Deficiency of the Fanconi anemia E2 ubiqitin conjugase UBE2T only partially abrogates Alu-mediated recombination in a new model of homology dependent recombination

Abstract The primary function of the UBE2T ubiquitin conjugase is in the monoubiquitination of the FANCI-FANCD2 heterodimer, a central step in the Fanconi anemia (FA) pathway. Genetic inactivation of UBE2T is responsible for the phenotypes of FANCT patients; however, a FANCT patient carrying a maternal duplication and a paternal deletion in the UBE2T loci displayed normal peripheral blood counts and UBE2T protein levels in B-lymphoblast cell lines. To test whether reversion by recombination between UBE2T AluYa5 elements could have occurred in the patient's hematopoietic stem cells despite the defects in homologous recombination (HR) in FA cells, we constructed HeLa cell lines containing the UBE2T AluYa5 elements and neighboring intervening sequences flanked by fluorescent reporter genes. Introduction of a DNA double strand break in the model UBE2T locus in vivo promoted single strand annealing (SSA) between proximal Alu elements and deletion of the intervening color marker gene, recapitulating the reversion of the UBE2T duplication in the FA patient. To test whether UBE2T null cells retain HR activity, the UBE2T genes were knocked out in HeLa cells and U2OS cells. CRISPR/Cas9-mediated genetic knockout of UBE2T only partially reduced HR, demonstrating that UBE2T-independent pathways can compensate for the recombination defect in UBE2T/FANCT null cells.


INTRODUCTION
Alu elements are the most abundant short interspersed elements (SINEs) in the human genome, numbering over one million copies. These repetitive sequences are hotspots for genetic intrachromosomal or interchromosomal recombination (1). The proximity of abundant Alu elements in the genome clearly favors deletions by RAD51-independent intrachromosomal single strand annealing (SSA) (2). Alumediated recombination (AMR) events contribute to multiple forms of cancer and other genetic disorders (3)(4)(5)(6)(7)(8), and are estimated to be responsible for 0.3% of human genetic diseases (4,9). These repeated elements also drive genomic evolution; it has been estimated that more than five hundred Alu-mediated deletion events have occurred since divergence of the human and chimpanzee genomes (9). Here, we modeled an unusual somatic reversion event in a Fanconi anemia (FA) patient who had inherited a partial genomic duplication in the FANCT/UBE2T gene from his mother. In the current model system, an in vivo double strand break leads to homology-dependent recombination between two UBE2T Alu elements, mimicking a contraction of the maternal duplication to restore the WT allele.
FA is a rare recessive or dominant DNA repair disorder characterized by genome instability, developmen-tal abnormalities, bone marrow failure and cancer predisposition (10)(11)(12). Loss-of-function mutations in one Xchromosomal (FANCB) and at least twenty autosomal recessive genes (FANCA to RFWD3/FANCW) as well as missense mutations in one dominant negative FA gene (RAD51A/FANCR) result in the typical defects associated with FA (13)(14)(15). At the cellular level in FA deficient cells, genome instability in combination with erroneous repair of DNA interstrand crosslinks (ICLs) and DNA double strand breaks often results in complex genome rearrangements (CGR), translocations and gene amplification (16)(17)(18). Among the known activities of FA proteins are replisome stabilization during replication stress (17,19), the removal of DNA ICLs caused by endogenous aldehydes (20), the resolution of R-loops (21), stimulation of the alternative end joining (Alt-EJ)/microhomology-mediated end joining (MMEJ) of DNA double strand breaks (22), regulation of the spindle assembly checkpoint (23,24), and autophagic clearance of damaged mitochondria or viruses (25).
The diagnosis of FA is based on the combination of typical clinical symptoms and the characteristic hypersensitivity of cells from affected patients to the ICL reagents diepoxybutane (DEB), mitomycin C (MMC), melphalan or cisplatin, which often are used to dissect the functions of individual FA proteins (18,26). A key step in activating ICL repair is the monoubiquitination of FANCI and FANCD2 in the FANCI-FANCD2 (ID2) protein complex by the thirteen subunit FA core complex containing FANCL as the E3 ubiquitin ligase (reviewed in (26,27)). The FANCT/UBE2T gene product is not part of this protein complex but encodes the major E2 ubiquitin conjugating enzyme used by the FANCL E3 ligase to modify and activate the DNAbound ID2 dimer (28)(29)(30)(31). Monoubiquitination of FANCI and FANCD2 is necessary for their co-localization into nuclear foci. Additional roles for FANCI and FANCD2 in the stabilization of replication forks and HR have also been reported (17,30,(32)(33)(34)(35).
Machida et al. (36) and Alpi et al. (37) have shown that UBE2T is the E2 conjugating ligase in the FA pathway and that genetic deficiency in UBE2T −/− DT40 cells leads to the classical cellular phenotypes of FA, including hypersensitivity to low doses of DNA ICL agents and high frequencies of chromosomal abnormalities. Subsequently, three groups including ours independently described three FA patients with germ-line defects in the UBE2T gene, now also designated FANCT (18,(38)(39)(40). The 16-year-old FA patient (100166/1) of Italian ancestry described by us (40) was born with bilateral malformations of both thumbs and radii, microcephaly, café-au-lait spots and left kidney abnormality. He was confirmed as being affected by FA due to high levels of DEB-induced chromosomal breakage in metaphases of peripheral blood lymphocytes at birth (40). We identified the patient's primary fibroblast cells as being defective in UBE2T by overexpression of the wildtype UBE2T cDNA as a candidate FA gene (RefSeq: NM 014176.3) which entirely corrected G2/M phase arrest and also other cellular phenotypes induced by MMC. Importantly, no mutation in the UBE2T locus could be detected in the patient's germ-line DNA by Sanger sequencing or next-generation sequencing of UBE2T, as he had inherited genomic rearrangements at the two identical 311-bp AluYa5 elements present in the same orientation in introns 1 and 6 of the human UBE2T gene.
Notably, three Alu-mediated recombination events were evident at the UBE2T locus In the FANCT −/− 100166/1 proband (40). From his heterozygous father, the patient had inherited a large genomic deletion of exons 2-6, resulting in an allele without any protein-coding transcript. From his healthy mother, the patient inherited a UBE2T allele in which a duplication of exons 2-6 had occurred, resulting in a UBE2T locus with three identical AluYa5 repeats. Importantly, this maternal allele was capable of expressing a transcript for a truncated UBE2T protein that contained the complete ubiquitin binding (UB) domain of UBE2T (40). When overexpressed, this shorter protein completely restored the defects in the FA pathway in UBE2T −/− cells (40). However, western blot analysis revealed that no mutant UBE2T protein was expressed from the duplicated maternal allele in either the patient's or his mother's cells, as the mRNA from this allele was subject to nonsense mediated RNA decay (40). The third recombination event in the UBE2T locus occurred somatically in utero in a hematopoietic stem cell, as the patient's peripheral blood lymphocytes were already a mixture of normal and FA-deficient cells when analyzed by chromosomal breakage three days after birth (40). Here, it is safe to hypothesize that the normal UBE2T allele was generated by intrachromosomal SSA or unequal sister chromatid homologous recombination between the maternally duplicated Alu elements (Figure 1A), as no normal allele that could serve as a recombination donor is present in the patient's cells. Sequencing of FANCT −/− 100166/1 proband genomic DNA PCR products corroborated that the reversion had occurred at the AluYa5 repeats within the UBE2T locus (40). Subsequently, this 'corrected' hematopoietic stem cell repopulated the entire hematopoietic system with normal progeny -a phenomenon known as somatic mosaicism in FA (41)--and the patient had normal peripheral blood counts for more than 15 years and never experienced bone marrow failure.
The two main branches of homology directed recombination (HDR) are RAD51-independent single strand annealing (SSA) (42) and RAD51-dependent homologous recombination (HR) (43). To develop a model to emulate Alumediated homology directed recombination events in the UBE2T locus and also for other loci in the genome, we generated dual fluorescent reporter constructs using two independent expression cassettes for green (eGFP) and red (dTomato) fluorescent proteins with three identical AluYa5 repeats in the same orientation. An exogenous I-Sce1 site was included at either of two distinct locations in the reporter constructs to allow introduction of a single sitespecific DNA double strand break (DSB). After stable integration of one copy of the dual fluorescent reporter construct into the genome of HeLa cells, we show that expression of the I-Sce1 protein in the cells promotes DNA breakage and homology-directed AMR that mimics the reversion which had happened in the patient's hematopoietic cells.
Using the dual fluorescence system, we find that UBE2T has a limited role in HR, and demonstrate the role of HDR in Alu-mediated recombination in UBE2T using inhibitors or knockdown of HDR or nonhomologous end joining (NHEJ)-related proteins. Our results show that the dual flu-  orescent HeLa cells are also a robust tool for the systematic study of Alu-mediated recombination events and their role in inducing human disease. Combined with knockdowns of specific genes of interest, the dual fluorescence system can quantitatively report on the contribution of specific proteins to HDR and NHEJ.

Dual Fluorescence (DF) plasmid constructs
The plasmids used in this work were constructed to avoid regions of homology with the resident vector at the ectopic integration site (44) Figure S1 and confirmed by the present results, there is insufficient homology between the red and green fluorescent protein genes to enable homologous recombination.

Cell sorting (FACS)
DF2 cell lines were subjected to I-Sce1 transfection and allowed 8 days of recovery before flow cytometry to allow turnover of preexisting fluorescent proteins. The heterogeneous population (∼5 × 10 6 cells) was then prepared for cell sorting by centrifugation at 300 × g for 3 min, 4 • C. The pellet was resuspended in 1 ml FACS buffer (Hank's Balanced Salt Solution, 25 mM HEPES, 1 mM EDTA, 1% BSA and 2% FBS) and filtered through a 35 m cell strainer tube (Falcon). Cell sorting was performed at the Cincinnati Children's Hospital Medical Center (CCHMC) Research Flow Cytometry Core (RFCC) on a BD FACS Aria II flow cytometer with two 96 well tissue culture plates (single cell per well) (Corning) for each of the four flow cytometry quadrants. Single cell clones from each well were transferred to 10 cm tissue culture dishes (Corning). Once the 10 cm dishes were ∼80% confluent the cells were harvested for DNA analysis. DNA was isolated using an EZNA tissue isolation kit (Omega Bio-Tek) to serve as template in PCR amplifications.

Flow cytometry
Flow cytometry was performed on the BD Accuri C6 flow cytometer. 5 × 10 5 adherent cells were trypsinized (Gibco) and centrifuged at 300 × g for 3 min. Supernatant was removed and the cell pellet was resuspended in 300 l PBS. The cells were centrifuged for another 3 min at 300 × g. The supernatant was aspirated and the cell pellet was resuspended in 200 l PBS. The number of events counted was set to 20 000 on medium flow. Color compensation was set at correcting FL2 by subtracting 6.3% of FL1. The color compensation and quadrant determination was determined empirically to minimize spectral overlap using marker DF6 and DF7 cell lines that produce only red fluorescent dTomato protein or green fluorescent eGFP protein respectively.

CRISPR-Cas9 knockdown
CRISPR guides that target UBE2T were selected using the design website http://crispr.mit.edu (51). Guide 2 (TTTGATACCTACGAGCTCGCAGG) was chosen after single cell cloning of transduced cells and western analysis. This guide RNA is complementary to intron 1, approximately 1.3 kb downstream of the UBE2T 5 Alu element. Briefly, LentiCRISPR v2 containing the puromycin resistance-mediating pac gene was a kind gift from Feng Zhang (Addgene plasmid #52961). Cloning of the guide RNA into the LENTICRISPR v2 vector was performed as specified by the Zhang laboratory protocol (52). Infectious replication-deficient lentiviral particles in the VSV-G pseudotype were generated as previously described (40). DF3 and U2OS DR-GFP cell lines were transduced with a lentivirus construct expressing Cas9 and the UBE2T CRISPR sgRNA at an MOI of <1 and single-cell clones were selected by puromycin treatment. Single cell clones were tested by western blot for loss of UBE2T expression (40). Control cell lines underwent the same procedures but were transduced with the empty LentiCRISPR v2 vector that does not produce the UBE2T CRISPR guide.

Retroviral UBE2T expression vectors
In order to demonstrate that UBE2T deficiency of UBE2T −/− HeLa cells generated through CRISPR/Cas9 was indeed responsible for the DNA repair defects, a retroviral vector was constructed that overexpressed the wildtype UBE2T cDNA. As the UBE2T −/− HeLa cells were already resistant to neomycin, puromycin and hygromycin, the cDNA of the blasticidin resistence gene was cloned from the pHAGE.CMV.EG SLX4 vector (a kind gift of Agata Smogorzewska, Rockfeller University, NY, USA) and inserted into the retroviral pS91 UBE2T-IRES-puroR vector using NcoI-ClaI. The final retroviral vector expressed the blasticidin resistence gene after the IRES site. Infectious recombinant retroviral particles with the VSV-V phenotype were produced in 293T cells and used to transduce the HeLa cells at a multiplicity of infection of <0.2, as described previously (40).

Construction of dual fluorescence (DF) reporter constructs for the maternal UBE2T locus
The wild-type UBE2T allele in humans comprises seven exons with the translation start in exon 2, and two identical AluYa5 elements in the same orientation (www.ensembl. org). The first Alu element is located ∼180 bp 3 of the exon1/intron 1 border, and a second Alu element is ∼180 bp 3 of the exon 6/intron 6 border ( Figure 1A). The mutant maternal allele contains a duplication of exons 2-6 and the mutant paternal allele exhibits a deletion of exons 2-6. Both mutant alleles seem to have occurred by Alu-mediated recombination. We have shown previously that the paternal allele is present as a founder mutation at low frequencies in the Italian and the German populations. However, we did not detect the maternal duplication in almost 2000 individuals (40) suggesting that it is much more restricted to the maternal lineage. In the proband, spontaneous deletion of the maternally duplicated exons 2-6 by recombination between neighboring Alu elements has been postulated to account for the outgrowth of normal blood cells and the reversion of the hematopoietic phenotypes (40).
To simulate Alu-mediated recombination in the UBE2T locus, we generated clonal cell lines in which one copy of a model UBE2T locus was integrated into the single FLP recombinase target (FRT) site in HeLa/406 cells (45)(46)(47)(48)(49)53) ( Figure 1B). Previous work has shown that Alu-mediated recombination is responsible for homozygous deletions in the STK11/LKB1 locus of HeLa cells to generate aberrant LKB1 fusion transcripts (54), for MLL/KMT2A duplications in normal and AML hematopoietic cells (55), and for nonallelic homologous recombination deletions in the FANCA, B, C and D2 genes (5,56) in FA patient cells.
To model the maternal UBE2T allele containing the duplication, three AluYa5 repeats and ∼180 bp of UBE2T flanking intron 1 (IVS1) or intron 6 (IVS6) sequences were separated by complete reporter gene cassettes encoding the red fluorescent protein (RFP) dTomato driven by the human phosphoglycerate kinase (PGK) promoter with a 3 bGH polyA site, and the eGFP protein driven by a modified viral SFFV U3 promoter region followed by the polyA sequences of a SIN lentiviral LTR (57).
There is no sequence overlap between the promotors and polyA sites of the expression cassettes. In addition, the dTomato and eGFP ORFs have minimal sequence overlap and only two identical nucleotide stretches at the 5 and 3 ends of 21 and 25 bp (Supplementary Figure S1), respectively. For clarity, we have numbered the AluYa5 elements Alu 1, 2 and 3 in our reporter constructs ( Figure 1B), although these elements comprise the identical 311 bp sequence. Dual Fluorescence 1 (DF1) cells do not contain a cleavage site for I-Sce1, while in Dual Fluorescence 2 (DF2) cells a cut site for I-Sce1 preceded the eGFP gene ( Figure  1B). In DF3 cells the I-Sce1 cut site was placed upstream of the dTomato gene ( Figure 1B).

Dual fluorescence cells to monitor the repair of DSBs in the presence of Alu repeats
DF1, 2 and 3 cells initially expressed both red dTomato and green eGFP markers, and appeared yellow under UV illumination ( Figure 2A). These cells were then transfected with a plasmid that expressed the I-Sce1 endonuclease maximally at 24 h post transfection (Supplementary Figure S2) to introduce a unique DSB within the DF2 and DF3 reporter constructs. As shown in Figure 2A, DF1 cells did not change color (yellow) following transfection with the I-Sce1 expression plasmid as the DF1 construct does not harbor an I-Sce1 site. In contrast, a substantial fraction of DF2 cells showed red fluorescence associated with the loss of eGFP (RFP+, eGFP-) and only a smaller proportion of cells turned green (RFP-eGFP+) after I-Sce1 transfection, while DF3 cells preferentially turned green (RFP-, eGFP+). These data suggest that a single DNA double strand break between identical Alu repeats can efficiently induce recombination events leading to loss of the intervening color reporter gene, consistent with Alumediated intrachromosomal SSA or unequal homologydependent recombination between sister chromatids.
To confirm the extent of I-Sce1 induced deletions, we performed PCR across the ectopic integration site in each of the reporter cell lines. Lanes 1 and 2 (DF1 and DF1+I-Sce1) showed no deletions in the reporter construct in the DF1 cells upon I-Sce1 expression ( Figure 2F), as anticipated. In contrast, I-Sce1 digested DF2 and DF3 cell DNAs displayed distinct lower molecular weight major bands ( Figure  2F, lanes 4, 6) corresponding to deletion products expected of recombination between the proximal (Alu 1/2) and distal (Alu 2/3) Alu sites, consistent with the changed flow cytometry patterns of these cells after I-Sce1 expression. Specifically, recombination between the homologous Alu 1 and Alu 2 sites was found to generate a major PCR band of approximately 3.5 kb after deletion of the dTomato gene in RFP-, eGFP+ (green) DF2 cells (3.4% of total cells) and DF3 cells (30.2% of total cells).
Recombination between Alu 2 and Alu 3 generated a major PCR band of approximately 3.5 kb and deleted the eGFP gene in RFP+, eGFP-(red) DF2 cells (33.8% of total cells) and DF3 cells (1.1% of total cells). Recombination between Alu 1 and Alu 3 generated a major 1.4 kb PCR band and deleted both the dTomato and eGFP genes in RFP-, eGFP-(double negative) DF2 (9.7% of total cells) or DF3 cells (14.6% of total cells). Thus, in I-Sce1 treated DF2 or DF3 cells, more than ∼45% of the cells underwent recombination (red, green, double negative cells).
As described in detail below, we estimated the percentage of DF2 double positive cells that had undergone recombination in vivo after I-Sce1 transfection, by in vitro I-Sce1 digestion of a full-length PCR product spanning the DF2 ectopic site. Since the DF2 full-length PCR product could only have come from double positive cells, and more than ∼60% of this PCR product was resistant to in vitro I-Sce1 cleavage ( Figure 4I), these results indicate that >75% of the ectopic sites (45% recombinant cells plus 60% of double positive cells (53.1% of total) had been cut by I-Sce1 in vivo and then further processed by DNA repair. Thus, Alumediated homology-directed recombination events had occurred in >30% of DF2 cells (red) and DF3 cells (green) following introduction of the I-Sce1 DSB within the reporter constructs. However, based on the resistance of the ectopic site PCR product to in vitro I-Sce1 digestion, additional repair mechanisms (e.g. NHEJ/MMEJ) were also active in >60% of cells where both fluorescent reporters were retained.

Single colony PCR characterization of recombinants
Because standard PCR on DNA from unsorted cultures may not disclose recombination products that arise from minor percentages of the total cell population, we took a single colony PCR approach to assess the fluctuation between cells following I-Sce1 digestion. Eight days after I-Sce1 plasmid transfection, DF2 cells were FACS sorted into individual double positive, red, green, and double negative cells and clonally expanded. Genomic DNA was harvested from randomly selected clones and amplified by PCR with primers across the ectopic site. In DNA from double positive (RFP+, eGFP+) DF2 cells, PCR across the single ectopic site ( Figure 3A) generated a major band of ca. 5.7 kb in addition to bands of lesser intensity that presumably resulted from recombination during clonal expansion ( Figure  3B). The majority of red (RFP+, eGFP-) cells displayed a major band of ca. 3.3 kb, consistent with recombination by sister chromatid HR or intrachromosomal SSA between the Alu 2 and Alu 3 elements flanking the I-Sce1 site ( Figure  3C).
None of the colonies that turned green (RFP-, eGFP+) showed the ca. 3.3 kb band predicted by homology-directed repair between Alu 1 and Alu 2, but apparently had undergone a more complex series of recombinations generating smaller and larger PCR products that retained the eGFP gene and PCR primer binding sites ( Figure 3D). Surprisingly, only a minority of the double negative (RFP-, eGFP-) cell clones showed the expected recombination between Alu 1 and Alu 3 ( Figure 3E, lane 9).
The differences in the PCR profiles of the DNA from unsorted vs. sorted cells was striking. Following PCR of DNA from unsorted cells ( Figure 2F), the most abundant cell population gave the expected ca. 3.5 kb PCR product, although these may still have comprised a minor percentage of the total cells. In contrast, in the PCR of DNA from sorted and cloned cells (Figure 3), even the lesser abundance green cells could give products that were different from the expected Alu 1/2 recombinant.
The data of Figures 2 and 3 argue that the most plausible mechanism of in vivo reversion of the maternally duplicated UBE2T gene resulted from a spontaneous double strand break that led to homology-directed recombination between Alu 2 and Alu 3 elements. Additionally, it is likely that a substantial percentage of breaks did not recombine by classical RAD51-dependent homologous recombination to restore the WT allele, but instead resulted in cells carrying alternative recombinations that were selected against in the patient's hematopoietic compartment.

Role of DNA repair proteins in Alu-mediated recombination at the model UBE2T locus
To test the mechanism of recombination in the model UBE2T locus further, DF2 cells were exposed to I-Sce1 while proteins involved in nonhomologous end joining (NHEJ) or HR were chemically inhibited or knocked down by siRNA. First, we treated I-Sce1 transfected DF2 cells with caffeine ( Figure 4A), which preferentially inhibits the major DNA repair kinases ATM (ataxia telangiectasia mutated) and ATR (ataxia telangiectasia and RAD3-related) (58)(59)(60). Compared to control cells digested with I-Sce1 ( Figure 4B), caffeine significantly decreased the percentage of red, green, and double negative cells, and consequently increased the fraction of double positive cells ( Figure 4C, H). This effect on recombination efficiency was more dramatic when using the ATM specific inhibitor KU60019 ( Figure 4D, H), consistent with the stimulation of DSB end resection and downstream steps in HR and NHEJ by ATM (58)(59)(60)(61)(62).
The CtIP (C-terminal binding protein 1 interacting protein) nuclease catalyzes an early step in end resection of DNA DSBs in conjunction with the MRN complex, thereby terminating NHEJ and initiating HR (63-66).
siRNA-mediated inhibition of CtIP ( Figure 4E, H) or chemical inhibition of the RAD51 recombinase (RAD51i B02, Figure 4F, H; RAD51i RI-1, RAD51i RI-2, Supplementary Figure S3) also significantly decreased the homology-directed recombination that produced red cells upon I-Sce1 digestion. In contrast, specific inhibition of the NHEJ protein DNA-PK cs by NU7026 (67) did not significantly change the percentage of red cells ( Figure 4G, H), as expected if RFP+, eGFP-cells resulted from HR or SSA. Inhibition of NHEJ by NU7026 decreased the percentage of double negative and green cells, but increased the percentage of double positive (yellow) cells, most likely by promoting precise religation of the I-Sce1 cut (68).
To distinguish between in vivo precise religation vs. mutagenic NHEJ of the cleaved I-Sce1 site, we carried out PCR across the ectopic site in DNA from DF2 cells transfected with I-Sce1. We then digested the PCR product with I-Sce1 in vitro ( Figure 4I). DF2 cells transfected with I-Sce1 and p<0.0028 ns treated with caffeine or ATMi showed increased in vitro cutting of the ectopic site PCR product relative to cells treated with siControl ( Figure 4I, lanes 1-3). Inasmuch as these cells displayed reduced homology-directed repair (i.e. red cells, Figure 4C, D, H), we conclude that the increased in vitro cutting of the PCR product reflected an increase in precise ligation in vivo. In contrast, the ratio of in vitro cut to uncut PCR product was reduced in DNA from cells treated with siCtIP or RAD51i ( Figure 4E, F, H). Finally, inhibition of NHEJ by DNA-PK cs i resulted in an increase in the ratio of cut vs. uncut PCR product ( Figure 4I, lane 6). Since DNA-PK cs i did not significantly change HDR ( Figure 4H), we presume this increased cutting in vitro is due to enhanced precise ligation when DNA-PK cs and NHEJ end processing are inhibited in vivo.
Based on the results of PCR in the FACS sorted cell clones ( Figure 3D, E), we hypothesize that NHEJ was involved in the aberrant recombination occurring in the double negative (RFP-, eGFP-) and green (RFP-, eGFP+) I-Sce1 transfected DF2 cells. While the results of CtIP knockdown and RAD51i treatment are consistent with a reduction in HR, CtIP is also required for SSA (69), and knockdown of CtIP end resection can also be compensated by increased NHEJ (70,71), which would reduce in vitro I-Sce1 cutting. Although small-molecule RAD51 inhibitors have been shown to reduce HR (72)(73)(74)(75), the possibility of offtarget effects of these compounds remains.
Therefore, we sought an additional test of whether RAD51-dependent HR was responsible for contraction of the model UBE2T locus that produced red (RFP+, eGFP-) DF2 cells. As shown in Figure 5, we assayed the effects of knocking down BRCA2 ( Figure 5A), which interacts with RAD51 and controls its translocation into the nucleus, nucleofilament formation and assembly of RAD51 foci in response to DNA damage (76)(77)(78)(79). I-Sce1 was expressed in DF2 control cells or cells treated with BRCA2 siRNA ( Figure 5B-F). The percentage of double positive (RFP+, eGFP+) cells increased significantly with BRCA2 knockdown ( Figure 5F). By contrast, BRCA2 knockdown had no effect on the fraction of red cells produced by I-Sce1 expression. Instead, the percentage of green (RFP-, eGFP+) and double negative (RFP-, eGFP-) cells both decreased when BRCA2 was depleted. Conversely, the fraction of double positive cells was increased by compensating NHEJ/MMEJ, and mutagenic recombinations which produce green and double negative cells (Figure 3) were also enhanced. Thus, the change in the percentages of double positive, double negative and green cells indicates the biological effectiveness of the BRCA2 knockdown, while the absence of an effect on the percentage of red cells argues that BRCA2-independent SSA is responsible for the DF2 Alumediated contraction, and that downregulation of HR by BRCA2 depletion promotes alternative, mutagenic forms of recombination (80)(81)(82)(83).

Cis-acting effects of the Alu 1/IVS1 sequence on recombination at the model UBE2T locus
Digestion of DF2 cells with I-Sce1 resulted in the expected 1.4 kb Alu 1/3 recombination product when DNA from unsorted cells was analyzed ( Figure 2F), however, only about 9% of the cells fell in the double negative population ( Figure  2D) and only 1 of 9 randomly selected double negative cell clones produced the 1.4 kb PCR product predicted to arise from Alu1/3 recombination ( Figure 3D). To determine if Alu 1/IVS1 was exerting an effect in cis on homology directed recombination between Alu2/3, two additional cell lines, DF4 and DF5, were constructed that deleted Alu 1/IVS1 ( Figure 6A). The DF4 cell line differs from DF5 in that DF4 does not contain an I-Sce1 site; consequently, I-Sce1 transfection did not change the flow cytometry profile of DF4 cells ( Figure 6B). I-Sce1 digestion of DF2 or DF5 cells resulted in similar percentages of double positive cells ( Figure 6C, D). Surprisingly, I-Sce1 expression in DF5 cells ( Figure 6C) produced approximately half as many red (RFP+, eGFP-) cells and twice as many double negative (RFP-, eGFP-) recombinants as in I-Sce1 digested DF2 cells ( Figure 6D). This result indicates that Alu 1/IVS1 acts in cis to promote Alu 2/3 homology-directed recombination to yield red cells, and to suppress recombination leading to double negative cells.
The decrease in the percentage of DF5 red cells after I-Sce1 transfection is not due to reduced efficiency of I-Sce1 digestion, as the total percentage of red, green and double negative cells was closely similar between I-Sce1 digested DF2 and DF5 cells, but the ratio of green:red:double negative recombined cells changed from approximately 1:10:2.5 in DF2 cells to approximately 1:2:3 in DF5 cells. To analyze the I-Sce1 induced deletions, we performed PCR across the ectopic site in DF4 and DF5 cells. The gel of Figure 6F confirmed that I-Sce1 digested DF5 cells displayed a lower molecular weight band of ∼3.1 kb corresponding to the deletion product expected for recombination at the Alu 2 and Alu 3 sites, consistent with the flow cytometry patterns of these cells ( Figure 6C, D; RFP+, eGFP-).
We also observed that ∼22% of the I-Sce1 digested DF5 cells were double negative (RFP-, eGFP-). Taken with the PCR results on double negative cell DNA in Figure 3C, we judged that amplified PCR products smaller than 3.1 kb retained the PCR primer binding sites and were the result of excessive nuclease digestion at the I-Sce1 break or spontaneous breakage and recombination at sites other than the Alu 2 and Alu 3 repeats.

Effect of UBE2T knockout on homology directed recombination
The conclusion that a recombination event with generation of a wild-type UBE2T allele in the FANCT patient is the direct consequence of homology directed recombination (HDR) and the reason for the phenotypic reversion in a hematopoietic cell containing the inherited partial duplication of the UBE2T gene, is based on the supposition that FANCT/UBE2T deficient cells can still perform HDR. Because several FA proteins, including UBE2T, have been implicated in multiple forms of DNA repair including homology directed recombination (26,84,85), we wished to test directly the effect of UBE2T loss on HDR efficiency. Thus, lentivirus-mediated CRISPR-Cas9 was used to knockout the UBE2T genes in DF3 cells (UBE2T Δ cells) ( Figure 7A) and then single cell clones were expanded. Homology directed repair of the I-Sce1 cut in DF3 cells ( Figure 7B) is      predicted to yield green cells (RFP-, GFP +; Figure 7C). To assess the efficiency of homologous recombination we compared the levels of green cells produced when naïve DF3 cells or UBE2T Δ DF3 cells were transiently transfected with the I-Sce1 vector. Compared to naïve cells, UBE2T Δ cells showed an approximate 50% decrease in the percentage of RFP-, eGFP+ (green) cells ( Figure 7C, D). We conclude that significant homology-directed recombination could still occur at the Alu repeats in the model UBE2T allele of UBE2T Δ cells when a DSB breaks occurred between flanking Alu repeats. For comparison, the UBE2T genes were also knocked out in U2OS cells using CRISPR-Cas9 lentivirus ( Figure 7A). The UBE2T −/− knock-out U2OS cells contain the direct repeat (DR-GFP) construct ( Figure 7E), which has been used widely as a reporter for HR (86)(87)(88)(89)(90)(91). In these cells, the upstream SceGFP gene is inactive due to the presence of a stop codon within the I-Sce1 cleavage site, while a second partial GFP fragment of 812 bp (iGFP) is present 3.7 kb downstream on the same chromosome. I-Sce1 cleavage of this construct can lead to a homology-dependent gene conversion event between the iGFP and the broken SceGFP sequence, and thereby restore a functional GFP protein which can be measured by flow cytometry.
Biallelic CRISPR-Cas9 knockout of the U2OS cell UBE2T genes (UBE2T −/− ) resulted in a significant decrease in GFP + cells after I-Sce1 cleavage compared to naïve control U2OS cells or the same U2OS UBE2T −/− cells corrected with a viral UBE2T cDNA expression vector (KO + UBE2T; Figure 7F). These results showed that the UBE2T protein contributes to the efficiency of homology dependent repair, however, it is not absolutely required for HR repair to occur. The observation that UBE2T −/− cells retained more than half of the HR activity of UBE2T +/+ cells implies that redundant pathways to homology directed repair are active in these engineered FA-T cells.

DISCUSSION
The E2 ubiquitin conjugases UBE2T and UBE2W can act in the ubiquitination of FANCI/D2. However, UBE2T is only the conjugase required for activation of the FA pathway in response to MMC/ICL, and biallelic loss-offunction mutations of UBE2T lead to the FA phenotype of chromosome instability (36)(37)(38)(39)(92)(93)(94). Our experiments to model recombination at the UBE2T locus were prompted by an FA patient with inactivating duplication and deletion mutations in UBE2T and whose fibroblasts were hypersensitive to the genotoxins DEB and cisplatin, and defective in monoubiquitination of FANCD2 in response to MMC (40). Surprisingly, the UBE2T duplication was almost completely absent in genomic DNA from the patient's peripheral blood, consistent with normal thrombocyte, leukocyte and platelet counts as well as normal bone marrow cellularity.
To test whether homology dependent recombination at the UBE2T AluYa5 elements could account for the genetic reversion in this patient, we integrated single copy model UBE2T alleles into HeLa/406 cells using FLP recombinase (45)(46)(47)(48)(49)(50). The model UBE2T alleles contained red (dTomato) and green (eGFP) reporter genes separated by an I-Sce1 cleavage site so that the recombination events after DNA double strand breakage could be monitored on protein levels by flow cytometry.
I-Sce1 expression in the reporter cell lines DF2 and DF3 resulted in the cleavage of greater than 75% of the model UBE2T alleles and changes in color that were monitored by fluorescence microscopy and flow cytometry. The loss of the dTomato or eGFP ORFs and the sizes of the major PCR products from the I-Sce1 treated cells were consistent with homology-dependent Alu-mediated recombination. From these results, we conclude that an endogenous DSB likely initiated recombinational reversion of the duplicated maternal UBE2T allele in the proband and possibly partial deletion of one UBE2T allele in the father.
The structures of the recombined reporter loci indicated that HDR had occurred primarily between pairs of Alu elements. However, analysis of flow sorted clonal cell populations indicated that more complex mechanisms leading to unexpected patterns of loss of both color reporter genes were also active in a minority (<15%) of cells. Knockdown or inhibition of proteins involved in DNA damage signaling and HR (RAD51, ATM, CtIP), but not NHEJ (DNA-PKcs), inhibited recombination between Alu elements, supporting the view that classical single strand annealing was responsible for Alu-mediated recombination. In DF2 cells, inhibition shifted approximately half of the recombinations from HDR (RFP+, eGFP-) to NHEJ to yield RFP+ eGFP+ cells that generated I-Sce1 resistant PCR products.
We also observed that Alu 1/IVS1 acts in cis to promote Alu 2/3 homology directed recombination to yield red cells. In DF2 cells, Alu 1 is 2 kb upstream of Alu 2, which is well within the length of DNA that is resected in advance of homology directed repair (119,120). In flow cytometry-sorted cells (Figure 3), only one of nine double negative clones showed the expected Alu 1/3 recombination product. This result suggests that there are sequences in the construct that allow homeologous recombinations that retain (lanes 3, 6) or delete (lanes 1, 2, 4, 5, 7, 8) the PCR primer sites. One possible explanation for the positive influence of Alu 1/IVS1 on HR is that the Alu 1 sequence or its binding proteins (121)(122)(123) on the sister chromatid aid in positioning the Alu 2/IVS6 to anneal to the Alu3/IVS6 sequence during intrachromosomal SSA, akin to the phenomenon of transvection (124)(125)(126)(127).
Since HR had been implicated in the reversion of the duplication in the UBE2T −/− patient's hematopoietic cells, we tested for this activity following UBE2T knockout in DF3 and U2OS cells. An approximate 50% decrease in RFP-eGFP+ cells in UBE2T knockout DF3 cells implied that redundant homology-dependent repair pathways are operable for Alu-mediated recombination. In UBE2T −/− U2OS cells, the DR-GFP assay showed a reproducible decrease of ∼40% in HR, suggesting that UBE2T plays a role in HR in addition to activation of FANCI/D2 for ICL removal, but that one or more UBE2T-independent parallel pathways also exist for residual HR in UBE2T −/− cells. An overview of the decrease in homologous recombination detected by the DR-GFP assay due to insufficiency of several FA or HR-related proteins (Supplementary Table S1) shows a wide range of effects, with knockdown of several known FA proteins showing incomplete inhibition of HR, as in the case of UBE2T knockout.
We propose that UBE2T-dependent homology directed recombination is one mechanism of Alu-mediated reversion of the model UBE2T locus and that in UBE2T null patient cells, a UBE2T-independent residual mechanism of HR such as SSA was responsible for contraction of the partially duplicated UBE2T locus. We conclude that a spontaneous DSB in the duplicated UBE2T locus of the FANCT patient that had occurred in an hematopoietic stem cell was sufficient for UBE2T independent, Alu-mediated recombination that restored a wild-type UBE2T gene and thereby provided survival advantage for that stem cell and its progeny.
Finally, we note that the dual fluorescence flow cytometric assay for recombination in DF cells is quantitatively responsive to DNA double strand breaks and the manipulation of DNA repair pathways, indicating that this is a robust gateway system that could be adapted to the analysis of Alu-mediated homologous recombination in the human BRCA1 locus in breast tumors (6) and other diseases including FA (9,128), or more generally to probe the causes and consequences of DSBs in human cells (129). Combined with knockdowns of specific genes of interest or testing of chemical/medicinal compounds, the dual fluorescence system can quantitatively report on the contribution of specific proteins to HR and NHEJ and is therefore well suited for high throughput systematic studies of Alu-mediated recombination events.