Skip to content

December 2025 — Monthly Briefing

1. What was read

637 research articles from 32 of the 33 journals on the roster.

A further 292 records (31% of the 929 this month's harvest returned) carry no abstract and are not research: news, correspondence, editor highlights and obituaries. PubMed types them Journal Article and nothing else, so the publication-type filter cannot see them — across the whole index they are 22.7% of it and 35% of general, and 3,801 of them are JAMA. Every count below is over the full harvest and is inflated by this; the figure above is not.

General Cancer Immune Bioinfo Sysbio
431 200 90 183 25

Volume swing against the median month: bioinfo -35%

Check the source before reading anything into this. Two benign causes account for most of these. bioinfo every January carries the Nucleic Acids Research database issue and runs high. And several journals simply deposit nothing in some months: across the 44 months read, 4 of 33 journals are absent from at least one, worst being Nat Mach Intell (20 months missing, 81 papers in total) and Bioinformatics (4 months missing, 2,905 papers in total). This month is missing Nucleic Acids Res. Queried directly, PubMed returns zero for those journal-months too — the holes are in the source, reproduced exactly by re-running the harvest. So a swing here is not evidence that a field went quiet, and is not usually evidence that the pipeline broke either.

Largest contributors: JAMA 113, Nature 71, Science 67, J Immunother Cancer 54, Nat Med 45, Nat Methods 41.

2. Net change on the thread curve

Thread This month Cumulative Status
agentic-ai −1 +17 / −4 forming
cell-therapy +1 / −1 +202 / −64 forming
ctdna-mrd +2 / −1 +74 / −24 forming
foundation-models +4 / −2 +132 / −84 splitting
immunotherapy-resistance −3 +220 / −85 forming
interception +2 +127 / −23 forming
method-credibility −8 +18 / −484 crisis
neoantigen-vaccines +1 +80 / −41 splitting
neuro-immune-microbiome-axis +2 +191 / −20 forming
next-gen-checkpoints −3 +68 / −47 splitting
single-cell-3d-genome +3 +77 / −17 forming
suppressive-population-inversion +1 / −6 +53 / −95 contested

Status is derived by rule, not asserted. Fewer than 5 cards → too-early; then refutes ≥ 2×supports → crisis; refutes > supports → contested; supports ≥ 2×refutes → forming; else splitting.

3. The evidence that moved it

The divergence criterion was published as a title. 41372473 (Nature Cancer): foundation models in oncology win benchmarks but miss the clinic. Every other instance of this pattern in the corpus had to be assembled by putting an enthusiastic preclinical block next to a negative or absent clinical one. Here the journal states it. It is the single most useful card in the file, because it is the only one where the reading and the finding are the same object.

The virtual cell arrived as a Nature Methods title (41360943), alongside Artificial intelligence agents for biology, Novae and Nicheformer. The vocabulary, the named models and the audits are in one journal in one month, which is the part that can be read off the rows. No interval is quoted from the Cell Virtual Cell Challenge any more: the phrase was already a Nature Biotechnology title before that challenge launched, so a span measured forward from it measures nothing.

Corrected 2026-08-26

This paragraph first said twelve months, and the surrounding briefings said the virtual-cell vocabulary had not reached these journals before November. Both were wrong, and the error was one of sampling: each month's reader searched only its own month. The 2025-01..07 backfill found the phrase in Nature Biotechnology in April 2025, two months before the challenge that supposedly introduced it, and again in Cell in August 2025. The vocabulary did not arrive in December; it recurred, and the reason December looks like an arrival is that December is when it arrived in Nature Methods specifically.

Agentic AI reached the immunology journals. 41348860 (Sci Immunol) is an assessment commentary rather than a result, and it is the earliest AI-as-discovery-agent framing in the nine immunology and systems-biology journals that this corpus has read — which is not the same claim as the one this paragraph used to make.

Corrected 2026-08-27

This paragraph carried a five-link dated chain — ESCARGOT on 4 February 2025 as the first LLM-agent paper, the word agentic four days later, oncology in June, wet-lab-adjacent in August, immunology in December, and a "ten months, bench to bench" diffusion figure resting on all of it.

Four of the five were already refuted by this corpus's own 2024 backfill and the paragraph was never updated. An LLM agent with tools is in a tracked journal in 2024-01, thirteen months before ESCARGOT, and an agent reaches an oncology journal in 2024-12, six months before the date given here. The chain was wrong at its origin and at its oncology waypoint, so the interval computed from it was never a measurement of anything.

The corrected version is not a better set of dates. It is that these are not datable from a bounded window at all — each "first" was the earliest month then being read. The corpus has now made this error six times.

Checkpoints regressed to targets that are not checkpoints. December's nominations are CLEVER-1, CD300e, TIPE2 and NEAT1 — macrophage receptors and a long non-coding RNA. Of August's roster, CD24 and TIM-3 survive; CD36, PYK2 and ICOS are gone; TIGIT persists only as its ligand CD155. The only clinical-stage non-PD-1 content is a B7-H3 review whose own title hedges (41401407) and an Fc-enhanced CTLA-4 — both from waves preceding the current preclinical set. The category is dissolving, not merely churning.

4. Refutations

(a) Landing on a tracked thread

Two Nature Immunology papers relocate CD47's relevant function from the macrophage don't-eat-me axis to T-cell-intrinsic exhaustion via thrombospondin-1 (41258436, 41249483). Five months after two negative CD47 phase 3s, this reads as a mechanistic post-mortem: the phagocytosis-checkpoint rationale the programme was built on may have been the wrong mechanism, not merely an under-powered one.

Corrected 2026-08-27

41423265 (JITC) was filed here as a refutation, on the argument that the safety carve-out in its title — "without inducing systemic toxicity" — was itself the finding: that systemic Treg depletion is understood to cause harm, and that the paper's selling point is avoiding it. That is carding a phrase in a title rather than a result in a paper, and the abstract attributes the systemic-toxicity problem to mogamulizumab, not to this agent.

The study is a straightforwardly positive one. A tumour-targeted anti-HER2xCCR4 bispecific depletes tumour-infiltrating and tumour-chemotactic Tregs, raises the CD8/Treg ratio, controls advanced HER2-positive tumours in humanised mice without reducing Tregs systemically, and synergises with anti-PD-1. Depleting a suppressive population improved the outcome once the depletion was confined to the tumour, which is the thread's claim, and the card supports it.

The question the paragraph went on to raise survives, and the corrected reading sharpens rather than weakens it. Read alongside 41401810 (Immunity, indexed 2026-01), which reports opposing functions for distinct Treg subsets in colorectal cancer, what a depletion programme has to answer is not whether the population is suppressive but whether the marker it selects separates the subset that should go from the one that should stay. Confining the depletion to the tumour is one answer to that, and this paper is evidence that the answer works.

41371211 is an Immunity preview of Herbrich et al. in Cancer Cell, a paper this corpus does not hold: TET2-mutant clonal haematopoiesis, an adverse host factor by assumption, reprograms tumour macrophages and boosts checkpoint therapy. The inversion belongs on the thread; the record carrying it is a commentary on another journal's work, which the earlier version of this line did not say.

41417925: Siglec-F+ neutrophils promote resolution of lung injury. 41272165: interferon-driven NK activation raises cytolysis and lowers the neutralising antibody response — the surrogate improves while the protective endpoint worsens, in one title.

(b) Method credibility

41331675 (Genome Biol): CRISPR-Cas9-induced double-strand breaks disrupt maintenance of epigenetic information. Together with November's 41196998, the two months say that the most widely used perturbation in molecular biology corrupts the readout of the field that uses it most.

The self-audit genre reached one paper in ten of the methods journals, its highest share in the corpus, and its object escalated again — from the model, to the pipeline, to the instrument (41476115), the wet-lab assay (41331675), and the generative model's own output (40523934, hallucination detection in virtual staining). Auditing what a model emitted, rather than how it was trained, is the newest object in the sequence — but it is not a new genre, and this paragraph originally called it one. Nature audits a generative model's own output in 2024-06, eighteen months earlier. That correction was published in the 2024-06 briefing on 2026-08-26 and should have been applied here the same day; it was not.

41102561 (Nat Med) names a frontier model and calls it brittle in medicine. In July the counterweight to clinical-AI enthusiasm was opinion; it is now system-specific.

41061703 (Cell): ascertainment, not biology alone, produces the variable expressivity reported for complex disorders. 41344327: retraction of the transcriptome-scale Cas13 screen catalogue of essential lncRNAs. 41327972: an Editor's Note attaches to a foundational longitudinal ICB immune-signature and resistance paper.

A hole in the month, recorded here so it is not read as biology. Nucleic Acids Research contributed zero rows to December; the bioinfo domain is 42% below its median and the volume warning in section 1 fires on it. Everything above about the methods journals is drawn from a month missing one of its largest contributors, so the self-audit share of one in ten is computed on a reduced denominator and is if anything an underestimate of the count.

Corrected 2026-08-26

This paragraph called the hole a harvest defect, which implied it was ours and could be repaired by re-pulling. It cannot. Queried directly, PubMed returns zero Nucleic Acids Research research articles with a December 2025 publication date — the gap is in the source, not in the pull, and re-running the harvest reproduces it exactly. The denominator caveat above stands unchanged; the diagnosis does not. The same check found three more months that had been silently empty and would have read as quiet fields: Bioinformatics in 2024-04, 2025-01, 2025-04 and 2026-03, and Genome Research in 2024-01. All return zero from PubMed too.

Separately, and worse: Nature Machine Intelligence contributes 64 papers to the entire 32-month corpus and is absent from fifteen months of it, because PubMed indexes only its deposited fraction. It is a flagship venue for precisely the foundation-model and agentic-AI material these briefings track most closely. Every statement this corpus has made about where machine-learning methods are published is made with that venue effectively missing.

5. Open items

Nothing carried forward.

6. Meetings

No abstract book published this month. The abstract layer runs 2023-02 to 2026-08 and is seasonal by nature — meetings are annual, so most months are empty here and that is the expected shape, not a gap in the harvest.

7. Regulatory

Event Count
trial-first-posted 673 (of which 602 genuinely oncology)
fda-approval 6
ema-authorisation 1

6 FDA actions — dated from each notice's own prose, not the table's posting date, which disagrees on 7 of 190 rows and once by enough to move an approval into the wrong month:

  • 2025-12-03 — FDA grants traditional approval to pirtobrutinib for chronic lymphocytic leukemia and small lymphocytic lymphoma
  • 2025-12-04 — FDA approves lisocabtagene maraleucel for relapsed or refractory marginal zone lymphoma
  • 2025-12-12 — FDA approves niraparib and abiraterone acetate plus prednisone for BRCA2-mutated metastatic castration-sensitive prostate cancer
  • 2025-12-15 — FDA approves fam-trastuzumab deruxtecan-nxki with pertuzumab for unresectable or metastatic HER2-positive breast cancer
  • 2025-12-17 — FDA approves amivantamab and hyaluronidase-lpuj for subcutaneous injection
  • 2025-12-17 — FDA grants regular approval to rucaparib for metastatic castration-resistant prostate cancer

8. Trade press

41 on-topic of 303 items across 3 feeds (n-stat 290, n-aacr-blog 10, n-acir 3). Headlines are selected by subject keyword, not recency — most of what these feeds carry is health policy and industry finance the corpus does not read.

  • 2025-12-29 — Editors’ Picks, December 2025: Potent Molecular Glues, Personalized CAR T-cell Therapies, and More — n-aacr-blog
  • 2025-12-22 — SABCS 2025 Session Makes Sense of Rapidly Evolving Therapeutic Landscape for Breast Cancer — n-aacr-blog
  • 2025-12-22 — FDA approves pill version of Novo Nordisk’s Wegovy, possibly widening access — n-stat
  • 2025-12-22 — An advocate who offered a window on the experience of living with brain cancer dies at 43 — n-stat
  • 2025-12-22 — Study revives debate about unpopular diet for cancer patients — n-stat
  • 2025-12-19 — STAT+: Cytokinetics heart drug wins FDA approval, the biotech’s first — n-stat
  • 2025-12-18 — Cancer in 2025: Funding, New Treatments, and Breakthrough Ideas — n-aacr-blog
  • 2025-12-17 — Can Some Breast Cancer Patients Skip MRI or Sentinel Lymph Node Biopsy? — n-aacr-blog
  • 2025-12-17 — STAT+: FDA approves brain tumor warning on Depo-Provera label as court battle grows over side effects — n-stat
  • 2025-12-17 — STAT+: Pharmalittle: We’re reading about cancer drug trials, Europe’s biotech push, and much more — n-stat
  • 2025-12-17 — Bringing in neutrophils to improve CAR-T efficacy in solid tumors — n-acir
  • 2025-12-15 — Apps and Acupuncture: Improving Breast Cancer Survivors’ Quality of Life — n-aacr-blog
  • … and 29 more on-topic items.