Colitis at molecular resolution

New joiner briefing

Ravi Kalia

2026-09-11

Thirty minutes, five acts

English Español

English-only briefing for new joiners. Gene names, enzymes, and software stay in their published form. Press s for timed notes. After each check, say the answer before you click.

The hour

1. Clinical problem — 6 min
2. Two molecular languages — 8 min
3. This study — 5 min
4. Lab and software — 8 min
5. How you start — 3 min

Pause 20 seconds: who is wet-lab, who is computational, who is both?

Ask the room split. 1 min.

Colombia, 2024: the cancer backdrop

2 min

Act 1 · Clinical problem

IARC estimates, both sexes, all ages. These are national figures, not this project’s sample. They explain why a Colombian colitis–CRC study exists.

Do not present these as study results. Source: IARC Colombia fact sheet 2024.

Ulcerative colitis is mucosal, chronic, and not Crohn’s

Ulcerative colitis (UC) is chronic inflammation of the colonic mucosa. Blood in stool, diarrhea, urgency, and abdominal pain are common. Extent and activity vary.

What it is

Continuous mucosal inflammation starting from the rectum in typical disease. Relapsing course.

What it is not

Not Crohn’s. This protocol excludes Crohn’s, prior colonic surgery, and contraindication to colonoscopy or biopsy.

Why eight years appears

AGA/ECCO surveillance often starts ~8–10 years after onset, or earlier with primary sclerosing cholangitis. That is a surveillance threshold, not a clock that says cancer arrives at year eight.

Progression is possible. It is not the default path.

Most people with UC never enter dysplasia or colitis-associated CRC. Colonoscopic surveillance remains the clinical backbone. This project does not replace it. It asks whether blood and tissue molecules add a second, biological layer.

The research layer

Paired plasma cfDNA, colonic tissue methylation, and serum sIgA — read together, not as a diagnostic test.

90 seconds. Most patients never reach panel 4.

Check 1

Most people with ulcerative colitis develop colorectal cancer.

A · True

B · False

False. Long-standing extensive UC raises risk. Most patients never enter dysplasia or colitis-associated CRC. Surveillance exists because cancer is possible, not because it is typical.

DNA methylation is a chemical mark, not a mutation

3 min

Act 2 · Two molecular languages

A sequencer reads C whether or not a methyl group is attached. The biology lives in that extra mark: 5-methylcytosine (5mC), usually at a CpG (cytosine then guanine on the same strand).

hypermethylation = more 5mC hypomethylation = less 5mC silenced = promoter 5mC + low RNA, measured together

Candidate genes are hypotheses, not the disease

Also around UC neoplasia

CDKN2A / p16, CDH1 (E-cadherin), FOXE1 / SYNE1.

How this project must test them

Keep methylated and total counts. Use DSS. Require coherent CpGs in a region. Compare tissue with paired plasma. Adjust for inflammation and cell mixture.

Secretory IgA is a mucosal border antibody

sIgA binds microbes and antigens at wet surfaces and limits epithelial contact without turning every encounter into inflammation.

This study measures serum sIgA by ELISA, calibration 1.250–0 µg/ml in the presented run (\(R^2=0.959\)). That \(R^2\) is assay development, not plate acceptance. Blanks, recovery, replicate CV, LLOQ/ULOQ, and parallelism still decide if a number is usable.

Serum is not stool, not mucus, not luminal coating. Do not report a serum value as intestinal secretion.

The two signals answer different questions

DNA methylation

Epigenetic state and cell identity. Plasma cfDNA + colonic tissue. CpG/region readout. High-dimensional. Fails under batch, cell mixture, and mixed tissue of origin.

Serum sIgA

Mucosal humoral immunity. Serum ELISA concentration. One interpretable number. Fails under infection, drugs, systemic production, and clearance.

Integration tests whether the joint pattern separates UC, CRC, and healthy groups better than either marker alone. That is still association, not a diagnostic test.

Check 2

Match the measurement to the specimen.

A · Plasma EM-seq reads serum sIgA concentration.

B · Tissue DNA is left unsheared so fragmentomics of the colon can be read.

C · Plasma cfDNA methylation, tissue methylation, and serum ELISA sIgA come from one participant.

C. EM-seq is DNA chemistry, not an antibody assay. Tissue DNA is intentionally sheared; only native plasma fragments can support fragmentomics.

The question this grant is built to ask

2 min

Act 3 · This study

Can differential methylation in plasma cfDNA and colonic tissue, read together with serum sIgA and inflammation markers, help characterize high-risk UC relative to CRC and healthy comparators?

PCI-2024-0016 project code

18 participants planned

36 paired DNA specimens

3 clinical groups

Operational lead: Angie Alejandra Benavides Rojas. Advisor: Dr. Sandra Perdomo. Co-advisor: Dr. María Consuelo Romero. Group: InMubo.

Pairing is within a person, not between groups

UC eligibility: confirmed UC >8 years or primary sclerosing cholangitis. Sites: Cobos Medical Center, HIC Bucaramanga, GastroAdvanced.

Ethics: Colombian Resolution 8430 of 1993; HIC Bucaramanga–Fundación Cardiovascular committee Minutes 005-2024. Public git stays de-identified. No participant-level clinical or genomic data in this repository.

What this design can claim — and what it cannot

1 · Association

Signals vary together in these 18 people. This design can test that.

2 · Discrimination

A model separates predefined groups in held-out data — only with locked features and honest resampling.

3 · Prognosis

Predicts future progression. Needs longitudinal follow-up. Not this cross-section.

4 · Clinical utility

Using the model improves care versus colonoscopic surveillance. Not in scope.

This study lives on rungs 1–2.

Blood, biopsy, enzymes, sequencer, plate

3 min

Act 4 · Tools

Stage	Instrument	Stop/go
Plasma cfDNA	PANAMAX 48	Yield without group-biased failure?
Tissue gDNA	PANAMAX 48 → UltraShear	Fragmentation documented?
Methylome library	NEBNext EM-seq	Conversion controls pass?
Sequencing	NovaSeq 6000, paired-end	Depth and identity usable?
sIgA	Serum ELISA	Replicates in range?

Plasma is not sheared. Tissue is.

Why EM-seq instead of bisulfite

Bisulfite converts unmodified cytosines but damages DNA. Plasma cfDNA is already scarce and short. EM-seq uses TET2 + T4-BGT to protect 5mC/5hmC, then APOBEC to deaminate unmodified C, which sequences as T.

Standard EM-seq reports 5mC and 5hmC together. Write “modified cytosine” unless a second assay splits them.

Fragmentomics is a second layer on the same plasma library

From one paired-end plasma BAM you may recover fragment length, ends, end motifs, and regional coverage. Histones are not sequenced; packing is inferred.

This is a feasibility extension. It dies if plasma was sheared, insert sizes were not retained, depth is too low, or conversion bias is uncontrolled.

From FASTQ to a region you can name

At one CpG the data are methylated reads \(M\) out of informative reads \(N\). 50% of 2 reads is not 50% of 100. DSS uses a beta-binomial model, then joins neighboring CpGs into DMRs.

Interpretation limit

KEGG/Reactome enrichment is a structured hypothesis. Record mapping rule, background universe, database version, and FDR.

Check 3

With 18 participants, the defensible near-term product is:

A · A clinical classifier that predicts UC→CRC for a new patient.

B · A quality-controlled feasibility analysis and a small locked set of candidate features.

C · A Galleri-style multi-cancer blood test for Colombia.

B. High-dimensional classifiers, prognosis, and clinical use need independent longitudinal validation. Galleri is not validated for UC progression surveillance.

Your first week in this repository

2 min

Act 5 · How you start

python3 -m venv .venv && source .venv/bin/activate
python -m pip install -r requirements-quarto.txt
python -m ipykernel install --user --name project-python \
  --display-name "Python (methylation project)"
quarto preview

The kernel name project-python is required even for prose-only pages.

Do not commit

Participant-level clinical or genomic data, identifiers, consent forms, linkage keys.

Freeze contract

After any edit to an R document, re-render locally and commit _freeze/. CI has no R packages.

Done today means a candidate signature, not a test

Read, in this order: study design, EM-seq, sIgA, three data streams, candidate genes, statistics, contribute.

Then take the 30-question quiz.

Closing line: this is a candidate epigenetic–immune signature with transparent uncertainty — not a diagnostic test.

Appendix · Protocol chain

1. Cohort and consent
2. Blood processing
3. Colonic biopsy
4. PANAMAX cfDNA
5. PANAMAX genomic DNA
6. UltraShear
7. EM-seq libraries
8. NovaSeq 6000
9. sIgA ELISA