation organized by AIF (Fig. 2K and 2L). This can point to possible interaction of DNA topoisomerase II α with AIF or it simply binds to condensed chromatin during apoptosis. Apart from cytoplasmic distribution, cyclophilin A surprisingly showed strong nuclear staining (Fig. 2G) maintained even during apoptosis (Fig. 2I), although its sequence does not contain any recognizable signals or motifs. Cyclophilin A can probably bind other protein that can translocate to nucleus even at non-apoptotic conditions. Immunostaining of HSP70-1 after strong heat shock showed significant nuclear localization of HSP70-1 into nucleoli (Fig. 2M) as was already observed [57]. Localization of HSP70-1 in non-apoptotic conditions was found to be cytoplasmic as predicted (Fig. 2N), but suprisingly during apoptosis HSP70-1 translocated into the cell nucleus (Fig. 2O), which even more strongly supports the possible role of this protein in apoptosis and it also corresponds to our prediction results. Cells of normal morphology expressing high levels of endoG-EYFP were identified in stably transfected clones. The signal comprised not only the mitochondrial fluorescence but also a strong confluent fluorescence in the cytoplasm and even in the nuclear chromatin (Fig. 2D). Such cells showed no morphological apoptotic changes. Thus the presence of endoG in the cytoplasm and nucleus was not sufficient to induce apoptotic chromatin degradation in the cells. This confirms our bioinformatic predictions about endoG, that the NLS of endoG can transport endoG-EYFP into the nucleus when the protein is highly overexpressed in the cytoplasm. These findings present the experimental evidence that the mere presence of endoG in the nucleus is not sufficient to initiate DNA cleavage and, although it is a nuclease, endoG apparently needs to be activated to degrade DNA in living cells [6, 60]. The fluorescence signal of AMID-tHcRed was distributed throughout the cytoplasm, although not diffuse as was suggested [24]. AMID was found to localize to unidentified regions throughout the entire cell (Fig. 2A and 2B). AMID-tHcRed localized close to the nuclear membrane, possibly with the Golgi apparatus and/or the endoplasmic reticulum; it was also associated with small vesicles in the cytoplasm and possibly with the plasma membrane and it does not translocate to the nucleus during apoptosis (Fig. 2C) [25, 26]. Cells overexpressing AMID-tHcRed were viable with normal morphology, and overexpression of AMID was not sufficient to induce apoptosis; thus, our results challenge a previous report that overexpression of AMID could induce apoptotic degradation of chromatin and the loss of structurally differentiated mitochondria [5].

We prepared 3D structural models for proteins without known 3D structure (Fig. 3). Both, these refined models and known protein structures, were used for in silico prediction of locations of binding residues for DNA and proteins. We studied especially molecules that were shown to interact with AIF or are possible candidates for such interaction. Especially we studied the interaction of AIF with B-DNA (Fig. 4A) and cyclophilin A (Fig. 4B) using molecular modeling. We also studied the hypothetical possible interaction of AIF with endoG (Fig. 4C), based on know interaction of their analogues WAH-1 and CPS-6 from Caenorhabditis elegans (Fig. 4D) [17]. The resulting values produced by FireDock server tool are suggesting that interaction of AIF with endoG is very probable due to the high negative value of global energy function, comparable to other known interactions. We visualized the predicted binding residues from servers cons-PPISP and DISPLAR in our molecular docking complexes (Fig. 5). This novel visualization clearly shows, that molecular docking results and binding residues predictions corresponds very well, thus supporting results of each other and verifying the binding residues locations and molecular binding spatial configuration of the studied proteins.

Our results show great possibilities of in silico methods and image analysis to understand the localization, interactions and functions of proteins. Our results present new data about the structure, localization, functions and interactions of proteins AMID, AIF, endonuclease G, and other apoptosis-related proteins.