Research organism

elife

eLife

2050-084X

eLife Sciences Publications, Ltd

101163

10.7554/eLife.101163

10.7554/eLife.101163.3

version of record

Research Article

Neuroscience

Human brain ancestral barcodes

Shibata

Darryl

https://orcid.org/0000-0002-4567-1639

dshibata@usc.edu11

https://ror.org/03taz7m60

Department of Pathology, University of Southern California, Keck School of Medicine

Los Angeles

United States

Konopka

Genevieve

Reviewing Editor

https://ror.org/05byvp690

University of Texas Southwestern Medical Center

United StatesBehrens

Timothy E

Senior Editor

https://ror.org/052gg0110

University of Oxford

United Kingdom

10062025

RP101163

14072024

This manuscript was published as a preprint.

17072024

This manuscript was published as a reviewed preprint.

02102024

The reviewed preprint was revised.

27122024

2024

Shibata

http://creativecommons.org/licenses/by/4.0/

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Dynamic CpG methylation ‘barcodes’ were read from 15,000–21,000 single cells from three human male brains. To overcome sparse sequencing coverage, the barcode had ~31,000 rapidly fluctuating X-chromosome CpG sites (fCpGs), with at least 500 covered sites per cell and at least 30 common sites between cell pairs (average of ~48). Barcodes appear to start methylated and record mitotic ages because excitatory neurons and glial cells that emerge later in development were less methylated. Barcodes are different between most cells, with average pairwise differences (PWDs) of ~0.5 between cells. About 10 cell pairs per million were more closely related with PWDs <0.05. Barcodes appear to record ancestry and reconstruct trees where more related cells had similar phenotypes, albeit some pairs had phenotypic differences. Inhibitory neurons showed more evidence of tangential migration than excitatory neurons, with related cells in different cortical regions. fCpG barcodes become polymorphic during development and can distinguish between thousands of human cells.

lineage-tracingneurondevelopment

Research organismHuman

http://dx.doi.org/10.13039/100000002

National Institutes of Health

P01CA196569

Shibata

Darryl

http://dx.doi.org/10.13039/100000002

National Institutes of Health

CA271237

Shibata

Darryl

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Author impact statement

A dynamic DNA methylation 'barcode' on the X chromosome can distinguish thousands of individual adult male neurons and may record aspects of their development.

publishing-route

prc

Introduction

Cell lineages outline tissue development. Complete fate maps are possible by direct observation for small organisms such as C. elegans, but various elegant experimental fate markers are employed for larger tissues and longer time intervals (McKenna and Gagnon, 2019). For human tissues, prior experimental manipulations are impractical, and genomic alterations are employed. Somatic mutations mark subclones and their fates can be reconstructed with DNA sequencing. Recent advances in single-cell technologies potentially allow fate map reconstruction at single-cell resolution.

Here, we show how fCpG DNA methylation (Gabbutt et al., 2022) could be used as dynamic barcodes to study human brain development using single-cell epigenomes annotated with their locations and phenotypes (Tian et al., 2023). DNA methylation patterns are usually copied between cell divisions, but replication errors are much higher compared to base replication, allowing for more differences between daughter cells. DNA methylation modulates expression and their patterns can be used to infer cell phenotypes (Tian et al., 2023; Loyfer et al., 2023), but most fCpG sites are present outside of genes or in unexpressed genes. Criteria for our fCpG barcode are as follows: (1) a defined initial pattern in a progenitor cell; (2) polymorphic changes upon cell division; (3) adequate polymorphism to distinguish between most cells; and (4) capability to record ancestry.

The brain has several features that facilitate barcode development and validation. Foremost, there is extensive single-cell methylation data, with thousands of cells annotated by locations and phenotypes (Tian et al., 2023). Although billions of cells are present in an adult brain, lineage trees are compact because growth is largely prenatal. The brain also allows for serial ‘stopwatch’ barcode sampling because development roughly follows a caudal to rostral pattern, and groups of neurons characteristically stop dividing and differentiate at different times and locations (Stiles and Jernigan, 2010). Brainstem neurons emerge early (Fan et al., 2020) and their barcodes should most resemble the initial progenitor state, whereas the stopwatch runs longer for excitatory neurons that appear later in development. To facilitate presentation, barcode performance is summarized as follows: The brain fCpG barcode initializes as predominately methylated in the progenitor cell and becomes polymorphic with more diverse barcodes in excitatory neurons that emerge later in development. The barcode becomes sufficiently polymorphic to uniquely distinguish between most sampled brain cells, and barcoded cells organize into lineage trees.

ResultsfCpG barcode identification

Barcode development was limited by the sparse single-cell data (Tian et al., 2023), with <5% of CpG sites sequenced, often with only a single read. Sparse coverage was mitigated with X-chromosome fCpG sites because only a single read can infer a binary (0, 1) state in male individuals. Autosomal CpG sites require at least 2 reads to infer three possible states (0, 0.5, 1). The X-chromosome also simplifies the identification of polymorphic fCpGs because many neurons have different binary states if average methylation is between 0.25 and 0.75 in bulk WGBS adult male neurons reference data (Loyfer et al., 2023).

CpG sites (N ~116,000), with average methylation between 0.25 and 0.75 in bulk neurons from seven males (Loyfer et al., 2023), were further filtered by discarding more stable CpG sites with average methylation less than 0.2 or more than 0.8 for all cells, inhibitory neurons, and excitatory neurons in brain H02. The ~79,000 CpGs were further filtered to remove sites with average methylation less than 0.3 or greater than 0.7 in brain H01, and ~31,000 fCpG sites were used for analysis.

fCpG site methylation appears neutral because they are predominately intergenic, with 16% within genes or promoters (Supplementary file 1). Epigenomes from 15,434–21,836 cells were downloaded from three male brains with a general criterion of allc.tsv.gz file sizes 90 mb or larger (Table 1). Neurons were preferentially sampled, whereas glial cells were sometimes excluded (Supplementary file 2). Analyzed cells had at least 500 fCpGs (average ~1100), with pairwise distances (PWDs) calculated between cell pairs when at least 30 fCpGs were comparable (average ~48 fCpGs per cell pair). Each cell, annotated by its provided phenotype and location, is characterized by its fCpG methylation level and its PWDs from other cells. A PWD of 0 is a perfect match and 0.5 indicates randomization. fCpG methylation was variable between cells with averages of ~58% for all three brains (Figure 1A). The 73–197 million possible cell pair comparisons revealed polymorphic barcodes with average PWDs of ~0.47 between cells (Figure 1B).

Table 1.Brain data.

brain	age	cells	ave fCpG per cell	ave Meth per cell	ave PWD between cell pairs	cell pairs per million	fCpG per pair	closely related pairs (PWD <0.05)	closely related pairs per million	fCpG per nearest neighbor pair*
H01	42y	21,836	1170	0.58	0.47	197	51	1385	7.0	35
H02	29y	16,161	1128	0.58	0.47	99	48	743	7.5	34
H04	58y	15,434	1060	0.58	0.47	73	45	1078	14.8	35
between brains
H02-H01					0.47	281	49	785	2.8	35
H02-H04					0.47	178	46	653	3.7	34
H04-H01					0.47	254	47	943	3.7	35

Figure 1.Fluctuating CpG (fCpG) barcode methylation 0.5.

(A) Barcode methylation was variable between cells with averages ~50% (B) Most cells had different barcodes with average pairwise differences (PWDs) ~0.5.

fCpG barcodes initialize methylated and change with cell division

fCpG barcode patterns were similar between the brains, and data are presented for H01, with H02 and H04 presented in supplemental figures. Methylation was variable between cells of the same type and average methylation was highest in the pons (PN) and thalamus (THM), intermediate for other inhibitory neurons, and lowest for excitatory neurons, glial cells, and cerebellar cells (Figure 2A). Outer layer cortical excitatory neurons (L2_3) that are made later during development were less methylated than inner cortical excitatory neurons (L4_6) that appear earlier. Predominantly methylated individual fCpG sites were common in subcortical neurons (PN, THM), less frequent in other inhibitory neurons, and rare in excitatory neurons (Figure 2B).

Figure 2.Higher fluctuating CpG (fCpG) barcode methylation in earlier emerging cells PN CB SubCt Foxp2 Lamp5 Vip Pvalb excitatory THM MSN Chd7 Sncg Sst AMY Hip-mc CA DG L4_6 L23 glial cell barcode methylation ASC OPC ODC ganglionic eminences mature.

(A) Average barcode methylation was higher in the brainstem and inhibitory neurons. Barcode methylation was lower for excitatory neurons, cerebellar, and glial cells. Notably, average methylation was lower for outer cortical (L2_3) compared to earlier appearing inner (L4_6) cortical excitatory neurons. Abbreviations are as in reference 3, with L2_3 all outer and L4_6 all inner cortical excitatory neurons, and NN are non-neuronal cells other than ASC, OPC, and ODC. (B) Most fCpGs appear to start methylated in a progenitor because nearly all individual fCpGs are methylated in inhibitory neurons in the subcortex (PN, THM, MSN). Many fCpGs in inhibitory neurons (pvalb, sst) are still predominately methylated. Few fCpGs in excitatory neurons that differentiate later in development are fully methylated. Glial cells that also emerge late in development, and hippocampal cells that may divide postnatally had variable methylation with both highly methylated and unmethylated fCpGs. (C) Barcodes are assumed to become fixed when their cells stop dividing and differentiate. Therefore, barcode methylation levels can indicate when neurons emerge during development, and can be correlated with a cartoon of physical caudal to rostral brain development. The x-axis indicates the barcode methylation of individual cells and is assumed to roughly correlate with calendar time. The y-axis indicates the cumulative proportion of cells of each type present at each methylation level. A value of 0 indicates that cells of given type are not yet present and a value of 1 indicates the adult content of this cell type has been reached. At the start of development, inhibitory neurons (PN) in the pons with highly methylated barcodes appear first. More inhibitory neurons, made in the ganglionic eminences, appear and reach their final adult contents before many cortical excitatory neurons and glial cells appear. Notably, barcode methylation indicates many lower cortical layer neurons appear earlier in life relative to outer cortical neurons that reach adult levels late in development. Brain contents inferred by adult barcodes may differ from actual neonatal brains because neurons that die during development are not sampled in adult brains.

Figure 2—figure supplement 1.H02 data.

(A) Barcode methylation for different cell types ASC OPC ODC mature (B) Barcode methylation versus final adult brain content indicates that inhibitory neurons appear first and reach their adult levels before excitatory or glial cells.

Figure 2—figure supplement 2.H04 data.

(A) Barcode methylation for different cell types (B) Barcode methylation versus final adult brain content indicates that inhibitory neurons appear first and reach their adult levels before excitatory or glial cells.

The methylation hierarchy is consistent with a barcode initialized with predominately methylated fCpGs in a brain progenitor cell. Barcodes become progressively demethylated and are fixed when their cells stop dividing and differentiate, which occurs at different times and places during brain development. Barcodes for each cell type had a range of methylation (Figure 2A), consistent with synchronous development rather than a strict stepwise process. Very simplistically, the hindbrain with mature neurons (Fan et al., 2020) forms early in development, followed by inhibitory neurons in the ganglionic eminences, and then excitatory neurons and glial cells in the cortex. Barcode methylation follows this temporal development and reconstructs when specific neuron types start to appear and reach their adult contents (Figure 2C). For example, after barcodes change from ~100 to~70% methylated, most adult brainstem (PN) and inhibitory neurons are present but excitatory neurons are fewer, with very few adult outer layer (L2_3) neurons. Outer excitatory and glial progenitor cells are present (Eze et al., 2021), but their barcodes continue to demethylate until they stop dividing and differentiate later in development. This stopwatch-like pattern, with more demethylated barcodes in later appearing cell types, was present in all three adult brains.

fCpG barcodes are polymorphic

A progenitor cell barcode is assumed to become increasingly polymorphic with subsequent divisions. This pattern was observed, with average barcode PWDs lowest in brainstem cells (PN), intermediate between other inhibitory neurons, and highest for excitatory neurons (Figure 3A). Most cells had different barcodes, with an overall average PWD of ~0.47 (Figure 1B). Cells of the same phenotype were more similar with lower average PWDs (Figure 3A and B), suggesting they are more related to each other and have common progenitors.

Figure 3.Related cell pairs.

(A) Most cells had different barcodes with average pairwise differences (PWDs) between cell pairs of ~0.5. Cell pairs of the same phenotype had different barcodes but were on average more related to each other. (B) Heatmap showing that cells of the same phenotype are more related. (C) Cells that emerge early in development are more related and more methylated. Closely related nearest neighbors (PWD <0.05) are numerically more common for more methylated cell types.

Figure 3—figure supplement 1.H02 data.

(A) Pairwise differences (PWDs) between cells of the same type. (B) PWDs between cell types (order of cell types is the same as in ‘A’).

Figure 3—figure supplement 2.H04 data.

(A) Pairwise differences (PWDs) between cells of the same type.(B) PWDs between cell types (order of cell types is the same as in ‘A’).

Figure 3—figure supplement 3.Brain tumor fCpG methylation: fCpG sites were matched with brain tumor data from methylation arrays.

Average methylation is displayed for each brain tumor sample. PA=pilocytic astrocytoma, GSE44684, 145 fCpGs covered; MB=medulloblastoma, GSE193646, 434f CpGs covered; MB=medulloblastoma, GSE63669, 883 fCpGs covered; GBM=glioblastoma multiforme, GSE109399, 1,138 fCpGs covered; Gliomas=various gliomas, GSE248471, 1,138 fCpGs covered. Includes both male and female tumors.

The human brain has billions of cells and relatively few cells were sampled from each region. Consistent with sparse sampling, cell pairs with nearly identical barcodes and smaller PWDs (<0.05) were rare. To help distinguish between ancestry and chance, cells within and between brains were compared (Table 1). Closely related cell pairs were ~2.9 times more frequent within a brain (average ~9.8 per million) compared to between brains (average ~3.4 per million). Closely related cells had fewer matching fCpG sites (~35 compared to ~48 for all cell pairs) and were more common early in development when barcodes are more methylated (Figure 3C), indicating that lower barcode complexity favors matching. Overall, fCpG barcodes are sufficiently polymorphic to distinguish between most adult brain cells.

Brain lineage trees

It should be possible to reconstruct human brain development if barcodes record ancestry. fCpG barcodes from ~1000 brain cells with different phenotypes yield trees with a standard phylogeny software that resemble caudal to rostral development (Figure 4). The trees are rooted by a progenitor with a fully methylated barcode, and branches progressively yield brainstem neurons (PN), a subset of excitatory lower (L4_6) neurons, thalamic neurons (THM), inhibitory neurons, cerebellar cells, and glial cells. Excitatory neurons branch last, and hippocampal neurons (CA, DG) that may divide postnatally (Gage, 2002) were at the terminus. Cells are generally grouped by phenotype, with some early appearing excitatory neurons admixed among inhibitory neurons. Similar trees were observed for H02 and H04, albeit with less separation between inhibitory and excitatory neurons for H04 (Figure 4A). Barcode lineage trees are largely consistent with expected sequential neuronal differentiation.

Figure 4.Brain trees.

(A) Barcodes from 960 cells form trees using IQtree (Nguyen et al., 2015) that are rooted by a fully methylated progenitor, and generally follow caudal to rostral brain development, with sequential branching of inhibitory neurons, cerebellar neurons, and excitatory neurons, with hippocampal neurons furthest from the start. Trees are similar between the brains, with H04 inferring less distance between inhibitory and excitatory lineages. The trees illustrate the ability to produce phylogenies with IQtree, but the phylogenies are limited by sparse cell sampling and that barcodes may be similar by chance. The degree of confidence was generally low, with bootstrap branch support typically less than 15%. (B) H01 tree with labeled cell types. Neuron types generally clustered by phenotypes with closely branching excitatory and inhibitory neurons more common earlier in development. (C) H01 tree with labeled cell locations. Related inhibitory and excitatory neurons can be found in different parts of the brain FC = frontal (red), TC = temporal, OC = occipital (blue), PC = parietal, HIP = hippocampus (orange), cere = cerebellum (yellow), SC = subcortical (black). (D) H01 tree with ~2853 cortical excitatory neurons has more evidence of localized radial migration because related neurons are more often found in the same cortical region. Excitatory neurons cluster by subtype, and closely related lower and upper excitatory neurons were still few. (E) H01 tree with ~2847 cortical inhibitory neurons still retains evidence of tangential migration with related neurons scattered throughout the cortex. Inhibitory neurons cluster by subtype with switching between some closely related pairs.

Figure 4—figure supplement 1.H02 data.

(A) Ancestral tree with 1001 cells rooted at a fully methylated progenitor shows sequential branching with excitatory, then brain stem, inhibitory and cerebellar neurons, then glial cells, and finally excitatory neurons with hippocampal neurons at the end. (B) Related cells colocalize for brain stem, cerebellar, and hippocampal neurons. Inhibitory neurons are more scattered. Excitatory neurons are also scattered with some localization within the cortex.

Figure 4—figure supplement 2.H04 data.

(A) Ancestral tree with 1033 cells rooted at a fully methylated progenitor shows sequential branching with excitatory, then brain stem, inhibitory and cerebellar neurons, then glial cells, and finally excitatory neurons with hippocampal neurons at the end. (B) Related cells colocalize for brain stem, cerebellar, and hippocampal neurons. Inhibitory neurons are more scattered. Excitatory neurons are also scattered with some localization within the cortex.

Cell lineage fidelity and cortical migration

Uncertain for mouse and human development is whether inhibitory and excitatory neurons originate from shared or distinct progenitors (Tian et al., 2023; Bandler et al., 2022; Delgado et al., 2022). Barcodes could potentially record neuronal differentiation patterns and lineage fidelity can be quantified by comparing most closely related cells or nearest neighbor cell pairs with PWDs <0.05. The approach remains speculative due to several factors: the absence of direct experimental validation, limited experimental cell sampling, and the possibility that barcode similarities may arise by chance rather than reflecting true biological relationships. Lineage fidelity was high (>90%) for inhibitory neurons (Figure 5A). Excitatory lineage fidelity was slightly lower, indicating that some excitatory and inhibitory neurons may share common progenitors (Delgado et al., 2022). Lineage trees indicate common progenitors are present earlier in development, and excitatory neurons that appear later do not have many closely related inhibitory neighbors (Figure 4B). The barcodes documented the known switching between inhibitory neuron subtypes (Figure 5B). Hindbrain, excitatory, and non-neuronal cells had more subtype lineage fidelity.

Figure 5.Lineage fidelity, migration, and differentiation.

(A) Inhibitory neurons have higher lineage fidelity because nearest neighbor pairs (pairwise difference, PWD <0.05) were nearly always both inhibitory neurons. Excitatory neurons had slightly less lineage fidelity because a nearest neighbor was more often an inhibitory neuron. Data are for all three brains. (B) Nearest neighbor inhibitory neuron pairs often had subtype differences. More lineage fidelity was generally present for subcortical and excitatory neurons. Numbers indicate percent lineage subtype fidelity. (C) Nearest neighbor inhibitory and excitatory neuron pairs showed evidence of tangential migration because they were found in different cortical regions. The data indicate greater evidence of inhibitory neuron tangential migration than for excitatory neurons. However, the extent of migration is uncertain because of sparse sampling and because barcodes may be similar by chance. Data are for all three brains. (D) Nearest neighbor neurons were scattered in the cortex. Numbers indicate percent location fidelity. (NonC = non-cortical location, Paleo = paleocortex).

Figure 5—figure supplement 1.H02 data.

(A) Lineage cell type fidelity between nearest neighbor pairs (pairwise difference, PWD <0.05). Numbers on the right (percentage) indicate how often a closely related cells has the same cell phenotype (B) Location fidelity between nearest neighbor pairs (PWD <0.05). Numbers on the right (percentage) indicate how often a closely related cell is located in the same cortical region. NonC indicates the closely related cell was found outside the cortex.

Figure 5—figure supplement 2.H04 data.

Barcodes can also potentially infer migration because their neurons are annotated by their adult locations. Daughter cells with similar barcodes could be sampled from the same region, or from different regions if migration occurred. Trees (Figure 4C) indicate that most neurons sampled from the brainstem and hippocampal regions are related and localized to their respective regions. Inhibitory neurons were scattered throughout the cortex, consistent with their differentiation in the ganglionic eminences and subsequent tangential migration to the cortex. Nearest neighbor inhibitory cortical neuron pairs were found in the same cortical region ~25% of the time (Figure 5C and D). Nearest neighbor excitatory neuron pairs were also scattered throughout the cortex, but less than inhibitory neurons, and were in the same cortical region ~50% of the time. The barcode data indicate more inhibitory rather than excitatory neuron tangential migration, but the extent of migration is uncertain due to sparse sampling and because barcodes can match by chance.

The poor ability to detect localized excitatory neuron radial cortical migration with ~1000 cell whole brain trees (Figure 4C) may reflect that sparse sampling is unlikely to include multiple neurons from the same small clonal region that originated from common subventricular progenitors (radial unit hypothesis Rakic, 1988). Greater localized excitatory neuron migration was seen when trees were reconstructed with more (~2800) neurons, while inhibitory neurons still showed scattered tangential migration (Figure 4D). Neurons of the same subtype were still more related. Hence, lineage trees appear to increase their resolution with more cells, albeit related lower and upper excitatory neuron pairs were still uncommon, which may reflect the unlikely chance of sampling very small radial clonal units.

Discussion

fCpG barcodes are potential markers of somatic cell ancestry and not cell type classifiers, although cells of the same phenotype are often related because they originate from common progenitors (see new Supplement for fuller discussion). Dynamic barcodes would be useful to study human tissues, but testing their performance is difficult. Ideally, samples obtained at different times would document how they change. The brain facilitates barcode validation because it periodically stores neurons that stop recording at relatively defined times and locations (McKenna and Gagnon, 2019; Finlay and Darlington, 1995). Specific neuron subsets recovered from the adult brain allow for sampling through time and before birth.

This serial sampling strategy facilitated fCpG barcode validation. The barcode appeared to start predominantly methylated in multiple individuals and became sufficiently polymorphic to distinguish between thousands of neurons. Barcode changes appear to represent replication errors because they reconstruct lineage trees roughly consistent with caudal to rostral brain development. Barcode methylation may indicate when different neurons that survive to adulthood appear in the neonatal brain (Figure 2C).

The current barcode indicates that most inhibitory and excitatory neurons have relatively distinct progenitors, consistent with the lineage dendrograms reconstructed with neuron-specific methylation of the same data (Tian et al., 2023). There was also evidence for common inhibitory and excitatory progenitors (Delgado et al., 2022), primarily for earlier emerging excitatory neurons. Tangential migration was also detected, manifested by inhibitory neurons with closely related barcodes in different cortical regions. Tangential excitatory neuron migration was also detected, albeit related excitatory neurons were more localized than inhibitory neurons. Tangential migration is also seen with sequencing studies that find neurons with specific mutations in multiple brain regions (Lodato et al., 2015; Breuss et al., 2022; Chung et al., 2024; Evrony et al., 2015).

fCpGs more efficiently distinguish between cells than mutations due to higher replication error rates. Although average methylation decreases with time, both demethylation and remethylation are likely because fully demethylated neurons were not observed, and balanced fluctuating methylation is inferred in other tissues when CpG sites are ~50% methylated in bulk tissues (Gabbutt et al., 2022). More adult divisions in brain cancers did not saturate the barcode with average fCpG methylation ~50% (Figure 3—figure supplement 3). Fluctuating methylation complicates lineage tracing but backmutations can be modeled for ancestral reconstructions. Lineage resolution could be improved by combining mutations and fCpGs.

Weaknesses of this study include very sparse cell sampling and lack of uniform CpG site comparisons between neurons. Like many human fate marker studies, it is difficult to independently verify accuracy. Interpretation of the barcodes relies on several untested assumptions, including relatively constant error rates between fCpG sites and through aging, neutrality, and a predominately fully methylated start in the progenitor cell. Epigenetic remodeling occurs after progenitors stop dividing (Ciceri et al., 2024), which could erase ancestral barcode information. However, neurons of the same type were both closely and remotely related, indicating that such epigenetic remodeling does not systematically alter the fCpG sites. In addition, fCpG barcodes appear to be relatively stable through aging (new Supplement). Inferred lineage trees (Figure 4) had relatively low statistical support for their branches and are presented to demonstrate that the barcodes are readily organized into trees with a commonly used phylogeny software.

Technical improvements such as targeted bisulfite sequencing of a limited number of informative fCpGs could lead to more consistent coverage and less expensive sequencing of more neurons. Single-cell measurements of small numbers of fCpGs, and snMCode cell type-specific CpG sites (Tian et al., 2023), could efficiently reconstruct human brain lineages, although back changes add complexity. A barcode of 100 fCpGs has enough complexity (2¹⁰⁰ or ~1 × 10³⁰) to potentially distinguish between most excitatory neurons, with less resolution early in development when cells are inherently more related.

The analysis of more brains can verify that a fCpG barcode starts predominately methylated in most individuals. A common initialized state could facilitate standardized human fate maps and comparisons between individuals. Many polymorphisms linked to brain abnormalities such as autism are in neuronal proliferation, migration, and maturation pathways (Pan et al., 2019), and this preliminary survey indicates lineage heterogeneity between individuals (Figure 4A). fCpG barcodes have been applied to the intestines, endometrium, and blood (Gabbutt et al., 2022), and could be found for multiple other tissue types, helping to unravel human development and aging.

MethodsBrain single-cells

Single-cells with their annotations and methylation at each fCpG site were read from single-cell files downloaded from GEO (GSE215353) and supplemental files from reference (Tian et al., 2023). Lists of fCpG sites and data summarized for the Figures are in Supplementary file 2. The cells and methylation at the fCpG sites are in Supplementary file 3 and Supplementary file 4; Supplementary file 5. PWDs were calculated between all cell pairs with at least 30 matching fCpG sites, with PWD data matrices in Supplementary file 6. Additional details and data in response to the Reviewers are provided in the Appendix.

Iqtree

IQtree (Nguyen et al., 2015) tree was downloaded and run on a server with 64 cpus and 32 GB of memory. The model (GTR2 + FO + G4) accounts for backmutation and binary data with missing values. Bootstraps were 1,000 per tree with 3000 iterations for whole brain (~1000 cells) trees and 1000 iterations for inhibitory or excitatory (~2,800 neurons) trees. Trees (.treefile) were displayed with FigTree (http://tree.bio.ed.ac.uk/software/figtree/) with truncation of long branches (generally fewer than 10) for display purposes. The cells used for the trees are in Supplementary file 6.

Additional information

Competing interests

No competing interests declared

Author contributions

Conceptualization, Data curation, Software, Formal analysis, Funding acquisition, Validation, Visualization, Methodology, Writing – original draft, Writing – review and editing

Additional files

MDAR checklist

Supplementary file 1.fCpG annotations.

Supplementary file 2.fCpG genomic locations.

Supplementary file 3.fCpG Brain H01 data.

Supplementary file 4.fCpG H02 brain data.

Supplementary file 5.fCpG H04 brain data.

Supplementary file 6.Cells used for IQtree analysis.

Source code 1.Python code used for analysis.

Data availability

The data used were obtained from GEO (GSE215353).

The following previously published dataset was used:

TianWBartlettALiuHAltshulJNeryJRChenHEckerJRGomezCR2023Epigenetic landscape of Human Brain by Single Nucleus Methylation SequencingNCBI Gene Expression OmnibusGSE215353

Acknowledgements

This work was supported by grants from the NIH (P01CA196569 and CA271237). I thank Drs. Trevor Graham and Heather Grant for useful discussions, and Omar Khan and Nikhil Krishnan for initial studies. The author thanks all of the researchers that helped produce the high-quality very valuable, and freely available data used for analysis.

References

Bandler

Vitali

Delgado

Dvoretskova

Ibarra Molinas

Frazel

Mohammadkhani

Machold

Maedler

Liddelow

Nowakowski

Fishell

Mayer

2022

Single-cell delineation of lineage and genetic identity in the mouse brain

Nature601404409

10.1038/s41586-021-04237-0

34912118

Breuss

Yang

Schlachetzki

JCM

Antaki

Lana

Chung

Chai

Stanley

Song

Newmeyer

Nguyen

O’Brien

Hoeksema

Cao

Nott

McEvoy-Venneri

Pasillas

Barton

Copeland

Nahas

Van Der Kraan

Ding

NIMH Brain Somatic Mosaicism NetworkGlass

Gleeson

2022

Somatic mosaicism reveals clonal distributions of neocortical development

Nature604689696

10.1038/s41586-022-04602-7

35444276

Chien

J-F

Liu

Wang

B-A

Luo

Bartlett

Castanon

Johnson

Nery

Osteen

Altshul

Kenworthy

Valadon

Liem

Claffey

O’Connor

Seeker

Ecker

Behrens

Mukamel

2024

Cell-type-specific effects of age and sex on human cortical neurons

Neuron11225242539

10.1016/j.neuron.2024.05.013

38838671

Chung

Yang

Hevner

Kennedy

Vong

Liu

Patel

Nedunuri

Barton

Noel

Barrows

Stanley

Mittal

Breuss

Schlachetzki

JCM

Kingsmore

Gleeson

2024

Cell-type-resolved mosaicism reveals clonal dynamics of the human forebrain

Nature629384392

10.1038/s41586-024-07292-5

38600385

Ciceri

Baggiolini

Cho

Kshirsagar

Benito-Kwiecinski

Walsh

Aromolaran

Gonzalez-Hernandez

Munguba

Koo

Sevilla

Goldstein

Levitz

Leslie

Koche

Studer

2024

An epigenetic barrier sets the timing of human neuronal maturation

Nature626881890

10.1038/s41586-023-06984-8

38297124

Delgado

Allen

Keefe

Mancia Leon

Ziffra

Crouch

Alvarez-Buylla

Nowakowski

2022

Individual human cortical progenitors can produce excitatory and inhibitory neurons

Nature601397403

10.1038/s41586-021-04230-7

34912114

Evrony

Lee

Mehta

Benjamini

Johnson

Cai

Yang

Haseley

Lehmann

Park

Walsh

2015

Cell lineage analysis in human brain using endogenous retroelements

Neuron854959

10.1016/j.neuron.2014.12.028

25569347

Eze

Bhaduri

Haeussler

Nowakowski

Kriegstein

2021

Single-cell atlas of early human brain development highlights heterogeneity of human neuroepithelial cells and early radial glia

Nature Neuroscience24584594

10.1038/s41593-020-00794-1

33723434

Fan

Zhou

Sun

Yang

Wang

Chen

Yong

Dong

Wen

Qiao

Wang

Tang

2020

Single-cell transcriptome analysis reveals cell lineage specification in temporal-spatial patterns in human cortical development

Science Advances6

eaaz2978

10.1126/sciadv.aaz2978

32923614

Finlay

Darlington

1995

Linked regularities in the development and evolution of mammalian brains

Science26815781584

10.1126/science.7777856

7777856

Gabbutt

Schenck

Weisenberger

Kimberley

Berner

Househam

Lakatos

Robertson-Tessi

Martin

Patel

Clark

Latchford

Barnes

Leedham

Anderson

ARA

Graham

Shibata

2022

Fluctuating methylation clocks for cell lineage tracing at high temporal resolution in human tissues

Nature Biotechnology40720730

10.1038/s41587-021-01109-w

34980912

Gage

2002

Neurogenesis in the adult brain

The Journal of Neuroscience22612613

10.1523/JNEUROSCI.22-03-00612.2002

11826087

Heffel

Zhou

Zhang

Lee

D-S

Hou

Pastor-Alonso

Abuhanna

Galasso

Kern

Tai

C-Y

Garcia-Padilla

Nafisi

Zhou

Schmitt

Haeussler

Wick

Zhang

Xie

Ziffra

Mukamel

Eskin

Nowakowski

Dixon

Pasaniuc

Ecker

Zhu

Bintu

Paredes

Luo

2024

Temporally distinct 3D multi-omic dynamics in the developing human brain

Nature635481489

10.1038/s41586-024-08030-7

39385032

Guo

Gao

Ren

Yuan

Yan

Lian

Gao

Wen

Tang

Qiao

2018

Single-cell multi-omics sequencing of human early embryos

Nature Cell BiologyPMID847858

10.1038/s41556-018-0166-4

Lister

Mukamel

Nery

Urich

Puddifoot

Johnson

Lucero

Huang

Dwork

Schultz

Tonti-Filippini

Heyn

Rao

Esteller

Haghighi

Sejnowski

Behrens

Ecker

2013

Global epigenomic reconfiguration during mammalian brain development

Science341

1237905

10.1126/science.1237905

23828890

Lodato

Woodworth

Lee

Evrony

Mehta

Karger

Lee

Chittenden

D’Gama

Cai

Luquette

Lee

Park

Walsh

2015

Somatic mutation in single human neurons tracks developmental and transcriptional history

Science3509498

10.1126/science.aab1785

26430121

Loyfer

Magenheim

Peretz

Cann

Bredno

Klochendler

Fox-Fisher

Shabi-Porat

Hecht

Pelet

Moss

Drawshy

Amini

Moradi

Nagaraju

Bauman

Shveiky

Porat

Dior

Rivkin

Hirshoren

Carmon

Pikarsky

Khalaileh

Zamir

Grinbaum

Abu Gazala

Mizrahi

Shussman

Korach

Wald

Izhar

Erez

Yutkin

Samet

Rotnemer Golinkin

Spalding

Druid

Arner

Shapiro

AMJ

Grompe

Aravanis

Venn

Jamshidi

Shemer

Dor

Glaser

Kaplan

2023

A DNA methylation atlas of normal human cell types

Nature613355364

10.1038/s41586-022-05580-6

36599988

McKenna

Gagnon

2019

Recording development with single cell dynamic lineage tracing

Development146

dev169730

10.1242/dev.169730

31249005

Messerschmidt

Knowles

Solter

2014

DNA methylation dynamics during epigenetic reprogramming in the germline and preimplantation embryos

Genes & Development28812828

10.1101/gad.234294.113

Nguyen

L-T

Schmidt

von Haeseler

Minh

2015

IQ-TREE: a fast and effective stochastic algorithm for estimating maximum-likelihood phylogenies

Molecular Biology and Evolution32268274

10.1093/molbev/msu300

25371430

Pan

Yuan

2019

Toward a better understanding of neuronal migration deficits in autism spectrum disorders

Frontiers in Cell and Developmental Biology7

205

10.3389/fcell.2019.00205

31620440

Rakic

1988

Specification of cerebral cortical areas

Science241170176

10.1126/science.3291116

3291116

Stiles

Jernigan

2010

The basics of brain development

Neuropsychology Review20327348

10.1007/s11065-010-9148-4

21042938

Tian

Zhou

Bartlett

Zeng

Liu

Castanon

Kenworthy

Altshul

Valadon

Aldridge

Nery

Chen

Johnson

Lucero

Osteen

Emerson

Rink

Lee

Siletti

Liem

Claffey

O’Connor

Yanny

Nyhus

Dee

Casper

Shapovalova

Hirschstein

Ding

S-L

Hodge

Levi

Keene

Linnarsson

Lein

Ren

Behrens

Ecker

2023

Single-cell DNA methylation and 3D genome architecture in the human brain

Science382

eadf5357

10.1126/science.adf5357

37824674

Appendix 1

The purpose of this appendix is to address the broad concerns raised by the three reviewers. The goals are to better explain the mechanics of fCpG barcodes, present new data illustrating their reproducibility and stability with aging, and demonstrate how barcodes can record ancestry after conception, even in the context of active demethylation.

Differences between fCpGs and traditional cell type classifiers

Both fCpG barcodes and traditional DNA methylation-based cell type classifiers can differentiate between cells, but their identification processes and underlying biological principles differ significantly. Traditional cell type classifiers focus on identifying CpG sites that are consistently differentially methylated across distinct cell types. Success in this approach yields a barcode, such as the snMCode CpG sites (Tian et al., 2023), where each cell of a given phenotype shares the same methylation pattern, distinct from those of other phenotypes.

In contrast, the identification of fCpGs is fundamentally different because fCpG methylation can vary among cells, even within the same type. In traditional screening for cell type classifiers, a CpG site is retained if its methylation is consistent across cells with the same phenotype and discarded if it varies. Conversely, when screening for fCpG sites, a CpG site is retained if its methylation differs among cells and discarded if it skews toward either methylated or unmethylated states. The fCpG barcode captures random errors accumulated throughout development, recording ancestry, while the methylation profiles used in cell type classifiers more accurately reflect terminal differentiation.

Lineage tracing: Cladistics and ancestry

Although fCpG selection and cell type classifier selection are fundamentally opposed, their barcodes may align if clades of cells sharing the same phenotypes arise from common progenitors (Appendix 1—figure 1A). This study supports the notion that neurons of the same type typically originate from common progenitors, as cells within a clade generally exhibit polymorphic yet more similar fCpG barcodes than those from different clades. fCpG barcodes may provide complementary insights into cell mitotic ages, diversity within a clade, and migration patterns of daughter cells—factors that are often more challenging to discern from RNA-seq data alone.

Appendix 1—figure 1.Single cell lineage tracing with dynamic fCpG barcodes.

(A) Trees can be reconstructed by comparing phenotypes or by comparing genomic differences such as fluctuating CpG (fCpG) barcodes. Ancestry and phenotypes may be discordant if progenitor cells produce cells of different phenotypes. More typically, ancestry and phenotypes align because cells with the same phenotypes tend to have common progenitors. For the single-cell brain data, ancestry and phenotype align because cells of the same type are generally more closely related. (B) fCpG barcodes appear to start predominately methylated in the progenitor cell. With division, random replication error occur and are propagated to daughter cells. Counting and then averaging the differences between fCpG sites yields an average pairwise distance (pairwise difference, PWD, range 0–1). More related daughter cells tend to have lower PWDs, but barcodes may also match by chance.

fCpG barcode mechanics

The fCpG barcode begins as predominately methylated, reflecting the observation that inhibitory neurons in the pons—emerging early in development—are predominantly methylated across the three brains examined. It is assumed that random errors occur when the barcode is replicated during cell divisions, as illustrated in a cartoon (Appendix 1—figure 1B). As errors are perpetuated and new ones accumulate, closely related cells tend to exhibit more similar methylation patterns than those that are less related. The differences among these patterns can be quantified by calculating the average PWDs between barcode fCpGs.

The barcode mechanics are more formally described by simulations that match the experimental data. These simulations start with a single-cell possessing a fully methylated barcode. Key parameters include fCpG methylation error rates, the number of cell divisions, and whether a cell division yields two, one, or zero daughter cells. A simulation that broadly matches the experimental data for excitatory neurons has 150 cell divisions and a fCpG error rate of 0.01 per division, applied equally to both methylated-to-unmethylated and unmethylated-to-methylation flips (Appendix 1—figure 2).

Appendix 1—figure 2.Barcode dynamics.

Simulations broadly consistent with the experimental data indicate a replication error rate of 0.01 per fluctuating CpG (fCpG) site per division, with equal probabilities of changes or flips from methylated to demethylated (1>0) and from 0>1. A simulation for excitatory neurogenesis is shown, where simplistically, excitatory neurons cease division and appear after 150 divisions. The graph displays how individual fCpG sites change through time. The fCpG barcode starts methylated, and barcode methylation decreases with divisions. Even with an error rate of 0.01, after 150 divisions only about 5% of fCpG sites experience four or more flips, and half have had zero or only a single flip. A fCpG barcode can still effectively distinguish between cells if the flips are random and multiple fCpG sites are compared between cells. Although backflips complicate analysis, the large numbers of replication errors facilitate comparisons between neurons that develop during a short prenatal interval.

Notably, even with a high error rate of 0.01 per division ancestry may be reconstructed because each fCpG sites experiences relatively few changes or methylation flips. By the end of 150 divisions, the average methylation level approaches approximately 50%, with most fCpG sites experiencing three or fewer changes in methylation. Among the methylated sites, around 20% remain unchanged (methylated), while approximately 25% of fCpG sites have undergone two flips (from 1 to 0 to 1). For unmethylated sites, about one-third experience one flip (from 1 to 0), and roughly 15% have had three flips (from 1 to 0 to 1–0). Only about 5% of sites experience more than three flips. Despite this, the randomness of these flips leads to highly polymorphic barcodes among the final cells, with more related cells having more similar barcodes. As noted by Reviewer 1, this barcode mechanism is essentially a mitotic clock that becomes polymorphic.

Simulations of neurogenesis

Simulations that align with experimental data can effectively illustrate the mechanics of the barcode. The first simulation, designed to model early neurogenesis in the hindbrain, features a limited number of divisions and a basic exponential expansion, beginning with a single-cell possessing a fully methylated barcode (Appendix 1—figure 3A). After 19 divisions, resulting in ~500,000 cells, the average methylation across the population declines to approximately 85%, while the average PWD between cells increases to about 0.28. The proportion of closely related neighboring cells (with PWDs <0.1) progressively decreases to around 45%.

Appendix 1—figure 3.Fluctuating CpG (fCpG) barcode simulations.

The simulations start with a single progenitor and a fully methylated barcode with 200 fCpG sites and an error rate of 0.01 per division. At each time point, up to 1000 cells are sampled from the population to calculate fCpG barcode average methylation, average pairwise difference (PWD), and cell proportions with a nearest neighbor with a PWD <0.1. Final tree expansions are truncated to allow for visualization. A range of simulations are broadly consistent with the experimental data. (A) Simulation of exponential growth where each cell yields two daughter cells broadly models early hindbrain neurogenesis. After 19 divisions, average barcode methylation is ~0.83, PWD is ~0.27, and among the 1000 sampled cells, for about 44% of cells there is another cell with a similar barcode (PWD <0.1). (B) Simulation of inhibitory neurogenesis with differentiation after 50 divisions. Early divisions are characterized by cell death (zero or one daughter, represented by dead ends in the tree), with terminal growth. After 50 divisions, average barcode methylation is ~0.69, average PWD is 0.35, and 20% of sampled cells have a nearest neighbor (PWD <0.1) C: Simulations of excitatory neurogenesis with differentiation after 150 divisions. As with inhibitory neurogenesis, cell death during early divisions limits population size before terminal expansion. After 150 divisions, average methylation is ~0.56, PWD is ~0.4, and ~6% of cells have a nearest neighbor (PWD <0.1) among the 1000 sampled cells.

The simulation of inhibitory neurogenesis extends to 50 divisions, also starting from a single-cell with a fully methylated barcode (Appendix 1—figure 3B). In this scenario, cell division is balanced by cell death to maintain a small population early in development, followed by subsequent expansion. By the end of 50 divisions, average cell population methylation falls to about 70%, and average PWD increases to around 0.35, with the proportion of closely related neighboring cells dropping to approximately 20%.

The simulation of excitatory neurogenesis involves 150 divisions, where early development similarly balances cell division and cell death, followed by expansion (Appendix 1—figure 3C). The average cell population methylation decreases to about 55%, and the average PWD among cells rises to around 0.4, with the proportion of closely related neighboring cells declining to about 5%. While these simulations do not fully capture the complexity of neurogenesis, such as the variation in differentiation times, they illustrate how fundamental barcode mechanics can yield neuron populations broadly consistent with experimental data.

Barcode cell division dynamics during early development

The public review requested additional experimental data, and Reviewer 1 raised concerns that active methylation processes early in development could obscure any signature of inaccurate fCpG methylation. Direct comparisons of immediate human daughter neurons pose challenges, and it is uncertain what a relevant cell culture study might be.

Whole genome bisulfite single-cell data are available for human germ cells, zygotes, 2–8 cell embryos, morulae, and ICM samples (Li et al., 2018). The data show that barcodes are more methylated in germ cells and gradually become less methylated as development progresses to the ICM (Figure 4). While fCpG barcodes are polymorphic between unrelated samples, they exhibit greater similarity among closely related cells in the 2–8 cell stage, with reduced similarity in morulae and ICM samples (Appendix 1—figure 4).

Appendix 1—figure 4.Fluctuating CpG (fCpG) barcodes at conception, when germline methylation is erased by active and passive demethylation.

Whole genome bisulfite single-cell sequencing data are from GSE100272. Male cells were inferred from a paucity of Y chromosome reads. (A) fCpG methylation generally decreases during early development. fCpG barcode methylation is highest in sperm, albeit sperm X chromosomes yield female zygotes. A brain cell progenitor with predominately methylated fCpGs was not evident. (B) Unlike at the start of brain development, fCpG barcodes early in life are polymorphic between unrelated embryos. However, fCpG barcodes are more similar between related cells in 2–8 cell embryos, and less similar between cells in morulae and the ICM. Dots indicate values of cell pairs, with a minimum of 25 comparable fCpG sites.

This additional data indicates that more closely related daughter cells have more similar barcodes, even amidst the chaotic background of early development, where both active and passive demethylation largely erases germ cell methylation (Messerschmidt et al., 2014). Notably, although a neurogenic lineage is present among the embryonic cells, none of the cells displayed fully methylated barcodes, suggesting that the anticipated remethylation in a neurogenic progenitor has not yet occurred. The activation of DNMT3A during early neurogenesis (Lister et al., 2013) may ultimately lead to the establishment of fully methylated barcodes that subsequently become polymorphic.

Neuron barcode reproducibility and stability with aging

New single-cell neuron datasets (Heffel et al., 2024; Chien et al., 2024) became available while the manuscript was under review. The core concept behind brain ancestry barcodes is that they can be used to compare different brains, assuming all brains start with a fully methylated barcode and share similar developmental trajectories. These new datasets offer valuable opportunities to test for barcode reproducibility and stability. One dataset (Appendix 1—figure 5) measured neurons in a 7-mo-old brain from the frontal cortex and hippocampus (Heffel et al., 2024). The adult fCpG barcode methylation levels observed are largely present in the infant brain, supporting the idea that fCpG methylation patterns are established prenatally through replication errors and remain stable once cell division ceases.

Appendix 1—figure 5.New single-cell data indicate fluctuating CpG (fCpG) barcode methylation at 7 mo of age is similar to adult levels (H02, 29 yo; H01, 42 yo; H04, 58 yo).

(A) Inhibitory, and lower and upper cortical excitatory neuron barcode methylation levels from the frontal cortex are similar. (B) Inhibitory and excitatory (CA and DG) neuron barcode methylation levels are similar between infant (three samples) and adult hippocampus.

A second dataset (Chien et al., 2024) sampled the same region of the frontal cortex (Brodmann area 46) from young (~25 y) and old (~70 y) brains. This study did not reveal extensive changes in DNA methylation with aging but identified a small number of differentially methylated CpG regions (Chien et al., 2024). Consistent with the overall stability of DNA methylation during aging, average barcode methylation (Appendix 1—figure 6A) and PWDs (Appendix 1—figure 6B) among the ten samples from six male brains and three adult brains (Brodmann area 46) in the manuscript were similar, indicating reproducibility. While average barcode methylation remained stable with age for inhibitory neurons, upper and lower cortical excitatory neurons exhibited significantly higher methylation levels in older brains. This trend suggests a preferential loss of barcode methylation in ‘older’ excitatory neurons, as significantly fewer neurons with less methylated barcodes were sampled from older brains (Appendix 1—figure 6C and D). Alternatively, if fCpG methylation remains stable after cell division ceases, the observed decrease in average barcode methylation could indicate that neurons with less methylated barcodes are preferentially lost with aging. This suggests that neurons arising later in development may have higher mortality during aging. While other explanations are possible, the data illustrate how barcodes could be potentially used to infer both brain development and aging. Overall, these two new datasets help illustrate the degree of fCpG barcode reproducibility, and stability during aging.

Appendix 1—figure 6.Fluctuating CpG (fCpG) barcode reproducibility and stability with aging.

New data are frontal cortex (Broadman area 46) WGBS single-cells from young (25 y-old, three individuals and five samples) and older (70–71 y-old, three individuals and fivesamples) males. Data from the manuscript are also shown (H02, 29 yo; H01, 42 yo; H04, 58 yo). (A) Average neuron barcode methylation levels were similar between different aged individuals. Excitatory neuron barcode methylation in older males (70–71 yo) was significantly greater than the 25 yo males (t-test comparing only the 25 yo and 70–71 yo groups). Both upper (green) and lower (blue) cortical excitatory neurons showed greater average barcode methylation with aging. (B) Average inhibitory and excitatory neuron barcode pairwise differences (PWDs) (comparing within subtypes in each brain) were similar between different aged individuals, indicating that barcodes remain polymorphic. (C) Composite histograms of individual neuron barcode methylation levels for younger (black, 25 yo) and older (red, 70-71 yo) brains. There is a preferential loss of neurons with less methylated barcodes, especially with excitatory neurons. (D) Younger brain neuron barcodes were used to define quantiles. Older brain neurons with less methylated barcodes were depleted in the less methylated quantiles, with significant differences for all quantiles (Mann Whitney U test, p<10–9).

Summary

This supplement addresses several concerns raised during the Public Review. The criteria for the fCpG barcode, as stated in the manuscript, are as follows: (1) a defined initial pattern in a progenitor cell; (2) polymorphic changes upon cell division; (3) sufficient polymorphism to distinguish between most cells; and (4) the capability to record ancestry. Among these criteria, only the adequacy of barcode polymorphisms to differentiate between most inhibitory and excitatory neurons has been well-established, indicating that while evidence for lineage tracing is supportive, it remains inadequate and dependent on assumptions that are difficult to validate.

The supplement provides the requested more detailed explanation of barcode mechanics. Simulations that begin with a single-cell containing a fully methylated barcode broadly align with experimental data, incorporating relatively straightforward cell divisions, deaths, and expansions alongside random barcode replication errors. The brain facilitates serial barcode examinations because neurons typically stop dividing at different times during development. New barcode data reveal that daughter cells exhibit greater relatedness during embryogenesis, even amidst active and passive demethylation. Additionally, new data demonstrate the degree of barcode reproducibility and stability throughout aging.

Tools to reconstruct human brain development are limited, and fCpG barcodes can potentially add complementary information such as mitotic ages, diversity within a clade, and migration patterns between immediate daughter cells. fCpG barcode complexity is required because most brain development occurs prenatally, necessitating very high replication error rates to distinguish between billions of adult cells. Such high error rates (~10^–2) would inherently lead to polymorphic methylation patterns, complicating the differentiation of ancestral signals from technical or biological noise, especially if methylation remodeling occurs independently of cell division. While the current evidence for fCpG barcode lineage tracing is inadequate, a functional DNA methylation-based barcode comprising as few as 50 fCpG sites (yielding 10¹⁵ unique binary patterns) with similar high replication error rates could feasibly reconstruct aspects of human brain development and aging.

Pipelines

(Files are in Pileline.zip)

1: Download GSE file from GEO

2: Filter out smaller files (typically keep 70 mB or larger tsv.gz files)

3: Use Program 1 (‘mergetsvgzcol123divide4by5.py’ with ‘locALLforfirstfilter.csv’) to select CpG sites at the chromosome locations (+and - strands) of the ~31 k fCpGs, and calculate methylation. A.tsv.gz file produces a.tsv file.

4: Use Program 2 (‘collectmasterlistfromsmalltsvfastermerge’ with ‘master_list.csv’) to make a csv file with the methylation of each fCpG site for each cell, with blank sites labeled as ‘NA’

5: Use Excel to annotate single-cells with their phenotypes from the published keys (1,5). Also calculate the fCpG methylation, and number of fCpGs with data for each cell.

6: Filter cells without relevant phenotypes, and cells with insufficient numbers of fCpG sites with data (500 for the manuscript and 400 fCpG sites for the new Supplement).

7: Calculate PWDs and the number of fCpGs that are compared between all possible cell pairs. Program 3 (‘pwd2.py’) compares between all cells in a single.csv file (no headers, with individual cells across and fCpGs down). Program 4 (‘pwdsquare.py’) calculates PWDs and the number of fCpGs compared between cells when two.csv files are compared. Programs 3 and 4 were used to segment the data into smaller chunks to shorten run times. Each program produces a csv file with PWDs and a csv file with numbers of fCpGs compared.

8: Excel was used to remove cell pairs with too few comparable fCpGs (minimum of 30 for the manuscript and 25 for the new Supplement).

9: Excel was used to calculate average PWDs between all comparable cell pairs and to find the minimum PWD of its nearest neighbor.

Simulations of neurogenesis

Program 5 (‘NumPydet_lin_fCpG_5_1.3arrayflipavemultrun_oneout_good_writeN_XeditDS.py’) is used to simulate different growth scenarios and is parameterized with two csv files. The first csv file (‘lookup5_table_Xonly.csv’ with four columns; lineage, array1, p0_1, p1_0) initializes the first progenitor cell. The number of rows is the number of fCpG sites, array1 is the starting methylation state (all 1’s), p0_1 is the probability of flipping from 0 to 1, and p1_0 is the probability of flipping from 1 to 0 and is 0.01 in the simulations. The other csv file (‘cellnumber_table.csv’ with five columns; division, desired_population, q, r, s) controls cell population size and cell death with each division. The desired_population provides the number of cells after each cell division. The program determines cell survival by randomly selecting (without replacement) for each mother cell a q, r, or s value, where (q+r + s) = (the number of mother cells). A cell with ‘q’ will have one surviving daughter cells, ‘r’ has 2 surviving daughter cells and ‘s’ has no surviving daughter cells.

The outputs are ‘sumlineage_arrays.csv’ that calculates the average methylation of the cell population at each fCpG site, and ‘last_run_lineage_arrays.csv’ which outputs the methylation of each fCpG site for each simulated cell. Program 6 (‘covertrowlosecommas.py’) produces ‘out.csv’ to format the data for Program 7 (‘pwdofNcellsNaN4sigAVE_Rruns.py’) that samples specific numbers of simulated cells to match the numbers of cells sampled with the experimental data. Program 8 (flip101010.py) was used for the simulations of Appendix 1—figure 2.

10.7554/eLife.101163.3.sa0

eLife Assessment

Konopka

Genevieve

Reviewing EditorUniversity of Texas Southwestern Medical CenterUnited States

Incomplete

Valuable

This study presents a valuable conceptual approach that cell lineage can be determined using methylation data. However, the evidence supporting the claims of the author remains incomplete after revision. If clarified further as described in the reviews, this approach could be of broad interest to neuroscientists and developmental biologists.

10.7554/eLife.101163.3.sa1

Reviewer #1 (Public review):

Reviewer

Summary:

In this manuscript, Shibata describes a method to assess rapidly fluctuating CpG sites (fCpGs) from single-cell methylation sequencing (sc-MeSeq) data. Assuming that fCpGs are largely consistent over time with changes induced by inheritable events during replication, the author infers lineage relationships in available brain-derived sc-MeSeq. Supplementing current lineage tracing through genomic and mitochondrial mosaic variants is an interesting concept that could supplement current work or allow additional lineage analysis in existing data.

However, the author failed to convincingly show the power of fCpG analysis to determine lineages in the human brain. While the correlation with cellular division and distinction of cell types appears plausible and strong, the application to detect specific lineages is less convincing. Aspects of this might be due to a lack of clarity in presentation and erroneous use of developmental concepts. However, without addressing these problems it is challenging for a reader to come to the same conclusions as the author.

On the flip side, this novel application of fCpGs will allow the re-use of existing sc-MeSeq to infer additional features that were previously unavailable, once the biological relevance has been further elucidated.

Strengths:

• Novel re-analysis application of methylation data to infer the status of fCpGs and the use as a lineage marker

• Application of this method to an innovative existing data set to benchmark this framework against existing developmental knowledge

Weaknesses:

• Inconsistent or erroneous use of neurodevelopmental concepts which hinders appropriate interpretation of the results.

• Somewhat confusing presentation at times which makes it hard to judge the value of this novel approach.

10.7554/eLife.101163.3.sa2

Reviewer #3 (Public review):

Reviewer

Summary:

Cell lineage tracing necessitates continuous visible tracking or permanent molecular markers that daughter cells inherit from their progenitors. To successfully trace cell lineages, it is essential to generate and detect sufficient new markers during each cell division. Thus, molecular cell lineages have been predominantly studied with stably inherited genetic markers in animal models and somatic DNA mutations in the human brain. DNA methylation is unstable across cell divisions and differentiation, and is hardly called barcodes. The use of "Human Brain Barcodes" in the title and across the whole paper lacks convincing evidence - it is questionable that CpG methylation is always stably inherited by daughter cells.

Strengths:

Analysis of DNA methylation.

Weaknesses:

The unstable nature of CpG methylation would introduce significant problems in inferring the true cell lineage. To establish DNA methylation as a means for lineage tracing, it is necessary to test whether the DNA methylation patterns can faithfully track cell lineages with in vitro differentiated & visibly tracked cell lineages.

The unreliable CpG methylation status also raises the question of what the "Barcodes" refer to in the title and across this study. Barcodes should be stable in principle and not dynamic across cell generations, as defined in the Reference #1. The CRISPR/Cas9 mutable barcodes or the somatic mutations may be considered barcodes, but the reviewer is not convinced that the "dynamic" CpG methylation fits the "barcodes" terminology. This problem is even more concerning in the last section of the results, where CpG status fluctuates in post-mitotic cells.

The manuscript frequently states assumptions in a tone of conclusions and interprets results without rejecting alternative hypotheses. For example, the title "Human Brain Barcodes" should be backed with solid supporting evidence. For another example, the author assumed that the early-formed brain stem would resemble progenitors better and have a higher average methylation level than the forebrain - however, this difference in DNA methylation status could well reflect cell-type-specific gene expression instead of cell lineage progression.

Other points:

(1) The conclusion that excitatory neurons undergo tangential migration is unclear - how far away did the author mean for the tangential direction? Lateral dispersion is known, but it is hard to believe that the excitatory neurons travel across different brain regions. More importantly, how would the author interpret shared or divergent methylation for the same cell type across different brain regions?

(2) The sparsity and resolution of the single-cell DNA methylation data. The methylation status is detected in only a small fraction (~500/31,000 = 1.6%) of fCpGs per cell, with only 48 common sites identified between cell pairs. Given that the human genome contains over 28 million CpG sites, it is important to evaluate whether these fCpGs are truly representative.

(3) While focusing on the X-chromosome may simplify the identification of polymorphic fCpGs, the confidence in determining its methylation status (0 or 1) is questionable when a CpG site is covered by only one read.

10.7554/eLife.101163.3.sa3

Author response

Shibata

Darryl

AuthorUniversity of Southern California

Los Angeles

United States

The following is the authors’ response to the original reviews.

eLife assessment:

Developing a reliable method to record ancestry and distinguish between human somatic cells presents significant challenges. I fully acknowledge that my current evidence supporting the claim of lineage tracing with fCpG barcodes is inadequate. I agree with Reviewer 1 that fCpG barcodes are essentially a cellular division clock that diverges over time. A division clock could potentially document when cells cease to divide during development, with immediate daughter cells likely exhibiting more similar barcodes than those that are less related. Although it remains uncertain whether the current fCpG barcodes capture useful biological information, refinement of this type of tool could complement other approaches that reconstruct human brain function, development, and aging.

Due to my lack of clarity, the fCpG barcode was perceived to be a new type of cell classifier. However, it is fundamentally different. fCpG sites are selected based on their differences between cells of the same type, while traditional cell classifiers focus on sites with consistent methylation patterns in cells of the same type. Despite these opposing criteria, fCpG barcodes and traditional cell classifiers may align because neuron subtypes often share common progenitors. As a result, cells of the same phenotype are also closely related by ancestry, and ex post facto, have similar fCpG barcodes. fCpG barcodes are complementary to cell type classifiers, and potentially provide insights into aspects such as mitotic ages, diversity within a clade, and migration of immediate daughters---information which is otherwise difficult to obtain. The title has been modified to “Human Brain Ancestral Barcodes” to better reflect the function of the fCpG barcodes. The manuscript is edited to correct errors, and a new Supplement is added to further explain fCpG barcode mechanics and present new supporting data.

Reviewer #1 (Public review):

I thank Reviewer 1 for his constructive comments. Major noted weaknesses were (1) insufficient clarity and brevity of the methodology, (2) inconsistent or erroneous use of neurodevelopmental concepts, and (3) lack of consideration for alternative explanations.

(1) The methodology is now outlined in detailed in a new Supplement, including simulations that indicate that the error rate consistent with the experimental data is about 0.01 changes in methylation per fCpG site per division.

(2) Conceptual and terminology errors noted by the Reviewers are corrected in the manuscript.

(3) I agree completely with the alternative explanation of Reviewer 1 that fCpGs are “a cellular division clock that diverges over 'time'”. Differences between more traditional cell type classifiers and fCpG barcodes are more fully outlined in the new Supplement. Ancestry recorded by fCpGs and cell type classifiers are confounded because cells of the same phenotype typically have common progenitors---cells within a clade have similar fCpG barcodes because they are closely related. fCpG barcodes can compliment cell type classifiers with additional information such as mitotic ages, ancestry within a clade, and daughter cell migration.

Reviewer #1 (Recommendations for the authors):

(1) A lot of the interpretations suffer from an extremely loose/erroneous use of developmental concepts and a lack of transparency. For instance:

a) The thalamus is not part of the brain stem

Corrected.

b) The pons contains cells other than inhibitory neurons in the data; the same is true for the hippocampus which contains multiple cell types

Corrected to refer to the specific cell types in these regions.

c) The author talks about the rostral-caudal timing a lot which is not really discussed to this degree in the cited references. Thus, it is also unclear how interneurons fit in this model as they are distinguished by a ventral-dorsal difference from excitatory neurons. Also, it is unclear whether the timing is really as distinct as claimed. For instance, inhibitory neurons and excitatory neurons significantly overlap in their birth timing. Finally, conceptually, it does not make sense to go by developmental timing as the author proposes that it is the number of divisions that is relevant. While they are somewhat correlated there are potentially stark differences.

The manuscript attempts to describe what might be broadly expected when barcodes are sampled from different cell types and locations. As a proposed mitotic clock, the fCpG barcode methylation level could time when each neuron ceased division and differentiated. The wide ranges of fCpG barcode methylation of each cell type (Fig 2A) would be consistent with significant overlap between cell types. The manuscript is edited to emphasize overlapping rather than distinct sequential differentiation of the cell types.

d) Neocortical astrocytes and some oligodendrocytes share a lineage, whereas a subset of oligodendrocytes in the cortex shares an origin with interneurons. This could confound results but is never discussed.

The manuscript does not assess glial lineages in detail because neurons were preferentially included in the sampling whereas glial cells were non-systematically excluded. This sampling information is now included in the section “fCpG barcode identification”.

e) Neocortical interneurons should be more closely related in terms of lineage-to-excitatory neurons than other inhibitory neurons of, for instance, the pons. This is not clearly discussed and delineated.

This is not discussed. It may not be possible analyze these details with the current data. The ancestral tree reconstructions indicate that excitatory neurons that appear earlier in development (and are more methylated) are more often more closely related to inhibitory neurons.

f) While there is some spread of excitatory neurons tangentially, there is no tangential migration at the scale of interneurons as (somewhat) suggested/implied here.

The abstract and results have been modified to indicate greater inhibitory than excitatory neuron tangential migration, but that the extent of excitatory neuron tangential migration cannot be determined because of the sparse sampling and that barcodes may be similar by chance.

g) The nature of the NN cells is quite important as cells not derived from the neocortical anlage are unlikely to share a developmental origin (e.g., microglia, endothelial cells). This should be clarified and clearly stated.

The manuscript is modified to indicate that NN cells are microglial and endothelial cells. These cells have different developmental origins, and their data are present in Fig 2A, but are not further used for ancestral analysis.

(2) The presentation is often somewhat confusing to me and lacks detail. For instance:

a) The methods are extremely short and I was unable to find a reference for a full pipeline, so other researchers can replicate the work and learn how to use the pipeline.

The pipeline including python code is outlined in the new Supplement

b) Often numbers are given as ~XX when the actual number with some indication of confidence or spread would be more appropriate.

Data ranges are often indicated with the violin plots.

c) Many figure legends are exceedingly short and do not provide an appropriate level of detail.

Figure legends have been modified to include more detail

d) Not defining groups in the figure legends or a table is quite unacceptable to me. I do not think that referring to a prior publication (that does not consistently use these groups anyway) is sufficient.

The cell groups are based on the annotations provided with each single cell in the public databases.

e) The used data should be better defined and introduced (number of cells, different subtypes across areas, which cells were excluded; I assume the latter as pons and hippocampus are only mentioned for one type of neuronal cells, see also above).

The data used are present in Supplemental File 2 under the tab “cell summary H01, H02, H04”.

f) Why were different upper bounds used for filtering for H01 and H02, and H04 is not mentioned? Why are inhibitory and excitatory neurons specifically mentioned (Lines 61-66)?

The filtering is used to eliminate, as much as possible, cell type specific methylation, or CpG sites with skewed neuron methylation. The filtering eliminates CpG sites with high or low methylation within each of the three brains, and within the two major neuron subtypes. The goal is to enrich for CpG sites with polymorphic but not cell type specific methylation. This process is ad hoc as success criteria are currently uncertain. The extent of filtering is balanced by the need to retain sufficient numbers of fCpGs to allow comparisons between the neurons.

g) What 'progenitor' does the author refer to? The Zygote? If yes, can the methylation status be tested directly from a zygote? There is no single progenitor for these cells other than the zygote. Does the assumption hold true when taking this into account? See, for instance, PMID 33737485 for some estimation of lineage bottlenecks.

A brain progenitor cell can be defined as the common ancestor of all adult neurons, and is the first cell where each of its immediate daughter cell lineages yield adult neurons. The zygote is a progenitor cell to all adult cells, and barcode methylation at the start of conception, from the oocyte to the ICM, was analyzed in the new Supplement. The proposed brain progenitor cell with a fully methylated barcode was not yet evident even in the ICM.

(3) I am generally not convinced that the fCpGs represent anything but a molecular clock of cell divisions and that many of the similarities are a function of lower division numbers where the state might be more homogenous. This mainly derives from the issues cited above, the lack of convincing evidence to the contrary, and the sparsity of the assessed data.

Agree that the fCpG barcode is a mitotic clock that becomes polymorphic with divisions. As outlined in the new Supplement, ancestry and cell type are confounded because cells of the same type typically have a common progenitor.

a) There appears little consideration or modeling of what the ability to switch back does to the lineage reconstruction.

fCpG methylation flipping is further analyzed and discussed in the new Supplement.

b) None of the data convinced me that the observations cannot be explained by the aforementioned molecular clock and systematic methylation similarities of cell types due to their cell state.

See above

(4) Uncategorized minor issues:

a) The author should explain concepts like 'molecular clock hypothesis' (line 27) or 'radial unit hypothesis' (line 154), as they are somewhat complex and might not be intuitive to readers.

The molecular clock hypothesis is deleted and the radial unit hypothesis is explained in more detail in the manuscript.

b) Line 32: '[...] replication errors are much higher compared to base replication [...]'. I think this is central to the method and should be better explained and referenced. Maybe even through a schematic, as this is a central concept for the entire manuscript.

The fCpG barcode mechanics are better explained in the new Supplement. With simulations, the fCpG flip rate is about 0.01 per division per fCpG.

c) Line 41: 'neonatal'. Does the author mean to say prenatal? Most of the cells discussed are postmitotic before birth.

Corrected to prenatal.

d) Line 96: what does 'flip' mean in this context? Please also see the comment on Figure 2C.

Edited to “chage”

e) Lines 134-135: I am not sure whether the author claims to provide evidence for this question, and I would be careful with claims that this work does resolve the question here.

Have toned down claims as evidence for my analysis is currently inadequate.

f) Lines 192-193: I disagree as the fCpGs can switch back and the current data does not convince me that this is an improvement upon mosaic mutation analysis. In my mind, the main advantage is the re-analysis of existing data and the parallel functional insights that can be obtained.

Lineage analysis is more straightforward with DNA sequencing, but with an error rate of ~10-9 per base per division, one needs to sequence a billion base pairs to distinguish between immediate daughter cells. By contrast, with an inferred error rate of ~10-2 per fCpG per division, much less sequencing (about a million-fold less) is needed to find differences between daughter cells.

g) Lines 208-209: I would be careful with claims of complexity resolution given many of the limitations and inherent systematic similarities, as well as the potential of fCpGs to change back to an ancestral state later in the lineage.

Have modified the manuscript to indicate the analysis would be more challenging due to back changes.

h) There seem to be few figures that assess phenomena across the three brains. Even when they exist there is no attempt to provide any statistical analyses to support the conclusions or permutations to assess outlier status relative to expectations.

The analysis could be more extensive, but with only three brains, any results, like this study itself, would be rightly judged inadequate.

Figure 2B: there appears to be a higher number of '0s' for, for instance, inhibitory neurons compared to excitatory neurons. Is that correct and worth mentioning? The changing axes scales also make it hard to assess.

Inhibitory neurons do appear to have more unmethylated fCpGs compared to excitatory neurons, but in general, most inhibitory fCpGs are methylated with a skew to fully methylated fCpGs, consistent with the barcode starting predominately methylated and inhibitory neurons generally appearing earlier in development relative to excitatory neurons.

j) Figure 2C: I have several issues with this. A minor one is the use of 'Glial' which, I believe, does not appear anywhere else before this, so I am unclear what this curve represents. Generally, however, I am not sure what the y-axis represents, as it is not described in the methods or figure legend. I initially thought it was the cumulative frequency, but I do not think that this squares with the data shown in B. I appreciate the overall idea of having 'earlier'/samples with fewer divisions being shifted to the left, but it is very confusing to me when I try to understand the details of the plot.

This graph is now better described in the legend. “Glial” cells are defined as oligodendrocytes and astrocytes. Other non-neuronal cells (such a microglial cells) have now been removed from the graph.

This graph attempts to illustrate how it may be possible to reconstruct brain development from adult neurons, assuming barcodes are mitotic clocks that become polymorphic with cell division. The X axis is “time”, and the Y axis indicates when different cell types reach their adult levels. The cartoon indicates what is visually present along the X axis during development--- brainstem, then ganglionic eminences with a thin cortex, and finally the mature brain with a robust cortex. Time for the X axis is barcode methylation and starts at 100% and ends at 50% or greater methylation. The fCpG barcode methylation of each cell places it on this timeline and indicates when it ceased dividing and differentiated.

The Y axis indicates the progressive accumulation of the final adult contents of each cell type during this timeline. Early in development, the brain is rudimentary and adult cells are absent. At 90% methylation, only the inhibitory neurons in the pons are present. At 80% methylation, some excitatory neurons are beginning to appear. Inhibitory neurons in the pons have reached their final adult levels and many other inhibitory neuron types are reaching adult levels. By 70% methylation, most inhibitory neurons have reached their adult levels, and more adult excitatory neurons (mainly low cortical neurons, L4-6) and glial cells are beginning to appear. By 60% methylation, inhibitory neurogenesis has largely finished. Adult excitatory neurons and glial cells are more abundant and reach their adult levels by 50% or greater cell barcode methylation levels.

The graph illustrates a rough alignment between mitotic ages inferred by barcode methylation levels and the physical appearances of different neuronal types during development. Many neurons die during development, and this graph, if valid, indicates when neurons that survive to adulthood appear during development.

k) Figure 4Bff: it is confusing to me that the text jumps to these panels after introducing Figure 5. This makes it very hard to read this section of the text.

The Figures appear in the order they are first referred to in the text.

l) Figure 5A: could any of this difference be explained by the shared lineage of excitatory neurons and dorsal neocortical glia?

Not sure

m) Figure 5B: after stating that interneurons have a higher lineage fidelity, the figure legend here states the opposite and I am somewhat confused by this statement.

The legend and text have been clarified. Fig 5A restricts fidelity to within inhibitory cell types. Fig 5B compares between neuron subtypes, and illustrates more apparent inhibitory subtype switching, albeit there are more interneuron subtypes than excitatory subtypes.

n) Figure 5E: generally, the use of tSNE for large pairwise distance analysis is often frowned upon (e.g., PMID 37590228), and I would reconsider this argument.

This analysis was an attempt to illustrate that cells of the same phenotype based on their tSNE metrics can be either closely or more distantly related. Although the tSNE comparisons were restricted to subtypes (and not to the entire tSNE graph), tSNE are not designed for such comparisons. This graph and discussion are deleted.

Reviewer #2 (Public review):

The manuscript by Shibata proposed a potentially interesting idea that variation in methylcytosine across cells can inform cellular lineage in a way similar to single nucleotide variants (SNVs). The work builds on the hypothesis that the "replication" of methylcytosine, presumably by DNMT1, is inaccurate and produces stochastic methylation variants that are inherited in a cellular lineage. Although this notion can be correct to some extent, it does not account for other mechanisms that modulate methylcytosines, such as active gain of methylation mediated by DNMT3A/B activity and activity demethylation mediated by TET activity. In some cases, it is known that the modulation of methylation is targeted by sequence-specific transcription factors. In other words, inaccurate DNMT1 activity is only one of the many potential ways that can lead to methylation variants, which fundamentally weakens the hypothesis that methylation variants can serve as a reliable lineage marker. With that being said (being skeptical of the fundamental hypothesis), I want to be as open-minded as possible and try to propose some specific analyses that might better convince me that the author is correct. However, I suspect that the concept of methylation-based lineage tracing cannot be validated without some kind of lineage tracing experiment, which has been successfully demonstrated for scRNA-seq profiling but not yet for methylation profiling (one example is Delgado et al., nature. 2022).

I thank Reviewer 2 for the careful evaluation. The validation experiment example (Delgado et al.) introduced sequence barcodes in mice, which is not generally feasible for human studies.

(1) The manuscript reported that fCpG sites are predominantly intergenic. The author should also score the overlap between fCpG sites and putative regulatory elements and report p-values. If fCpG sites commonly overlap with regulatory elements, that would increase the possibility that these sites being actively regulated by enhancer mechanisms other than maintenance methyltransferase activity.

As mentioned for Reviewer 1, fCpGs are filtered to eliminate cell type specific methylation.

(2) The overlap between fCpG and regulatory sequence is a major alternative explanation for many of the observations regarding the effectiveness of using fCpG sites to classify cell types correctly. One would expect the methylation level of thousands of enhancers to be quite effective in distinguishing cell types based on the published single-cell brain methylome works.

As mentioned above, the manuscript did not clearly indicate that the fCpG barcode is not a cell type classifier. The distinctions between fCpG barcodes and cell type classifiers are better explained in the new Supplement.

(3) The methylation level of fCpG sites is higher in hindbrain structures and lower in forebrain regions. This observation was interpreted as the hindbrain being the "root" of the methylation barcodes and, through "progressive demethylation" produced the methylation states in the forebrain. This interpretation does not match what is known about methylation dynamics in mammalian brains, in particular, there is no data supporting the process of "progressive demethylation". In fact, it is known that with the activation of DNMT3A during early postnatal development in mice or humans (Lister et al., 2013. Science), there is a global gain of methylation in both CH and CG contexts. This is part of the broader issue I see in this manuscript, which is that the model might be correct if "inaccurate mC replication" is the only force that drives methylation dynamics. But in reality, active enzymatic processes such as the activation of DNMT3A have a global impact on the methylome, and it is unclear if any signature for "inaccurate mC replication" survives the de novo methylation wave caused by DNMT3A activity.

Reviewer 2 highlights a critical potential flaw in that any ancestral signal recorded by random replication errors could be overwritten by other active methylation processes. I cannot present data that indicates fCpG replication errors are never overwritten, but new data indicate barcode reproducibility and stability with aging.

New data are also present where barcodes are compared between daughter cells (zygote to ICM) in the setting of active and passive demethylation, when germline methylation is erased. This new analysis shows that daughter cells in 2 to 8 cell embryos have more related barcodes than morula or ICM cells. The subsequent active remethylation by a wave of DNMT3A activity may underlie the observation that the barcode appears to start predominately methylated in brain progenitors.

(3) Perhaps one way the author could address comment 3 is to analyze methylome data across several developmental stages in the same brain region, to first establish that the signal of "inaccurate mC replication" is robust and does not get erased during early postnatal development when DNMT3A deposits a large amount of de novo methylation.

See above

(4) The hypothesis that methylation barcodes are homogeneous among progenitor cells and more polymorphic in derived cells is an interesting one. However, in this study, the observation was likely an artifact caused by the more granular cell types in the brain stem, intermediate granularity in inhibitory cells, and highly continuous cell types in cortical excitatory cells. So, in other words, single-cell studies typically classify hindbrain cell types that are more homogenous, and cortical excitatory cells that are much more heterogeneous. The difference in cell type granularity across brain structures is documented in several whole-brain atlas papers such as Yao et al. 2023 Nature part of the BICCN paper package.

As noted above, fCpG barcode polymorphisms and cell type differentiation are confounded because cells of the same phenotype tend to have common progenitors. The fCpG barcode is not a cell type classifier but more a cell division clock that becomes polymorphic with time. Although fCpG barcodes could be more polymorphic in cortical excitatory cells because there are many more types, fCpG barcodes would inherently become more polymorphic in excitatory cells because they appear later in development.

(5) As discussed in comment 2, the author needs to assess whether the successful classification of cell types (brain lineage) using fCpG was, in fact, driven by fCpG sites overlapping with cell-type specific regulatory elements.

Although unclear in the manuscript, the fCpG is not a cell classifier and the barcode is polymorphic between cells of the same type. fCpG barcodes can appear to be cell classifiers because cell types appear at different times during development, and therefore different cell types have characteristic average barcode methylation levels.

(6) In Figure 5E, the author tried to address the question of whether methylation barcodes inform lineage or post-mitotic methylation remodeling. The Y-axis corresponds to distances in tSNE. However, tSNE involves non-linear scaling, and the distances cannot be interpreted as biological distances. PCA distances or other types of distances computed from high-dimensional data would be more appropriate.

The Figure and discussion are deleted (similar comment by Reviewer 1)

Reviewer #3 (Public review):

Summary:

In the manuscript entitled "Human Brain Barcodes", the author sought to use single-cell CpG methylation information to trace cell lineages in the human brain.

Strengths:

Tracing cell lineages in the human brain is important but technically challenging. Lineage tracing with single-cell CpG methylation would be interesting if convincing evidence exists.

Weaknesses:

As the author noted, "DNA methylation patterns are usually copied between cell division, but the replication errors are much higher compared to base replication". This unstable nature of CpG methylation would introduce significant problems in inferring the true cell lineage. The unreliable CpG methylation status also raises the question of what the "Barcodes" refer to in the title and across this study. Barcodes should be stable in principle and not dynamic across cell generations, as defined in Reference#1. It is not convincing that the "dynamic" CpG methylation fits the "barcodes" terminology. This problem is even more concerning in the last section of results, where CpG would fluctuate in post-mitotic cells.

I thank Reviewer 3 for his thoughtful and careful evaluation. I think the “barcode” terminology is appropriate. Dynamic engineered barcodes such as CRISPR/Cas9 mutable barcodes are used in biology to record changes over time. The fCpG barcode appears to start with a single state in a progenitor cell and changes with cell division to become polymorphic in adult cells. Therefore, I think the description of a dynamic fCpG barcode is appropriate.

Reviewer #3 (Recommendations for the authors):

(1) As the author noted, "DNA methylation patterns are usually copied between cell division, but the replication errors are much higher compared to base replication". This unstable nature of CpG methylation would introduce significant problems in inferring the true cell lineage. To establish DNA methylation as a means for lineage tracing, one control experiment would be testing whether the DNA methylation patterns can faithfully track cell lineages for in vitro differentiated & visibly tracked cell lineages. Has this kind of experiment been done in the field?

These types of experiments have not been performed to my knowledge and an appropriate tissue culture model is uncertain. New single cell WGBS data from the zygote to ICM indicate that more immediate daughter cells have more related barcodes even in the setting of active DNA demethylation.

(2) The study includes assumptions that should be backed with solid rationale, supporting evidence, or reference. Here are a couple of examples:

a) the author discarded stable CpG sites with <0.2 or >0.8 average methylation without a clear rationale in H02, and then used <0.3 and >0.7 for a specific sample H01.

The filtering was ad hoc and was used to remove, as much as possible, CpG sites with cell type specific or patient specific methylation. CpG sites with skewed methylation are more likely cell type specific, whereas X chromosome CpG sites with methylation closer to 0.5 in male cells are more likely to be unstable. The ad hoc filtering attempted to remove cell specific CpGs sites while still retaining enough CpG sites to allow comparisons between cells.

b) The author assumed that the early-formed brain stem would resemble progenitors better and have a higher average methylation level than the forebrain. However, this difference in DNA methylation status could reflect developmental timing or cell type-specific gene expression changes.

This observation that brain stem neurons that appear early in development have highly methylated fCpG barcodes in all 3 brains supports the idea that the fCpG barcode starts predominately methylated. Alternative explanations are possible.

(3) The conclusion that excitatory neurons undergo tangential migration is unclear - how far away did the author mean for the tangential direction? Lateral dispersion is known, but it would be striking that the excitatory neurons travel across different brain regions. The question is, how would the author interpret shared or divergent methylation for the same cell type across different brain regions?

As noted with Reviewer 1, this analysis is modified to indicate that evidence of tangential migration is greater for inhibitory than excitatory neurons, but the extent of excitatory neuron migration is uncertain because of sparse sampling, and because fCpG barcodes can be similar by chance.

(4) The sparsity and resolution of the single-cell DNA methylation data. The methylation status is detected in only a small fraction (~500/31,000 = 1.6%) of fCpGs per cell, with only 48 common sites identified between cell pairs. Given that the human genome contains over 28 million CpG sites, it is important to evaluate whether these fCpGs are truly representative. How many of these sites were considered "barcodes"?

fCpG barcodes are distinct from traditional cell type classifiers, and how fCpGs are identified are better outlined in the new Supplement.

(5) While focusing on the X-chromosome may simplify the identification of polymorphic fCpGs, the confidence in determining its methylation status (0 or 1) is questionable when a CpG site is covered by only one read. Did the author consider the read number of detected fCpGs in each cell when calculating methylation levels? Certain CpG sites on autosomes may also have sufficient coverage and high variability across cells, meeting the selection criteria applied to X-chromosome CpGs.

In most cases, a fCpG site was covered by only a single read

(6) The overall writing in the Title, the Main text, Figure legends, and Methods sections are overly simplified, making it difficult to follow. For instance, how did the author perform PWD analysis? How did they handle missing values when constructing lineage trees?

There is not much introduction to lineage tracing in the human brain or the use of DNA methylation to trace cell lineage.

These shortcomings are improved in the manuscript and with the new Supplement. The analysis pipeline including the Python programs are outlined and included as new Supplemental materials. IQ tree can handle the binary fCpG barcode data and skips missing values with its standard settings.

Line 80: it is unclear: "Brain patterns were similar"

Clarified

Line 98: The meaning is unclear here: "Outer excitatory and glial progenitor cells are present" What are these glial progenitor cells and when/how they stop dividing?

The glial cells are the oligodendrocytes and astrocytes. The main take away point is that these glial cells have low barcode methylation, consistent with their appearances later in development.

Line 104: It is unclear if this is a conclusion or assumption -- "A progenitor cell barcode should become increasingly polymorphic with subsequent divisions." The "polymorphic" happens within the progenitors, their progenies, or their progenies at different time points.

The statement is now clarified as an assumption in the manuscript.

Similarly line 134 "Barcodes would record neuronal differentiation and migration." Is this a conclusion from this study or a citation? How is the migration part supported?

The reasoning is better explained in the manuscript. Migration can be documented if immediate daughter cells with similar barcodes are found in different parts of the adult brain, albeit analysis is confounded by sparse sampling and because barcodes may be similar by chance.

Line 148 and 150: "Nearest neighbor ... neuron pairs" in DNA methylation status would conceivably reflect their cell type-specific gene expression, how did the author distinguish this from cell lineage?

As noted above, because cells with similar phenotypes usually arise from common progenitors, cells within a clade are also usually related. However, the barcodes are still polymorphic within a clade and potentially add complementary information on mitotic ages, ancestry within a clade, and possible cell migration.

Figure 3C: "Cells that emerge early in development" Where are they on the figure?

Hindbrain neurons differentiate early in development and their barcodes are more methylated. The figure has been modified to label some of the values with their neuron types. Also, the older figure mistakenly included data from all 3 brains and now the data are only from brain H01.

Figures 4D and 4E, distinguishing cell subtypes is challenging, as the same color palette is used for both excitatory and inhibitory neurons.

Unfortunate limitations due to complexity and color limitations

Figures 4 and 5, what are these abbreviations?

The abbreviations are presented in Figure 1 and maintained in subsequent figures.