We found a 15% prevalence of rPBP3 in a nationwide collection of 795 eye, ear and respiratory isolates of H. influenzae in Norway. The prevalence of clinical resistance to ampicillin due to rPBP3 was 9%, compared to 2.5% in a similar study three years earlier . Despite methodological differences between the two studies, we conclude with a significant increase from 2004 to 2007. National phenotypic surveillance data indicate a further increase to 17% rPBP3 in respiratory isolates in 2011  and a prevalence at 15% rPBP3 in invasive isolates in 2012 (n = 73, 77% nontypeable) , consistent with observations in other European countries and in Canada [2, 4, 12, 14].
As expected, group II low-level resistant isolates predominated. Notably, group III high-rPBP3 was identified for the first time in Northern Europe. The genotypic distinction between low-level and high-level beta-lactam resistance is clinically relevant: As resistance to cefotaxime is mainly seen in high-rPBP3 , cefotaxime is suitable for empiric treatment of severe disease only in regions where high-rPBP3 is rare. However, 6% of group II isolates in the present study were resistant to cefotaxime and 20% were non- susceptible to meropenem in case of meningitis. These observations underline the importance of confirming susceptibility to beta-lactams in severe infections such as meningitis and septicemia.
When the prevalence of low-rPBP3 in Japanese respiratory isolates reached 17% in the mid 1990s, group III isolates increased from zero to 29% in six years . This was followed by a rapid increase in group III isolates in meningitis (predominantly Hib) from zero to 70% . A recent report revealed a shift from low-level to high-level resistance in respiratory tract isolates in South Korea during the last decade, with an increase in the prevalence of group III isolates from 1% to 21% in five years [16, 22].
A similar development in other parts of the world would seriously compromise current empiric antibiotic therapy in severe infections. So far, single group III isolates have been reported from France  and Canada  whereas group III-like isolates are slightly more frequent [9, 14, 18, 20, 21, 24]. Clusters of group III and group III-like high-level resistant isolates were recently observed in Norway (Skaare et al., manuscript in preparation).
The current epidemiologic situation in Europe and Canada, with a gradually increase in low-rPBP3 and sporadic reports of high-rPBP3 isolates, strongly resembles the situation in Japan and South Korea prior to the shifts in resistance genotypes. Continuous monitoring of susceptibility to cefotaxime and meropenem is necessary to ensure safe empiric treatment.
By comparing the study isolates with isolates from a comparable population collected in 2004 , we were able to study the clonal dynamics of PBP3-mediated resistance. The increasing prevalence of rPBP3 in Norway is due to expansion of a few clones. Four STs with characteristic ftsI alleles accounted for 61% of the rPBP3 isolates in the present study. Two of these strains were the main contributors to PBP3-mediated resistance in Norway three years earlier . Interestingly, the replacement of ST14 by ST367 as the most prevalent rPBP3 strain did not cause a shift in PBP3 type nor phylogroup, as both STs carried PBP3 type A and belong to eBURST group 2.
We have previously suggested the existence of one or more widely disseminated rPBP3 clones . This is supported by later reports of PBP3 type A and compatible substitution patterns (identical to PBP3 type A as far as comparison is possible) being common in Europe [4, 18, 23–25], Canada [3, 12], Australia  and South Korea [16, 22], and by the present study.
PBP3 type A is frequently linked to ST14 and ST367 in the limited number of previous reports on the molecular epidemiology of rPBP3. Studies on invasive H. influenzae in Canada in the periods 2000–2006 [2, 12, 42] and 2008–2009  revealed an increasing prevalence of rPBP3 in NTHi, with PBP3 type A being common in both sampling periods [3, 12]. ST14 and ST367, respectively, were the most common STs in NTHi from two different regions and sampling periods [3, 42]. PBP3 type A was by far the most frequent substitution pattern in ST14 and also appeared in some ST367 isolates (R. Tsang, personal communication).
Furthermore, a study on invasive H. influenzae in Sweden  identified a cluster of seven NTHi isolates of ST14 and related STs (hereunder ST367), all carrying PBP3 type A and collected in the period 2008–2010 (F. Resman, personal communication). Finally, in two recently published Spanish studies, ST14 and/or ST367 isolates with substitution patterns compatible with PBP3 type A were reported in invasive disease (ST367, n = 2)  and pneumonia (ST14, n = 2; ST367, n = 1)  in the period 2000–2009.
The ftsI alleles encoding type A in this and our previous study  had high genetic similarity and alleles in separate clusters rarely encoded identical PBP3 types. Thus, despite the lack of cross-study comparison of ftsI DNA sequences, the examples above indicate that clonal distribution is a more likely explanation for the occurrence of PBP3 type A and compatible patterns in separate studies from four continents [3, 4, 9, 11, 12, 16, 18, 20],[22–25] than independent development of this substitution pattern by convergence.
Importantly, an invasive high-level resistant rPBP3 isolate with the same combination of MLST allelic profile (ST155) and PBP3 substitution pattern as the two group III-like isolates in the present study was recently reported from Spain . A single-locus variant (ST1118) with an identical substitution pattern was also reported.
These observations are notable and support the need of global surveillance initiatives. We here show that combining MLST and PBP3 typing provides a tool for cross-study identification of rPBP3 strains and clones. The previously suggested system for subgrouping of group II isolates  does not separate PBP3 types [11, 16] and is unsuitable for this purpose.
Preferably, MLST should be combined with ftsI DNA sequencing. The ftsI gene is nearly 200 kb from its nearest MLST neighbor (mdh) and distortion of the MLST results due to linkage is thus very unlikely. With recent technological development reducing both costs and analysis time of whole-genome sequencing, and smaller bench-top sequencers becoming readily available, MLST-ftsI typing will probably be possible to perform for surveillance purposes in the near future.
We are aware of a number of previous studies where MLST and ftsI sequencing was performed [3, 4, 12, 23–25, 43–45]. To our knowledge, four reports have linked MLST data and PBP3 substitution patterns: one presented the allelic profiles of 83 group III respiratory isolates from Japan ; another presented the substitution pattern of a single group II ST368 NTHi isolate causing meningitis in Italy ; and two most recent publications presented the substitution patterns and STs of 95 respiratory  and 18 invasive isolates  from Spain. However, the present study is to our knowledge the first to connect STs to ftsI alleles.
PFGE is highly discriminative and generally considered suited for assessment of relatedness between epidemiologically connected isolates, particularly in populations with high recombination rates such as NTHi [39, 46]. In this study, PFGE clusters correlated well to MLST clonal complexes. Band patterns were stable over time and also traced phylogenetic relationship not detected by MLST and parsimony analysis. Combining MLST and PFGE for typing of NTHi may thus increase both sensitivity and resolution of clone detection.
Development of resistance
As discussed above, clonal expansion is important for the spread of rPBP3. However, the PBP3 type A-encoding, highly divergent ftsI allele lambda-2 was distributed among several unrelated STs. Similar observations are reported previously but only in studies using PFGE for strain characterization [11, 26]. Exchange of complete alleles by HGT seems the most likely explanation, and has been demonstrated in vitro . The mechanisms for HGT of ftsI sequences in H. influenzae are not completely resolved but involvement of classical transformation and homologous recombination has been suggested [26, 47].
Transformational competence varies extensively between H. influenzae strains . This implies that the ability to acquire mutant ftsI alleles encoding rPBP3 will vary correspondingly, which may explain the differences in ST and phylogroup distribution between rPBP3 and sPBP3 isolates. It has been suggested that phylogroups are maintained by restriction barriers, preventing recombination between isolates of different heritage . This is challenged by the distribution of lambda-2 to several phylogroups. A simple explanation may be that restriction barriers prevent recombination between some phylogroups and allow recombination between others.
Recent studies applying whole-genome sequencing have revealed that transformation in competent strains of H. influenzae is more extensive than previously recognized  and that transformational exchange may cause allelic variation involving complete genes between strains of identical STs . However, transfer of complete ftsI alleles is probably less common than exchange of shorter sequences, causing mosaicism [26, 28]. Preliminary multiple sequence alignment analysis of ftsI sequences in this study indicated intrageneic recombination (data not shown).
PBP3-mediated resistance and virulence
The association between rPBP3 and virulence is poorly described. One experimental study reported increased ability of a group III NTHi strain to invade bronchial epithelial cells, and the authors hypothesized that rPBP3 may enhance virulence by acting as an adhesion molecule . A more recent retrospective epidemiological study concluded with no difference in pathogenicity between rPBP3 and sPBP3, but an association between rPBP3 and underlying respiratory disease was observed . Molecular strain characterization was not performed in any of the two studies.
In the present study, regression analysis (without adjustment for ST) suggested that rPBP3 is associated with increased risk of eye infection and hospitalization. However, ST-specific analysis indicated that pathogenicity is correlated with STs rather than with resistance genotypes. For instance, ST395, ST396 and ST201 were significantly associated with eye infections but only the two latter STs were associated with PBP3-mediated resistance.
Irrespective of resistance genotypes, a number of STs in this study had clinical characteristics recognizable from previous reports: ST159 was predominantly isolated from the respiratory tract of hospitalized elderly patients, consistent with ST159 being adapted to infection in chronic obstructive pulmonary disease (COPD) ; and the high proportion of ST57 in children is consistent with a previously reported association with acute otitis media (AOM) . Finally, the high hospitalization rate of patients with ST14-PBP3 type A corresponds well with the potential of this strain to cause pneumonia  and invasive disease [3, 4, 42].
These observations are in accordance with a recent population study suggesting association between population structure and disease . In conclusion, the association between rPBP3 and pathogenicity suggested by the regression analysis most likely reflects that some of the most frequently occurring rPBP3 strains in this study also possessed strain-associated virulence properties. Identification of virulence determinants is beyond the scope of this study. However, our observations underline that studies on the correlation between resistance genotypes and pathogenicity should include molecular strain characterization. Accordingly, the previously reported association between PBP3-mediated resistance and clinical characteristics [17, 51] may be spurious.