Oncofusion

start / graph / gene-cxcl12

Genestrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q33-niche-protein-layer

CXCL12 (SDF-1) - the marrow retention chemokine this graph already had a mechanism entry for and no entry for the protein

also called SDF-1 · SDF1 · SDF1A · SDF1B · stromal cell-derived factor 1 · PBSF

CXCL12 is a small signal protein that marrow support cells release to hold blood stem cells - and leukaemia cells - in place inside the bone. It is named in twenty-one entries of this graph and had no entry of its own until now.

The same thing, for a scientist

CXCL12/SDF-1 (UniProt P48061, 93 aa, secreted) is the ligand half of the canonical marrow retention axis. Compartment-specific deletion in mouse partitions the niche: perivascular stromal CXCL12 maintains HSCs, osteoblast CXCL12 maintains early lymphoid progenitors [E2]. In KMT2A-fusion disease the evidence is genuinely two-directional and both arms are murine myeloid MLL-AF9/MLL::AF9. Prx1+ mesenchymal CXCL12 maintains MLL::AF9 leukaemia stem cells through redox and energy balance and its deletion is rescued by antioxidant therapy [E3]; in a syngeneic MLL-AF9 model CXCL12 was dispensable for leukaemia development while CXCR4 signalling was essential [E4]. In the index fusion the only CXCL12 experiment is an applied SDF-1 gradient on RS4;11 (KMT2A::AFF1), where knockdown of MEIS1 or HOXA7/9/10 each reduced migration toward SDF-1 and abolished SDF-1-augmented proliferation [E5]. CXCL12 is not a cell-intrinsic dependency in any of 1,254 DepMap lines [E6], which is the expected result for a ligand made by another cell and is the reason a monoculture screen cannot see this biology.

What it looks like
CXCL1293 aa
Signal peptideSignal peptide · residues 1–21Mature SDF-1 chainMature SDF-1 chain · residues 22–93Receptor activation motif · residues 22–23SDF-1-alpha(3-67)SDF-1-alpha(3-67) · residues 24–88SDF-1-beta(3-72)SDF-1-beta(3-72) · residues 24–930

No three-dimensional model is available for this protein, so the map above is the whole picture. That is itself informative: AlphaFold does not model chains this long, and the parts that have been crystallised are held under the entries for the individual domains.

Go deeper
What it is and where it comes from

A 93-residue secreted chemokine (signal peptide 1-21, mature chain 22-93) with two principal processed forms, SDF-1-alpha(3-67) and SDF-1-beta(3-72), and a two-residue receptor activation motif at 22-23 1. Its cellular sources in marrow are not interchangeable: Ding and Morrison deleted Cxcl12 from endothelium, from perivascular stromal cells and from osteoblasts separately and got three different phenotypes - endothelial deletion depleted HSCs but not progenitors, perivascular deletion depleted HSCs and mobilised restricted progenitors, and osteoblast deletion depleted early lymphoid progenitors without touching HSCs 2. That result is why 'the niche makes CXCL12' is not a usable statement: for a B-lineage disease the osteoblast/endosteal source is the one implicated in lymphoid progenitor biology, and no KMT2A::AFF1 experiment has asked which compartment supplies the chemokine the blasts respond to.

In KMT2A-fusion disease the two published arms disagree, and both are murine myeloid

In MLL::AF9 mouse AML, CXCL12 from Prx1+ mesenchymal cells (but not from mature osteolineage cells) maintains leukaemia stem cells by modulating energy metabolism and redox balance; Cxcl12 deletion raised ROS and DNA damage in LSCs and impaired serial transplantation, and the defect was attenuated by antioxidant therapy 3. In a syngeneic MLL-AF9 model, a cell-surface CRISPR screen made CXCR4 a top dependency - and the ligand was dispensable: CXCL12 was not required for leukaemia development in recipient mice, while mutant-receptor experiments showed CXCR4 signalling was 4. Rule 73 applies to both: neither is infant B-lineage KMT2A::AFF1, and the first clause of any claim built on them must say MLL-AF9 mouse.

The only index-fusion CXCL12 experiment is a gradient applied to RS4;11

Orlovsky and colleagues knocked down MEIS1, HOXA7, HOXA9 and HOXA10 separately in RS4;11, a human B-precursor line carrying MLL-AF4 (KMT2A::AFF1), and transplanted into NOD/SCID. Homing was similar but colonisation was slowed, and the CXCR4/SDF-1 axis was hampered in every knockdown - reduced migration toward an SDF-1 gradient and loss of SDF-1-augmented proliferation in culture 5. This is the single strongest link between the niche layer of this graph and its core: the fusion's canonical menin-dependent output program (HOXA/MEIS1) is upstream of the blast's ability to use CXCL12. It is written up as hyp-cxcr4-function-is-a-hoxa-meis1-output-of-the-fusion.

Epigenetics: CXCL12 is better documented as an INPUT to chromatin than as a locus under fusion control

In T-ALL cells, CXCL12 exposure promotes fast H3K9 methylation and nuclear-shape changes that accompany invasiveness 7 - the chromatin change is in the responding cell, not at the CXCL12 gene. At the locus level the record is empty in this disease: CXCL12[tiab] AND (H3K79[tiab] OR H3K4me3[tiab] OR DOT1L[tiab]) returns 3 records (querytranslation: "CXCL12"[Title/Abstract] AND ("H3K79"[Title/Abstract] OR "H3K4me3"[Title/Abstract] OR "DOT1L"[Title/Abstract]), 2026-09-22), none of them KMT2A. No H3K79me2, H3K4me3 or menin occupancy measurement at CXCL12 in any KMT2A-rearranged system has been reported in a title or abstract; per rule 80 that bound is over abstracts and supporting figures were not swept.

Facts and figures
gene symbolCXCL12
uniprotP48061
length AA93
accession resolutiongene_exact:CXCL12 AND organism_id:9606 AND reviewed:true returned exactly ONE reviewed human record (n=1), P48061 / SDF1_HUMAN, primary gene name read back as CXCL12, length 93 aa. Fetched live 2026-09-22. Synonyms carried by the record: SDF1, SDF1A, SDF1B.
localizationSECRETED. UniProt P48061 records a single subcellular location comment, 'Secreted', with ECO:0000269 experimental support. There is no membrane, cytoplasmic or nuclear compartment on the record. Functionally the secreted protein is retained on cell-surface and matrix proteoglycans, which UniProt notes inhibits formation of SDF-1-alpha(3-67) and preserves local activity - so the biologically relevant pool is an immobilised extracellular gradient rather than a freely soluble one, and that is what an SDF-1 gradient experiment such as E5 reconstitutes in vitro.
subnuclear localizationNOT APPLICABLE, and this is a real answer rather than a gap. CXCL12 is a 93-residue secreted chemokine whose entire mature chain is extracellular; UniProt records no nuclear pool, and there is no NLS, no chromatin-binding module and no reported nuclear isoform. A subnuclear compartment cannot be assigned to a protein that is not in the nucleus. WHAT IS NOT NOT-APPLICABLE: CXCL12 SIGNALLING changes chromatin in the responding cell - fast H3K9 methylation and nuclear-shape change in T-ALL [E7] - so the nuclear biology of this axis is real and lives in the receiving nucleus, not in the ligand.
subnuclear absence classnot-applicable-by-topology
cell typeSOURCE cells (where the protein is made), which is the only sense in which a secreted ligand has a cell type: perivascular stromal cells as the principal source, with lower expression in endothelium, osteoblasts and some haematopoietic cells, each source deleted separately with a distinct phenotype [E2]; Prx1+ mesenchymal cells as the leukaemia-supporting source in MLL::AF9 [E3]; and CXCL12-abundant reticular (CAR) cells, which this graph holds as cell-type-car-cell. In infant KMT2A::AFF1 no experiment has identified which marrow compartment supplies the CXCL12 the blasts respond to.
cell stateThe state CXCL12 maintains in the normal compartment is HSC QUIESCENCE and pool size [E2, and the receptor-side result in gene-cxcr4 E1]. In MLL::AF9 the state it maintains is a LOW-ROS, OXPHOS-competent leukaemia stem cell state, evidenced by the ROS and DNA-damage phenotype on deletion and its rescue by antioxidant therapy [E3]. This graph holds the matching state entries via mechanism-niche-chemoprotection-kmt2a.
epigenetic readoutTWO DIFFERENT QUESTIONS, and only one has an answer. (1) CXCL12 AS AN INPUT TO CHROMATIN - answered: in T-ALL, CXCL12 promotes fast H3K9 methylation accompanying nuclear change and invasiveness [E7]. (2) THE CXCL12 LOCUS AS A TARGET OF FUSION-DRIVEN CHROMATIN - unanswered. CXCL12[tiab] AND (H3K79[tiab] OR H3K4me3[tiab] OR DOT1L[tiab]) returns 3 records, none in a KMT2A-rearranged system; no H3K79me2, H3K4me3 or menin occupancy measurement at CXCL12 has been reported in a KMT2A context. Bounded over titles and abstracts only (rule 80). Note the locus question is only partly interesting here, because CXCL12 is transcribed by the STROMAL cell, which does not carry the fusion - the fusion-controlled locus in this axis is the receptor, CXCR4, in the blast.
shown for fusionPARTIALLY, and which fusion matters. MLL::AF9 mouse: Cxcl12 deletion from Prx1+ mesenchymal cells impairs LSC maintenance [E3]. MLL-AF9 mouse: CXCL12 DISPENSABLE for leukaemia development [E4] - the two point opposite ways. KMT2A::AFF1 (the index fusion): only as an applied SDF-1 gradient on RS4;11, where the response depends on MEIS1/HOXA [E5]. NO for any measurement of CXCL12 protein or transcript in the marrow of an infant KMT2A::AFF1 patient, and NO for any identification of the source compartment in B-lineage disease.
intact union membershipABSENT from the 277-partner IntAct REST union of the six core baits (KMT2A, MEN1, PSIP1, AFF4, DOT1L, AFF1) recorded in analyses/interactome-union-migrated-to-rest/results.json, generated 2026-09-22. Checked by substring on both the symbol and the accession P48061. This is the expected result and it is worth stating: the niche layer is a transcriptional and microenvironmental layer, not a physical-interaction layer of the fusion complex, so interactome coverage metrics can never surface it.
collision noteCXCL12 has no row in data/symbol-collisions.json (registry read 2026-09-22, case-sensitive, iterating d['collisions'], 177 rows) and the token needs none: CXCL12 and SDF-1 are unambiguous. The alias SDF1 is safe; 'SDF' alone was not searched here.
domains
Signal peptide1.000Signal peptide: 1Mature SDF-1 chain22.0Mature SDF-1 chain: 22Receptor activation motif22.0Receptor activation motif: 22SDF-1-alpha(3-67)24.0SDF-1-alpha(3-67): 24SDF-1-beta(3-72)24.0SDF-1-beta(3-72): 24
domains — the numbers behind the chart
[
  {
    "name": "Signal peptide",
    "start": 1,
    "end": 21
  },
  {
    "name": "Mature SDF-1 chain",
    "start": 22,
    "end": 93
  },
  {
    "name": "Receptor activation motif",
    "start": 22,
    "end": 23
  },
  {
    "name": "SDF-1-alpha(3-67)",
    "start": 24,
    "end": 88
  },
  {
    "name": "SDF-1-beta(3-72)",
    "start": 24,
    "end": 93
  }
]
depmap essentiality
{
  "depmap_common_essential": false,
  "tissue_blocks": 29,
  "n_screens": 1254,
  "mean_gene_effect": -0.0286,
  "min_gene_effect": -0.6693,
  "max_gene_effect": 0.4595,
  "n_lines_at_or_below_minus_1": 0,
  "open_targets_data_version": "year 26, month 06",
  "open_targets_api_version": "26.6.3",
  "fetched": "2026-09-22",
  "control": "Live zero-screen control run in the same call and the same code path: MEN1 returned 29 tissue blocks, 1,258 screens, mean gene effect -0.3616, minimum -2.5262 and 79 lines at or below -1.0. So a zero count of dependent lines below is a measured zero and not a silent empty payload. DepMap release recorded as Open Targets dataVersion year 26 month 06 (rule 83: the payload carries no DepMap release id; this is the OT version), API 26.6.3, fetched 2026-09-22. NO median-split proportion is reported (rule 84): the statistic is an absolute count of lines at or below a gene-effect of -1.0, whose null is not 50%."
}
How this could bite us back
  • Blocking this axis pharmacologically means blocking the receptor, not the ligand, and the receptor is required by normal HSCs for pool maintenance - the same experiment that establishes the axis (gene-cxcr4 E1) is also the toxicity argument.
  • If the leukaemia-supporting function of CXCL12 is redox and energy balance rather than adhesion [E3], then removing it mobilises blasts AND leaves them in a higher-ROS state, which is chemosensitising in one reading and mutagenic in another. The antioxidant rescue in E3 means the direction is assay-dependent.
  • CXCL12 is dispensable in one MLL-AF9 model [E4]. A ligand-directed strategy could therefore fail while a receptor-directed strategy works, and a preclinical programme that tests only the ligand arm can produce a false negative for the whole axis.
Still open
  1. Which marrow compartment supplies the CXCL12 that infant KMT2A::AFF1 blasts respond to? Ding and Morrison's osteoblast arm is the one that matters for early lymphoid progenitors and nobody has asked it in B-lineage leukaemia.
  2. Are the two MLL-AF9 results reconcilable - is the CXCL12 that matters a local, contact-range pool that a recipient-wide deletion does not remove?
  3. Does CXCL12 transcript in marrow stroma change as a KMT2A::AFF1 xenograft establishes, i.e. is the ligand side part of the remodelling that mechanism-leukaemia-remodels-its-own-niche describes?
Every source behind this entry (7)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    The reviewed human CXCL12 record is UniProt P48061 (SDF1_HUMAN), 93 aa, secreted, with signal peptide 1-21, mature chain 22-93 and processed forms SDF-1-alpha(3-67) at 24-88 and SDF-1-beta(3-72) at 24-93.

    What this evidence cannot show

    Resolved by gene_exact:CXCL12 AND organism_id:9606 AND reviewed:true, which returned exactly ONE record; the primary gene name was read back as CXCL12 and the length as 93 aa (rule 62 / accession-resolution rule). Subcellular location 'Secreted' carries ECO:0000269, but per rule 91 the evidence code says a curator saw an experiment, not that the cited paper is the founding observation.

    Evidence that argues against this (1)
    • A UniProt record is curation, not measurement, and the processed-form boundaries SDF-1-alpha(3-67) and SDF-1-beta(3-72) are annotations of proteolytic products rather than species any cited experiment measured. CXCL12 also has multiple isoforms (alpha, beta, gamma and others) differing at the C-terminus, so '93 aa' describes one isoform; almost no paper in this field states which isoform it used, and the recombinant protein sold as 'SDF-1' is usually the alpha form.
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    In mouse, conditional Cxcl12 deletion from endothelium, perivascular stromal cells and osteoblasts gives three different phenotypes - HSCs occupy a perivascular niche and early lymphoid progenitors an endosteal niche - so the cellular source of marrow CXCL12 is not interchangeable.

    PMID 23434755DOI
    What this evidence cannot show

    Mouse, normal haematopoiesis, no leukaemia. The lymphoid-progenitor result is about normal early lymphoid progenitors, not about a transformed B-precursor clone, and the inference to infant KMT2A::AFF1 is an analogy and nothing more.

    Evidence that argues against this (1)
    • Mouse, and normal haematopoiesis - nothing in this experiment concerns leukaemia, let alone KMT2A-rearranged leukaemia. Conditional-deletion niche phenotypes also depend on Cre-driver specificity, and the drivers used to label endothelium, perivascular stroma and osteoblasts label overlapping populations, so assigning a phenotype to a cell type is weaker than assigning it to a driver.
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In murine MLL::AF9 AML, CXCL12 from Prx1+ mesenchymal cells maintains leukaemia stem cells through energy metabolism and redox balance; Cxcl12 deletion raises ROS and DNA damage and impairs serial transplantation, and the defect is attenuated by antioxidant therapy.

    The bone marrow niche regulates redox and energy balance in MLL::AF9 leukemia stem cells · Vinado AC, Calvo IA, Cenzano I, et al. · Leukemia 2022
    PMID 35618797DOI
    What this evidence cannot show

    Murine myeloid MLL::AF9, not infant B-lineage KMT2A::AFF1 (rule 73). The antioxidant rescue means the mechanism is partly redox rather than adhesion, so a pure 'mobilise the blast' reading of the axis does not follow from this paper.

    Evidence that argues against this (1)
    • The sibling MLL-AF9 study reports CXCL12 itself DISPENSABLE for leukaemia development while CXCR4 signalling is essential, which is hard to reconcile with a ligand-dependent LSC-maintenance mechanism unless the relevant ligand pool is local and the recipient-derived pool is not. Ramakrishnan R et al. 2020
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In a syngeneic murine MLL-AF9 AML model, CXCL12 was dispensable for leukaemia development in recipient mice while CXCR4 signalling was essential, and Cxcr4 deletion eradicated leukaemia without impairing marrow homing.

    CXCR4 Signaling Has a CXCL12-Independent Essential Role in Murine MLL-AF9-Driven Acute Myeloid Leukemia · Ramakrishnan R, Pena-Martinez P, Agarwal P, et al. · Cell Rep 2020
    PMID 32460032DOI
    What this evidence cannot show

    Murine myeloid. 'Dispensable' here means dispensable in the recipient for leukaemia development; it does not exclude a requirement for a locally produced or leukaemia-autonomous ligand pool, which is what E3 reports. This entry does not resolve the contradiction and should not be read as doing so.

    Evidence that argues against this (1)
    • THIS IS THE STRONGEST COUNTER-EVIDENCE IN THIS ENTRY AND IT IS ABOUT THE ENTRY'S OWN SUBJECT. PMID 32460032 shows CXCL12 was DISPENSABLE for leukaemia development while CXCR4 signalling was essential, in a murine MLL-AF9 model. If the receptor is required where the ligand is not, then 'the CXCL12/CXCR4 axis' is the wrong unit of analysis, and an entry for CXCL12 should not inherit importance from CXCR4's.
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In RS4;11 (human MLL-AF4 / KMT2A::AFF1 B-precursor line), separate knockdown of MEIS1, HOXA7, HOXA9 or HOXA10 each hampered the CXCR4/SDF-1 axis - reduced migration toward an SDF-1 gradient and loss of SDF-1-augmented proliferation - and slowed marrow colonisation after similar homing in NOD/SCID mice.

    PMID 21518888DOI
    What this evidence cannot show

    Read from the abstract, which states the CXCR4/SDF-1 result but not the magnitude, the number of replicates or whether CXCR4 surface level itself fell. Whether the defect is receptor level, receptor trafficking or downstream signalling is not resolved by the abstract, and the underlying figures were not read (rule 80).

    Evidence that argues against this (1)
    • A single human line (RS4;11), knockdown rather than knockout, with an in-vitro migration readout. MEIS1 and HOXA knockdown are strongly pleiotropic in KMT2A-rearranged cells - they reduce proliferation and engraftment generally - so a reduced SDF-1 chemotactic response is compatible with a general fitness defect rather than a specific effect on the CXCR4/SDF-1 axis.
  6. E6established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    CXCL12 is not a cell-intrinsic dependency in any DepMap line: 29 tissue blocks, 1,254 screens, mean gene effect -0.0286, range -0.6693 to 0.4595, and zero lines at or below a gene effect of -1.0.

    What this evidence cannot show

    Live zero-screen control run in the same call and the same code path: MEN1 returned 29 tissue blocks, 1,258 screens, mean gene effect -0.3616, minimum -2.5262 and 79 lines at or below -1.0. So a zero count of dependent lines below is a measured zero and not a silent empty payload. DepMap release recorded as Open Targets dataVersion year 26 month 06 (rule 83: the payload carries no DepMap release id; this is the OT version), API 26.6.3, fetched 2026-09-22. NO median-split proportion is reported (rule 84): the statistic is an absolute count of lines at or below a gene-effect of -1.0, whose null is not 50%. And the result is uninformative about the biology by construction: DepMap screens cells in monoculture, so a chemokine supplied by a stromal cell cannot score, and the contrast with the in vivo syngeneic screen that made CXCR4 a top dependency [E4] is a property of the assay, not of the gene.

    Evidence that argues against this (1)
    • A NICHE LIGAND CANNOT SCORE AS A DEPENDENCY IN THIS ASSAY, SO THIS IS AN ABSENCE BY CONSTRUCTION. DepMap screens cell lines in monoculture without stroma; CXCL12 is made by stromal cells and acts on the blast in trans. A mean gene effect of -0.0286 therefore measures that a leukaemia cell does not need to make its own CXCL12, which is not in question, and says nothing about whether CXCL12 in a marrow matters. Reading it as evidence against the biology would be a category error.
  7. E7moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    In T-cell acute lymphoblastic leukaemia cells, CXCL12 promotes fast H3K9 methylation that contributes to nuclear changes and invasiveness.

    PMID 34999734DOI
    What this evidence cannot show

    T-ALL, not KMT2A-rearranged B-ALL, and read from the title and abstract only. Author names are deliberately omitted rather than typed from memory (the fetched record was read for the title, DOI and journal only). The chromatin change is in the CXCL12-responding cell; this is not a measurement at the CXCL12 locus and must not be cited as one.

    Evidence that argues against this (1)
    • T-cell ALL, not the B-precursor KMT2A::AFF1 disease this graph is built around, and a single study. It is also the only line of evidence here connecting CXCL12 to an epigenetic readout, so it carries the whole weight of that link with no independent replication.
Connected to