bio-chipseq-cut-and-run-tag

What this skill does

## Version Compatibility

Reference examples tested with: SEACR 1.4+, MACS2 2.2.9+, MACS3 3.0.4+, samtools 1.19+, bowtie2 2.5+, bedtools 2.31+, deepTools 3.5+, GoPeaks 1.0+, LanceOtron (pip).

# CUT&RUN / CUT&Tag

**"Analyze CUT&RUN or CUT&Tag chromatin profiling data"** -> Use the lower-background, lower-input alternatives to traditional ChIP. CUT&RUN tethers MNase to an antibody via Protein A; CUT&Tag tethers Tn5 via Protein A/G. Both bypass cross-linking, fragmentation, and IP washes — producing 10-100× lower background, allowing 100-1000× lower cell input, and shifting the peak-calling problem from "find signal in noise" to "find signal in near-zero background."

- Aligner: bowtie2 (CUT&RUN/Tag standard) or bwa-mem; chromap optional
- Peak calling (CUT&RUN/Tag): SEACR (Meers 2019), MACS2 with `-f BAMPE --keep-dup all`, or both for consensus
- Spike-in: E. coli carryover from bacterially-produced pA-MNase/Tn5 (automatic, variable)
- Control: IgG-only (no input control; native chromatin has no meaningful "input")

CUT&RUN/CUT&Tag has different QC thresholds, different peak calling defaults, different spike-in protocols, and different antibody requirements than traditional ChIP. Treating it as ChIP fails silently.

## Protocol Variant Taxonomy

| Variant | Chimera | Year | Use case | Failure mode |
|---------|---------|------|----------|--------------|
| **CUT&RUN** (Skene Henikoff) | pA-MNase | 2017 | Native chromatin profiling; broad antibody compatibility | Native (no fixation) — gentler; MNase digest needs careful Ca²⁺ control |
| **CUT&Tag** (Kaya-Okur Henikoff) | pA-Tn5 (rabbit only) | 2019 | Lower cell input (~5000); faster; library-ready output | Rabbit-only antibody; PCR cycles can over-amplify |
| **CUT&Tag-IT** (Active Motif) | pA-Tn5 commercial | 2020 | Standardized lots; reproducible | Cost; vendor-locked |
| **pAG-Tn5 CUT&Tag** | pAG-Tn5 | 2020 | Binds both rabbit AND mouse IgG | More versatile; identical performance otherwise |
| **AutoCut&Tag** | pAG-Tn5 plate-based | 2021 | High-throughput (96-well) | Throughput at the cost of per-sample optimization |
| **CUTAC** (CUT&Tag-then-ATAC) | pAG-Tn5 + protocol modification | 2020 | Chromatin accessibility variant of CUT&Tag | Less common; not standard CUT&Tag |
| **scCUT&Tag** | pAG-Tn5 in droplets | 2021 | Single-cell histone mark profiling | Very sparse (~1000-5000 reads/cell) |

## Algorithmic Taxonomy

| Tool | Model | Strength | Fails when |
|------|-------|----------|------------|
| **SEACR** (Meers 2019) | Empirical threshold on signal block totals; IgG-aware "stringent" mode | Designed for sparse CUT&RUN data; "stringent + norm + IgG" is the recommended default | Wrong mode (top-X% without IgG; "non" mode if no upstream spike-in normalization); broad mark with very flat signal landscape |
| **MACS2 `-f BAMPE --keep-dup all`** | Local Poisson | Familiar; integrates well with downstream tools (DiffBind) | Default `-q 0.05` may be too lenient for low-background CUT&Tag; consider `-q 0.01` |
| **GoPeaks** (Yashar 2022) | Sliding-window thresholding | Broad-mark-oriented; faster than SEACR on broad data | Newer; smaller user base |
| **LanceOtron** (Hentges 2022) | CNN trained on ENCODE peaks | Parameter-free; handles both narrow and broad | Less validated for CUT&RUN/Tag specifically; web-only or pip |
| **MACS2 + SEACR consensus** | Intersection | Highest confidence; per 2025 btaf375 benchmark, best for cross-paper reproducibility | Most conservative; may miss true peaks at marginal regions |

**2025 benchmark (Bioinformatics btaf375):**
- MACS2 better for sharp peaks (H3K4me3, TFs)
- SEACR better for broad signal (H3K27me3)
- MACS2 + SEACR consensus best for publication-grade peak sets
- "norm stringent" + IgG is the recommended SEACR default

## SEACR Workflow (Canonical CUT&RUN/Tag Caller)

**Goal:** Call CUT&RUN/CUT&Tag peaks from aligned BAMs using SEACR with IgG-aware threshold.

**Approach:** Align with Henikoff parameters, convert BAM to fragment bedGraph via bamtobed-bedpe, then invoke SEACR with `norm stringent` mode and IgG control.

```bash
# 1. Align with bowtie2 (Henikoff lab standard parameters)
bowtie2 --local --very-sensitive --no-mixed --no-discordant \
    --phred33 -I 10 -X 700 \
    -x hg38 -1 reads_R1.fq -2 reads_R2.fq \
    -S aln.sam

# 2. Convert SAM to BAM, sort, index
samtools view -bS aln.sam | samtools sort -o aln.bam
samtools index aln.bam

# 3. Generate bedGraph for SEACR (paired-end fragments)
samtools view -bS -F 0x04 aln.bam | bedtools bamtobed -bedpe -i - > aln.bedpe
awk '$1==$4 && $6-$2 < 1000 {print $0}' aln.bedpe > aln.clean.bedpe
cut -f 1,2,6 aln.clean.bedpe | sort -k1,1 -k2,2n -k3,3n > aln.fragments.bed
bedtools genomecov -bg -i aln.fragments.bed -g hg38.chrom.sizes > aln.bedgraph

# 4. Same for IgG control
# (... produce igg.bedgraph similarly ...)

# 5. SEACR with stringent + norm + IgG control (recommended default).
# Final argument is the OUTPUT PREFIX; SEACR appends ".stringent.bed" / ".relaxed.bed".
bash SEACR_1.4.sh aln.bedgraph igg.bedgraph norm stringent target_peaks
# Output file: target_peaks.stringent.bed

# Alternative: no IgG control, use top 1% of peaks
# bash SEACR_1.4.sh aln.bedgraph 0.01 non stringent target_peaks
```

**SEACR mode selection:**
- `norm` (recommended): scales target to IgG distribution
- `non`: use ONLY if upstream spike-in normalization was applied; otherwise use `norm`
- `stringent`: top-half of signal blocks (recommended default)
- `relaxed`: full distribution (use only for very sparse signal)
- IgG control: `bash SEACR_1.4.sh target.bg igg.bg norm stringent out_prefix` -> writes `out_prefix.stringent.bed`
- Top-X% without IgG: `bash SEACR_1.4.sh target.bg 0.01 non stringent out_prefix` -> writes `out_prefix.stringent.bed`

## E. coli Spike-In Carryover (Automatic Spike-In)

CUT&RUN/CUT&Tag spike-in is "free" because the pA-MNase or pA-Tn5 carries E. coli DNA from bacterial production. Carryover is variable across batches but stable within a batch.

```bash
# Align reads to combined hg38 + E. coli genome (or sequential)
bowtie2 -x hg38_ecoli_combined -1 R1.fq -2 R2.fq -S aln.sam

# Count E. coli reads per sample
ECOLI_READS=$(samtools view -c -F 4 aln.bam ecoli_chr1)

# Scaling factor: smallest E. coli read count / per-sample E. coli reads
# Apply BEFORE peak calling for cross-condition comparison
```

**Expected E. coli alignment fractions:**
- Target ChIP samples: 0.5-2% of total reads
- IgG control: 2-5% (higher because no target chromatin to dilute)
- Below 0.1% E. coli: spike-in carryover lost; cross-condition normalization unreliable
- Above 10%: target ChIP failed (mostly E. coli)

The carryover is variable between batches of bacterial production; single-experiment carryover spike-in is noisier than deliberate Drosophila spike-in (ChIP-Rx). For high-stakes cross-condition claims, add deliberate Drosophila spike-in despite the E. coli carryover. See chip-seq/spike-in-normalization.

## QC Differences from Traditional ChIP

| Metric | Traditional ChIP | CUT&RUN/CUT&Tag |
|--------|------------------|------------------|
| FRiP (TF) | > 0.05 | > 0.10 (often > 0.25) |
| FRiP (histone) | > 0.10 | > 0.25 |
| Library size requirement | 20-50M | 3-10M (often sufficient) |
| Input control | Required | IgG only (no input meaningful) |
| Fragment size (CUT&Tag) | Sub-nucleosomal or mono-nucleosomal | Sharp peak at 25-75 bp (Tn5 staggered cuts) |
| Fragment size (CUT&RUN) | Variable | Mono- + di-nucleosomal pattern |
| Duplicates | Remove (MarkDuplicates) | **Keep for CUT&Tag** (low PCR cycles, dups have biology) |
| Spike-in alignment | Deliberate (Drosophila); 0.5-5% | Automatic E. coli; 0.5-5% |
| Cell input | 1-10M | 5,000-100,000 |

**Critical: `--keep-dup all` in MACS for CUT&Tag.** PCR cycles are 12-15 (vs 5-8 for ChIP); duplicates at high-coverage TF binding sites contain biology. Standard ChIP convention `--keep-dup auto` will over-deduplicate CUT&Tag data.

## Fragment Size as Diagnostic (Critical for