Dissecting non-coding TAL1 mutations in T-ALL

Dissecting non-coding TAL1 mutations in T-ALL#

T-cell acute lymphoblastic leukemia (T-ALL) often arises when somatic mutations ectopically activate the transcription factor TAL1 in thymocyte progenitors where the locus is normally silent. Whole-genome sequencing of leukemias from individuals with T-ALL and T-ALL cell lines has mapped recurrent non-coding insertions, deletions, and single-nucleotide variants across three cis-regulatory sites flanking TAL1 (Mansour et al., Science 2014; Liu et al., Nature Genetics 2017; Liu et al., Nature Genetics 2020; Smith et al., Haematologica 2023).

Following Avsec et al., Nature 2026, we use AlphaGenome 2 to ask three questions:

Where do published T-ALL non-coding mutations fall relative to TAL1?
How does a single oncogenic insertion (chr1:47239296:C>ACG) remodel RNA-seq, DNase-seq, and histone modification tracks across the locus?
Is the effect sequence-specific, or would random insertions of the same length at the same positions produce comparable activation?

# @title Install AlphaGenome

# @markdown Run this cell to install AlphaGenome.
from IPython.display import clear_output
! pip install alphagenome
clear_output()

Set up#

Store your API key in Colab Secrets for persistent access across sessions (see Installation). We initialize the AlphaGenome 2 client (dna_client.create) and load canonical MANE Select protein-coding transcripts from GENCODE v46 (GRCh38) to visualize surrounding genes.

# @title Imports
import io

from alphagenome import colab_utils
from alphagenome.data import gene_annotation
from alphagenome.data import genome
from alphagenome.data import transcript
from alphagenome.models.v2 import dna_client
from alphagenome.models.v2 import variant_scorers
from alphagenome.visualization import plot_components
import matplotlib.pyplot as plt
import matplotlib.ticker as ticker
import numpy as np
import pandas as pd

dna_model = dna_client.create(colab_utils.get_api_key())

gtf = pd.read_feather(
    'https://storage.googleapis.com/alphagenome/reference/gencode/hg38/gencode.v46.annotation.gtf.gz.feather'
)
gtf_mane = gene_annotation.filter_to_mane_select_transcript(gtf)
transcript_extractor = transcript.TranscriptExtractor(gtf_mane)

Published non-coding T-ALL variants at the TAL1 locus#

We compile 32 somatic T-ALL mutations (15 unique alleles) across three regulatory sites around TAL1, which is transcribed on the minus (-) strand from chr1:47,232,373 to chr1:47,216,290:

5’ MuTE neo-enhancer (chr1:47,239,291 to chr1:47,239,296, ~7 kb upstream of the transcription start site): short 1-bp to 3-bp insertions and longer 4-bp to 18-bp insertions/indels (Mansour et al., Science 2014; Liu et al., Nature Genetics 2017; Liu et al., Nature Genetics 2020).
Intronic SNV (chr1:47,230,639:C>T): a recurrent single-nucleotide variant inside intron 1 (Liu et al., Nature Genetics 2020).
3’ neo-enhancer (chr1:47,212,072 and chr1:47,212,074, ~4 kb downstream of the 3’ end): 6-bp and 21-bp insertions (Smith et al., Haematologica 2023).

# @title Define published non-coding TAL1 variants across four T-ALL cohorts
VARIANT_TSV = '''ID	CHROM	POS	REF	ALT	Study
Jurkat	chr1	47239296	C	CCGTTTCCTAACC	Mansour_2014
MOLT-3	chr1	47239296	C	ACC	Mansour_2014
Patient_1	chr1	47239296	C	AACG	Mansour_2014
Patient_2	chr1	47239291	CTAACC	TTTACCGTCTGTTAACGGC	Mansour_2014
Patient_3-5	chr1	47239296	C	ACG	Mansour_2014
Patient_6	chr1	47239296	C	ACC	Mansour_2014
Patient_7	chr1	47239295	AC	TCAAACTGGTAACC	Mansour_2014
Patient_8	chr1	47239296	C	AACC	Mansour_2014
UPNT802	chr1	47212072	T	TGGGTAAACCGTCTGTTCAGCG	Smith_2023
UPNT613	chr1	47212074	G	GAACGTT	Smith_2023
SJALL043861_D1	chr1	47230639	C	T	Liu_2020
SJALL018373_D1	chr1	47230639	C	T	Liu_2020
SJALL040467_D1	chr1	47239296	C	AACC	Liu_2020
PATBGC	chr1	47239296	C	AACC	Liu_2017
PATBTX	chr1	47239296	C	ACGGATATAACC	Liu_2017
PARJAY	chr1	47239296	C	ACGGAATTTCTAACC	Liu_2017
PARSJG	chr1	47239296	C	AACC	Liu_2017
PASYAJ	chr1	47239296	C	AACC	Liu_2017
PATRAB	chr1	47239293	TTA	CTAACGG	Liu_2017
PAUBXP	chr1	47239296	C	ACC	Liu_2017
PATENL	chr1	47239296	C	AACC	Liu_2017
PARNXJ	chr1	47239296	C	ACG	Liu_2017
PASXSI	chr1	47239296	C	AACC	Liu_2017
PASNEH	chr1	47239296	C	ACC	Liu_2017
PAUAFN	chr1	47239296	C	AACC	Liu_2017
PARASZ	chr1	47239296	C	ACC	Liu_2017
PARWNW	chr1	47239296	C	ACC	Liu_2017
PASFKA	chr1	47239293	TTA	ACCGTTAATCAA	Liu_2017
PATEIT	chr1	47239296	C	AC	Liu_2017
PASMHF	chr1	47239296	C	AC	Liu_2017
PARJNX	chr1	47239296	C	AC	Liu_2017
PASYWF	chr1	47239296	C	AC	Liu_2017'''

variants_df = pd.read_table(io.StringIO(VARIANT_TSV))
variants_df['variant'] = [
    genome.Variant(r.CHROM, int(r.POS), r.REF, r.ALT, name=r.ID)
    for r in variants_df.itertuples(index=False)
]
(
    variants_df.groupby(['CHROM', 'POS', 'REF', 'ALT'], as_index=False)
    .agg(samples=('ID', 'count'), example_sample=('ID', 'first'), study=('Study', 'first'))
)

	CHROM	POS	REF	ALT	samples	example_sample	study
0	chr1	47212072	T	TGGGTAAACCGTCTGTTCAGCG	1	UPNT802	Smith_2023
1	chr1	47212074	G	GAACGTT	1	UPNT613	Smith_2023
2	chr1	47230639	C	T	2	SJALL043861_D1	Liu_2020
3	chr1	47239291	CTAACC	TTTACCGTCTGTTAACGGC	1	Patient_2	Mansour_2014
4	chr1	47239293	TTA	ACCGTTAATCAA	1	PASFKA	Liu_2017
5	chr1	47239293	TTA	CTAACGG	1	PATRAB	Liu_2017
6	chr1	47239295	AC	TCAAACTGGTAACC	1	Patient_7	Mansour_2014
7	chr1	47239296	C	AACC	8	Patient_8	Mansour_2014
8	chr1	47239296	C	AACG	1	Patient_1	Mansour_2014
9	chr1	47239296	C	AC	4	PATEIT	Liu_2017
10	chr1	47239296	C	ACC	6	MOLT-3	Mansour_2014
11	chr1	47239296	C	ACG	2	Patient_3-5	Mansour_2014
12	chr1	47239296	C	ACGGAATTTCTAACC	1	PARJAY	Liu_2017
13	chr1	47239296	C	ACGGATATAACC	1	PATBTX	Liu_2017
14	chr1	47239296	C	CCGTTTCCTAACC	1	Jurkat	Mansour_2014

Chromatin and transcriptional remodeling by a 5’ MuTE insertion#

To see how a single mutation from an individual with T-ALL alters local chromatin state and transcription, we inspect chr1:47239296:C>ACG (Patient_3-5), the 2-bp 5’ MuTE insertion highlighted in Avsec et al., Nature 2026. We predict across a 1 Mb context (dna_client.SEQUENCE_LENGTH_1MB) in CD34-positive bone marrow common myeloid progenitors (UBERON:0002371_CL:0001059), requesting RNA_SEQ, DNASE, and CHIP_HISTONE tracks and keeping experimentally observed tracks via filter_to_observed() (see Imputation).

# @title Predict REF and ALT tracks for chr1:47239296:C>ACG in CD34+ common myeloid progenitors
lead_variant = genome.Variant(
    'chr1', 47239296, 'C', 'ACG', name='chr1:47239296:C>ACG'
)
condition = 'UBERON:0002371_CL:0001059'  # Common myeloid progenitors in bone marrow.

output = dna_model.predict_variant(
    interval=lead_variant.reference_interval.resize(
        dna_client.SEQUENCE_LENGTH_1MB
    ),
    variant=lead_variant,
    condition=condition,
    requested_outputs=[
        dna_client.OutputType.RNA_SEQ,
        dna_client.OutputType.DNASE,
        dna_client.OutputType.CHIP_HISTONE,
    ],
)

ref_obs = output.reference.filter_to_observed()
alt_obs = output.alternate.filter_to_observed()

# @title Plot predicted track differences (ALT - REF) across RNA-seq, DNase-seq, and histone ChIP-seq
tal1_interval = genome.Interval('chr1', 47_209_255, 47_242_023)

fig = plot_components.plot(
    [
        plot_components.TranscriptAnnotation(
            transcript_extractor.extract(tal1_interval)
        ),
        plot_components.Tracks(
            tdata=alt_obs.rna_seq - ref_obs.rna_seq,
            ylabel_template='{biosample_name} ({strand})\n{Assay title}',
            filled=True,
        ),
        plot_components.Tracks(
            tdata=alt_obs.dnase[:, :1] - ref_obs.dnase[:, :1],
            ylabel_template='{biosample_name} ({strand})\n{Assay title}',
            filled=True,
        ),
        plot_components.Tracks(
            tdata=alt_obs.chip_histone - ref_obs.chip_histone,
            ylabel_template='{biosample_name} ({strand})\n{histone_mark}',
            filled=True,
        ),
    ],
    annotations=[plot_components.VariantAnnotation([lead_variant])],
    interval=tal1_interval,
    fig_width=10,
    xlabel='Chromosome 1 Coordinate (GRCh38)',
    title='Predicted Effect (ALT - REF) of chr1:47239296:C>ACG in CD34+ Common Myeloid Progenitors',
)
fig.axes[-1].xaxis.set_major_formatter(
    ticker.FuncFormatter(lambda x, _: f'{int(x):,}')
)
plt.show()

../../_images/ccc33ae0f5145995e8b9d9ed3c0df9c73bbcdfa14d9842ba3865c8d9c26d9641.png

Comparing T-ALL mutations with length-matched random alleles#

We pair each of the 15 unique T-ALL alleles with up to six length-matched random alternate alleles at the same genomic position (excluding REF and observed T-ALL alleles) and score every candidate across a 1 Mb context using variant_scorers.RECOMMENDED_VARIANT_SCORERS['RNA_SEQ'].

# @title Score unique T-ALL alleles and length-matched random alleles at the same positions
unique_df = variants_df.drop_duplicates(subset=['CHROM', 'POS', 'REF', 'ALT'])
rng = np.random.default_rng(42)
bases = np.array(list('ACGT'))
seen = set(zip(unique_df['POS'], unique_df['ALT']))

rows = []
for r in unique_df.itertuples(index=False):
  rows.append({
      'CHROM': r.CHROM,
      'POS': r.POS,
      'REF': r.REF,
      'ALT': r.ALT,
      'is_oncogenic': True,
  })
  alt_len = len(r.ALT)
  pool = (
      list('ACGT')
      if alt_len == 1
      else [''.join(rng.choice(bases, size=alt_len)) for _ in range(15)]
  )
  added = 0
  for alt in pool:
    if alt != r.REF and (r.POS, alt) not in seen:
      rows.append({
          'CHROM': r.CHROM,
          'POS': r.POS,
          'REF': r.REF,
          'ALT': alt,
          'is_oncogenic': False,
      })
      seen.add((r.POS, alt))
      added += 1
      if added == 6:
        break

eval_df = pd.DataFrame(rows)
eval_df['variant'] = [
    genome.Variant(r.CHROM, int(r.POS), r.REF, r.ALT)
    for r in eval_df.itertuples(index=False)
]

scores = dna_model.score_variants(
    intervals=[
        v.reference_interval.resize(dna_client.SEQUENCE_LENGTH_1MB)
        for v in eval_df['variant']
    ],
    variants=eval_df['variant'].tolist(),
    variant_scorers=[variant_scorers.RECOMMENDED_VARIANT_SCORERS['RNA_SEQ']],
    conditions=[condition] * len(eval_df),
    max_workers=5,
)
tal1_idx = scores[0][0].obs.query('gene_name == "TAL1"').index[0]
track_idx = (
    scores[0][0]
    .var.query(
        '`Assay title` == "polyA plus RNA-seq" and endedness == "paired" and'
        ' data_pipeline == "encode"'
    )
    .index[0]
)
eval_df['raw_score'] = [
    float(s[0][tal1_idx, track_idx].X[0, 0]) for s in scores
]

# @title Plot predicted TAL1 RNA-seq log-fold change across T-ALL and random alleles
def _variant_group(row):
  alt_len = len(row.ALT)
  if row.POS >= 47239290:
    return (
        "5' MuTE (>3 bp indels)"
        if alt_len > 4
        else f"5' MuTE ({alt_len - 1} bp ins.)"
    )
  if row.POS == 47230639:
    return 'Intronic SNV (chr1:47230639)'
  return f"3' Enhancer ({alt_len - 1} bp ins., chr1:{row.POS})"


eval_df['group'] = eval_df.apply(_variant_group, axis=1)

TOP_GROUPS = [
    "5' MuTE (>3 bp indels)",
    "5' MuTE (3 bp ins.)",
    "5' MuTE (2 bp ins.)",
    "5' MuTE (1 bp ins.)",
    'Intronic SNV (chr1:47230639)',
]
BOTTOM_GROUPS = [
    "3' Enhancer (6 bp ins., chr1:47212074)",
    "3' Enhancer (21 bp ins., chr1:47212072)",
]

fig, (ax_top, ax_bot) = plt.subplots(
    2,
    1,
    figsize=(7.6, 4.8),
    gridspec_kw={'height_ratios': [5, 2], 'hspace': 0.38},
)
rng_plot = np.random.default_rng(0)

for ax, groups in [(ax_top, TOP_GROUPS), (ax_bot, BOTTOM_GROUPS)]:
  sub = eval_df[eval_df['group'].isin(groups)]
  y_map = {g: i for i, g in enumerate(groups)}
  bg = sub[~sub['is_oncogenic']]
  onc = sub[sub['is_oncogenic']]

  ax.scatter(
      bg['raw_score'],
      [y_map[g] + j for g, j in zip(bg['group'], rng_plot.uniform(-0.14, 0.14, len(bg)))],
      color='#6b7280',
      alpha=0.55,
      s=34,
      edgecolors='none',
      label='Random length-matched alleles',
      zorder=2,
  )
  ax.scatter(
      onc['raw_score'],
      [y_map[g] + j for g, j in zip(onc['group'], rng_plot.uniform(-0.10, 0.10, len(onc)))],
      color='#e65100',
      alpha=0.92,
      s=62,
      edgecolors='white',
      linewidths=0.6,
      label='T-ALL alleles (individuals and cell lines)',
      zorder=4,
  )
  ax.axvline(0.0, color='#374151', linestyle='--', linewidth=1.0, zorder=1)
  ax.set_yticks(range(len(groups)))
  ax.set_yticklabels(groups, fontsize=9.5)
  ax.set_ylim(-0.45, len(groups) - 0.55)
  ax.invert_yaxis()
  ax.spines['top'].set_visible(False)
  ax.spines['right'].set_visible(False)

ax_top.set_title(
    'T-ALL Mutations vs. Length-Matched Random Alleles at the Same Coordinates',
    fontsize=11,
    pad=8,
)
ax_top.legend(
    frameon=True,
    facecolor='white',
    edgecolor='#e5e7eb',
    fontsize=8.8,
    loc='lower right',
)
ax_bot.set_xlabel('Predicted TAL1 RNA-seq log-fold change (ALT vs. REF)', fontsize=10)
plt.show()

../../_images/502412177860519857eded6804e1e76f5f40d468678e8c4886adba48d3b048d7.png

Across all seven variant groups, every somatic T-ALL mutation is predicted to increase TAL1 transcription, and by more than the length-matched random alleles at the same coordinate:

5’ MuTE neo-enhancer (chr1:47,239,291 to chr1:47,239,296): Short 1-bp to 3-bp insertions (C>AC, C>ACC, C>ACG, C>AACC, C>AACG) and longer insertions from individuals with T-ALL create de novo MYB recognition motifs (5'-ACCGTT-3' / 5'-CAACG-3') that recruit MYB together with hematopoietic transcription factors (GATA3, RUNX1, TAL1, and CBP) to establish an H3K27ac/H3K4me1 enhancer, clear repressive H3K27me3 across the locus, and drive TAL1 transcription (Mansour et al., Science 2014; Avsec et al., Nature 2026).
Intronic SNV (chr1:47,230,639:C>T) and 3’ neo-enhancer insertions (chr1:47,212,072 and chr1:47,212,074): Independent mutations in individuals with T-ALL inside intron 1 (Liu et al., Nature Genetics 2020) and ~4 kb downstream of TAL1 (Smith et al., Haematologica 2023) converge on the same transcriptional outcome, while matched random insertions at the same coordinates scatter around zero.