Variation in zygotic CRISPR/Cas9 gene editing outcomes generates novel reporter and deletion alleles at the Gdf11 locus

Recent advances in CRISPR/Cas gene editing technology have significantly expanded the possibilities and accelerated the pace of creating genetically engineered animal models. However, CRISPR/Cas-based strategies designed to precisely edit the genome can often yield unintended outcomes. Here, we report the use of zygotic CRISPR/Cas9 injections to generate a knock-in GFP reporter mouse at the Gdf11 locus. Phenotypic and genomic characterization of founder animals from these injections revealed a subset that contained the correct targeting event and exhibited GFP expression that, within the hematopoietic system, was restricted predominantly to lymphoid cells. Yet, in another subset of founder mice, we detected aberrant integration events at the target site that dramatically and inaccurately shifted hematopoietic GFP expression from the lymphoid to the myeloid lineage. Additionally, we recovered multiple Gdf11 deletion alleles that modified the C-terminus of the GDF11 protein. When bred to homozygosity, most of these alleles recapitulated skeletal phenotypes reported previously for Gdf11 knockout mice, suggesting that these represent null alleles. However, we also recovered one Gdf11 deletion allele that encodes a novel GDF11 variant protein (“GDF11-WE”) predicted to contain two additional amino acids (tryptophan (W) and glutamic acid (E)) at the C-terminus of the mature ligand. Unlike the other Gdf11 deletion alleles recovered in this study, homozygosity for the Gdf11WE allele did not phenocopy Gdf11 knockout skeletal phenotypes. Further investigation using in vivo and in vitro approaches demonstrated that GDF11-WE retains substantial physiological function, indicating that GDF11 can tolerate at least some modifications of its C-terminus and providing unexpected insights into its biochemical activities. Altogether, our study confirms that one-step zygotic injections of CRISPR/Cas gene editing complexes provide a quick and powerful tool to generate gene-modified mouse models. Moreover, our findings underscore the critical importance of thorough characterization and validation of any modified alleles generated by CRISPR, as unintended on-target effects that fail to be detected by simple PCR screening can produce substantially altered phenotypic readouts.


Results
Generation of a Gdf11 reporter construct and founder mice. We sought to target a fluorescent reporter gene to the mouse Gdf11 locus using CRISPR/Cas9 and first tested this approach in cell culture. We began by designing single guide RNAs (sgRNAs) compatible with Streptococcus pyogenes Cas9 (spCas9) to target the mouse Gdf11 locus. Based on the location of the protospacer adjacent motif (PAM) for each sgRNA, spCas9 was predicted to cut after the stop codon and at the beginning of the 3′UTR (Fig. S1A). To test the ability for each sgRNA to direct spCas9 to the intended genomic location, we transfected mouse C2C12 cells with an spCas9-2A-mCherry expression vector, together with individual sgRNA expression plasmids. Three days after transfection, we FACS-purified mCherry + and mCherry − cells (Fig. S1B) and PCR-amplified a 797-base pair (bp) amplicon surrounding the target sequence (Fig. S1C). T7 Endonuclease I (T7EI) mismatch repair assays 43 indicated the introduction of indels at the target site for all four sgRNAs in the mCherry + cells, confirming that each could effectively target the Gdf11 locus (Fig. S1C). We also detected faint bands indicating indels at the target site in the mCherry − cells (Fig. S1C), which likely reflects impurities in sorting the mCherry − fraction or that this fraction included cells that had already expressed Cas9 and targeted the Gdf11 locus, but had not yet robustly expressed mCherry.
We next developed a strategy to introduce a GFP reporter into the mouse Gdf11 locus using HDR. As GDF11 is a secreted ligand whose activity requires multiple proteolytic processing events 44 , we decided against a direct protein fusion approach, which could cause fusion-related disruptions in GDF11 expression, stability or signaling capacity. Instead, we pursued a bicistronic approach to drive GFP expression from an internal ribosome entry site (IRES), a strategy used previously for the generation of analogous reporter lines [45][46][47] . We cloned a plasmid donor containing IRES-GFP sequences flanked by 2.2 kb homology arms (Fig. 1A) and transfected this construct Red boxes indicate founder animals harboring large deletions in Gdf11. Gel image is uncropped. Positive and negative control reactions for PCR amplification were run on a separate gel, which is presented in Supplementary Fig. 12A. knockout mice 26,33 , we hypothesized that these mutations likely render GDF11 non-functional. We also detected multiple Gdf11 indels in the 2 dead pups that did not lack tails (pups E-F; Fig. S2D). These mutations either eliminate 1-2 amino acids at the C-terminus or produce non-coding disruptions in the 3′UTR (Fig. S2D). Given that these pups do not phenocopy the tailless phenotype of the Gdf11 null mouse, these indels may preserve the function of GDF11, although expression of some loss-of-function phenotypes in these zygotically-injected mice could also have been suppressed by genetic mosaicism.

Characterization of progeny from zygotic CRISPR injections.
Screening of the 57 live pups produced from our CRISPR/Cas injections using internal and knock-in specific PCR primers identified 6 founder animals harboring the Gdf11-IRES-GFP allele (Round #1: 1 founder; Round #2: 0 founders; Round #3: 3 founders; Round #4: 2 founders) (Figs. 1B,D and S2A,B). In founder #1 from Round #1, we only detected the presence of a knock-in-sized amplicon, suggesting biallelic targeting (Fig. 1B). Sanger sequencing of the knock-in sized amplicon from this founder (Fig. 1C) confirmed the presence of the silent mutation (C→T) included in the HDR template, but additionally showed a peak for the native C nucleotide on the chromatogram (Fig. 1C). These results indicate mosaicism of founder #1 from Round #1, as has been reported previously 18,19 . During our PCR screening, we also identified 4 mice with a discernably smaller amplicon than the wild type-sized amplicon (red boxes; Fig. 1D).
Collectively, 4 of the 6 Gdf11-IRES-GFP targeted founders survived to adulthood, and we used these mice to establish independent lines for further validation studies. Southern blot analysis revealed that while all 4 founders contained the Gdf11-IRES-GFP at the intended locus, founders #1 and #13 also contained an additional integration (AI) event (Figs. 2A, S3A and S12B). When founder #1 was outbred to a C57BL/6J mouse, the F1 progeny either contained only the correctly targeted allele (see mouse 1B in Fig. 2B) or both the targeted allele and an additional integration event (see mouse 1A in Fig. 2B). When founder #13 was outbred to a C57BL/6J mouse, the additional integration event segregated with the correctly targeted event (Figs. 2B, S3B and S12C). Next, we performed Targeted Locus Amplification (TLA) sequencing 48,49 using two independent primer sets to identify the genomic location of each targeting event. TLA results confirmed that line 1B contained the correctly integrated IRES-GFP transgene at the Gdf11 locus (Figs. 2C and S3C). TLA sequencing also demonstrated targeting at the Gdf11 locus in lines #1A and #13; however, the integration events in these lines incorporated the intended donor sequence along with the plasmid backbone, likely as a concatemer (data not shown). Thus, the additional integrations in lines #1A and #13 were not off-target integrations of the transgene, but rather aberrant integration events at the intended Gdf11 target site. In summary, while all founders that screened positive by PCR were indeed targeted at the Gdf11 locus, only a fraction of these alleles contained the correct insertion of the IRES-GFP cassette behind the Gdf11 coding sequence.
The knock-in Gdf11-GFP gene accurately reports Gdf11 expression in correctly targeted lines.
We next performed flow cytometry to detect GFP fluorescence within peripheral blood cells of progeny from the 5 different Gdf11-IRES-GFP alleles. Interestingly, the pattern of GFP fluorescence varied substantially between the 2 categories of targeted mice (Fig. 2D). In the correctly targeted lines (line #1B, #11 and #12), GFP fluorescence was detected primarily in T and B lymphocytes, with minimal fluorescence in monocytes and neutrophils (Fig. 3A,B). These results are consistent with publicly available datasets reporting enrichment of Gdf11 expression in lymphocytes relative to other hematopoietic cell types [50][51][52] . Our own analysis further confirmed enrichment of Gdf11 mRNA within splenic CD19+ B-lineage cells relative to CD19-cells in 2-month old mice (Fig. 3C). Conversely, the 2 incorrectly targeted lines (line #1A, #13) exhibited robust GFP fluorescence in monocytes and neutrophils and minimal GFP fluorescence in lymphoid cells (Fig. S4A,B). While we cannot explain the molecular mechanisms underlying this striking lymphoid to myeloid switch in GFP expression in the incorrectly targeted GFP reporter mouse lines, these findings emphasize the need to rigorously validate new CRISPR-generated mouse lines using multiple molecular techniques, as aberrant integration events at the target site can dramatically alter reporter gene expression and phenotypic readouts.
Focusing our further efforts on only the correctly targeted Gdf11-IRES-GFP mouse lines, we performed additional studies to determine whether GFP fluorescence accurately reports Gdf11 mRNA expression within these animals. To this end, we FACS-purified GFP high and GFP low splenocytes from homozygous animals and performed real-time PCR analysis for Gdf11 using two independent primer sets. We observed approximately 5-fold greater levels of Gdf11 mRNA in GFP high relative to GFP low splenocytes for all 3 Gdf11-IRES-GFP reporter lines (Figs. 2E,F and S3D-G). Next, to determine if IRES-GFP integration altered Gdf11 expression in the reporter mice, we compared splenic Gdf11 mRNA and serum GDF11 protein levels in homozygous reporter animals to age-and sex-matched C57BL/6J controls. To specifically measure GDF11 protein levels, we employed a recently developed liquid chromatography tandem mass spectrometry assay that selectively discriminates between GDF11 and GDF8 by measuring two divergent peptide fragments 53 . Splenic Gdf11 mRNA expression (Fig. 2G,H) and serum GDF11 protein levels ( Fig. 2I) were not significantly different among the 3 reporter lines and C57BL/6J controls, indicating that introduction of the IRES-GFP reporter sequence did not perturb endogenous GDF11 production. GDF8 serum levels were also unchanged in the reporter mice (Fig. 2J). Lastly, we analyzed the spatial expression pattern of Gdf11 and Gfp mRNA in reporter animals by in situ hybridization and compared these results to the pattern of endogenous Gdf11 mRNA expression in age-matched C57BL/6J mice. Patterns of Gdf11 expression were indistinguishable in C57BL/6J and Gdf11-IRES-GFP line #1B heterozygous embryos at E10.5 (Fig. 2K,L) and consistent with published data indicating enrichment of Gdf11 expression in somites, tail tip, limb bud, and forebrain at this developmental stage 26,27 . Hybridization with antisense Gfp probes further revealed an expression pattern similar to that of Gdf11 in Gdf11-IRES-GFP line #1B heterozygous embryos and a lack of signal in C57BL/6J embryos (Fig. 2K,L), confirming that Gfp expression recapitulates Gdf11 expression. , Gdf11-IRES-GFP F1 progeny. Nco1-digested genomic DNA was hybridized with the internal probe. Expected fragment size: WT = n/a; T (targeted) = 3.5 kb. AI: Additional integration. Blot images were cropped to focus on the target bands. Uncropped blots are presented in Supplementary Fig. 12B,C. (C) TLA sequencing coverage and analysis plots from line 1B using outward facing primers residing in the GFP transgene. (D) Flow cytometry analysis of GFP expression in live (7AAD − ) peripheral blood cells in left: mice exhibiting correct targeting (lines 1B, 11 and 12) and right: mice exhibiting incorrect targeting (lines 1A and 13). (E,F) Real time PCR analysis of Gdf11 levels in FACS-purified GFP high and GFP low splenocytes from line 1B using (E), primers spanning exons 1-2 and (F), primers spanning exons 2-3. βactin was used as a housekeeping gene. Transcript levels were normalized to levels in GFP low splenocytes. N = 4 males (blue), 4 females (red). Data are presented as individual data points overlaid with mean ± SD. (G,H) Real time PCR analysis of Gdf11 levels in whole spleen from correctly targeted lines (1B, 11 and 12) and age-and sex-matched C57BL/6J mice. Relative Gdf11 expression levels were assayed using (G), primers spanning exons 1-2 and (H), primers spanning exons 2-3. β-actin was used as a housekeeping gene. Transcript levels were normalized to levels in C57BL/6J mice. N = 3-4 males (blue), 3-4 females (red). Data are presented as individual data points overlaid with mean ± SD. (I,J), Quantification of (I), GDF11 protein levels, and (J), GDF8 protein levels, in serum from correctly targeted lines (1B, 11 and 12) and age-and sex-matched C57BL/6J mice. (K,L) Whole mount in situ hybridization for Gdf11 (top) and Gfp (bottom) in E10.5 Gdf11 +/+ and Gdf11 KI/+ embryos from line 1B. For each embryo, the right-most images show the dissected forelimb. mb: midbrain, fb: forebrain, psm: pre-somitic mesoderm, fl: forelimb, hl: hindlimb, s: somite. Scale bar: 0.5 mm.
www.nature.com/scientificreports www.nature.com/scientificreports/ Analysis of Gdf11-GFP fluorescence reveals dynamic expression of Gdf11 during aging. We utilized our Gdf11-IRES-GFP line #1B animals to evaluate the dynamics of Gdf11 mRNA expression during aging. Given that hematopoietic lineage cells express high levels of Gdf11 during youth relative to other tissues 37 , we chose to profile Gdf11 expression in discrete subsets of peripheral blood cells from 3 weeks of age to 25 months of age. Approximately 40-50% of T cells in Gdf11-IRES-GFP mice exhibited detectable GFP signal throughout most of life with a transient drop to ~20% of T cells at 6-8 months of age (Fig. 3D). The mean fluorescence intensity of GFP, reflecting the relative level of Gdf11 expression in individual cells, was also largely constant in T cells during aging, with a slight spike at the 6-8 month timepoint (Fig. 3F). These results indicate that both the level of Gdf11 expression and the proportion of Gdf11-expressing peripheral T cells remain largely steady during aging. However, as T cell frequency among peripheral blood mononuclear cells (PBMCs) decreases from ~45% at 3 weeks of age to ~10% at 22-24 months of age, while B cell frequency increases from ~25% at 3 weeks to ~51% at 22-24 months of age (Fig. 3H), it is likely that T cells become proportionately lesser contributors to hematopoietic production of GDF11 with increasing age.
The profile of Gdf11-IRES-GFP expression during aging was slightly different for B lymphocytes. From 3 weeks to 4 months of age, the frequency of GFP+ B cells increased from ~35% to ~70%, but then decreased after 4 months of age, eventually dropping to ~60% at 24 months of age (Fig. 3E). However, GFP mean fluorescence intensity showed an overall increase in B cells during aging (Fig. 3G). Thus, while the fraction of peripheral blood B cells that express Gdf11 gradually decreases in adult mice after 4 months of age, those Gdf11-expressing B cells that remain in older animals appear to increase Gdf11-GFP expression. This change in Gdf11-GFP expression may compensate to some degree for the observed age-related loss of Gdf11-producing B cells; however, bulk analysis of Gdf11 mRNA levels in splenic CD19+ B cells or CD19− non-B cells from young or aged mice using real-time PCR suggests that the total abundance of Gdf11 mRNA in B lymphocytes still declines with age (Fig. 3C). These apparently divergent trajectories of Gdf11 expression in aging B cells could reflect differences in the anatomical site of analysis (peripheral blood versus spleen), aging-related changes in Gdf11 mRNA translational efficiency, or other as yet undiscovered regulatory mechanisms. Regardless, these data underscore the importance of applying single cell resolution, as achieved by flow cytometric analysis of the fractional representation of Gdf11-GFP-expressing B cells, to dissect the complexities of gene expression dynamics in aging lymphocyte subsets.
Gdf11 deletion alleles recapitulate developmental phenotypes of Gdf11 −/− embryos. Following analysis of the Gdf11-IRES-GFP knock-in reporter animals, we revisited the Gdf11 deletion alleles that we also recovered after targeting of the Gdf11 locus (red boxes; Fig. 1D). Founder #4 from Round #3 contained two distinct Gdf11 deletion alleles that segregated in the F1 generation, termed 4A and 4B (data not shown). Sanger sequencing of all 4 deletion alleles recovered from 3 founders indicated deletions ranging in size from 89 bp up to 381 bp (Fig. 4A). In silico translation of these sequences yielded predicted peptide sequences with an additional 2-156 amino acids at the C-terminus of the protein due to partial translation of the 3′UTR (Fig. 4B). To evaluate the impacts of these novel deletion alleles, we assayed their effects on GDF11 function in vivo. Given that our zygotic injections produced several dead pups that lacked a tail and harbored frameshift mutations leading to extensive translation of the Gdf11 3′UTR, we hypothesized that these recovered deletion alleles would likewise induce skeletal abnormalities when bred to homozygosity. For each deletion line, we established timed matings between heterozygous mice and harvested E17.5 or E18.5 stage embryos. While wild type (Gdf11 +/+ ) and heterozygous (Gdf11 indel/+ ) embryos from lines 4A, 4B, and 11 exhibited an external tail, homozygous (Gdf11 indel/indel ) embryos from all 3 of these lines lacked a tail (Fig. 4C), similar to engineered Gdf11 null mice 26,33 . Whole mount skeletal preparations from these embryos further confirmed the skeletal abnormalities associated with these alleles, as all heterozygous embryos contained one extra thoracic vertebra and all homozygous embryos contained five additional thoracic vertebrae relative to wild type controls ( Fig. 4D,H,I). Altogether, embryos homozygous for deletion 4A, 4B or 11 showed a significant decrease in total vertebrae number (Fig. 4F,G), with expansions in the number of both thoracic (T) and lumbar (L) vertebrae ( Fig. 4H,I). These skeletal phenotypes are consistent with previously reported phenotypes from Gdf11 knockout mice 26,33 , which frequently exhibit a C7/T18/L8 pattern 26 , and strongly suggest that these 3 genomic deletions render the Gdf11 allele non-functional.
Interestingly, during our analyses of the skeletons of these Gdf11 indel mice, we also observed an additional skeletal phenotype within the otic capsule of the skull, a bone that surrounds the inner ear 54 . Both heterozygous and homozygous embryos exhibited a discernable hole within the otic capsule ( Fig. 4E; top panels). This phenotype had not been reported previously for Gdf11 null mice, thus prompting us to interrogate embryos heterozygous or homozygous for the previously generated Gdf11 null allele 26,33 for this phenotype. We identified this feature also in the previously described genetically engineered Gdf11 heterozygous and knockout mice ( Fig. 4E; bottom panels), confirming that this phenotype indeed arises due to the disruption of Gdf11.
In addition to skeletal phenotypes, Gdf11 knockout mice exhibit a spectrum of renal defects, particularly unilateral and bilateral kidney agensis 29 . We therefore quantified kidney number in E18.5 stage Gdf11 +/+ , Gdf11 indel/+ , and Gdf11 indel/indel embryos from lines 4A and 11 to determine if kidney agenesis was also a feature of these newly generated Gdf11 loss-of-function alleles. 100% of Gdf11 +/+ embryos (5 out of 5 examined) exhibited two kidneys (Fig. S5A,B), as did most Gdf11 indel/+ embryos. 4% (1 out of 25 examined) of Gdf11 indel/+ embryos exhibited unilateral renal agenesis (Fig. S5A,B), consistent with previously reported results from Gdf11 heterozygous mice 29 . In contrast, 63.6% (7 out of 11 analyzed) of Gdf11 indel/indel embryos exhibited bilateral renal agenesis and 36.4% (4 out of 11 analyzed) exhibited unilateral renal agenesis (Fig. S5A,B), again consistent with previous results in Gdf11 knockout mice 29 . Taken together, these findings provide additional supportive evidence that the Gdf11 deletion alleles recovered in this study represent null alleles of Gdf11. Gdf11 WE deletion allele generates a functional GDF11 protein variant. In contrast to the other deletion alleles generated in this study, and to Gdf11 null mice, homozygous progeny from line 7 were viable and survived through adulthood, suggesting that the Gdf11 mutation in this line may preserve GDF11 function. In silico translation of this Gdf11 gene variant is predicted to add a tryptophan (W) and a glutamic acid (E) to the C-terminus of GDF11; thus, we named this variant GDF11-WE. We further investigated the functionality of www.nature.com/scientificreports www.nature.com/scientificreports/ GDF11-WE by examining skeletal preparations at embryonic day 18.5 (E18.5) from GDF11-WE intercrosses. In contrast to the Gdf11 mutants from lines 4 A, 4B, and 11 ( Fig. 4C-E), and the previously reported Gdf11 null mouse 26   www.nature.com/scientificreports www.nature.com/scientificreports/ in Gdf11 knockout mice. For example, Gdf11 WE/+ and Gdf11 WE/WE embryos exhibited a normal number of ribs (Fig. 5B), a developed otic capsule (Fig. 5C), and had a tail (Fig. 5A,B). However, upon further analysis of the vertebrae, we observed one fewer lumbar vertebra in all heterozygous embryos and most homozygous embryos analyzed (Fig. 5D,E), resulting in a predominant C7/T13/L5 vertebral pattern that is not seen in Gdf11 +/− or Gdf11 −/− mice 26 . Although the number of lumbar vertebrae can vary by strain [55][56][57] , this pattern is also notably different from the predominant pattern in the Gdf11 WE background strain-C57BL/6J-which primarily exhibits 6 lumbar vertebral segments (Fig. 5E). Altogether, these data strongly indicate that the deletion 7 allele codes for a new variant of GDF11, GDF11-WE, that maintains largely normal function during development.
We next investigated whether adult mice harboring one or two alleles encoding GDF11-WE showed any differences in circulating levels of GDF11 or GDF8. Interestingly, 5-8-week old Gdf11 WE/+ and Gdf11 WE/WE males, but not females, exhibited statistically significant increases in circulating GDF11 levels relative to age-matched Gdf11 +/+ controls (Fig. 5M). GDF8 levels, in contrast, were not significantly different in either males or females (Fig. 5N). These results suggest that there may be changes in ligand production and/or stability in mice possessing the GDF11-WE variant.
To assess the signaling capacity of GDF11-WE, we generated plasmid constructs enabling expression of either the mature GDF11-WE or wild-type GDF11 ligand, and tested the activity of these recombinant proteins in a well-established HEK293 SMAD3-responsive ((CAGA) 12 -promoter) luciferase assay [58][59][60][61][62][63] . In this assay, exogenous addition or co-transfection of TGF family ligands such as GDF11, together with the processing enzymes furin and tolloid-like 1, activates the SMAD2/3 signaling pathway, resulting in an increase in luciferase activity [58][59][60][61][62][63] . We transfected reporter cells with increasing amounts of plasmid encoding wild-type GDF11 or GDF11-WE (Fig. S6A,B). Transfection with GDF11-WE activated the SMAD3 reporter, but the activity was significantly less compared to cells transfected with an equal amount of plasmid encoding wild-type GDF11 (Fig. S6E). Because C-terminal modification or extension of TGF-β has been shown to negatively impact ligand dimerization and activity 64,65 , we wondered if GDF11-WE might dimerize ineffectively. We evaluated this possibility first by Western analysis and found that, whereas WT recombinant GDF11 protein is detected only as a dimer under non-reducing conditions, conditioned media from cells transfected with the GDF11-WE variant produced both the dimer and monomer signal (Fig. S6F, left). Under reducing conditions, the monomer but not the dimer, was detected for both WT recombinant GDF11 protein and GDF11-WE conditioned media (Fig. S6F, right). These data suggest that dimerization may be disrupted in the GDF11-WE variant. We further explored this idea by mutating GDF11 cysteine 73 in the mature ligand, the residue responsible for the intermolecular disulfide bond 58,66 , to a serine in both wild-type GDF11 (GDF11 C73S) and GDF11-WE (GDF11-WE C73S) (Fig. S6C,D), in order to test whether the activity of GDF11-WE, like wild-type GDF11, requires the intermolecular disulfide bond. Ligand activity was assessed using the same transfection assay described above. As expected, transfection with GDF11 C73S resulted in little to no detectable luciferase activity. Transfection with GDF11-WE C73S similarly showed little to no luciferase activity, suggesting that dimerization is still required for GDF11-WE activity (Fig. S6E). Thus, the diminished activity of the variant GDF11-WE protein relative to wild-type GDF11 may be explained by its reduced efficiency of dimerization, although alternative mechanisms related to specific effects of the added C-terminal tryptophan and glutamic acid on ligand activity cannot be excluded.
Although the activity of GDF11-WE is clearly sufficient to support largely normal development and survival after birth (Fig. 5A-F), we investigated whether adult mice harboring the GDF11-WE variant might exhibit postnatal phenotypes due to the altered levels and/or signaling activity of this variant protein. As Gdf11 expression is upregulated is discrete immune cell lineages (Fig. 3A,B) and GDF11 functions in adulthood have previously been implicated in cardiac hypertrophy 37,67-69 , erythropoiesis 70,71 , and skeletal muscle biology 39,40 , we focused our analyses on the heart, skeletal muscle, and immune organs. We did not detect significant differences in body weight (Fig. 5G), normalized heart weight ( Absence of overt immunological phenotypes in mice harboring Gdf11 loss-of-function or Gdf11 WE alleles. Given that Gdf11 expression is upregulated is immune cell lineages (Fig. 3A,B), we examined whether mice containing Gdf11 deletion alleles might exhibit any overt immunological phenotypes. We did not observe any obvious changes in the abundance or composition of circulating blood leukocytes in mice containing the Gdf11 deletion alleles (Fig. S7A-G). Profiling of immune cell activation markers within T and B cell subsets similarly revealed no differences in mice harboring Gdf11 deletion alleles (Fig. S8-10). In addition, the vast majority of circulating immune and inflammatory cytokines did not differ between Gdf11 +/+ and Gdf11 indel/+ mice (pooled from lines 4A, 4B, and 11), although IL-2 levels were significantly decreased (~6-fold) and M-CSF levels significantly increased (~2-fold) in Gdf11 indel/+ mice relative to Gdf11 +/+ mice (Table S1). No differences in serum cytokine levels were detected among age-matched Gdf11 +/+ , Gdf11 WE/+ , and Gdf11 WE/WE mice (Table S1). Mice with Gdf11 deletion alleles also did not exhibit differences in the levels of immunoglobulin sub-classes or Ig light chain types (Table S2). Finally, we compared the relative sizes of immune organs, including the spleen, thymus, and inguinal lymph node, for cohorts of adult mice with various Gdf11 deletion alleles. Mice heterozygous for Gdf11 indel alleles from lines 4A, 4B, and 11 exhibited reduced spleen weight compared to Gdf11 +/+ mice (Fig. S11A); however, we did not detect significant differences in spleen weight within the Gdf11 WE cohort or in thymus or lymph node weights among mice in either cohort (Fig. S11A-C). Taken together, our analyses do not suggest substantial differences in homeostatic immune parameters in mice containing Gdf11 deletion alleles, a www.nature.com/scientificreports www.nature.com/scientificreports/ conclusion that is further supported by two recent studies 72,73 in which lineage-specific ablation of Gdf11 in the hematopoietic system failed to produce overt hematoimmune phenotypes.

Discussion
CRISPR/Cas9 gene editing tools have substantially accelerated the pace at which researchers can generate gene-modified animal models; however, reports of off-target editing events and complex genomic rearrangements following CRISPR-mediated gene editing emphasize the importance of properly validating CRISPR-modified alleles prior to phenotypic analysis. In this report, we present a series of observations stemming from our production of a knock-in Gdf11-IRES-GFP reporter mouse using zygotic CRISPR injections. While we succeeded in generating multiple founder animals harboring correctly targeted Gdf11-IRES-GFP alleles, we also generated additional alleles containing mistargeting events that produced a dramatically altered pattern of GFP expression among hematopoietic lineage cells. We further recovered multiple site-specific deletion alleles that disrupted the endogenous Gdf11 stop codon and induced partial translation of the 3′UTR. While most of these deletions abolished GDF11 function, we also unexpectedly produced a variant allele that codes for a GDF11 protein containing two additional amino acids at its C-terminus (termed "GDF11-WE") and retains substantial activity in vitro and in vivo. Altogether, the variety of unintended genetic alterations generated as a "side effect" of our targeted genome engineering project emphasizes both the power of CRISPR/Cas9 gene editing tools to provide important and even unexpected biological insights into gene expression and function and the absolute necessity for researchers to carefully validate proper integration of knock-in alleles. Such validation is particularly crucial when generating fluorescent reporter alleles, as additional-and in some cases difficult to detect-integration events at the target site can drastically alter phenotypic readouts, as demonstrated here.
Our results also provide new insights into the regulation of Gdf11 expression during aging. Previous work from our group reported enriched Gdf11 expression in the spleen of young mice and found that splenic Gdf11 mRNA levels decline during aging 37,38 . Here, using our validated Gdf11-IRES-GFP reporter mouse, we extend these initial observations to track the dynamics of Gdf11 expression with single cell resolution in peripheral blood cells of aging mice. Our findings that Gdf11-IRES-GFP expression is enriched within T and B lymphocytes during youth (Fig. 3A,B) suggest that lymphoid cells may represent a cellular source that contributes to the total pool of GDF11 in circulation and in tissues. In the T cell compartment, the frequency of Gdf11-expressing cells and the mean fluorescence intensity of Gdf11-IRES-GFP in the periphery do not change significantly during aging. Thus, as the overall frequency of T cells in the periphery declines during aging (Fig. 3H), the contribution to GDF11 production from T cells in the periphery likely declines with increasing age. Our findings from B cells in the periphery are more complex. While the mean fluorescence intensity of Gdf11-IRES-GFP in B cells increases with age, the proportion of Gdf11-expressing peripheral B cells increases only up to 4 months of age, and then gradually decreases. As the overall frequency of peripheral blood B cells increases during aging in mice (Fig. 3H) and the total output of Gdf11 mRNA by B cells decreases (Fig. 3C), these results suggest that the relative abundance of B cells that do not express Gdf11 increases with age, with partial compensation from those Gdf11-expressing B cells that remain. Overall, in comparison to T cells, B cells likely become more significant contributors to the total pool of GDF11 as age increases. Moving forward, the validated Gdf11-IRES-GFP reporter mouse described here will provide a useful tool to further interrogate Gdf11 expression dynamics and the regulation of Gdf11-expressing cells in multiple physiological contexts, including development, homeostasis, aging, and disease.
Our findings that Gdf11-IRES-GFP expression in validated reporter mice is enriched in lymphoid cells are consistent with published data sets assessing Gdf11 expression levels [50][51][52] . A peculiar finding from our analysis of mice harboring alternative Gdf11-IRES-GFP integration events, which remains unresolved, is why the lymphoid-biased GFP expression pattern observed in the correctly targeted reporter lines switched to a myeloid-biased GFP expression pattern in the incorrectly targeted reporter lines. One possible explanation is that the integration of sequences from the plasmid backbone at the target site in the incorrectly targeted lines structurally alters a regulatory element that controls gene expression selectively within either the lymphoid or myeloid lineage. The striking differences in the GFP expression pattern between the different alleles suggest that DNA-based elements may exist that have the capacity to switch gene expression patterns from the lymphoid to the myeloid lineage, which may be interesting to dissect in future studies.
In addition to the intended generation of a Gdf11-IRES-GFP knock-in allele, zygotic targeting at the Gdf11 locus also produced several Gdf11 indel variants, which revealed unexpected insights into how modifications at its C-terminus can alter GDF11's biological function. Previous attempts to add epitope tags to the C-terminus of TGF-β molecules have disrupted protein processing and significantly diminished ligand bioactivity 64,65 . Thus, we expected that adding exogenous amino acids to the GDF11 C-terminus would similarly impair its function. Consistent with this notion, the deletions detected in lines 4A, 4B, and 11, which add 156, 14 and 32 additional amino acids, respectively, to the C-terminus of the protein, each exhibited skeletal and renal phenotypes consistent with Gdf11 null phenotypes 26,33 (Figs. 4C,D and S5A,B). These results suggest that extension of the GDF11 mature ligand by as few as 14 amino acids renders the protein non-functional. On the other hand, the GDF11-WE variant, which is predicted to extend the GDF11 mature ligand by only two amino acids, does not recapitulate GDF11 loss-of-function phenotypes in vivo, as mice homozygous for this allelic variant showed no tail malformations or ectopic ribs (Fig. 5A,B) and survive to adulthood with no major physiological defects. Interestingly, we did detect a decrease in the number of lumbar vertebrae in all heterozygous embryos and a majority of homozygous Gdf11 WE embryos analyzed (Fig. 5D,E). While this feature has been reported in other mutant animals [74][75][76][77] , it has not been reported in mice harboring Gdf11 null alleles 26,33 . Indeed, the loss of lumbar vertebra seen in Gdf11 WE/+ and Gdf11 WE/WE mice appears opposite to that of Gdf11 +/− and Gdf11 −/− mice, which show expanded numbers of lumbar (and thoracic) vertebrae 26 www.nature.com/scientificreports www.nature.com/scientificreports/ or Gasp2, known inhibitors of GDF11, have similarly reported five lumbar vertebrae 34,78 , along with other skeletal abnormalities. Further investigation will be necessary to determine if and how the addition of a tryptophan and glutamic acid to the C-terminus of GDF11 may alter its biological functions and to tease apart whether the specific addition of these two amino acids, as opposed to any others, is key to the preservation of many of GDF11's normal in vivo functions by this variant ligand. Another interesting finding emerging from this study is the presence of a discernable hole within the otic capsule of E18.5 embryos containing one or two copies of a Gdf11 null allele 26 , as well as in several of the Gdf11 indel embryos (Fig. 4E). While the mechanism by which this hole forms, or is retained, during development in the absence of two functional copies of Gdf11 remains unclear, it would be interesting to investigate whether this phenotype has any biological effects.
Finally, in addition to the novel insights into the regulation and function of GDF11, our study has broad implications for utilizing CRISPR/Cas9 genome editing approaches to generate precise gene-modified animal models via HDR. From our efforts to generate a knock-in GFP reporter mouse by CRISPR/Cas9, we generated alleles that were targeted to the intended genomic locus while also containing an additional on-target integration event that strikingly disrupted reporter gene expression. These findings are consistent with a prepublication report from another group that also detected head-to-tail insertions of donor templates at the target site following zygotic CRISPR/Cas9 injections in mice 79 . Disturbingly, these additional integration events were not detected by PCR analysis, emphasizing the need to perform additional screening methods (e.g. Southern blotting, TLA sequencing) to confirm appropriate targeting. For exploratory and pre-clinical CRISPR/Cas9-mediated HDR studies, we recommend verifying correct CRISPR targeting events using these additional methods, as incorrect on-target events can occur in a significant percentage of founder animals and can substantially alter phenotypic readouts.
Four different sgRNAs were designed to target exon 3 of Gdf11 immediately after the stop codon and at the beginning of the 3′UTR. sgRNA sequences were designed using the online tool http://www.genome-engineering. org, and the sgRNA target sequences are listed in Table S3. For each sgRNA, a pair of oligos were designed containing BbsI overhangs and a G nucleotide to facilitate U6-driven expression. Oligo nucleotides were annealed, digested with BbsI (New England Biolabs) and ligated into a BbsI-digested plasmid backbone using T4 ligase (New England Biolabs). The backbone was generated by removing the GFP sequence from the gRNA_GFP_T1 plasmid to contain 2 BbsI sites. gRNA_GFP-T1 was a gift from George Church (Addgene plasmid # 41819; http:// n2t.net/addgene:41819; RRID Addgene_41819) 9 . The plasmid was transformed into Top10 cells and confirmed by Sanger sequencing. Plasmids were purified using the EndoFree Plasmid Maxi Kit (Qiagen).
To profile immune cell activation markers and T and B cell subsets in peripheral blood from mice containing Gdf11 deletion alleles, ACK-lysed peripheral blood cells were incubated with rat anti-mouse CD16/32 (Mouse BD Fc Block, 2.4G2, BD Biosciences, 1:50 dilution) for 5 minutes on ice. One half of the sample was incubated with the following antibodies for 30 minutes on ice to evaluate T cells: anti-CD3-PECy7 ( Magnetic activated cell sorting of CD19+ and CD19− cells. Spleens were dissociated into single cell suspensions using the gentleMACS dissociator in manufacturer provided C-Tubes containing 3 mL MACS buffer (PBS, pH 7.2, 0.5% BSA, 2 mM EDTA). This suspension was passed through 40 μM filters into 15 ml Falcon tubes and placed on ice. Cells were counted using a hemocytometer, resuspended to a density of 10 8 cells/mL, and incubated with a proportional volume of MACS CD19 microbeads (10 μL beads/10 7 cells) at 4 °C for 15 minutes. Using the MACS Separator under a positive selection program, CD19+ cells bound to microbeads were held to a magnetic column while CD19-cells flowed through. Each sample was double sorted using this selection method. CD19+ and CD19− fractions were counted using a hemocytometer with Trypan blue to determine the proportion of live CD19+ cells separated from whole spleen. Approximately 4,000 cells from the CD19+ and CD19− fractions were assessed by flow cytometry to determine purity of fractions and efficacy of magnetic separation. Cells were resuspended in Trizol and stored at −80 °C. T7 Endonuclease 1 (T7E1) mismatch detection assay. Genomic DNA was extracted from FACS-purified mCherry + C2C12 cells with 30 μL of DNA Quick Extract Buffer (Epicentre). The Gdf11 target region was PCR amplified (see Table S4 for primer sequences) using Q5 polymerase (New England Biolabs) according to the manufacturer's protocol. PCR products were purified using a PCR purification kit (Qiagen), denatured at 95 °C for 5 minutes and re-annealed from 95 °C-85 °C at a rate of −2 °C/second, and from 85 °C-25 °C at a rate of -0.1 °C/second to form heteroduplexes. Reannealed heteroduplexes were incubated with 25 units of T7E1 enzyme (New England Biolabs) at 37 °C for 15 minutes, and products from mismatch assay were visualized on a 2% agarose gel.
In vitro transcription. The T7 promoter was added to the sgRNA by PCR amplification of the sgRNA plasmid template using the primers listed in Table S4. The T7-sgRNA3 amplicon was gel purified and used as the template for in vitro transcription using the MEGAscript T7 kit (Life Technologies). The in vitro transcribed sgRNA3 was purified using the MEGAclear Transcription Clean-Up Kit (Thermo Fisher) and eluted in nuclease-free water. The sgRNA was ethanol precipitated with ammonium acetate, rinsed 5 times with 70% ethanol and re-suspended in nuclease-free water.
Evaluation of CRISPR-mediated HDR at Gdf11 locus. For analysis of Gdf11-IRES-GFP targeting in C2C12 cells and mice, genomic DNA was extracted using DNA Quick Extract Buffer (Epicentre) and PCRs were performed with Q5 polymerase (New England Biolabs) according to the manufacturer's protocols. The targeted region of the Gdf11 locus was PCR amplified by semi-nested PCR (see Table S4 for primers used).
F 0 founders were first screened for the presence of the Gdf11-IRES-GFP using an internal PCR amplicon, which is the same amplicon used for the T7E1 assay (see Table S4). Of those mice containing the Gdf11-IRES-GFP (2019) 9:18613 | https://doi.org/10.1038/s41598-019-54766-y www.nature.com/scientificreports www.nature.com/scientificreports/ sequence, the F 0 founders were subsequently screened using two different amplicons that spanned the junction between the genomic DNA and homology arms (see Table S4 for PCR primer sequences). Southern blot. Genomic DNA was isolated from mouse tail in DNA digestion buffer (10 mM Tris pH 8.0, 5 mM EDTA, 0.1 M NaCl, 1% SDS) containing 100 μg proteinase K overnight at 56 °C followed by phenol-chloroform extraction, ethanol precipitation and re-suspension in TE buffer. Genomic DNA was digested overnight with 100 units of NcoI-HF (New England Biolabs) at 37 °C, separated on a 0.8% agarose gel and transferred to a positively charged nylon membrane (Roche). Membranes were hybridized with digoxigenin (DIG)-labelled internal probes (Roche) and visualized by chemiluminescence using an AP-anti-DIG antibody (Roche) 81 and CDP-Star (Roche). Membranes were stripped using 0.2 M sodium hydroxide with 1% SDS at 37 °C, rehybridized with DIG-labelled external probes and visualized using the same chemiluminescence protocol. DIG-labelled probes were generated by PCR using the DIG probe synthesis kit (Roche) according to manufacturer's instructions (see Table S4 for primer sequences).
Targeted locus amplification (TLA) sequencing. Spleens from Gdf11-IRES-GFP mice were homogenized and subjected to ACK lysis. Splenocytes were frozen and shipped to Cergentis (Utrecht, the Netherlands) for TLA sequencing analysis as described previously 48,49 . Briefly, splenocyte genomic DNA was crosslinked, digested and re-ligated. Genomic DNA was subsequently purified and circular TLA templates were amplified using two independent sets of inverse primers complementary to the transgene (see Table S4 for primer sequences). Following amplification of the targeted locus, PCR amplicons were purified and sequencing libraries were prepared for Illumina sequencing.
RNA isolation, cDNA synthesis and real-time PCR. Total RNA was isolated using Trizol (Thermo Fisher) according to the manufacturer's protocol. RNA concentration was determined using a NanoDrop (Thermo Fisher) and equal amounts of RNA were added to cDNA synthesis reactions. cDNA was synthesized using the SuperScript III First-Strand Synthesis SuperMix for qRT-PCR (Thermo Fisher) according to the manufacturer's protocol. Real-time PCR was performed on an ABI Prism 7900HT sequence detection system using RT 2 SYBR Green/ROX FAST master mix (Qiagen) using the primers listed in Table S4. Relative quantification was calculated as 2( −ΔΔCT ) using β-actin as a reference gene. For analysis of Gdf11 expression from the MACS-purified CD19+ and CD19-splenic cells, real-time PCR analysis was performed with Taqman probes for Gdf11 (Mm01159973_m1, Taqman Gene Expression Assays, Thermo Fisher) and Hprt (Mm01545399_m1, Taqman Gene Expression Assays, Thermo Fisher). Relative quantification was calculated as 2( −ΔΔCT ) using Hprt as a reference gene. in situ Hybridization. Antisense DIG-labeled riboprobes against Gdf11 and Gfp were in vitro transcribed from PCR fragments (Roche, 1175025910) that were generated using primers listed in Table S4. PCR products were sequenced verified prior to in vitro transcription. In situ hybridization was performed on four embryos per genotype at E10.5. For whole-mount in situ hybridization, embryos were fixed in 4% paraformaldehyde for 18 hours at 4 °C, washed 3x for 10 minutes in 1X PBS, and then dehydrated through a graded series of 25%, 50%, 75% methanol/0.85% NaCl incubations. Embryos were stored in 100% methanol at −20 °C before in situ hybridization. Embryos were rehydrated through a graded series of 75%, 50%, 25%, methanol/0.85% NaCl incubations and finally washed in 2x PBS with 0.1% tween-20 (PBST). E10.5 embryos were treated with 10 mg/mL proteinase K in PBST for 30 minutes at room temperature. Samples were then fixed in 4% paraformaldehyde/0.2% glutaraldehyde in PBST for 20 minutes at room temperature and washed 2x in PBST. Embryos were incubated in pre-hybridization solution for 1 hour at 68 °C, and then incubated in 500 ng/mL of riboprobe at 68 °C for 16 hours. Post-hybridization, samples were washed in stringency washes and incubated in 100 μg/mL RNase A at 37 °C for 1 hour. Embryos were washed in 1X maleic acid buffer with 0.1% tween-20 (MBST) and then incubated in Blocking Reagent (Roche) with 10% heat inactivated sheep serum (Sigma # S2263) for 4 hours at room temperature. Anti-DIG antibody (Roche, 11093274910) was used at 1:5000 and samples were incubated for 18 hours at 4 °C. Samples were washed 8x with MBST for 15 minutes at room temperature, 5x in MBST for 1 hour at room temperature, and then 1x in MBST for 16 hours at 4 °C. Prior to development, embryos were incubated 3x in NTMT (100 mM NaCl, 100 mM Tris-HCl (pH9.5), 50 mM MgCl2, 0.1% tween-20, and 2 mM levamisole), and the in situ hybridization signal was developed by adding BM Purple (Roche, 11442074001) for 5 hours at room temperature. After the colorimetric development, samples were fixed in 4% paraformaldehyde and cleared through a graded series of glycerol / 1X PBS and stored in 80% glycerol. Skeletal preparation protocol. E17.5 and E18.5 embryos were dissected, deskinned, eviscerated, fixed in 100% ethanol for 24 hours at room temperature and dehydrated in 100% acetone for 24 hours at room temperate. Skeletons were incubated in staining solution (0.3% Alcian blue 8GS (Sigma), 0.1% Alizarin red S (Sigma), 70% ethanol and 5% acetic acid) for 72 hours. Skeletons were cleared in 1% potassium hydroxide solution for 24 hours, followed by additional clearing increasing the percentage of glycerol from 20% to 50% to 80% after every 24-hour period. Embryos were imaged and stored in 80% glycerol with 1% potassium hydroxide. (2019) 9:18613 | https://doi.org/10.1038/s41598-019-54766-y www.nature.com/scientificreports www.nature.com/scientificreports/ Grip strength assay. Mice were allowed to grasp the metal grid of the grip strength meter (Bioseb) with only their forelimbs. Mice were pulled backwards, and the force applied to the grid was recorded in Newtons. Raw force measurements were performed in duplicate, recorded and averaged. The force normalized to the body weight for each animal was calculated.
Collection of peripheral blood, serum, and tissues. Peripheral blood samples were collected from adults via the lateral tail vein into PBS (Gibco) containing 10 mM EDTA and stored on ice. Red blood cells were sedimented using 2% Dextran (Sigma) in PBS for 30 minutes at 37 °C, followed by washing with Staining Media (HBSS containing 2% FBS), incubation with ACK lysis buffer (Thermo Fisher) and filtration through a 40 μM cell strainer for flow cytometry analysis.
For serum collection, peripheral blood was collected via the tail vein into Microtainer tubes with a serum separator (BD) and incubated for at least 30 minutes at room temperature. Samples were centrifuged at 2000xg for 10 minutes at room temperature. The upper serum layer was transferred to a new tube and stored at −80 °C prior to analysis.
For tissue collection from adults, mice were euthanized with CO 2 prior to tissue collection. The heart was dissected, cleaned in PBS and dried on a paper towel prior to weighing. The spleen, thymus, and inguinal lymph node were dissected, weighed and either stored in staining media or snap-frozen in liquid nitrogen. The tibialis anterior and extensor digitorum longus (TA and EDL) muscles were harvested and weighed. The tibia bone was cleaned and measured using calipers.
For tissue collection from E18.5 embryos, pregnant females were euthanized with CO 2 and embryos were dissected in PBS. Urogenital tracts were dissected in PBS and imaged. Embryos were scored for the presence of 0, 1 or 2 kidneys by visual inspection.

Complete blood count (CBC) analysis. Peripheral blood was collected via tail vein in EDTA-coated
Microtainer tubes (BD) and kept on ice. CBCs were analyzed on an Element HT5 (HESKA) instrument. cytokine analysis. Serum was isolated and shipped to Eve Technologies for cytokine analysis using the Mouse Cytokine Array/Chemokine Array 31-Plex (MD31). To enable statistical analyses, values reported as out of range below the 4 or 5 parameter logistic standard curve were inputted as 0.02 pg/mL, the lowest extrapolated value that can be calculated by the standard curve mathematical formula.
Immunoglobulin isotyping array. Serum was diluted 100-fold in Sample Diluent and hybridized to the Rapid Mouse Ig Isotyping Array (Ray Biotech, Norcross, GA, USA) according to the manufacturer's protocol. Following hybridization, slides were shipped to Ray Biotech for scanning and data extraction. Normalized fluorescence values were calculated using the AAM-ISO-G1 Analyzer program (Ray Biotech, Norcross, GA, USA).
Liquid chromatography tandem mass spectrometry of GDF11 and GDF8. A minimum of 100 μL of mouse serum was submitted for quantitative liquid chromatography tandem mass spectrometry to detect GDF11 and GDF8 protein concentrations 53 at the Brigham and Women's Hospital Brigham Research Assay Core (BRAC). Briefly, mouse serum was denatured, reduced and alkylated, followed by pH-based fractionation using cation ion exchange SPE; the appropriate elution fraction was digested with trypsin. After desalting and concentrating of the tryptic digest, the peptide mixture was separated and eluted by liquid chromatography followed by mass spectrometric analysis operated in positive electrospray ionization mode. The most intensive and unique proteotypic peptides from GDF11 and GDF8 as surrogate peptides along with heavy-labeled unique peptides as internal standards were used for quantitative determination of GDF11 and GDF8.
Site-directed mutagenesis. Primers used for site-directed mutagenesis were designed to target nucleotides in our previously described chimeric DNA template that consisted of human GDF8 prodomain attached to the mature GDF11 protein sequence, annotated GDF8pro/GDF11mature in the pRK5 vector (see Table S4 for primer sequences) 58 . Successful mutagenesis for all constructs was confirmed by sequencing. HEK293 (CAGA) 12 luciferase assay. As previously described, HEK293-(CAGA) 12 luciferase reporter cells were plated at ~2 × 10 4 cells/well and grown overnight in 96-well plates and transiently co-transfected with plasmids containing the GDF8pro/GDF11mature constructs (25-100 ng; pRK5), human tolloid-like 1 (TLL1; 50 ng; pcDNA3), and human furin (50 ng; pcDNA4) using Mirus LT-1 transfection reagent. Empty pRK5 vector was added for a total of 200 ng DNA transfected/well. Cells were transfected for 6 hours followed by removal of growth medium, replaced by serum-free medium and cultured for an additional 24 hours. Cells were lysed and luminescence was recorded immediately using a BioTek Synergy H1 Hybrid plate reader. Values from triplicates were averaged and each average was normalized to the negative control cells transfected with empty pRK5 vector, furin, and TLL1. The activity data was imported into GraphPad Prism for graphing and statistical analysis.
Western blot. Conditioned media was collected from Expi293 cells (Invitrogen; #A14527) four days following transfection with the GDF11-WE plasmid. Increasing volumes of conditioned media (5 μL, 10 μL, or 20 μL) or 10 ng of recombinant GDF11 (rGDF11) protein (Peprotech) were processed under non-reducing or reducing conditions. For reducing conditions, samples were denatured in the presence of β-mercaptoethanol prior to separation on a 4-15% SDS gel (Bio-Rad; #4561086). For non-reducing conditions, samples were denatured, separated on an SDS gel, and the gel was reduced after separation by electrophoresis in the presence of 575 mM β-mercaptoethanol diluted in SDS running buffer (25 mM Tris pH 8.5, 192 mM glycine, 0.1% SDS) and incubated for 20 minutes at room temperature. Proteins were transferred to a PVDF membrane using the Bio-Rad Trans-Blot Turbo ™ Transfer System. The membrane was blocked with 5% milk diluted in TBST (10 minutes, (2019) 9:18613 | https://doi.org/10.1038/s41598-019-54766-y www.nature.com/scientificreports www.nature.com/scientificreports/ rocking) and incubated with primary antibody (Rabbit Anti GDF11/GDF8 Abcam Cat# EPR4567 38 ) at a 1:1000 dilution in 1.5% milk in TBST (2 hours, rocking). The membrane was washed 5 times with TBST (5 minutes per wash) and incubated with secondary antibody (Anti Rabbit R&D antibody Cat#HAF008 82 ) at a 1:3000 dilution in 1.5% milk in TBST (1 hour, rocking). The membrane was washed 5 times with TBST (5 minutes per wash), incubated using Pierce ECL Plus Western Blotting Substrate, and developed. Equipment and settings. Agarose gel electrophoresis images (Figs. 1B,D, 5F, S1C,D, S2A,B and S12A,D) were acquired on a Biorad Molecular GelDoc XR imaging system using Biorad ImageLab software. For all images other than Fig. S12A, digital images were imported into Photoshop CC 2017 (Adobe), cropped when indicated, and no other post-image processing was performed. For the image in S12A, the gel image was printed from ImageLab software onto Thermal Paper (110 mM × 20 M, Bio Doc-IT accessory, Genesee Scientific, #K65HMCE) using a Mitsubishi P95DW printer and digitally scanned using a Scanjet 8300 (HP). The image was rotated and cropped to focus on the agarose gel using Photoshop CC 2017 (Adobe), and no other post-image processing was performed.
Southern blot images (Figs. 2A,B, S3A,B and S12B,C) were acquired by exposing membranes to Biomax Maximum Sensitivity (MS) autoradiography film (Kodak). Films were fixed and developed using an M35A X-Omat Processor (Kodak) and digitally scanned using a Scanjet 8300 (HP). Brightness and contrast levels were adjusted uniformly across the entire scanned images using Photoshop CC 2017 (Adobe).
For in situ hybridization images (Fig. 2K,L), embryos were placed in a 10 cm 2 tissue culture dish filled with 80% glycerol in PBS. Embryos were imaged on a Leica M216FA stereomicroscope (Leica Microsystems) equipped with a DFC300 FX digital imaging camera. Images were acquired with Leica Application Suite v2.3.4 R2 software. Forelimbs were acquired at 51x magnification and embryo images were acquired at 17x magnification. In situ hybridization images were processed in Adobe Bridge, where the following adjustments were applied to all images: exposure +1.10, tint +13, temperature +3.
For mouse embryo images (Figs. 4C and 5A), embryos were photographed using a Samsung Galaxy S9+ camera. Brightness and contrast levels were adjusted uniformly across the entire scanned images using Photoshop CC 2017 (Adobe). Skeletal preparations (Figs. 4D,E and 5B-D) were photographed using a Nikon D7000 with a Nikon 28-105 lens with macro. Images were captured in Nikon Electronic Format (NEF) and adjusted in Adobe Camera Raw plugin to a + 0.70 exposure and color temperature of 6700 K.
For embryonic urogenital tract images (Fig. S5A), tissues were imaged using a Leica MZ16 FA fluorescence stereomicroscope equipped with a Leica DFC7000 T camera. Images were captured in.lif format using Leica Application Suite X and exported as.tif files with no image post-processing performed.
Transfected C2C12 cells (Fig. S1F) were imaged on a Zeiss Observer.D1 inverted fluorescence microscope using AxioVision Rel 4.7 software. Western blot images (Fig. S6F) were acquired by exposing membranes to GeneMate blue autoradiography film. Films were fixed and developed using an M35A X-Omat Processor (Kodak) and digitally scanned using an Epson GT-1500. Images were flipped horizontally, converted to grayscale, and brightness levels were adjusted uniformly across the entire scanned images using Photoshop CC 2017 (Adobe).

Statistical analysis.
For the Gdf11-IRES-GFP mice, we performed statistical analyses to assess the association between aging and GFP+ cell frequency as well as GFP mean fluorescence intensity (MFI). For the percentage analyses, we fit the logistic regression model and applied the generalized estimating equations (GEE) approach 83,84 using the gee package in the statistical software R (v3.5.3) 85 . The GEE estimation method allows the analysis to account for correlations between repeated measurements. For normalized MFI responses, we used linear regression models with the GEE approach. For each of the analyses, the adequacy of the statistical models was assessed through the visual inspection of standardized residuals and fitted values. Both the residual plot and time plot of the mean response in Fig. 3E,G suggested that the B cell percentage and MFI data required multiple linear trends over time. Therefore, we used a piecewise regression model, which allows two linear models to be fit to the B cell data, one for ≤4 months of age and the other for age >4 months of age. Two-sided p-values for all tests are presented in Fig. 3D-G.
For all other analyses, comparisons between three or more groups were performed by One-Way ANOVA followed by Bonferroni post-test correction. Observed differences with p < 0.05 were considered as statistically significant.

Data availability
For original data, please contact Amy Wagers at amy_wagers@harvard.edu.