Hidden in plain sight: illuminating the tRNA landscape by sequencing.

Introduction
Transfer RNAs, or tRNAs, are essential components of the translational machinery, acting as adaptors that decode messenger RNA (mRNA) into amino acids during protein synthesis. tRNAs are historically considered as house-keeping molecules. However, recent discoveries have shown that tRNAs are more than passive players in this process, with the help of high-throughput sequencing technologies. The biogenesis and function of tRNAs are governed by a series of intricately coordinated processes. In the beginning, tRNAs are transcribed by RNA Polymerase III as precursor transcripts that require extensive remodeling to become functional [1, 2]. These processing steps include the removal of 5’ leader and 3’ trailer sequences, intron splicing, and the addition of a CCA sequence to the 3’ end [2–4]. Additionally, tRNAs undergo a variety of chemical modifications that can impact their structure, stability, and efficiency [5–7]. tRNAs are among the most heavily modified RNA species, with over 100 known post-transcriptional modifications, such as methylation and pseudouridylation [5]. These modifications occur at specific positions and can profoundly influence tRNA stability, and decoding accuracy [8–12]. Recent studies have revealed that dysregulation of tRNA modifications, termed “tRNA modopathies”, is associated with various diseases including chemoresistant cancers, neurological disorders, and viral infections [5, 13]. Another key step in tRNA maturation is aminoacylation (charging), where each tRNA is linked to the correct amino acid by a specific aminoacyl-tRNA synthetase — a process that must be highly accurate to avoid mistakes in protein production [14, 15]. Aberrant regulation of tRNA charging has emerged as a critical factor in disease pathology, with disruptions in aminoacylation contributing to cancer progression, neurodegenerative disorders, mitochondrial dysfunction, and stress-induced translational reprogramming [5, 16]. Beyond their traditional role, tRNAs can also play diverse non-canonical roles in metabolism and cellular regulation, such as acting as essential cofactors in bacterial tetrapyrrole biosynthesis or as substrates for non-ribosomal peptide synthesis and aromatic amine biosynthesis [17–19]. Furthermore, tRNAs can also be cleaved into smaller fragments, especially under stress or in disease conditions. These tRNA fragments (tRFs) have been implicated in regulating gene expression, stress granule formation, and cell fate decisions [20, 21]. Together, these interconnected layers of regulation highlight tRNAs as dynamic molecules with roles that extend far beyond protein synthesis. Unraveling these complex regulatory layers requires precise and sensitive tools, and advances in sequencing technologies have been instrumental in enabling comprehensive analyses of tRNA expression, processing, modifications, and function. Here in this review, we aim to cover the recent development of sequencing technologies to reveal the different layers of tRNA regulation, including expression, modifications, charging and processing/fragmentation (Fig. 1). We highlight the latest applications of high-throughput tRNA sequencing, in particular Nanopore direct RNA sequencing, that reveal dynamic regulation of tRNA biology in diverse biological contexts including cancer, neuronal function, and the immune system. These insights pave the way for novel diagnostic and therapeutic strategies targeting tRNA biology.

Analytical framework for high-throughput tRNA sequencing
Despite being one of the most expressed RNA types, sequencing tRNAs poses several unique challenges due to their extensive chemical modifications, stable structure, and high sequence similarity among tRNA genes. tRNAs are the most modified RNAs in cells, with an average of 13 modifications per molecule [22]. These modifications frequently stall reverse transcriptase enzymes or lead to nucleotide misincorporation, resulting in truncated or error-prone cDNAs [23]. In addition, the strong secondary structure of tRNAs impedes enzymatic processing, and their short length makes adapter ligation difficult [24]. To overcome these challenges, several tRNA-specific next-generational sequencing methods have been developed that share a common general pipeline: pre-treatment of RNA, adapter ligation, cDNA synthesis and PCR amplification. Together, these allow for researchers to investigate global tRNA gene expression, base modifications, charging status, and tRNA processing and fragmentation (Fig. 2). The collection of high-throughput tRNA sequencing pipelines allows researchers to select one based on their specific emphasis to interrogate tRNA expression, base modifications, charging status and processing/fragmentation, starting RNA amount, and experimental time (Table 1). Here we provide a general framework that is commonly used by different pipelines, more technical details and comparisons can be found in recent reviews [24].
For measuring tRNA gene expression (Fig. 2A), nearly all methods (including ARM-seq, DM-tRNA-seq, YAMAT-seq, LOTTE-seq, Hydro-tRNAseq, QuantM-tRNA-seq, CHARGE-seq, AQRNA-seq, mim-tRNAseq, OTTR-seq, DAMM-seq, MSR-seq, tRNA structure-seq, ALL-tRNAseq, DORQ-seq, LIDAR, and Induro tRNA-seq) employ cDNA libraries combined with reverse transcription and PCR amplification [25, 27–29, 31–40, 42–44]. Adapter ligation strategies vary, with some methods relying on ligation of distinct 5′ and 3′ adapters, while others, like DM-tRNA-seq, use a single adapter followed by circularization [37]. The cDNA synthesis step is often carried out using reverse transcriptases such as SuperScript III or other M-MLV (Moloney Murine Leukemia Virus) reverse transcriptases, though more recent methods, like DM-tRNA-seq and Induro tRNA-seq, use TGIRT and Induro thermostable group II intron RTs (reverse transcriptases) with superior modification tolerance and processivity [37, 39]. PCR amplification follows, enabling the generation of sufficient material for Illumina-based sequencing. Spike-in controls, typically synthetic or in vitro transcribed tRNAs, are incorporated in many protocols to normalize expression and control for potential global tRNA changes [35]. For analyzing tRNA gene expression, several computational pipelines have been developed, but challenges remain in accurately quantifying expression levels. A critical step is reference curation, which often involves using custom tRNA annotations from tools like tRNAscan-SE 2.0 or GtRNAdb 2.0, as many standard genome annotations are incomplete or imprecise for tRNAs [45, 46]. Given sequence similarity between tRNA isodecoders and isotypes, multi-mapping reads are a major concern. Some pipelines, like mim-tRNAseq, discard multi-mapped reads [33], while others do not. Moreover, the mismatch tolerance must be carefully controlled to account for a moderate mismatch percentage. This can help retain reads that include biological variation and modification-induced misincorporations. However, too high a threshold can risk false alignments [24, 33, 47]. It is important to note that most tRNA-seq methods are designed to capture mature tRNAs, and due to sequence redundancy, a single mature tRNA can map to multiple genomic loci. To resolve the ambiguity inherent in tRNA-seq, integrative approaches using Pol III ChIP-seq and chromatin accessibility assays (e.g., ATAC-seq) have been employed. These studies reveal that only a subset of annotated tRNA genes are actively transcribed, with many loci being epigenetically silenced or differentially regulated in a tissue- and species-specific manner [48, 49].
To assess base modifications (Fig. 2B), several approaches incorporate pre-treatment steps that enzymatically remove modifications or exploit RT signature patterns. tRNAs are rich in diverse post‑transcriptional modifications such as pseudouridine (Ψ), dihydrouridine (D), wybutosine (yW), and various methylations (e.g., m1A, m1G, m2₂G, m5C) many of which cause common reverse transcriptases to stall or misincorporate nucleotides. The pre-treatment step is particularly crucial for addressing modification-induced RT failure. For example, Hydro-tRNAseq and ARM-seq employ AlkB, an E. coli demethylase that removes certain methyl modifications including m1A and m3C, thus improving read-through during reverse transcription [29, 31]. The comparison between treated and untreated samples then allows identification of modified positions based on changes in RT efficiency or termination. AQRNA-seq, DAMM-seq, mim-tRNAseq, and ALL-tRNAseq analyze misincorporation rates and drop-off signatures of the RT to infer the presence and location of modifications, while MSR-seq and DM-tRNA-seq combine demethylation with RT signature mapping to improve modification detection sensitivity and specificity [25, 27, 28, 33, 36]. To address this, many methods employ highly processive or engineered RTs, such as thermostable group‑II intron RTs like TGIRT or Induro, and retroviral variants like SuperScript IV, each chosen for their ability to read through structured or modified regions with different fidelity and temperature tolerances [25, 39, 42, 50]. However, these methods are limited to detecting certain types of modifications – primarily those that disrupt Watson–Crick base pairing, such as certain methylations – while other modifications that do not alter RT behavior require specialized protocols (discussed below) or orthogonal detection strategies for reliable identification. Annotation and interpretation of modification sites are facilitated by resources like MODOMICS and tModBase, which are databases for known tRNA modifications [51, 52]. However, these databases are not yet complete, and next-generation sequencing–based approaches provide only indirect evidence of modifications, typically inferred from reverse transcriptase stops or misincorporation signatures. Importantly, NGS cannot reveal the chemical identity of a modification, which still requires orthogonal validation by mass spectrometry or other biochemical methods.
Several modification-specific tRNA sequencing methods have been developed to map individual chemical modifications at single-nucleotide resolution. Notably, Pseudo-seq employs CMC (N-Cyclohexyl-N′-(2-morpholinoethyl)-carbodiimide methyl-p-toluenesulfonate)-mediated detection to identify pseudouridine sites via RT stops [53]. m1A-seq uses a simple AlkB treatment to demethylate m1A, which normally blocks RT, and relies on an untreated control comparison [54]. The AlkAniline-Seq method uses alkaline hydrolysis and aniline cleavage to profile both 7-methylguanosine (m7G) and 3-methylcytosine (m3C) in tRNAs and other RNAs [55]. Building on this, TRAC-Seq applies AlkB demethylation, sodium borohydride reduction, and aniline cleavage to achieve mapping of m7G specifically across tRNAs [56]. For m3C, HAC-Seq provides transcriptome-wide detection via hydrazine and aniline cleavage [57]. In D-seq, dihydrouridine residues are treated with sodium borohydride, converting them into tetrahydrouridine, which specifically impedes reverse transcription [58]. Most recently, PAQS-Seq uses periodate treatment to cause deletion signatures corresponding to queuosine modifications [59]. These modification-specific methods are important addition to the generic tRNA sequencing approach, since some modification types (e.g. Ψ, m7G, D, Q) do not give rise to RT signatures without additional chemical conversion.
Charging status (Fig. 2C), which reflects whether a tRNA is aminoacylated or not, can be assessed with various methods. mim-tRNA-seq, CHARGE-seq, Charged DM-tRNA-seq and MSR-seq are methods to specifically target this question. These methods differentiate charged from uncharged tRNAs based on their susceptibility to periodate oxidation and β-elimination, which only cleaves uncharged tRNAs at the 3’ end [25, 30, 33, 35]. Notably, this analysis can only reveal whether a tRNA is charged or not, but can’t distinguish mischarging from correct charging event. In these approaches, charging fraction is obtained by counting the fraction of CCA-ending reads versus CCA/CC-ending reads.
Given the rising interests in tRNA processing and fragmentation, several methods also capture tRNA processing and fragmentation patterns (Fig. 2D), important for studying tRFs and maturation dynamics. ARM-seq, AQRNA-seq, Hydro-tRNA-seq, OTTR-seq, MSR-seq, ALL-tRNAseq, and LIDAR, retain information on cleavage sites, adapter-ligation biases, or strandedness, which can be used to map processing intermediates or tRNA fragments [25, 27–29, 31, 32, 34, 39]. Notably, LIDAR and OTTR-seq use a template-switching method and do not rely on adapter ligation techniques, allowing for unbiased capture of tRNA fragments regardless of terminal modifications [32, 34]. This enables detection of fragments with blocked or chemically altered 3′ ends that are typically missed by ligation-dependent protocols, while also preserving strandedness and full-length sequence information. tRFs are often classified into types such as tRF-5, tRF-3, tRF-1 and i-tRF based on their position relative to the parent tRNA [60]. For example, tRF-1 s represent poly-U ending trailer sequences that are cleaved by RNase Z during maturation from pre-tRNAs [60]. However, naming conventions vary, and standardization remains an issue [61]. Computational tools like MINTmap, tRFexplorer, tRAX, and unitas have been developed to systematically identify and classify tRFs, often using curated tRNA references and specific alignment strategies that account for the small size and potential for post-transcriptional modifications [62–65]. These pipelines often allow for multi-mapping and tolerate certain mismatches to accommodate biological variation, but careful parameter selection is essential to ensure biologically meaningful interpretation.
Unlike traditional cDNA-based methods above, Nano-tRNA-seq (Fig. 2E) utilizes Oxford Nanopore direct RNA sequencing to analyze full-length, native tRNA molecules without cDNA amplification [41, 66]. Direct RNA sequencing works by passing individual RNA molecules through a nanopore, where changes in ionic current are measured to identify each nucleotide in real time without reverse transcription or amplification [66]. This approach circumvents many of the challenges posed by chemical modifications and strong secondary structures by directly threading intact RNAs through a nanopore, enabling accurate measurement of tRNA gene expression and potential to measure processing or fragmentation based on read length and sequence coverage. Importantly, base modifications are preserved in the native RNA and inferred through altered ionic current signatures or systematic base-calling errors, offering a powerful strategy for modification mapping across the entire tRNA body [26, 41, 67, 68]. For instance, Nano-tRNA-seq applied to S. cerevisiae detects coordinated T-loop modifications (including Ψ55, m1A58. and m5U54) as reproducible base-calling error and ionic current signatures, and these signals are notably absent or altered in corresponding writer knockout strains (e.g., PUS4, TRM2, TRM6/TRM61 deletions), confirming the specificity of modification detection by comparing wild-type versus mutant backgrounds [41, 69]. Computational tools like Tombo, EpiNano, MoDorado, and Nanocompore analyze these deviations to infer modification sites, although accurate identification remains challenging without matched controls [68, 70–72]. Meanwhile, aa‑tRNA‑seq extends this strategy by chemically ligating charged tRNAs via their esterified amino acids to sequencing adapters [26]. Then, during nanopore translocation, the amino acid creates distinct ionic current perturbations that, when analyzed with signal‑based machine learning models, enable single‑molecule classification of charging status and even amino acid identity (Fig. 2E). However, current challenges of direct tRNA sequencing include relatively high error rates, limited read depth, high amounts of input RNA (~ 250 ng of tRNA enriched from total RNA), and difficulty in confidently identifying specific modifications due to overlapping signal patterns [73]. We expect the usage of Nano-tRNA-seq will continue to increase, while for samples with limited quantity, Illumina-based approach will remain popular.

tRNA sequencing facilitates novel discovery of key translation regulators in human biology
Having outlined the diverse toolkit of tRNA-seq methods, we will now explore how dysregulation of tRNAs contributes to disease, such as cancer, neurological disorders, and viral infections. A unifying theme across these studies is translation reprogramming, whereby aberrant tRNA expression, charging, processing, or modification reshapes the translational landscape. tRNA sequencing has been instrumental to pave a road that allows these studies to occur. Table 2 summarizes what novel tRNA biology was illuminated by each tRNA sequencing method (not limited to human biology). In these examples below, tRNA-seq provides important mechanistic insight to pinpoint which specific aspect of tRNA dysregulation contributes to translation reprogramming, with a particular focus on human biology.

(1) tRNA-enhanced translation programs in cancer biology
Recent studies have underscored the critical role of tRNA modifications in modulating cellular functions, particularly in oncogenesis. A common theme here is the translation reprogramming of codon-enriched transcripts due to altered modification on specific tRNAs (Fig. 3A). Among these, the N7-methylguanosine (m7G) modification on tRNAs has emerged as a key regulatory mark. METTL1, the methyltransferase catalyzing m7G formation, is implicated in various cancers, including glioblastoma, sarcoma, and acute myeloid leukemia [74]. METTL1 depletion reduces m7G modification levels on tRNAs, resulting in cell cycle perturbations. Notably, METTL1 loss attenuates the proliferative capacity and colony formation in cholangiocarcinoma cell lines [75]. In contrast, METTL1 amplification is frequently observed across cancers and promotes oncogenic transformation via m7G -modified tRNAs (particularly Arg-TCT-4–1), as revealed by TRAC-Seq. This technique pinpointed specific m7G sites and enabled a quantitative profiling of modified tRNAs, demonstrating that METTL1 enhances the translation efficiency of mRNAs that regulate cell cycle progression [74, 76]. Moreover, in lung cancer contexts, TRAC-Seq data correlate METTL1 expression with selective translation of mRNAs containing m7G-modified codons, linking modification status directly to translational output and oncogenic progression [76]. These studies suggest that METTL1 reshapes the cellular translatome by modifying specific tRNAs with m7G, therefore stabilizing those tRNAs and biasing translation toward growth-promoting proteins.

The TRMT6/61A complex installs N1-methyladenosine (m1A) on a subset of cytosolic tRNAs. This m1A mark also regulates translation of key oncogenes. In T cells, TRMT6/61A is strongly induced upon activation and confer m1A at the 58 position on a pool of “early-wave” tRNAs needed for rapid protein synthesis, including MYC [54]. MYC is a master regulator of proliferation, apoptosis, and cellular transformation [77]. To map this modification landscape, researchers have applied a quantitative m1A-seq approach [54]. In mice, the conditional deletion of Trmt61a in T cells abolishes MYC protein accumulation, causing cell cycle arrest and blocking T cell expansion. By extension, elevated TRMT6/TRMT61A in T-cell leukemias or lymphomas could contribute to MYC-driven growth, as shown in normal T cells [54]. Additionally, in hepatocellular carcinoma (HCC), a similar outcome of m1A has been reported using m1A-seq. In HCC tumors, global tRNA m1A levels are strikingly high and correlate with poor patient prognosis [78]. TRMT6/61A itself is also overexpressed in advanced HCC, and functional experiments show that TRMT6/61A mediated m1A is required for liver tumorigenesis [78]. It has been shown that these enzymes methylate specific tRNAs to enhance translation of metabolic regulators, like PPARD mRNA. PPARδ protein drives cholesterol synthesis and activates Hedgehog signaling in liver cancer stem cells which can sustain HCC self-renewal and growth [78]. Notably, inhibiting TRMT6/61A or preventing tRNA m1A deposition suppresses tumorigenesis, underscoring that m1A-seq not only uncovered this dysregulation but also pinpointed a modification-dependent translational program [78]. Together, these findings highlight the critical role of TRMT6/61A-mediated m1A in supporting cancer cell proliferation across multiple tumor types and highlight its value as a potential therapeutic target.
N3-methylcytidine (m3C) modifications occur at position 32 in the anticodon loop of certain tRNAs, installed by METTL2A, METTL2B and METTL6 [79]. Recent work has shown that these m3C marks are essential for decoding specific codons and for cell-cycle/DNA-damage programs. Using HAC-seq (a method to comprehensively profile m3C across the transcriptome) after CRISPR knockout of the three writer enzymes, a study found that loss of METTL2A/2B/6 greatly reduces m3C on tRNA-Ser-GCT, and it becomes a poor decoder [80]. Because of the lack of modification in the anticodon loop, ribosomes stall at the serine AGU codons and translation efficiency drops [80]. This codon specific deficit disproportionately affects mRNAs involved in cell-cycle control and DNA repair. In cells with complete knockout, translation of many DNA damage response genes is impaired, leading to slow cell growth and S-phase delay [80].
To summarize, growing evidence shows that cancer cells exploit tRNA epitranscriptomic changes to favor oncogenic translation. m7G, m1A and m3C tRNA modifications are frequently dysregulated in tumors, with each pathway enhancing translation of specific growth-related transcripts. Importantly, these studies highlight how new sequencing-based approaches have been pivotal in revealing the codon- and transcript-specific consequences of these modifications, uncovering selective translational programs that drive malignancy.

(2) tRNA-reduced translation programs in neuronal disorders
tRNA-reduced translation programs in neuronal disorders: tRNAs are also emerging as dynamic regulators of neuronal function. One key modification is N5-methylcytidine (m5C), catalyzed by NSUN2. To study this, researchers applied YAMAT-seq, a specialized high-throughput sequencing approach optimized for mature tRNAs. YAMAT-seq revealed that conditional knockout of Nsun2 causes a dramatic drop in tRNA m5C and a depletion of most tRNAGly species [81]. This leads to widespread proteome changes where glycine-rich synaptic proteins are lost and glutamatergic signaling is impaired. These molecular effects correlated with behavioral changes including decreased fear conditioning and impaired contextual memory. Additionally, there was reduced depression-like behavior in specific tests, while overexpression of Nsun2 correlates with increased depressive-like behaviors and preserved fear memory [81]. In a later study, it was shown that increased levels of tRNAGly were observed in individuals who had died by suicide [82]. Thus, the findings enabled by tRNA sequencing illustrate that while tRNA dynamics play a role in regulating glycine tRNA levels and consequently affect synaptic protein synthesis, these relationships are not straightforward and can have nuanced impacts on neurological functions like mood and cognition.
Another critical example of tRNA modification impacting neuronal function involves TRMT1, the enzyme responsible for installing N2, N2-dimethylguanosine (m22G) at position 26 in specific tRNAs. A homozygous missense mutation in TRMT1 was identified in an individual presenting with intellectual disability, developmental delay, and epilepsy [83]. To characterize the molecular impact, researchers used a m22G RT readthrough analysis, which enabled detection of this modification based on altered RT signatures. This approach revealed that this mutation resulted in loss of function and a global loss of m22G modifications in patient derived cells. Although the exact downstream pathways remain to be understood, the loss of m22G is believed to destabilize tRNA structure and compromise translational fidelity [83, 84]. This can cause potential affects in protein synthesis required for neuronal function. Using Nano-tRNA-seq, another study found TRMT1L (a paralog of TRMT1) is responsible for catalyzing m22G on position 27 in tyrosine tRNA [85]. Building on this, recent studies in human cells further clarified the division of labor between TRMT1 and TRMT1L paralogs. TRMT1 broadly modifies tRNAs carrying guanosine at position 26, whereas TRMT1L specifically catalyzes m22G at position 27 of tyrosine tRNAs [86]. Nano-tRNA-seq provided the resolution needed to capture these modification events, showing that tyrosine and serine tRNAs are particularly dependent on these marks for stability and efficient translation [86]. From these studies, it was suggested that the hypomodified tRNAs contributed to dysregulated translation programs during neuronal development (Fig. 3A). These studies highlight how applying modification-sensitive sequencing strategies can link a single tRNA modification defect to clinically observable neurodevelopmental phenotypes.

(3) tRNA-induced frameshifting in host immune responsse
Viral infection can dramatically reshape the tRNAome. For example, herpes simplex virus 1 (HSV-1) infection causes a massive surge in host tRNA transcription. DM-tRNA-Seq showed that the total amount of nuclear-encoded tRNA increased two-fold after HSV-1 infection [87]. This upregulation occurs despite overall host transcriptional shutoff – HSV-1 seems to specifically activate RNA Polymerase III at tRNA genes, boosting the pool of tRNAs available for translation. Such dynamic remodeling of tRNA abundance likely helps the virus commandeer the host translation machinery while host mRNAs are degraded. 
Immune cell activation also involves tRNA reprogramming (Fig. 3B). During mouse CD4 + T cell activation, both tRNA abundance and tRNA modifications change over time. Notably, two specialized tRNA modifications, wybutosine and 2-methylthio-N6-threonylcarbamoyladenosine (ms2t6a) are downregulated as T cells exit quiescence, as detected by DM-tRNA-seq [88]. These modifications normally stabilize codon-anticodon pairing (particularly on “slippery” codons prone to frameshifting), so their loss increases −1 translational frameshifting [88]. Similar phenomenon has also been recorded in yeast and E. coli, where changes in tRNA modification cause + 1 frameshifting [89, 90]. This suggests that proliferating T cells may sacrifice some translational fidelity to speed up protein synthesis, which could influence viral tropism. Another paper also reported that loss of wybutosine in cancer cells, via TYW2 knockout, induces −1 ribosomal frameshifting at UUU codons, generating out-of-frame neoantigens that enhance tumor immunogenicity and sensitivity to checkpoint blockade [91].
Together, these studies demonstrate how tRNA sequencing provides a powerful lens into dynamic changes in both tRNA abundance and modification states during stress, infection, and immune activation. By revealing how viruses and immune cells reprogram the tRNAome to favor their respective translational needs, these findings highlight tRNAs as active players in host–pathogen interactions and cellular adaptation. This emerging view sets the stage for exploring tRNAs as potential therapeutic targets in infection and immunity.

(4) tRNA-derived fragments in stress response
tRNA-derived fragments (tRFs) have been shown as abundant and versatile regulators of gene expression that operate across diverse biological contexts. Methods such as LIDAR and OTTR-seq (as well as others mentioned above) can be used to detect and highlight tRFs under different contexts. tRFs can act through multiple molecular mechanisms more broadly relevant to stress adaptation and cellular regulation (Fig. 3C). Firstly, tRFs can function in a micro-RNA like fashion, guiding silencing of target transcripts or, in some cases, enhancing translation through sequence complementarity [92, 93]. Additionally, tRFs can modulate the activity of RNA-binding proteins, many of which are translation factors, thereby generally repressing protein synthesis under stress [94, 95]. Finally, tRFs can also form distinct structural conformations that drive their activity, such as nicked forms that alter ribosome dynamics [96], or G-quadruplex assemblies that influence RNA stability and translation [97].

tRFs have also emerged as key carriers of epigenetic information in the context of transgenerational inheritance. In sperm, tRFs comprise a major component of the small RNA transcripts delivered at fertilization [98], as OTTR sequencing in mouse sperm has shown a tRF complex that was previously underappreciated. Both the 5’ and 3’ halves from the majority of tRNAs are present, including many fragments with unusual 3’ end chemistry that evade standard protocols. This hidden diversity suggests that sperm carry a rich repertoire of tRF signals into the zygote, potentially influencing early embryonic gene regulation. Importantly, paternal environment can shape sperm tRF profiles. Factors like diet, stress, and physical activity are reflected on sperm RNA populations, providing a conduit for epigenetic inheritance, influencing early embryogenesis and offspring physiology [99]. Another study demonstrated that sperm acquire tRFs during epididymal maturation through vesicular transfer, and using small RNA-seq were able to identify that specific tRFs (such as tRF-Gly-GCC) can regulate endogenous retroelements in the early embryo [100]. Complementary experiments show that injection of sperm tRFs from high-fat diet fathers into normal zygotes is sufficient to induce glucose intolerance and metabolic defects in the offspring [101]. Emerging evidence suggests that such tRF-mediated epigenetic effects may impact metabolic pathways, stress responses, and even behavioral traits in progeny, highlighting a crucial role for sperm RNA beyond genetic information in heredity and development [100].

(5) tRNAs and tRFs as promising pathology and biomarker applications
A recent innovation, low-input methylation sequencing (LIME-seq), enable the profiling of modification signatures in circulating cell-free RNA (cfRNA) from plasma, capturing a diverse array of tRNAs and small non-coding RNAs of both human and microbiome origin [102]. Remarkably, LIME-seq revelated the microbiome-derived cfRNA, rather than host RNA, harbors distinctive RNA modification patterns that sensitively reflect host-microbiota dynamics and show strong potential for early detection of colorectal cancer [102]. Complementing these liquid-biopsy approaches, the newly developed Patho-DBiT platform brings spatial resolution to archival formalin-fixed, paraffin-embedded (FFPE) tissues by enabling in situ polyadenylation and barcoding of RNAs. This makes it possible to capture and map short RNAs such as tRFs and non-polyadenylated RNAs like tRNA. Since tRNAs and tRFs are increasingly recognized as biomarkers of cancer progression, stress responses, and therapeutic resistance [5], Patho-DBiT provides a powerful framework to not only detect them in difficult samples but also to place them in a spatial and histopathological context, greatly expanding their potential clinical utility. These additions highlights how tRNA-seq methods can be used to serve in biomarker applications.

Outlook and challenges
Building on the biomedical insights gained from understanding tRNA’s role in health and disease, it is essential to consider the technological advances that are driving this field forward. Among these, Nanopore sequencing stands out as a promising platform that enables direct RNA analysis, offering new opportunities for more comprehensive and precise characterization of tRNAs and their modifications. Particularly, its ability to directly measure RNA rather than cDNA represents a transformative advance that is likely to drive broader adoption of this method. Recent developments in machine learning approaches, particularly those trained on synthetic RNA standards, have enhanced sensitivity of specificity of modification detection [103, 104]. While these methods have shown promise for various RNA species, their application to tRNAs remains to be thoroughly explored.
Despite the advantages of Nanopore sequencing, Illumina-based short-read sequencing remains the preferred method for studies involving limited or precious biological samples, such as patient-derived materials, due to its higher throughput and cost-effectiveness [105, 106]. Additionally, it is important to note that LC–MS/MS remains the only method that directly determines the chemical structures of RNA modifications, rather than inferring them from sequencing. An emerging advancement is to apply LC–MS/MS on full-length or partial tRNA sequencing, even in complex mixtures, making it a critical orthogonal approach [107–110]. However, significant knowledge gaps persist in the field of tRNA sequencing and modification analysis. Large-scale genomics datasets frequently omit tRNA information, limiting insights into tRNA biology at a systems level [111, 112]. Validation of modification calls is often challenging, owing to the complexity of tRNA structures and modifications. Moreover, sequencing data on precursor tRNAs (pre-tRNAs) are scarce, hindering the understanding of tRNA maturation dynamics. Finally, single-cell resolution approaches for RNA modification mapping have yet to be developed, restricting our ability to study tRNA modifications in heterogeneous cell populations. Addressing these challenges will be critical for advancing our understanding of tRNA function and regulation through direct RNA sequencing technologies.