A Hidden Fourth Type of Lung Tumour May Explain Why Some “Low-Grade” Cancers Turn Aggressive

A large international study has uncovered a previously underappreciated form of lung neuroendocrine tumour that appears to blur the boundary between relatively slow-growing carcinoids and deadly small-cell lung cancers. The tumours, known as “supra-carcinoids,” can look deceptively mild under a microscope while carrying molecular features associated with rapid growth, invasion and treatment resistance. In an analysis of 319 lung neuroendocrine tumours, researchers identified four biologically distinct groups rather than the two broad grades used in current clinical classification. The findings suggest that some lung tumours now labelled mainly by how many cells are dividing may instead need to be understood through their genetic programs, cellular identity and surrounding immune environment. The study, led by scientists at the International Agency for Research on Cancer and collaborators across Europe, Australia, Chile and the United States, combines whole-genome sequencing, RNA analysis, DNA methylation profiling, spatial transcriptomics, proteomics and artificial intelligence. Its central message is striking: two tumours that look similar on a pathology slide may be fundamentally different diseases, while a tumour that looks relatively indolent may be poised to evolve into a much more aggressive cancer.

Lung neuroendocrine tumours, often called pulmonary carcinoids, arise from cells with hormone-producing and neuron-like characteristics. They are currently divided by the World Health Organization into typical carcinoids, classified as grade 1, and atypical carcinoids, classified as grade 2. The distinction depends largely on mitotic count—the number of cells caught dividing—and the presence or absence of necrosis, a form of tumour-cell death caused by inadequate blood supply or extreme biological stress. Typical carcinoids generally have favourable outcomes, with five-year overall survival around 90%, whereas atypical carcinoids have a five-year survival near 60%. But these criteria do not always predict how a tumour will behave or which treatment it might respond to. Different molecular subtypes can appear in both grade 1 and grade 2 categories, and pathologists may disagree when assessing borderline features. The new study argues that histological grade remains useful for estimating tumour progression, but it does not fully capture the tumour’s origin, internal circuitry or therapeutic vulnerabilities.

To expose that hidden diversity, the researchers integrated several layers of molecular information. Gene-expression data reveal which cellular programs are active; DNA methylation profiles record chemical marks that regulate gene activity; whole-genome sequencing identifies mutations, structural rearrangements and copy-number changes; and copy-number analysis shows where chromosomes or chromosome segments have been gained or lost. The team used Multi-Omics Factor Analysis, a statistical framework that compresses these data types into shared axes of biological variation, followed by archetype analysis to identify the major molecular “corners” of the tumour landscape. Three groups corresponded to previously described categories called Ca A1, Ca A2 and Ca B. The fourth, called the supra-carcinoid-enriched or sc-enriched group, contained 14 of 16 supra-carcinoid tumours in the molecular analysis. Unlike the other groups, it had no clear preference for patients’ age, sex or tumour location. It was, however, more likely to invade the pleura—the membrane surrounding the lungs—and was associated with poorer survival. Ten-year overall survival was estimated at 86% for Ca A1, 83% for Ca A2, 70% for Ca B and only 60% for the sc-enriched group.

The genomic profile of supra-carcinoid-enriched tumours offered another warning sign. These cancers carried higher burdens of small mutations, large structural variants and deleted sections of the genome than the other tumour groups. Their mutation patterns were also more varied, suggesting that several different DNA-damaging processes may have contributed to their development. Their active genes were enriched for multiple cancer hallmarks, including persistent growth signals, the ability to keep dividing indefinitely and mechanisms that help tumours evade immune attack. The tumours also showed a higher proliferation index even when the analysis was restricted to grade 2 cancers. Yet the apparent contradiction remained: only one supra-carcinoid in the examined pathology subset clearly crossed the proposed mitotic and Ki-67 thresholds for a new grade 3 category. Ki-67 is a protein expressed by proliferating cells and is commonly used to estimate how quickly a tumour is growing. The result illustrates why molecular testing may be needed: a supra-carcinoid can be biologically aggressive without displaying the textbook microscopic appearance of a high-grade tumour.

The researchers found that the defining biology of these tumours was not confined to the cancer cells themselves. RNA deconvolution and spatial transcriptomics showed that supra-carcinoids were unusually rich in immune and stromal cells, particularly classically activated M1-like macrophages and cancer-associated fibroblasts. Macrophages are immune cells that can either attack tumours or, under certain conditions, support their growth, invasion and repair-like behaviour. In supra-carcinoids, regions with an immature, lower-airway-progenitor-like state were located alongside macrophages and fibroblasts. Cell-interaction analyses highlighted macrophage migration inhibitory factor and collagen signalling, as well as a potential interaction between ICAM1 on tumour progenitor-like cells and ITGAX and ITGB2 on macrophages. ICAM1 is a cell-adhesion molecule involved in inflammation, tissue repair and metastasis. The spatial arrangement does not prove that macrophages create supra-carcinoids, but it suggests that communication between tumour cells and their microenvironment may help sustain or promote this unusual state. The finding also challenges the idea that cancer progression is driven only by mutations inside tumour cells; in some cases, the surrounding tissue may help determine what those mutations become.

Cell-state analysis provided further clues to the tumours’ origins. Most lung neuroendocrine tumours contained mixtures of differentiated neuroendocrine cells and club cells, airway epithelial cells involved in repair and chemical defence. Ca A1 tumours were richest in mature neuroendocrine cells and expressed proneural transcription factors such as ASCL1 and MYT1L. Ca B tumours had more club-cell features and showed signals consistent with exposure-related biology, including a stronger smoking-associated mutational signature. Supra-carcinoids were different: they contained the highest proportion of lower-airway progenitor cells, which are capable of giving rise to neuroendocrine and other airway cell types during lung development. Spatial analysis showed that progenitor-like and mature neuroendocrine-like regions occupied separate areas within the same tumours. This pattern suggests that at least some supra-carcinoids may contain an undifferentiated or stem-like population that has lost much of its classic neuroendocrine identity. Such plasticity could help explain why these tumours resemble both carcinoids and high-grade neuroendocrine carcinomas.

One patient’s disease offered an especially vivid example of possible transformation. The patient, a 26-year-old never-smoker, was initially diagnosed with a stage IA atypical carcinoid. The tumour carried a BRAF V600E mutation and a highly rearranged region of chromosome 9 resembling chromothripsis, a catastrophic event in which a chromosome is shattered and reassembled in a chaotic pattern. The patient initially responded to the BRAF and MEK inhibitors dabrafenib and trametinib, but later developed metastases in the bone and liver. A liver biopsy was reclassified as large-cell neuroendocrine carcinoma, and sequencing revealed a deletion affecting exons 2–8 of BRAF, a known mechanism of resistance to BRAF-inhibitor therapy. A patient-derived tumour organoid grown from the original cancer became progressively faster-growing in culture, accumulated more mutations and increased expression of MKI67, a gene associated with proliferation. Over successive passages, its cell profile shifted from carcinoid-like toward neuroendocrine-carcinoma-like, even without immune cells or fibroblasts present. When the team compared supra-carcinoids with atypical small-cell lung cancers carrying chromothripsis, the atypical small-cell tumours clustered more closely with supra-carcinoids than with conventional small-cell lung cancer, raising the possibility that they represent successive stages of the same evolutionary route.

The study also points toward ways of bringing the four-group classification into clinical practice. Deep-learning models trained on whole-slide pathology images learned to distinguish molecular groups from visual features such as spindle-shaped cells, nested or organoid architecture, fibrosis, clear-cell changes and cartilage-like tissue. Surprisingly, the models predicted molecular groups more accurately than they predicted conventional histological grades. A simpler laboratory approach also showed promise: staining tumour samples for three proteins—ASCL1, HNF1A and OTP—correctly assigned 77 of 90 additional samples when combined with characteristic morphology. For research laboratories with sequencing capacity, the investigators developed PACMOS, a publicly available software package that classified tumours with 98.8–100% accuracy in cross-validation, depending on whether RNA sequencing, DNA methylation data or both were used. The molecular groups also differed in expression of potential treatment markers. Ca A1 had higher DLL3 and EGFR, Ca A2 had higher FGFR4, Ca A2 and Ca B showed stronger SSTR2, and supra-carcinoids had relatively high EGFR, FGFR3 and MGMT. They also displayed a five- to sevenfold higher T-cell-inflammation signature than other groups, a signal that may justify testing immunotherapy strategies—but not yet proof that such treatments will work.

The researchers stress that the results do not immediately replace the WHO grading system or establish a new treatment standard. The cohort was large for this rare cancer, but the supra-carcinoid group itself contained only 16 tumours in the main analysis, and many conclusions about transformation come from genomic patterns, spatial correlations and a particularly informative patient case rather than from long-term prospective observation. Future studies will need to validate the classification in routine biopsy samples, determine how reproducibly pathologists can recognise the morphological signatures and test whether the molecular groups predict response in clinical trials. Even so, the work redraws the map of lung neuroendocrine cancer. Ca A1, Ca A2 and Ca B appear largely to be established early by group-specific genomic events, whereas supra-carcinoids may sometimes arise from one of the other groups and acquire additional alterations later. That possibility could explain why a tumour with apparently low-grade architecture occasionally behaves like small-cell lung cancer. By combining molecular biology with pathology and machine learning, the study turns a mysterious clinical problem into a testable evolutionary model—and offers the prospect that future patients will be treated according to the biology of their tumour, not simply the way it looks under a microscope.