POLE mutation spectra are shaped by the mutant allele identity, its abundance and mismatch repair status

SUMMARY

Human tumors with exonuclease domain mutations in POLE, the gene encoding DNA polymerase ε, have incredibly high mutation burdens. These errors arise in four unique mutation signatures occurring in different relative amounts, the etiologies of which remain poorly understood. We used CRISPR/Cas9 to engineer human cell lines expressing POLE tumor variants, with and without mismatch repair (MMR). Whole exome sequencing of these cells after defined numbers of population doublings permitted analysis of nascent mutation accumulation. Unlike an exonuclease active site mutant that we previously characterized, POLE cancer mutants readily drive signature mutagenesis in the presence of functional MMR. Comparison of cell line and human patient data suggests that the relative abundance of mutation signatures partitions POLE tumors into distinct subgroups dependent on the nature of the POLE allele, its expression level and MMR status. These results suggest that different POLE mutants have previously unappreciated differences in replication fidelity and mutagenesis.

Graphical Abstract

eTOC Blurb

Hodel et al. demonstrate that POLE cancer variants generate intense POLE hotspot error mutagenesis in human cells even in the face of functional post-replication MMR. The relative abundance of mutation signatures stratifies human POLE mutant tumors into discrete subgroups based on specific alleles and MMR status.

INTRODUCTION

The eukaryotic nuclear genome is normally replicated with high fidelity largely due to three major processes: the high intrinsic base selectivity of each replication DNA polymerase (Pol); a 3' to 5' exonuclease proofreading activity (Exo) contained in both Pol ε and Pol δ; and post-replicative mismatch repair (MMR) (Burgers and Kunkel, 2017; Ganai et al., 2015; Preston et al., 2010). Compromising one or more of these activities can lead to increased mutagenesis.

A number of studies in the decade have revealed a set of mutations affecting the catalytic subunit of Pol ε (POLE) in human tumors. These mutations cluster in the sequence encoding the exonuclease proofreading domain of POLE and are present in cancers from many tissue types, including high incidence in colorectal (3%) and endometrial (8%) cancers (Campbell et al., 2017; Heitzer and Tomlinson, 2014; Kandoth et al., 2013; Palles et al., 2012; Shinbrot et al., 2014; TCGA, 2012; Yoshida et al., 2011). POLE variants are present in tumors with significantly elevated tumor mutation burdens (TMB), ranging from ≥10 to as many 500 Mutations per Megabase (Mut/Mb), in several unique signatures. Importantly, regardless of the mutation burden, all POLE tumor variants are heterozygous, with no apparent loss of heterozygosity (LOH). Even with these strong correlations, important questions remain regarding the degree to which the mutant POLE drives tumor development.

Model organisms have provided important insights into the role that Pol ε infidelity plays in mutagenesis and tumorigenesis. Studies in yeast have shown that mutation rates are elevated in haploid and diploid strains carrying mutant Pol e alleles that cause inactivated exonuclease activity (Morrison et al., 1991; Morrison et al., 1993; Morrison and Sugino, 1994; Ohya et al., 2002; Shcherbakova and Pavlov, 1996; Tran et al., 1999). As mutation rates increase, cells can adapt by evolving ways of suppressing this mutagenesis (Dennis et al., 2017; Herr et al., 2011b; Williams et al., 2013). Murine models have similarly shown that inactivation of Pol ε or δ proofreading elevates mutation rates and accelerates cancer mortality (Albertson et al., 2009; Goldsby et al., 2002; Goldsby et al., 2001). Curiously, cancer mortality in mice heterozygous for a proofreading-deficient allele of Pol ε, unlike what is seen in human tumors, were indistinguishable from wild-type mice.

POLE cancer mutant alleles now appear to be distinct from exonuclease active site mutants. The mutation rate in the orthologous yeast strain with the P286R mutant polymerase is significantly more elevated than in strains with proofreading mutant polymerase (Xing et al., 2019) by virtue of the arginine residue preventing primer annealing at the exo active site (Parkash et al., 2019). This residue change triggers a hyperactive polymerization state, which is hypothesized to drive the observed mutagenesis (Xing et al., 2019). Further cementing the unique status of the cancer alleles, a mouse with the P286R variant is able to drive highly mutagenic tumor development when that variant is heterozygous (Li et al., 2018). However, neither the yeast nor mouse encoding heterozygous P286R variants are able to fully recapitulate the unique POLE mutation signatures, suggesting additional unknown factors are essential. Whether other mutant POLE alleles, like V411L and S459F for example, employ similar mechanisms remains unknown.

MMR plays a critical, though not well understood, role in POLE mutagenesis and tumor development. Roughly 25% of hypermutated (≥10 Mut/Mb) cancers are associated with mutations in MMR genes while a large number of tumors bearing an ultra-hypermutated phenotype (≥100 Mut/Mb) possess mutations affecting both MMR and replicative DNA polymerases, predominantly POLE. Indeed, some tumors from patients with biallelic mismatch repair deficiency (bMMRD) that later acquire an additional somatic heterozygous variant in the POLE exonuclease domain, expand aggressively and show rapid mutagenesis, accumulating some of the highest mutation burdens measured to date (Campbell et al., 2017; Shlien et al., 2015). Interestingly, however, a subset of ultra-hypermutated POLE tumors does not have a MMR gene mutation or evidence of MSI. How these particular cancers acquire this level of mutagenesis in the face of functional MMR remains to be described.

In addition to possessing significantly elevated tumor mutation burdens, POLE-mutant tumor mutation spectra are enriched for several specific signature errors, including three trinucleotide hotspot mutations: C>A transversions in TCT context (C>A-TCT), C>T-TCG and T>G-TTT. Sophisticated mutation signature analyses have largely separated these into the three distinct single base substitution signatures (SBS), 10a (C>A-TCT), 10b (C>T-TCG) and 28 (T>G-TTT) (Alexandrov et al., 2013a; Alexandrov and Stratton, 2014; Campbell et al., 2017; Haradhvala et al., 2018). In addition, SBS 14, which is largely comprised of C>A transversions in NCT contexts, has been identified in tumors with mutations in POLE and inactivation of MMR, sometimes termed POLE/MMR (Haradhvala et al., 2018; Tate et al., 2019). The degree to which the mutator phenotype and its associated signatures are present appears to depend on multiple factors including the position and identity of the amino acid substitution, whether this residue change precedes or follows MMR inactivation and the duration that each deficiency has been present (Campbell et al., 2017).

We developed a conditional CRISPR/Cas9 knock-in system to introduce two clinically relevant POLE variants, P286R and S459F, into human cell lines with and without functional MMR, and then compared nascent mutations in these cells to those from human POLE tumors. We show here that a mutant POLE cancer allele can generate POLE signature mutagenesis in MMR-proficient cells. In MMR-deficient cells, the POLE S459F mutant allele generates a SBS14-like mutation signature at rates approaching MMR-deficient POLE tumors (Shlien et al., 2015). Interestingly, cells expressing mutant POLE proteins are also able to suppress mutagenesis by suppressing transcription of the active mutant allele.

POLE tumors can be stratified into distinct groups based on the relative abundance of POLE and POLE/MMR mutation signatures. Further, the two most frequent POLE mutations, P286R and V411L, have different compatibilities with MMR and drive different amounts of 10a and 10b signature mutations. Taken together, these findings suggest that Pol ε mutants can, by themselves, drive a significant mutator phenotype independent of MMR inactivation but that the final tumor mutation spectrum is shaped by the particular mutant allele, its abundance in the cell and the status of mismatch repair.

RESULTS

Cancer-associated POLE variants cause a significant increase in tumor mutation burden (TMB; >100 Mut/Mb) that is comprised of four distinct single base substitution (SBS) mutation signatures: SBS10a, 10b, 28 and 14. The first three are generally associated with fully functional mismatch repair, while SBS 14 is associated with concurrent mutation in POLE and loss of MMR. While yeast (Xing et al., 2019) and mouse (Li et al., 2018) models have demonstrated elevated mutagenesis and tumorigenesis, the mutation spectra from these models have not faithfully modeled the mutation signatures seen in human tumor samples. To address this, we used a CRISPR/Cas9 knock-in system to generate several diploid human cell lines containing heterozygous or homozygous cancer-associated POLE variants.

Cell line construction

Previously we adapted a synthetic exon promoter trap and recombinant adeno-associated virus (rAAV) method (Rago et al., 2007) to generate cell lines in which we replaced the POLE ExoI catalytic acidic residues with alanines (Hodel et al., 2018). We further adapted this system to permit CRISPR/Cas9-driven homology-directed repair (HDR) and conditional Cre-driven knock-in mutations at two different POLE target loci, P286R and S459F (Figure S1A, B). Both are known patient POLE variants associated with a hypermutated tumor phenotype.

Briefly, we cotransfected HCT-116 cells with a gRNA/Cas9-expressing construct and a neomycin-resistance cassette-containing homology donor. Single cells were cloned using limiting dilution and neo^R clones were selected via growth in neomycin. Targeted integration was confirmed via PCR (Figure S1A,B, primers a/b). Aliquots of these “pre-Cre” clones were then grown and infected with Cre-encoding AAV. Single cells were cloned by limiting dilution and screened for HDR using PCR and Sanger sequencing (Fig. S1A,B, primers c/d). We directly investigated predicted off-target genetic alterations (Fu et al., 2013; Hsu et al., 2013; Schaefer et al., 2017) in our CRISPR/Cas9-treated cell lines and none were observed (Table S1).

Mutation Spectra of POLE mutant cells

We carried out population-level whole-exome sequencing (WES) at two timepoints for each cell line: at an arbitrary starting point, defined here as population doubling level (PDL) 0, and then again after the indicated number of PDLs (Fig. 1A,B). This strategy permits identification of de novo mutations arising in a defined period of time (Hodel et al., 2018). We did this for MMR-deficient (MMR−) POLE^wt/wt and POLE^wt/S459F cells and also for MMR-proficient (MMR+) POLE^wt/wt, POLE^wt/S459F and POLE^S459F/S459F cells (Fig. 1B & Fig. S1C,D).

FIGURE 1. Whole-exome sequencing nascent mutations from engineered POLE mutant human cell lines shows POLE signature mutations.

(A) Schematic for sequencing mutations that occurred during culturing of human cells expressing mutant POLE alleles. (B) Cell populations were subjected to an initial round of whole-exome sequencing (WES) at an initial arbitrary timepoint (PDL = 0) and again after a defined number of population doublings (PDL_x, where x is indicated for each cell line). Nascent mutations in the population were defined as occurring in PDL_x relative to PDL₀. (C) At an intermediate passage (PDL₅₄, red star), the population was again subjected to WES. Single cells were simultaneously isolated from the PDL₅₄ culture by limiting dilution. These subclones were expanded to a confluent well of a 6 well plate followed by WES. Nascent subclone mutations were defined as occurring in each independent subclone relative to bulk PDL₅₄ sequencing. MuTect2 and VarScan2 were used to determine only those mutations arising during passaging or subclone expansion. Nascent mutations for each indicated sample are shown in the 96 possible trinucleotide contexts. The proportion of each base pair substitution (y-axis) in a specific trinucleotide context to the total SNVs (indicated) in a given sample is plotted. Each exome (60 x 10⁶ bp) was sequenced to a mean depth of ~116x (population studies) and ~154x (subclone studies).

In all POLE-S459F mutant samples, regardless of MMR status, there was a clear increase in C>A transversions in a TCT trinucleotide context (C>A-TCT) (Fig. 1B & Fig. S1C,D), one of the canonical POLE signature mutations. An additional POLE signature mutation, T>G-TTT, was also increased in MMR+ POLE-S459F cells. C>T-TCG, the third hallmark POLE signature error, was not an apparent hotspot in any sample.

The high degree of non-POLE mutagenesis observed in these cell lines could represent variants pre-existing in the parental cells below the levels of mutagenic detection. Our method could not accurately detect variants below 5% allelic frequency. Cells containing these variants that also have a competitive growth advantage could then expand to dominate the culture, appearing in the final sequencing as though they arose de novo. To address this issue, we isolated several subclones from a single cell line at PDL54, expanded these cells to outgrown colonies and then sequenced their exomes, (Fig. 1A,C), similar to other studies (Saini et al., 2016; Zou et al., 2018). Variants were identified relative to exome sequencing the bulk population at PDL₅₄, thus removing from the analysis any variants that accumulated due to growth advantage. Each subclone showed a clear enrichment in POLE signature C>A-TCT and T>G-TTT mutations (Fig. 1C). Also evident are a much smaller, but still detectable, number of C>T-TCG mutations that were likely masked by the increased background in the passaged culture. These results clearly show that a cancer variant in POLE is both necessary and sufficient to drive POLE mutation signatures.

POLE Signature Mutations Found in POLE-Mutant Engineered Cell Lines

While mutation spectra provide useful descriptive information, we used mutation signature analyses to better understand contributions of the distinct underlying mutational processes. Non-negative matrix factorization (NMF) (Alexandrov et al., 2013a; Nik-Zainal et al., 2012; Petljak et al., 2019) was used to extract de novo mutation signatures in a mutation spectra dataset containing: our control and engineered cell lines, the tumor-derived HCC2998 cell line known to express POLE-P286R and 64 TCGA samples containing known POLE driver mutations. Tumors with POLE driver mutations were identified as described (Campbell et al., 2017; Park and Pursell, 2019), defined as TMB>10 Mut/Mb and a POLE mutation in the exonuclease domain. NMF revealed twelve distinct signatures from this dataset (NMF1-12, Fig. S2). NMF1-12 can be further grouped according to similarities to existing COSMIC mutation signatures, including the POLE-associated mutation single base substitution signatures (SBS) 10a, 10b, 28 as well as the POLE/MMR signature 14 (Fig. S2C).

Mutagenesis increased 3.1-fold in POLE^wt/S459F MMR-deficient cells over POLE^wt/wt control cells during passaging (Fig. 2A). Almost half of this observed increase was due to mutations bearing strong similarity to those identified in tumors with mutant POLE exo domain combined with loss of MMR (NMF4, green, Fig. 2A) (Haradhvala et al., 2018). These include C>A transversions at NCT motifs and C>T transitions with a strong preference for GCG. Approximately 5% of the nascent mutations were from a signature with maximal cosine similarity (>0.9) to POLE sig 10a, due to the strong C>A-TCT and C>T-TCG peaks, but also containing extra C>A-NCT and other C>T mutations resembling SBS 14 (NMF8, blue-gray, Fig. 2A). The signature of the remaining increased nascent mutations was also observed in the parental POLE^wt/wt cell line (NMF12, orange, Fig. 2A) and is consistent with DNA damage arising from culturing cells in vitro (Petljak et al., 2019). These results demonstrate that cells engineered to express heterozygous POLE mutant alleles in cells lacking MMR were able to recapitulate POLE/MMR mutagenesis (Haradhvala et al., 2018; Shlien et al., 2015).

Figure 2. NMF-derived signature reconstruction identifies POLE and POLE/MMR mutation signatures from engineered human POLE cell lines.

Non-negative matrix factorization was used to extract mutation signatures from exomic SNV profiles from POLE mutant tumors (n = 64) and cell lines (n = 5) and POLE wildtype cell lines (n = 2). SNV/PDL (total nascent SNVs/# of PDLs) is shown for MMR-deficient (A) and –proficient (B) cell lines. Each NMF-derived percent signature contribution was multiplied against the SNV/PDL in the sample to estimate signature-specific mutation accumulation. Mutation spectra and de novo NMF ID (boxed) of signatures identified in the sample are shown.

Human tumors that acquire somatic heterozygous mutations in POLE without any apparent functional defect in MMR are microsatellite stable (MSS) but nonetheless can develop hyper and ultra-hypermutation. A critical unanswered question is whether a single mutant POLE allele alone is capable of driving this mutagenesis or if other factors are necessary. To investigate the mutation spectrum in these cells, we performed a PDL sequencing experiment on POLE^wt/S459F and POLE^S459F/S459F MMR-proficient cells in the same manner as described for the MMR-deficient cells. Mutagenesis decreased 15-fold in the parental cells, consistent with functional MMR correcting replication errors. In MMR-proficient heterozygous POLE^wt/S459F and MMR-proficient homozygous POLE^S459F/S459F cells, mutation accumulation increased 29- and 27-fold, respectively, as compared to the MMR-proficient wild type POLE cells (Fig. 2B). Interestingly, the homozygous S459F allele did not increase mutagenesis over the heterozygous allele. To our knowledge no homozygous mutant POLE tumors have been identified in humans.

Analysis of the mutation signatures in POLE^wt/S459F and POLE^S459F/S459F cells showed that 14% and 13% came from SBS10a-like signatures, respectively (NMF6&11, light blue, Fig. 2B), demonstrating for the first time that a mutant POLE allele can drive POLE-dependent mutagenesis in human cells. Additionally, each line showed an equivalent increase in POLE/MMR signature mutations (NMF4, green, Fig. 2B). These mutations suggest that some cells in the population experience reduced or lost mismatch repair during passaging. As with the MMR-deficient cells, the largest increase in mutation signature is consistent with DNA damage induced by cell culture conditions (NMF12, orange, Fig. 2B).

POLE Mutant Allele Expression

After continuously passaging the POLE^wt/S459F clone for over one year after the PDL and exome sequencing experiment, we retested it using an HPRT1 reporter assay and surprisingly found it was no longer a strong mutator (Fig. 3A). Suppressor mutations are known to reduce mutation rates in yeast (Herr et al., 2011a; Herr et al., 2011b; Williams et al., 2013) and would be predicted to accumulate in this cell line. Additionally, mutant allele transcripts can be under- or overexpressed in the absence of genomic copy number alteration (Govindan et al., 2012; Liu et al., 2018; Rhee et al., 2017). We reasoned that reduced expression of the mutant POLE mRNA could lead to lowered levels of the mutant protein and thus reduced mutagenesis. To test this, we made cDNA from POLE mRNA surrounding each engineered mutation. High depth sequencing of POLE cDNA (2,000X-450,000X average) showed that heterozygous and homozygous POLE-S459F transcripts were expressed at levels consistent with the genotype at the timepoints used for WES experiments, regardless of MMR status (Fig. 3B,C).

Figure 3. Sequencing POLE variant mRNA shows variable allelic expression.

(A) HPRT mutant frequencies were measured for parental MMR-deficient HCT116 cells (POLE^wt/wt), cells that had undergone minimal passaging after Cre recombinase-mediated activation of the POLE-S459F (POLE^wt/S459F-Post Cre) and cells that had undergone extensive (> 1 year) passaging after Cre recombinase-mediated activation of the POLE-S459F (POLE^wt/S459F-Passaged). Individual mutant frequency values are plotted, along with mean and 95% CIs for each group; p-values are shown for each comparison (two sample, unpaired t test; ****, p<0.0001) (B) Total RNA was extracted from POLE variant cell lines and used to prepare cDNA. Loci containing S459 and P286 were amplified from cDNA using PCR and sequenced to high depth using Next-Generation Sequencing. Wild type (gray) and mutant (black) POLE allele frequency is shown for clones that had been passaged for at least 54 passages. * indicates a S459F clone that was re-sequenced after being passaged extensively for over one year. (C) Fresh clones, defined as within 22 cell doublings after Cre excision and single cell selection, were prepared and sequenced as in B. MMR status, POLE genotype and total reads are shown. P286R-l and P286R-r indicate independent clones derived from two distinct CRISPR/Cas9 cleavage sites to the left or right of the P286R codon, respectively. Hatched boxes represent fraction of reads incorrectly spliced into intron 9. (D) Example reads from properly spliced and mis-spliced POLE transcripts spanning exons 9 and 10 are shown.

When we tested several clones engineered to express POLE^wt/P286R, however, the total levels of POLE-P286R mRNA were consistently less than 2.5% of total POLE (Fig. 3B,C). Closer inspection of these reads showed that only 1.6% were from normally spliced POLE-P286R. The remaining POLE-P286R reads resulted from a missplicing event into intron 9 that introduced several in-frame nonsense codons (FIG. 3D), generating a truncated and inactive mutant catalytic subunit. A very small, but still measurable, number of wild type POLE reads also contained this same missplicing event (0.03-1.3%), suggesting that a pre-existing cryptic splice site was activated, but other possibilities cannot be ruled out.

In order to test the possibility that the mutant POLE was initially expressed at 50% but gradually reduced during passaging, we re-derived each mutant and then sequenced within two passages after Cre-dependent mutant allele activation (Fig. 3C). While the S459F mutant allele was expressed at heterozygous levels early, total P286R allele was reduced relative to wild type, representing only 4.6-13.5% of total POLE reads. The misspliced product was again abundant, with the properly spliced P286R allele present at only 1.5-4.4% of the total POLE levels (Fig. 3C). While we cannot formally exclude the possibility that this downregulation is due to DNA mutations in POLE regulatory elements, the observed reduced P286R allele expression came from four independent neo^R CRISPR clones. Since there was no full-length POLE-P286R transcript, we were unable to find any distinguishable mutation signature in these cells. These results suggest the possibility that POLE-P286R expression is sufficiently disruptive to cellular fitness such that its expression is minimized.

The reduction in full-length POLE-S459F expression during passaging raises the possibility that, in addition to acquisition of suppressor mutations (Dennis et al., 2017; Herr et al., 2011a; Herr et al., 2011b; Williams et al., 2013), cells may act to suppress excessive mutation accumulation by reducing mutator allele expression. We tested the possibility that the observed reduction in mutant allele expression arose at least partly in response to stress. In the presence of functional MMR, mutant POLE had no discernable effect on proliferation or cell cycle profile regardless of how long the cells had been passaged (Fig. S3A). In the absence of MMR, however, passaged POLE^wt/S459F mutant cells grew more slowly than POLE ^wt/wt cells and showed reduced populations of S phase cells (Fig. S3A,B). These defects were observed in freshly derived clones as well, suggesting this stress occurs rapidly. Consistent with this, MMR-deficient POLE ^wt/S459F cells showed Cre-dependent, and thus POLE^S459F-dependent, activation of p53 targets, p21 and PUMA, that persisted in the MMR-deficient passaged cells (Fig. S3C). One of the POLE^wt/S459F clones we examined did not show activation of p53 targets, but rather had an extremely large increase in p53 protein levels. Since this typically indicates presence of a p53 inactivating mutation, we sequenced the p53 transcript and found a substitution in the DNA-binding helix H2 (P278L) that has been seen in over 60 human tumors and in a Li-Fraumeni-like family (Bougeard et al., 2001). Taken together, these results suggest that mismatch repair is sufficient to prevent mutant POLE-induced checkpoint activation but not accumulation of POLE-dependent mutagenesis.

POLE tumor sample stratification based on relative mutation signature abundance

Signature extraction from all TCGA tumors with mutant POLE and hypermutation allowed us to distinguish three groups of patients based on the relative abundance of POLE, POLE/MMR and MMR signature mutations (Fig. S4). When these groups were separated and sorted by TMB within each group, several distinctions became clear (Fig. 4A). Patients for whom the POLE/MMR signature dominated (green bars, Fig. 4A,B) had a higher overall TMB than those with little evidence of MMR loss (blue bars, Fig. 4A,B).

Figure 4. Relative abundance of POLE and POLE/MMR mutation signatures assigns human POLE tumors to three distinct groups.

(A) Human POLE tumors with TMBs ≥ 10 Mut/Mb from TCGA were assigned to groups based on the relative abundance of POLE (blue) and POLE/MMR (green) mutation signatures, or lack thereof (red), then each group was sorted from low to high TMB. Engineered (this study) and existing (HCC2998) mutant POLE cell lines are also shown (right). MSI status and MS-indel number are shown below individual patients when known. (B) Mean TMB and 95% CI are plotted for each group and p-values shown for each comparison (two sample, unpaired t test). (C) Mean microsatellite indel number and 95% CI is plotted for each group and p-values are shown for each comparison (two sample, unpaired t test).

Two patients with hypermutant POLE tumors showed less than 5% POLE mutation signature (V411L and L424I mutations, red bars, Fig. 4A-C). While this suggests minimal contribution of these mutant POLE alleles to overall mutagenesis, several features of these tumors point to an alternative explanation: early loss of MMR in the tumor with associated MMR-deficient mutation signature acquisition followed only later by mutation in POLE. Both samples have significant amounts of canonical MMR-like mutation signatures combined with high levels of indels (Fig. 4C). Each tumor has a frameshift mutation in a critical MMR factor (MLH1 with POLE-V411L and PMS2 with POLE-L424I) at higher variant allele frequencies than those of the POLE mutations. They both also have several additional missense mutations of unknown significance in multiple MMR proteins.

Across all patients, the relative abundance of signatures with high similarity to SBS 10a, 10b and 28 was variable (compare light blue [NMF6,11], dark blue [NMF1,2,10] and blue-gray [NMF3,8] bars, Fig. 4A). To investigate this further we focused solely on the POLE signatures in each patient sample (Fig. S5). The POLE patient samples tended towards roughly twice as much 10a than 10b, while the POLE/MMR patients had roughly equal 10b and 10a (Fig. 5A). The two MMR-like patients were essentially devoid of 10a signature errors.

Figure 5. POLE mutant allele-dependent distribution of POLE mutation signatures, independent of MMR status.

(A) NMF-derived POLE signatures (this study) were grouped according to their similarity to existing SBS signatures 10a (NMF6/11, light blue), 10b (NMF10/1/2, dark blue) and 28 (NMF3, gray blue). Mean signature contribution as a proportion of POLE signatures only, 95% CI and p-values are shown. Differing distributions of POLE signatures in human POLE tumor subgroups (as in Figure 4) are shown. Statistical comparisons made on SBS 10a contribution (two sample, unpaired t test). (B) Mean POLE signature contributions are shown for MMR+/− engineered POLE mutant cell lines and variant-specific POLE mutant tumors as in A. Statistical comparisons made on SBS 10a contribution (ANOVA with post-hoc test). (C) Signature contributions to individual trinucleotide base pair substitutions are shown for MMR+/− engineered POLE mutant cell lines and select POLE mutant tumors.

Grouping samples by amino acid change showed distinct, allele-dependent differences in the 10a:10b signature ratio. The most common alleles, P286R and V411L, showed inverse ratios (Fig. 5B): 10a errors predominated in P286R tumors (2.2:1), while the ratio flipped in V411L tumors (1:2.3). In contrast, the S459F tumors showed strong 10a:10b bias (4.6:1) bias. While this manuscript was under review another study observed a similar phenomenon, strengthening these results (Fang et al., 2020). Taken together, these results suggest the possibility that the origins of each POLE hotspot replication error may depend more strongly on the mutant allele than previously believed.

In our POLE-S459F engineered cells, we noted a lack of the 10b C>T-TCG hotspot mutations, with almost all POLE errors assigned to 10a. Strikingly, we observed one patient, AG-3892, with an almost identical distribution of 96% POLE errors due to 10a (Fig. 5B). Like our engineered cells, this tumor contained an MSS POLE-S459F mutant. This is the first engineered POLE mutant cell line to recapitulate the mutation signature from a patient sample.

Signature extraction and reconstruction also allows for a more detailed understanding of specific mutation signature contributions to specific single base pair changes in individual samples. For example, sample E6-A1LX, an endometrial tumor with the P286R allele, is an example of a canonical mutant POLE tumor (Fig. 5C). The three hotspot mutations were prominent, with 10a responsible for the majority of C>A-TCT, 10b responsible for T>C-TCG and 28 responsible for T>G-TTT. Another example is EO-A22U, a POLE/MMR classified P286R endometrial tumor, which contains one of the weaker overall POLE mutation signatures. Individual base pair analysis showed that the POLE hotspot mutations were clearly due to the mutant POLE allele. This analysis was extended to the engineered POLE cell lines. While the overall POLE hotspot mutations were relatively modest in the overall mutation spectra (Fig. 1B & Fig. S1C,D), individual base pair signature analysis (Fig. 5C) clearly showed that the POLE-S459F mutant was responsible for 60-90% of these nascent mutations in the mutant cell lines. This was further confirmed in the subclone analysis (Fig. 1C). Signature reconstruction also demonstrates that the very low levels of C>A-TCT, T>C-TCG and T>G-TTT mutations seen in wild type polymerase cells are clearly not due to POLE mutagenesis (Fig. 5C).

DISCUSSION

We show that human POLE tumor variants are able to produce the POLE mutator signatures when expressed in human cells. Contrary to previous models of exonuclease inactivation, cancer-associated POLE alleles are capable of generating a hypermutant phenotype even when MMR is functional. Cells can lose expression of mutant polymerase alleles over time, resulting in loss of the mutator phenotype. We further show that a POLE^wt/S459F cell line drives mutagenesis most similar to a tumor with a heterozygous S459F mutation. Analysis of patient mutation spectra reveals that POLE tumors can be subdivided into distinct classes, depending on their relative abundance of POLE, POLE/MMR and MMR-only mutation signatures.

Restricting mutation analysis to just POLE signature mutations shows that the relative abundance of the POLE signature hotspot mutations is heavily influenced by the particular POLE mutant allele. For example, S459F tumors have a significantly higher contribution of 10a relative to 10b errors than do V411L tumors. However, this ratio is not absolute, varying somewhat among tumors with a particular mutant allele. For example, among the four tumors with the S459F mutant alleles, the contributions of 10a errors to POLE mutagenesis are 53%, 58%, 69% and 92% (Fig. 4A). This is likely driven in some part by variance intrinsic to the mutant enzyme but also in part by other factors, especially the particular genetic background of an individual tumor. Further studies with mutant POLE alleles expressed in different cell lines will prove helpful in distinguishing these effects. Collectively, these data provide evidence that the unique mutation signatures found in POLE tumors are shaped by several factors including the nature of the mutant allele, its expression level, MMR status and likely the genetic background.

POLE mutations found in hypermutated tumors are all located in the exonuclease proofreading domain. Tumors bearing somatic or germline POLE mutations are heterozygous with no evident LOH (Briggs and Tomlinson, 2013; Kandoth et al., 2013; Palles et al., 2012; Shinbrot et al., 2014; Shlien et al., 2015; TCGA, 2012). These tumors are all associated with four distinct mutation signatures, SBS10a, 10b, 14 and 28. While the mutation signatures and tumor-driving properties were believed to be driven by proofreading deficiencies, recent results in yeast have questioned this model (Barbari and Shcherbakova, 2017; Kane and Shcherbakova, 2014; Xing et al., 2019). To help resolve these issues we introduced POLE mutations into human cells to more closely model the situation in human tumors. Unlike the POLE^{wt/D275A;E277A} cells we engineered previously (Hodel et al., 2018), some POLE cancer variants can drive mutagenesis even when MMR is functional. Similar to results in yeast, these mutants appear to be intrinsically different, and much more mutagenic, than the ExoI metal-binding mutant allele. Whether this is also due to hyperactivated DNA synthesis for the S459F mutant as is seen in the yeast P286R equivalent (Xing et al., 2019) remains to be studied.

Whole-genome and -exome sequencing coupled with computational mutation signature extraction have proven incredibly powerful tools to help define global mutation patterns and tumor etiologies in cancer patients (Alexandrov et al., 2013a; Alexandrov et al., 2013b; Alexandrov and Stratton, 2014; Helleday et al., 2014; Nik-Zainal et al., 2016). When applied to our engineered POLE mutant cell lines along with POLE patient tumors, we are able to more precisely define the contributions of mutant POLE to mutagenesis. In the absence of MMR, cancer mutant POLE drives the acquisition of 38 exome-wide SNVs per population doubling (Fig. 2A). This considerably exceeds the mutation rate of engineered proofreading inactivation (Hodel et al., 2018). We calculated that 10.2 SNVs/50 Mb/PDL could be assigned to SBS14, the POLE/MMR signature. This is on the same order of mutagenesis measured in cMMRD/POLE mutant brain tumors (Campbell et al., 2017; Shlien et al., 2015).

When MMR is proficient, the heterozygous S459F mutant allele population accumulates 2.4 SNV/50 Mb/PDL that are SBS 10a. This is likely only a minimal estimate in light of the subclone data (Fig. 1C). Each subclone contains an average of 510 mutations over the 54 population doublings plus the first several detectable doublings during subclone expansion, 90.2% of which are unique to each individual clone. Of these, 8.0%, or 40 per subclone, are C>A-TCT. Thus, at the time of sequencing, the entire population of cells may have accumulated significantly more POLE mutations than were measured in the population doubling experiment. For similar reasons it is likely that the 38 SNVs per population doubling measured in the MMR-deficient cells (Fig. 2A) is also a minimal estimate and would require careful subclone analyses. Future experiments will be designed to test this more quantitatively.

The lack of NMF12 signature mutations in the subclones also suggests that oxidative mutagenesis is not contributing substantially to ongoing mutagenesis in MMR-proficient POLE^wt/S459F cells. The presence of NMF12 in the MMR+ populations of POLE^wt/S459F coupled with their absence in populations of POLE^wt/wt cells (Fig. 2B) provides further evidence for clonal expansion of pre-existing mutations. The POLE^wt/S459F cells were not subject to single cell bottleneck prior to passaging, while POLE^wt/S459F cells were bottlenecked during POLE mutant isolation, providing an explanation for these mutations appearing to dominate in the POLE^wt/S459F cells.

These results suggest that POLE mutants can easily drive the ultrahypermutant TMB with one allele expressed even in the face of functional post-replication MMR. This particular variant, at least in a human cell culture model, also does not generate large amounts of SBS10b signature mutations. Coupled with our patient tumor analyses, this suggests that these mutations might require the presence or absence of additional factors.

Cells are tuned to tolerate mutagenesis up to a certain threshold, above which evolution favors cells with mutations or epigenetic changes that reduce this stress (Dennis et al., 2017; Herr et al., 2011a; Herr et al., 2011b; Morrison and Sugino, 1994; Pavlov et al., 2004; Williams et al., 2013). Our results suggest that this stress may present in the form of persistent checkpoint activation. In response to DNA polymerase mutator alleles, suppressor mutations arise that act to reduce the stress, as well as the mutation rate, either in cis (within the polymerase gene itself) or in trans (in genes affecting mutation acquisition). Allelic imbalance, in which a mutant transcript is over- or under-expressed relative to its wild type copy, is known to occur in tumor cells (Gardner, 2010; Lindeboom et al., 2016; Mort et al., 2008). Nonsense mediated decay in particular is effective at inactivating a mutant copy that might prove deleterious (Brogna and Wen, 2009). Based on a number of in vitro biochemical, yeast and human tumor studies, POLE-P286R expression is predicted to drive strong mutagenesis in human cells, even when heterozygous. The strong activation of the misspliced P286R transcript, which introduces several in-frame nonsense codons, would decidedly counteract this mutator phenotype. While we cannot rule out construct-specific effects, the fact that the misspliced product occurred in the wild type allele and with two different P286R constructs suggests that this could be a natural phenomenon that is strongly induced to suppress mutation acquisition.

It has been noted that POLE-P286R tumors designated as MSI might be false-positives generated due to base pair substitutions within the microsatellite sequences (Haradhvala et al., 2018). Our results further strengthen this argument. Here we find that P286R tumors are heavily biased towards the POLE signatures. Only 2/25 POLE-P286R tumors (EO-A22U and IB-7651) in this set are classified as POLE/MMR. IB-7651 has some of the highest POLE/MMR and lowest POLE signature contributions observed. It is of note that IB-7651 has a nonsense mutation in MSH6 at a higher variant allelic frequency than POLE P286R. EO-A22U is unique in that it has a very modest POLE signature contribution, small MMR signature, has no POLE/MMR signature and is the sole pancreatic POLE tumor in this set.

In contrast, 93% of P286R tumors are classified here as POLE, with 60-90% of SNVs arising from SBS 10a, 10b and 28. P286R also does not arise spontaneously in bMMRD tumors with preexisting lack of MMR. By and large P286R appears to be incompatible with loss of MMR. V411L mutants are evenly distributed between POLE and POLE/MMR categories, indicating that this mutant better tolerates, or is perhaps indifferent to, MMR contribution. This is intriguing in light of the allele-specific POLE signature mutation distribution. V411L mutants are more heavily biased towards the 10b, C>T-TCG dominated signature. Haradhvala and colleagues noticed a similar phenomenon with V411L having a higher fraction of what they termed E3 mutation signature, which is dominated by C>T-TCG and lacking C>A-TCT (Haradhvala et al., 2018). V411L mutants also frequently co-occur with loss of MMR and MSI. One possibility is that this mutant drives a slower accumulation of POLE signature mutations, thus allowing more time for the 10b signature and loss of MMR to occur. This is consistent with results in yeast showing a much more modest mutator effect for this allele as well as its physical location being removed from the exonuclease active site (Parkash et al., 2019; Xing et al., 2019). This is also consistent with the possibility that 10b signature mutations are facilitated by an additional process. Another possibility is that P286R and V411L differ fundamentally in their biochemical activities such that they drive different mutations, but this has yet to be tested.

An additional alternative possibility, supported by our cell model results, is that the POLE/MMR error signature is closer to the “true” intrinsic mutant POLE replication fidelity. These are errors made by the polymerase during replication that arise only when they are not corrected by MMR. The 10a, 10b and 28 error signatures are then the result of MMR correcting some errors, while other errors escape MMR with perhaps an additional subset induced by other factors.

The ability to quantify the level of contribution for each signature to specific trinucleotide mutations in individual patients is also extremely helpful to distinguish mutant allele-specific effects, particularly for samples containing multiple complex signatures. For example, IB-7651, which is dominated by POLE/MMR mutagenesis, shows small but clear POLE mutation signature contributions, suggesting POLE mutation preceding MMR loss. EO-A22U shows well-defined POLE signature errors that are dominated by 10a, MMR loss errors and yet no POLE/MMR errors. B5-A3FC provides a clear example of possible very early MMR loss followed much later by the POLE mutation (Fig. 5C). In another example, the background mutagenesis in our cells is high, likely due to effects stemming from culturing in high oxygen tension (Petljak et al., 2019). POLE signature errors are seen in all our cell samples, regardless of MMR status. However, individual signature contributions from SBS14 (POLE/MMR in MMR-deficient cells) and SBS10a (in MMR-proficient cells) can clearly be seen (Fig. 5C). Additionally, a small contribution of SBS14 can be seen in POLE-S459F cells even when MMR has been restored, indicating a small fraction of cells that did not fully restore MMR.

Combining more sophisticated mutation analyses of human tumor samples with engineered model systems is a powerful approach to understanding how mutagenesis shapes tumor progression. With recent advances in sequencing normal human tissues, it will be of great interest to apply this combined approach to measuring how normal cells start down and commit to the path of tumor development.

STAR METHODS

RESOURCE AVAILABILITY

Lead Contact

Further information and requests for resources and reagents should be addressed to the Lead Contact, Zachary F. Pursell (zpursell@tulane.edu), Department of Biochemistry & Molecular Biology, Tulane University School of Medicine. All unique reagents generated in this study are available from the Lead Contact.

Materials Availability

Plasmids and cell lines generated in this study are all available upon request via the Lead Contact.

Data and Code Availability

The datasets generated during this study are available at the NCBI Sequence Read Archive (SRA), accession code: PRJNA589337. The code used to analyze the WES data in this study are available from the corresponding author (Z.F.P) upon request. Unprocessed Western blot images are available through Mendeley (doi: 10.17632/cdgs6fmpsn.2).

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Human cell lines and cell culture

Human colorectal HCT-116 cells were cultured in DMEM supplemented with 10% fetal bovine serum (VWR), 1X sodium pyruvate and 1X MEM-NEAA. All cell lines were cultured using standard growth conditions (10% FBS complete media 37°C, 5% CO₂).

METHOD DETAILS

Generation of CRISPR-Cas9 Constructs

Suitable targets for Cas9 cleavage were identified in silico at crispr.mit.edu. The following guide sequences were used to target POLE ExoI: Guide 3: 5’-ATGATTTCCTACATGATCGA-3’ and Guide 13: 5’-AGGAAACTTGAGGGGCAGTT-3’. The following guide sequences were used to target POLE ExoIII: Guide 2: 5’-GGTGGACGTACTTCATGTAC-3’ and Guide 4: 5’-AGCATCTGACACAGAATACG-3’ Guide 5: 5’-CTTCATGTACAGGTAGTAAG-3’. Oligos (Invitrogen) designed to express gRNA targeting the guide sequences above were cloned into the prelinearized CRISPR OFP Nuclease Vector (Invitrogen) as per the manufacturer’s protocol and verified by Sanger-sequencing.

Generation of Homology Donor Constructs

To facilitate the introduction of mutations leading to knock-in of Pol ε cancer alleles, we used a synthetic exon promoter trap (SEPT) flanked by ~1kb homology arms (HA). For the homology donor targeting P286, a 1029bp fragment containing exon 9 from POLE (termed HA1) was PCR amplified from HCT-116 genomic DNA using primers designed to add KpnI and SacI sites to the 5’ and 3’ ends, respectively. Concurrently, a 1000bp fragment containing exons 10, 11 & 12 as well as introns 10 and 11 from POLE (termed HA2) was PCR amplified from HCT-116 genomic DNA using primers designed to add EcoRI and NotI sites to the 5’ and 3’ ends, respectively. HA1 and HA2 were cloned into a pCR2.1-TOPO vector and sequence verified. The HA1-containing plasmid was then subjected to site-directed mutagenesis (SDM) to change proline 286 to arginine and sequence verified. The P286R-modified HA1 plasmid was then subjected to a second round of SDM to introduce silent mutations into the PAM and PAM-proximal nucleotides in the guide sequence and sequence verified. HA1 (KpnI/SacI), HA2 (EcoRI/NotI) and SEPT (SacI/EcoRI) were cloned in three sequential steps into an empty pCR2.1-TOPO vector to obtain the full homology donor.

The homology donor targeting S459 was generated in the same manner as described for the P286 homology donor with the following differences: HA1 was 1033bp including exons 13 and 14 as well as intron 13. HA2 was 924bp spanning exons 15 and 16 as well as intron 15. All restriction site adaptors added during PCR amplification were identical except for the 3’ end of HA2, which used XbaI. SDM performed on the HA1-containing plasmid to introduce mutations leading to the serine 459 to phenylalanine amino acid substitution was performed. The second round of SDM altered the respective guide sequences targeting this locus in a similar manner as above. HA1, HA2 and SEPT were sequentially cloned as described above.

CRISPR-mediated knock-in cell lines

Prior to transfection, 1.25 x 10⁵ HCT-116 cells were seeded into 24-well plates and incubated at 37°C/5% CO₂ for ~24 hours. Cotransfection of 500ng of both the homology donor and CRISPR/Cas9-expressing plasmids were performed using lipofectamine 2000 (Invitrogen) according to the manufacturer’s instructions. 72 hours post-cotransfection, media was replaced and supplemented with geneticin to a final concentration of 400 μg/mL. Cells were incubated under selection for an additional 14 days with media changes every 3 days. Geneticin-resistant colonies were counted at this time. Colonies were trypsinized, pooled, counted and diluted into two 96-well plates at a density of 0.5 cells per well and incubated for an additional 10-12 days. Wells visually confirmed to contain a single, well-defined colony were transferred to 24-well plates for expansion and subsequent screening. Genomic DNA was isolated through TNES lysis, phenol:chloroform extraction, ethanol precipitation and finally resuspended in 100 μL TE (10mM, 0.1mM). Successfully targeted clones were identified by the presence of a PCR amplicon using a primer set where the forward primer annealed outside HA1 and the reverse primer annealed inside the SEPT cassette (See Fig. S1A).

Cre-mediated excision of SEPT cassette

1.9 x 10⁵ cells from the positive clones identified in the previous section were seeded into 6-well plates and incubated at 37C°/5% CO₂ for ~24 hours. Media was removed and replaced with 1.6 mL media containing 1 uL Cre-expressing adenovirus (1 x 10¹⁰ PFU/mL, Vector Biolabs). After 24 hours cells were trypsinized, pooled, counted and single-cell diluted into two 96-well plates at a density of 0.5 cells per well and incubated for 10-12 days in nonselective media. Wells visually confirmed to contain only one well-defined colony were transferred to 24-well plates for expansion and subsequent screening. Genomic DNA was isolated as above. Colonies with successful excision of the SEPT cassette were screened with PCR amplification with primers containing HA1 and a novel SacI site introduced during the cloning process with subsequent digestion with SacI. Positive clones retaining a wild type SacI-resistant band and two SacI cleavage products emanating from the targeted and excised allele were identified and knockin allele zygosity determined by sequencing.

Assessment of off-target effects from CRISPR

Off-target sites were identified in silico at crispr.mit.edu. For a given guide sequence the two top scoring off-target sites (regardless of coding status) and the top scoring exonic off-target site were chosen to investigate. PCR primers capturing the potential off-target DSB site were designed. Genomic DNA from geneticin-resistant clones was amplified, PCR-purified (Invitrogen) and digested with T7 Endonuclease I (New England Biolabs) in order to assess indel formation at these loci. Given the clonal source of the post-CRISPR gDNA, any positive indel formation would be readily detectable at ~50%. Primers used to capture guide-specific off-target DSB sites are as follows. For Guide 3: 5’-TCCCCAGCTTCTGAATTCTTTT-3’ and 5’-TCCTCCATTCATGCAGCACTT-3’, 5’-AGCAAGAGTGATGGGAACGG-3’ and 5’-TTGCTCAGGTTCCAGTTTCCT-3’, 5’-CACAGCATCCCTTTGCACCA-3’ and 5’-GTGGTCAGGATAGCTGCCTTTA-3’. For Guide 13: 5’-GAGCAACATGGGTTATGCAG-3’ and 5’-GCAGCTATTTGAGCCCAGAG-3’, 5’-GACAGGGAGGACTGGAAGGAA-3’ and 5’-TCAAAGAGAGAGGGGAGCAAG-3’, 5’-CAGCCCATCCCCCATATTCC-3’ and 5’-AGCAATGTGAAGAGGAAAGTGC-3’. For Guide 2: 5’-GGAATTGGGACTGGGAGTGA-3’ and 5’-GAATGGGGCTGGGTTACGAT-3’, 5’-CCTGGCCAATATGGTGAAAC-3’ and 5’-GAATGCAACTCTTGAGCCAAC-3’, 5’-GGGCTAAGGGAACTTCTGGG-3’ and 5’-GGTTTGCCCTGAGACTCCAA-3’. For Guide 4: 5’-TGGGCAGAACAAATAAGAATCGG-3’ and 5’-AGCTGGTTGCACACATACAA-3’, 5’-ACATGTTAGGCCAAGGAGTAGAG-3’ and 5’-ATTCAAGTGTGCTTCAGTGCC-3’, 5’-TCCAATCACCTTTTCTGCCCAA-3’ and 5’-GAGAGTGCAGCACGGACATC-3’.

Population Doubling Level Experiments

MMR-deficient POLE^wt/wt, POLE^wt/S459F and MMR-proficient POLE^wt/wt, POLE^wt/S459F, POLE^S459F/S459F cell lines were seeded into T75 flasks and grown at 37°C/5% CO₂ until 80% confluency was reached. Cells were counted and 1 x 10⁶ cells were seeded into new T75 flasks and incubated until 80% confluency was reached. The above protocol was repeated at regular intervals (3-4 days) and population doubling level (PDL) was calculated using the following equation: PDL = [ln(N_t)-ln(N₀*PE)]/ln2. N_t = Number of viable cells counted after passage; N₀ = Number of cells seeded prior to passage; PE = plating efficiency. Genomic DNA was isolated from cells as described previously at PDL0 and again at 21.0, 63.8, 77.6 and 72.1 for POLE^wt/wt, POLE^wt/S459F MMR-deficient and POLE^wt/S459F & POLE^S459F/S459F MMR-proficient cell lines, respectively. The data for MMR-proficient POLE^wt/wt cells at PDL0 and PDL70 was taken from (Hodel et al., 2018).

Whole exome sequencing

Genomic DNA from the specified PDLs (above) were sent to Genewiz for whole exome sequencing (WES) performed on Illumina HiSeq 2000 (2x150 bp) with libraries prepared using Agilent SureSelect Human Exome Library Preparation V5 kit and sequenced to an average depth of 112.7 (104.9x-119.7x). Raw reads were aligned to the human genome (hg38) using the Burrows Wheeler Aligner (Li and Durbin, 2009). Variant calling was performed using Mutect 2 ((Cibulskis et al., 2013)) with the following additional filtration steps to retain only high confident mutations: VAF < 0.05, SNV per gene ≥ 3, dbSNPs were all removed.

MMR-proficient POLE^wt/S459F subclone WES analysis

Mismatch repair-proficeint POLE^wt/S459F cells from PDL₅₄ were obtained from the population doubling level experiment described previously. Cells were trypsinized, counted and cloned to single cells by limiting dilution into two 96-well plates at a density of 0.5 cell per well and incubated for 7 days. After wells visually validated to contain only a single well-defined colony, these subclones were expanded to 24-well plates, then to 6-well plates. No cells were discarded during this expansion. Four daughter clones were analyzed. Total RNA was extracted to confirm mutant POLE allele expression. Genomic DNA was extracted using TNES lysis buffer followed by phenol:chloroform extraction and ethanol precipitation, and finally resuspended in TE. Genomic DNA from these four daughter clones was prepared and then sent to BGI for whole exome sequencing (WES). Libraries were prepared with the Agilent SureSelect Human All Exon V6 kit and sequenced on the BGI DNBseq platform. Average sequencing depth was 154x. Quality control was performed on the raw FASTQ reads with FASTQC. The raw reads were aligned to the human genome version GRCh38/hg38 with the Burrows-Wheeler Aligner (Li and Durbin, 2009) and duplicates were removed with GATK4. Somatic variant calling was performed with VarScan2 Basic Protocol 1 (Koboldt et al., 2013) using the parental colony as a matched normal for each daughter clone. Bcftools (SAMtools/BCFtools, RRID:SCR_005227) and the R packages tidyverse and MutationalPatterns were used to analyze mutation data.

Non-negative matrix factorization and signature cosine similarity

VCF files from patients with POLE exonuclease domain mutations (n = 64) confirmed to by associated with a hypermutant TMB were downloaded from the GDC (https://portal.gdc.cancer.gov/). Variants for each sample were classified by mutation type (C>A, C>G & C>T or T>A, T>C & T>G) as well as 5’ and 3’ flanking bases immediately adjacent (+1 and −1) to the altered base and quantified as previously described. Mutation signatures were generated via non-negative matrix factorization (NMF) using the python package, nimfa (Zitnik and Zupan, 2012). The optimal number of signatures was determined via rank test (Fig. S2A). Briefly, NMF was run attempting to deconvolute the mutation matrix into anywhere between 2..40 signatures, with each factorization performed 1000 times. A corresponding cophenetic coefficient (signature stability) and RSS (error) were calculated for each set of factorizations, with the optimal number of signatures being the quantity with a maximum cophenetic coefficient and minimum RSS. Using the empirically determined optimal number of signatures, we performed NMF on our mutation matrix. Each signature was compared to version 2 of the COSMIC Mutational Signatures (https://cancer.sanger.ac.uk/cosmic/signatures_v2) via cosine similarity (Alexandrov et al., 2013b).

Mutant Frequency Measurements

Prior to the measuring of mutant frequencies, pre-existing HPRT1 mutants were purged from the population through 5 passages in the presence of HAT (1X Hypoxanthine-Aminopterin-Thymidine). For each MMR-deficient cell line, 2.5 x 10⁵ cells were seeded in 6-well plates in triplicate in media containing 6-TG (5 μg/mL). Concurrently, 100 cells were seeded in triplicate in the absence of 6-TG in order to assess plating efficiency (PE). After 10 days, PE plates were stained with crystal violet and counted. After 14 days, plates being selected with 6-TG were stained with crystal violet and counted. Mutant Frequency was calculated with the following equation: (# of 6-TG resistant colonies/(# of cells seeded x [# of colonies on PE/100])). Colonies were defined as ≥ 50 cells.

POLE variant transcript sequencing and analysis

Total RNA was extracted using Trizol (Invitrogen), and reverse transcription reactions were performed using the First-Strand cDNA Synthesis Kit (GE Healthcare Life Sciences). S459 or P286 locus was amplified using the following primers: 5'-TACAATGTCAGATACCGAGGAAATG-3' and 5'-CAATCTCCCTGTTGGTGATGAG-3' for S459; 5'-TTCCTGTGGGCAGTCATAATC-3' and 5'-AGCTTATTGAACTCCTGCTCTT-3' for P286. PCR products were then purified using the PureLink PCR Purification Kit (Invitrogen) and sequenced to high depth using Amplicon-EZ Sequencing (Genewiz). Sequencing data was uploaded to the Galaxy web platform and we conducted the following analysis with the public server at usegalaxy.org (Afgan et al., 2010): quality control was performed on all raw reads by FastQC (Wingett and Andrews, 2018), and bases with quality scores below 28 were trimmed using Trimmomatic (Bolger et al., 2014); the high-quality reads were then aligned to the human reference genome (hg38) by RNA STAR (Dobin et al., 2013) and visualized by IGV. Finally, normal spliced reads and mis-spliced reads were quantified with SAMtools (Li et al., 2009) and BEDtools (Quinlan and Hall, 2010).

Cell proliferation analysis

For each HCT116 clone, 2,500 cells were seeded in triplicates in 96 well plates. After the cells adhered, 4 images were taken per well for every 6 hours by IncuCyte S3 (Sartorius) (10x objective). Images were analyzed and percent confluency was measured using IncuCyte Software (Sartorius).

Cell cycle profiles and checkpoint activation

To determine the cell cycle profiles and checkpoint activation of the POLE variants, cells were plated in 6-well plates and cultured using standard growth conditions (10% FBS complete media 37°C, 5% CO2). For cell cycle analysis, at approximately 80% confluency BrdU was added at 30 μM final concentration for 1 hr. Cells were permeabilized with 70% ethanol for 1 hr and incubated with 2N HCl/0.5%Triton X-100 for 0.5 hr. Cells were stained with a-BrdU (mAb) for 1 hr, then goat a-mouse IgG-FITC for 0.5 hr. Finally, cells were incubated with RNase A and propidium iodide for 0.5 hr. FACS was performed immediately and data were analyzed using FlowJo.

Western Blot

Cells were lysed with radioimmunoprecipitation (RIPA) buffer (150mM NaCl, 50 mM Tris, 1% TritonX-100, 0.5 % deoxycholate, 0.1% SDS) supplemented with protease inhibitors at 4°C for 0.5 hr. Following centrifuge at 12,000 rpm for 20 min, the protein concentration of the supernatant was measured using Bio-Rad Protein Assay Dye Reagent Concentrate (Bio-Rad). Equal amounts (50μg) of the protein samples were electrophoresed on 4-20% linear gradient SDS-PAGE gels and transferred to PVDF membranes (Bio-Rad). The membranes were blocked with 5% skim milk in Tris-buffered saline containing 0.1% Tween 20 (TBST) for 1 hr at room temperature and incubated with indicated primary antibodies at 4°C overnight. The membranes were incubated with corresponding secondary antibody for 1 hr, washed with TBST, and incubated in western ECL substrate chemiluminescent detection reagent (ThermoFisher Scientific) for 5 min prior to image acquisition. The chemiluminescent blots were imaged with the ChemiDoc MP imager (Bio-Rad). The following antibodies were used in these assays: phospho-p53(Ser15) (Cell Signaling Technology), p53 (Santa Cruz Biotechnology), p21 (Abcam), PUMA (Santa Cruz Biotechnology) and MDM2 (Santa Cruz Biotechnology).

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical analyses were all performed using GraphPad Prism. Data are represented as mean and 95% CIs. Comparisons made between groups were performed with unpaired t test or ANOVA. P-values are indicated throughout (****, p<0.0001).

ADDITIONAL RESOURCES

Table S1. Related to STAR Methods. List of DNA oligos used for CRISPR-mediated gene editing, off-target analyses, and variant transcript sequencing.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Goat anti-mouse IgG	Sigma-Aldrich	Cat# F0257; RRID:AB_259378
Purified anti-BrdU	Biolegend	Cat# 317902; RRID:AB_604040
Phospho-p53 (Ser15) mouse	Cell Signaling Technology	Cat# 9286; RRID:AB_331741
Anti-p53 (DO-1)	Santa Cruz Biotechnology	Cat# SC-126; RRID:AB_628082
Anti-p21	Abcam	Cat# CP74; RRID:AB_1603138
Anti-MDM2 (SMP14)	Santa Cruz Biotechnology	Cat# SC-965; RRID:AB_627920
Anti-PUMAα/β (H-136)	Santa Cruz Biotechnology	Cat# SC-28226; RRID:AB_2064827
Bacterial and Virus Strains
Subcloning Efficiency DH5α Competent Cells	Invitrogen	Cat# 18265017
Ad-Cre-GFP Adenovirus, Cre Recombinase	Vector Biolabs	Cat# 1700
Biological Samples
N/A
Chemicals, Peptides, and Recombinant Proteins
UltraPure Phenol:Chloroform:Isoamyl Alcohol (25:24:1, v/v)	Invitrogen	Cat# 15593031
KpnI-HF	New England Biolabs	Cat# R3142S
SacI-HF	New England Biolabs	Cat# R3138S
EcoRI-HF	New England Biolabs	Cat# R3101S
NotI-HF	New England Biolabs	Cat# R3189S
XbaI	Invitrogen	Cat# IVGN0126
T7 Endonuclease I	New England Biolabs	Cat# M0302S
Lipofectamine 2000	Invitrogen	Cat# 11668030
Geneticin	Life Technologies	Cat# 10131027
HAT Supplement (50X)	Life Technologies	Cat# 21060017
6-Thioguanine	Sigma	Cat# A4882-250MG
Crystal Violet	ThermoFisher Scientific	Cat# C581-100
MEM Non-Essential Amino Acids Solution (100X)	Life Technologies	Cat# 11140050
0.05% Trypsin-EDTA	Life Technologies	Cat# 25300120
Trizol Reagent	Invitrogen	Cat# 15596018
DEPC-Treated Water	Invitrogen	Cat# AM9915G
5-Bromo-2′-deoxyuridine	Sigma	Cat# B5002
Propidium iodide	Sigma	Cat# P4170
SuperSignal™ West Pico PLUS Chemiluminescent Substrate	ThermoFisher Scientific	Cat# 34578
Critical Commercial Assays
SsoAdvanced Universal SYBR Green Supermix	Bio-Rad	Cat# 1725270
PureLink PCR Purification Kit	Invitrogen	Cat# K310002
GeneArt CRISPR Nuclease Vector with OFP Reporter Kit	Invitrogen	Cat# A21174
TOPO TA Cloning Kit for Subcloning	Invitrogen	Cat# 451641
First-Strand cDNA Synthesis Kit	GE Healthcare Life Sciences	Cat# 27926101
Bio-Rad Protein Assay Dye Reagent Concentrate	Bio-Rad	Cat# 5000006
Deposited Data
Genomic Data Commons Data Portal	National Cancer Institute GDC Data Portal	https://portal.gdc.cancer.gov/
Whole Exome Sequencing Raw Datasets	This paper	SRA accession number: PRJNA589337
Unprocessed Western blot images	Mendeley	doi: 10.17632/cdgs6fmpsn.2
Experimental Models: Cell Lines
Human: HCT116	Other	RRID: CVCL_0291
Human: HCT116+Mlh1	This paper (also Hodel et al 2018 eLife)	N/A
Human: HCT116 POLEwt/S459F	This paper	N/A
Human: HCT116 POLEwt/S459F+Mlh1	This paper	N/A
Human: HCT116 POLES459F/S459F+Mlh1	This paper	N/A
Human: HCT116 POLEwt/P286R+Mlh1	This paper	N/A
Experimental Models: Organisms/Strains
N/A
Oligonucleotides
POLE guide sequences, see Supp. Table 1		N/A
POLE guide off-target sequences, see Supp. Table 1		N/A
Primers for guide-specific off-target DBS sites, see Supp. Table 1		N/A
Primers for POLE variant transcript sequencing, see Supp. Table 1		N/A
Recombinant DNA
N/A
Software and Algorithms
Optimized CRISPR Design Tool	Zhang Lab	crispr.mit.edu
Image Lab Software	Bio Rad	https://www.bio-rad.com
Incucyte Software	Sartorius	https://www.essenbioscience.com
GraphPad Prism 8	GraphPad	https://www.graphpad.com
Burrows-Wheeler Aligner	Li et al. 2009	http://bio-bwa.sourceforge.net
Mutect 2	Cibulskis et al. 2013	http://www.broadinstitute.org/cancer/cga/mutect
Python package: NIMFA	Zitnik and Zupan. 2012	http://www.jmlr.org/papers/volume13/zitnik12a/zitnik12a.pdf
Galaxy	Afgan et al. 2016	http://usegalaxy.org
IGV	Robinson et al 2011	https://igv.org
SAMtools	Li et al. 2009	http://samtools.sourceforge.net
BEDtools	Quinlan et al. 2010	https://github.com/arq5x/bedtools2
VarScan 2	Koboldt et al. 2012	http://varscan.sourceforge.net
FASTQC	Babraham Bioinformatics	https://www.bioinformatics.babraham.ac.uk/projects/fastqc/
GATK4	Auwera et al. 2013	https://gatk.broadinstitute.org/hc/en-us
BCFtools	Li 2011	https://www.htslib.org
R package: Tidyverse	Wickham et al. 2019	https://tidyverse.tidyverse.org/index.html
R package: MutationalPatterns	Blokzijl et al. 2018	https://bioconductor.org/packages/release/bioc/html/MutationalPatterns.html
FlowJo v10.6.2	FlowJo	https://www.flowjo.com
Other
N/A

Highlights.

• POLE cancer variants are sufficient to drive signature mutation accumulation in cells

• Signature mutations are made even in the presence of functional mismatch repair

• Mismatch repair plays a critical role in shaping the ultimate mutation spectrum

• POLE tumor subgroup classifications are made from relative signature mutation amounts