Unraveling the selective antibacterial activity and chemical composition of citrus essential oils

Post-weaning diarrhea (PWD) is an often disease affecting piglets. It is caused mainly by enterotoxigenic Escherichia coli (ETEC) colonization in pig gut. Antibiotics has been used to prevent, combat and control PWD and its negative impact on the productivity of pig breeding sector. Nonetheless, antibiotics due to their wide antibacterial spectrum also can reach beneficial gut bacteria, such as Lactobacillus. Lately, essential oils (EOs) have emerged as a potential alternative to using antibiotics in animal breeding because of their effect on bacterial growth. Commonly, citrus EOs are by-products of food industry and the availability of these EOs in the worldwide market is huge. Thus, six commercials citrus EOs were evaluated on ETEC strains, as model of pathogenic bacteria, and on Lactobacillus species, as models of beneficial bacteria. In overall, citrus EOs exhibited a selective antibacterial activity with higher effect on pathogenic bacteria (ETECs) than beneficial bacteria (Lactobacillus). Brazilian orange terpenes (BOT) oil presented the highest selective performance and caused higher disturbances on the normal growth kinetic of ETEC than on Lactobacillus rhamnosus. The action was dose-dependent on the maximal culture density (A) and the lag phase duration (λ) of the ETEC. The highest sub-inhibitory concentration (0.925 mg/mL) extended the λ duration to ETEC eight times (14.6 h) and reduced A in 55.9%. For L. rhamnosus, the λ duration was only extended 1.6 times. Despite the fact that limonene was detected as the major compound, the selective antibacterial activity of the citrus EOs could not be exclusively attributed to limonene since the presence of minor compounds could be implicated in conferring this feature.

colistin is still allowed following the current recommendation of the European Medicine Agency, which restrict the rational use of colistin to treat clinical cases in livestock animals 14 . Moreover, the wide spectrum activity of some antibiotics can affect gut microbiota since they can kill or inhibit both pathogenic and beneficial bacteria. Thus, longtime antibiotic use can provoke decreased microbiota diversity and increased chances of pathogens colonizing the gut 2 . On the other hand, it is well known that Lactobacillus is the major group of beneficial bacteria presented in pig gut microbiota, which has been identified as an important group of bacteria able to prevent gut diseases 15 . Possible Lactobacillus mechanisms to fight post-weaning infections in piglets have been described as (i) the direct inhibition of pathogen growth and its virulence by secretion of antimicrobial metabolites as bacteriocins, (ii) the modulation of microbiota composition and its activity and, (iii) the stimulation of the host immune system and improvement of intestinal barrier integrity 2 . In this scenario, finding an antimicrobial feed additive with a selective antibacterial activity, high spectrum activity on pathogenic bacteria and a reduced or not effect on beneficial bacteria like Lactobacillus would be very desirable. In the last decade, phytogenic compounds like essential oils (EOs) have received more attention as potential alternatives to replace antimicrobial growth promoters (AGP) in animal production due to their known biological properties: antimicrobial, antioxidant and anti-inflammatory 16,17 . A few studies have reported that some EOs can suppress pathogenic bacteria while stimulating beneficial microorganisms such as Lactobacillus in the pig gut 18,19 . Specifically, the citrus EOs, which are by-products of orange juice production 20 , could be an excellent alternative for that purpose since they have shown good potential to fight pathogenic bacteria such as Listeria spp. 21 , Salmonella spp. 22 , E. coli, Staphylococcus aureus and Bacillus cereus 23 . Furthermore, the use of citrus EOs in animal feed could become feasible since there is a huge availability of these oils in the worldwide market. Therefore, the aim of this study was to evaluate the selective antibacterial activity of six commercial citrus EOs on ETEC strains isolated from pig gut and on two Lactobacillus species belonging to ATCC, as well as to determine the chemical composition of these citrus EOs.

Results
Antibacterial activity. Screening by disc diffusion. The antibacterial activity screening of the six citrus EOs on the ETECs and Lactobacillus species is shown in Table 1. The association of inhibition zone diameter (IZD) means of six citrus EOs, when evaluated on ETECs and Lactobacillus species, by principal component analysis (PCA), showed that the first principal component explained 96.62%, and the second component 2.80% of the total variance (Fig. 1). Therefore, a good representation of antibacterial activity of these citrus oils was obtained. The IZD data showed that all the citrus oils had high antibacterial activity on all the ETECs, while low activity on the two Lactobacillus species was observed (Table 1). Consequently, it is possible to highlight these citrus oils as having a selective antibacterial activity. Furthermore, in contrast to the antibiotic colistin, an antibiotic that presented a selective performance (antibacterial activity on ETECs and no activity on Lactobacillus species), the citrus EOs showed superior performance on ETECs. Looking at the PCA (Fig. 1), four of the six citrus EOs, BOT, OOPE, CT and OPO, were the most selective oils since a closer association of their antibacterial activity with ETECs than with Lactobacillus species was observed. IZDs of these oils on ETECs were >18 mm. However, from these four citrus EOs, BOT stood out by exhibiting the best selective antibacterial activity (p < 0.05), since it presented the largest IZDs to all ETECs and considerable low IZDs (p < 0.05) to the two evaluated Lactobacillus species. Therefore, this oil was selected to continue the study.
Regarding the susceptibility of ETECs, E. coli U7 was the most sensitive strain (p < 0.05) to the activity of the citrus EOs, since the largest IZDs were observed for this ETEC. Conversely, E. coli U21 was the least sensitive or the most resistant ETEC, since were gotten the lowest IZDs (p < 0.05) on this bacterium. In the case of Lactobacillus species, it was observed that L. plantarum was the more resistant beneficial bacterium to the citrus EOs activity, since lowest IZDs were gotten on this bacterium. This behavior of the sensitivity for the bacteria tested is also represented in the PCA (Fig. 1), where the farthest association of E. coli U21 to the EOs is observed, and the closer association of L. rhamnosus than of L. plantarum to EOs can be observed. Therefore, E. coli U21 www.nature.com/scientificreports www.nature.com/scientificreports/ was selected as the most resistant ETEC and L. rhamnosus selected as the most sensitive beneficial bacterium to the antibacterial activity of the citrus oils.

Determination of minimal inhibitory concentration (MIC) and minimal bactericidal concentration (MBC).
The MIC and MBC determinations were made only using the most selective EO, which was BOT, the most resistant E. coli strain (E. coli U21), and most sensitive Lactobacillus species (L. rhamnosus), as determined in the screening phase. The MIC and MBC values for BOT, as determined by survival curves and resazurin test, are shown in Table 2. The MIC for E. coli U21 was 1.85 mg/mL (Fig. 2a), which was also the MBC for this bacterium. The MIC for L. rhamnosus was 3.70 mg/mL (Fig. 2b) and the MBC was 7.40 mg/mL (Table 2). Thus, these results reaffirm the selective antibacterial activity of this citrus oil, since to totally inhibit the growth of the beneficial bacterium L. rhamnosus required an EO concentration equivalent to twice the MIC observed for E. coli U21, and to kill it necessitated a concentration of four times the E. coli U21 MIC. Therefore, pathogenic bacterium was more sensitive to BOT than beneficial bacterium.
Bacterial growth modeling and calculation of kinetic parameters. The curves representing the bacterial growth kinetic for E. coli U21 and L. rhamnosus, when they were exposed to different concentrations of the BOT, are shown in Fig. 2. The Gompertz model modified by Zwietering et al. 24 was used to fit the data of the bacterial growth (Eq. 6). The parameters obtained of this model for both bacteria are shown in Table 3.
The survival curves or growth kinetics of E. coli U21 (Fig. 2a) showed that the four highest EO concentrations totally inhibited the growth of this bacterium and the growth was only observed for concentrations equal to or under 0.925 mg/mL of BOT oil. In the case of L. rhamnosus (Fig. 2b), bacterial growth was observed up to the concentration of 1.85 mg/mL of BOT and the three highest concentrations caused complete inhibition of this bacteria. Furthermore, we observed that EO concentration was able to provoke higher disturbances on the normal growth kinetic of E. coli U21 than L. rhamnosus. The modified Gompertz model allowed us to evaluate these disturbances with more accuracy through the three main biological parameters that it considers: maximal bacterial culture density (A), maximum specific growth rate (µ max ) and lag phase duration or adaptation time (λ).
The values of parameter A were significantly affected (p < 0.05) for E. coli U21, since A was greatly reduced as the EO concentration was increased (Table 3). For instance, A was reduced in 38.9% at the lowest concentration and in 55.9% at the highest subinhibitory concentration (0.925 mg/mL), in contrast with the control (0 mg/mL of BOT). The parameter A had an inverse sigmoidal behavior as function of the EO concentration (Fig. 3a) and the mathematical function that describes this behavior to E. coli U21 is shown in Eq. (1), which had a good fit (R 2 = 0.91). Regarding L. rhamnosus, this parameter was not affected by EO concentrations (Table 3), no significant differences were detected (p < 0.05), and it was considered the average in the general model of this bacterium, which was A LR = 1.191 ± 0.028.     6)). BOT = Brazilian orange terpenes  Table 3. Calculated parameters* of Modified Gomperzt Model (Eq. (6) 24 ) for each evaluated concentration of Brazilian orange terpenes (BOT). *A = maximal bacterial culture density (OD600 nm), µ max = the maximum specific growth rate (h −1 ), λ = the lag phase duration (h). (−)Undetermined parameters due to total inhibition. c Concentration of essential oil. a There are significant differences in the growth kinetics parameters after exposure to essential oil concentrations (p < 0.05). b No significant differences were observed (p < 0.05).
The maximal growth rate, that is, the parameter µ max for both bacteria, E. coli U21 and L. rhamnosus, was not affected by the EO concentrations since no significant differences were detected among the concentrations tested (p < 0.05). Consequently, it was considered an average growth rate of each bacterium for the general model, 0.076 ± 0.014 h −1 to E. coli U21 and 0.189 ± 0.026 h −1 to L. rhamnosus. As observed, the growth rate was quite higher for L. rhamnosus than E. coli U21.
Regarding parameter λ, which is the lag phase duration, we observed that for both bacteria this parameter increased as the EO concentration was increased. For E. coli U21, λ was slightly longer than control at the three lowest EO concentrations, but from 0.463 mg/mL to above, the lag phase was notably increased. For instance, λ increased approximately eight times at the highest subinhibitory concentration of BOT oil (0.925 mg/mL), in contrast with the control (0.00 mg/mL). Therefore, this parameter had an exponential behavior as function of the EO concentration (Fig. 3b) and the mathematical function that describes it is shown in Eq. (2), which had a good fit (R 2 = 0.97). For L. rhamnosus, λ only suffered a significant increase, approximately 1.6 times at the highest effective subinhibitory BOT concentration (1.85 mg/mL). The behavior of λ to this beneficial bacterium was lineal as function of the EO concentration (Fig. 4); the mathematical function that describes it is shown in Eq. (3) and it had a good fitting (R 2 = 0.93). λ = .

LR EO
Finally, the modified Gompertz model to describe the bacterial growth kinetics as function of EO concentration (0.00 < C EO < 14.80 mg/mL) and time of exposure (in hours), is shown in Eq. (4) for E. coli U21, and in Eq. (5) for L. rhamnosus. Additionally, the model for E. coli U21 was plotted in 3D and is shown in Fig. 3c. The surface obtained highlights the shifts in the bacterial growth kinetic of E. coli U21 as function of initial EO concentration.
EO Therefore, evaluation of growth bacterial kinetic parameters confirmed that the citrus oil, BOT, had a stronger effect on the pathogenic bacterium than on the beneficial bacterium, provoking higher disturbances in its growth kinetics. This proves the selective antibacterial features of the BOT oil. chemical composition characterization of essential oils. The chemical composition of the six citrus EOs is shown in Table S1 (Supplementary information). Overall, the identification of the chemical composition of each citrus EO by both columns was quite similar, as is observed in the individual factor map of the MFA (Fig. 5a), the polar and non-polar identification points were very close to each other. In addition, a good representation of the chemical composition data was obtained, since the first dimension of the MFA explained 53.89% and the second dimension explained 33.84% of the total variance. For the six citrus oils, limonene was detected as the major compound; however, TLOP had a negative association with this compound in both dimensions (Fig. 5b), since it presented the lowest relative amount of limonene (46.53%/47.46%).
Moreover, it was observed that BOT, CT, OOPE and OPO had a positive association with limonene in the two dimensions of the MFA and OPOFF had that positive association in the second dimension, showing all these EOs presented higher amount of limonene (Fig. 5b). However, OPOFF presented a fully different chemical profile than BOT, CT, OOPE and OPO oils (Fig. 5a) and it was the oil that presented the weakest antibacterial activity. BOT, CT, OOPE and OPO oils presented a very close chemical composition profile and they were characterized by having the highest selective antibacterial activity. Therefore, this allows us to suggest that the antibacterial activity of the citrus EOs could not be attributed specifically to limonene. Minor common compounds such as cis-limonene oxide, trans-carveol, carvone, trans-limonene oxide, cis-p-Mentha-2,8-dien-1-ol, trans-p-Mentha-2,8-dien-1-ol, perrilla alcohol, cis-carveol, cis-p-Mentha-1(7),8-dien-2-ol and 1,8-menthadien-4-ol (Fig. 5b) were detected in these four citrus EOs and they were either present exclusively in these four EOs or in higher amounts than TLOP and OPOFF oils. Therefore, these minor compounds could be implicated to confer the selective and highest antibacterial activity of BOT, CT, OOPE and OPO oils.
In addition. we observed that TLOP and OPOFF had a different chemical profile. The TLOP profile is shown in the third quadrant of the MFA (Fig. 5b); besides limonene, this oil had a high amount of p-cymene and γ-terpinene. The OPOFF profile is shown in the second quadrant of the MFA (Fig. 5b), and linalool was detected as second major compound in it.

Discussion
Citrus EOs are one of the main by-products of the orange juice industry and could have several applications besides cosmetology and the food industry. Citrus oils have been reported as having antimicrobial properties [20][21][22]25 , but testing their antimicrobial activity on bacteria affecting animal farms has not been greatly mentioned. E. coli causing post-weaning diarrhea in piglets during their early life is a critical problem for the pig industry 2 . The antibacterial activity of commercial citrus EOs evaluated in this study on E. coli strains isolated from pig gut and on Lactobacillus bacteria showed that, in overall, citrus EOs presented a selective antibacterial activity, having higher activity on pathogenic bacteria E. coli than on beneficial bacteria Lactobacillus spp. www.nature.com/scientificreports www.nature.com/scientificreports/ The selectivity towards bacterial pathogens by EOs or EO compounds rather than beneficial bacteria have been already reported 18,26,27 , EOs or single EO compounds were shown to have higher inhibitory effect on E. coli O157:H7, E. coli K88, S. typhimurium than on Lactobacillus and Bifidobacterium spp 18 . As an alternative to AGPs, the selectivity of citrus oils could be an important feature of antimicrobial spectrum, which commonly has not been considered to conventional AGPs long used, since the aim is to have an effect on the pig gut. Thus, current search for antimicrobial substances as alternatives to AGPs should consider selectivity aspects between pathogenic and beneficial gut bacteria. The lower effect on beneficial bacteria by an AGP and its potential alternatives (such as citrus EOs) would be desirable since beneficial bacteria such as Lactobacillus spp. contribute to fighting pathogen colonization in the gut and preventing gut infections. Thereby, reinforcing gut microbiota and contribute to improve animal health 27 .
The capability of citrus EOs to inhibit pathogenic bacteria has been well reported in several studies. For instance, tangerine EO (Citrus reticulata) was reported as having an inhibitory effect on S. aureus, Bacillus subtilis, and Pseudomonas aeruginosa 28 . Another study proved the high effectiveness of the lemon EO (Citrus limon www.nature.com/scientificreports www.nature.com/scientificreports/ L. Burm) to inhibit several strains of Listeria monocytogenes, S aureus and Salmonella enterica associated with foodborne diseases 25 Also, it has been reaffirmed the anti-salmonella 22 and anti-listeria 21 activities of several commercial citrus oils. Specifically, the antibacterial activity of citrus oils on E. coli has been investigated. Tangerine EO (Citrus reticulata) was shown to be effective to produce an inhibition of 14.6 ± 1.1 mm on E. coli 28 . A total inhibition of E. coli growth by this oil was found at 1.96 mg/mL 29 . A close value to this MIC was found for BOT oil in our study (1.85 mg/mL = 0.21%v/v). Other citrus EOs such as bergamot, orange and lemon were also effective to produce inhibitions ≥18 mm on E. coli O157, but full inhibition of the growth of this bacterium was reported at higher MICs than the BOT oil, between 0.5-1.0%v/v 30 . Similarly, another study reported higher MICs than the BOT to mandarin and lemon EOs on E. coli, 5 and 30 µL/mL, respectively 31 . In addition, the EO of sweet orange (Citrus sinensis Osbeck) has presented an inhibitory effect on E. coli at 18.8 µL/mL 32 , also considered as a high MIC in contrast to the BOT oil. Conversely, the non-effectiveness of several citrus oils to fight E. coli affecting animals, such as E. coli associated with poultry colibacillosis, has been reported 33 . In comparison, the citrus EOs tested in our study were quite effective in treating E. coli affecting pigs.
Furthermore, some studies highlighted that citrus EOs have higher effectiveness to inhibit Gram-positive pathogenic bacteria than Gram-negative pathogenic bacteria 30,34 . However, in our study, the opposite was observed, since the Gram-negative E. coli was more sensitive than the Gram-positive Lactobacillus spp. to the activity of the citrus oils. The difference in the sensitivity to EOs between these two groups of bacteria has been hypothesized to be the consequence of differences in the cell wall structure, since Gram-positive bacteria lack an outer membrane (OM), which Gram-negative bacteria have. This OM contains lipopolysaccharides (LPS) with polar ends (O-polysaccharides) and transmembrane channels (porins), which permit the passage of hydrophilic solutes and make difficult for hydrophobic compounds to diffuse such as EOs components into the cell. Therefore, this would allow Gram-negative bacteria be more resistant to EOs 35 . Nonetheless, the antibacterial spectrum of EOs depends on the specificity of the functional groups of EO compounds to single or multiple targets. Some EO compounds have the ability to disintegrate the OM of Gram-negatives as E. coli, release the material associated to this membrane and penetrate the cell, provoking a disruptive effect 36 . Probably, the compounds present in citrus EOs may have this ability due to their higher effectiveness observed on this Gram-negative bacterium, E. coli. On the other hand, the antibacterial activity of citrus oils on Gram-positive beneficial bacteria has been little reported. Orange, lemon, mandarin and grapefruit EOs had a low inhibitory effect on Lactobacillus sakei and Lactobacillus curvatus, exhibiting the orange oil the lowest effect on these bacteria (12.8 ± 0.5 and 13.5 ± 0.2 mm, respectively) 37 . The authors demonstrated that the inhibitory effect of these four oils was dose-dependent causing inhibition of those Lactobacillus species only at the highest concentrations tested 37 . This was also noticed in our study, where the citrus oils exhibited IZDs lower than 11 mm to Lactobacillus species. Moreover, L. rhamnosus was inhibited at a high BOT concentration and killed even at an upper concentration, thus showing BOT had a low antibacterial activity on L. rhamnosus. Although general structures and biosynthesis pathways among Gram-positive bacteria are conserved, some Gram-positive bacteria, such as Lactobacillus spp., could show low sensitivity to antimicrobials, such as EOs, since the cell wall of Gram-positive lactic acid bacteria (LAB) as Lactobacillus spp. possess unique properties that could confer intrinsic resistance to some antimicrobial agents 38 . For instance, the intrinsic resistance to antibiotics of some Lactobacillus (e.g. to vancomycin) would be related to the fact of having a D-lactate instead of D-alanine as the last amino acid in the peptidic chain of the peptidoglycan layer of their cell wall 38,39 , which would avoid the antibiotic binds to the peptidic chain and cause the inhibition of these bacteria 40 .
Additionally, it was observed that BOT oil caused higher disturbances on the growth kinetics of E. coli than L. rhamnosus, significantly affecting its maximal culture density and the lag phase duration. Both parameters were dose-affected and changed as function of the BOT concentration. The higher dose-dependent effect of some EOs and single EO compounds on the growth kinetic of E. coli than on Lactobacillus spp. has already been observed 27 . Oregano, thyme and rosemary EOs, carvacrol, eugenol and thymol provoked higher reduction on the maximal culture density of E. coli strains than Lactobacillus fermentum and Lactobacillus reuteri with increasing of the concentration of EOs/EO compounds 27 . Also, the dose-dependent effect of carvacrol to extend the lag phase of E. coli by increasing the concentration of this compound has been proved 41 . In addition, some combinations of EOs have been reported as more efficient to cause an increase of the E. coli lag phase. For instance, combinations of oregano with basil EOs and oregano with lemon balm EOs were able to significantly increase the E. coli lag phase, approximately 7.4 h and 3.6 h longer, respectively, compared to when oregano EO was used alone 42 . Regarding L. rhamnosus, in our study, we observed only the lag phase duration was extended as the BOT concentration was increased. This effect on L. rhamnosus has been previously observed with the oil of Melaleuca armillaris, which additionally reduced the growth rate and final culture density with increasing of the EO concentration 43 . Therefore, this oil had a higher dose-dependent effect on the growth kinetic parameters of L. rhamnosus than the citrus oil (BOT) tested in our study. Moreover, a recent study observed that Eucalyptus globulus and Pimenta pseudocaryophyllus EOs presented a dose-dependent effect on the lag phase of L. rhamnosus as well; however, P. pseudocaryophyllus oil caused higher extension of this parameter in comparison to E. globulus oil at the same sub-MICs 44 .
Limonene has been shown as a major compound of citrus EO composition and most of their biological activities have been attributed to this compound. All citrus oils evaluated in our study presented limonene as the major compound. However, the mismatching between limonene content and antibacterial activity of these oils suggested that their antibacterial activity cannot be attributed exclusively to limonene. Some studies have already reported the lack of antibacterial activity of limonene individually tested. For instance, the pompia EO (Citrus limon var. pompia), which presented limonene as major compound (28%), at a concentration of 256.3 mg/mL, presented an antibacterial effect on several pathogenic bacteria, but when limonene was evaluated alone, it did not exhibit any antibacterial effect 45 . Thus, this proved that limonene would not be the compound responsible for the antibacterial activity observed for this oil. Nonetheless, coexisting minor compounds in citrus oils could contribute to conferring the antibacterial property of these oils. In mandarin EO, compounds like octanal, decanal, www.nature.com/scientificreports www.nature.com/scientificreports/ citral, citronellal, linalool, α-sinensal and thymol were suggested as possible collaborators to the antibacterial activity, when this oil (with 56.8% of limonene) was tested against Gram-negative and Gram-positive bacteria 28 . Other minor compounds, belonging to oxygenated monoterpenes class, such as 4-terpineol, α-terpineol, cis-geraniol, β-citral, nerol and α-citral, might also be implicated in conferring the antibacterial activity of citrus oils, since they have been detected in high amounts in the composition of the citrus EO that presented high antibacterial activity 25 . Minor oxygenated monoterpenes compounds (cis-limonene oxide, trans-carveol, carvone, trans-limonene oxide and perrilla alcohol) were also detected in the group of the most selective citrus oils of our study. Possibly, these compounds might play an important role in conferring the selective antibacterial activity of citrus EOs. In addition, an orange cold pressed EO rich in limonene (85.3%) presented an antibacterial activity on E. coli ten times higher than limonene alone, and even minor compounds, such as linalool, pinene and terpineol, presented a higher activity than limonene 46 . Furthermore, the antibacterial activity of limonene has been shown to be variable and depending on its stereoisomeric form present in the EO. The (−) stereoisomer of limonene could inhibit E. coli at a lower concentration (8 mg/mL) than the (+) stereoisomer (11 mg/mL). On the other hand, limonene alone has been proved to stimulate the growth of beneficial bacteria as L. fermentum, instead of having any inhibitory effect on it 27 . Likewise, limonene has been reported as not effective to inhibit several Lactobacillus species including L. rhamnosus ATCC 7469 45 , the bacteria also evaluated in our study, and which showed be the more resistant to the antibacterial activity of the citrus oils proved. Therefore, it would be possible to infer that limonene could collaborate with the selective activity of citrus oils when it is present in the gut, promoting the beneficial bacteria while other minor compounds could act in inhibiting pathogenic bacteria. It has been reported that in an EO, major and minor components probably act in synergism to confer the biological properties of the EO 47 . When an EO compound is proved, individually, its effect may differ from the effect that this compound may have in combination with the other compounds inside the EO. Thus, it would be recommended the use of the whole EO instead of single EO compounds, since every compound inside an EO could exert a different mechanism of action on the bacteria cell 48 , and this could reduce the chance to bacteria develop easily resistance to the EO. Contrariwise, bacteria could develop a rapid and easy mechanism of resistance to a single EO compound, as in the case of an antibiotic, which consist of a single compound. Several mechanisms of action of EOs have been proposed in the literature. The mechanism of action comprises a serie of events on the bacterial cells. Initially, they can destabilize the cellular architecture, leading to the breakdown of membrane integrity and thus increased permeability of cellular constituents. This disrupts many cellular activities, including energy production, membrane transport, and other metabolic regulatory functions 48 . In addition, EOs can alter the membrane fatty acids composition, the membrane proton motive force and affect proteins in the cytoplasmatic membrane. Additionally, EOs can interfere with the quorum sensing activity, decreasing the proteolytic activity, biofilm formation, and virulence factors expression and their functions, as well as to affect the metabolome of bacteria 35 .
In conclusion, our study highlights as an important feature of antimicrobial spectrum the selectivity between pathogenic and beneficial gut bacteria, which should be considered when searching for antimicrobial substances as alternatives to conventional AGPs, since the aim is to have an effect on the pig gut. Our study has proved, by a screening, MIC determination, and growth kinetic parameters evaluation, the selective antibacterial activity of citrus EOs on E. coli and Lactobacillus spp., thus suggesting these EOs as potential alternatives to AGPs. Consequently, based on the selective performance and the huge viability in the global market of citrus EOs, the possible application of these oils in the pig production sector could turn feasible. In addition, chemical composition characterization showed that minor compounds present in these citrus oils would be implicated in conferring their selective activity, instead of limonene, the major present compound, playing this role exclusively. Finally, our results motivate further research to clarify, for instance, the possible mechanism of action that citrus oils would have on pathogenic and beneficial bacteria as well as their direct effect on the pig gut and on the microbiota resident in it.  www.nature.com/scientificreports www.nature.com/scientificreports/ bacterium were transferred to tubes containing sterile saline solution (0.85%) until reaching an optical density within 0.08 to 0.1 abs, at 625 nm, which corresponds to 0.5 McFarland standard, therefore containing ~1-2 × 10 8 CFU/mL 49 . After this, Mueller Hinton (MH) agar plates (E. coli) and MRS agar plates (Lactobacillus spp.) were inoculated and the bacterial inoculum spread. Seven microliters of each EO solution (90% v/v) were placed on 6-mm diameter sterile paper discs (Whatman N° 3), which was transferred to the inoculated agar plates. Three discs with the same EO solution were placed in each plate; one disc of colistin (15 µg/disc) was used as a positive control, and one disc of acetone (10 µL/disc) was used as a negative control since its non-antimicrobial activity was proved. Then, the agar plates were incubated at 37 °C for 24 h (E. coli) and at 30 °C for 48 h (Lactobacillus spp.). Inhibition zone diameters (IZD) were measured after incubation with the aid of a caliper rule. The experiment was carried out in three independent replicates.

Material and
The EO presenting the highest IZDs for the four ETECs strains and the lowest IZDs for the two Lactobacillus species was considered as the best selective EO between pathogenic and beneficial bacteria and was thus selected for further investigation. Referring to the bacterial strains, the most resistant ETEC strain and the most sensitive Lactobacillus specie to the activity of EOs were selected to continue with the study.

Determination of minimal inhibitory concentration (MIC).
The determination of the MIC of the selected EO was performed by microdilution assay in a 96-well microplate following the standard protocol M07-A9 from the Clinical and Laboratory Standards Institute, with some modifications 50 . For the assay, the standard inoculum was prepared in sterile saline (0.85% w/v) from living colonies of the selected bacteria above contained in plates of TSA agar (E. coli) or MRS agar (Lactobacillus spp.) at the optical density equivalent to 0.5 McFarland Standard (0.08-0.13 at 625 nm) as described previously. Subsequently, this inoculum was diluted at 1:100 to obtain an inoculum of 10 6 CFU/mL (final inoculum). The EO stock solution was prepared at 29.6 mg/mL (3.29%) with MH or MRS broth using Tween 80 as emulsifier. From the stock solution, two-fold serial dilutions were made in a range from 14.80 to 0.116 mg/mL along the Y-axis of the microplate. Twenty microliters from the final inoculum were added to each well containing 180 µL of several EO concentrations, being the final volume in each well of 200 µL and bacterial population of approximately 10 5 CFU/mL. The following controls were used: culture medium control (200 µL of MH or MRS broth); growth control (180 µL of MH or MRS broth +20 µL of inoculum); Tween 80-emulsifier control (200 µL of MH or MRS broth with Tween 80) and growth control containing the emulsifier (180 µL of MH or MRS broth with Tween 80 + 20 µL of inoculum). Finally, microplates were incubated in a microplate reader (Vitor TM X3, PerkinElmer) at 37 °C for 24 h for E. coli and at 30 °C for 36 h for Lactobacillus spp.
The MIC was established as the lowest EO concentration that inhibited visible bacterial growth. The existence or not of bacterial growth was evaluated by construction of survival curves and by resazurin test at the end of the incubation period. The lowest concentration that did not produce detectable absorbance values (at 600 nm) until the end of incubation was considered as the MIC obtained by survival curves. For resazurin test, 25 µL of resazurin (R7017; Sigma-Aldrich) solution at 0.0135% m/v were used per well. Thus, after visual inspection the presence of viable cells was evidenced through a change in the resazurin color from blue resazurin to pink resofurin 51 , after further incubation at 37 °C (E. coli) or 30 °C (Lactobacillus spp.) for 1 h. Assays were carried out in triplicate in three independent replicates.
Bacterial growth modeling and calculation of kinetics parameters. Bacterial growth kinetics (or survival curves) for each tested EO concentration was built from absorbance readings at 600 nm of the wells of the microplate configured as above, carried out every hour during the total incubation period, 24 h for E. coli and 36 h for Lactobacillus spp. Bacterial growth kinetics were modeled using the Gompertz model modified by Zwietering et al. 24 (Eq. (6)), since this model considers the three main biological parameters of bacterial growth. The data were fitted to the mathematical model with a confidence level of 95% using the Levenberg-Marquardt algorithm in Statistica 12.0 (StatSoft, Inc., Tulsa, OK, USA) software.
max where: y represents the relative population size against time, the A, µ max and λ are the three parameters that described three phases of the bacterial growth curve 24 . The asymptote A is the maximal bacterial culture density (OD 600 nm ), µ max represents the maximum specific growth rate (h −1 ) and it is the tangent of the log phase curve, λ is the lag phase duration (h) and is defined as the x-axis intercept of this tangent. e represents the number e = 2.7183. Finally, the goodness of fit for the model was measured based on the mean square error (MSE) and on the corrected determination coefficient (corrected R 2 ) for each set of data.
Minimal bactericidal concentration (MBC). The determination of MBC was performed from wells containing EO concentrations where there was no visible bacterial growth. So, an aliquot of 100 µL was taken from each well and seeded in MH or MRS agar. Plates were incubated for 24 h at 37 °C for E. coli and for 48 h for Lactobacillus spp. The MBC was defined as the lowest concentration of EO able to cause total bacterial death, represented by the visible absence of colonies on the agar plates. chemical composition of essential oils. The chemical composition characterization of the EOs was performed by gas chromatography coupled with mass spectrometry (GC/MS) using non-polar and polar columns.
The analysis on non-polar column was carried out using an Agilent Technology gas chromatograph 6890 Plus Series (Santa Clara, CA, USA) coupled to a selective Mass Spectrometry Detector 5973 and an Auto Sampler 7893. www.nature.com/scientificreports www.nature.com/scientificreports/ film thickness coated with 5%-phenyl polydimethylsiloxane was used. The oven temperature was set as follows: initial oven temperature was held at 45 °C for 5 min, then raised to 150 °C at 4 °C/min for 2 min, one more time raised to 250 °C at 5 °C/min, and finally to 300 °C at 10 °C/min, which was kept for 60 min. The injector temperature was 250 °C, 2.0 µL of samples diluted in dichloromethane was injected in the "split" mode at a ratio of 30:1. EIMS, electron energy was 70 eV. The mass detector operated in full scan mode in the range of 40 to 350 m/z. The temperature of the ion source and transfer line was 230 °C and 285 °C, respectively. Helium gas was used as the carrier gas with an inlet pressure of 16.97 psi. The retention index (RI) was calculated for all the volatile compounds using a homologous series of C7-C30 n-alkanes (49451-U Sigma-Aldrich), according the linear equation of Van den Dool and Kratz 52 .
The analysis on polar column was carried out using an Agilent Technology gas chromatograph 7890 a Plus Series (Palo Alto, CA, USA) coupled to a selective Mass Spectrometry Detector 5975 C. A fused-silica capillary column DB-WAX (J&W Scientific, Folsom, CA, USA) of 60 m × 0.25 mm i.d × 0.25 μm of film thickness coated with polyethylene glycol was used. The oven temperature was set as follows: initial oven temperature was held at 50 °C for 5 min, then raised to 150 °C at 4 °C/min for 7 min and finally to 230 °C at 4 °C/min, which was kept for 40 min. The injector temperature was 250 °C, 2.0 µL of samples diluted in dichloromethane was injected in the "split" mode at a ratio of 30:1. EIMS, electron energy was 70 eV. Helium gas was used as the carrier gas with an inlet pressure of 16.91 psi. Mass detector operated in full scan mode in the range of 40 to 350 m/z. The retention index (RI) was calculated for all the volatile compounds as described above.
The identification of the components was performed by comparing their RI and mass spectra with data published in the literature 53,54 and in the computer libraries (NIST 107 and WILEY 8).
Data analysis. The IZD data from the initial screening was evaluated by analysis of variance (ANOVA) followed by the Tukey test for pairwise comparison at 5% of significance using XLSTAT (Addinsoft, New York, NY, USA). Based on this, the susceptibility of ETECs strains and Lactobacillus species was determined. Also, principal component analysis (PCA) using the correlation matrix was performed based on IZDs means using XLSTAT (Addinsoft, New York, NY, USA).
Furthermore, an ANOVA to detect significant differences in the growth kinetics parameters A, λ and µ max of E. coli and Lactobacillus spp. after exposure to EO concentrations was performed (p < 0.05) using R software. If significant differences were detected in those parameters, as effect of the EO concentrations, the behavior of each parameter was modeling by nonlinear regression.
In addition, the multiple factor analysis (MFA) was performed on polar and non-polar data (two tables) to describe and contrast the chemical composition profile of the six citrus EOs obtained by GC-MS, with this analysis running in the XLSTAT software.