A non-disruptive and efficient knock-in approach allows fate tracing of resident osteoblast progenitors during repair of vertebral lesions in medaka

In bone remodeling and fracture repair, osteoclasts resorb old or damaged bone while osteoblasts synthesize mineralized matrix and mediate the formation of new bone (Einhorn and Gerstenfeld, 2015; Hadjidakis and Androulakis, 2006). Defects in the recruitment of osteoblasts to sites where bone formation is needed result in reduced bone density, increasing the risk of low bone mass disorders such as osteoporosis as well as leading to compromised fracture repair (Dirckx et al., 2013; Thiel et al., 2018). It has been shown that several chemotactic factors can guide the migration of osteoblasts and their progenitors (Colciago et al., 2009; Hengartner et al., 2013; Nakasaki et al., 2008; Otsuru et al., 2008). However, molecular and cellular mechanisms regulating the recruitment of osteoblasts to bone formation sites have not been completely elucidated (Bragdon and Bahney, 2018; Thiel et al., 2018). Identification of molecular cues driving osteoblast trafficking remains challenging, as the origin of recruited osteoblasts is not clearly defined (Bragdon and Bahney, 2018).

Zebrafish (Danio rerio) and medaka (Oryzias latipes) have been widely used as models in bone research (Lleras-Forero et al., 2019; Rosa et al., 2021; Tonelli et al., 2020). Components of the mineralized bone matrix are similar in teleosts and mammals (Lleras-Forero et al., 2019; Witten et al., 2017). In addition, molecular pathways governing differentiation of osteogenic cells are highly conserved (Valenti et al., 2020; Witten et al., 2017). We have previously generated transgenic medaka where inducible overexpression of receptor activator of NF-κB ligand (Rankl) leads to degradation of bone matrix in the medaka vertebral column (To et al., 2012). Subsequently, collagen10a1 (col10a1)-expressing osteoblast progenitors accumulated at the bone lesion sites and differentiated into osterix (osx)-expressing mature osteoblasts, facilitating re-mineralization of induced lesions (Renn et al., 2013). The origin of these recruited col10a1 osteoblast progenitors, however, remained unknown.

Cre/loxP recombination-mediated cell lineage tracing is a popular technique used to identify cellular sources contributing to tissue repair and regeneration (Dasyani et al., 2019; Fu et al., 2020; Geurtzen et al., 2014; Hu et al., 2017; Tu and Johnson, 2011). As Cre/loxP recombination is irreversible and heritable, all recombined cells and their progeny can be permanently labeled, allowing tracking of migration, proliferation and differentiation of cell lineages in vivo (Centanin et al., 2014; Livet et al., 2007; Pan et al., 2013). The precision of this approach depends highly on spatiotemporal control of Cre expression in the targeted population of cells (Song and Palmiter, 2018). In zebrafish and medaka, several transgenic Cre driver lines have been generated using either the Tol2 transposon system or I-SceI meganuclease (Dasyani et al., 2019; Knopf et al., 2011; Lee et al., 2014). However, these transgenesis approaches are limited as transgenes are often randomly integrated into the genome, potentially resulting in position effects and ectopic transgene expression (Kondrychyn et al., 2009; Thermes et al., 2002). For accurate cell lineage analyses, Cre driver lines are needed that recapitulate endogenous gene expression patterns with high fidelity. In zebrafish, two previous studies have reported the generation of endogenous Cre drivers by CRISPR/Cas9-mediated knock-in (Almeida et al., 2021; Kesavan et al., 2018). CRISPR/Cas9 was used to induce a site-specific double-stranded break in the genome and the inherent DNA repair mechanisms of the cell were exploited to insert CreERT2 into the endogenous gene loci (Almeida et al., 2021; Kesavan et al., 2018). However, these knock-in approaches resulted in a deleterious effect on the endogenous gene of interest, either reducing gene expression (Kesavan et al., 2018) or disrupting the gene-coding sequence (Almeida et al., 2021).

Here, we therefore used a novel CRISPR/Cas9 knock-in approach to generate a medaka col10a1 Cre driver line. A p2a linker was used for bicistronic expression of col10a1 and a 4-hydroxytamoxifen (4-HT)-inducible CreERT2, and homology-directed repair (HDR) was used to replace the col10a1 stop codon with a p2a-CreERT2 knock-in cassette. The knock-in cassette also included a cardiac myosin light chain 2 (cmlc2):EGFP reporter, which allowed efficient screening of knock-in founders. To show reliability of this approach, we also knocked in p2a-CreERT2 with similar efficiencies at loci of two other genes, odd-skipped related transcription factor 1 (osr1) and collagen2a1a (col2a1a). In mammals, Col10a1 is expressed specifically in hypertrophic chondrocytes (Shen, 2005), while col10a1 in teleosts is expressed in osteoblast progenitors, osteoblasts and chondrocytes (Eames et al., 2012; Renn et al., 2013; Renn and Winkler, 2012). Here, we show that CreERT2 expression in col10a1^p2a-CreERT2 medaka recapitulated endogenous col10a1 expression spatiotemporally. In addition, skeletal development and survivability were unaffected in both heterozygous and homozygous col10a1^p2a-CreERT2 medaka. By crossing col10a1^p2a-CreERT2 medaka with transgenic loxP reporter lines, we show that CreERT2 in col10a1^p2a-CreERT2 is functional and can be induced by 4-HT to catalyze efficient recombination and specific labeling of col10a1⁺ cells in larvae and adult fins. Finally, using col10a1^p2a-CreERT2 cell labeling, we tracked the dynamics of col10a1⁺ osteoblast progenitors after Rankl induction. We show that col10a1⁺ osteoblast progenitors residing at neural arches are recruited to bone lesion sites in the vertebral centra where they contribute to re-mineralization. This identifies col10a1⁺ cells as a resident source of osteoblasts that facilitate bone repair.

To knock-in p2a-CreERT2 into the medaka col10a1 locus, a CRISPR/Cas9-mediated homology-directed repair (HDR) strategy was used. A sequence positioned 37 base pairs (bp) upstream of the endogenous col10a1 stop codon was selected as the knock-in target site for the guide RNA (gRNA) (Fig. 1, indicated in magenta). To facilitate HDR, a 292 bp sequence upstream and a 215 bp sequence downstream of the col10a1 stop codon were cloned into the donor plasmid as 5′ and 3′ homology arms, respectively. These homology arms flank a 4.8 kb sequence, which consists of the p2a-CreERT2 coding sequence, as well as a secondary fluorescence reporter with a cmlc2 promoter driving the expression of EGFP. At the 3′ ends of both EGFP and CreERT2, a stop codon and SV40 polyA (pA) signal were inserted. To prevent CRISPR/Cas9-cleavage at the col10a1 gRNA site in the donor plasmid, PCR primers with template-primer mismatches were used to incorporate wobble bases (Fig. 1, magenta asterisks). Two mbait gRNA sites, previously reported to enhance knock-in efficiency (Kimura et al., 2014), were also included in the donor plasmid (Fig. 1).

To generate the col10a1^p2a-CreERT2 knock-in line, wild-type medaka embryos were injected with the donor plasmid, Cas9 nuclease, col10a1 crRNA (targeting the col10a1 gRNA site), mbait crRNA (targeting the mbait gRNA site) and tracrRNA at the one-cell stage (Fig. 2A). Injected F0 embryos were raised to 12 days post-fertilization (dpf) and screened for cmlc2:EGFP expression. Twenty-four out of 338 (7.1%) injected embryos showed mosaic cmlc2:EGFP expression and were raised as potential founders (Fig. 2B). Potential founders were out-crossed to wild-type fish and F1 progenies were screened for cmlc2:EGFP expression. Three out of seven (42.9%) potential founders transmitted the cmlc2:EGFP transgene to the next generation (Fig. 2B,C). To confirm precise knock-in at the col10a1 locus, PCR was performed using a forward primer that binds to medaka genomic DNA at the col10a1 locus and a reverse primer that binds to the donor plasmid (Fig. 1, yellow arrows, FP1, RP1). For cmlc2:EGFP-postive F1 offspring from two out of three potential founders, the expected 505 bp band was observed (Fig. 2B,D). Sequencing of the 505 bp amplicon revealed precise integration of the donor template into the col10a1 locus at the 5′ homology region (Fig. 2E), identifying two col10a1^p2a-CreERT2 founders. Sequencing analysis also confirmed precise knock-in at the 3′ homology region (Fig. S1). Compared with wild-type, col10a1^p2a-CreERT2 medaka had single wobble base changes at the col10a1 gRNA target site as expected (Fig. 2E, magenta asterisks). In addition, the endogenous col10a1 stop codon was replaced with p2a-CreERT2;cmlc2:EGFP in-frame (Fig. 2E). Protein sequence prediction using the ExPaSy web tool (Swiss Institute of Bioinformatics; SIB) revealed that the amino acid sequence of Col10a1 in col10a1^p2a-CreERT2 medaka remains undisrupted until immediately before its endogenous stop codon (Fig. 2E).

To demonstrate the reproducibility of our knock-in approach, we used the same strategy for a targeted knock-in of p2a-CreERT2 at the osr1 (Fig. S2A) and col2a1a (Fig. S3A) gene loci. Thirteen out of 138 injected embryos (9.4%) and 10 out of 122 injected embryos (8.2%) showed mosaic cmlc2:EGFP expression at 12 dpf for osr1 and col2a1a, respectively (Figs S2B, S3B). To confirm precise knock-in at the osr1 and col2a1a loci, cmlc2:EGFP-positive F0 larvae were fin clipped at 12 dpf for PCR genotyping and DNA sequencing analysis. Correct knock-in sequences were detected in four out of 13 cmlc2:EGFP-positive F0 (30.8%) and three out of 10 cmlc2:EGFP-positive F0 (30.0%) for osr1 and col2a1a, respectively (Figs S2C-D, S3C-D). These findings show that our knock-in approach is an effective and efficient method for generating knock-ins with precise integration.

For accurate cell fate and lineage analyses, it is important that Cre driver lines recapitulate endogenous gene expression patterns spatiotemporally. We previously generated col10a1:CreERT2-p2a-mCherry transgenic medaka and showed that the transgene expression matched endogenous col10a1 expression in the adult medaka fin (Dasyani et al., 2019). However, in the larval vertebral column, the col10a1:CreERT2-p2a-mCherry transgene is ectopically expressed (Fig. S4). Hence, the col10a1:CreERT2-p2a-mCherry transgenic line is not suitable for lineage tracing of col10a1 osteoblast progenitors in the vertebral column. To test whether CreERT2 transcription in col10a1^p2a-CreERT2 knock-in medaka recapitulates endogenous col10a1 transcription, whole-mount RNA in situ hybridization was performed with a CreERT2 riboprobe and compared with col10a1 expression in wild-type medaka at 5, 7 and 9 dpf (Fig. 3A-B″′). At all three stages, CreERT2 expression was identical to col10a1 expression. At 5 dpf, both col10a1 and CreERT2 expression were observed in intramembranous bones such as the parasphenoid, cleithrum, operculum and branchiostegal ray (Fig. 3A,B). At 7 dpf, in addition to intramembranous bones, col10a1 and CreERT2 were also expressed in the ceratohyal, which is a chondral bone (Fig. 3A′,B′). At 9 dpf, col10a1 and CreERT2 expression were detected in cartilage structures such as the ceratobranchials (Fig. 3A″,B″). Similar col10a1 and CreERT2 expression patterns were also observed in neural arches and vertebral bodies of the vertebral column, as well as in the hypural and fin rays of the caudal fin (Fig. 3A″′,B″′; CreERT2 sense control in Fig. S5).

Next, to examine whether CreERT2 protein is expressed specifically in col10a1 cells, the col10a1^p2a-CreERT2 line was crossed to a previously established col10a1:GFP transgenic line (Renn et al., 2013) and CreERT2 immunohistochemistry was performed on 20 µm transverse cryosections of the 9 dpf vertebral column (position indicated by the red dotted line in Fig. 3B″′; Fig. 3C-D′). Similar to col10a1 and CreERT2 expression patterns in the vertebral column (Fig. 3A″′,B″′), CreERT2 protein was detected at neural arches (Fig. 3C,C′, yellow arrows). In addition, at the neural arches, CreERT2 protein expression overlapped with col10a1:GFP expression (Fig. 3C,C′, yellow arrows). In contrast, no staining was observed in neural arches in negative controls (Fig. 3D,D′, yellow arrows). In both CreERT2 staining and negative controls, non-specific staining was observed at the mineralized matrix of vertebral centra (Fig. 3C,D, white arrows) and col10a1:GFP expression was observed in cells lining the mineralized matrix (Fig. 3C,D, green arrows). CreERT2 was detected in a subset of col10a1:GFP cells lining the mineralized matrix (Fig. 3C, cyan arrows). This is in line with low endogenous expression levels of col10a1 and CreERT2 at the vertebral centra, as shown in Fig. 3A″′ and Fig. 3B″′, respectively. Together, this shows that CreERT2 expression in col10a1^p2a-CreERT2 medaka faithfully recapitulates endogenous col10a1 expression at both the RNA and protein level.

In knock-in fish, nucleotide changes were created at the col10a1 gRNA target site and p2a was used to link the col10a1- and CreERT2-coding sequences (Fig. 2E). P2a is a 19 amino acid linker that self-cleaves between glycine and proline at its C-terminal end during translation (Donnelly et al., 2001). In col10a1^p2a-CreERT2 medaka, the p2a self-cleavage is predicted to result in Col10a1 having 18 additional amino acids at its C-terminal end and CreERT2 having an additional proline residue at its N-terminal end (Fig. 2E, black arrowhead indicates p2a cleavage site). To test whether alterations at the C-terminal end of Col10a1 result in any deleterious effects on protein function, we had previously generated a medaka mutant with a 5 bp deletion at the col10a1 gRNA site, leading to a frameshift of the last 11 amino acids and elongation of its protein by one amino acid (W.H.T. and C.W., unpublished; Fig. 1, col10a1 gRNA site indicated in magenta). Heterozygous col10a1 medaka mutants exhibited aberrant skeletal formation, while homozygous mutants had severe bone and cartilage defects, and did not survive beyond 4 weeks (W.H.T. and C.W., unpublished; Fig. 4A). This suggests that distinct alterations at the C-terminal end of Col10a1 can principally result in severe skeletal defects. To elucidate whether the specific nucleotide alterations and additional 18 amino acids introduced in col10a1^p2a-CreERT2 medaka had any effect on skeletal development, Alizarin Red bone staining and Alcian Blue cartilage staining were performed on heterozygous [wild type (WT)/knock-in (KI)] and homozygous (KI/KI) col10a1^p2a-CreERT2 medaka and compared with homozygous col10a1 medaka mutants (Mut/Mut) (Fig. 4A,B). Genotyping of WT/KI and KI/KI medaka was performed by multiplex PCR using three primers, as indicated in Fig. 1 (FP1, RP1 and RP2 in Fig. 1; Fig. S6). In the vertebral column at 12 dpf, mineralized neural arches were absent in Mut/Mut larvae when compared with WT/WT controls (Fig. 4A, top panel, yellow arrows). On the other hand, mineralized neural arches formed normally in WT/KI and KI/KI larvae compared with WT/WT controls (Fig. 4B, top panel). In addition, Alcian Blue staining at 12 dpf showed that ceratobranchials were deformed in Mut/Mut larvae compared with WT/WT controls (Fig. 4A, bottom panel, white arrows). In contrast, no deformities were observed in ceratobranchials of WT/KI and KI/KI larvae compared with WT/WT controls (Fig. 4B, bottom panel, white arrows). Alizarin Red and Alcian Blue staining also showed that other skeletal structures where col10a1 is expressed, such as the branchiostegal rays, cleithrum, hypural and fin rays (Fig. 3), formed normally in WT/KI and KI/KI larvae compared with WT/WT controls (Fig. S7). At adult stage (3 months post-fertilization; mpf), no phenotypic abnormalities were observed in WT/KI and KI/KI fish compared with WT/WT controls (Fig. 4C). In addition, compared with col10a1 Mut/Mut medaka in which lethality was observed at 1 mpf (T.W.H. et al., unpublished), genotyping of 3 mpf offspring from an in-cross of WT/KI carriers revealed a ratio of 6 WT/WT (20.7%):16 WT/KI (55.2%):7 KI/KI (24.1%; Fig. 4C), which is close to a 1:2:1 Mendelian ratio. Together, this indicates that the integration of p2a-CreERT2;cmlc2:EGFP at the medaka col10a1 locus was non-disruptive and did not impair skeletal development or survivability. Hence, the col10a1^p2a-CreERT2 knock-in line is suitable for cell lineage analyses during skeletal development and repair in medaka.

To test whether CreERT2 in col10a1^p2a-CreERT2 medaka can be induced to catalyze recombination between loxP sites in col10a1 cells, the col10a1^p2a-CreERT2 line was crossed to two transgenic loxP reporter lines, GaudiRSG and GaudiBBW2.1, respectively (Fig. 5A; Centanin et al., 2014). In GaudiRSG, DsRed fluorescent protein is ubiquitously expressed by default and can be switched to nuclear localized nlsEGFP upon Cre/loxP recombination (Fig. 5A; Centanin et al., 2014). In GaudiBBW2.1, default membrane-tagged ECFP expression can be switched to either dTomato, YFP or nlsEGFP expression upon Cre/loxP recombination (Fig. 5A; Centanin et al., 2014). For induction of CreERT2 activity, double heterozygous col10a1^p2a-CreERT2|GaudiRSG or col10a1^p2a-CreERT2|GaudiBBW2.1 embryos were treated with 4-hydroxytamoxifen (4-HT; 10 µM; 6 h) at 9 dpf, when col10a1 is expressed in bone and cartilage cells (Fig. 3A″,A″′). After 4-HT treatment, Cre/loxP recombination-mediated cell labeling was observed in a pattern in line with col10a1 expression (Fig. 5B-G). One day post 4-HT treatment (dp4HT), Cre/loxP recombination-mediated fluorescent reporter switching was observed in subsets of chondrocytes (Fig. 5B,C, white arrows) and perichondral osteoblasts (Fig. 5B,B′, yellow arrows) at the ceratohyal. At 7 dp4HT, a switch in fluorescent reporter expression was observed in cells lining the neural arches (Fig. 5D,E, white arrows) and vertebral bodies (Fig. 5D,E, yellow arrows) in the vertebral column. These cell labeling patterns matched endogenous col10a1 expression (Fig. 3A″,A″′) as well as GFP expression in the col10a1:GFP transgenic line, which marks all col10a1-expressing cells (Renn et al., 2013; Fig. 5F,G). For col10a1^p2a-CreERT2|GaudiRSG and col10a1^p2a-CreERT2|GaudiBBW2.1, Cre/loxP recombination-mediated cell labeling was observed in 22 out of 36 (61.1%) and 12 out of 21 (57.1%) 4-HT-treated embryos, respectively. To test whether higher rates of Cre/loxP recombination can be achieved with the col10a1^p2a-CreERT2 line, we extended the 4-HT treatment (at 10 µM) of double heterozygous col10a1^p2a-CreERT2|GaudiRSG embryos from 6 to 24 h at 9 dpf. Under these conditions, 20 out of 20 (100%) 4-HT-treated embryos showed Cre/loxP recombination-labeled col10a1 cells in the vertebral column at 13 dpf (Fig. 5H). Compared with col10a1:GFP, where 54±3 cells were labeled per vertebral body (Fig. 5I, K; n=7 fish, 3 vertebral bodies per fish), 31±5 cells (57.4%) were labeled in the vertebral body of 4-HT-treated col10a1^p2a-CreERT2|GaudiRSG larvae at 13 dpf (Fig. 5H,K; n=7 fish, 3 vertebral bodies per fish). In addition, no cell labeling was observed in larvae not treated with 4-HT (>100 larvae each for GaudiRSG and GaudiBBW2.1), indicating that Cre/loxP recombination did not occur in the absence of 4-HT. This shows that CreERT2 in col10a1^p2a-CreERT2 medaka is functional, non-leaky and, depending on the treatment duration of 4-HT, can catalyze different rates of recombination between loxP sites for specific labeling of col10a1 cells.

We previously used col10a1:CreERT2-p2a-mCherry transgenic medaka for lineage tracing of col10a1 cells in the regenerating adult fin and showed that labeled col10a1 cells migrate beyond the amputation plane, contributing to blastema formation and fin regeneration (Dasyani et al., 2019). To test whether the col10a1^p2a-CreERT2 line is also functional at adult stages for cell lineage analyses, we performed lineage tracing of col10a1 cells in the regenerating adult fin of col10a1^p2a-CreERT2|GaudiRSG knock-in medaka. Similar to transgenic col10a1:CreERT2-p2a-mCherry|GaudiRSG medaka, Cre/loxP recombination-mediated cell labeling was observed specifically in joint cells and osteoblasts proximal to the amputation plane in col10a1^p2a-CreERT2|GaudiRSG knock-in medaka (Fig. 5L,M, yellow and white arrows). In addition, in 2 dpa regenerates of both col10a1^p2a-CreERT2|GaudiRSG and col10a1:CreERT2-p2a-mCherry|GaudiRSG fins, labeled cells were observed in the blastema (Fig. 5L,M, cyan arrows), indicating that labeled col10a1 cells contribute to fin regeneration. Together, this shows that the col10a1^p2a-CreERT2 line can be used for col10a1 cell fate and lineage analyses at both larval and adult stages.

We previously generated rankl:HSE:CFP transgenic medaka and showed that, upon heat-shock, ectopic expression of Rankl led to excessive formation of osteoclasts and the formation of osteoporotic lesions in the mineralized matrix of the vertebral column (To et al., 2012; also see Fig. 6A). Subsequently, col10a1 osteoblast progenitors were found to aggregate at lesion sites and differentiated into osx-positive mature osteoblasts, facilitating de novo mineralization (Renn et al., 2013; also see Fig. 6A, red arrows). At the same time, col10a1 osteoblast progenitors lining the neural arches vanished (Renn et al., 2013; also see Fig. 6A, white arrows), raising the possibility that these cells could migrate to the lesion sites in vertebral bodies in order to repair bone. Time-lapse imaging over a 12 h period using the col10a1:GFP transgenic line showed that in the absence of Rankl induction, col10a1:GFP cells located at the tip of the neural arch exclusively migrated dorsally (Fig. S8A, red arrows; n=5 fish; Movie 1). In contrast, after Rankl induction, a subset of col10a1:GFP cells at the tip of the neural arch switched migration direction and moved ventrally towards the vertebral centra (Fig. S8B, yellow arrows; n=5 fish). However, 12 h time-lapse videos were unable to detect migration of col10a1 cells all the way from the neural arches to the vertebral centra. Displacement tracks generated from time-lapse videos showed that in a 12 h period, col10a1 cells were displaced less than 10 µm from their original position (Fig. S8A,B).

Thus, to trace the fate of col10a1 osteoblast progenitors at neural arches after Rankl induction over a longer time period, we crossed col10a1^p2a-CreERT2|GaudiBBW2.1 fish with rankl:HSE:CFP fish (Fig. 6B). Embryos heterozygous for col10a1^p2a-CreERT2, GaudiBBW2.1 and rankl:HSE:CFP were treated with 4-HT at 7 dpf for col10a1 cell labeling and heat-shocked at 9 dpf for Rankl induction (Fig. 6C). At 11 dpf, live confocal imaging was performed for a selection of larvae with labeled cells at the neural arch. Out of 44 4-HT-treated larvae, 26 larvae (59.1%) had labeled cells in the vertebral column. Twelve larvae (27.3%) had a single labeled cell at a neural arch (Fig. 6D,E, arrows) and were raised until 13 dpf for imaging (Fig. 6D′,D″,E′,E″). In all 12 larvae, there was no increase in the number of labeled cells in the vertebral column from 11 to 13 dpf, indicating that de novo cell labeling did not occur and that labeled cells did not proliferate.

In the absence of Rankl induction, labeled cells at the neural arches remained at roughly the same position from 11 to 13 dpf and never migrated ventrally to the vertebral centra (Fig. 6D-D″, white arrows; n=5/5 fish). This is consistent with observations from time-lapse imaging of transgenic col10a1:GFP larvae at 11 dpf (Movie 1; Fig. S8A, red arrows; n=5 fish). In stark contrast, after Rankl induction, labeled cells that were previously at the neural arch were observed at the vertebral centra where mineralized matrix was degraded (Fig. 6E-E″, yellow arrows; n=5/7 fish). In addition, dynamic changes in the morphology of labeled cells were observed (Fig. 6F-H). Cell protrusions formed (Fig. 6G, yellow arrow) and retracted (Fig. 6H, yellow arrow) as labeled cells migrated to lesion sites where re-mineralization occurred (Fig. 6I-L, red arrows). These observations indicate that col10a1 osteoblast progenitors residing at neural arches serve as an activatable source that can be triggered to facilitate bone repair upon formation of bone lesions in the medaka vertebral column.

The Cre/loxP recombination system has been widely used for genetic cell lineage tracing and fate-mapping studies in teleosts and mammals (Centanin et al., 2014; Harris et al., 2014; Kretzschmar and Watt, 2012; Tu and Johnson, 2011). For precise tracking of cell fate and lineages, it is crucial that Cre/loxP recombination is induced only in the specific cell type of interest. The use of transgenic Cre drivers with ectopic Cre expression may confound data interpretation and lead to false conclusions (Chakraborty et al., 2019; Song and Palmiter, 2018). In this study, we used CRISPR/Cas9 to generate col10a1^p2a-CreERT2 knock-in medaka where CreERT2 expression is driven by the endogenous col10a1 promoter. We show that knock-in fish developed bone and cartilage normally and survived without overt defects up to adulthood. In addition, CreERT2 was expressed in a pattern identical to col10a1 and could be induced by 4-HT for specific Cre/loxP recombination-labeling of col10a1 cells. Using col10a1^p2a-CreERT2, we identified col10a1 osteoblast progenitors at neural arches as a resident cellular source for recruited osteoblasts to repair osteoporotic bone lesions in the vertebral column. This highlights col10a1^p2a-CreERT2 as a useful tool for col10a1 cell fate and lineage analyses during bone remodeling and repair in a vertebrate model suitable for live imaging.

Tol2 transposon and I-SceI meganuclease are commonly used tools for the generation of transgenic Cre driver lines in zebrafish and medaka (Carney and Mosimann, 2018; Centanin et al., 2014). However, these transgenesis approaches make use of random integration of transgenes into the genome, often resulting in position effects. Depending on the regulatory environment at the location of transgene integration, transgene expression may be ectopically enhanced or silenced (Kondrychyn et al., 2009; Thermes et al., 2002). To overcome position effects, the CRISPR/Cas9 system has been used for targeted knock-in of transgenes (Kimura et al., 2014; Watakabe et al., 2018). In CRISPR/Cas9 knock-in approaches, DNA sequences are inserted into specific gene loci via non-homologous end joining (NHEJ) or homology-directed repair (HDR). Although NHEJ is more efficient compared with HDR, it is error-prone, and indels are often observed at knock-in sites (Albadri et al., 2017). On the other hand, HDR allows precise knock-in but has limited efficiency (Liu et al., 2019).

Two recent zebrafish studies have reported the generation of Cre driver lines via CRISPR/Cas9-mediated knock-in (Almeida et al., 2021; Kesavan et al., 2018). However, these approaches disrupted either the coding sequence or the expression level of target genes. Kesavan et al. inserted CreERT2 upstream of the endogenous start codon via NHEJ and they observed a reduction in expression of the target gene in knock-in fish. Almeida et al. knocked in CreERT2 at a coding exon via HDR, causing a loss-of-function mutation upon successful knock-in. Hence, the application of these approaches is limited for genes where knockdown or haploinsufficiency leads to a phenotype.

In comparison, our knock-in approach resulted in minimal interruption to the gene of interest. By targeting knock-in at the 3′ end of col10a1, we avoided disruption of regulatory elements near its transcription initiation site. In addition, making use of HDR for precise integration of the donor template, we inserted p2a-CreERT2 immediately before the stop codon of col10a1, preserving its coding sequence. Although bases at the col10a1 gRNA site were edited to avoid CRISPR/Cas9 cleavage at the donor template 5′ homology arm, these modifications were silent as the protein-coding sequence of col10a1 remained unchanged in knock-in fish. As the efficiency of HDR has been reported to be low (Liu et al., 2019), we included a cmlc2:EGFP secondary fluorescence reporter to accelerate the screening process. We observed high germline transmission rates when screening via cmlc2:EGFP expression (3/7, 42.9%). In addition, we found that cmlc2:EGFP expression was a good indicator of precise knock-in at the col10a1 locus (2/3, 66.7%). With our screening strategy, the germline transmission rate of precise knock-in was 28.6% (2/7). This is markedly higher than the rate of 3% reported in Kesavan et al. (2018) and comparable with rates of 5% to 33% reported by Almeida et al. (2021). Using our knock-in strategy, we also successfully knocked in p2a-CreERT2 into the endogenous osr1 and col2a1a loci, with similar knock-in efficiencies to col10a1 at the F0 generation. Our approach therefore provides an efficient and non-disruptive method for generating knock-in Cre driver lines in medaka and zebrafish.

For effective use of the Cre/Lox system to trace cell fate and lineages, it is important that cells are labeled in a specific and efficient manner. We previously generated transgenic col10a1:CreERT2-p2a-mCherry medaka for lineage tracing of col10a1 cells in the regenerating adult fin (Dasyani et al., 2019). col10a1:CreERT2-p2a-mCherry expression matched endogenous col10a1 expression in adult fin joint cells and osteoblasts, and could be induced to label col10a1 cells with an efficiency of 61.5% (Dasyani et al., 2019; data not shown). However, compared with endogenous col10a1, expression of the transgene in the cranial skeleton and vertebral column was delayed during early development (data not shown). In addition, ectopic transgene expression was observed in spinal cord neurons (Fig. S4). Hence, the col10a1:CreERT2-p2a-mCherry transgenic line was not suitable for col10a1 cell fate and lineage analyses in these structures. Here, we found that CreERT2 RNA and protein expression in col10a1^p2a-CreERT2 medaka precisely recapitulated endogenous col10a1 expression in bone and cartilage structures at early developmental time points. In addition, similar to col10a1:CreERT2-p2a-mCherry, col10a1^p2a-CreERT2 is functional for Cre/loxP recombination-mediated cell labeling and lineage tracing of col10a1 joint cells and osteoblasts in the adult fin. In heterozygous col10a1^p2a-CreERT2 medaka larvae, cell labeling efficiencies were 59.6% (34/57 fish) and 100% (20/20 fish) for 4-HT treatment durations of 6 h and 24 h, respectively. Hence, depending on the objectives of future studies, the treatment duration and concentration of 4-HT can be adjusted to control the extent of cell labeling. In addition, to further enhance cell labeling efficiency, homozygous col10a1^p2a-CreERT2 can be used. The col10a1^p2a-CreERT2 medaka line therefore serves as a versatile tool that can be used to accurately label col10a1 cells for cell fate and lineage analyses in medaka.

For bone growth, homeostasis and repair, timely recruitment of osteoblasts and their progenitors to sites of bone deposition is crucial. Previous studies have shown that cellular sources of recruited osteoblasts include mesenchymal stem cells (Méndez-Ferrer et al., 2010; Park et al., 2012), periosteal osteoblast progenitors (Kegelman et al., 2021; Maes et al., 2010), pericytes (Crisan et al., 2008; Kalajzic et al., 2008) and hypertrophic chondrocytes (Giovannone et al., 2019; Yang et al., 2014; Zhou et al., 2014). However, molecular factors driving the migration of these cells to bone remodeling or repair sites remain poorly understood (Thiel et al., 2018). In vitro studies have identified chemoattractants [such as CXC motif chemokine ligand 16 (CXCL16; Ota et al., 2013)], bone morphogenetic protein 7 (BMP7; Lee et al., 2006), clastokines [such as the slit guidance ligand 3 (SLIT3; Kim et al., 2018)] and platelet-derived growth factors (PDGFs; Colciago et al., 2009) as stimulants of osteoblast migration. However, whether these molecules play a role in the recruitment of osteoblasts in vivo remains to be shown.

We have previously shown that col10a1:GFP osteoblast progenitors accumulate at Rankl-induced bone lesion sites and facilitate bone re-mineralization in the medaka vertebral column (Renn et al., 2013). However, the origin of these recruited osteoblast progenitors and the mechanism guiding their recruitment remain unclear. Owing to the limitations of live time-lapse imaging, i.e. photobleaching of fluorophores and reduced survival of larvae after 12 h of imaging, recruitment of col10a1 osteoblast progenitors from neural arches to bone lesion sites could not be detected using the col10a1:GFP transgenic line. Here, by using the col10a1^{p2a- CreERT2} knock-in line to track col10a1 cells over 48 h, we found that col10a1 osteoblast progenitors residing at neural arches can be triggered to migrate to bone lesion sites at vertebral centra for bone repair. Without Rankl induction, migration of col10a1 osteoblast progenitors from neural arches to vertebral centra was never observed and these cells appeared stationary. Hence, we speculate that, upon formation of Rankl-induced bone lesions, specific chemotactic factors are released from the lesion sites. Chemotactic gradients then activate migratory behavior of col10a1 osteoblast progenitors, which form cell projections and move towards the chemical cues. Future analysis on cell-surface receptors and molecular networks activated in these recruited col10a1 cells will provide valuable insights into the mechanisms driving osteoblast recruitment. Our medaka model also serves as a useful in vivo platform for the study of molecules previously proposed to regulate osteoblast migration in vitro. Such studies would greatly facilitate the development of bone anabolic therapies for treatment of low bone mass disorders such as osteoporosis, as well as for the stimulation of bone fracture repair.

Medaka lines were maintained under a controlled 14 h light/10 h dark cycle in the fish facilities at the National University of Singapore. Fish breeding and experiments were performed in accordance with protocols approved by the Institutional Animal Care and Use Committee of the National University of Singapore (protocol numbers BR19-0120 and R18-0562). The medaka Cab strain was used as wild type. Transgenic col10a1:GFP (Renn et al., 2013), rankl:HSE:CFP (To et al., 2012), GaudiRSG and GaudiBBW2.1 (Centanin et al., 2014) lines have been previously described.

For cloning of the knock-in donor plasmids, 5′ and 3′ col10a1, osr1 and col2a1a homology arms were PCR amplified from genomic DNA of wild-type medaka. p2a-CreERT2-pA was PCR amplified from plasmid col10a1:CreERT2-p2a-mCherry (Dasyani et al., 2019) and cmlc2:EGFP-pA was PCR amplified from plasmid pBS/I-SceI/Hsp70::Cre-NLS (kindly provided by Lazaro Centanin, Centre for Organismal Studies, Heidelberg, Germany; Centanin et al., 2014). All PCR primers used are listed in Table S1. The mbait guide RNA (gRNA) sequence used has been described previously (Kimura et al., 2014) and is listed in Table S2. col10a1, osr1 and col2a1a gRNA target sequences were selected using the CCTop web tool (Stemmer et al., 2015; Table S2). Wobble bases at the col10a1, osr1 and col2a1a gRNA sites in the respective donor plasmids were introduced by PCR using primers with mismatched nucleotides (Table S1). The homology arms, mbait, p2a-CreERT2-pA and cmlc2:EGFP-pA sequences were combined in the order as shown in Fig. 1, Figs S2A and S3A for col10a1, osr1 and col2a1a, respectively, by PCR using overlapping primers (Table S1). For the col10a1 donor plasmid, the mbait-5′HA-p2a-CreERT2-pA-cmlc2:EGFP-pA-3′HA-mbait insert was ligated to a pCR-bluntII-TOPO vector (Invitrogen). For osr1 and col2a1a donor plasmids, the mbait-5′HA-p2a-CreERT2-pA-cmlc2:EGFP-pA-3′HA-mbait insert were ligated to a pJET1.2/blunt vector (ThermoFisher Scientific). All donor plasmids can be obtained from Addgene (Addgene 184874, 184875 and 184876 for col10a1, osr1 and col2a1a, respectively).

A solution containing the donor plasmid (20 ng/µl), Cas9 protein (0.25 µg/µl; Integrated DNA Technologies, IDT), mbait crRNA (36 ng/µl; IDT), tracrRNA (67 ng/µl; IDT) and the respective col10a1, osr1 or col2a1a crRNA (36 ng/µl; IDT) was injected into wild-type medaka embryos at the one-cell stage. Injected embryos were screened for cmlc2:EGFP expression at 12 days post-fertilization (dpf). To establish a stable col10a1^p2a-CreERT2 line, cmlc2:EGFP-positive larvae were raised to adulthood and out-crossed to wild-type medaka. F1 offspring positive for cmlc2:EGFP expression were raised and subsequently fin clipped for PCR genotyping to determine whether knock-in occurred at the endogenous col10a1 locus (primers listed in Table S1). DNA sequencing was performed to confirm precise integration of the donor template at the 3′ and 5′ col10a1 homology regions (primers listed in Table S1). Primers used for genotyping of heterozygous and homozygous col10a1^p2a-CreERT2 knock-in fish are listed in Table S1. To confirm precise knock-in of the donor template in osr1^p2a-CreERT2 and col2a1a^p2a-CreERT2 potential founders (F0), 12 dpf cmlc2:EGFP-positive larvae were fin clipped for PCR genotyping, and DNA sequencing was performed (primers listed in Table S1).

The col10a1 riboprobe was generated as previously reported (Renn and Winkler, 2010). To generate the CreERT2 riboprobe, a 637 bp CreERT2 sequence was amplified from the knock-in donor plasmid using primers CreERT2_366_FP (5′-GAAAACGTTGATGCCGGTGAAC-3′) and CreERT2_1003_RP (5′-TTGCCCCTGTTTCACTATCCAGG-3′). This sequence was then ligated to a pCR-bluntII-TOPO vector (Invitrogen) containing flanking T7 and SP6 promoter sequences. To synthesize CreERT2 sense and antisense riboprobes, plasmid linearization was performed using the respective SpeI and XbaI restriction enzyme (New England Biolabs) and in vitro transcription was performed using T7 and SP6 RNA polymerase (DIG RNA Labeling Mix, Roche), respectively. Whole-mount RNA in situ hybridization was performed as described by Renn and Winkler (2009). Immunohistochemistry on medaka cryosections (20 µm) was performed as previously described (Dasyani et al., 2019). Mouse anti-Cre Recombinase antibody (1:400, Merck MAB3120, Clone 2D8, lot 3543072) and secondary goat anti-mouse Alexa Fluor 633 (1:500, Invitrogen A-21052) were used.

Live bone staining using Alizarin complexone (ALC; Sigma A3882) or Calcein (Sigma C0875) was performed as described by Renn et al. (2013). For fixed samples, staining of bone using Alizarin Red S (Sigma A5533) and cartilage using Alcian Blue 8GX (Sigma A3157) were performed as described by Renn and Winkler, 2009.

To induce Cre/loxP recombination at embryonic stages (7 to 9 dpf), col10a1^p2a-CreERT2|GaudiRSG or col10a1^p2a-CreERT2|GaudiBBW2.1 embryos were incubated in 10 µM 4-hydroxytamoxifen (4-HT; Sigma T176) for 6 or 24 h in the dark at 28°C. After 4-HT treatment, fish were washed thrice with 1× fish medium [19.3 mM NaCl, 0.23 mM KCl, 0.13 mM MgSO₄, 0.2 mM Ca(NO₃)₂, and1.7 mM HEPES (pH 7.0)]. To induce Cre/loxP recombination in col10a1^p2a-CreERT2|GaudiRSG adults (5 months old), fish were treated with 0.5 µM 4-HT for 72 h in the dark. After 4-HT treatment, fish were washed thrice with fish system water.

Embryos from a cross of col10a1^p2a-CreERT2|GaudiBBW2.1 and rankl:HSE:CFP were used. At 7 dpf, Cre/loxP recombination-mediated col10a1 cell labeling was performed via 4-HT treatment (10 µM, 6 h, 28°C). At 9 dpf, Rankl induction was conducted via a heat-shock at 39°C for 2 h. After heat-shock, embryos were kept in a 30°C incubator and live confocal imaging was performed at 11 and 13 dpf.

Amputation of the adult medaka fin and imaging of the regenerating fin were performed as described by Dasyani et al. (2019).

Quantification of col10a1:GFP cells was performed as described by Phan et al. (2020). To quantify nlsEGFP-labelled cells in col10a1^p2a-CreERT2|GaudiRSG medaka, cells were counted manually.

Images were captured using the Olympus FluoView FV3000 confocal microscope, Zeiss LSM900 confocal microscope and Nikon SMZ18 stereomicroscope. For live fluorescence imaging, larvae were anaesthetized in 0.016% Tricaine (Sigma MS-222) and mounted in 1.2% low melting agarose on a glass-bottom dish. Image processing was performed using Fiji ImageJ, Bitplane Imaris and Adobe Photoshop.

We thank Lazaro Centanin (Centre for Organismal Studies, University of Heidelberg) for sharing the pBS/I-SceI/Hsp70::Cre-NLS plasmid as well as the GaudiRSG and GaudiBBW2.1 transgenic lines. We also thank the Centre for Bioimaging Sciences (CBIS) confocal unit and the fish facility at Department of Biological Sciences, National University of Singapore for continued support.

The peer review history is available online at https://journals.biologists.com/dev/article-lookup/doi/10.1242/dev.200238