Study 3 · Computational reanalysis · 19 September 2026

Nearby GATC additions have larger expression gains in a plant reporter library

Across 153 held-out DNA fragments, adding a regulatory motif near another copy was associated with an 8–10% larger stepwise expression ratio. The association survived sequence and saturation controls in two reporter backbones.

Scope: a new, reproducible analysis of published experimental data. Novelty is provisional. This is an association within synthetic addition trajectories; it does not establish a universal spacing rule, optimal distance, physical cooperation between proteins, or improved crop performance. A separate native-motif removal check was inconclusive.

The contribution. The source paper established that GATC copies enhance expression. This analysis asks whether the location of the next copy relative to existing copies carries information beyond copy number, the parent sequence and baseline-dependent saturation. It provides a held-out estimate, exact sequence checks and a reusable table of adjacent mutation steps.

What was measured

Voichek and colleagues (2024) tested 160-base Arabidopsis fragments in plant reporter constructs. Their engineered library includes successive replacements of native eight-base segments with CAGATCTG, measured inside a reporter intron in Arabidopsis protoplasts using 35S and pTRP1 backbones. Supplementary Tables 3–4 supply sequences and activity values. The study reports three experimental replicates; we use the deposited aggregate activities, not individual replicate measurements.

For each eligible step we subtract the preceding construct’s log₂ activity from the current construct’s value. We locate all matches to the published consensus YVGATCBR directly in both sequences, including overlapping matches. A step must retain every existing motif and add exactly one. We compare new motifs whose start lies within 16 bases of the nearest existing motif against those farther away. Distances are start-to-start: 16 bases corresponds to an eight-base gap between non-overlapping eight-base motifs.

All eligible changes lie inside the intended eight-base replacement. Their coordinates agree with the synthesis labels, which use zero-based coordinates. The original label convention was corrected during an auxiliary audit; the primary analyses always derived positions directly from sequence.

A failed prediction came first

We initially predicted that close copies would interfere. The chromosome-1 pilot contradicted that prediction: estimated close-copy effects were +0.136 and +0.159 log₂ units. We retained the failed hypothesis as journal entry 023 and explicitly labeled the opposite hypothesis as exploratory. Before examining chromosomes 2–5, we committed and pushed the validation protocol, preserving the same distance threshold and model. Both backbones had to show an effect of at least +0.10 log₂ with a parent-cluster t statistic of at least 3.

The pilot contained 451 eligible steps in 65 fragments. Confirmation used 1,065 steps in 153 different fragments from 150 genes: 457 nearby additions and 608 more distant additions. Together, the analyzed trajectories comprise 1,516 steps in 218 fragments. The original table contains 220 addition-series parents; sequence and paired-outcome filters determine eligibility. All exclusions are available as a table.

The held-out result

BackboneEffect, log₂95% intervalStep-ratio differenceTwo-sided p
35S0.1350.081 to 0.1909.8%2.08e-06
pTRP10.1130.043 to 0.1828.1%0.00165

Intervals and p values above cluster by source gene, the more conservative unit when multiple fragments share a gene. The frozen parent-cluster test also passed in both backbones. These two backbone measurements come from the same experimental system and are not independent species replications. P values are nominal for the single frozen contrast in each backbone; the discovery process included earlier rejected hypotheses.

The percentages are 100 × (2ᵝ − 1), where β is the close-versus-distant contrast in log₂ step ratios. They are not a universal increase in whole-plant expression or a percentage of the motif’s total benefit.

Points show close-copy coefficients; bars show 95% intervals. The chromosome-1 row is exploratory. Confirmation and control analyses use untouched chromosomes 2–5, except where the sampling unit changes to source genes.
Points show close-copy coefficients; bars show 95% intervals. The chromosome-1 row is exploratory. Confirmation and control analyses use untouched chromosomes 2–5, except where the sampling unit changes to source genes.

Controls and an important limit

The primary model uses parent-fragment fixed effects, addition-step categories, GC-count change and new-motif position. Parent clustering accounts for adjacent steps sharing measurements. An independently implemented within-parent estimator reproduces the original full-dummy regression coefficients to numerical precision.

Baseline-dependent saturation controls use native-fragment activity from the separate MIX1 experiment, interacted with addition step, and native motif count interacted with step. Further controls account for the overwritten eight-base sequence’s mono- and dinucleotide composition, the number of changed bases, curved position effects and proximity to the fragment edge. The resulting estimates are +0.148 log₂ for 35S and +0.122 for pTRP1. Parent bootstrapping with 2,000 resamples yields intervals [0.100, 0.201] and [0.057, 0.182]. Source-gene clustering gives similar uncertainty.

The association remains positive in the zero-native-motif subset and after leaving out each validation chromosome. All predeclared alternative distance cutoffs (12, 20, 24 and 32 bases) produce positive estimates. These are sensitivity analyses, not independent discoveries or evidence that 16 bases is an optimum.

Secondary analysis: distance-bin coefficients relative to additions more than 40 bases from the nearest motif, with full covariate adjustment and source-gene intervals. The curve describes this library’s sampled distances; it does not isolate spacing experimentally.
Secondary analysis: distance-bin coefficients relative to additions more than 40 bases from the nearest motif, with full covariate adjustment and source-gene intervals. The curve describes this library’s sampled distances; it does not isolate spacing experimentally.

We also examined exact single-base GATC-to-GATA losses in native multi-motif fragments, using a separately frozen extension. Among 124 parent fragments absent from the addition set, the nearby-loss contrasts were −0.036 log₂ for 35S (95% interval −0.138 to +0.066) and −0.096 for pTRP1 (−0.199 to +0.007). Neither interval excluded zero. Restricting to the spacing range represented in the synthetic design did not establish replication.

The complementary removal experiment is inconclusive. Negative estimates point in the predicted direction, but the non-overlapping-parent confidence intervals include zero in both backbones. This prevents a claim that the addition result generalizes to native regulatory architecture.
The complementary removal experiment is inconclusive. Negative estimates point in the predicted direction, but the non-overlapping-parent confidence intervals include zero in both backbones. This prevents a claim that the addition result generalizes to native regulatory architecture.

Why the result may be useful

For this library, motif count alone misses a reproducible association with the incremental response. A plant-design study can therefore include nearest-copy distance as a candidate predictor and explicitly vary it in a future controlled experiment. The modest estimate also supplies an effect-size target for planning that experiment. The released paired-step table can support evaluation of sequence models without treating thousands of correlated mutations as independent biological samples.

Spacing was not isolated through matched randomization: replacements also change native sequence, and the design’s existing motif positions constrain later placement. Statistical adjustment cannot exclude every sequence feature or assay-specific technical effect. There are no new wet-lab measurements, direct binding measurements or independent laboratory validation here. The author code archive references replicate-level MPRA tables that are not included in that archive; this reanalysis starts from the published supplementary aggregates.

Novelty and closest prior work

Two targeted searches, source-paper and supplementary-method review, an audit of the deposited mutation notebook, and a screen of 45 Europe PMC citing records found no earlier report of this exact paired-step spacing estimate. That supports provisional novelty of the narrow reanalysis, not proof that nobody has done it.

The complete review records exact queries, scope and access limitations. Non-indexed or inaccessible work could overlap. Publication readiness and value still require external scientific review.

Reproduce and inspect

Download the CPU reproduction package · Archive checksum and inventory. The package downloads the 12.7 MB source workbook, verifies its SHA256, and regenerates the pilot, confirmation, controls, native-removal check and figures. It includes the report and instructions. No GPU is needed.

python3 -m venv .venv
.venv/bin/pip install -r requirements-study.txt
PYTHON=.venv/bin/python BIO_DISCOVERY_CACHE=/path/to/cache ./run_study.sh

Exact sequences and paired measurements · Excluded steps · All model estimates · Distance-curve data · Native-removal data · Completion verification.

Data credit: Yoav Voichek, Gabriela Hristova, Almudena Mollá-Morales, Detlef Weigel and Magnus Nordborg. Source article/data: CC BY 4.0. This derivative analysis preserves attribution and does not imply author endorsement. No paper has been submitted and no researcher has been contacted. Both earlier completed studies and the first 20 journal entries are preserved.