Kidney organoid reproducibility across multiple human iPSC lines and diminished off target cells after transplantation revealed by single cell transcriptomics

Human iPSC-derived kidney organoids have the potential to revolutionize discovery, but assessing their consistency and reproducibility across iPSC lines, and reducing the generation of off-target cells remain an open challenge. Here, we used single cell RNA-Seq (scRNA-Seq) to profile 415,775 cells to show that organoid composition and development are comparable to human fetal and adult kidneys. Although cell classes were largely reproducible across iPSC lines, time points, protocols, and replicates, cell proportions were variable between different iPSC lines. Off-target cell proportions were the most variable. Prolonged in vitro culture did not alter cell types, but organoid transplantation under the mouse kidney capsule diminished off-target cells. Our work shows how scRNA-seq can help score organoids for reproducibility, faithfulness and quality, that kidney organoids derived from different iPSC lines are comparable surrogates for human kidney, and that transplantation enhances their formation by diminishing off-target cells.


Introduction
Kidney diseases affect ~800 million people worldwide (Coresh et al., 2007). Despite the enormous disease burden, therapeutic innovation has lagged (Inrig et al., 2014), owing in part to the lack of appropriate models that reflect the cellular complexity of the human kidney.
To harness the full potential of iPSC derived kidney organoid technology, we must address critical unsolved questions about their reproducibility, faithfulness, and quality. First, we must establish organoid reproducibility: the comparability and range of variability in cellular composition and state between different iPSC lines from normal individuals across replicates and protocols (building on previous efforts to draw comparisons between iPSC and embryonic stem cell (ESC) derived organoids (Wu et al., 2018) or bulk RNASeq data comparing iPSC derived organoids (Phipson et al., 2019)). This is critically important, because individual patient iPSCs offer significant advantages over ESCs for precision medicine and drug development projects (Boreström et al., 2018;Burrows et al., 2016;Takasato et al., 2016b). Second, we must define their faithfulness: how well organoids across many iPSC lines recapitulate kidney development and disease-associated genes at single cell resolution. Third, we need to define their quality. Since off-target cells interfere with organoid quality, we must understand how to drive their removal to more faithfully reproduce the human kidney (Wu et al., 2018). portion of a tumor nephrectomy from an adult male (with no known kidney diseases) (Fig. 1C).
We determined the cellular composition of organoids using unsupervised graph-based clustering followed by post-hoc annotation (Methods) with signatures of cell types ( Supplementary Fig. S2, Supplementary Table 1) and cell cycle genes (Kowalczyk et al., 2015). (Na-Cl symporter; Supplementary Fig. 2), a canonical marker of the distal convoluted tubule (DCT). The organoid single cell profiles retained the proximal (podocyte) to distal axis of the human nephron (Fig. 1C, left) on visualization of the data using t-distributed Stochastic Nonlinear Embedding (tSNE), unlike the discrete clusters seen in adult kidney (Fig. 1C, right).

Mature organoids most closely resemble fetal kidney
We related the cell clusters from D29 human kidney organoids (ThF line) to human tissue by comparing them to fetal kidney from the first (8 weeks) and second (17 weeks) trimesters (Lindström et al., 2018b;, and to adult kidney (Fig. 1C, right). We used a classification based approach (random forest classifier) (Pandey et al., 2018) to assess the relation between each cell cluster in these tissues to each organoid cluster ( Fig. 1E-G,

Methods).
Overall, cells were most similar to those from first and second trimester fetal kidneys, largely consistent with previous studies using bulk RNA-Seq data (Takasato et al., 2015b). All nephron lineages were accurately classified by the algorithm (Fig. 1E-G). Compared to those in adult human kidney (Fig. 1C), all organoids contained cells of human nephron segments, but with poorly differentiated distal tubular cells. In contrast, in the adult kidney, we identified distinct proximal, loop of Henle, DCT and CD sub-clusters, including principal cells (PC) and α-and βintercalated cells (IC), as well as endothelial and immune cells (Fig. 1C, Supplementary   Fig.3B).

Cell proportions vary depending on iPSC line
Cell type proportions in mature D29 organoids were consistent between replicates of independent clones, but they varied between iPSC lines (Fig. 3A, B). We quantified the For example, podocytes were captured across all experimental conditions ( Fig. 3A and Supplementary Fig. 5B), but varied between 1.33% (N1, averaged across replicates) to 2.58% (ThF) between lines on the ML protocol. AS, passage 1 (AS-1) was an outlier, with lower overall nephron numbers (0.3%). The JB protocol captured an average of 0.81% podocytes. Similarly, the distal nephron compartment (including TAL and GATA3+ distal nephron cells) ranged in average proportion from 2.34% (N1) to 6.95% (ThF), with an average 1.45-fold-change between protocols. In general, N1 organoids had the lowest average nephron cell proportions, followed by AS, N2 and ThF. GATA3+ distal-like cells were more abundant in ThF iPSCs, independent of protocol; GATA3 expression in AS and N1 was lower than in ThF and N2 (Supplementary Fig.   2), as confirmed by IF (Fig. 2C).
To make higher level comparisons, we looked at 4 groups: nephron, mesenchymal, off-target and endothelial compartments (Fig. 3B, Supplementary Table 3). The nephron compartment was on average 16.7%, of all cells (9.89% in N1 to 24.1% in ThF). The N1 line had the largest average relative proportion of endothelial cells (0.1%), with a global average of 0.05% across all lines. Mesenchymal cells were 82.8% of all cells in AS and 88.6% in N1 organoids, but only 64.2% in ThF and 39.2% in N2. Off-target cells varied markedly by iPSC line, from 1.6% in AS-1 and N1 to 43.7% in N2 (Fig. 3B). To determine variability within each compartment, we again computed the JSD (Fig. 3D). The nephron and mesenchymal compartments were consistent (nephron mean JSD between lines = 0.05, sd = 0.038; mesenchymal mean JSD = 0.07, sd = 0.05), whereas the off-target compartment was variable between lines (mean JSD = 0.14, SD = 0.09) and protocols (Fig. 3D). N2 organoids were most divergent from AS in off-target composition. Notably, the ratio of mesenchymal to nephron cells was inversely related (Spearman correlation r = -0.84) to the proportion of off-target cells. Higher proportion of offtargets (N2, ThF:11.7%, JB ThF: 12.2%) resulted in lower mesenchymal:nephron ratios (N2: 2.29, ThF: 2.66, JB ThF: 4.05) and vice-versa (Off-targets -N1, AS:1.59%; mesenchymal: nephron -N1: 8.96, AS: 5.31). In contrast, the mesenchyme proportions were lower in both adult and fetal kidney, both overall (15.7% in adult, 19.7% in fetal week 17) and relative to nephron cells (0.65 in adult, 0.47 in fetal). In summary, we noticed greater organoid heterogeneity between iPSC lines than between replicates within a line, or between protocols, with the off-target compartment contributing the most variability.

Variability in cell type proportions detected by scRNA Seq in D15 organoids
To explore whether the iPSC-line associated differences were associated with their basal state (Burrows et al., 2016;Féraud et al., 2016) at D0 or the process of differentiation, we compared the single cell profiles collected at D0,7, and 15 for each organoid culture.
Analysis of 42,433 single cells from the 4 undifferentiated iPSC lines at D0 (Fig. 3E, H and Supplementary Fig. 6A), showed they were comparably pluripotent. To determine if there were clusters of cells primed towards a particular developmental germ layer, we scored organoids using known transcriptional signatures for the three germ layers. We did not observe subclusters with signatures for any specific germ layer (Tsankov et al., 2015), or primed for differentiation (Nguyen et al., 2010) (Supplementary Fig. 7A). The four cell clusters had representation from each of the iPSC lines (avg JSD = 0.05, sd = 0.04, Fig. 3H, I), and from all cell-cycle phases (Supplementary Fig. 7B).
Similarly, at D7, developing organoids from all iPSC lines expressed appropriate markers of mesodermal differentiation with actively proliferating cells (Kowalczyk et al., 2015) ( Supplementary Fig. 7C, D) and little variability between lines (Fig. 3F, H there were notable differences in composition between iPSC lines (avg JSD = 0.14, sd = 0.07;

Fig. 3G, H, I).
Heterogeneity in endothelial progenitors was consistent with D29: N1 had the highest average relative proportions (0.32% vs 0.05% among other lines; Fig. 3J). The proportion of nephron-like cells ranged from 0.08% in AS to 56.9% in N2 D15 organoids ( Fig.   3H, J). Although, no distinct off-target cells were found at D15, N2 and ThF organoids had a small population of cells expressing the neuronal progenitor SOX2 within the mesenchymal compartment ( Supplementary Fig. 8B). The mesenchyme:nephron ratios were also lower in N2 and ThF D15 organoids (Fig. 3J). Hence, organoids with higher nephron proportions at D15 had a distinct pool of SOX2+ cells, and went on to develop a higher proportion of off-target cells at D29, associated in turn with lower D29 mesenchyme:nephron ratios (Fig. 3B, J). Taken together, variability in organoid differentiation was evident by D15.

Concordant expression of developmental programs across organoids from 4 human iPSC lines
Next, we tested whether known transcription factors (TFs) and critical genes involved in nephrogenesis (Little and McMahon, 2012;Little et al., 2016) are appropriately expressed in the respective clusters in organoids from 4 different iPSC lines, compared to adult kidney ( Fig. 4A,   Supplementary Fig 9,10). Overall, key developmental programs were expressed at expected time points and transitions, in a comparable manner across different iPSC-derived organoids ( Fig. 4A, Supplementary Fig. 9,10). First, core pluripotency TFs were expressed during the iPSC stage and decreased subsequently (D7-D29). Notably, SOX2, a neuronal progenitor marker (Graham et al., 2003), re-appeared in D29 organoids in off-target neuron cells (Fig. 4A).
Organoids from all lines reproducibly expressed the majority of genes associated with progressive kidney diseases (including adult onset diseases) in at least one compartment.
Genes associated with embryonic and early childhood abnormalities (CAKUT and HRC) were broadly expressed in developing organoids from all iPSC lines as early as D7 (i.e. CHD1L(Brockschmidt et al., 2012) at D7). Appropriately, developmental genes enriched in organoids were absent in adult kidney (i.e. RET, HPSE2). Some genes achieved high levels of expression and cell-type specificity in mature organoids, in a pattern similar to adult kidney. For example, PTPRO (from CKD GWAS and monogenic glomerular diseases) was appropriately expressed in D29 and adult human podocytes (Wharram et al., 2000) (Supplementary Fig. 13,   14). Similarly, PAX2 and MUC1 (from CAKUT and cystic diseases, respectively) were highly expressed in D29 distal tubules, similar to their human adult expression pattern (Leroy et al., 2002) (Fig. 5A, B; Supplementary Fig. 11,12). We validated these markers derived from scRNA-Seq analysis by IF staining, which confirmed the co-expression of PAX2 and MUC1 in CDH1+ distal tubular epithelial cells in mature organoid sections (Fig. 5C). In summary, kidney organoids from 4 different iPSC lines could serve as reasonably faithful surrogates of human kidney tissue for the study of a broad array of kidney diseases.

Organoid transplantation in mice diminishes off-target cells and enhances organoid quality and composition
We hypothesized that the reduction of off-target cells may improve organoid composition and overall quality. First, we determined that prolonged organoid culture in vitro, up to 51 days, did not reduce the portion of off-target cells, as assessed by scRNA-seq of organoids grown in vitro at day 32 (D32) and day 51 (D51) (Fig. 6A, B; Supplementary Fig. 15). D32 and D51 organoids had most populations present in D29 organoids, including muscle, neuronal and melanoma off-target cell clusters. In particular, we noted 2 clusters of neuronal off-target cells at D29, D32 and D51: STMN2+ neuron-like cells (D29, D32) and SOX2+NTRK2+ neuronalprecursor cells (D29, D32, D51; Supplementary Fig. 2, 15B).
The transplanted organoid epithelial cells had increased expression of genes suggestive of more mature states compared to in vitro D32, D29, or D51 controls. Specifically, both KLF6, a TF involved in the development of the ureteric bud and the kidney collecting duct (Fischer et al., 2001), and the Notch effector HES1, which plays a role in proximal to distal patterning in kidney nephrogenesis (Piscione et al., 2004), were upregulated in MAL+ distal cells from D32 transplanted organoids (Fig. 6E).
Strikingly, we found little off-target expression of SOX2 (neuronal precursor cells) or PMEL (melanoma cells) in D32 transplanted organoids compared to controls, suggesting that transplantation diminished off-target cells ( Fig. 6F; Supplementary Fig. 17A). MYOG positive muscle cells persisted in D32 transplanted organoids (Fig. 6F, Supplementary Fig. 17A). We also noted that a rare STMN2+ neuronal cluster persisted in D32 transplants. Interestingly, this cluster was uniquely and highly correlated with the neuronal cluster in week 17 fetal kidney (Spearman r = 0.82; Fig. 1F, Supplementary Fig. 17B). A set of enriched genes (GAL, CHGA, CHGB(Dressler, 2006); Supplementary Fig. S18) was shared between this cluster in fetal kidney and in D32 transplanted organoids (Supplementary Fig. S18A). Further analysis revealed that in fact CHGA and CHGB were detectable in a small number of cells within the STMN2+ cluster in D29 control organoids (Supplementary Fig. S18B) and transplantation selected for this STMN2+/CHGA+/CHGB+ cluster (Supplementary Fig. S18A), suggestive of a more fetal-kidney-like state.
We applied a classifier to further assess the relation of all cell clusters in transplanted organoids to organoids grown in vitro (Fig. 6G). Consistently, no transplanted organoid cells

Discussion
In this study, we present a comprehensive atlas of human kidney organoids in comparison to human adult and fetal kidneys at single cell resolution spanning multiple iPSC cell lines, time points, protocols, and replicates including transplantation into mice, at average sequencing depths of 10,000 reads per cell in a total of 415,775 cells. This analysis provided answers to several critical questions regarding organoid reproducibility, faithfulness and quality. Second, the applicability and faithfulness of organoids as a tool for discovery was bolstered by the fact that the vast majority of genes associated with kidney diseases was expressed in organoids across all four iPSC lines in the expected, corresponding cell types. For example, we validated the expression of MUC1 in the distal tubular compartment of D29 organoids.
Mutations in MUC1 are a cause of autosomal dominant tubulointerstitial kidney disease (Kirby et al., 2013;Živná et al., 2018), a rare kidney disease without a cure. Hence, efforts to find cures for genetically defined diseases may benefit greatly from our ability to study their mechanisms in patient iPSC-derived organoids, and this study provides an important foundational step in this direction. Specifically, our work suggests that organoids derived from individual patients can indeed be used as a tool to fuel biological discovery and therapeutics. However, single cell transcriptomics using comprehensive atlases such as this may be an essential reference when trying to make meaningful comparisons between iPSC-derived kidney organoids from different patients.
Finally, we addressed the critical question of organoid quality, with a focus on reduction of offtarget cells. Recent reports focused on eliminating neuronal off-target cells from kidney organoids by using an NTRK2 blocker (Wu et al., 2018). However, NTRK2 is not only expressed in off-target neuronal cells, but it is also abundantly expressed in podocytes (Caroleo et al., 2015;Li et al., 2015) in both fetal human kidneys (Metsuyanim et al., 2009) and in developing organoids (Supplementary Fig. 18A), raising the concern that NTRK2 blockers applied to organoid cultures may adversely affect podocyte differentiation and function. Here we identified organoid transplantation as an alternate approach to diminish off-target cells. Remarkably, organoid transplantation under the mouse kidney capsule diminished SOX2+ neuronal precursors and PMEL+ melanoma cells, in contrast to organoids grown in vitro, that retained a neuronal precursor SOX2+ off-target population. While some of these observations merit further studies, these data show that transplantation reduced off-target cells and improved organoid quality and maturity. Future studies will be required to test if earlier organoid transplantation, as early as D7, may eliminate off-target cells and enhance organoid formation, especially ureteric bud and collecting duct development (building on gene programs such as uniquely upregulated KLF6 and HES1 in transplanted organoids). As it stands, our protocols may benefit from incorporating organoid transplantation into rodents. Since endothelial progenitors were identified in D15 organoids, and endothelial cells were detected in D29 organoids, we can further speculate that improved vascularization may be achieved by preserving and expanding human endothelial cells in organoids in the hopes of eliminating all off-target cells and generating even more faithful, adult-kidney-like organoids.
In conclusion, these studies provide unprecedented single cell resolution into iPSC-derived kidney organoids, illuminating their reproducibility, pinpointing the source of variability, and showing that transplantation is a novel method for reducing off-target cells and improving organoid quality. This comprehensive atlas, uniquely enabled by scRNA-Seq technology, should serve as a foundational resource for the community, fueling the development of much needed therapies for patients with kidney diseases.