The study cohort comprised patients at MD Anderson Cancer (MDACC) who had VGS isolated from their blood between July 1, 2011, and December 1, 2012. MDAAC is a 600-bed referral cancer hospital in Houston, Texas, USA. We used a standardized data collection form to abstract clinical data from the comprehensive electronic medical records of patients with blood culture results positive for VGS. Antimicrobial drug resistance was determined in accordance with guidelines of the Clinical and Laboratory Standards Institute (www.clsi.org/standards/). VGS are known to contaminate blood cultures and to cause clinically minor, transient bacteremia, and differentiating between contamination and infection is problematic (20). Thus, for the purpose of this study, we considered that patients without signs or symptoms of infection had clinically minor bacteremia, even though they may represent cases of blood culture contamination.

Severity of infection, as measured by the Pitt bacteremia score, was determined as described (21). Pitt bacteremia scores were not determined for patients with polymicrobial bacteremia. VGS shock syndrome was defined by using the accepted definition for septic shock (i.e., hypotension refractory to fluid replacement in the setting of an infection) (22). A focus of the bloodstream infection was defined as isolation of a VGS species from a nonsterile site (e.g., liver abscess) at the same time that VGS were isolated from the blood, with the exception of infective endocarditis, which was defined according the modified Duke criteria (23,24). Neutropenia was defined as an absolute neutrophil count of <500 cells/μL.

Some patients had signs and symptoms of a lower respiratory infection and x-ray findings compatible with a pneumonic process that could not be explained (i.e., no known respiratory pathogens were isolated and no other alternative explanation, e.g., congestive heart failure, was found). Such patients were defined as having unexplained pulmonary infiltrates. Because VGS are considered normal flora, isolation of these organisms from a respiratory specimen would not have been considered clinically meaningful by the clinical microbiology laboratory and thus would not have been reported. The study protocol was approved by the MDCC institutional review board.

Bacterial isolates were identified as VGS on the basis of the following: presence of α-hemolysis, gram-positive reaction, coccus morphology arranged in chains, negative catalase test results, and exclusions of pneumococcus and enterococci by routine biochemical tests (i.e., optochin, bile solubility, and pyrrolidonyl arylamidase tests) (25). VGS species was determined as described (19). In brief, concatenated sequences of 7 housekeeping genes were used for phylogenetic tree construction in MEGA5 (megasoftware.net/); strains were assigned to VGS species on the basis of their distance from species type strains (19). For whole-genome sequencing of 9 Streptococcus mitis strains and 1 S. oralis strain, we fragmented 3 μg of genomic DNA to 350 bp (mean fragment size) and prepared barcoded sequencing libraries. The 03/10libraries were sequenced on the HiSeq 2000 sequencing System (Illumina, San Diego, CA, USA) by using 76-bp, paired-end sequencing. The raw reads in FASTQ format were aligned to the S. mitis B6 (GenBank accession no. NC_013853.1) and S. pneumoniae TIGR4 (GenBank accession no. NC_003028.3) genomes by using Mosaik (26). There was an average of 250× coverage per base, indicating extremely high confidence for base calls. Contigs were generated by feeding the raw genome sequence data into the A5 pipeline (27). Gene annotations were obtained by uploading contigs to the Rapid Annotation using the Subsystem Technology server at the National Microbial Pathogen Data Resource website (28). (Individual gene sequencing data have been deposited into GenBank. Short-read sequencing data have been deposited to the Short Read Archive (www.ncbi.nlm.nih.gov/Traces/sra/sra.cgi?view=announcement) under accession no. PRJNA240080.)

Experiments in mice were performed according to a protocol approved by the MDACC Institutional Animal Care and Use Committee. To induce neutropenia, we injected 5-week-old female Balb/C mice intraperitoneally with 100 mg/kg of cyclophosphamide (Sigma, St. Louis, MO, USA) on days −4 and −1 before bacterial injection. On day 0, mice (10 per bacterial dose) were injected with 100 μL of phosphate-buffered saline (PBS) containing 10-fold increments of bacteria ranging from 10³ to 10⁷ CFUs. As a control, 10 mice were injected with PBS alone. Mice were monitored over 7 days for near-death status. The dose at which 50% of the mice nearly died (hereafter referred to as LD₅₀) was calculated by using the probit method. Neutropenia was confirmed in select mice on postinfection days 1, 3, and 6.

Differences between categorical variables were assessed by using the χ² test; Fisher exact test was used when at least 1 category had <5 occurrences. The relationship between VGS species and Pitt bacteremia scores was analyzed by using the Mann-Whitney U test. The Bonferroni method was employed to account for multiple comparisons when appropriate. All tests of significance were 2-sided, and statistical significance was defined at p<0.05. SPSS Statistics version 19 (IBM, Armonk, NY, USA ) was used for statistical analysis.

A total of 118 consecutive patients with VGS-positive blood cultures were included in the study cohort; ≈80% of the patients had neutropenia and hematologic malignancies (Table 1). Most patients had bacteremia without a defined focus, but several other clinical scenarios were observed, including skin/soft tissue infections, gastrointestinal infections, and infective endocarditis. Most patients had clinically mild infections (Pitt bacteremia score of 0 or 1), but 25% of patients had moderate to severe infections (Pitt bacteremia scores of >2), including 12 patients who had VGS shock syndrome.

To gain insight into the species of VGS causing bacteremia in the study cohort, we performed MLSA of 7 housekeeping genes, as described (19). Strains were assigned to species by comparing their position on the phylogenetic tree with those of established type strains (19). The 118 strains could be confidently assigned to 11 distinct species (Figure 1; Technical Appendix 1). The most commonly observed species were S. mitis (68 strains), S. oralis (22 strains), and S. parasanguinis (12 strains). For classification purposes, various VGS species are often placed into distinct groups; the association between VGS strains causing bacteremia and group assignment is shown in Figure 1, using the scheme from Sinner et al. (29). In total, 80% of strains were from the Mitis group, and the remaining strains were from the Sanguinis group (14%), Anginosus group (3%), and Salivarius group (3%).

Because of the diverse genetic nature of the various VGS species, we next tested the hypothesis that distinct VGS species cause different clinical syndromes. Given the number of strains for each species, we analyzed the Mitis group species (i.e., S. mitis and S. oralis) individually and analyzed species comprising the Sanguinis, Anginosus, and Salivarius groups by group (Table 2). Compared with strains of other VGS species, S. mitis strains were significantly more likely to cause primary bacteremia (p<0.01) and less likely to cause polymicrobial bacteremia (p = 0.01) and clinically minor bacteremia (p<0.01). S. oralis strains were more likely to cause polymicrobial infection (p = 0.02), Sanguinis group strains were more likely to cause clinically minor bacteremia (p<0.01), and Anginosus group strains were significantly associated with bacteremia with a gastrointestinal focus (p<0.01). When we only considered patients with neutropenia or cases of monomicrobial bacteremia, we observed the same statistically significant species–clinical disease relationships (data not shown).

We next sought to determine if there was a relationship between VGS species and disease severity (as determined by Pitt bacteremia score). Organ dysfunction, such as hypotension, begins to occur at Pitt bacteremia scores of >2 (21). The distribution of Pitt bacteremia score by infecting species is shown in Figure 2, panel A. Patients infected with S. mitis were significantly more likely to have a higher Pitt bacteremia score (p<0.01). One possible explanation for this observation is that S. mitis strains mainly caused infections in patients with neutropenia who, compared with patients without neutropenia, might be more likely to have serious infections. Thus, we repeated the analysis, including only patients with neutropenia. Again, the Pitt bacteremia scores were significantly higher for patients infected with S. mitis (p<0.01; Figure 2, panel B). Of the 12 cases of VGS shock syndrome, 11 were caused by S. mitis strains and 1 was caused by an S. constellatus strain (Anginosus group).

Most cases of bacteremia and severe disease occurred in patients infected with S. mitis; thus, we focused on S. mitis strains and strains from the closely related S. oralis species. In contrast to what we observed for the S. oralis strains, several distinct groupings could be visualized within the S. mitis strains, which we arbitrarily labeled as clusters 1, 2, and 3 (Figure 3, panel A). S. mitis cluster 1 comprised 10 strains, including 2 that were genetically identical by MLSA, and cluster 2 comprised 22 strains, including 6 that were genetically identical. Cluster 3 comprised 29 strains and may contain additional strain groupings, but further phylogenetic delineation of this cluster could not be done with sufficient confidence. We did not observe substantial differences, in terms of distinct disease types or severity of infection, between patients from whom the S. mitis cluster strains were derived (Technical Appendix 2). However, there was a predominance of unexplained pulmonary infiltrate cases among patients infected with cluster 2 strains (p<0.01 for cluster 2 strains vs. noncluster 2 strains) (Figure 3, panel B). This variable was investigated because, given the close genetic relationship between S. mitis and S. pneumoniae, we hypothesize that S. mitis strains may cause pneumonia in severely immunocompromised persons (Figure 1).

To determine whether the MLSA data accurately represented the entire genetic content of the Mitis group strains, we performed whole-genome sequencing of 9 S. mitis and 1 S. oralis isolates (Figure 4, panel A). For the 9 S. mitis strains, the reads mapped to ≈70% coverage of the only completely finished S. mitis genome (S. mitis strain B6 [30]), Technical Appendix 2, Table 1). This considerable level of intraspecies genetic diversity for S. mitis strains has been observed previously in sequencing and DNA:DNA hybridization studies and meant that we could not use whole-genome analysis of single-nucleotide polymorphisms to determine strain relatedness (30,31). Thus, we next sought to identify regions of genetic similarity among the strains that could be analyzed for interstrain comparisons.

All 9 S. mitis strains contained operons encoding a putative polysaccharide capsule. The first 4 genes of the operon, corresponding to cpsA–cpsD in S. pneumoniae, were relatively well conserved among the 9 strains. Concatenated alignment of cpsA–cpsD showed a close relationship for the 4 cluster 2 strains and strain Shelburne VGS (SVGS) 003, whereas the cpsA–cpsD genes from the remaining 4 strains were more closely related to the S. pneumoniae strain TIGR4 (SVGS004 and SVGS019) or to the S. mitis type strain SK142 (SVGS002 and SVGS011) (Figure 4, panel B).

In addition to the capsule operons, multiple other comparisons arising from our whole-genome analysis confirmed the idea that the 4 cluster 2 strains and strain SVGS003 were closely related. All of the S. mitis strains contained the gene encoding the fibrinogen-binding protein, PavA (pneumococcal adherence and virulence protein A). However, there was a different gene 5′ to the pavA gene in SVGS003 and the 4 cluster 2 strains than in the other 4 S. mitis strains (Figure 4, panel C). In a similar manner, the LytB protein in cluster 2 strains and SVGS003 grouped separately from the LytB protein in the other strains, and a LytB paralog in strain SVGS003 and the 4 cluster 2 strains was distinct from other forms of the LytB protein and from LytB paralogs encoded by SVGS011 and SVGS019.

LytB is part of a group of choline-binding proteins that are involved in cell-wall turnover, some of which have been shown to be important for virulence in S. pneumoniae (32). When the presence or absence of choline-binding proteins was determined for the various strains, substantial strain-to-strain heterogeneity was observed (Technical Appendix 2 Table 2). The only repeating pattern of gene content was 1 that occurred for 3 different choline-binding protein–encoding genes: cbpE, cbpI, and lytC. These 3 genes, which are present in diverse chromosomal locations, were absent in the 4 cluster 2 strains, SVGS003, and the S. oralis strain SVGS021 but present in the other noncluster 2 S. mitis strains. Thus, the whole-genome data support MLSA data, assigning 4 of the fully sequenced strain to S. mitis cluster 2; strain SVGS003, a group 3 strain by MLSA, appears to have genetic characteristics of the cluster 2 strains by whole-genome analysis.

Given the apparent differences in genetic content among S. mitis strains, we sought to develop an animal model for testing VGS virulence that would approximate the disease observed in cancer patients with neutropenia. No neutropenia model of VGS infection exists, so we used serial 10-fold CFU dilutions of 5 S. mitis strains to determine the LD₅₀ of organisms for the endpoint of being near death. The S. mitis challenge strains included isolates from the major S. mitis clusters (Figure 5; Table 3), and we also injected PBS as a control. None of the mice injected with PBS became ill, indicating that neither the neutropenia nor the injection itself caused major disease (Figure 5, panel B). All of the S. mitis strains could cause near-death status, and a dose–response relationship was observed for all strains (see example in Figure 5, panel B), but the LD₅₀ varied by 100-fold among the strains (Table 3). Strain SVGS016, which caused the most severe clinical disease (i.e., it was isolated from a patient with the highest Pitt bacteremia score) also was the most virulent in the mouse model. Thus, we suggest that S. mitis strains cause disease in mice with neutropenia and that there is differential virulence in this mouse model among genetically diverse S. mitis strains.