What a positive marine eDNA result actually says
Environmental DNA, usually shortened to eDNA, is genetic material recovered from an environment rather than directly from a captured and identified organism. In marine water, the material can include cells, fragments, organelles and extracellular DNA. Metabarcoding amplifies a chosen genetic marker from mixed environmental DNA and sequences many resulting molecules. The primary outcome is a set of sequence observations under a particular sampling, laboratory and computational workflow; it is not a literal census of living animals at the sampling point.
Imagine a marine water sample whose sequencing table contains 320 reads assigned to a fish taxon. This means that 320 retained sequencing reads were classified that way by the stated pipeline. It does not mean 320 fish were present, 320 fish contributed DNA, or the fish were physically under the vessel at the instant of collection. The defensible interpretation requires the markers, assignment thresholds, controls, reference database and possible transport pathways.
An accessible translation begins: “DNA matching the selected taxon was detected in this water sample using the stated assay, subject to laboratory controls and classification rules.” That statement is useful without inventing an organism count. Claims about local presence, relative biomass or population change need additional evidence. This distinction is especially important because marine currents can transport DNA away from its source and because DNA can persist after organisms have moved.
Why the field sample is already a selective observation
A water bottle collects one small part of a changing sea. The collection point, depth, time, water volume and vessel handling determine the physical sample. Marine DNA is unevenly distributed, especially near boundaries, river plumes, particle aggregations and biologically patchy habitats. Sampling a single bottle and assuming it represents an entire bay is a spatial inference that the laboratory cannot repair by obtaining more sequence reads from that one bottle.
Suppose two one-litre samples are collected metres apart and only one detects a rare target. That result might reflect patchy DNA distribution, chance sampling, laboratory stochasticity or contamination. It does not automatically demonstrate that the animal occupied one metre of water but not the next. Repeated collections, environmental measurements and an explicit sampling design help separate those explanations. A good report states whether specimens were separate field samples or aliquots of the same bottle.
If a station is sampled three times on different days, those are temporal observations. If three PCR wells are amplified from one extraction, those are technical observations. Treating the wells as three independent days artificially inflates evidence about persistence. A simple metadata table with station, collection time, extraction batch, filter and amplification identifiers can prevent this mistaken translation of replication.
Filtration volume, pore size and retention
Water may be filtered to concentrate DNA-bearing particles and material, but the operational definition depends on the filter, pore size and processing. Larger water volumes can sometimes increase the number of captured target molecules, while turbidity and clogging can make processing difficult. A high volume is not unconditionally better. A filter may retain cell-associated DNA differently from smaller extracellular fragments, and the retained fraction does not represent every molecule in the sea.
In a hypothetical comparison, one filter processes 0.25 litres and a second processes 1.00 litre. If both procedures have equal target concentration and recovery, the second volume might deliver four times as much original target material. But if the larger filter clogs and its extraction efficiency falls, the observed sequence output need not follow a fourfold rule. Sequencing read counts are further shaped by amplification and sequencing allocation.
An honest result needs the volume actually filtered, not the bottle’s nominal capacity. A one-litre bottle yielding only 600 millilitres through a clogged filter must not be described as one full litre processed. Volume is relevant to detection design and quantitative conversion; the read table alone rarely reveals it. Similar-looking sequence tables can arise from very different amounts of starting seawater.
Preservation and DNA degradation
After collection, enzymes, microbes, sunlight and temperature can affect DNA integrity. Filter preservation and extraction timing matter, and validated procedures vary with the study. The interpreter should not assume every sample preserves the same fraction of genetic material. A preservation deviation can affect the probability of recovery even when the original biological community was unchanged. Recording this is part of the scientific result, not optional clerical detail.
Consider samples collected in the same bay on consecutive mornings. One was preserved promptly under the validated method; the other remained warm for several hours because transport failed. If the second shows fewer target detections, one cannot confidently call that a population decline. Differences in preservation create a competing explanation. The correct next step is to flag the handling difference and assess whether a validated comparison remains possible.
Even intact DNA does not prove a source organism is alive. DNA may come from waste, carcasses, prey remains, transported particles or organisms in transit. The ecological question must therefore be stated carefully. Detection can be strong evidence that target DNA reached the sampled water, while evidence of live local occupancy may remain indirect.
Field blanks and their distinctive purpose
A field blank is a control exposed to elements of the collection workflow without intentionally collecting environmental target material. It can reveal DNA introduced through water handling, containers, pumps, equipment, personnel or surrounding activities. Exact blank designs differ by study. A field blank is not a substitute for a sample replicate, and its contamination pathways may differ from those in real seawater.
Imagine a survey with ten environmental samples and two field blanks. A fish sequence appears in six environmental samples and one blank. The blank detection should trigger investigation of contamination pathways and how the study’s predetermined decision rule handles affected taxa and samples. Deleting the blank from the results so the survey appears clean would conceal information essential to interpretation.
The absence of a sequence in a small number of blanks is reassuring but not a universal proof that contamination never occurred. Sporadic contamination may escape limited negative controls. Scientific prose should acknowledge the role and limits of controls, as emphasized by USGS research on contamination in aquatic eDNA studies. A negative blank provides evidence under its design, not a guarantee about every tube.
Extraction blanks and amplification negatives
An extraction blank passes through DNA extraction without an environmental sample and monitors contamination introduced at that stage. A no-template PCR control checks for DNA introduced during amplification preparation. These controls are complementary. A clean PCR control with a contaminated extraction blank points toward a different location in the workflow than contamination appearing only during PCR setup. Treating every negative control as an interchangeable check erases this diagnostic value.
If an unexpected sequence appears in an extraction blank, investigators may compare batch membership, reagent lots and sample handling order. If the same sequence appears only in a no-template PCR control, they examine amplification reagents, plates and transfer. Any decision to discard or qualify samples should follow documented rules and evidence. It should not be based on a convenient desire to preserve a narrative about ecological change.
The translated report should name each control type and the level at which it was processed. “All controls were negative” is insufficient if the study never included field blanks. “The controls appropriate to sampling, extraction and amplification showed no detections above the stated thresholds” is more informative when supported by actual results and the specific thresholds.
Positive controls and assay competence
A positive control contains a known target or sequence construct expected to be detected. It helps establish that the amplification and analytical workflow can recognize relevant material. An internal amplification control can help identify inhibition in individual samples. Neither makes every negative environmental result automatically reliable, because target abundance, sample-level inhibition and stochastic capture still affect detection.
Suppose a run’s positive control works but one seawater sample contains substances that inhibit the polymerase. The control tube may amplify while that sample produces no target reads. Without an inhibition check, the sample can be incorrectly declared target-free. A failed internal positive control suggests that the negative result cannot be interpreted in the same way as a technically sound negative sample.
When inhibition mitigation such as dilution is used, investigators must account for the trade-off: diluting inhibitors also dilutes rare target molecules. A successful reanalysis should report both the change in assay performance and any consequence for sensitivity. The correct plain-language explanation is not simply “the DNA disappeared”; it is that the measurement system was unable, or less able, to detect it under those conditions.
Primer selection changes what can be seen
Metabarcoding relies on primers that bind chosen regions of DNA so the intervening marker can be amplified. Different primer sets target different organisms or taxonomic resolution. Primer mismatches reduce amplification efficiency for some taxa, and a marker that works well for one group may be unsuitable for another. Consequently, the list of recovered taxa depends partly on the assay design, not only on the organisms contributing DNA to the water.
In a fictional community containing equal original target molecule numbers from taxa A and B, imagine the selected primers amplify A ten times more efficiently than B through the net workflow. The output might have many more A reads even when original molecule numbers were equal. The exact amplification ratio is example-only; real bias can depend on marker, taxon, PCR conditions and competition. Read proportions cannot simply be equated to organism proportions.
NOAA-affiliated research comparing marker choices has shown how metabarcoding views of biodiversity can differ across genetic markers. The right title for a result is “taxa recovered using the selected marker and pipeline,” not “all species in the sea.” A second marker may reveal groups missed by the first, and combining markers requires thoughtful taxonomic reconciliation rather than naively concatenating read counts.
PCR is nonlinear evidence of original DNA abundance
Polymerase chain reaction repeatedly copies target DNA. Early cycles may appear approximately exponential, but efficiency differences, reagent depletion, competition and stochastic effects complicate quantitative interpretation. Small starting differences can become substantial differences in amplicon products. A final sequencing read count is therefore not a direct measure of how many original DNA copies were present unless a validated quantitative design supports that conversion.
Consider two taxa with 100 template copies each. If the net amplification yields are 1,000-fold for A and 200-fold for B, the product pools contain 100,000 and 20,000 copies before downstream steps. A sequencing instrument sampling that pool will tend to reflect the amplification imbalance. The calculations illustrate why equal starting material need not yield equal read abundance; they are not a universal biological amplification law.
A strong technical description records primer sequences or names, target marker, reaction conditions, cycles and replicate structure. “More reads means more fish” is generally too strong. A defensible alternative is that a taxon was more represented in the sequenced amplicon pool under the selected workflow. Additional quantitative calibration may support more, but it has to be demonstrated.
Sequence library preparation and indexing
After amplification, sequence libraries may be assigned sample indexes and pooled for sequencing. The indexes help distinguish samples computationally. Index misassignment, cross-talk and carryover can still produce low-level reads in unexpected samples. Thus a single unexpected read should not automatically become a confident presence claim. Library preparation, indexing strategy and contamination controls are part of the interpretive context.
Imagine the run generates two million reads across multiple samples. One sample obtains 50,000 reads and another obtains 5,000. The second has one tenth the sequencing depth; it is not necessarily one tenth as biologically diverse. Rare taxa are more likely to be missed with fewer observations, and differences in sample allocation can affect detection. A proper comparison checks read depth and processing thresholds.
The word “library” is especially treacherous in cross-language writing. It can mean a prepared collection of DNA molecules ready for sequencing, not a physical library of books or a taxonomic reference database. A report may mention both a sequencing library and a reference sequence library. Clarify their separate roles; one is material being sequenced, the other is information used for classification.
Reads, molecules, cells and organisms
A read is a sequence observation generated by a sequencing platform, often after filtering and merging rules. It is not automatically one original DNA molecule. PCR duplicates and library preparation can create many reads from the same original template. Nor does one molecule equal one cell, because genomes have different copy numbers and organisms shed DNA at different rates. The chain from reads to organisms contains several unresolved conversion factors.
Suppose a table contains 2,400 reads assigned to a fish and 600 assigned to another taxon. The first represents 80 percent of those 3,000 selected reads. This is not evidence that 80 percent of fish individuals or biomass belonged to the first species. The denominator is the selected sequence read pool; it is not the population of fish in the sampled ecosystem.
When communicating this result, specify the post-quality-control read set and any filtering that removed unassigned reads. If the 3,000 reads were only part of a much larger dataset, the 80 percent proportion applies to the chosen subset. Failing to name the denominator can make a precise percentage completely misleading.
Bioinformatics quality filtering
Raw sequence output may include errors, low-quality bases, adapter sequences, chimeras and other artifacts. A processing pipeline filters and transforms those reads before taxonomic assignment. Exact methods differ. Error correction can infer amplicon sequence variants, or ASVs, while other pipelines cluster sequences into operational units. These are computational products with defined criteria, not guaranteed unique biological species.
An illustrative run starts with 100,000 raw reads, retains 82,000 after quality filtering and obtains 78,000 after chimera removal. The overall retained fraction is 78 percent of raw reads. Reporting “100,000 biological observations” as if every raw read supported taxon inference would ignore the filtering loss. Reporting the 78,000 without saying what was removed also limits reproducibility.
A reproducible report names the software, version, quality parameters and classification method. The vocabulary “denoised,” “merged” and “assigned” denotes distinct actions. Technical translations that replace all three with “cleaned” lose information about where errors might arise. A reader should be able to follow the journey from raw reads to the final taxon table.
ASVs are not automatically species
An amplicon sequence variant is a distinct inferred sequence in a chosen marker region under a particular processing approach. Two species can share the same marker sequence, and one species can sometimes contain variant sequences. Taxonomic resolution depends on the marker and available reference data. Therefore, a count of ASVs and a count of species are conceptually different statistics.
Suppose 18 ASVs are assigned to one genus and only six to named species. It would be incorrect to say that 18 species in that genus were conclusively found. Some ASVs may reflect within-species variation, unresolved taxa or classification limitations. Conversely, species sharing a barcode can be hidden within one sequence assignment. A report should distinguish ASV richness from supported species-level richness.
Researchers may choose ASVs, taxonomic groups or other observational units for different questions. The important point is to keep the unit explicit. If a visualization switches from species to ASVs between panels without explanation, a reader can infer a nonexistent diversity change. Precise language protects the meaning of the measurement even when the underlying data matrix remains untouched.
Reference databases and taxonomy confidence
Taxonomic assignment compares observed sequences against reference sequences and follows chosen similarity or probabilistic rules. A reference database can lack species, contain misidentified entries or have uneven coverage. A top match does not necessarily establish a unique species identity when related species share the marker. A short sequence can provide strong group-level support while providing little species-level discrimination.
Consider a sequence that matches two closely related species equally well over the amplified region. The scientifically honest output may be their shared genus or another broader group. Selecting the first database entry because it appears at the top of a list would turn a tie into an unsupported species claim. Such errors can propagate into conservation or invasive-species alerts.
A high-quality result includes the reference database name, version or release, classification thresholds and treatment of ambiguous assignments. “Unassigned” means the pipeline did not place the sequence at the required level; it does not mean the DNA must come from an unknown life form. New reference entries can change later classifications without changing the original environmental sample.
Negative detections and the false-absence problem
A negative eDNA observation is not proof of ecological absence. A rare species may shed little DNA, the bottle may miss it, extraction may lose molecules, PCR may fail or the marker may amplify poorly. The U.S. Geological Survey explicitly emphasizes replicates and detection probability when interpreting eDNA. That principle applies strongly to marine settings where water exchange and DNA fate vary.
Suppose a particular method has a per-sample detection probability of 0.60 when the target DNA is truly present under comparable conditions. If three independent environmental replicates each have that probability, the chance all three miss it is (1 − 0.60)^3 = 0.064, or 6.4 percent. This is a teaching probability model, not a published sensitivity estimate for all marine eDNA assays.
Independence is a crucial assumption. Samples drawn from the same failed extraction batch may have correlated failures, so the 6.4 percent calculation can be overoptimistic. Researchers should estimate detection under their own conditions and model dependence where appropriate. The lesson is that “not detected in three samples” is an evidentiary statement, not a universal declaration that a species is absent from a region.
Occupancy and detection are separate probabilities
Ecological occupancy models distinguish whether the target occupies a sampling unit from whether a survey detects it when present. Repeated observations can help estimate both processes, but the design must define the spatial and temporal unit. In a marine environment, transported eDNA can weaken the relationship between local biological occupancy and DNA detection, making the model’s interpretation especially important.
For an invented model, let occupancy probability be 0.40 and detection probability conditional on occupancy be 0.70. If false-positive detection is ignored for simplicity, the probability of a positive observation is 0.40 × 0.70 = 0.28. The 28 percent positive rate does not imply that only 28 percent of animals survived or that DNA chemistry changed in 28 percent of bottles. It mixes two different processes.
If false positives are possible, the simple product needs an expanded observation model. Controls and field validation help constrain that risk, but cannot remove it by declaration. When communicating occupancy estimates, state whether “occupancy” means organism occurrence or DNA occurrence as operationally defined. In moving water those are not always interchangeable.
Quantification limits and zero inflation
Targeted quantitative PCR, digital PCR and metabarcoding are different methods with different quantitative strengths. Even where quantitative copy-number estimates are defensible, limits of detection and quantification are assay-specific. USGS guidance recommends documenting how these limits were derived and checking for inhibition. A number below the validated quantification range should not be copied into a report as though it were a precise concentration.
Imagine a qPCR calibration that reliably quantifies from 20 copies per reaction upward. A sample yields an estimated 9 copies with intermittent amplification. The defensible label may be detected below quantification, depending on the assay’s predeclared criteria. Calling it “exactly nine copies” exaggerates confidence; calling it zero removes potential evidence. The distinction between detectable and quantifiable is central to technical explanation.
Metabarcoding read counts typically require even greater caution. Without an appropriate internal standard or calibration, ten reads cannot automatically be translated into ten copies. NOAA-affiliated research demonstrates that standards can improve quantitative interpretation for specific marker designs, while PCR and sample-level biases remain method-dependent. An article should describe the validated method rather than market one read-count conversion as universally correct.
Relative abundance and compositional effects
Sequencing data are often compositional: percentages sum to one hundred within a selected read pool. If one taxon contributes more amplified DNA, another’s percentage can fall even when its absolute DNA amount is unchanged. This denominator effect is easy to overlook. A relative read proportion is therefore not an absolute population trend and should not be interpreted as one without independent quantitative evidence.
In a synthetic sample A has 100 reads and B 100, so each has half of the selected pool. In a later sample A has 100 and B 300, so A falls to 25 percent even though its observed read count remains 100. Changes in sequencing depth and amplification may further affect these numbers. The percentage drop alone cannot establish an absolute decline in A’s source DNA.
A report must also say whether the denominator includes all quality-filtered reads, only assigned reads, or only reads in a chosen biological group. Different choices yield different proportions. Ratios can be analytically useful if the denominator and model are clear, but they should not be phrased as “percent of fish in the sea” merely because that sounds intuitive.
Marker choice and taxonomic blind spots
Short markers can be robust for degraded DNA yet provide limited discrimination among relatives. Longer markers may offer additional differences but can be harder to recover from degraded material. Different taxonomic groups require different primer designs, and a marker set optimized for fish may poorly recover marine invertebrates or microorganisms. A biodiversity survey has a scope defined by its assays as well as its geographic map.
Consider a report with 45 detected fish taxa using a fish-focused marker. It cannot be rewritten as “45 marine species lived at the site,” because the assay may not be intended to detect corals, algae, bacteria or many invertebrates. Even within fishes, incomplete reference coverage and bias affect the count. Scope and taxonomic resolution are necessary qualifiers.
Multi-marker approaches can broaden observation but create new reconciliation problems. Two markers may detect the same organism under different taxonomic names or at different levels of resolution. Counting all assignments together can double count. Keep marker-specific outputs and describe the harmonization rule before estimating combined richness.
Transport, mixing and the source-location problem
DNA moves with currents, particles, tides and waves. Its distribution reflects production, release, transport, degradation and removal. A sample may contain DNA originating outside the station’s immediate vicinity. The spatial scale of plausible source regions depends on environmental context and molecular form. Therefore, a detection can be strong laboratory evidence yet weak evidence for exact animal location.
Suppose a fish eDNA signal occurs in a current flowing from a nearby estuary. The source might be local fish, upstream organisms or material carried from the estuary. A simultaneous current measurement, repeated upstream-downstream sampling and biological surveys can help investigate these alternatives. The DNA measurement alone does not tell which source explanation is correct.
Avoid phrases such as “the fish was swimming underneath the sampling vessel” unless independent observation supports that claim. “Fish DNA was detected in water collected at this position” keeps the sampling coordinate attached to the bottle rather than attaching a definite coordinate to an unseen animal. That change in wording preserves the distinction between a sample location and a biological source location.
Persistence and the meaning of recent presence
DNA degradation can occur through enzymatic activity, sunlight, microbial processes and other environmental effects, but no single universal decay time applies to every marine setting. Material associated with cells or particles may behave differently from freely dissolved fragments. Sediment resuspension can also reintroduce older DNA into the water column. The clock for organism presence is not stamped onto every detected sequence.
Consider a storm followed by an unusual DNA pulse. It could reflect organisms present during the storm, changing water masses, sediment resuspension or altered sampling efficiency. Without controls or independent measurements, assigning the pulse to “newly arrived animals today” may be unjustified. An experiment quantifying decay in one environment need not transfer directly to another with different temperature, salinity or sunlight.
A cautious summary says the observation is consistent with target DNA being present in sampled water at collection time. It separates that fact from assumptions about when or where the DNA originated. More precise recency claims require validated decay and transport models and an understanding of particle or sediment storage.
Mock communities reveal observation bias
A mock community contains known DNA constituents or biological material and allows a method to be tested against known input composition. It can reveal failures to recover certain taxa or distortions in read proportions. But a mock mixture cannot perfectly reproduce every inhibitor and physical DNA form in wild seawater. Its role is calibration and diagnosis, not an unconditional certificate that all future samples are unbiased.
Suppose a mock mixture contains equal input marker copies from four taxa, yet the results are 40, 30, 20 and 10 percent of selected reads. That is evidence of net differential recovery under the test conditions. It does not automatically mean the same bias ratios hold at all concentrations or with every co-occurring species. The method team should investigate primer mismatch, amplicon properties and PCR settings.
NOAA-affiliated research has examined species-specific sequence observation bias in mock communities. A faithful translation calls those outputs “relative read observations under the tested protocol,” not direct animal proportions. When correction factors are proposed, the report should state their calibration domain, the marker and the uncertainty of the corrected estimates.
Internal DNA standards and quantitative calibration
Some metabarcoding designs add a known amount of external DNA as an internal standard. Comparing sample-derived reads with standard-derived reads may support a quantitative estimate under validated assumptions. Standards can help correct certain effects of variable sequencing allocation and amplification competition. They do not automatically correct every primer mismatch, extraction loss or differential DNA shedding in the environment.
In a simplified teaching mixture, 1,000 known standard copies yield 200 reads while a target yields 100 reads. Under an explicitly assumed equal detection and amplification response, the target might be estimated at 500 copies in the analytical input. That calculation fails if the target and standard have different amplification efficiencies. The ratio must be calibrated rather than treated as a universal stoichiometric law.
A marine fish metabarcoding study archived by NOAA evaluated a 12S internal-standard approach under specified conditions. Its quantified range and performance should not be generalized to unrelated primers or other taxa. Good scientific communication describes the demonstrated scope of the method, not an invented promise that any number of reads can be converted into an accurate fish population size.
Rare reads, thresholds and false-positive trade-offs
A study may set a minimum read count, replicate support or frequency threshold before accepting a detection. Raising a threshold can reduce some spurious positives but can also remove genuine rare taxa. Lowering it can capture more weak signals while increasing the need to investigate contamination and assignment errors. No threshold is universally optimal; it should follow the assay’s validation and intended use.
Imagine a fictional acceptance rule requiring at least ten reads in two independent PCR replicates. A taxon with 50 reads in only one replicate would not satisfy that rule, while one with 12 and 14 would. The rule does not prove the second taxon is truly present; it defines a reproducible analytical acceptance criterion. Its error rates should be assessed through positive and negative controls.
A result table should preserve both underlying read observations and the final detection call when practicable. Publishing only “present” or “absent” hides the reason for the choice and may make later reanalysis impossible. Conversely, presenting every single unfiltered read as a biological record makes technical artifacts look ecological. Transparency about the threshold connects rigor with useful plain language.
Diversity metrics need sampling context
Species richness, ASV richness and diversity indices answer different mathematical questions. Observed richness is sensitive to sampling effort and sequencing depth, while abundance-weighted indices use proportions in the selected observational units. Changes in any metric can reflect ecological shifts, marker and bioinformatics effects or different detection effort. The report should not treat an index as a direct inventory of all local biodiversity.
Suppose Sample A contains 20 assigned taxa and 50,000 quality-filtered reads, while Sample B contains 15 assigned taxa and 5,000 reads. The difference in observed richness may partly reflect sampling depth. A richness comparison needs a strategy that recognizes detection effort and compositional limitations. Rarefaction or model-based approaches have assumptions and trade-offs; they are analytical decisions, not magical corrections.
Even when two samples have equal read totals, variation in original water volume, extraction efficiency and amplification can remain. A map of richness should state whether the metric counts ASVs, species assignments or higher taxa. “More biodiversity” may be too broad when the underlying evidence is “more classified sequence variants under this workflow.”
When to use detection probabilities
A sequence table describes observations, but policy questions often concern the chance that a target is present or absent. Repeated surveys with validated detection models can estimate uncertainty around such claims. The framework must account for sample independence, false positives, and transport where relevant. A detected signal can be strong enough to justify further investigation without proving every detail of habitat occupancy.
Take a hypothetical assay with a ten percent false-negative chance per independent collection when DNA is present. With two truly independent collections, the probability both miss is 0.1 × 0.1 = 0.01, or one percent. But if the same sample-handling failure affects both, their detection errors are correlated, so the calculation underestimates uncertainty. Models need realistic dependencies.
In a conservation briefing, a negative result can be described as “not detected under a method with stated sensitivity,” and a positive as “detected and confirmed under the specified decision rule.” Both sentences invite readers to ask about evidence. Neither turns the assay into an all-seeing census. This calibrated language is especially valuable when management consequences are substantial.
A complete worked exercise with sequencing proportions
An invented sequencing table has 12,000 quality-filtered reads. Of these, 9,000 are assigned to the target animal group, 2,000 to other groups and 1,000 remain unassigned. Within the full quality-filtered pool, the target represents 9,000 / 12,000 = 75 percent. Within the assigned pool alone, it represents 9,000 / 11,000, about 81.8 percent. Both percentages are numerically correct, but their denominators differ.
If a summary reports “82 percent fish,” first ask which pool was used and whether assignment quality supported that group. It still must not be translated as 82 percent of animals or biomass. The analysis has quantified the composition of recovered reads. Sample DNA copies and organism abundance require separate, validated inferential steps.
A useful troubleshooting habit is to reconstruct every category omitted from the denominator. Unassigned sequences, controls, chimeras, low-quality reads and off-target groups may have been removed at different stages. Without this provenance, two percentages can appear contradictory even though they summarize the same original run in different ways.
A worked exercise with PCR replicates
A seawater extract is amplified in four PCR wells. The taxon is detected above the study’s read threshold in wells one, two and four, but not in well three. The observed amplification-replicate detection frequency is three out of four, or 75 percent. It is not automatically a 75 percent chance the taxon exists at the field station. It describes replicate outcomes from one extraction.
If the study’s decision rule requires at least two positive wells, this sample passes under that rule. To claim station-level occupancy, however, the report must consider whether other field samples were collected and whether the detected DNA could be transported. A single extraction divided among wells cannot provide independent evidence about three separate locations or dates.
A technician who merges all four wells before archiving results may lose information about stochastic detection. Preserve replicate identifiers and counts. An original detail that looks cumbersome in a spreadsheet often becomes decisive when investigating contamination or low-concentration targets. The most useful technical translation is the one that keeps the evidence reconstructible.
A worked exercise with field contamination
A set of eight environmental bottles has target detections in seven; the field blank also shows the same target sequence. A naive summary might celebrate seven detections and omit the blank. That is unacceptable because a control detected material it should not have contained. The immediate question is whether contamination was systemic, sporadic, or linked to handling. No fixed subtraction of one contaminated read from every bottle follows automatically.
Investigators might review batch controls, read abundances, known positive sources, index cross-talk and field practice. A predeclared quality policy may classify some observations as indeterminate and require recollection. If the evidence supports a particular correction, it must be documented. The result of this investigation should be a revised confidence assessment, not a story that ignores uncomfortable controls.
A transparent sentence is: “The target sequence appeared in seven environmental samples and one field blank; therefore, the ecological interpretation remains conditional pending contamination assessment.” That sentence accurately conveys the data and what they do not yet establish. It is more useful to decision-makers than asserting seven sites were unquestionably occupied.
A worked exercise with inhibitor dilution
A molecular assay’s internal positive control fails in an undiluted extract but succeeds after the extract is diluted tenfold. This suggests inhibition was reduced, but it also means the target molecules were diluted. If the undiluted extract originally contained an estimated 200 target copies in the tested aliquot under an idealized assumption, an otherwise equivalent tenfold dilution would contain about 20 copies in the same aliquot volume before other losses.
The detection sensitivity after dilution depends on the validated assay. A negative after dilution cannot automatically be called ecological absence, particularly if the target was rare. Conversely, the successful internal positive control does provide evidence that one important technical barrier was removed. State both consequences so readers do not confuse control recovery with proof of target absence.
If an analyst converts diluted qPCR concentration back to original extract concentration, the dilution factor must be applied once and in the correct direction. For a measured 15 copies per microlitre in the tenfold-diluted extract, the inferred original concentration is approximately 150 copies per microlitre under complete mixing and valid quantification. This arithmetic does not repair differential extraction loss.
Independent transfer problems for the reader
Problem one: an experiment sequences 30,000 quality-filtered reads; 18,000 are assigned to a particular taxonomic group. What share of the filtered reads does this represent? The answer is 60 percent. Explain why it is not a count of animals. Problem two: a target is detected in one of three PCR wells from a single water sample. How many independent field samples is that? The answer is still one.
Problem three: a model assumes independent per-sample detection probability 0.5 when DNA is present. What is the probability of missing the target in four independent samples? It is 0.5^4 = 0.0625, or 6.25 percent. Explain why shared inhibition or transport may invalidate the independence assumption. Problem four: a 12S marker does not distinguish two sister species. What should the report say? Assign the shared supported taxonomic level, not an arbitrary species.
Problem five: two surveys each report 500 target reads, but the total read pools are 5,000 and 50,000. The relative proportions are ten percent and one percent. Does that mean target population density fell tenfold? No. The denominator changed and the workflow may have changed. These exercises teach the central skill: preserve the measured unit and the supported inference, not the most dramatic headline.
How to make results understandable without overstating them
A responsible summary names sampling location and date, marker and primer, field and laboratory controls, filtering and classification pipeline, replicate support, and limits on interpreting reads as abundance. It can be conversational: “We detected a fish-associated DNA sequence in this water sample, and repeat amplifications supported the call. This does not tell us exactly how many fish were nearby.” The straightforward wording is scientifically stronger than an unqualified population estimate.
The first question to ask of a surprising result is not “How many animals does this prove?” but “What material was collected, which marker was sought, what passed the laboratory controls, and which taxonomic name is defensible?” A second question is whether currents or sample handling could offer an alternative account. These questions do not undermine eDNA. They make it a credible, reusable scientific tool.
If readers can distinguish physical water samples, independent field replicates, extraction replicates, PCR wells, sequencing reads, ASVs and taxonomic assignments after reading the report, the translation has achieved its central purpose. Each term refers to a different stage or unit. Protecting those units allows scientists, students and policy readers to understand what a result establishes and where additional evidence is needed.
Primary evidence, protocols and related learning
First-party and specialist sources: USGS eDNA field, laboratory, controls and interpretation guidance; USGS environmental DNA monitoring and detection probability; NOAA-hosted research on 12S metabarcoding internal DNA standards; NOAA-hosted research on marker-dependent observations of biodiversity; and USGS study of controls and contamination. Every numerical scenario is independently constructed for teaching and is not a field observation.