September 2023 — Monthly Briefing¶
1. What was read¶
684 research articles from 32 of the 33 journals on the roster.
A further 225 records (25% of the 909 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 |
|---|---|---|---|---|
| 449 | 132 | 85 | 225 | 18 |
Volume swing against the median month: cancer -29%, sysbio -25%
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 Nat Mach Intell. 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: Nature 108, Science 100, JAMA 92, Nucleic Acids Res 80, Bioinformatics 56, Lancet 46.
2. Net change on the thread curve¶
| Thread | This month | Cumulative | Status |
|---|---|---|---|
cell-therapy |
+14 / −1 | +71 / −19 | forming |
ctdna-mrd |
+1 | +18 / −10 | splitting |
foundation-models |
+4 / −1 | +24 / −12 | forming splitting → forming |
immunotherapy-resistance |
+7 / −3 | +80 / −23 | forming |
interception |
+5 / −1 | +51 / −5 | forming |
method-credibility |
−5 | +1 / −95 | crisis |
neoantigen-vaccines |
+1 / −2 | +28 / −12 | forming |
neuro-immune-microbiome-axis |
+1 | +47 / −4 | forming |
next-gen-checkpoints |
+1 / −1 | +14 / −12 | splitting |
single-cell-3d-genome |
+3 / −1 | +18 / −6 | forming |
suppressive-population-inversion |
+5 / −4 | +22 / −22 | splitting contested → splitting |
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¶
cell-therapy closes +14 / −1, and the cards that carry the month leave the
CD19 template in three different directions.
37314354
(NEJM) gives base-edited allogeneic CAR7 T cells for relapsed T-cell ALL — off
the B-cell lineage and off autologous manufacturing in a single trial.
37648862 (Nature) edits the epitope
out of a surface antigen on healthy haematopoietic cells so that a CAR can be
aimed at it on leukaemia, and an independent group publishes the same shielding
logic on CD45 in the same month
37651540 (Sci Transl Med), with
pooled knock-in screening 37714135
(Cell) making such designs searchable rather than hand-built.
36658341 (Nat Biotechnol) points
chimeric autoantibody receptor T cells at the autoantigen-specific B cells of
myasthenia gravis — off oncology entirely. Manufacturing, antigen space and
indication all moved in four weeks.
The nine remaining supports are the engineering that has to exist for those moves to be repeatable, and they are spread across six journals: massively parallel knock-in engineering of human T cells 36702900 (Nat Biotechnol); a trifunctional NKp46–CD16a engager pointing NK cells at CD123 36635380 (Nat Biotechnol); microfluidic immunomagnetic isolation of tumour-reactive lymphocytes from peripheral blood rather than from tumour 37037966 (Nat Biomed Eng); universal redirection of CAR T cells against solid tumours by membrane-inserted ligands 37291434 (Nat Biomed Eng); a CTLA-4 tail fusion that raises CAR-T antitumour activity 37500885 (Nat Immunol); allogeneic double-negative CD4-CAR T cells against T-cell malignancies 37678917 and AKT inhibition generating clinical-grade AUTO1 with better function 37709295 (both JITC); reductive carboxylation instructing T-cell differentiation 37730993 (Nature); and YTB323, an autologous CAR-T manufactured to preserve T-cell stemness 37249512 (Cancer Discov). Several of them are about where the cells come from and how they are made rather than about what they are aimed at, which is the constraint the headline results run into next.
Corrected 2026-09-04
This paragraph opened "cell-therapy gains three cards" and then cited five.
The month is +14 / −1. September 2023 had been carded from titles; the
2026-09-04 abstract re-scan added the nine further supports listed above. The
three-directions reading is unchanged and still correct — it was a claim about
those papers, not about the size of the month, and the month was nearly three
times the size stated.
Three groups, one ligand family, on next-gen-checkpoints.
37460871 (Nat Cancer) nominates
Siglec-9 as a macrophage checkpoint restricting T-cell priming in glioblastoma;
37709296 (JITC) nominates Siglec-9
macrophages as an immunosuppressive subset with therapeutic vulnerability in
ovarian cancer in the same month; and
37467314 (Sci Transl Med) targeted
the Siglec–sialic acid axis in glioblastoma two months ago. The thread's recorded
pattern is that surface receptor nominations have a half-life of a few months and
ligand-side ones last longer. This is a ligand-side nomination arriving from
three directions at once.
Two proteome-wide variant-effect models within weeks, each flagged by its own journal. 37563329 (Nat Genet, News & Views 37563330) scores disease variant effects genome-wide from a protein language model; 37733863 (Science, perspective 37725046) does proteome-wide missense prediction. Two labs reaching clinical variant interpretation from self-supervised sequence models at the same time, and both editorial boards thought it worth annotating. Separately 37592105 (Nat Med) builds a vision–language model for pathology out of scraped public images and presents it in a clinical journal as a foundation model.
Three screening-effectiveness readouts in one month.
37478397 (JCO) compares two
blood-based approaches to nasopharyngeal carcinoma screening head to head on
performance and operational feasibility;
37667043 (Nat Cancer) reports a
prospective multicentre hepatocellular carcinoma screening cohort; and
37379494 (JCO) reports a mortality
benefit for a blood biomarker panel in the PLCO cohort. interception also gains
a model that predicts myeloid neoplasia risk before the disease exists
37620601 (Nat Genet).
What is absent, having looked. Across the 2,855 records this quarter's
harvests returned there is not one paper describing an LLM agent that plans, calls
tools or acts. The
nearest items are answerers: a chat model benchmarked on biomedical text
generation and mining 37682111
(Bioinformatics), and a clinical-journal pair reporting that large language
models answer medical questions accurately without matching clinicians' knowledge
37548971 and asking how such models
should be built and adopted
37548965. agentic-ai holds one
earlier card in this corpus, a commentary on a chatbot rather than an agent
36988602, and the months between
that and this one are empty. This matters more than a positive would: eleven of
the thirteen threads open in January 2023, the month the window itself opens,
which is a fact about where reading started rather than about the world.
agentic-ai does not, and its silence here was measured rather than assumed. The
2026-09-04 re-read of this quarter against abstracts rather than titles did not
disturb either finding: agentic-ai still takes no card in this quarter, and
clinical-practice still holds none anywhere in the corpus. The journal index is
the wrong instrument for the second one, and it was left alone rather than padded.
4. Refutations¶
(a) Landing on a tracked thread¶
Four of the month's refutations land on cell populations that had been nominated for removal.
37713507 (Sci Immunol) — lung tumour-infiltrating regulatory T cells split into transcriptionally and functionally divergent subsets that track differently with checkpoint blockade response, with 37713508 mapping the network that makes some of them suppressive and 37563309 (Nat Immunol) finding the same divergence in resident macrophages. Depleting Tregs as a population removes both arms of something that has two.
37525015 (Nat Cancer, with 37553501) — IL-17 signalling, routinely nominated as tumour-promoting, is associated with clinical benefit from combined CTLA-4 and PD-1 blockade in patients. Uncarded but pointing the same way, 37572655 (Immunity) has CTLA-4 blockade recruiting microglia — a population nominated for depletion in glioblastoma — into the effector arm.
37537301 (Nat Cancer) — CD8 T cells keep killing MHC-I-negative tumour cells through NKG2D and its ligands. Loss of antigen presentation is scored as an escape route throughout this literature; here it is not sufficient on its own.
37653344 (Nat Med) — myeloma escapes both BCMA- and GPRC5D-directed immunotherapy through defined antigen-loss mechanisms. Each new target inherits the failure mode, which is the argument against reading the platform's reach as a function of how many antigens can be found. Section 3's three supporting cards and this one are the same thread in the same month, and both readings are correct.
(b) Method credibility¶
Five cards, and one of them lands on a second thread.
37723583 (Genome Biol) — methods
for calling spatially variable genes disagree with each other, reported as the
reason to benchmark them. It refutes single-cell-3d-genome as well, which is
why that thread reads +1 / −1: the claim is that spatial multi-omics becomes a
routine modality that downstream pipelines ingest, and the pipelines do not yet
agree on what they are reading.
37734322 (Cell Systems) — the two-dimensional embeddings that single-cell papers are read from misrepresent the data they display. The figure and the finding are in different places.
37580132 (Genome Res) — gene regulatory network inference from single-cell transcriptomes has struggled to beat random guessing, and the reason given is that mature mRNA cannot report upstream regulator activity. Intronic reads improve it, so published GRNs inherit the wrong input rather than a tuning problem. 37674217 (Genome Biol) — a seven-algorithm assessment against reference materials finds batch-effect correction over-corrects as readily as it under-corrects, and that most algorithms fail outright when batch is confounded with biology. Only ratio-based scaling was broadly reliable.
37669666 (Cell) is a retraction notice; 37616332 (Science) and 37548412 (NAR) are editorial expressions of concern in the same four weeks, and 37717594 (Lancet) reports an institution halting projects over a fraud investigation. This thread is a ledger of things found unreliable and so it almost only refutes: +1 / −95 cumulative, where the single support is June's likelihood-ratio re-analysis of 169 statistically nonsignificant trial outcomes 37338877 (JAMA), which found 91% of them favouring the null over the prespecified effect — a method vindicated rather than faulted. One in ninety-six describes the instrument and not a decline. What the quarter does show is where the audit has got to: the clustering step, the embedding the clusters are viewed in, the spatial caller, the network inferred from the counts, the batch correction applied before any of it, and the correction notice itself.
Corrected 2026-09-04
This read "As every month, this thread can only refute, so +0 / −22 cumulative describes the instrument and not a decline." Two things in it are wrong. The cumulative figure is +1 / −95 — the 2026-09-04 abstract re-scan more than quadrupled the refutation count on this thread. And can only is too strong: the thread carries one support by September 2023, and by 2026 it carries several, all of them audits that a method passed. The structural point stands, but the ledger is capable of recording a pass and occasionally does.
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 | 484 (of which 429 genuinely oncology) |
| ema-authorisation | 12 |
| fda-approval | 2 |
2 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:
- 2023-09-14 — FDA approves new and updated indications for temozolomide under Project Renewal
- 2023-09-26 — FDA approves bosutinib for pediatric patients with chronic myelogenous leukemia
6 EMA oncology authorisations:
- 2023-09-15 — Orserdu
- 2023-09-15 — Inaqovi
- 2023-09-15 — Tevimbra
- 2023-09-15 — Enrylaze
- 2023-09-22 — Tepkinly
- 2023-09-29 — Degarelix Accord
8. Trade press¶
No items. The press archive covers 2024-01 to 2026-08; feeds that cannot be paginated backwards contribute only from the first harvest forward.