Mutations in the gdpP gene are a clinically relevant mechanism for β-lactam resistance in meticillin-resistant Staphylococcus aureus lacking mec determinants

In Staphylococcus aureus , resistance to β-lactamase stable β-lactam antibiotics is mediated by the penicillinbinding protein 2a, encoded by mecA or by its homologues mecB or mecC. However, a substantial number of meticillin-resistant isolates lack known mec genes and, thus, are called meticillin resistant lacking mec (MRLM). This study aims to identify the genetic mechanisms underlying the MRLM phenotype. A total of 141 MRLM isolates and 142 meticillin-susceptible controls were included in this study. Oxacillin and cefoxitin minimum inhibitory concentrations were determined by broth microdilution and the presence of mec genes was excluded by PCR. Comparative genomics and a genome-wide association study (GWAS) approach were applied to identify genetic polymorphisms associated with the MRLM phenotype. The potential impact of such mutations on the expression of PBP4, as well as on cell morphology and biofilm formation, was investigated. GWAS revealed that mutations in gdpP were significantly associated with the MRLM phenotype. GdpP is a phosphodiesterase enzyme involved in the degradation of the second messenger cyclic-di-AMP in S. aureus . A total of 131 MRLM isolates carried truncations, insertions or deletions as well as amino acid substitutions, mainly located in the functional DHH-domain of GdpP. We experimentally verified the contribution of these gdpP mutations to the MRLM phenotype by heterologous complementation experiments. The mutations in gdpP had no effect on transcription levels of pbp4; however, cell sizes of MRLM strains were reduced. The impact on biofilm formation was highly strain dependent. We report mutations in gdpP as a clinically relevant mechanism for β-lactam resistance in MRLM isolates. This observation is of particular clinical relevance, since MRLM are easily misclassified as MSSA (meticillin-susceptible S. aureus ), which may lead to unnoticed spread of β-lactam-resistant isolates and subsequent treatment failure.


INTRODUCTION
β-Lactam antibiotics, in particular isoxazolyl penicillins and first-and second-generation cephalosporins, are still the first choice for the treatment of staphylococcal infections. Their effect is based on allosteric binding of the β-lactam ring to native penicillin-binding proteins (PBPs). This results in the formation of an acyl-enzyme complex that prevents cross-linking of the peptidoglycan in the bacterial cell wall. This disrupts cell wall integrity, ultimately resulting in cell death. Meticillin-resistant Staphylococcus aureus (MRSA) usually carries an additional PBP, PBP2a, encoded by mecA or rarely by its homologues mecB or OPEN ACCESS mecC. It is characterized by reduced affinity to most β-lactams; thus, allowing cross-linking in the presence of these antibiotics [1][2][3]. However, meticillin-resistant lacking mec (MRLM) isolates have been described, but only limited data is available on their prevalence, characteristics and underlying resistance mechanisms [4][5][6][7][8]. Although these isolates seem to be rare at present, they nevertheless pose a serious threat to public health and therapeutics as they are easily misclassified as meticillin-susceptible S. aureus (MSSA) based on the exclusive detection of mec genes or PBP2a [9][10][11][12][13][14].
The MRLM phenotype has been previously associated with (i) a hyperproduction of blaZ-encoded β-lactamase (BORSA phenotype) [15], (ii) reduced affinity of native PBPs to β-lactams due to pbp mutations (MODSA phenotype) [16][17][18][19] or (iii) increased peptidoglycan cross-linking due to mutations in the pbp4 promoter resulting in overexpression of PBP4 [20]. Additionally, it has been shown that loss-of-function mutations in gdpP, a gene involved in the metabolism of the second messenger cyclic-di-AMP, can lead to reduced susceptibility of S. aureus to β-lactams [4,21,22]. However, it remains unclear how common gdpP mutations are in clinical isolates, as previous reports include only limited numbers of clinical MRLM isolates (n max =20) [4,5,7,23], or such with only slightly elevated minimum inhibitory concentration (MIC) values [23].
The use of genome-wide association studies (GWASs) to identify molecular mechanisms of antibiotic resistance has been successfully employed previously [24,25]. In combination with comparative genomics, we used this approach to link phenotypes and genotypes of 141 clinical MRLM isolates collected across Germany between 2006 and 2019 and a control set of isolates, in order to identify lineage-independent polymorphisms associated with the MRLM phenotype.

Bacterial isolates
All isolates (n=283) originated from submissions to the German National Reference Centre (NRC) for Staphylococci between 2006 and 2019. They were subjected to antibioticsusceptibility testing, PCR for the detection of mecA or mecC, and spa-typing [26,27]. A total of 141 isolates phenotypically resistant towards oxacillin (OXA) (MIC >2 mg l −1 ) and cefoxitin (CXI) (MIC >4 mg l −1 ) but lacking a mec gene were selected and classified as 'meticillin resistant lacking mec' (MRLM, according to work published elsewhere [5]). In addition, 142 MSSA isolates were chosen as controls. Besides their resistance phenotype, isolates were selected to represent the current German S. aureus population with respect to clonal lineage and geographical origin (Fig. S1, available with the online version of this article) [28][29][30][31].

Whole-genome sequencing (WGS) and sequence assembly
Genomic DNA was extracted from overnight cultures in TSB (tryptic soy broth) using the DNeasy blood and tissue kit (Qiagen) and quantified with the Qubit dsDNA HS assay kit (Thermo Fisher Scientific) according to the manufacturers' instructions. Sequencing libraries were generated with the Nextera XT DNA library preparation kit (Illumina) and paired-end sequencing was performed using either a MiSeq or a HiSeq instrument with the 2×300 MiSeq v3 and the 2×250 HiSeq Rapid SBS v2 reagent kit (Illumina), respectively. The quality of raw reads was checked using FastQC v01.11.5 [34] and putative contaminations were assessed with Kraken v1.0 [35]. FQStat(S. Fuchs, personal communication) revealed a mean coverage of 226±116 fold. Trimmed raw reads (Trimmomatic v0.36) [36] were de novo assembled using SPAdes v3.12.0 [37] and the quality of assemblies was verified with quast v5.0 [38].

Impact Statement
The study described here is dedicated to the phenomenon of mec-independent β-lactam resistance, which has recently gained increased attention in the literature. It contributes additional insights and discussion points at the interface of microbial diagnostics, microbial genomics, antibiotic therapy and infection control. Using comparative genomics in combination with a genome-wide association study (GWAS) approach, we identified mutations in the c-di-AMP phosphodiesteraseencoding gdpP as causative for the resistance phenotype in a unique, comprehensive strain collection comprising exclusively patient isolates.
(six identical loci define a CC) [43]. For comparison reasons, phylogeny was further reconstructed based on whole-genome SNPs. Trimmed paired-end reads were aligned to the reference genome of S. aureus COL (NC_002951.2) using the in-house pipeline batchMap v2.0.0, as previously described [44]. The resulting consensus alignment was reduced to variant positions using SNPfilter v3.2.3 with default parameters and using an exclusion distance of 150 bp [45]. Based on the resulting SNP alignment, a neighbour joining tree was calculated in Geneious Prime 2020.0.5 (Biomatters).

SNP and indel analysis
De novo assembled contigs of all sequenced isolates were aligned to the MSSA reference strain NCTC 8325 (NC_007795) using Geneious Prime 2020.0.5 (Biomatters). Amino acid substitutions/indels in the four native pbp genes (pbp1-pbp4), the pbp4 promoter and the gdpP gene of each strain were compared, and non-silent polymorphisms were classified into lineage-specific and putatively MRLMassociated polymorphisms. All corresponding polymorphisms are listed in Fig. S1. Further sequence comparison of closely related MRLM-MSSA pairs (n=2) was carried out by aligning trimmed paired-end reads of the MSSA isolate to the concatenated, de novo assembled contigs of the corresponding MRLM isolate using the in-house pipeline batchMap as described above.

GWAS and gene enrichment
The microbial pan-GWAS tool scoary v1.6.16 was used to link patterns of gene presence or absence with the MRLM phenotype [46]. For this, a presence/absence table of 2567 core and accessory genes based on de novo assembled contigs of all 283 isolates was generated using Seqsphere + v7.1.0 (Ridom). A gene was defined as present if the corresponding allele was at least 95 % identical to the reference gene from S. aureus COL (NC_002951.2) and neither contained any frameshift nor truncation. Additionally, a binary trait matrix input was created containing the isolate's resistance phenotype (MRLM vs MSSA). scoary was run with default parameters. The presence or absence of a gene was classified as associated with the MRLM phenotype if (i) the Benjamini-Hochberg adjusted P value was <0.05, (ii) the worst pairwise comparison P value was <0.05, and (iii) the empirical P value (based on 100 permutations) was <0.05. Finally, a Manhattan plot illustrating the significance levels of resistance-associated genes in relation to their position in the S. aureus COL reference genome was created. For illustration purposes, SNPs with a -log 10 P value of 0 were excluded. In addition, a gene enrichment analysis was used to test whether some genes possessed significantly more mutations in MRLM isolates compared to MSSA isolates (with respect to the reference genome). For each gene, a negative binomial count model was created based on isolates possessing the MSSA phenotype. This model was used as a null model to test for enrichment of mutations in the MRLM isolates. The resulting P values were then adjusted using a Benjamini-Hochberg correction.

Transcription levels of pbp4
TSB overnight cultures were diluted 1 : 100 into TSB medium and grown at 37 °C to an OD 600 of 0.6. Aliquots of 20 ml were centrifuged for 10 min at 7000 g and 4 °C. Pellets were resuspended in 800 µl 1× RNA protection reagent (NEB) and cells were lysed mechanically for 3 min using 0.1 mm glass beads on a MM400 mixer mill (Retsch). Cell debris and glass beads were separated from the lysate by centrifugation at 12 000 g for 5 min, and RNA was extracted using the Monarch total RNA miniprep kit (NEB) following the manufacturer's protocol. RNA integrity was verified using the Agilent 2100 Bioanalyzer with the 6000 Nano Kit (Agilent). Real-time PCR analyses were performed with the Luna universal one-step RT-qPCR kit (NEB) according to the manufacturer's instructions and using the CFX connect real-time PCR detection system (Bio-Rad). Transcription levels of pbp4 were determined as the mean of three biological replicates for each strain, utilizing the primers pbp4_forward (5´-CTGCATAC-GAACCGACGAGT-3´) and pbp4_reverse (5´-GTCA TAGA CGCT GGAT TCCACT-3´). ΔCq values were determined by normalization to the transcription levels of the 16S rRNA gene for each strain utilizing the primers 16S_forward (5´-TCAACCTTGCGGTCGTACTC-3´) and 16S_reverse (5´-CACGCCGTAAACGATGAGTG-3´). Finally, the pbp4 transcription level of the reference strain NCTC 8325 (NC_007795) was defined as baseline and fold-change values for each strain were calculated. Unpaired t-tests were done to determine significant differences in transcription levels of pbp4 using GraphPad Prism (v8.4.0).

Biofilm formation
TSB overnight cultures were diluted 1 : 100 into fresh TSB medium supplemented with 1 % (w/v) glucose. Cell suspension (200 µl) was aliquoted into sterile 96-well flat-bottom plates and incubated at 37 °C for 24 h. Adherent bacteria were washed three times with 1x PBS, dried for 1 h at room temperature and stained with 100 µl of a 0.1 % (w/v) crystal violet solution for 5 min. The wells were washed three times with 1x PBS and adhering cells were resuspended with 100 µl of 5 % (v/v) acetic acid and the A 620 was measured. Values for each strain were calculated relative to the MRSA reference strain 38887 (10-03022) [47]. Biofilm levels were determined in three biological replicates with three technical replicates each. Unpaired t-tests were done to determine significant differences in biofilm formation using GraphPad Prism (v8.4.0).

Analysis by electron microscopy
TSB overnight cultures were diluted 1 : 100 into fresh TSB medium and grown at 37 °C to an OD 600 of 0.6. Aliquots (20 ml) were centrifuged at 5000 g for 5 min. The pellet was resuspended in 0.05 M HEPES buffer (pH 7.5) and centrifuged at 5000 g for 2 min. Finally, the pellet was dissolved in 1 ml clinical fixative (2.5 % glutaraldehyde, 1 % paraformaldehyde, 0.05 M HEPES buffer pH 7.5) and incubated for at least 48 h.
For transmission electron microscopy (TEM) imaging, staphylococcal cells were embedded in agarose by mixing equal volumes of cell suspension and low-melting-point agarose (3 % in distilled water). After post-fixation (1 % OsO4 in distilled water, 1 h), staining (2 % uranyl acetate in distilled water, 1 h), stepwise dehydration in a graded ethanol series and embedding in LR White resin (Science Services), the blocks were polymerized at 60 °C overnight. Ultra-thin sections were prepared using a ultramicrotome (UCT; Leica Microsystems) and then counterstained with uranyl acetate (2 % in distilled water) for 10 min and lead citrate for 3 min. All sections were examined in a transmission electron microscope (Tecnai12, FEI) operated at 120 kV.
For scanning electron microscopy (SEM) imaging, circular 12 mm coverslips were coated with 1 % Alcian Blue to facilitate bacterial adhesion, loaded with 100 µl S. aureus cell suspension and incubated for 30 min at room temperature in a moisture chamber. Subsequently, the coverslips were washed, dehydrated in a graded ethanol series, critical point dried, mounted on aluminium stubs, sputter coated with an 8 nm layer of gold/palladium and examined with a scanning electron microscope (ZEISS 1530 Gemini; Carl Zeiss Microscopy) operating at 3 kV with the in-lens electron detector at 10 000-fold magnification.
The diameters of 100 individual cells per strain were measured at their widest circumference using Fiji [48]. Only those bacteria were measured that presented themselves in a not tilted, unambiguous angle to the viewer. Unpaired   Fig. S1). We used the anhydrotetracycline (ATc)-inducible expression vector pRAB11 [49], in which the bla gene was deleted to prevent hyperproduction of β-lactamase leading to OXA resistance (pRAB11∆bla). The gdpP gene was amplified from NCTC 8325 genomic DNA with primers gdpP-pRAB11-for (5´-TTGA TAGA GTAT GATG GTAC CACT GACA CCTA CGAC ACATAT TG-3´) and gdpP-pRAB11-rev (5'-TTGT AAAA CGAC GGCC AGTG AATT CTAC CTTTACCT TTACCTT-TAACATC-3'). The resulting PCR product was cloned into pRAB11∆bla that had been digested with EcoRI and KpnI by Gibson assembly according to the manufacturer's protocol (NEB). The resulting plasmid pRAB11∆bla-gdpP was passed through Escherichia coli IM01B or IM30B [50] depending on the CC of the final recipient, and then used to electroporate the clinical S. aureus MRLM isolates. Expression of gdpP was induced with 100 ng ATc ml −1 . OXA MICs were determined by broth microdilution and were compared to the corresponding isolates containing the empty plasmid pRAB11∆bla.

Characterization of the strain collection Clinical origin and phenotypic susceptibility
A total of 141 MRLM and 142 MSSA were obtained from human (n=280) and animal (n=3) samples, originating from various clinical diseases including bacteraemia (n=28), wound infections (n=39), pneumonia (n=6), urinary tract infections (n=6) and abscesses (n=12). A substantial number of isolates originated from screenings (n=123) and from samples with unknown clinical origin (n=47) (Fig. 1a). The MRLM phenotype did not appear to be associated with any clinical appearance, as MRLM and MSSA strains were encountered across clinical diseases with similar frequencies (Fig. 1a, b). Therefore, we do not have evidence that MRLM isolates occur more frequently in severe chronic infections and might be selected as a consequence of prolonged β-lactam therapy, as has been suggested recently [23]. Based on our results, it appears that the selection of MRLM occurs in the community and is potentially due to a hitherto unknown selection pressure (e.g. the use of antiseptics or disinfectants as recently proposed [51]). However, we have not systematically collected the isolates investigated in this study and are not in possession of the full meta-data, so this conclusion remains speculative. Further research on the conditions of the MRLM selection process is, therefore, necessary.
Most MRLM isolates showed narrow resistance patterns (Fig. S2) and borderline resistant MICs for OXA (MIC 4-8 mg l −1 ; n=125) and CXI (MIC 8 mg l −1 ; n=132). Only a few isolates exhibited OXA MICs from 16 to 256 mg l −1 (n=16) and CXI MICs from 16 to 64 mg l −1 (n=9) (Fig. 1c). All MRLM isolates were resistant to OXA in the presence of oxacillin/sulbactam (OXA-SU), suggesting that β-lactamase overproduction is not involved in MRLM phenotype expression. Furthermore, within our MRLM collection, a large proportion of isolates were blaZ negative. Therefore, a link between MRLM phenotype development and blaZ presence could not be confirmed as suggested in another report (Fig. S2) [23]. Also, based on the genome sequences, the presence of mecB could be excluded as a cause for the MRLM phenotype [1].

Phylogeny
Extraction of spa-types (n=73) and MLSTs (n=46) from WGS data confirmed that the strain collection included a wide range of clonal lineages (Figs 1b and 2, Fig. S1).
MRLM and MSSA strains were evenly distributed throughout a neighbour-joining tree based on 49 192 SNPs in 1074 cgMLST loci (Fig. 2). This result could be confirmed by the use of SNP-based neighbour joining trees within individual CCs (data not shown). Based on these results, we concluded that the strain collection is suitable for the application of GWAS. It has previously been shown that the reliability of GWAS is highly dependent on the quality of the strain collection. The use of a well-balanced strain collection minimizes the risk of detecting polymorphisms associated with clonal lineage rather than with the phenotype of interest. Moreover, including a wide range of clonal lineages enables the detection of convergent mutations in different clonal lineages as a consequence of selection [52,53].
cgMLST analysis revealed two MRLM-MSSA pairs within the strain collection. Isolates of each pair were obtained from a single patient at one time point, indicating the presence of a heterogenous population within the patients. The isolates showed no genetic differences within the 1074 cgMLST loci employed for phylogenetic reconstruction. Isolates of the first pair belonged to sequence type (ST)59 [12-03487 (MRLM) and 12-03488 (MSSA)] and were isolated from an abscess of a human host, while the other two isolates were assigned to ST1 [12-00973 (MRLM) and 12-00975 (MSSA)] and originated from a horse suffering from an eczema. Subsequent read mapping revealed that both MRLM isolates harboured non-synonymous mutations in gdpP when compared to the corresponding MSSA isolate. MRLM isolate 12-03487 additionally harboured a T616A substitution in the fnbA gene.

Mutations in gdpP
gdpP encodes the GGDEF-domain protein phosphodiesterase, GdpP, consisting of two N-terminal trans-membrane domains followed by three intracellular domains [54]. The  N-terminal DHH-domain is required for the degradation of the second messenger cyclic-di-AMP into phosphadenylyladenosine (pApA) [22]. In the MRLM isolate 12-00973 of pair one, GdpP was truncated at position R397*, while the isolate 12-03487 of pair two revealed a H442Q mutation in the active-site DHH-domain of GdpP (Fig. 3).
Comparative analysis of all available datasets revealed mutations in gdpP in 131 of the MRLM isolates of our collection (92.9 %). This included truncations (n=37), insertions or deletions (n=61), mostly leading to frameshifts (n=58), as well as diverse amino acid substitutions (n=33) mainly located in the functional DHH-domain of the protein (Fig. 3, Fig.  S1). Further analysis of respective amino acid substitutions using sift (Sorting Intolerant from Tolerant) [55] revealed that at least 24 (72.7 %) of the 33 amino acid substitutions were predicted to affect protein function (Fig. 3). In contrast, most susceptible control isolates carried a GdpP variant that matched the GdpP of the meticillin-sensitive S. aureus strain NCTC 8325 (Fig. S1). This indicated that loss-of-function mutations in the gdpP gene might be causal for the meticillinresistance phenotype in the majority of MRLM isolates investigated. Mutations in gdpP causing reduced susceptibility to β-lactams have been previously reported in S. aureus laboratory mutants [21][22][23], as well as in smaller clinical strain collections [4,5,7]. In detail, the following polymorphisms found in MRLM isolates in our strain collection have been previously associated with β-lactam resistance: T104I [16], T260K [21], E468K, Q56*, Q163*, Q258*, E314* [5], E334* and R397* [4]. The amino acid substitutions D105N [5,16], P392S [5], I456V and D561E [56] have also been associated with β-lactam resistance previously, but were found in our collection amongst MRLM as well as control isolates, suggesting that they are rather associated with clonal lineage than with the MRLM phenotype (Fig. S1).
In addition to β-lactam resistance in S. aureus, gdpP mutations have also been implicated in DAP non-susceptibility in enterococci. However, in a previous study applying GWAS to a collection of DAP-resistant S. aureus, no association to mutational changes in gdpP was observed [57]. Furthermore, in the strain collection described here, none of the MRLM isolates examined were DAP resistant.
Additionally, 34 amino acid substitutions previously described to be associated with β-lactam resistance were found in MRLM isolates and in controls, again suggesting that they are associated with a clonal lineage rather than the MRLM phenotype (Fig. S1). This finding highlights the need for case and control isolates from a wide range of epidemiologically relevant clonal lineages to exclude lineage-specific polymorphisms.
Six MRLM isolates contained neither mutations in the native pbp1-pbp4 genes nor in gdpP. Among these, a single isolate carried a G-to-A mutation 9 bp upstream of the start codon of gdpP, which could reduce the expression of GdpP and eventually lead to phenotypic β-lactam resistance, in a manner similar to loss-of-function mutations in GdpP. In two other isolates, deletions in the pbp4 promoter were identified. One of these isolates exhibited a 17.4-fold increased transcription of pbp4 ( Table 2). Overexpression of PBP4 has previously been described to cause β-lactam resistance by increasing the amount of cross-linked peptidoglycan [20,58]. Increased amounts of cross-linked peptidoglycan, as well as increased transcription levels of the native pbp4, were also reported in a gdpP laboratory mutant [59]. However, in this study, comparative analysis of pbp4 transcription levels revealed no differences between MRLM isolates with mutations in gdpP and their corresponding control isolates (Table 2). Similarly, no differences in expression levels of PBP4 were found in a dacA (encoding c-di-AMP cyclase) mutant strain expressing decreased levels of c-di-AMP [60]. These results indicate that in clinical gdpP mutants increased transcription of pbp4 is not involved in the MRLM phenotype. Thus, the link between increased c-di-AMP levels and the MRLM phenotype remains unclear and putative mechanisms involving c-di-AMP-dependent regulation of cell wall homeostasis and cell wall stress response need to be further investigated.
In 140 of 142 control isolates, gdpP was identified as being present (Fig. S1). In contrast, in 96 of 141 MRLM isolates gdpP was absent, which is consistent with results from the detailed gdpP sequence analysis. Manual sequence analysis of the remaining 45 MRLM revealed 32 different amino acid substitutions in GdpP in 33 isolates (Fig. 3). These amino acid substitutions did not result in gdpP being assessed as 'not present' in these isolates, although amino acid substitutions might have an influence on the functionality of GdpP and, therefore, on the resistance phenotype.
To identify putative resistance determinants aside from gdpP, we initially also applied the reference-free, k-mer based dbGWAS algorithm [61] and the SNP-based GWAS-tool plink [62]. None of these approaches revealed k-mers or SNPs significantly associated with the MRLM phenotype (data not shown). In the case of the gdpP and pbp genes, this is most likely due to the large diversity of polymorphisms across the collection of MRLM isolates. The high diversity of mutations in the gdpP gene is likely to complicate the establishment of rapid genotypic tests for the detection of gdpP-mediated β-lactam resistance. Ultimately, only sequencing of the gene or the entire genome can provide reliable information on the MRLM status based on comparison to MSSA isolates. Furthermore, based on our data, a small proportion of the phenotype may also be due to other only partially known polymorphisms, which makes easy and rapid molecular diagnostics even more difficult.
We conducted a gene-enrichment analysis to determine whether some genes possessed significantly more mutations in MRLM isolates compared to control isolates (Fig. 5). The gdpP gene had, on average, 4.6 times more mutations in MRLM strains compared to MSSA controls. In addition, eight other genes were significantly enriched with P values <0.01, but the increase in mutation ratios were below 1.5fold. These genes also included pbp1 and pbp2, which were previously associated with β-lactamresistance [4,5,7,19].
Our GWAS results highlight the importance of determining the appropriate GWAS tool, as identification might be highly dependent on the type of genetic differences (e.g. horizontal acquisition of genes, mobile genetic elements vs mutations vs indels, truncations) and their distribution [25]. It is important to keep in mind that GWAS generally does not reveal causality but statistical associations with the phenotype of interest. However, in this study, GWAS results support the findings of comparative genomics and mutation enrichment studies. In contrast to our results, Giulieri et al. [23] used various approaches to find a variety of different genes associated with the MRLM phenotype. However, due to differences in strain collections and analyses, the results from the comparative analysis of clinical isolates, selection experiments and GWAS were only partially congruent, but also pointed to gdpP being associated with the MRLM phenotype [23].

Genetic complementation
To experimentally test whether the proposed mutations in gdpP indeed mediate the observed β-lactam resistance, we cloned a functional gpdP gene from S. aureus 8325 into an inducible expression plasmid. The complementation plasmid was transferred to five MRLM strains containing different gdpP mutations (from two distinct CCs) and resistance to β-lactam antibiotics was tested as before.
MIC values of the WT strains with added ATc were slightly lower than without the addition, presumably due to low basal toxicity of the inducer. Induced heterologous expression of the functional gdpP gene alongside the native, mutant variants in the MRLM strains led to a complete restoration of the β-lactamsensitive phenotype (Table 1). While the induced wild-type MRLM isolates containing the empty plasmid displayed MIC values from 4 to 8 mg l −1 , MICs of the complemented isolates ranged from 0.25 to 1 mg l −1 . This corresponds to a reduction of 3 to 5 dilution steps. Thus, in addition to showing association by GWAS, we could also experimentally demonstrate that the identified gdpP mutations are responsible for the MRLM phenotype in these strains.

Biofilm formation
The biofilm phenotype of clinical isolates strongly contributes to the virulence and resistance properties of S. aureus [63].
For several bacterial species, it has been demonstrated that an intracellular accumulation of c-di-AMP promotes biofilm formation [64]. However, reported results for gdpP mutants in S. aureus were contradictory [22,65,66]. Within our strain collection we analysed possible effects of gdpP mutations on biofilm formation in 11 MRLM and 11 MSSA isolates.
The subset of strains covered different clonal lineages, as well as diverse mutations in gdpP (Table 3). We found no consistent link between the MRLM phenotype and biofilm-forming abilities (Fig. 6a, Table 3), which is in line with previous studies on biofilm formation in gdpP mutants. While Corrigan et al. [22] reported inconsistent effects of gdpP mutations on the ability to form biofilms depending on the genetic background, Chung et al. [66] demonstrated an increase in biofilm formation with a mutation in gdpP. In contrast, DeFranscesco et al. [65] reported a reduction of biofilm formation along with a reduced eDNA (extracellular DNA) release in gdpP-mutated strains. In summary, these results suggest that the influence of c-di-AMP levels on biofilm formation is likely dependent on the strain background.

Cell size and morphology
Analysis of the same subset of 22 strains via SEM revealed clinical MRLM isolates with mutations in gdpP (n=11) to be reduced in cell size in comparison to MSSA isolates (n=11) (MRLM mean diameter=839±29 nm, MSSA mean diam-eter=964±38 nm; Fig. 6b, Table 3). MRLM isolates were further analysed by thin-section TEM to reveal additional morphological effects of gdpP mutations. In contrast to MSSA isolates, MRLM isolates contained membrane-associated vesicles in the cytoplasm (Fig. 6c). The reduced cell size and presence of membrane-associated vesicles in a gdpP deletion mutant were also described by Corrigan et al. [22]. This supports the hypothesis that mutations in gdpP could be causal for those morphological differences. However, the role of membrane-associated vesicles is still unexplained.
The influence of intracellular c-di-AMP levels on cell shape has been investigated previously. Increased c-di-AMP levels have been shown to inhibit the uptake of osmolytes, finally leading to a reduction of cell turgor [22,67,68]. Cell turgor reduction has been shown to cause decreased growth rates and cell sizes in Gram-positive bacteria, which might explain the smaller cell sizes reported [69,70]. Additionally, the cell turgor reduction of gdpP mutants might be associated with the MRLM resistance phenotype, since it prevents cell burst mediated by cell-walltargeting antibiotics like β-lactams [70]. A decrease in growth rates, suggesting a relevant fitness cost of the MRLM phenotype, was seen in many isolates of our collection and also was reported by Giulieri et al. [23].

Conclusions
In summary, we report mutations in gdpP as an important mechanism for low-level β-lactam resistance in clinical MRLM isolates. Based on statistical analyses, we could demonstrate a significant association between gdpP mutations and resistance phenotype, and could experimentally verify this for a selection of strains. However, there is currently no knowledge of how this particular β-lactam resistance genotype has been selected. The current guidelines for MRSA diagnostics require confirmation of an MRSA phenotype by detecting either a mec gene or the corresponding protein PBP2a. This is not possible for MRLM isolates, which leads to inconclusive results in clinical MRSA diagnostics. However, a correct identification of β-lactam resistances is essential in order to provide reliable therapy recommendations and to decide on adequate measures for infection prevention and control. For conventional MRSA, rapid mec gene genotyping tools are available, which enable diagnosis typically 24-48 h before phenotypic OXA-resistance results. Currently, however, this is not possible for MRLM strains due to the high gdpP diversity. Clinical studies investigating the actual effectiveness of β-lactams (and alternative antibiotics) in controlling MRLMs are, therefore, essential.

Funding information
This study was supported by a PhD grant from the Robert Koch Institute. The German Reference Centre for Staphylococci and Enterococci is funded by the German Federal Ministry of Health.