Skip to main content

Thank you for visiting You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

The role and robustness of the Gini coefficient as an unbiased tool for the selection of Gini genes for normalising expression profiling data


We recently introduced the Gini coefficient (GC) for assessing the expression variation of a particular gene in a dataset, as a means of selecting improved reference genes over the cohort (‘housekeeping genes’) typically used for normalisation in expression profiling studies. Those genes (transcripts) that we determined to be useable as reference genes differed greatly from previous suggestions based on hypothesis-driven approaches. A limitation of this initial study is that a single (albeit large) dataset was employed for both tissues and cell lines. We here extend this analysis to encompass seven other large datasets. Although their absolute values differ a little, the Gini values and median expression levels of the various genes are well correlated with each other between the various cell line datasets, implying that our original choice of the more ubiquitously expressed low-Gini-coefficient genes was indeed sound. In tissues, the Gini values and median expression levels of genes showed a greater variation, with the GC of genes changing with the number and types of tissues in the data sets. In all data sets, regardless of whether this was derived from tissues or cell lines, we also show that the GC is a robust measure of gene expression stability. Using the GC as a measure of expression stability we illustrate its utility to find tissue- and cell line-optimised housekeeping genes without any prior bias, that again include only a small number of previously reported housekeeping genes. We also independently confirmed this experimentally using RT-qPCR with 40 candidate GC genes in a panel of 10 cell lines. These were termed the Gini Genes. In many cases, the variation in the expression levels of classical reference genes is really quite huge (e.g. 44 fold for GAPDH in one data set), suggesting that the cure (of using them as normalising genes) may in some cases be worse than the disease (of not doing so). We recommend the present data-driven approach for the selection of reference genes by using the easy-to-calculate and robust GC.


In a recent paper1, we introduced the Gini index (or Gini coefficient, GC)2,3,4,5 as a very useful, nonparametric statistical measure for identifying those genes whose expression varied least across a large set of samples (when normalised appropriately6 to the total expression level of transcripts). The GC is a measure that is widely used in economics (e.g.4,7,8,9,10,11,12) to describe the (in)equality of the distribution of wealth or income between individuals in a population. However, although it could clearly be used to describe the variation in any other property between individual examples13,14,15,16), it has only occasionally been used in epidemiology17,18,19 and in biochemistry1,5,20,21,22,23,24,25. Its visualisation and calculation are comparatively straightforward (Fig. 1): individual examples are ranked on the abscissa in increasing order of the size of their contribution, and the cumulative contribution is plotted against this on the ordinate. The GC is given by the fractional area mapped out by the resulting ‘Lorenz’ curve (Fig. 1). For a purely ‘socialist’ system in which all contributions are equal (GC = 0), the curve joins the normalised 0,0 and 1,1 axes, while for a complete ‘autocracy’, in which the resource or expression is held or manifest by only a single individual (GC = 1), the ‘curve’ follows the two axes (0,0 → 1,0 → 1,1).

Figure 1
figure 1

Graphical indication of the means by which we calculate the Gini coefficient.

Since the early origins of large-scale nucleic acid expression profiling, especially those using microarrays26,27,28, it has been clear that expression profiling methods are susceptible to a variety of more or less systematic artefacts within an experiment, whose resolution would require or benefit from some kind of normalisation (e.g.29,30,31,32,33,34,35,36,37,38,39). By this (‘normalisation of the first kind’), and what is typically done, we mean the smoothing out of genuine artefacts within an array or a run, that occur simply due to differences in temperature or melting temperature or dye binding or hybridisation and cross-hybridisation efficiency (and so on) across the surface of the array. This process can in principle use reference genes, but usually exploits smoothing methods that normalise geographically local subsets of the genes to a presumed distribution.

Even after this is done, there is a second level of normalisation, that between chips or experiments, that is usually done separately, not least because it is typically much larger and more systematic, especially because of variations in the total amount of material in the sample analysed or of the overall sensitivity of the detector (much as is true of the within-run versus between-run variations observed in mass spectrometry experiments40,41). This kind of normalising always requires ‘reference’ genes whose expression varies as little as possible in response to any changes in experimental conditions. The same is true for expression profiling as performed by qPCR42,43,44,45,46,47, where the situation is more acute regarding the choice of reference genes since primers must be selected for these a priori. Commonly, the geometric mean of the expression levels of that or those that vary the least is selected as the ‘reference’. The question then arises as to which are the premium ‘reference’ genes to choose.

Data-driven and hypothesis-dependent science are complementary, though when a field is data-rich but hypothesis-poor, as is genomics, data-driven strategies are to be preferred48. Perhaps surprisingly48, rather than simply letting the data speak for themselves, choices of candidate reference genes were often made on the basis that reference genes should be ‘housekeeping’ genes that would simply be assumed (‘hypothesised’) to vary comparatively little between cells, be involved in nominal routine metabolism and also that they should have a reasonably high expression level (e.g.49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66). This is not necessarily the best strategy, and there is in fact (and see below) quite a wide degree of variation of the expression of most standard housekeeping genes between cells or tissues (e.g.53,62,65,67,68,69,70,71,72,73,74,75,76,77,78,79). Indeed, Lee et al.69 stated explicitly that housekeeping genes may be uniformly expressed in certain cell types but may vary in others, especially in clinical samples associated with disease.

It became obvious that an analysis of the GC of the various genes was actually precisely what was required to assess those ‘housekeeping’ (or any other) genes that varied least across a set of expression profiles, and we found 35 transcripts for which the GC was 0.15 or below when assessing 56 mammalian cell lines taken from a wide variety of tissues1. These we refer to as the ‘Gini genes’. Most of these were ‘novel’ as they had never previously been considered as reference genes, and we noted that their Gini indices were significantly smaller (they were more stably expressed) than were those of the more commonly used reference genes66. However, this analysis was done on only one (albeit large) dataset of gene expression profiles. While some of the compilations (e.g.65,80) contain massive amounts of expression profiling data, many of these, especially the older ones, may well be of uncertain quality. Thus, especially since the GC is very prone to being raised by small numbers of large outliers, we decided for present purposes that we should compare our analyses of candidate Gini genes using a smaller but carefully chosen set of expression profiling experiments. The more modern RNA-seq (e.g.81,82,83,84,85), in which individual transcripts are simply counted digitally via direct sequencing, is seen as considerably more robust81,86,87 and sensitive88,89, and so we selected additional large and recent datasets that used RNA-seq in cell lines and tissues (Table 1). We note too that the precision of these digital methods (as with other, digital, single-molecule strategies90,91,92), means that the requirement for reasonably high-level expression levels is much less acute.

Table 1 Studies used for assessing proposed stable reference genes.

In a similar vein (Table 2), we selected a small number of reasonably detailed studies in which particular housekeeping genes had been proposed as reference genes.

Table 2 Studies used for expression profiling data.

To our knowledge, there are no large-scale studies to determine housekeeping genes in large, cell-line cohorts; the present paper serves to provide one. In addition, we include an experimental RT-qPCR analysis of a subset of the Gini genes.


The Gini Coefficient as a robust measure of gene expression stability in multiple cell-line data sets

We previously identified a number of genes in the Human Protein Atlas (HPA) cell line data set93 with very low expression variability and thus potential for use as reference genes1. However, we did not compare these Gini genes to other genes that have previously been proposed as housekeeping genes. We therefore performed a similar analysis using the potential housekeeping genes we proposed in1 as well as other reference genes proposed in other studies (Table 2) with additional large RNA-Seq cell line data sets (Table 1).

Figure 2A shows a plot of the GC of a variety of candidate Gini genes versus their median expression level in the HPA cell lines dataset set93. It is clear that genes we identified previously have much lower GC values in the HPA dataset than do any of the others (just two, VPS29 and CHMP2A, were also identified by Eisenberg and Levanon and another, RPL41, by Caracausi). This is not at the expense of an unusually low expression (Fig. 2A), a finding broadly confirmed when we look at the median expression levels for the CCLE dataset (Fig. 2B) and of the Klijn dataset (Fig. 2C).

Figure 2
figure 2

Gini coefficient and median expression levels of proposed reference genes in the HPA cell-line dataset. (A) GC versus median expression level of HPA dataset. (B) Median expression levels of CCLE vs HPA datasets. Line of best linear fit (in log space) shown is y = 0.991 + 0.827 × (r2 = 0.606). (C) Median expression levels of CCLE vs Klijn datasets. Line of best linear fit (in log space) shown is y = 0.998 + 0.804 × (r2 = 0.593). Colour coding: red, GeneGini reference genes; blue Eisenberg & Levanon; yellow Vandesompele; green Lee; lilac both GeneGini and Eisenberg and Levanon.

Figure 3 shows the GC values for the various genes in two other datasets, viz CCLE and Klijn. Our previous Gini genes have a lower GC than that of any of the other housekeeping genes in 25 out of 38 cases in Klijn (all under 0.2) and in 26 out of 40 cases for CCLE (all under 0.22). In confirmation of this, and of the correlation found above between the median expression levels in CCLE and Klijn, the GC values are also well correlated with each other for the two datasets (Fig. 3). Thus, although the absolute numbers are slightly larger than are those for the HPA dataset (unsurprisingly, given the much larger number of examples), the trend is still very clear: the GiniGenes remain the best among those variously proposed as reference genes in a variety of large and quite independent datasets. It also suggests that variations in the total amount of mRNA are not an issue either.

Figure 3
figure 3

Gini coefficient of candidate reference genes in CCLE and Klijn/Genentech cell-line datasets. Left panel shows all proposed housekeeping genes considered in this study, with the right panel showing labels of those genes with a GC < 0.25. The line of best fit is y = −0.171 + 0.829 × (r2 = 0.909). Colour code as in Fig. 2.

Another common statistical measure, more resistant to individual outliers, is the interquartile ratio (the ratio between the 25th and 75th percentile when expression levels are ranked); by this measure too, the Gini genes that we uncovered previously stand out as being the least varying (Fig. 4 A,B). This suggests that, as a measure of gene expression stability, the GC is robust: the GiniGenes have the lowest ratio between their maximum and minimum expression values in the HPA dataset (Fig. 4C) and also the lowest interquartile ratio in their levels of expression in all three cell line data sets explored here (Fig. 4B,C) with good correlation between these two datasets.

Figure 4
figure 4

Robustness of the Gini coefficient. (A) IQR of different genes in Klijn/Genentech vs HPA cell-line dataset. Left panel shows all genes considered in this study, with right panel showing genes with IQR < 2 in both datasets. Line of best linear fit (in log space) shown is y = 0.01 + 1.11 × (r2 = 0.937). (B) IQR of different genes in CCLE vs HPA cell-line dataset. Left panel shows all genes considered in this study, with right panel showing genes with IQR < 2 in both datasets. Line of best linear fit (in log space) shown is y = 0.04 + 0.99 × (r2 = 0.930). (C) Min vs Max: Median expression levels in HPA data set. Colour code as in Fig. 2.

Use of the Gini Coefficient to find GiniGenes in an unbiased manner in cell-line data sets

Up to now, our analyses of these data sets have used a set of predefined genes to look at expression stability. We next sought to investigate whether the GC would highlight genes with high expression stability that have been reported by others or by ourselves when performing this analysis in a data-driven manner. To that end, we found 115 genes shared between the three data sets with a GC ≤ 0.2 (Figs. 5, 6). This value for the GC was chosen since reducing this to ≤0.15 meant no or very few genes were found in some data sets (e.g. no genes in the CCLE data set had a GC ≤ 0.15) and going above this meant the number of genes were unmanageable (e.g. 1051 genes with a GC ≤ 0.21 in the Klijn data set). Of the 115 genes shared between the datasets with GC < 0.2, 13 were GiniGenes and two were housekeeping genes defined by Caracausi and colleagues (Fig. 5B). When we selected the top 20 expressing genes in each data set, only 13 of these were common across these data sets; Table 3 shows some descriptive statistics of 13 of these, with descriptive statistics of all 115 genes found in Supplementary Table S1. Of these genes, two (HNRNPK and PCBP1) are GiniGenes and one (SLC25A3) is a gene previously reported by Caracausi et al. Seven out of the 13 genes (HNRNPK, HNRNPC, PCBPB, SF3B1, SRSF3, EDF1 and EIF4H) here share important roles in RNA transcription, translation and stability94,95,96,97,98,99,100,101,102, are implicated in a number of diseases, including cancer94,97,103,104,105,106,107,108,109,110,111,112,113, and some, such as SRSF3 are essential for embryo development114. Given their pivotal functions, it may be unsurprising that the expression of these genes are tightly regulated across cell lines of different tissue origins, even where these are cancer cell lines. Overall, the distribution, expression stability and important functional roles of these genes suggest that these are excellent housekeeping genes across different cell types.

Figure 5
figure 5

Shared and unique genes in HPA, CCLE and Klijn/Genentech cell-line data sets. (A) Genes with a GC < 0.2 .(B) Housekeeping genes in Table 2 with GC < 0.2.

Figure 6
figure 6

GC vs Median for 115 genes in. (A) HPA, (B). CCLE and C. Klijn/Genentech cell-line data sets. Colour coding: Blue, Caracausi; Green, GeneGini reference genes; Grey, neither. Shape coding: Circle, other; Triangle, SLC coding gene.

Table 3 Descriptive statistics of 13 genes common across cell-line data sets with GC < 0.2.

Of particular interest to us was finding one gene encoding a mitochondrial phosphate transporter protein (SLC25A3115) to be within this list of the top expressing stably expressed genes. This might seem logical since mitochondrial ATP synthesis is required by all cell types and tissues.

Figure 7 shows the robustness of the GC for the subset of 115 genes common between the three data sets studied here with a low GC (<0.2). Lower Gini coefficients correlate with lower IQR and Max:median ratios (Fig. 7: only results for the Klijn data set are shown). The range of IQR values of these genes was smaller in the larger two data sets (CCLE, 1.42–1.67; Klijn, 1.30–1.64) than in the HPA data set (1.26–1.84) suggesting the measured expression values were more stable in the larger data sets (Supplementary Table S1). This may, however, be due to a larger number of cell lines in these two large datasets (934 and 622 in CCLE and Klijn) compared with the HPA data set (56 cell lines).

Figure 7
figure 7

Robustness of GC for finding stably expressed genes using shared genes between HPA, CCLE and Klijn/Genentech cell-line data sets with GC < 0.2. Shown are the results for the Klijn/Genentech dataset. (A) IQR vs GC, (B). Max:Mean vs Min. Colour coding: Blue, Caracausi; Green, GeneGini reference genes; Grey, neither. Shape coding: Circle, other; Triangle, SLC coding gene.

Application of the Gini coefficient to human tissue RNA-Seq data sets

The results presented thus far are representative of human cell lines. Most reports in the literature regarding housekeeping genes refer to tissue expression data. This may be due to the cell lines being “dedifferentiated” with respect to the tissues from which they are derived116.

In our previous report1 we also analysed RNA-Seq data from tissues93 and found 22 genes with a GC < 0.15, of which 3 (CHMP2A, VPS29 and PCBP1) were also found in cell line data with a GC < 0.15. The median expression level and GC of these and other candidate GiniGenes in this tissue data set are shown in Fig. 8. As with cell line data, the genes we previously identified (GGs, green dots in Fig. 8) have much lower GCs in this tissue data set than do any of the other candidate GiniGenes, with only two of these genes (VPS29 and CHMP2A) identified previously by Eisenberg & Levanon49. The low GC value of these GiniGenes is not at the expense of low expression: of the 22 GiniGenes, 13 are expressed at a median level of between 40 and 200 TPM (see Supplementary Table S2). Moreover, the GC was also representative of the variation in expression of these genes (albeit influenced to a lesser extent by outliers), as shown in Fig. 9A,B, with all GiniGenes having a GC < 0.15 and the lowest RSD (relative standard deviation), ranging from 24.096% to 28.66% and IQR (1.26 to 1.44) of this list of housekeeping genes. The expression of other housekeeping genes such as GAPDH, ACTB, RPL13A, SDHA, B2M was quite varied according to these measures. For example, the GC of GAPDH (a commonly used HKG) was 0.33, with a RSD of 72.4% and IQR of 2.24, and for ACTB (another commonly used HKG) these values were 0.29, 55.24%, and 2.11.

Figure 8
figure 8

Gini coefficient and median expression levels of proposed reference genes in the HPA tissue dataset. Colour coding: blue, Caracausi; purple, Eisenberg and Levanon; green, GeneGini reference genes; yellow, both GeneGini and Eisenberg and Levanon; orange, Lee; black, Vandesompele.

Figure 9
figure 9

Robustness of the Gini coefficient in the HPA tissue data set. (A) RSD versus Gini coefficient of candidate reference genes. Line of best linear fit (in log space) shown is y = 2.45 + 1.24 × (r2 = 0.938) (B). IQR versus Gini coefficient of candidate reference genes. Line of best linear fit (in log space) shown is y = 0.87 + 0.96 × (r2 = 0.566). Colour code as in Fig. 8.

The median expression levels of the proposed reference genes show a similar level of correlation between the data sets as was found with the cell line data (Fig. S1A–C), and GiniGenes displayed a mid-range level of expression. The GC of the tissue GiniGenes we proposed however, tended to be higher and more variable in their GC values than in the HPA dataset (Fig. S2,A–C) suggesting that those genes may be representative of the HPA data set only. As an example, in the GTEx dataset only 28 genes had a GC < 0.2, of which the majority (17) were those reported by Caracausi and colleagues, and 7 were GiniGenes. The results here are likely influenced by the number and status (disease or normal) of the tissues analysed in the various data sets compared; for example, the GTEx data come from 53 different, normal human tissues, whereas the HPA tissue data include a mixture of disease and normal tissue samples. In addition, compared to the cell line data where hundreds (in the case of the Cancer Cell Line Encyclopedia) of cell lines were analysed, the number of tissues in these data sets was fewer than 100.

In the case of the data set used by Eisenberg and Levanon49, viz. the Illumina Human Body Map (E-MTAB-513), 10 of the 11 housekeeping genes proposed here (which included 2 Gini Genes, CHMP2A and VPS29) had a GC ≤ 0.2 and were reasonably well expressed (with median expression levels between 50–270 TPM, see Supplementary Table S2 and Supplementary Fig. S4). This may be compared to the 5 other GGs with GC < 0.2 in this data set whose expression value was lower, with median expression between 19–35 TPM. This suggests that finding suitable HKGs may be dependent on the data set itself, and the type of tissue under investigation.

We next sought to perform a more comprehensive and integrative analysis by filtering the tissue data sets to only include genes with a GC ≤ 0.2 to find common genes across these data sets with reasonable expression stability (Supplementary Table S3). As shown in Fig. 10 only 15 genes were shared between the four data sets with a GC ≤ 0.2, none of which has been reported previously as a housekeeping genes. Table 4 shows some descriptive statistics of these genes. In any case, the names of the proteins encoded by these 15 genes suggest these play important and essential roles. The median expression values of these genes varied from around 10–450 TPM, with SNX3 (Sorting nexin-3 (Protein SDP3)) and COX4I1 (Cytochrome c oxidase subunit 4 isoform 1) being consistently the two highest-expressing genes.

Figure 10
figure 10

UpSetR139 plot showing genes with a GC < 0.2 that are variously shared and unique across the PCAWG, HBM, GTEX and HPA tissue data sets. The data underpinning this plot can be found in Supplementary Table S4.

Table 4 Descriptive statistics of 15 common genes across tissue data sets with a GC < 0.2.

Sorting nexins are a group of cytoplasmic and membrane-associated proteins involved in the regulation of intracellular trafficking117. SNX3 has been reported to play a role in receptor recycling and formation of multivesicular bodies118, and its dysregulation has been implicated in disorders of iron metabolism and the pathogenesis of some neurodegenerative diseases119,120.

The COX4I gene encodes the nuclear-encoded cytochrome c oxidase subunit 4 isoform 1, the terminal enzyme in the mitochondrial respiratory chain. Given the key role of the mitochondrial respiratory chain in all human cells (except red blood cells), stable expression of such a gene in all tissues may not be a surprising result. Increased RNA COX4I1 levels have been reported in sperm of an obese male rat model121 and thus may play a role in obesity-related fertility problems, and reduced expression of this subunit leads to a reduction in mitochondrial respiration as well as sensitising cells to apoptosis122.

The small number of genes shared between these data sets with a GC < 0.2 indicates that the data in these studies are more variable compared to cell lines alone. The cause of this variation may be due to the tissue data having been obtained from different subjects123. Moreover, tissues are themselves a mixture of cell types with varying levels of gene expression in each cell type124, while cell lines are nominally clonal.

Our results suggest that in the case of RNA-seq tissue data sets, where gene expression tends to be more variable, an unbiased approach, using the Gini coefficient, may be more fruitful in the search for stably expressed genes with which to perform normalisation, than the other commonly used methods used until now123,125.

RT-qPCR analysis of gene expression stability of some housekeeping genes in 10 cell lines

In order to illustrate the utility of the GC to find suitable housekeeping genes, we next chose to assess this experimentally by RT-qPCR using a small subset of candidate reference genes (40; top 32 genes from genes ordered by GC and expression value from94, plus 8 of the most commonly used from the literature, including seven from66 and one (RPL32) from126,127, and 10 cell lines from a range of tissues (see Tables 5 and 6). We first set a Cq value (which is inversely proportional to expression level) cut-off of 32, above which no expression is observed, and subsequently used the Cq values of genes in cell lines as a relative expression level (Cq cut off/Cq value of gene). Descriptive statistics of the expression of each gene in individual cell lines were then calculated. As a final step, the median expression value of each gene in individual cell lines was used to calculate descriptive statistics, including the GC, of gene expression across these cell lines. Figure 11 illustrates a KNIME workflow128,129,130 that we wrote for this purpose. The raw data and descriptive statistics extracted are provided in Supplementary Tables S5 and S6 respectively, and the KNIME analysis workflow in Supplementary File 1.

Table 5 Details of human cell lines used for the assessment of expression of candidate reference genes by RT-qPCR.
Table 6 Candidate reference genes used to assess expression stability experimentally by RT-qPCR.
Figure 11
figure 11

The KNIME workflow described here to calculate descriptive statistics and the Gini coefficient from RT-qPCR data. This workflow can be adapted for use with large RNA-Seq Data sets.

Figure 12 uses RT-qPCR data to plot the GC of the candidate reference genes analysed here versus their relative median expression level. Three GiniGenes94 (RBM45, TRNT1 and CNOT2) had very low and variable expression. Most of the other genes analysed showed low GC values with a range of (relative) expression values; the inset in Fig. 12 shows genes with a GC < 0.2 including a mix of 35 genes: 26 GiniGenes and 6 housekeeping genes referenced by Vandesompele and colleagues66, one referenced by Caracausi65 and one by Lee et al.131. Two of these GiniGenes, HNRNPK and PCBP1, which we also found to be stably expressed in the cell line data suggesting these may be potential stable housekeeping genes. As shown in Fig. 13 and inset, the GC is well correlated with the % RSD.

Figure 12
figure 12

Gini coefficient and median expression levels of candidate reference genes assessed by RT-qPCR. Left panel shows all genes considered in this study, with right panel showing genes with GC < 0.2. Colour coding: green, GeneGini reference genes; red, both GeneGini and Caracausi reference genes; yellow, GeneGini and Eisenberg and Levanon; orange, Lee, yellow; black, Vandesompele; purple, Zhang and Kriegova.

Figure 13
figure 13

Robustness of the Gini coefficient in assessed experimentally by RT-qPCR using a small subset of proposed reference genes. Left panel shows Gini coefficient vs % RSD for all genes considered in this study, with right panel showing the same with genes with a GC < 0.2 and % RSD <10. Line of best linear fit shown is y = 0.002 + 0.004 × (r2 = 0.988). Shape coding as in Fig. 12.

More importantly, the GC of our GiniGenes was particularly low (Fig. 12). The low absolute magnitude reflected the fact that Cq value is based on a logarithmic scale. Various commonly used housekeeping genes (HPRT1, GAPDH, ACTB, SDHA, HMBS and B2M) displayed higher % RSDs and GC than other genes studied here in spite of their higher relative expression levels. This was also the case when inspecting the interquartile ratio against the GC of these (Fig. S3).

The above results suggest that the GC is also applicable to RT-qPCR data, with GiniGenes having good potential (as novel “housekeeping” genes) for the normalisation of such data.


Reference genes are commonly used to normalise gene expression data, so as to account for bias resulting from both biological and technical variability, and to enable quantification of gene expression changes or differences in the system under study. It is generally considered that such reference genes should come from pathways that are required for general metabolism, using only one gene per ‘pathway’ to avoid co-regulation which might make the gene expressions look very stable.

Such reference genes are commonly referred to as ‘housekeeping’ genes (HKGs) because they are considered to participate in essential cellular functions, are ubiquitously expressed in all cells and tissue types, and their expression is considered to be stable49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66). A number of such genes have been proposed over the years, and genes such as GAPDH, ACTB, RPL13A, SDHA, B2M are frequently used in such studies66. However, the expression levels of these and other proposed HKGs have in fact been shown to vary widely between cells and tissues (e.g.53,62,65,67,68,69,70,71,72,73,74,75,76,77,78,79) and their expression has also been reported to be affected by a number of factors relating to the experiment such as cell confluence132, pathological, experimental and tissue specific conditions133. As highlighted by Huggett et al.134, despite the reports of the potential variability of expression of ‘classic’ references genes such as GAPDH and ACTB, these are still used without mention of any validation processes. Our GiniGenes are selected as reference genes through different, data-driven, criteria.

Various tools have been developed to evaluate and screen reference genes from experimental datasets; these include geNorm66, NormFinder135, Best Keeper136 and the comparative ΔCT finder52. RefFinder ( and RefGenes ( can integrate these to enable a comparison and ranking of any tested candidate reference genes137.

These tools assess expression stability of genes in different ways:

  • geNorm determines gene stability through a stepwise exclusion or ranking process followed by averaging the geometric mean of the most stable genes from a chosen set. Python implementation:

  • BestKeeper also uses the geometric mean but using raw data rather than copy numbers. BestKeeper136 can be used as an Excel-based tool. It can accommodate up to 10 housekeeping genes in up to 100 biological samples. Optimal HKGs are determined by pairwise correlation analysis of all pairs of candidate genes, and the geometric mean of the top-ranking ones.

  • NormFinder measures variation, and ranks potential reference genes between study groups. NormFinder135 has an add-in for Microsoft Excel and is available as an R programme. It recommends analysis of 5–10 candidate genes and at least 8 samples per group.

  • The comparative ΔCT finder requires no specialist programmes since this involves comparison of comparisons of ΔCTs between pairs of genes to find a set of genes that show least variability.

  • RefGenes allows one to find genes that are stably expressed across tissue types and experimental conditions based on microarray data, and a comparison of results from geNorm, NormFinder and Best Keeper to find a set of reference genes. However, this is not a free service unless one searches for one gene at a time. Furthermore, the site for this tool is no longer available. Moreover, all these tools require the user to make a prior selection of such HKGs (introducing bias and potential errors) and most are cumbersome to understand and calculate.

We have here shown how via a simple calculation, the GC, we can find potential reference genes, and illustrated its utility in large-scale cell-line, tissue RNA-Seq data sets and RT-qPCR data. The expression of a number of classical HKGs from a number of carefully selected publications do in fact vary much more substantially between large RNA-Seq data sets, both for tissues and cell lines.

Whilst not all studies will involve large data sets such as those we have analysed here; the GC should also be of use for smaller-scale studies to select a subset of genes in a panel of cell lines or tissues relevant to the study in question.

Overall we find that (i) two of these genes, HNRNPK and PCBP1, seemed to be particularly robustly and stably expressed at reasonable levels in all cell lines studied, and (ii) a data-driven strategy based on the GC represents a useful and convenient method for normalisation in gene expression profiling and related studies.


The datasets used are described and referenced below. The data, in transcripts per million (TPM) units were downloaded from the EBI expression atlas as a .tsv file. As previously1, the Gini Index was calculated using the ineq package (Achim Zeileis (2014). ineq: Measuring Inequality, Concentration, and Poverty. R package version 0.2–13. in R ( These calculations were incorporated into KNIME via KNIME’s R integration R Snippet node. A spreadsheet giving the extracted analyses is provided as Supplementary Tables (Tables S7 and S8).

Cell lines and culture conditions

A panel of 10 cell lines were grown in appropriate growth media: K562, PNT2 and T24 in RPMI-1640 (Sigma, Cat No. R7509), Panc1 and HEK293 in DMEM (Sigma, Cat No. D1145), SH-SY5Y in 1:1 mixture of DMEM/F12 (Gibco, Cat No. 21041025), J82 and RT-112 in EMEM (Gibco, Cat No. 51200–038), 5637 in Hyclone McCoy’s (GE Healthcare, Cat No. SH30270.01) and PC3 in Ham’s F12 (Biowest, Cat No. L0135-500). All growth media were supplemented with 10% fetal bovine serum (Sigma, Cat No. f4135) and 2 mM glutamine (Sigma, Cat No. G7513) without antibiotics. Cell cultures were maintained in T225 culture flasks (Star lab, CytoOne Cat No. CC7682-4225) kept in a 5% CO2 incubator at 37 °C until 70–80% confluent.

Harvesting Cells for RNA Extraction

Cells from adherent cell lines were harvested by removing growth media and washing twice with 5 mL of pre-warmed phosphate buffered saline (PBS) (Sigma, Cat No. D8537), then incubated in 3 mL of 0.025% trypsin-EDTA solution (Sigma Cat No. T4049) for 2–5 min at 37 °C. At the end of incubation cells were resuspended in 5–7 mL of respective media when cells appeared detached to dilute trypsin treatment. The cell suspension was transferred to 15 mL centrifuge tubes and immediately centrifuged at 300 × g for 5 min. Suspended cell lines were centrifuged directly from cultures in 50 mL centrifuge tubes and washed with PBS as above. The cell pellets were resuspended in 10–15 mL media and cell count and viability was determined using a Nexcellom Cellometer Auto 1000 Cell Viability Counter (Nexcellom Bioscience) set for Trypan Blue membrane exclusion method. Cells with >95% viability were used for downstream total RNA extraction.

RNA Extraction

Total RNA was extracted from 2–5 × 106 cells using the Qiagen RNeasy Mini Kit (Cat No. 74104) and DNAse treated using Turbo DNA-free kit (Invitrogen, Cat No. AM1907) according to the manufacturer’s instructions. Briefly, 1 X DNA buffer was added to the extracted RNA prior to adding 2U (1 µL) of DNAse enzyme. The reaction mixture was incubated at 37 °C for 30 min and inactivated for 2 min at room temperature using DNAse inactivating reagent. The mixture was centrifuged at 10,000 × g for 1.5 min and the RNA from the supernatant was transferred to a clean tube. The RNA concentration was determined using a NanoDrop® ND-1000 spectrophotometer and further validated using an Agilent 2100 bio-analyser coupled with 2100 Expert software system. Only RNA samples with an RIN (RNA Integrity Number) between 9–10 were selected for cDNA synthesis.

Reverse Transcription and cDNA Synthesis

1 µg of RNA was reverse transcribed into cDNA. Briefly, a 20 µL reaction was setup by adding 1 µL each of oligodT (50 µM, Invitrogen, cat No. 18418020) and dNTP mix (10 mM, Invitrogen, Cat No. 18427-013) followed by adding an appropriate volume for 1 µg of RNA. Nuclease free water (Ambion, Cat No. AM9937) was then added to make the volume up to 13 µL and incubated at 65 °C for 5 min then cooled on ice for 1 min. To initiate transcription 4 µL of 5 X first strand buffer (Invitrogen, Cat No. 1889832) and 1 µL each of 0.1 M DTT (Invitrogen, Cat No. 1907572), RNaseOUT™ (Invitrogen, Recombinant RNase Inhibitor, Cat No. 1905432) and SuperScript™ III RT (200 units/µL, Invitrogen, Cat No. 1685475) reverse transcriptase enzyme were added, mixed gently then incubated at 50 °C for 60 min followed by inactivation at 70 °C for 15 min. The cDNA was diluted 1:100 to be used in RT-qPCR experiment.

Validation of gene expression by geNorm

A set of candidate reference genes (40; top 32 genes from genes ordered by GC and expression value from94, plus 8 of the most commonly used from the literature including seven from66). RNAseq data were selected for validation of stable gene expression using geNorm66. First, a typical qPCR protocol was prepared from a master mix for each gene to be tested per cell line in triplicate. This consisted of 10 µL/well made by adding 0.8 µL of nuclease free water (Ambion), 5 µL of LC480 SYBR Green I Master (2 X conc. Roche, Product No. 04887352001), 0.1 µL each of forward and reverse primers (20 µM) (for primer and amplicon sequences see Supplementary Table S9) and 4 µL of 1:100 diluted cDNA in a 384 well qPCR plate (Starlab Cat. No. E1042-9909-C). The no template controls (NTC) for each gene were produced by replacing cDNA with 4 µL of nuclease free water. Thermal cycling conditions used were: one cycle of 95 °C for 10 min followed by 40 cycles of 95 °C for 10 sec and 60 °C for 30 sec. qPCR was performed using Roche LightCycler LC480 qPCR platform. The fluorescence signals were measured in real time during amplification cycle (Cq) and also during temperature transition for melt curve analysis.

The mean Cq values were converted into relative values for a gene across all cell lines using ΔCq method138. Briefly, the lowest Cq value in a panel of cell lines for a gene was subtracted from all the values in that panel using the equation: \(R={2}^{({C}_{{q}_{sample}}-{C}_{{q}_{control}})}\), where \({C}_{{q}_{sample}}\) is the mean Cq value obtained for a gene in each of the cell lines and \({C}_{{q}_{control}}\)is the lowest Cq value in that panel. The relative values for each gene in a panel were then obtained by applying \(R={2}^{-{\Delta C}_{q}}\). These relative values were applied in geNorm Visual Basic applet for Microsoft Excel®66 that determines the most stable reference genes from a set of genes in a given panel of cell lines.

Validation of gene expression using the Gini coefficient

To the raw RT-qPCR data a Cq value (which is inversely proportional to expression level) cut-off of 32 was set, above which no expression is observed. The Cq values of genes in cell lines were subsequently converted to a relative expression level (Cq cut off/Cq value of gene). Descriptive statistics of the expression of each gene in individual cell lines were then calculated. As a final step, the median expression value of each gene in individual cell lines was used to calculate descriptive statistics, including the GC, of gene expression across these cell lines. Figure 11 illustrates a KNIME workflow128,129,130 for this purpose. The raw data and descriptive statistics extracted are provided in Supplementary Tables S5 and S6 respectively, and the KMNIME analysis workflow in Supplementary File 1.

Data availability

All data generated or analysed during this study are included in this published article (and its Supplementary Information Files). The original datasets used are referenced throughout and are summarised in Table 2.


  1. 1.

    O’Hagan, S., Wright Muelas, M., Day, P. J., Lundberg, E. & Kell, D. B. GeneGini: assessment via the Gini coefficient of reference “housekeeping” genes and diverse human transporter expression profiles. Cell systems 6, 230–244, (2018).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  2. 2.

    Gini, C. Concentration and dependency ratios (in Italian). English translation in: Rivista di Politica. Economica 87(1997), 769–789 (1909).

    Google Scholar 

  3. 3.

    Gini, C. Variabilità e Mutabilità. Contributo allo Studio delle Distribuzioni e delle Relazioni Statistiche. (C. Cuppini, 1912).

  4. 4.

    Ceriani, L. & Verme, P. The origins of the Gini index: extracts from Variabilità e Mutabilità (1912) by Corrado Gini. J Econ Inequal 10, 421–443, (2012).

    Article  Google Scholar 

  5. 5.

    Jiang, L., Tsoucas, D. & Yuan, G. C. Assessing Inequality in Transcriptomic Data. Cell systems 6, 149–150, (2018).

    CAS  Article  PubMed  Google Scholar 

  6. 6.

    Wagner, G. P., Kin, K. & Lynch, V. J. Measurement of mRNA abundance using RNA-seq data: RPKM measure is inconsistent among samples. Theory Biosci 131, 281–285, (2012).

    CAS  Article  PubMed  Google Scholar 

  7. 7.

    Wilkinson, R. & Pickett, K. The spirit level: why equality is better for everyone. (Penguin Books, 2009).

  8. 8.

    Kondo, N. et al. Income inequality and health: the role of population size, inequality threshold, period effects and lag effects. J Epidemiol Community Health 66, e11, (2012).

    Article  PubMed  Google Scholar 

  9. 9.

    Pickett, K. E. & Wilkinson, R. G. Income inequality and health: a causal review. Soc Sci Med 128, 316–326, (2015).

    Article  PubMed  Google Scholar 

  10. 10.

    Darkwah, K. A., Nortey, E. N. & Lotsi, A. Estimation of the Gini coefficient for the lognormal distribution of income using the Lorenz curve. Springerplus 5, 1196, (2016).

    Article  PubMed  PubMed Central  Google Scholar 

  11. 11.

    Kohler, T. A. et al. Greater post-Neolithic wealth disparities in Eurasia than in North America and Mesoamerica. Nature 551, 619–622, (2017).

    ADS  CAS  Article  PubMed  PubMed Central  Google Scholar 

  12. 12.

    Nishi, A., Shirado, H., Rand, D. G. & Christakis, N. A. Inequality and visibility of wealth in experimental social networks. Nature 526, 426–429, (2015).

    ADS  CAS  Article  PubMed  Google Scholar 

  13. 13.

    Damgaard, C. & Weiner, J. Describing inequality in plant size or fecundity. Ecology 81, 1139–1142, 10.1890/0012-9658(2000)081[1139:Diipso]2.0.Co;2 (2000).

  14. 14.

    Sadras, V. & Bongiovanni, R. Use of Lorenz curves and Gini coefficients to assess yield inequality within paddocks. Field Crops Res 90, 303–310, (2004).

    Article  Google Scholar 

  15. 15.

    Weidlich, I. E. & Filippov, I. V. Using the gini coefficient to measure the chemical diversity of small-molecule libraries. J Comput Chem 37, 2091–2097, (2016).

    CAS  Article  PubMed  Google Scholar 

  16. 16.

    Wren, J. D. Bioinformatics programs are 31-fold over-represented among the highest impact scientific papers of the past two decades. Bioinformatics 32, 2686–2691, (2016).

    Article  PubMed  Google Scholar 

  17. 17.

    LEE, W.-C. Analysis of Seasonal Data Using the Lorenz Curve and the Associated Gini Index. International Journal of Epidemiology 25, 426–434, (1996).

    MathSciNet  CAS  Article  PubMed  Google Scholar 

  18. 18.

    Lee, W.-C. Characterizing Exposure–Disease Association in Human Populations Using the Lorenz Curve and Gini Index. Statistics in Medicine 16, 729–739, 10.1002/(SICI)1097-0258(19970415)16:7<729::AID-SIM491>3.0.CO;2-A (1997).

    CAS  Article  Google Scholar 

  19. 19.

    Lee, W.-C. Probabilistic analysis of global performances of diagnostic tests: interpreting the Lorenz curve-based summary measures. Statistics in Medicine 18, 455–471, 10.1002/(SICI)1097-0258(19990228)18:4<455::AID-SIM44>3.0.CO;2-A (1999).

    CAS  Article  Google Scholar 

  20. 20.

    Ainali, C. et al. Transcriptome classification reveals molecular subtypes in psoriasis. BMC Genomics 13, 472, (2012).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  21. 21.

    Tran, Q. N. Improving the Accuracy of Gene Expression Profile Classification with Lorenz Curves and Gini Ratios. Software Tools and Algorithms for Biological Systems 696, 83–90, (2011).

    CAS  Article  Google Scholar 

  22. 22.

    Jiang, L., Chen, H., Pinello, L. & Yuan, G. C. GiniClust: detecting rare cell types from single-cell gene expression data with Gini index. Genome Biol 17, 144, (2016).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  23. 23.

    Torre, E. et al. A comparison between single cell RNA sequencing and single molecule RNA FISH for rare cell analysis. bioRxiv, 138289, (2017).

  24. 24.

    Shaffer, S. M. et al. Rare cell variability and drug-induced reprogramming as a mode of cancer drug resistance. Nature 546, 431–435, (2017).

    ADS  CAS  Article  PubMed  PubMed Central  Google Scholar 

  25. 25.

    Torre, E. et al. Rare Cell Detection by Single-Cell RNA Sequencing as Guided by Single-Molecule RNA FISH. Cell systems 6, 171–179 e175, (2018).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  26. 26.

    Schena, M. et al. Parallel human genome analysis - microarray-based expression monitoring of 1000 genes. Proc. Natl. Acad. Sci. 93, 10614–10619 (1996).

    ADS  CAS  Article  Google Scholar 

  27. 27.

    Spellman, P. T. et al. Comprehensive identification of cell cycle-regulated genes of the yeast Saccharomyces cerevisiae by microarray hybridization. Mol. Biol. Cell 9, 3273–3297 (1998).

    CAS  Article  Google Scholar 

  28. 28.

    Schena, M. et al. Microarrays: biotechnology’s discovery platform for functional genomics. Trends Biotechnol. 16, 301–306 (1998).

    CAS  Article  Google Scholar 

  29. 29.

    Hoyle, D. C., Rattray, M., Jupp, R. & Brass, A. Making sense of microarray data distributions. Bioinformatics 18, 576–584 (2002).

    CAS  Article  Google Scholar 

  30. 30.

    Quackenbush, J. Microarray data normalization and transformation. Nat Genet 32(Suppl), 496–501, (2002).

    CAS  Article  PubMed  Google Scholar 

  31. 31.

    Knight, C. G. et al. Array-based evolution of DNA aptamers allows modelling of an explicit sequence-fitness landscape. Nucleic Acids Res 37, e6 (2009).

    Article  Google Scholar 

  32. 32.

    Walsh, C. J., Hu, P., Batt, J. & Santos, C. C. Microarray Meta-Analysis and Cross-Platform Normalization: Integrative Genomics for Robust Biomarker Discovery. Microarrays (Basel) 4, 389–406, (2015).

    Article  Google Scholar 

  33. 33.

    Do, J. H. & Choi, D. K. Normalization of microarray data: single-labeled and dual-labeled arrays. Mol Cells 22, 254–261 (2006).

    CAS  PubMed  Google Scholar 

  34. 34.

    Steinhoff, C. & Vingron, M. Normalization and quantification of differential expression in gene expression microarrays. Brief Bioinform 7, 166–177, (2006).

    CAS  Article  PubMed  Google Scholar 

  35. 35.

    Dabney, A. R. & Storey, J. D. A new approach to intensity-dependent normalization of two-channel microarrays. Biostatistics 8, 128–139, (2007).

    Article  PubMed  MATH  Google Scholar 

  36. 36.

    Kreil, D. P. & Russell, R. R. There is no silver bullet–a guide to low-level data transforms and normalisation methods for microarray data. Brief Bioinform 6, 86–97 (2005).

    CAS  Article  Google Scholar 

  37. 37.

    Rahman, M. et al. Alternative preprocessing of RNA-Sequencing data in The Cancer Genome Atlas leads to improved analysis results. Bioinformatics 31, 3666–3672, (2015).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  38. 38.

    Lin, Y. et al. Comparison of normalization and differential expression analyses using RNA-Seq data from 726 individual Drosophila melanogaster. BMC Genomics 17, 28, (2016).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  39. 39.

    Li, X. et al. A comparison of per sample global scaling and per gene normalization methods for differential expression analysis of RNA-seq data. PLoS One 12, e0176185, (2017).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  40. 40.

    Dunn, W. B. et al. Procedures for large-scale metabolic profiling of serum and plasma using gas chromatography and liquid chromatography coupled to mass spectrometry. Nat Protoc 6, 1060–1083 (2011).

    CAS  Article  Google Scholar 

  41. 41.

    Zelena, E. et al. Development of a robust and repeatable UPLC-MS method for the long-term metabolomic study of human serum. Analytical chemistry 81, 1357–1364 (2009).

    CAS  Article  Google Scholar 

  42. 42.

    Heckmann, L. H., Sørensen, P. B., Krogh, P. H. & Sørensen, J. G. NORMA-Gene: a simple and robust method for qPCR normalization based on target gene data. BMC Bioinformatics 12, 250, (2011).

    Article  PubMed  PubMed Central  Google Scholar 

  43. 43.

    Hruz, T. et al. RefGenes: identification of reliable and condition specific reference genes for RT-qPCR data normalization. BMC Genomics 12, 156, (2011).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  44. 44.

    Khanna, P., Johnson, K. L. & Maron, J. L. Optimal reference genes for RT-qPCR normalization in the newborn. Biotech Histochem, 1–8, (2017).

    CAS  Article  Google Scholar 

  45. 45.

    Ling, D. & Salvaterra, P. M. Robust RT-qPCR data normalization: validation and selection of internal reference genes during post-experimental data analysis. PLoS One 6, e17762, (2011).

    ADS  CAS  Article  PubMed  PubMed Central  Google Scholar 

  46. 46.

    Sang, J. et al. ICG: a wiki-driven knowledgebase of internal control genes for RT-qPCR normalization. Nucleic Acids Res, (2017).

    Article  Google Scholar 

  47. 47.

    Vanhauwaert, S. et al. RT-qPCR gene expression analysis in zebrafish: Preanalytical precautions and use of expressed repetitive elements for normalization. Methods Cell Biol 135, 329–342, (2016).

    CAS  Article  PubMed  Google Scholar 

  48. 48.

    Kell, D. B. & Oliver, S. G. Here is the evidence, now what is the hypothesis? The complementary roles of inductive and hypothesis-driven science in the post-genomic era. Bioessays 26, 99–105 (2004).

    Article  Google Scholar 

  49. 49.

    Eisenberg, E. & Levanon, E. Y. Human housekeeping genes, revisited. Trends Genet 29, 569–574, (2013).

    CAS  Article  PubMed  Google Scholar 

  50. 50.

    Hoerndli, F. J., Toigo, M., Schild, A., Götz, J. & Day, P. J. Reference genes identified in SH-SY5Y cells using custom-made gene arrays with validation by quantitative polymerase chain reaction. Anal Biochem 335, 30–41 (2004).

    CAS  Article  Google Scholar 

  51. 51.

    Ohl, F. et al. Gene expression studies in prostate cancer tissue: which reference gene should be selected for normalization? J Mol Med (Berl) 83, 1014–1024, (2005).

    CAS  Article  Google Scholar 

  52. 52.

    Silver, N., Best, S., Jiang, J. & Thein, S. L. Selection of housekeeping genes for gene expression studies in human reticulocytes using real-time PCR. BMC Mol Biol 7, 33, (2006).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  53. 53.

    de Jonge, H. J. M. et al. Evidence based selection of housekeeping genes. PLoS One 2, e898, (2007).

    ADS  CAS  Article  PubMed  PubMed Central  Google Scholar 

  54. 54.

    Tatsumi, K. et al. Reference gene selection for real-time RT-PCR in regenerating mouse livers. Biochem Biophys Res Commun 374, 106–110, (2008).

    CAS  Article  PubMed  Google Scholar 

  55. 55.

    Bustin, S. A. et al. The MIQE guidelines: minimum information for publication of quantitative real-time PCR experiments. Clin Chem 55, 611–622, (2009).

    CAS  Article  PubMed  Google Scholar 

  56. 56.

    Gur-Dedeoglu, B. et al. Identification of endogenous reference genes for qRT-PCR analysis in normal matched breast tumor tissues. Oncol Res 17, 353–365 (2009).

    Article  Google Scholar 

  57. 57.

    Li, Y. L., Ye, F., Hu, Y., Lu, W. G. & Xie, X. Identification of suitable reference genes for gene expression studies of human serous ovarian cancer by real-time polymerase chain reaction. Anal Biochem 394, 110–116, (2009).

    CAS  Article  PubMed  Google Scholar 

  58. 58.

    Thellin, O., ElMoualij, B., Heinen, E. & Zorzi, W. A decade of improvements in quantification of gene expression and internal standard selection. Biotechnol Adv 27, 323–333 (2009).

    CAS  Article  Google Scholar 

  59. 59.

    Chervoneva, I. et al. Selection of optimal reference genes for normalization in quantitative RT-PCR. BMC Bioinformatics 11, 253, (2010).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  60. 60.

    Wang, F., Wang, J., Liu, D. & Su, Y. Normalizing genes for real-time polymerase chain reaction in epithelial and nonepithelial cells of mouse small intestine. Anal Biochem 399, 211–217, (2010).

    CAS  Article  PubMed  Google Scholar 

  61. 61.

    Zampieri, M. et al. Validation of suitable internal control genes for expression studies in aging. Mech Ageing Dev 131, 89–95, (2010).

    CAS  Article  PubMed  Google Scholar 

  62. 62.

    Casadei, R. et al. Identification of housekeeping genes suitable for gene expression analysis in the zebrafish. Gene Expr Patterns 11, 271–276, (2011).

    CAS  Article  PubMed  Google Scholar 

  63. 63.

    Jacob, F. et al. Careful selection of reference genes is required for reliable performance of RT-qPCR in human normal and cancer cell lines. PLoS One 8, e59180, (2013).

    ADS  CAS  Article  PubMed  PubMed Central  Google Scholar 

  64. 64.

    Oturai, D. B., Sondergaard, H. B., Bornsen, L., Sellebjerg, F. & Christensen, J. R. Identification of Suitable Reference Genes for Peripheral Blood Mononuclear Cell Subset Studies in Multiple Sclerosis. Scand J Immunol 83, 72–80, (2016).

    CAS  Article  PubMed  Google Scholar 

  65. 65.

    Caracausi, M. et al. Systematic identification of human housekeeping genes possibly useful as references in gene expression studies. Mol Med Rep 16, 2397–2410, (2017).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  66. 66.

    Vandesompele, J. et al. Accurate normalization of real-time quantitative RT-PCR data by geometric averaging of multiple internal control genes. Genome Biol 3, RESEARCH0034 (2002).

    Article  Google Scholar 

  67. 67.

    Butte, A. J., Dzau, V. J. & Glueck, S. B. Further defining housekeeping, or “maintenance,” genes Focus on “A compendium of gene expression in normal human tissues”. Physiol Genomics 7, 95–96 (2001).

    CAS  Article  Google Scholar 

  68. 68.

    Hsiao, L. L. et al. A compendium of gene expression in normal human tissues. Physiol Genomics 7, 97–104, (2001).

    CAS  Article  PubMed  Google Scholar 

  69. 69.

    Lee, P. D., Sladek, R., Greenwood, C. M. & Hudson, T. J. Control genes and variability: absence of ubiquitous reference transcripts in diverse mammalian expression studies. Genome Res 12, 292–297, (2002).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  70. 70.

    Eisenberg, E. & Levanon, E. Y. Human housekeeping genes are compact. Trends Genet 19, 362–365, (2003).

    CAS  Article  PubMed  Google Scholar 

  71. 71.

    Dheda, K. et al. Validation of housekeeping genes for normalizing RNA expression in real-time PCR. Biotechniques 37, 112–114, 116, 118–119 (2004).

    CAS  Article  Google Scholar 

  72. 72.

    Barber, R. D., Harmer, D. W., Coleman, R. A. & Clark, B. J. GAPDH as a housekeeping gene: analysis of GAPDH mRNA expression in a panel of 72 human tissues. Physiol Genomics 21, 389–395, (2005).

    CAS  Article  PubMed  Google Scholar 

  73. 73.

    Rubie, C. et al. Housekeeping gene variability in normal and cancerous colorectal, pancreatic, esophageal, gastric and hepatic tissues. Mol Cell Probes 19, 101–109, (2005).

    CAS  Article  PubMed  Google Scholar 

  74. 74.

    Szabo, A. et al. Statistical modeling for selecting housekeeper genes. Genome Biol 5, R59, (2004).

    Article  PubMed  PubMed Central  Google Scholar 

  75. 75.

    Mane, V. P., Heuer, M. A., Hillyer, P., Navarro, M. B. & Rabin, R. L. Systematic method for determining an ideal housekeeping gene for real-time PCR analysis. J Biomol Tech 19, 342–347 (2008).

    PubMed  PubMed Central  Google Scholar 

  76. 76.

    Teste, M. A., Duquenne, M., François, J. M. & Parrou, J. L. Validation of reference genes for quantitative expression analysis by real-time RT-PCR in Saccharomyces cerevisiae. BMC Mol Biol 10, 99, (2009).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  77. 77.

    Robinson, M. D. & Oshlack, A. A scaling normalization method for differential expression analysis of RNA-seq data. Genome Biol 11, R25, (2010).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  78. 78.

    Kozera, B. & Rapacz, M. Reference genes in real-time PCR. J Appl Genet 54, 391–406, (2013).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  79. 79.

    De Spiegelaere, W. et al. Reference gene validation for RT-qPCR, a note on different available software packages. PLoS One 10, e0122515, (2015).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  80. 80.

    Papatheodorou, I. et al. Expression Atlas: gene and protein expression across multiple studies and organisms. Nucleic Acids Res 46, D246–D251, (2018).

    CAS  Article  PubMed  Google Scholar 

  81. 81.

    Mortazavi, A., Williams, B. A., McCue, K., Schaeffer, L. & Wold, B. Mapping and quantifying mammalian transcriptomes by RNA-Seq. Nat Methods 5, 621–628, (2008).

    CAS  Article  PubMed  Google Scholar 

  82. 82.

    Wang, Z., Gerstein, M. & Snyder, M. RNA-Seq: a revolutionary tool for transcriptomics. Nat Rev Genet 10, 57–63, (2009).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  83. 83.

    Oshlack, A., Robinson, M. D. & Young, M. D. From RNA-seq reads to differential expression results. Genome Biol 11, 220, (2010).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  84. 84.

    Xu, J. et al. Comprehensive Assessments of RNA-seq by the SEQC Consortium: FDA-Led Efforts Advance Precision Medicine. Pharmaceutics 8, (2016).

    Article  Google Scholar 

  85. 85.

    Bray, N. L., Pimentel, H., Melsted, P. & Pachter, L. Near-optimal probabilistic RNA-seq quantification. Nat Biotechnol 34, 525–527, (2016).

    CAS  Article  PubMed  Google Scholar 

  86. 86.

    Grabherr, M. G. et al. Full-length transcriptome assembly from RNA-Seq data without a reference genome. Nat Biotechnol 29, 644–652, (2011).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  87. 87.

    Schulz, M. H., Zerbino, D. R., Vingron, M. & Birney, E. Oases: robust de novo RNA-seq assembly across the dynamic range of expression levels. Bioinformatics 28, 1086–1092, (2012).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  88. 88.

    Tang, F. et al. mRNA-Seq whole-transcriptome analysis of a single cell. Nat Methods 6, 377–382, (2009).

    CAS  Article  PubMed  Google Scholar 

  89. 89.

    Macosko, E. Z. et al. Highly Parallel Genome-wide Expression Profiling of Individual Cells Using Nanoliter Droplets. Cell 161, 1202–1214, (2015).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  90. 90.

    Rissin, D. M. & Walt, D. R. Digital concentration readout of single enzyme molecules using femtoliter arrays and Poisson statistics. Nano Lett 6, 520–523, (2006).

    ADS  CAS  Article  PubMed  Google Scholar 

  91. 91.

    Salehi-Reyhani, A. et al. Scaling advantages and constraints in miniaturized capture assays for single cell protein analysis. Lab Chip 13, 2066–2074, (2013).

    CAS  Article  PubMed  Google Scholar 

  92. 92.

    Hudecova, I. Digital PCR analysis of circulating nucleic acids. Clin Biochem 48, 948–956, (2015).

    CAS  Article  PubMed  Google Scholar 

  93. 93.

    Thul, P. J. et al. A subcellular map of the human proteome. Science 356, (2017).

    Article  Google Scholar 

  94. 94.

    Wu, Y. et al. Function of HNRNPC in breast cancer cells by controlling the dsRNA-induced interferon response. The EMBO Journal 37, e99017, (2018).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  95. 95.

    Bomsztyk, K., Denisenko, O. & Ostrowski, J. hnRNP K: One protein multiple processes. BioEssays 26, 629–638, (2004).

    CAS  Article  PubMed  Google Scholar 

  96. 96.

    Makeyev, A. V. & Liebhaber, S. A. The poly (C)-binding proteins: a multiplicity of functions and a search for mechanisms. Rna 8, 265–278 (2002).

    CAS  Article  Google Scholar 

  97. 97.

    Huo, L.-R. & Zhong, N. Identification of transcripts and translatants targeted by overexpressed PCBP1. Biochimica et Biophysica Acta (BBA)-Proteins and Proteomics 1784, 1524–1533 (2008).

    CAS  Article  Google Scholar 

  98. 98.

    Cho, S.-J., Jung, Y.-S. & Chen, X. Poly (C)-binding protein 1 regulates p63 expression through mRNA stability. PloS one 8, e71724–e71724, (2013).

    ADS  CAS  Article  PubMed  PubMed Central  Google Scholar 

  99. 99.

    Lardelli, R. M., Thompson, J. X., Yates, J. R. & Stevens, S. W. Release of SF3 from the intron branchpoint activates the first step of pre-mRNA splicing. Rna (2010).

  100. 100.

    Kfir, N. et al. SF3B1 Association with Chromatin Determines Splicing Outcomes. Cell Reports 11, 618–629, (2015).

    CAS  Article  PubMed  Google Scholar 

  101. 101.

    Effenberger, K. A., Urabe, V. K., Prichard, B. E., Ghosh, A. K. & Jurica, M. S. Interchangeable SF3B1 inhibitors interfere with pre-mRNA splicing at multiple stages. RNA 22, 350–359, (2016).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  102. 102.

    He, X. & Zhang, P. Serine/arginine-rich splicing factor 3 (SRSF3) regulates homologous recombination-mediated DNA repair. Molecular Cancer 14, 158, (2015).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  103. 103.

    Gallardo, M. et al. hnRNP K Is a Haploinsufficient Tumor Suppressor that Regulates Proliferation and Differentiation Programs in Hematologic Malignancies. Cancer Cell 28, 486–499, (2015).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  104. 104.

    Barboro, P. et al. Heterogeneous nuclear ribonucleoprotein K: altered pattern of expression associated with diagnosis and prognosis of prostate cancer. British Journal Of Cancer 100, 1608, (2009).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  105. 105.

    Park, Y. M. et al. Heterogeneous Nuclear Ribonucleoprotein C1/C2 Controls the Metastatic Potential of Glioblastoma by Regulating PDCD4. Molecular and Cellular Biology 32, 4237, (2012).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  106. 106.

    Lee, E. K. et al. hnRNP C promotes APP translation by competing with FMRP for APP mRNA recruitment to P bodies. Nature structural & molecular biology 17, 732–739, (2010).

    CAS  Article  Google Scholar 

  107. 107.

    Zarnack, K. et al. Direct Competition between hnRNP C and U2AF65 Protects the Transcriptome from the Exonization of Alu Elements. Cell 152, 453–466, (2013).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  108. 108.

    Wang, H. et al. PCBP1 Suppresses the Translation of Metastasis-Associated PRL-3 Phosphatase. Cancer Cell 18, 52–62, (2010).

    CAS  Article  PubMed  Google Scholar 

  109. 109.

    Zhang, T. et al. PCBP-1 regulates alternative splicing of the CD44 gene and inhibits invasion in human hepatoma cell line HepG2 cells. Molecular Cancer 9, 72, (2010).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  110. 110.

    Liu, Y. et al. Expression of poly(C)-binding protein 1 (PCBP1) in NSCLC as a negative regulator of EMT and its clinical value. International journal of clinical and experimental pathology 8, 7165–7172 (2015).

    CAS  PubMed  PubMed Central  Google Scholar 

  111. 111.

    Zhang, Z.-Z. et al. HOTAIR Long Noncoding RNA Promotes Gastric Cancer Metastasis through Suppression of Poly r(C)-Binding Protein (PCBP) 1. Molecular Cancer Therapeutics 14, 1162, (2015).

    CAS  Article  PubMed  Google Scholar 

  112. 112.

    Wagener, R. et al. The PCBP1 gene encoding poly(rc) binding protein i is recurrently mutated in Burkitt lymphoma. Genes, Chromosomes and Cancer 54, 555–564, (2015).

    CAS  Article  PubMed  Google Scholar 

  113. 113.

    Ji, F.-J. et al. Expression of both poly r(C) binding protein 1 (PCBP1) and miRNA-3978 is suppressed in peritoneal gastric cancer metastasis. Scientific reports 7, 15488–15488, (2017).

    ADS  CAS  Article  PubMed  PubMed Central  Google Scholar 

  114. 114.

    Jumaa, H., Wei, G. & Nielsen, P. J. Blastocyst formation is blocked in mouse embryos lacking the splicing factor SRp20. Current Biology 9, 899–902, (1999).

    CAS  Article  PubMed  Google Scholar 

  115. 115.

    Palmieri, F. The mitochondrial transporter family SLC25: Identification, properties and physiopathology. Mol Aspects Med 34, 465–484, (2013).

    CAS  Article  PubMed  Google Scholar 

  116. 116.

    Schnabel, M. et al. Dedifferentiation-associated changes in morphology and gene expression in primary human articular chondrocytes in cell culture. Osteoarthritis and Cartilage 10, 62–70, (2002).

    CAS  Article  PubMed  Google Scholar 

  117. 117.

    Cullen, P. J. Endosomal sorting and signalling: an emerging role for sorting nexins. Nature Reviews Molecular Cell Biology 9, 574, (2008).

    CAS  Article  PubMed  Google Scholar 

  118. 118.

    Naslavsky, N. & Caplan, S. The enigmatic endosome – sorting the ins and outs of endocytic trafficking. Journal of Cell Science 131, jcs216499, (2018).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  119. 119.

    Chen, C. et al. Snx3 Regulates Recycling of the Transferrin Receptor and Iron Assimilation. Cell Metabolism 17, 343–352, (2013).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  120. 120.

    Xu, S., Nigam, S. M. & Brodin, L. Overexpression of SNX3 Decreases Amyloid-β Peptide Production by Reducing Internalization of Amyloid Precursor Protein. Neurodegenerative Diseases 18, 26–37, (2018).

    CAS  Article  PubMed  Google Scholar 

  121. 121.

    Binder, N. K., Sheedy, J. R., Hannan, N. J. & Gardner, D. K. Male obesity is associated with changed spermatozoa Cox4i1 mRNA level and altered seminal vesicle fluid composition in a mouse model. MHR: Basic science of reproductive medicine 21, 424–434, (2015).

    CAS  Article  PubMed  Google Scholar 

  122. 122.

    Li, Y., Park, J.-S., Deng, J.-H. & Bai, Y. Cytochrome c oxidase subunit IV is essential for assembly and respiratory function of the enzyme complex. Journal of Bioenergetics and Biomembranes 38, 283–291, (2006).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  123. 123.

    Storey, J. D. et al. Gene-Expression Variation Within and Among Human Populations. The American Journal of Human Genetics 80, 502–509, (2007).

    CAS  Article  PubMed  Google Scholar 

  124. 124.

    Lonsdale, J. et al. The Genotype-Tissue Expression (GTEx) project. Nature Genetics 45, 580,, (2013).

    CAS  Article  Google Scholar 

  125. 125.

    Pickrell, J. K. et al. Understanding mechanisms underlying human gene expression variation with RNA sequencing. Nature 464, 768,, (2010).

    ADS  CAS  Article  Google Scholar 

  126. 126.

    Zhang, X., Ding, L. & Sandford, A. J. Selection of reference genes for gene expression studies in human neutrophils by real-time PCR. BMC Mol Biol. 18, 4 (2005).

    CAS  Article  Google Scholar 

  127. 127.

    Kriegova, E. et al. PSMB2 and RPL32 are suitable denominators to normalize gene expression profiles in bronchoalveolar cells. BMC Mol Biol. 31, 69 (2008).

    Article  Google Scholar 

  128. 128.

    Mazanetz, M. P., Marmon, R. J., Reisser, C. B. T. & Morao, I. Drug discovery applications for KNIME: an open source data mining platform. Curr Top Med Chem 12, 1965–1979, (2012).

    CAS  Article  PubMed  Google Scholar 

  129. 129.

    Fillbrunn, A. et al. KNIME for reproducible cross-domain analysis of life science data. J Biotechnol, (2017).

    CAS  Article  Google Scholar 

  130. 130.

    O’Hagan, S. & Kell, D. B. The KNIME workflow environment and its applications in Genetic Programming and machine learning. Genetic Progr Evol Mach 16, 387–391, (2015).

    Article  Google Scholar 

  131. 131.

    Lee, S., Jo, M., Lee, J., Koh, S. S. & Kim, S. Identification of novel universal housekeeping genes by statistical analysis of microarray data. J Biochem Mol Biol 40, 226–231 (2007).

    CAS  PubMed  Google Scholar 

  132. 132.

    Greer, S., Honeywell, R., Geletu, M., Arulanandam, R. & Raptis, L. Housekeeping genes; expression levels may change with density of cultured cells. Journal of Immunological Methods 355, 76–79, (2010).

    CAS  Article  PubMed  Google Scholar 

  133. 133.

    Li, R. & Shen, Y. An old method facing a new challenge: Re-visiting housekeeping proteins as internal reference control for neuroscience research. Life Sciences 92, 747–751, (2013).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  134. 134.

    Huggett, J., Dheda, K., Bustin, S. & Zumla, A. Real-time RT-PCR normalisation; strategies and considerations. Genes Immun 6, 279–284, (2005).

    CAS  Article  PubMed  Google Scholar 

  135. 135.

    Andersen, C. L., Jensen, J. L. & Orntoft, T. F. Normalization of real-time quantitative reverse transcription-PCR data: a model-based variance estimation approach to identify genes suited for normalization, applied to bladder and colon cancer data sets. Cancer Res 64, 5245–5250, (2004).

    CAS  Article  PubMed  Google Scholar 

  136. 136.

    Pfaffl, M. W., Tichopad, A., Prgomet, C. & Neuvians, T. P. Determination of stable housekeeping genes, differentially regulated target genes and sample integrity: BestKeeper–Excel-based tool using pair-wise correlations. Biotechnol Lett 26, 509–515 (2004).

    CAS  Article  Google Scholar 

  137. 137.

    Xie, F., Xiao, P., Chen, D., Xu, L. & Zhang, B. miRDeepFinder: a miRNA analysis tool for deep sequencing of plant small RNAs. Plant Mol Biol, (2012).

    CAS  Article  Google Scholar 

  138. 138.

    Livak, K. J. & Schmittgen, T. D. Analysis of Relative Gene Expression Data Using Real-Time Quantitative PCR and the 2−ΔΔCT Method. Methods 25, 402–408, (2001).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  139. 139.

    Conway, J. R., Lex, A. & Gehlenborg, N. UpSetR: an R package for the visualization of intersecting sets and their properties. Bioinformatics 33, 2938–2940, (2017).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  140. 140.

    Uhlen, M. et al. Proteomics. Tissue-based map of the human proteome. Science 347, 1260419, (2015).

    CAS  Article  PubMed  Google Scholar 

  141. 141.

    Barretina, J. et al. The Cancer Cell Line Encyclopedia enables predictive modelling of anticancer drug sensitivity. Nature 483, 603–607, (2012).

    ADS  CAS  Article  PubMed  PubMed Central  Google Scholar 

  142. 142.

    Klijn, C. et al. A comprehensive transcriptional portrait of human cancer cell lines. Nat Biotechnol 33, 306–312, (2015).

    CAS  Article  PubMed  Google Scholar 

  143. 143.

    Consortium, G. T. Human genomics. The Genotype-Tissue Expression (GTEx) pilot analysis: multitissue gene regulation in humans. Science 348, 648–660, (2015).

    CAS  Article  Google Scholar 

Download references


All authors thank the BBSRC (grant BB/P009042/1) and the Novo Nordisk Foundation (grant NNF10CC1016517) for financial support.

Author information




D.B.K. highlighted the utility of the G.C. as shown in reference 1. M.W.M. adapted the Gini method and analyses workflows developed by S.O. from reference 1 and performed most of the analyses that were done using KNIME. P.J.D. contributed in particular to the analysis of the housekeeping genes. F.M. performed the RT-qPCR analyses. All authors contributed to the writing and approval of the manuscript.

Corresponding authors

Correspondence to Marina Wright Muelas, Philip J. Day or Douglas B. Kell.

Ethics declarations

Competing interests

The authors declare no competing interests.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Wright Muelas, M., Mughal, F., O’Hagan, S. et al. The role and robustness of the Gini coefficient as an unbiased tool for the selection of Gini genes for normalising expression profiling data. Sci Rep 9, 17960 (2019).

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI:

Further reading


By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.


Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing