Quantitative measurements of gene expression are increasingly important in understanding biological processes and research in general. Knowledge of the expression profile of a gene can, for instance, provide evidence about its regulation and its function. With genomes of several organisms already sequenced and much more on the way, researchers are able to use this information to elucidate the transcription of a gene in a relatively straightforward manner. Using the quantitative real-time RT-PCR (qPCR) technique the expression levels of a gene can be investigated in different cells, tissues and organisms and in different conditions during development or over a preferred period of time. qPCR is also widely used to verify microarray datasets or the knockdown of a gene in RNA interference experiments and is of great value in disease diagnostics 123. The analysis of qPCR data requires normalization relative to an active reference or endogenous control, which compensates for differences in sample preparation, cDNA and DNA synthesis and in the amount of the starting material. Such an internal control gene ideally has an equal transcript level in all cells at every developmental stage and is unaffected by experimental conditions. Traditionally, it is assumed that housekeeping genes (HKGs) meet these criteria, since they are necessary in every cell. Nowadays, in transcript analyses any gene with a seemingly constitutive and stable expression level is defined as a housekeeping gene. In practice, however, stable gene expression occurs only rarely and it was even suggested that such genes do not exist 4. In line with this, it was shown that transcript levels, normalized to a single HKG, can differ more than 20-fold from the actual expression 5. To circumvent this problem tests to validate the stability of HKGs and the use of multiple genes are supported. To this end, different software programs were developed to make a selection of housekeeping genes that are most suited for normalization 467.

Surprisingly, in almost all transcript profiling studies in insects so far, analyses to validate the choice of a set of HKGs have been overlooked. In this study we identified and examined seven HKGs in brain tissues of desert locusts (Schistocerca gregaria) during the last molt and the reproductive cycle. The aim of this study was to assess which of these were the most stable and therefore represent the best choice for qPCR experiments. To this end the freely available normalization programs geNorm 4 and Normfinder 6 were used. The selected genes were orthologs of the commonly used actin (Sg-ACT) 8 (and references therein), RP49 (Sg-RP49) 9 (and references therein), GAPDH (Sg-GAPDH), ubiquitin (Sg-Ubi) and EF1a (Sg-EF1a) genes, supplemented with the orthologues of TubA1 (Sg-TubA1) and CG13220 (Sg-CG13220). For the remainder of the text the prefix Sg will be omitted.

Using the database of an ongoing EST project we characterized seven housekeeping genes in the genome of S. gregaria (Table 1; Additional file 1); protein sequences of commonly used HKGs were blasted against the translated EST database of S. gregaria. In addition, TubA1 and CG13220 were selected from Flyatlas, since they showed invariant expression across the different tissues in Drosophila melanogaster 10. Primers were then designed and tested (Table 2). Only primer pairs with efficiency values (E) between 95 and 105% were used. This resulted in seven primer sets for the analysis of ACT, EF1a, GAPDH, RP49, TubA1, Ubi, and CG13220 as potential reference genes.

GeNorm and Normfinder were used to evaluate the stability of the HKGs in different brain samples taken during the fifth nymphal stage and/or the adult stage. For each control gene, geNorm calculates the pairwise variation with all other control genes and the gene-stability measure, M, is defined as the average pairwise variation. Consequently, genes with a low M value have a low variation and therefore, a stable expression. Then, in a stepwise manner it eliminates the gene with the highest M value and recalculates the M value of the remaining genes, eventually yielding the two most stable genes. This is represented on a chart, showing from the left (all genes included) to the right (two genes) the mean of the M values of all remaining genes. In addition, geNorm calculates a normalization factor for every sample. Normfinder, on the other hand, uses a model-based approach. It calculates a stability value based on the intragroup variance, and includes the intergroup variance if applicable. This program selects the genes with the least expression variation over the samples and takes into account systematic differences between sample subgroups.

In recent years it has become clear that the accuracy of quantitative RT-PCR and microarray analyses depends strongly on the choice of the normalization genes. Numerous studies have already been performed in the search for good HKGs in a wide variety of species and tissues 1112131415. To our knowledge this is one of the very first reports on a methodical housekeeping gene analysis in insects 1617. In this study we identified seven putative HKGs from the desert locust Schistocerca gregaria. To determine the most stable genes in adult and fifth larval stage brains, samples were taken at strict time intervals covering developmental changes from the last molting cycle to the transition from juvenile to sexually mature adults and data were analyzed by two different software programs.

A first difference between the two programs is the use of a stepwise exclusion process by geNorm. This stepwise elimination of the least stable gene is helpful because of the relative nature of the determination of variability; the M value of every gene is interdependent on every other gene. This means that the ranking based on the M values (and not the AESM scores) can change after the stepwise elimination of a gene, especially when the selected genes are all relatively stable.

The geNorm principle is based on the assumption that two ideal HKGs have identical expression ratios regardless of the conditions. The software thus provides the two genes that have the most similar expression profile throughout the samples. It does this in respect to all other genes included in the survey (e.g. when two couples of similar genes are present, the two genes will be chosen that resemble the additional genes most closely). This implies, however, that co-regulated genes will always appear to be more stable. To avoid this problem we selected genes from different functional classes, as they are more likely to be independently regulated (Table 1). In addition, Normfinder was employed as an extra control. Normfinder is more resistant to the presence of co-regulated genes, because it uses a different algorithm to establish the stability of the genes. This software presents a stability value, which is directly related to the intragroup variance (when no subgroups are present) and is independent of the gene and sample. It basically calculates which gene has the smallest variation over all samples.

We applied both software programs to our data as complementary analyses to obtain the most suitable genes for our experiments. Both algorithms resulted in an overall comparable order of genes. Three of the four best genes were always presented by both programs. Peculiarly, GAPDH was ranked as one of the two most stable genes (together with EF1a) by geNorm in both tissue samples, while Normfinder ranked it as best gene in adults and as worst in fifth instars. This suggests that, while GAPDH and EF1a have a similar expression pattern (low pairwise variation) regardless of the developmental stage, the transcript levels of both genes show greater fluctuations between different brain samples (intragroup variation) from fifth instars than from adult locusts.

A third popular, freely available reference gene validation software program, termed BestKeeper 7, can also be used to analyse HKGs. However, since this program is founded on the same principle as geNorm, it was opted not to use it 18.

When comparing our results to the study of HKGs in brains of honey bees 16, it is observed that four of the genes analysed in Apis mellifera are putative orthologues of genes in this study (UBQ, GAPDH, actin, αTUB). Under the conditions used in this other study actin, RPS18 and GAPDH were most suited. In our experiments, actin, GAPDH and another ribosomal protein (RP49) were also characterized as good reference genes in either one or the other setup. Unfortunately EF1a, which systematically scored well in our tests, was not included in the analyses. When comparing the stability measurements of the honey bee study with ours, we can conclude that under the given conditions the genes are more unstable then in our setup. This underlines the necessity for validation of the HKGs prior to an experiment in different insect species. This is also indicated by the work of Lourenço and co-workers who tested the stability of four reference genes in different developmental stages, tissues and after an experimental treatment in the honey bee. In each case a different ranking of the genes was observed 17.

Also in locusts additional real time test experiments to validate appropriate control gene combinations should be performed when using different experimental conditions. The same is true for different tissues, tissue combinations (e.g. to measure differential expression of a gene across different tissues) or developmental stages. However, based on the identification of seven HKGs from S. gregaria, including a set of primer sequences to perform real-time PCR analyses, as presented in this study (Table 2), these validation tests can be performed in a relatively straightforward manner. Moreover, the genes examined in our study seem to be very stable. Even the least stable gene still has an AESM well below the cut-off of 1.5 in all three conditions, which makes this gene set a possible interesting starting point for experiments in other insects.

The desert locust is the most harmful locust species due to its ability to rapidly increase in number and to migrate over large distances 19. This voracious insect usually exists in a harmless solitary phase, but due to changes in environmental conditions, it can transform into a gregarious phase. In this state the animals aggregate into enormous swarms comprising billions of individuals and form a threat to the food supply of millions of people. It is obvious that, when studying the behaviour of this insect, the brain is of critical importance. It is the primary neurohormone producing gland and coordinates all major processes, such as metabolism, growth and ecdysis, reproduction and behaviour. This study is important to obtain progress in our understanding of neuronal and neuro-endocrine processes at the molecular level in these insects. Foremost, it provides the right tools for the accurate elucidation of the expression profiles of genes in signaling pathways regulating developmental processes in the desert locust.

We identified seven HKG in the desert locust and validated RP49, EF1a and ACT as the most stable genes in fifth instar nymphs and GAPDH, Ubi and EF1a as the most stable genes in adult locusts using the software programs geNorm and Normfinder. Overall, all tested genes proved to be rather stable, offering a good initial set of genes to be used in future validation experiments in locusts and other insects.

MV contributed in the design of the study, performed the dissections and qPCR studies in L5 locusts, executed the geNorm and Normfinder analyses and drafted the manuscript. PV performed the dissections and qPCRs in adult locust. LT provided the geNorm and Normfinder expertise and contributed in drafting the manuscript. SV performed the primer test qPCRs. KV contributed in the selection of HKGs, the design of primers and the primer test qPCRs. RH participated in the design of the study. JV contributed in the drafting of the manuscript and supervised the process. GS designed the study, selected the HKGs, supervised the process and contributed in drafting of the manuscript. All authors read and approved the final manuscript.