Large‐scale single‐cell RNA sequencing atlases of human immune cells across lifespan: Possibilities and challenges

Single‐cell RNA sequencing technologies have successfully been leveraged for immunological insights into human prenatal, pediatric, and adult tissues. These single‐cell studies have led to breakthroughs in our understanding of stem, myeloid, and lymphoid cell function. Computational analysis of fetal hematopoietic tissues has uncovered trajectories for T‐ and B‐cell differentiation across multiple organ sites, and how these trajectories might be dysregulated in fetal and pediatric health and disease. As we enter the age of large‐scale, multiomic, and integrative single‐cell meta‐analysis, we assess the advances and challenges of large‐scale data generation, analysis, and reanalysis, and data dissemination for a broad range of scientific and clinical communities. We discuss Findable, Accessible, Interoperable, and Reusable data sharing and unified cell ontology languages as strategic areas for progress of the field in the near future. We also reflect on the trend toward deployment of multiomic and spatial genomic platforms within single‐cell RNA sequencing projects, and the crucial role these data types will assume in the immediate future toward creation of comprehensive and rich single‐cell atlases. We demonstrate using our recent studies of human prenatal and adult hematopoietic tissues the importance of interdisciplinary and collaborative working in science to reveal biological insights in parallel with technological and computational innovations.


Introduction
The study of the immune system has historically used model organisms such as the mouse due to barriers to human tissue access [1].The resource landscape of immunological research has evolved, with access to human tissue, including developmental tissue, accelerated by national tissue biobanks [2][3][4], and with systems such as organoids allowing for molecular perturbation of human cells [5,6].Advancements in both molecular biology and computational tools have also revolutionized the resolution and scale of cellular profiling [7,8], allowing researchers to Correspondence: Prof. Muzlifah Haniffa and Dr. Simone Webb e-mail: mh32@sanger.ac.uk; simone.webb@newcastle.ac.uk confidently ascribe multiomic information to single cells at high throughput.These technological advances have fueled international consortium efforts to map single cells in the human body such as the Human Cell Atlas (HCA) [9], which now provides crucial coordination and infrastructure for immunology research into both prenatal and postnatal human tissues.
Blood and immune cells differentiate from early progenitors arising in the extra-embryonic yolk sac (YS) and subsequently from long-lived hematopoietic stem cells in the aorta-gonadmesonephros (AGM) [10][11][12][13].AGM and YS-derived blood and immune cells then seed the embryonic liver [14] and fetal bone marrow (BM) [10], which become the dominant sites of hematopoiesis in humans in the second trimester and postnatally, respectively (Fig. 1).The subsequent seeding by early The major hematopoietic tissues that contribute to lifelong blood and immune cell formation in humans.The aorta-gonad-mesonephros (AGM), yolk sac, liver, and bone marrow are the primary site of hematopoietic production, with seeding of blood and immune cells to peripheral organs such as the thymus and spleen.Arrows indicate the approximate post-conception week (PCW) age or stage at which the relevant tissues are dominant contributors to blood and immune cell formation.Image created using Biorender.com.lymphoid/thymic progenitors of the thymus and naive B cells of the spleen are crucially important for adaptive T-and B-lymphocyte differentiation and maturation (Fig. 1).
When considering the fundamental differences between human and mouse hematopoiesis, it is clear that the extent to which model organisms might recapitulate human immunobiology are limited; for example, human fetal hematopoiesis begins in the BM from the second trimester and is dominant in the BM in the third trimester of gestation [15], whereas mouse BM hematopoiesis only begins peri-or postnatally [16].The cellular and molecular characterization of the developing human immune system has been accelerated by single-cell technologies and computational biology approaches but how the developing human immune system is educated and equipped for life after birth remains to be resolved [17].

Immunological insights from single-cell atlasing: In human fetal and adult tissues
One way to investigate human prenatal immunity is through organ-level single-cell profiling.We can now benefit from com-prehensive organ-specific human single-cell atlases [7,18], which have described the presence and function of immune cells within a range of unique tissue microenvironments, including the aforementioned hematopoietic organs such as the YS, AGM, liver, BM, spleen, and thymus (Fig. 1).These studies have enabled us to construct a dynamic framework of the developing human immune system across organs and gestation time, which we summarize below (Table 1).

The yolk sac and aorta-gonad-mesonephros form the first blood and immune progenitors
Hematopoietic stem and progenitor cells (HSPCs) from the AGM have been recently profiled at single-cell resolution [13] and shown to express the definitive HSPC gene signature of RUNX1 + HOXA9 + MLLT3 + MECOM + HLF + SPINK2 + .The human YS has been profiled at single-cell resolution by multiple groups [12,19,20], impacting our understanding of developmental stem and myeloid cell biology.The comprehensive YS immune lineage was inspected in our 2022 preprint [12].By integrating 3-8 postconception week (PCW) YS single-cell RNA sequencing (scRNA-seq)

Single-cell study
Human fetal tissue Biological finding from single-cell study Botting et al. [12] Yolk sac Primitive to definitive hematopoietic stem and progenitor predominance after five postconception weeks Jardine et al. 2021 [15] Bone marrow Innate myeloid cell diversification and differentiation in second trimester Botting et al. [12]; Garcia-Alonso et al. [27]; Dann et al. [28]; Calvanese et al. [13] Yolk sac; testes; skin; aorta-gonad-mesonephros Microglial-like macrophage presence in non-CNS tissues Popescu et al. [20] Fetal skin Physiological erythropoiesis in skin Popescu et al. [20] Fetal liver Dominance of erythropoiesis in human fetal liver Popescu et al. [20] Fetal liver Inferred dual myeloid and lymphoid origin for pDCs Jardine et al. [15] Bone marrow From 11 postconception weeks, enrichment in B-cell progenitors in fetal bone marrow in contrast to adult bone marrow where naive B cells, memory B cells, and plasma cells dominate Jardine et al. [15] Bone marrow No evidence of significant clonal expansion of B-lineage cells in response to antigen Khabirova et al. [38] Bone marrow KMT2A-rearranged infant B-ALL was dominated by an early lymphocyte precursor state Jardine et al. [15] Bone marrow Neutrophils and their precursors, and DC subsets (pDC, tDC, and DC3) are present in fetal BM scRNA-seq data, but not in data from yolk sac or fetal liver Jardine et al. [15] Bone marrow Identification of neutrophils in human fetal bone marrow, independent of eosinophils and basophils Jardine et al. [15] Bone marrow Cytotoxic genes are expressed in fetal bone marrow natural killer cells, but not in fetal liver counterparts Dann et al. [28] Spleen Highly vascularized splenic structure with overlapping T-and B-cell zone aggregates Dann et al. [28] Liver; skin; spleen; thymus; bone marrow; gut; yolk sac; kidney; mesenteric lymph node Presence of B1 cells in human tissues for the first time, and detected B-cell progenitors in almost all prenatal hematopoietic organs data [12], we observed a transition from primitive to definitive YS HSPCs at 5 PCW, and the persistence of definitive HSPC signature in the fetal liver (FL) [12].Our in vivo findings of primitive and definitive HSPCs were also recapitulated in external induced pluripotent stem cell scRNA-seq datasets [21,13].In addition, we demonstrated a role for the human YS in metabolism, coagulation, and erythropoietin production to support the early embryo.When taken at large, independent interrogations of early hematopoietic organs can lead to unexpected findings.Conventional wisdom from fate mapping studies in mice holds that brain-resident macrophages, or microglia, are derived from YS erythro-myeloid progenitors at E7.5 [22], which seed the brain where they become tissue-adapted and repopulate the brain primordium by E9.5 [23].In humans, a similar process is expected to occur, with macrophage progenitors in the 2-3 PCW YS [20,24] demonstrated to seed the forebrain and telencephalon by 3 and 4 PCW, respectively [25,26].However, in our human YS atlas [12], macrophages co-expressing microglia-associated transcripts P2RY12, CXCR3, and OLFML3 were identifiable in the YS from 4 to 8 PCW.Notably, microglia-like macrophages were reported independently in other non-CNS peripheral organs such as fetal testes [27], fetal skin [28], and AGM [12,13].This revealed the possibility of a broader role for microglia beyond function in the brain, for example, in supporting YS endothelial cell angiogenesis [12] or in the protection of maturing germ cells in the testes through immunomodulatory cell-cell interactions [27].Although the understanding of the role for these microglia in these peripheral tissues remains to be validated, this illustrates the power of open science facilitating data sharing for cross-tissue analysis.

Fetal liver hematopoiesis dominates in second trimester of human development
Blood and immune progenitors from the AGM and YS seed the embryonic liver, which becomes the dominant site of hematopoiesis in the second trimester (Fig. 1) [10].In our 2019 profiling of the human FL [20], we observed erythropoiesis dominating to support organism-wide oxygen supply as a priority at this time.We examined hemoglobin usage profiles of YS and FL erythrocytes and noted a switch from embryonic (HBZ and HBE1) to fetal hemoglobin subunit usage (HBA1 and HBG2) between YS and FL.We uniquely showed physiological erythropoiesis in the skin at the same gestational age, which warrants further study for better understanding of distributed contributions to erythropoiesis.
In addition, myeloid cells were identified in the human FL scRNA-seq dataset.Of particular interest, we predicted a dual myeloid and lymphoid origin for plasmacytoid DCs (pDCs) consistent with reports in mice [29], in vitro cultures [30] and as later confirmed using scRNA-seq, mass cytometry, in vivo fate mapping, and in vitro clonal assays [29,31,32], with implications for our understanding of cellular differentiation hierarchies.More broadly, an improved understanding of DC origins and differentiation trajectories will clarify the links between DC ontogeny and function, thereby aiding future efforts to engineer DC-based cancer vaccines.

The BM is the main source of blood and immune cells from the third trimester
Following FL hematopoiesis, the human fetal BM begins to take over this role from 11 to 12 PCW into childhood and adulthood (Fig. 1) [33,34].Building on a wealth of single-cell research from postnatal mouse BM [35][36][37], our human fetal BM study revealed the rapid establishment of hematopoiesis in fetal femur within weeks and the complete differentiation of all cellular lineages including granulocytes [15].We observed enrichment in Bcell progenitors in fetal BM in contrast to adult BM where naive B cells, memory B cells, and plasma cells dominated.B-cell progenitor differentiation from early lymphoid progenitors, pre-pro B, pro B, pre-B, immature B, and naive B cells was evident in fetal BM from 11 PCW.Leveraging matched single-cell B-cell receptor (BCR) data, we found no evidence of significant clonal expansion in response to antigen.However, it would now be timely to generate multiomic maps of organs implicated in B-cell differentiation (such as the fetal BM and spleen) in later gestation and after birth in order to further understand the foundations of the adaptive immune response including B-cell somatic hypermutation and maturation in the germinal center.It would be advantageous from a clinical perspective to clarify how lymphoid receptor clonal expansion is patterned across fetal organs during human development, and what utility clonal expansion might offer against potential antigen challenge in utero and after birth.
Identifying the genetic basis of diseases implicated in dysregulation of B-lymphopoiesis might have an impact on clinical therapy.To this end, we used the fetal BM atlas to characterize the temporal and cell state-restricted patterns of expression of genes implicated in childhood cancers [15].Specifically, we integrated our reference data with bulk RNA-seq from KMT2A-rearranged infant B-cell acute lymphoblastic leukemia (B-ALL) to identify the cell states driving disease [38].We found that KMT2A-rearranged infant B-ALL was dominated by an early lymphocyte precursor state [38].We were then able to identify statistically upregulated cell surface antigens in early lymphocyte precursors from B-ALL versus reference human fetal BM [38], representing a therapeutic avenue for B-ALL, and more broadly, a methodology by which studies of this kind can be interrogated for identification of selective cancer immunotherapy targets.
In addition, we demonstrated that the human fetal BM is a crucial site for innate myeloid cell diversification and differentiation [15].We showed for the first time with scRNA-seq data that neutrophils and their precursors, and DC subsets (pDC, tDC, and DC3) are present in fetal BM scRNA-seq data, but not in data from YS or FL.This was also the first scRNA-seq study to capture neutrophils in human fetal BM, and clearly identify them independently of eosinophils and basophils.With regard to the innate lymphoid cells present in the human fetal BM, we find evidence that cytotoxic genes are expressed in fetal BM natural killer (NK) cells, but not in FL NK cell counterparts, indicating poised innate immunity in the fetal BM early as the second trimester.
There are a few unresolved questions from the fetal BM characterization, for example, whether the DC progenitor in the YS seeds the FL and gives rise to DC1 and DC2, and further, what the role of the pDC progenitor cell could be within this differentiation paradigm (i.e., whether the pDC progenitor is a more primitive or mature state of the DC progenitor).In future, resolution of these questions could be aided by lineage tracing studies, or by the application of multimodal profiling to study gene regulation; scATAC-seq of human fetal BM tissue has now been performed, indicating the potential in this area [39].In addition, further investigation into the extrinsic regulation of B-lineage cells by stroma and how this may be altered in disease warrants further study.

The human fetal thymus and spleen are important sites of T-and B-cell selection
T-cell differentiation occurs exclusively in the thymus from early thymic progenitor differentiation into double negative thymocytes, double positive thymocytes, and single-positive naive T cells.T-cell receptor (TCR) selection for αβ and ɣδ TCR occurs during the transition of thymocytes across these stages.The repertoire of T-cell formation was investigated in a large-scale study in 2020, which profiled human fetal, pediatric, and adult thymus samples, identifying over 50 distinct cell states and revealing human VDJ usage bias in various T-cell subsets [40].The understanding of the T-cell repertoire as proffered by this landmark paper will better our understanding of T-cell selection, and aid future development of thymic in vitro systems, with implications in engineering T-cell-based immunotherapies.
Beyond assessment of VDJ chain usage by cell state, lymphocyte receptor analysis would greatly benefit from the availability of robust TCR/BCR receptor:antigen databases [41]; however, the curation of these poses complex problems.The provision of such a database (which is admittedly perhaps more relevant to the study of adult tissues that are subject to heavy burden of antigen challenge) is not just restricted to academic research interests.Commercial organizations are currently exploring the benefits of TCR sequencing, for example, Microsoft and Adaptive Biotechnology's Antigen Map Project [42]: a far-spanning project like the HCA, which aims to use machine learning to better understand TCR sequences and predict disease correspondence.Toward this effort, publicly available single-cell transcriptomic data could be re-analyzed in future with additional chromatin and/or proteome data to help further understand the role of thymus in acquired and congenital T-cell deficiency syndromes, such as Wiskott Aldrich syndrome.
BCR selection occurs in the BM but final maturation of B cells into antibody secreting cells following antigen exposure occurs in the spleen.The 16-18 PCW spleen has been profiled using scRNAseq and spatial genomics methods, revealing a highly vascularized splenic structure with overlapping T-and B-cell zone aggregates [28].Analysis of myeloid cells in the fetal spleen suggested that iron-recycling erythrophagocytic macrophages might have conserved features with their adult counterparts (unlike other prenatal macrophages), indicating that this iron-recycling profile is retained with great fidelity from fetal into adult life.Surprisingly, the meta-analysis of human fetal hematopoietic tissues in this study [28] also revealed the presence of B1 cells in human tissues for the first time and detected B-cell progenitors in almost all prenatal hematopoietic organs, underscoring the power of integrative analysis to capture emergent biology.This study [28] was included in a Science journal bundle alongside other pan-human single-cell publications [43,44], foregrounding a wider trend in the single-cell community toward population-scale analysis for emergent biological insights.

Adult immune cell profiling across tissues
To interrogate the gene expression profile of immune cells in adult life, Domínguez et al. analyzed large-scale single-cell transcriptomic data from 12 adult organ donors covering 16 different tissues [43].Their data leveraged a new logistic regression tool for automated cell type classification in order to detect over 100 immune cell states, with both the tool (CellTypist) and the dataset now serving as an aid and robust reference to the immunology community.Similar large-scale projects have contributed to our understanding of immune cells, such as that of the Tabula Sapiens consortium, which profiled single cells from 15 donors spanning 24 different tissues, identifying over 400 cell states [45].Around half of the Tabula Sapiens dataset comprised immune cells, representing a large resource for future re-analysis and biological insights in the field of immunology.
Of note, analysis of adult and fetal human tissues using singlecell technologies has not been performed in a silo, and some teams have focused their efforts on profiling the shared maternal and fetal microenvironment.For example, the maternal-fetal interface has now been characterized [46], with the research team's computational focus on cell-cell interactions further endowing the single-cell community with the CellPhoneDB tool [47], allowing for fuller appreciation of cellular crosstalk within scRNA-seq datasets that are derived from cell suspensions.The endometrium has also been profiled across all stages of the menstrual cycle at single-cell resolution [48] and has defined a small compartment of immune cells including mainly NK cells, T cells, monocytes, and DCs.Although studies profiling peripheral and cord blood are numerous, further characterization of umbilical cord tissue would enable research into mesenchymal stem cell origins and maintenance, with impacts on mesenchymal stem cell-based therapies.
The human immune system is understood to decline in function with age, leading to impaired response to vaccinations [49] and increased risk of autoimmunity [50].In ageing adults, hematopoietic stem cell loss and dysfunction is observed and may be implicated in lifespan and healthspan [51].Thymic production of T lymphocytes is seen to decrease significantly with age [52], although a decrease in B-lymphopoiesis is less unanimously evidenced in humans [53,54].The prevalence for somatic mutations in blood or BM cells (also known as clonal hematopoiesis of indeterminate potential) increases with age, with a prevalence of ∼10% for those aged 70-80 years [55].Given the striking differences that have been observed in the immune system of older individuals, taking age metadata into formal consideration when designing single-cell experiments and analyzing the resultant data would be advantageous.Ageing tissues in mice have comprehensively been profiled with scRNA-seq by the Tabula Muris Consortium [56], and other murine studies [57] have identified lineage biases [58], clonal expansion [59], and epigenetic [59,60] and niche-related changes [36,37] with age; there would be a clear benefit to taking this to the human context and for concerted efforts toward atlasing of tissues from ageing human adults, which will benefit from cross-species expertise and coordination now spearheaded the Aging Atlas Consortium [61].

Future directions and challenges in building human immune atlases FAIR data sharing
Reference atlases, such as those mentioned above, are increasingly being re-used by the community for further insights across tissues and across disease conditions.However, steps must be taken to ensure that data are truly re-usable for the public (Fig. 2).Data repositories such as EBI ArrayExpress [62] and NCBI GEO [63] are widely used for deposition of raw genomics data, but since data deposition is not mandatory all for published researched articles, the degree of data availability is often more of a spectrum in practice.For example, some studies might deposit raw FASTQ files, transformed count matrices, and associated metadata, while others might simply note that data are available upon request.Web portals that allow for intuitive visualization of single-cell data are also increasingly being provided alongside atlasing publications, allowing for data exploration and re-use by noncomputational beneficiaries [64].Code availability is also crucial for adherence to FAIR (Findable, Accessible, Interoperable, and Reusable) data principles [65]; GitHub and Zenodo platforms are leading the charge in this respect.A combination of bioinformatics training and clear guidelines on data stewardship within both research groups and publishing bodies will need to be prioritized to overcome these issues-toward this aim, some are working to provide platforms to alleviate this burden, such as the HCA Data Coordination Portal [66] and EMBL-EBI HCA project catalogue [67].In addition, community-based efforts such as the scverse consortium for maintenance of foundational tools in single-cell omics analysis [68] and others [69] contribute to guidelines and support for FAIR data principles.

A unified cell ontology language
Work toward a common cell label language or ontology would particularly benefit the immunology community when considering reuse of immune cell atlases [70].Integration efforts are especially difficult where independent researchers have annotated single cells based on nomenclature that are used within small communities, with no associated resources to clarify nomenclature, such as a marker gene list with references.This can lead to the unfortunate instance where a novel and highly interesting and unique cell state in one dataset is removed during the integration process due to having less perceived ontological evidence, than say, a less specific but more understandable broad cell type annotation in another dataset.A unified ontology would aid this effort, where, for example, every cell state annotated in a scRNAseq dataset was linked to an ontological ID from a widely used database; the EBI have developed the Ontology Lookup Service to perform this very function [71].Finally, computational expertise to define unique and high-quality cells in scRNA-seq data will be invaluable and will be required in conjunction with immunological expertise for insight into (i) traditional hierarchies and ontologies that might be recapitulated in a given dataset, and (ii) a common cell label language that can contribute to robust reference profiles for immune cells.

Multiomic techniques
The field of single-cell immunology is increasingly embracing multiomic experimental design in efforts to characterize cells in a manner more reflective of RNA/DNA/protein regulation [7,8].Computational methods have been developed [72] in order that the cell-state-discriminative protein combinations from multiomic RNA and surface protein technologies such as CITE-seq [73] can inform gating strategies widely used in flow sorting for immune cells, which is particularly helpful for rare cell states and less characterized tissues.Technologies for combined nuclear RNA and chromatin accessibility profiling such as scATAC-seq [74] hold promise to uncover the role of genomic structure in immune lineage priming, although integration of these datasets with cytoplasmic RNA counts in scRNA-seq datasets, unsurprisingly, poses some challenges [75].This example provides a compelling case for multiomic experimental design, where nuclei from previously inaccessible tissues and from cell states too large for droplet-based whole-cell technologies (e.g., adipocytes and myocytes) might be better captured with nuclear technologies [44].With richer multiomic single-cell data now available for a range of immune lineages and tissues, functional assay technologies will need to keep pace in order that a range of molecules in single cells can readily be validated in vitro, as has recently been performed for myelopoiesis with an organoid culture system [21].

Meta-analysis of data
Meta-analysis of large and even population-scale scRNA-seq datasets is another emerging trend in the field and will provide valuable insights into disease (e.g., source of genetic variants) and demographic differences in immune profiles (e.g., age and sex).Although many studies are limited in biological replicates for true population-scale meta-analysis, a recent surge in tools to link scRNA-seq datasets with genome-wide association study summary statistics may provide a bridge between fields in the meantime [76,77].In addition, perhaps owing to the size of developmental samples being more feasible, whole organism atlases for developing organisms have been produced, including for mouse gastrula [78], monkey gastrula [79], rabbit embryo [80], and human embryos [81], with immune cells profiled in all.These whole organism atlases will empower us toward immunological insights with tissues from the same donor, elevating our understanding of distributed systems such as hematopoiesis.

Spatial genomics
On a final note, an important final frontier for the single-cell immunology community will be the adoption and development of spatial technologies, which already allows researchers the ability to position RNA molecules to subcellular resolution across tissue biopsies [82].Spatial genomics will allow for unprecedented interrogation of cell-cell colocalization and lend further evidence to putative cell-cell interaction from receptor-ligand-based algorithms currently applied to suspension datasets [47,83].Computational prediction of transcriptomic profiles using deep learning on histology images is also an active field of research, with implications in clinical practice and disease stratification [84].A blue-sky end-goal might be integration of spatial transcriptomic data from sequential tissue sections toward creation of a three-dimensional transcriptomic map of a whole tissue, or even a whole organism, although adaptations of current technologies might be required in order that such a dataset would be most beneficial for the immunology community specifically (e.g., provision of TCR or BCR RNA enrichment in spatial genomics protocols).

Conclusions and future perspectives
The immunology community has seized the tools and opportunities made available through single-cell technologies to profile immune cells in human fetal and adult life, and in health and disease.There now exists a wealth of single-cell data on immune cells, alongside community initiatives to fund, coordinate, and guide these efforts.To date, studies have resulted in breakthroughs in myeloid, lymphoid, and stem cell biology, with the potential for meta-analysis of tissue studies and new whole organism studies promising to elevate these biological findings further.
In future, the ability conferred by meta-analysis of tissue atlases to investigate immune cell interactions with stromal cells and the neuro-immune axis will lead to unprecedented findings and shifts in our understanding of immune cell function.It would be greatly beneficial to the single-cell immunology community to move toward this new era in a manner that views immune cells not just in the dissociated cell suspension as we have become accustomed within the world of scRNA-seq, but using multiomic approaches as well as new spatial transcriptomics methods [85] for a view that is rich, inclusive of environment, extrinsic signals, and places immune cells within their context.

Figure 1 .
Figure 1.The major hematopoietic tissues that contribute to lifelong blood and immune cell formation in humans.The aorta-gonad-mesonephros (AGM), yolk sac, liver, and bone marrow are the primary site of hematopoietic production, with seeding of blood and immune cells to peripheral organs such as the thymus and spleen.Arrows indicate the approximate post-conception week (PCW) age or stage at which the relevant tissues are dominant contributors to blood and immune cell formation.Image created using Biorender.com.

Figure 2 .
Figure 2. Key challenges to overcome when producing single-cell immune atlases in the near future.A visual summary of the topics discussed for the future perspectives component of this review article, including FAIR data sharing, a common cell ontology language, meta-analysis of data for continued biological insights, and the expanded use of multiomic and spatial genomics techniques.Image created using Biorender.com.
Table containing a summary of the biological insights gained from single-cell studies on human fetal tissues