Brain Proteomic Responses to Glucocorticoids and Their Relationship With Transcriptome: A Systematic Meta‐Analysis

ABSTRACT

Although there are several proteomic studies testing brain responses to glucocorticoids, there were no attempts to integrate these data and compare them with responses at the level of mRNAs. Furthermore, the utility of available data is compromised by changes in nomenclature and usage of different types of identifiers. Therefore, the aim of this study was to identify the most consistent changes in protein expression in standardized mouse, rat, and human datasets and compare them with transcriptomic responses to glucocorticoids. The analysis showed that the two most frequently and consistently detected proteins were ATP synthase F1 subunit beta (Atp5f1b) and aldolase, fructose‐bisphosphate C (Aldoc), while the most consistent proteomic and transcriptomic findings included Aldoc, Plin4, Aqp4, Endod1, Glul, Anln, Aldh1l1, Parp1, Trf, Fermt2, Tmem63a, and Trim2. The study also revealed limitations of available proteomic data indicating significant gaps in knowledge. Finally, the study provides an integrated dataset with updated protein nomenclature and a complete set of major identifiers to facilitate usage of proteomic data.

An overlap between proteomic and transcriptomic findings in experiments testing brain responses to glucocorticoids. A shows that 455 proteins responsive to glucocorticoids were not confirmed at the transcriptomic level while 4 proteins were matched with significant differences in mRNA in all 8 datasets derived from transcriptomic experiments. B shows glucocorticoid‐responsive proteins matched with transcriptomic responses that were reported in at least 5 transcriptomic datasets. GCs, glucocorticoids; No., number.

1. Introduction

Endogenous glucocorticoids are hormones that regulate various processes during the sleep–wake cycle and in response to stress [1, 2, 3], whereas synthetic glucocorticoids are used in pharmacology because of their potent anti‐inflammatory properties [4, 5]. The release of glucocorticoid hormones is tightly regulated [6] because both insufficient [7, 8] and excessive levels of glucocorticoids [9] have serious health consequences. Although it is known that the effects of glucocorticoids are mediated by glucocorticoid and mineralocorticoid receptors acting as transcription factors [10, 11], the signaling cascades resulting from altered gene expression are still poorly understood [12]. The most important end products of gene expression are proteins which are building blocks of life due to their central role in energy production, control of chemical reactions, information exchange, transportation, cell structure and movement of cellular components. However, despite the importance of proteins, the majority of experiments focus on the intermediate step of protein synthesis (mRNA) due to methodological developments enabling a relatively straightforward analysis of mRNAs [13]. This approach was also based on the assumption that the level of mRNAs is a reliable proxy for biological processes [14]. This simplistic view has been, however, challenged by the studies showing more complex relationships between mRNAs and proteins that are affected not only by the rate of mRNA synthesis but also by subsequent translation, secretion and degradation [15, 16]. Disparities between transcriptome and proteome are observed not only in basal conditions [15, 16] but also during evoked responses. For example, data from peripheral [17] and central tissues [12] indicate disparities between mRNA and proteins during responses to glucocorticoids. It is hypothesized that some transcriptomic responses to glucocorticoids may constitute only a first preparatory stage of response that is not translated to proteins until the organism experiences a specific threat to its homeostasis [12]. On the other hand, some proteins may not be effectively regulated at the level of mRNAs due to limitations imposed by the speed of mRNA synthesis [12]. Therefore, it is important to understand genomic responses both at the transcriptomic and proteomic level. However, although there are several proteomic studies testing brain responses to glucocorticoids, there were no attempts to integrate these data and compare them with responses at the level of mRNAs. Furthermore, the utility of available data is compromised by changes in nomenclature and usage of different types of identifiers [18, 19]. Therefore, the aim of this study was to identify the most consistent changes in protein expression in standardized datasets and compare them with transcriptomic responses to natural and synthetic glucocorticoids (Table 1). The study also revealed limitations of available proteomic data indicating significant gaps in knowledge. Finally, the study provides an integrated dataset with updated protein nomenclature and a complete set of major identifiers to facilitate usage of data that accumulated over time.

TABLE 1.

Characteristics of glucocorticoids used in the analyzed studies.

Glucocorticoid	Characteristics
Cortisol	The primary glucocorticoid hormone in the majority of mammalian species [20, 21], including humans [22]. It exhibits 13.4‐fold a greater potency in activating mineralocorticoid receptors than glucocorticoid receptors in an invitro assay [23] but its mineralocorticoid activity is still 18.6‐fold a lower than that of aldosterone, the natural mineralocorticoid. Additionally, the mineralocorticoid activity of cortisol is diminished in some tissues (kidney, parotid and colon) by enzymatic transformation to an inactive steroid [24]
Corticosterone	The primary glucocorticoid hormone in mice [25], rats, and rabbits [20]. It binds both to glucocorticoid and mineralocorticoid receptors. Its affinity for mineralocorticoid receptors is comparable to that of aldosterone [26] and is 6‐ to 10‐fold higher than its affinity for glucocorticoid receptors [27]. The mineralocorticoid activity of corticosterone is diminished in some tissues (kidney and parotid) by enzymatic transformation to an inactive steroid [28]
Dexamethasone	A synthetic steroid used in medicine as a very potent glucocorticoid with no mineralocorticoid activity [29] In an in vitro assay [23], it exhibits 9.1‐fold a greater potency in activating glucocorticoid receptors compared with mineralocorticoid receptors. Furthermore, its mineralocorticoid activity is 5.7‐fold a lower than that of cortisol, the natural glucocorticoid, and 106‐fold a lower than that of aldosterone, the natural mineralocorticoid, while its glucocorticoid activity is 21.4‐fold a higher than that of cortisol
Methylprednisolone (6α‐Methylprednisolone)	A synthetic steroid used in medicine as a potent glucocorticoid with minimal mineralocorticoid activity [29]. Although it exhibits 1.26‐fold a greater potency in activating mineralocorticoid receptors than glucocorticoid receptors in an in vitro assay [23], its mineralocorticoid activity remains lower than that of endogenous hormones. Specifically, its mineralocorticoid activity is 2.6‐fold a lower than that of cortisol, the natural glucocorticoid, and 48.1‐fold a lower than that of aldosterone, the natural mineralocorticoid, while its glucocorticoid activity is 4.1‐fold a higher than that of cortisol
Prednisone	A synthetic steroid used in medicine as a potent glucocorticoid with minimal mineralocorticoid activity [29]. Prednisone is an inactive prodrug [23, 30] lacking affinity to glucocorticoid receptors [31, 32]. Under in vivo conditions, it is converted to prednisolone [33], which is responsible for its biological activity. Although prednisolone is a 1.8‐fold a more potent activator of mineralocorticoid receptors than glucocorticoid receptors in an in vitro assay [23], its mineralocorticoid activity remains lower than that of endogenous hormones. Specifically, its mineralocorticoid activity is 4.2‐fold a lower than that of cortisol, the natural glucocorticoid, and 78.8‐fold a lower than that of aldosterone, the natural mineralocorticoid, while its glucocorticoid activity is 1.7‐fold a higher than that of cortisol

^a Assessment based on EC₅₀ values (half maximal effective concentration) derived from [23].

2. Methods

2.1. Literature Search

This study was conducted in accordance with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta‐Analyses) guidelines [34] (File 1, https://data.mendeley.com/datasets/tw95jjrpxr/3). The study was not registered, and no review protocol was published elsewhere. The aim of the literature search was identification of original proteomic studies that tested changes in protein expression in brain tissue or in vitro brain cultures after treatment with glucocorticoids without a drug washout period. The author searched the PubMed database with words ‘brain’ and ‘proteome’ that were combined consecutively with one of the following: corticosterone, cortisol, dexamethasone, betamethasone, methylprednisolone, prednisone, and glucocorticoids (Figure 1). The initial screening of the literature was based on the data available in abstracts and was followed by inspection of full texts of selected papers to identify studies that provided lists of proteins with altered expression within about 24 h after the last treatment with glucocorticoids. The final search performed between 14 and 16 March 2025 revealed 10 studies (Table 2) that met our selection criteria. One of these studies [39] turned out to be a methodological paper describing a detailed protocol with the reanalysis of the data from Skynner et al. [44]. Therefore, the data from Guest [39] were restricted to new findings while the remaining overlapping results were discarded and included only in the dataset from Skynner et al. [44]. The selection of data was performed after the final update of the nomenclature.

FIGURE 1.

The flow diagram of the article screening and selection process. GCs, glucocorticoids; n, number of papers.

TABLE 2.

Summary of proteomic studies included in the present study.

Paper	Species	N	Drug	Route	Time	Brain area/cell type
Al‐Mayyahi et al. [35]	Mice CD1	4	Methylprednisolone 10 μM	In vitro medium	2 days	Neural stem cells from SVZ
Chitu et al. [36]	Male WT and Csf1r+/‐C57BL/6Jmice	4	Prednisone 1.8 mg/kg/day	s.c. pellets	12 months	Microglia and oligodendrocytes
Feldmann et al. [37]	Male Wistar rats	7 rats/3 samples after pooling	Corticosterone 26.8 mg/kg (about 15 mg/day)	s.c. injections	60 days	Hippocampus
Gong et al. [38]	Mice	3	Corticosterone (100 μmol/L)	In vitro medium	7 days	Neuronal C17.2 stem cells
Guest [39]	Male CD1 mice	8	Corticosterone (4 × 5 mg)	s.c. pellets	14 days	Hypothalamus Hippocampus
Kuznetsova et al. [40]	Female WT and 5‐HTT KO C57BL/6J mice	6–8	Corticosterone 25 mg/L	Drinking water	4 weeks	Hippocampus
Malkawi et al. [41]	Male SD rats	5	Dexamethasone 2.5 mg/kg twice a week	i.m.	14 weeks	Whole brain
Niu et al. [42]	Male rats Sprague–Dawley	5	Dexamethasone 10 mg/kg	i.p.	2 days	Hippocampus
Notaras et al. [43]	Male and female humans	4	Cortisol 10 μg/mL	In vitro medium	7 days	Dorsal forebrain organoids grown from Pluripotent Stem Cells
Skynner et al. [44]	Male CD1 mice	8	Corticosterone 20 mg (40 mg/kg/day)	s.c. pellets	14 days	Hypothalamus Hippocampus Cerebral cortex

Abbreviation: SVZ, subventricular zone.

2.2. Summary of Data Retrieval and Standardization

Proteomic data that included protein identifiers and direction of responses to glucocorticoids were retrieved from papers or supplementary files listed in Table 3.

TABLE 3.

Source of proteomic data used in the analysis.

Paper	Source of data
Al‐Mayyahi et al. [35]	Text and figure 5
Chitu et al. [36]	Tables S1 and S2
Feldmann et al. [37]	Paper, table 2
Gong et al. [38]	Paper, figures 3C and 5AB
Guest [39]	Paper, tables 2 and 3
Kuznetsova et al. [40]	Table S14
Malkawi et al. [41]	Table S1
Niu et al. [42]	Paper, table 1
Notaras et al. [43]	Table S2
Skynner et al. [44]	Paper, table 3

Additionally, data retrieved from Chitu et al. [36] were reanalyzed with two‐tailed Student's t‐test because the original data provided in supplementary files contain counterintuitive absolute values of log₂ transformed p‐values (−log₂ p‐values) obtained with this test. P‐values range from 0 to 1, with smaller values indicating greater statistical significance. The −log₂ transformation changes both the range of possible values and their interpretation because a p‐value of 1 is transformed to 0, while highly significant findings are indicated by values much greater than 1 (e.g., a p‐value of 0.0001 is transformed to 13.3). The results of the reanalysis were checked for consistency with the original data and, next were used to select significant differences indicated by p < 0.05.

A preliminary inspection of retrieved data revealed that the analyzed studies differed largely in the reported type of protein identifiers, preventing any direct comparison between collected datasets. Furthermore, identifiers may change over time due to the update and standardization of the nomenclature. Therefore, the data required standardization of protein and gene identifiers to enable the comparison between various datasets. The standardization was preferentially carried out based on unique identifiers such as UniProt or NCBI gene IDs. The second choice was protein and gene symbols (abbreviated names) while full names that are least reliable [18] were used as a last resort to support disambiguation of symbols. The preference for unique IDs over symbols results from the fact that some symbols are ambiguous and can be assigned to more than one gene product [19]. The update included retrieval of a complete set of IDs assigned by UniProt, NCBI, and Ensembl to facilitate search for additional data stored in the major databases. The data from the UniProt and NCBI databases were downloaded manually and then imported with the R scripts to the R studio environment (R version 4.4.2; RStudio 2024.12.0, Build 467) while Ensembl data were downloaded automatically by the scripts with the help of the biomaRt R package [45, 46] following the procedure used previously [19].

2.3. Databases

Data associated with UniProt accession numbers were retrieved from Retrieve/ID mapping online tool available at https://www.uniprot.org/id‐mapping with the search performed from UniProtKB AC/ID database to UniProtKB database. The selected output data were entry name, gene names (primary), organism, organism (ID), protein names (from UniProt Data/Names & Taxonomy directory), GeneID (from External Resources/Genome annotation directory) and one of the IDs assigned by nomenclature committees (mouse MGI, rat RGD or human HGNC) (from External Resources/Organism‐specific directory). The search results were downloaded in a tsv file for further work (saved as UniProt_data.tsv).

NCBI data for all available genes were downloaded from https://www.ncbi.nlm.nih.gov/datasets/gene/taxon/10116/ (rat genome), https://www.ncbi.nlm.nih.gov/datasets/gene/taxon/10090/ (mouse genome) and https://www.ncbi.nlm.nih.gov/datasets/gene/taxon/9606/ (human genome). Downloaded NCBI data (saved as ncbi_dataset.tsv) contained either all data available in the “Select columns” panel (mouse and rat genome) or only selected columns (NCBI GeneID, Symbol, Description, Gene Type, Nomenclature ID, Taxonomic ID, SwissProt Accessions) in case of human genome due to the problem with parsing the complete dataset.

The Ensembl data (release 113) were downloaded with the biomaRt R package that imported all mouse, rat, or human genes depending on the species investigated in individual papers. Retrieved data included official gene symbols, Ensembl stable gene IDs, NCBI IDs, nomenclature committee IDs (either mouse MGI, rat RGD, or human HGNC IDs), gene description, and gene type.

2.4. Updating Data Containing UniProt IDs

The update procedure had to be adjusted for data available in individual papers. The most common approach was the usage of the UniProt IDs from papers to download a set of UniProt data including NCBI gene IDs and identifiers assigned by a specialized nomenclature committee (mouse MGI, rat RGD and human HGNC IDs) that were next used to link the datasets with NCBI and Ensembl genomic data. This strategy was used in case of data retrieved from six studies [36, 40, 41, 42, 43, 44]. Preliminary tests showed that some UniProt IDs reported in papers were not recognized by the UniProt search tool https://www.uniprot.org/id‐mapping and were not found in the list of withdrawn IDs (https://www.uniprot.org/help/deleted_accessions). Such problematic IDs were manually checked by using other protein IDs included in the original data. The manual search enabled correction in the case of UniProt ID 81128 [44] that was changed to P81128 and UniProt ID BC027791 [44] that turned out to be a GenBank accession number for nucleotide (corrected to P16546). The joining of curated paper data with data available in UniProt, NCBI and Ensembl was performed with R scripts adjusted for analyzed species (R_Mouse_Uni_NCBI_Ens, R_Rat_Uni_NCBI_Ens and R_Human_Uni_NCBI_Ens) and deposited at https://github.com/Grzegorz‐R‐Juszczak/Scripts‐for‐Brain‐proteomic‐responses‐to‐GCs. The scripts matched data retrieved from different sources and compared gene symbols to indicate differences between databases and ambiguous IDs that can be assigned to more than one gene. The script also identifies cases when UniProt ID searched in UniProt database retrieves two new IDs assigned to different species due to the update of the nomenclature. An example of such ID is P50517 that has been split into mouse and rat variants of the same protein with new IDs P62814 and P62815 assigned in a species‐specific manner. The R scripts contain an optional code for the removal of such redundant entries assigned to other species than investigated in the analyzed dataset. Running the scripts requires installation of tidyr, biomaRt, readr, dplyr and purr R packages (Table 4).

TABLE 4.

R packages used in data analysis.

R package	Version	URL/Citation
tidyr	1.3.1	https://CRAN.R‐project.org/package=tidyr
dplyr	1.1.4	https://CRAN.R‐project.org/package=dplyr
purrr	1.0.2	https://CRAN.R‐project.org/package=purrr
readr	2.1.5	https://CRAN.R‐project.org/package=readr
naniar	1.1.0	https://CRAN.R‐project.org/package=naniar [47]
biomaRt	2.62.0	https://www.bioconductor.org/packages/release/bioc/html/biomaRt.html [45, 46]

2.5. Updating Data Containing GI IDs or Protein Symbols Together With Full Names

The second approach relied on manual assignment of NCBI and UniProt IDs to data retrieved from papers that provided either GeneBank GI (GenInfo) identifiers [37] or protein symbols together with full names [35, 39]. The NCBI GI numbers (www.ncbi.nlm.nih.gov/genbank/sequenceids/) and protein symbols were manually searched in the NCBI Protein database (https://www.ncbi.nlm.nih.gov/protein/) to assign UniProt protein IDs and NCBI gene IDs while full names were used to resolve ambiguities. The NCBI and UniProt IDs were retrieved directly from the NCBI Protein database whenever possible or, less frequently, from other resources based on UniProt, RGD or MGI IDs listed in NCBI protein database. Next, the paper data with manually added NCBI and UniProt IDs were joined with complete set of data retrieved from UniProt, NCBI and Ensembl, following the procedures described in section “Databases”. The integration of data retrieved from different sources was performed with R scripts R_Rat_manual_and_databases and R_Mouse_manual_and_databases (https://github.com/Grzegorz‐R‐Juszczak/Scripts‐for‐Brain‐proteomic‐responses‐to‐GCs), depending on the analyzed species in the original studies [35, 37, 39]. Running the scripts requires installation of tidyr, biomaRt, readr, dplyr and purr R packages (Table 4).

2.6. Updating Data Containing Only Gene Symbols

Finally, one study [38] provided only gene symbols associated with detected proteins without any other information enabling disambiguation of the data. Unfortunately, gene nomenclature changed over time, leading to ambiguity of some gene symbols that can be assigned now to more than one gene [19]. Therefore, I have used R script ‘R_Ensembl gene symbol search mouse’ (https://github.com/Grzegorz‐R‐Juszczak/Protein‐coding‐gene‐IDs‐human‐mouse‐rat‐pig) dedicated to updating gene symbols based on the data available in the Ensembl database [19]. The script performs a double Ensembl search for current official symbols followed by data integration, identification of ambiguous symbols and downloading stable IDs. The script used in this study was additionally customized to download UniProt IDs and saved as ‘R_Ensembl gene symbol search Gong et al. 2019’ (https://github.com/Grzegorz‐R‐Juszczak/Scripts‐for‐Brain‐proteomic‐responses‐to‐GCs). Running the script requires installation of tidyr, biomaRt, readr, dplyr, naniar and purr R packages (Table 4). The data obtained from Ensembl with the R script were screened for missing items and missing UniProt IDs were manually added based on the search of the UniProt database with identifiers downloaded from the Ensembl. The final list containing both UniProt IDs and Ensembl IDs was saved as Ensembl_UniProt_final.csv file. Next, the list of UniProt IDs (Ensembl_UniProt_final.csv) was combined with data from the paper (Paper_data.csv), Ensembl (Final_search_results.csv), NCBI (ncbi_dataset.tsv) and UniProt (UniProt_data.tsv) with the R script R_Mouse_Ens_Uni_NCBI (https://github.com/Grzegorz‐R‐Juszczak/Scripts‐for‐Brain‐proteomic‐responses‐to‐GCs) to obtain the complete set of data. Running the script requires installation of tidyr, readr and dplyr R packages (Table 4).

2.7. Integration of Data From Different Studies

Each updated dataset was first screened for missing items resulting from the lack of matching data in NCBI and Ensembl datasets used for automatic data update. The inspection revealed that some protein IDs reported in the analyzed datasets [37, 41, 44] belong to other species than animals used in experiments and, therefore, these IDs were not automatically matched with NCBI and Ensembl data that were restricted only to species used in experiments. Such unmatched IDs were searched in NCBI and Ensembl databases after the selection of proper species and missing data were filled manually whenever it was possible. Next, the finally updated data from individual studies were assembled into single datasets and analyzed with the R script R_Data_summary (https://github.com/Grzegorz‐R‐Juszczak/Scripts‐for‐Brain‐proteomic‐responses‐to‐GCs) to calculate the number of papers reporting each gene product, up fraction and expression score separately for NCBI and Ensembl gene symbols.

The up fraction (number of cases of increased expression divided by the total number of significant changes in expression) was calculated in two steps. First, the up fraction was calculated separately for each paper to avoid overestimation due to inclusion of papers presenting data from multiple groups and brain areas. Next, the up fractions from individual studies were used to calculate an average value of the parameter separately for each reported protein. Additionally, the down fraction was calculated by subtracting the average up fraction from 1.

The mean values of up and down fractions were next used to calculate the expression score that is a parameter representing jointly the frequency of appearance of each gene in separate papers and consistency of reported direction of altered expression [48]. The expression score is the result of multiplying the total number of papers reporting each protein by the mean up fraction in case of proteins that are more frequently upregulated or mean down fraction in case of proteins that are more frequently down‐regulated [48]. In case of proteins with an equal number of increased and decreased expression in different experiments, the multiplication by either up fraction or down fraction gives exactly the same value of the expression score. The expression score gives priority to genes that display the same direction of expression in different studies while lowers the rank of genes displaying a variable direction of responses that are more likely to result from signal noise than consistently regulated genes [48].

2.8. Statistics

The right‐sided binomial test [49] was used to assess the probability that the detected changes in protein expression across independent studies occurred more frequently than expected by chance. The calculations were performed using R code [50]:

probability = binom.test(x, n, p = 0.05, 
alternative = “greater”).

The terms used in the calculations were defined as follows: x—the number of independent studies that reported altered expression of a specific protein; n—the total number of independent studies (n = 9); p—the probability of detecting a significant effect in an individual experiment by chance.

The two‐sided binomial test [49] was used to assess the probability that the increased expression occurred either more (preferential up‐regulation) or less frequently (preferential down‐regulation) than expected by chance. The calculations were performed using R code [50]:

probability = binom.test(x, n, p = 0.5, 
alternative = “two.sided”)probability

The terms used in the calculations were defined as follows: x—the number of independent studies that reported significant upregulation of a specific protein; n—the total number of independent studies that reported a significant change in the expression of a specific protein; p—the probability of occurrence of one of two alternative states (up‐ or down‐regulation) in a single trial by chance.

The joint probability of detecting changes in expression of a specific protein and finding defined frequency of up‐regulation was calculated by multiplying the individual probabilities, For example, the probability of detecting a significant change in expression of a specific protein in two independent experiments (p‐value = 0.07121) that reported exclusively up‐regulation (p‐value = 0.5) is 0.035605 (0.07121 × 0.5).

Pearson's correlation and the exact two‐sided significance test were performed using R code [51, 52]:

cor(Data frame$Variable1, Data frame$Variable2, method = “pearson”)test <‐ cor.test(Data frame$Variable1, Data frame$Variable2, “two.sided”, method = “pearson”, exact = TRUE)test

3. Results

3.1. General Characteristics of the Analyzed Experiments

The analysis was performed on data retrieved from ten studies (Table 2) including one protocol paper [39] containing a reanalysis of earlier data [44]. In this case [39] only new findings were included in the analysis while the remaining overlapping results were discarded and included only in the dataset from Skynner et al. [44]. The studies summarized in Table 2 were performed on mice (6 studies), rats (3 studies), and human brain organoids (1 study). The experiments tested both natural (corticosterone and cortisol) and synthetic glucocorticoids (dexamethasone, methylprednisolone, and prednisone) applied for periods ranging from 2 days to 12 months (median = 14 days) both in in vivo (7 studies) and in vitro models (3 studies). The number of analyzed samples in groups ranged from 3 (including poolings) to 8 with mean n value = 5.1.

3.2. Glucocorticoid Responsive Proteins

The data retrieved from 10 analyzed studies contained jointly 1310 observations of significantly altered expression of proteins with various types of identifiers provided in original papers. Updating protein nomenclature was successful in most cases resulting in 1303 proteomic observations with a complete set of protein and gene identifiers retrieved from UniProt, NCBI, and Ensembl databases (Supporting Information 1, https://data.mendeley.com/datasets/tw95jjrpxr/3). Grouping the dataset by gene symbols revealed 1043 proteins with assigned unique NCBI gene symbols (Figure 2, Supporting Information 2, https://data.mendeley.com/datasets/tw95jjrpxr/3) and 1041 proteins with unique Ensembl gene symbols (Supporting Information 3, https://data.mendeley.com/datasets/tw95jjrpxr/3). The majority of symbols retrieved from the NCBI and Ensembl were the same (1031) with few exceptions including some Ensembl symbols that were updated since the time of the analysis (Ensembl IDs ENSRNOG00000016470, ENSRNOG00000018630, ENSRNOG00000036701, ENSRNOG00000015438, ENSRNOG00000010306, and ENSRNOG00000007048) or were associated with Ensembl stable IDs withdrawn from the most recent Ensembl release 115 (ENSRNOG00000047098, ENSRNOG00000005389, and ENSRNOG00000064025). Therefore, I focused on the data grouped by NCBI symbols while the Ensembl‐grouped data (Supporting Information 3, https://data.mendeley.com/datasets/tw95jjrpxr/3) are mentioned to make the readers aware of minor differences between databases and their editions that can affect specific genes.

FIGURE 2.

Summary of data processing and the final dataset grouped by NCBI gene symbols. The complete dataset is available in Supporting Information 2 (https://data.mendeley.com/datasets/tw95jjrpxr/3).

The analysis of the data showed that the majority of proteins with assigned unique genes (90.89%) were reported by single papers while only 0.96% were significant in at least 3 studies (Figure 2, Supporting Information 2, https://data.mendeley.com/datasets/tw95jjrpxr/3). Furthermore, the minority of proteins displayed a consistent direction of altered expression in independent studies. For example, in the group of 9 proteins reported in three independent studies, there were only two proteins displaying the same direction of expression in all experiments while 34 proteins displayed a consistent direction of responses in the case of proteins reported in two studies (Figure 2, Supporting Information 2, https://data.mendeley.com/datasets/tw95jjrpxr/3). The expression score parameter (Supporting Information 2, https://data.mendeley.com/datasets/tw95jjrpxr/3) representing jointly the number of reporting papers and consistency of responses showed that the two most frequently and consistently reported proteins were ATP synthase F1 subunit beta (Atp5f1b) and aldolase, fructose‐bisphosphate C (Aldoc).

The largest overlap in glucocorticoid‐responsive proteins was found between studies testing expression in mouse glial cells [36] and rat whole brains [41] (21 common proteins) and between data obtained in mouse glial [36] and neuronal stem cells [38] (16 common proteins). However, there was also a high correlation (r = 0.76, p = 0.0004) between the total number of differentially expressed proteins in individual datasets and their contribution to the pool of proteins common to at least two independent studies (Figure 3). This means that studies reporting a larger number of differentially expressed proteins are more likely to include overlapping findings. Therefore, it is difficult to draw firm conclusions from similarities among a small number of datasets, which may be affected by various factors, including methodological issues and random processes that are especially important in experiments with a small number of samples.

FIGURE 3.

Relationship between the number of common findings and the total number of differentially expressed proteins in individual proteomic comparisons performed for specific genotypes, brain areas, or cell types. Common findings are defined as differentially expressed proteins reported by at least two independent papers. wt, wild type.

The estimation of probability using a binomial test showed that all proteins detected in at least three different papers were reported more frequently than expected by chance in repeated experiments, while proteins reported in two studies reached significance only when both papers reported the same direction of altered expression (Table 5). It should be noted, however, that an assumption of the binomial test is that the probability of the outcome is the same in each trial [49]. The analysis described in Section 3.4 showed that proteomic data obtained from different experimental models are not comparable due to the narrow range of detected proteins and a bias toward proteins with the highest expression in the analyzed samples. As a result, the data are likely to contain a considerable number of false‐negative findings, limiting the replicability of results obtained from heterogeneous biological material (different species, brain areas, isolated cell types, and in vitro models). Therefore, the results of the binomial test should be treated cautiously.

TABLE 5.

Probabilities assessed using a binomial test.

Protein detection probability			Up‐regulation probability			Joined probability
x1	n1	p	x2	n2	p	p
1	9	0.37	0	1	1	0.37
			1	1	1	0.37
2	9	0.071	0	2	0.5	0.036
			1	2	1	0.071
			2	2	0.5	0.036
3	9	0.0084	0	3	0.25	0.0021
			1	3	1	0.0084
			2	3	1	0.0084
			3	3	0.25	0.0021
4	9	0.0006	0	4	0.125	0.00008
			1	4	0.625	0.0004
			2	4	1	0.0006
			3	4	0.625	0.0004
			4	4	0.125	0.00008

Note: x1—the number of independent studies that reported altered expression of a specific protein; n1—the total number of independent studies; x2—the number of independent studies that reported significant up‐regulation of a specific protein; n2—the total number of independent studies that reported a significant change in the expression of a specific protein.

3.3. Comparison Between Proteomic and Transcriptomic Responses to Glucocorticoids

The level of proteins depends on different mechanisms affecting gene transcription, translation and rate of protein degradation. Therefore, the proteomic dataset was compared with transcriptomic data including a set of most frequently and consistently reported genes in response to glucocorticoids [53], a large hippocampal dataset from an experiment testing the effect of overnight corticosterone treatment followed by three different resting periods [54] and hippocampal data from an experiment testing prolonged treatments ranging from 5 to 28 days [12]. The significant results from Jaszczyk et al. [54] were previously integrated with data from Juszczak et al. [12] and included in Supplementary data 2 published with previous study; [12] that were used as a reference transcriptomic dataset for comparison with proteomic data. These three transcriptomic datasets [12, 53, 54] comprise jointly 8 glucocorticoid groups. The reference lists of glucocorticoid‐responsive genes derived from different sources [12, 53, 54] are provided in Supporting Information 4 (https://data.mendeley.com/datasets/tw95jjrpxr/3) published with this study. The comparison between proteomic and transcriptomic data with gene symbols updated according to the Ensembl release 113 revealed that 455 (43.6%) differentially expressed proteins were not matched with any significant changes in the expression of mRNAs (Figure 4A, Supporting Information 5, https://data.mendeley.com/datasets/tw95jjrpxr/3) including 16 proteins reported by two different studies with a consistent direction of response to glucocorticoids (ACADM, AP2A2, CAPN5, EPHX1, FMNL2, HSPA9, MRPL43, NDUFA7, NDUFS2, PDCD6, PSMA1, RAB3A, RPS14, THOP1, C3 and NEFL). In turn, 56.4% of proteins were matched with significant changes in the mRNA level detected in at least one transcriptomic group (Figure 4A, Supporting Information 5, https://data.mendeley.com/datasets/tw95jjrpxr/3). The small group of most frequently reported transcriptomic changes together with corresponding proteomic responses to glucocorticoids are shown in Figure 4B. The figure also shows that in many cases (Aldoc, Plin4, Aqp4, Endod1, Glul, Anln, Aldh1l1, Parp1, Trf, Fermt2, Tmem63a and Trim2) the direction of altered expression of mRNAs and proteins is highly consistent. On the other hand, only 5 core glucocorticoid‐responsive genes (Aldoc, Ppp5c, Mgst1, Mt2, Gap43) identified in transcriptomic data [53] displayed altered expression at the level of proteins. This means that only 5.7% of the most frequently and consistently reported genes in various transcriptomic experiments [53] displayed significant changes at the proteomic level. Similarly, only 17 out of 249 most frequently and consistently regulated genes after prolonged treatment with corticosterone (transcripts significant in at least 2 out of 3 treatment periods) [12] displayed significant differences in proteomic data (AIF1L, EZR, PLIN4, ALDOC, AQP4, ENDOD1, GLUL, PARP1, PLP1, AK3, ANLN, MBP, UNC5B, ALDH1L1, CAR2, PDE8A, TTYH2). This indicates that the most consistently detected transcriptomic responses to glucocorticoids are poorly represented in proteomic data.

FIGURE 4.

An overlap between proteomic and transcriptomic findings in experiments testing brain responses to glucocorticoids. The transcriptomic reference data were derived from Juszczak and Stankiewicz [53] (core/extended dataset), Jaszczyk et al. [54] (12 h treatments), and Juszczak et al. [12] (5–28 days of treatment). (A) shows that 455 proteins responsive to glucocorticoids were not confirmed at the transcriptomic level while 4 proteins were matched with significant differences in mRNA in all 8 datasets derived from transcriptomic experiments [12, 53, 54]. (B) shows glucocorticoid‐responsive proteins matched with transcriptomic responses that were reported in at least 5 transcriptomic datasets. GCs, glucocorticoids; ref., references.

3.4. Limitations of Proteomic Data

A small overlap between the most consistently detected transcriptomic responses and proteomic data can result either from insufficient translation of some RNAs due to regulatory mechanisms or from failure to detect and quantify some proteins. Unfortunately, the problem is difficult to resolve because most of the studies list only significant findings. Some insight can be gained, however, from the difference between the number of expected and detected proteins in individual studies. The total number of proteins in each analyzed species is determined by the number of protein‐coding genes. The data available in Ensembl (release 113) indicate that there are more than 20 000 protein‐coding genes in mouse, rat and human genomes (Figure 5A) as indicated by the number of official gene symbols and the number of novel genes with assigned stable Ensembl IDs. Not all proteins are, however, expressed in each tissue. Therefore, the data from recent mouse proteome atlas [55] were used to make a more accurate estimate of an expected number of proteins. The data retrieved from this resource (Supplementary data 2 published with previous study, from Giansanti et al. [55]) show that brain samples express about 10 800 different proteins (Figure 5B) that can be detected and quantified with the most advanced methods. In turn, the proteomic experiments testing brain responses to glucocorticoids identified about 2800 proteins or distinguishable protein spots (Figure 5C). This means that available proteomic data concerning brain responses to glucocorticoids (Figure 5C) failed to detect about 74% of proteins. Therefore, there is a high probability that many glucocorticoid‐responsive proteins were not reported due to failure to detect their expression. The final confirmation for this assumption was provided by two recent studies [36, 43] that reported all quantified proteins in glucocorticoid experiments. The analysis of these datasets [36, 43] revealed that each study detected only 15 proteins coded by core glucocorticoid‐responsive genes [53] including both significant and insignificant findings (Figure 6). This means that both studies failed to detect and quantify 83% of proteins coded by genes known to be the most frequently and consistently reported in transcriptomic studies [53] including Sgk1, Ddit4, Nfkbia, Tsc22d3 and Nr3c1 (glucocorticoid receptor).

FIGURE 5.

Number of detected proteins in comparison with genomic and transcriptomic data. (A) number of protein‐coding genes derived from Ensembl. The values were calculated as a sum of unique gene symbols and the total number of novel genes with assigned stable Ensembl IDs. (B) Number of proteins detected and quantified in selected mouse tissues [55]. (C) number of identified proteins or distinguishable protein spots [37, 44] in studies testing the effect of glucocorticoids on brain proteome. The values were derived from papers or supplementary data. In case of Notaras et al. [43] the value was calculated as a number of proteins with assigned unique gene symbols plus the number of UniProt IDs without gene symbols.

FIGURE 6.

(A) Number of most frequently and consistently detected genes displaying significant differences in brain or in vitro brain cultures after treatment with glucocorticoids [53]. The core glucocorticoid‐responsive genes are defined as genes that displayed the same direction of change in at least four studies [53]. (B) Number of detected proteins [36, 43] that are coded by the core glucocorticoid‐responsive genes [53]. GC, glucocorticoid.

An additional insight into the properties of proteins detected in glucocorticoid experiments was gained from the match between proteins and transcriptomic dataset providing an expression level for all quantified genes in mouse hippocampus (Supplementary data 3 published with previous study; [12]). The transcriptomic data were used to calculate mean mRNA expression level for each gene and the results were matched with proteomic data based on Ensembl IDs or Ensembl gene symbols. The proteomic data from Supporting Information 1 (https://data.mendeley.com/datasets/tw95jjrpxr/3), were filtered for proteins detected in mouse hippocampus to ensure comparability with transcriptomic datasets. The selected hippocampal data (Supporting Information 6, https://data.mendeley.com/datasets/tw95jjrpxr/3) reported by Skynner et al. [44], Guest [39] and Kuznetsova et al. [40] contained 54 unique proteins that were matched with the level of hippocampal mRNAs derived from Juszczak et al. [12]. The mRNAs assigned to these proteins have an average expression of 385 counts per million of reads (CPM) and in most cases (61.1%) were classified to the group of transcripts with at least high expression in mouse hippocampus (Figure 7A,B). In contrast, mean expression of most consistent transcriptomic responses reported by Juszczak et al. [12] was 48.4 CPM with a large contribution of genes with a low or lower medium level of expression constituting jointly 47.8% of transcripts while genes with at least high expression constituted only 10.4% (Figure 7C,D). This indicates that proteomic methods are biased for proteins with higher abundance of mRNAs that are less likely to be regulated at the transcriptomic level.

FIGURE 7.

Mouse hippocampal expression of mRNA based on data from Juszczak et al. [12]. (A and B) show an expression level of mRNAs assigned to proteins responsive to glucocorticoids in mouse hippocampus. (C and D) show an expression level of 249 genes displaying the most consistent changes at the level of mRNA after prolonged treatment with corticosterone [12]. Consistency was defined as a significant change in expression observed after at least two out of three treatment durations. All data are available in Supporting Information 2 at https://data.mendeley.com/datasets/w3df8dhfwb/3 [12]. (D) is a modified (A) published previously [12] under an open access Creative Common CC BY license allowing for reuse provided that it is clearly cited (www.mdpi.com/openaccess). Data in panels A and C are presented as mean ± SEM (column bar graphs) overlaid on scatter plots. Individual data points represent expression levels of individual genes. The classification of mRNA expression ranging from residual to top is derived from Juszczak et al. [12].

A narrow range of detected molecules and a detection bias toward proteins with higher mRNA abundance, together with known differences in basal gene expression across brain regions [56] and cell types [57], raise questions about the comparability of data obtained from different experimental models. In other words, proteomic methods may detect different sets of proteins in samples that differ in the composition of highly expressed proteins, which may in turn lead to disparities between experimental models in the sets of differentially expressed proteins. Verification of this hypothesis, however, requires information on all detected proteins, which is rarely reported. Among the analyzed publications, only two [36, 43] provided identifiers for all quantified proteins. Comparison of these data after nomenclature standardization revealed a limited overlap between individual datasets, confirming the assumption of restricted comparability between data obtained from different models, such as organoids and isolated cell types datasets (Figure 8).

FIGURE 8.

Venn diagram showing the overlap between complete proteomic datasets (including both significant and non‐significant results) derived from Notaras et al. [43] and Chitu et al. [36]. The numbers indicate the number of genes matched to the detected proteins.

4. Discussion

The standardization of protein nomenclature revealed that the majority of differentially expressed proteins were reported only by single papers while only 10 proteins (0.96%) were reported in at least 3 independent studies. Furthermore, not all proteins displayed a consistent direction of responses in individual studies. Such a low overlap between studies is not unique for this dataset because a similar pattern was also observed in case of transcriptomic data from glucocorticoid and stress experiments [48, 53]. A restricted similarity of the results is caused by many factors including tested models (different species [58, 59], brain areas [60], in vivo or in vitro cultures [61], treatment durations [48, 62, 63] etc.), applied analytical and statistical methods [64], selective publishing of data based on various cut‐off criteria [65], scientific progress in identification of genes and their products [66], low number of analyzed samples leading to small statistical power [67] and random processes occurring during sample preparation and analysis that lead to signal noise in the acquired data [48, 65]. These factors are also present in analyzed proteomic data that were published during two decades of research performed in different models (Table 2) usually with a small number of samples (mean n = 5.1) typical for omics methods. Furthermore, the problem of proteomic data replicability is exacerbated by methodological limitations, which lead to the detection of largely non‐overlapping sets of proteins across different experimental models (Figure 8).

Despite a small overlap between individual datasets, the collected data allowed for identification of proteins displaying consistent changes in expression after treatment with glucocorticoids. The two most frequently and consistently detected proteins were ATP synthase F1 subunit beta (Atp5f1b) [37, 39, 41] and aldolase, fructose‐bisphosphate C (Aldoc) [37, 41, 44]. Aldoc regulates glycolysis of fructose and decreases competition between metabolism of glucose and fructose [68]. Although it is consistently upregulated in proteomic and transcriptomic datasets (Figure 3) in various models of brain responses to glucocorticoids [12, 53, 54], it is still understudied as indicated by the fact that the PubMed database lists only one paper [53] relevant for the words “Aldoc glucocorticoids”. The second most frequently and consistently detected protein, Atp5f1b, is a subunit with a catalytic site for ATP synthesis [69]. Although consistent changes in expression of Atp5f1b mRNA were detected after overnight treatment with corticosterone (Figure 3) in mouse hippocampus [54], this gene and its protein product were overlooked in glucocorticoid research as indicated by the fact that the PubMed database lists no study relevant for the words “Atp5f1b glucocorticoids”.

Although there is a group of overlapping findings in proteomic and transcriptomic experiments (Figure 3), there were also proteins that displayed consistent regulation in proteomic analyses but without a match in transcriptomic datasets. On the one hand, such discrepancies can be explained by methodological issues such as timing of experiments that is not always suitable for detecting changes in mRNAs. On the other hand, some changes in the expression of proteins can result from post‐transcriptional regulation of protein synthesis [70, 71, 72]. It should be noted that synthesis of mRNA is a slow process [70] that effectively increases the number of mRNA copies only in case of genes with a small number of transcripts. Therefore, proteins with abundant copies of mRNA are likely to be regulated mainly during translation of genomic information [12]. Finally, some detected differences in expression can result from changed abundance of proteins in soluble and insoluble fractions of proteins. Such a possibility is indicated by the experiment performed by Skynner et al. [44] who detected opposed changes in expression of GAPD (GAPD) in hippocampus and FSCN1 in hypothalamus depending on the analyzed fractions. Such changes can result from an interaction with different molecules and movement of proteins between different cellular compartments. Therefore, some changes in the levels of proteins detected in experiments testing only one fraction can be wrongly interpreted as changes in expression. The frequency of such cases in the entire analyzed dataset is unknown because the majority of studies tested only a single fraction of the protein extract.

Finally, an important issue is the limitation of available proteomic data. In fact, the studies included in our analysis failed to detect about 74% of proteins expected to be present in brain samples including many proteins coded by the most well‐known genes responsive to glucocorticoids. Furthermore, our analysis showed that proteomic studies are biased toward proteins with higher mRNA abundance while significant changes in the levels of transcripts are most common in case of genes with a smaller number of mRNA copies. This bias restricts the comparability between proteomic and transcriptomic data and between proteomic data obtained in different models. It is because proteins that are below the detection threshold in extracts obtained from whole tissues can be abundant enough in extracts obtained from the selected cell lines. The heterogeneity of the data together with the small number of papers available for analysis constitutes a limitation of the present study. The limitations of the available proteomic data also mean that there is a need for proteomic studies using more sensitive methods to fill the existing gaps in knowledge.

Author Contributions

Grzegorz R. Juszczak: conceptualization, methodology, data analysis investigation, writing manuscript.

Funding

This work was supported by Narodowe Centrum Nauki (NCN), 2017/27/B/NZ2/02796; IGAB PAN Intramural grant, STAT/GRZJUS/2025/01.

Conflicts of Interest

The author declares no conflicts of interest.

Data Availability Statement

Data are available at https://data.mendeley.com/datasets/tw95jjrpxr/3 while R scripts were deposited at https://github.com/Grzegorz‐R‐Juszczak/Scripts‐for‐Brain‐proteomic‐responses‐to‐GCs.