Reassessing Forebrain Organization: Recent Evidence Challenges the Prosomere Model

Abstract

For over three decades, the prosomere model has influenced our understanding of vertebrate brain development by proposing that the forebrain consists of several transverse segments, termed prosomeres. However, advances in molecular genetics and imaging have challenged key assumptions of this model, including the existence of clear segmental boundaries and the positioning of forebrain axes. Here we critically review evidence from gene expression, fate mapping, and morphogenesis studies, finding limited support for prosomeric segmentation and questioning the grouping of the hypothalamus with the telencephalon. We propose the tripartite hypothalamus model as an alternative framework that better aligns with recent data, restoring the hypothalamus’s relationship to the diencephalon and recognising the anterior forebrain as a uniquely complex, non-segmented unit with a posterior boundary within the diencephalon. These findings suggest a need to revise our understanding of forebrain structure, patterning and evolution.

Introduction

Modern technologies are transforming our understanding of brain development. Building on earlier high throughput approaches, single cell and spatial omics technologies have generated high-resolution cell and molecular atlases of the developing and adult brain [1– 11]. These not only provide insight into how brain cells are specified and differentiate, but also re-open two of the oldest and most fundamental questions in neuroscience: how is the brain laid out, and how do its parts relate to one another?

Over a century ago, pioneering anatomists tackled these questions by studying embryonic brain morphology. The embryonic brain vesicles inspired the major brain subdivisions still used today, while ventricular sulci were interpreted as revealing the brain’s anteroposterior (A-P, length) and dorsoventral (D-V, transverse) axes [12]. For most of the 20th century, these early observations shaped the ‘columnar’ model of the forebrain (Figure 1A). In this model the thalamus (then dorsal thalamus), prethalamus (formerly ventral thalamus), subthalamus, hypothalamus, and eye field were all considered parts of the diencephalon, with the telencephalon located rostral (anterior) to these structures [13]. Since the 1990s, however, an alternative framework has gained wide acceptance.

Figure 1:

Forebrain models and the updated prosomere model. A) Example of a columnar model. Major forebrain regions shown on left, mid-/forebrain axes indicated on right. The subthalamus (not shown) is located between the prethalamus and hypothalamus. B) The updated prosomere model. Prosomeres m1 to hp2, and their boundaries, are indicated. The slim m2 prosomere is not shown. C) Prosomere model axes. A = anterior, P = posterior, D = dorsal, V = ventral. Asterisks = anterior end of roof plate (orange) and floor plate (green). D) Detail showing the location of the hypothalamic mamillary and paraventricular regions. Hypo = hypothalamus; post. pit. = posterior pituitary. E) Gridlike arrangement of progenitor domains in the mammalian central nervous system. Based on Figure 17 in [14]. F) The four principal dorsoventral domains common to all prosomeres. Roof plate and floor plate coloured as per (C). Alar and basal plates are separated by the alar-basal boundary.

The prosomere model was first proposed in the early 1990s based on a number of gene expression patterns suggesting a transverse segmental organisation to the forebrain, like the hindbrain rhombomeres [14,15,20]. It underwent several major revisions as additional gene expression information accumulated [16–19]. Proponents proposed that the entire nervous system is composed of transverse segments termed neuromeres, or ‘prosomeres’ in the forebrain (prosencephalon), delineated by adjacent domains of characteristic marker genes (Figure 1B, current model shown). In addition, the neural tube A-P axis was interpreted as following the cephalic flexure, a prominent curvature at the base of the brain (Figure 1C). The curved expression pattern of the transcription factor Nkx2-2 became a molecular marker of the alar-basal boundary (ABB) in the forebrain (Figure 1D, F).

This revised positioning of the axes and prosomeres led to the striking conclusion that the hypothalamus and eye are not part of the ventral diencephalon, but instead are situated anterior to the diencephalon, and ventral (not posterior) to the telencephalon (compare Figures 1B, C to Figure 1A). The hypothalamus and telencephalon were thus grouped as a major new entity, the secondary prosencephalon (Figure 1B).

Although the number and boundaries of these segments have been modified repeatedly, these core premises persist in the most recent iteration of the prosomere model [14], first presented in 2012 [19] and revised slightly in 2015 [17].

Features of the prosomere model

From rostral to caudal, the 2012 ‘updated prosomere model’ divides the forebrain into two secondary prosencephalic prosomeres, hp2 and hp1, and three diencephalic prosomeres, encompassing the prethalamus (p3), thalamus (p2) and pretectum (p1), together with associated ventral regions (Figure 1B). More recently, two midbrain prosomeres (m1/m2) have been reclassified as forebrain [18].

The neural tube A-P axis terminates anteriorly at the hp2 roof plate dorsally, and hp2 floor plate ventrally (Figure 1C). The hp2 roof plate fate maps to the telencephalic septal region and anterior commissure. The hp2 floor plate occupies the ventral midline of the hypothalamic mamillary region (Figure 1D). The hp2 prosomere houses a specialised acroterminal region extending between the anterior ends of the roof plate and floor plate (Figure 1C, D), interpreted as the anterior ‘end wall’ of the neural tube. Structures within this region are equally anterior but span the D-V axis, resulting in an approximately 90° rotation of A-P and D-V axes relative to columnar models (compare Figures 1A, C). Acroterminal derivatives include the optic chiasm and the hypothalamic median eminence and neurohypophysis. Meanwhile, prosomere hp1 gives rise to the evaginated telencephalic pallium, subpallium, amygdala and retromamillary nucleus of the hypothalamus (supramamillary in conventional terminology). The hypothalamic paraventricular (PV) region is divided between hp2 and hp1 within the alar plate (Figure 1D).

While prosomeres are separated from each other by transverse ‘interprosomeric’ boundaries, several longitudinal zones extend continuously across them. Together, these two sets of orthogonal boundaries divide the forebrain into a grid of homologous zones (Figure 1E), explicitly equating D-V subdivisions across the forebrain, hindbrain and spinal cord. Accordingly, all prosomeres share a common D-V organization comprising the floor plate, basal plate, alar plate and roof plate (Figure 1F), with additional microzones further subdividing the basal and alar plates [14].

Drawing on recent evidence, this review critically evaluates core concepts of the updated prosomere model and considers a recently proposed alternative: the tripartite hypothalamus model.

Core concept 1: the forebrain is segmented

The posterior nervous system is organized into discrete transverse segments that repeat along the A-P axis. This is most clearly exemplified in the hindbrain, where transverse rhombomeres are demarcated by sharply defined, lineage-restricted boundaries [20–23]. Rhombomere boundaries exhibit distinctive morphology, a characteristic molecular signature, specialized functional properties, and sharply segregated gene expression domains (Table 1).

Table 1:

Evidence for prosomere segment boundaries. Prosomere boundaries are listed from posterior (rhombomere) to anterior (hp1-hp2). (a) Axon tracts are slightly offset from boundaries; see [14]. (b) The majority of hindbrain tracts course longitudinally or obliquely; however, axons accumulate at rhombomere boundaries during development, which act as guidance cues. Blood vessel arrangement is also raised as evidence but is not considered here as vessels do not yet penetrate the brain at patterning stages.

Type of evidence	Rhombomere (generic)	Boundary
		MB–p1	p1–p2	p2–p3	p3–hp1	hp1–hp2
Morphological
Bump / ridge / groove	complete	complete	complete	dorsal only (ZLI)	No	No
Molecular
Sharp gene expression limits on either side	complete	dorsal only	dorsal only	dorsal only (ZLI)	dorsal only	ventral (retromammillary) only
Specialised ECM, cell junctions	Yes	Not identified	Not identified	Not identified	Not identified	Not identified
Cellular
Enlarged extracellular space	Yes	Yes	No	dorsal only (ZLI)	Not identified	Not identified
Atypical interkinetic nuclear migration	Yes	Yes	Yes	dorsal only (ZLI)	Not identified	Not identified
Lineage restriction	Yes	Yes	Not identified	Yes	Not identified	Not identified
Axon tracts	(a) Limited examples (b)	Posterior commissure	Retroflex tract	Mammillothalamic tract	Cerebral peduncle	Fornix

See text for references. ZLI = zona limitans intrathalamica. MB = midbrain; p1–p3 = prosomeres 1–3; hp1, hp2 = hypothalamic prosomeres 1 and 2.

Whether comparable segmentation exists in the forebrain remains contentious. An early critique of the prosomere model was that the proposed prosomeres and interprosomeric boundaries lacked defining features of classical embryonic segments. Morphological ridges, with specialised cellular features, are seen only at the luminal surfaces of the putative p1-p2 and dorsal p2-p3 interprosomeric boundaries, the latter forming the zona limitans intrathalamica (ZLI, a key dorsal diencephalic signalling center [13,24–26]. While sharply segregated gene expression domains are observed on either side of the p1-p2, p2-p3 and p3-hp1 interprosomeric boundaries [14], no gene has ever been shown to be expressed uniformly across an entire prosomere, despite large-scale two-color in situ hybridization [27] and, more recently, spatial transcriptomic techniques capable of simultaneously profiling hundreds of genes with cellular resolution [6,11,28,29]. Likewise, single-cell RNA-Seq analysis of developing hypothalamus and prethalamus have failed to identify molecular features defining proposed prosomeric divisions [2–4]. This contrasts with hindbrain rhombomeres, which show complete transverse bands of Hox gene expression by mid-somitogenesis, and clear compartmentalization in scRNA-Seq data [1,5].

Gene expression patterns around the hp2-hp1 boundary have received particular scrutiny. Reanalysis of candidate markers reported to discriminate between hp2 and hp1 prosomeres in the hypothalamus, using multiplexed assays in wholemount chicken embryos, found that only a minority displayed sharply segregated expression, at the anterior border of the (ventral) retromammillary domain as early as HH14 (mid-somite stages) (Table 1) [31,2]. Other genes showed graded or late-stage expression differences (E15.5 or later i.e. late gestation), questioning the interpretation of these as true developmental boundaries [31]. These findings also underscore the difficulty of judging subtle spatial relationships from single-color in situ hybridizations on sections.

Lineage tracing in chicken embryos has found no evidence for lineage restriction at most interprosomeric boundaries, with the sole exception of the ZLI (dorsal p2-p3 interprosomeric boundary) [26]. Reported lineage restricted boundaries at the telencephalon-eye [32] and the pallium-subpallium [33] do not align with interprosomeric boundaries. Fate mapping further shows that growth across regions anterior to the ZLI is strongly directed towards the eye and appears unimpeded by putative hp2-hp1 and hp1-p3 boundaries, whereas telencephalic and hypothalamic lineages (from HH10) remain largely segregated across the retinal field [31,63].

Given the paucity of evidence for complete D-V interprosomeric boundaries, their existence is often inferred indirectly, for example from blood vessel arrangement or axon tract trajectories rather than early developmental markers (Table 1) [14]. However, axon tracts form relatively late, frequently change direction, and may run obliquely. As a result, prosomeres -- if present – are now described as “cryptic segments”, inferred primarily through incomplete gene expression patterns and axon tract trajectories. But while the notion of cryptic segments could explain why prosomeres do not display all the expected features of segments, it does not establish their existence in the absence of unambiguous morphological, functional, molecular or comparative evidence.

Overall, evidence accumulated since the introduction of the prosomere model challenges the idea of a segmented forebrain. While p1 and p2 show some boundary-like features, their basal plate limits remain unclear, and there is little experimental support for the p3–hp2 and hp2–hp1 boundaries [4,27,31].

Core concept 2: homologous arrangement of progenitor zones

The prosomere model portrays the central nervous system as a fundamentally equivalent structure along its entire A-P length. Accordingly, the brain and spinal cord are subdivided into transverse neuromeres, each containing a common arrangement of D-V zones comprising floor plate, basal plate, alar plate, and roof plate (Figure 1E, F). These four D-V domains are proposed to form continuous longitudinal zones, creating a grid-like organization.

This serial repetition of homologous units (metamery) is foundational to the prosomere model: a segment qualifies as a neuromere only if it contains all four D-V domains. The model is therefore constrained by the assumption that D-V structure is homologous throughout the nervous system, which is debatable. This premise underlies several defining features of the model, including (1) anterior termination of the neuraxis at the rostral limit of the Shh/Foxa1/Foxa2^+ve floor plate; (2) inferred orientation of the D–V and A–P axes; (3) the acroterminal concept; (4) positioning of the prechordal mesendoderm anterior, rather than ventral, to the neural tube [17,19]; and (5) placement of hypothalamic segments and boundaries (Figure 1C,D).

How has metamery shaped interpretation of hypothalamic segment boundaries? In 2003, proponents of the prosomere model acknowledged the absence of definitive evidence for a segmental boundary within the hypothalamus [16]. Later identification of a small floor plate domain in the murine mammillary region was used to justify distinct hp1 and hp2 prosomeres (Figure 2A) [17,19]. However, the existence of a bona fide hp2 floor plate remains questionable. The proposed hp2 floor plate resembles the retromamillary (hp1) region molecularly, with both expressing Shh and Foxa1 but not Nkx2-1. In mouse, the hp2 floor plate is distinguished largely by morphology, appearing as a small, ventral extension of the retromamillary region (Figure 2B, green) [17,19]. In contrast, molecular analyses in catshark and chicken fail to identify a sustained Shh/Foxa1^+ve midline zone beyond the retromamillary region (Figure 2C, D) [34,31]. Even the retromamillary Shh/Foxa1^+ve midline lacks classic floor plate markers such as En1 and Arx, distinguishing it from definitive floor plate [9,31,35].

Figure 2:

Recent evidence challenges core prosomere concepts. A) Prosomere model detail with selected hypothalamic domains indicated. Hypo = hypothalamic zones. B) Morphology of the ventral hypothalamus in the mouse embryo from E13. The retromamillary domain and mamillary floor plate have a similar gene expression profile (green) but occupy morphologically distinct parts of the basal hypothalamus. C) Ventral hypothalamus morphology and gene expression domains in the stage 29 catshark embryo. D) Ventral hypothalamus morphology and gene expression domains in the E7 chicken embryo. Dotted lines = estimated interprosomeric (blue) and longitudinal (magenta) boundaries. E) Prethalamic/hypothalamic Arx and Meis2 expression in control and Nkx2-1 mutant mouse embryos, E13.5. PTh = prethalamus. Arrow = expanded hypothalamic expression. F) Gene expression in control (DMSO) and SHH inhibited (cyclopamine or sonedigib-treated) chicken embryos. Arrows = ectopic FOXG1, expanded ARX. G) BARHL2/IRX3 expression in the HH20 chicken brain. PTh = prethalamus. H) Prosomere growth model. Crossed arrows indicate higher proliferation in the dorsal versus ventral neural tube. Right - prosomere model axes detail. A = anterior, P = posterior, D = dorsal, V = ventral. I) Updated growth model from [31]. Green/red arrows - direction of predominant growth. J) Partial vertebrate phylogeny showing forebrain morphology and gene expression for selected species at one or two embryonic stages. Genes represented: Nkx2-1/2-4B, HhA/HhB, Pitx2 (Lamprey [3638], note: Foxa not available); Nkx2-1, Shh, Foxa1/2, Pitx2 (Catshark [34,39–40]); nkx2.1, shha/shh, foxa1, (Zebrafish [41–44]); NKX2-1, SHH, FOXA1, PITX2 (Chicken [2,31]); Nkx2-1, Shh, Foxa1, Pitx2 (Mouse [27,31]).

Comparative analysis further challenges the existence of prosomeric hypothalamic zones. In mouse, the curved dorsal edge of the retromamillary and mamillary regions abuts Otp-expressing periretromamillary and perimamillary domains, creating an apparent grid-like organization. In chick, however, these domains are narrow and linear, with transverse and longitudinal boundaries converging rather than intersecting orthogonally [31].

Retromamillary (SHH/FOXA1) and peri(retro)mamillary (OTP) markers abut, while mamillary markers overlap both within a narrow domain (Figure 2D). More broadly, the scarcity of discriminating basal plate markers complicates verification of prosomeric boundaries.

Together, these observations suggest that reliance on late-stage mouse gene expression patterns has produced a model tailored to mouse-specific morphology that is difficult to generalize across vertebrates. Comparative studies challenge the idea that the forebrain can be divided into a universal grid of homologous transverse and longitudinal domains. While appealing in its simplicity, this view is increasingly undermined by insights from evolutionary developmental biology, which emphasises non-homology between parts of the forebrain and more caudal regions of the nervous system.

The chimeric brain hypothesis proposes that the bilaterian nervous system was formed by the fusion of two ancestral nervous systems, generating novel anterior gene combinations and cell types corresponding to the Six/Rax^+ve hypothalamus and anterior forebrain [45,46]. New data backs this up: single cell profiling of the basal chordate amphioxus shows that its central nervous system is built from three distinct embryonic lineages. One of these is unique to the anterior forebrain and its derivatives include both several Nkx2–1^+ve hypothalamic-like and Pax6^+ve prethalamic cell types [47,48]. It has also been suggested that the telencephalon arose through co-option of regulatory genes from sensory placodes, including FoxG, Dlx and Sp8 [49]. Consistently, Dlx/Arx/Sp8/Sp9^+ve GABAergic neurons originate exclusively in the telencephalon, hypothalamus and prethalamus, with a subset migrating into the thalamus [4,8,50]. The anterior forebrain therefore appears unique not only in complexity but also in developmental and evolutionary origin.

Core concept 3: the secondary prosencephalon

Grouping the hypothalamus with the telencephalon was a major innovation of the prosomere model, upending the traditional view of the hypothalamus as ventral diencephalon. This grouping predicts a close developmental relationship between hypothalamus and telencephalon, potentially reflected in shared differentiation states, fate interconversion potential, shared genetic regulatory control, and/or close spatial proximity of progenitors. Data from modern studies allows these predictions to be directly tested. In more caudal regions, manipulation of transcription factors controlling A–P and D–V identity produces inter- and intrasegmental fate transformations [30,51], providing a useful analogy for regulation of forebrain patterning.

Functional studies, however, offer little support for the predictions of the prosomere model. It has been claimed that Six3-deficient mice [52] exhibit a selective loss of telencephalic and hypothalamic identities, and preservation of prethalamic identity, thereby supporting the existence of a discrete secondary prosencephalon [17]. However, the markers used in this study – which include Nkx2-2, Fezf2, and Dlx2 – do not definitively distinguish prethalamus from posterior hypothalamic and/or midbrain structures [3–4,27]. Furthermore, in mice, loss of the hypothalamic markers Lhx2 [53] or Nkx2-1 [3] results in expansion of prethalamic neuroepithelium and loss of hypothalamic structures, consistent with regional fate transformations (Figure 2E). Conversely, loss of the prethalamic markers Dlx1/2 or Nkx2-2 leads to a loss of prethalamic identity, with mutant cells instead expressing posterior hypothalamic markers [4]. Single-cell RNA-Seq and ATAC-seq further demonstrate that transcription factors specifying prethalamic and hypothalamic identity mutually repress one another [4]. By contrast, the prosomere model predicts cross-inhibition between telencephalic and hypothalamic determinants, for which little evidence has been identified.

Additional studies clarify mechanisms coordinating prethalamic and hypothalamic patterning. ScRNA-Seq analysis of early chicken embryos revealed that prospective hypothalamus at neurulation (HH8) resembles a SHH/FOXA1/FOXA2^+ve floor plate-like population (HypFP cells), surrounded by PAX6/OLIG2^+ve prethalamus-like cells [2]. Trajectory inference suggested hypothalamic cells arise from prethalamic-like progenitors under ventral BMP/TGFβ signalling, constrained dorsally by follistatin, a model which was validated experimentally [2]. Complementary perturbation of ventralizing signals, including SHH, led to dorsalization, with expansion of the prethalamic marker ARX into posterior tuberal and anterior mammillary hypothalamus, while the telencephalic marker FOXG1 was ectopically expressed only in the anteriormost tuberal hypothalamus (Figure 2F) [31].

Together, these studies support a close developmental relationship between hypothalamus and prethalamus, rather than between hypothalamus and telencephalon. Molecular evidence reinforces this view. Many genes form contiguous transverse bands spanning hypothalamus and prethalamus [2,3,27], or the ZLI and retromamillary hypothalamus (Figure 2G). The retromamillary domain expresses Irx and Barhl family genes—posterior determinants shared with the ZLI—while lacking the anterior Six and Fezf family markers characteristic of prethalamic and telencephalic regions (Figure 2G) [54–58]. Single-cell and spatial transcriptomic analyses in mouse and human place Barhl1/2 expression at the interface between anterior forebrain markers and posterior forebrain/midbrain programs [1,8,9]. Dorsally, these domains meet at the ZLI, positioning the retromamillary region posterior to the prethalamus, contrary to the prosomere model.

Notably, this Irx/Barhl versus Six/Fezf subdivision is conserved in amphioxus [47,56], and homologous ZLI-like organizers have been identified not only in amphioxus [48] but even in the non-chordate deuterostome Saccoglossus kowalevskii [59]. These findings indicate that this A–P subdivision predates vertebrates and reflects an ancient patterning system. Comparative data therefore support a revised forebrain topology, separating telencephalon, prethalamus, and most of the hypothalamus from more posterior diencephalic and midbrain territories, including the retromamillary hypothalamus.

Core concept 4: forebrain axis positions

A longstanding argument supporting the prosomere axis positions is that, once D–V patterning is established in the neural plate, differential growth causes the neural tube to pivot around the cephalic flexure, preserving original A–P and D–V positional values (Figure 2H). Early fate-mapping studies in chicken supporting this view focused largely on dorsal regions, with limited analysis of ventral domains. Recent dye-based fate-mapping instead reveals region-specific, anisotropic growth, with distinct patterns in alar versus basal regions and in pre- versus post-ZLI territories [31,60]. Dorsal regions do expand extensively, contributing to neural tube bending, but concomitantly, ventral regions exhibit strong anterior extension toward the eye, and ventral midline cells (future tuberal, mammillary, and retromamillary progenitors) are anteriorly displaced relative to dorsal cells (Figure 2I). Consequently, cells ventral to future diencephalic progenitors at HH10 later lie beneath telencephalic, prethalamic, and ZLI cells at HH18 [31,60].

This anterior movement of ventral cells raises the possibility that the D–V axis becomes distorted over time. At late stages, the PITX2^+ve chicken retromamillary region aligns ventrally with the ZLI, but between HH20 and E7 an angle develops between them at the ZLI base (Figure 2G). Comparative analyses across vertebrates reveal similar early transverse alignment of ZLI and retromamillary markers and a relatively straight boundary with mammillary hypothalamus (Figure 2J) [31,57], although this organization is often obscured at later stages.

Fate-mapping and gene expression analyses demonstrate that midline HypFP cells giving rise to tuberal, mammillary, and retromamillary progenitors express early floor plate markers (Shh, Foxa1, Foxa2), confirming their ventral identity [2,60–61]. HypFP cells subsequently downregulate these markers in tuberal and mammillary regions, retaining expression only in the retromamillary domain. The ventral identity of future tuberal hypothalamic cells contradicts the prosomere model’s classification of these cells as acroterminal (anterior). Similarly, classifying the prechordal mesendoderm as anterior is surprising, given its expression of ventral markers and its role in inducing ventral midline characteristics, including floor plate gene expression [2,62–65].

Taken together, evidence accumulated over recent years undermines key assumptions of the prosomere model. Despite early promise and repeated refinements, its defining features—the existence of multiple forebrain segments, serial homology of D–V domains, the secondary prosencephalon concept, and proposed axis positions—are challenged by new data enabled by modern technologies. While some aspects may prove partially correct, the organization of the anterior forebrain—hypothalamus, prethalamus, and telencephalon—clearly requires a fundamentally new interpretation.

The Tripartite hypothalamus model

In the recently proposed tripartite hypothalamus model of forebrain organisation, the forebrain is divided into the anterior and posterior forebrain (Figure 3A) [31]. The anterior forebrain is a uniquely complex, non-segmented region that expresses floor plate markers (Shh/Foxa) only transiently during early development and lacks definitive floor plate markers such as Arx (Figure 3B). It expresses a distinct set of patterning genes [1] and specific cell types, generating Dlx/Arx^+ve GABAergic neurons, for example [4,8,50]. It encompasses the telencephalon, retina, prethalamus and most of the hypothalamus, excluding the retromamillary hypothalamus (Figure 3C).

Figure 3:

the tripartite hypothalamus model of forebrain developmental organization A) Anterior and posterior forebrain divisions. B) Ventral midline domains near the meeting point of anterior and posterior forebrain. C) Major forebrain and hypothalamic regions. The hypothalamus is indicated with a yellow border. D) The anterior and posterior forebrain (FB) meet at a specialised boundary region. E) Boundary (dotted line) between the anterior and posterior forebrain in selected species at the embryonic stages indicated. See Figure 2J for colour key. F) Transverse levels and axes within the forebrain. The hypothalamus (yellow border) is situated mainly within the diencephalon and is ventral to three major regions. A = anterior, P = posterior, D = dorsal, V = ventral. G) Tripartite model axes translated to the mouse embryo.

The posterior forebrain contains the thalamus, pretectum and diencephalic tegmentum (Figure 3C) and shares early determinants with the midbrain. The anterior and posterior forebrain meet at the ZLI and the retromamillary hypothalamus, which are contiguous (Figure 3C, D); gene expression links both to posterior, rather than anterior, forebrain identity. Selective depletion of the proliferation marker PCNA in catshark [34] raises the possibility of a complete boundary along the anterior edge of the ZLI and retromamillary hypothalamus (red dotted line in Figure 3A, approximated by BARHL2 expression in HH20 chicken; Figure 2G), potentially conserved across species (Figure 3E). Factors involved in SHH, WNT, Notch and Ephrin signalling pathways, which help to establish and maintain segment boundaries elsewhere, are differentially expressed along or across this putative boundary [8,31,66].

We further speculate that the retromamillary hypothalamus, together with flanking tissues, forms a broad ventral ‘boundary region’. Rich in signalling interactions and patterning diversity, this region lies at the intersection of anterior and posterior forebrain and extraneural tissues including the notochord, prechordal mesendoderm, Rathke’s pouch and gut (Figure 3D). Its distinctive floor plate expresses Shh, Foxa1/2, and Pitx2 but is Arx-negative (Figure 3B), and may be lineage-restricted from the Arx^+ve, Pitx^−ve ‘definitive’ floor plate [35].

Importantly, the anterior mammillary (defined differently to the prosomere mamillary domain; see [2,3,27,31]) and most of the tuberal hypothalamus are positioned ventral to the prethalamus (Figure 3C, F), restoring most of the hypothalamus to the diencephalon. Only the anterior tuberal hypothalamus lies ventral to the telencephalon. The hypothalamus thus relates to three major dorsal regions—the ZLI, prethalamus and telencephalon—hence ‘tripartite’. The diencephalon itself is split across the anterior and posterior forebrain at the ZLI, implying that it may not be a single developmental unit.

The basal and paraventricular hypothalamus and the prethalamus relate genetically and morphogenetically to the temporal retina, all sharing Foxd1 expression, while the Foxg1^+ve telencephalon and nasal retina are likewise associated. The model therefore posits a split telencephalic/diencephalic identity to the retina, loss of which, in eyeless Rax and Lhx2 mutants, results in expansion of markers of both regions [32,53].

The D–V axis passes through the ZLI and retromamillary hypothalamus, and the ventral midline continues beyond the end of the definitive floor plate into the tuberal hypothalamus, although its anterior limit is not yet defined (Figure 3B, F; compare Figure 3G with Figure 1C). Nkx2-2 is taken to mark the ABB, as in the prosomere model, in mouse and chicken, though it is less useful in the zebrafish tuberal hypothalamus where expression is broad.

The tripartite hypothalamus model complements current views on forebrain evolution [45–49] by suggesting another explanation for the extraordinary diversity of hypothalamic cell types. We speculate that the region beneath the ZLI constitutes a highly derived boundary where, ancestrally, tissue interfaces created an environment enriched in secreted molecules and heterophilic cell–cell interactions. This mixture of cues may have driven iterative evolution of novel cell phenotypes within this restricted forebrain territory.

Unresolved questions include where the A–P axis ends anteriorly, whether it converges to a single anteriormost point or includes an acroterminal-like domain. The relatively understudied paraventricular progenitor region also remains uncertain but is grouped with the prethalamus in a diencephalic unit resembling the zebrafish optic recess region.

Evolutionary approaches using simpler nervous systems of early-branching chordates may be most informative for interpreting topology at the nervous system’s most anterior limits. At these limits, however, different lines of enquiry may yield different conclusions, since orthogonal A–P and D–V axes are ultimately abstractions rather than physical properties of living organisms.

Conclusion

The prosomere model is now widely accepted in developmental neurobiology, appearing in textbooks [19,67] and the Allen Brain Atlas, and bringing forebrain organisation to broad attention. Its proponents pioneered the use of molecular information and made major contributions to precision and consistency in neuroanatomical terminology. Nonetheless, evidence accumulated since the last major update in 2012 contradicts several central claims, indicating that its most striking predictions, such as reassigning the hypothalamus from the diencephalon to the secondary prosencephalon, are no longer tenable. While the model revived interest in forebrain topology and shaped the field’s conceptual trajectory, emerging technologies and new insights now warrant serious consideration of alternative models.