1184 Lif Helgerson et al. J Dent Res 90(10)2011 only reason given for non-participation was lack of time. The or later than gestational wk 37 only was also run. Variables were study was approved by The Regional Ethical Review Board, autoscaled to unit variance, and cross-validated prediction ofY Umea, Sweden, and participating mothers signed informed con- calculated(Wold, 1978). Cross-validation was done by a sys- sent at recruitment tematic prediction of 1/th of the data by the remaining 6/7th of Mode of delivery(C-section or vaginal), intravenous treat- the data. The importance of each x-variable in the model is ment with antibiotics during delivery, and body weight and given by a variable importance in projection(VIP)value. VIP> length were checked against medical records. The mothers com- 1.0 was considered influential and ViP> 1.5 highly influential leted a questionnaire on other possible confounders, such as ( Sjostrom et al, 1986). The R2-and Q -values give the capacity health issues(allergy, infections, stomach problems ) the infant's of the x-variables to explain(R)and predict (Q)the outcome use of antibiotics, feeding mode(breast-or bottle-fed), use of a pacifier, and the presence of teeth. RESULTS Microbiota by 16S rRNA Probes in HOMIM Microarray Cohort Descripti We collected oral biofilm samples by carefully swabbing the There were no differences by gender or by other characteristics, cheeks,tongue, and alveolar ridges. DNA was purified from including breast-feeding, between infants born vaginally and samples with the use of the Gen Elute Bacterial Genomic DNA those bom by C-section(Table 1). Two infants were born in ges- kit(Sigma Aldrich, St. Louis, MO, USA) to obtain 60-1220 ng tational wk 35(one delivered by C-section and one by the vaginal DNA, which exceeded the amount required for the microarray route), whereas all other infants were born in gestational wk or assay. Four samples were excluded because of low yield after later. All infants were healthy at birth and at 3 mos of age. None DNA extraction, leaving 38 and 25 of the samples from of the infants had ever received antibiotic treatment, and none was C-section and vaginal delivery groups, respectively ever given supplements containing probiotic bacteria. With the The purified DNA of samples was assayed with 422 oligo- exception of 15 mothers who received intravenous antibiotics in nucleotide probes to the 16S rRNa gene targeting more than association with a C-section because of an acute clinical compli 300bacterialtaxaintheHomiMmicroarray(http://mim.forcationnonehadantibioticsatdeliveryTherewerenosignificant syth. org/homim. html). Samples were analyzed at the homim differences between participating and non-participating infants in microarray facility at The Forsyth Institute, Cambridge, MA, length and weight at birth and at 3 mos of age, or in the socio- USA(Colombo et al, 2009). Hybridization signals were read on economic variables of their families(data not shown) a six-level scale(0-5), with a lower limit of detection of 10 cells Colombo et al., 2009) Bacteria Detected by HOMIM Microarray Statistical procedures There was reactivity to 85 of the 300 taxa in the HOMIM micro- array in oral biofilms of three-month-old infants(Appendix body length and weight at birth and at 3 mos were averaged among Table). Bacteria detected belonged to 6 phyla or divisions, and infants delivered by the two birth delivery modes. Differences approximately half of the taxa detected belonged in Firmicutes, Dichotomized scores from the HOMIM microarray analyses were infants(Table 2). Other genera detected in 80 to 99%of all used. Lack of signal was set to 0, and all signal levels I to children were Actinomyces. Gemella and Veillonella. a smaller Differences n prevalence distribution between groups were tested proportion of the infants (<15% of the combined groups) had th a Chi? test. The False Discovery Rate method was used to species in the genera Bacteroides, Selenomonas, Aggregatibacter identify a p-value with less than one false rejection of HO when true Kingella, Neisseria, and the TM7 division. (p <0.005). Thus, a p-value < 0.005 was considered statistically Species or species clusters detected in all infants were Streptococcus Cluster Il, Streptococcus Cluster Ill, Streptococcus Multivariate partial least-squares discriminant analysis anginosus/intermedius, Streptococcus oralis(Append (PLS-DA)modeling was performed (SIMCA P+, version 12.0, Table). Species detected in 2 80% of all children were Umetrics AB, Umea, Sweden)as described(Sjostrom et al, Streptococcus Cluster I, Streptococcus mitis biovar 2, 986: Bylesjo et al, 2006). In contrast to traditional regression Streptococcus australis, Streptococcus parasanguinis I and models, the PLS-DA technique, which defines the maximum Actinomyces gerensceriae, Gemella hemolysans, veillonella separation between class members(here mode of delivery) in atypical, and Veillonella parvula Species detected in only a few the data, is suitable for data where the number of observations is (<15%)infants included Streptococcus sanguinis and smaller than the number of variables, and where the independent Streptococcus mutans, species of Neisseria, Aggregatibacter, variables co-vary. Dichotomous HOMIM signals, and the and Kingella, and Actinomyces naeslundii genospecies and selected individual characteristics, gender, weight and length at (Actinomyces clusters I and Il, respectively)(Appendix Table) birth and 3 mos, gestational wks at delivery, treatment antibiotics during delivery, feeding mode(bottle- or breast-fed), use of pacifier, presence and number of teeth, and town of resi- Species Distribution by Mode of Delivery dence built the X-block, and mode of delivery the Y-block(out- There were higher numbers of taxa detected by the microarray come). An identical model including breast-fed infants born in in swabs from infants delivered vaginally (79 species/clusters)1184 Lif Holgerson et al. J Dent Res 90(10) 2011 only reason given for non-participation was lack of time. The study was approved by The Regional Ethical Review Board, Umeå, Sweden, and participating mothers signed informed consent at recruitment. Mode of delivery (C-section or vaginal), intravenous treatment with antibiotics during delivery, and body weight and length were checked against medical records. The mothers completed a questionnaire on other possible confounders, such as health issues (allergy, infections, stomach problems), the infant’s use of antibiotics, feeding mode (breast- or bottle-fed), use of a pacifier, and the presence of teeth. Microbiota by 16S rRNA Probes in HOMIM Microarray We collected oral biofilm samples by carefully swabbing the cheeks, tongue, and alveolar ridges. DNA was purified from samples with the use of the Gen Elute Bacterial Genomic DNA kit (Sigma Aldrich, St. Louis, MO, USA) to obtain 60-1220 ng DNA, which exceeded the amount required for the microarray assay. Four samples were excluded because of low yield after DNA extraction, leaving 38 and 25 of the samples from C-section and vaginal delivery groups, respectively. The purified DNA of samples was assayed with 422 oligonucleotide probes to the 16S rRNA gene targeting more than 300 bacterial taxa in the HOMIM microarray (http://mim.forsyth.org/homim.html). Samples were analyzed at the HOMIM microarray facility at The Forsyth Institute, Cambridge, MA, USA (Colombo et al., 2009). Hybridization signals were read on a six-level scale (0-5), with a lower limit of detection of 104 cells (Colombo et al., 2009). Statistical Procedures Body length and weight at birth and at 3 mos were averaged among infants delivered by the two birth delivery modes. Differences between means were tested by two-sided, independent t tests. Dichotomized scores from the HOMIM microarray analyses were used. Lack of signal was set to 0, and all signal levels ≥ 1 to 1. Differences in prevalence distribution between groups were tested with a Chi2 test. The False Discovery Rate method was used to identify a p-value with less than one false rejection of H0 when true (p < 0.005). Thus, a p-value < 0.005 was considered statistically significant to account for multiple comparisons. Multivariate partial least-squares discriminant analysis (PLS-DA) modeling was performed (SIMCA P+, version 12.0, Umetrics AB, Umeå, Sweden) as described (Sjöström et al., 1986; Bylesjö et al., 2006). In contrast to traditional regression models, the PLS-DA technique, which defines the maximum separation between class members (here mode of delivery) in the data, is suitable for data where the number of observations is smaller than the number of variables, and where the independent variables co-vary. Dichotomous HOMIM signals, and the selected individual characteristics, gender, weight and length at birth and 3 mos, gestational wks at delivery, treatment with antibiotics during delivery, feeding mode (bottle- or breast-fed), use of pacifier, presence and number of teeth, and town of residence built the X-block, and mode of delivery the Y-block (outcome). An identical model including breast-fed infants born in or later than gestational wk 37 only was also run. Variables were autoscaled to unit variance, and cross-validated prediction of Y calculated (Wold, 1978). Cross-validation was done by a systematic prediction of 1/7th of the data by the remaining 6/7th of the data. The importance of each x-variable in the model is given by a variable importance in projection (VIP) value. VIP > 1.0 was considered influential and VIP ≥ 1.5 highly influential (Sjöström et al., 1986). The R2 - and Q2 -values give the capacity of the x-variables to explain (R2 ) and predict (Q2 ) the outcome. Results Cohort Description There were no differences by gender or by other characteristics, including breast-feeding, between infants born vaginally and those born by C-section (Table 1). Two infants were born in gestational wk 35 (one delivered by C-section and one by the vaginal route), whereas all other infants were born in gestational wk 37 or later. All infants were healthy at birth and at 3 mos of age. None of the infants had ever received antibiotic treatment, and none was ever given supplements containing probiotic bacteria. With the exception of 15 mothers who received intravenous antibiotics in association with a C-section because of an acute clinical complication, none had antibiotics at delivery. There were no significant differences between participating and non-participating infants in length and weight at birth and at 3 mos of age, or in the socioeconomic variables of their families (data not shown). Bacteria Detected by HOMIM Microarray There was reactivity to 85 of the 300 taxa in the HOMIM microarray in oral biofilms of three-month-old infants (Appendix Table). Bacteria detected belonged to 6 phyla or divisions, and approximately half of the taxa detected belonged in Firmicutes, particularly Streptococcus species, which were detected in all infants (Table 2). Other genera detected in 80 to 99% of all children were Actinomyces, Gemella, and Veillonella. A smaller proportion of the infants (< 15% of the combined groups) had species in the genera Bacteroides, Selenomonas, Aggregatibacter, Kingella, Neisseria, and the TM7 division. Species or species clusters detected in all infants were Streptococcus Cluster II, Streptococcus Cluster III, Streptococcus anginosus/intermedius, and Streptococcus oralis (Appendix Table). Species detected in ≥ 80% of all children were Streptococcus Cluster I, Streptococcus mitis biovar 2, Streptococcus australis, Streptococcus parasanguinis I and II, Actinomyces gerensceriae, Gemella hemolysans, Veillonella atypical, and Veillonella parvula. Species detected in only a few (< 15%) infants included Streptococcus sanguinis and Streptococcus mutans, species of Neisseria, Aggregatibacter, and Kingella, and Actinomyces naeslundii genospecies1 and 2 (Actinomyces clusters I and II, respectively) (Appendix Table). Species Distribution by Mode of Delivery There were higher numbers of taxa detected by the microarray in swabs from infants delivered vaginally (79 species/clusters)