Evaluation of non-canonical p53 functions in DNA replication and recombination for variant classification.

Introduction
TP53 is the most frequently mutated gene in cancer [1, 2], affecting ~30% of all BC cases [3, 4] and ~90% of ovarian cancer (OC) patients [5]. Germline P/LP TP53 variants cause LFS characterized by highly penetrant cancer, occurring in three age-related phases: brain tumors in childhood, sarcomas and BC during early adulthood and other epithelial cancer types at later stages [6–8]. Genetic testing of LFS as well as HBOC patients and families not meeting classic LFS criteria identified P/LP TP53 variants predominantly of the missense type [9], whereby the widespread use of next-generation sequencing (NGS) has been accompanied by a growing number of variants of unknown significance (VUS) [10]. Two major approaches first applied to BRCA1 and BRCA2 are used to comprehensively interpret the probability of pathogenicity: First, multifactorial prediction models integrate direct genetic evidence and clinical data, bioinformatic prediction of splicing, and protein functionality based on structural features and evolutionary conservation [11, 12]. Second, functional assays have gained significance for classification of HBOC variants, reaching sensitivities close to 100% [13]. Significant progress was even made regarding moderate and low-penetrance variants through development of gene-specific, mostly cell-based assays [14–17]. In the case of TP53, these assays focused on tetramer formation [18, 19] and transcriptional activation (TA) of p53 target genes [20, 21]. More recently, TP53 variant-specific fitness scores were calculated, reflecting transcription-dependent induction of either cell death or cell-cycle arrest and DNA-damage removal [22–26]. Such datasets were assembled in public databases, pioneered at the International Agency for Research on Cancer (IARC) [27]. Guidelines for classification of HBOC variants into P, LP, B, LB and VUS have been developed by expert panels from the ACMG/AMP [28], ENIGMA [29], and the GC-HBOC [30, 31].
Although a wealth of data exists on TP53 variants regarding TA and downstream effects, reliance on a single biochemical activity that can genetically be separated from other genome-stabilizing functions of this multifaceted tumor suppressor might fall short [32–35]. Moreover, when annotating existing information on TP53 missense variants in the GC-HBOC database, we noticed discrepancies between functional classifications from different reports [20, 23, 24, 26]. Accumulating evidence indicates a direct regulatory role of p53 in DNA double-strand break repair and recombination [32, 36, 37]. Recently, we and others unraveled previously unknown genome-protecting functions of human p53, altering DNA replication dynamics, particularly in stem cells undergoing self-renewal [38–42]. More specifically, p53 forms an idling complex with the specialized polymerase iota (POLɩ) at DNA replication barriers to slow down replication and to promote the safe DNA-damage tolerance (DDT) pathways of fork reversal and recombination-mediated bypass [34, 39]. Given that loss-of-function (LOF) in TA, homologous recombination and DNA replication of other high-penetrance HBOC gene products like BRCA1 have been linked to pathogenicity, we decided to evaluate non-canonical functions for TP53 variant annotation [43, 44]. Here, we employed two cell-based assays monitoring bypass of replication barriers by a recombination reporter-based approach and the dynamics of nascent DNA synthesis by the DNA fiber spreading assay. Our results show clear discrimination between P/LP and B/LB control variants in the recombination test, whereas DNA replication dynamics showed a more heterogeneous pattern, reflecting the complexity of biochemical processes underlying this readout. Analyzing such non-canonical activities of TP53 VUS identified in the GC-HBOC provides a refined view of pathogenicity-associated phenotypes and demonstrates the power of recombination assays to capture dysfunction caused by moderate structural changes of p53.

Materials and methods

Recombination measurements
Following the recommendations for functional assay development [45], we performed ≥3 independent experiments per variant. Experiments were run in a blinded fashion and in randomized batches together with three positive and three negative controls each, enabling normalization to internal references to exclude inter-experimental differences. Details are provided in Supplementary Materials and Methods.

DNA fiber spreading assay
This assay was performed following the guidelines by Brnich et al. [45], as detailed in Supplementary Materials and Methods.

Molecular Modeling
Structures of p53 cancer variants were modeled using AlphaFold3 [46].
Structural figures were prepared using PyMOL [47].

Statistical analysis and graphs
Graphic presentations of data and statistical analyses were carried out using GraphPad Prism version 9 (La Jolla, CA, USA). For calculation of statistically significant differences, the Kruskal-Wallis H test was applied to the non-parametric k-sample comparisons. In case of statistical significance, the two-tailed Mann-Whitney U test was applied for pairwise comparisons.
Additional details are provided in Supplementary Materials and Methods.

Results

Identification of TP53 VUS in the GC-HBOC
To evaluate the usefulness of testing non-canonical functions for TP53 variant classification, we selected 23 VUS identified in the germline of BC patients and family members counseled at one of the GC-HBOC centers before February 2022 (Fig. 1a; Table 1). We focused on TP53 VUS altering a single amino acid (aa) via missense or in-frame-deletion without predicted effect on RNA splicing (Fig. 1b). As controls, we selected eight B/LB and ten P/LP TP53 variants from the same GC-HBOC cohort. The P/LP variants included truncation and missense variants; the latter comprised cancer hotspot mutations affecting protein conformation (e.g. p.R175H) or specific DNA binding (e.g. p.R273H)[48]. In Fig. 1c, heatmaps illustrate canonical functions of the investigated TP53 variants, comparing transcriptional transactivation (TA) activities of eight p53 targets [20], of WAF1/p21 separately [20], and the p21 protein expression data from our study (Supplementary Fig. 1). Reddish-colored regions in the p53 molecule reflect enrichment for LOF variants.

Measurements of DNA recombination activities to evaluate the impact of TP53 variants
Separation-of-function (SOF) TP53 variants have helped distinguish non-canonical functions of p53 in DNA replication and recombination from its canonical functions in TA and growth control [34, 39]. To test whether such non-canonical functions associate with pathogenicity, we used two functional assays monitoring bypass of replication barriers [39], following the recommendations for functional assay development [45].
First, we determined recombination frequencies in the human, p53-negative K562(HR3) reporter cell line, established for p53-dependent recombination analysis [37, 39, 49]. To compare different TP53 variants, cDNA-based expression plasmids were introduced into K562(HR3) cells by electroporation (Fig. 2a, b). K562 was chosen to exclude apoptosis-related effects [50], and transient expression to minimize protein level changes, as can be seen with stabilized oncogenic mutant proteins [51]. Quantification of p53 proteins in test samples demonstrated changes with on average 1.1-fold augmented wild-type (WT) level for B/LB and 2.2-fold for P/LP variant expression, excluding the truncated variant p.R110Pfs* from this calculation (Supplementary Fig. 1). Expression of transcriptional p53 target p21 was reduced to on average 30% of the WT level in samples expressing P/LP variants. B/LB variants showed an average of 1.3-fold expression.
Recombination measurements were performed by flow cytometry, monitoring the fraction of green fluorescent live cells resulting from reconstitution of wild-type Enhanced Green Fluorescent Protein (EGFP) over a cultivation period of 72 h [37, 39, 49]. Transfection efficiencies ranged from 75 to 94% (Supplementary Fig. 2a), i.e., were uniformly high for the different TP53 variants, as was observed for viabilities of 72–82% according to SSC/FSC gating (Supplementary Fig. 2b). Strikingly, plotting recombination fold changes for each tested variant relative to the WT-specific value in a waterfall plot showed clear discrimination of B/LB variants (green) with a high number of EGFP+ events and P/LP variants (red) with low event numbers (Fig. 2c; Supplementary Fig. 2c). The majority of VUS-specific values (15/23, blue) were found in the intermediate range of >20% to ≤65% relative to the WT. However, five VUS-specific values were within the range of functional B/LB variants, and three were within the range of non-functional P/LP variants. Consistently, statistical calculations showed that all mean recombination frequencies in the range of B/LB variants ( >65%) were significantly different from the mean value for the empty vector (EV) control (b), and all means in the P/LP range ( ≤20%) from the mean for the WT (a) (Fig. 2c; Supplementary Table 1). Significant differences to both controls were found only in the intermediate range (a,b). These data reveal association of p53-mediated recombination with BC suppression. Thresholds for categorization could be defined, underscoring the power of this assay monitoring a non-canonical function to classify TP53 VUS.

Analysis of nascent DNA synthesis of TP53 variants
Second, we examined whether expression of these TP53 variants affects the speed of nascent DNA synthesis, which was previously discovered to reflect p53-mediated induction of a DDT pathway, leading to recombination at the fork [34, 39]. Therefore, we performed DNA fiber spreading assays in p53-deficient K562 cells transiently expressing the TP53 variants analyzed in recombination (Fig. 3a). Replication speed was monitored via measurements of the track lengths of sequential 20 min pulses with thymidine analogs CldU and IdU, stained by green and red fluorescent antibodies, respectively (Fig. 3b). The waterfall plot depicting the mean fold changes of IdU track lengths relative to the means of two internal references (p.R110Pfs*, p.R213X) showed enrichment of P/LP variant-specific values in the range of long tracks, and of B/LB variant-specific values in the range of short tracks (Fig. 3c; Supplementary Fig. 3a). Thresholds for categorization were defined between the values for P/LP variant p.R213X and the WT, leaving a narrow intermediate range between <97% and ≥90% relative to the references. However, P/LP variant p.Q331fs induced a short track length in the range of B/LB, P/LP variant p.R158H and three B/LB variants (p.N235S, p.G360A, p.R202H) in the intermediate range. Reflecting the wide distribution of controls, we calculated significant differences to both EV- and WT-values (a,b) for several B/LB-specific track lengths (Fig. 3c; Supplementary Table 1). To test replication fork stalling as a possible mechanism for track shortening [52], we measured asymmetries of tricolored forks and long/short ratios of CldU and IdU tracks. Fork asymmetries did not reveal any significant changes caused by specific variants (Supplementary Fig. 3b). Track ratios showed <8% increases compared to WT, reaching significance for only four variants (p.R213X, p.M246I, p.R290C, p.C229G) (Supplementary Fig.3c; Supplementary Table 1). It has been reported that the speed of DNA synthesis is higher when origin firing is inhibited [53]. Calculating the percentages of bidirectional forks, i.e. three-colored origins which fired during the first pulse of labelling [52], we observed a highly significant change only for TP53 p.R175H, namely a decrease (Supplementary Fig. 3d). Therefore, and in agreement with our previous work [34, 39], we do not consider fork stalling or changes in origin firing major mechanisms contributing to track shortening.
Side-by-side comparison of the outcomes of the DNA recombination and fiber spreading assays showed that categorization into functional groups matches only for 21 out of 43 TP53 variants (Fig. 4). Consistently, Spearman correlation analysis was not statistically significant (Supplementary Fig. 4). Altogether, analysis of TP53 variants by DNA fiber spreading assays generates a more heterogeneous picture than recombination measurements.

Recombination mediated by TP53 variants correlates with canonical functions
To quantitatively compare our results with functional data obtained by previously established assays for TP53 variant classification (Table 2), we performed correlation analyses (Fig. 5a). Comparing our recombination data with TA monitored by Kato et al. [20] and allele enrichment Z-scores after etoposide treatment by Giacomelli et al. [24] revealed positive correlations, whereas relative fitness scores (RFSs) in Kotler et al. [23] and a recent study by Funk et al. [26] revealed negative correlations, reaching an rs close to -0.80. When comparing the four data sets reflecting canonical p53 functions with replication track lengths, only Z-scores by Giacomelli et al. [24] showed a significant correlation (Supplementary Fig. 5a). Expectedly, mean expression levels of the transcriptional p53 target p21 in our samples significantly correlated with all four previously reported data sets (Supplementary Fig. 5b). Interestingly, our recombination but not replication data correlated with p21 protein levels (Fig. 5b). These observations confirmed the validity of recombination measurements, rather than replication track length measurements, for functional categorization of TP53 variants, as already indicated by the clear separation of recombination frequencies specific to P/LP and B/LB variants (Fig. 2c).
To mimic heterozygosity in the germline, we co-expressed TP53 WT with four control variants (P: p.R110Pfs*, p.R175H, p.R273H; B: p.G360A) and the eight VUS showing functionality or LOF in recombination (Supplementary Fig. 6). When comparing recombination frequencies measured in these samples with the control lacking TP53 co-expression (WT/EV), we did not observe dominant-negative effects (DNEs) predicted to reduce the recombination frequency below that of WT/EV (Supplementary Fig. 6a). Co-expressing p.S121P or p.G360A resulted in significantly elevated recombination compared with WT/EV (and even with WT/WT in the case of p.S121P; Supplementary Fig. 6b), supporting functionalities observed after individual expression in Fig. 2c. Interestingly, co-expressing p.M237T also increased recombination significantly, suggesting that the recombination defect seen with this variant in Fig. 2c can be rescued by WT p53. We conclude that recombination measurements detect LOF but not DNE of TP53 variants and intriguingly, a DNE is not as representative in LFS-associated tumors as loss-of-heterozygosity and gains in copy number of the mutant allele [54].
Of further interest, we observed diametrically opposed phenotypes in recombination (functional) and replication (non-functional) for p.G108S and p.S121P (Fig. 4). When we explored their impact on p53-POLɩ complex formation via proximity ligation assay (PLA), we observed a 35% reduction in PLA foci with TP53 p.S121P (vs. B counterpart p.V31I) and a 3.2-fold increase with p.G108S (Fig. 5c). Although both VUS were categorized functional in recombination (Fig. 2c; Table 2), we noticed that p.G108S expressing cells were at the upper, and p.S121P expressing cells at the lower limit within this functional category. Our results suggest that analysis of individual mechanistic steps of the p53-induced DDT pathway, such as PLA-based detection of p53-POLɩ complexes, may not be as strictly connected with pathogenicity as recombination, which has the potential to add another functional assay to VUS classification.

Discussion
Our study firstly provides a systematic analysis of preclassified TP53 variants observed in HBOC patients regarding non-canonical p53 functions in DNA replication and recombination. While canonical p53 functions guard genomic integrity indirectly via TA of target genes [25, 55], non-canonical functions ensure stability directly via safe bypass of replication barriers [39, 42, 55]. Our reporter-based recombination measurements correctly separate all P/LP and B/LB TP53 variants, whereas measurements of replication speed by DNA fiber spreading failed to unequivocally categorize 25% of these controls. Our recombination-based analyses of 23 TP53 VUS and 18 P/LP plus B/LB controls, which emerged in the genetic testing program of the GC-HBOC, showed highly significant correlations with the results from four systematic studies using assays for canonical p53 functions [20, 23, 24, 26]. Given that previous work demonstrated that canonical and non-canonical functions of p53 can be separated genetically [32, 39], measurements of recombination frequencies added independent functional data for TP53 VUS classification.
The DNA fiber spreading assay is a powerful method for multiparametric analysis of different aspects of nascent DNA synthesis [52]. Here, replication speed was elevated, i.e. correctly indicated LOF, in only 9/11 P/LP/EV controls. Analysis of asymmetries and fiber track ratios excluded the possibility that deviating results in two cases (p.R158H, p.Q331fs) could be explained by slow-down of replication due to replication fork stalling. Moreover, track lengths in B/LB/WT controls were widely distributed, preventing a clear separation of functional and non-functional groups. These observations contrasted with the robust assessment of p53 functionality via recombination measurements. There are both technical and biological explanations for the inferiority of the DNA fiber assay. First, this assay involves multiple steps of experimental manipulations and therefore cannot be performed in a high-throughput format, despite careful optimization and standardization of the protocol. Differently, reporter-based recombination measurements largely rely on sensitive and specific detection of EGFP+ cells by flow cytometry. Second, recombination frequencies showed ≤14.5-fold differences, while track lengths offered a much narrower window of ≤1.4-fold differences. Third, replication can be slowed down by p53 via idling in complex with POLɩ, via HLTF- and ZRANB3-mediated fork reversal and possibly other fork remodeling mechanisms such as involving PRIMPOL and POLζ [34]. This recently refined concept of p53-POLɩ complex-dependent and –independent mechanisms is supported by our observation of diametrically opposed PLA results for p.G108S and p.S121P despite common B/LB recombination phenotype. Involvement of different fork remodelers may also explain why we found SOF variants, which appeared non-functional in replication slow-down but functional in recombination stimulation (p.G108S, p.S121P, p.Q167P) and vice versa (p.P191del, p.M237T, p.Q331fs, p.A347T) (Fig. 4).
Consequently, we focused on recombination measurement as candidate functional assay for TP53 variant classification. When including functional data from our work in an updated standard classification of TP53 VUS following ClinGen TP53 Expert Panel Specifications to the ACMG/AMP Variant Interpretation Guidelines for TP53 Version 2.3.0 [56], our recombination data did not support reclassifying any TP53 VUS as LB/LP (Fig. 2c; Table 2). Most remarkably, VUS p.M237T with LOF in recombination and in Funk et al. [26] as well as p.A347T with LOF in recombination and a defect in tetramerization [18, 19] still remain VUS due to the design of the ACMG/AMP guidelines (Table 2). In particular, criterion pathogenic strong 3 (PS3) cannot be applied because the strict definition of this criterion is primarily based on the functional assay of Kato et al. [20], which showed only partial impairment but not LOF. Criterion PS3 moderate (PS3_mod) cannot be applied either, because retained functions were reported in Kotler et al. [23] and/or Giacomelli et al. [24], which generates conflicting evidence with our results and those of others [18, 19, 26]. We noticed that all VUS investigated here retained functionality in Giacomelli et al. [24] and all but p.P191del in Kotler et al. [23], whereas (partial) LOF was detectable in 17/23 VUS tested in Kato et al. [20], 9/16 VUS in Funk et al. [26], and 18/23 VUS in our recombination assay. Altogether, six variants (p.R196Q, p.P219L, p.C229G, p.I232L, p.M246I, p.R282Q) predicted to be moderately destabilized [26] and tested in all five assays showed retained functions only in the studies by Kotler [23] and Giacomelli [24]. Therefore, there is a possibility that these outgrowth assays are not sufficiently sensitive for detection of only mildly/moderately compromised p53 functions. Disadvantages of TA assays in yeast [20] are test performance at 30 oC rather than at body temperature. p53 has a low intrinsic stability, a feature exploited for development of conditional mutants [57, 58], so that such experimental conditions are predicted to underestimate LOF. In support, AlphaFold Protein Structure Database Heatmap data indicate that both p.M237T and p.A347T are bona fide LP [59]. In light of these limitations, current guidelines may give too much weight to the assays described by Kato [20], Kotler [23], and Giacomelli [24]. We propose that our recombination test may serve as a valid additional assay, particularly as it monitors a non-canonical p53 function, which previously was genetically separated from canonical functions [35, 39]. All other assays analyze canonical p53 functions, which bears the risk of circularity.
Revisiting classification of TP53 variants that show partial function in Kato et al. [20] will be of particular interest in light of recent work by Montellier et al. [8], Kasper et al. [60], and Müntnich et al. [61]. These studies refined the resolution of genotype-phenotype correlations. Kasper et al. [60] argue that TP53 variants contribute to two clinical entities: While classical LFS associated with DNE missense variants, TP53-related BC rather associated with null variants. Montellier et al. [8] and Müntnich et al. [61] were able to cluster missense variants phenotypically that allowed grouping them in four classes, A-D, with A having the lowest and D having the highest TA. TP53 variants clustering in class C were found to be associated with predisposition to BC, moderate penetrance and onset of disease around 40–50 years. Groups A and B meet classical LFS criteria with predisposition to brain, bone, soft tissue or hematological tumors, high penetrance and disease onset around 20–40 years. Class C variants often show partial function and non-deleterious Grantham biophysical prediction scores, class A and B variants show LOF. This phenotypic heterogeneity is not yet reflected in current variant interpretation guidelines for TP53. While P/LP controls in our study all belong to classes A and B, half of the VUS were grouped in class C (Table 2). Among class C variants, VUS p.A347T is of particular interest because it has been detected in four index patients from three BC families in the GC-HBOC, matching reports of six unrelated probands meeting revised Chompret criteria and two families meeting classic LFS criteria according to the variant curation expert panel (VCEP) [62–65]. p.A347T shows SOF between canonical and non-canonical functions, i.e., it may drive BC via recombination defect rather than via LOF in TA/growth control. Our findings suggest that recombination may serve as an independent functional assay for an updated version of variant classification guidelines.
To gain further insights into the structure-function relationships underlying variant pathogenicity, we analyzed the distribution of selected variants across the structure of the DBD and their potential structural impact. Class A and B variants were reported to cluster in major structural motifs of the DBD [8]. The VUS investigated here were mostly, but not exclusively, located in solvent-exposed regions, often in loop structures (Fig. 6a). Variants p.S121P, and to a lesser extent p.G108S and p.Q167P, were functional in recombination stimulation and compromised in TA (Tables 1 and 2). These variants affect different loop regions of the DBD. p.G108S reduces the flexibility of the loop region preceding the first beta-strand, at a site distant from known functional interfaces, whereas p.Q167P alters the short helical turn within L2 on the opposite end of the DBD (Fig. 6a; Supplementary Fig. 7a). Interestingly, this site is located at the edge of the p53-53BP1 interface [66]. The p.H115N variant, which is located between p.G108S and p.S121P, was previously demonstrated to cause the opposite pattern, with LOF in p53´s intrinsic 3´-5´exonuclease activity and, consequently, idling in complex with POLɩ, replication slow-down, and recombination, while TA remained unaffected [39, 67]. Like H115, S121 is located on L1. However, S121 lies at the tip of the loop, directly adjacent to the DNA-contacting residue K120 [68, 69], which is subject to acetylation that modulates p53 binding specificity [70, 71]. The substitution of this serine with proline in p.S121P introduces rigidity into L1, thereby reducing its conformational flexibility and potentially affecting its role in DNA recognition (Fig. 6a; Supplementary Fig. 7b). Replication track lengths further reveal a GOF of p.S121P compared to p53-negative cells, which we speculate is explained by a combination of track lengthening due to reduced idling, analogous to p.H115N [39], and intermediate p21 expression unleashing PRIMPOL-mediated repriming [72]. PRIMPOL generates ssDNA gaps, which can be sealed by recombination [73], and this might explain why the p.S121P variant still shows WT-like recombination despite reduced formation of the p53-POLɩ idling complex (Fig. 5c). Intriguingly, a recent study has revealed that WT p53 can decelerate forks in two ways: either through the formation of a p53-POLɩ complex or independently of this complex but involving HLTF, ZRANB3, and PRIMPOL [34].
p.R175H is the most frequent TP53 variant in cancer and the prime example of p53 GOF in binding novel interaction partners, activating novel pathways, inducing genomic instability, tumor initiation, promoting metastasis and drug resistance [74, 75]. Interestingly, some GOF phenotypes may depend on additional genetic alterations [26, 76]. It remains to be seen whether altered chromatin remodeling, gained interaction with MRE11, augmented aggregation with family members p63/p73 and/or itself [74, 75] are causal to the GOF of p.R175H in repressing origin firing seen here. Given that p.R175H shows severe loss of both canonical and non-canonical functions (Table 2), it is also conceivable that a synergistic defect in origin firing could arise from the loss of replication slow-down and compromised TA-dependent metabolic regulation, which normally ensures histone methylation and prevents R-loop formation [77].
The three SOF variants p.P191del, p.M237T, and p.A347T showed the opposite behavior in our functional assays, with low numbers of recombination events (non-functional) and short replication tracks (functional). The p.P191del variant shortens L2 near R175. In the wild-type structure, the backbone of P191 interacts with the guanidinium group of R175. In the deletion variant, P190 shifts into the position of P191, but this rearrangement disrupts the original hydrophobic packing interactions of P190 with H193 and Y205 (Fig. 6b). The p.M237T variant is located in L3 close to the zinc-binding site, which is critical for protein stability and positioning L3 to enable DNA binding via R248 (Fig. 6a, c). The more frequent variant p.M237I impairs both protein stability and zinc binding [78], and the p.M237T variant is also expected to reduce the conformational stability of the DBD, consistent with the reported LOF in Funk et al. [26]. All three tested variants affecting the OD (p.Q331fs, p.R333C, p.A347T) show a similar SOF pattern in replication versus recombination, highlighting the OD as an interesting domain for future studies. The p.A347T variant, for example, directly perturbs the interface between two primary dimers (Fig. 6d), preventing assembly into stable tetramers [79].

Conclusions
Altogether, evaluating replication-associated p53 functions demonstrates a high discriminatory power of recombination measurements, with complete separation of 20 analyzed P/LP and B/LB TP53 variants. Our analysis suggests sensitive detection of hypomorphic features associated with subtle changes of the protein structure via recombination measurements. Thus, we provide an independent assay with strong potential for the refined classification of TP53 variants with lower penetrance predisposing to BC.

Supplementary information