Cell-of-Origin in Diffuse Large B-Cell Lymphoma: The Hans Algorithm as a Prognostic Compass

Diffuse large B-cell lymphoma (DLBCL) is the most common aggressive lymphoma, and it isn't one disease. Under the microscope two tumors can look nearly identical — sheets of large transformed B-cells effacing the node — yet behave very differently. Part of that difference traces back to where in the B-cell's life cycle the malignancy arose. The concept of cell-of-origin (COO) captures this, and the Hans algorithm is the workhorse immunohistochemical tool that most laboratories use to approximate it [1].

What the test measures

Normal B-cells pass through the germinal center of a lymph node, where they mature, mutate their immunoglobulin genes, and are selected for high-affinity antibody production. Gene-expression profiling showed years ago that DLBCL falls into at least two molecular groups reflecting that journey: a germinal-center B-cell (GCB) group with a transcriptional signature resembling normal germinal-center cells, and an activated B-cell (ABC) group resembling B-cells that have exited the germinal center and switched on survival programs. These groups differ prognostically.

Gene-expression assays aren't available in most routine labs, so Hans and colleagues asked a practical question: can a handful of antibodies stand in for the microarray? Their answer, published in 2004, was the three-marker algorithm now bearing the senior author's name [1]. It measures the protein expression of CD10, BCL6, and MUM1 (IRF4) — markers that track germinal-center versus post-germinal-center biology.

How it's tested

The assay is immunohistochemistry on formalin-fixed, paraffin-embedded (FFPE) tissue — the same block used for the diagnostic workup. That's a real strength: no fresh or frozen tissue is needed, and the stains can be run on the original biopsy. Preanalytic quality still matters. Underfixation, prolonged cold ischemia, and decalcified bone marrow specimens can all degrade nuclear antigens like BCL6 and MUM1, and a weak or patchy stain can push a case across the decision threshold.

Scoring uses a 30% cutoff for each marker — a cell population is called positive when at least 30% of tumor nuclei (or membranes, for CD10) stain [1]. The decision tree runs in sequence. If CD10 is positive, the case is GCB. Full stop. If CD10 is negative, the pathologist looks at BCL6: a BCL6-negative case is non-GCB. If BCL6 is positive, MUM1 breaks the tie — MUM1-positive means non-GCB, MUM1-negative means GCB.

In practice, the 30% threshold sounds crisp but the staining rarely is. Pathologists often find themselves counting a borderline case twice, because a tumor sitting at 25% versus 35% BCL6 positivity lands in different buckets. That interobserver softness is worth remembering when you read a report.

What each result state means

The algorithm returns one of two states: GCB or non-GCB. The non-GCB bin is the IHC surrogate for the molecularly defined ABC group, though the two aren't identical — more on that below.

Prognostically, the two point in opposite directions. GCB tumors have historically shown more favorable outcomes; non-GCB (ABC-like) tumors have historically fared worse in the immunochemotherapy era [1]. This isn't a small statistical footnote — it was one of the first molecular distinctions in DLBCL to survive independent validation and enter routine reporting.

A caution for trainees: this is a prognostic marker, not a diagnostic one. A GCB result doesn't make the diagnosis of DLBCL, and it doesn't rule anything in or out. It refines the risk conversation for a lymphoma already diagnosed on morphology and a broader immunophenotype.

What it determines for treatment eligibility

Here's where the biology becomes clinically interesting, and where care is needed in how we frame it. The GCB/non-GCB split isn't just prognostic bookkeeping; the two groups depend on different survival machinery, and that machinery is drug-targetable in principle.

The ABC (non-GCB) subgroup is characterized by constitutive activation of the NF-κB survival pathway, often driven downstream of chronic B-cell receptor (BCR) signaling. That mechanistic dependence is why agents that interrupt BCR signaling — for example, inhibitors acting on kinases in that pathway — and agents targeting the NF-κB axis have been studied preferentially in non-GCB disease, where the rationale is strongest. GCB tumors, by contrast, lean on different programs, including germinal-center transcription factors and, in some cases, distinct genetic lesions. So COO can inform eligibility for, or enrichment into, trials and regimens built around a drug class whose activity differs by subgroup. It doesn't tell any individual patient which drug they should take — that's a clinical decision made with the full picture, and this article isn't the place for it.

The honest summary of the evidence: COO has helped rationalize which mechanisms of action ought to matter in which subgroup, and it has shaped how targeted-agent studies are designed. Translating that into consistent survival gains has been harder than the biology promised.

Caveats and what's evolving

The Hans algorithm is a surrogate, and a surrogate has limits. Compared against gene-expression profiling, IHC-based COO misclassifies a meaningful minority of cases — the non-GCB bin, in particular, isn't a clean stand-in for molecular ABC. The three-antibody design was a pragmatic compromise, and it shows.

Several things soften the tool's reliability in day-to-day use. The 30% cutoffs are convention, not biological bright lines. Antibody clones, detection systems, and fixation vary between laboratories, so a case called GCB in one lab could conceivably read non-GCB in another. And the binary output collapses what's really a spectrum; a subset of tumors sits in an unclassified molecular zone that the Hans tree has no box for.

The field has also moved on in how it thinks about DLBCL. The current WHO classification incorporates COO alongside a growing list of genetic and molecular features, reflecting a more granular view than a two-way split can capture [2]. More reproducible, RNA-based assays run on FFPE — the class of tools exemplified by digital gene-expression platforms — were developed precisely to deliver the molecular COO call without the interobserver drift of IHC. Where such assays are available, they offer a firmer molecular assignment than the antibody surrogate.

What the Hans algorithm can't tell you is often as important as what it can. It won't identify double-hit genetic lesions, it won't resolve the molecularly unclassified cases, and it won't capture the finer genetic subgroups that increasingly shape how aggressive lymphomas are understood. What it does — quickly, cheaply, on tissue you already have — is give a first-pass biological orientation that has held prognostic value for two decades. Read alongside its limits, that's a genuinely useful thing for a report to say.

---

References

1. Hans CP, Weisenburger DD, Greiner TC, et al. Confirmation of the molecular classification of diffuse large B-cell lymphoma by immunohistochemistry using a tissue microarray. Blood. 2004. doi:10.1182/blood-2003-05-1545 (PMID:14695068).

2. Alaggio R, Amador C, Anagnostopoulos I, et al. The 5th edition of the WHO Classification of Haematolymphoid Tumours: Lymphoid Neoplasms. Leukemia. 2022. doi:10.1038/s41375-022-01620-2.