Variants in mismatch repair genes are overrepresented in clade A isolates with TR34/L98H
Eukaryotic MMR consists of two major recognition complexes: MutSɑ (Msh2-Msh6), which recognises base–base mismatches and small loops, and MutSβ (Msh2-Msh3), which recognises larger loops, with a bias towards deletion loops34. The MutLɑ protein complex (Pms1-Mlh1) directs downstream protein–protein interactions, is required for daughter strand recognition, and has endonuclease activity. We screened 218 previously sequenced A. fumigatus isolates20 65 originating from environmental and 153 from clinical sources in the United Kingdom (Supplemental dataset 1), of which 91 contain TR34/L98H and 7 contain TR46/Y121F/T289A cyp51A azole resistance mutations, for variants in MMR genes msh2 (AFUB_039320, AFUA_3G09850), msh3 (AFUB_090020, AFUA_7G04480), msh6 (AFUB_065410, AFUA_4G08300), pms1 (AFUB_029050, AFUA_2G13410) and mlh1 (AFUB_059270, AFUA_5G11700). In total, across all five genes, 212 non-synonymous point mutations were present relative to the Af293 reference strain (Supplemental dataset 1). No frameshift, nonsense, or truncation mutations were observed in any of the isolates; thus, variants were expected to alter rather than abolish the activity of the MMR systems. Of these variants, non-synonymous mutations in msh2 (c.A2435G, p.E812G), msh6 (c.G698C, p.G233A) and pms1 (c.A1331G,p.E444G) were significantly associated with clade A (Fig. 1a–c, msh2: d.f. = 1, χ2 = 8.88, P = 0.0029, msh6: d.f. = 1, χ2 = 141.64, P < 2.2 × 10−16, pms1: d.f. = 1, χ2 = 19.12, P = 1.2 × 10−5); however, only the G233A variant in msh6, which occurs prior to the annotated N-terminal MutS domain responsible for mismatch recognition (Fig. 1e), was significantly associated with the presence of the azole resistance mutation TR34/L98H in cyp51A (d.f. = 1, χ2 = 122.27, P < 2.2 × 10−16). With the exception of the variants we have identified within clade A of A. fumigatus, the G233 amino acid is perfectly conserved across over 150 million years of evolutionary history35 within the Trichocomaceae family (Fig. 1f). In total, 85% (105/123) of clade A isolates contained the G233A variant allele of msh6, while the variant is only present in 3% of isolates in clade B (3/95). Of the 86 TR34/L98H azole-resistant genotypes, 96.5% contained G233A, whereas. of the 7 isolates containing the TR46/Y121F/T289A resistance haplotype, only 4 harbour the msh6-G233A variant (Fig. 1d). Notably, msh6 is encoded on chromosome 4 0.36 Mbs away from the cyp51A gene, however, previous population genomic studies have shown that genetic linkage can decay rapidly even within the resistant cluster21. The cyp51A gene has a high average fixation index (FST) of 0.1127 (standard error (se) = 0.025) between isolates within clades A and B, implying population subdivision at this locus (average genome-wide FST = 0.086, se = 0.00026). In comparison, msh6 also has a high average FST of 0.1386 (se = 0.0422). A two-tailed t-test assuming unequal variances between chromosome 4 (where both msh6 and cyp51A are located) and the whole of the genome recovers a significant p-value of 1.367e−67, suggesting multiple loci across chromosome 4 are associated with azole drug resistance. In addition, previous analysis shows this region has a significant association with itraconazole resistance20 (treeWAS P < 0.001). The association between G233A, clade A and TR34/L98H was also evident in a global collection of isolates36 (Fig. S1, Clade A: d.f. = 2, χ2 = 463.93, P < 2.2 × 10−16, TR34/L98H: d.f. = 1, χ2 = 326.23, P < 2.2 × 10−16). Thus, the presence of the non-synonymous variant G233A in msh6, which is an essential component of MutSɑ, responsible for recognising base–base mispairing, is strongly associated with the presence of azole resistance allele TR34/L98H in clade A.
Occurrence of MMR variant alleles in clade A (red) and clade B (blue) for a msh2 (E812G), b msh6 (G233A) and c pms1 (E444G). d An unrooted maximum-likelihood phylogenetic tree using genome-wide SNPs relative to Af293 of 218 WGS UK isolates20. The presence of variants in msh2, msh6 and pms1 are highlighted in the black boxes, the presence of cyp51A resistance variants and clade are coloured. e Domain structures of Msh6 in A. fumigatus, G233A variant labelled, labels show predicted domain positions in protein sequence. f G233 locus homology across Trichocomaceae, alignments of 140 isolates spanning Aspergillus ssp., 27 Talaromyces and Paecilomyces, and 122 Penicillium Msh6 protein sequences. G233 highlighted in yellow, G233A variants are only present in Aspergillus fumigatus. Cladogram shows the hierarchical clustering of Msh6 protein sequences. Source data are provided as a Source Data file.
MutS and MutL null mutants result in a hypermutator phenotype
Given variant alleles in msh2, msh6 and pms1 are significantly associated with clade A, and in the case of msh6 TR34/L98H azole resistance genotypes, we first asked whether each of the three genes influence mutation rate. The three genes were independently deleted from the wild-type strain MFIG001, a laboratory strain that clusters within clade B37. The minimal inhibitory concentrations to voriconazole, a current antifungal used to treat A. fumigatus infections, or the phase III clinical trial compound olorofim in the orotomide class38 were not altered in the MMR defective strains relative to the parental MFIG001 strain (Fig. S2) indicating no direct effect of these alleles on azole or orotomide sensitivity. To measure mutation rates, a modified Luria–Delbrück fluctuation test39 was implemented in which mononucleated spores from replicate cultures grown without selection were challenged with voriconazole to determine the probability that spores would spontaneously gain mutations that provide resistance (see the “Methods” section and Fig. 2a). The rate of spontaneous mutation in the wild-type MFIG001 strain to voriconazole was 2.78 × 10−10 (±6.9 × 10−11) per spore, similar to rates measured in other fungal species40,41,42. The modified Luria–Delbrück method was validated by treatments with the mutagen ethyl methanesulfonate during growth. The method detected the linear increase in mutation rate in MFIG001 to voriconazole with increasing concentrations of the mutagen (linear regression, R2 = 0.92, F1,13 = 153, P < 0.001, Fig. S3). Mutation rates for voriconazole resistance in the MMR deletion strains Δmsh2, Δmsh6 and Δpms1 were ~85-, ~47 and ~173-fold higher than the parental wild-type strain (Fig. 2b, Two sample ML-test, MFIG001-Δmsh2 T = −3.83, P < 0.001, MFIG001-Δmsh6 T = −3.82, P < 0.001, MFIG001-Δpms1 T = −4.12, P < 0.0001). The MICs of spontaneous resistant mutants to voriconazole ranged from 4 to 32 µg/ml and were not dependent upon the genetic background (Kruskal–Wallis, d.f. = 2, χ2 = 4.25, P = 0.119). Sequencing of the cyp51A gene showed that 37% (10/27) of the randomly selected spontaneously resistant isolates had the known voriconazole resistance allelic variant G448S (4/8 MFIG001, 2/7 Δmsh2, 3/6 Δmsh6, 1/6 Δpms1), but these exclusively occurred within isolates with MICs ≥ 16 µg/ml. No mutations within cyp51A were observed in the other resistant isolates and tandem repeats in the promotor of cyp51A were never observed.
a Workflow of fluctuation tests to measure mutation rates in A. fumigatus. Clonal isolates were cultured in the absence of antifungal selection to generate genetic diversity. Resistant mutants were selected on lethal concentrations of antifungals. Counts of resistant mutants were fitted to the Luria–Delbrück distribution to calculate the number of mutational events. b Mutation rates for resistance to voriconazole for MMR-deficient mutants. Each point shows the calculated mutation rate from a single independent fluctuation test using 12 replicate cultures. Error bars show 95% confidence intervals, cross bars show the median mutation rate across fluctuation tests. Fold differences show median fold change in mutation rate from the parental MFIG001 strain. Triangle points represent mutation rates measured using independently constructed deletion mutants. c Mutational frequency to olorofim resistance. d Mutational frequency to itraconazole resistance. Each point shows the mutational frequency of an individual population (N = 6). Error bars show SEM and cross bars show median mutational frequency. Fold differences show median fold change in mutation frequency from the parental MFIG001 strain. Source data are provided as a Source Data file.
Whole genome sequencing (WGS) of a further five randomly selected spontaneous voriconazole-resistant isolates from the wild-type MFIG001 strain revealed that only one mutational event was detectable in the entire genome of each resistant strain, all generating a G448S variant in cyp51A. In contrast, sequencing of five resistant isolates derived from the msh2 null mutant and 6 from the msh6 and pms1 null mutants resulted in 28 (s.e. 4), 48 (s.e. 6), and 35 (s.e. 3.2) mutations (synonymous, non-synonymous and intergenic) per genome respectively (Fig. S4a), including canonical azole resistance mutations cyp51A G448S (2/5 Δmsh2, 3/6 Δmsh6, 2/6 Δpms1) and HMG-CoA s in 2/5 Δmsh2 resistant mutants (Supplemental dataset 2). The deletion of msh2 resulted in higher frequencies of transversions relative to the deletion of msh6 or pms1 (Fig. S4b, Tukey multiple comparisons, Δmsh6-Δmsh2 P < 0.05, Δpms1-Δmsh2 P < 0.05) mirroring previously published results showing that the loss of function of Msh2 results in a bias towards transversions43. The specific base substitutions showed that the bias towards transversions in the Δmsh2 strain was due to an over-representation of C>A mutations (Fig. S5a), however there were no clear trinucleotide signatures that were associated with the transversions (Fig. S5b). The proportion of intergenic mutations also differed significantly between deletion mutants (ANOVA, F2,13 = 33.37, P < 0.001, Fig. S4c). While the deletion of msh6 resulted in a similar ratio of intergenic mutations (Δmsh6 48.4% intergenic, s.e. 2.23) to the proportion of intergenic regions in the A. fumigatus genome (48.76%), the Δpms1 and Δmsh2 mutations were overrepresented by intergenic mutations (Δpms1 76.2% intergenic, s.e. 2.45, Δmsh2 59.8% intergenic, s.e. 2.68) suggesting that the elevated mutation rates within these strains result in deleterious or lethal intragenic mutations which are purged by negative selection, and suggesting that a high fitness costs may be associated with the deletion of these genes. Mutations in msh2 have previously been associated with elevated mutation rates and the acquisition of antifungal resistance in Cryptococcus deuterogattii30 with an overrepresentation of mutation occurring with homopolymer nucleotide runs29. WGS data from the MMR null mutants showed that while single nucleotide variants did not occur within homopolymer runs (Fig. S6a), single base-pair indels were strongly associated with homopolymer nucleotide runs (Fig. S6b), with indel events occurring in homopolymer runs with a mean length of 9 base pairs. However, indels were not clearly associated with resistance, with no mutations showing parallelism between independent resistant mutants. Moreover, only 11% of indel events occurred within protein-coding regions, despite 49% of the A. fumigatus A1163 genome being protein-coding.
The frequency of resistance to another azole-class antifungal, itraconazole and the dihydroorotate dehydrogenase (DHODH) inhibitor olorofim showed similar significant increases in mutation rate in all three MMR deficient strains, with the largest increases in Δpms1 and the lowest increases in mutation rate in Δmsh6 (pairwise Wilcoxon tests, P < 0.05, Fig. 2c, d). The probability of resistance arising differed between antifungals (ANOVA, F2,225 = 42.49, P < 0.001), with resistance arising between 2 and 15 times more frequently to itraconazole than olorofim or voriconazole (Tukey post hoc test, P < 0.001), however, there was no difference between the frequency of resistant mutants to voriconazole and olorofim (Tukey post hoc test, P = 0.828). Of the randomly selected olorofim spontaneous resistant mutants, 28/28 had mutations in the pyrE resistance hotspot G11944,45, which provided high levels of olorofim resistance (>2 µg/ml). Together, these results show that msh2 and pms1 null mutants, which abolish the activity of the MMR system, result in highly elevated mutation rates that facilitate the emergence of resistance. Moreover, although MutSβ recognition complexes remain functional when disrupting msh634, the loss of function of msh6 still results in significant increases in mutation rate.
Defective MMR results in significant reduction in fitness
Uncontrolled mutation can result in the accumulation of deleterious mutations, which decrease the fitness of hypermutator strains within stable environments46,47. We therefore asked whether such costs were associated with the MMR defects in A. fumigatus. Though we detected non-synonymous variants in msh2, msh6 and pms1 no predicted loss of function mutations within the MMR genes were observed in the clinical or environmental isolates sequenced, suggesting that the complete loss of MMR could be associated with a significant cost in A. fumigatus. However, no defect in radial growth rates was observed over 96 h in the MMR deletion mutants relative to their parental strain in either nutrient-rich or minimal growth conditions (Fig. S7). Interestingly, morphological sectoring, likely to occur due to mutations during hyphal growth, occurred within the MMR mutants but not the parental strain. To determine whether fitness costs manifested over longer periods of growth, MMR-deficient mutants directly competed with the parental strain over five serial transfers on complete (rich) and minimal solid media (Fig. 3) in the absence of antifungal selection. The frequency of MMR deleted strains decreased through time and was dependent upon both the MMR mutant and the environment (Mixed effects linear model, Transfer:Media F1 = 7.8447, P < 0.01, Transfer:Strain F2 = 7.8528, P < 0.001, Fig. 3a). Growth medium had a significant effect on the relative fitness in all MMR deficient strains, with fitness being consistently lower in minimal media compared to rich media (Wilcoxon test, W = 233, P < 0.05, Fig. 3b). Deletion of msh2 resulted in the highest overall cost relative to the parental strain, displaying 30% cost in rich media (T.test, T5 = −13.6, P < 0.001, Holm adjusted for multiple testing) and a 75% cost in minimal media (T.test, T5 = −30, P < 0.001, Holm adjusted for multiple testing, Fig. 3b). Deletion of pms1 also resulted in significant cost (Rich: 13% fitness cost, T.test, T5 = −4.9, P < 0.05, Minimal: 25% fitness cost, T.test, T5 = −15.5, P < 0.001, Holm adjusted for multiple testing), although lower than Δmsh2 (two sample t.test, T9.68 = −4.8, P < 0.0001). In contrast, the msh6 null mutant did not have a significant decrease in relative fitness measured at transfer 5 when competed in either rich or minimal media (Fig. 3), however, the fraction of Δmsh6 had reduced significantly by transfer 4 when competed in minimal media (Wilcoxon test, W = 0, P < 0.01). Thus, over longer periods of time there are significant costs associated with the loss of function of MutS or MutL complex.
a The fraction of Δmsh2, Δmsh6 and Δpms1 through time when in direct competition with the parental MFIG001 strain on solid agar faceted by media type (rich = aspergillus complete media, minimal = aspergillus minimal media). The coloured lines show the mean of 6 independent competitions, presented by individual grey lines. The horizontal dashed lines show the starting fraction of the MMR deletion strain. b Mean fitness of Δmsh2, Δmsh6 and Δpms1 relative to MFIG001 across the five transfers presented in panel a, the cross bar shows the median (N = 6 independent competitions), the lower and upper hinges correspond to the first and third quartiles and the whiskers extend to 1.5*IQR. Box plots coloured by media type. The horizontal dashed line shows equal fitness of zero, p-values show significant difference from zero using two-sided t-tests, using Holm correction for multiple testing, Δmsh2 rich P = 0.00023, minimal P = 4.8e−6, Δmsh6 rich P = 1, minimal P = 0.97, Δpms1 rich P = 0.025, minimal P = 0.00012, asterisks represent significance (*p < 0.05, **p < 0.01, ***p < 0.0001). Source data are provided as a Source Data file.
Msh6 G233A increases mutation rate and is correlated with increased mutation rates in clade A but not a fitness cost
Since we find deletion of MMR components to be costly, as well as increasing rates of anti-fungal resistance, we asked whether the more subtle variants we see in our strain collection have a similar effect. As we did not have access to the rare clade B strains harbouring the msh6-G233A variant, we reconstructed the G233A variant of msh6 within MFIG001 (which clusters within clade B) through marker-less CRISPR-Cas9 mediated transformation48 to determine the effect of the variant on mutation rate. The msh6 variant resulted in a modest but significant increase in rates of resistance mutation to olorofim of 3.6-fold relative to the isogenic parental strain in three independent transformants (Fig. 4a, two sample ML-test, MFIG001 vs. msh6-G233A T = −2.9576, P < 0.01). A similar increase in mutation rate in the MFIG001 vs. msh6-G233A relative to its isogenic parental strain was observed when voriconazole was used as the selective marker (two sample ML-test, MFIG001 vs msh6-G233A T = −3.2727, P < 0.01, Fig. S8). Moreover, the increase in mutation rate was not associated with a significant increase in cost in either rich or minimal media (Fig. 4b, one sample T.test mu = 0, ACM T5 = −0.60839, P = 1, AMM T5 = −1.8148, P = 0.51, Holm adjusted for multiple testing). We, therefore, hypothesise that this variant is associated with variation in mutation rates in natural isolates. To test this, we assayed the rate of mutation to olorofim resistance in 18 isolates with a range of variants in msh2, msh6 and pms1, with a combination of cyp51A resistance variants including wild-type, TR34 and TR46from clade A and clade B (Table 1). Only the novel antifungal olorofim could be used for these fluctuation tests, as some isolates already had mutations in cyp51A that conferred resistance to azoles. The olorofim MICs of the isolates were not significantly different between isolates, and all fell at least 4-fold below the concentration used to select for resistant mutants (Fig. S9), enabling direct comparisons of mutation rates to be made. The isolates from clade B without variants in the MMR genes did not have significantly different mutation rates compared to MFIG001, including a clade B isolates with cyp51A TR46/Y121F/T289A or TR34/L98H azole resistance mutations (Fig. 4c, Two sample ML-test, MFIG001-clade B isolates, all P > 0.05). In contrast, isolates from clade A with the msh6-G233A variant had between a 1.3- and 5.1-fold increase in mutation rate (median 4.6-fold), significantly increasing the likelihood of spontaneous olorofim resistance arising (Fig. 4d, Two sample ML-test, MFIG001 vs msh6-G233A isolates, P < 0.05 in 7/8 strains). The mean difference in mutation rate to olorofim resistance between the msh6-G233A and msh6-WT populations was 1.22e−08 (T-test, T8.82 = 5.76, P > 0.001). These isolates encompassed both TR34/L98H and TR46/Y121F/T289 cyp51A resistance genotypes and isolates with clinical and environmental origins (Table 1). However, two clade A isolates, C89 without msh6-G233A, and C21 with msh6-G233A, did not display similarly elevated mutation rates. Previous studies44,45, together with Sanger sequencing 28 spontaneous olorofim-resistant mutants, show that olorofim resistance is mediated by a single point mutation in the drugs’ target site; therefore, it is unlikely that the mutational target size for resistance is different between the two clades. Reverting the msh6-G233A allele to its WT form in a clade A isolate (C6) also resulted in a halving of mutation rate in three independent marker-less transformants (two sample ML-test, P < 0.001, Fig. S10). These results show that the presence of the G233A variant in msh6, unique to clade A, is associated with elevated rates of mutation to resist a novel antifungal in natural isolates. Moreover, this elevated mutation rate was not influenced by the genotype of the linked azole resistance locus.
a Each point shows the calculated mutation rate from a single independent fluctuation test using 12 replicate cultures. Each of the three points for the MFIG001 msh6-G233A variant is a separate independent transformant. Error bars show 95% confidence intervals. Fold change shows the median fold change in mutation rate from the parental MFIG001 strain. b Mean fitness of msh6-G233A relative to MFIG001 across five transfers, the cross bar shows the median (three independent transformants, to replicates each for a total of N = 6), the lower and upper hinges correspond to the first and third quartiles and the whiskers extend to 1.5*IQR. Box plots coloured by media type. The horizontal dashed line shows equal fitness. c Mutation rate of natural genotypes to olorofim. Each point shows the calculated mutation rate from a single independent fluctuation test using 12 replicate cultures. Error bars show 95% confidence intervals. Fold difference shows the median fold change in mutation rate of isolates with G233A allele from clade B isolates. Points coloured by clade, red clade A, blue clade B. Key below plot shows the presence of azole resistance mutation cyp51A TR34/L98H in orange, TR46/Y121F/T289A in blue, and the presence of msh6-G233A in grey. Source data are provided as a Source Data file.