In this study we used real-time PCR to obtain a quantitative estimation of the presence of the four predominant Lactobacillus species known to occupy the vaginal econiche and of Atopobium vaginae and Gardnerella vaginalis.
Real-time PCR has been applied previously for the description of changes in the vaginal microflora. Zariffard et al.  and Sha et al. [28, 29] used real-time PCR on frozen cervicovaginal lavage samples to quantify the presence of Mycoplasma hominis, G. vaginalis and the combined presence of L. crispatus and L. jensenii, in comparison with the microscopic interpretation of the Gram stain according to the criteria of Nugent . Bradshaw et al.  applied real-time PCR for A. vaginae for a follow-up study of recurrent BV before and after treatment with oral metronidazole. Ferris et al.  applied real-time PCR for A. vaginae to samples of six BV patients before and after treatment with a topical metronidazole gel. Thus far, only one group (Byun et al. ), studying advanced dental caries, used real-time PCR to quantify the presence of specific Lactobacillus species.
Compared to microscopy and culture, molecular methods such as PCR may provide less observer-dependent and culture medium dependent information and a more direct and detailed view of the quantity of certain species in a sample. E.g., A. vaginae was only recently recognized as strongly linked to disturbed vaginal microflora because of its fastidious growth requirements in vitro. A possible additional advantage of PCR, compared to culture, is that it permits detection of dead or metabolically inactive bacteria, and as such it can be applied for the detection of bacteria in a dormant state, as is the case in biofilms, which occur also in BV as has been shown by means of FISH .
Besides culture-associated bias, difficulties with regard to the standardization of vaginal sampling make quantification of the vaginal microflora cumbersome and make bacterial counts reported in different publications difficult to compare. Some groups have attempted to standardize sampling by cervicovaginal lavage (CVL), whereby the volume of saline that is recovered from the vagina can be measured and used for calibration of the bacterial load/ml. Drawbacks are that CVL samples also from the cervix and does not represent a pure vaginal sample. Also it cannot be excluded that biofilm, that may be present in cases of BV, is recalcitrant with regard to dissolving in the saline that is used and is not sampled in the same manner as loosely associated microflora. In the condition of BV there may be more vaginal cell debris so it is feasible that there is more non-bacterial material present in the sample. In conclusion, vaginal sampling remains difficult to standardize. In this study we used unweighted vaginal swabs taken by the same gynecologist so there is no standardization for the volume of the sample, but standardization of the way samples are taken.
Besides sampling bias and culture bias also real-time PCR may be biased, because the conversion of the PCR positivity threshold value to a concentration of bacteria, depends on the manner the standard curve is calibrated. The standard curve, which is prepared by extracting DNA from a dense bacterial suspension and by serial tenfold dilution of this DNA extract, can be compared to i) the bacterial colony count, as determined by plating out tenfold serial dilutions of the initial suspension, or to ii) the number of bacterial genomes in the initial suspension, determined on the basis of measurement of the DNA-concentration present in the extract from the initial suspension. We found differences between calculation of number of genomes by measurement of DNA-concentration and determination of number of cells by dilution culture from less to one log10 unit (L. jensenii) to more than 3 log10 units (L. crispatus) (data not presented). A possible explanation may be that species like L. jensenii are more easily cultured on blood plates than for example L. crispatus and therefore show a lower discrepancy between DNA-concentration and culture. In conclusion, when preparation of the standard curve in this study had been based on cultured cells, L. jensenii would have been overestimated relative to the other species.
L. crispatus and L. jensenii
L. crispatus can be found in samples of all grades but the log10 median concentration/ml (MC) of this species is much higher in grade I (8.8) compared to grades II (4.2) and III (5.2). Although the difference between the log10 quantity of L. crispatus in grade Ia and grade Ib is not significantly different (p = 0.096), possibly because of the low sample size, a different profile is observed for these grades (Figure 2 and 3).
L. crispatus is present in high inoculum (i.e. at least) in all 8 grade Ia samples (log10 MC 9.0), in most (i.e. 6/10) grade Iab samples (log10 MC 8.9), but only in 3/12 grade Ib samples (log10 MC = 5.6). Our study confirms that a grade Ia microflora is characterized by high concentrations of L. crispatus, whereas grades Iab and Ib consist of a more diverse microflora, with different Lactobacillus species present in comparable amounts.
Also, in grade Ib, high concentrations of L. crispatus can be found but on Gram stain especially long and small Lactobacillus cell types, which were suggested to be more typical for L. gasseri , are observed.
Based on cloning of the 16S rDNA library from the indigenous microbiota of three women with normal vaginal microflora, Verhelst et al.  found an almost pure population of L. jensenii in one of them and concluded that this species was associated with a normal vaginal microflora, but in this study (71 swabs) we can not correlate L. jensenii with either normal or disturbed vaginal microflora since it is present sporadically in samples of all grades, and in both high and low numbers. Its presence in 8/10 grade Iab samples, compared to 1/8 grade Ia samples is striking (p = 0.024).
G. vaginalis and A. vaginae
A positive correlation between G. vaginalis and A. vaginae could be established (p < 0.0001). G. vaginalis can be found in all grades but in much higher concentration in a disturbed vaginal microflora. Delaney et al.  performed a comparative study in which they quantified G. vaginalis, Lactobacillus spp., Prevotella spp. and Peptostreptococcus spp. in vaginal swabs by means of culture. The mean log10 number of G. vaginalis in BV was determined to be 9.64 cfu/g, which may compare well to the value we obtained, i.e. 9.2 cells/ml in grade III.
A. vaginae is found less frequently than G. vaginalis in non-grade III samples, but in high concentrations in 10 of the 12 grade III samples, confirming that the presence of A. vaginae seems to be a diagnostically more valuable marker for BV than the presence of G. vaginalis, as suggested by Verhelst et al.  and confirmed by several other studies [31–33]. A recent study using T-RFLP for the characterization of normal (n = 20) and BV (n = 50) microflora found that the terminal restriction fragment (TRF) with the length to that corresponding of A. vaginae was present in the vagina of 48 of the 50 women with BV (grade II and III lumped together) and absent from all 20 women with normal microflora . Bradshaw et al.  and Ferris et al.  monitored the changes in the concentration of A. vaginae before and after treatment with metronidazole and both found that very high vaginal concentrations of A. vaginae were predictive for patients for whom treatment failed partially or completely, although Bradshaw et al.  found a significant reduction of A. vaginae after metronidazole treatment. This might be due to a lower susceptibility of some strains of A. vaginae for metronidazole  but a more plausible explanation might be the existence of biofilm as shown by Swidsinski et al. , and in which G. vaginalis was shown to be associated with A. vaginae.
L. gasseri and L. iners
Overall L. gasseri and L. iners are abundantly present in most grades. Strikingly, L. gasseri is virtually absent in grade III whereas L. iners is virtually absent in grades Ia and II.
In our study we found a high concentration of L. iners to be clearly associated with low concentrations of L. gasseri, which is more prevalent in normal (grade Ia, grade Ib, grade Iab) and mildly disturbed flora (grade II). This was confirmed by Spearman rank testing, which showed that L. gasseri and L. iners are significantly negatively correlated to each other in grades II and III (r = -0.793, p < 0.0001), but also when all samples are considered together (r = -0.397, p = 0.001).
It can be noted that even the single grade III sample with a high concentration of L. gasseri was also one of the two grade III samples with low concentration of L. iners. In addition, we found a positive correlation between G. vaginalis and L. iners. The distribution frequencies over the different grades of L. iners and G. vaginalis are quite similar, i.e. present in most grades but predominantly in grade III, but different from A. vaginae, i.e. present almost exclusively in grade III, which may explain why L. iners does not correlate with A. vaginae. This distribution of L. iners was also observed in the T-RFLP study of Thies et al. , with 32 out of 50 BV samples and 11 out of 20 normal microflora samples containing the L. iners TRF. The L. gasseri group TRF was observed only twice, in women with normal microflora. Unfortunately, these authors did not distinguish between grade II and III.
According to some research groups , L. iners is the most abundant vaginal Lactobacillus species, not only in black African women (64% of 241 healthy premenopausal Nigerian women) but also in white women from Sweden, the US and Canada and is considered as even more typical for normal vaginal microflora than L. crispatus. Although L. iners was present in high numbers in grades Iab and Ib, we found L. crispatus and L. gasseri to be more abundant in grade Ia. Ferris et al.  concluded that L. iners is a more transitional species, present in recently cured patients. They reported the predominance of L. iners in 5 patients with BV after metronidazole treatment, whereas for the sixth patient, the only one with a complete treatment failure, L. iners was present but not predominant. These results were confirmed by Jakobsson and Forsum .
Interestingly, it has been observed by the group of Taylor-Robinson that grade II microflora responds poorly to clindamycin treatment whereas most subjects with grade III microflora readily reverted to grade I [37, 38] and also that women with clindamycin treated grade III microflora had a better pregnancy outcome than women with clindamycin treated grade II microflora . In agreement, the predominance of clindamycin susceptible L. iners in grade III might explain the observations of Taylor-Robinson et al. .
Finally, in case grade II would be an intermediate stage in between transition from grade I to grade III, one would expect to see an increase in L. iners and a decrease in L. gasseri, since the former is present in high numbers in grade III and the latter almost absent, but we observe exactly the opposite, i.e. nearly complete absence of L. iners and high numbers of L. gasseri in grade II.
In order to establish whether this could be related to the differential presence of L. gasseri and L. iners in these grades, with L. gasseri predominant in grade II, we determined the MIC for clindamycin for 10 strains of each. We found that 8 out of 10 L. iners strains were inhibited by less than 0.25 μg/ml, whereas 9 out of 10 L. gasseri strains had a MIC value of 2 or more μg clindamycin/ml. Although L. gasseri cannot be considered definitely as the causative micro-organism responsible for the adverse pregnancy outcome that has been associated with grade II, our data indicate that it can be considered as contributing to the poor response to clindamycin treatment.
To test the hypothesis that L. gasseri and L. iners are mutually exclusive, e.g. by the production of bacteriocins, we carried out inhibition tests between both species. However, with the method we used we could not observe mutual inhibition between both species.
With the use of molecular techniques it becomes clear that the definition for bacterial vaginosis in which it is claimed that the lactobacilli are replaced by anaerobic bacteria is not quite correct. Our data indicate that rather L. crispatus, L. gasseri and L. jensenii present in a healthy vaginal microflora and grade II microflora are replaced largely by L. iners in grade III vaginal microflora. Accordingly, the qualitative culture study by Delaney & Onderdonk  found a weak negative correlation between Nugent score and numbers of all lactobacilli, meaning that the number of lactobacillar cells did not decrease with high Nugent score, i.e. disturbed microflora. On the other hand, Zariffard et al.  and Sha et al. [28, 29] who quantified lactobacilli in cervicovaginal lavage samples, using a real-time PCR format which picked up only L. crispatus and L. jensenii, found a decline in the number of lactobacilli in BV. The discrepancy may be explained because we take also L. iners and L. gasseri into account.
To further resolve these discrepancies we carried out Gram staining of several strains of each L. gasseri and L. iners. It is of importance to note that L. iners has not only long escaped from our attention due to its inability to grow on MRS agar, but also hides as a Lactobacillus from Gram stains, because it stains Gram-negative and its cell morphology is rather coccobacillar than bacillar. Since it is the predominant Lactobacillus species in BV microflora – even present in high numbers – this has led to the false assumption that the BV microflora is devoid of lactobacilli.
The very long lactobacillar cells observed for some of the L. gasseri isolates remind of earlier observations. Pahlson & Larsson  reported the occurrence of very long fusiform lactobacilli in 7.4% of the vagina of 981 women studied. Horowitz et al.  reported the occurrence of 'vaginal lactobacillosis', a pathological condition characterized by the presence of extremely long lactobacilli (up to 60 μm).
Grade I-like was described only recently by Verhelst et al.  as a grade that, on Gram stain, resembles grade I because of the Gram positive rods, but with the use of molecular methods it was shown that these rods were Bifidobacterium spp. instead of Lactobacillus spp.
Again this study confirms the separate nature of grade I-like vaginal microflora. A significantly lower concentration of L. crispatus (log10 MC = 5.5 cells/ml) is present compared to grade I (log10 MC = 8.8 cells/ml, p = 0.014) and much lower concentrations of G. vaginalis (log10 MC = 4.0 cells/ml, p < 0.0001) and A. vaginae (log10 MC = 1.6 cells/ml, p = 0.003) are present in grade I-like microflora compared to grade III (respectively log10 MC = 9.2 cells/ml and log10 MC = 9.0 cells/ml).