Who Coordinates Two Metres of DNA?

1. Four years, 53 minutes, ten hours

The question began with a scale analogy: if DNA were a railway, its length would be enormous relative to the proteins working on it. Who synchronizes all those workers? How could information travel quickly enough for the entire job to finish?

These calculations use 6.4 billion base pairs, a circular representation, and a constant fork speed of 1,500 base pairs per minute. Both 40,000-origin examples are fully synchronized. Placement alone changes their result. The genome length and eight-hour endpoint are comparison settings, not measurements of the particular HCT116 samples later examined. [2]

99.98% replicated. Zero of 200 genomes complete.

That separate result comes from the finite-resource model at eight hours: 400,000 candidate origins, capacity of 16,000 effective fork units, a 30-minute recovery wait, and a constant initiation hazard when capacity is available. The percentage is the mean across 200 synthetic layouts; zero of 200 counts layouts that finished every base. This contrast became the central lesson of the investigation. [1]

2. The explanation that the calculation overturned

The initial argument divided genome length by origin count, then divided half that average spacing by fork speed. Because this gave about an hour, it seemed to show that an eight-hour replication process could not be fully synchronized. The calculation was correct for regular spacing. The inference was wrong: average spacing had silently replaced the longest interval that needed to be copied.

For simultaneous initiation on the idealized circle, T_complete = max(g_i)/(2v), where g_i are neighboring-origin gaps. For N uniformly and independently positioned origins, the expected largest gap is (L/N)H_N, with H_N the harmonic sum. At N=40,000, that gap is about 1.79 million base pairs and the expected completion time is 9.93 hours. The saved 200-layout simulation had a median of 9.71 hours and only 2/200 completions before eight hours. The analytical expectation and the finite-sample median are different statistics. [1] [2]

A second control placed one origin at a random position inside each equal-sized segment. Its median completion time was 1.77 hours. This stratified arrangement reduces large gaps, but it is not the perfectly equally spaced arrangement responsible for the 53-minute calculation.

The correction matters beyond this example: a completion time cannot, by itself, identify whether origins synchronized. The model had exposed a hidden assumption in the explanation. It had not discovered the random-completion problem, which was already addressed in the replication literature. [3]

3. Methods: separate geometry, initiation, and capacity

The initial comparison evaluated seven conditions over 200 paired layouts, or 1,400 trials. Candidate origins were fixed before replication, could initiate only while unreplicated, and were passively inactivated when an approaching fork copied them. Fork speed was constant; the genome had no chromosome ends, stalls, or explicitly resolved nuclear geometry. No origin inspected the global completion fraction or received a deadline instruction.

The finite-resource extension used 400,000 candidates. Initiation consumed two capacity units; when opposing forks terminated, their units entered a recovery wait before becoming available again. Nine capacities, nine waits, and two initiation laws, plus an effectively unlimited control, produced 32,600 trials. One law maintained an initiation hazard of 0.001 per minute whenever two units were available. The other multiplied that hazard by the free fraction of capacity. A unit represented effective fork-support capacity, not a measured protein molecule or an intact replisome recycled unchanged. [1]

Geometry and resource shortage were then checked separately. A candidate-free gap longer than 1.44 million base pairs cannot be filled in eight hours at the assumed speed even if both ends start immediately. None of the 200 dense, 400,000-candidate layouts violated this bound, and their paired unlimited-resource controls all completed. In the separate 40,000-candidate random model, 198/200 layouts were geometrically unable to finish by the deadline. The spacing of candidates must not be confused with the spacing of origins that actually fired. [2]

4. Almost complete is a different endpoint

At capacity 16,000 with a 30-minute wait under the constant-hazard availability gate, the mean replicated fraction at eight hours was 99.9795%. Nevertheless, all 200 layouts still contained unreplicated DNA. The remaining total averaged 1.31 million base pairs; the largest remaining gap in each layout averaged about 229,000 base pairs. Median full completion occurred at 551.00 minutes, or 9.18 hours. The striking percentage is therefore neither an observed cellular replication fraction nor a claim that exactly 0.02% remained in every trial. [1]

Total work provides a useful lower bound. Copying L bases at speed v requires L/v, approximately 4.267 million fork-minutes. To finish in 480 minutes requires about 8,889 active forks on average. But this accounting identity does not explain how origins schedule themselves, nor does it supply enough capacity for startup delays, time spent waiting for recovery, or low utilization while the final gaps close.

Required capacity also depended strongly on the initiation law. With immediate reuse, the smallest tested capacity achieving at least 95% observed completion was 24,000 for the availability gate and 80,000 for the free-fraction law. Requiring the lower limit of a Wilson 95% interval to exceed 95% raised the latter threshold to 128,000. These are discrete tested boundaries, not universal protein requirements. In the free-fraction law, changing capacity while preserving the initial hazard does not hold a microscopic binding coefficient fixed. [1]

5. What resource reuse can explain

Recycling creates an indirect interaction: finishing one region returns capacity to a shared pool, changing the initiation opportunities elsewhere. This supplies a computational way for distant activities to coordinate without individual commands. The pool is well mixed, so it does not establish actual nuclear transport times, target-search times, or concentration gradients.

A subsequent factorial comparison varied initiation hazard, initial resource loading, and return delay across 12 conditions and the same 200 layouts. At capacity 24,000, full-pool initiation hazard 1 per minute, and a 30-minute initial-loading time scale, immediate reuse gave 200/200 completions by eight hours and a median of 4.27 hours. Changing only the return wait to 480 minutes gave 0/200 completions by that deadline. This is a different initiation law and parameter set from the 99.98% example above. The long wait prevents reuse before the deadline; it does not prohibit reuse forever. [2]

Reuse was insufficient with slower initiation. Moreover, none of the 12 conditions passed both the predefined supported-range hump descriptor and the observed 95% completion criterion. Successful fast conditions did show a raw decline after a sharp peak, but that decline occurred outside the predefined range with at least 1% unreplicated-DNA exposure. The negative result concerns that specific joint criterion, not the existence of any peak. The detailed comparison remains in A Peak Is Not a Mechanism.

Arbona and colleagues had already shown how localized candidate origins and recycling can account for initiation-rate dynamics under appropriate assumptions. Fan and colleagues supplied human-cell evidence that impaired replisome disassembly can retain components on chromatin and compromise replication. These are prior results supporting the relevance of reuse. They do not identify the effective capacity, initiation kinetics, or normal recovery wait in this model. [3] [4]

“Late acceleration” also needs an explicit denominator. I(t) measures new initiations per unit time per unit length of unreplicated DNA. Increasing I(t) is not equivalent to accelerating total DNA synthesis, and a bell-shaped curve eventually declines. This study did not establish that recycling alone reproduces late-S dynamics and eight-hour completion in matched HCT116 conditions.

6. A missing denominator and a missing clock

The public-data audit examined HCT116 flow-cytometry files (FCS) and Nanopore data associated with Hossain and colleagues. Across 12 Nanopore samples, it recovered 362 origin calls, 270 termination calls, and 3,544 filtered fork records, and reproduced the 24 published origin/fork and termination/fork ratios. The relative metrics were reproducible. The question was whether they could support an absolute initiation rate. [2] [5]

For a time bin, the required rate is new initiations within the bin / ∫ U(t)dt, where U(t) is the relevant unreplicated DNA length. The origin-containing reads and the filtered fork reads used in the reported denominator were different selected sets. An origin call may also represent an origin that fired before the labeling window. Neither fact makes the published relative metric erroneous; both prevent a direct reinterpretation as initiations per unreplicated DNA length per minute.

Searching the processed distributions did not recover a complete analog-negative and analog-positive read population. The FCS audit likewise did not establish a same-aliquot mapping to the Nanopore samples. Most decisively, the asynchronous data lacked an identified clock measuring time since S-phase entry. Acquisition timestamps and DNA content could not simply be relabeled as that clock. Static total or chromatin-bound factor measurements also did not identify the time required for a released factor to become functionally reusable. [2]

A separate synthetic control demonstrates why the clock matters. One layout was replayed after all kinetic time scales were doubled. The curves of I(t)T against t/T remained identical, while completion changed from 4.394 to 8.788 hours. One met the deadline and one missed it. I(t)T has units of inverse length; it is not dimensionless. These were two replays of one layout, not independent biological experiments. An absolute clock or appropriately matched speed measurement could constrain this freedom. Relative curve shape alone could not. [2]

7. Biological correspondence, communication, and rescue

The last-gap problem has a biological counterpart. Letessier and colleagues found a roughly 700-kilobase region lacking initiation events at FRA3B in lymphoblastoid cells, together with mid-S firing in flanking regions and vulnerability to incomplete replication when forks slowed. The initiation pattern and fragility were cell-type specific. This supports the relevance of long distances between active starts; it does not mean the present random-circle model reproduces that fragile site or explains all fragile-site biology. [6]

Similarly, “no individual progress controller” does not mean “no communication.” DNA contour length is not the same as transport distance through the folded nucleus. Local molecular interactions, shared factors, cell-cycle regulation, and checkpoint signaling can all matter. ATR–CHK1 regulation of replication factories is one example of real communication beyond an isolated fork. The present construction establishes that global progress commands are unnecessary for its own completion behavior, not that cells lack global regulation. [7]

When difficult regions remain underreplicated beyond S phase, mitotic DNA synthesis (MiDAS) can provide a rescue route. Minocherhomji and colleagues showed MUS81-associated, POLD3-dependent DNA synthesis at common fragile sites after replication stress, reducing chromosome segregation defects. This does not guarantee recovery of every remaining gap and cannot establish that S phase itself finished within eight hours. MiDAS was not simulated in this investigation. [8]

8. Why the investigation stopped

The final audit located a recovery route for about 25 GB of original BAM data in cold storage. More records might improve the observable DNA population and, with suitable calibration, support an exposure-weighted average initiation rate in an asynchronous population. They would not by themselves add an S-phase age axis. Without that axis, the proposed recovery would not decide whether initiation accelerated late enough to close the final gaps on time.

The investigation therefore stopped. No author inquiry was sent, no cold-storage restoration was initiated, and no new model fit was added to make the remaining uncertainty look resolved. This was a decision about the information needed for the original question, not a claim that every possible future experiment would be uninformative. [2]

The process had drifted twice. First, the genome-wide question narrowed into identifying coordination inside a single replication machine from short observations. Later, the absence of a usable clock became a sequence of increasingly detailed file-location tasks. Each step could be technically reasonable while the overall work moved away from the question. Human critique restored the scope and supplied the stopping decision. The separate observation-design article records the narrower synthetic study; it is not evidence for genome-wide completion.

This is a documented case of AI-assisted research, not a controlled evaluation of AI research performance. Its practical lesson is to state, before another task, which possible result would change the answer to the original question. If neither a positive nor a negative outcome would change that answer, the task may add detail without adding discrimination.

9. What this study established

No new biological mechanism or scientific novelty was established. Resource recycling, stochastic completion, and late unreplicated regions already have substantial prior literature. The useful outcomes here were narrower: a compelling but incorrect explanation was caught by calculation; the difference between near-completion and full completion was quantified; and public measurements were audited before being assigned an absolute interpretation they could not support.

The answer to the original question is consequently conditional but substantial. Many local processes can coordinate through shared resources and cellular regulation without a controller issuing an instruction to every origin. In the model, sufficient candidates and initiation opportunity allowed full-length completion; inadequate timing or resources left terminal gaps. Biological evidence supports a role for reuse and for rescue pathways, but this work did not jointly validate late-S I(t), recovery time, and whole-genome completion under one HCT116 condition.

The article is a synthesis of retained computational records, a bounded public-data audit, and published biological context. It publishes reviewed aggregate supplements and source hashes, not the complete raw run archive or an executable reproduction environment. Hashes identify retained inputs; they do not make unavailable raw data independently inspectable. No new simulation, biological experiment, or empirical fit was performed for this update. Earlier model summaries remain available in the references. [1] [2]