Defects in XRCC4 and KU80 differentially affect the joining of distal nonhomologous ends

Abstract

XRCC4-null mice have a more severe phenotype than KU80-null mice. Here, we address whether this difference in phenotype is connected to nonhomologous end-joining (NHEJ). We used intrachromosomal substrates to monitor NHEJ of two distal double-strand breaks (DSBs) targeted by I-SceI, in living cells. In xrcc4-defective XR-1 cells, a residual but significant end-joining process exists, which primarily uses microhomologies distal from the DSB. However, NHEJ efficiency was strongly reduced in xrcc4-defective XR-1 cells versus complemented cells, contrasting with KU-deficient xrs6 cells, which showed levels of end-joining similar to those of complemented cells. Nevertheless, sequence analysis of the repair junctions indicated that the accuracy of end-joining was strongly affected in both xrcc4-deficient and KU-deficient cells. More specifically, these data showed that the KU80/XRCC4 pathway is conservative and not intrinsically error-prone but can accommodate non-fully complementary ends at the cost of limited mutagenesis.

DNA double-strand breaks (DSBs) are harmful lesions generated by a variety of endogenous or exogenous stresses, potentially leading to genomic rearrangement. Nonhomologous end-joining (NHEJ) is a prominent pathway for DSB repair (1). Canonical NHEJ involves the successive intervention of the KU80-KU70 heterodimer, DNA-PKcs-Artemis, and, finally, ligase IV (Lig4) associated with its cofactors XRCC4 and Cernunnos/Xlf (2, 3). KU-independent NHEJ (KU-alt) has been described in vitro in both acellular extracts and cultured cells (1, 4–6).

Alternative, XRCC4-independent DSB repair pathways (XRCC4-alt) have also been described, using episomic plasmid in cultured cells, using pulse field gel electrophoresis, or in in vitro biochemical experiments (5–10). One hypothesis could propose that XRCC4 and KU are implicated in the same canonical NHEJ pathway, whereas the alternate pathway is independent of both KU and XRCC4. However, in transgenic mice, the inactivation of XRCC4 or Lig4 results in a more severe phenotype than the inactivation of KU (11), suggesting that XRCC4 might have an additional essential function; however, studies show that XRCC4 and Lig4 do not have roles outside of NHEJ, whereas in contrast, KU acts in other processes such as transcription, apoptosis, and responses to the cell microenvironment (12–14).

Alternatively, these varying phenotypes in mice may actually result from differences in DSB repair efficiencies, indicating that defects in XRCC4 might be more deleterious for DSB repair than defects in KU. What challenges this hypothesis, however, is that substantial class switch recombination (CSR) has recently been shown to occur in mouse B cells without XRCC4, whereas no CSR was recorded in cells devoid of KU (15–18).

The relative contributions of XRCC4 and KU80 versus the XRCC4-alt and KU-alt pathways, respectively, to DSB repair remain unclear in wild-type cells. Contrasting results were obtained in living cells, using an episomic plasmid-based reactivation assay. Defects in either KU80 or XRCC4 had no effect on end-joining efficiency in hamster cells, whereas defects in Lig4 decreased the efficiency of end-joining in human cells (5, 19). The basis for this discrepancy could be because of species specificity or because these assays did not monitor intrachromosomal DSB repair.

Finally, xrcc4/p53-null mice harbor “complicons,” complex translocations that arise from high chromosomal instability (20, 21). Thus, the ensuing XRCC4-alt DSB repair pathways might culminate in homologous recombination (22, 23) or XRCC4-alt NHEJ.

Here, we investigated the impact of XRCC4 on the efficiency and accuracy of distal-end joining, comparing these data with those obtained in KU80-deficient cells, using the same NHEJ substrate (present data and ref. 1). We also compared the relative efficiencies of the XRCC4 and XRCC4-alt pathways. We used the substrates depicted in Fig. 1, chromosomally integrated in xrcc4-deficient XR-1 or KU80-deficient xrs6 cells. These substrates allowed us to: (i) study NHEJ of DNA ends on a precise molecular level in the context of chromatin in living cells and (ii) mimic the local events of break-induced genomic rearrangement. Using this strategy, we have shown that although a defect in KU80 barely affects the efficiency of end-joining, it drastically reduces the accuracy of end-joining of fully complementary ends (1). Here, we measured the impact of KU80 on non-fully complementary ends and of XRCC4 on both kinds of ends.

Fig. 1.

The NHEJ substrate. (A) The reporter genes are H2Kd and CD4, coding for membrane antigens. Before expression of the meganuclease I-SceI, only H2Kd is expressed, because CD4 is too far from the pCMV promoter. I-SceI cleaves the two sites on either side of an internal fragment containing H2Kd. The deletion of this fragment and the joining of the two distal ends places the CD4 gene directly downstream of the promoter, and CD4 is then expressed. The expression of the H2Kd and CD4 antigens can be monitored by FACS or by immunofluorescence. Cells having undergone NHEJ events can be selected by magnetic cell sorting, using antibodies against CD4; the repaired junctions can be amplified by PCR and sequenced. Importantly, the two colinear I-SceI cleavage sites of the NHEJ substrate are located in noncoding sequences, and the events are not selected for their viability. Therefore, this strategy allows analysis of the accuracy of joining of distal ends (1). (B) Two substrates were constructed. In pCOH-CD4, the two I-SceI sites are in direct orientation, and I-SceI cleavage generates fully complementary ends for CD4 expression (Left). End-joining either is accurate or leads to extended deletion (sometimes associated with DNA capture). In pINV-CD4, the I-SceI sites are in inverted orientation and generate non-fully complementary ends for CD4 expression (Right). DSB repair either leads to extended deletion or uses the annealing of 2 of the 4 protruding nucleotides (in blue with blue dot), according to 3 classes of intermediates: class I contains 1 nucleotide gap (green -) and one mismatch (red A/A); class II contains two A/A mismatches (in red) and two nicks (black triangles); and class III contains two 3′ nonannealed tails (in red). (C) Names and descriptions of the cell lines used.

These data indicate the existence of an XRCC4-independent pathway for the repair of DSBs induced by I-SceI in a chromosomal context but imply a differential impact of defects in KU and xrcc4 on NHEJ efficiency. Finally, these findings shed light on the accuracy of the NHEJ process itself, which does not appear to be intrinsically error-prone.

Results

Cell Lines and Strategy.

The substrates depicted in Fig. 1 were integrated into Ku80-defective xrs6 or xrcc4-defective XR-1 cells. These substrates can be used to monitor, in a chromosomal context, the efficiency and accuracy of the NHEJ that induces local genomic rearrangement. Because they represent the principal events (1), the excision-deletion events leading to CD4 expression were focused on in this study. Two substrates were devised with respect to the orientation of the two I-SceI sites, generating fully complementary or non-fully complementary ends. Depending on the type of DNA ends generated, joining occurs according to different classes of events or extended deletions at the junction (Fig. 1B) (1). The different cell lines used are listed in Fig. 1C.

Defects in XRCC4 but Not KU80 Reduce NHEJ Efficiency, but an Alternative Pathway Exists in a Chromosomal Context.

In the absence of I-SceI, the frequency of nonspecific CD4⁺ cells was close to background levels (Fig. 2A Left, B, and C). Transfection of xrcc4-deficient cells with I-SceI stimulated a 3- and 7.6-fold increase in the frequency of CD4⁺ cells in XCO-3 and XCO-11 clones, respectively (P < 0.05) (Fig. 2A), and a 6.9- and 7.7-fold increase in XINV4 and XINV5 clones, respectively (P < 0.05) (Fig. 2 B and C). Thus, in xrcc4-defective cells, end-joining of I-SceI-induced DSBs occurs at a significant frequency in a chromosomal context, independently of the structure of the DNA double-strand ends (fully or non-fully complementary ends).

Fig. 2.

Frequency of NHEJ. (A) Example of FACS analysis of I-SceI-induced NHEJ in xrcc4-defective cells in absence of I-SceI (Left), after I-SceI transfection but in absence of complementation by XRCC4 (Center); and after both I-SceI transfection and XRCC4 complementation (Right). (B) Excision/deletion (CD4⁺) involving fully complementary ends. (C) Excision/deletion (CD4⁺) involving non-fully complementary ends. The values correspond to at least five experiments. *, Significant statistical difference (P < 0.05) between the control (without I-SceI) and induced by I-SceI. **, Significant statistical difference (P < 0.05) by complementation with XRCC4. Error bars depict the SEM. The names of the clones used are indicated in the figure.

However, compared with complemented cells, the absence of XRCC4 reduced the efficiency of excision/deletion events involving complementary ends 5.8- and 7.1-fold in two independent clones (Fig. 2 A and B). These differences were not due to stimulation of NHEJ by XRCC4 overexpression. Indeed, overexpression of XRCC4 in wild-type CHO cells did not affect the frequency of CD4⁺ cells (data not shown).

Conversely, the absence of XRCC4 reduced the frequency of excision/deletion events (CD4⁺) involving non-fully complementary ends; in both independent clones, the frequency of CD4⁺ cells declined 3.9- and 5.8-fold in XINV4 and XINV5 clones, respectively, compared with complemented cells (Fig. 2C).

The present data derived from xrcc4-defective cells contrast sharply with those showing that defects in KU80 barely affect end-joining efficiency using the same substrates in hamster cells (comparing data presented here with that presented in ref. 1). For a more direct comparison, we performed the same experiments in KU80-defective xrs6 cells in parallel, this time comparing with complemented cells.

As shown in ref. 1, I-SceI strongly stimulated the frequency of CD4⁺ cells 64- and 61-fold in XD5 and XD11 KU-deficient cells (fully complementary ends), respectively (P < 0.05) (Fig. 2B), and 65- and 13-fold in XU3 and XU7 clones (non-fully complementary ends), respectively (P < 0.05) (Fig. 2C). Importantly, the frequency of CD4⁺ cells among KU-deficient cells was not statistically different from that among KU80-complemented or in wild-type cells (Fig. 2 B and C). These data confirm that the defect in KU80 does not significantly affect the efficiency of distal end-joining, leading to CD4⁺ cells, in contrast with the defect in XRCC4.

We used several independent clones in each case, with two different kinds of DNA ends—fully or non-fully complementary ends—showing no major interclone variability. Moreover, we compared the mutant cells with respective complemented cells. Thus, it is unlikely that the differences in end-joining efficiency between KU-deficient and xrcc4-deficient cells result from position effects of the substrate or from differences in the cell lines.

Taken together, these data show that, in vivo and in a chromosomal context, there is a significant alternative XRCC4 pathway (see Fig. 2A Middle), which retains 14–26% of the efficiency of the XRCC4 pathway, regardless of the structure of the ends. Nevertheless, XRCC4 controls most NHEJ-mediated excision/deletion events (compare Fig. 2A Center and Right and Fig. 2 B and C), whether or not the ends are fully complementary. Importantly, a defect in XRCC4 affects the efficiency of end-joining more strongly than the defect in KU80.

KU80/XRCC4 Is a Conservative Pathway.

To define the respective implications of KU and XRCC4 in NHEJ on a more precise molecular level, we sorted CD4⁺ cells and used specific primers to PCR-amplify and sequence the repair junctions. With complementary ends, error-free events dropped from 15 of 26 events in the XRCC4-complemented cells to 2 of 25 events in xrcc4-defective cells (Fig. 3A). The difference in repair patterns between xrcc4-deficient and XRCC4-complemented cells confirms, on a molecular level, the existence of the XRCC4-alt pathway and shows that this pathway is an error-prone process for I-SceI-induced DSBs, in the context of native chromatin. The mutagenic repair events primarily corresponded to extended deletions. Importantly, most of these events involved microhomologies distal from the DSB (87.5% of the deletion events). On some occasions, deletions were associated with DNA capture (Fig. 3C).

Fig. 3.

Sequence analysis of the junctions in xrcc4-defective cells. (A) Junction sequences from fully complementary ends, in xrcc4-defective cells. (B) Junction sequences from non-fully complementary ends, xrcc4-defective cells. Sequences were performed by using at least two independent sets of experiments. The I-SceI sites are in bold. Squares indicate the locations of internal microhomologies. The numbers of nucleotides involved in microhomology annealing are indicated on the right part of each sequence. Parentheses indicate the number of identical sequences. (C) Frequencies of the different events. The values between the parentheses correspond to the percentage of deletion, using microhomologies among the deletion events.

Imperfect complementary ends can be joined by using partial pairing of four protruding nucleotides (4Pnt) generated by I-SceI (see Fig. 1B). Such was the case in 22 of 29 (76%) XRCC4-complemented cells. But no such events were detected in xrcc4-deficient cells (Fig. 3B), which exhibited mostly deletions. Again, a majority of events presented microhomologies at the boundaries (85%). Some additional deletion events associated with DNA capture were also recorded (Fig. 3C). Thus, in xrcc4-defective cells, the joining patterns were similar between non-fully and fully complementary ends.

The present data show that, in vivo, the XRCC4-dependent pathway is able to ligate non-fully complementary ends. They also show on a molecular level that: (i) XRCC4-dependent NHEJ is mostly a conservative pathway; (ii) XRCC4-alt, which can arise in vivo on one single intrachromosomal end-joining event, is highly nonconservative; and (iii) XRCC4-alt leads primarily to deletions at the junctions and uses mostly microhomologies distant from the double-strand ends, indicating that XRCC4 is not implicated in the use of microhomologies for DSB repair.

In contrast with XRCC4, KU80 deficiency did not substantially affect the efficiency of the joining, but, similar to XRCC4, it affected the accuracy of end-joining of fully complementary ends (1).

To obtain a complete view, we analyzed the impact of KU80 on non-fully complementary ends. Twenty of 21 events sequenced showed that wild-type cells mainly used the 4Pnt for end-joining, i.e., 95% of the events (Fig. 4). In KU-deficient cells, this value decreased to 4%, and nearly all of the events (22 of 23 sequenced) corresponded to extended deletions. Distal microhomologies were efficiently used in KU-deficient cells.

Fig. 4.

Junction sequences of non-fully complementary ends in KU80-defective cells (xrs6). (A) The structures of the DNA ends involved are shown at Top. In bold is the I-SceI cleavage site. Underlined are protruding nucleotides after I-SceI cleavage. Squares indicate the locations of internal microhomologies. Sequences were performed by using at least two independent sets of experiments. The I-SceI sites are in bold. The numbers of nucleotides involved in microhomology anneling are indicated on the right part of each sequence. Parentheses indicates the number of identical sequences. (B) Frequencies of the different events. The values between the parentheses correspond to the percentage of deletion, using microhomologies among the deletion events.

Discussion

XRCC4-null mice show a more severe phenotype than KU-null mice (11). Here, we show, in a chromosomal context and on a precise molecular level, that the defect in XRCC4 affects the efficiency of end-joining more severely than the defect in KU80. Our data are highly consistent with the in vivo viability phenotypes, suggesting that the differences observed between these different mouse models indeed correlate with the respective efficiencies of DSB repair.

Our data confirm existence of the alternative XRCC4 pathway (see Fig. 2), which is able to join distal DNA ends. We show that the XRCC4-alt pathway is highly mutagenic on one single end-joining event, leading to deletion and primarily using microhomologies.

Finally, we compared the relative efficiencies of the XRCC4 versus the XRCC4-alt pathways. These data are highly consistent with the recent in vivo description of significant CSR efficiency in xrcc4- or Lig4-defective mouse B cells by an alternative pathway, using microhomologies and robust XRCC4-alternative V(D)J recombination (17, 18, 24). These results indicate that the XRCC4-alt pathway is not specific to differentiated, specialized cells (B cells), or to a particular process (CSR), but more generally acts in DSB repair and in different cell types.

Interestingly, whereas substantial CSR occurs without XRCC4, no CSR is detected in the absence of KU. That the defect in XRCC4 affects NHEJ more strongly than the defect in KU in our studies raises two alternative hypotheses: (i) the defect in CSR in KU-deficient cells is not due to impaired NHEJ but to a defect in another function of KU (for review, see refs. 12 and 14), or (ii) the CSR events in xrcc4-deficient cells are KU-dependent.

Although their joining efficiencies differ, defects in KU80 or in XRCC4 both abrogated error-free end-joining. In non-fully complementary ends, wild-type cells primarily use the annealing of two of the 4Pnt, in a process consistent with in vitro biochemical data (25, 26). PolX polymerases participate in the synthesis of nucleotide gaps (27, 28).

In wild-type cells, use of the 4Pnt represents 76–96% of the events in non-fully complementary ends. In fully complementary ends, error-free repair restores one I-SceI site that can be cleaved again by the remaining I-SceI molecules, thus increasing the probability of mutagenic repair. In a non-fully complementary substrate, annealing of two of the 4Pnt does not restore a cleavable I-SceI. Thus, use of the 4Pnt can be considered a hallmark of the canonical NHEJ pathway. In Saccharomyces cerevisiae, imperfect end-joining involves small deletions and fill-ins in a KU-dependent manner (29, 30). In accordance with our data, it suggests that extended deletion associated with the use of microhomologies distant from the end is a hallmark of the NHEJ-alt pathway and is thus a process different from that using the 4Pnt, as also suggested by experiments using episomic plasmids (6). Consistent with our findings, microhomology-mediated repair in S. cerevisiae is KU-independent and only partially Dnl4-dependent (31).

Thus, canonical NHEJ is not intrinsically an error-prone process but a process that is as conservative as possible, wherein the occurrence of mutations depends on the structure of the DNA ends rather than on the accuracy of the NHEJ machinery itself. This model is consistent with results obtained in yeast (32). Even when ends are not fully complementary, instead of impairing end-joining completely, NHEJ permits approximate joining of the ends at the cost of limited mutagenesis at the junctions but protects against extensive nonconservative degradation. These conclusions are summarized in the model shown in Fig. 5.

Fig. 5.

The canonical and NHEJ-alt pathways. (A) The canonical NHEJ pathway, involving KU and XRCC4, can seal double-strand ends, even distal and non-fully complementary ends, in a conservative fashion. (B) In the KU-alt pathway, the main event is extended deletion at the junction, generally associated with the use of internal microhomologies distant from the ends. This supports the hypothesis that the NHEJ-alt pathway is initiated by single-strand resection (or by a helicase generating single-strand tails), followed by the joining and annealing of internal sequences. The maturation of the intermediate structures should reseal the DNA, leading to extended deletions at the junction.

Finally, our data show that the KU-alternative pathway is very efficient for end-joining, in the absence of KU protein. This contrasts with the fact that KU-deficient cells repair DSBs induced by ionizing radiation poorly, as measured by pulse field gel electrophoresis (33, 34). Taken together, these findings suggest either that KU is required for multiple DSBs or that the KU-alternative pathway is inefficient with regard to radiation-induced DSBs.

If the signature of the KU-dependent pathway—i.e., use of the 4Pnt—is prominent in wild-type cells (76–95%), this shows that the KU-alt pathway is less efficient than the KU-pathway in wild-type cells. We speculate that KU impairs all alternative pathways involving single-strand resection. Consistent with this hypothesis are the observations that defects in KU80/KU70 result in robust stimulation of homologous recombination (35), which is initiated by single-strand resection, and that KU-defective yeast have increased end-resection (36). Because a defect in XRCC4 affects end-joining efficiency, in contrast to a defect in KU, one hypothesis proposes that XRCC4 acts in both the KU- and KU-alt pathways. In an alternative hypothesis, KU80/KU70 are present and able to bind DNA ends in the xrcc4-defective cells, as has been shown in vitro (37). In the absence of XRCC4, the process cannot be completed, thus leading to the inhibition of the canonical NHEJ pathway. However, the KU80-KU70 heterodimer and DNA-PKcs (still present) would remain bound to the DNA ends, inhibiting the KU-alt pathway by limiting access to the DSBs. According to this hypothesis, KU-alt and XRCC4-alt would correspond to the same pathway, despite the apparent difference in efficiency. This hypothesis is consistent with the knockout of KU restoring the viability of Lig4-null mice (38).

The intrachromosomal substrates used here also constitute a useful working model, mimicking local genetic rearrangements (excision/deletion) generated by the joining of distal double-stranded ends. A substrate with a closer I-SceI site has also been described (39). We have shown that the I-SceI-excised fragment can be inverted or translocated and captured at a third DSB (1). Another model with two I-SceI sites on two different chromosomes has demonstrated the stimulation of chromosomal translocations by I-SceI (40). These results show that the wild-type NHEJ process is able to promote DSB-induced genomic rearrangement.

However, DSB-induced genomic rearrangements have been found to occur very efficiently in KU-defective cells (1, 41). The present data show that XRCC4 is a prominent pathway, accounting for a high frequency of NHEJ of distal ends leading to excision-deletion (CD4⁺). Although less efficient, the XRCC4-alt pathway can actually promote such events.

These data, however, cannot account for xrcc4-null mice showing high levels of complex genetic rearrangements (20, 21). Because the XRCC4 pathway is more efficient than XRCC4-alt in the joining of distal ends, it is likely that additional mechanisms act to maintain genomic stability in wild-type mice and that a deficiency in XRCC4 leads to escape from these processes. Chromosome structure, distance between the two breaks (here, the two I-SceI cleavage sites are 3.2 kb apart), and nuclear organization can affect the joining of more distant ends. Of note is the recent report demonstrating that KU appears to be involved in the immobilization of broken DNA ends (42).

Most uncontrolled DSBs, such as those that develop after ionizing radiation or replication inhibition, infrequently generate fully complementary ends. Canonical NHEJ, which allows rejoining of non-fully complementary ends, should generate genetic diversity at the junctions. Importantly, translocations leading to Ewing's sarcoma involve nonhomologous sequences, and the breakpoint junctions are highly modified (43). The ability to join different kinds of nonhomologous ends underlies the high capacities of genomic plasticity. Therefore, these mechanisms should account for much of the genomic diversity and instability that occur, which are essential in fundamental processes such as molecular evolution, genetic adaptation, and neoplastic development.

Materials and Methods

DNA Manipulation.

All DNA manipulations were performed as described in ref. 44.

Cells and Transfection.

XR-1 cell lines and their derivatives were cultured in DMEM (− pyruvate), and CHO-K1, xrs6, and their derivatives were cultured in α-MEM, supplemented with 10% FCS, 2 mM glutamine, and 200 international units/ml penicillin at 37°C with 5% CO₂. ScaI-linearized NHEJ vectors were electroporated into cells to favor single copy, which was verified by Southern blot as described in ref. 1. Clones were selected with blasticidin (5 μg/ml) added 48 h after electroporation, then scored for H2kd expression.

Cells (2 × 10⁵ or 3 × 10⁵) were plated one day before I-SceI transfection. Expression of the meganuclease I-SceI in the cell lines was achieved by transient transfection of the expression plasmid pCMV-I-SceI (45), using Jet-PEI, under conditions specified by the manufacturer (Q-BIOgene); 2.5 × 10⁻¹³ mol of I-SceI plasmid was used, and for complementation experiments, 7.5 × 10⁻¹³ mol of KU or 7.5 × 10⁻¹³ mol of XRCC4 expression plasmid was used. Expression of KU80 or XRCC4 was verified by Western blot, using specific antibodies as published in ref. 46 for XRCC4 and α-KU86 (Santa Cruz Biotechnology).

Immunostaining and FACS Analysis.

Seventy-two hours after transfection, cells were dissociated with PBS/10 mM EDTA, washed, and fixed in PBS/2% PAF for 15 min at room temperature. Cells (10⁶) were washed in PBS (without Ca²⁺, Mg²⁺). After saturation by PBS/5% BSA, cells were stained for 45 min with 0.2 μg of anti-CD4-FITC (BD-PharMingen). Cells were washed twice in PBS before FACS analysis.

Microscope Analysis.

Cells were washed in PBS and fixed in PBS/2% PAF for 15 min at room temperature. Cells were stained with CD4-FITC (1:500) for 30 min at room temperature in PBS/0.5% BSA.

Enrichment of CD4⁺-Expressing Cells.

Cells were dissociated with PBS and 10 mM EDTA and washed with PBS (without Ca²⁺, Mg²⁺), 0.5% BSA, and 2 mM EDTA. Cells (10⁷) were stained for 15 min with 1.5 μg of anti-CD4-FITC. Cells were then incubated with 20 μl of goat anti-rat IgG-coated beads (Miltenyi Biotec) in 80 μl of PBS, 0.5% BSA, and 2 mM EDTA for 15 min at room temperature. After washing, the positively stained cells were separated by magnetic cell sorting over miniMACS columns and enriched by ≈80%.

Junction Sequence Analysis.

PCR, using the primers CMV-1 (5′-TGGCCCGCCTGGCATTATGCCCAG-3′) or CMV-3 (5′GTACGGTGGGAGGTCTATA3′) and CD4-int (5′-GCTGCCCCAGAATCTTCCTCT-3′), was performed by using genomic DNA from CD4⁺ cells as template. PCR products were cloned into pGEM-T (Promega), which allows for isolation of individual clones, and sequenced.

Statistical Analysis.

Two-tailed Mann–Whitney tests were done by using GraphPad Prism 3.0.