A Backup Role of DNA Polymerase κ in Ig Gene Hypermutation Only Takes Place in the Complete Absence of DNA Polymerase η¹

Somatic mutation at the Ig locus is initiated by the action of activation-induced cytidine deaminase (AID),⁴ which generates uracils from cytidine deamination in the DNA domain encoding the rearranged V gene segments (1, 2, 3). Uracils produced by DNA deamination are potent mutagens if carried over replication, giving rise to G to A or C to T transition mutations at the targeted site. However, only a fraction of the mutations observed at the H and L chain Ig loci corresponds to such a passive mutagenesis, implying that additional repair processes involving error-prone DNA synthesis, i.e., mutagenic DNA polymerases (Pol), are mobilized to spread and enlarge the mutation spectrum (4).

The first experimental evidence for the implication of mutagenic polymerases in hypermutation came from the observation of Ig gene mutations in memory B cells from patients with the variant form of xeroderma pigmentosum (XPV) (5). Classical xeroderma pigmentosum diseases correspond to inactivation of one of the genes involved in nucleotide excision repair, whereas XPV corresponds to the inactivation of Pol η, responsible for the error-free bypass of UV-induced thymidine dimers during replication (6, 7). Accordingly, XPV patients show an increased sensitivity to UV light with a high frequency of skin cancers. Analysis of Ig gene mutations in XPV patients showed a marked reduction, but not a suppression, of mutations at A/T bases, clearly suggesting a role for Pol η in the generation of these mutations (5, 8).

Studies performed in the mouse have further confirmed the major role of Pol η in A/T mutagenesis (9, 10). We have moreover established that Pol η is the only contributor of A/T mutations during the normal process in the mouse, the residual mutagenesis observed at A/T bases in Pol η-deficient B cells being generated by another enzyme not normally involved in the process and acting as substitute. This enzyme was proposed to be Pol κ, which belongs to the same polymerase family, on the basis of its characteristic mutation signature, biased for A to C transversions (9, 11).

Most actual models of hypermutation are derived from the work of Rada et al. (12), which showed that only two repair factors are involved in the processing of AID-induced lesions, as follows: uracil glycosylase (UNG) and MutS-homolog (MSH)2-MSH6, the mismatch binding moiety of the mismatch repair complex. The consensus view is that processing of uracils is split into a pathway that recognizes it as a mismatch and mobilizes the MSH2-MSH6 complex together with Pol η, and generates mostly (but not exclusively) mutations at A/T bases through an error-prone patch repair. A second pathway recognizing uracils as a foreign base mobilizes UNG and generates mutations at the site of deamination, i.e., G/C bases, most probably through the mutagenic bypass of the abasic sites produced, performed by polymerases involved in DNA damage tolerance. An unknown fraction of G/C transition mutations may also result from uracils carried over replication. Models of hypermutation essentially differ on how repartition occurs between these two pathways: alternative choices, competition, and/or distribution according to the cell cycle (4, 13, 14).

To gain insight into the A/T mutation pathway and the mode of recruitment of Pol η during this process, we studied various conditions in which Pol η and/or Pol κ expression levels are altered in either mouse or human B cells. We describe in particular a new type of mutation of the POLH gene in two related patients with XPV syndrome, mutation that allows the residual production of a normal Pol η protein. From these different approaches, Pol η appears to play a dominant role in hypermutation, Pol κ being unable to be recruited or to compete even in cases in which Pol η availability is reduced.

Pol η-deficient mice have been described previously (9). Pol κ-deficient mice, described previously (15), were obtained from E. Friedberg (University of Texas Southwestern Medical Center, Dallas, TX), and were bred for two generations to obtain Pol η × Pol κ double-deficient animals. All of the animals studied lacked the Pol ι mutation of the 129/Ola background. Animal experiments were performed according to the Institut National de la Santé et de la Recherche Médicale guidelines, and were approved by the Scientific Commitee of the Necker Animal Facility.

The patient XPV-nc1 (referenced as XP606VI), born in Algeria, was diagnosed as XPV at the age of 32. She developed at least three malignant melanomas on partially exposed skin. Familial melanoma was excluded by sequencing the CDKN2A, CDK4, and P14 genes. Her sister (XPV-nc2), 4 years older, exhibited numerous nevi and hyperpigmentations, but did not develop skin tumors yet. Blood samples were obtained for these two XPV patients and from their parents. Primary fibroblasts were isolated from unexposed skin biopsies of patient XPV-nc1 and of her mother. The patient XPV-2, described previously (8) (referenced as XP422VI), presents a stop codon at aa 303 on one POLH allele, and a 4-bp deletion inducing a frameshift at aa 408 on the other allele. The patient XPV-1 (XP127VI), described previously (16), shows a homozygous insertion of one C at position 1091, resulting in a frameshift at aa 364. XPV fibroblasts used as controls have been isolated from a XPV patient presenting a homozygous frameshift mutation at position 358 (XPV-3, registered as XP546VI). Blood and fibroblast samples were obtained after informed consent of the patients. This work was conducted under the control of the Ethical Committee of the European Geneskin project. Control blood samples were obtained from healthy donor volunteers, and from the Etablissement Français du Sang.

Total RNA was extracted from patient’s fibroblasts using the Qiagen RNeasy kit and reverse transcribed by random priming using the ProSTAR first-strand RT-PCR kit (Stratagene), and a 2494-bp fragment encoding Pol η (wild-type size) was amplified using the following primers: Polη-5′cDNA, GATCCCTTCTCGGTTTCTCC and Polη-3′cDNA, TCCATGCCTGTGAAGAGATG (35 cycles, 10 s at 98°C, 30 s at 63°C, and 90 s at 72°C, with Phusion Pol (Finnzymes)). The sequence of exon 1 was obtained by genomic amplification of a 328-bp fragment from fibroblast DNA, using the Polη-5′cDNA primer and a primer located in the intronic region, CTTAGAAGACAGTACCAGGG (same conditions, but 30 cycles).

For quantitative RT-PCR, germinal center B cells (B220⁺ peanut lectin (PNA)^high) from mouse spleen were obtained 4 days after a secondary immunization with SRBC (Eurobio; 10⁹ cells injected i.p.). The spleen cell suspension was subjected to RBC lysis by resuspension in 0.75% Tris-buffered NH₄Cl (pH 7.2), and sorted with a FACSAria (BD Biosciences) using PE-conjugated anti-B220 (BD Pharmingen) and FITC-conjugated PNA (Vector Laboratories) labeling. For mutation analysis, germinal center B cells from Peyer’s patches of 4- to 6-mo-old mice were isolated similarly.

For quantitative RT-PCR, human naive peripheral blood B cells were enriched successively by negative selection with the RosetteSep B cell enrichment mixture (StemCell Technologies) and by CD27⁺ depletion with anti-CD27 magnetic beads and LD depletion columns (Miltenyi Biotec). Naive B cells were activated for 48 h at 37°C at 500,000 cells/well in 24-well plate with 1 μg/ml trimeric CD40L, 5 ng/ml IL-10 (gifts from Amgen, Thousand Oaks, CA), and 10 ng/ml IL-4 (gift from Novartis Pharmaceuticals, East Hanover, NJ) in RPMI 1640 (Invitrogen), 10% FCS (HyClone), 10 mM HEPES, 10 mM sodium pyruvate, and 10% nonessential amino acids (Invitrogen).

For mutation analysis of healthy controls, B cells enriched by negative selection using the RosetteSep B cell enrichment mixture (StemCell Technologies) were sorted into IgD⁺CD27⁺ and IgD⁻CD27⁺ subsets, as described (17). For XPV patients, CD27⁺ B cells selected with anti-CD27 magnetic beads and retained on LD depletion columns were eluted for DNA extraction.

Primary human fibroblasts were isolated from skin biopsies, grown in DMEM (Invitrogen), 10% FCS, 100 U/ml penicillin, and 100 μg/ml streptomycin (Invitrogen) (37°C, 5% CO₂). This procedure allows the collection of naive B cells (for in vitro activation) and CD27⁺ B cells (for mutation analysis) from the same blood sample.

RNA extracted and reverse transcribed as described above was submitted to real-time quantitative PCR using the following TaqMan gene expression assays (Applied Biosystems). For human RNA, β₂-microglobulin (β₂m) HS99999907_m1 and Pol η Hs00197814_m1; for mouse RNA, β₂m Mm00437762_m1, bcl6 Mm00477633_m1, and Pol η Mm00453162_m1 (located in exons 3–4, exon 4 being deleted in the mouse Pol η knock-out). The equivalent of 15,000 cells (mouse) or 25 ng of RNA (human) was used for each amplification.

Intronic mouse J_H4 sequences were amplified using a mixture of five V_H primers covering most mouse V_H sequences and a primer downstream of J_H4 allowing the determination of 490 bp of intronic sequences, as described (9). For human J_H4 sequences, a set of six primers amplifying all human V_H genes, together with a primer upstream from J_H5 allowing the sequence determination of 283 bp, was used as described (17). Amplified sequences were cloned using the Zero blunt ligation kit (Invitrogen), and sequences were determined on DNA extracted from plasmids grown in a 96-well format using an ABI Prism3130xl Genetic Analyzer. Mutations were identified using sequences aligned with the AutoAssembler program (Applied Biosystems).

A 640-bp region at the beginning of the BCL6 gene was amplified using the primers 5′-BCL6, ATGCCGAAGATTAGTCCCAC and 3′-BCL6, ACGATACTTCATCTCATCTGG (45 s at 94°C, 1 min at 56°C, and 3 min at 72°C, for 35 cycles with PFU Pol), and mutations were determined in a core 480-bp fragment.

Pol η was shown to play an essential role during hypermutation of Ig genes in germinal center B cells. We therefore wanted to know whether Pol η could be a rate-limiting factor for this process through the analysis of Polh^+/− mice. First, its mRNA expression level was determined by quantitative PCR in splenic germinal center B cells isolated 4 days after a secondary anti-SRBC immunization, this immunogen being selected due to the large number of cells it mobilizes (10–20% of splenic B cells harboring the PNA^high activation marker). Bcl6, a gene with 100-fold higher expression in germinal center B cells, was used to control the quality of the cell purification procedure, and showed expression levels that parallel the RNA content estimated by β₂m expression (data not shown). In this assay, Pol η mRNA expression is reduced ∼2-fold in heterozygous mice compared with wild-type mice, and appears thus to reflect the copy number of the gene (Fig. 1).

Somatic mutations in the unselected intronic sequence downstream of rearranged V_HDJ_H4 genes were analyzed in Peyer’s patch PNA^high B cells from Polh^+/− animals, and compared with mutation data obtained from both wild-type and Pol η-deficient mice. No difference in the mutation pattern was observed, notably concerning the relative proportion of mutations at A/T bases, thus indicating that there is no haploinsufficiency of Pol η for A/T mutagenesis (Fig. 2 and Table I).

The residual A/T mutagenesis observed in Polh^−/− B cells showed a striking bias for transversion mutations, notably A to C and T to G changes, which correspond to the in vitro signature of Pol κ (18, 19). This led us to suggest that, in the absence of Pol η, another enzyme, not normally involved in hypermutation, was mobilized as a substitute. This proposition was addressed by analyzing mutations in Pol η × Pol κ double-deficient animals. Mutations at A/T bases appeared further reduced, with no specific nucleotide substitution bias among them (Fig. 2). However, in contrast to what is observed in Msh2 × Polh^−/− mice (11), A/T mutations were not completely abolished, suggesting that a second backup exists. No obvious polymerase candidate emerges from the residual A/T mutation pattern of these double-deficient mice. Quantitative analysis of mRNA in germinal center B cells failed to detect any modification in Pol κ expression level in Pol η-deficient or haploinsufficient animals (Fig. 1), indicating that mobilization of Pol κ is not regulated by a modulation of its mRNA expression.

The sequence of Pol η was determined for two sisters diagnosed clinically as XPV. A 2.5-kb Pol η sequence was amplified from patient’s fibroblast RNA and shown to lack 15 bp of the first noncoding exon, adjacent to the 5′ splice site (Fig. 3). Amplification of exon 1 from genomic DNA revealed a homozygous mutation in the intronic consensus donor site (GT to CT) for both patients, a mutation also found at the heterozygous state in both parents. This mutation resulted in the use of an alternative splice site, 15 bp upstream from the normal one (Fig. 3). These patients are thereafter referred to as XPV-nc (for noncoding).

Because the Pol η enzyme bore no deleterious mutation in these two patients, the XPV clinical diagnostic was confirmed by the analysis of the sensitivity of patient’s fibroblasts to UV light-induced DNA damages. As expected, quantification of the UV-induced nucleotide excision repair pathway by unscheduled DNA synthesis was normal (data not shown). Two other distinctive features of Pol η deficiency were analyzed, as follows: the delay in recovery of replicative DNA synthesis, and the arrest in S phase induced after UVC light exposure in the presence of caffeine (Fig. S1).⁵ Both features were found to be qualitatively and quantitatively similar to those of XPV fibroblasts with an inactivating mutation of Pol η.

The level of Pol η mRNA expression was estimated in two different cell types, fibroblasts and B cells. For B cells, naive B cells (enriched as CD27⁻) were analyzed, with or without in vitro stimulation with IL-4, IL-10, and CD40L. The Pol η expression level was 2-fold higher in activated B cells of healthy controls compared with fibroblasts, and 7-fold higher than in resting B cells (Fig. 4). Surprisingly, very different ratios in Pol η expression levels between XPV-nc patients and controls were observed in different tissues. In XPV-nc fibroblasts, the expression was 20-fold lower than in controls. This low expression level accounts for the UV sensitivity of these cells and to the XPV phenotype of these patients. However, the difference with the controls was only 2-fold for resting B cells, and 4-fold for activated B cells, suggesting either a higher efficiency in the use of the alternative splice site of exon 1 in this cell lineage or an increased stability of the mRNA. Interestingly, Pol η expression in fibroblasts from the patients’ mother (with one normal and one mutated allele) was slightly reduced, representing approximately two-thirds of the control value (Fig. 4).

We determined mutations within 283 bp of rearranged J_H4-J_H5 intronic sequences in memory B cells from six healthy donors to get insight into the individual variability of the Ig gene mutation pattern, and to allow for comparison with mutations observed from XPV patients. Two B cell subsets with mutated Ig genes exist in humans, in roughly equal proportions, characterized by the presence of the CD27 surface marker and expressing either the IgM and IgD or the IgG (or IgA) isotypes (20). The mutation frequency differs on average by a factor of two between these two subsets, but the mutation pattern does not differ (Table S1 and Fig. S2).⁵ To allow for comparison with Ig gene mutations from patients that were obtained for total CD27⁺ B cells for reasons of limited sample availability, Ig gene mutations from healthy controls were determined for both subsets separately on a roughly equal number of sequences, and data from both subsets were pooled for each individual (between 400 and 600 mutations per individual). Sequence data are listed in Fig. S2 and Table S1.⁵ The relative mutation distribution is summarized for these six donors in Fig. 5,A and shows interestingly a 10–25% interindividual variation in mutation frequency for major categories of mutations, e.g., G/C vs A/T, or transitions vs transversions (Fig. 5 B).

Two XPV patients (XPV-1 and XPV-2) were selected on the basis of strict inactivating mutations in their POLH gene (see Materials and Methods) and mutations determined in rearranged J_H4-J_H5 intronic sequences from CD27⁺ peripheral B cells (individual data are shown in Fig. S3,⁵ and pooled data in Fig. 5). Mutations showed a marked reduction in targeting of A/T bases (12.5 and 10%, respectively), described previously for coding and noncoding Ig gene sequences (5, 8, 21). In agreement with data reported in the mouse, these residual mutations showed the characteristic bias for transversion mutations (notably A to C and T to G transversions) that corresponds to the mutagenic preference of Pol κ, thus suggesting that Pol κ could be mobilized as backup for Ig gene mutations in human Pol η-deficient cells.

Mutations were analyzed in CD27⁺ peripheral B cells from both XPV-nc patients. Results were similar in both cases (individual data are shown in Fig. S3,⁵ and pooled data in Fig. 5). Mutations at A/T bases were reduced to ∼25% of total, a result in the upper range of all XPV patients studied. However, when A/T mutations were analyzed, no increase in transversions was observed, notably of A to C and T to G, the overall pattern being very similar to controls, and therefore strictly attributable to the enzymatic specificity of Pol η (22). The 4-fold reduced Pol η mRNA expression thus impacts on the frequency of mutations at A/T bases, but by only a 2-fold factor. Nevertheless, despite this decreased mutagenesis, the residual presence of Pol η does not allow another polymerase to compete and to substitute.

Several non-Ig genes have been shown to be targeted for hypermutation in normal germinal center B cells (23, 24, 25, 26). Among them, the BCL6 gene shows the highest mutation frequency, albeit 10- to 20-fold lower than the one observed for V_H genes. Mutations were determined within 350 bp of the BCL6 5′ promoter proximal part, and displayed, as previously described, most characteristics of mutations at the Ig locus (G/C vs A/T and transition vs transversion ratios) (Fig. 6). Some differences were nevertheless noticeable, the overall mutation pattern of any given gene depending obviously upon its specific nucleotide composition and the distribution of mutation hotspots. Among these differences is a marked increase in mutations at T nucleotides (even after correction for the base pair composition of this T-rich sequence), opposite to the A over T bias observed for Ig genes, thus suggesting a different strand preference for A/T mutations at the BCL6 locus.

The mutations observed for the patient XPV-1 showed the characteristic reduction in A/T mutations (14 vs 54% for healthy controls) associated with a deficit in Pol η. Although A/T mutations were biased for transversions as expected if contributed by Pol κ, the sample size was obviously too small for these data to be considered significant. For the two XPV-nc patients, the reduction in A/T mutations was much less severe (34%), and, as observed for the Ig locus, these mutations showed a bias for transitions similar to controls and thus corresponding to the mutation profile of Pol η (Fig. 6, B and C).

We previously established that Pol η is the sole polymerase generating mutations at A/T base pairs during the processing of AID-induced uracils at the Ig locus (11). AID expression was recently reported to be rate limiting for the DNA transactions it initiates, notably in AICDA^+/− mice in which its expression is reduced, or through modulation of expression induced by micro-RNA control (27, 28, 29, 30). Considering the major role of Pol η in hypermutation, we therefore wanted to ask whether this enzyme could be rate limiting as well, and what conditions, notably related to its expression level, could allow another polymerase to substitute, or possibly compete, during hypermutation.

The analysis of Pol η heterozygous mice did not show any alteration of the mutation profile, notably in the A/T vs G/C mutation ratio, despite ∼2-fold reduction in Pol η mRNA expression in germinal center B cells. A previous report described a similar mutation profile, but suggested that the haploinsufficiency of Pol η resulted in a quantitative reduction of mutations (31). We did only observe a very minor, probably not significant, reduction of mutation in activated B cells from Peyer’s patches, thus suggesting that a quantitative effect, if existing, would be at the most an indirect consequence of a reduced Pol η mRNA level. The lack of impact of a complete Pol η deficiency on B cell proliferation in vitro (10, 11) cannot obviously preclude more subtle effects during the germinal center reaction. It has been reported that mouse Pol η^+/− skin cells have an increased susceptibility to UV-induced tumorigenesis (32, 33). The several thousands of pyrimidine dimers introduced during such an irradiation procedure are, however, more likely to be saturating for the nucleotide excision repair pathway as well as for the catalytic activity of the enzyme than the number of uracils introduced by AID per cell cycle during the germinal center reaction (even with the inclusion of a number of non-Ig genes with lower mutation levels (26)). It appears anyway from both studies that there is no haploinsufficiency of Pol η for establishing the A/T vs G/C ratio during hypermutation.

In the absence of Pol η, we previously suggested that another polymerase could be used as a substitute. The residual A/T mutation pattern, biased toward transversions, led us to suggest Pol κ as candidate (9). This enzyme would only be used as backup in the absence of the main player, because its deficiency was reported to have no impact on hypermutation (15, 34). We show in this study, through the analysis of Pol η × Pol κ double-deficient animals, that this is indeed the case. Interestingly, whereas the residual A/T mutagenesis was further reduced in the absence of Pol κ, it was not totally abolished. This would suggest that a second backup may exist in Pol η × Pol κ-deficient B cells, the nature of the polymerase involved remaining to be established. This situation contrasts with the observation made for Pol η × MSH2-deficient mice in which no A/T mutations over background could be detected in germinal center B cells, which indicated that the residual A/T mutagenesis observed in Msh2 or Msh6^−/− mice was contributed by Pol η mobilized by UNG in an error-prone repair pathway (11). None of these substitute polymerases appears thus able to participate in A/T mutagenesis in MSH2 × Pol η-deficient mice, but it is difficult at that stage to establish whether they fail to be mobilized in the UNG pathway, or whether their error rate is too low to be discerned from background.

To gain insight into the hierarchical recruitment of error-prone polymerases during hypermutation, we studied in this work two related XPV patients (named XPV-nc for noncoding), harboring an unusual mutation in the POLH gene. Despite a clear XPV diagnosis, based on both clinical features and characteristic UV sensitivity of derived primary fibroblast lines, both alleles of the gene presented an intact coding sequence. A homozygous mutation was detected that altered the donor splicing site of the first noncoding exon, forcing the use of an alternative site located 15 bases upstream in exon 1, while preserving the coding sequence. This altered splicing reduced the level of Pol η mRNA produced in fibroblasts to only a twentieth of its normal value, thus accounting for the XPV phenotype. Most splicing mutations observed in genetic diseases occur in internal exons and result in either exon skipping (using evolutionary selected splicing sites), or in the use of cryptic sites selected among large intronic sequences. The restricted possibilities offered for alternative splicing in this specific configuration (exon 1 being only 233 bp long) may account for such a nonoptimal processing pathway and the resulting low yield of mRNA produced.

Pol η mRNA expression is increased in control human B cells activated in vitro compared with fibroblasts. Surprisingly, the reduction in Pol η mRNA was less pronounced in activated B cells from XPV-nc patients as compared with healthy controls, the difference being only 4-fold. Although only two types of tissues were examined, this variable reduction in mRNA expression suggests that a variable deficiency in Pol η occurs in different cell types, thus resulting in an unusual mosaic defect of expression at the level of the individual, a genetic deficiency, to our knowledge, not described to date.

As observed for Pol η-deficient mice, the residual A/T mutagenesis observed in XPV patients at the Ig locus harbors the transversion mutation signature of Pol κ (9), further documented in this study on a large database of mutations from two patients with strict inactivating mutations in the Pol η protein sequence. For XPV-nc patients in contrast, mutations at A/T bases were quantitatively reduced at the Ig locus to about half of the normal frequency (25% mutations at A/T bases compared with ∼50% in healthy controls), but their pattern strongly suggested that they were contributed by Pol η alone. A similar mutation pattern was observed at the BCL6 locus, together with sequence-specific features. The preferential recruitment of Pol η appears thus to be maintained even in conditions of limiting enzyme supply attested by a lower A/T mutation frequency, and is thus unlikely to proceed by a competition between enzymes fulfilling equivalent roles.

In the initial description of the impact of Pol η deficiency on the A/T mutation frequency of Ig genes, the A/T mutations analyzed did not reveal a polymerase signature differing from Pol η for two of three of the XPV patients studied (5). Although this initial analysis was performed on functional V_H6 sequences, a similar profile was later reported for J_H4-J_H5 intronic sequences (21). Interestingly, both patients for which a Pol η mutation profile was still manifest in these studies harbored mutations compatible with a residual Pol η activity, i.e., a homozygous missense mutation (Gly263Val) for one patient and a 1-aa deletion (together with a frameshift mutation for the second allele) for the second patient (16). Such a residual activity was not detected in an in vitro translesion DNA synthesis assay using cellfree extracts from fibroblasts (35), but it could possibly be manifest in vivo in activated B cells up-regulating Pol η expression. For the third patient described, in contrast, the Pol η mutation, recently characterized (XP31BE (36)) is an inactivating mutation resulting in a truncation at aa 221.

In conclusion, Pol η appears as a dominant factor in Ig gene hypermutation that is not competed out by other translesion DNA synthesis polymerases, even in conditions of reduced availability. Its replacement by Pol κ as a first line backup, and by a to date unknown enzyme when Pol κ is missing, has been only observed in its complete absence. These data thus reinforce the notion of a preferential and active recruitment of Pol η in an error-prone repair process that seems to be restricted to the specific differentiation stage of hypermutating B cells (37).

We thank Agathe Van der Linden and Floriane De Rosa for excellent technical assistance, and Jérome Mégret and Corinne Garcia for cell sorting. This work has benefited from the mouse-breeding expertise of the Service d’expérimentation animale et de transgénèse, Villejuif, directed by Cécile Goujet. We gratefully acknowledge Amgen for the gift of trimeric CD40L and of IL-10, and Novartis Pharmaceuticals for IL-10. We thank Errol Friedberg and Klaus Rajewsky for Pol κ-deficient mice. We thank very much the patients and their families for understanding of research needs. We are very grateful to the late Dr. Agnès Champret and to Dr. Marie-Françoise Avril (Institut Gustave Roussy, Villejuif, France) for help in this study.

The authors have no financial conflict of interest.