53BP1 and the LINC complex promote microtubule-dependent DSB mobility and DNA repair

SUMMARY

Increased mobility of chromatin surrounding Double Strand Breaks (DSBs) has been noted in yeast and mammalian cells but how it is driven and whether it contributes to DSB repair remain unclear. Here, we use a telomere-based system to track DNA damage foci with high resolution in living cells. We find that the greater mobility of damaged chromatin requires 53BP1, SUN1/2 in the Linker of the Nucleoskeleton and Cytoskeleton (LINC) complex and dynamic microtubules. The data further demonstrate that the excursions promote non-homologous end joining of dysfunctional telomeres and implicated Nesprin-4 and kinesins in telomere fusion. 53BP1/LINC/microtubule-dependent mobility is also evident at irradiation-induced DSBs and contributes to the mis-rejoining of drug-induced DSBs in BRCA1-deficient cells showing that DSB mobility can be detrimental in cells with numerous DSBs. In contrast, under physiological conditions where cells have only one or a few lesions, DSB mobility is proposed to prevent errors in DNA repair.

INTRODUCTION

The integrity of eukaryotic genomes is perpetually threatened by the formation of DSBs, which can arise due to errors in DNA metabolism or genotoxic insults, such as chemotherapeutic agents. The repair of DSBs is a critical aspect of genome maintenance, despite the fact that non-cycling cells experience only a few DSBs per day ((Fumagalli et al., 2012); U. Herbig, pers. comm.). In G1, DSBs are repaired by nonhomologous end-joining (NHEJ) whereas replicating cells can also use a second pathway, homology-directed repair (HDR), to restore genome integrity. NHEJ and HDR are highly regulated to avoid ectopic repair, which can generate translocations, multicentric chromosomes, and other deleterious chromosome rearrangements.

The role of the DNA damage response factor 53BP1 in DSB repair and its contribution to cell cycle appropriate execution of NHEJ and HDR has been studied extensively (reviewed in (Escribano-Diaz et al., 2013; Panier and Boulton, 2014; Zimmermann and de Lange, 2014)). 53BP1 accumulates at sites of DNA damage through a dual interaction between its Tudor domain with constitutively dimethylated histone H4 (H4K20diMe) and its UDR domain with ubiquitylated histone H2A (H2AK15Ub), which marks sites of DNA damage. Many of the functions of 53BP1 are mediated by binding partners that associate with the 53BP1 N-terminus upon phosphorylation of ST/Q sites by the ATM and ATR kinases.

A critical role of 53BP1 is to limit the 5′ resection of the broken ends in a cell cycle dependent manner. Whereas inappropriate resection in G1 will impede the repair of DSBs by NHEJ, resection is needed for HDR in S/G2. Inhibition of 5′ end resection in G1 is primarily mediated by the 53BP1-bound Rif1 and Rev7/MAD2L2, but the mechanism by which resection is blocked is unknown (Chapman et al., 2013; Di Virgilio et al., 2013; Escribano-Diaz et al., 2013; Feng et al., 2013; Zimmermann et al., 2013; Boersma et al., 2015; Xu et al., 2015). In S/G2, the action of Rif1 and Rev7/MAD2L2 are counteracted by BRCA1, allowing resection and generating the 3′ overhangs required for HDR. A second 53BP1-interacting factor, PTIP, has an auxiliary role that involves end trimming by the Artemis nuclease (Munoz et al., 2007; Callen et al., 2013; Wang et al., 2014).

The contribution of 53BP1 to DSB repair pathway choice has received considerable attention in the context of the treatment of BRCA1-deficient cancers with poly(ADP-ribose) polymerase inhibitors (PARPi) (reviewed in (Banerjee et al., 2010)). PARP inhibition results in a large number of persistent single-stranded (ss) gaps that are converted into DSBs by DNA replication. In absence of BRCA1, the inefficiency of 5′ end resection allows NHEJ to dominate the repair. When many broken ends persist, NHEJ can promote mis-rejoining of broken chromatids, forming radial chromosomes and chromosome aberrations that have lethal consequences. This mis-repair of DSBs determines the synthetic lethality of PARP inhibition and HR deficiency. Removal of 53BP1 in this setting blocks the formation of mis-repaired chromosomes, in part by alleviating the inhibition of resection and hence restoring HDR (Cao et al., 2009; Bouwman et al., 2010; Bunting et al., 2010; Chapman et al., 2013; Di Virgilio et al., 2013; Zimmermann et al., 2013; Xu et al., 2015). Indeed, absence of Rif1 or MAD2L2 also minimizes the formation of mis-repaired chromosomes in PARPi-treated BRCA1-negative cells. However, 53BP1 has a greater effect than Rif1 (Zimmermann et al., 2013), suggesting a second mechanism by which 53BP1 promotes mis-rejoining.

We have used dysfunctional telomeres to investigate the second, Rif1-independent function of 53BP1. Mammalian telomeres are protected from the DNA damage response (DDR) by the six-member shelterin protein complex residing on the telomeric TTAGGG repeats (reviewed in (Palm and de Lange, 2008)). Removal of TRF2 from shelterin unleashes two pathways that normally are repressed at telomeres. Telomeres lacking TRF2 activate ATM kinase signaling, leading to Chk2 phosphorylation and the accumulation of 53BP1 at telomeres. In addition, TRF2 loss from telomeres renders them highly susceptible to Ku70/80- and DNA ligase IV(lig4)-dependent classical(c)-NHEJ.

In addition to blocking resection at dysfunctional telomeres, 53BP1 alters their mobility. After loss of TRF2, telomeres travel greater distances and roam larger subnuclear territories than functional telomeres (Dimitrova et al., 2008). This effect was also observed upon telomere deprotection with a TIN2 shRNA (Chen et al., 2013). The altered mobility of dysfunctional telomeres is strictly dependent on 53BP1 but not influenced by Rif1 (Zimmermann et al., 2013) or Rev7/MAD2L2 (Boersma et al., 2015). Given that in G1, the fusion of two telomeres involves chromosome ends that are spatially separated, we speculated that 53BP1-dependent mobility could stimulate c-NHEJ by increasing the chance that two ends become juxtaposed. Indeed, 53BP1 is required for telomere-telomere fusions (Dimitrova et al., 2008) and this dependency can not be fully explained by the ability of 53BP1 to block resection (Zimmermann et al., 2013).

In budding yeast, increased chromatin mobility occurs near an I-Sce-induced DSB and, to lesser extent, at the level of global chromatin (Dion et al., 2012; Mine-Hattab and Rothstein, 2012; Seeber et al., 2013), possibly enhancing the homology search needed for HDR (Agmon et al., 2013). Similarly, in fission yeast, DSBs associate with the LINC complex in a process that promotes HDR (Swartz et al., 2014). However, the data on the mobility of DSBs in mammalian cells has been equivocal (reviewed in (Dion and Gasser, 2013)). IR-induced DSBs show an ATM-dependent increase in mobility (Neumaier et al., 2012; Becker et al., 2014); lesions induced by α-particles or I-PpoI have been inferred to move (Aten et al., 2004; Falk et al., 2007; Gandhi et al., 2012); and directed movement occurs during telomere recombination in the context of the Alternative Lengthening of Telomeres (ALT) pathway (Cho et al., 2014). However, other findings have argued against an altered mobility of DSBs (Kruhlak et al., 2006; Soutoglou et al., 2007; Jakob et al., 2009).

Using time-lapse imaging of conditional TRF2 knockout (KO) mouse embryonic fibroblasts (MEFs) as a model system, we demonstrate here that 53BP1-dependent chromatin mobility is mediated by microtubules and the LINC complex. The LINC complex spans the inner and outer membranes (INM and ONM, respectively) of the nuclear envelope (NE) and connects components of the cytoskeleton, including microtubules, with the inside of the nucleus such that cytoskeletal forces are transferred to the nuclear content (reviewed in (Starr and Fridolfsson, 2010; Wilson and Foisner, 2010; Chang et al., 2015)). The key components of the mammalian LINC complex are the trans-membrane SUN-domain proteins, SUN1 and SUN2, which span the INM and interact with the KASH-domain Nesprin proteins in the lumen of the NE. Nesprins cross the ONM and connect to cytoplasmic filaments, including microtubules. Using microtubule poisons in combination with SUN1/2 and kinesin KO MEFs, we show that the 53BP1-dependent mobility of dysfunctional telomeres is a LINC/microtubule-dependent process that promotes NHEJ. Furthermore, we document that 53BP1/LINC/microtubule promote the mobility of IR-induced DSBs and contributes to their mis-repair in PARPi-treated BRCA1-deficient cells. These results establish a feature of the DDR that can lead to aberrant DNA repair when cells sustain large numbers of breaks. We argue that this potentially dangerous system is adaptive in the context of the physiological DDR, which has evolved to ensure correct DNA repair in cells with few DSBs.

RESULTS

A standardized method for analysis of dynamic behavior of DNA damage foci

The mechanism of 53BP1-dependent mobility was studied using immortalized TRF2^F/FCre-ER^T1 MEFs expressing an mCherry-53BP1 fusion protein that contains the Tudor, UDR, and oligomerization domains of 53BP1 (Figure 1A). This mCherry fusion accumulates at DSBs and deprotected telomeres but is neither functional nor interferes with the function of the endogenous 53BP1 (Dimitrova et al., 2008).

Figure 1. Microtubule dynamics promote mobility of dysfunctional telomeres.

(A) Schematic of the imaging approach. mCherry-BP1-2 foci at deprotected telomeres after TRF2 deletion were traced for 10 min by time-lapse microscopy.

(B) Immunoblot for TRF2 and phosphorylation of Chk2 in TRF2^F/F RsCre-ER^T1 MEFs at 55-62 h after addition of 4-OH tamoxifen (4-OHT).

(C) Images of mCherry-53BP1-2 foci with microtubule visualized with YFP-αTubulin (with γ-correction).

(D) Examples of traces of mCherry-53BP1-2 foci as described in (B) and (C) and shown in Movies S1A-D.

(E) and (F) Distribution of the cumulative distance travelled and MSD with SDs of all the mCherry-BP1-2 foci detected in the conditions as (C). Data obtained from three independent experiments with >10 cells/condition. Numbers below the data points are averages and SDs of the three median values from three independent experiments. Bars represent the median of all the foci (>1000) traced. P values are from two-tailed Mann-Whitney test. Symbols: (****) p<0.0001, (***) p<0.001, (**) p<0.01, (*) p<0.05, (ns) not significant.

(G) Percentage of cells discarded (means and SDs from three independent experiments). P values were based on unpaired t-test. Symbols as in (F).

See also Figure S1 and Table S1.

As expected, mCherry-BP1-2 formed foci at the dysfunctional telomeres generated by Cre-mediated deletion of TRF2 (Figure 1A-C), allowing detection of the dynamic behavior of mCherry-marked dysfunctional telomeres using 3D time-lapse microscopy and automated tracing in deconvolved images (Movie S1A and Figure 1D). Since MEF nuclei are flat (2-4 μm in the z direction compared to 15-20 μm in x and y), the data was analyzed in 2D-maximum intensity projected images.

Although the resulting traces can be corrected for the nuclear translocation and rotation (Dimitrova et al., 2008), large-scale nuclear deformation, such as expansion, contraction, folding, and twisting, also confounds the analysis. We therefore developed a standardized method to select nuclei that do not display overt distortions. The method is based on three parameters (Figure S1) applied to the data after correcting for the translocation and rotation of the nuclei as described previously (Dimitrova et al., 2008). First, because extensive distortion of a nucleus will usually shift the geometrical center (Figure S1A, Type I, Movie S2A), the Maximal Movement of the Geometrical Center (MMGC) of the nucleus was evaluated (Figure S1B). Second, to identify nuclei undergoing expansion or contractions (Figure S1A, Type II, Movie S2B), the Maximal Difference between the Average Distances of the foci from the geometrical center (MΔAD) was determined (Figure S1C). Third, we identified nuclei with groups of foci moving in the same direction, which could indicate the nucleus was folding, twisting, or rotating (Figure S1A Type III, Movie S2C). For this determination, the percentage of foci moving in the four different quadrants of the XY projections (Upper Right (UR); Lower Right (LR); Upper Left (UL); Lower Left (LL)) was determined (Figure S1D and S1E). Sets of foci that show concerted movement will over-populate one of these quadrants, allowing detection of nuclei with distortions. Similarly, over-population of half the space in the projections (Lateral, Vertical, Diagonal) was used to detect nuclear rotation. Using arbitrarily set thresholds for these parameters (Figure S1; see Experimental Procedures), nuclei were discarded from the analysis. In most experiments, approximately half the nuclei passed these selection criteria and were deemed to retain their shape.

Analysis of the selected nuclei showed that dysfunctional telomeres travelled a median cumulative distance of 2.5 μm in 10 min (Figure 1D and 1E, Movie S1A and Table S1), which is consistent with previous data (Dimitrova et al., 2008). The Mean Square Displacement (MSD) increased over time, with a final MSD of 0.3 μm² after 10 min (Figure 1F and Table S1). Fitting of the MSD measured for dysfunctional telomeres to MSD=A+Γt^α showed an Anomalous Diffusion Coefficient (α) of close to 1.0 (Table S1), indicating diffusive motion. The calculated diffusion coefficient was 0.37.10⁻³ μm² (Table S1) which is in the range observed by others for dysfunctional mammalian telomeres (Chen et al., 2013; Cho et al., 2014), DNA damage lesions formed after UV and IR irradiation of mammalian cells (Kruhlak et al., 2006; Falk et al., 2007; Mahen et al., 2013; Becker et al., 2014), and a locus next to an I-SceI induced DSB in yeast (Mine-Hattab and Rothstein, 2012; Dion et al., 2012).

53BP1-dependent mobility requires dynamic microtubules

We previously showed that the movement of dysfunctional telomeres is not affected by the actin drug Latrunculin A (Dimitrova et al., 2008). However, when cells were incubated with the microtubule poisons Taxol or nocodazole, which stabilize and depolymerize microtubules, respectively (Figure 1C), there was a striking reduction in the mobility of the dysfunctional telomeres and the distance travelled by the telomeres was significantly smaller (Figure 1D-1F, Movie S1A-C and Table S1). The effect of Nocodozale was completely reversed within one hour of its removal from the media, showing that the lack of dynamic behavior was not due to a permanent toxic effect of the drug (Figure 1D-1F, Movie S1D and Table S1). Both microtubule poisons also affected the extent to which the nuclei were distorted (Figure 1G and Table S1), indicating that much of the nuclear deformation observed in these fibroblasts is microtubule dependent.

SUN1 and SUN2 promote the mobility and NHEJ of dysfunctional telomeres

Since the involvement of microtubule dynamics suggested a link between the dysfunctional telomeres and the cytoplasm, we tested the role of the LINC complex in the movement of dysfunctional telomeres. To this end, we used SUN1 and SUN2 KO mice (Ding et al., 2007; Lei et al., 2009) to generate immortalized conditional TRF2^F/F SUN1^−/−SUN2^−/− MEFs. The absence of the two SUN proteins did not interfere with Chk2 phosphorylation or the formation of Telomere dysfunction Induced Foci (TIFs) containing γH2AX, 53BP1, and Rif1 after deletion of TRF2 and 53BP1 was detected by ChIP at dysfunctional telomeres in SUN1/2 DKO cells (Figure 2A-2C; Figure S2A and 2B). Nonetheless, the SUN1/2-deficient cells showed a significant reduction in the mobility of the dysfunctional telomeres (Figure 2D-2F, Figure S2C, Movie S3A and Table S1). The effect of removal of SUN1 and SUN2 was at least as strong as the effect of absence of 53BP1 monitored in parallel experiments (Figure 2D-F, Movies S3C and S3B and Table S1).

Figure 2. SUN1 and SUN2 promote mobility of dysfunctional telomeres.

(A) Immunoblots for TRF2, SUN1, SUN2, 53BP1, and phosphorylated Chk2 in the indicated MEFs at 72 h after Hit&Run Cre.

(B) TIF (Telomere dysfunction Induced Foci) assay on the MEFs described in (A). Telomeres were detected by FISH with FITC-(CCCTAA)₃ probe (green). Phosphorylated H2AX (top panel), 53BP1 (middle panel), and Rif1 (bottom panel) were detected by IF (red). DAPI: DNA (blue).

(C) Quantification of TIF response after Cre as assayed in (B). Cells with >10 TIFs were scored. Values are means and SDs of three independent experiments. P values were from an unpaired t-test (see legend to Figure 1).

(D) Examples of traces of mCherry-53BP1-2 foci at 66-72 h after Cre (see Movies S3A-C).

(E) and (F) Distribution of the cumulative distance travelled and MSDs with SDs of mCherry-BP1-2 foci in the analyzed MEFs (as (D)) in four experiments, as described in Figure 1.

See also Figure S2 and Table S1.

The percentage of nuclei that were discarded due to deformation was reduced in the absence of SUN1 and SUN2 (Figure S2C), implicating the LINC complex in the microtubule-mediated changes in nuclear shape. In contrast, 53BP1 had no effect on nuclear deformation (Figure S2C).

Importantly, in the TRF2 SUN1/2 TKO cells the diminished mobility of the dysfunctional telomeres was accompanied by a reduction in their fusion (Figure 3A and B). Metaphase spreads of cells lacking SUN1 and SUN2 showed a 2-fold decrease in the NHEJ of dysfunctional telomeres at 84 h (Figure 3B, Figure S2B). The reduction in telomere fusions was also apparent from the diminished appearance of fused telomeric restriction fragments (Figure S2D and S2E). The difference in telomere fusion frequency with and without the SUN proteins was negligible when the assay was saturated at a later time-point (108 h). In contrast, in 53BP1-deficient cells telomere fusions remained infrequent at later time points, consistent with 53BP1 promoting telomere fusions through inhibition of resection as well as SUN1/2-dependent mobility. Since the absence of SUN1 alone affected telomere fusions less than absence of both SUN1 and SUN2 (Figure S2F), we conclude that the SUN proteins have partially overlapping functions in this pathway.

Figure 3. The LINC complex promotes NHEJ of dysfunctional telomeres.

(A) Metaphases showing telomere fusions in the indicated MEFs at 84 h after Hit&Run Cre. Telomeres were detected by FISH with a FITC-(CCCTAA)₃ probe (green). DNA: DAPI (red).

(B) Distribution of telomere fusions as in (A) at 84 and 108 h after Cre. Dots represent % fusions in individual metaphases. Bars represent the median of telomere fusions in 15 metaphases for three independent experiments (45 metaphases). P values from unpaired t-test (see legend to Figure 1).

(C) In-gel assay for single-stranded telomeric DNA. Telomeric overhangs detected in situ with end-labeled ³²P-(AACCCT)₄ in MboI-digested genomic DNA from the indicated MEFs at 84 and 108 h after TRF2 deletion (top panel). Bottom: the DNA was denatured in situ and rehybridized with the same probe to determine the total telomere DNA.

(D) Quantification of relative overhang signal as detected in (C). Values represent means for four independent experiments with SDs. The ss telomeric signal was normalized to the total telomeric DNA in the same lane. For each MEF line, the normalized no Cre value of cells was set at 100 and the post-Cre values are given relative to this value. 2way ANOVA for multiple comparisons were used to perform statistical analysis. For p value symbols see legend to Figure 1.

(E) Schematic of the LINC complex and microtubules.

(F) and (G) Quantification of telomere fusions in TRF2^F/F MEFs treated with shRNAs to Nesprin-4 or Kif5B 96 h after Cre and analyzed as in (A) and (B). Bars represent the median % of telomeres fused in three independent experiments (20 metaphases each).

(H) Quantification of telomere fusions in TRF2^F/F RsCre-ER^T1 and TRF2^F/F Kif3A^F/F RsCre-ER^T1 MEFs 72 and 90 h after 4-OHT, as in (A) and (B).

See also Figure S2 and Figure S3.

We verified that the deficiency in telomere fusion in the SUN1/2 KO was not due to increased resection using a quantitative assay for the amount of ssTTAGGG repeats after deletion of TRF2 from SUN1/2 deficient cells. In Lig4^−/− MEFs (TRF2/SUN1/SUN2/Lig4 quadruple KO), which are a good system for detection of resection because the telomeres remain free, there was no great increase in the overhang signal after TRF2 deletion (Figure 3C and 3D), indicating that resection remained repressed. Parallel deletion of TRF2 from 53BP1^−/− Lig4^−/− cells showed the substantial increase in overhang signal expected from the role of 53BP1/Rif1 in repression of resection (Lottersberger et al., 2013). These data, together with the normal localization of Rif1 at dysfunctional telomeres in SUN1/2 DKO cells (Figure 2B and 2C), supports the idea that SUN1 and SUN2 are dispensable for the protection of DSBs from resection and act independently from Rif1. We propose therefore that SUN1 and SUN2 promote the c-NHEJ of telomeres by increasing their dynamic behavior.

Nesprin-4 and kinesins contribute to NHEJ of dysfunctional telomeres

Since the SUN proteins are connected to the cytoskeleton through nesprins (reviewed in (Starr and Fridolfsson, 2010); see Figure 3E) and SUN1/2-deficient cells lack Nesprin-1, -2, -3, and -4 at the NE (Crisp et al., 2006; Padmakumar et al., 2005; Ketema et al., 2007; Lei et al., 2009; Roux et al., 2009), we tested shRNAs to Nesprins for an effect on NHEJ of dysfunctional telomeres. Two shRNAs targeting Nesprin-4 lowered the frequency of telomere fusions without affecting cell proliferation or the DDR (Figure 3F, Figure S3A and S3B).

As Nesprin-4 is known to interact with the plus-end directed microtubule motor kinesin-1 (Figure 3E), we tested shRNAs to the Kif5B subunit of kinesin-1 for an effect on the NHEJ of dysfunctional telomeres. Two shRNAs to Kif5B lowered the frequency of telomere fusions at an early time point without affecting the proliferation or the DDR upon telomere deprotection (Figure 3G, Figure S3C and S3D). In addition, two shRNAs to the kinesin-2 subunit Kif3A resulted in a reduced frequency of telomere fusions (Figure S4E-G). Since kinesin-2 had not previously been shown to cooperate with Nesprin-4, we generated TRF2^F/FKif3A^F/F MEFs to further verify the shRNA data. Consistent with the shRNA results, MEFs lacking Kif3A showed a significant reduction in the efficiency of telomere fusions after TRF2 deletion (Figure 3G, Figure S3H and S3I). These data suggest that 53BP1-mediated mobility of dysfunctional telomeres likely involves redundant action by the microtubule motors kinesin-1 and kinesin-2, as well as Nesprin-4 and possibly other Nesprins.

Phosphorylation sites in 53BP1 required for telomere mobility

As a version of 53BP1 lacking its N-terminal S/TQ sites (53BP1-28A) fails to induce chromatin mobility (Lottersberger et al., 2013), we determined which S/TQ sites are involved in this process. We generated a collection of S or T to A mutations at the S/TQ positions in a C-terminally truncated version of 53BP1 that lacks the BRCT domain (53BP1DB; Figure 4A; (Bothmer et al., 2011)) and behaves like wild type 53BP1 in the context studied here (Lottersberger et al., 2013; Zimmermann et al., 2013). Through the analysis of the mutants, we identified one mutant, referred to as 53BP1ΔMOB, which appeared to be a separation-of-function mutant specifically deficient in the ability of 53BP1 to promote mobility but proficient in blocking resection (Figure 4A). 53BP1ΔMOB recruited Rif1 to sites of DNA damage and was able to interact with PTIP, which was expected since the region of mutated S/TQ sites falls outside the previously mapped Rif1 and PTIP interacting regions (Figure 4A and Figure S4A-C) (Munoz et al., 2007; Escribano-Diaz et al., 2013). Consistent with its binding to Rif1, the 53BP1ΔMOB mutant was proficient in repressing hyper-resection at telomeres after TRF2 deletion in TRF2^F/F 53BP1^−/− Lig4^−/− cells (Figure S4D and S4E).

Figure 4. The mobility domain of 53BP1, but not PTIP, is required for mobility of dysfunctional telomeres.

(A) Schematic of 53BP1, S/TQ site mutations, and their phenotypes.

(B) Quantification of telomere fusions in the indicated MEFs complemented with the indicated 53BP1 alleles 96 h after TRF2 deletion with Hit&Run Cre (as in Figure 3). Data from >70 metaphases analyzed in four independent experiments. For each experiment, the median fusion frequency for 53BP1DB was set to 100 and all other values were normalized to this frequency.

(C) MSDs with SDs of mCherry-BP1-2 foci detected in the TRF2-deleted 53BP1^−/− MEFs expressing the indicated 53BP1 alleles. Data from three independent experiments.

(D) Examples of traces of mCherry-53BP1-2 foci at 66-72 h after Cre in the indicated MEFs (see Movies S4A-C).

(E) and (F) Distribution of the cumulative distance travelled and MSDs with SEMs of mCherry-BP1-2 foci in the indicated MEFs (as in Figure 1). Bars represent medians of the cumulative distance travelled >500 foci in the two experiments and numbers indicate the averages and SEMs of the two median values obtained in the two independent experiments.

(G) Quantification of telomere fusions in the indicated MEFs at 84 and 108 h after Cre (as in Figure 3).

See also Figure S4, Figure S5 and Table S1.

Despite the normal interactions with Rif1 and PTIP, the ability of 53BP1ΔMOB to promote telomere fusions upon complementation of 53BP1 deficiency was significantly reduced (Figure 4B). However, 53BP1ΔMOB promoted telomere fusion similar to 53BP1DB in SUN1^−/− SUN2^−/− 53BP1^−/− cells (Figure 4B and Figure S4F), suggesting that the 53BP1ΔMOB is only deficient in a function that requires SUN1/2. Time-lapse imaging showed that 53BP1ΔMOB is completely defective in promoting the increased mobility of dysfunctional telomeres resulting in dynamics that are indistinguishable from cells transduced with the empty vector or the 53BP1Δ28A mutant (Figure 4C, Figure S4G and Table S1). In contrast, 53BP1ΔPTIP showed no defect in promoting mobility of dysfunctional telomeres (Figure 4C, Figure S4G and Table S1). Thus, the ability of 53BP1 to promote mobility of dysfunctional telomeres likely involves an interaction that depends on phosphorylation of one or more of the ST/Q sites in the MOB domain. The identity of the MOB domain interacting partner is unknown. It is not excluded that this domain interacts with SUN1 and SUN2 but this interaction was not detected by in mass spectrometry (Di Virgilio et al., 2013) and ChIP failed to reveal SUN1 and SUN2 at dysfunctional telomeres (Figure S2A).

PTIP is not required for 53BP1-dependent mobility

To determine whether PTIP contributes to the 53BP1-dependent mobility, TRF2 and PTIP co-deletion in SV40LT immortalized TRF2^F/F PTIP^F/F MEFs was analyzed. Absence of PTIP did not affect cell proliferation or the DDR at the dysfunctional telomeres (Figure S5A-D). In the PTIP-deficient setting, the distances traveled and MSD of the dysfunctional telomeres was equal to that of PTIP containing control cells (Figure 4D-F and Figure S5E, Movie S4A-C and Table 1S). Moreover, the analysis of the telomere overhangs showed that PTIP deficiency did not affect the resection at dysfunctional telomeres (Figure S5F and S5G), supporting the previous conclusion that 53BP1-dependent protection from resection is primarily dependent on Rif1 (Zimmermann and de Lange, 2014). Nonetheless, as previously shown (Callen et al., 2013), telomere fusions appeared slightly delayed when PTIP was deleted (Figure 4G, Figure S5F and Figure S5G). Consistent with these results, the 53BP1ΔPTIP mutant displayed a mild defect in promoting telomere fusions but appeared unaffected with regard to protection from resection and the induction of mobility (Figure 4B and Figure S4D and S4E).

53BP1/LINC/microtubule dependent mobility of IR-induced DSBs

Despite their resemblance to DSBs, dysfunctional telomeres could be argued to be different from chromosome-internal DNA breaks. We therefore tested whether genome-wide DSBs are subject to the 53BP1/LINC/microtubule dependent changes in dynamics. To this end, we analyzed the mobility of the mCherry-BP1-2 foci after induction of ~100 DSBs with 2.75 Gy IR (Rothkamm and Lobrich, 2003) in wild type, SUN1^−/−SUN2^−/−, and 53BP1^−/− MEFs. As expected, the Chk2 phosphorylation and formation of γ-H2AX foci were not affected by the genotype of the cells (Figure 5A-C). The IR-induced mCherry-53BP1-2 foci showed a cumulative distance travelled and an MSD comparable to the MSD of dysfunctional telomeres. This dynamic behavior was strongly diminished in absence of 53BP1 or the SUN proteins and upon treatment with Taxol (Figure 5D-G, Movie S5A-D and Table S1). Therefore, we conclude that the 53BP1/LINC/microtubules pathway promotes the mobility of chromosome-internal DSBs as it does at dysfunctional telomeres.

Figure 5. 53BP/LINC/microtubule promoted mobility of IR-induced DSBs.

(A) Immunoblot for phosphorylation of Chk2 (as in Figure 2A) in the indicated MEFs at 1 h after 2.75 Gy IR.

(B) IF for γH2AX (green) and 53BP1 (red) for cells treated as in (A). DAPI: DNA (blue).

(C) Quantification of IR-induced γ-H2AX and 53BP1 foci as assayed in (B).

(D) Examples of 10 min traces of mCherry-53BP1-2 foci at 1 h after IR of the cells described in (A) with or without 20 μM Taxol (see Movies S5A-D).

(E-G) Percentage of cells discarded, distribution of the cumulative distance travelled, and MSDs with SDs of mCherry-BP1-2 foci detected as (D) and (E) (as in Figure 1). Data from three independent experiments.

See related Figure S6 and Table S1.

Undamaged chromatin is minimally affected by DSBs

We next asked whether the presence of mobile DSBs changes the dynamics of the global chromatin. To address this question, we monitored the mobility of fully functional telomeres, marked with eGFP-TRF1 in cells with and without IR-induced DSBs. The IR was delivered at 2.75 Gy, which induces ~1 DSB/60 Mb (~100 DSBs per cell, see above). Since the 80 telomeres of the mouse genome represent ~4 Mb (~0.1% of the genome), telomeres are not expected to contain DSBs after 2.75 Gy. Nonetheless, the eGFP-marked telomeres showed a very slight but statistically significant increase in the cumulative distance traveled (Figure S6). Moreover their MSD and diffusion coefficient were slightly increased, although much less than when the telomeres were dysfunctional (Figure S6, Movie S6A-D and Table 1S). These results indicate that while the chromatin dynamics primarily affects sites of DNA damage, there is also a minor increase in the mobility of undamaged chromatin, consistent a previous report (Zidovska et al., 2013).

When the eGFP-TRF1 marker was used to detect nuclear deformations, the incidence of distorted nuclei was not affected by deletion of TRF2 (Figure S6B and Table S1), indicating that microtubule dynamics distort nuclei regardless of the presence of DNA damage.

Chromatin mobility promotes DSB mis-repair in BRCA1-deficient cells

We considered that for genome-wide DSBs, the increased mobility of the chromatin could promote the joining of unrepaired DNA ends that are at a distance. One setting in which this process may be relevant is the formation of radial chromosomes in PARPi-treated BRCA1-deficient cells. Radial formation involves the joining of a DNA end from one chromosome with a break in another chromosome, which may be at a distance and therefore would require spatial exploration for joining. We therefore tested whether the 53BP1-dependent mobility contributes to the mis-rejoining when many S phase DSBs are induced with PARPi and HDR is impaired. Experiments with cells containing fluorescently labeled Geminin to reveal their cell cycle stage showed that IR-induced DSBs become mobile in S/G2 as well as in G1 (Figure S7A-D).

As previously shown, when BRCA1 shRNA treated cells were incubated with Olaparib, a significant number of mis-rejoined chromosomes was formed and this phenotype was repressed by deletion of 53BP1 (Figure 6A-C). Importantly, SUN1^−/− SUN2^−/− MEFs also diminished the formation of aberrantly repaired chromosomes (Figure 6A-C) and the mis-rejoining events were strongly reduced by Taxol (Figure 6D). The effect of Taxol was not due to diminished PARP inhibition, since PARPi/Taxol treated cells showed no parsylation in response to H₂O₂ (Figure S7E). Taxol did not further reduce either the mobility or the chromosome mis-rejoining events in absence of SUN1 and SUN2 (Figure 6E, Figure S7F-H and Table S1), supporting the view that the SUN proteins and microtubules act in the same pathway to promote chromatin mobility and aberrant DNA repair. Importantly, SUN1/2 deficiency also diminished the lethality of PARPi treatment in BRCA1 deficient cells (Figure 6F and Figure S7I). As expected, the absence of 53BP1 rescued the lethality of PARPi treatment to a greater extent, consistent with the multiple mechanisms by which 53BP1 affects DSB repair (Figure 6G).

Figure 6. SUN1/2 and dynamic microtubules promote radial formation.

(A) Immunoblots for BRCA1 and γ-tubulin in the indicated MEFs (as in Figure 2A) at 144 h after infection with BRCA1 shRNA or empty vector. Olaparib was added 16 h before analysis.

(B) Representative mis-rejoined chromosomes (arrowheads). DNA stained with DAPI.

(C) Quantification of mis-rejoined chromosomes in the indicated MEFs (as in (A)), analyzed as in (B). Each dot represents a metaphase. Bars represent the median of mis-rejoined chromosomes in three independent experiments (10 metaphases each). P values as in Figure 3B.

(D) Quantification of mis-rejoined chromosomes in the indicated MEFs 18 h with or without Taxol as in (C).

(E) Quantification of mis-rejoined chromosomes in each metaphase in the indicated MEFs with or without Taxol as described in (C) and (D). All cells used in (A-F) are TRF2^F/F.

(F) Quantification of colony formation in the indicated cells infected with BRCA1 shRNA and treated with or without Taxol for 7 days. The curves represent the average and SEMs of two independent experiments.

(G) Schematic of the role of 53BP1 in NHEJ of distant DSBs. In addition to controlling of DNA end processing, 53BP1 can affect NHEJ by increasing the mobility of DSBs. The mobility of DSBs is dependent on the LINC complex and microtubule dynamics. Dashed arrows indicate the possibility that the DDR affects the LINC complex and microtubules independent of 53BP1.

See related Figure S7.

DISCUSSION

These results establish that DSBs show altered dynamic behavior in mammalian nuclei. The mobility and roaming of damaged chromatin requires the MOB domain in 53BP1, the SUN1/2 components of the LINC complex, and dynamic microtubules. In addition, data on telomere fusions implicated plus-end directed microtubule motors (kinesin-1 and kinesin-2) and at least one of the Nesprin proteins in this process. The LINC complex contributes to the dynamic behavior of specific chromosomal loci, including telomeres, during bouquet formation in many eukaryotes (reviewed in (Shibuya and Watanabe, 2014)). However, the process acting on DSBs is different from bouquet formation. While the bouquet configuration bundles loci at one area of the NE in preparation for meiosis I, the DSB mobility recorded here is not overtly associated with clustering or NE targeting.

In the experimental settings analyzed here, the spatial exploration of DSBs promotes their pathological joining by NHEJ. DSB mobility enhanced telomere-telomere fusions forming dangerous dicentric chromosomes and similarly, it promoted the mis-repair of PARPi-induced DSBs generating lethal radial chromosomes. Given these fatal outcomes, a major question is why this pathway is allowed to act on DSBs. Below we propose that the enhanced mobility of DSBs represents a mechanism to restore the connection between DNA ends that have lost their proper interaction. We argue that this mechanism can counteract ectopic repair when DSBs are rare, as is the case under physiological conditions. On the other hand, DSB mobility will promote mis-repair under experimental conditions when a high number of DSBs are generated at the same time.

How DSB mobility could prevent repair errors in G1 and S/G2

It is reasonable to assume that 53BP1 did not evolve to promote the fusion of dysfunctional telomeres and mis-repair of DSBs in PARPi-treated BRCA1-deficient cells. Instead, we propose that 53BP1 has gained the ability to promote DSB mobility to facilitate correct repair (Figure 7). We imagine two settings where increased chromatin mobility at a DSB would be advantageous. The first setting is in G1 when a DSB is formed and its repair by Ku70/80-dependent c-NHEJ is the preferred mechanism to reestablish the integrity of the genome (Figure 7A). If Ku loading fails or synapsis does not occur, the DNA ends might become spatially separated. For instance, chromatin-remodeling and nucleosome eviction at DSBs (reviewed in (Peterson and Almouzni, 2013)) may drive the two DNA ends apart. If the separated ends are mobile, their increased spatial exploration could reconnect them and promote their joining.

Figure 7. Proposed function and mechanism of 53BP1-dependent mobility of DSBs.

(A) and (B) Proposed function for 53BP1-dependent mobility in promoting correct DSB repair.

(A) G1: Mobility of DNA ends that have lost their association could promote their rejoining, thereby promoting NHEJ.

(B) S/G2: If a DNA end loses connection with the sister chromatid and invades an ectopic locus, DSB mobility could disrupt this aberrant interaction and promote correct HDR. If the DSB is being repaired correctly using HDR on the sister chromatid, mobility will not dissociate the ends because of the presence of cohesin and basepairing.

(C) Proposed models for the mechanism of 53BP1/LINC/microtubule dependent mobility of DSBs. The enlarged part of the nucleus shows 53BP1 (red) at a DSB with the ends separated. One end (top) portrays a model in which 53BP1 has a physical connection with the LINC complex (green). The LINC complex connects to dynamic microtubules and thereby moves the LINC-bound 53BP1-covered DNA end. The other end (bottom) portrays a model in which there is no physical connection between the LINC complex and 53BP1. The LINC complex associates with microtubules that ‘poke’ the nucleus. The 53BP1 associated chromatin moves more readily even when not at the periphery, perhaps because 53BP1 alters the flexibility of the chromatin fiber. See text for discussion.

The second setting in which mobility of damaged chromatin could prevent repair errors is after DNA replication (Figure 7B). In S/G2, DSBs can be repaired by HDR using the sister chromatid as the template. However, if the DNA topology is unfavorable, one DNA end (or both) could lose its attachment to the sister chromatid and initiate ectopic repair on a different locus (Figure 7B). Mobility of the chromatin near the DSB could help to disconnect the wandering DNA end from an ectopic locus where it is not held down by cohesin and where basepairing will be limited. In contrast, chromatin mobility of DSBs is less likely to interrupt HDR on the sister because of the stabilizing effects of cohesion and base-pairing.

The proposed role of DSB mobility in counteracting ectopic interactions is analogous to what has been proposed for the mobility of the chromosome pairing centers in C. elegans meiosis (Sato et al., 2009). Dernburg and colleagues argued that this process preferentially disrupt pairing of non-homologous chromosomes since paired homologs will have a greater ability to resist forces. Although the system described here is different from the meiotic events, both regulatory pathways may have evolved to provide a mechanism aimed to distinguish weak non-homologous interactions from the stronger connection afforded by homology.

A key consideration with regard to the role of 53BP1 in DSB repair is that the mammalian DDR did not evolve to handle hundreds of DSBs occurring at the same time. In vivo, the majority of cells in primate brain and liver show no evidence of DSBs and only 10% of the cells have one or two 53BP1 foci ((Fumagalli et al., 2012); U. Herbig, pers comm.), indicating that the occurrence of multiple DSBs in one nucleus is rare in post-mitotic tissues. Furthermore, in MEFs that are in S phase, where DSBs are expected to be more frequent, less than 20% of the nuclei have 5 or more 53BP1 foci and none showed more than 10 (Wu et al., 2010). This number of potential S phase DSBs may be an overestimate because 53BP1 foci can form at a variety of DNA lesions. These observations argue that the 53BP1-mediated mobility of DSBs is unlikely to cause chromosomal aberrations unless cells experience an exogenous genotoxic insult.

Models for the mechanism by which DSB mobility is generated

We are considering two general types of models for how 53BP1, the LINC complex, and microtubules promote mobility (Figure 7C). In the first model, there is a physical connection between the 53BP1-marked chromatin and a LINC complex that interacts with microtubules. In the second model, no such connection exists.

Although we have not been able to establish a physical interaction between 53BP1 and the SUN proteins, it is not excluded that 53BP1 directs DSBs to the LINC complex. If 53BP1 interacts with the LINC complex, kinesin- and microtubule-dependent mobility of the LINC complex could alter the dynamic behavior of DSBs. The lack of clear peripheral localization of DSBs is not a strong argument against this model since the nuclei we have studied are flat, positioning most of the chromatin fairly close to the NE. Furthermore, NE invaginations could allow a connection of a non-peripheral DSB with the LINC complex. We note that the recorded trajectories and the diffusive behavior of DSBs gleaned from the MSD curves argue against the direct interaction model. However, if the engagement is short-lived and takes place in iterative rapid steps, the outcome may resemble diffusive behavior rather directed movement.

Nonetheless, we favor a second type of model in which no physical connection occurs between 53BP1 and the LINC complex. In this model, the role of the LINC complex is to transduce microtubule forces onto the chromatin in an untargeted manner. This process may be analogous to the microtubule-mediated fenestration of the nuclear envelope in prophase, which is in part mediated by the SUN proteins (Turgay et al., 2014). Random ‘poking’ of the nucleus in response to DNA damage would explain why the global chromatin becomes slightly more dynamic in cells with DSBs but how this process is activated by the DNA damage response remains to be determined. It is also unclear whether the visco-elastic properties of chromatin and the resistance of the lamin network allow force propagation over the required distance.

How could microtubule forces specifically increase the mobility of DNA damaged loci in absence of a connection between 53BP1 and the LINC complex? The simplest explanation would be that 53BP1, through a factor that binds to the MOB domain, changes the flexibility of the chromatin fiber containing the DSB. Increased flexibility of the large chromatin domain containing 53BP1 could render it more sensitive to the microtubule forces transduced through the NE. Indeed, chromatin that contains DSBs shows a decreased density as determined by EM and appears to expand (Kruhlak et al., 2006), attributes that could be consistent with a change in the flexibility of the chromatin fibers.

Implications

This study revealed that mammalian cells use microtubules in the cytoplasm to promote the mobility of sites of DNA damage in the nucleus. Although some of the molecular details of this process remain to be determined, the main players, including the MOB domain of 53BP1, the LINC complex, kinesins, and microtubules are now known, allowing further investigation. The results show that in cells with many DSBs, the induced mobility of the damaged chromatin can promote aberrant DSB repair events, including the fusion of dysfunctional telomeres and formation of radial chromosomes in PARPi-treated BRCA1-deficient cells. Two main issues warrant attention in the near future. First, one prediction from our findings is that curbing microtubule dynamics with taxanes might limit the efficacy of PARPi-treatment of HR-deficient cancers. Thus, when a combination of taxanes with olaparib or other DNA damaging agents (e.g. platin drugs) is being considered, the effect of taxanes on the efficacy of genotoxic drugs merits further testing. Second, it will be of interest to test our proposal that the 53BP1-dependent mobility of DSBs can prevent DNA repair errors under normal physiological settings when DSBs are rare.

EXPERIMENTAL PROCEDURES

Live-cell imaging and identification of distorted nuclei

Dysfunctional telomeres were visualized using mCherry-BP1-2 as described previously (Dimitrova et al., 2008). Images were deconvolved and 2D-maximum intensity projection images were obtained using SoftWoRx software. Tracking of mCherry-BP1-2 foci was performed with ImageJ software on at least 10 cells per condition. Cells were registered by the StackReg plugin using Rigid Body (Thevenaz et al., 1998) and particles were tracked using the Mosaic Particle Detector and Tracker plugin (Sbalzarini and Koumoutsakos, 2005) with the following parameters for particle detection and tracking: radius=1-2 pixels; cutoff=1-2 pixels; percentile=1-6; link range=1; displacement=5 pixels. The x and y coordinates of each trajectory were used for further calculation. All mCherry-BP1-2 foci in a cell that were continuously tracked for at least 19 out of 20 frames were analyzed. The analysis of the eGFP-TRF1-marked telomeres was similarly conducted using the following parameters: radius=1 pixels; cutoff=1 pixels; percentile=8-12; link range=1; displacement=5 pixels.

The average x and y values of all the foci was calculated in each frame as the Geometrical Center (GC) and normalized over the GC_t=0. The distance traveled by the GC between each time points t=b and t=a was calculated as Movement of Geometrical Center

$${\mathit{MGC}}_{b-a}=\sqrt{{(x_{t=b}^{\mathit{GC}}-x_{t=a}^{\mathit{GC}})}^2+{(y_{t=b}^{\mathit{GC}}-y_{t=a}^{\mathit{GC}})}^2}$$

and the maximal MGC (MMGC) for each cell was identified. Cells were discarded if their MMGC exceeded the arbitrary threshold of 2, or if their MMGC exceeded the secondary threshold of 1 and another parameter was also above threshold.

The Difference of the Average Distances of all the i foci in the cell and GC_t=0 (ΔAD) between each time points t=b and t=a was calculated as

$$\Delta {\mathit{AD}}_{b-a}=\mid \left(\frac{\sum _{i=1}^n\sqrt{{(x_{t=b}^i-x_{t=0}^{\mathit{GC}})}^2+{(y_{t=b}^i-y_{t=0}^{\mathit{GC}})}^2}}{n}\right)-\left(\frac{\sum _{i=1}^n\sqrt{{(x_{t=a}^i-x_{t=0}^{\mathit{GC}})}^2+{(y_{t=a}^i-y_{t=0}^{\mathit{GC}})}^2}}{n}\right)\mid$$

and the maximal ΔAD (MΔAD) for each cell was identified. Cells were discarded if MΔAD exceeded the arbitrary threshold of 2, or if MΔAD exceeded the secondary threshold of 1 and another parameter was also above threshold.

Finally, the trajectories travelled by each focus i per cell, relatively to the GC, were normalized to the coordinates ${\mathrm{x}}_{\mathrm{t}=0}^{\mathrm{i}}$ and ${\mathrm{y}}_{\mathrm{t}=0}^{\mathrm{i}}$ and projected together on a XY plane. The percentage of foci in each quadrant was calculated for each time frame: Upper Right (UR(%)), Lower Right (LR(%)), Upper Left (UL(%)), Lower Left (LL(%)) and the average of these values during the time-lapse was derived. Laterality (LAT (%)), Verticality (VER (%)) and Diagonality (DIA (%)) were calculated for each time frame as:

$$\begin{array}{cc}\hfill \mathit{LAT}(\%)& =\mid \left(\right(\left(\mathit{UR}\right(\%)+\mathit{LR}(\%\left)\right)÷100)-0.5)÷0.5\mid \times 100\hfill \\ \hfill \mathit{VER}(\%)& =\mid \left(\right(\left(\mathit{UR}\right(\%)+\mathit{UL}(\%\left)\right)÷100)-0.5)÷0.5\mid \times 100\hfill \\ \hfill \mathit{DIA}(\%)& =\mid \left(\right(\left(\mathit{UR}\right(\%)+\mathit{LL}(\%\left)\right)÷100)-0.5)÷0.5\mid \times 100\hfill \end{array}$$

and the average of these values during the time-lapse were derived. Cells were discarded if UR, LR, UL, LL, LAT, VER or DIA exceeded the arbitrary threshold of 40%, or if they exceeded the secondary threshold of 30% and another parameter was also above threshold.

The Cumulative Distance traveled in 10 min by each of the foci i (CDⁱ) was calculated relative to the GC, as previously described (Dimitrova et al., 2008), as

$${\mathit{CD}}^i=\sum _{t=1}^{20}\sqrt{{\left(\right(x_t^i-x_t^{\mathit{GC}})-(x_{t-1}^i-x_{t-1}^{\mathit{GC}}\left)\right)}^2+{\left(\right(y_t^i-y_t^{\mathit{GC}})-(y_{t-1}^i-y_{t-1}^{\mathit{GC}}\left)\right)}^2}.$$

Mean Square Displacement (MSD) was calculated as

$$\mathit{MSD}\left(\Delta t\right)=\frac{1}{n}\times \sum _{i=1}^n{D}_i{\left(\Delta t\right)}^2,$$

where

$$D_i\left(\Delta t\right)=\sqrt{{\left(\right(x_t^i-x_t^{\mathit{GC}})-(x_{t-\Delta t}^i-x_{t-\Delta t}^{\mathit{GC}}\left)\right)}^2+{\left(\right(y_t^i-y_t^{\mathit{GC}})-(y_{t-\Delta t}^i-y_{t-\Delta t}^{\mathit{GC}}\left)\right)}^2}.$$

All data output in pixels (standard ImageJ output) were converted to meters by the formula, 1 pixel = 0.215 μm, based on the characteristics of the objective. Diffusion Coefficient D was calculated as

$$D=m∕4$$

where m is the slope of the MSD after fitting to a linear curve. The Anomalous Diffusion Coefficient α was derived using MATLAB by the fitting of MSD to the diffusion model function:

$$\mathit{MSD}=A+\Gamma t^{\propto }$$

For cumulative distance, statistical analysis was performed using Prism Software applying the Mann-Whitney test.

Other experimental procedures

All procedures for derivation of MEFs, cell treatments, plasmids, shRNAs, immunoblotting, IF, IF-FISH, analysis of metaphase chromosomes, in-gel analysis of telomeric DNA, co-immunoprecipitation, ChIP, and mutagenesis were performed using previously published standard procedures. The mutated 53BP1 alleles were as follows: 53BP1ΔPTIP (S6A, S13A, S25A, S29A) and 53BP1ΔMOB (S674A, T696A, S698A, S784A, S831A, T855A, S892A, S1068A, S1086A, S1104A, T1148A, S1171A, S1219A).

Detailed experimental procedures are given in the supplemental materials.