Neural lineage tracing in the mammalian brain

Abstract

Delineating the lineage of neural cells that captures the progressive steps in their specification is fundamental to understanding brain development, organization, and function. Since the earliest days of embryology, lineage questions have been addressed with methods of increasing specificity, capacity, and resolution. Yet, a full realization of individual cell lineages remains challenging for complex systems. A recent explosion of technical advances in genome-editing and single-cell sequencing has enabled lineage analysis in an unprecedented scale, speed, and depth across different species. In this review, we discuss the application of available as well as future genetic labeling techniques for tracking neural lineages in vivo in the mammalian nervous system.

The lineage of a particular cell refers to the history of cellular divisions that sequentially tracks its mother cell up to the original ancestor. Resolving cell lineage relationships of a tissue or organism is important because it provides direct information on how particular cells progressively acquire their final identities. It also helps in the understanding of how gene mutations and environmental factors can perturb the process and result in malformed cells. One of the most successful examples of cell lineage analysis is the full description of the cell lineage tree of C. elegans (Figure 1) [1–3]. Taking advantage of the organism’s transparency, John Sulston and others used Normarski differential interference contrast (DIC) microscopy and traced the complete genealogy of all cells in a nematode by direct visualization and manual drawings. This landmark achievement not only immediately informed the comprehensive lineage relationships between cells, but also served as a fundamental framework instructing subsequent experimental design and data interpretation.

Figure 1. Lineage analysis of C. elegans by direct visualization.

(a) DIC images of the progressively developing nematode embryos (adapted based on Ref. 1). Scale bar: 10 µm. (b) A complete lineage tree of the C. elegans (adapted based on Wikipedia).

The mammalian nervous system such as the cerebral cortex contains a large number of neural cells that are highly diverse with complex progenitor cell origins (Figure 2). Commanding a complete census of all neural cell types in different brain regions as well as their developmental origins is a prerequisite to understanding their development, organization, and function. Since the ground-breaking work of Ramón y Cajal, neural cells have been increasingly defined; yet, our understanding of neural cell diversity, especially the origin of cells, remains limited. A fundamental source of neural cell diversity is likely rooted in the process of genesis during development. Moreover, the developmental history and lineage relationship may influence the structural and functional organization of neural networks [4–15]. Therefore, it is important to systematically delineate the lineage history and relationship of neural cells in the mammalian nervous system.

Figure 2. Neural lineages in the mammalian cerebral cortex.

(a) A schematic representation of the progressive development of the cerebral cortex. As time progresses, neuroepithelial cells (NEs) proliferate and give rise to radial glial progenitor cells in the ventricular zone (vRGs), which account for the major neural progenitor cells in the developing cortex. vRGs initially undergo symmetric division to amplify the progenitor pools and then undergo asymmetric division to produce distinct cortical neurons (SPNs, subplate neurons; CThPN, cortico-thalamic projection neurons; SCPN, subcerebral projection neurons; GNs, granular neurons; CPN, callosal projection neurons) in a temporally progressive manner via intermediate progenitor cells (IPs) and outer subventricular zone radial glial progenitor cells (oRGs). New-born neurons migrate radially in a birth date-dependent inside-out fashion to constitute the future cortex. Towards the end of neurogenesis, a fraction of RGs proceeds to gliogenesis, generating astrocytes, oligodendrocytes, ependymal cells (EpCs), and adult subventricular zone stem cells (adult SCs). CR, Cajal-Retzius cell; MZ, marginal zone; OPC, oligodendrocyte precursor cell; APC, astrocyte precursor cell; (b) The progenitor cell origins of excitatory principal neurons, inhibitory interneurons, and glial cells in the cerebral cortex. Excitatory neurons are generated by PAX6-expressing RGs in the dorsal telencephalon, migrate radially, and form ontogenetic radial units in the cortex. Inhibitory interneurons are produced by NKX2.1-and GSX2-expressing RGs that often contact periventricular blood vessels in the ventral telencephalon, including the medial ganglionic eminence (MGE), preoptic area (PoA), and caudal ganglionic eminence (CGE), migrate tangentially, and frequently form ontogenetic clusters in the cortex. Oligodendrocytes arise from both the dorsal and ventral telencephalon in different waves, whereas astrocytes are mostly generated by progenitors in the dorsal telencephalon. LGE, lateral ganglionic eminence; CP, cortical plate.

In complex, non-transparent species, mosaic marking of cells is essential for tracking cells of interest. This was initially fulfilled by the development and application of various vital dyes or tracers. While informative, the non-selective nature of these dyes and tracers in labeling cells limits their use. With the development of genetically encoded enzymatic (e.g. β-galactosidase, β-gal) and fluorescent (e.g. green fluorescence protein, GFP) cellular markers, it became theoretically possible to selectively label any ancestor or progenitor cells and directly trace their lineage progression and progeny output. Below we discuss the major genetic labeling technologies that allow lineage analysis in the mammalian nervous system in vivo (Figure 3).

Figure 3. Genetic labeling techniques for in vivo neural lineage tracing in the mammalian nervous system.

(a) Plasmid transfection, such as in utero electroporation, introduces the marker gene into progenitor cells either episomally (i) or integrated into the genome via a transposase (ii) for progeny labeling and lineage tracing. (b) Sparse retrovirus infection introduces the maker gene without (i) or with (ii) the barcodes into the genome of dividing progenitor cells for progeny labeling and lineage tracing. (c) Mouse genetics approaches for engineering the marker gene into the genome of progenitor cells via site-specific recombination (i), multi-color labeling (ii), or MADM (iii) for progeny labeling and lineage tracing. (d) CRISPR-Cas9-based genome-editing such as GESTALT (i) or MEMOIR (ii) introduces traceable genomic edits that are progressively and stably accumulated over cell division for lineage reconstruction. (e) Next-generation single cell whole-genome sequencing allows the detection of somatic mutations such as L1 retrotransposon, CNVs, SNVs, and Microsatellites for lineage reconstruction.

Plasmid transfection for lineage tracing

Conceptually, using genetic markers for lineage tracing is rooted in the classic cell transplantation and chimeric embryo experiments that rely on native genetic differences [16,17]. It is worth noting that cell lineage under transplantation condition can be different from that in native condition due to the changes in cellular context.

The emergence of recombinant DNA technology enabled the introduction of exogenous genetic markers into animals for cell labeling and lineage tracing. In utero electroporation has emerged as an effective technique for transfecting one or more enzymatic or fluorescent protein coding plasmids into neural progenitors in vivo (Figure 3a) [18,19]. By adjusting the direction of electrical pulses and the operation time, it is possible to achieve a degree of spatial (i.e. location) and temporal control of DNA transfection. Target cell specificity can also be implemented by using specific promoters or mouse lines to drive marker gene expression [20].

While the labeling density can be adjusted, in utero electroporation of plasmid often transfects a large number of cells and thereby is not suitable for lineage analysis at the individual progenitor cell (i.e. clonal) level. Another limitation is the nonpermanent nature of marker gene expression. Upon transfection, the marker gene plasmid is carried episomally and passed on to the subsequent daughter cells passively at the completion of cell division. It becomes progressively diluted in progeny at each round of division. As a result, the late-born progeny of a lineage may not be reliably labeled due to the low abundance of the marker gene plasmid. Therefore, plasmid transfection techniques usually do not label the entire lineage.

The transposon system offers a solution to plasmid loss due to cell divisions. A class II transposable element (i.e. DNA sequence) can jump in the genome through a cut-and-paste mechanism in the presence of a transposase protein such as TOL2, Sleeping Beauty, or Piggyback [21]. Thus, it offers a strategy to integrate the exogenous marker gene into the genome of transfected cells. The transposon system is an attractive tool for transgenesis and lineage tracing, especially in non-genetically tractable animals, such as ferrets and primates. Of note, however, random genomic integration can lead to disruptions of endogenous genes at or near the integration site (i.e. mutagenesis) that might confound experimental results. On the other hand, the rate of mutagenesis is reduced by a limited round of transposition events due to the transient expression of the transposase and the heterozygous state of most mutations.

Retrovirus infection for lineage tracing

Retroviruses can be considered as transposable elements and thereby act as an agent of gene transfection to deliver an exogenous marker gene for permanent lineage tracing (Figure 3b) [22,23]. While some retroviruses (e.g. Lentiviruses) infect both dividing and non-dividing cells [24], other retroviruses (e.g. Maloney murine leukemia virus) infect only dividing cells and subsequently allow genomic integration of exogenous DNA into one of the two daughter cells [25,26]. The integrated marker gene is then stably inherited by all the descendants of the labeled cells. The infection time and rate (i.e. density) can be easily adjusted depending on the experimental needs. For example, it is possible to serially dilute the retrovirus to achieve infection of a single or few dividing progenitor cells for lineage analysis at the clonal level. Notably, the broad tropism of retroviruses can also be achieved by envelope pseudotyping. These remarkable features have made retroviruses attractive in studying cell lineages in vivo, particularly in the mammalian nervous system [27,28].

There are, however, several limitations of this method that should be kept in mind. First, retrovirus infection-mediated marker gene genome integration is considered to be mostly random [29], and thus may result in insertional mutagenesis of endogenous genes. The invention of replication-incompetent retroviral vectors limits additional infection and mutagenesis in the host cell genome. Second, a stable expression of the marker gene is required to detect the infected cells. Retroviral vectors can suffer gene silencing that compromises the detection of transfected cells [30]. Retroviral silencing occurs stochastically, in an individual locus-specific manner. It may lead to an underestimation of the overall lineage size and complexity, but should not systematically skew the experimental results. To circumvent this limitation, retroviral vectors can be optimized to reduce the probability of silencing. For example, insulator and locus control region elements can be employed to reduce vector insertion site position effects and epigenetic-mediated silencing [31]. In addition, high-throughput next-generation sequencing may be considered to detect the marker gene, instead of the marker protein, for lineage tracing.

The third main disadvantage of using retroviruses for lineage analysis is that they lack the intrinsic resolution of true clonality (i.e. progeny originated from the same dividing progenitor cells), as all cells are labeled similarly. In a typical clonal analysis experiment, a low titer retrovirus is used to sparsely infect individual dividing progenitor cells. In this way, the labeled progenitors and their progeny (i.e. individual clones) are scattered throughout a tissue or organism and may therefore be distinguished. In other words, spatial segregation of labeled cells is used to infer clonal relationship. However, a degree of uncertainty always remains, especially when cells within a clone can disperse widely. To gain additional resolution in lineage relationship, retroviral libraries harboring a marker gene as well as a short but variable DNA fragment as a barcode tag have been developed [32–38]. The rich repertoire of DNA barcode sequences in principle allows the clonality of labeled cells to be inferred at the single-cell resolution. On the other hand, the accuracy and capacity of using barcoded retroviral libraries for a definitive clonal analysis critically depend on the single representation of individual barcodes in the library and the success of barcode recovery [38,39]. Any overrepresentation of individual barcodes would compromise data interpretation (e.g. two clones being considered as one). Current methods for recovering barcodes through PCR amplification from individual cells dissected from tissue sections also appear to be challenging [34,35,38]. With the rapid advances in next-generation sequencing, it is now possible to access barcodes by high-throughput single cell sequencing.

Sparse retroviral labeling has been extensively used for clonal analysis of the mammalian nervous system, especially in the retina and the cerebral cortex [14,26,40–43]. Sibling cells arising from individual progenitor cells have been found to form prominent radially aligned clusters in the developing retina as well as the cortex, supporting the concept of ontogenetic units [15]. It has also been reported that some clones in the cortex disperse widely [36,44,45]. The exact nature of these dispersed clones remains to be investigated.

Mouse genetics for lineage tracing

Mouse genetic engineering is a powerful means to drive the expression of a marker gene in a cell and temporal specific manner for lineage tracing (Figure 3c). Cell specificity can be achieved by using specific promoters alone in transgenics or in the context of gene recombination mediated by two sequence-specific recombinases, Cre that recognizes a loxP sequence and Flp that recognizes FRT sequence [46–48]. Mice are engineered to express the recombinase under the control of a cell type- or region-specific promoter. These mouse lines are then crossed with a reporter line in which a marker gene, such as β-Gal or GFP, is preceded by a loxP or FRT-flanked transcriptional/translational stop cassette. The conditional marker gene is usually inserted into a ubiquitously expressed locus, such as Rosa26. Upon recombinase-driven recombination, the stop sequence is excised and the marker gene is expressed. Additional temporal control of recombination (e.g. genetic inducible fate mapping, GIFM) can be achieved by using an inducible Cre or Flp system, in which the Cre or Flp recombinase is fused with a mutant form of estrogen receptor (CreER or FlpER) that binds Tamoxifen, but not its endogenous ligands [49]. Upon Tamoxifen binding, Cre recombinase translocates from the cytoplasm to the nucleus and triggers recombination and reporter gene expression. The timing and dose of Tamoxifen administration can be adjusted to label progenitor cells at a defined time point in different densities for lineage analysis.

Neural tissues are exceedingly complex and often demand additional cellular resolution. Intersectional approaches have been developed to achieve greater cell type specificity. The split-Cre system expresses two inactive but complementary Cre fragments [50,51]. Alternatively, Cre and Flp recombinases that recognize distinct sequences can be combined [52–54]. These intersectional approaches not only provide additional cell type specificity but also reduce background labeling due to any leakage in reporter gene expression in the absence of recombination.

More sophisticated multi-color reporter gene designs have also been incorporated by mouse engineering to improve the capacity and resolution of lineage tracing. The Brainbow mice takes advantage of stochastic Cre-mediated recombination and incompatible loxP sites to drive the combinatorial expression of four fluorescent reporter genes, including CFP, GFP, YFP, and RFP, under the control of the Thy1 promoter [55]. Upon Cre activation, cells expressing a particular combination of fluorescent proteins share a common progenitor cell origin. As many as 90 different colors can be generated. Sophisticated microscopic analysis is, however, necessary to distinguish the precise expression of different fluorescent proteins at distinct levels. The Confetti mice are an example of a modified version of the Brainbow mice for analysis in any tissue [56]. Stochastic Cre-mediated recombination generates the expression of four distinct markers for lineage tracking; however, the recombination frequency of each marker is not equal.

Mosaic Analysis with Double Markers (MADM) is another remarkable mouse genetic technique for lineage tracing. Analogous in principle to Drosophila Mosaic Analysis with a Repressible Cell Marker (MARCM) [57], MADM relies on the Cre-loxP system to catalyze inter-chromosomal recombination to reconstitute two otherwise nonfunctional split marker genes (i.e. GFP and RFP) [58]. Depending on the chromosomal segregation pattern, the two reconstituted fluorescent marker genes are inherited by the same or two different daughter cells of the labeled progenitor cells, resulting in the generation of yellow or green/red cell lineages, respectively. A great advantage of this approach is that in addition to conventional lineage tracing, the permanent and separate green or red labeling of two daughter cells of one initial dividing progenitor cell offers important insights into progenitor cell division patterns, such as whether they are symmetric or asymmetric.

In conjunction with the available inducible CreER mouse lines, MADM can be implemented in a cell type and temporally specific manner [59–62]. Moreover, MADM can be used to delete endogenous genes located on the same chromosome as the MADM allele selectively in certain labeled cells at the clonal level, whereas the other labeled cells serve as a wild type control. Notably, a recent study utilizing MADM revealed that radial glial progenitors (RGPs) in the developing mouse neocortex exhibit highly deterministic behaviors in proliferation, neurogenesis, and gliogenesis [59]. An alternative method for combining Flp-inducible lineage analysis and Cre-mediated cell mutagenesis is mosaic mutant analysis with spatial and temporal control of recombination (MASTR), which has the advantage that a loxP-flanked gene can be mutated on any chromosome [63,64].

CRISPR-Cas9 genome-editing for lineage tracing

The rapid advance of the clustered, regularly interspaced, short palindromic repeats (CRISPR)-Cas9 genome-editing technology has offered new strategies for large scale lineage tracing even at the whole-organism level (Figure 3d). Several methods have been developed to harness diverse, permanent edits generated by genome-editing at designated genomic regions for lineage reconstruction, such as genome editing of synthetic target arrays for lineage tracing (GESTALT) [65], memory by engineered mutagenesis with optical in situ readout (MEMOIR) [66], and Scartrace [67]. GESTALT introduces synthetic arrays of 9–12 CRISPR/Cas9 target sites as the barcode, which can be progressively and stochastically edited over cell divisions. The accumulated combinatorial barcode edits are systematically examined by targeted single cell sequencing of either DNA or RNA. Lineage relationships can then be reconstructed by relating the patterns of recovered edits. GESTALT has been successfully applied to the zebrafish to evaluate the lineage relationship of ~200,000 cells. A similar approach has recently been applied in C. elegans [68].

While powerful, GESTALT does not preserve the information about the actual anatomical location of each queried cell. On the other hand, MEMOIR provides a versatile platform for CRISPR/Cas9-mediated barcode editing and in situ, single-cell level of readout through multiplexed single-molecule RNA fluorescence hybridization (smFISH). It is also compatible with same-cell measurements of endogenous gene expression through additional rounds of smFISH or other in situ sequencing technologies [69]. Therefore, it has the potential to provide both lineage and endpoint cell state (e.g. cell identity) information for the queried cells.

Additional optimizations on the design of barcode targets, the delivery of editing reagents including Cas9 and guide RNAs (gRNAs), and the efficiency and accuracy in recovering and analyzing edits will further empower genome-editing based techniques towards faithfully constructing complete maps of cell lineages in complex nervous tissues. Furthermore, integrating them with other single cell analysis methods such as RNA sequencing (RNA-Seq) or assay for transposase-accessible chromatin (ATAC)-Seq will add rich cellular information to lineage studies to build comprehensive atlases of neural cell types and brain development.

Naturally occurring somatic mutations for lineage tracing

Naturally occurring somatic mutations in individual precursor cells are transmitted into their progeny, thereby serving as indelible genetic signatures for lineage reconstruction (Figure 3e). For example, highly mutated microsatellite DNA sequences have been used to build cell lineage trees in normal and disease human tissues [70]. A challenge in using naturally occurring somatic mutations as lineage markers is to discover the mutations reliably in individual cells. Due to the low frequency, it is often difficult to identify somatic mutations by sequencing a mixture of cells at conventional depths. Recent advances in next-generation single cell sequencing have made it possible to discover rare mutations or variants at the single cell level that carry lineage information [71–73]. These variants, from the least to the most frequently occurring, include retrotransposons, copy-number variants (CNVs), single-nucleotide variants (SNVs), and microsatellites. The rates at which these variants occur in somatic tissues dictate the resolution of lineage relationship reconstruction.

In principle, this strategy allows lineage tracing in any organisms, including humans. Endogenous retroelements such as long interspersed nuclear element 1 (LINE-1 or L1) retain the mobilization ability to insert into a new genomic location during somatic cell division [74–76]. While the actual rate of L1 retrotransposition in the human brain is debated [77–80], L1 insertion has been used to assess lineage relationship of neurons in the human cerebral cortex [81]. Similarly, somatic CNVs and SNVs are potentially useful lineage-tracing markers [82–85]. The relative readiness in identifying CNVs on the basis of low-coverage sequencing makes the sequencing of many single cells for CNV variant discovery and lineage analysis possible. Microsatellites are the most frequently mutated somatic loci [86]. The analysis of all microsatellite locations in the genome may have the potential for being used to reconstruct the complete cell lineage tree of an entire organism.

One important condition of using somatic mutations for reliable lineage analysis is that the mutations are sufficiently abundant and functionally neutral. In addition, whole-genome single cell sequencing currently requires genome amplification prior to sequencing to generate sufficient material, which can introduce technical artifacts and complication. It is therefore necessary to consider the types and frequencies of potential errors. Finally, single-cell sequencing based lineage tracing does not provide actual information on the precise progenitor cell origin, as the exact location and timing at which the somatic mutations are introduced is unknown.

Concluding remarks

Lineage tracing provides fundamental information on cell production and specification that cannot be obtained by other studies. When designing a lineage tracing experiment, it is important to consider the strengths and weaknesses of all possible techniques and select one that is best suited to the biological question under study. Notably, different techniques are not mutually exclusive but can be integrated. Viral or transposon labeling can be combined with the Cre-loxP system and/or the multi-color reporter system; for example, retroviral labeling has been combined with the Cre-loxP system for lineage tracing of defined progenitor cells [11,87], and the transposon and Brainbow transgene methods have been integrated to develop a multiaddressable genome-integrative color (MAGIC) marker toolkit [88]. Future efforts harnessing multiple technologies in concert will permit systematic analyses of lineage relationships at an ever greater scale and with single cell resolution. In particular, rapid advances in genome-editing and single cell sequencing make it possible to decipher lineage relationships of the entire population of neural cells in a brain region under normal or pathological conditions.

Highlights.

• Cell lineage underlies the developmental steps taken to acquire cell identity.

• Genetic labeling allows lineage tracing with cell type and temporal specificity.

• Genomic-editing and single-cell sequencing enable large scale lineage studies.

• Somatic mutations offer signatures for lineage reconstruction in the human brain.