Monoubiquitination by the Fanconi Anemia core complex locks FANCI:FANCD2 on DNA in filamentous arrays

FANCI:FANCD2 monoubiquitination is a critical event for replication fork stabilization by the Fanconi anemia (FA) DNA repair pathway. It has been proposed that at stalled replication forks, monoubiquitinated-FANCD2 serves to recruit DNA repair proteins that contain ubiquitin-binding motifs. Here we have reconstituted the FA pathway in vitro to study functional consequences of FANCI:FANCD2 monoubiquitination. We report that monoubiquitination does not promote any specific exogenous protein:protein interactions, but instead stabilizes FANCI:FANCD2 heterodimers on dsDNA. This locking of FANCI:FANCD2 complex on DNA requires monoubiquitination of only the FANCD2 subunit. We further show that purified monoubiquitinated FANCI:FANCD2 forms filament-like arrays on long dsDNA using electron microscopy. Our results reveal how monoubiquitinated FANCI:FANCD2 is activated upon DNA binding and present new insights to potentially modulate monoubiquitinated FANCI:FANCD2/DNA filaments in FA cells.


Introduction
Fanconi anemia (FA) is a devastating childhood syndrome that results in bone marrow failure, leukemia and head and neck cancers (1,2). FA is caused by inheritance of one of 22 dysfunctional FA genes (FANCA-FANCW) (3). Absence of any one member of the pathway causes genome instability during DNA replication, which results in mutagenic (cancer-causing) DNA damage and hypersensitivity to chemotherapeutic (normal and cancer-killing) DNA damage (4). Central to the FA pathway is the conjugation of ubiquitin to FANCI:FANCD2 (ID2) complexes (5,6). ID2 monoubiquitination is critical to prevention of bone marrow failure in FA, but it is currently unknown how ID2-ub differs in its function to ID2. Several proteins have been proposed to specifically bind FANCI Ub or FANCD2 Ub but not the unubiquitinated proteins (7,8). For example, FAN1 nuclease was proposed to interact with FANCD2 Ub via its ubiquitin-binding domain (UBZ) (7), whereas recruitment of SLX4 endonuclease to the interstrand crosslink (ICL) site was shown to be dependent on FANCD2 ubiquitination (9). However, support for these interactions is limited to analysis of ubiquitination deficient (K>R) mutants, rather than evidence for direct ubiquitin-mediated protein interactions.
The retention of FANCD2 in chromatin foci is dependent on its monoubiquitination by a "core complex" of Fanconi anemia proteins (10). FANCI and the FA core complex are required to generate FANCD2foci that mark the location of double strand breaks, stalled replication forks and R-loops (11)(12)(13) in the nucleus, and protect nascent DNA at these sites from degradation by cellular nucleases (14). The ubiquitinated form of FANCD2, and also its ubiquitinated partner protein FANCI, become resistant to detergent and high-salt extraction from these foci (15,16), leading to speculation about the existence of a chromatin anchor or altered DNA binding specificity post-monoubiquitination (17).
A recent electron microscopy study revealed a DNA interacting domain that is required for FANCI:FANCD2 binding to DNA (18). The crystal structure of the non-ubiquitinated FANCI:FANCD2 shows that the monoubiquitination sites of FANCI:FANCD2 are buried and therefore inaccessible in the dimer interface of the complex (19), suggesting that DNA binding might be required to expose the ubiquitin binding sites. Based on biochemical analyses non-ubiquitinated FANCI and FANCD2 preferentially bind to branched DNA molecules which mimic DNA replication and repair intermediates (20)(21)(22), however how that activates monoubiquitination of FANCI:FANCD2 remains poorly understood.
DNA is a cofactor for maximal ubiquitination (17,23) Here we have reconstituted the FA pathway using recombinant FA core complex and fluorescently labelled DNA oligomer substrates. We show that once monoubiquitinated, FANCI:FANCD2 forms a tight interaction with double-stranded containing DNA. We report the successful purification of monoubiquitinated FANCI:FANCD2 complex bound to DNA using an Avi-ubiquitin construct, and show that the monoubiquitination does not promote any new protein:protein interactions with other factors in vitro. Instead, we reveal a new role of monoubiquitinated FANCI:FANCD2 in forming higher order structures and demonstrate how monoubiquitinated FANCI:FANCD2 interacts with DNA to initiate DNA repair. Our work uncovers the molecular function of the pathogenetic defect in most cases of FA.

Monoubiquitination does not promote association of FANCI:FANCD2 with a panel of proteins previously hypothesized to bind the ubiquitinated form
Mono-ubiquitinated FANCI:FANCD2 (henceforth I Ub D2 Ub ) is the active form of the complex in repair of DNA damage. Many previous studies have speculated about the existence of DNA repair proteins that specifically associate with I Ub D2 Ub . A summary of these proteins is presented in Table 1. Using recombinant ID2 or I Ub D2 Ub prepared by a novel purification method (Figure 1a) (24) we sought to directly compare the binding of this panel of ID2-associated proteins. Each of the partner proteins was expressed using reticulocyte extracts (Figure 1b), and the majority bound to the ID2 complex as predicted based on previously identified associations (Figure 1c). The strongest binding proteins in terms of fraction of protein recovered were SLX4, SMARCAD, FANCJ, PSMD4, SF3B1, TRIM25, MCM5 and BRE. Luciferase protein was used as a control 35S-labeled prey-protein, and this protein did not bind to ID2 (Figure 1c). Surprisingly, we discovered that none of the proteins showed any increased affinity for I Ub D2 Ub over ID2 (Figure 1c-d).

Monoubiquitination locks FANCI:FANCD2 on DNA
An alternative explanation for the observed increased in association between ID2 and its associated proteins after DNA damage is that I Ub D2 Ub has an increased affinity for DNA, which brings the protein into closer proximity to these partners. The majority of ID2 associated proteins are chromatin localized.
In order to explore the stability of I Ub D2 Ub on DNA, we performed in vitro monoubiquitination reactions in the presence of IR-dye700 labelled double-stranded DNA (dsDNA). As previously characterized (23), we observed DNA-dependent appearance of monoubiquitinated forms of FANCD2 and FANCI when using recombinant FA core complex components (Figure 2a-b). ID2 monoubiquitination readily lead to DNA mobility shifts using EMSA (electromobility shift assay) even at low concentrations, but this was not observed for the unmodified (apo)-ID2 complex in the absence of the enzymatically FA core complex, or when monoubiquination-defective K-to-R mutants of ID2 were used in the reaction ( Figure   2c, lanes 2-4). increased retention of an EMSA shifted band ( Figure 3). Conversely, ssDNA very weakly stimulated monoubiquitination but did not cause an EMSA shift.
Both FANCI ub and FANCD2 ub are associated with a "locked" protein:DNA complex Previous studies reported that monoubiquitination of ID2 complex may lead to dissociation of the heterodimer to its individual subunits, as measured by loss of co-immunoprecipitation of FANCI with FANCD2 (35,36). In contrast, we did not observe any Ub-mediated dissociation of ID2 in vitro. First, Western blotting of the EMSA gels confirmed that the gel shifted DNA band contains both FANCI and FANCD2 proteins ( Figure 4a). Second, FANCI Ub still co-immunoprecipitated with FANCD2 Ub at the plateau of the in vitro ubiquitination reaction ( Figure 4b).
To determine the contribution of each of FANCD2 Ub and FANCI Ub to the locking of I Ub D2 Ub complex to DNA, we used ubiquitination-deficient (KR) mutants in the ubiquitination reaction. FANCI KR :FANCD2 WT or FANCI WT :FANCD2 KR mutant results in decrease in EMSA shift, and FANCI KR :FANCD2 KR did not bind to DNA (Figure 4c). However, this retention on DNA correlated with the extent of FANCD2 monoubiquitination retained by these mutant complexes. Western blotting the EMSA gels confirmed that both FANCD2 and FANCI are found in the EMSA shifted product, although in higher amounts when both proteins are capable of being monoubiquitinated (Figure 4d).

Mutant forms of ubiquitin can still lock ID2 onto DNA
We postulated that the altered affinity for DNA induced by monoubiquitination must result from either a conformational change in the ID2 heterodimer after monoubiquitination, or participation of the conjugated ubiquitin directly in protein:DNA or protein:protein binding. To help distinguish these possibilities we utilized mutants of ubiquitin that have previously been shown to mediate the known protein:ubiquitin or protein:DNA interactions in other ubiquitinated protein interactions ( Figure 5a) (37).
Each of these Ub mutants were conjugated to ID2 by the FA core complex with similar efficiency ( Figure   5b) and their locking onto DNA was then measured. Mutations in surface patch 1 (F4A, D58A), surface patch 2 (I44A, V70A), a DNA binding residue (K11R) or a tail mutant (L73P) had no apparent effect on DNA locking (Figure 5c). This result suggests that no canonical surface or region of ubiquitin is critical for DNA locking of ID2, and instead ubiquitin conjugation to ID2 probably induces a conformational rearrangement of the heterodimer.

Purification of monoubiquitinated FANCI:FANCD2 complex bound to dsDNA reveals a filamentous architecture
In order to examine the architecture of purified recombinant I Ub D2 Ub complex in the presence of dsDNA plasmid, we utilized a recombinant Avi-tag ubiquitin construct containing a 3C protease site between the biotinylated Avi-tag and the N-terminus of ubiquitin ( Figure 6a). This tagged ubiquitin is incorporated onto FANCI:FANCD2 by the FA core complex, allowing Avidin-Sepharose purification of monoubiquitinated ID2 that is then eluted by 3C protease cleavage. We recovered monoubiquitinated FANCI:FANCD2 complex only when FANCI is monoubiquitinated, suggesting that the N-terminus of D2-attached ubiquitin may be buried within the di-ubiquitinated complex, but the N-terminus of ubiquitin attached to FANCI is accessible for streptavidin binding (Figure 6b).
Using this purified protein, we compared FANCI ub :FANCD2 ub to unmodified FANCI:FANCD2 using electron microscopy (EM). Surprisingly, we observed that FANCI ub :FANCD2 ub forms filament-like oligomers when bound to dsDNA plasmid ( Figure 6c). Such filaments were not observed in the When smaller DNA molecules were used as the substrate for ID2 binding, we either observed no filament-like structures (60bp, Figure 7a) or shorter filament-like structures (150bp, Figure 7b) compared to structures that were on average 7-8x longer than the characteristic double saxophone structure of ID2 heterodimer in the non-ubiquitinated state ( Figure 7c).
The observation that filament length correlated with the size of DNA available for ID2 binding strongly suggested that the association between heterodimer subunits in the array was DNA-mediated. To test whether the array of I ub D2 ub is also dependent upon binding to the same DNA molecule we examined the plasmid-stimulated ubiquitination reaction products after treatment with the non-specific endonuclease, Benzonase. It is apparent from EM images that addition of Benzonase breaks the long filaments formed by I ub D2 ub complex into very short or heterodimer-sized units ( Figure 7d). This finding is consistent with Benzonase cleaving exposed DNA between I ub D2 ub units, leading to destabilization of the filamentous arrays. Together our results show that, in vitro, ubiquitination of ID2 leads to a ubiquitinand DNA-stabilized filamentous structure.

Single I ub D2 ub heterodimers on short 60bp DNA have an altered architecture
Due to variability in the length and shape of filament-like I ub D2 ub structures on longer DNA molecules we have not been able to uncover the shape or subunit rearrangement of the individual units of the arrays. However, examination of I ub D2 ub purified together with short 60bp DNA allowed us to collect sufficient images of individual particles for analysis. These particles were similar in size to nonubiquitinated ID2, but it is clear from individual molecule and class average views that the I ub D2 ub complex forms a distinct architecture from that of ID2 ( Figure 8). In particular, the overall shape of individual particles and their class averages reveal a twisting that repositions the solenoid arms of one or both of the subunits bringing them into closer proximity. The conformational change induced appears to reduce the size of ID2 in an X but not Y direction, similar to that predicted in a previously model prediction that placed DNA in a channel between FANCI and FANCD2 post DNA binding (17). These images support the view that monoubiquitination induces a conformational change in the ID2 complex that locks it upon DNA.

Discussion
The protection of stalled forks by DNA repair factors is essential for proper DNA replication and the maintenance of genome stability. The primary mechanism of replication fork stabilization at ICLs, and We propose that I Ub D2 Ub does not demonstrate restricted interaction with any specific protein partner.
None-the-less, its retention in chromatin after ubiquitination would be much more likely to bring the complex into proximity of these other DNA repair factors, where it could still influence their activity.
Locking onto DNA occurs through a ubiquitin-mediated conformational change in the ID2 complex. A ubiquitin binding-domain (UBD) in FANCD2 has previously been shown necessary for the retention of the protein in the chromatin faction, and for strong binding to FANCI (46). This UBD domain sits in the FANCD2 structure opposite to where ubiquitin is likely to reside after its conjugation on to FANCI by the FA core complex, and most likely mediates the locking function and conformational rearrangement.
Relocation of a FANCD2 tower domain upon DNA binding most likely stimulates the conformational rearrangement that is then stapled in place by ubiquitin:UBD association. Other DNA binding proteins such as histone H2A show an increased association with DNA after monoubiquitination (47) and monoubiquitination also increases the DNA occupancy of transcription factors such as FOXO4 and CIITA (48). It's possible that ubiquitin to UBD mediated locking is a general mechanism of protein:DNA target stabilization.

I Ub D2 Ub locked in nucleoprotein filaments
In addition to a conformational change in ID2 induced by monoubiquitination (that has also been concurrently discovered and reported by the Paveltich and Passmore labs (49, 50)) we found that monoubiquitinated I Ub D2 Ub formed large filament-like arrays when it was purified together with plasmid DNA, but not short 60bp DNA fragments. On average, the length of plasmid-associated structures is 7-8x that of that associated with 60bp DNA. Larger or longer arrays may potentially be obscured from view because the purification strategy makes elution exponentially more difficult with increasing numbers of conjugated ubiquitin-molecules. Steps to remove "aggregates" may have also inadvertently removed larger arrays. However, as the number of potential plasmid DNA binding sites for ID2 was in large excess the concentration of ID2 used to stimulate reaction, their appears to be some purpose to creation of these filamentous arrays.
There is evidence that I Ub D2 Ub locked in nucleoprotein filaments exist in cells. Antibodies against FANCD2 have long been used as a marker of double strand breaks, stalled replication forks and Rloops because the protein forms large, intensely staining foci during S-phase that are increased after treatment with DNA damaging agents (11,12,51). We suspect that these intense foci are due to coating of DNA around damaged forks, potentially in filamentous arrays similar to those we observed by EM.
Support for the large size and extent of DNA binding reflective of filamentous arrays also comes from chromatin immunoprecipitation and sequencing (ChIP-Seq) using anti-FANCD2 (13). FANCD2, and two other damage markers MRE11 and γH2AX, showed no specific localization in a bulk population of cells, but strongly localized adjacent to a Cas9-induced site-specific DNA break. Both γH2AX and FANCD2 produced a broad peak centred at the target site kilobases (kb) to megabases (mb) in length.
In contrast, MRE11 is located within a very tight peak within ~100bp of the break. Chromatin within 1-2 kb of the DSB showed reduced occupancy by γH2AX, consistent with dechromatinization around break sites (52), but FANCD2 was present right up to the DSB. Accumulation of FANCD2 increases at the DSB early after cleavage, and accumulates more distant from the DSB progressively with time postcleavage. This is suggestive of a polymerisation of the FANCD2 signal away from the break site, as hypothesised would occur for a protein that forms a growing filament at broken DNA (13). The conserved function of FANCD2 as a histone chaperone (53, 54) may even be directly linked to displacement of nucleosomes as filamentous arrays extend into break-adjacent chromatin.
In this study, we also observed direct association of two ID2 heterodimers by co-immunopurification only after the protein becomes monoubiquitinated. This approach, if performed in cells, could be used to further delineate the mechanism and cellular factors required for the extension of I Ub D2 Ub arrays during fork protection. Of particular interest will be determining the role of BRCA1 in locking and/or array extension. BRCA1:BARD1 was initially thought to be the E3 for FANCD2 monoubiquitination, because it co-immunoprecipitates FANCD2, and FANCD2 does not form nuclear foci after damage in BRCA1deficient cells [31]. However, in various assays it was later shown that FANCD2 monoubiquitination does occur in BRCA1-deficient cells, but it is uncoupled from FANCD2 foci formation [32,33].

How would a locked ID2 filament mediate fork protection?
Filamentous structures on DNA play a genome protective role in prokaryotes: eg DAN protein forms a rigid collaborative filament that reduces accessibility during anoxia (55), while the Vibrio cholera protein ParA2 forms protective filamentous structures on DNA during segregation (56). Structural characterization has demonstrated how these filaments function and, in the case of ParA2, can be targeted therapeutically (57). The coating of ssDNA by RPA in eukaryotes, also protect DNA from the activity of nucleases, and directs the specific activity of others (58)(59)(60). We propose that a FANCI ub :D2 ub filament may have a similar stabilising role on newly synthesised dsDNA at a stalled replication fork.
This property would explain why stalled forks are prone to degradation in FA and BRCA patient cells (14,43). In particular, we hypothesise that filamentous DNA-locked I ub :D2 ub could prevent access to DNA by MRE11 and DNA2 nucleases and prevent aberrant ligation of broken DNA to other parts of the genome by non-homologous end-joining.
Second, the tight binding of FANCI ub :D2 ub to dsDNA, when localized to stalled replication forks, may also prevent the branch migration of replication forks and prevent their spontaneous or helicasemediated reversal (61). Reversed forks are the substrate for degradation by DNA2 and WRN nuclease activities, providing a hypothetical link between the activities of FANCD2-monoubiquitination and the nuclease activity of DNA2 and WRN (62,63) Third, I ub :D2 ub arrays may also locally suppress non-homologous end-joining (NHEJ) factors, and/or delineate the newly synthesized chromatin from unreplicated regions during the promotion of templated repair processes such as homologous recombination. FANCD2, FANCI, and components of the FA core complex were identified amongst relatively few other factors, in a genome-wide screen for genes that promote templated repair over NHEJ (64). Stabilization of RAD51 filaments, required for HR, is also an in vitro property of ID2 (65), suggesting I ub :D2 ub filamentous arrays may exist adjacent to or coincident with RAD51 filaments in cells, in order to provide a polarity to the homologous recombination reaction without loss or gain of genomic sequences.

Role of FANCI-monoubiquitination
Fanci and Fancd2 have common and distinct functions in mouse models of Fanconi anemia (66), while the double knockout of FANCI and FANCD2 has an unexpectedly distinct phenotype compared to single knockouts in human cells (67). But FANCI K523R expressing cells are less sensitive to DNA damage than FANCI knockout in human cells (15), so what is the role of FANCI monoubiquitination?
Previous studies demonstrated that FANCI monoubiquitination is always subsequent to FANCD2 monoubiquitination, both in cells (68) and in biochemical assays (23). FANCI also likely plays a role in recruiting the FA core complex to the substrate (69). In this study, we show that FANCImonoubiquitination is not necessary for locking of the ID2 complex onto DNA (Figure 4). However, in vivo it is likely that FANCI monoubiquitination plays a critical role in regulating deubiquitination of the ID2 complex. FANCI recruits the deubiquitinating enzyme USP1:UAF1 (70), which prevents trapping of monoubiquitinated FANCD2 at non-productive DNA damage sites, but only ID2 Ub but not I Ub D2 Ub is a substrate (23). It is also clear from our EM investigations that FANCI must play an important role in the structural integrity of I Ub D2 Ub filamentous arrays on DNA, possibly creating an asymmetry necessary for a specific polarity to array assembly.

Implications for understanding the deficiency of Fanconi anemia
Onset of progressive bone marrow failure occurs at a median age of 7 in children with FA (71). Almost all of these patients lack FANCD2 and FANCI monoubiquitination, due to mutation in either FANCD2 or FANCI or one of the 9 other FANC proteins required for their monoubiquitination (10). for nearly 20 years. The approach is likely to be formidable in drugging the FA pathway in future studies.

Protein purification
Flag-FANCI and StrepII-FANCD2 were expressed using the pFastBac1 vector (Life Technologies). For His-UBE1 was purchased from Boston Biochem.

In vitro Transcription/translation pull down of 35 S-labeled proteins
Flag-tagged FANCI:FANCD2 and monoubiquitinated FANCI:FANCD2 was prepared by incubating purified FANCI:FANCD2 or monoubiquitinated FANCI:FANCD2 on Flag beads for 2 hr followed by extensive washes in buffer A (20 mM TEA pH 8.0, 150 mM NaCl, 10% glycerol). 35 S-labeled proteins containing UBZ or other ubiquitin domains (Table 1) were generated using the TNT Quick Coupled T7 Transcription/Translation System (Promega) and 35

Electrophoretic mobility shift assay
Oligonucleotides used to create fluorescently labeled DNA were IRDYE-700-labelled X0m1 (IDTDNA) and other oligos with the sequences shown in Supplementary Table 1. Assembly of the different DNA structures was performed exactly as previously described (Supplementary Table 2

Purification of monoubiquitinated FANCI:FANCD2 complex
Di-monoubiquitinated FANCI:FANCD2 complex was purified as described (24). DNA molecules of 60bp or 150bp (dsDNA from oligonucleotides) or 2.6kb (circular plasmid DNA) were used to stimulate the reaction for different experiments, as indicated.

Mass spectrometry analysis of monoubiquitinated FANCI:FANCD2 complex
Gels containing monoubiquitinated FANCI and FANCD2 bands were excised and in-gel digested with trypsin and subjected to LC/MS analysis on ESI-FTICR mass spectrometer at Bio21. The analysis program MASCOT was used to identify ubiquitination sites on FANCI and FANCD2.

Negative stained electron microscopy
Freshly purified monoubiquitinated or non-ubiquitinated FANCI:FANCD2 complex was applied to glowdischarged, carbon/formvar grids and allowed to adsorb for 60 s. Specimen was then stained with 2% uranyl formate for 60 s. Specimen were imaged at a magnification of 73,000 x with camera (corresponding to a pixel size of 1.9 A) in Tecnai 120 kV.

Single-particle image processing
Monoubiquitinated or non-ubiquitinated FANCI:FANCD2 particles were semi-automatically picked using XMIPP3 (73). The parameters of the contrast transfer function (CTF) for negative stained data was estimated on each micrograph using CTFFIND3 (74). Finally, reference free 2D alignment and averaging were executed using XMIPP3 or CisTEM (75).