Using reference genes that have a stable expression between the compared groups is crucial in gene expression studies. Several studies have shown that the use of different reference genes can change the outcome and conclusions of a study [21–25]. The aim of the present study was therefore to validate, for the first time, reference genes for studies in BALF of horses with IAD, irrespective of treatment with corticosteroids. We found that using GAPDH, HPRT, SDHA and RPL32 as a combination of reference genes is the most appropriate normalization approach in this experimental design and that GAPDH is the single most stably expressed gene in the BALF of horses treated with corticosteroids. Although the present study only used 7 horses, which is less than typical studies using mice or human samples, it is comparable to other clinical studies in equine medicine using PCR on 3 to 10 horses [17, 19, 20, 26, 27]. Compared to other techniques measuring gene expression, the PCR technique is better suited for samples of smaller size and the number of horses used here allowed obtaining significant findings that are relevant for studies in the field of equine medicine.
To our knowledge, there are only two studies evaluating the stability of reference genes in horse tissues, one being in the skin  and the other being in the peripheral blood ; no data is available on the stability of reference genes in the lungs of horses to compare our results with. However, Cappelli at al. found GAPDH as the least stably expressed gene in the panel of candidate reference genes they tested in equine blood lymphocytes during exercise-induced stress . This emphasizes the importance of appropriate reference gene validation for every tissue and experimental protocol, even when using the same species. The discrepancy is probably due in this case to the difference in tissues tested as well as to the effect of corticosteroids on cellular metabolism. In addition, a few studies assessed the validity of proposed housekeeping genes in the bronchoalveolar cells of humans with various pathologies [21, 28, 29]. Using a different method than reported here, a study found that GAPDH was the most stable reference gene in the bronchoalveolar samples of people with nonsmall cell lung cancer . Another study used Genorm to test candidate housekeeping genes that were mostly different from those described here and found that GNB2L1, HPRT1 and RPL32 were the most stably expressed genes in alveolar macrophages from 22 subjects with chronic obstructive pulmonary disease (COPD) ; they also described that GAPDH was inappropriate for these studies. In agreement, one study found that GAPDH and ACTB were not suitable as reference genes in asthma and illustrated it by showing that the use of GAPDH vs ACTB as reference genes would lead to conflicting results . Lastly, a study used equivalence test as well as the statistical tools BestKeeper, GeNorm and NormFinder to assess the most suitable housekeeping genes in the lungs of a large number of people (2 cohort studies) irrespective of gender, smoking, lung pathologies, treatments, and BAL cytology . This study only shared 3 reference genes with the data presented here, but found that only RPL32 along with the proteasome subunit 2 (PSMB2) were stably expressed among BAL samples . In this sense, RPL32 came as the second and third most stably expressed gene in the NormFinder and GeNorm analysis presented here.
Numerous methods are available to validate reference genes for relative quantification by QPCR. Studies will often use only one method, but one study compared two reference genes validation methods in horse samples  and another compared four methods in humans BAL samples . Similarly, to ensure consistency and for comparison purpose, the data was analyzed here using three different softwares (GeNorm, NormFinder and qBasePlus). GeNorm and qBasePlus use a pairwise comparison model, while NormFinder uses a model-based approach. In our study, GeNorm 1) identified GAPDH, SDHA, HPRT and RPL32 as the most stably expressed reference genes by calculating a stability parameter (M) (see methods) and 2) determined that the optimal number of reference genes to be used was 4 (GAPDH, SDHA, HPRT and RPL32), by calculating pair-wise variation (V values) (see methods). NormFinder defined GAPDH as the best reference gene when using the treatment with corticosteroids as group identifiers to calculate a stability value for each candidate reference gene. NormFinder takes into account variation across subgroups, thus avoiding artificial selection of co-regulated genes by analyzing the expression stability of candidate genes independently from each other . Lastly, qBasePlus confirmed that GAPDH is the best reference gene in our study design. Similarly to previous studies [19, 30], we found a good agreement in the reference genes ranking between GeNorm and NormFinder as they both ranked GAPDH as the most stable expressed gene, which was confirmed by the qBasePlus analysis. We also found that the first three most stable reference genes were consistently the same when using GeNorm and NormFinder, even if they were not in the exact same ranking order. There was a slight difference in the top four most stably expressed genes as the four most stably expressed genes ranked by NormFinder were, in decreasing order, GAPDH, RPL32, HPRT and B2M, while they were GAPDH, SDHA, HPRT and RPL32 with GeNorm. Very similar discrepancy between the different algorithms has been observed in other studies comparing statistical analysis methods: The only study using horse samples and comparing GeNorm with NormFinder found also that the best three reference genes were ranked differently by the two algorithms and that there was disagreement on the fourth most stable gene . Another study using BAL samples on a large cohort of human patients also found different ranking order and genes identification for the top four most stable reference genes . Such discrepancy could be explained by genes' co-regulation. Indeed, co-regulated genes may become highly ranked independently of their expression stabilities with GeNorm software . In contrast, results obtained with NormFinder are not significantly affected by co-regulation of candidate reference genes. In our study, the main difference in ranking involved SDHA, which ranked as the fifth most stable gene with NormFinder, but ranked second with GeNorm. Although co-regulation has been described for ACTB, B2M, GAPDH and HPRT, we did not find evidence for the possible co-regulation of SDHA in the literature. To check for possible co-regulation of SDHA, we analyzed the data again with GeNorm, this time excluding SDHA from the candidate reference gene panel. We found that removing SDHA from the analysis did not resolve the discrepancy in ranking of the candidate reference genes between NormFinder and GeNorm (data not shown). We thus concluded that the discrepancy in ranking was not caused by co-regulation of SDHA.
ACTB and UBB were ranked by both softwares as unstably expressed genes and therefore should not be used as reference in gene expression studies in bronchoalveolar lavage (BAL) cells obtained from horses with IAD, which is similar (for ACTB) as another study using humans BAL samples . It is however in contrast with a study done in normal horses' skin in which ACTB and UBB came out as the most stably expressed genes from the panel tested with GeNorm . GAPDH was however not evaluated in this previous study . This shows again that the choice of reference genes cannot be transposed from on study to the other without validation for the specifics of each experimental protocol.
As described above, GeNorm also provides a measure for the best number of reference genes that should be used for optimal normalization. In agreement with several previous studies, we found that the use of more than one reference gene allows for a more accurate normalization than the use of only one reference gene [10, 16, 31]. Based on a cut-off point of 0.15 for the V value, as described by Vandesompele et al , a combination of the four most stable reference genes was calculated as being optimal for gene expression studies in BAL cells of horses with IAD treated with corticosteroids (figure 4).