Comparison of known VAG and ARG characteristics studied in the targeted approach.

(A-C) Comparison of the distribution of VAGs per strain among the six main functional classes of virulence of the E. coli commensal and BSI collections isolates. We show the odds ratio (OR with 95% CI) for the risk of infection associated with the number of VAGs (logistic model of infection status as a function of the number of VAGs), for (A) all the strains (912 BSI strains), (B) BSI strains with urinary portal of entry (PE) to commensals (498 BSI strains) and (C) BSI strains with digestive portal of entry to commensals (310 BSI strains). Functional classes of virulence are ordered by increasing associated odds ratio for all strains. (D-F) Comparison of the distribution of resistant strains for eight antibiotics of clinical importance of the E. coli commensal and BSI collections isolates. We show the odds ratio (OR with 95% CI) for the risk of infection associated with the resistance of strains (logistic model of infection status as a function of the resistance of strains), for (D) all the strains (912 BSI strains), (E) BSI strains with urinary portal of entry (PE) to commensals (498 BSI strains) and (F) BSI strains with digestive portal of entry to commensals (310 BSI strains). Categories of antibiotics are ordered by increasing associated odds ratio for all strains. For AMK, FQ and FEP categories, we only show the lower bound of the CI because the estimated odds ratios are huge as none of the commensal isolates were resistant to these antibiotics. AMK, amikacin; AMP, ampicillin; CTX/CAZ, cefotaxime/ceftazidime; FEP, cefepime; FQ, fluoroquinolones; GEN, gentamicin; SXT, cotrimoxazole; TZP, piperacillin/tazobactam. (G) Repertoire of adhesins and (H) iron capture systems in BSI and commensal strains. We selected the four most significant systems in terms of complete genes and/or alleles when comparing commensal to BSI strains (S4 Table) and evaluated their combinations. A system was considered present when all its genes were detected. (I) Predictor coefficients of pathogenicity among adhesins and (J) iron capture systems determined with a lasso regression. We calculated confidence intervals using 1000 bootstrap resamples. Unselected genes and/or alleles (lasso coefficient close to zero) are shown in gray.