Results of QTL detection are dependent, amongst other parameters, on the statistical method used for analysis. In this study, we used maximum likelihood analyses with the QTLMap software , whereas regression analysis with the QTLExpress software  was used in our previous studies [3, 4]. The two approaches were expected to give partially different results. Although we used the same data as previously described , genotypes were enriched with eleven SNP markers flanking QTLs for resistance to Salmonella colonization.
Numerous QTLs identified using QTLMap
QTLExpress assumes the F0 lines to be inbred, particularly with the F2 inbred option of QTLExpress used previously . This contrasts to QTLMap which takes into account the alleles segregating within the lines, both between and within sire families. This is of particular importance since the N and 61 lines are highly but not totally inbred, possibly accounting for the reason why more genome-wide QTLs were found in this study than previously observed . Nadaf et al. achieved a similar result with the same two software packages . They analyzed a F2 cross between lines largely divergent for a phenotype of growth, but not fixed for the QTL alleles. They obtained similar results: with QTLMap, they found 5 genome-wide and 4 chromosome-wide significant QTLs, while with QTLExpress there were only 3 and 2 respectively. Two of the QTLs found to be of chromosome-wide significance with QTLExpress were estimated as genome-wide significant with QTLMap. As suggested by the authors, these differences probably originate from the allele segregation within each line. In our case, although segregation was expected to be less frequent since the lines were partly inbred, QTLMap allowed identification of more QTLs with higher significance.
Numerous QTLs were detected with the none vs two QTLs analysis. The latter deals with the possibility that two linked QTL located within a very short interval could not be detected because they are exerting antagonistic effects. Except for CSW4 on GGA5, for the other traits and chromosomes the distance between the two QTLs was rather short and ranged from 7 to 28 cM. Moreover, on several chromosomes, QTLs controlling different traits were observed in close proximity, which supports the reliability of our results. For instance, on chromosome 9, both QTLs controlling CAEC (46 cM; 56 cM) were very close to those involved in the control of CSW5 (45 cM; 52 cM). On microchromosome 22, two pairs of QTLs were identified for CAEC (0 cM; 28 cM) and for CSW5 (29 cM; 42 cM). On chromosome 24, one of the QTLs controlling CAEC and one of the QTLs controlling CSW4 co-localized at 51 cM. All these co-localisations suggest that these regions may indeed have an impact on the control of Salmonella carrier-state.
QTLMap software makes it possible to test binary data, such as the presence/absence of bacteria. The distribution of such traits is far from being Gaussian, while normality is assumed for most methods of QTL research. In this study, this question was of special interest for cloacal swabs at 4 and 5 weeks p.i., for which only 41.6 % and 33 % of animals respectively were positive for the presence of bacteria. When analyzing the binary trait CSW4d, with the model dedicated to discrete data, only one significant QTL was found on chromosome 23 at 1 cM, close to the QTL detected for CAEC at 10 cM. This QTL, although highly significant at the chromosome-wide level, was not significant at the genome-wide level. Another QTL was detected on chromosome 21 with CSW4d after deleting the three sire families which had the lowest likelihood ratio. It was only significant at the chromosome-wide level, most probably because of a lack of power due to the loss of information when categorizing the data. Since this QTL co-localizes with QTL identified for CSW4 with the one-QTL analysis, it is probably not a statistical artefact.
Comparison with previous results obtained with QTLExpress
Although a high number of QTLs were identified with each of the software packages, only one QTL on chromosome 22 was identified with both of them. On this chromosome, the QTL for CSW5 identified at 29 cM by the two-QTLs analysis is located at the same position as the QTL found by . We could not identify the dominant QTLs observed by  with QTLExpress, since QTLMap does not take dominance effects into account. In particular, for the most significant QTL detected on chromosome 14 for CAEC, dominance effects were very high (d = 0.90) while additive effects were very low (a = −0.11), with a ratio d/2a equal to −4.25. This is probably why it could not be detected by the additive model with QTLMap. Moreover, some of these dominant QTLs could also be false positives as observed by . When analyzing, with a large set of methods, a set of simulated data with only additive QTLs, several dominant QTLs were observed . Conversely, using QTLMap we detected a QTL on chromosome 2 at 86 cM, which was not observed previously  using the same set of SNP markers with QTLExpress (except the 3 additional markers close to QTLs observed by Fife et al. ). However, a QTL for CSW4 was previously detected at 87 cM using microsatellites [2, 3]. The lack of detection of this QTL in the previous QTLExpress analysis  could be due to the poor information content of several sire families at markers surrounding the QTL region, which is best taken into account by the QTLMap software. The map density and the information content of markers are lower in this region of chromosome 2 than in other regions. For all other QTLs, it must be assumed that either they are statistical artefacts or they display effects so weak that they are significant with only one of the two methods of analysis tested.
Similarly to Calenge et al., we identified a very low number of QTL in P2. This may be as a result of the lower number of traits measured (one versus three). However, even when considering only the trait CAEC measured in P2, there were fewer QTLs detected in P2 than in P1. This might be related to the phenotypic distributions, which are highly different in P1 and P2. In P1 most animals were free of Salmonella (76.2 %), while in P2 most animals were infected (92.3 %). Despite the fact that, in both P1 and P2, the logarithm of numbers of colony forming units was measured, the biological meaning of both traits differs. The former trait is related to capacity for bacterial clearance: more resistant animals are free from bacteria while the susceptible individuals remain infected. The latter refers to the level of infection: the more resistant animals can maintain the infection at a lower level. Mechanisms underlying both traits may thus be quite different and the same holds for the genes controlling them .
Are Salmonella carrier-state and colonization controlled by common loci?
Although most QTLs identified in the present study were not identified in the previous QTLExpress analysis, some of them co-localize with QTLs or candidate genes identified in other, independent studies, which strengthens their interest and reliability.
The QTL on chromosome 3 is of particular interest. Although it was not significant at the genome-wide level, it was the more significant of the two QTLs detected in P2. In addition, it co-localizes with a QTL for resistance to Salmonella colonization identified at 96 Mb (vs 94.5 Mb) for the same measure (caecal bacterial count) assessed after infection with another Salmonella serotype at an earlier interval post-inoculation (5 days p.i.) . Interestingly, when SNP markers surrounding the QTLs for resistance to colonization were added to the genetic map, the significance of the QTL for resistance to carrier-state reaches the genome-wide significance level. This might be due to the additional information brought by these markers, one of which may be close to the causal gene for the QTL. Nevertheless, two of the three additional SNP markers were in strong segregation distortion, and it therefore cannot be excluded that skewed segregations cause an artificial increase of the QTL significance. Indeed, one of the distorted markers was mapped at an incorrect position when considering physical positions of markers (inversion with an adjacent marker). However, this result is coherent with the co-localisation of this QTL and several candidate genes involved in innate immunity like the avian beta-defensins AvBD1 to AvBD14 between 110.20 Mb and 110.27 Mb. Moreover, polymorphisms in four genes within the beta-defensin cluster (AvBD3, 11, 12 and 13) were associated with caecal bacterial load in chickens orally infected with S. Enteritidis, while AvBD5 was associated with spleen bacterial load [17, 18]. Other genes like the interleukins IL-17A and IL-17 F are also located in close proximity to the QTL (at 110.36 Mb and 110.37 Mb respectively) and were associated with caecal and spleen bacterial loads, as well as antibody response to Salmonella Enteritidis vaccine [17, 18]. The role of this immune pathway in resistance to salmonellosis is strengthened by the detection in chromosome 2 of RFTN1, which modulates T cell receptor signals and which is necessary for the fine-tuning of T cell-mediated immune responses and especially the Th17 immune response .
Numerous genes related to immunity can be identified between the markers flanking the QTL present on chromosome 2 in P1, which is located adjacent to a QTL associated with the presence of a hardened caseous caecal core after infection with Salmonella Typhimurium . Although the QTL peaks are located 10 Mb apart and their confidence intervals do not overlap, QTL locations are known to vary according to many parameters. In this meta-analysis, infection protocols and Salmonella serotypes were different, as were the parents of the progenies. Although belonging to the same lines, separation of these flocks may have, over time, resulted in divergence, as strongly suggested by the incomplete segregation of some of the new SNP markers added to our original SNP set. The number of markers used in both of these studies and the strong linkage disequilibrium in these mapping populations may also impact on the resolution of the QTLs. It is therefore possible that both QTLs for these disparate traits do co-localize. Moreover, the gene IL-6 (interleukin-6), whose product drives induced innate responses in chickens and especially the Th17 pathway , is located close to the QTL peak (30.9 Mb with the QTL peak at 38.5 Mb). It is interesting to note that numerous genes identified between the markers flanking the peaks of the QTL present in chromosome 2 are involved in the immune response. Some genes play a role in the Toll-like receptor (TLR) signaling cascade, such as the TLR4 interactor with leucine-rich repeats (TRIL) gene, known to interact with the TLR4 protein. TLR4 is involved in stimulation of immune responses after interaction with bacterial lipopolysaccharide ,. Moreover, higher TLR4 expression has been observed in caecal cells from an uninfected resistant chicken line compared to those from the susceptible line . Other genes are related to stimulation of this innate immune response pathway, such as TAX1BP1, a key regulator of the NF-κB and IRF3 signaling and involved in anti-apoptotic activities . Interestingly, other genes located on other chromosomes and close to QTL are also involved in these immune signaling cascades. This is the case for TRAF3IP3, which interacts with TRAF3, a regulator of the JUN N-terminal kinase (JNK) and NF-κB which is a transcriptional factor important for interleukin production . Similarly, HIVE3, a member of the ZAS family, which interacts with TRAF1 and TRAF2 to regulate IL-2 expression, was also detected .
Surprisingly, numerous genes identified between the markers flanking the peaks of the QTL are involved in cell signaling leading to cell migration and cell proliferation and thus involved in the immune response. This is the case of the miRNA mir148A which promotes cell proliferation and cell cycle progression by targeting p27, a key inhibitor of the cell cycle. The roles of the other miRNAs (mir1732, mir196-2, mir 205a) detected in our study are not known. However, very recent studies in mice suggest that miRNAs are involved in the specific host response to bacterial pathogens such as Salmonella.. In a similar way, SKAP2, which is an adaptor protein, plays an important role in cellular functions, such as cell proliferation, migration, and apoptosis . It is important to note that several genes flanking the peaks of our QTL and involved in cell proliferation/migration signaling pathways belong to the Ras GTPases superfamily, also called small GTPases. These small GTPases generally serve as molecular switches for a variety of cellular signaling events but are also modulated by Salmonella to drive its entry into cells [26, 27]. The Rab family, a subfamily of the Ras GTPases superfamily, regulate many steps of membrane traffic, including vesicle formation and vesicle movement along actin and tubulin networks. These proteins are also crucial for intracellular survival of Salmonella. Two genes (RAB5A and RABEP1) corresponding to different QTL, located on different chromosomes, have similar function and belong to the same signaling cascade. Rab5A, indeed, interacts with RABEP1 . Rab5a is a key molecule for the IFN-γ promoted clearance of pathogenic bacteria but it has been also described that the Salmonella effector protein SopB promotes phosphatidylinositol 3-phosphate formation on Salmonella vacuoles by recruiting Rab5. Within this signaling cascade we could also introduce PLC2 which is a phospholipase C involved in phosphatidylinositol 3-kinase signalling. CHN2 or beta2 chimaerin is a member of the RhoGAP protein family, another Ras GTPase subfamily, showing specific GTPase-activating protein (GAP) activity toward Rac which induces the formation of actin-rich lamellipodia protrusions involved in cell migration ; ANKRD28 (a protein that contains twenty-six ankyrin domain repeats) interacts with DOCK180, a guanine-exchange factor of Rac, to promote cell migration , and NKIRAS1 is a small GTPase modulating NF-κB activity . Further investigations are needed to elucidate their putative role in the control of resistance to Salmonella carrier-state in these populations, with reference to the traits measured in this study.
These co-localisations on chromosomes suggest that these regions may be involved in general mechanisms of resistance, acting on both colonization and carrier-state and effective on both serotypes (Enteritidis and Typhimurium). This result is not surprising: limiting bacteria colonization shortly after infection could be one of the ways to limit bacteria carriage several weeks after infection. Another way of limiting carriage could be, for instance, high bacterial clearance ability. Genes involved in the immune response are interesting candidates for the causative genes underlying these QTLs and warrant further research. Nevertheless, due to the high number of putative candidate genes, a finer mapping of the QTL will be necessary before studying the actual implication of one or several of these genes in resistance to carrier state.
On chromosome 5 the QTL identified at 104 cM for CSW4 in the two-QTL analysis is close to the QTL detected by Tilquin et al. at 111 cM for CSW5 . In addition, similarly to both Tilquin et al. and Calenge et al., we observed a suggestive QTL for CAEC at the same location (108 cM). Nevertheless, these QTL effects are too weak to be considered of interest for further research.
Even with QTLMap, no QTL could be detected on chromosome 7 close to the SLC11A1 gene (previously named NRAMP1), which was shown to be involved in the resistance to both acute salmonellosis ‐ and Salmonella carrier-state . This result is consistent with previous observations ‐. It can be assumed that either this gene does not segregate in this cross or its effect is too weak to be detected.