Sampling

Specimens of Leucosolenia corallorrhiza Haeckel, 1870 (Leucosolenida, Calcarea, Porifera) [25] were collected in Kandalaksha Bay, the White Sea, at the outskirts of the Pertsov White Sea Biological Station of Lomonosov Moscow State University (66°34′ N, 33°8′ E) from June to August 2023. Animals were collected in the upper subtidal zone (0–2 m) during low tides and kept in a laboratory aquarium with natural seawater for no longer than five days.

For the regeneration assay in L. corallorrhiza, we distinguished three time points: intact tissues (IT); body fragments with a growing regenerative membrane (CR as in ‘crescent’), 24 hours post-operation (hpo); body fragments with a fully sealed regenerative membrane (RM), 48 hpo. IT samples (N = 4) were retrieved by processing whole sponges and clearing debris with minimal damage. As a regeneration model, we used ring-shaped body fragments from oscular tubes. They were cut with Castroviejo scissors, collected, and cleared of debris. To obtain ring-shaped body fragments, oscular tubes were transversely cut into rings 0.5–1.5 mm in width (osculum rims were excluded). All operational procedures were performed manually under a stereomicroscope in 0.22 µm-filtered seawater (FSW). The fragments were sustained in 30 mm plastic Petri dishes with 5 ml of FSW, half of which was changed every 12 h. Body fragments were maintained at physiological temperatures for the species during the summer period (8–12°C). Regular observations under a stereomicroscope were performed to exclude contaminated samples and evaluate the stage of regeneration. Each replicate contained more than 100 ring-shaped fragments obtained from oscular tubes of 1–3 sponges (for CR samples, N = 6, and for RM samples, N = 4). The differences in sample size were due to technical limitations encountered during sample processing.

Specimens of marine sponge Halisarca dujardinii Johnston, 1842 (intact tissues, N = 6) were also collected from the upper subtidal zone of Kandalaksha Bay in February 2025. Animals were kept in a laboratory aquarium with natural seawater for no longer than two days.

Gemmules of the freshwater Ephydatia fluviatilis Linnaeus (1759) were collected manually from an unnamed small river flowing from the Vaulino Reservoir (v. Vaulino, Sergievo-Posadsky District, Moscow region) in November, 2023. The gemmules were immediately placed in cold water and stored at 2°C until further use. To start germination, gemmules were placed in sterile Mineral-medium, M-medium (1 mM CaCl₂x6H₂O, 0.5 mM MgSO₄x7H₂O, 0.5 mM NaHCO₃, 0.05 mM KCl, and 0.25 mM Na₂SiO₃) [26] at room temperature (RT, 22°C). Each IT sample (N = 5) consisted of 40–50 young (7 days after hatching) fully developed sponges.

The culture of the marine carnivorous sponge Lycopodina hypogea Vacelet & Boury-Esnault (1996) is kept at 2.5–3 L containers with artificial seawater (14°C) at the Marine Station d’Endoume (Marseille). Under such conditions (weekly feeding with Artemia salina nauplii and water change), the population grows, thrives, and is fully capable of maintaining itself. Adults unfed for at least three weeks were processed as IT samples (N = 5).

All specimens used in the study were typical representatives of the corresponding species in sense of overall body structure, physiological condition (only specimens considered fully healthy were used in the study), and habitat. Culturing conditions for L. hypogea and gemmules of E. fluvatilis are well accepted and routinely used [27–29].

RNA extraction and reverse transcription

Prior to RNA extraction, all samples were dabbed at absorbent paper for a few seconds to dispose of excessive water and lysed in 1 mL of ExtractRNA reagent (BC032, Evrogen, Moscow, Russia), homogenized with polypropylene pestle, incubated for 20 min at RT and stored at −80°C at most for two months.

RNA was extracted using the acid guanidinium thiocyanate-phenol-chloroform method [30]. The samples were then incubated with DNAse I (EN052, Thermo Fisher Scientific, Waltham, MA, USA) and the RNAse inhibitor RiboCare (EK005S, Evrogen, Moscow, Russia) for 10 min at 37°C. RNA was purified and concentrated using a ‘Clean RNA Standard’ spin-columns kit (BC033, Evrogen, Moscow, Russia) according to the manufacturer’s protocol. The RNA concentration was measured spectrophotometrically using a NanoPhotometer (Implen, München, Germany). RNA purity was verified by A260/A280 ratio (median 2.101, range 1.75–2.667) and A260/A230 ratio (median 2.017, range 0.44–3.39). Normalization of samples within each species was performed at this stage by using equal amounts of total RNA (0.1–1 µg) for reverse transcription.

cDNA was synthesized using Random Primers from RevertAid First Strand cDNA Synthesis Kit (K1621; Thermo Fisher Scientific, Waltham, MA, USA). The required volume of RNA, 1 µL of the manufacturers’ Random Hexamer Primers, and nuclease-free water (PB207S, Evrogen, Moscow, Russia) up to 12 µL were mixed. Next, 4 µL of 5x Reaction Buffer, 1 µL of RiboLock RNAse Inhibitor (20 units/µL), 2 µL of 10 mM dNTP mix, and 1 µL RevertAid M-MuLV reverse transcriptase (200 units/µL) were added per reaction. The final reaction volume was 20 µL. The tubes were gently mixed and briefly centrifuged. The reaction was incubated for 5 min at 25°C, followed by 60 min at 42°C, and terminated by heating at 72°C for 5 min. Nuclease-free water (80 µL) was added to each reaction mixture, resulting in a final volume of 100 µL. cDNA was stored at −20°C.

Identification of target sequences and primers design

Prospective target sequences were retrieved using tblastn search [31] against the reference databases: a) de novo transcriptome of L. corallorrhiza; b) genome-derived transcriptome of H. dujardinii; с) de novo transcriptome of E. fluviatilis; d) genome-derived transcriptome of L. hypogea (for details on the assemblies see Supporting Information and Table S1 in S1 File). The query amino acid sequences of Mus musculus were obtained using the UniProt database (accessed October-November 2024). The retrieved Porifera transcripts were translated using Expasy Translate tool [32] and studied using web-based InterProScan v5.69-101.0 (accessed October-November 2024) to assess the conservation of the encoded protein domain structure. The identification was also validated using reciprocal pblast against the UniProtKB + SwissProt databases.

For lamins, nuclear intermediate filaments, providing structural support to the nucleus and contributing to chromatin organization, we used a phylogeny-based approach. Amino acid sequences of intermediate filaments from animals other than sponges were retrieved from the UniProt database using manual or blast searches. The resulting set of amino acid sequences was aligned using MUSCLE [33] with default settings; the alignment was trimmed using TrimAl v1.4.1 [34] and used to build maximum-likelihood phylogenetic tree using IQTree web server [35]. The ML trees were built using ultrafast bootstrap (N = 1000) and SH-aLRT branch support, with automatic assessment of the optimal substitution model. The resulting trees were visualized using FigTree v1.4.4. Nucleic acid sequences of target genes are listed in the ‘Nucleotide sequences from Table S2.fasta’ file in S2 File(see Supporting information section).

Primer sequences were designed using the Primer–BLAST Tool (based on Primer3, version 2.5.0) with standard settings for target sequences. All primers were synthesized by Evrogen JCS (Moscow, Russia). The quality of synthesis was manufacturer-controlled by analytical electrophoresis using PAAG and mass spectral analysis. The cDNA samples used in this study were also used in x1, x10, x100, x1000, and x10000 dilutions in duplicates to create a linear regression plot for each gene (the final efficiency percentage is the mean from three independent calculations). Primer details (sequences, amplicon sizes, annealing temperatures, PCR efficiencies) are listed in Table S3 in S1 File.

Real-time quantitative polymerase chain reaction

Real-time qPCR was performed in 96-well CFX96 Touch Real-Time PCR Detection System (Bio-Rad, Hercules, CA, USA) or in 96-well DTprime 4М1 PCR Detection System (DNA-Technology, Moscow, Russia) with SYBR Green I dye and reference dye ROX. All reagents were provided by Evrogen (Moscow, Russia) (5X qPCRmix-HS SYBR+High/LowROX, PK155L/PK156L) and the manufacturer’s protocols were followed. The reaction volume was 20 µL and contained 4 µL 5x qPCRmix-HS-SYBR+High/LowROX buffer (includes 3 mM Mg²⁺ ions and 0.12 mM of each dNTP per each reaction), 25 pM of each primer, and 2 µL of x5 cDNA dilution. The reaction protocol included denaturation at 95°C followed by 40 amplification cycles — 95°C for 15 s, T_a°C (individual for each pair of primers) for 30 s, 72°C for 30 s, — and further melting curve analysis (55–95°C, step 0.5°C). All samples were processed in triplicates. The same cDNA samples were placed repeatedly into each PCR run and served as the inter-run calibrator.

Data analysis and statistics

Relative cDNA quantity (RQ, i.e., normalized gene expression) was calculated as RQ = CPE^ΔCT, where

• CPE = 10^{(−1/Eslope)} – 1; converted PCR efficiency;
• ΔCT = Ct_n– Ct_i; averaged triplicate Ct_n value for each sample, Ct_i – averaged triplicate Ct value for one chosen sample.

The obtained datasets were tested for normality using the Shapiro-Wilk test. Subsequent comparisons were performed using the appropriate statistical tests. The test used is indicated in each case individually. All statistical procedures and data visualization were performed using RStudio version 4.0.3 (RStudio, USA). The significance level (α) was set at < 0.05. The following R packages were used: ‘tidyverse’, ‘ggpubr’, ‘NormqPCR’, ‘pastecs’. The raw runs’ data and R script is available in the Mendeley Data repository (see “Data availability” section).

Identification of target sequences

The identification of GAPDH, TBP, RPS3A, RPL13A, HPRT1, and RHOA did not need to be supplemented by phylogenetic trees since a single copy for each gene/transcript was found in the transcriptomes. The results of reciprocal BLAST searches and the predicted domain structures of proteins encoded by the studied genes/transcripts (Fig 1a–1h and Table S2 in S1 File) unambiguously validated the identification.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Domain structures of proteins encoded by the identified transcripts.

GAPDH (a), RPS3A (b), RPL13A (c), HPRT1 (d), RHOA (e), LMN1 (f), ACT1 (g), TBP (h). Hdu – Halisarca dujardinii, Lhy – Lycopodina hypogea, Efl – Ephydatia fluviatilis, Lco – Leucosolenia corallorrhiza.

https://doi.org/10.1371/journal.pone.0349507.g001

For lamins, several prospective sequences were identified. All of them clearly fell into the clade containing lamins, rather than cytoplasmic intermediate filaments (Fig 2). Poriferan lamins tended to cluster together with sequences from cnidarians, and based on the topology of the tree, it seems that calcarean and demosponge lamins are highly divergent (Fig 2). We found a single lamin-coding sequence in the L. hypogea and E. fluviatilis transcriptomes. For H. dujardinii and L. corallorrhiza, we chose transcripts that encoded proteins with less predicted disorder (namely, Hdu_evm.model.contig32.21 and Lco_TRINITY_DN9368_99323) as targets for RT-qPCR.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Phylogenetic relationships among intermediate filament (IF) proteins.

A maximum likelihood tree was constructed using IQTree with ultrafast bootstrap (N = 1000) and SH–aLRT branch support. A full-length alignment trimmed with TrimAl was used to construct the tree. The substitution model LG + G4 was chosen automatically. Node labels represent ultrafast bootstrap/SH–aLRT support values. Names of sponge sequences used for PCR analysis in the study are highlighted in magenta. Abr – Astropecten brasiliensis, Aca – Aplysia californica, Apl – Acanthaster plancii, Aqu – Amphimedon queenslandica, Ate – Actinia tenebrosa, Alu – Ascaris lumbricoides, Aam – Amphibalanus amphitrite, Bgl – Biomphalaria glabrata, Bne – Bugula neritina, Cas – Cornu aspersum, Cel – Caenorhabditis elegans, Cbr – Caenorhabditis briggsae, Che – Clytia hemisphaerica, Cin – Ciona intestinalis, Csa – Ciona savignyi, Cte – Capitella teleta, Cvi – Crassostrea virginica, Ddi – Dictyostelium discoideum, Dme – Drosophila melanogaster, Dpu – Daphnia pulex, Dim – Dirofilaria immitis, Dja – Dugesia japonica, Emu – Ephydatia muelleri, Efl – Ephydatia fluviatilis, Ero – Euperipatoides rowelli, Esp – Eoperipatus sp. LH–2012, Gpy – Glottidia pyramidata, Ham – Haemopis marmorata, Hex – Hypsibius dujardini, Hdu – Halisarca dujardinii, Hgl – Heterodera glycines, Him – Hirudo medicinalis, Hpo – Helix pomatia, Hvu – Hydra vulgaris, Isc – Ixodes scrapularis, Lco – Leucosolenia corallorrhiza, Lgi – Lottia gigantea, Lun – Lingula unguis, Lte – Lumbricus terrestris, Lhy – Lycopodina hypogea, Lsa – Lineus sanguineus, Lvi – Lineus viridis, Min – Myxicola infundibulum, Mye – Mizuhopecten yessoensis, Nve – Nematostella vectensis, Nsl – Nototodarus sloanii, Ohi – Ooperipatus hispidus, Osp – Oscarella sp., Ovo – Onchocerca volvulus, Pel – Parasagitta elegans, Phi – Principapillatus hitoyensis, Pca – Priapulus caudatus, Pij – Phoronis ijimai, Pst – Phascolion strombus, Pvi – Polysphondylium violaceum, Sci – Sycon ciliatum, Spi – Stylopora pistillata, Sph – Sepia pharaonis, Sko – Saccoglossus kowalewski, Uca – Urechis caupo.

https://doi.org/10.1371/journal.pone.0349507.g002

We identified several actin-coding sequences in each database. Actin-like proteins were filtered using reciprocal blast. However, the precise phylogenetic relationships between the remaining actins remained unclear, as the attempt to construct ML trees resulted in low node support values. Thus, we aligned candidate transcripts to β- and γ -cytoplasmic isoforms of vertebrate actins; contigs Lco_TRINITY_DN24_95078 (L. corallorrhiza), Hdu_evm.model.28.48 (H. dujardinii), Lhy_evm.model.OZ017783.1.600 (L. hypogea), and Efl_m.16996_g.16996 (E. fluviatilis) encoded proteins with the highest similarity to vertebrate ACTB and were chosen for the subsequent analysis with ACT1 name (Fig 3a–3c). Notably, the predicted E. fluviatilis sequence was fragmented and lacked the proper N-end of the amino acid chain. To validate that the identified contigs encode cytoplasmic actins in sponges, L. corallorrhiza regenerative membranes were immunostained with antibodies targeting various actins: monoclonal mouse anti-actin-β (BioRad MCA5775GA, clone 4C2), monoclonal mouse anti-actin-γ antibodies (BioRad MCA 5776GA, clone 2A3), and polyclonal rabbit anti-actin-β antibodies (ab8227, Abcam). Subsequent imaging revealed that cytoplasmic filamentous structures and stress fibers were stained with a certain level of differentiation between the different actin isoforms (Fig S1 in S1 File). For detailed immunostaining protocol see [36].

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Alignment of amino acid sequences encoded by the identified transcripts.

(a, b) – N-end of the alignment with human β-cytoplasmic actin and γ-cytoplasmic actin set as reference sequences; (c) – the whole alignment with human β-cytoplasmic actin set as the reference. Colored residues outside the reference sequence represent residues that deviate from the reference sequence. Asterisks indicate amino acid positions essential for distinguishing between β- and γ-actin in vertebrates.

https://doi.org/10.1371/journal.pone.0349507.g003

Expression variability and stability assessment of candidate reference genes in Leucosolenia corallorrhiza

We conducted a comprehensive assessment and evaluation of candidate reference genes in L. corallorrhiza using four established algorithms: geNorm, NormFinder, BestKeeper, and RefFinder. Based on the analysis of existing data on the assessment of gene expression in invertebrates using real-time PCR, we selected seven candidate reference genes, ACT1, RPL13A, GAPDH, HPRT1, RPS3A, TBP, and LMN1, which are commonly used in normalization studies because of their involvement in essential cellular processes, such as cytoskeletal organization (ACT1), energy metabolism (GAPDH), ribosomal structure (RPL13A, RPS3A), transcription regulation (TBP), nucleotide metabolism (HPRT1), and nuclear integrity (LMN1).

To assess the variability in the expression of the selected genes, we first analyzed their cycle threshold (Ct) values (N = 14). Expression analysis of the seven candidate reference genes across all samples from regeneration experiment on L. corallorrhiza revealed significant differences in both expression levels and variability (Fig 4a and Table 1). The coefficients of variation (CV) for all genes remained below 0.04, suggesting relatively stable expression in biological replicates. The lowest Ct values, corresponding to the highest expression, were observed for ACT1, whereas TBP exhibited the highest Ct values, indicating a lower transcript abundance. RPL13A and RPS3A showed intermediate and relatively stable expression, as reflected by their narrow interquartile ranges and limited number of outlier points. In contrast, HPRT1 and TBP demonstrated pronounced variability, with more outliers (in case of TBP), suggesting less consistent expression across the samples. Next, we plotted histograms with fitted curves to show the distribution of the Ct values for each gene (Fig 4b). ACT1, RPL13A, and GAPDH displayed approximately normal (unimodal and symmetric) Ct value distributions, indicating uniform gene expression within the dataset. Meanwhile, HPRT1 and TBP showed more skewed or multimodal distributions, reflecting greater instability and potential condition-dependent expression fluctuations.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Cycle threshold (Ct) values and descriptive statistics for 7 candidate endogenous referent genes assayed across 14 samples from Leucosolenia corallorrhiza tissues.

https://doi.org/10.1371/journal.pone.0349507.t001

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Expression variability and stability assessment of candidate reference genes in Leucosolenia corallorrhiza.

(a) Boxplots showing the distribution of Ct values for the seven candidate reference genes across all samples, with lower Ct values indicating higher transcript abundance. Upper whisker extends from the top of the box (third quartile, Q3) to the most extreme data point within 1.5 times the Interquartile Range (IQR) above Q3. Lower whisker extends from the bottom of the box (first quartile, Q1) to the most extreme data point within 1.5 times the IQR below Q1. (b) Histograms with fitted curves (red) illustrating the frequency distribution of the Ct values for each gene. (c) Stability ranking of the reference genes calculated using four algorithms: geNorm (M values), NormFinder (stability values), and BestKeeper (standard deviation of C_t-values). Lower values correspond to higher stabilities. (d) Pairwise variation (V values) between combinations of reference genes across experimental groups (IT, intact tissue; CR, ‘crescent’/growing regenerative membrane; RM, fully sealed regenerative membrane; All, total dataset), as calculated using geNorm.

https://doi.org/10.1371/journal.pone.0349507.g004

To evaluate the stability of the selected reference genes, we performed an analysis using the geNorm, NormFinder and BestKeeper applications (Fig 4c). The geNorm algorithm evaluates the stability of the reference gene by calculating the M value, which represents the average pairwise variation of a gene relative to all the other tested genes [5]. Genes with lower M values are considered more stable. Our results showed that RPL13A, RPS3A and GAPDH were among the most stably expressed, whereas HPRT1 and LMN1 exhibited the highest variability. In addition to ranking genes, geNorm calculates pairwise variation (V value, the average variation between the normalization factors using n and n + 1 reference genes, indicating the effect of adding an extra gene on normalization accuracy) to determine the minimum number of genes required for reliable normalization. The algorithm assumes that the most stably expressed genes should have minimal variation relative to each other, with a recommended threshold of 0.15 [5]. The analysis of pairwise variation coefficients demonstrated consistently low V values across all experimental groups (Fig 4d and Table S4 in S1 File). The V2/3 values for all conditions, including intact tissues, growing regenerative membranes, and fully sealed regenerative membranes, were well below the 0.15 threshold, indicating that the use of only two reference genes was sufficient for reliable normalization. This confirms that the selected reference genes exhibit strong co-expression stability and that further inclusion of additional genes does not significantly improve the normalization accuracy.

The NormFinder algorithm ranks candidate reference genes based on their stability value, which is determined by estimating intragroup (within experimental conditions) and intergroup (between conditions) variations [22]. Unlike geNorm, which relies on pairwise comparisons, NormFinder calculates the systematic variance of each gene independently and considers the effects of experimental grouping. Genes with lower stability values were more reliable for normalization. According to NormFinder, ACT1 was identified as the most stable gene, followed by GAPDH and TBP, whereas RPL13A and LMN1 showed higher variability (Fig 4c and Table S5 in S1 File).

The BestKeeper algorithm assesses reference gene stability based on their standard deviation (SD) and coefficient of variance (CV). Genes with lower SD and CV were considered more stable [23]. Unlike geNorm and NormFinder, BestKeeper does not assume equal expression ratios between genes but instead evaluates absolute Cq values directly. In our analysis, GAPDH, RPL13A, and ACT1 had the lowest SD values, suggesting that they were the most stable genes according to BestKeeper, whereas HPRT1 and LMN1 had the highest SD values, indicating significant variability (Fig 4c). Given that RPL13A, ACT1, and GAPDH consistently appeared among the top three across three independent algorithms (ranked top-3 at least by two of the three algorithms) (see Table 2), and considering their normal distribution and low variability in expression, using these genes provides robust and reliable normalization across different experimental conditions in Leucosolenia corallorrhiza.

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Stability ranking of candidate reference genes in sponge species according to GeNorm, NormFinder, and BestKeeper analysis.

https://doi.org/10.1371/journal.pone.0349507.t002

Cross-species evaluation of reference gene stability

To assess the universality and robustness of the identified candidate reference genes, we extended our analysis to three additional sponge species representing diverse taxonomic and ecological backgrounds: Halisarca dujardinii (marine demosponge), Ephydatia fluviatilis (freshwater demosponge), and Lycopodina hypogea (carnivorous marine demosponge). This comparative approach enabled us to determine whether consistent reference genes could serve as reliable normalizers across different sponge taxa, thereby providing insights into the broader applicability of our findings for quantitative gene expression studies in Porifera.

In H. dujardinii, geNorm identified RPS3A and HPRT1 as the most stable genes, ACT1 and RPL13A also demonstrated high stability. NormFinder ranked RPL13A, RPS3A, and HPRT1 as the top candidates, whereas BestKeeper analysis confirmed the high stability of RPL13A, RPS3A, and TBP. Collectively, RPS3A, RPL13A, and HPRT1 were the most reliable reference genes for this species (Fig 5a and Table 2).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Stability ranking of candidate reference genes in the three sponge species based on three statistical algorithms.

The expression stability of the seven candidate reference genes was evaluated in (A) Halisarca dujardinii, (B) Ephydatia fluviatilis, and (C) Lycopodina hypogea using geNorm (first column), NormFinder (second column), and BestKeeper (third column). Each graph shows the stability values or ranking scores for the tested genes, with lower values indicating a higher expression stability.

https://doi.org/10.1371/journal.pone.0349507.g005

Both geNorm and NormFinder indicated that GAPDH, ACT1, and RPS3A were the most stably expressed genes in E. fluviatilis. The BestKeeper analysis was consistent, highlighting RPS3A, ACT1, and HPRT1 as the top performers. Thus, ACT1, RPS3A, and GAPDH were determined as the most suitable reference genes for this species (Fig 5b and Table 2).

For L. hypogea, geNorm suggested GAPDH and RPL13A as the most stable pair, followed by ACT1 and RPS3A. NormFinder ranked RPS3A, HPRT1, and GAPDH as the most stable genes, while BestKeeper analysis favored RPL13A, ACT1, and LMN1. Based on these results, GAPDH, RPL13A, RPS3A, and ACT1 were identified as the most stable genes in L. hypogea (Fig 5c and Table 2).

To compare our multi-algorithm approach with a commonly used consensus tool, we also analyzed candidate reference genes using RefFinder. The resulting rankings are shown in Fig S2 in S1 File. RefFinder integrates the outputs of four commonly used algorithms (geNorm, NormFinder, BestKeeper, and the ΔCt method) by calculating their geometric mean to generate a consensus stability ranking of candidate reference genes [24]. The RefFinder consensus rankings of candidate reference genes showed some species-specific variation and, in several cases, diverged from the stability patterns obtained with individual algorithms. In L. corallorrhiza, RefFinder placed RPL13A and ACT1 at the top, while RPS3A appeared least stable. However, in our multi-algorithm evaluation, both RPS3A and RPL13A were consistently among the most stable genes.

In H. dujardinii, RefFinder ranked LMN1 and RPL13A as the most stable genes. This contrasted all other outputs, as none of the algorithms ranked LMN1 among the most stable.

In E. fluviatilis, RefFinder identified TBP as the most stable genes, but surprisingly placed RPL13A and HPRT1 at the very bottom. This was biologically implausible, since TBP displayed relatively high variability across all other algorithms, and none of the independent methods supported such instability. These comparisons demonstrate that RefFinder may over- or under-estimate the stability of certain genes, producing rankings that contradict other algorithmic outputs.

Taken together, comparative analysis across all three species indicated that the most consistently stable reference genes included a ribosomal gene (RPL13A or RPS3A), ACT1, and GAPDH. This combination provides a reliable basis for normalization in gene expression studies involving diverse sponge species (Table 2).

Effect of single vs. multiple reference gene normalization on RHOA expression

Using the three selected reference genes individually for normalization resulted in notably different RHOA expression values, whereas the combined normalization factor (NF) yielded more consistent measurements. As shown in Fig 6a, single-gene normalization can over- or underestimate RHOA levels for the same sample. For example, the RHOA expression normalized to ACT1 was higher than that normalized to GAPDH or RPL13A. Such disparities indicate how much using a single reference gene can bias the results. In contrast, when using the combined NF, the RHOA expression of each sample fell between the extremes of the single-gene values, resulting in a more moderate and uniform estimate (see Table S6 in S1 File). This pattern was consistent across all experimental conditions in the regeneration assay for L. corallorrhiza.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 6. Impact of normalization strategy on RHOA expression estimated across sponge species.

(a) Normalized RHOA gene expression in 3 experimental groups of regeneration model (L. corallorrhiza); (b) scatter plots showing normalized RHOA mRNA expression for individual experiments in regeneration model of L. corallorrhiza; red dots represent mean value; whiskers represent standard deviation (SD); (c) Normalized RHOA mRNA expression for 3 other sponge species. Chosen NF is normalization factor, calculated as geometric mean of relative quantities of GAPDH, ACT1 and RPL13A. Normalization was performed according to Material & Methods section, IT1 sample was chosen as a control one.

https://doi.org/10.1371/journal.pone.0349507.g006

The scatter plot panels reinforce that normalization with multiple reference genes reduces variability. For L. corallorrhiza (Fig 6b), normalization to a single gene, even to a commonly used and relatively stable gene such as ACT1, results in greater variability and can lead to significant differences in gene expression between treatments. In L. hypogea, the expression values for the same treatment can differ by up to 1.5-fold, depending on the reference gene used (Fig 6c). This variability reflects the sensitivity of single-gene normalization to inter-sample variation and occasional outliers. In contrast, normalization with combined NF produces values that exhibit significantly smaller standard deviations (see Table 3). Similar trends were evident in H. dujardinii and E. fluviatilis: the single-reference normalized values varied considerably, whereas the NF-normalized values clustered tightly, indicating consistent expression.

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Normalized RHOA mRNA expression for individual experiments in regeneration model of L. corallorrhiza (as shown in Fig 6b). Mean ± SD. Note the instability of results in case of normalization using only one gene.

https://doi.org/10.1371/journal.pone.0349507.t003