Main

The neocortex is responsible for carrying out many higher-order cognitive functions, such as decision making, language comprehension and sensory perception. During neural development, progenitors in the cortex and ganglionic eminences produce the diverse array of neurons and glia that comprise the neocortex. Extensive developmental studies in rodents have demonstrated that the two principal types of cortical neuron—glutamatergic excitatory neurons and GABAergic inhibitory interneurons—are produced by two distinct groups of progenitors1,2,3,4,5. Excitatory neurons are generated by progenitors located dorsally in the developing cortex and migrate radially to occupy their terminal positions in the cortical plate. By contrast, cortical interneurons are generated by progenitors located in the ganglionic eminences of the ventral forebrain and then migrate dorsally into the developing neocortex.

A few limited studies have suggested the possibility that human cortical progenitors might also give rise to cortical interneurons, but the findings have been conflicting6,7,8,9,10. Short-term clonal labelling of progenitors in the human cortex revealed the local generation of newborn GABAergic neurons7, but did not determine whether they were cortical interneurons or another type of interneuron such as those of the olfactory bulb, which have previously been shown to derive from cortical progenitors in mice11,12,13. Similarly, in vitro cultures derived from human cortical progenitors have been shown to generate GABAergic inhibitory neurons10. Conversely, a later study that labelled dividing cells in short-term cultures of human organotypic tissue slices did not find a substantial fraction of newborn inhibitory neurons in the cortex8. The overwhelming consensus remains that human cortical progenitors give rise to excitatory neurons but not cortical inhibitory neurons; however, this has not been thoroughly examined and the developmental potential of individual human cortical progenitors remains largely unknown.

Design and validation of lineage tracer

In order to carry out high-throughput clonal lineage tracing of primary human neural progenitors, we developed STICR—an ultra-high-complexity barcoding strategy that allows us to permanently label cells and their progeny. This approach uses a lentivirus encoding a heritable, transcribed molecular barcode within the 3′-untranslated region of the enhanced green fluorescent protein (EGFP) reporter gene (Fig. 1a). The combinatorial, single-molecule barcode design of STICR allows for a predefined, error-correctable barcode library with a maximum diversity of 125 million sequences (Extended Data Fig. 1a; see Methods). Deep sequencing of STICR plasmid and lentiviral libraries confirmed their ultra-high barcode diversity (approximately 50 million to 65 million unique barcodes per library) without any overrepresented barcodes (Extended Data Fig. 1b). Using the observed barcode diversity and frequency from each STICR library, we modelled the rate of barcode ‘collision’—the event in which two different cells are independently labelled with the same barcode—and found that STICR could be used to label more than 250,000 cells before reaching an estimated barcode collision rate of around 0.5% (Extended Data Fig. 1c). To confirm the accuracy of STICR barcode recovery from single cells, we carried out a cell-mixing, or ‘barnyard’, experiment in which we labelled mouse and human cells with different STICR libraries that were readily distinguishable by a constant ‘viral index’ sequence unique to each library (Extended Data Fig. 1d). Following single-cell RNA sequencing (scRNA-seq), recovered STICR barcodes were 100% concordant with the STICR library used to infect each population (Extended Data Fig. 1e, f), indicating that our method is accurate and can be applied robustly to carry out high-throughput clonal lineage tracing.

Fig. 1: STICR-labelled progenitors generate all three principal cortical cell types.
figure 1

a, Design of the STICR lentiviral vector. CAG, cytomegalovirus/chicken β-actin promoter; WPRE, woodchuck hepatitis virus post-transcriptional regulatory element; pA, polyadenylation signal. b, The experimental design used to label and capture samples. GW, gestational week; Rep, replicate. c, Uniform manifold approximation and projection (UMAP) embedding and Leiden clustering of STICR-labelled cells following scRNA-seq. IPC, intermediate progenitor cell. d, Feature plots showing the principal-cell-trajectory marker genes DLX2 (GABAergic inhibitory neurons and DLX2+ IPCs), NEUROD2 (excitatory neurons), GFAP (glia), and EOMES (EOMES+ IPCs). e, Histogram showing clone sizes within each sample. Left, clone sizes of 1–75 cells. Right, clone sizes of more than 75 cells, in 25-cell bins.

Full size image

Lineage tracing of human neural progenitors

To determine the developmental potential of individual human cortical progenitors, we derived primary human cell cultures from the cortical germinal zone of three different specimens at stages of peak neurogenesis (gestational weeks 15 and 18, GW15 and GW18) (Fig. 1b). Before clonal labelling, one of the specimens (GW18) was further dissected on the basis of known anatomical landmarks, allowing us to generate region-specific cultures from the germinal zones of the prefrontal cortex (PFC), primary visual cortex (V1) and medial ganglionic eminence (MGE). We then labelled cells with STICR and cultured them for six weeks in vitro before carrying out scRNA-seq. Transcriptome analysis of 121,290 cells identified three principal cortical cell type trajectories— excitatory neurons, GABAergic inhibitory neurons and glia—on the basis of differential gene expression, including that of marker genes NEUROD2, DLX2 and GFAP, respectively (Fig. 1c, d and Extended Data Figs. 2, 3). We identified intermediate progenitor cells (IPCs) within both the inhibitory and the excitatory neuron trajectories, which we refer to as DLX2+ IPCs (inhibitory trajectory) and EOMES+ IPCs (excitatory trajectory) (Extended Data Fig. 3). Cluster correlation analysis of STICR datasets with an scRNA-seq atlas of the developing primary human cortex at comparable developmental time points and regions14 further supported these cell-type designations (Extended Data Fig. 2d).

We recovered STICR barcodes in 63 ± 23% (mean ± s.d., n = 5 libraries) of cells per culture, including cells of each principal cell type (Supplementary Table 1 and Extended Data Fig. 3g). In total, we identified 1,461 unique clonal barcodes, 1,324 of which belonged to multicellular clones with a median size of 23 cells per clone (Fig. 1e). Although there is very little known about the output of human cortical progenitors over this time scale, we observe a maximum clone size of 1,209 cells, which is congruent with a prior study that measured the output from three individual human outer radial glia15. Notably, no STICR barcodes were shared between experimental groups (Extended Data Fig. 3h), further indicating that cells identified as members of multicellular clones here are not the result of barcode collision. Although all three principal cortical cell types (excitatory glutamatergic neurons, GABAergic interneurons and glia) were found in multicellular clones in both GW15 samples and the GW18 V1 sample, GW18 PFC clones contained relatively more GABAergic inhibitory neurons than excitatory neurons or glia (Extended Data Fig. 3i). Consistent with previous studies, the GW18 MGE culture completely lacked excitatory neurons and instead comprised almost entirely interneurons (Extended Data Fig. 3i). Thus, STICR reveals clonal lineage relationships of all principal cell types derived from human forebrain progenitors.

Clonal relationship of human cortical neurons

In order to determine the lineage relationships of cells born from cortical progenitors, we first analysed the relative proportions of principal cell types in multicellular clones. The majority (66%; 829 of 1,252) of multicellular clones from cortical cultures contained at least 1 excitatory neuron, with these clones containing a median of 3 excitatory neurons per clone (Extended Data Fig. 4a, b and Supplementary Table 1). Although our study of regional cortical progenitors is limited to a single GW18 sample, we found that PFC clones contained proportionally fewer excitatory neurons than clones in the V1 sample (Extended Data Fig. 4b, c). This is consistent with the neurodevelopmental gradient in the cortex, where developmental milestones are reached in the PFC several weeks before the V1 region16. Across all samples, only 1.5% (19 of 1,252) of cortical clones were composed entirely of excitatory neurons, and 51% (635 of 1,252) contained a combination of excitatory neurons, inhibitory neurons and glia (Fig. 2a–c, Extended Data Fig. 4c and Supplementary Table 1). Owing to extensive aggregation of excitatory neurons within in vitro cultures and the gentle dissociation used (Extended Data Fig. 4d, e), production of excitatory neurons may be underestimated in the dataset derived from cell culture experiments. Nevertheless, across GW15 samples, the coincidence of excitatory and inhibitory neurons within the same clone occurred across the range of clone sizes (Fig. 2b). Together, these results suggest that individual human cortical progenitors reproducibly generate both excitatory and inhibitory neurons.

Fig. 2: Individual human cortical progenitors can generate both excitatory and inhibitory cortical neurons in vitro.
figure 2

a, Stacked boxplot depicting the average (mean ± s.d.) proportion of GABAergic inhibitory neurons, DLX2+ IPCs, EOMES+ IPCs, excitatory neurons and glia in different sized clones. b, Bar graph depicting the proportions of clones of different sizes that contain both excitatory and inhibitory neurons. c, Representative 6-cell (GW15, replicate 1, clone 113) and 61-cell (GW15, replicate 1, clone 732) clones depicted in UMAP space. Cells within each clone are coloured in red. Dashed lines depict borders of principal cell types from Fig. 1c. d, Pseudotime transcriptional trajectories of subclustered inhibitory neurons. Three different interneuron transcriptional trajectories (IN.1, IN.2 and IN.3) are indicated with arrows. e, Violin plots depicting expression of IN.1 marker genes (SST, NPY, TAC3 and NXPH1) and general CGE marker genes (SCGN, SP8, PCDH9 and BTG1) in IN.1, IN.2 and IN.3 trajectories. f, Volcano plot comparing differences in gene expression between IN.2 and IN.3 trajectory cells. Not sig, genes with log2 fold-change less than 0.125 and/or adjusted P value greater than 0.1. g, Venn diagram depicting the number of multicellular cortical clones that contain IN.3 cells or excitatory neurons.

Full size image

Analysis of cortically born inhibitory neurons

We next sought to determine the transcriptional identities of cortically born GABAergic neurons. In addition to DLX2 (Fig. 1d), inhibitory-trajectory cells were broadly enriched for general markers of interneurons, including GAD1, ARX, SLC32A1 and DLX6AS-1 (Extended Data Fig. 3e). Iterative subclustering and transcriptional trajectory analysis of the inhibitory cells along with the DLX2+ IPCs revealed three distinct subgroups of GABAergic inhibitory neurons that we termed IN.1, IN.2 and IN.3 (Fig. 2d and Extended Data Fig. 5a–c). IN.1 cells were enriched for markers of SST+ cortical interneurons, including SST, NPY, TAC3 and NXPH1 (Fig. 2e). Consistent with developmental studies in mice which showed that SST+ cortical interneurons derive primarily from the MGE17, 73% (492 of 671 cells) of IN.1 cells are produced by MGE progenitors (Extended Data Fig. 5d). By contrast, IN.1 cells comprised only 0.3% (179 of 56,244 cells) of progeny born from cortical progenitors (Extended Data Fig. 5e). Furthermore, MGE-derived IN.1 cells expressed canonical MGE-born interneuron genes, including LHX6, NKX2-1, ACKR3 (CXCR7), PDE1A and MAF, while cortically born IN.1 cells did not (Extended Data Fig. 5f). Together, these data suggest that IN.1 cells are transcriptionally similar to SST+ cortical interneurons and that the majority of IN.1 cells derive from the MGE. In contrast to IN.1 trajectory cells, IN.2 and IN.3 trajectory cells were transcriptionally similar to cells born from the caudal ganglionic eminence (CGE) based on their expression of marker genes such as SCGN, SP8, PCDH9 and BTG1 (Fig. 2e)18. Furthermore, IN.2 and IN.3 cells differed from IN.1 cells in that they were derived entirely from cortical progenitors, with no contribution from MGE progenitors (Extended Data Fig. 5d). Top IN.2 markers included TSHZ1, PBX3, MEIS2, CALB2, CDCA7L, SYNPR and ETV1, which are enriched in mouse olfactory bulb interneurons (Fig. 2f)19,20. By contrast, top IN.3 marker genes included NR2F1, NFIX, PROX1 and NR2F2,which are enriched within the CGE, as well as SOX6 and CXCR4, which are enriched in cortical interneurons (Fig. 2i), suggesting that these cells are transcriptionally similar to CGE-derived cortical interneurons14,21,22,23,24,25. Comparison of IN.2 and IN.3 marker genes with orthogonal datasets, including those from the Allen Brain Atlas26, similarly support this distinction (Extended Data Fig. 5g, h). Thus, although there are currently no marker genes that can unequivocally distinguish cortical interneurons from olfactory bulb interneurons, our transcriptome-wide data suggest that IN.2 cells resemble olfactory bulb interneurons while IN.3 cells are similar to CGE-born cortical interneurons.

Previous fate-tracing studies in mice have shown that cortical progenitors can produce a subset of embryonically born olfactory bulb interneurons11. In agreement with these findings, we found that multicellular cortical clones frequently contain olfactory-bulb-like (IN.2 branch) GABAergic inhibitory neurons (Extended Data Fig. 5d, e). Notably, many clones containing olfactory-bulb-like GABAergic inhibitory neurons also contained excitatory neurons (61%; 321 of 530 clones) (Extended Data Fig. 5j). Such clones were found in all cortical samples but were especially common in both GW15 samples (Supplementary Table 1). GW18 PFC clones contained the highest proportion of olfactory-bulb-like GABAergic neurons (41%; 1,637 of 3,949 cells) out of any sample analysed, and although relatively few of the GW18 PFC clones contained excitatory neurons (9 of 141 clones), 8 of these also contained at least one olfactory-bulb-like interneuron (Extended Data Fig. 5d and Supplementary Table 1). These results suggest that, similar to mice13, human cortical progenitors can generate olfactory bulb interneurons during embryogenesis, and that some excitatory neurons and embryonically born olfactory bulb interneurons are clonally related.

Surprisingly, we found that most (79%; 655 of 829) multicellular clones that contained excitatory neurons also included putative cortical interneurons (IN.3 cells) (Fig. 2g), indicating that some human cortical progenitors can generate both excitatory neurons and cortical interneurons. Many of these clones contained multiple cells of both types, and mixed clones were especially abundant in GW15 and GW18 V1 samples. Further subclustering analysis of excitatory neurons revealed both deep-like and upper-like excitatory neuron subgroups, both of which were frequently clonally related to IN.3 cells (Extended Data Fig. 6). Notably, every cortical sample we analysed contained clones with both excitatory and inhibitory cortical neurons (Fig. 2g). These mixed clones also frequently contained olfactory bulb interneurons as well as glial-trajectory cells (Extended Data Fig. 5i, j). Taken together, our results suggest that human cortical progenitors cultured in vitro are unexpectedly multipotent in their ability to generate a wide variety of principal neural cell types, including both excitatory neurons and putative cortical interneurons—two cell types previously thought to be produced by different pools of spatially restricted progenitors in the developing forebrain.

Mixed excitatory–inhibitory clones in xenografts

To confirm that the observed lineage relationship between excitatory neurons and cortical interneurons was not an artefact of in vitro culture, we STICR-labelled cortical germinal zone cells from three additional GW15 specimens as above, and transplanted them into the cortex of early postnatal immunodeficient mice, where we allowed them to develop for six weeks before analysis (Fig. 3a). Immunohistochemistry revealed xenografted human cells at the injection site itself as well as distributed throughout the adjacent tissue (Figs. 3b and Extended Data Fig. 7a–c). To quantify the proportion of the principal cell types in the xenografts, we carried out immunohistochemistry for the excitatory-neuron marker NEUROD2, the interneuron marker GABA, and the glia markers SOX9 and OLIG2. Excitatory neurons were the most common cell type, accounting for 75.7 ± 8.4% of xenograft-derived cells, followed by glia at 16.8 ± 6.3%, and interneurons at 7.9 ± 3.2% (mean ± s.d.; n = 7 recipient mice; Extended Data Figs. 7a–c). In parallel, we isolated EGFP+ xenograft cells by fluorescence-activated cell sorting (FACS), and carried out scRNA-seq to determine their transcriptional identities and clonal relationships. Similar to our in vitro cultures, xenograft cells formed distinct transcriptional clusters of GABAergic neurons, excitatory neurons and glia (Fig. 3c and Extended Data Fig. 7d–f). The proportions of excitatory and inhibitory neurons within multicellular clones from xenograft experiments differed from those observed in our in vitro cultures (Fig. 2g). This was likely to be due to the specific loss of excitatory neurons from in vitro culture during dissociation, as these cells frequently associated in tightly formed masses that were difficult to dissociate (Extended Data Figs. 4d, e). Quantification of principal cell types by scRNA-seq closely matched the cell proportions observed by immunohistochemistry, and was reproducible across biological replicates (Extended Data Fig. 7g). These data suggest that xenograft-derived STICR-labelled cells analysed by scRNA-seq accurately reflected the cell proportions produced within the xenograft.

Fig. 3: Xenografted human cortical progenitors generate both excitatory and inhibitory cortical neurons in the same clone.
figure 3

a, Experimental design and analysis of STICR-labelled progenitors by immunohistochemistry (IHC) and scRNA-seq following transplantation into the postnatal murine cortex. b, Representative image of transplanted human cortical cells. EGFP expression from STICR is in green. Scale bar, 500 μm. c, UMAP embedding and Leiden clustering of xenografted cells following scRNA-seq. d, Histogram of clone sizes within each xenograft sample. Left, clone sizes from 1–25 cells. Right, clone sizes of greater than 25 cells, in 25-cell bins. e, Stacked barplot depicting the relative proportions of principal cell types within multicellular clones of each sample. f, UMAP embedding of both cultured and xenograft-derived inhibitory neurons. Xenograft-derived cells that are members of multicellular clones are highlighted in red. IN.2 and IN.3 trajectories are depicted with arrows. g, Venn diagrams showing the number of multicellular clones containing excitatory neurons and/or IN.3 neurons.

Full size image

We recovered STICR barcodes from 76 ± 10% (n = 3 libraries) of xenografted cells and identified 660 multicellular clones that ranged in size from 2 to 101 cells, with a median clone size of 5 cells (Fig. 3d). Within multicellular clones, the proportions of principal cortical cell types were highly similar across biological samples (Fig. 3e). We then further analysed cells categorized as either inhibitory neurons or DLX2+ IPCs in order to determine their transcriptional identities and clonal relationships. Subclustering analysis revealed several distinct groups of GABAergic cells (Extended Data Fig. 8a), including one that was enriched for genes found in cortical interneuron-like IN.3 cells—such as NR2F1, KLHL35, NFIX and SCGN—and was similar to reference GABAergic cortical inhibitory neurons (Extended Data Fig. 8b, c). In order to directly compare the transcriptomic identities of GABAergic neurons from xenografts to those observed following in vitro culture, we integrated scRNA-seq data of GABAergic neurons from both sets of experiments, using the in vitro cultured cells as a reference. GABAergic neurons from the xenografts clustered well with their in vitro counterparts (Fig. 3f, Extended Data Fig. 8d) and integrated entirely within the previously defined transcriptional trajectories (Fig. 2f). Consistent with their marker-gene expression, 85% (211 of 249) of GABAergic xenograft cluster-1 cells clustered within the IN.3 trajectory (Extended Data Fig. 8d), suggesting that they had the transcriptional identity of cortical interneurons. Few (4%; 9 of 249 cells) GABAergic xenograft cluster-1 cells clustered within the IN.2 trajectory (Extended Data Fig. 8d), consistent with previous transplantation studies which found that the cortical environment is not conducive to the generation of olfactory bulb interneurons27. Xenograft-derived GABAergic IN.3 neurons were found in 56 multicellular clones, 43 (77%) of which also contained excitatory neurons (Fig. 3g, Extended Data Fig. 8e–g and Supplementary Table 2). Mixed IN.3/excitatory-neuron clones were found in all three xenografted samples, including GW15 replicate 5, which was transplanted immediately after transduction with STICR and was never cultured (see Methods). Thus, as we had previously observed in vitro, cortical progenitors can generate mixed excitatory/inhibitory-neuron clones when xenografted into the perinatal mouse brain, and these inhibitory neurons are transcriptionally similar to cortical interneurons.

Finally, to determine whether human cortical-progenitor-derived GABAergic inhibitory neurons had additional features consistent with cortical interneurons, we characterized their morphology and distribution in two additional xenograft experiments from GW17 samples. For these experiments, we used a FACS-based isolation strategy to enrich for human progenitor cells before transducing them with STICR (Fig. 4a, Extended Data Fig. 9). Given the protracted time scale of human brain development, we waited for 12 weeks after transplantation before analysing the xenografts by immunohistochemistry. We observed STICR-labelled cells with well-elaborated processes distributed throughout the cortical layers and also within the olfactory bulbs (Fig. 4b and Extended Data Fig. 9). Of STICR-labelled cells located in the cortex, 8.3% (89 of 1,071) expressed GABA, and these were also broadly dispersed across the cortical layers (Fig. 4c, d and Extended Data Fig. 10). Although most STICR-labelled GABA+ cells were located at a similar rostral–caudal level as the transplantation site, we did observe cells more distant from the transplantation site in both directions. Together, these morphological characteristics of cortically derived GABAergic inhibitory neurons are consistent with their classification as cortical interneurons. Thus, individual human cortical progenitors can give rise to both excitatory neurons and cortical GABAergic inhibitory neurons (Fig. 4e).

Fig. 4: Xenografted human cortical progenitors generate GABAergic inhibitory neurons that distribute across the cortical laminae.
figure 4

a, Experimental design and analysis of PTPRZ1-enriched, STICR-labelled progenitors by IHC following transplantation into the postnatal murine cortex. b, Representative images of transplanted human cortical cells analysed by IHC. EGFP expression from STICR is depicted in green, NEUROD2 in red, and DAPI counterstain in blue. Arrows indicate NEUROD2+/STICR+ double-positive cells. Scale bar, 50 μm. c, Representative images of STICR-labelled GABA+ cells throughout the cortical plate. Red arrows depict soma of two cells in the same image. Scale bars, 10 μm. d, Relative laminar positions of GABA+/STICR+ cells in the cortical plate from the corpus callosum (value of 0.0) to the pial surface (value of 1.0) in each host mouse brain. Host brains 1 and 2 were transplanted with GW17 (replicate 1); host brains 3 and 4 were transplanted with GW17 (replicate 2). e, Schematic showing the differences in the developmental potential of human and mouse cortical progenitors. OB, olfactory bulb.

Source data

Full size image

Discussion

By carrying out high-throughput lineage tracing of approximately 1,900 human cortical progenitors, we have been able to demonstrate a lineage relationship between human excitatory neurons and putative cortical interneurons—two populations that had been widely believed to arise from distinct progenitor populations. Although existing scRNA-seq informatic tools have been used to describe the transcriptional trajectories between progenitors and mature cell types (that is, pseudotime analysis28), extensive analysis of the developing human brain at this time has not uncovered this lineage relationship29, nor would further analysis be likely do so given the degree to which the transcriptional trajectories of these cell types differ. Furthermore, pseudotime analysis itself does not provide direct evidence of clonal relationships, and the lineages of transcriptionally similar progenitors can differ. At present, approaches that retrospectively infer developmental lineage relationships from somatic mutations predicted to have arisen during development30,31,32 lack the resolution to confidently assign daughter cells to individual progenitors, and thus would not have revealed this relationship.

Previous studies suggested that cortical progenitors generate cortical interneurons in primates, but were limited by their use of a small set of putative ‘marker’ genes that do not directly link progenitors to their progeny6,7. In these studies, differences in the expression of GABAergic marker genes between different germinal zones were presumed to persist across differentiation from progenitors to inhibitory neurons. These markers were thus used to retrospectively infer the origin of cortical interneurons. However, this type of approach cannot account for potential changes in gene expression that can occur over development, or for potentially unknown sources of cells expressing the same putative ‘marker’ genes. Additionally, although one study identified cortically born interneurons using a gammaretorviral labelling approach7, their approach did not determine whether these cells were cortical interneurons or olfactory bulb interneurons, which our work and others have found to be produced from progenitors in the cortex11,12,13. Our prospective approach of clonal labelling by STICR combined with scRNA-seq-based transcriptome-wide analysis has allowed us to distinguish between these two types of interneurons, and we have found, surprisingly, that both types can be clonally related to excitatory neurons (Figs. 2–4). As with any prospective labelling study carried out in humans, we cannot definitively rule out the possibility that our methodologies influenced progenitor behaviour. However, the observation of clonally related excitatory and inhibitory neurons in all of our experimental contexts provides strong evidence that human cortical progenitors have the capacity to generate both types of neuron.

In this study, we investigated the clonal relationships of cells born from GW15–GW18 human cortical progenitors over a six-week developmental window. A previous study in which newborn neurons were labelled with 5-bromo-2′-deoxyuridine (BrdU)8 quantified BrdU+;DLX2+ cells in organotypic slice culture derived from GW17.5–GW20.5 human cortex. After roughly eight to ten days of BrdU labelling, that work8 did not find a substantial amount of BrdU+;DLX2+ cells from cortical progenitors. Given the longer window of our experiments, these data suggest that the production of interneurons by cortical progenitors does not occur throughout the entire duration of corticogenesis, but instead begins at some point after midgestation. Consistent with this interpretation, six weeks after labelling we observed a higher proportion of inhibitory neurons within clones derived from GW18 progenitors than within those from GW15 progenitors (Supplementary Table 1). Furthermore, extensive chains of migrating interneurons have been observed in the perinatal human cortex33. Thus, production of cortical interneurons from cortical progenitors may extend beyond the period in which excitatory neurons are born. It is also possible that the different experimental systems used in these studies (that is, in vitro cell culture and xenograft here, and ex vivo organotypic slice culture in ref. 8) might contribute to some of the observed incongruencies between these three systems. Future studies aimed at detailing the output of cortical progenitors over a broader developmental period and throughout different cortical regions will help to further elucidate the contribution of this phenomenon to human brain development.

In a companion study34, STICR was used to perform in vivo clonal labelling of embryonic mouse forebrain progenitors, and STICR-labelled cells were analysed postnatally by scRNA-seq. Although both glutamatergic excitatory neurons and cortical GABAergic inhibitory neurons were recovered, they did not occur within the same clone. The lineage relationship that we observe between cortical excitatory and inhibitory neurons herein thus raises new questions regarding the development of the human cerebral cortex. First, what are the implications of a single progenitor producing both excitatory neurons and cortical interneurons? Evolutionary expansion of the primate neocortex has been attributed to the increased proliferative capacity of cortical neural progenitors. Adaptations in cortical progenitor competence to produce both principal types of cortical neuron could help to ensure the appropriate inhibitory/excitatory balance, despite the dramatic increase in the pool of cortical excitatory neurons35. Recent studies have revealed that although the inhibitory/excitatory balance increases from mice to humans, the relative composition of cortical interneuron types remains relatively constant across evolution. Although existing technical limitations prevent us from confidently estimating the precise cellular contributions of cortical progenitors to the mature human brain, future studies that quantify the relative contributions of progenitors from the cortex and ganglionic eminences will be helpful in understanding the cellular basis of how normal human cortical function is achieved.

Our study opens many avenues for future investigation. The molecular mechanisms that regulate the production of locally born cortical inhibitory neurons are unknown at present. In mice, signalling through the Sonic hedgehog pathway36 is required for individual cortical progenitors to undergo a GABAergic ‘switch’ and to generate inhibitory neurons37 that migrate to the olfactory bulb13. Does a similar molecular mechanism govern the production of cortically-derived cortical interneurons in humans? Furthermore, what molecular markers, if any, can distinguish them from cortical interneurons born in the CGE or MGE? Previous studies have found that NR2F1 and NR2F2 are expressed not only in the CGE but also in cortical progenitors9,10. Given the transcriptional similarity of cortically-derived IN.3 cells to CGE-derived interneurons observed here, it is possible that a similar developmental program is used. These are just some of the questions raised by our new understanding of the human cortical lineage, and addressing these will help to further decipher the origins and mechanisms underlying human brain development.

Methods

STICR barcode design

STICR barcode fragment sequences were generated using the Barcode Generator script written by L. Comai and T. Howell (http://comailab.genomecenter.ucdavis.edu/index.php/Barcode_generator), with a sequence length of 15 base pairs (bp) and a minimum hamming distance of 5. Sequences containing the restriction-enzyme sites matching STICR’s multicloning site (MCS) or homopolymer repeats longer than 4 bp were excluded. In total, 3 non-overlapping sets of 500 sequences meeting these design criteria were selected (Supplementary Table 3).

STICR library creation

STICR barcode libraries were created using a modified pSico lentiviral plasmid (Addgene, catalogue number 11578) in which the sequence between the central polypurine tract (cPPT)/central termination sequence (CTS) and 3′ long terminal repeat (LTR) was replaced with a DNA fragment containing the hybrid cytomegalovirus/chicken β-actin promoter (CAG) promoter, EGFP transgene, WPRE, multiple cloning site (MCS), and bGH polyadenylation signal. The MCS consisted of three adjacent pairs of restriction-enzyme sites (EcoRI–BamHI–NheI–XhoI) between which the STICR barcode fragments would be added. Each double-stranded DNA (dsDNA) STICR barcode fragment was synthesized as a pair of single-stranded DNA (ssDNA) oligonucleotides (Genewiz), annealed, and then pooled together with all the other barcode fragments belonging to its set. Each set of barcode fragments were added into the STICR MCS individually over three rounds of restriction-enzyme cloning. After each round of barcode fragment addition, a ‘negative selection’ digestion was performed using a restriction enzyme that targeted the stuffer sequence in the MCS that should have been replaced with a barcode fragment, so that undigested/barcoded molecules were removed from the library. Following each ‘negative selection’, the resulting STICR barcode plasmid library was amplified by transformation into MegaX DH10B electrocompetent Escherichia coli (Thermo, C640003) and grown overnight on LB agar plates at 37 °C. In order to maintain even sequence distribution, transformed bacteria were plated at high density (roughly 75 million colonies per large-format plate), which helped to restrict colony size and make them grow more uniformly. In order to maximize the diversity of barcode sequences, we transformed enough barcoded plasmid to get at least ten times as many colonies as there were potential barcode sequences. Additionally, the STICR plasmid contained a sequencing primer site upstream of the STICR barcode. Each STICR library contained a 3-bp viral ‘index’ immediately downstream of the sequencing primer binding site. The viral index is unique to each library and allowed us to differentiate STICR barcodes from different libraries.

Generation of STICR NGS libraries for modelling

To generate a STICR plasmid library for next-generation sequencing (NGS), we first digested 1 μg of STICR plasmid library with XhoI and then ligated a polymerase chain reaction (PCR) adaptor containing a unique molecular identifier (UMI) to this site (Supplementary Table 4). Ligation products were amplified by PCR using a Q5 Hot Start high fidelity 2× master mix (NEB, catalogue number M0494) using primers targeted to the STICR sequencing primer site and the adaptor sequence (Supplementary Table 4), using the following program: 1, 98 °C, 30 s; 2, 98 °C, 10 s; 3, 62 °C, 10 s; 4, 72 °C, 10 s; 5, repeat steps 2–4 15 times; 6, 72 °C, 2 min; 7, 4 °C, hold. Following PCR amplification, a 0.8–0.6× dual-sided size selection was performed using Ampure XP beads (Beckman Coulter, catalogue number A63881). The resulting libraries were sequenced to the depth of roughly 30 million reads.

To generate a STICR lentiviral library for NGS, we carried out an RNA extraction from 1/20 of the total lentiviral prep using 300 μl Trizol (Thermo Fisher, catalogue number 15596026). After incubating for 5 min, 60 μl chloroform was added, incubated for 3 min, then centrifuged at 12,000g at 4 °C for 15 min. The aqueous phase was extracted and mixed with an equal volume of 100% ethanol, then loaded onto a Zymo Direct-zol RNA Microprep (Zymo, R2061) column. The Zymo protocol was followed from there to bind and rinse the RNA, which was eluted in a final volume of 7 μl. Complementary DNA (cDNA) was generated from 5 μg of template RNA with the SuperScript IV kit (Thermo Fisher, catalogue number 18090010) using 1 μl of 2 μM STICR viral library reverse transcription (RT) primer (Supplementary Table 4). To add a UMI and a primer-binding handle to the individual cDNA molecules, one-cycle PCR was performed with cDNA and STICR viral library cDNA UMI primer (Supplementary Table 4) with 25 μl Q5 high fidelity 2× MasterMix (NEB, catalogue number M0492S), 2.5 μl primer (10 μM), 2.5 μL H2O, and 20 μl cDNA with the following program: 1, 98 °C, 40 s; 2, 62 °C, 20 s; 3, 72 °C, 2 min; 4, 4 °C, hold. Primers were removed with a left-sided 0.8× SPRISelect cleanup (Beckman Coulter, catalogue number B23318). Finally, cDNA was amplified using the same methods as for the lentiviral library above, and the library was sequenced to a depth of roughly 100 million reads.

Diversity and collision modelling of STICR libraries

STICR barcode sequences were extracted from fastq files with custom scripts that removed PCR duplicate reads using the UMI (for a general description, see the section ‘scRNA-seq analysis and STICR barcode analysis’ below). As it is prohibitively expensive to sequence high-diversity libraries to saturation, we extrapolated the total number of unique STICR barcodes using the Preseq38 command lc_extrap using default settings. Together with the measured relative barcode abundances, we used the extrapolated STICR barcode library size to model barcode collisions with the R (v4.0.1) programming language. Using base R functions, we simulated the labelling of a starting population of cells with a range of sizes from 101 to 106 and repeated each simulation 20,000 times. We then quantified the mean number of unique barcodes chosen for each starting cell population size. The difference between the starting cell population size and the number of unique barcodes present represented the number of collisions that happened at that population size.

Cell lines

We used NIH/3T3 cells (ATCC) and Lenti-X HEK293T cells (Takara Bio) here; we did not test them for mycoplasma or authenticate them.

Lentivirus production

STICR lentivirus was produced using a third-generation lentivirus packaging system: pMDLg/pRRE (Addgene, catalogue number 12251), pRSV-Rev (Addgene, 12253) and VSVG envelope (Addgene, 12259). Plasmids were transfected into Lenti-X HEK293T cells (Takara Bio, 632180) using jetPRIME (Polyplus, 114-15). In order to improve the viral titre, which is reduced in part because of the reverse orientation of the STICR EGFP-barcode transcript relative to the external lentiviral promoter, we also co-transfected pcDNA3.1 puro Nodamura B2 plasmid (Addgene, 17228) along with the other plasmids. Lenti-X 293T cells were grown and transfected in Dulbecco’s modified Eagle’s medium (DMEM) (Fisher, MT10017CV) supplemented with 10% fetal bovine serum (FBS) (Hyclone, SH30071.03) and 1% penicillin streptomycin (Fisher, 15070063). Twenty-four hours after transfection, media was replaced with Ultraculture media (Lonza, BE12-725F) supplemented with sodium pyruvate (final concentration 0.11 mg ml−1, Sigma P2256-25G) and sodium butyrate (0.005 M final molarity, Sigma B5887-1G) and penicillin streptomycin. Seventy-two hours after transfection, media were collected, passed through a 0.45 μM filter (Corning, 431220), and then ultracentrifuged at 22,000g for 2 h. Pellets were resuspended in 100 μl of sterile phosphate-buffered saline (PBS) (Thermo, 14190250) overnight at 4 °C and then aliquoted and stored at −80 °C.

Tissue procurement and STICR transduction

Deidentified tissue samples were collected with previous patient consent in strict observance of the legal and institutional ethical regulations. Protocols were approved by the Human Gamete, Embryo, and Stem Cell Research Committee (Institutional Review Board) at the University of California, San Francisco. In order to visualize tissue for microdissection, samples were embedded in 3% low-melting-point agarose (Fisher, catalogue number BP165-25) and then cut into 300-μm sections perpendicular to the ventricle on a Leica VT1200S vibrating blade microtome in oxygenated artificial cerebrospinal fluid containing 125 mM NaCl, 2.5 mM KCl, 1 mM MgCl2, 1 mM CaCl2 and 1.25 mM NaH2PO4. The germinal zone was then isolated by microdissection using a scalpel and fine forceps under a Leica MZ10F dissecting microscope. In order to dissociate cells into a single-cell suspension, microdissected tissue was incubated in 200 μl of 0.25% trypsin (Fisher, catalogue number 25200056) and 2,000 units ml−1 of DNase I (NEB, M0303) for 20 min at 37 °C and then gently mechanically triturated with a 1,000 μl pipetteman 10 times. We then added 800 μl of DMEM supplemented with 10% FBS to the sample to neutralize the papain, and removed the trypsin/DNase/FBS solution by centrifuging the sample for 5 min at 300g. The sample was rinsed in Dulbecco’s PBS (Thermo, 14190250) and then centrifuged again for 5 min at 300g to remove the rinse.

Cells were then immediately resuspended in NES media (defined in ref. 39), supplemented with ROCK inhibitor (Tocris, 1254) at a final concentration of 10 μM to reduce cell death and 1% penicillin streptomycin (Fisher, 15070063), and then plated on a 24-well tissue culture dish coated with 0.01% poly-l-ornithine (Sigma, P4957), 5 μg ml−1 laminin (Invitrogen, 23017-015) and  μg ml−1 fibronectin (Corning, 354008) at a density of 500,000 cells per well. STICR lenitivirus was added to culture media at a roughly 1:250 to 1:500 dilution so that approximately 30% of cells were infected. After 24 h, virus-containing medium was removed and replaced with fresh medium. Seventy-two hours after infection, cultures were dissociated using papain (Worthington, LK003163), and EGFP+ cells were isolated by FACS and then used in a barnyard assay or in vitro culture assay. In order to maximize cell viability in the in vivo mouse transplantation assay, EGFP+ cells were not FACS-isolated before transplantation, but instead the entire culture (containing both EGFP− and EGFP+ cells) was used. See below for descriptions of each assay. In an effort to label subclones, the GW15 replicate-2 in vitro sample was initially infected with a STICR viral library derived from a fully barcoded STICR plasmid encoding a GFP-T2a-TVA transgene and a viral index ‘E’ and then plated on mouse astrocytes as above. This culture was infected two and four weeks later with EnvA-pseudotyped (Addgene, 74420) STICR libraries with viral index ‘1’ (two weeks) and ‘3’ (four weeks) with a 1:100 dilution of virus:medium. During analysis, we found that labelling with EnvA-pseudotyped libraries was poor, so STICR barcodes with a ‘1’ or ‘3’ index were not considered during analysis. Therefore, clonal analysis of GW15 replicate 2 was carried out using only the initial ‘E’ index STICR barcodes transduced six weeks before analysis, as for all the other in vitro libraries.

Barnyard experiment

To confirm that transcribed STICR barcodes can be accurately recovered using scRNA-seq, we carried out a ‘barnyard experiment’ in which we infected separate cultures of human cortical cells (GW18 sample) and mouse 3T3 cells (ATCC) with different STICR libraries. These libraries could be distinguished from each other by a constant sequence unique to each library (‘viral index’) (Fig. 1a). After three days, we dissociated cultures with papain and FACS-isolated EGFP+ cells. EGFP+ cells from both species were then mixed together and processed using a Chromium Single Cell 3ʹ GEM, Library & Gel Bead Kit v3 (10X Genomics, PN-1000075). Following sequencing (see below), transcript libraries were aligned with Cell Ranger (version 3.0.2) to a hybrid mouse/human genome and droplets were determined to be either a mouse cell, a human cell, or a multiplet. Quantification of recovered STICR viral index (see below) for mouse, human and multiplate droplets is included in Extended Data Fig. 1f.

In vitro culture assay

Long-term in vitro experiments were performed using an astrocyte co-culture system. Primary mouse cortical astrocytes were isolated from CD-1 mice at postnatal day 1 and cultured as described40, but with additional subdissection to remove the subventricular zone surrounding the lateral ventricles. Astrocytes were plated at a density of 400,000 cells per 3.5 cm2 in 12-well cell-culture plates containing DMEM (Fisher, MT10017CV) supplemented with 10% FBS (Hyclone, SH30071.03) and 1% penicillin streptomycin (Fisher, 15070063) 3–5 days before the addition of human cells.

For each experiment, roughly 1,000–2,000 STICR-labelled (EGFP+) cells were added to a 12-well plate already containing mouse cortical astrocytes and cultured in DMEM, 1% B-27 supplement (Invitrogen, 12587-010), 1% N-2 supplement (Invitrogen, 17501-048) and 1% penicillin streptomycin. Cell culture medium was half-changed every three to four days for six weeks. Cultures were then dissociated into single-cell solution using papain, and EGFP+ cells were isolated by FACS. Following FACS isolation, EGFP+ cells were concentrated by centrifugation (300g for 10 min), and prepared for scRNA-seq.

PTPRZ1-based FACS

For enrichment of progenitors before xenograft transplantation, we used an adaptation of the protocol in ref. 41 in order to isolate cells that express the cell-surface protein PTPRZ1, which is enriched on the surface of cortical progenitors15,42. Cortical tissue was dissociated to a single-cell suspension as above, then resuspended in a solution of 1% bovine serum albumin (BSA) (Sigma Aldrich, A7979-50ML) and 0.1% glucose in Hanks’ balanced salt solution (HBSS) (Life Technologies, 14175-095) for staining. Cells were incubated with mouse anti-PTPRZ1 primary antibody (Santa Cruz Biotechnology, sc-33664) at 1:50 dilution for 20 min on ice, washed with HBSS/BSA/glucose, incubated in goat anti-mouse IgM 488 secondary antibody (Thermo Fisher Scientific, A-21042) at 1:500 dilution for 20 min on ice, washed, and resuspended in HBSS/BSA/glucose. PTPRZ1+ cells were then isolated via FACS, plated, labelled with STICR, and cultured as above. Representative FACS plots are shown in Extended Data Fig. 9a.

Xenograft transplantation assay

Mouse transplantation assays were performed in CB17.Cg-PrkdcscidLystbg-J/Crl mice (Envigo) at postnatal day 3–5. STICR-labelled cultures were dissociated with papain, centrifuged at 300g for 5 min, rinsed once with DPBS, and then resuspended in ice-cold L15 media (Fisher, 11-415-064) with 180 Kunitz per millilitre of DNase (Fisher, 50-100-3290). Following anaesthetization, 100 nl of cell mixture (roughly 40,000 to 80,000 cells) was injected through a bevelled glass needle using a stereotactic rig into L −1, A 2.5, D −0.8 mm from lamda. At least five mice were injected with cells from each human specimen. Both male and female mice were used. In order to minimize clumping of xenograft cells, 20 mM EGTA (Sigma, E4378) was added to the cell mixture for GW17 replicate-1 and replicate-2 samples as well as GW15 replicate 3.

After 6 weeks, mice were euthanized and one brain from each set of transplantations was cut into 1-mm coronal sections using a brain mould (Stoelting, 51386). EGFP+ regions of the cortex were dissected from slices using a fluorescent dissecting scope and then dissociated into a single-cell solution using papain. EGFP+ cells were then isolated by FACS, concentrated by centrifugation (300g for 10 min), and prepared for scRNA-seq. Mice were housed in a barrier facility with a 12-h light/12-h dark cycle and temperature and humidity control (70 °F, 50% rack humidity). All protocols and procedures followed the guidelines of the Laboratory Animal Resource Center at the University of California, San Francisco, and were conducted with approval of the Institutional Animal Care and Use Committee (IACUC).

IHC of in vitro cultures and xenografts

At the experimental endpoint, transcardiac perfusion of sterile PBS followed by 4% PFA (Fisher, 50-980-487) was used to rinse and then fix the specimens. Brains were dissected out and drop-fixed overnight in 4% PFA at 4 °C. Brains used for cryosections (Fig. 3 and Extended Data Fig. 7) were then cryopreserved in a 1:1 solution of OCT (VWR, 25608-930) and 30% sucrose, embedded in cryomolds containing the same solution, frozen on dry ice, and stored at −80 °C. Brains were then cryosectioned at 12 μm onto glass slides and stored at −80 °C. Blocking and permeabilization were performed using a blocking solution consisting of 10% normal donkey serum, 1% Triton X-100 and 0.2% gelatin in PBS for 1 h. Primary and secondary antibodies were diluted and incubated in this same blocking solution. Cryosections were incubated with primary antibodies at 4 °C overnight, washed 3 × 10 min with washing buffer (0.1% Triton X-100 in PBS), incubated with secondary antibodies for 2 h at room temperature, and washed 3 × 10 min with washing buffer; coverslips (Azer Scientific, 1152460) were then mounted using Prolong Gold Antifade reagent (Invitrogen, P36930).

Brains used for morphological analysis of GABAergic cells (Fig. 4 and Extended Data Figs. 9, 10) were fixed as above but stored in PBS at 4 °C. Brains were then sectioned on a Leica VT1000 S vibrating blade microtome to 40 μm and slices were stored in PBS. Slices were incubated in blocking solution composed of 10% normal donkey serum and 0.1% Triton-X in PBS at room temperature for 2 h and then incubated in primary antibodies diluted in blocking solution overnight at 4 °C, washed 5 × 30 min in 0.1% PBST, incubated in secondary antibodies in blocking solution overnight at 4 °C, washed 5 × 30 min in 0.1% PBST, mounted on glass slides, and coverslipped as above using ProLong Gold Antifade reagent.

In vitro cultures comprising human cortical cells co-cultured with mouse astrocytes were prepared as described above, but without FACS isolation of EGFP+ STICR-labelled cells, and plated on eight-well chamber slides (Thermo Scientific, 154534PK). After 6 weeks, cultures were fixed with 4% PFA for 1 h at 4 °C. Cultures were then washed three times and stored in PBS. Immunohistochemistry was performed as described for the 40-μm mouse brain sections but with 10-min washes. Slides were coverslipped with ProLong Gold Antifade reagent.

The antibodies used include: chicken anti-GFP (Aves, GFP-1020; 1:1,000), mouse anti-human nuclear antigen (Novus, NBP2-34342; 1:100), rabbit anti-GABA (Millipore Sigma, A2052-100ul; 1:250), rabbit anti-NEUROD2 (Abcam, ab104430; 1:500), guinea pig anti-DCS (Millipore Sigma, AB2253; 1:200), rabbit anti-GFAP (abcam, ab7260; 1:1,500), rabbit anti-SOX9 (Abcam, ab104430; 1:250) and mouse anti-OLIG2 (Millipore Sigma, MABN50; 1:200). Secondary antibodies used include AlexaFluor anti-chicken 488 (Jackson Immunoresearch 703-545-155; 1:500), anti-mouse 488 (ThermoFisher A-21042; 1:500), anti-rabbit 594 (ThermoFisher A-21207; 1:500), anti-guinea pig 647 (Jackson Immunoresearch 706-605-148; 1:500), anti-mouse IgG1 488 (ThermoFisher A-21121; 1:500), and anti-mouse IgG2a 647 (ThermoFisher A-21241; 1:500).

Confocal imaging was performed using a Leica SP8 confocal microscope with either a 10× or a 20× air objective; a 2-μm optical z-step was used for all images. Images were processed using ImageJ/Fiji. For quantification of the major cell types in xenografted mice (Fig. 3 and Extended Data Fig. 7), tilescans of the transplanted region were z-projected with average intensity; channel intensity was normalized across images; and cells expressing EGFP and/or human nuclear antigen in addition to cell-type markers were counted manually using the CellCounter plugin for ImageJ/Fiji.

GABA+ STICR-labelled cells in the cortex of host mice (Fig. 4) were quantified by imaging four 40-μm-thick brain slices derived from within 400 μm of the transplantation site of each animal. Two mice for each of the two GW17 specimens were analysed in total. GFP+ cells in the cortex were counted using the CellCounter plugin for ImageJ/Fiji. The relative laminar positions of GABA+/GFP+ double-positive cells were measured using the Measure tool from ImageJ/Fiji to draw a line from the top of the corpus callosum straight up to the pial surface through the soma of each cell. The relative location of the cell’s soma to the top of the corpus callosum was then divided by the total length of the line drawn from the corpus callosum to pial surface.

scRNA-seq library preparation

Preparation of scRNA-seq libraries was carried out using a Chromium Single Cell 3ʹ GEM, Library & Gel Bead Kit v3 (10x Genomics, #PN-1000075). Libraries were prepared according to the manufacturer’s protocol.

STICR barcode library recovery

STICR barcodes were subamplified from each 10X cDNA library using the Q5 Hot Start High Fidelity 2× master mix (NEB, M0494). In brief, 10 μl of cDNA was used as template in a 50 μl PCR reaction containing STICR barcode read 1 and 2 primers (0.5 μM, each) targeting the region immediately upstream of the STICR viral index/barcode as well as the partial Illumina Read1 sequence added during cDNA library preparation (Supplementary Table 4), using the following program: 1, 98 °C, 30 s; 2, 98 °C, 10 s; 3, 62 °C, 20 s; 4, 72 °C, 10 s; 5, repeat steps 2–4 11 times; 6, 72 °C, 2 min; 7, 4 °C, hold. Following PCR amplification, a 0.8–0.6 dual-sided size selection was performed using Ampure XP beads (Beckman Coulter, #A63881).

Sequencing

10X transcriptomic libraries and STICR barcode libraries were sequenced using Illumina Novaseq 6000 or Illumina HIseq 4000 machines. 10x transcriptomic libraries were sequenced to the average depth of approximately 50,000 reads per cell (2,200 genes per cell). STICR barcode libraries from lineage-tracing experiments were sequenced to a depth of roughly 30 million reads per library. STICR plasmid libraries for diversity and collision modelling were sequenced to a depth of approximately 30 million reads per library, while a single STICR lentiviral library was sequenced to a depth of approximately 100 million reads.

scRNA-seq analysis and STICR barcode analysis

10X transcriptomic libraries were aligned to the hg38 genome using CellRanger (version 3.0.2). Aligned cell/transcript counts were processed by Seurat43 (version 3.2.0.9014 for initial in vitro cultures and version 4.0 for subsequent xenografted cultures and integration of xenograft data into in vitro data) to remove cells containing fewer than 1,000 genes, fewer than 1,250 transcripts, or a high abundance of mitochondrial reads (greater than 7% of total transcripts). Cells passing these thresholds were then processed with Cellbender44(version 0.1) in order to identify and remove background reads and instances of barcode swapping. We then identified and remove multiplets that arose during cell capture using Solo45(version 0.1). Additionally, xenograft libraries were aligned to a chimaeric hg38/mm10 genome using CellRanger (version 3.0.2) in order to identify potential cross-species multiplets; cells identified as such were removed from analysis. Libraries were integrated using Seurat’s SCTransform and FindIntegrationAnchors functions, to identify integration features. First, in vitro cultured STICR experiments were integrated together. Subsequently, transcriptomic libraries form xenograft experiments were integrated with in vitro cells using integration anchors identified from in vitro cultures. Leiden cell clustering, pseudotime analysis and data visualizations (that is, creation of UMAP images) were performed using Monocle346 (version 0.2.1.9). Pseudotime analysis of cells of the inhibitory-neuron trajectory was performed by setting the root node within mitotic inhibitory neuron clusters. IN.1, IN.2 and IN.3 trajectories were defined as major branches of the principal graph that led to distinct sets of clusters. Within subclustered inhibitory neurons, cluster 17 appeared to be at the beginning of both IN.2 and IN.3 trajectory cells in pseudotime analysis and was thus termed ‘IN.early’.

Differential gene expression analysis was conducted using Seurat FindMarkers/FindAllMarkers functions. With the exception of cluster 34 from the in vitro culture data (Fig. 1c), we identified marker-gene expression consistent with previously described cell types. Cluster 34 expressed markers of multiple cell types and did not show strong cluster correlation with reference data (Extended Data Fig. 2d). Thus, we refer to cells in this cluster as ‘Unknown’ in Supplementary Table 1.

Iterative subclustering of excitatory neuron and EOMES+ IPC trajectory cells from in vitro STICR cultures revealed ‘upper-like’ (subclusters 3, 4, 7, 9), ‘deep-like’ (subclusters 1, 5), ‘newborn excitatory neurons’ (subcluster 6), ‘EOMES+ IPCs’ (subcluster 2), and ‘EN.other’ (subcluster 8) subgroups based on gene expression. The EN.other group did not show a strong correlation with excitatory neurons, and thus cells in this category were removed from further clonal analysis. Within xenografted excitatory-neuron and EOMES+ IPC trajectory cells, iterative subclustering revealed ‘excitatory neuron’ and ‘EOMES.IPC’ subclusters.

STICR barcode analysis was performed using custom scripts. First, BBMap (BBMap, Bushnell B; sourceforge.net/projects/bbmap/) was used to remove low-quality reads and then extract reads containing STICR barcode sequences. Then, BBMap was used to extract individual STICR barcode fragments that were then aligned to our predefined fragment reference sets using Bowtie47 (version 5.2.1), allowing for up to two mismatches per fragment. Aligned STICR barcodes were compiled into a file containing their corresponding 10x cell barcode and 10x UMI sequences using Awk. Finally, UMI-tools48 (version 0.5.1) was used to remove duplicate STICR barcode/cell barcode (CBC) reads by UMI, allowing for 1-bp mismatches in the UMI. STICR barcodes/CBC) pairings with at least five distinct UMIs were retained. Cells with a single STICR barcode meeting this criterion were retained for clonal analysis. Possible instances of STICR barcode superinfection (multiple STICR barcodes per starting progenitor) were identified by calculating jaccard similarity indexes of all STICR barcodes pairings found to co-occur within a single cell. Those pairings with a jaccard similarity index of 0.55 or greater that occurred in 10 cells or more were considered to be a valid superinfection clone and retained for clonal analysis. Cells that contained multiple STICR barcodes that were not determined to be valid superinfections were further analysed for the relative abundance of individual STICR barcodes. Cells that contained a ‘dominant’ STICR barcode with five or more times the number of barcode counts compared with the next most abundant STICR barcode, as determined by UMI, were retained and assigned that dominant barcode. Those cells that did not contain barcodes meeting these criteria were not considered for clonal analysis.

Validation of marker genes with Allen Brainspan data

Expression data (Z-score) for IN.2 and IN.3 marker genes were downloaded from the Allen Brainspan database (https://www.brainspan.org). Biological samples from the cortex, basal ganglia and rostral migratory stream/olfactory bulb were retained for further analysis. Hierarchical clustering was performed using the R package pheatmap(v.1.0.12), with the ward.D clustering method. For visualization purposes, the data ranges were divided into 20 quantiles.

Transcriptional cluster correlation

Marker genes from each cluster in the Nowakowski primary human brain reference atlas14 were calculated using Seurat FindAllMarkers, restricting genes to those present in at least 25% of cells of that cluster. The top 100 marker genes by fold-expression for each cluster were then retained for further analysis. The average expression for each cluster’s top marker genes was calculated for each cluster in the Nowakowski reference atlas as well as in each cluster within the STICR transcriptional datasets. We then calculated pairwise Pearson correlations between every reference and STICR cluster, and depicted the result using a heatmap.

Statistics and reproducibility

Images shown in figures are representative of results found in multiple replicates: Fig. 3b (two specimens each xenografted into three mice), Fig. 4b, c (two specimens each xenografted into two mice), Extended Data Fig. 4d, e (four specimens), and Extended Data Figs. 9, 10 (two specimens each xenografted into two mice).

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this paper.