DSB Sites

The boundaries got stronger, and not just next to the breaks

Computational analysis & wet-lab supportMcCord Genome Labwith Heng LiFall 2021 – Summer 2022AURA-funded analysisHi-C insulation-score analysis, cell culture, qPCR, gel electrophoresis

158 windows across 79 AsiSI cut sites72% read stronger in the damaged conditionDirection independently confirmed against Heng Li's dissertation

Context

This was the first thing Heng handed me — before the CTCF poster, before I knew what any of it meant. I worked with him on how the 3D genome structure responds to DNA damage from radiation, aimed at understanding how these disruptions relate to diseases like cancer.

Heng's research compares two DNA states — mOHT (undamaged) and pOHT (damaged) — where damage is induced at specific double-strand break (DSB) sites by the AsiSI restriction enzyme, so the same ~80 places break every time. The research examines how CTCF and the cohesin complex maintain TAD (topologically associating domain) boundary strength around those sites.

Question

When you break the DNA, does the boundary next to the break change — and which way?

Not whether something happens. The direction. A boundary could tighten as repair machinery arrives and holds the region still, or loosen as the structure comes apart. Both are plausible before you look — and the direction is the part I got wrong for years.

Approach & rationale

I split my time between data analysis and wet-lab support. For the analysis, I compared insulation scores between the damaged and undamaged states across matched genomic windows, using Excel and R to build box plots.

In the wet lab, I supported a parallel starvation arm of the project — culturing GM12878 cells and running qPCR and gel electrophoresis to test whether starving cells before DNA damage changes their response. That arm is what later became the CTCF poster.

Methodology & iteration

For each of the ~80 AsiSI-cut DSB sites I took the two flanking TAD-boundary windows — the nearest boundary and the next one out — and pulled the Insulation Score for each in both conditions, from Hi-C data at 40 kb resolution. 79 sites × 2 windows = 158 differences. No modelling, no fitting: a subtraction, done 158 times.

Attempt — read the direction off the box plot and call it a result

TriedTaking the majority lean across 158 windows as the finding, and reporting it as a weakening of TAD boundaries near damage.

AssumedThat a clear majority leaning one way was the same thing as an effect — and that the subtraction ran in the direction my write-up described.

Fell shortBoth were wrong. 72% of 158 is a description of a sample, not a test against chance. And the spreadsheet computes mOHT − pOHT, undamaged minus damaged, so the 114 negative values mean the damaged condition read higher — the opposite of what I reported.

Held upRe-deriving the sign from the formula in the sheet, then checking the corrected result against Heng's own analysis of the same model.

In short

Across 158 windows the differences lean one way: 114 of them (72%) read higher in the damaged condition, against 44 that read lower.

Results

Across 158 windows the differences lean one way: 114 of them (72%) read higher in the damaged condition, against 44 that read lower. The median shift is 0.051, the mean 0.065 — small numbers on scores that sit around 1.0.

Higher boundary strength means fewer contacts crossing, which means a stronger boundary. So the answer is that after the DNA is broken, the boundaries around the break tighten rather than come apart.

Splitting by which window it was is where it gets interesting. The boundary nearest the break strengthens in 66% of sites; the one further out, in 78%. If the effect were driven by proximity to the break you would expect the opposite ordering — and you don't get it.

Significance

Proposed for presentation at the EUReCa undergraduate research exhibit under the AURA award.

Reflection

This ran alongside the glucose-starvation work as part of the same AURA-funded collaboration with Heng Li — moving between the wet-lab and computational sides of one shared question about how DNA damage reshapes the genome around it.

It is also the project where I learned the difference between a number and a finding. I had 158 differences and a clear lean, and what I did not have was any right to call it an effect.

BackAll work NextCTCF →