HopeatrarelabsVisit Website →
← Back to blog

Intron Aware ASO Off Target Workflow for Labs With SPR Checks

September 4, 2026
Intron Aware ASO Off Target Workflow for Labs With SPR Checks

The minimal defensible ASO off-target assessment is a three-layer stack: a transcriptome-wide in silico scan that includes introns and splice isoforms; transcriptome-wide experimental confirmation by RNA-seq; and focused qPCR or protein-level validation of the hits that survive. Add biophysical accessibility assays when a predicted target is structured or when the ASO chemistry changes binding kinetics enough to make Tm unreliable. This is the sequence OSWG guidance and FDA draft guidance point toward, and it is the sequence that catches what any single method misses. The mechanisms behind it dictate exactly where to tune the workflow.

TL;DR:

  • In silico transcriptome scans must include both spliced and pre-mRNA models, especially when using high-affinity chemistries like LNA, to predict off-target effects accurately.
  • Hybridization-dependent off-target effects can cause intronic cleavage by RNase H1, while chemistry-driven effects involve protein binding and immune activation, requiring different assessment methods.
  • Accurate prediction involves scoring mismatches with attention to position, wobble pairing, duplex stability, and conservation across species, especially for highly expressed, disease-relevant genes.
  • RNA-seq confirmation with multiple replicates and proper controls is essential to distinguish true off-target effects from noise, with qPCR used for validation of flagged transcripts.
  • Incorporating structured RNA biophysical assays like surface plasmon resonance helps evaluate actual binding behavior, especially for targets buried in RNA secondary structures.

Table of Contents

Mechanisms of ASO Off-Target Effects and What They Mean for Testing

Off-target activity from antisense oligonucleotides splits into two categories that require different detection strategies. Hybridization-dependent effects happen when the ASO base-pairs with an unintended RNA sequence, triggering RNase H1 cleavage or steric block at that site. Hybridization-independent effects come from the chemistry itself, protein binding, immune activation, or aptamer-like interactions that have nothing to do with sequence complementarity.

RNase H1 works in the nucleus as well as the cytoplasm, so gapmer ASOs designed against mature mRNA can also cleave pre-mRNA at intronic sequences that never make it into the final transcript. Studies confirming intronic binding show this is not a theoretical risk. It shows up in real profiling data.

Chemistry shifts the whole risk profile. Locked nucleic acid (LNA) bases raise binding affinity and thermal stability, which improves potency but also widens the mismatch tolerance window where off-target hybridization becomes possible. Phosphorothioate (PS) backbones increase protein binding and can drive hybridization-independent toxicity. 2′-O-methoxyethyl (MOE) modifications sit in between, generally trading some potency for a narrower off-target window.

What this means for testing:

  • Predict both hybridization-dependent and independent liabilities, since a clean in silico match score says nothing about protein-binding toxicity.
  • Search pre-mRNA and mature transcript models separately for any gapmer expected to recruit RNase H1.
  • Treat higher-affinity chemistries (LNA-heavy designs in particular) as needing a wider mismatch tolerance in prediction, not a narrower one.

How Do You Run In Silico Off-Target Prediction Correctly?

Sequence prediction sets the boundaries of what you will later confirm experimentally, so getting the reference set and scoring rules right up front saves weeks downstream. Start with both spliced mRNA and pre-mRNA transcript models pulled from a current annotation build (RefSeq or Ensembl), because a scan against mature transcripts alone will miss every intronic RNase H1 substrate.

The scoring itself has to reflect how ASOs actually bind, not just how many bases match. A central mismatch destabilizes a duplex far more than one at the 3′ or 5′ terminus, and a wobble pairing (G:U) partially rescues binding that a strict Watson-Crick mismatch count would flag as safe. Seed-region weighting, borrowed conceptually from siRNA off-target work, applies here too: mismatches clustered near the presumed RNase H1 cleavage window matter more than mismatches at the flanks.

A workable prediction pipeline looks like this:

  1. Align the candidate ASO against the full transcriptome (spliced plus pre-mRNA) using a BLAST-like short-sequence aligner or a purpose-built oligo screening pipeline.
  2. Score each hit for mismatch count and position, applying reduced penalties for wobble pairs and terminal mismatches.
  3. Filter by predicted duplex stability (a thermodynamic affinity estimate, not raw Tm) to drop matches unlikely to form a stable enough duplex to trigger cleavage.
  4. Cross-reference surviving hits against expression data for the relevant tissue or cell type.
  5. Rank the remaining candidates by combined binding-strength score and expression level, then flag any hits in genes under strong evolutionary constraint for extra scrutiny.

Sequence conservation across species adds a useful, often overlooked signal. A predicted off-target in a gene conserved from mouse to human raises the stakes of getting it wrong in a way a species-specific paralog does not.

Pro Tip: Don't stop at genome coordinates. Pull the actual local secondary structure around each predicted off-target site. A perfect sequence match buried in a tight RNA hairpin may never be accessible to your ASO in a cell, and this is exactly the kind of false positive biophysical assays catch before you waste RNA-seq budget on it.

Why Transcript Models and Expression Filters Matter for Candidate Lists

A raw sequence-match list from a naive BLAST search is usually enormous and mostly noise. The job of a good search strategy is narrowing that list to something a lab can actually confirm.

Intronic sequences belong in the search for any RNase H1-active gapmer design, full stop. Experimental profiling of 13-mer LNA gapmers confirmed off-target knockdown at mismatch distances up to d=2, including hits inside intronic regions that a mature-transcript-only search would never surface.

Expression filtering does most of the practical narrowing work. A perfect sequence match in a gene that is silent in your target tissue or cell line is a lower priority than a two-mismatch match in a gene expressed at high copy number in the same system. Matched cell-line RNA-seq data or a public expression atlas lets you rank predicted hits by actual biological plausibility rather than sequence identity alone.

Splice-isoform awareness catches a subtler problem: a predicted match that looks clean against one transcript isoform of a gene may sit inside an exon that a different, more abundant isoform skips entirely, changing the real-world risk substantially.

Practical mismatch cutoffs depend on ASO length and chemistry:

  • Shorter gapmers (13 to 16 mers) with LNA wings need tighter cutoffs, generally d≤2, given their higher intrinsic affinity.
  • Longer, lower-affinity chemistries can often tolerate a looser d≤3 cutoff without missing biologically relevant hits.
  • Any hit meeting the cutoff in a highly expressed, disease-relevant gene should move forward regardless of where it falls in that range.

Designing RNA-Seq and qPCR Experiments to Confirm Off-Targets

Prediction defines the universe of possible off-targets. Most of them turn out to be nothing. Industry technical assessments of off-target screening workflows consistently note that the majority of predicted sites show no measurable change on unbiased transcriptome profiling, which is exactly why RNA-seq confirmation is not optional.

RNA-seq is the right default for most programs because it is unbiased. It catches modulated transcripts you never predicted, not just the ones on your candidate list. Microarray remains usable when cost is the binding constraint and the relevant transcript set is well annotated, but it will not surface novel isoforms or unannotated transcripts the way sequencing does.

A defensible experimental design includes:

  1. At least three biological replicates per treatment condition, with sequencing depth sufficient to detect moderate expression changes in low-abundance transcripts, not just highly expressed genes.
  2. A dose and time-course structure that separates immediate primary knockdown (early time point, on-target effect dominant) from secondary and downstream transcriptional changes that accumulate later.
  3. A scrambled-sequence control matched for length and chemistry, run in parallel, to distinguish sequence-specific effects from generic chemistry-driven stress responses.
  4. A second, independently designed ASO targeting the same gene through a different sequence, since a change reproduced by two unrelated sequences targeting the same transcript is far more likely to be a true on-target or shared off-target effect than an artifact.
  5. Where feasible, a rescue or overexpression construct to confirm that a phenotype tracks with the target transcript level.

Once RNA-seq flags candidate off-targets, qPCR validation on an independent RNA preparation is what turns a statistical hit into a confirmed finding. Design primers spanning exon junctions where possible to avoid genomic DNA contamination artifacts, and always normalize against multiple reference genes rather than one, since a single reference gene can itself shift under ASO treatment.

A commonly cited screening result found off-target modulation surviving statistical correction in gapmer profiling experiments even at two-mismatch distances, a reminder that "close enough" sequence similarity is often close enough biologically too.

When Do You Need Biophysical Binding Assays?

Tm calculations assume the target RNA is a simple linear strand available for pairing. Real transcripts fold into hairpins, pseudoknots, and higher-order structures that can block an ASO from ever reaching a sequence that looks perfectly complementary on paper. Surface plasmon resonance analysis of structured RNA targets shows binding behavior at physiological temperature that frequently diverges from what melting-curve predictions alone would suggest.

SPR earns its place here because it measures kon and koff directly, not just an equilibrium snapshot. That kinetic detail matters: two sequences with identical predicted Tm can have dramatically different association and dissociation rates, and a fast koff means transient binding that may never translate into meaningful cleavage in a cell. Immobilizing the RNA target on the chip (rather than the ASO) generally preserves native-like secondary structure better, which is the whole point of running the assay.

ITC gives you full thermodynamic profiling, enthalpy and entropy separately, but it needs more material and more time per sample than SPR, which limits it to a shortlist rather than a broad screen. DSF trades some of that thermodynamic detail for throughput, making it a reasonable first-pass filter before committing sequences to SPR or ITC.

  • Reach for SPR when a predicted off-target sits inside a folded RNA structural element the sequence scan flagged.
  • Use DSF as a higher-throughput early filter across a larger candidate set before narrowing to SPR or ITC.
  • Reserve ITC for the handful of sequences where full thermodynamic characterization actually changes a design decision.

Pro Tip: A predicted off-target with strong sequence complementarity but weak or fast-dissociating SPR binding is a strong candidate to deprioritize before it ever reaches a cell-based assay. Kinetics catch what a static mismatch score cannot.

Setting Thresholds to Separate Real Off-Targets from Noise

A commonly used starting threshold is a fold change of roughly 1.5 to 2 times control with statistical support after multiple-testing correction, but that number is a starting point, not a rule. A modest fold change in a dose-sensitive gene tied to a known disease pathway can matter more than a larger change in a transcript with no plausible link to the phenotype you are chasing.

Distinguishing a direct hybridization event from a downstream ripple effect is where independent-ASO and kinetic data both earn their keep. If two unrelated ASO sequences targeting the same primary gene both produce the same secondary change, that change is more likely downstream of on-target knockdown than a shared off-target hit. A change that shows up with one sequence but not the other, at a site your prediction pipeline flagged, points toward direct hybridization.

Functional consequence assessment is the step teams skip under time pressure and regret later. A confirmed off-target transcript change only matters clinically if it moves a protein level or a pathway readout. Western blot or targeted proteomics on flagged genes, followed by pathway enrichment analysis on the broader RNA-seq change set, tells you whether an off-target hit is biologically silent or a genuine liability.

  • Flag hits with confirmed knockdown plus a plausible pathway link for immediate re-design.
  • Downgrade statistically significant but functionally silent hits to a monitoring list rather than a stop signal.
  • Reserve program stops for off-targets in genes with known toxic loss-of-function phenotypes.

Roughly experimental profiling work has repeatedly found that off-target knockdown at two-mismatch distances is common enough in gapmer chemistries that treating d=2 hits as automatically negligible is a mistake many programs make early and correct later.

Building Off-Target Assessment Into Lead Design and Prioritization

The workflow only pays off when it runs in the right order and early enough to influence sequence choice rather than just document it after the fact. A practical pipeline looks like this:

  1. Generate a broad candidate sequence set during initial ASO development and immediately run the in silico transcriptome scan, introns included, before any wet-lab work starts.
  2. Narrow the surviving candidates with expression filtering and, where structure is a concern, a DSF or SPR pass.
  3. Move the top candidates into RNA-seq confirmation with proper replicate and control design.
  4. Validate flagged hits by qPCR and, where warranted, protein-level assays.
  5. Feed the results back into sequence re-design: shift the ASO a few bases along the target, swap a chemistry at the flagged mismatch position, or adjust the gap design in a gapmer to reduce RNase H1 access at the problem site.

Re-design based on off-target hits is rarely a full restart. Shifting binding position by even a handful of bases, or changing wing chemistry to alter affinity at the specific mismatch causing trouble, often resolves the liability without sacrificing on-target potency.

Resource planning deserves a real line item in the program budget: assay panels, replicate counts, and biophysical instrument time all need to be scoped before the first candidate sequence is synthesized, not after RNA-seq comes back messy. Documentation matters as much as the data itself, since a regulatory review package needs a clear, reproducible record of what was predicted, what was tested, and what threshold triggered a re-design decision.

What Do Regulators and Industry Groups Expect From This Process?

The Oligonucleotide Safety Working Group and FDA draft guidance now frame combined in silico and experimental off-target assessment as an expected part of an oligonucleotide development package, not an optional add-on for cautious teams. That expectation has moved from best practice to standard practice over a short span, and regulatory harmonization trends suggest international alignment, plausibly including ICH-level coordination, is heading the same direction.

For a preclinical package, reviewers generally expect to see documentation of the sequence-dependent prediction approach used, the transcriptome-wide confirmation method and its statistical thresholds, and the qPCR or functional validation applied to any flagged hits.

  • Document reference transcript models used (spliced and pre-mRNA) and search parameters.
  • Report RNA-seq design details: replicate number, depth, and control structure.
  • Include the decision rule applied to distinguish a true off-target from statistical noise.

How RareLabs Applies This Workflow in Patient-Derived Models

A combined approach uses in silico off-target screening with biophysical testing and RNA-seq confirmation directly in patient-derived induced pluripotent stem cell (iPSC) models, rather than in a generic cell line that may not reflect the disease context. That matters for rare-disease programs where the relevant transcript expression pattern is tissue-specific and a generic line can miss the off-target entirely.

Running the confirmation step in cells that carry the patient's own genetic background also surfaces isoform-level and expression-level nuances a standard reference cell line would not show. Teams working on custom ASO projects for ultra-rare and undiagnosed diseases use this approach to accelerate lead selection when there is no existing precedent to lean on. Further background on how these models get built lives in the RareLabs Knowledge hub.

The Missing Pieces the Field Still Needs to Fix

Prediction tools still lean too heavily on mismatch counting and not enough on structure and kinetics, and that gap is where most false positives and false negatives both hide. Shared, standardized datasets combining sequence, structure, kinetic, and transcript-context data would let the field train prediction models that actually improve on today's heuristics, rather than each lab re-deriving its own rules of thumb in isolation.

Intron-aware and isoform-aware prediction should stop being a nice extra and become the default search mode, given how often confirmed off-targets turn out to sit outside the mature transcript. And standardized reporting formats for off-target packages would speed up regulatory review by giving reviewers a consistent structure to evaluate rather than a different homemade format from every sponsor.

None of this requires waiting on a new assay. It requires the field agreeing on how to talk about the data it already generates.

— John

Contracted Off-Target Assessment Through RareLabs

If teams would rather not build the full prediction-to-confirmation pipeline in-house, some service providers run the combined workflow directly on patient-derived cells instead of a generic surrogate line, which is the detail that matters most for rare-disease programs with no established model system.

Hopeatrarelabs

The service maps onto the pipeline described above: transcriptome-wide sequence screening including intronic search, biophysical accessibility testing (SPR/DSF) for structured targets, and RNA-seq confirmation run in iPSC-derived cells relevant to the specific disease. Expect a ranked off-target report, standard volcano and MA plots from the RNA-seq analysis, and a validated shortlist rather than a raw data dump you have to interpret yourself. For programs facing a synthesis deadline or a foundation-funded timeline, that turnaround difference is often the deciding factor. Visit the RareLabs Knowledge page to see current service scope and start a conversation about your candidate sequences.

Sources

Start with the FDA draft guidance on oligonucleotide specificity, the gapmer off-target confirmation study, and the biophysical screening strategy paper for methodology. For sequence work, NCBI BLAST remains the practical starting point for transcriptome-wide alignment searches.