A positional barcode for the lung, and what it exposes in organoid models

What the work claims

HOX genes are a set of 39 homeobox transcription factors arranged in four chromosomal clusters (A to D). During development they switch on in spatial order, with genes at the 3-prime end of a cluster active earlier and in more anterior body regions, and 5-prime genes activated later and more posteriorly. The long-standing idea is that adult cells retain the HOX expression pattern they acquired in the embryo, a cell-specific and tissue-specific "HOX code" that records positional identity for life.¹ This study sets out to define that code for the adult human lung and to ask which cell types carry it.

Working from normal human lung specimens, the authors combine bulk expression profiling and quantitative RT-PCR with reanalysis of published single-cell RNA sequencing datasets and their own immunohistochemistry and immunofluorescence. They report a specific anterior signature: HOXA3 to HOXA7, HOXB5, HOXB7 and HOXB8, HOXC6 to HOXC9, and HOXD1 to HOXD8 are expressed, while posterior paralogs (HOX10 to HOX13) are absent, consistent with an earlier microarray survey of human lung. The headline is not just which genes but where: expression is predominantly stromal and vascular, and it is sparse in the alveolar compartment. They then track HOX expression across iPSC-derived lung organoid differentiation and argue the organoid "confirms" the adult code.¹

This is a mapping and localization study, not a functional one. It generates a candidate code and a cell-type atlas; it does not test what any of these HOX factors do in human lung by loss or gain of function. That framing should set the weight a reader gives each claim.

How it works

The cell-type resolution is the interesting part. HOXA3 and HOXA5 localize primarily to sub-bronchial and vascular smooth muscle cells; HOXA4, HOXB5 and HOXB7 to vascular mural cells; HOXA6 and HOXA7 to sub-bronchial and adventitial connective tissue; and HOXA7, HOXB5, HOXB7 and HOXD1 to the endothelium of larger vessels. Bronchial epithelial cells stain for almost all of the HOX proteins examined, though in some cases only in scattered single cells. The alveolar compartment, by contrast, shows only a thin set (HOXA6, presumably in connective tissue, plus HOXB6, HOXC6 and HOXC8). The picture that emerges is of a positional code written mostly into the mesenchymal and vascular scaffold of the conducting airways, with the gas-exchange region comparatively HOX-quiet.¹

The organoid arm uses a matrix-free iPSC lung organoid system.² By quantitative RT-PCR across days 0, 4, 7, 14, 28, 38 and 45, anterior cluster A and B HOX genes rise over differentiation in a pattern the authors read as recapitulating the adult code. But their own single-cell analysis of matured organoids tells a more complicated story: the organoids also express posterior HOX genes, including HOXA10, HOXA11, HOXC10 and HOXA13 in the fibroblast and alveolar-like clusters, and the abstract flags rising HOXD10 and HOXD13 that "contradict" the normal-lung values. Those posterior paralogs are precisely the ones the adult lung does not express.

The strongest version of the argument

Taken at face value, this is the first cell-type-resolved HOX atlas of the adult human lung, and it triangulates across three independent readouts (bulk expression, single-cell transcriptomics and protein-level staining) rather than resting on one. The anterior cluster A and B dominance it reports agrees with the older literature, which is reassuring for a field where antibody artifacts are common. And the temporal organoid series is a genuine attempt to connect a static adult map to a dynamic in vitro system, showing that at least the anterior code is not random noise but develops in a reproducible direction. If the localization holds, it hands the field a concrete, measurable identity signature rather than a vague appeal to "lung-likeness."

Where a skeptic should push

The single most load-bearing assumption is antibody specificity. HOX paralogs share a highly conserved homeodomain, and commercial HOX antibodies are notorious for cross-reacting across paralogs. The most granular claims here (this HOX in that cell type, sometimes in single cells) rest on immunohistochemistry, the readout most vulnerable to that problem. Without an orthogonal, paralog-specific method such as RNA in situ hybridization or a tagged allele, the cellular assignments should be read as strong hypotheses, not settled facts.

Second, the transcriptomic backbone is reanalysis. The adult single-cell data come from published datasets generated on 10x Genomics, while the organoid single-cell data were produced on a different platform (BD Rhapsody), and the authors explicitly decline to integrate the two. That makes the organoid-to-adult comparison qualitative by construction. Third, sample sizes are modest (n of 4 to 8 per organoid timepoint) and the whole design is descriptive: there is no perturbation showing that any HOX factor is required for a lung cell's identity in human tissue. The developmental necessity arguments are imported from mouse knockouts. Finally, "the organoid confirms the code" is generous. The organoid reproduces the anterior half and adds a posterior half the adult lung lacks, while under-representing the stromal and endothelial cells where the adult code actually lives. That is partial, directional concordance, not identity. One caveat cuts the other way and is worth stating plainly: the adult absence of posterior HOX rests largely on bulk expression, qRT-PCR and antibody staining, while the organoid presence is partly a single-cell transcriptomic call, and single-cell sequencing detects low-level and rare-cell transcripts that bulk and protein methods miss, so some of what looks ectopic may be newly detectable rather than newly expressed. The organoid qRT-PCR time course does, however, also register rising posterior HOXD10 and HOXD13, so the effect is not purely a platform artifact.

What a HOX miscode means for lung models

For anyone building lung organoids as models of human airway biology or as substrates for respiratory drug discovery, the location of the code is the uncomfortable finding. If positional identity is written predominantly into smooth muscle, mural, connective-tissue and endothelial cells, then an epithelium-centric lung organoid (the common design for a screening assay) is stroma-poor in exactly the compartment that carries regional identity. The organoid may express epithelial lineage markers convincingly and still be positionally under-specified: a generic "lung epithelium" without a defined proximal-to-distal address. Drug responses that depend on airway region (proximal conducting airway pharmacology versus distal alveolar, or fibrosis biology, where HOX deregulation is itself implicated) could be mis-assigned if the model has no stable regional coordinate. This is a hypothesis rather than a demonstrated property: localizing HOX protein to stroma does not prove that stroma is required to set epithelial positional identity, since epithelial-to-mesenchymal signaling could still transmit it, and the study runs no perturbation to test the point.

The opportunity is that a HOX code is measurable. Unlike "maturity," which is usually asserted, an anterior cluster A and B signature is something a lab can assay by RT-PCR and use as an objective identity and quality-control metric, or even a modulation target to push an organoid toward a defined region. The genuine threat sits alongside it, though it needs stating carefully. These organoids switch on posterior HOX genes (HOX10 to HOX13) that healthy adult lung does not express. The most parsimonious reading is developmental: iPSC organoids are fetal-like, and posterior paralogs that were never silenced or were mis-set during patterning are a plausible immaturity artifact rather than a pathological program. It is worth flagging, but not overreading, that some of those same posterior cluster C and D genes are the ones an older and correlative literature reports re-expressed in chronic lung disease such as emphysema and pulmonary hypertension. Genes seen in a disease are not a disease signature, and this source offers no mechanism linking the two. The honest statement is narrower: a lung organoid can carry a positionally off-pattern HOX profile whose meaning, whether developmental immaturity, a technical detection difference, or something disease-relevant, is unresolved, and telling those apart needs the paralog-specific validation the study does not provide. That unresolved ambiguity is itself the reason to characterize a baseline before trusting it in a disease-modeling comparison.

The bottom line

Treat this as a hypothesis-generating atlas. What is reasonably established: anterior cluster A and B HOX genes are expressed in adult human lung, concentrated in stromal and vascular cells and sparse in alveoli, and iPSC lung organoids develop a broadly concordant anterior signature over roughly six weeks. What remains hypothesis: that these patterns constitute a functional cell-type-specific code, that the antibody-based cell assignments are paralog-specific, and that the organoid's ectopic posterior HOX is a fidelity defect rather than a resolvable immaturity. The claim would be confirmed by paralog-specific in situ validation, by loss or gain of function that actually shifts a human lung cell's regional identity, and by profiling organoid and matched adult stroma on one platform. It would be weakened if antibody cross-reactivity explains the localization, or if the posterior HOX simply reflects a developmental stage that matures away. Either way, the practical message for model builders is concrete: if you cannot state your lung organoid's HOX address, you cannot be sure which lung you are modeling.