Unraveling the miRNA landscape in tsetse fly saliva and salivary glands during Trypanosoma brucei infections

Abstract

MicroRNAs (miRNAs) are emerging as potential mediators of cross-species gene regulation in vector-borne diseases. Using small RNA-sequencing, this study investigated the presence of miRNAs in the tsetse fly (Glossina morsitans)-trypanosome parasite system, focusing on the fly’s salivary glands and saliva during Trypanosoma brucei infections. To ensure annotation accuracy, we implemented a robust pipeline incorporating de novo prediction, comparative analyses and manual curation to minimize false positives. This resulted in the detection of 54 unique miRNAs in tsetse saliva and salivary glands, among which 12 were novel to tsetse flies, 5 were putative novel miRNAs and none were originating from T. brucei. Salivary gland samples contained most detected miRNAs, with a large subset also present in saliva, some exhibiting high relative abundance. Interestingly, two miRNAs appeared more represented in saliva as compared to salivary glands (let-7 and miR-8) and one miRNA (miR-184) was only detected in uninfected saliva. The presence of a T. brucei infection in the salivary glands significantly reduced miRNA diversity and abundance in saliva. Target analysis of tsetse salivary miRNAs identified numerous potential host targets, but functional enrichment was limited. In conclusion, this is the first description of miRNAs in the salivary glands and saliva of a tsetse fly species, a blood-feeding vector of African trypanosomes. Our study emphasizes the importance of a robust bioinformatics pipeline in de novo detection of miRNAs in non-model organisms and unconventional matrices like insect saliva. Our work highlights the challenges and limitations of miRNA target prediction, underscoring the need for rigorous experimental validation of miRNAs in host–vector–pathogen interactions.

Graphical Abstract

Graphical Abstract.

Introduction

The mutual modulation of gene expression impacting biological pathways is one of the mechanisms in the arms race between pathogens and their hosts. In recent years, microRNAs (miRNAs) have been implicated as cross-species gene expression regulators [1]. MiRNAs consist of 19–25 nucleotides and have been documented as endogenous regulators of gene expression in eukaryotes. They silence target genes by pairing with a (partially) complementary sequence of the respective messenger RNAs (mRNAs). Gene silencing is mediated by the argonaute protein and involves translational repression and/or mRNA decay. Within the context of host–parasite interactions, miRNAs have been documented as cross-species regulators for the blood stage of the Plasmodium falciparum malaria parasite [2, 3] and for several helminths, like the liver fluke Clonorchis sinensis, and the rodent intestinal nematode Heligmosomoides polygyrus [3–5]. Recent research has also suggested a possible modulating role of miRNAs in infections borne by blood-feeding arthropods (BFAs). Here, cross-species interactions may occur not only between pathogen and host but also between the vector and pathogen, and the vector and host. At the vector-to-host transmission stage, through the bite of the BFAs, miRNA-containing saliva is injected into the host dermis, together with the pathogen. Some of these miRNAs, originating from the vector and/or the pathogen, are hypothesized to modulate the host physiology and immune response at the bite site, affecting the establishment of the pathogen infection in the host [6–12].

MiRNA-mediated actions within the pathogen–vector–host triad have been investigated for Chikungunya virus (transmitted by Aedes mosquitoes) [6, 11], for tick-borne infections (with as vectors Ixodes scapularis and Ixodes ricinus) [7, 10], and in Plasmodium infections (transmitted by Anopheles mosquitoes) [8]. Interesting to note here is the recurrence of identical miRNA families in the saliva of diverse arthropods, possibly implying their functional conservation [8, 11]. Other outstanding findings include the differential salivary expression of specific miRNA families according to vector infection status [6–8, 10], the enrichment of host immune-related gene targets of these salivary miRNAs [6–8, 11], and infection-enhancing effects of enriched miRNAs in cell cultures [6, 10].

However, several important methodological challenges remain. The first of these is the reliable identification and quantification of miRNAs in small-RNA sequencing data in real-life biological samples. MiRNA prediction carries significant risks of misinterpretation and over- and under-detection [13, 14]. Studies of cross-species interactions in BFA rely on small RNA (sRNA) sequencing data in complex matrices like arthropod saliva, possibly adding variables that affect miRNA predictions. Discrepancies in the number of miRNAs identified in studies of BFA-borne infections are striking. For example, in two methodologically similar studies of Ixodes ticks’ saliva, respective numbers of miRNA identifications amounted to 67 miRNAs (of which 35 novel) and 379 miRNAs (of which 338 novel) [7, 10]. In the protozoan parasite Trypanosoma brucei, estimates of the miRNA repertoire ranged from none [15, 16] to 26 [17], to over one thousand [18].

Second, questions remain regarding the mode of action of miRNAs in the extracellular environment. Besides miRNAs and their target host cells, the bite site contains large volumes of lysed-cell constituents, tissue exudate, and extravasated blood. Molecules that act as cross-species regulators must attain a concentration and mode of transportation that enables them to reach and modulate their targets. Exosomal miRNA delivery has been proposed for counteracting dilution and binding competition by other RNA species. However, exosomal sorting signatures of miRNAs have been only inconsistently found in the saliva of BFAs [7, 8, 11]. Additionally, it has been questioned whether the quantities of miRNAs within exosomes are sufficient for modifying targets in recipient cells [19].

The third challenge concerns the complexity of the interaction of miRNAs with their targets. MiRNAs exhibit target abundance: each given miRNA family binds to target sites on many different mRNAs (e.g. for canonical human miRNAs this amounts to >300 gene targets per miRNA family) [19]. Therefore, target analysis of any given miRNA deals with networks comprising hundreds of genes, to which assigning a discrete biological effect is challenging. This is even more critical when done across biologically unrelated species (protozoan, arthropod, and vertebrate species). Indeed, analyses that identify cross-species-modulated pathways exhibit high false discovery rates [8] and often result in predicted target genes/pathways that cannot be directly linked to pathogen infection and disease progress in the host [7, 8, 11].

In this study, we have investigated the miRNA presence in the tsetse fly-trypanosome parasite system, focusing on the salivary glands and saliva of the fly when infected with T. brucei. African trypanosomes are protozoan parasites transmitted by tsetse flies and cause Animal and Human African Trypanosomiasis (AAT and HAT). These diseases cause a serious burden of human suffering and socio-economic losses in sub-Saharan Africa [20]. This is the first time that this important vector-parasite infection model is studied in the context of miRNA-mediated cross-species interactions. We used small-RNA sequencing data of the saliva and salivary glands of T. brucei-infected and uninfected tsetse flies. To increase accuracy of analyses, we used a consensus approach of two state-of-the-art miRNA prediction models: miRDeep2 and sRNAtoolbox. Next, based on those high-confident results, we aimed to address the following questions: (i) are miRNAs present in the saliva of tsetse flies? And if yes, does an infection with trypanosomes affect the miRNA diversity and abundance? (ii) if present, do these miRNAs target specific biological pathways of the host that could impact host physiology at the bite site or the early parasite infection? In complement to the experimental work, we aimed to critically address the above-mentioned challenges of miRNA prediction in biological samples by an in-depth review of the existing literature and used methodologies.

Materials and methods

Tsetse fly infection, sample collection, and sRNA sequencing

Tsetse flies (Glossina morsitans morsitans) were obtained from the colony at the Institute of Tropical Medicine, Antwerp. These flies are maintained on defibrinated bovine/sheep blood using an in vitro feeding system, at 25 ± 0.5°C and a relative humidity of 65%–75%. To obtain infected flies, freshly emerged male tsetse flies were fed for their first blood meal with defibrinated sheep blood (E&O Laboratories LTD-UK) containing the T. brucei brucei AnTat1.1 bloodstream parasites (∼5.10⁶ tryps./ml) and supplemented with 10 mM reduced L-glutathione to enhance trypanosome infection establishment in the tsetse midgut [21]. The bloodstream trypanosomes were collected by heart puncture from cyclophosphamide-immune suppressed mice at day 7 after the initial trypanosome inoculation. Then, the flies were further maintained for 4 weeks while feeding three times a week on uninfected defibrinated sheep blood. Thirty days after the initial blood meal, flies that carried a metacyclic trypanosome infection in the salivary glands were selected by induced probing on a microscopic slide and subsequent examination of the fly spit by phase-contrast microscopy (400×). Salivary glands were dissected in ice-cold sterile physiological H₂O, and the saliva was allowed to flow out for at least 2 h. Then, salivary glands, saliva, and metacyclic parasites were obtained separately by sequential centrifugation. Salivary glands and saliva of uninfected male flies (control group) of the same age and with the same feeding history were also sampled at this stage.

All samples (infected and uninfected) were obtained from two independent infection-collection cycles, yielding samples each derived from a pool of 35 flies, i.e. one pool of 35 flies is considered as a biological replicate. At the end, six pools of salivary gland, saliva, and metacyclic parasite samples were obtained from the infected flies. From the uninfected flies, seven pools of salivary gland and saliva samples were obtained. Total RNA from these 32 pooled samples (Fig. 1) was extracted using the miRNeasy Micro kit (Qiagen, Germany), adapter-ligated using the NEXTflex Small RNA-Seq Kit (Perkin-Elmer, Waltham, MA, USA), and sequenced by GenomeScan (The Netherlands) on the NovaSeq 6000 platform (Illumina) (Fig. 1). Raw sequencing data have been deposited in the NCBI GEO database under the accession number GSE288319.

Figure 1.

Overview of obtaining samples. Figure made in biorender.com https://BioRender.com/m19n282

Quality control, read trimming, and genome mapping

Adapter trimming was performed using Cutadapt according to trimming instructions provided for the NEXTflex kit [22]. Read mapping was performed using Bowtie1 (-k 100 –best –strata) [23]. For samples from uninfected flies, mapping was performed using the G. morsitans Yale genome (GmorY1), obtained from VectorBase - release 63 [24]. For samples isolated from the infected flies, mapping was performed using a merged genome index comprising the GmorY1 genome and the T. brucei TREU927 (TbT927) genome, obtained from TriTrypDB—release 63 [25]. To annotate reads per read length (for visualizing the distribution of RNA type like rRNA, sRNA, mRNA, etc. according to their length), reads were sorted for each category, mapped, and annotated per read length. For this, gff annotation files for both reference genomes were downloaded from VectorBase and TriTrypDB. Gff files were merged and harmonized using scripts from the AGAT suite (https://github.com/NBISweden/AGAT). The featureCounts program was subsequently used for assigning reads to gene features (allowing overlapping annotations and multiple feature assignment; -M -O arguments) [26]. Additional mapping statistics were obtained using Qualimap v. 2.3 [27]. Mapping results were summarized using MultiQC [28]. Taxonomic read classification to check for potential contamination or check the presence of symbiotic species, was done using Kraken2 [29], with the pre-built Kraken2-Microbial database (version September 2018, https://lomanlab.github.io/mockcommunity/mc_databases.html). Output analysis and visualization was done with KronaTools [30].

De novo miRNA annotation by sRNAtoolbox and miRDeep2

To increase the accuracy of our miRNA annotations, two highly-cited general-purpose miRNA annotation tools were used: sRNAtoolbox [31, 32] and miRDeep2 [33]. To avoid biases stemming from low-quality miRNA annotations existent in the literature, the identification of miRNAs was performed by de novo prediction, rather than by using available annotations for G. morsitans and T. brucei [17, 18, 34–36]. For all 32 samples, sRNAtoolbox (version 2.0, July 2020) and miRDeep2 (version 2.0.1.2) were run in genome mode, implying that reads are mapped against the entire genomes of the organisms of interest. Both tools implement algorithms for dealing with multimapping reads and reads originating from non-miRNA genes, and therefore no prior read clean-up was performed. The G. morsitans GmorY1 genome was downloaded in the sRNAbench module of sRNAtoolbox using the built-in populate script. The T. brucei TREU927 genome and its annotations (obtained from TriTrypDB) were manually added to the database, according to instructions of the tool’s creators. Since multi-species analyses are not supported in miRDeep2, samples from infected flies were independently mapped on both the GmorY1 and T. brucei TREU927 genome. Results were collated afterward. For homology-supported scoring of putative miRNA loci, miRDeep2 was supplied with mature miRNA sequences of Drosophila melanogaster from miRGeneDB.

Comparative analyses and curation of miRNA annotations

MiRNA annotations of both tools were compared using the BEDtools multiIntersectBed and intersect functions (defining identical putative miRNAs as those for which the precursor miRNA coordinates overlapped) [37]. To establish validity of detected putative miRNAs, various sources were used. Obtained putative miRNA sequences were compared with existing literature sources in the field of tsetse flies and African trypanosomes [17, 34–36]. For this, comparison was done using the mature sequence of the miRNA, since precursor coordinates were either not available [17, 34, 36] or were reported for the genome scaffolds of Glossina pallidipes (another tsetse fly species), precluding direct comparison [35]. An early study that reported 1162 putative miRNA genes in T. brucei does not report the miRNA mature sequences or genome coordinates per miRNA gene, and was therefore excluded from analysis [36]. In the process of matching miRNAs based on their mature sequences, a shift of up to two nucleotides was allowed to accommodate small differences in how the mature sequence was delineated. For further analysis, RNAcentral was queried using the mature sequence of each putative miRNA. Finally, as an alternative approach to miRNA prediction, we used mirMachine, a state-of-the-art online miRNA annotation tool that predicts likely miRNA genes from the genome rather than from sRNA sequencing data [38]. For this, the GmorY1 genome was uploaded, and as a reference phylogenetic node Drosophila was selected.

Differential expression analysis

For differential expression analysis of miRNAs detected by both miRDeep2 and sRNAtoolbox, miRDeep2’s quantifier.pl script was used. Per mature sequence, miRDeep2 may report multiple read counts, according to different likely precursor genes assigned to each subset of reads for that mature sequence. For these miRNAs, we averaged the read count obtained across several precursor genes. A read count matrix was imported in DeSeq2 for count normalization and differential expression analysis. When a given miRNA was undetected in any of the sample types included in differential expression analysis, the count was set to 1, assuming the presence of the miRNA but below the read count threshold required for its annotation. The threshold for significant miRNA expression was set to a P-value of 0.05. Plotting was performed using the ggplot2 and pheatmap packages in the R programming language. In the heatmaps, hierarchical clustering was performed using the default complete linkage method and Euclidean distance as implemented in the pheatmap package. Differential expression analysis between infected and uninfected glands and saliva was performed using Student’s t-test. A pseudocount of 1 was added to calculate the log2 fold change. Visualization of differentially expressed miRNAs was carried out using the EnhancedVolcano R package.

Exosomal sorting signatures

We investigated the presence of exosomal sorting signatures in three ways. First, similar to Hackenberg et al. [7] and Arcà et al. [8], we summed the total number of adenine and uracil non-template additions in miRDeep2 reads for differentially expressed salivary miRNAs and compared the U/A ratio. For this, individual reads were extracted per miRNA annotation from raw miRDeep2 files (MRD files). First, non-template additions were counted relative to the 3′ end of the mature miRNA sequence. Second, the presence of exosomal sorting motifs, extracted from two studies [39, 40], was screened and their proportion determined in saliva- or salivary gland-enriched miRNAs. We tested for significance of differences in proportions using a Chi square test. Third, the Improbizer web server [41] was used to find enriched motifs in RNA sequences.

Target analysis

We investigated if target genes of miRNAs enriched in tsetse saliva demonstrated enrichment of immunological pathways in the human host. For this, we performed target analysis using four different target prediction tools: miRanda [42], PITA [43], TargetSpy [44], and TargetScan [45]. The first three tools are run using the miRNAconsTarget module of sRNAToolbox [32] (webserver version, November 2023) with default settings, using the Homo sapiens 3′-UTR catalog for target prediction. TargetScan was run using the standalone version 7.0, also with default parameters and using the 3′ UTR-database as downloaded from the TargetScan website.

Filtering of the target prediction results was performed using criteria specific to each prediction tool. Binding affinity (in kcal/mol) was considered for miRanda, PITA, and TargetSpy, the reported score was used for TargetSpy, and seed match types were evaluated for TargetScan. Only gene targets predicted by all four tools were retained for downstream analysis. Three filtering strategies—lenient, normal and stringent—were applied, each combining the specified cutoffs to reflect different levels of stringency.

For lenient filtering, the binding affinity threshold was set to < −20 kcal/mol for miRanda and PITA, < −15 kcal/mol for TargetSpy, with a TargetSpy score >0.9950, and seed match types restricted to 8mer-1a or 7mer-m8 for TargetScan. Normal filtering required a binding affinity < 25 kcal/mol for miRanda and PITA, < −20 kcal/mol for TargetSpy, a TargetSpy score >0.9950, and seed match types limited to 8mer-1a or 7mer-m8 for TargetScan. For stringent filtering, the binding affinity had to be < −27.5 kcal/mol for miRanda and PITA, < −20 kcal/mol for TargetSpy, with a TargetSpy score exceeding 0.9975, and seed match types restricted to 8mer-1a for TargetScan. Target gene transcript identifiers were converted to gene names using the Ensembl BioMart tool [46]. Resulting gene names were then uploaded to the WebGestalt [47] server. Over-representation analysis of functional categories among salivary-expressed miRNAs was performed using the Homo sapiens protein-coding genome as the reference. The analysis was conducted on the WebGestalt platform, targeting Gene Ontology (GO) Biological Processes, KEGG pathways, PantherDB, and WikiPathways. Functional category analysis was performed on the predicted target genes of all salivary-expressed miRNAs, as well as on a subset of miRNAs selected based on their expression levels (measured by read counts), using cut-off thresholds of 2500 and 10 000 reads.

Results

miRNA sequencing data

We generated miRNA sequencing data across five biological conditions: salivary glands and saliva from both T. brucei infected and uninfected flies, as well as metacyclic stage parasites isolated from the saliva of the infected flies (see Fig. 1). Samples were collected from two independent infection cycles conducted several months apart, with each sample derived from a pool of 35 flies. In total, we obtained six pooled biological replicates for salivary gland, saliva, and metacyclic parasite samples from infected flies, and seven pooled biological replicates for salivary gland and saliva samples from uninfected flies, resulting in total in 32 samples. Read counts and mapping/annotation rates per biological condition are shown in Table 1. For each sample type (infected and uninfected saliva and salivary glands, and metacyclics), ∼150–200 × 10⁶ reads were obtained. Overall, we obtained mapping rates > 95% for all samples, using Bowtie1 with -k 100 –best –strata parameters (a commonly used mapping strategy for sRNAs) [23]. However, applying a mapping strategy that discards multimapping reads (-k 1 -m 1 –best –strata), results in removal of up to one third of reads, a phenomenon more pronounced in infected versus non-infected samples, and salivary versus gland-derived samples. The increase in multimapping reads in these samples is compatible with an increase of small, likely partially degraded RNA fragments (see further), which are likely to be more abundant in saliva samples compared to salivary gland samples. Taxonomic classification demonstrates negligible read counts of symbiotic organisms like the bacterium Sodalis glossinidius (<1%) for all samples.

Table 1.

Read counts and mapping rates per sample type

	Pools	Raw read count (106 reads)	Read count post trimming (106 reads)	Total % mapped (retain multimapping reads)	% Mapped and assigned to feature	% Mapped but unassigned to feature	% Mapped (discard multimapping reads)
Salivary gland, infected	6	158.4	156.2	99.3%	62.0%	37.3%	92.3%
Salivary gland, uninfected	7	191.5	187.2	95.6%	75.8%	19.8%	94.1%
Metacyclic trypanosomes	6	213.9	211.4	98.6%	73.4%	25.2%	61.5%
Saliva, infected	6	217.4	216.0	99.1%	69.2%	29.9%	73.6%
Saliva, uninfected	7	251.9	241.4	96.6%	45.8%	50.8%	86.9%

Annotation of miRNA reads

First, we investigated whether the reads mapped to known genomic features and—if possible—annotated them using the genome annotations of G. morsitans and T. brucei. Despite high mapping rates, significant portions of the reads could not be assigned to annotated genes of either organism using the featureCounts program (e.g. up to 50.8% of reads in uninfected saliva, see Table 1). Further analyses using Qualimap v. 2.3 determined that many of these reads map to intergenic regions (data not shown).

To determine the distribution of RNA species present in our samples, we excluded ambiguous, multimapping reads (Bowtie1 with -k 1 -m 1 –best –strata arguments) and plotted the distribution of different types of RNA molecules (e.g. ribosomal RNA (rRNA), mRNA, and miRNA) per read length, including information on the species it originated from (Fig. 2). For the salivary gland samples and negative saliva samples, we observe a clear enrichment of reads mapping to previously known miRNA genes for the reads with an anticipated miRNA length between 19 and 25 nucleotides (red bars in Fig. 2), confirming that our protocol is optimized for the identification of miRNAs. In infected salivary samples, the proportion of miRNA gene mapping reads of this length is overshadowed by mRNA-mapping fragments of tsetse and trypanosomal origin. The reads mapping to G. morsitans RNA species in the metacyclic trypanosome samples are most likely originating from individual salivary glands cells that were co-pelleted when collecting the metacyclic cells by centrifugation of the infected saliva, prior to RNA extraction.

Figure 2.

Alignments per sample type (multimapping reads discarded), annotated per read length, using the G. morsitans Yale (GmorY1) and T. brucei TREU927 (TbT927) genomes. Lnc_RNA, long non-coding RNA; mRNA, messenger RNA; ncRNA, non-coding RNA; pre_miRNA, precursor miRNA; rRNA, ribosomal RNA; snRNA, small nuclear RNA; snoRNA, small nucleolar RNA; tRNA, transfer RNA. Gland_neg, uninfected gland samples; gland_pos, infected gland samples; saliva_neg, uninfected salivary samples; saliva_pos, infected salivary samples; metacyclics, metacyclic T. brucei parasites.

Comparison of miRDeep2 and sRNAtoolbox miRNA annotations

Two different de novo miRNA prediction algorithms were applied on all 32 miRNA sequencing data sets, i.e. miRDeep2 and sRNAtoolbox. Predictions by miRDeep2 are by default unfiltered, and amount to 2.5 to almost 13 times the number of predictions by sRNAtoolbox (Table 2, columns “All miRDeep2 predictions” and “sRNAtoolbox annotations”). This difference is particularly observed in the infected saliva and metacyclic trypanosome samples, which contain the highest number of multimapping reads. After filtering miRDeep2 predictions (using a confidence score threshold of 4 and by exclusion of predictions with miRDeep2-flagged similarity to tRNA/rRNAs), the number of miRDeep2 miRNA predictions remained on average two times higher than those of sRNAtoolbox (Table 2 and Fig. 3). Comparing genomic regions and coordinates of each miRNA, 75% of sRNAtoolbox-predicted miRNAs are also predicted by miRDeep2, whereas only 33% of miRDeep2 miRNA predictions are also in the output of sRNAtoolbox (Fig. 3). This suggests that, even when miRNA predictions are filtered to only include the highest-confidence predictions, miRDeep2 still generates more predictions than sRNAtoolbox does. For both tools, median confidence scores and read counts are lower for predictions exclusive to each tool than for those common between both tools. Examining reasons for non-overlap, only seven of 190 sRNAtoolbox-exclusive annotations were found in read mapping (MRD) files of miRDeep2. This suggests that the main part of the observed prediction discrepancy stems from the read mapping process preceding the de novo identification of miRNAs [33] rather than from the confidence scoring process.

Table 2.

The number of miRNAs predicted by miRDeep2 (with different filtering strategies) and sRNAtoolbox per sample type, averaged across the 6 or 7 individual samples belonging to each sample type

	miRDeep2 miRNA predictions
Type	All miRDeep2 predictions	Predictions with score > 2	Predictions with score > 4	Predictions with score > 4 and Rfam alert negative	sRNAtoolbox annotations	Average ratio predictions miRDeep2 score >4/ sRNAtoolbox
Salivary glands, infected	98.5	78.2	64.5	64.5	35.7	1.8
Salivary glands, uninfected	89.3	73.7	65.7	65.6	36.3	1.8
Metacyclic trypanosomes	172.5	72.2	57.8	57.7	21.7	2.7
Saliva, infected	148.0	50.3	40.7	40.3	11.7	3.5
Saliva, uninfected	73.9	53.9	45.3	45.1	19.7	2.3

Figure 3.

Overlap of miRNA predictions by miRDeep2 (filtered for confidence score > 4 and Rfam alert negative) and sRNAtoolbox summed per sample type (i.e. not averaged per sample).

Curated miRNA predictions

A flowchart of selecting and filtering miRNA predictions that overlap between miRDeep2 and sRNAtoolbox is shown in Fig. 4. After deduplication per sample type, the number of unique miRNAs per sample type are shown in Supplementary Table S1, ordered per miRNA family, and returning in total 59 putative miRNAs. For each of the 59 miRNAs, manual curation was performed as outlined in the “Materials and methods” section, utilizing detailed miRNA predictions in tsetse flies from previous publications for annotation [17, 34–36]. Details on the information sources used to obtain this manual annotation are shown in Supplementary Table S2. A manual annotation search in five entries with no homology to known miRNA families demonstrated that these are likely false positives as they map to rRNA genes or coding genes. In total, 54 high-confidence annotations were obtained, all of them mapping to loci of the G. morsitans genome, and none to the T. brucei genome. For five of these 54, we could not find homology with any of the known miRNA families, or with entries in the RNA families database; we considered these therefore as putative novel miRNAs. The characteristics of these five putative novel miRNAs are shown in Table 3. Additionally, 12 miRNAs were not previously described in any of the non-salivary miRNA studies in tsetse flies. When performing genomic miRNA gene prediction using MirMachine (a tool for predicting miRNA genes solely based on the organism’s genome sequence, instead of from experimental small-RNA sequencing data), 45 of our 54 sRNA-sequencing-based miRNA predictions were also predicted by MirMachine in the G. morsitans genome (MirMachine output is given in Supplementary Table S3). Additionally, we examined how many of the MirMachine-predicted miRNA loci (based solely on genome sequence and conservation) are supported by small RNA sequencing reads in our dataset. Of the 83 miRNA predictions from MirMachine, 45 overlapped with the miRNA list obtained from the intersection of miRDeep2 and sRNAtoolbox. Among the remaining 38 MirMachine predictions, we observed read coverage for five additional loci—specifically, one miR-14, one miR-96, and three miR-2-like miRNAs. These miRNAs are highlighted in Supplementary Table S3.

Figure 4.

Filtering and curating miRNA annotations. A flowchart is shown of the number of annotations obtained by both tools, and subsequent deduplication and filtering steps. Image created in BioRender https://BioRender.com/5ikdi39.

Table 3.

Characteristics of five putative novel miRNAs

GmorY1 scaffold and approximate coordinates	Mature sequence (miRDeep2)	Precursor sequence (miRDeep2)	miRDeep2 score	sRNAtoolbox score	No. of samples found
scf7180000638986: 38617–38 667	Aaagugauuggaugauuaucga	aauaaucagucaaucgcuuuggcg ucauaaagugauuggaugauuaucga	33.3	5/6	1
scf7180000645283: 176258–176 315	ucacggcaaccacauucagugg	ucacggcaaccacauucagugguugg uauauguccacggaaauugguugucguaau	0.2	5/6	5
scf7180000648447: 12063–12 109	uuuacggucuugucaauauca	gaugaagagacugaauuggauuu uauuuacggucuugucaauauca	106.2	5/6	1
scf7180000648877: 58199–58 260	uugaagaugauagauuaaaug	uugaagaugauagauuaaaugcuuacuacau aacauugcacauaaucucuuaucuacauau	2.5	6/6	3
scf7180000651863: 23513–23 571	acuacugcugaagcacaacgug	aauugugcuucagcagucagcgggggaauu ucaacuacuacugcugaagcacaacgug	59.5	6/6	11

The distribution of the occurrence (absence/presence) of miRNAs across the 32 samples is shown in Fig. 5. Overall, this figure suggests that most of the miRNAs are predicted in the salivary gland samples (whether infected or non-infected), with a large subset of these miRNAs also present in uninfected saliva. A smaller subset is found in saliva originating from infected flies. The consistency of the predicted miRNAs across different biological replicates for each condition underscores the robustness of our approach. To further validate the robustness of our approach, we conducted hierarchical clustering and principal component analysis (PCA) on the miRNA count matrix (Supplementary Fig. S1A and C, respectively). These analyses confirmed the clear grouping of samples according to tissue type (saliva or salivary glands) and revealed that T. brucei infection notably influenced the miRNA profiles in saliva, with a more modest effect observed in the salivary glands.

Figure 5.

(A) The distribution of occurrences of 54 miRNA predictions (rows) across samples and sample types (columns). A green square indicates that the miRNA was predicted by both miRNA detection algorithms. (B) Volcano plots comparing uninfected glands to uninfected saliva (left), and infected glands to infected saliva (right). miRNAs highlighted in red have a P-value below 0.05 and an absolute log2 fold change >1, with their names labeled on the plot. Points shown in blue but without labels indicate miRNAs with a P-value below 0.05 only, points in green have an absolute fold change higher than 1 only.

To identify miRNAs differentially expressed between salivary glands and saliva, we performed a differential expression analysis using DESeq2. However, this approach yielded skewed results, likely due to the characteristics of the dataset and the assumptions underlying advanced normalization methods. Although all 32 samples were sequenced to comparable depths and produced similar numbers of raw reads, the number of reads mapping specifically to miRNAs varied widely after processing, ranging from ∼1000 to several hundred thousand (see Supplementary Figs S1A and S2, and Supplementary Data S2). As a result, normalization methods that assume most features are not differentially expressed—such as those used in DESeq2—may produce unreliable size factors and biased results when applied to small and sparse miRNA datasets like ours. As could be expected, the number of reads per sample mapping to miRNAs was higher in the salivary gland samples compared to saliva samples. Within the latter, samples infected with T. brucei showed lower coverage than non-infected samples. To ensure that this observation in infected samples was not due to a lower proportion of sequencing reads mapping to G. morsitans and a higher proportion mapping to T. brucei, we examined the absolute number of reads mapping to G. morsitans in both infected and uninfected samples, which revealed similar counts. (Supplementary Table S4)

As an alternative, albeit less statistically robust approach, we visualized the raw read counts (log10-transformed) and the relative percentages of reads per miRNA per sample in a heatmap (Supplementary Fig. S1B and Supplementary Data S2). Examining the percentages of miRNA-mapping reads per sample revealed that miRNAs with a high proportion of reads are also, relatively speaking, the most abundant in the salivary samples. Notably, an interesting pattern emerged for Let-7 and Mir-8: the percentage of reads mapping to these miRNAs was higher in saliva samples compared to salivary gland samples, increasing on average from 7.0% to 24.6% for Let-7 and from 2.5% to 16% for miR-8. Additionally, miR-184 showed an average coverage of 8.5% in all uninfected saliva samples, whereas no reads mapping to this miRNA were detected in salivary gland samples. These observations are further supported by volcano plots comparing infected glands to infected saliva, as well as uninfected glands to infected saliva (Fig. 5B), confirming the statistically significant upregulation in saliva of Let-7 and miR-8 for both comparisons and miR-184 for uninfected saliva.

For the two miRNAs (Let-7 and miR-8) with higher percentages of reads in saliva compared to salivary glands, this pattern may suggest an active export mechanism. However, we found no evidence supporting exosomal sorting in the saliva-enriched miRNAs analyzed. Specifically, we were unable to (i) detect a high fraction of uridylation in the miRNAs, (ii) identify the four-nucleotide sequence motifs associated with exosomal or cytosolic enrichment, and (iii) identify via a de novo motif detection tool (Improbizer) conserved motifs that might act as a novel recognitions site for exosomal sorting (see Supplementary Text S1 and Supplementary Fig. S7).

When analyzing the heatmap of absolute read counts (Supplementary Fig. S1A), we observed that overall, miRNAs with a high read abundance in salivary gland samples also tended to recur in saliva, with the exception of miR-184. Conversely, miRNAs only present in salivary gland samples and absent in saliva generally exhibited already low read abundance in the salivary gland samples.

Literature comparison of miRNA families

A comparison of the obtained miRNA families in our salivary samples of tsetse flies with miRNA predictions found in studies of Ixodes ticks [7], Aedes aegypti [11], Aedes Albopictus [6], and Anopheles [8] mosquitoes is shown in Fig. 6, along with the phylogenetic node of origin per family. Across these six studies, miRNAs belonging to 152 unique miRNA families are predicted. Our analysis shows that 56 out of 152 miRNA families recur in at least three of these six salivary BFA datasets. When considering the phylogenetic node of origin of each miRNA family, a significant proportion of recurring miRNAs originate at the node of Bilateria, an early evolutionary group that comprises all arthropods and their possible mammal hosts (Supplementary Figs S3 and S4).

Figure 6.

Presence or absence of miRNA families detected in saliva [6, 8, 11, 7] of Glossina morsitans, Ixodes ricinus, Anopheles coluzzi, Aedes aegypti (two studies) and Aedes albopictus (limited to miRNA families detected in the present study of G. morsitans). The phylogenetic node of origin is shown between brackets, as was obtained from miRGeneDB. If not provided in miRGeneDB, a dash (-) is shown. Information on data extraction is provided in Supplementary Fig. S4.

Target analysis

To explore the potential role of saliva-injected miRNAs at the tsetse bite site for interacting with the host hemostatic and/or immune response, we performed target analysis on the group of miRNAs encompassing all salivary miRNAs, consisting of 34 miRNAs. We employed the target detection strategy as described in the “Materials and methods” section, performing target analysis with four distinct tools: miRanda, PITA, TargetSpy, and TargetScan. Based on the cut-off parameters outlined in the “Materials and methods” section, target sets were generated using lenient, normal, or stringent filtering approaches. Only targets identified by all four tools were retained for downstream analysis, which resulted in gene sets consisting of 8345, 902, and 177 genes for the lenient, normal, and stringent filtering approach, respectively. Overrepresentation analysis using WebGestalt on the three gene sets derived from different stringency levels primarily showed statistically significant enrichment (False Discovery Rate (FDR) < 0.05) for some functions and pathways in the gene set generated using the lenient filtering criteria (Supplementary Figs S5 and S6). In contrast, the other two gene sets showed either no or a limited number of enriched pathways or functional classes. Additionally, when including results with an FDR > 0.05, no consistent pattern of enriched functions emerged when integrating the three different stringency levels. Most of the pathways identified were challenging to associate unambiguously with hemostatic or immune response related processes. We repeated the analysis, this time focusing exclusively on the most abundant miRNAs in the saliva samples, as determined by their read counts. Applying a threshold of 2500 reads resulted in 8 miRNAs, leading to a number of target genes of 6871, 2183, and 133 genes for the lenient, normal, and stringent filtering criteria, respectively. When the threshold was increased to 10 000 reads, 17 miRNAs were selected and the number of genes was reduced to 2881, 180, and 25 for the same respective stringency levels. Overall, the results were consistent with the initial analysis: statistically significant pathways were only identified under the lenient criteria, and none of these could be clearly associated with immunology-related processes.

Discussion

MiRNAs are involved in the post-transcriptional regulation of almost all cellular processes and consequently play important roles in the development of many diseases [48]. Moreover, they have been investigated as potential cross-species immunological regulators in the pathogenesis of infectious diseases. In vector-borne infections, it has been hypothesized that miRNAs originating from the vector or pathogen, delivered through the vector’s saliva, could modulate the host physiology and immunity at the inoculation site to enhance infection establishment following the bite of an infected vector. However, reliably predicting and annotating these miRNAs within this complex experimental framework remains a significant challenge. In this study, we developed a robust bioinformatics pipeline to explore the presence of miRNA molecules in a previously unexplored infection model, i.e. the tsetse fly (G. morsitans)-trypanosome transmission model. Our approach systematically addresses key challenges in the detection, annotation, and target prediction of salivary miRNAs in G. morsitans, paving the way for deeper insights into their potential roles in host–tsetse–trypanosome interactions.

Studies of miRNA annotation are known to be vulnerable to biases, particularly in non-model species (such as those involved in arthropod-borne infections) and unconventional biological matrices like insect saliva. Even in model species, such as Homo sapiens or Drosophila melanogaster, false miRNA annotations abound in non-curated databases, with two-thirds of human annotations in miRBase taken to be false positives [49, 50]. Indeed, as of today, only ∼5% of the human miRNA entries appear to be experimentally verified. Additionally, systematic evaluations indicate that many of the annotated sequences likely do not represent true miRNAs, but other RNA species or even sequencing artifacts [48]. We therefore emphasized methodological robustness in our miRNA prediction analyses, using comprehensive quality control and a consensus approach to prediction using two state-of-the-art miRNA prediction tools (sRNAtoolbox and miRDeep2). Using this methodology, we identified 54 unique miRNAs in salivary samples derived from uninfected and trypanosome-infected G. morsitans tsetse flies, all originating from G. morsitans and none from the T. brucei parasite. To our knowledge, 12 of these have not yet been described in tsetse flies. Five miRNAs, predicted by two independent prediction algorithms, lacked homology with known miRNA families, and therefore possibly represent novel miRNAs (Table 3). Comparing miRNA predictions using sRNAtoolbox and miRDeep2, the number of predictions by miRDeep2 is two to thirteen times that of sRNAtoolbox, depending on filtering thresholds used in miRDeep2. Most of the sRNAtoolbox-exclusive predictions could not be found in raw read signature data of miRDeep2, suggesting that these predictions result from reads that are not mapped by miRDeep2. This implies that variables in mapping algorithms between both tools cause most discrepancies. Indeed, methodologically similar studies that use sRNAtoolbox or miRDeep2 report a vastly different number of miRNA predictions [7, 10]. Notably, even when using a single tool, miRDeep2, variable filtering thresholds in use in the literature [the default score 0(8, 11) compared to 2(37) or 4(10)] result in a 43% reduction and a 52% reduction in the number of predictions in our dataset, respectively. Characterization of the composition of small RNA species in our dataset demonstrates the presence of many small, non-specifically mapping RNA fragments. The presence of such spurious RNA fragments is a well-known challenge in miRNA annotation, which whole-genome mapping algorithms in sRNAtoolbox and miRDeep2 try to mitigate. Therefore, studies that rely on custom, “in house” miRNA prediction algorithms, rather than validated tools like miRDeep2 or sRNAtoolbox, are at an even higher risk of reporting erroneous predictions.

In our dataset, using validated miRNA prediction tools, we could only retrieve curated miRNA predictions originating from G. morsitans. As such, despite using a de novo prediction approach, we did not obtain curated miRNA predictions in T. brucei. Indeed, the presence of miRNAs in unicellular organisms remains an open question, as highlighted by a study re-annotating published miRNAs among seven unicellular eukaryotes, including three human pathogens, which finds that only 28 of the 159 miRNAs published among these species meet validated miRNA annotation criteria, and are all occurring in green and brown algae [51]. In T. brucei, specifically, various methods for miRNA identification have been used, yielding discrepant estimates of the miRNA repertoire, ranging from none [15, 16] to 26, [17] to over one thousand [18]. Two of these studies use small RNA-sequencing data, but do not perform de novo miRNA prediction, limiting sensitivity of the analysis to those miRNA annotations already previously known [15, 16]. Other authors screen public domain genomic sequences of T. brucei and describe 26 and 1162 putative miRNA genes in its genome, respectively [17, 18]. Both use custom-made algorithms based on homology-searches in miRBase of hairpin structures in the T. brucei genome. They use relatively lenient cutoffs for defining homological similarity and report a distribution of homological matches that is challenging to reconcile with evolutionary patterns. For example, their top matches include species as diverse as Caenorhabditis elegans, Oryza sativa (rice), Arabidopsis thaliana (thale cress), and Marek’s disease virus [18]. In addition to the methodological variability in miRNA prediction, T. brucei possesses a small interfering RNA system (siRNA), which mechanistically functions similarly as the miRNA system, and might further confound miRNA predictions [8, 11].

We have identified 51 miRNAs that are present in the tsetse salivary gland cells, of which 32 miRNAs could also be demonstrated in the corresponding saliva. Here, it remains unclear how these intracellularly produced miRNAs are transported to the saliva, as we could not detect exosomal sorting signatures in any of these miRNAs. The presence of a T. brucei parasite infection significantly impacted the abundance and diversity of miRNAs in the tsetse saliva, with only 20 miRNAs that could be detected above the threshold. This significant decrease in saliva miRNAs is probably linked to the perturbated functioning of the salivary gland cells in the trypanosome-infected flies. Indeed, previous studies have shown that the parasite infection in the tsetse salivary gland induces a local tissue response to tolerate the infection in the glands. However, it also strongly suppresses the constitutive cellular functioning of the glands, resulting in, for example, a highly reduced production of saliva proteins [52, 53] hampering the fly’s feeding performance. Saliva of tsetse flies is a complex potion and essential for its blood feeding behavior by counteracting host hemostatic responses such as vasoconstriction, platelet aggregation and coagulation reactions and modulating the local immune responses [54, 55]. Here, it can be hypothesized that some of the tsetse saliva miRNAs could also play an active biological role. The observation that a significant number of these miRNAs are also found in the saliva of other BFAs strengthens this hypothesis (Fig. 6 and Supplementary Fig. S3).

However, identifying the miRNA–host “target” genes remains challenging and speculative but is pivotal for elucidating the complex miRNA–host target gene regulatory networks. The high complexity of the latter is more and more getting acknowledged and many supposedly reliable findings must be regarded as preliminary, as they are grounded on far from a clear and uniform definition of a miRNA-“target” [48]. Moreover, miRNA–target networks have an inherent fuzziness since they are subject to constant dynamic changes, not at least in response to varying cellular conditions [48]. In addition, the many-to-many nature of posttranslational gene regulation by miRNAs, complicates computational target analysis. To address this challenge, we used a combination of available tools (i.e. miRanda [42], PITA [43], and TargetSpy [44], all three run via sRNAtoolbox [32], as well as TargetScan [45]) to enhance the prediction of possible miRNA targets. Despite employing different input filtering strategies and databases to identify enriched functions for the miRNAs detected in tsetse saliva, our analysis did not reveal any clear association with blood-feeding and/or immunology-related processes, or any other specific biological function. It is worth noting that previous publications also found enriched immunology-related targets only when allowing a false discovery rate above 0.05, or when reporting a high number of non-immunology-related targets alongside the immunological ones [7, 8, 11].

Exploring functional validation reports in the literature on miRNAs that we have identified n tsetse saliva, yields some interesting leads for their putative activity in the tsetse fly interaction with the mammalian host. Several miRNAs (such as Let-7, miR-1, miR-9, miR-10, miR-184, miR-190, miR-375, among others) have been demonstrated to play a role in the mammalian anti-pathogen response, in the modulation of cellular development, and in the regulation of different immune cells and/or inflammatory reactions in various vertebrate tissues and diseases [56–64]. Moreover, some miRNAs (miR-9 family) were also shown to affect processes such as angiogenesis. Some of these miRNA activities could be of relevance in the context of the tsetse fly bite and blood feeding event. Indeed, tsetse flies pierce the skin with their mouthparts during probing, while disrupting the surrounding tissue and capillaries generating an intradermal blood pool. During this event, the locally injected saliva is essential to prevent blood clotting by inhibiting the complex hemostasis cascade and also modulates the induced inflammatory response [54, 55, 65]. It can be hypothesized that some of miRNAs present in the tsetse saliva (especially the saliva-enriched miRNAs let-7 and miR8) contribute to facilitate the blood-feeding process by interacting with specific hemostatic reactions, or to the modulation of the early intradermal inflammatory response that is initiated by the bite. This immunomodulatory activity could also affect the early development of the tsetse-injected trypanosomes at the dermal site [66, 67].

We confirmed the recurrence of several miRNA families in Glossina saliva (such as let-7 and miR-8, among others) that have already been reported in saliva and/or salivary glands of Aedes and Anopheles mosquitoes, and Ixodes ticks. This recurrence of specific miRNA families in saliva of various BFAs might be indicative of a functional conservation [8, 11], suggesting a role in the blood feeding process by, for example, modulating hemostasis and/or host immune response. However, so far, experimental validation is currently missing.

In conclusion, our study offers the first insight into the miRNA landscape in tsetse fly saliva, potentially relevant in the tsetse fly–trypanosome–host transmission. It also underscores the challenges of miRNA prediction and annotation, particularly in non-model species. We developed a robust methodology allowing us to identify several novel miRNAs in the tsetse fly salivary glands and saliva. Despite methodological complexities, including the risk of false positives and biases in miRNA prediction, we demonstrate the presence of >30 different miRNAs in the tsetse saliva, with a significant decrease of diversity and abundance linked to the presence of a trypanosome infection. However, despite the use of various target prediction tools, we could not find clear indications for a functional role of these saliva miRNAs in the tsetse–trypanosome–host interplay at the bite site. This underscores the need for rigorous experimental validation of identified miRNAs as a challenging but only way forward to assess their role in modulating host biological pathways that are of relevance in the tsetse–trypanosome transmission.

Supplementary data

Supplementary data is available at NAR Molecular Medicine online.

Conflict of interest

There are no conflicts of interest to disclose.

Funding

Peter Odhiambo is funded by an individual PhD scholarship funded by the DGD/ITM Framework agreement. This work was funded by the FWO-senior research project n° G0A3922N (INNATEBITE). The computational resources used in this work were provided by the VSC (Flemish Supercomputer Center), funded by FWO and the Flemish Government.

Data availability

Raw sequencing data have been deposited in the NCBI GEO database under the accession number GSE288319.