Emergence of KPC-3- and OXA-181-producing ST13 and ST17 Klebsiella pneumoniae in Portugal: genomic insights on national and international dissemination

Abstract

Background

Carbapenem-resistant Klebsiella pneumoniae (CRKP) strains are of particular concern, especially strains with mobilizable carbapenemase genes such as bla_KPC, bla_NDM or bla_OXA-48, given that carbapenems are usually the last line drugs in the β-lactam class and, resistance to this sub-class is associated with increased mortality and frequently co-occurs with resistance to other antimicrobial classes.

Objectives

To characterize the genomic diversity and international dissemination of CRKP strains from tertiary care hospitals in Lisbon, Portugal.

Methods

Twenty CRKP isolates obtained from different patients were subjected to WGS for species confirmation, typing, drug resistance gene detection and phylogenetic reconstruction. Two additional genomic datasets were included for comparative purposes: 26 isolates (ST13, ST17 and ST231) from our collection and 64 internationally available genomic assemblies (ST13).

Results

By imposing a 21 SNP cut-off on pairwise comparisons we identified two genomic clusters (GCs): ST13/GC1 (n = 11), all bearing bla_KPC-3, and ST17/GC2 (n = 4) harbouring bla_OXA-181 and bla_CTX-M-15 genes. The inclusion of the additional datasets allowed the expansion of GC1/ST13/KPC-3 to 23 isolates, all exclusively from Portugal, France and the Netherlands. The phylogenetic tree reinforced the importance of the GC1/KPC-3-producing clones along with their rapid emergence and expansion across these countries. The data obtained suggest that the ST13 branch emerged over a decade ago and only more recently did it underpin a stronger pulse of transmission in the studied population.

Conclusions

This study identifies an emerging OXA-181/ST17-producing strain in Portugal and highlights the ongoing international dissemination of a KPC-3/ST13-producing clone from Portugal.

Introduction

Data on Klebsiella pneumoniae antimicrobial resistance (AMR) in Europe show that in 2019 approximately one-third (36.6%) of all invasive strains were resistant to at least one antimicrobial agent under surveillance, and that 7.9% harboured resistance to carbapenems.^1–3 Resistance to this β-lactam class is associated with increased mortality and frequently co-occurs with resistance to other antimicrobial classes: 4.9% of all isolates reported showed concomitant resistance to carbapenems, third-generation cephalosporins, fluoroquinolones and aminoglycosides.³ Moreover, outbreaks involving resistance to carbapenems and colistin (a last-resort drug) have been reported.⁴ Carbapenem resistance is usually driven by loss of permeability or production of carbapenemases. The latter entails additional concern due to the potential for lateral transfer of bla_KPC, bla_NDM or bla_OXA-48 genes via plasmids coupled with other mobile genetic elements such as Tn4401.⁵ In addition to the K. pneumoniae resistome diversity, several virulence factors such as those associated with capsule production, type 1 and 3 fimbriae and siderophore production can provide selective advantages in the host environment.

From a phylogenetic perspective, early studies enabled the identification of three K. pneumoniae phylogroups (KpI–III), now regarded as separate species, with K. pneumoniae comprising more than 150 deeply branching lineages.^6,7 The differential ability of some of these lineages to successfully disseminate is notorious and well documented. Such is the case of the highly successful and epidemic K. pneumoniae ST258 strain that, along with its clonal group (CG258), is thought to have emerged in the mid-1990s and acted as a key driver for the worldwide dissemination of bla_KPC.⁸

Understanding not only the molecular drivers of AMR but also which clones are acquiring further resistance determinants and its dissemination dynamics is critical. Comparative genomics of multiple clinical isolates from Lisbon enabled the dissection of a polyclonal outbreak scenario including distinct strains belonging to the same sequence type.⁹ However, the scope of prior genomic epidemiological surveillance in Lisbon has so far been limited to a single hospital. Nothing is known regarding the genomic diversity of carbapenemase-producing K. pneumoniae in major hospital centres formed by several tertiary care hospitals. Here, we study the genomic diversity of clinical K. pneumoniae isolates obtained from tertiary care hospitals in Lisbon and their spread in Europe. The data obtained provide further evidence for the emergence and international dissemination of two carbapenemase-producing non-CG258 clades.

Methods

Clinical isolates and drug susceptibility testing

The study included a set of 20 carbapenem-resistant K. pneumoniae (CRKP) clinical isolates, all from different patients and consecutively isolated between June 2018 and September 2019 from clinically relevant biological specimens at the central Microbiology Laboratory of Centro Hospitalar de Lisboa Central. The latter comprises a tertiary care university hospital encompassing five different hospitals, designated hospitals A–E, with approximately 1352 beds.

Initial identification and drug susceptibility testing was carried out using the automated VITEK2 system (bioMérieux, La Balme-les-Grotte, France). Colistin susceptibility was determined by broth microdilution using the Sensititre Gram Negative MIC Plate (Thermo Scientific, Waltham, MA, USA). All isolates were subcultured on blood agar plates and sent to the Bacterial Pathogenomics and Drug Resistance Laboratory at the Research Institute for Medicines for further genotypic and phenotypic characterization and storage.

Two additional genomic datasets were included afterwards for comparative purposes: (i) a total of 26 isolates belonging to ST13, ST17 and ST231 available at the iMed.ULisboa strain collection and obtained between 1982 and 2019 in multiple hospitals in Lisbon (n = 23) and Coimbra (n = 2), and a single isolate obtained from wastewater samples¹⁰; and, (ii) 64 international genomic assemblies of ST13 over 14 countries from the period 1999–2020.

WGS and de novo assembly

Genomic DNA was extracted from overnight growth on Mueller–Hinton agar using the NZY Tissue gDNA Isolation kit (NZYTech, Portugal) and sequencing performed on an Illumina HiSeq 4K high-throughput platform (paired-end 2  × 151 bp reads) by the Applied Genomics Centre at the London School of Hygiene and Tropical Medicine.

De novo assembly was carried out using the Unicycler pipeline (v0.4.4),¹¹ and SPAdes v.3.8 for k-mer selection optimization assembler. Assembly quality was evaluated using QUAST: N50, 128 969–405 372 bp; contig count, 66–173; largest contig, 923 944 bp. Annotation was carried out using Prokka (v1.14.6).¹² Roary (v.3.13.0) was used to investigate the core and pan genomes of ST13 (n = 84) using a 95% cut-off.¹³

For the analysis of plasmidic genomes, three Dutch isolates (29778, 30778 and 37937) were sequenced by a Dutch in-house-developed protocol of long-read third-generation sequencing.¹⁴ In brief, the Oxford Nanopore protocol SQK-LSK109 and the expansion kit for native barcoding EXP-NBD104 were used (Oxford Nanopore Technologies, Oxford, UK). The final library was loaded onto a MinION flow cell (MIN-106 R9.4.1). Base calling and demultiplexing was performed using Albacore 2.3.1, and a single FASTA file per isolate was extracted from the FAST5 files using Poretools 0.5.1.¹⁵ Fifty base pairs were trimmed at both sides and only reads larger than 5000 were used in further analyses. Illumina and Nanopore data were used in a hybrid assembly performed by Unicycler.¹¹

Phylogenetic analysis

Phylogenetic analysis was performed by mapping raw sequencing reads to the reference genome of K. pneumoniae F93-2 (GenBank Accession NZ_CP026157). Prior to mapping, raw sequencing reads were trimmed using Trimmomatic v0.36. Subsequently, reads were mapped to the reference genome using the Burrows Wheeler Aligner tool (BWA-MEM algorithm).^16,17 Following deduplication and local indel realignment by Picard Tools and GATK v.3.6, variant calling was performed by both SAMtools/BCFtools and GATK (UnifiedGenotyper) software (mapping quality >23, depth of coverage >10).^18,19 We retained only concordant variants between SAMTools/GATK, and an additional coverage validation step was performed where a missing call was assigned if the coverage depth did not reach a minimum of 20 reads or none of the nucleotides reached 75% of the total coverage. SNP positions were removed if they: (i) showed an excess of 10% missing calls; (ii) were within 10 bp of other SNP positions; and (iii) yielded <49 bp unique k-mers.²⁰ This approach produced an alignment of 36 400 sites across all 21 genomes (including reference genome). Next, the best fit nucleotide substitution model was determined in R using the phangorn package and ranked upon Akaike Information Criterium correction for the number of estimated parameters. A maximum-likelihood phylogeny was constructed using Seaview (PhyML), and the statistical robustness of each clade assessed by the approximate likelihood ratio test (aLRT).^21,22 The Interactive Tree of Life online tool was used to annotate and visualize the tree (iTOL, https://itol.embl.de/).²³ A minimum spanning tree was constructed using the Phyloviz (goeBURST) online tools.²⁴

The additional ST13 genomic dataset was integrated using the procedure described above but based on concatenated SNP pseudo-molecules of 38 024 bp. Further assessment was carried out via principal component analysis (PCA) with four principal components using the adegenet package in R.²⁵ The fixation index (F_ST) was calculated using the pegas R package for each individual SNP used in the phylogenetic analysis as a metric of its ability to identify specific ST13 subclades.²⁶

Phylogeography, dating and transmission modelling

BEAST2 software was used to estimate the divergence dates and geographical dissemination of the ST13/GC1/KPC-3 clade using the year of isolation as tip dates and country of origin as a discrete character state. Initial runs of 50 million iterations were carried out for different permutations of tree priors and clock models and compared by marginal likelihood estimation using both stepping-stone and path sampling methods. A constant tree prior distribution under an uncorrelated relaxed lognormal molecular clock was found to best fit the data upon comparison by Bayes factor estimation. Using the best fit models and a general time-reversible substitution model, BEAST was run for 300 million iterations with sampling at each 1000 iterations. Tracer was used to evaluate the convergence of multiple runs as well as the effective sample sizes (ESS). All statistics showed a value well above the minimum ESS threshold considered (ESS = 100) and convergence was observed for all statistics over three independent runs under the same priors. LogCombiner was used to merge log and tree files from independent runs, and TreeAnnotator was used to produce a final maximum clade credibility tree.

Transphylo was used to reconstruct transmission events for the ST13/GC1/KPC-3 branch and investigate transmission dynamics over time.²⁷ The time-scaled maximum clade credibility tree was used as input for Transphylo. A gamma distribution with shape parameter 1.2 and scale 1.0 for the generation time prior, a distribution mean of 1.2 years and an SD of 1.096 years were used as previously reported.²⁸

In silico typing, AMR genes and capsular types

MLST was determined using Institut Pasteur’s MLST (https://bigsdb.pasteur.fr/klebsiella/). AMR genes were detected and identified using AMRFinder (https://ncbi.nlm.nih.gov/pathogens/antimicrobial-resistance/AMRFinder/) and the NCBI Bacterial Antimicrobial Resistance Reference Gene Database (Accession PRJNA313047) using 60% coverage and 95% identity thresholds. Capsular type (KL) and O-antigen (O) types were predicted using Kaptive as implemented in Kleborate software (https://github.com/kelwyres/Kaptive-Web).

Results and discussion

Between June 2018 and September 2019, a total of 20 CRKP clinical isolates underlying distinct clinical conditions were isolated from Centro Hospitalar de Lisboa Central (Table S1, available as Supplementary data at JAC Online). Four isolates showed consistently lower resistance levels to all carbapenems tested (ertapenem, imipenem and meropenem) and only met the resistance threshold for ertapenem (>0.5 mg/L) with MIC values of 2.0 mg/L. All the remaining isolates (n = 16) showed a MIC of ≥8.0 mg/L. Notably, all isolates showed resistance to second- and third-generation cephalosporins, aminopenicillins and piperacillin/tazobactam, and 11 isolates were also resistant to cefepime. Phenotypic resistance to other drugs was also noted (see Table S2).

Genomic population structure and diversity of K. pneumoniae

To gain deeper insights into the resistance determinants and the population structure for this dataset, all isolates (n = 20) were subjected to WGS. The MLST, KL and O-antigen types were then determined in silico. A total of four STs were detected in this sample set: ST13 (n = 12), ST17 (n = 5), ST231 (n = 2) and ST4446 (n = 1). All ST13 isolates were KL3/O1v2 whereas all but one ST17 isolates were KL25/O5. Both ST231 isolates were classified as KL51/O1v2, and the single ST4446 isolate showed an undetermined KL (KL3/KL102) and an O1v2 antigen type (Table S1 and Figure 1).

The phylogenetic tree based on 36 400 SNPs subsequently constructed denotes a topology that is congruent with the ST-based classification (Figure 1). All ST clades were assigned to monophyletic branches although Kp5509 shared a more recent common ancestor with the reference strain used (NZ_CP026157; ST17). In fact, Kp5509 showed a deeper branching pattern within the ST17 clade and a high pairwise SNP distance from other ST17 isolates (>2900 SNPs). These findings are consistent with the divergence observed at the KL and O-antigen types and suggest both capsular and O-antigen switching events associated with this specific clade. We further identified two genomic clusters (GCs) by imposing a 21 SNP cut-off^29,30 on pairwise comparisons: ST13/GC1 (n = 11) and ST17/GC2 (n = 4), which may highlight recent ongoing spread between patients and hospitals. A high pairwise SNP distance (231 SNPs) was found between both ST231 isolates therefore precluding any association to ongoing recent transmission events involving both patients. Also, Kp5526 (ST13) did not meet the inclusion criteria relative to GC1 by a marginal difference and should therefore be regarded as a GC1 potentially related isolate. The phylogenetic position of this isolate, nested within GC1, lends further support to this notion. It is also noteworthy that each GC is composed of clinical isolates obtained from patients in different hospitals: GC1 comprised 11 patients from four hospitals whereas GC2 comprised four patients, all from different hospitals (Figure 1; Table S1). These observations likely reflect a scenario of ongoing dissemination of ST13 and ST17 strains between different hospitals in Lisbon and stress the importance of the adoption of strict infection control measures.

Resistome diversity, virulence and phylogenetic distribution

Upon screening for AMR genes, we determined that all isolates included in this study were classified as carbapenemase producers bearing one of two genes: bla_KPC-3 (n = 16/20; 80.0%) or bla_OXA-181 (n = 4/20; 20.0%). All bla_OXA-181 isolates also carried the bla_CTX-M-15, aac(6′)-Ib-cr5, aac(3)-IIa and arr-3 genes, which did not occur in the KPC-3-producing isolates (Figure 1; Table S1). The Tn4401d transposon was determined as the genetic structure at the origin of the dissemination of all bla_KPC-3 genes detected. This and other Tn4401 isoforms have been identified as key drivers of global bla_KPC-3 dissemination.^31–33

The OXA-181-producing isolates were restricted to the ST17 clade and showed a complete overlap with the four isolates with lower carbapenem resistance levels. Only one isolate from the latter clade, Kp5509, carried the bla_KPC-3 gene instead of bla_OXA-181, which is consistent with its phylogenetic distance and unrelatedness with the GC2/ST17 isolates producing OXA-181. No ESBL other than blaCTX-M-15 (in ST17) was identified. Two ST17 isolates harboured the blaDHA-1 gene. Six aminoglycoside resistance-conferring genes were identified: aac(3)-IIa and aac(6′)-Ib-cr only among ST17 isolates; aadA in all isolates; aph(3′′)-Ib and aph(6)-Id in all isolates except Kp5521; and aac(6′)-Ib in all non-ST17 isolates (Figure S1; Table S1).

None of the isolates contained genes coding for the regulator of mucoid phenotype A (rmpA, rmpA2), which is associated with hypervirulence and the hypermucoviscosity phenotype. Consistent with its phylogenetic distance from the remaining ST17 isolates, the Kp5509 isolate showed a virulence genotypic profile that contrasts with the profile observed for GC2/ST17 isolates, namely, the presence of glf, kfoC, wbbMNO, wzm and wzt coding genes, all part of the rfb locus responsible for O-antigen biosynthesis and diversity. Also, Kp5509 does not bear the yersiniabactin gene cluster (fyuA, irp and ybtAEPQSTUX). All isolates harboured the gene clusters coding for type I and III fimbriae (fim and mrk gene clusters) along with other genes involved in capsule synthesis, enterobactin siderophore and type VI secretion system. Production of colibactin toxin was restricted to ST13 isolates. Overall, comparing the number of resistance and virulence coding genes by drug class or function, ST17 isolates appear to be more strongly associated with drug resistance due to the association with an increasing number of AMR genes, whereas ST13 shows an increased potential for more virulent phenotypes (Table S1; Figure 1).

In fact, K. pneumoniae ST17 strains are increasingly regarded as some of the most successful non-CG258 clones, and ST17 has been proposed as a high-risk clone driving the emergence and/or burden of CTX-M-like genes.³⁴ OXA-181-producing ST17-K. pneumoniae had been already described in 2017 in a single Lisbon hospital and was herein detected across four other hospitals in Lisbon denoting its emergence also as an important carbapenemase-producing clone.³⁵ The importance of KPC-3- and OXA-181-producing ST13 and ST17 K. pneumoniae strains, respectively, is further strengthened by the findings of Mendes et al.³⁶ in a different tertiary care hospital in Lisbon, which also found these strains, albeit in lower proportions (25.0% and 22.1%, respectively), although only six of the KPC-3-producing ST13 isolates were also KL3/O1v2.

National and international dissemination of ST13 K. pneumoniae

We next investigated the geographical scope and reach of the disease caused by the main clades investigated by searching for ST13, ST17 and ST231 isolates available from sequenced strain collections in Portugal. A total of 26 isolates were identified (13 ST13, 3 ST17 and 10 ST231): 23 were obtained from Lisbon hospitals between 1982 and 2015; 2 were obtained from a major tertiary care Hospital in Coimbra (in 2017); and 1 (from 2019) was previously isolated from a recent study conducted to investigate the burden of AMR pathogens in wastewater samples from treatment plants (Table S3). The inclusion of this additional dataset enabled the expansion of ST17/GC2 to six isolates with generally similar resistome compositions, albeit Kp3HUC lacked the aac3-IIa, bla_OXA-1, bla_TEM-1 and catB3 genes. GC1/ST13 remained unchanged since the additional ST13 isolates were over 100 SNPs distance. A third smaller cluster, highlighting possible recent interregional dissemination event(s) between Lisbon and Coimbra, was identified and composed of two ST231/KL51/O1v2 KPC-3-producing isolates (distance: 18 SNPs) (Figure S1).

Afterwards we focused our attention on the GC1/ST13 strains identified as the most important disseminating threat in the investigated setting. By reconstructing a genome-wide phylogenetic tree (38 024 sites) that encompassed the 12 ST13 isolates identified in this study along with 13 additional ST13 isolates from Portugal and 64 publicly available K. pneumoniae ST13 genomes, we were able to identify K. pneumoniae GC1/ST13/KL3/O1v2/KPC-3-producing strains as an international emerging clone. The latter is found to be part of one of two major clades of ST13 strains (herein designated clade A) and one of its subclades (clade A1), whose phylogenetic distinctiveness finds support in the populational structure analysis by PCA identifying six and four barcode SNPs with perfect differentiation (fixation index F_ST= 1) for clades A and the GC1/KPC-3 clone, respectively (Figure 2; Table S4; Figures S2 and S3).

In this new scenario, GC1 is found to encompass 23 isolates but limited to three countries: Portugal, France and the Netherlands. The six French KPC-3-producing isolates were identified in a recent report and were isolated from Portuguese immigrant patients.²⁹ Likewise, the Dutch isolates were collected from patients who previously travelled to Portugal and stayed at Portuguese hospitals. The global ST13 phylogenetic tree denotes the importance of the GC1/KPC-3-producing clone along with its clonal expansion across these countries. Multiple carbapenemase-coding genes were detected—bla_KPC-2 (n = 4), bla_KPC-3 (n = 28), bla_OXA-48 (n = 2) and bla_OXA-181 (n = 1)—but showed a scattered distribution across the phylogenetic tree, except for the GC1/KPC-3 clade. The polyphyletic distribution of isolates per country of origin denotes multiple dissemination events across the tree topology, which is also congruent with the fact that the French isolates were recovered from patients from four distinct regions in France (Figure 2).²⁹

ST13 K. pneumoniae is an internationally disseminated clone associated with the spread of several ESBL and carbapenemase genes: bla_OXA-48 in Finland and Ireland,^37,38bla_TEM-3 in France³⁹ and bla_DHA-1 in Spain.⁴⁰ The prevalence of ST13 KPC-3-producing K. pneumoniae, on the other hand, is low but is increasingly recognized as these have been identified as an emerging non-CG258 clone herein and in French and Dutch patients with epidemiological links to Portugal.²⁹

ST13 isolates have a highly conserved core genome and a bla_KPC-3-carrying IncFIA plasmid

The genomic content of ST13 isolates and of its GC1/KPC-3 clade was further characterized by determining the number of core and pan genes associated with both clades. Genomic annotation and comparative gene content analysis enabled the identification of a total of 11 202 genes, 4436 (39.6%) of which conserved across at least 95% of the ST13 isolates. In contrast, GC1/KPC-3 strains showed a smaller gene pool comprising 5660 genes and a total of 4936 (87.2%) core genes, therefore highlighting a more conserved core genome, congruent with the recent clonal expansion already shown by the phylogenetic analysis (Figure S4).

Hybrid assembly of long-read and short-read sequencing data of Dutch isolates enabled the identification of three different replicon types across all three isolates, including multiple ColRNAI plasmids (2.5–24.6 kb) and one IncFIB(K) plasmid that was only detected in a single isolate and bore, among other resistance genes, a bla_CTX-M-15 coding gene. In addition to these and common to all three isolates, were the IncFIA(HI1) plasmids with sizes varying between 129.9 and 141.8 kb and similar gene content, containing the bla_KPC-3 gene along with additional resistance genes. All K. pneumoniae ST13 isolates, except one Portuguese ST13 isolate (Kp5519) showed ∼100% coverage for the bla_KPC-3-bearing IncFIA(HI1) plasmid (Figure 3) denoting its presence across all ST13/KPC-3 isolates obtained in Portugal. These data suggest an overall similar plasmid content between isolates obtained in Portugal, France and the Netherlands with increased variability observed mostly for ColRNAI plasmids, which, except for one plasmid (RIVM_C0307788_3), do not appear to be associated with drug resistance.

Phylogeography and dating the emergence of the European K. pneumoniae ST13/KPC-3 clone

To gain further insight into the disseminating history of the GC1/KPC-3 clone in Europe we next obtained a consensus time-scaled phylogenetic tree along with ancestral discrete character reconstruction as an approach for identifying the geographical locations associated with each ancestral node for this subtree. The estimates obtained place the emergence of the GC1/ST13/KPC-3 clone in 2008 (2008.7; 95% highest posterior density: 2003.7–2012.8) in Portugal based on an inferred clock rate of 8.35 × 10⁻⁵ substitutions per site per year. Ancestral reconstruction is therefore congruent with multiple cross-border dissemination events having Portugal at the origin, occurring over at least one decade before the initial isolates studied, and along with within-country spread between hospitals. Furthermore, reconstruction of the transmission tree suggests that a stronger pulse of transmission initiated in 2013–2014 with most patients unsampled by the present study, although much of these are likely associated with asymptomatic carriage but are nonetheless potential sources of transmission (Figure 4). Other high-risk clones such as those belonging to ST147 and ST15 from 2014/2015 were previously associated with the dissemination of bla_KPC-3, from non-hospitalized patients in Portugal.³² These results demonstrate that bla_KPC-3 acquisition can happen out of the hospital settings, supporting the asymptomatic carriage of these genes.

Conclusion

In conclusion, the present study revealed dissemination of ST13 and ST17 strains carrying bla_KPC-3 and bla_OXA-181 carbapenemase genes between different hospitals in Lisbon. By performing WGS-based analysis we were able to identify that KPC-3-producing K. pneumoniae clones (ST13) from France and the Netherlands are epidemiologically linked to the clones from the Portuguese hospital settings. These observations stress the importance of adopting strict infection control measures and shows the importance of WGS-based analysis in monitoring the dynamics of high-risk clones.

Data availability

Sequence data have been submitted to the European Nucleotide Archive (https://www.ebi.ac.uk/ena/) under study accession ERP125389.

Funding

This study was supported in part by UID/DTP/04138/2019 from Fundação para a Ciência e Tecnologia (FCT), Portugal and by Associação para o Desenvolvimento do Ensino e Investigação da Microbiologia (ADEIM). R.E. is supported by the FCT through the PhD Fellowships (Grant Reference: 2021.08701.BD). J.P. (CEECIND/00394/2017) is supported by Fundação para a Ciência e Tecnologia through Estímulo Individual ao Emprego Científico. T.G.C. received funding from the Medical Research Council UK (grant nos. MR/K000551/1, MR/M01360X/1, MR/N010469/1, MR/R020973/1, MR/R025576/1, MR/S01988X/1, MR/S03563X/1, MR/X005895/1) and Biotechnology and Biological Sciences Research Council (BBSRC) UK (BB/R013063/1). S.C. received funding from the Bloomsbury Set, Medical Research Council UK (grants MR/R020973/1, MR/R025576/1, MR/S01988X/1, MR/S03563X/1, MR/X005895/1) and the BBSRC UK (BB/R013063/1).

Transparency declarations

None.

Supplementary data

Figures S1–S4 and Tables S1–S4 are available as Supplementary data at JAC Online.

References

Author notes