Oncofusion

start / graph / fusion-npm1-alk

Fusionestablished confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.Q11-comparative

NPM1::ALK (NPM-ALK), t(2;5)(p23;q35)

also called NPM-ALK · NPM1-ALK · p80 NPM/ALK · t(2;5) ALCL · NPM::ALK

In this lymphoma a gene for a busy nuclear shuttle protein, NPM1, is fused to the business end of a receptor enzyme called ALK, and because NPM1 naturally clumps together it forces the enzyme to switch itself on permanently. That enzyme has a real pocket where ATP normally binds, so a pill that fills the pocket shuts the cancer off - which is why this fusion is largely treatable and KMT2A::AFF1 is not.

The same thing, for a scientist

NPM1::ALK, generated by t(2;5)(p23;q35), was the first ALK fusion described and defines ALK-positive anaplastic large-cell lymphoma (ALCL) [E1]. The chimera joins the **N-terminal 117 residues of nucleophosmin - its oligomerisation and nucleolar-shuttling segment - to the entire cytoplasmic portion of ALK**, giving a 75-80 kDa protein [E2]. The mechanism is textbook **constitutive kinase** activation by forced oligomerisation: NPM-ALK forms multimers of >=200 kDa that also contain normal NPM, and this oligomerisation is *absolutely required* - three different NPM-segment deletion mutants each failed to multimerise, lacked kinase activity in vivo and failed to transform [E2]. Decisively for the comparative question, NPM could be **functionally replaced** by the unrelated TPR leucine zipper with almost equal transforming efficiency, proving the 5' partner's only requirement is dimerisation and that the partner interface is not itself the target [E2]. NPM1-ALK accounts for roughly 90% of paediatric ALK-positive ALCL (58/66 in one consecutive series; variants TPM3-, ATIC- and MYH9-ALK make up the rest) [E3]. Because the retained ALK module is a classical ATP-site tyrosine kinase, crizotinib produced overall response rates of 83% and 90% in relapsed/refractory paediatric ALK-positive ALCL with complete responses in 80-83% [E4]. Front-line chemotherapy already achieves 10-year PFS 70% and OS 90% [E5]. This node exists in the graph as the positive control: the fusion whose partner supplies oligomerisation and whose effector supplies a drug pocket.

What it looks like
loading 2XP2…
PDB 2XP2 ↗
shape

This is real experimental data — atoms placed by crystallography or cryo-EM, deposited by the people who did the experiment.

I have never read one of these pictures before

A protein is one long chain of amino acids that folds up. The ribbon traces that chain from one end to the other; it is not a wire, it is a simplification of where the backbone runs. Drag to rotate, scroll to zoom.

Domains are the stretches of chain that do a particular job, and they are what the coloured blocks above the picture mark out. Click one and the view flies to it. In this project the important question about any domain is usually whether the fusion still has it — the labels say lost where it does not.

Anything in yellow sticks is a bound small molecule, which is usually the drug.

Go deeper
What is kept and what is lost on each side1 deeper

Kept from NPM1 (the 5' partner): the N-terminal 117 amino acids, comprising the core oligomerisation domain (an eight-stranded beta-barrel that assembles as a pentamer, and pentamer-of-pentamers decamers) together with a nucleolar-localisation/shuttling capability 2.

Lost from NPM1: everything C-terminal of residue 117 - the acidic histone-chaperone regions, the nuclear-localisation signal, the nucleic-acid-binding C-terminal domain, and the tryptophan residues whose mutation causes NPM1-mutant AML. So the fusion loses all of NPM1's chaperone and ribosome-biogenesis chemistry and keeps only the self-assembly module.

Kept from ALK (the 3' partner): the entire cytoplasmic portion, including the complete tyrosine kinase domain with its canonical bilobal fold, glycine-rich loop, gatekeeper, hinge and ATP-binding cleft, plus the C-terminal tail with its PLCgamma, SHC and IRS-1 docking tyrosines 12.

Lost from ALK: the entire extracellular ligand-binding region (MAM domains, LDLa, glycine-rich region) and the transmembrane domain. The consequence is that the kinase is now cytoplasmic/nucleoplasmic and ligand-independent - it cannot be regulated by its normal extracellular cue.

The architectural contrast with KMT2A::AFF1. Both fusions keep a targeting/assembly module from the 5' partner and an effector from the 3' partner. The difference is entirely in the effector: ALK's kinase domain is a folded enzyme with a deep, conserved, evolutionarily druggable ATP pocket; AFF1's pSER/CHD contribution is intrinsically disordered scaffold with no pocket at all. Same grammar, different vocabulary - and the vocabulary is what determines curability.

Subcellular localisation is a red herring (and the experiment that proved it)

NPM-ALK is found in cytoplasm, nucleoplasm and nucleolus, because the retained NPM segment carries it there and it also heterodimerises with endogenous full-length NPM 2. It is tempting to build a mechanism on that nucleolar localisation. Bischof and colleagues killed the idea cleanly: an engineered TPR-ALK hybrid, using the TPR leucine zipper instead of NPM, is purely cytoplasmic yet transformed cells almost as efficiently as NPM-ALK 2. Nuclear/nucleolar trafficking is therefore dispensable; oligomerisation is not.

This is one of the most useful negative results in the fusion field, and it has a direct bearing on the lab's central hypothesis. For NPM1::ALK, the partner interface is necessary but interchangeable - so drugging that interface would be pointless, because any other dimerisation module substitutes. Contrast menin-KMT2A, where the interface is necessary and (as far as anyone knows) not interchangeable, because it is not a generic dimerisation surface but a specific adaptor that also delivers LEDGF and chromatin targeting.

Mechanism: constitutive kinase, in the strict sense

Class: constitutive-kinase.

  1. The NPM 1-117 segment self-assembles, bringing two or more ALK kinase domains into proximity.
  2. Proximity permits trans-autophosphorylation of the activation loop, locking the kinase in its active conformation without any ligand.
  3. The active kinase phosphorylates itself and substrates, recruiting adaptors through phosphotyrosine motifs and firing JAK/STAT3, PI3K-AKT, RAS-MAPK and PLCgamma. STAT3 activation is the dominant and most therapeutically relevant output in ALCL.
  4. Downstream, STAT3 drives a program including CD30 and IL-2R expression and suppression of T-cell identity genes, producing the characteristic ALCL phenotype (CD30-bright, frequently with loss of CD3 and other T-cell antigens).

The requirement for oligomerisation is not inferred - it is demonstrated by loss-of-function: NPM-segment deletion mutants failed to form complexes, had no in vivo kinase activity and did not transform 2.

Why the mechanism is druggable. The ATP cleft of a tyrosine kinase is a pre-evolved small-molecule binding site. Crizotinib occupies it, and there is a crystal structure of human ALK bound to crizotinib (PDB 2XP2). Nothing analogous exists or can exist for a disordered elongation scaffold. The lesson the comparative graph should take is not 'kinase fusions are easier' but 'fusions whose retained effector is a folded enzyme inherit a drug pocket for free; fusions whose retained effector is a disordered scaffold must borrow one from a partner.'

Clinical phenotype, cell of origin and age

Disease: ALK-positive anaplastic large-cell lymphoma, a mature (post-thymic) T-cell lymphoma, plus a minority of ALK-positive diffuse large B-cell lymphoma and rare other presentations. The same ALK kinase fused to other partners (EML4-ALK) drives lung adenocarcinoma, and fused to TPM3/RANBP2 drives inflammatory myofibroblastic tumour - a single effector reused across tissues.

Cell of origin: a mature, activated, post-thymic CD4+ T lymphocyte (frequently expressing cytotoxic granule proteins and having lost several pan-T antigens). This is the most mature cell of origin anywhere in this comparison set - the opposite extreme from KMT2A::AFF1's fetal lympho-myeloid progenitor.

Age and presentation: a disease of children, adolescents and young adults, with a paediatric peak around 10-12 years. Presentation is typically systemic: peripheral and mediastinal lymphadenopathy, B symptoms (fever, weight loss), frequent extranodal involvement of skin, bone, soft tissue and lung, and a characteristic waxing-and-waning course before diagnosis.

Frequency: roughly 90% of paediatric ALK-positive ALCL carries NPM1-ALK - 58 of 66 in a consecutive series in which only one case was ALK-negative, with TPM3-ALK (n=2), ATIC-ALK (n=1) and MYH9-ALK (n=1) accounting for most of the remainder; nuclear-plus-cytoplasmic ALK staining perfectly predicted NPM1-ALK while cytoplasm-only staining predicted a variant partner 3.

Outcome: good, and good before targeted therapy. In 420 children and young adults treated on the international ALCL99 protocol, 10-year PFS was 70% and OS 90%, with rare late relapses and no secondary malignancies reported 5. Risk stratification by small-cell/lymphohistiocytic morphology plus minimal disseminated disease separated 10-year PFS of 40%, 75% and 86% 5.

Druggability and therapeutic status

Clinically validated, on the fusion protein itself.

Crizotinib. In the Children's Oncology Group study, 26 patients with relapsed/refractory ALK-positive ALCL received crizotinib; overall response rates were 83% at 165 mg/m2 and 90% at 280 mg/m2, with complete responses in 83% (5/6) and 80% (16/20) respectively, and NPM-ALK transcript levels fell during therapy in most patients 4. The commonest drug-related adverse event was neutropenia (33% and 70% by dose level) 4. Median duration of therapy reached 2.79 years in the lower-dose group 4.

Next-generation ALK inhibitors (ceritinib, alectinib, brigatinib, lorlatinib) were developed largely in ALK-rearranged lung cancer and address ATP-site resistance mutations and CNS penetration; they are increasingly used in ALCL and inflammatory myofibroblastic tumour.

The comparison that matters for this graph.

NPM1::ALKKMT2A::AFF1
Targetthe fusion protein's own ATP pocketa partner protein's peptide-binding pocket (menin)
Best single-agent response in R/R diseaseORR 83-90%, CR 80-83% 4CR+CRh 22.8%, ORR 63.2%
Front-line cure fraction before targeted therapy10-y PFS 70%, OS 90% 56-y EFS 46.1%
Partner interface as a drug targetpointless - interchangeable with any dimeriser 2the only validated node

Bite-back is real even here. ATP-site resistance mutations (the ALK equivalents of BCR-ABL1 T315I) arise under selection; ALK is required in normal neural development; and in ALCL the disease is already ~90% curable with chemotherapy, so a targeted agent has to beat a high bar rather than rescue a hopeless one.

Analogy: two ways of forcing a kinase on, and the axis where the class stops explaining anything

Closest analogue, and the axis. fusion-etv6-ntrk3, on the axis of shared mechanism class: a 5-prime partner that contributes nothing but forced self-association, joined to a retained receptor tyrosine kinase domain stripped of its ligand-binding ectodomain. mechanism-constitutive-kinase-activation holds the class definition; this section is about what the pair inside it settles and what it does not.

What the analogy predicts, and the test that has been run on both sides. If the 5-prime partner's only contribution is oligomerisation, it should be interchangeable. That is the single most load-bearing experiment in this entry: NPM-ALK forms multimers of at least 200 kDa, three different NPM-segment deletion mutants each failed to multimerise and each lost kinase activity in vivo and transforming ability, and NPM could be functionally replaced by the unrelated TPR leucine zipper with almost equal transforming efficiency 2. The same logic holds for ETV6::NTRK3, whose 5-prime module is the ETV6 SAM/PNT self-association domain. The prediction and its confirmation are why this node exists in the graph as a positive control: the partner interface is not the target; the kinase is.

Where the two differ on the mechanism axis itself, which is the fine-grained half. NPM1 self-associates as a closed oligomer; ETV6's SAM domain forms an open-ended head-to-tail helical polymer through a mid-loop/end-helix interface, which is a mechanistically stronger form of forced self-association than a closed oligomer. Same class, different stoichiometric ceiling. Nobody has asked whether that difference has a consequence - for instance whether polymer-driven fusions tolerate lower expression, or resist partner-directed disruption differently - and it is answerable with the swap constructs that already exist.

Where the analogy breaks, and this is the disanalogy worth carrying out of this entry. The two fusions are in the same mechanism class and their tissue distributions are not comparable:

NPM1::ALKETV6::NTRK3
diseasesessentially one - ALK-positive anaplastic large-cell lymphoma, ~90% of paediatric casescongenital fibrosarcoma, cellular mesoblastic nephroma, secretory breast carcinoma in 12 of 13 cases
lineagelymphoidmesenchymal and epithelial
what sets the diseasenot the mechanismnot the mechanism

Identical mechanism, one fusion confined to a single lymphoid entity and the other recurring across three unrelated tissues. So mechanism class predicts druggability extremely well and predicts which disease you get not at all. The variable that does the work is the cell the fusion lands in, and neither fusion's mechanism contains any information about it.

And the reason this entry is in a KMT2A graph at all: it is the contrast class. Both kinase fusions are drugged to response rates in the 57-90% range because the retained effector is an enzyme with a deep, conserved ATP cleft. KMT2A fusions always lose the KMT2A SET domain at the breakpoint and retain no catalytic module of any kind, so the drug has to engage a protein-protein interface on a cofactor (menin) instead of an active site on the driver. Reading the kinase class as the template for what a fusion oncoprotein is has repeatedly set the wrong expectation for chromatin fusions - see the resistance analogy in fusion-bcr-abl1, where the escape geometry transfers and the rescue route does not.

Facts and figures
gene5NPM1
gene3ALK
cytobandt(2;5)(p23;q35)
frequency note88% is 58/66 of all childhood ALCL in the cited consecutive series; expressed as a share of ALK-POSITIVE paediatric ALCL the figure the authors give is 'almost 90%'.
prognosisGood and already good before targeted therapy: international ALCL99 10-year PFS 70% and OS 90% in 420 children and young adults, with biological risk groups at 40%, 75% and 86% 10-year PFS. In relapsed/refractory disease crizotinib gave ORR 83-90% with CR in 80-83%.
uniprot5P06748
uniprot3Q9UM73
uniprot5 verifiedP06748 = NPM_HUMAN, gene NPM1, nucleophosmin, 294 aa, Homo sapiens (UniProt REST, checked 2026-09-18)
uniprot3 verifiedQ9UM73 = ALK_HUMAN, gene ALK, ALK tyrosine kinase receptor, 1620 aa, Homo sapiens (UniProt REST, checked 2026-09-18)
PDB noteNo structure of the intact NPM1::ALK fusion protein or of its oligomeric assembly exists. 2XP2 = the human ALK kinase domain in complex with crizotinib - the retained effector module with its drug in the ATP cleft, i.e. the actual therapeutic target; 4N8M = structural polymorphism in the N-terminal oligomerisation domain of NPM1, the retained 5' module whose self-assembly forces kinase activation. Together the two entries cover both halves of the mechanism as separate fragments. Both verified against the RCSB entry API on 2026-09-18.
druggability classclinically-validated
druggability noteValidated ON THE FUSION PROTEIN ITSELF, at a pre-evolved ATP-binding cleft - the opposite situation from KMT2A::AFF1, which must borrow a pocket from a partner. Note that the partner INTERFACE here is explicitly NOT a useful target: NPM's oligomerisation segment is functionally interchangeable with an unrelated leucine zipper, so any interface-directed agent would be bypassed by a different dimeriser.
stage of hematopoiesismature, activated post-thymic CD4+ T lymphocyte (frequently cytotoxic-granule-positive with loss of several pan-T antigens) - the most mature cell of origin in this comparison set
age peak~10-12 y (paediatric/adolescent peak); ALK-positive ALCL is predominantly a disease of children and young adults
developmental order rank10
developmental order noteRanked 10 of 10 - the latest and most mature cell of origin in the set. ALCL arises in a fully differentiated peripheral T cell, in direct contrast to KMT2A::AFF1 (rank 1, fetal lympho-myeloid progenitor). This node therefore anchors the mature end of the developmental axis the dashboard orders fusions along.
other drugs noteNext-generation ALK inhibitors (ceritinib, alectinib, brigatinib, lorlatinib) were developed mainly in ALK-rearranged lung cancer and address ATP-site resistance mutations and CNS penetration; they do not yet have their own nodes in this graph.
diseases
[
  "ALK-positive anaplastic large-cell lymphoma",
  "ALK-positive diffuse large B-cell lymphoma (rare)"
]
frequency
{
  "paediatric_alk_positive_alcl_pct": 88,
  "paediatric_alcl_npm1_alk_n_over_total": "58/66",
  "paediatric_alcl_alk_negative_n_over_total": "1/66",
  "variant_x_alk_partners_n": 7
}
breakpoints
[
  {
    "id": "bp-npm1-e4-alk-e20",
    "gene5_exon_last": 4,
    "gene3_exon_first": 20,
    "frequency_pct": 100,
    "population": "NPM1::ALK-positive ALCL; the fusion is essentially monomorphic at the transcript level - the breakpoint reproducibly joins NPM1 codon 117 to the start of the ALK cytoplasmic domain, which is why a single RT-PCR assay detects it and can be used for minimal-disseminated-disease monitoring",
    "domains_retained_5": [
      "NPM1 oligomerisation domain (core beta-barrel, residues ~1-117)",
      "nucleolar-localisation / shuttling capability"
    ],
    "domains_lost_5": [
      "acidic histone-chaperone regions A1/A2",
      "nuclear localisation signal",
      "C-terminal nucleic-acid-binding domain",
      "Trp288/Trp290 nucleolar-retention residues"
    ],
    "domains_retained_3": [
      "complete tyrosine kinase domain with ATP-binding cleft",
      "juxtamembrane cytoplasmic segment",
      "C-terminal tail with PLCgamma / SHC / IRS-1 docking tyrosines"
    ],
    "domains_lost_3": [
      "entire extracellular region (two MAM domains, LDLa, glycine-rich region)",
      "transmembrane domain",
      "signal peptide"
    ],
    "phenotype_notes": "Loss of the extracellular region makes the kinase ligand-independent; retention of NPM1 1-117 makes it constitutively oligomeric and therefore constitutively autophosphorylated. Nuclear-plus-cytoplasmic ALK immunostaining is pathognomonic for this partner, whereas cytoplasm-only staining indicates a variant X-ALK fusion. Exon numbers follow the standard NPM1/ALK transcript conventions and correspond to the well-established codon-117 junction."
  }
]
PDB IDS
[
  "2XP2",
  "4N8M"
]
mechanism class
[
  "constitutive-kinase"
]
drugs
[
  "drug-crizotinib"
]
key interactors
[
  "STAT3",
  "JAK3",
  "PI3K-AKT",
  "RAS-MAPK",
  "PLCgamma",
  "wild-type NPM1 (heterodimerisation partner)"
]
trials
[]
How this could bite us back
  • Resistance: ATP-site mutations in the ALK kinase domain (the ALK analogues of BCR-ABL1 gatekeeper mutations) arise under inhibitor selection, which is why four generations of ALK inhibitors exist. A drug pocket is not a durable solution, merely a good one.
  • Compensation: ALK signalling is redundant downstream (STAT3, PI3K-AKT, RAS-MAPK in parallel), so bypass activation of STAT3 or AKT can sustain the leukemic/lymphoma program with the kinase inhibited.
  • Essential-function toxicity: ALK has roles in neural development and crizotinib is a multi-kinase agent (MET, ROS1), producing neutropenia in up to 70% of children at the higher dose level plus visual disturbance and hepatotoxicity. NPM1 itself is haploinsufficient and essential, so any strategy touching wild-type NPM1 is off the table.
  • Clinical-equipoise hazard: with 10-year OS already 90% on chemotherapy, a targeted agent tested front-line risks doing harm where there is little room for benefit - the mirror image of the KMT2A::AFF1 situation.
  • Delivery: CNS relapse in ALCL and the limited CNS penetration of crizotinib drove the development of later-generation inhibitors; sanctuary sites remain a practical limit on a well-drugged target.
  • Conceptual bite-back for the lab's hypothesis: this node is the strongest evidence AGAINST 'obligate partner interfaces are generalisable drug targets'. Here the partner interface is obligate and demonstrably useless as a target, because it is generic. Obligate does not imply druggable.
  • If the NPM1 moiety is swappable for an unrelated leucine zipper, then 'obligate oligomerisation interface' is a description of a requirement, not of a target: there is nothing partner-specific for a molecule to bind.
Still open
  1. The NPM oligomerisation segment is functionally interchangeable with an unrelated leucine zipper, so this fusion's partner interface is necessary but not specific. Is the menin-KMT2A interface genuinely different in kind - a specific adaptor that also delivers LEDGF and chromatin targeting - or is it also, in principle, substitutable? An experiment replacing MBM1/MBM2 with a heterologous targeting module in a KMT2A fusion would settle whether the menin lesson generalises or is a lucky special case.
  2. ALK-positive ALCL is already ~90% curable with chemotherapy. Can ALK inhibition replace rather than supplement cytotoxic therapy, reducing late effects in children, and what is the relapse rate on inhibitor withdrawal given that the fusion is not eradicated?
  3. Why does this fusion arise in a fully mature T cell while every KMT2A fusion arises in a progenitor? Is a constitutive kinase simply able to transform a differentiated cell in a way that a chromatin-program hijack cannot, and if so, does that mean the KMT2A fusions are permanently tied to a developmental window?
  4. The same ALK kinase domain fused to EML4 makes lung adenocarcinoma and fused to TPM3/RANBP2 makes inflammatory myofibroblastic tumour. What determines which tissue tolerates the fusion - the partner promoter's expression pattern, or a tissue-intrinsic permissiveness that the KMT2A family would also obey?
  5. Does ALK have any allosteric or autoinhibitory pocket analogous to the ABL1 myristoyl site? That, rather than another ATP-competitive generation, is what the CML history says determines whether a kinase fusion stays tractable.
  6. Does the geometry of forced self-association matter, or only its existence? NPM1 gives a closed oligomer and ETV6's SAM domain an open-ended polymer, and the TPR-leucine-zipper swap shows the NPM module is replaceable - but no experiment has swapped a POLYMERISING module into NPM1::ALK or a closed oligomeriser into ETV6::NTRK3 and measured transforming efficiency, kinase output or sensitivity to the respective inhibitor. If the polymer is functionally superior, partner geometry becomes a variable in kinase-fusion prognosis; if it is not, oligomerisation is a switch rather than a dial.
Every source behind this entry (6)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    The t(2;5)(p23;q35) of non-Hodgkin's lymphoma fuses the nucleolar protein gene NPM to the kinase gene ALK, creating a chimeric tyrosine kinase.

    Fusion of a kinase gene, ALK, to a nucleolar protein gene, NPM, in non-Hodgkin's lymphoma · Morris SW, Kirstein MN, Valentine MB, Dittmer KG, Shapiro DN, Saltman DL, Look AT · Science 1994
    PMID 8122112DOI
    What this evidence cannot show

    A cloning and characterisation paper; it identifies the fusion and the novel ALK gene but does not establish the oligomerisation mechanism, the cell of origin or the therapeutic implication. ALK's normal physiological ligand and role were unknown at the time.

    Evidence that argues against this (1)
    • The transcript is present in people who do not have lymphoma, which bounds what cloning it from a tumour establishes. Trumper et al. detected NPM/ALK fusion cDNA by RT-PCR, confirmed by hybridisation with a breakpoint-specific oligonucleotide, in peripheral blood cells of 14 of 29 healthy donors (48%); the positive-cell frequency was too low to assign to any defined subpopulation. t(2;5) thereby joins t(14;18) and t(9;22) as a tumour-associated translocation product recoverable from normal blood. Two consequences for this node: the founding paper identifies a fusion, not a sufficient cause; and any diagnostic or minimal-residual-disease application of NPM1-ALK PCR needs a quantitative threshold rather than a binary positive/negative call. Trumper L et al. 1998
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    NPM-ALK joins the N-terminal 117 residues of nucleophosmin to the entire cytoplasmic portion of ALK; oligomerisation mediated by the NPM segment is absolutely required for kinase activation and transformation, and NPM can be functionally replaced by the unrelated TPR leucine zipper with almost equal transforming efficiency despite purely cytoplasmic localisation.

    PMID 9121481DOI
    What this evidence cannot show

    Fibroblast transformation assays and sedimentation-gradient biochemistry with overexpressed constructs, not endogenous-level lymphoma cells; 'almost as efficiently' for TPR-ALK is a qualitative comparison. Whether nuclear/nucleolar localisation modulates the disease phenotype in patients (as opposed to being required for transformation) is not addressed.

    Evidence that argues against this (1)
    • This result is usually quoted as proof that the NPM1 oligomerisation interface is the essential, targetable element. It shows the opposite about targetability: NPM can be functionally replaced by the unrelated TPR leucine zipper with almost equal transforming efficiency and purely cytoplasmic localisation. What is required is oligomerisation as a generic property, not the NPM1 surface in particular -- so there is no specific interface here for a drug to recognise, and the tractable site remains the ALK ATP pocket. Bischof D et al. 1997
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Almost 90% of paediatric ALK-positive anaplastic large-cell lymphomas express NPM1-ALK: in 66 consecutive childhood ALCLs, 58 were NPM1-ALK, one was ALK-negative, and the remaining seven were cytoplasm-only ALK-positive with TPM3-ALK (n=2), ATIC-ALK (n=1), MYH9-ALK (n=1) or an unidentified partner (n=3).

    PMID 19545284DOI
    What this evidence cannot show

    n=66 from a single trial-based German cohort; the RT-PCR panel screened only TPM3, TFG, ATIC, CLTC and MYH9, so three cases remained unassigned and rarer partners could be under-counted. Adult ALCL has a higher ALK-negative fraction, so this figure should not be generalised beyond paediatrics.

    Evidence that argues against this (1)
    • The 90% is conditioned twice - on childhood and on ALK positivity - and both conditions carry most of the weight. In the International Peripheral T-Cell Lymphoma Project, 1314 consensus-reviewed cases across 22 institutions yielded 181 ALCL (13.8%), of which 159 were systemic (12.1%), and ALK-NEGATIVE ALCL formed a large share of them with markedly worse outcome: 5-year failure-free survival 36% versus 60% and overall survival 49% versus 70% for ALK-positive (p=.015 and p=.016). In an unselected, predominantly adult ALCL population the NPM1-ALK share of all anaplastic large-cell lymphoma is therefore far below 90%, and the favourable prognosis this node attaches to the fusion belongs specifically to the ALK-positive, largely paediatric slice. A 66-case consecutive childhood series is the best case for the fusion, not the population case. Savage KJ et al. 2008
  4. E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    Crizotinib produced overall response rates of 83% (165 mg/m2) and 90% (280 mg/m2) with complete responses in 83% (5/6) and 80% (16/20) in 26 children with relapsed/refractory ALK-positive ALCL, and NPM-ALK transcript levels fell during therapy in most patients.

    PMID 28787259DOI
    What this evidence cannot show

    Single-arm phase I/II with 26 ALCL patients split across two dose levels, so the 83% versus 90% difference is not a meaningful dose comparison; 12 patients left protocol therapy for transplantation, which confounds duration-of-response interpretation. Neutropenia affected 70% at the higher dose. No randomised comparison against salvage chemotherapy.

    Evidence that argues against this (1)
    • Crizotinib's response rate did not change the standard of care, because chemotherapy had already largely solved this disease: on the international ALCL99 protocol 10-year progression-free survival was 70% and overall survival 90% without any ALK inhibitor. The kinase class's reputation for 'reliably producing drugs' partly reflects diseases where the drug is added to an already-good backbone. Mussolin L et al. 2020
  5. E5established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    In 420 children, adolescents and young adults treated on the international ALCL99 protocol, 10-year progression-free survival was 70% and overall survival 90%, with biological risk groups separating 10-year PFS of 40%, 75% and 86%.

    PMID 32987765DOI
    What this evidence cannot show

    ALCL99 enrolled ALCL by histology, so the cohort is ALK-positive-dominant rather than NPM1-ALK-confirmed throughout; minimal disseminated disease was assessable in only 162 of 420 patients, so the biological risk groups rest on a subset. This outcome reflects intensive multiagent chemotherapy, not ALK inhibition.

    Evidence that argues against this (1)
    • Risk stratification in this disease has not yet converted into a therapy that changes the failure rate, which is the part a 70%/90% headline hides. In the randomised ALCL99-vinblastine trial, 217 children and adolescents with high-risk ALCL were assigned to the same 4-month regimen with or without vinblastine plus weekly maintenance to one year. Vinblastine cut the hazard of an event during the first year (HR 0.31, 95% CI 0.15-0.67, p=.002), giving 1-year EFS of 91% versus 74% - and then the hazard REVERSED (HR 4.98, 95% CI 1.65-15.0, p=.003), so 2-year EFS was 73% versus 70% and the overall EFS curves did not differ (HR 0.91, 95% CI 0.55-1.5, p=.71). The authors' conclusion was explicit: the intervention 'significantly delayed the occurrence of relapses but did not reduce the risk of failure'. Biological risk groups separating 10-year PFS at 40%, 75% and 86% describe who fails; no randomised intervention in this disease has yet moved that. Le Deley MC et al. 2010
  6. E6established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    The retained ALK module is a classical ATP-site tyrosine kinase whose cleft is occupied by a clinically used small molecule, as shown by the crystal structure of the human ALK kinase domain in complex with crizotinib.

    What this evidence cannot show

    The structure covers the isolated ALK kinase domain, not the NPM1::ALK fusion protein or its oligomeric assembly, so it demonstrates that the effector module is ligandable rather than depicting the oncoprotein. Verified against the RCSB entry API on 2026-09-18.

    Evidence that argues against this (1)
    • And an occupied ATP pocket is still an occupancy drug. The founding fusion of this class, BCR::ABL1, showed that escape comes by point mutation at the drug-binding surface with the mechanism intact, and that the field's rescue required a second, orthogonal site (the ABL1 myristoyl autoinhibitory pocket, asciminib) rather than more chemotypes at the same site. No equivalent orthogonal site is described for ALK. Wylie AA et al. 2017
Connected to