Temporal controls over cortical projection neuron fate diversity

During neocortex development, cortical projection neurons (PN) are sequentially produced and assemble into circuits underlying our interactions with the environment. Cortical PN are heterogeneous in terms of birthdate, layer position, molecular identity, connectivity, and function. This diversity increases in evolutionarily most recent species, but when and how it emerges during corticogenesis is still debated. While time-locked expression of determinant genes and early stochasticity allow the production of different types of PN, temporal differences in unfolding similar transcriptional programs, rather than fundamental differences in these programs, further account for anatomical variability between PN subtypes and across species. Altogether, these mechanisms, which will be discussed here, participate in increasing cortical PN diversity.


Main text
During development, sequential steps must occur with high precision both in time and space for organs to form correctly. The mammalian brain is a complex organ composed of different structures made of thousands of different cell types with distinct identities, anatomies, and functions [1e4]. In the brain, the neocortex is the evolutionarily most recent part and is highly expanded in primates, underlying their higher cognitive abilities. It displays a tremendous diversity of cell types organized into 6 layers, which form specific functional circuits: in the mouse, hundreds of cell types expressing distinct gene sets (i.e. with distinct transcriptional identities) have been described, with cortical excitatory (glutamatergic) and inhibitory (GABAergic) neurons accounting for the majority, among which about half are glutamatergic projection neurons (PN) [5].
In terms of connectivity and function, cortical PN are divided into two main groups: (1) neurons projecting into the cortex (intracortically projecting, ICPN), such as neurons projecting to the contralateral cortex through the corpus callosum; (2) neurons projecting subcortically, such as motor neurons projecting to the spinal cord [6]. ICPN are predominantly located superficially in layers 4 and 2/3. While they have diverse projections to distinct cortical areas, related to distinct functions in sensory information processing (e.g. primary somatosensory cortex neurons projecting to the motor cortex or to the secondary somatosensory cortex), layer 2/3 ICPN are transcriptionally relatively homogeneous in mice [7] d in humans, a greater heterogeneity is found, particularly in layer 3, which is largely expanded [1]. Neurons projecting subcortically are located deeper in layers 5 and 6, and are very heterogeneous both in terms of projections (e.g. to the thalamus, the pons, the superior colliculus, and the spinal cord) and transcriptional identities (w20 transcriptional subtypes in mice) [5].
While sequential and transient expression of transcription factors in progenitors specifies birthdate-defined neuron identities in invertebrates [8e10], it is less clear how the diversity of neuron types emerges during cortical development in mammals [11,12]. The diversity and proliferative capacity of progenitor types increase with evolution, together with the increasing size of the cortex and number and diversity of cortical PN produced (in particular in superficial layers) [3,13]. However, a linear relationship has not been found between one type of progenitors and one type of PN produced; most progenitor types can produce most PN types. Instead, some "master" genes are expressed in postmitotic neurons early during corticogenesis and define their identity and function, such as the transcription factors Tbr1/Foxp2 for layer 6 neurons, Fezf2/Bcl11b for layer 5 neurons, Cux1/ Cux2 for superficial layer neurons with Rorb defining layer 4 and Pou3f2 layer 2/3 neurons [14].
This review aims to provide recent insights into the temporal molecular programs controlling neuron fate acquisition. Using select examples and focusing on cortical excitatory PN, I will first discuss the temporal controls over the production of distinct neuron types from common progenitors. I will then discuss the temporal controls over the maturation of distinct cortical PN types and present recent work showing that controlling the time of gene action is critical for the emergence of connectivity-defined neuron types.
Temporal controls over the production of cortical PN types a. Time-locked production of different cortical PN.
From embryonic day (E) 8.5e9.5 in mice and posteconception week (pcw) 4 in humans, progenitors of the dorsal pallium undergo self-replicating divisions before generating radial glial progenitors, which will sequentially produce the different types of cortical PN over several days in mice (from E11 to E17) [15] and weeks in humans (from pcw8 to 24) [16] (Figure 1a). Neurons located in deep layers are generated first, followed by neurons located in superficial layers. Radial glial progenitors also give rise to intermediate progenitors, expanding the neuronal output as corticogenesis proceeds (Figure 1b).
While different types of cortical inhibitory neurons come from distinct progenitors in distinct proliferative zones (such as the ganglionic eminences or the preoptic area), contributing to distinct identities (i.e. spatial patterning plays a predominant role in their diversity [17]), different types of cortical PN are born from progenitors located in the most proximal ventricular and  subventricular zones just below the cortical plate. In mice, for example, apical radial glial progenitors can produce all PN types [18].
The competence of progenitors to generate temporally defined neuron types results from the interplay of both cell-autonomous mechanisms and surrounding cues, as their environment changes during development [19]. For example, intrinsic controls over fate specification allow embryonic stem cells cultured without morphogen to recapitulate the sequential generation of the different cortical PN types in vitro [20]. Fate specification is further regulated by environmental cues, and apical radial glia transplanted into a younger brain can produce the neurons normally produced at this developmental stage by sensing this earlier environment and re-entering past molecular, physiological and neurogenic states [21]. Additionally, epigenetic regulations by histone modifications and chromatin remodeling have been shown to influence the cycling properties of progenitors and their neuronal output [19]. For more details on the transcriptional, environmental, and epigenetics influences on progenitor temporal patterning, see recent reviews by Koo et al. and Oberst et al. [19,22].

b. Stochasticity in the emergence of PN heterogeneity
In the "primordial" cortex, fate variability may have been selected over fate reliability to generate diverse neuron types in a limited time. This is for example the case in the evolutionarily conserved piriform cortex in mice, where Dbx1 progenitor lineage produces neurons at E11.5, which will differentiate in diverse types, like the semilunar or the superficial pyramidal neurons, with distinct molecular identities, layer position, morphologies, projections, and functions [23,24]. This may also be the case for neocortical deep layers, which are enriched in PN projecting to distinct targets, like the thalamus or the spinal cord, and expressing distinct marker genes, such as Tbr1 or Bcl11b, respectively. Additionally, a substantial proportion of ICPN expressing Satb2 are found in deep layers, intermingled with other PN types [25]. Those diverse types of PN are simultaneously born from apical radial glia during early corticogenesis [26], suggesting that early on, the same pool of progenitors exposed to the same environment can produce neurons which will acquire very different identities (Figure 1c). While adult deep-layer neurons display a high transcriptional heterogeneity, corresponding to the different types described above, early apical radial glial progenitors do not. Instead, the molecular variability appears to emerge postmitotically, with some immature early born neurons expressing lateborn neuron markers [18,26].
Lineage barcoding strategy further showed that single cortical progenitors can generate both excitatory and inhibitory cortical neurons in the human brain [27]. Altogether, this supports that similar progenitors in a given environment can be multipotent, and that stochasticity (i.e. random generation of PN types, but see the comment by Klingler & Jabaudon [28]) may participate in the generation of a range of PN types from similar progenitors [29]. The sequential expression of master regulators determining PN fate identity may remain plastic, and some stochastic expression may allow the emergence of various projection types at a given developmental time. Early during neocortex development, this variability may persist, yet at later stages relatively homogeneous superficial layer PN are generated to be later diversified (see section 2).

c. Origin of intracortically projecting neurons
Few evidences suggest that distinct types of progenitors may produce distinct types of PN. For example, CUX2 þ progenitors would produce superficial layer ICPN [30]. However, such populations of progenitors have not been identifiable using single-cell transcriptomics approaches and could instead reflects a transient expression of CUX2 protein rather than a genuine subtype [18,31e33].
Cortical progenitors remain capable of producing ICPN throughout corticogenesis, directly or indirectly through intermediate progenitors [25,26,34]. These different progenitor types have been proposed to be at the origin of ICPN subtype diversity. Genetic fate-mapping using Eomes (a transcription factor enriched in intermediate progenitors) showed that amongst primary somatosensory cortex (S1) ICPN, Lhx2-expressing ones, projecting to the ipsilateral motor cortex, seem to be generated through direct neurogenesis, while Plxnd1-expressing ones, projecting to the striatum, come from indirect neurogenesis [34].
Here, clonal analyses [29] or lineage barcoding [27] combined with axon tracing would be critical to determine whether single progenitors could give rise to one type of PN only throughout corticogenesis (Figure 1d), or if the PN type of given progenitors is determined later during development. Thus, as of now, the molecular logic between progenitor identities and the production of a type of cortical PN remains to be identified. In the next part, I will discuss the temporal controls over the differentiation and the maturation of cortical PN, which start embryonically and last during the postnatal period.
Temporal controls over the differentiation of cortical projection neuron types a. Transcriptional programs guiding neuron differentiation Once cortical PN are produced, they migrate to their final location within cortical layers, extend their axon toward their target(s), and develop their dendritic tree and synapses, before final refinements (Figure 2a). The transcriptional programs at play during each of these steps in different types of PN have been difficult to identify, as these are very dynamic processes that can involve redundant molecular pathways. For example, axon guidance proteins can also play a role in progenitor proliferation, neuron migration and synaptogenesis [35].
Single-cell RNA sequencing performed at sequential developmental stages have identified waves of gene expression during the differentiation of cortical PN types both in mouse and human developing cortex [18,31e33,36e40] (Figure 2b). Furthermore, unspliced/ spliced mRNA ratio allows to predict the future state of individual cells on a hours-timescale [41], opening doors for identifying gene expression regulations with even higher temporal resolution. In particular, the expression of receptors, ligands and downstream pathway components is highly dynamic throughout neuron differentiation [40]. Local translation allows restricted and dynamic expression of those proteins at the axon growth cones, leading to specific responsiveness to surrounding signals at specific developmental stages and correct axon pathfinding [42,43].
During cortical development, genetic programs first define neuronal differentiation stage rather than mature projection types. Early on, developmental programs are conserved between neuron types (and even across species), while neuron-type specific programs emerge later during maturation [44] (Figure 2b). This may reflect the fact that once any type of cortical PN is born, it must first undergo common developmental steps (i.e. radial migration, neuritogenesis) before projection-type specific processes, like axon guidance to specific targets. This involves, for example, the expression of genes like Dcx or Tubb3, when neurons are migrating, followed by Map2 when they start growing neurites [45,46] (www. humous.org). The PN type transcriptional identity may therefore initially be "embedded" in strong genetic programs encoding early common steps or only emerge at later developmental times. However, some molecular determinants of cortical PN are expressed early on, in newborn neurons, or even at the progenitor level [40,47]. For example, deep-layer neuron determinants such as Sox5, Bcl11b and Fezf2 are expressed in one-day-old earlyand late-born neurons. They are subsequently repressed in the late-born ones, allowing them to acquire their superficial layer identity [18]. The co-expression of progenitor and neuronal differentiation programs makes the definition of cell type identity challenging (e.g. is a cell co-expressing Eomes and Bcl11b an intermediate progenitor expressing neuronal programs or an immature neuron still expressing progenitor programs?), as is the interpretation of fate-mapping experiments based on the expression of a single gene in a "specific" cell type. Altogether, at which precise stage these neuronal determinants become critical for neuron fate acquisition remains largely unclear.
The transcription factor LHX2 represses corticospinal PN identity by regulating Fezf2 expression [48]. While progenitors expressing Lhx2 between E10.5 and E12.5 give rise to all types of cortical PN, postnatal day (P)5 neurons expressing Lhx2 only become ICPN, suggesting differential roles for LHX2 at different developmental stages/in different cell types [49]. For example, when LHX2 is absent from early subplate progenitors, thalamocortical axons fail to develop correctly [50]. In contrast, the areal patterning nuclear receptor NR2F1 (COUP-TF1) is critical in both progenitors and postmitotic neurons for correct cortical patterning [51]. Ablation of Nr2f1 induces the premature generation of corticospinal PN (when layer 6 corticothalamic PN are normally born) and the abnormal differentiation of latergenerated genuine corticospinal neurons generated after [52]. Altogether, these examples show how a single molecular determinant can play critical roles throughout the generation and differentiation of specific types of PN.
Finally, after a period where the PN identity remains plastic, molecular determinants lose their ability to control PN fate. Although some are still expressed, their functional partners may not be, or their target DNA may have become inaccessible because of epigenetic modifications, preventing them to execute their developmental function. For example, P1 overexpression of the transcription factor Fezf2 -critical for the acquisition of layer 5 corticospinal PN identity [53] -in layer 4 neurons reprograms them into corticospinal-like PN, while from P10 layer 4 neurons are only partially reprogrammed [54].
Controlling the time of gene action is therefore key to understanding how genetic regulations allow different PN to be specified at precise developmental stages. Recent technologies using inducible CRISPR screens will likely enable transient manipulation of gene expression with higher temporal resolution [55], and therefore investigation of function at specific steps of PN fate acquisition.
b. Intrinsic programs control the pace of neuron maturation The maturation of cortical PN (i.e. axon branching, dendritic tree and synapse development) highly depends on activity and input [6]. However, the pace of neuron maturation is primarily encoded by cell-intrinsic programs and varies across species [19], ranging from 5 weeks in mice to several years in humans. This longer maturation in evolutionarily more recent species includes prolonged dendritic morphogenesis and synaptogenesis, and synaptic pruning, and is conserved in in vitro models [56], including organoids [57]. Furthermore, in vivo, xenotransplanted human neurons into the mouse cortex conserve their human-like prolonged development [58] (Figure 2c). Differential expression of specific genes does not explain such difference in timing, which has instead been related to different speeds in protein turnover between species, with slower kinetics in humans than in mouse cells [59,60].
Such molecular heterochronies have also recently been reported between species with cortices of comparable size. During corticogenesis, marsupials show more precocious but protracted maturation programs than eutherians (see below; [61]). Interestingly, even within an organism or a given organ, different cell types have different maturation rhythms [56], but to which extent this rhythm instructs the fate of different cell types only starts to be addressed.

c. Temporal controls over cortical projections
The expression time window of generic sets of genes controls the establishment of cortical connectivity. While in marsupials, neocortical hemispheres are interconnected via the ventrally located anterior commissure, in eutherian mammals this connection is made via a novel tract, the dorsally located corpus callosum [62]. SATB2 is a critical determinant of callosal PN: Satb2 knockout mice display an agenesis of the corpus callosum with rerouting of axons to subcerebral targets and the anterior commissure [63]. In dunnart marsupials, Satb2 is expressed from stage 20 (corresponding to E12 in mice) but not before E14 in the mouse neocortex. In mice, heterochronic expression of Satb2 from E12 on (stage at which deep layer subcortically projecting neurons are born) induces a marsupial-like phenotype with increased anterior commissure projections [64] (Figure 2d). Thus, temporal differences in unfolding similar transcriptional programs, rather than fundamental differences in these programs, can account for anatomical variability across species.
Although in mice, superficial layer ICPN are transcriptionally relatively homogeneous, their projections are quite diverse, with, for example, subpopulations of S1 neurons projecting to the motor cortex and others to the secondary somatosensory cortex (S2) [65]. While very similar in terms of gene expression, neurons projecting to the motor cortex display a prolonged expression of developmental genes compared to those projecting to S2, and have a delayed maturation both in terms of axon and dendrite extension, and circuit function [7].
Altogether, this shows that dynamic differences in the expression of a conserved set of genes may account for the development of distinct axonal projections and participate in the emergence of cortical PN diversity.

Concluding remarks
Unraveling the temporal controls over the specification of cortical PN types is critical to understanding how their diversity emerges during development and allows the formation of functional circuits. This has further implications in the era of in vitro model development.
Indeed, to recapitulate brain development in a dish using organoids, understanding the sequence of gene action is critical to producing neuron types comparable to what is observed in vivo [66]. As of now, generating dorsal or ventral forebrain neurons from pluripotent cells in vitro has been made possible thanks to the addition of cell-fate factors (e.g. by modulating the Sonic Hedgehog pathway) [67]. The temporality of the molecular cocktails needed to recapitulate the cellular diversity of the neocortex largely remains to be elucidated to address more functional questions and improve disease modeling.

Conflict of interest statement
Nothing declared.

Data availability
No data were used for the research described in the article.