Beyond a platform protein for the degradosome assembly: The Apoptosis-Inducing Factor as an efficient nuclease involved in chromatinolysis

Abstract The Apoptosis-Inducing Factor (AIF) is a moonlighting flavoenzyme involved in the assembly of mitochondrial respiratory complexes in healthy cells, but also able to trigger DNA cleavage and parthanatos. Upon apoptotic-stimuli, AIF redistributes from the mitochondria to the nucleus, where upon association with other proteins such as endonuclease CypA and histone H2AX, it is proposed to organize a DNA–degradosome complex. In this work, we provide evidence for the molecular assembly of this complex as well as for the cooperative effects among its protein components to degrade genomic DNA into large fragments. We have also uncovered that AIF has nuclease activity that is stimulated in the presence of either Mg2+ or Ca2+. Such activity allows AIF by itself and in cooperation with CypA to efficiently degrade genomic DNA. Finally, we have identified TopIB and DEK motifs in AIF as responsible for its nuclease activity. These new findings point, for the first time, to AIF as a nuclease able to digest nuclear dsDNA in dying cells, improving our understanding of its role in promoting apoptosis and opening paths for the development of new therapeutic strategies.

sites of the pET-28a(+) plasmid. AIF∆101 variants and CypA were produced and purified using Ni 2+ affinity chromatography as previously described (1). For H2AX overexpression, E. coli cultures were grown at 37 ºC and 180 rpm in LB supplemented with 30 mg/L of kanamycin (Sigma-Aldrich). When cultures reached OD600nm ~0.6, 1 mM IPTG was added to induce protein expression, and they were incubated for 3 additional hours. Cells were then harvested, Fractions were pooled and the imidazole content was removed in a PD-10 column as before. In all cases protein purity was assessed by SDS-PAGE and molecular exclusion chromatography. Additionally, purity of AIF∆101 was further confirmed by mass spectrometry, discarding any adventitious nuclease contamination.
Clear Native (CN) and 2D denaturing electrophoresis CN gradient electrophoresis was first run at 80 V and 4 ºC for 25-30 min, and then the amperage was set to 12 mA/gel and the voltage limited to 300 V until the sample front reached the bottom of the gel (~120 min in total). For 2D analysis, the first-dimension lane was cut out from the gel and incubated for 1 hour at 25 ºC in 1% SDS and 1% β-mercaptoethanol. The supernatant was then run in a 15% second-dimension denaturing gel (2D SDS-PAGE) at 4 ºC and 30 V for 25-30 min, and after that the voltage was set in the 80-120 V range until the dye reached the bottom of the gel. Gels were electroblotted onto Hybond-P PVDF membranes (Amersham) and then probed with specific antibodies against AIF and His-tag for CN-PAGE and SDS-PAGE respectively. Detection of immunoreactive proteins was performed using HRP-conjugated secondary antibodies (Thermo Fisher Scientific). Signals were detected using the EZ-ECL Chemiluminescence Detection kit from HRP (Pierce™), and immunoblot images were obtained in an automated WB processor Amersham TM Imager 600 (GE Healthcare).

AFM imaging measurements
Samples of protein complexes were prepared by mixing AIF101 (0.5 µM) with CypA or/and H2AX in 1:1 molar ratios for 10 min at 4 ºC under mild stirring. These mixtures, as well as the free proteins, were also mixed with 0.05 ng/µl of the pET-28a(+) plasmid -linearized with EcoRI-to visualize protein binding to dsDNA.
Final concentrations were chosen to ensure the observation of individual features and thus, to facilitate complex identification and further analysis (3). All samples were prepared in PBS pH 7.0 (Thermo Scientific). For dsDNA degradation assays, mixtures were prepared in the presence of 5 mM CaCl2 and 5 mM MgCl2 to stimulate nucleolytic activity. Samples were incubated on fresh cleaved mica pieces (Electron Microscopy Sciences) for 10 min at room temperature to achieve molecular immobilization. In the case of samples involving dsDNA, a pretreatment with 200 mM MgCl2 was applied for 2 min to favor attachment of its strands to the negative hydroxyl groups at the mica surface (4). Then, the mica pieces were washed three times with the same buffer to prevent non-desirable interactions among free biomolecules and the AFM tip, which might disturb image acquisition. At least 10 representative images from 10 different areas of 200 x 200 nm and 400 x 400 nm were analyzed for protein and protein-DNA samples, respectively. The resolution of all AFM images was at least of 512 x 512 pixels and the acquisition rate was defined at 0.5 Hz. Estimation of percentages and their associated errors were calculated as previously described for the different association states (5). Raw AFM images were analyzed using the WSxM free software (6).

ITC measurements
dsDNA samples (0.5 mM) for ITC titrations were prepared by annealing 1 mM solutions of the forward and reverse ssDNA 15-bp oligonucleotides (5′-GGT TAG TTA TGC GCG -3′ and 5′-CGC GCA TAA CTA ACC -3′; synthetized by Integrated DNA Technologies) upon incubation at 99 ºC for 1 min, followed by a 3-hour temperature gradient from 95 to 25 ºC (decreasing 1 ºC every 3 min). This length of dsDNA was designed as an appropriate model for a one to one interaction, since dsDNA was predicted to interact with AIF through no more than 12 bp (7).
where Ka app,titrating-ligand is the association constant for the ligand in the syringe at each concentration of the cell-ligand, Ka cell-ligand is the association constant for the ligand in the cell in binary complex with AIFΔ101 or in ternary complex with AIFΔ101:CypA, and [cell-ligand] is the concentration of free cell-ligand in the calorimetric cell.

Quantification of DNA degradation by densitometry
Densitometry was employed to quantitate DNA degradation from the solution nuclease assays. Agarose gels stained with ethidium bromide were scanned with a Gel Doc EQ (Bio-Rad) system and subsequently quantitated using the Quantity One (Bio-Rad) software. To calculate the percentage of non-degraded DNA, the area comprising each peak was delimited and the intensity of signal within it was measured. Background noise was calculated likewise, delimiting an area of identical dimensions in the same well but with no peak, and subsequently subtracted from the intensity of samples and control signals. The measured intensity of the control (DNA without nuclease) was considered to be 100%. To determine the relative rate of DNA degradation, the estimated amount of degraded DNA (ng) was divided by the time of degradation (seconds) and the amount of AIF (ng).  (14). MD was then performed on the best 3 poses. For MD simulations, complexes were protonated to pH 7.0 using PROPKA (15). Protein and FAD parameters were generated using respectively pdb2gmx and Gaussian 09 (16).

Analysis of the degradosome assembly by AFM
When taking images of AIFΔ101, CypA and H2AX by AFM ( Figure S1A-C), their respective average heights of 6.3 ± 0.9 nm, 3.9 ± 0.6 nm and 4.6 ± 0.7 nm agreed well with the dimensions of their corresponding PDB structures, indicating they were visualized mostly as monomers (95-98%). Besides that, no features bigger than occasional homo-dimers were observed (Table S3). When similarly evaluating mixtures containing AIFΔ101 and either CypA or H2AX, monomers corresponding well to the dimensions of isolated proteins were identified, but additional imaging features were also observed ( Figure S1D-E, Figure S2).  (Table S3).
Moreover, overall height of the degradosome assemblies remained similar to those of the monomeric isolated features, suggesting that association of the three proteins takes place at the mica plane.

Structural modelling of the degradosome assembly
A sequential routine of rigid body docking and MD simulations was followed to produce energetically optimized assembles, first for the AIF:

Binary interactions between the components of the degradosome
Interaction within the binary complexes was evaluated among all of its components -namely, AIF∆101, CypA, H2AX and dsDNA-through ITC. Binding isotherms were best fitted to a single binding site model with a Kd within the micromolar range for all binary complexes ( Figure S5 and Table S4). Control ITCs were additionally carried out with the ligand in the syringe being titrated into the buffer, ensuring that there was no significant dilution effect on the heat change ( Figure S5A-C, in red). Thermograms for the titration of AIF∆101 with its protein partners, CypA and H2AX (Figure S5A, B), demonstrated that the interaction of AIF∆101 with CypA is enthalpically guided, whereas binding of AIF to H2AX is driven by a favorable entropic contribution. This suggests that the AIF∆101:CypA binary complex rises from specific polar interactions between both proteins, which is in agreement with previous results in the literature (1, 13). On the contrary, the interaction between AIF∆101 and H2AX appears to lack such electrostatic specificity. Nevertheless, both binary complexes displayed a significant affinity between their components, with Kd values of 0.7 and 0.8 µM respectively (Table S4).  Binding of dsDNA to each protein was also separately evaluated ( Figure S5C-E). dsDNA binding was entropically driven with an unfavorable enthalpic contribution in all assayed binary complexes. The interaction was characterized in all cases by a moderate affinity (Table S4). Nonetheless, the interactions of AIF∆101 and H2AX with dsDNA were stronger in comparison with the CypA:dsDNA one (the later shows a considerable increase in Kd, Table S4). In the case of H2AX, this is in particular due to a considerably less unfavorable enthalpic contribution to the binding, despite a just milder favorable entropic contribution. Moreover, these data are consistent with the proposed mechanism of interaction between dsDNA and AIF, which is expected to take place within the clusters of positive charges present throughout the protein's surface (1, 28).
Additional assays were performed with the AIF∆101:dsDNA complex in order to estimate the buffer-independent binding enthalpy and the net number of protons  Figure S5F). Additionally, the interaction was observed to be strongly associated with proton exchange between the complex and the buffer, confirming that the affinity is pH-dependent and that about one proton is released into the bulk solution upon complex formation (at least one ionizable group is involved). Values obtained from ITC assays at 15 ºC and at 150 mM ionic strength of the abovementioned buffers. N is the calculated binding stoichiometry, usually interpreted as a fraction of binding-competent or active protein. The thermodynamic parameters were calculated using well-known relationships: Kd = (Ka) −1 , ΔG = RT . lnKd and -TΔS = ΔG -ΔH. Errors considered in the measured parameters (± 30% in Ka and ± 0.4 kcal/mol in ΔH and -TΔS) were taken larger than the standard deviation between replicates and the numerical error after the fitting analysis.

Electrophoretic-mobility-shift assays and cooperativity effects
Electrophoretic-mobility-shift assays were performed to further evaluate potential differing mechanisms in the formation of the DNA-degradosome ( Figure S7).

Dissecting the DNA-degradosome assembly by AFM
AFM images of mixtures of each isolated protein with dsDNA showed that all proteins remained mainly monomeric (Table S6), while topography profiles suggestive of dsDNA interacting with the proteins were observed. Heights corresponding to the protein monomeric features bound to dsDNA, ∼9 nm ( Figure S8A), agreed well with AIFΔ101 (6.3 ± 0.9 nm; Figure S1A) plus free dsDNA (1.5 ± 0.5 nm) ( Figure S8J). Moreover, AIFΔ101 induced the stretching and opening of dsDNA strands ( Figure S8A). Nonetheless, the binding did not appear to be sequence-specific because distinct sequences along the strands bound AIFΔ101 with similar efficacy. The AIF∆101:dsDNA interaction also appeared to display cooperativity, since several AIFΔ101 molecules were attached to a DNA strand in a clustered fashion, like "beads on a necklace", while no condensation of isolated dsDNA strands was observed. Figure S8B and C also confirmed binding of dsDNA to both CypA and H2AX monomers, according to the height profiles perpendicular to dsDNA molecules (Table S6).
When the AIF∆101:CypA and AIF∆101:H2AX systems were assayed in the presence of dsDNA, percentages of protein-protein association modes remained in similar ranges as when dsDNA was absent (Table S6). Moreover, the morphology and angle of these hetero-dimers were maintained, showing only a few ratios with a smaller angle. Binding of dsDNA to the degradosome increased the percentage of hetero-trimers by nearly two-fold (from 32 up to 52 %), with a modest decrease of hetero-dimers and monomers (32 % and 16 %, respectively; Table S6). Hardly any hetero-trimer was found attached to the mica instead of to the dsDNA, as observed by general color scales and verified by height profiles. This observation can be explained by the net charge exhibited by the proteins forming the degradosome, due to differences in terms of their isoelectric points. At working pH, the AIFΔ101 electrical charge must be minimal because its isoelectric point is neutral (29), while CypA is slightly negatively charged (pI 6.4-6.5) and H2AX is positively charged (pI 10.7). Such differences reflect different surface electrostatic potentials inducing protein recognition in specific orientations to enable the hetero-trimer formation. Figures S8E and 2E show the hetero-trimer bound to dsDNA in detail.

AIF preparations have no adventitious nuclease contamination
Purity of AIFΔ101 protein samples was confirmed by SDS-PAGE ( Figure S10A) and mass spectrometry (MS) assays. Most peptides identified by MS correspond to human AIF, and no nuclease peptide from any other sources was identified.
Additionally, the purity of the AIF protein samples was confirmed through CN-PAGE, proving that there was no nuclease contamination. To do so, a purified AIF∆101 sample was separated by high-resolution CN-PAGE in duplicate lanes.
One lane was subsequently stained with Coomassie blue (Figure S10B)

Human and mouse AIF share nuclease activity
The detected AIF nuclease activity was also confirmed with purified mouse AIF (mAIF) using both plasmid and genomic DNA substrates under similar conditions to those for the human protein ( Figure S11). These assays confirmed that mAIF also presents a certain degree of nuclease activity, although being significantly less efficient than human AIF. An assessment of different enhancing divalent ions (Ca 2+ , Mg 2+ and Mn 2+ ) demonstrated that the nuclease activity of mAIF becomes optimal with 0.1 mM of Mn 2+ (Figure S11A), differing again from human AIF.

The influence of partners and key residues on AIF nuclease activity
The protein partners' influence on AIF∆101 nuclease activity was further investigated using genomic DNA ScreenTape (Agilent) to determine the size of the remaining DNA, the concentration of intact dsDNA and the DNA Integrity Table S7). These data are fully discussed in the main text.   Values obtained from ITC assays at 15 ºC in 50 mM potassium phosphate, pH 7.4. N is the calculated stoichiometry for binding. The thermodynamic parameters were calculated using well-known relationships: Kd = (Ka) −1 , ΔG = RT.lnKd and -TΔS = ΔG -ΔH. Errors considered in the measured parameters (± 30% in Kd and ± 0.4 kcal/mol in ΔH and -TΔS) were taken larger than the standard deviation between replicates and the numerical error after the fitting analysis.

Insights into the molecular mechanism of AIF nuclease activity
Since mutations at DEK and TopIB motifs do not have a major effect on dsDNA binding, their negative effect on the direct AIF ability to degrade genomic DNA (Table SP7) has to relate with these residues being somehow implicated on the nuclease catalytic process. To better illustrate this possibility at the molecular level, relative orientations of residues at the DEK and TopIB motifs regarding dsDNA were evaluated in dsDNA:AIF docking models built by HADDOCK as above described. Top panel in Figure S14A shows the potential organization upon binding of dsDNA to the AIF DEK motif through divalent cations. In AIF this motif appears to occur in a β-hairpin plus an α-helix. DEK is a divalent cation dependent-motif that has been shown to adapt to diverse surrounding tertiary structures in different nuclease activities and pathways by being diverse and permissive in primary sequence (30). In this motif, negatively charged residues contribute to fit the positions of divalent cations that bind the target dsDNA. On its side, the third residue, Lys in AIF, usually H-bonds to a nucleophilic water and to the dsDNA, and is attributed to couple the recognition of the target DNA sequence with the cleavage reaction (31). As shown in medium and bottom panels of Figure S14A, Ala replacements of any of these three residues will surely alter either the DEK motif nuclease catalytic step or the achievement of the competent geometry for it to occur. Regarding the AIF TopIB active site, the top panel in Figure S14B shows that it contains the expected Arg, Lys, His and nucleophile Tyr residues for this motif, with the basic residues oriented to neutralize the DNA backbone (32,33). The model shows that the highly AIF conserved R449 can H-bond both the scissile DNA phosphate and the postulated nucleophile Y443, while the other charged residues might contribute as general acids to protonate the 5' leaving group of the DNA. Again, as shown in the middle and bottom panels of Figure S14B, Ala replacements at any of these residues will break the sequential events proposed for a TopIB motif nuclease activity. Removal of any of the three side-chains of the DEK motif will negatively impact the structural DEK-DNA-divalent cation organization. (B) Model for a potential organization of the AIF TopIB motif and the target DNA during TopIB nuclease activity. The top panel shows the WT model, while medium and bottom panels represent potential impact of the R449A and Y443A mutations. Removal of these side-chains will prevent achievement of catalytic competent geometry expected for TopIB nuclease activity. Residues of the DEK and TopIB motifs are highlighted in sticks with carbons respectively colored in green and magenta. Target nucleotides from the docked DNA chain to each motif are in sticks CPK colored with carbons in dark green. The DEK motif shows in pale blue spheres a potential position for the two divalent cations (placed as observed in other DEK motifs to compensate the acidic residues), while in the TopIB the top WT panel highlights as dashed lines the proposed interplay among R449, Y443 and the target DNA phosphate occurring during nuclease activity. dsDNA:AIF models are shown as produced by the HADDOCK 2.4 web server using dsDNA of 15 to 20 bp sequences as ligands and the conformation of AIF in the energetically optimized degradosome model as receptor (Figures S3 and S4).

A B
In agreement with the mutated residues at the TopIB and DEK motifs being at the protein surface, Ala substitutions do not alter protein conformation ( Figure  S14). Therefore, mutant models directly produced on the energetically optimized WT AIF molecular model are adequate to evaluate the impact of the mutations on the protein electrostatic surface potential (ESP). As shown in Figure S15, replacements to Ala produced very minor changes in the overall protein ESP, with subtle changes being only observed for some of the mutants at the position of the introduced mutation. Considering that the binding of dsDNA to AIF is nonspecific and contributed by several residues and regions on the protein surface, this agrees with none of the mutations preventing dsDNA binding and with some even favoring it (Table S8).