Introduction

Species geographic distributions or global range maps, based on species occurrences and expert knowledge, are the cornerstones of conservation strategies, such as the classification of threatened species on the IUCN Red List. Linking species geographic distributions to ecological conditions, known as species niches [1], is fundamental for anticipating the effect of multiple threats on biodiversity, ecosystem functioning and nature’s contributions to people in the face of accelerating global change [2,3]. Species geographic distributions also serve as the foundation of Species Distribution Models (SDMs) [4], which help reveal the drivers underpinning both past and current species occurrences [5] and predict future species distributions under varying socioeconomic pathways [6,7] and climate change scenarios [8]. Given that anthropogenic stressors also shape global biodiversity distribution [9], local species extirpations and colonizations [10], and can be considered as dimensions of species ecological niches [11], they can no longer be ignored in SDMs [12].

The accuracy of these approaches depends on the quantity, quality, and representativeness of species occurrences, obtained from both opportunistic records and systematic surveys while sampling bias remains a pervasive issue that can hinder our perception and prediction of species–human–environment relationships [13,14]. Species geographical distributions and ecological niches are indeed notoriously incomplete due to field surveys and species occurrences often being sparse and over-aggregated in taxonomic, geographic, environmental, and anthropogenic dimensions [15]. For example, only 6.74% of the Earth has been sampled for animals [14]. Moreover, less than 2% of species recorded in international databases such as the Global Biodiversity Information Facility (GBIF) and the Ocean Biodiversity Information System (OBIS) account for more than 50% of occurrence records in most taxonomic groups [14]. Furthermore, the inherent challenges of accessing and sampling the poles, remote areas, and deep waters have always restricted our capacity to detect species in their entire ranges and niches. Meanwhile, the occurrence data aggregated in GBIF and OBIS often fail to capture the full extent of biodiversity, particularly in the tropics where half of the cryptobenthic reef fishes remain missed by conventional surveys [14,16]. Detection biases and accessibility issues contribute to a strong latitudinal, longitudinal, and vertical sampling bias, resulting in incomplete geographic and ecological niche ranges for most species, given the environmental and anthropogenic uniqueness of the poles and the tropics [17].

Moreover, estimating accurate species geographical distributions and ecological niches depends on our detection capacity, minimizing the risk of false negatives (species being present but not detected). In the vast ocean, this becomes particularly difficult when dealing with small or elusive species [9,18]. For fish, this estimation is even more critical since they are sensitive to both environmental and anthropogenic pressures (e.g., habitat degradation) [10], and play important roles in linking marine ecosystem functioning and human societies [19], as well as supporting key contributions to people [20]. Conventional monitoring methods, such as underwater visual or camera surveys, often miss sharks, rays, and cryptobenthic fishes in habitats such as coral reefs or kelp forests [21–23]. Consequently, these species may suffer from silent local extirpations in the face of global change [10] since their geographic distributions and ecological niches are partly unknown. Thus, there is an urgent need for more comprehensive and non-destructive methods able to detect species across wide environmental and anthropogenic gradients [24].

Environmental DNA (eDNA) metabarcoding has recently been incorporated into the ecologists’ toolkit for accurately assessing fish occurrences [25,26]. By amplifying and sequencing DNA filtered in water samples, and comparing sequences to a genetic reference database, this non-invasive method can detect even the most elusive and rare fish species, including elasmobranchs [18] with a restricted spatial and temporal signature [27–31]. Yet, the extent to which eDNA surveys can complement classical surveys and global databases of marine fish occurrences in undersampled regions and ecological conditions to ultimately unveil unsuspected geographic and niche ranges has not been documented.

Here, we used over 6 years of eDNA surveys and nearly one thousand samples across the world’s oceans, including polar regions and tropical remote islands, to reveal sampling biases in species geographic ranges, as well as in environmental and anthropogenic ranges (hereafter ecological niche ranges) of 478 species. Rather than relying on using SDMs with limited additional occurrences, we compared species geographic and ecological niche ranges with and without the new eDNA occurrences. We also applied a machine learning and a null model approach to assess the benefits of sampling new locations and using eDNA metabarcoding to fill gaps in species geographic and niche ranges.

Results

GBIF/OBIS and eDNA records

The analysis of 936 eDNA samples collected from 542 locations worldwide (Fig 1a) detected 1,225 fish species that showed 100% identity on the teleo marker, whose primers amplify the DNA of both teleost and elasmobranch species [32]. In order to only consider species with accurate assignations, and thus avoid false positives (species detected but not being present), we removed species from genera in which fewer than 95% of species had been sequenced in neighboring regions, reducing the dataset to 478 species (see Methods). The likelihood of misassigning unknown species to other genera was low, as only 463 out of 19,777 unique sequences across the entire species pool were shared between different genera (2.3%).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Map of both GBIF/OBIS and eDNA surveys, and the distribution of their associated environmental and anthropogenic variables.

(a) Sampling intensity of GBIF/OBIS (i.e., the number of sampling events per 0.1° cell) overlaid with eDNA samples (orange dots). Cells with a dark background have not been sampled. Made with Natural Earth (public and free vector and raster map data at naturalearthdata.com). (b) Distributions of the five environmental and anthropogenic variables across the surveys: Sea Surface Temperature (SST; °C), Net Primary Productivity (NPP; mgC m⁻³ day⁻¹), Sea Bottom Dissolved Oxygen (SBDO; mmol m⁻³), Sea Bottom Temperature (SBT; °C), and Human Impact [28]. The values of these variables were extracted from the sampled cells, separating GBIF/OBIS-only cells (blue) and eDNA-only cells (orange). The data underlying this figure can be found in https://doi.org/10.6084/m9.figshare.30138220.

https://doi.org/10.1371/journal.pbio.3003432.g001

We extracted and downloaded occurrences from GBIF [33] and OBIS [34] databases for the 478 retained fish species, with 90% of these records dating from 1980 to 2023. Next, we gridded all eDNA and GBIF/OBIS species occurrences at 0.1° resolution (~10 km × 10 km at the equator) globally. Four types of cells can be distinguished: (i) cells with only GBIF or OBIS records (hereafter GBIF/OBIS-only), (ii) cells only sampled with eDNA (hereafter eDNA-only), (iii) cells with both GBIF/OBIS and eDNA samples (hereafter GBIF/OBIS + eDNA), and (iv) cells with no sample (hereafter unsampled). Out of the 4,335,949 global ocean cells, GBIF/OBIS fish occurrences were obtained in 223,956 cells (~5%). Our eDNA sampling effort yielded fish records in 239 cells (~0.006% of the ocean), with 110 of these cells sampled exclusively with eDNA (eDNA-only). The distribution of GBIF/OBIS occurrences reveals some vast unsampled areas, mainly in the Southern Hemisphere, the poles, the Pacific coasts of South America, and the coasts of Asia (Fig 1a). Several of the eDNA-only cells were in these previously undersampled areas (Tanzania, Lengguru in West Papua/Indonesia, the Arctic, and the Antarctic Oceans) and thus may fill gaps in our empirical knowledge of species distributions and niches.

We assigned each grid cell a value for five variables known to influence fish distributions [35] and available across the ocean: Sea Surface Temperature (SST), Sea Bottom Temperature (SBT), Sea Bottom Dissolved Oxygen (SBDO), Net Primary Productivity (NPP), and the Cumulative Human Impact Index [36]. We found significant differences in the mean value of these variables between eDNA-only and GBIF/OBIS-only cells (Fig 1b). More specifically, eDNA sampling effort covers almost the same SST range as GBIF/OBIS records but is more focused on areas with cooler SST (t test = 3.78, df = 106.09, p < 0.001), equal SBDO (t test = −1.11, df = 100.09, p = 0.27) and SBT (t test = 0.18, df = 106.09, p  = 0.85), lower NPP (t test = 6.05, df = 101.78, p < 0.001), and lower Human Impact (t test = 12.78, df = 110.2, p < 0.001).

Benefits of eDNA surveys

In order to assess the benefits of eDNA surveys, we used Boosted Regression Trees (BRT) to model the number of species gained per cell as a function of the eDNA sampling effort, the sampling effort estimated from GBIF/OBIS, the species richness obtained from GBIF/OBIS, bathymetry, and the environmental and anthropogenic variables across the 239 eDNA sampled cells in this study. The BRT model showed very high predictive power, with a correlation of 0.96 between predicted and observed values, and a correlation of 0.83 based on a 10-fold cross-validation. Partial dependence plots highlighted that the number of species gained (over 18) plateaued after 10 eDNA samples (S1 Fig). In contrast, species gains decreased sharply with the increasing GBIF/OBIS sampling effort, reaching its minimum at around 130 GBIF/OBIS sampling records (S1 Fig). We then generated predictions with this accurate BRT model to estimate how many additional species 10 eDNA samples would yield in each cell: we found a mean gain of 23 (s.d. ± 14.52) species and a maximal gain of 98 species in undersampled, remote tropical areas (Fig 2). In our study, the cost of analyzing 10 eDNA samples corresponds to approximately 5,000 euros, excluding the cost of transport to the sampling point. So, on average, the detection cost of a new species is worth 277 euros, but is highly variable across regions.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Map of the predicted number of fish species gained in a cell if 10 eDNA samples were added to GBIF/OBIS sampling records, if any.

Predictions obtained from a Boosted Regression Tree using the eDNA sampling effort, the sampling effort estimated from GBIF/OBIS, the species richness obtained from GBIF/OBIS, bathymetry, Sea Surface Temperature, Net Primary Productivity, Sea Bottom Dissolved Oxygen, Sea Bottom Temperature, and the Cumulated Human Impact as predictors across the 239 eDNA sampled cells in this study. Made with Natural Earth (public and free vector and raster map data at naturalearthdata.com). The data underlying this figure can be found in https://doi.org/10.6084/m9.figshare.30138520.

https://doi.org/10.1371/journal.pbio.3003432.g002

Geographic range filling

Aggregating all these data revealed species geographic (i.e., distance between a new detection and its nearest GBIF/OBIS occurrence) and ecological niche range (the relative increase in the niche breadth) filling with our eDNA samples (Fig 3). Species range fillings detected by eDNA in cells that had previously been sampled using GBIF/OBIS address a methodological sampling bias, since the species had been missed in the surveys reporting past non-eDNA records. At the opposite, we considered that range fillings detected by eDNA in cells that had never previously been sampled address a geographic sampling bias, since we cannot quantify the extent to which eDNA facilitated the filling in these cases.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Description of the framework used to analyze geographic and ecological niche range filling for each marine fish species.

(a) Example of species occurrences recorded in GBIF/OBIS on grid cells. New eDNA species detections (orange) were added to existing GBIF/OBIS occurrences (black). (b) Geographic range filling for a given fish species was measured as the maximum distance between new eDNA detections and the nearest GBIF/OBIS occurrence. (c) Ecological niche range filling for a given fish species was measured as the increase in the anthropogenic and environmental niche breadth, based on the minimum and maximum values encountered before and after updating the niche with new eDNA detections.

https://doi.org/10.1371/journal.pbio.3003432.g003

For each species, we computed the geographic range filling as the longest distance by sea between each of its eDNA detections and the nearest OBIS/GBIF occurrence (Fig 3b), allowing passage through the Panama and Suez canals. If no new occurrence was detected by eDNA beyond the OBIS/GBIF geographic range, the distance was set to 0 for this species. We obtained geographic range filling for 445 out of 478 fish species, i.e., 93% of the species pool, at a maximum distance ranging from 7.4 to 10,636 km from the nearest OBIS/GBIF occurrence (S2 Fig). The mean geographic range filling for the 445 species was 472 km (±893 s.d.). The distribution of the range fillings (S2 Fig) showed that they did not result from an arbitrary choice of the cell size, since 84% of them were above 20 km, and 73% above 50 km.

We identified a total of 57 species at locations more than 1,000 km from their nearest OBIS/GBIF occurrence. Six cases of particular interest have been chosen to illustrate these exceptional geographic range fillings. A species of crocodile icefish (Chionodraco hamatus), that is only known in the Antarctic, was detected in Patagonia using eDNA (Fig 4a). A sampling campaign in the Falkland Islands had also detected this species [31], but GBIF classified the occurrence as “presumed swapped coordinates”. Our detection confirms that this occurrence was indeed a true positive. We detected the Large-scale Mullet (Planiliza macrolepis) off the island of Corsica, although this species is usually found in the Indian Ocean and the Red Sea (Fig 4b), suggesting a hidden long-range colonization (>4,000 km) and a new non-indigenous fish species in the Mediterranean Sea. The Big Roughy (Gephyroberyx japonicus) lives at a depth of more than 300 m in temperate waters across the Western Pacific Ocean. However, a conserved specimen was found in Mauritius in 2014 [37], and we detected this species in eDNA samples in Lengguru, West Papua (3 samples), and in Tanga, Tanzania (Fig 4c) suggesting a more cosmopolitan distribution than previously thought of this species across the Indopacific. The Phaeton Dragonet (Synchiropus phaeton) is usually found in the deep waters of the Eastern Atlantic Ocean. However, a Phaeton Dragonet was recorded in the Gulf of Mexico [38]. We detected this species near Malpelo, in the Eastern Pacific Ocean, Columbia, in nine different eDNA samples (Fig 4d). The only referenced congeneric species in the area that could match eDNA assignment to the reference database, the Antler Dragonet (Synchiropus atrilabiatus), has its complete 12S sequence available on GenBank which is different. It is therefore possible that the Phaeton Dragonet also occurs in the Gulf of Mexico, as well as in the Caribbean Sea, and may have passed the Panama Canal in an interoceanic colonization event. The Shortfin Mako (Isurus oxyrinchus), an endangered shark species commonly found in all temperate and tropical seas across the world [39], was detected in our eDNA samples near the Arctic circle in the Baffin Sea (Fig 4e) suggesting a wider geographic range and niche than previously thought, thus questioning its IUCN status. Finally, the Bluehead Wrasse (Thalassoma bifasciatum) is usually detected in the Caribbean Sea, but five individuals were sampled and photographed in Tonga in 2006 while another study detected the species in New Caledonia through eDNA surveys in 2015 [40]. We added a detection in the Western Pacific through eDNA surveys performed in West Papua, Indonesia (Fig 4f), suggesting a wider geographic distribution across coral reefs.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Maps of the distance to the nearest GBIF/OBIS occurrence for six species detected by eDNA far from their known spatial range and qualified as geographic range filling: the darker the background, the further the cell is from a referenced GBIF/OBIS occurrence.

Detections via eDNA in our surveys are represented with orange dots. Distances are calculated by sea, and passage through the Suez and Panama canals is permitted. Made with Natural Earth (public and free vector and raster map data at naturalearthdata.com), icons made by hand. The data underlying this figure can be found in https://doi.org/10.6084/m9.figshare.30138226.

https://doi.org/10.1371/journal.pbio.3003432.g004

Ecological niche range filling

After associating all GBIF/OBIS fish records located in all grid cells with the corresponding environmental and anthropogenic values, we extracted all species ecological niche ranges, expressed as the range (maximum value - minimum value) of each variable for each species. Subsequently, we updated these species niche ranges using eDNA surveys and assessed the percentage of ecological niche range filling for each variable and each species, if any. We recorded a total of 66 niche fillings across 35 species (6.9% of the retained species), with 30 fillings at the lower margin of the range and 36 at the upper margin, ranging from 0.41% to 475% (Fig 5). The largest niche fillings were a 9.5 °C positive range filling in SST for the crocodile icefish (Chionodraco hamatus), a 12.6 mgC m⁻³ day⁻¹ negative range filling in NPP for Ophiogobius jenynsi, a 56 mmol m⁻3 positive range filling in SBDO for the same species, an 11 °C negative range filling in SBT for the red-fin goby (Evorthodus minutus), and a 1 unit positive range filling in Human Impact for the zebra goby (Zebrus zebrus).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Ecological niche range fillings detected with eDNA surveys for 35 fish species, calculated as the filling percentage relative to the known range, for each of the five variables.

Range fillings at the upper and lower margins of each variable are represented by black dots, and their distributions with red and green boxplots, respectively. Icons made by hand. The data underlying this figure can be found in https://doi.org/10.6084/m9.figshare.30138235.

https://doi.org/10.1371/journal.pbio.3003432.g005

We found that 63% of these 66 ecological niche fillings were detected by eDNA surveys in cells where species were not recorded in GBIF/OBIS records. These eDNA-facilitated fillings, addressing the methodological bias (Fig 3), predominantly corresponded to cryptobenthic species (31 out of 42 niche fillings), from families such as Gobiidae, Tripterygiidae, Syngnathidae, and Apogonidae. We also found that when filling ranges by addressing the geographic bias, the maximum body length of fish species was greater than when addressing the methodological bias (Wilcoxon test, W = 772.5, p < 0.001; Fig 6).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Violin plots showing species maximum body length associated with each ecological niche range filling on a logarithmic scale (66 range fillings across 35 fish species).

Range fillings addressing the methodological sampling bias (cells that were previously sampled in GBIF/OBIS) and those addressing the geographic sampling bias (cells that were not previously sampled in GBIF/OBIS) are shown in light and dark orange, respectively. The data underlying this figure can be found in https://doi.org/10.6084/m9.figshare.30138244.

https://doi.org/10.1371/journal.pbio.3003432.g006

Discussion

Despite the extensive sampling efforts in recent decades, which together encompass all the world’s oceans, there is still a significant bias and major gaps in the assessment of marine biodiversity based on existing global occurrence databases such as GBIF and OBIS. These biases and gaps hinder the correct estimation of species niches [13] and are critical in the context of the ongoing biodiversity crisis under accelerating global change [10]. Here, by using worldwide eDNA samples to complement GBIF/OBIS occurrences, we demonstrate geographic range filling for most of the species targeted in our study (93%) and ecological niche range filling for a few species (7%). This range filling of various magnitudes reveals geographic (e.g., species detected in new cells), ecological niche (e.g., species found at unexpected values of their environmental and anthropogenic boundaries), and taxonomic (e.g., cryptobenthic fishes exhibited more niche range filling) biases. Most of these range fillings address a methodological sampling bias, occurring when eDNA detects a species where it was not previously recorded in GBIF/OBIS, despite the location having already been sampled. Simulations demonstrate that an average 23 species could be gained in a cell for 10 eDNA samples. Those gains would be higher in cells without GBIF/OBIS occurrences.

Species geographic range fillings were detected in all directions: poleward, equatorward, and longitudinally. This multidimensional range filling was mostly expected for rare fishes because detecting new eDNA occurrences for these species increases the likelihood of finding them far from previous records without eDNA, given that the number of records for such species is essentially small. Surprisingly, we also detected geographic range filling for species with a large number of occurrences, such as the shortfin mako shark (Isurus oxyrinchus), indicating that even data-rich species may have been missed by OBIS/GBIS records, specifically in undersampled areas such as the polar regions. The high proportion of species with geographic range filling (93%) corresponds to new occurrences increasing their area of occupancy by at least one cell (i.e., ~ 10 km) for nearly all studied species. This does not necessarily reflect an increase in these species’ extent of occurrence since geographic range filling of a few kilometers most likely corresponds to gap filling in the true area of occupancy (which was certainly underestimated in previous OBIS/GBIF records) or to spatial mismatch due to the fine scale of our study. However, we show that this effect is certainly marginal since most species display range filling at a coarser scale. Indeed, the large number of species found at hundreds or more than a thousand kilometers from the nearest GBIF/OBIS occurrence more likely reflects major underestimations in species’ extents of occurrence suggesting that species biogeography in the ocean remains poorly understood and that more locations should be investigated with new technologies [41]. This lack of knowledge in unsampled regions has already been highlighted for deep sea species in OBIS [41], and for all species in GBIF since 79% of GBIF data come from only 10 countries, and 37% from the USA [14]. The OBIS community acknowledges these widespread gaps across taxonomy and space [42]. Yet, no study has investigated the knowledge gaps at the species level, considering ecological niche ranges with an alternative sampling approach.

The BRT model revealed that eDNA surveys can potentially increase the number of species detected in tropical regions and that this increase depends on the number of sampling events per cell, with diminishing gains after more than one hundred GBIF/OBIS records. This result shows that in very diverse and unexplored areas, eDNA can be a valuable tool for estimating marine fish diversity. However, while a small number of eDNA samples may capture a substantial portion of the species pool, recent studies suggest that eDNA and conventional methods often detect different species subsets [43]. This underscores the value of integrating multiple sampling methods to obtain more accurate estimates of marine biodiversity and propose conservation strategies, particularly in tropical regions [25]. In polar regions, the predicted number of species gains remains quite low, likely due to incomplete genetic databases [44] and lower diversity. This highlights the urgent need to increase research efforts in these rapidly changing areas [45], which are home to unique fauna [46] due to recent speciation rates [47].

Our eDNA sampling efforts in remote, previously unsampled locations broaden niche conditions beyond those which have been extensively sampled in recent decades, adding unexpected species in some geographical areas and ecological conditions. With a rather limited number of sampled cells (n = 239) compared to GBIF/OBIS (n = 223,956), and a majority of coastal species detected by our eDNA sampling design [9], we filled the geographic and ecological niche ranges of all fish kinds, from large sharks with circumpolar distributions to small endemic and cryptobenthic gobies. For example, the shortfin mako (Isurus oxyrinchus) was detected in an eDNA-only sampled cell, 3 °C lower in SST than previously recorded, resulting in a 10% filling of its lower thermal niche margin. This finding suggests unseen poleward range shifts and represents crucial knowledge to efficiently preserve such threatened species, especially since fast range shifters often present declining population dynamics [48]. On the other side of the fish size spectrum, we detected the Liechtenstein’s goby (Corcyrogobius liechtensteini), which is endemic to the Mediterranean Sea, by addressing the methodological sampling bias in the Mediterranean, therefore increasing its NPP range by 110%. The detections of the shortfin mako shark and of the Liechtenstein’s goby illustrate that sampling in remote areas and performing eDNA surveys in over-sampled areas may both increase fish ecological niche ranges toward unexpected values. Moreover, since locations with particularly low human impact or high temperature also exist in remote tropical areas such as in West Papua or in Tanzania, we detected range fillings at both low and high niche margins, highlighting the importance of tropical samples in addition to polar ones, since future range projections from climate change could be biased due to uncharted species territories or niches [35,49]. Our results relating to temperature and oxygen are crucial since marine fishes are mostly ectotherms and may shift their ranges faster than terrestrial species [35]: a 10% filling in a fish species’ thermal range is a key parameter to set up effective conservation strategies and fisheries resource management in the face of climate change. They are also of paramount importance regarding the only human-related variable included in the study: the anthropogenic niche remains poorly quantified for most species, although key in modeling and management [12].

The null model approach allowed us to quantify which ecological niche range fillings may simply have been caused by a small number of species occurrences. It reveals that half of them were significantly higher than expected under the random allocation of sampling effort. Nevertheless, this range filling still indicates that these species lack records in global databases. The main limitation of this model is that it only generates random occurrences within previously sampled locations. Since some eDNA samples occur in locations where no GBIF/OBIS records were available, the model cannot generate random occurrences in these unsampled locations. As a result, when eDNA contributes to range filling in previously unsampled regions, we cannot determine whether this effect is due to eDNA’s unique ability to detect species or simply because we sampled a new overlooked location. Assessing the contribution of eDNA surveys in depth would require a null model that spatially mimics our sampling effort. However, this was not achievable because our sampling was deliberately focused in remote, unsampled areas such as the poles or Patagonia. A null model would randomly sample species in locations where they cannot occur, thereby artificially inflating the likelihood of attributing our range fillings to random events. Alternative approaches would be to either obtain samples through conventional methods in locations where only eDNA surveys were performed, or increase the number of eDNA surveys where OBIS/GBIF data is already available [50,51], thus enabling a fair comparison between eDNA and conventional methods.

Although cryptobenthic fishes are the building blocks of reef trophodynamics [21], they are challenging to detect and underestimated by OBIS/GBIF, thus hindering the assessment of their ecological niche, geographic range, and potential extinction risk [52]. In remote locations, where space, time, and resources represent major sampling constraints, eDNA facilitates the detection of cryptobenthic species, particularly in environments rich in organic matter, where other techniques may have missed them [53]. Accordingly, our results show a significant difference in maximum body length between species range fillings addressing a methodological bias and those addressing a geographic bias [54]. The logistic regressions also indicate a lower chance of detecting range fillings for larger species, notably in highly sampled cells, even if eDNA allows the detection of large elusive species such as sharks [18]. Our result suggests that fish species of all sizes can still be detected in remote undersampled locations, while mainly small-bodied fishes can be discovered from eDNA surveys in GBIF/OBIS cells owing to eDNA’s superior detection capacity over conventional methods even in the same locations [55]. While eDNA appears promising in terms of gap-filling in occurrence data, some limits persist. We may have detected unknown species that have never been recorded in a given location, or known species that have unreported intraspecific variability, and that share an identical sequence with the species we report here (false positives), or even species that diverged very recently (e.g., Chaetodon family). This is particularly true in hyper-diversified regions where the available genetic reference database may cover around 40%–50% of the known checklist [44]. In addition, we know the approximate error rate for GenBank (~5%) [56,57], but we cannot precisely determine an identification error rate for GBIF and OBIS at a global scale and across so many species. This implies that, on the one hand, a small fraction of our detections may be erroneous, and on the other hand, that some ecological niches or geographic ranges computed with GBIF and OBIS may be overestimated, thus hindering the capacity to detect true range fillings when adding our eDNA samples. Nonetheless, we may have also discarded several true positives since the sequencing coverage in the region was below our 95% threshold, leading to an underestimation of our range filling. In this case, we point out the importance of better sampling remote areas, of completing genetic reference databases, and advocate the use of eDNA multi-metabarcoding primers [58]. Indeed, short markers increase the likelihood of detecting rare species and degraded eDNA whereas longer fragments could allow for better resolution at the species identification step, and finer spatial resolution since they degrade more rapidly [32,55,59]. Considering that polar and tropical regions typically have low sequencing coverage [44], which is likely to induce species range underestimations, future sampling campaigns should combine multi-primer eDNA surveys whenever possible, and species sequencing to complete genetic reference databases. Some geographic or niche range fillings may also have been missed since our first eDNA samplings date from 2017, and some of eDNA detections may have been referenced in GBIF/OBIS when data were included in a scientific publication. Some of the reported range fillings could also actually correspond to species range shifts, but our framework does not allow us to separate range shifts from range fillings. Such range shifts may happen during extreme events, triggering a new colonization event, and thus need to be documented [60]. Another reason species occurrences were certainly underestimated in our eDNA surveys is that we usually sampled around 30 L of seawater in duplicates, so we may have missed a large portion of fish assemblages since our eDNA sampling effort was relatively low. This highlights that adapted sampling strategies, including more field replicates, are required if eDNA is to be used [61]. Moreover, eDNA optimizes detection capacity locally, but cannot provide exhaustive species inventories, since only a limited area and depth layer are sampled [62]. Future sampling campaigns should, where possible, conduct eDNA surveys at different depth levels (from mesophotic zone to bathypelagic and abyssopelagic zones, etc.) in addition to shallow waters [63].

Although a limited quantity of eDNA samples collected over a short time period is sufficient to demonstrate biogeographical and ecological biases and pinpoint the crucial need for more data, it does not allow for effective integration into SDMs. For example, the shortfin mako (Isurus oxyrinchus) has more than 8,500 occurrences reported in GBIF/OBIS, while we detected the species only three times with eDNA, one of them corresponding to a range filling. A novel species distribution model (SDM) which simply adds some new eDNA occurrences would have little to no impact on the estimated species niche in most modeling frameworks, especially if these occurrences represent ecological outliers.

Our observed pervasive geographic and ecological niche range fillings underline the need for more comprehensive sampling surveys to enhance conservation policies, assess sensitivity to climate change and anthropogenic stressors, improve SDMs, and most importantly, fill gaps in areas of occupancy. Since current SDMs incorporate spatial factors into their predictions [64], and threatened species are partly assessed based on their population size [65], geographic range filling is just as important as ecological niche range filling for future conservation strategies.

Materials and methods

Supporting information

Acknowledgments

Data from this paper come from the projects and expeditions listed below. We thank Laurent Ballesta, Patrick Durville, Florian Holon, and Laurent Pouyaud, for participating in the eDNA sampling.

eREF

Project led by the University of Montpellier and co-financed by Agence de l ’eau Rhône Méditerranée Corse (French Water Agency), Andromède Océanologie and Spygen

Gombessa Lapérouse

The Lapérouse expedition is a Galaxea initiative and was co-financed by the French State (Ministry of Overseas France, represented by the Prefect of Reunion Island), under the Convergence and Transformation Contract 2019–2022, measure 3.5.1.1 of the “Resilient Territories” section. Sponsors and patrons also participated in the financing: Vie Océane, Aquarium de La Réunion, Comité Régional d ’Etudes et de Sports sous-marins de La Réunion, Inset, O Sea Bleu, Fondation du Crédit Agricole Réunion Mayotte, Fondation d ’Entreprises des Mers Australes, Air France, Odyssée Créateur de voyages, Entropie. The financial partners of Andromède Océanologie and Gombessa Expéditions are: Blancpain, Ocean Commitment Blancpain, RGBlue, Keldan, Topstar, Bigblue, Ap Diving, Molecular, Subspace, Nikon, Aqualung, Seacam.

ANGE

This work was funded by the ANGE project led by the University of Montpellier and co-financed by Agence de l ’eau Rhône Méditerranée Corse (French Water Agency), Office Français de la Biodiversité (OFB), Parc Naturel Marin du Cap Corse et de l ’Agriate (PNMCCA), Andromède Océanologie, and Bastia Offshore Fishing. Fieldwork was also part of Gombessa 6 expedition.

Gombessa 6 et 6+ “CAP CORSE”

Data provided by the scientific expedition Gombessa 6 and Gombessa 6+ CAP CORSE led by Andromède Océanologie and supported by Manufacture de Haute Horlogerie Suisse Blancpain and Blancpain Ocean Commitment, the Prince Albert II de Monaco Foundation, the Société des explorations de Monaco, Office Français de la Biodiversité, Agence de l’eau Rhône-Méditerranée-Corse (French Water Agency), National Geographic Society, Iris Foundation and Lemarchand Foundation. The mission was also supported by La Marine Nationale (French Navy) and Préfecture Maritime de la Méditerranée. The authors also want to thank ARTE, INPP (Institut National de Plongée professionnelle), Les Gens Bien Production, CNC, Parc Naturel Marin du Cap Corse et de l’Agriate (France Relance program), Ushuaïa TV, AP diving, Aqualung, Nikon, Molecular, Seacam, Yamaha, Paralenz, Bigblue, Neotek, Seaowl, Marlink, Subspace, Suex, Dive System and Francqueville without whom this expedition would not have been possible.

PISCIS

Data provided from the PISCIS project led by Andromède océanologie and co-funded by Agence de l ’eau Rhône Méditerranée Corse (French Water Agency) and the University of Montpellier.

Lengguru

We thank the National Research and Innovation Agency (BRIN) in Indonesia and IRD in France for promoting our international collaboration with the strong support of the Sorong Polytechnic of Marine and Fisheries (Politeknik KP Sorong), in West Papua, Indonesia. We thank the International Joint Laboratory ‘Sentinel Laboratory of the Indonesian Marine BiodiversiTy’ (IJL SELAMAT) for its support. The Lengguru fieldwork was conducted according to relevant guidelines by the government of the Republic of Indonesia and under a research permit issued by RISTEK (Indonesia) (permit no. 3179/FRP/E5/Dit.KI/IX/2017) and the relevant Indonesian government collecting permit. Fieldwork in Indonesia and laboratory activities were supported by the Lengguru 2017 Project, conducted by the French National Research Institute for Sustainable Development (IRD), National Research and Innovation Agency (BRIN) with the Research Center for Oceanography (RCO), the Politeknik Kelautan dan Perikanan Sorong, the University of Papua (UNIPA) with the help of the Institut Français in Indonesia (IFI), funding from Monaco Explorations, and corporate sponsorship from the Total Foundation and TIPCO company.

Malpelo

We thank Monaco Explorations for funding fieldwork and laboratory analysis, the Yersin crew for assistance with at-sea operations, the Malpelo Foundation for the coordination of the expedition and National Parks of Colombia and the Navy of Colombia for the permits.

"We thank the José Benito Vives de Andréis Marine and Coastal Research Institute Invemar for their institutional support, and to Giomar H. Borrero-Pérez and Andrea Polanco F. for coordinating the fieldwork in Colombia, performing sample collection, and managing the process of obtaining research permits."

Shift-eDNA

We thank the Nunavut Research Institute and all local communities for allowing us to obtain samples in the Baffin Sea, and sharing their knowledge with our teams. We also thank Eric Brossier (Vagabond) and Sébastien Roubinet (Nagalaqa) for the fieldwork, and we have a special thought for Louis Bernatchez, who helped us prepare fieldwork in Nunavut. This work was financed by the ANR International Shift-eDNA (Call CE2-2021).

TOPtoTOP Global Climate Expedition

We thank Dario Schwörer for his participation in the TOPtoTOP Global Climate Expedition.

MPA poverty

This study was also supported by the MPA-POVERTY project (ANR-19-CE03-0005).

LabCom DiagADNe

This study was also supported by the LabCom DiagADNe (ANR-20-LCV1-0008) funded by the Agence Nationale de la Recherche (ANR).