Metabarcoding Characterization of Fungal Communities in Spanish Cereals with a Special Focus on Fusarium Species

Abstract

Cereal safety is a critical public health concern, as crops are frequently compromised by mycotoxin-producing fungi, particularly those from the genus Fusarium. Although metabarcoding is a powerful tool for characterizing these fungal communities, the universal internal transcribed spacer 2 (ITS2) marker lacks the resolution required to distinguish closely related species with distinct toxigenic profiles. The objective of this work was to accurately characterize the mycobiota and specific Fusarium community structure in Spanish cereal field samples. We propose a high-resolution metabarcoding workflow within the QIIME 2 environment to optimize Fusarium identification using the translation elongation factor 1-α (TEF1) gene. We developed a TEF1 Naive Bayes classifier trained on a curated database derived from FUSARIUM-ID v.3.0, expanded with eukaryotic sequences to prevent false-positive assignments. This approach was validated through direct comparison with ITS2 in the same samples. As expected, ITS2 significantly underestimates Fusarium diversity, whereas TEF1 enables precise species-level resolution within this genus. General mycobiota analysis performed using ITS2 revealed that fungal communities are shaped primarily by geographical location rather than host cereal species. Crucially, our approach confirmed the persistence of key toxigenic species, such as F. langsethiae and F. graminearum, and revealed broader diversity through the consistent detection of species often overlooked by traditional methods, including F. equiseti, F. acuminatum, and F. culmorum. We conclude that our metabarcoding framework reveals a high Fusarium diversity in Spanish cereal grains, and this knowledge is essential for designing targeted strategies to predict and mitigate mycotoxin contamination in these crops.

Graphical Abstract

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1007/s00248-026-02833-z.

Introduction

Cereals are the foundation of human and animal diets worldwide, and the global demand for these crops is expected to increase substantially. Currently, external pressures such as geopolitical conflicts and extreme weather events pose a significant threat to universal access to grains [1]. This vulnerability is already evident; for instance, cereal yields in the European Union decreased by 5.1% by 2024, largely because of extreme temperatures and heavy rainfall [2]. In addition to these supply limitations, the safety of the available grain remains a critical concern. Specifically, cereal production is frequently compromised by mycotoxin contamination [3]. Mycotoxins are low-molecular-weight, stable, and toxic secondary metabolites produced by filamentous fungi, which belong primarily to the genera Aspergillus, Fusarium and Penicillium [4]. Chronic dietary exposure to these compounds is linked to carcinogenic, mutagenic, and teratogenic effects [5]. Consequently, the contamination of cereals constitutes a major public health challenge, particularly since these crops are the primary source of human exposure to such toxic metabolites [6–9].

Cereals and their processed food products are frequently contaminated by a broad spectrum of mycotoxins, including aflatoxins, ochratoxin A, citrinin and patulin [10–12]. Notably, a significant proportion of the mycotoxins found in cereals are produced by the genus Fusarium [13]. Within this group, toxigenic potential is highly species dependent. Type A trichothecenes (T-2 and HT-2) are produced primarily by F. langsethiae and F. sporotrichioides, whereas species like F. equiseti and F. acuminatum are considered minor producers [14, 15]. Conversely, type B trichothecenes such as deoxynivalenol (DON) and nivalenol (NIV) are predominantly synthesized by F. culmorum, F. graminearum, with F. poae also being a notable producer of NIV [16, 17]. The former species (F. culmorum and F. graminearum), along with F. cerealis, are also the main producers of zearalenone (ZEA) [18]. Finally, fumonisin production is attributed primarily to specific members of the Fusarium fujikuroi species complex (FFSC), including F. verticillioides and F. proliferatum [19].

The current global climate change scenario creates ideal conditions for the emergence of mycotoxin-producing fungi, including specific Fusarium species that were previously considered rare or absent in cereals. In this context, obtaining an accurate picture of the crop contamination profile is critical [20–22]. To address this challenge, metabarcoding (also known as metataxonomics), which is based on the high-throughput sequencing of taxonomic marker genes directly from environmental DNA, is among the most robust approaches. Unlike traditional techniques, this method allows for the study of complex fungal communities, including groups that are difficult to culture [23]. For example, the application of a metabarcoding approach recently enabled the first detection of the trichothecene producer F. langsethiae in Spanish oats, providing evidence of a geographical shift in Fusarium species likely driven by climate change [21].

The internal transcribed spacer 2 (ITS2) region remains the most widely used marker for fungal metabarcoding. However, its limited interspecific variability limits the discrimination of closely related species with distinct mycotoxigenic profiles, which is a particularly critical issue for the genus Fusarium [24]. In addition, many members of this genus possess nonorthologous copies of the ITS2 region [25]. Given these limitations, the use of an alternative target becomes necessary to ensure accurate identification of Fusarium species. In this context, the translation elongation factor 1-α (TEF1) gene has proven to be efficient for distinguishing closely related species within this group [26, 27]. Moreover, TEF1 is a single-copy gene in Fusarium [24], avoiding the issues associated with intragenomic heterogeneity.

Metabarcoding approaches to identify Fusarium species with TEF1 were carried out in previous studies [20, 25, 28, 29]. However, these methods are either (i) reliant on lower-resolution clustering methods (OTUs) instead of the amplicon sequence variant (ASV) standard, (ii) dependent on basic homology search tools (BLAST) for taxonomic classification, or (iii) limited by the use of TEF1 databases restricted solely to Fusarium or fungal sequences. In addition, amplicons generated by most of these works are either too long to be joined into a single sequence from paired-end reads or include significant proportions of the exon sequence, which is significantly less variable [29]. The primary objective of this work was to characterize the grain mycobiota and specific Fusarium community structure across cereal-growing regions in Spain. To achieve this, we build upon previous approaches and propose a comprehensive optimization of Fusarium identification with TEF1-based metabarcoding. A bioinformatics workflow was implemented using the QIIME 2 platform [30], a current state-of-the-art platform for microbiome analysis. Unlike most prior methods that rely on clustering OTUs or BLAST-based taxonomic classification, we employed a Naive Bayes classifier to assign taxonomy to ASVs. To train this classifier, we constructed a curated TEF1 database derived from the updated FUSARIUM-ID v.3.0 [24], expanding it to include non-Fusarium and non-fungal eukaryotic TEF1 sequences to prevent false-positive assignments. Finally, this TEF1 approach was complemented by the universal ITS2 marker to provide a comprehensive characterization of the general cereal grain mycobiota. This allowed us to assess fungal diversity and elucidate the environmental drivers of Fusarium communities in Spanish cereal grains.

Methods

Sample Collection and DNA Sequencing

In 2023, a total of 30 samples (approx. 3 kg) of small-grain cereals (12 wheat samples, 11 barley samples, and 7 oat samples) grown in fields were collected from grain stores located in 11 Spanish cereal-producing regions. The sampling locations are provided in Figure S1.

First, the grains were milled using an IKA A11 Basic Mill (IKA, Staufen, Germany). DNA extraction was conducted using the DNeasy Plant Pro Kit (QIAgen, Düsseldorf, Germany), starting from 100 mg of grain flour. For each sample, six DNA extractions were conducted. The extracted DNA was subsequently mixed in pairs to yield a total of three biological replicates. They were then concentrated to a volume of 50 µL using a vacuum concentrator (Concentrator plus, Eppendorf, Hamburg, Germany).

The libraries were amplified using ITS3/ITS4 primers (GCATCGATGAAGAACGCAGC/TCCTCCGCTTATTGATATGC) [31] for the analysis of internal transcribed spacer 2 (ITS2) and custom primers EFseqF350/EFseqR (CGAACCATCGAGAAGTTCGAGA/GGTGATACCACGCTCACGCT) for the translation elongation factor 1 alpha (TEF1) gene. TEF1 primers were designed to amplify the highly variable second intron of this gene on the basis of sequence alignments of this region [32]. The dataset included sequences of more than 150 Fusarium spp. isolates from different origins obtained in several works carried out in our laboratory or retrieved from databases. Using the alignments of the complete dataset generated by the UGENE programme (Unipro, Novosibirsk, Russia) via the MUSCLE algorithm [33], we identified identical conserved regions to design primers that would enable amplification of the target region in all Fusarium species. The amplification fragment generated in all cases would approximately be 350 bp long. Although the TEF1 primers target the entire Fusarium genus, the potential amplification of non-specific fungal taxa is resolved during downstream bioinformatic processing (see Methods 2.3.2). PCR products of the proper size were selected by 2% agarose gel electrophoresis. Equal amounts of PCR products from each sample were pooled, and sequencing libraries were generated using the NEBNext Ultra II DNA Library Prep Kit (New England Biolabs, USA). Libraries were sequenced on the Illumina NovaSeq 6000 platform to generate 250 bp paired-end raw reads.

Development of a TEF1 Naive Bayes Taxonomic Classifier

FUSARIUM-ID v3.0 [24], a FASTA-formatted database that includes 1,107 curated TEF1 sequences representative of the genus Fusarium and other members of the Nectriaceae family, was retrieved from the FUSARIUM-ID GitHub repository (https://github.com/fusariumid/fusariumid). The FASTA file was subsequently parsed to create a taxonomy file formatted for compatibility with the QIIME 2 environment. Both the taxonomy file and the TEF1 sequences were then imported into QIIME 2 (v2024.2.0). Additionally, 93,977 TEF1 sequences from non-Fusarium fungi and other eukaryotes were sourced from NCBI GenBank [34, 35] using a modified query based on the work of Boutigny et al. [28]. These sequences are employed to provide the classifier with TEF1 sequences that are either non-Fusarium or nonfungal, thus reducing the probability of false positives in taxonomic classification (i.e., incorrectly assigning Fusarium to a sequence that does not belong to this genus).

Low-quality NCBI GenBank sequences were removed, and the remaining sequences were dereplicated using the QIIME 2 RESCRIPt plugin [36]. The sequences from NCBI GenBank were combined with the sequences from FUSARIUM-ID into a single sequence file. To develop an amplicon-specific classifier, the amplicon region was extracted from the sequences using the TEF1 PCR primers specified in Sect. 2.1. After another round of dereplication, a total of 6,327 sequences were preserved. Finally, a Naive Bayes classifier was trained via the q2-feature-classifier plugin [37] through the fit-classifier-naive-bayes method.

Metabarcoding Analysis of ITS2 and TEF1 Sequences

Quality Control

A preliminary evaluation of the quality of the demultiplexed FASTQ files was performed using FastQC (v0.12.1) [38]. The primer sequences were subsequently trimmed using Cutadapt (v4.6) [39]. After the trimming process, another quality assessment was carried out using FastQC. MultiQC (v1.21) was employed to compile the outputs from FastQC and Cutadapt [40]. The sequences were imported into QIIME 2 (v2024.2.0) [30] for downstream analysis.

ASV Count

Reads were denoised through the DADA2 algorithm (version 1.26.0) [41] using the q2-dada2 plugin. Both forward and reverse reads were truncated to a length of 220 nucleotides. The amplicon sequence variants (ASVs) derived from the resultant feature tables were taxonomically annotated using Naive Bayes machine-learning classifiers using the q2-feature-classifier plugin [37]. The classifiers were trained on the UNITE v10 (04.04.2024) 99% identity, all eukaryotes, and singletons as in the RefS database for ITS2 [42] and against the custom database for TEF1 (see Sect. 2.2). A confidence threshold of 0.7 was established to mitigate taxonomic inaccuracies. The feature tables were subsequently filtered to remove any ASVs that lacked phylum-level annotation or were not classified as Fungi (ITS2) or Fusarium (TEF1).

Diversity Analysis

The Chao1 [43], Shannon [44], and inverse Simpson [45] indices were used to evaluate within-sample (alpha) diversity. To evaluate between-sample (beta) diversity, we employed robust Aitchison principal component analysis (rPCA), an ordination technique designed for microbiome data that combines sample and ASV information into a biplot [46–48]. Diversity metrics were calculated using the q2-diversity (alpha indices) and Gemelli (rPCA) QIIME 2 plugins. Alpha diversity metrics were computed on a feature table rarefied to 7,000 (ITS) and 500 (TEF1) sequences per sample. These thresholds were selected on the basis of the plateau observed in the rarefaction curves (Figure S2). The rPCA was computed on the unrarefied table. Notably, the variance percentages displayed on the rPCA axes represent the proportion of variance explained by the underlying low-rank matrix approximation rather than the total variance of the original dataset. The Gemelli QIIME 2 plugin utilizes matrix completion to address the inherent sparsity of microbiome data, focusing on a reduced-rank model [48]. Specifically, since we utilized the default rank of 3, the three resulting principal components collectively explain 100% of the variance captured by the reconstructed low-rank model. Consequently, these values appear markedly higher than the variance proportions typically reported in standard PCA or PCoA, which are calculated on the basis of the total original variance.

Differential Abundance Analysis

Differential abundance analysis was performed using the Analysis of Compositions of Microbiomes with Bias Correction (ANCOM-BC) method [49] via the q2-composition plugin. ANCOM-BC adjusts for compositional effects and accounts for sampling variability in microbiome data by incorporating a sample-specific bias-correction factor. The following comparisons were tested: barley versus wheat, oat versus wheat, and oat versus barley.

Statistical Analysis

Comparisons of alpha diversity values were performed using the nonparametric Kruskal-Wallis test [50]. P values were adjusted for multiple testing using the Benjamini & Hochberg method [51] to control the false discovery rate (FDR). For beta diversity, the Adonis implementation of permutational multivariate analysis of variance (PERMANOVA) was conducted via the q2-diversity plugin to test the effects of cereal type and sample location [52, 53]. Permutational analysis of multivariate dispersion (PERMDISP) [54] was performed to assess the homogeneity of group dispersions to determine whether the observed group differences were because of variations in dispersion or centroid differences [55]. Both Adonis and PERMDISP were run on the basis of 999 permutations. Convex hulls and group centroids were included in the rPCA biplot to visually assess differences between the groups. Differential abundance was assessed using a threshold of log₂ fold change (|log2FC| ≥ 2). The significance threshold for all the statistical analyses was set at P ≤ 0.05.

Computational Details

This work was performed using computational resources (Brigit HPC server) provided by the Complutense University of Madrid. Files in the QIIME 2 format were imported into R (v4.2.3) [56] with the R package qiime2R (v0.99.6) [57]. Plots were created with the R packages ggplot2 (v3.5.0) [58], ggpubr (v0.6.0) [59], and EnhancedVolcano (v1.16.0) [60]. The sample location map was plotted using the R packages sf (v1.0.6) [61] and mapSpain (v0.9.1) [62]. The R package patchwork (v1.2.0) was used to assemble the plots [63]. The computational method parameters that were not explicitly stated were maintained at their default values.

Results

In this study, 90 environmental DNA extractions were performed from cereals (wheat, barley, and oat), and two marker genes were subsequently amplified and sequenced, ITS2, to profile the inherent fungal communities, and TEF1, to achieve species-level taxonomic assignment of Fusarium spp. Next, we constructed ASVs and assigned a taxonomy to them. For TEF1 taxonomic assignment, we constructed a Naive Bayes classifier using a database of curated Fusarium sequences and other eukaryotic sequences retrieved from NCBI GenBank. Finally, we measured sample diversity and compared it across the different cereal types. See Methods for details of the analysis.

Taxonomic Classification of ITS2 ASVs

The results of the taxonomic classification of ITS2 ASVs at the genus level, represented in terms of relative abundance, are shown in Fig. 1. Overall, the presence or absence of fungal genera was consistent across the sample replicates, with variations mostly observed among samples. Cladosporium and Alternaria were the most abundant genera in each sample. In terms of cereal grouping, Cladosporium exhibited higher relative abundances in oat samples, whereas Alternaria showed a similar trend in wheat samples. In addition to Alternaria, other genera with toxigenic potential were present in almost all the samples. However, whereas we identified Fusarium, Aspergillus, and Penicillium ASVs, none of these could be resolved to specific toxigenic species using the ITS2 marker.

Fig. 1.

Overview of the taxonomic classification of the ITS2 ASVs. Bar plots of the relative abundance of fungal genera, grouped by cereal and sample replicate. “Other genera” aggregates the remaining taxa with lower relative abundance grouped for legibility. “Unassigned” represents ASVs that could not be classified at the genus level

Fusarium Identification with TEF1

Although the custom TEF1 primers also amplified non-Fusarium genera, these non-target sequences were filtered out prior to downstream analysis (see Methods Sect. 2.3.2). The relative abundance (within ASVs classified as Fusarium) of Fusarium species detected by ITS2 and TEF1 sequencing, aggregated by cereal and species complex, is shown in Fig. 2. Overall, TEF1 enabled the identification of a diverse array of Fusarium species. The relative abundances of F. clavum, F. langsethiae, and F. acuminatum were high in the barley samples. Oat samples shared these predominant species, with a reduced relative abundance of F. clavum and an increase in F. equiseti. Wheat samples were primarily dominated by F. clavum, accompanied by a more uniform relative abundance of the remaining Fusarium species. With respect to ITS2, the species and relative abundances obtained with this marker differed from those obtained with TEF1. For all cereals, the relative abundance of Fusarium ITS2 ASVs without a species-level assignment was among the greatest, whereas for TEF1, a relatively low percentage of Fusarium ASVs corresponded to Fusarium sp. In terms of species-level Fusarium ASVs, all ITS2 ASVs belonging to the F. tricinctum species complex (FTSC) were classified as F. tricinctum, whereas TEF1 ASVs of the FTSC were also classified as species such as F. acuminatum, F. avenaceum, and F. flocciferum. The same occurs with the F. sambucinum species complex (FSAMSC), with ITS2 ASVs classified only as F. sporotrichioides and F. poae, whereas TEF1 FSAMSC ASVs were also classified as relevant mycotoxigenic species such as F. langsethiae, F. graminearum and F. culmorum. In the case of the F. chlamydosporum species complex (FCSC), F. fujikuroi species complex (FFSC) and F. incarnatum-equiseti species complex (FIESC), only TEF1 allowed the identification of ASVs with a species-level classification included in one of these species complexes. This is particularly relevant for the FIESC, given the high relative abundance of F. clavum in all cereals.

Fig. 2.

Bubble plot of the relative abundances (within ASVs classified as Fusarium) of Fusarium species detected with ITS2 and TEF1 sequencing, aggregated by cereal and by species complex. FCSC: Fusarium chlamydosporum species complex, FFSC: Fusarium fujikuroi species complex, FIESC: Fusarium incarnatum-equiseti species complex, FOSC: Fusarium oxysporum species complex, FRSC: Fusarium redolens species complex, FSAMSC: Fusarium sambucinum species complex, FTSC: Fusarium tricinctum species complex. FFSCundesc, FSAMSC25, FSAMSC26, and FTSCundesc refer to the ad hoc nomenclature used in the FUSARIUM-ID database to designate undescribed species within each complex

Diversity Analysis

Focusing first on ITS2, Fig. 3 shows three different alpha fungal diversity indices (left) and a beta diversity rPCA biplot of Aitchison distances (right) aggregated by cereal. With respect to alpha diversity, the Chao1, Shannon, and inverse Simpson indices provided an overview of the richness, evenness, and dominance of the fungal communities within each sample. No significant differences were found between barley and oat in any of the three indices. However, the Shannon and inverse Simpson values corresponding to the wheat samples were significantly greater than those corresponding to the other cereals (Shannon Kruskal-Wallis wheat vs. barley P = 0.0325, wheat vs. oat P = 0.0144; inverse Simpson Kruskal-Wallis wheat vs. barley P < 0.0001, wheat vs. oat P = 0.0011). In contrast, alpha diversity indices from Fusarium communities calculated from TEF1 ASVs were not significantly different between any of the cereals (Figure S3, left).

Fig. 3.

Fungal diversity of ITS2 ASVs aggregated by different cereal samples. Box plots of the Chao1, Shannon and inverse Simpson indices, with different letters indicating statistically significant differences between cereal types within each index based on Kruskal-Wallis tests (P < 0.05) (left). Aitchison robust principal component analysis (rPCA) biplot, highlighting distances to centroids and convex hulls, along with the top 10 ASVs (taxonomy and first 6 characters of the ASV ID) that influence differences on the basis of the magnitude of the vector (right). Axis percentages (*) indicate the explained variance of the low-rank matrix and do not represent the total variance of the original data (see Methods). Adonis R²: proportion of variance explained by host cereal species. Adonis adjusted R²: proportion of variance explained by host cereal species after controlling for geographical location

With respect to beta diversity, the type of cereal accounted for 24.84% of the variance in Aitchison distances when considered alone (Adonis R² = 0.2484, P = 0.001, 999 permutations) and for 14.63% of the variance when adjusted for sample location (Adonis adjusted R² = 0.1463, P = 0.001, 999 permutations). These variance values were attributed to intergroup differences rather than data dispersion (PERMDISP P = 0.339, 999 permutations) (see Methods: Statistical analysis). The rPCA biplot showed a marginal separation between cereal groups. The main ASVs driving the variance in the rPCA space were taxonomically classified as Alternaria, Aureobasidium, Cladosporium, Nigrospora, and Vishniacozyma. Conversely, for the TEF1 marker, the type of cereal accounted for 4.3% of the variance when considered alone (Adonis R² = 0.043, P = 0.001, 999 permutations) and for 5.8% of the variance when adjusted by sample location (Adonis R² = 0.058, P = 0.001, 999 permutations). In this case, the main ASVs driving the variance were assigned to F. clavum, F. acuminatum, and F. langsethiae (Figure S3, right).

Finally, sample location accounted for 61.61% of the variance in ITS2 Aitchison distances when considered alone (Adonis R² = 0.6161, P = 0.001, 999 permutations) and for 51.39% of the variance when adjusted for the type of cereal (Adonis adjusted R² = 0.5139, P = 0.001, 999 permutations). However, this significant effect was confounded by heterogeneous dispersion among the location groups (PERMDISP P = 0.011, 999 permutations) (Figure S4).

Differential Abundance Analysis

The species-level taxonomic classification of the differentially abundant (DA) TEF1 Fusarium ASVs between the cereals is shown in Fig. 4. Overall, although some ASVs were DA (p ≤ 0.05), none of them met the fold change threshold of 2. All the DA ASVs belonged to the F. sambucium species complex (FSAMSC) and, in the case of oat, to the F. tricinctum species complex (FTSC). One ASV classified as F. culmorum was DA in wheat and oat compared to barley. Another ASV belonging to the FSAMSC was DA in oat compared with both wheat and barley. No DA ASVs were detected in barley compared with oat (Fig. 4, right). A similar differential abundance analysis for ITS2 ASVs is shown in Figure S5. DA ASVs on wheat were predominantly classified within the Pleosporales group, whereas oat comparisons were characterized by Cladosporiales, and barley ASV corresponded to Cystofilobasidiales and Tremellales. At the genus level, Cladosporium, Alternaria, and Vishniacozyma were the most frequently identified DA taxa.

Fig. 4.

Differential abundance analysis of TEF1 ASVs: barley versus wheat (left), oat versus wheat (centre), and oat versus barley (right). Bar plots of species-level taxonomic classification of differentially abundant (DA) ASVs between cereals (P ≤ 0.05). FSAMSC: Fusarium sambucinum species complex, FTSC: Fusarium tricinctum species complex. FSAMSC26 refers to the ad hoc nomenclature used in the FUSARIUM-ID database to designate undescribed species within each complex

Discussion

Cereals are an essential component of animal and human diets and require comprehensive study, particularly regarding health risks such as their contamination by toxigenic fungi. Among these, species of the genus Fusarium are of particular concern, as they are not only major plant pathogens responsible for significant agricultural losses [64, 65] but also produce a broad spectrum of mycotoxins, including fumonisins, trichothecenes, and zearalenone. In a context in which climate change is shifting fungal distribution patterns, distinguishing among these species is essential for precise risk assessment. High-throughput amplicon sequencing (metabarcoding) offers valuable insights into fungal diversity and the distribution shifts of mycotoxigenic species, as demonstrated by our previous studies [21, 66]. However, the standard ITS2 region lacks the resolution required to discriminate among Fusarium species. To overcome this limitation, in this study, we used TEF1-based metabarcoding for Fusarium species-level classification and compared its performance against that of the universal ITS2 marker on a set of Spanish cereal grain samples.

The TEF1 marker provides robust species-level resolution for the genus Fusarium because of the high intrinsic variability of this marker within the genus [27]. Consequently, the fungal profile obtained with TEF1 in this work is highly targeted. The primers used were specifically selected to ensure the comprehensive amplification of Fusarium species, without consideration for maintaining universality across other fungal taxa. This can potentially lead to the underamplification of some fungal groups. Given that the TEF1 approach is intentionally specialized, the results obtained from the universal ITS2 marker perfectly complement this targeted profiling. ITS2 fulfils the fundamental prerequisite of resolving broader microbial diversity, as evidenced by the minimal percentage of unassigned ASVs at the genus level (Fig. 1).

With respect to the genus Fusarium in particular, the taxonomic resolution provided by the two markers differed significantly. Compared with the ITS2 marker, TEF1 achieved high resolution by identifying a diverse array of Fusarium species. Considering that TEF1 is widely recognized as the optimal barcode for Fusarium species discrimination [24, 26, 27], the discrepancies observed suggest that numerous ITS2 ASVs were either misclassified or lacked sufficient sequence variation to be resolved. Indeed, as shown in Fig. 2, ITS2 ASVs were predominantly assigned to a few specific taxa, such as F. sporotrichioides and F. tricinctum, or remained unresolved as Fusarium sp. These species-level assignments must be interpreted with caution, as they are likely influenced by a combination of confounding factors, including (i) ITS copy number variation across Fusarium lineages, (ii) reference database composition, and (iii) bioinformatic processing choices. Consequently, these assignments underscore the lack of reproducibility and the fundamental limitations of the universal ITS2 barcode for species-level resolution within this genus [24, 25]. Conversely, TEF1 analysis of the same samples revealed a much more comprehensive taxonomic profile, with a substantial reduction in the relative abundance of unresolved Fusarium sp.

Within the F. sambucinum species complex (FSAMSC), ITS2 was largely limited to identifying F. sporotrichioides (all cereals), a species known for producing type A trichothecenes [67]. In contrast, TEF1 retrieved many additional species within the FSAMSC, revealing a much broader risk profile. Notably, TEF1 identified major producers of ZEA and type B trichothecenes, such as F. culmorum (all cereals) and, with a lower relative abundance, F. graminearum (barley and wheat), alongside the ZEA producer F. cerealis (barley). It also detected the T-2/HT-2 producer F. langsethiae across all cereals, with a particularly high relative abundance in oat and barley. This highlights a severe limitation of ITS2 for food safety assessments, as relying solely on this marker would result in a failure to detect toxigenic threats such as F. culmorum, leaving potential ZEA and DON contamination unassessed. Furthermore, the limited variability of ITS2 not only masks these critical species but also leads to misclassifications within the complex. Specifically, the majority of F. sporotrichioides assignments made in oat by ITS2 are likely artefacts, as TEF1 only detects this type A trichothecene producer in barley and wheat [15] and not oat samples. Conversely, our data suggest that F. poae abundance is underestimated by ITS2, given that this species is also detected by TEF1 in barley. Interestingly, some of the species detected within the FSAMSC by TEF1 are consistent with previous studies conducted in Spain by our research group. For example, the detection of F. graminearum aligns with an earlier survey in barley [68], which linked this species to significant contamination with type B trichothecenes (DON) and zearalenone (ZEA) [16]. On the other hand, TEF1 also allowed the identification of species that were only sporadically reported or absent in previous culture-dependent surveys. A clear example is the DON and ZEA producer F. culmorum [17], identified here in all cereals but not reported in previous surveys of Spanish grains.

Among these findings within the FSAMSC, the detection of F. langsethiae is of significant relevance given the challenges associated with its isolation via traditional culture techniques because of its slow growth rate and atypical colony morphology, compounded by the often-symptomless nature of its plant infection [69, 70]. Furthermore, it is a primary producer of T-2 and HT-2 toxins in northern Europe [67, 71]. Climate change is reportedly driving its distribution shift towards warmer climates; indeed, its presence in southern Europe has been previously reported in barley and wheat [70, 71]. In addition, our research group reported the first occurrence of this fungus in Spanish oats [21]. This is particularly important given that the European Union recently introduced new reduced maximum levels for T-2 and HT-2 toxins in oat-based products [72]. Consistent with previous findings, the present work identifies F. langsethiae not only in oats but also in Spanish barley and wheat, reinforcing the evidence that this ecological shift affects a broader spectrum of cereals. Consequently, further surveillance studies across diverse cereal matrices are essential to fully evaluate the potential health risks associated with this species and its mycotoxins.

A similar pattern was observed within the F. tricinctum species complex (FTSC). ITS2 exclusively detected F. tricinctum, a known producer of enniatins (ENNs) [73], across all cereals, whereas TEF1 resolved a broader diversity of species within this complex. Crucially, this included not only F. acuminatum (present in all cereals), a known producer of type A trichothecenes and ENNs [4, 74] but also other significant taxa, such as F. avenaceum (all cereals) and F. flocciferum (wheat). Considering the heterogeneous toxigenic profile of the FTSC, accurate discrimination within this species complex is essential [75]. For instance, both F. avenaceum and F. flocciferum are increasingly recognized as major producers of emerging mycotoxins, such as ENNs [76, 77], and are not known to produce trichothecenes. Consequently, aggregating these species under a single F. tricinctum label, as occurs with ITS2, severely hinders food safety risk assessment. This overestimation by ITS2 is evidenced by the TEF1 results, where the relative abundance of F. tricinctum was markedly lower and even absent in oat samples. Considering these findings within the FTSC in a broader context, the identification of F. avenaceum and F. tricinctum across our samples is consistent with the findings of TEF1-based metabarcoding studies in other crops such as maize [20].

Beyond the FSAMSC and FTSC, TEF1 successfully resolved species within other complexes that were overlooked by ITS2. For instance, only TEF1 allowed the identification of members of the F. fujikuroi species complex (FFSC), such as F. subglutinans in barley samples. Similarly, F. equiseti, a representative of the F. incarnatum-equiseti species complex (FIESC), was detected across all the cereal types. However, these results must be interpreted with caution. For instance, only a few F. subglutinans isolates can produce fumonisins [19]. Similarly, although some F. equiseti isolates can produce type A and B trichothecenes and ZEA, this species is characterized as a common saprophyte within cereal crops [78], and its detection does not necessarily imply mycotoxin production. The identification of these species, along with those belonging to the F. sambucinum and F. tricinctum species complexes, highlights that the Fusarium community colonizing these cereal grains might comprise both potential mycotoxin producers and saprophytes.

Finally, in order to check for potential ecological interactions, we explored whether the relative abundances of fungal genera derived from ITS2 correlated with specific Fusarium ASVs derived from TEF1. However, no significant correlations were detected at either the ASV or the species/genus levels (data not shown).

In order to evaluate whether fungal diversity is influenced by host species in Spanish cereal grains, a diversity analysis was conducted using both ITS2 and TEF1 ASVs. The alpha and beta diversity metrics derived from the ITS2 data suggest that the cereal type (barley, oat, or wheat) is not a dominant driver of fungal community structure. With respect to within-sample (alpha) diversity, only the Shannon and inverse Simpson indices significantly differed, with the values for the wheat samples being significantly greater than those for the barley and oat samples. Overall, the alpha diversity ranges observed for ITS2 align with previous findings from our research group. For instance, in a prior characterization of oat grains and their associated soils in Spain and the UK, we reported alpha diversity values for grain samples within the same range as those reported here [21]. Furthermore, these findings are consistent with those of other recent metabarcoding surveys on cereals. For instance, Cao et al. [79] reported comparable Shannon and inverse Simpson indices for barley grains, although their estimated Chao1 values were higher than those observed in our dataset. A similar trend was observed for the alpha diversity indices derived from TEF1 ASVs (Figure S3, left), where no significant differences were observed for any of the indices across the three cereal types.

ITS2 beta diversity analysis failed to show a distinct clustering by cereal type in the rPCA. Instead, a clearer separation emerged when the samples were grouped by location (Figure S4). This clustering, coupled with the fact that statistical analyses attributed a lower percentage of explained variability to the cereal type (Adonis adjusted R² = 0.1463, P = 0.001, 999 permutations) than to the location (Adonis adjusted R² = 0.5139, P = 0.001, 999 permutations), indicates that the geographical location is the primary force shaping the fungal diversity of the grain. However, these findings should be interpreted with caution; the significant PERMDISP test (P = 0.011, 999 permutations) for this grouping indicates that these results could be partially confounded by heterogeneous dispersion and unbalanced sampling sizes among locations, which may influence the sensitivity of the Adonis test. Regardless, given the remarkably high R² value obtained, the influence of geographical location remains the most substantial factor, even when potential dispersion effects are accounted for. In this context, it is essential to clarify that ‘location’ should be understood not only as geographic coordinates but also as a complex proxy for site-specific factors, including edaphoclimatic conditions, soil properties, and individual agronomic management practices (such as tillage, crop history, and fertilization). This multitude of factors defining ‘location’ are broadly reported in the scientific literature as major drivers of fungal diversity, outweighing other variables such as host species across a wide range of habitats [80–83]. Consequently, future research should focus on identifying the specific drivers, such as temperature, moisture, and local management practices, that most critically contribute to this location-driven variance in fungal and Fusarium communities.

Furthermore, according to the rPCA biplot vectors, no consistent trend was observed among specific ASVs driving diversity changes towards a particular cereal. For instance, the vectors for Alternaria ASVs point towards both wheat and barley zones. Similarly, the ASVs classified as Cladosporium herbarum exhibited vectors pointing in diametrically opposite directions. These results support the conclusion that there is no host-specific preference for these taxa under the conditions analysed in this study. Overall, the predominant effect of location found in this work is consistent with a previous study by our research group, in which we reported that the fungal and bacterial diversity of cereal field soils is better explained by geographical location than by the farming system employed (conventional versus organic) [66]. Complementing the ITS2 beta diversity results, the rPCA computed for the TEF1 samples showed a similar trend with respect to cereal type, with a lower percentage of explained variability assigned to the host species (Adonis adjusted R² = 0.058, P = 0.001, 999 permutations) (Figure S3, right).

Consistent with these findings, the results of the differential abundance analysis of TEF1 ASVs (Fig. 4) further support the observation that the grain mycobiota is independent of the cereal host. Whereas a marginal trend was observed at the species complex level, specifically with members of the FSAMSC appearing differentially abundant in all comparisons, this did not translate into broad taxonomic shifts. Indeed, after the fold-change threshold was applied, the number of differentially abundant ASVs was reduced to zero. At the species level, when only statistically significant ASVs were considered regardless of the fold change magnitude, certain ASVs corresponding to potentially toxigenic species, such as F. culmorum, F. avenaceum, and F. langsethiae, were identified. Notably, one ASV found to be differentially abundant in oat compared with wheat and barley was annotated as FSAMSC26, a designation that represents an ad hoc nomenclature from the FUSARIUM-ID database for a previously uncharacterized fungus within the F. sambucinum species complex [84]. A recent taxonomic revision formally described this clade (formerly “novel 26”) as the newly proposed species F. cuspidatum [85]. F. cuspidatum is phylogenetically nested within the F. brachygibbosum clade and presents a relevant toxigenic profile, including the in vitro production of type A trichothecenes such as diacetoxyscirpenol (DAS) and neosolaniol (NEO) [84]. Furthermore, the detection of this species in oat grains is particularly relevant, given that members of the F. brachygibbosum clade are typically associated with soil environments rather than grains [21]. Broadening the scope to the general mycobiota, the differential abundance analysis performed on ITS2 ASVs showed a similar pattern to that of TEF1, with only a negligible number of ASVs found to be differentially abundant (Figure S5) [86, 87].

Conclusion

Using a metabarcoding approach involving the development of a TEF1 Naive Bayes classifier, we resolved the Fusarium community in Spanish cereal grains to the species level. Our findings not only align with those of previous studies, confirming the persistence of key species such as F. langsethiae or F. graminearum but also reveal broader toxigenic potential through the consistent detection of species often overlooked by traditional methods, such as F. equiseti, F. acuminatum, and F. culmorum. Furthermore, using the ITS2 marker, we demonstrated that the grain mycobiota is shaped primarily by geographical location rather than by host cereal species. Overall, our results underscore that the characterization of fungal and Fusarium communities in cereal grains, achieved by integrating different metabarcoding marker regions, is essential for accurate health risk assessments and the development of robust food safety strategies in a changing global landscape, with profound implications for both human and animal health.

Supplementary Information

Below is the link to the electronic supplementary material.

Author Contributions

S. A-S.: Data curation, Investigation, Methodology, Software, Visualization, Writing – original draft, Writing – review & editing; B.P.: Conceptualization, Funding acquisition, Project administration, Resources, Writing – review & editing; J.G-S.: Conceptualization, Investigation, Methodology, Resources, Writing – review & editing.

Funding

The authors acknowledge the grant PID2022-136803OB-I00 funded by MICIU/AEI/https://doi.org/10.13039/501100011033 and FEDER, UE. Sergio Alías-Segura was funded by a predoctoral fellowship by Complutense University of Madrid and Banco Santander (CT25/24).

Data Availability

Raw sequencing data have been deposited in the European Nucleotide Archive at EMBL –EBI under accession number PRJEB108986. Feature tables have been archived on Zenodo (https://doi.org/10.5281/zenodo.18848848). Metataxonomic analysis was performed with code from [https://github.com/SergioAlias/sporeflow](https:/github.com/SergioAlias/sporeflow). TEF1 Naive Bayes taxonomic classifier was developed with code from [https://github.com/SergioAlias/fusariumid-train](https:/github.com/SergioAlias/fusariumid-train). Visualizations were created with code from [https://github.com/SergioAlias/fungi-ITS-TEF1](https:/github.com/SergioAlias/fungi-ITS-TEF1).

Declarations

Competing Interests

The authors declare no competing interests.