Enhanced gene targeting to evaluate Lynch syndrome alterations

The identification of gene alterations that predispose individuals to disease has resulted in an explosion in genetic diagnostics. Because diseases like cancer susceptibility may result in debilitating surgery that has physical, emotional, and realized costs, it is important to get the diagnosis right. The obvious solution is to combine simple DNA genotype analysis with evocative phenotype analysis (1). However, a simple phenotypic analysis for the complex progressions of cancer that often includes dozens of attendant mutations (2) seemed largely destined to fail. Enter Lynch syndrome, a dominant predisposition to colorectal, endometrial, ovarian, and upper urinary tract cancers that is caused by heterozygous germ-line mutations in the mismatch repair (MMR) system (3). MMR recognizes and repairs mismatched nucleotides that principally arise from polymerase misincorporation during replication (4). When MMR is defective, spontaneous mutation rates dramatically increase, accelerating the evolution of tumors (mutator hypothesis) (5). The simplicity of Lynch syndrome cause-and-effect belies the enigma of genetic diagnostics, which has uncovered abundant missense variants of unknown significance (VUS) that pepper the entire coding regions of the core human MMR genes (Fig. 1). In PNAS, two papers provide useful solutions to the genotype–phenotype problem of Lynch syndrome diagnostics. In the first, van Ravesteyn et al. (6) developed an improved rapid gene-targeting methodology for mouse embryonic stem cells that then allowed Houlleberghs et al. (7) to advance functional analysis of Lynch syndrome VUS in mammalian cells.

Fig. 1.

The location of MSH2 variants of unknown significance on the MSH2–MSH6 protein structure. The MSH2 class 3 VUS from the InSIGHT Colon Cancer Gene Variant Database (9) were mapped to their location (green dots) on the MSH2 (purple)–MSH6 (blue) structure bound to a DNA mismatch (22).

MMR is a relatively simple excision–resynthesis process that requires MutS homologs (MSH) to recognize the mismatch and transmit this recognition along the DNA to MutL homologs (MLH/PMS) that ultimately activate excision (4). The replication polymerase complex then completes resynthesis of the resulting ssDNA gap. In human cells, the MSH2–MSH6 heterodimer is responsible for mismatch recognition and the MLH1–PMS2 heterodimer for downstream excision activation (4). Mutation of these core MMR genes MSH2 (33%), MSH6 (18%), MLH1 (42%), and PMS2 (7%) accounts for the vast majority of Lynch syndrome cases, with most of the alterations being nonsense or frameshift mutations with obviously defective function (8).

Because genetic diagnostics have become commonplace and Lynch syndrome is relatively frequent in the human population, the numbers of VUS have become significant. As a consequence, members of the International Society for Gastrointestinal Hereditary Tumors (InSiGHT) proposed guidelines to classify the disease significance of MMR gene variants (9). A five-tiered system was developed that assigned variants into a significance category based on multiple lines of evidence including population genetics, variant segregation data, disease phenotype, and any available functional analysis.

The functional analysis of MMR missense alterations has taken two tacks (10). The first was biochemical analysis of known protein functions and/or interactions. For example, when MSH2–MSH6 heterodimer recognizes a mismatch, it binds ATP, resulting in the formation of a stable sliding clamp on the DNA (11). Thus, examination of mismatch-dependent ATP activation or heterodimer subunit protein–protein interactions were relatively simple early assays (10). Once the complete human MMR reaction was reconstituted, it became the method of choice for examining the effect of MMR protein variants (10). Although valuable screens, it was always possible that VUS activities in vitro might not fully reflect the functions important for tumorigenesis.

A second tack was to develop cellular assays. One of the first gauges of a cellular MMR defect was the insertion or deletion of nucleotides in simple repeat (microsatellite) sequences [termed microsatellite instability (MSI)] (12, 13). A number of simplified reporter constructs have been developed that place an out-of-frame microsatellite within a gene-coding region that with MSI results in a clear phenotypic expression. In addition, nearly 35 y ago, Karran and Marinus (14) showed that MMR-deficient bacteria were resistant to chemical methylation. A similar methylation resistance was found with MMR-defective human cells (15), which was extended to cancer chemotherapeutic drug resistance such as 6-thioguanine (6-TG) (16). Therapeutic drug resistance ultimately linked MMR to the cellular DNA damage response that appears to be an extremely important function because, among other things, its timely loss likely provides the selective advantage for inactivating of the remaining wild-type MMR allele that drives Lynch syndrome tumorigenesis (17).

Because the MMR system is highly conserved, the development of cell-based functional analysis began in yeast where gene targeting and repair assays are straightforward. Of course, the problem is that yeast do not get cancer. In a previous iteration, the te Riele group (18) showed that they could introduce single-nucleotide gene alterations into MMR-deficient (Msh2^−/−) mouse embryonic stem (mES) cells with a frequency of ∼0.001% using single-stranded DNA oligonucleotides (ssODNs). The frequency of mutations in wild-type mES cell was at least 100-fold less efficient (below detection), suggesting that mutagenesis likely relies on unusual ssODN-directed replication priming that escapes detection in the MMR-deficient cells. The simplicity and efficiency of ssODN mutagenesis was interesting but not particularly useful for examining Lynch syndrome VUS in the already MMR-defective cells.

van Ravesteyn et al. (6) introduced locked nucleic acids (LNAs) into the ssODN mutagenesis scheme. LNAs modify the sugar with a 2′-C,4′-C-oxymethylene linker that effectively confines the ribose ring to the 3′-endo conformation reducing the mobility of the LNA–nucleotide with respect to its nearest neighbors. Because the structures of MSH proteins invariably showed a bend at the site of the bound mismatch, the idea was to impede this recognition with an inflexible LNA that would effectively make ssODN invisible to the MMR machinery. The results were extraordinary. Using the reconstitution of a neomycin start site as a reporter, van Ravesteyn et al. show that the frequency of ssODN mutagenesis with an LNA at the targeted nucleotide was identical in Msh2^−/− and wild-type mES cells. As expected, the Escherichia coli MutS did not bind mismatched LNA oligonucleotides, although the more revealing mismatch-dependent ATP activation was not tested (11). van Ravesteyn et al. (6) then determined that the sweet spot for LNA location was the nucleotides immediately adjacent to the mismatch: a result consistent with previous studies that showed MSH ATP activation was dependent on flexibility of adjacent 5′-nucleotides and not nucleotide stacking behavior (19). Including a 5′-acridine to increase oligonucleotide-target annealing, a donor length of 25 nt and at least one LNA 5′ of the mismatch ultimately increased the mutagenesis frequency in wild-type mES cells to 0.35% (of seeded cells). Such a targeting efficiency clearly begins to rival yeast.

The results of van Ravesteyn et al. (6) set the stage for the utilization of LNA–ssODN mutagenesis to examine Msh2 VUS by Houlleberghs et al. (7). These studies used heterozygous Msh2^+Pur/Δ mES cells that contained a single allele of Msh2 with a closely linked

van Ravesteyn et al. developed an improved rapid gene-targeting methodology for mouse embryonic stem cells that then allowed Houlleberghs et al. to advance functional analysis of Lynch syndrome VUS in mammalian cells.

puromycin resistance gene that could be monitored to assure that the ssODN mutagenesis scheme introduced the variant alteration alone without a potentially confounding gene deletion resulting in loss of heterozygosity (LOH). As a reduction to practice, they examined 10 pathogenic and 10 nonpathogenic MSH2 variants. LNA–ssODN variants were transfected into Msh2^+Pur/Δ mES cells followed by 6-TG drug selection (termed method 1). Resistant cell clones were sequenced to determine whether the targeted variant was present. They found that ∼70% of the 6-TG–resistant cell clones contained the pathogenic ssODN variant (ranging from 33% to 100%; n = 104, 5–15 cell clones each), whereas none of the resistant clones included a nonpathogenic ssODN variant (n = 77, 3–13 cell clones each). Two partially pathogenic variants that were not detected by method 1 were found in 25% and 80% of resistant clones when the 6-TG selection concentration was significantly reduced and the LOH puromycin selection was included (n = 4 and 20, respectively; termed method 2). These observations suggested that, although 6-TG drug selection may result in resistant cell clones by a variety of mechanisms, the improved LNA–ssODN targeting efficiency appeared to statistically enrich for pathogeneic MMR variations when present in the population.

Houlleberghs et al. (7) then used method 2 to screen 19 VUS. In addition to the initial 6-TG selection/sequencing screen, single variant-containing cell clones from each of the 19 VUS were examined for the range of 6-TG resistance as well as protein stability by Western analysis and MSI rate with an out-of-frame (G₁₀)-neo system integrated into the Rosa26 locus. The 6-TG survival of VUS clones ranged from complete to intermediate resistance, compared with the known 6-TG–resistant Msh2^P622L/Δ pathogenic variant and the sensitive parent Msh2^+Pur/Δ mES cells. High (H) MSI almost always correlated with complete 6-TG resistance, whereas medium (M) MSI correlated with intermediate 6-TG resistance (see below). Most of the 19 VUS displayed very low expression of the variant protein (1–4%) compared with 60% of wild-type levels displayed by the parent cells: a surprising protein instability effect that should be studied further. There were notable inconsistencies including the Msh2(G674A) variant that has been reported to dissociate the DNA damage response from MMR activity (20). This variant was expressed at 33% of wild-type levels, displayed intermediate 6-TG resistance, and yet it was MSI-H. Although dissociation of the DNA damage response and MMR appears rare, such a phenotype could complicate the functional analysis of VUS.

In the end, Houlleberghs et al. (7) were able to classify all 19 VUS as displaying substantial or at least highly suspected pathogenic significance. They conclude that the LNA–ssODN system with the Msh2^+Pur/Δ mES cells could be incorporated into diagnostic laboratories as a functional analysis of MMR VUS. Because most diagnostic laboratories are fundamentally DNA sequence based, this seems like a bit of a stretch. To incorporate such an analysis, the laboratory would have to be fully competent at manipulating and maintaining mES cells. That said, the Msh2 mES cell functional assay appears to be 97.6% specific and 92.3% sensitive, which is a significant improvement over existing cellular methods. The specificity and sensitivity have yet to be determined for the other Lynch syndrome MMR genes.

Finally, it is not clear whether the LNA–ssODN methodology will be useful for any other disease or cancer susceptibility syndrome. MMR defects are unique in that they result in easily selectable resistance to DNA damaging agents as well as MSI indicative of replication-associated MMR defects. To our knowledge, no other disease or cancer susceptibility has similar positive selection and comparative phenotypes. If disease development becomes the functional endpoint for assessing a VUS, then targeting mES cells where LNA–ssODN works best may not be the best system. Recent advances in synthetic human tissues derived from induced pluripotent cells seem more likely to become relevant systems for future disease functional analysis, in which case clustered regularly interspaced short–palindromic-repeat (CRISPR) ribonucleoprotein particle targeting that in worms can approach 10–80% will likely become the genome-editing standard (21).