Oncofusion

start / graph / noncoding-hoxa-as2-as3

Non-codingweak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.Q10-genome

HOXA-AS2 and HOXA-AS3 — mid-cluster antisense transcripts with essentially no KMT2A-r evidence

also called HOXA-AS2 · HOXA cluster antisense RNA 2 · HOXA-AS3 · HOXA cluster antisense RNA 3 · HOXA3as

HOXA-AS2 and HOXA-AS3 are two more non-coding RNAs made from the opposite strand in the middle of the HOXA gene cluster. They have been studied a bit in adult AML and in solid tumours, but nobody has shown they do anything specific in KMT2A-rearranged leukemia, so this entry is mostly a record of what is missing.

The same thing, for a scientist

HOXA-AS2 (ENSG00000253552; GRCh38 chr7:27,107,777-27,134,302, + strand, spanning the HOXA3-HOXA4 interval) and HOXA-AS3 (ENSG00000254369; GRCh38 chr7:27,129,977-27,155,928, + strand, spanning the HOXA5-HOXA6 interval) are overlapping antisense lncRNAs in the middle of the HOXA cluster [E5]. **These two elements are grouped into one node deliberately, because the KMT2A-rearranged-specific evidence for both is effectively nil and separate nodes would imply a precision the literature does not support.** HOXA-AS2 has a real, if low-quality, leukemia literature: it is a retinoic-acid-induced apoptosis repressor in NB4 promyelocytic cells [E1], is up-regulated in adult AML where high expression associates with worse overall survival in TCGA (HR 1.640, 95% CI 1.046-2.573, log-rank P = 0.0284) [E2], and is reported to act as an oncogene by binding EZH2 and repressing LATS2 [E3]. In ALL, one paper reports HOXA-AS2 is high in prednisone-insensitive ALL cell lines and patient samples and confers glucocorticoid resistance through HOXA3/EGFR/Ras/Raf/MEK/ERK [E4]. HOXA-AS3 has **37 PubMed records in total and zero that involve leukemia of any kind** [E6] — its literature is vascular smooth muscle, lung and gastric cancer. Neither element has been perturbed in a KMT2A-rearranged cell line, model or cohort. Confidence for any KMT2A-r claim here is **weak**, and the node is retained chiefly so that the graph records the hole rather than leaving a plausible-looking blank.

Go deeper
What is actually known, ranked by how much it should be trusted1 deeper

Tier 1 — reasonably solid, but not KMT2A-r. HOXA-AS2 is expressed in NB4 promyelocytic cells and in normal human peripheral blood neutrophils, is induced by all-trans retinoic acid, and shRNA knockdown reduced viable cells and increased apoptosis measured by annexin V and by activity and cleavage of caspases-3, -8 and -9, accompanied by elevated TRAIL 1. This is a careful, mechanistically modest paper from the same Newburger group that described HOTAIRM1, and its conclusion is narrow: ATRA induction of HOXA-AS2 fine-tunes apoptosis during myeloid differentiation.

Tier 2 — bioinformatic, hypothesis-generating. In the TCGA LAML dataset HOXA-AS2 is up-regulated in AML bone marrow and high expression associates with poorer overall survival (HR 1.640, 95% CI 1.046-2.573, log-rank P = 0.0284) 2. The confidence interval nearly touches 1 and the P value is 0.0284 on a single univariate test with no multiplicity correction and no validation cohort. This is a weak signal presented as a biomarker.

Tier 3 — the mechanistic-axis genre, treat with scepticism. Three papers report HOXA-AS2 acting through a specific downstream axis: EZH2/LATS2 in AML 3; SOX4/PI3K/AKT in AML; miR-520c-3p/S100A4 in AML chemoresistance; and HOXA3/EGFR/Ras/Raf/MEK/ERK in glucocorticoid-resistant ALL 4. Each proposes a different, non-overlapping mechanism for the same transcript in the same disease. That pattern — one lncRNA, four incompatible single-axis mechanisms, each in a different journal, each from a different group with no cross-replication — is characteristic of a literature that has not converged, and it should be read as a signal of low reliability rather than of rich biology.

The ALL paper deserves specific caution

The glucocorticoid-resistance paper 4 is the one most tempting to import into an infant-ALL project, because prednisone poor response is a defining KMT2A-r phenotype. Three reasons to hold back:

  1. Its cell model is Jurkat, a T-ALL line, and KMT2A-r infant leukemia is B-lineage. The patient samples are described as paediatric ALL without a stated fusion breakdown or n.
  2. The proposed mechanism (HOXA-AS2 → HOXA3 → EGFR → Ras/Raf/MEK/ERK) is a generic proliferation cascade with no KMT2A-specific element.
  3. There is a near-identical published claim for a different HOXA-cluster lncRNA: HOTAIRM1 also enhances glucocorticoid resistance in leukemia, but through ARHGAP18/RHOA/ROCK1 (see noncoding-hotairm1). Two neighbouring antisense transcripts, both said to cause steroid resistance by unrelated routes, in cell models derived the same way, is more consistent with a locus-level or assay-level artefact than with two independent mechanisms.

If the lab wants this question answered properly, the experiment is to ask whether HOXA-locus transcription as a whole correlates with prednisone response in a genotyped infant-ALL cohort, not to chase one antisense transcript.

HOXA-AS3: a genuine blank

A PubMed search for HOXA-AS3 returns 37 records in total, and a search for HOXA-AS3 AND leukemia (in myeloid form or otherwise) returns zero 6. The published HOXA-AS3 literature is about vascular smooth-muscle phenotype switching, lung adenocarcinoma, gastric and colorectal cancer, and endothelial biology.

So for this project: HOXA-AS3 is a named, annotated, HOXA-cluster antisense lncRNA that overlaps HOXA5/HOXA6 and about which nothing haematological is known. It sits inside the domain that the KMT2A fusion opens, it is certainly transcribed in a HOXA-high blast, and it has never been looked at. That is the honest statement, and it is exactly the sort of thing the PI's question was aimed at.

It is worth noting why it has never been looked at: HOXA-AS2 (27,107,777-27,134,302) and HOXA-AS3 (27,129,977-27,155,928) overlap by about 4.3 kb, and both overlap the long HOXA3 isoform (27,106,184-27,152,610) on the opposite strand 5. Short-read RNA-seq cannot cleanly assign reads among them, so most published quantifications of 'HOXA-AS2' in leukemia are of uncertain identity. Any serious element-level work in this region needs long-read or strand-specific targeted sequencing, which nobody has applied here.

Intervention read

Neither element is a credible target today, and saying otherwise would be dishonest.

  • There is no KMT2A-r loss-of-function experiment for either.
  • The HOXA-AS2 mechanisms in the literature do not agree with each other.
  • The overlapping annotation means an ASO designed against HOXA-AS2 will very likely also hit HOXA-AS3 and the HOXA3 antisense region, so 'specific knockdown' is not currently achievable by sequence design alone.
  • HOXA-AS2's best-supported function 1 is anti-apoptotic during differentiation, which at least has the right polarity for a therapy (knockdown promotes apoptosis), unlike HOTAIRM1.

The tractable version of this question is not a drug. It is an annotation and quantification problem: resolve which transcripts in the chr7:27.10-27.16 Mb window are actually expressed in KMT2A::AFF1 blasts, with strand-specific long-read data, before anyone designs a reagent.

Facts and figures
cytoband7p15.2
annotation overlap bp gene level4326
annotation problemHOXA-AS2 and HOXA-AS3 overlap each other and both overlap the long HOXA3 model on the opposite strand; short-read RNA-seq cannot unambiguously assign reads, so published 'HOXA-AS2' quantifications in leukemia are of uncertain transcript identity
kmt2a r specific evidencenone published for either element
elements
[
  {
    "element_symbol": "HOXA-AS2",
    "class": "lncRNA",
    "genome_build": "GRCh38",
    "locus": "chr7:27,107,777-27,134,302",
    "chromosome": "7",
    "start": 27107777,
    "end": 27134302,
    "strand": "+",
    "ensembl_gene_id": "ENSG00000253552",
    "ensembl_canonical_transcript": "ENST00000669815",
    "hgnc_id": "HGNC:43745",
    "position_in_cluster": "mid-cluster, antisense, spanning the HOXA3-HOXA4 interval",
    "reported_partners_or_axes": [
      "EZH2 / LATS2 (AML)",
      "SOX4 / PI3K-AKT (AML)",
      "miR-520c-3p / S100A4 (AML chemoresistance)",
      "HOXA3 / EGFR-Ras-Raf-MEK-ERK (ALL, glucocorticoid resistance)",
      "TRAIL / caspase-8 (NB4 apoptosis)"
    ],
    "direction_of_change_in_kmt2a_r": "unknown",
    "leukemia_pubmed_records": 8
  },
  {
    "element_symbol": "HOXA-AS3",
    "class": "lncRNA",
    "genome_build": "GRCh38",
    "locus": "chr7:27,129,977-27,155,928",
    "chromosome": "7",
    "start": 27129977,
    "end": 27155928,
    "strand": "+",
    "ensembl_gene_id": "ENSG00000254369",
    "ensembl_canonical_transcript": "ENST00000518848",
    "hgnc_id": "HGNC:43748",
    "position_in_cluster": "mid-cluster, antisense, spanning the HOXA5-HOXA6 interval",
    "reported_partners_or_axes": [
      "EZH2 (vascular smooth muscle, non-haematological)"
    ],
    "direction_of_change_in_kmt2a_r": "unknown",
    "leukemia_pubmed_records": 0
  }
]
cohorts with direction evidence
[
  {
    "cohort": "TCGA LAML (adult de-novo AML), HOXA-AS2 only",
    "finding": "up-regulated in AML bone marrow; high expression associated with worse OS, HR 1.640 (95% CI 1.046-2.573), log-rank P=0.0284",
    "pmid": "33754013",
    "n": "not stated in abstract; TCGA LAML RNA-seq cohort",
    "title": "Comprehensive investigation of the clinical significance of long non-coding RNA HOXA-AS2 in acute myeloid leukemia using genome-wide RNA sequencing dataset.",
    "authors": "Huang R, Liao X, Wang X, et al.",
    "venue": "J Cancer",
    "year": 2021,
    "title_provenance": "Title/authors/venue backfilled from PubMed esummary on 2026-09-20 because this citation stored a bare PMID with no title, which made it unscoreable by verify-citations.py. BACKFILLING IS NOT VERIFICATION: it confirms the PMID resolves to a real paper, not that this is the right paper for the claim. That still requires reading."
  }
]
intervention assessment
{
  "tractability": "none at present",
  "best_modality": "not applicable — no validated KMT2A-r dependency exists to target",
  "modalities_ruled_out": [
    "small molecule (no pocket)",
    "ASO (cannot be made element-specific given the overlapping annotation; an ASO against HOXA-AS2 would probably also hit HOXA-AS3 and the antisense HOXA3 region)"
  ],
  "blocking_problem": "transcript identity, not chemistry. The chr7:27.10-27.16 Mb window needs strand-specific long-read RNA-seq in KMT2A::AFF1 blasts before any reagent is designed.",
  "first_experiment": "strand-specific long-read (Nanopore/PacBio) RNA-seq of the HOXA cluster in 3-5 KMT2A::AFF1 B-ALL samples and matched KMT2A-germline B-ALL, to establish which antisense transcripts exist and at what level. Cheap, definitive, and nobody has published it."
}
How this could bite us back
  • Delivery of specificity, not drug: any ASO here would hit multiple overlapping transcripts including the antisense strand of HOXA3, so an apparent phenotype could not be attributed to a single element. This is a reagent-specificity trap rather than a biological one.
  • Essential-function toxicity: these transcripts are part of normal HOXA collinear regulation in development and in myeloid maturation; perturbing them is perturbing the locus control region of a developmental master cluster.
  • Literature risk: importing the HOXA-AS2 glucocorticoid-resistance claim into an infant-ALL programme without replicating it in B-lineage KMT2A-r cells would build a research direction on a single unreplicated T-ALL-line result.
  • Compensation: the HOXA cluster contains at least five separately annotated non-coding elements that are co-activated. Silencing one is the least likely perturbation in this locus to produce a phenotype.
Still open
  1. Which antisense transcripts in the chr7:27.10-27.16 Mb window are genuinely expressed in KMT2A::AFF1 blasts, and at what level? Short-read data cannot answer this and long-read data has never been generated for this locus in KMT2A-r leukemia.
  2. HOXA-AS3 has zero leukemia literature despite sitting in the middle of the domain the KMT2A fusion opens. Is it transcribed in KMT2A-r blasts at all? A single strand-specific RNA-seq lookup answers it.
  3. Two neighbouring HOXA antisense lncRNAs (HOXA-AS2 and HOTAIRM1) have each been claimed to cause glucocorticoid resistance by unrelated mechanisms. Is the real variable HOXA-locus transcriptional output rather than any individual transcript? This is testable and would reframe several papers.
  4. Does the KMT2A fusion protein bind within the HOXA-AS2/AS3 interval, or does antisense transcription there merely follow domain opening? Answerable from the existing Kerry 2017 and Harman 2021 ChIP-seq tracks.
  5. Is antisense transcription across the HOXA cluster functionally required for the fusion's HOXA9/HOXA10 output, or is it a byproduct of a permissive chromatin state? No experiment in the literature separates these.
Every source behind this entry (6)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    HOXA-AS2 is expressed in NB4 promyelocytic leukemia cells and in normal human peripheral blood neutrophils, is increased by all-trans retinoic acid, and shRNA knockdown decreases viable cell number and increases apoptosis (annexin V binding; activity and cleavage of caspases-3, -8 and -9) with elevated TRAIL, indicating that ATRA induction of HOXA-AS2 suppresses ATRA-induced apoptosis.

    HOX antisense lincRNA HOXA-AS2 is an apoptosis repressor in all trans retinoic acid treated NB4 promyelocytic leukemia cells · Zhao H, Zhang X, Frazão JB, Condino-Neto A, Newburger PE · Journal of Cellular Biochemistry 114(10):2375-83 2013
    PMID 23649634DOI
    What this evidence cannot show

    Single cell line (NB4, an APL line with PML-RARA) plus primary neutrophils, in an ATRA-differentiation paradigm only. shRNA knockdown without rescue. The TRAIL link is described by the authors themselves as 'possibly'. No B-lymphoid and no KMT2A-rearranged context. Effect sizes are modest.

    Evidence that argues against this (1)
    • Later AML papers assign HOXA-AS2 entirely different mechanisms (EZH2/LATS2; SOX4/PI3K/AKT; miR-520c-3p/S100A4), none of which involves TRAIL or caspase-8, so the apoptosis mechanism reported here has not been corroborated. Feng Y et al. 2020
  2. E2weak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.paper

    In the TCGA acute myeloid leukemia RNA-sequencing dataset HOXA-AS2 is significantly up-regulated in AML bone marrow and high HOXA-AS2 expression is associated with poorer overall survival (hazard ratio 1.640, 95% CI 1.046-2.573, log-rank P = 0.0284).

    PMID 33754013DOI
    What this evidence cannot show

    Purely re-analysis of public TCGA LAML level-3 data; the cohort n is not stated in the abstract. Confidence interval lower bound 1.046 is marginal, P = 0.0284 uncorrected, no validation cohort, no multivariable adjustment reported in the abstract. TCGA LAML is an adult, de-novo AML cohort with very few KMT2A-rearranged cases, so it cannot speak to KMT2A-r disease. The 'predicted drugs' (megestrol, carmustine, cefoxitin) are connectivity-map style outputs and should not be treated as leads.

    Evidence that argues against this (1)
    • The direction conflicts with the only functional myeloid study of this transcript, in which HOXA-AS2 is induced BY differentiation therapy and restrains apoptosis during maturation, which would predict association with a more differentiated, not a worse-prognosis, phenotype. Zhao H et al. 2013
  3. E3weak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.paper

    HOXA-AS2 is up-regulated in AML cell lines and tissues with overexpression negatively correlated with patient survival; silencing HOXA-AS2 inhibits proliferation and induces differentiation of AML cells in vitro and in vivo, reportedly by binding EZH2 and thereby repressing LATS2.

    PMID 33268767DOI
    What this evidence cannot show

    Single-group study, no independent replication. RNA-EZH2 pulldown is notoriously prone to non-specific association (EZH2 binds many RNAs promiscuously in vitro), so 'binding to EZH2' is a low-specificity claim. The AML cell lines and patient numbers are not stated in the abstract. No KMT2A-r stratification. Graded weak. **Resolved:** the lines are **NB4 (PML::RARA, 32 mentions) and THP-1 (28)**. Only THP-1 is KMT2A-rearranged, and the dominant line is an acute promyelocytic leukaemia model, which is **HOXA-low** - the opposite of the state this entry reasons about. PMC7710717. Resolved 2026-09-19 by the Q18b unstated-systems sweep from the study's deposition, not from its text; bounded as a statement about the deposited data rather than about the whole study.

    Evidence that argues against this (2)
    • Competing single-axis mechanisms for the same transcript in the same disease (SOX4/PI3K/AKT; miR-520c-3p/S100A4) were published by other groups in the same period, and none has been reconciled or replicated, which argues the EZH2/LATS2 axis is not established. Qu Y et al. 2020
    • A HOXA-antisense result whose dominant cell line is HOXA-low is being read into a HOXA-high disease. The direction of any effect could reverse with the underlying HOXA state, and nothing in the source separates the two lines' contributions. Oncofusion knowledge graph (internal analysis) 2026
  4. E4weak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.paper

    HOXA-AS2 is reported to be highly expressed in prednisone-insensitive ALL cell lines and patient samples and to enhance glucocorticoid resistance by up-regulating HOXA3 and thereby activating EGFR/Ras/Raf/MEK/ERK signalling, with TCF7L2 activating HOXA-AS2.

    TCF7L2 activated HOXA-AS2 decreased the glucocorticoid sensitivity in acute lymphoblastic leukemia through regulating HOXA3/EGFR/Ras/Raf/MEK/ERK pathway · Zhao Q, Zhao S, Li J, Zhang H, Qian C, Wang H, Liu J, Zhao Y · Biomedicine & Pharmacotherapy 109:1640-1649 2019
    PMID 30551418DOI
    What this evidence cannot show

    Graded weak. The MeSH-indexed cell model is Jurkat, a T-ALL line, whereas KMT2A-rearranged infant leukemia is B-lineage; patient sample n and fusion genotypes are not stated. No fusion stratification whatsoever. The mechanism is a generic proliferation cascade. Published in a journal with a high volume of unreplicated single-axis lncRNA papers. A near-identical glucocorticoid-resistance claim exists for the neighbouring lncRNA HOTAIRM1 by an unrelated mechanism, which raises the possibility of a locus-level or assay-level artefact rather than two independent mechanisms.

    Evidence that argues against this (1)
    • A separate group attributes leukemia glucocorticoid resistance to HOTAIRM1, a different HOXA-cluster antisense lncRNA ~12 kb away, acting through ARHGAP18/RHOA/ROCK1. Two neighbouring antisense transcripts each claimed to cause the same resistance phenotype by unrelated routes is more parsimoniously explained by HOXA-locus transcription tracking the resistant phenotype than by two mechanisms. Liang L et al. 2021
  5. E5established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    HOXA-AS2 is ENSG00000253552, lncRNA, GRCh38 chr7:27,107,777-27,134,302 plus strand, HGNC:43745; HOXA-AS3 is ENSG00000254369, lncRNA, GRCh38 chr7:27,129,977-27,155,928 plus strand, HGNC:43748. The two annotations overlap by approximately 4.3 kb, and both overlap the long HOXA3 gene model (chr7:27,106,184-27,152,610, minus strand), so short-read RNA-seq cannot unambiguously assign reads between them.

    What this evidence cannot show

    Gene-level Havana model coordinates only; the overlap figure is computed by us from the two gene-level intervals and does not account for exon structure, so exonic overlap may be smaller than 4.3 kb. The read-assignment claim is an inference from the annotation geometry, not an experimental measurement. Annotations change between Ensembl releases, and HOXA-AS2's boundaries in particular have been revised.

    Evidence that argues against this (1)
    • Two annotations that overlap each other by about 4.3 kb and both overlap the HOXA3 gene model are, from the point of view of any RNA-seq quantification, not separable. Treating HOXA-AS2 and HOXA-AS3 as distinct entities with distinct expression is an assumption the annotation itself does not support. Ensembl, EMBL-EBI 2026
  6. E6moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database

    As of 2026-09-18, PubMed contains 37 records for HOXA-AS3 in total and zero records for HOXA-AS3 in any leukemia context; the HOXA-AS2 leukemia literature comprises 8 records, none of which stratifies by or studies KMT2A rearrangement.

    What this evidence cannot show

    Symbol-based literature searches miss papers that use alternative names (HOXA3as, HOXA-AS3 has been called HOXA3os in some rodent work) or that report the transcript only in a supplementary table. Absence from title/abstract indexing is not proof that no data exist. Preprints are not indexed. This supports a claim about the published record, not about the biology.

    Evidence that argues against this (1)
    • A PubMed count is a measure of indexing and of naming, and this project has a standing rule that alias searches fail in both directions. HOXA-AS3 has been published under other designations, antisense transcripts are frequently discussed inside papers about the sense gene without appearing in title or abstract, and zero records therefore bounds the SEARCHABLE literature rather than the literature. NCBI 2026
Connected to