Abstract
Growth differentiation factor 11 (GDF11) and GDF8 (MSTN) are closely related TGF-β family proteins that interact with nearly identical signaling receptors and antagonists. However, GDF11 appears to activate SMAD2/3 more potently than GDF8 in vitro and in vivo. The ligands possess divergent structural properties, whereby substituting unique GDF11 amino acids into GDF8 enhanced the activity of the resulting chimeric GDF8. We investigated potentially distinct endogenous activities of GDF11 and GDF8 in vivo by genetically modifying their mature signaling domains. Full recoding of GDF8 to that of GDF11 yielded mice lacking GDF8, with GDF11 levels ∼50-fold higher than normal, and exhibiting modestly decreased muscle mass, with no apparent negative impacts on health or survival. Substitution of two specific amino acids in the fingertip region of GDF11 with the corresponding GDF8 residues resulted in prenatal axial skeletal transformations, consistent with Gdf11-deficient mice, without apparent perturbation of skeletal or cardiac muscle development or homeostasis. These experiments uncover distinctive features between the GDF11 and GDF8 mature domains in vivo and identify a specific requirement for GDF11 in early-stage skeletal development.
Introduction
The TGF-β superfamily of proteins is well known for regulating embryological development, wound healing, and adult tissue maintenance. In recent years, two highly homologous TGF-β proteins—growth differentiation factor 11 (GDF11) and GDF8 (also known as myostatin/MSTN)—have garnered substantial interest with evidence of their roles in aging and regenerative processes (Loffredo et al, 2013; Biesemann et al, 2014; Sinha et al, 2014; Poggioli et al, 2016; Walker et al, 2016; Du et al, 2017). Due to the 89% amino acid sequence identity in their C-terminal signaling domains, GDF11 and GDF8 have been viewed as serving redundant functions in vivo (McPherron et al, 2009; Poggioli et al, 2016; Walker et al, 2016) (Fig 1A). Yet, growing evidence suggests that GDF11 and GDF8 have distinct potencies and different spatiotemporal functions in vivo. As members of the activin subclass of TGF-β, both GDF8 and GDF11 signal through the type I receptors ALK4, ALK5, and ALK7 (Rebbapragada et al, 2003; Andersson et al, 2006; Walker et al, 2017; Lee et al, 2020). Molecularly, they are synthesized as precursors that remain in an inactive, latent complex until a Tolloid-like (TLD) protease cleaves the ligand prodomain to relieve the mature domain from inhibition (Lee & McPherron, 2001; Rebbapragada et al, 2003; Wolfman et al, 2003; Ge et al, 2005; McFarlane et al, 2005; Anderson et al, 2008). Mature GDF11 and mature GDF8 each consist of two monomers linked by disulfide bonds to form a homodimer of propeller-like shape (Fig 1B), which creates symmetrical concave and convex surfaces used for receptor binding (Yadin et al, 2016; Walker et al, 2017). To signal, the ligands assemble a combination of two type II and two type I Ser/Thr kinase receptors that have a single extracellular ligand-binding domain (Allendorph et al, 2006; Weber et al, 2007). Assembly of this complex allows the type II receptor to phosphorylate the type I receptor, which initiates the SMAD signaling cascade (Weiss & Attisano, 2013).
(A) Gdf11 unique amino acid residues are highlighted in orange and Gdf8 unique residues in blue. (B) Surface representation of GDF8 (monomer A in blue and monomer B in grey) showing location of unique GDF8 amino acid residues in orange, with G89 and E91 highlighted in red. (C) Schematic of CRISPR/Cas9 strategies for changing residues in Gdf11 and Gdf8 native loci. In Strategy 1, the D89 and Q91 amino acid residues within Gdf11 are changed for the analogous G89 and E91 amino acid residues from Gdf8, and the G89 and E91 amino acid residues within Gdf8 are changed for the analogous D89 and Q91 residues from Gdf11. (D) In Strategy 2, the full mature domain of Gdf8 is changed for the full mature domain from Gdf11. Amino acid residues of Gdf11 are shown in orange, and those of Gdf8, in blue.
GDF8 is expressed postnatally, predominantly by skeletal and cardiac muscles, and is a well-recognized negative regulator of muscle growth (McPherron & Lee, 1997; McPherron et al, 1997; Lee, 2012). GDF11 is expressed more broadly across multiple tissues, with noted involvement in early development (McPherron et al, 1999; Wu et al, 2003; Kim et al, 2005; Liu, 2006; McPherron et al, 2009). Genetic mutation of Mstn (Gdf8) results in hypermuscular, hypo-adipose phenotypes in numerous animal species, including humans (McPherron & Lee, 1997; McPherron et al, 1997; Schuelke et al, 2004; Clop et al, 2006; Mosher et al, 2007), whereas homozygous deletion of Gdf11 leads to axial skeletal malformation and defects in organ development in mice (McPherron et al, 1999). In addition, recent evidence suggests that genetic loss of GDF11 function in humans causes multisystem pathology with variable impact on the skeleton, nervous system, heart, muscle, and/or connective tissue (Ravenscroft et al, 2021). Importantly, Gdf11-null mice exhibit perinatal lethality, whereas Mstn-null (Gdf8−/−) mice do not (McPherron et al, 1999), and lower levels of Mstn in Gdf8+/− heterozygotes may actually extend lifespan (Mendias et al, 2015). These differences in postnatal survival following genetic manipulation make comparative studies of Gdf11 versus Gdf8 activities in vivo particularly difficult, while reports on GDF11’s essential functions in adulthood—most of which have relied on the use of exogenous recombinant proteins—are incompletely defined. Nevertheless, interest persists from pharmaceutical and biotechnology companies in the potential effects of GDF11 in age-related organ dysfunction (Loffredo et al, 2013; Katsimpardi et al, 2014; Sinha et al, 2014), and several studies support the notion that exogenous GDF11 may regulate cardiac hypertrophy and skeletal muscle repair in older animals (Loffredo et al, 2013; Sinha et al, 2014; Du et al, 2017).
In a prior study, we demonstrated that GDF11 and GDF8 differ in their signaling properties in multiple cell lines and cultured primary myoblasts, with GDF11 signaling at lower concentrations than GDF8 and more efficiently using the type I receptors ALK4, ALK5, and ALK7 (Walker et al, 2017). We define this ability to activate downstream pathways at lower concentrations as having greater potency. We further showed that administration of GDF11 in vivo more potently induces SMAD phosphorylation in the myocardium compared with GDF8 (Walker et al, 2017). These differences implicate residue differences between GDF11 and GDF8, particularly those clustered around the type I binding interface, in determining signaling potency, likely via effects on dimer stability and stability of receptor interactions. Consistent with this possibility, structural analysis and mutational studies of the ternary complex of GDF11 with type I receptor Alk5 and type II receptor ActRIIB revealed that different mechanisms regulate specificity and binding with type I receptor, compared with TGF-β, providing an explanation for how GDF11 and the TGF-β activin class more effectively facilitate low-affinity type I interactions (Goebel et al, 2019). These biochemical and structural studies indicate that GDF11 and GDF8 are unlikely to be functionally equivalent, especially when ligand concentrations are low, as they typically are in vivo (Walker et al, 2017; Goebel et al, 2019).
In this study, we evaluate GDF11 and GDF8 functional equivalence in vivo by using the CRISPR/Cas9 system to introduce GDF8-like amino acid substitutions into GDF11, and GDF11-like substitutions into GDF8. These sequence alterations in GDF11, which previously were shown to diminish signaling potency of the resulting protein (Walker et al, 2017), caused a perturbation of the axial skeletal structure of mutant mice during development that persists into adulthood. In contrast, the sequence alterations introduced into GDF8, which previously were shown to increase the signaling potency of the resulting ligand (Walker et al, 2017), did not produce observable developmental phenotypes. As such, we generated a third line of mutant animals, in which the entire GDF8 mature domain was replaced with the corresponding mature domain sequence of GDF11, resulting in full replacement of the endogenous GDF8 signaling domain with that of GDF11. These mature domain (MD) mutants had up to 50-fold greater levels of GDF11 in circulation, with concomitant depletion of GDF8 to undetectable levels and showed modestly decreased skeletal muscle mass, with no apparent impact on postnatal survival, total adult body weight, or the development and function of other organ systems.
While we were performing our study, the Se-jin Lee group published a study that used a similar genetic approach as our third mouse line to replace the Mstn gene sequences encoding the mature C-terminal peptide with the full mature domain of Gdf11 (Lee et al, 2022). Their characterization of these mice (Lee et al, 2022) supports our data here, showing that GDF11 entirely replaced circulating MSTN and increased GDF11 levels ∼30–40-fold (Lee et al, 2022). However, our findings extend these observations, addressing the converse hypothesis as well and showing that diminution of GDF11 potency—through targeted replacement of two key amino acids from GDF8—causes a significant developmental defect in osteogenesis, distinct from that seen with modulation of GDF8. Taken together, our findings elucidate precise, differential molecular mechanisms underpinning the biological actions of GDF11 and GDF8 that cannot be explained solely by differences in in vivo ligand concentrations and patterns of expression. They also provide direct evidence that structural and biochemical differences in these ligand’s mature signaling domains contribute significantly to their unique roles in mammalian development and organ physiology.
Results
Generation and characterization of chimeric amino acid GDF11 and GDF8 mice
We previously reported that substitution of two residues located in the fingertip of GDF11 (D89 and Q91) into the analogous region of GDF8 (in place of G89 and E91) enhanced SMAD signaling activity of the hybrid GDF8 molecule by ∼50% (Walker et al, 2017). This result indicates that sequence differences in the mature GDF11 and GDF8 proteins are likely responsible for differences in ligand signaling and function. To address whether such sequence-determined signaling differences impact in vivo activities of GDF11 and GDF8, we used CRISPR/Cas9 to create two lines of chimeric mice (Strategy 1; Fig 1C), in which we replaced D89 and Q91 residues within the Gdf11 locus with the analogous G89 and E91 residues from Gdf8, or conversely replaced G89 and E91 within native Gdf8 with D89 and Q91 from Gdf11. The third chimeric line we generated replaced the full mature domain region within the Gdf8 locus with the corresponding region from Gdf8 (Strategy 2; Fig 1D). Based on our prior in vitro studies, substitution of all the unique GDF11 residues into GDF8 in this manner is able to enhance signaling of the resulting protein to be ∼5-fold more potent than WT GDF8 (Walker et al, 2017).
To generate the two amino acid–modified lines, we constructed genetically modified Gdf11 (Fig S1A) and Gdf8 (Fig S1B) single-stranded DNA (ssDNA) donor plasmids containing the mutant codons, flanked by ∼80-bp homologous arms. In generating the full mature domain replacement line, we constructed a genetically modified Gdf8 double-stranded DNA (dsDNA) donor plasmid (Fig S1C) containing the GDF11 mature domain sequences, flanked by ∼4-kb homologous arms. After homologous recombination in embryonic stem cells, targeted microinjections into C57BL/6J zygotes, and implantation of zygotes into C57BL/6J surrogate females, we produced F0 founders with the chimeric allele incorporated in the germline. The ssDNA donor template incorporating Gdf8-like G89 and E91 residues into the native Gdf11 locus (Fig S1A) also introduced an AseI restriction enzyme unique to Gdf8 as a genetic marker for downstream genotyping. Likewise, the ssDNA donor template incorporating Gdf11-like D89 and Q91 residues into the native Gdf8 locus (Fig S1B) and the dsDNA donor template containing the Gdf11 full mature domain sequences (Fig S1C) removed the same AseI site from the Gdf8 locus. We verified successful integration of the chimeric constructs at the Gdf11 and Gdf8 loci via Sanger sequencing (Fig S1A–C) and identified F0 founders through PCR validation and subcloning for Gdf11Gdf8aa (Fig S1D), Gdf8Gdf11aa (Fig S1E), and Gdf8Gdf11MD (Fig S1F) lines. Downstream genotyping confirmed the presence of the unique AseI restriction enzyme site in the modified Gdf11 locus (Fig S1G) and the absence of the same AseI site at the modified Gdf8 loci (Fig S1H and I). We further confirmed integration of silent mutations included in the donor templates, whose purpose was to mutate the PAM sequence to prevent further cutting after donor construct integration (Fig S1A–C). Collectively, these results validated our genetic modification strategy to integrate Gdf11-like and Gdf8-like changes into the Gdf8 and Gdf11 loci, respectively.
To generate Gdf11Gdf8aa, Gdf8Gdf11aa, and Gdf8Gdf11MD mutant lines, DNA donor plasmids were designed to contain the desired amino acid or full mature domain mutations. (A, B, C) For the amino acid mutant lines (A) Gdf11Gdf8aa and (B) Gdf8Gdf11aa, single-stranded donor templates were flanked by 80-bp homologous arms, whereas a double-stranded donor template was flanked by 4-kb homologous arms, which was constructed for the full mature domain mutant (C) Gdf8Gdf11MD. Schematics of the CRISPR strategy depict the native Gdf11 and Gdf8 loci, with locations of the gRNA cut site (top), the donor templates containing the codons to incorporate the Gdf8-like and Gdf11-like amino acid mutations (middle), and the resulting chimeric mutants (bottom). (A, B, C, D, E, F) Successful transgene modification was confirmed via Sanger sequencing (A, B, C) and PCR validation and subcloning, identifying F0 founders for (D) Gdf11Gdf8aa, (E) Gdf8Gdf11aa, and (F) Gdf8Gdf11MD mouse lines. (G) In the process of genotyping, digested bands from incubation of DNA with restriction enzyme AseI were shown for Gdf11+/8aa and Gdf118aa/8aa samples, which acquired the AseI site from the donor template. (H) Gdf8+/11aa and Gdf811aa/11aa DNA could no longer be digested fully after incubation with AseI, indicating the absence of the excised native AseI site. (I) Gdf8+/11MD and Gdf811MD/11MD DNA also could not be digested fully after incubation with AseI. Amino acid residues of Gdf11 are represented in orange, and those of Gdf8, in blue.
Through this process, we successfully generated the following:
(1) Gdf11Gdf8aa mice (with Gdf8 amino acid residues G89 and E91 replacing the corresponding residues in Gdf11) (Fig S1A).
(a) Mono-allelic (Gdf11+/8aa), bi-allelic (Gdf118aa/8aa), WT (Gdf11+/+).
(2) Gdf8Gdf11aa mice (with Gdf11 amino acid residues D89 and Q91 replacing the corresponding residues in Gdf8) (Fig S1B).
(a) Mono-allelic (Gdf8+/11aa), bi-allelic (Gdf811aa/11aa), WT (Gdf8+/+).
(3) Gdf8Gdf11MD mice (with the Gdf11 mature domain replacing the Gdf8 mature domain in the Gdf8 locus) (Fig S1C).
(a) Mono-allelic (Gdf8+/11MD), bi-allelic (Gdf811MD/11MD), WT (Gdf8+/+).
We backcrossed the knock-in alleles five generations (to F5) in each chimeric line before analyses to breed out potential off-target modifications and confirmed Mendelian ratios of allele inheritance to rule out potential embryonic lethality resulting from the genetic modifications (Table S1). Targeted locus amplification (TLA) sequencing (de Vree et al, 2014), performed on bone marrow DNA harvested from F5 mono-allelic and bi-allelic offspring from Gdf11Gdf8aa (Fig S2A and B), Gdf8Gdf11aa (Fig S2C and D), and Gdf8Gdf11MD (Fig S2E and F) mice and aligned to the mouse mm10 genome sequence, further confirmed correct integration of the desired mutant sequences into the native Gdf11 and Gdf8 loci, with no evidence across the whole genome of structural variation surrounding the integration site or within the insert, of incorrect or off-target integration events, or of a locus duplication at the integration site matching the WT allele. Taken together, these results indicate that the Gdf11Gdf8aa and Gdf8Gdf11aa chimeric lines successfully integrated the intended Gdf8 or Gdf11 nucleotide alterations leading to the anticipated amino acid changes, with no other genomic off-target mutations detected, and that the Gdf8Gdf11MD line successfully integrated the full mature domain sequences of Gdf11 and replaced the native mature domain region of Gdf8, with no evidence of incorrect targeting.
(A) In Gdf11Gdf8aa mice, detected mutations were in exon 3 of Gdf11 on chr10 (RefSeq). (B) In Gdf118aa/8aa samples, two mutated bases (G→C = D89G and T→C = Q91E) were found at 100% frequency, indicating bi-allelic mutation. In Gdf11+/8aa samples, the mutations occurred at ∼50% frequency, indicating mono-allelic mutation. Three additional silent mutations—one for the insertion of the AseI site unique to Gdf8 and two for the mutations—were also confirmed. (C) In Gdf8Gdf11aa mice, mutations were in exon 3 of Gdf8 on chr1 (RefSeq). (D) In Gdf811aa/11aa samples, two mutated bases (G→A = G89D and G→C = E91Q) were found at 100% frequency, indicating bi-allelic mutation. In Gdf8+/11aa samples, the mutations occurred at ∼50% frequency, indicating mono-allelic mutation. Two additional base changes, accounting for the PAM site silent mutation, and the absence of the AseI site were also confirmed. (E) In Gdf8Gdf11MD mutants, the full GDF11 mature domain sequence was found on chr1, in addition to the native Gdf8 locus on chr10. No structural variants, off-target integration, or locus duplication was detected. (F) Gdf811MD/11MD samples lacked WT Gdf8 mature domain sequences at the integration site and in the deleted region, confirming bi-allelic replacement of native GDF8, whereas Gdf8+/11MD samples had WT reads at the integration site and in the deleted region on one allele, confirming mono-allelic replacement of native GDF8. The mouse mm10 reference sequence was used for comparisons. Amino acid residues of Gdf11 are shown in orange, and those of Gdf8, in blue.
Circulating GDF11 concentration increases 50-fold in Gdf8Gdf11MD mutants, whereas GDF11 and GDF8 levels in Gdf11Gdf8aa and Gdf8Gdf11aa mutants remain unchanged
To determine whether full replacement of the mature domain of native Gdf8 with Gdf11 altered circulating protein levels in vivo, we collected serum from Gdf8+/+, Gdf8+/11MD, and Gdf811MD/11MD mice at 10–14 wk of age and measured endogenous GDF11 and GDF8 levels, using a liquid chromatography–tandem mass spectrometry assay that distinguishes between GDF11 and GDF8 by detecting two differential peptide fragments between the two proteins (Garbern et al, 2019) (Fig 2). In the bi-allelic Gdf811MD/11MD mutants (n = 10), circulating GDF11 concentrations increased ∼50-fold above normal levels (Fig 2A), whereas circulating GDF8 concentrations decreased below the level of detection (<LOD) (Fig 2B). The same trend in GDF11 and GDF8 concentrations was observed in male and female mice separately, with significant differences in ligand concentration varying according to allelic dosage across the three genotypes (Gdf8+/+, n = 10; Gdf8+/11MD, n = 8; and Gdf811MD/11MD, n = 10). The levels of GDF11 peptides measured in Gdf811MD/11MD mutants (Fig 2A) were comparable to the levels of GDF8 in Gdf8+/+ mice (Fig 2B) and to the combined concentrations of GDF11 + GDF8 in Gdf8+/11MD mice (Fig 2A and B), suggesting a direct replacement of mature GDF11 for GDF8, with expression levels ultimately determined by either cis-regulatory elements in the Gdf8 locus or association with the GDF8 prodomain, or a combination of these two factors. In support of this interpretation, the total pool of GDF11 + GDF8 did not differ for any of the chimeric genotypes (Fig 2C). Mass spectrometry data from the Se-jin Lee group of the Gdf8 full coding region–replaced chimeras corroborate our findings (Lee et al, 2022). They also reported no detectable MSTN and a ∼30–40-fold increase in circulating GDF11 in MstnGdf11/Gdf11 mice (Lee et al, 2022), with Mstn+/Gdf11 mice having intermediate levels of the two ligands (Lee et al, 2022). Our study further expands this analysis of the impact of mature domain sequence on ligand expression by measuring endogenous GDF11 and GDF8 in serum from Gdf11Gdf8aa and Gdf8Gdf11aa mutants as well, showing that serum GDF11 and GDF8 protein concentrations are not significantly altered in either mono-allelic or bi-allelic Gdf11Gdf8aa or Gdf8Gdf11aa mutants, compared with WT mice (Fig 2A–C). Based on the biochemistry of the ligands, we do not anticipate that the matrix binding of GDF11 and GDF8 would be affected, though we have not studied this. These data indicate that the dual amino acid substitutions alone did not impact endogenous expression or circulation of GDF11 or GDF8. Therefore, despite some reports that high levels of GDF11 in humans are associated with adverse health consequences (Egerman et al, 2015: Hinken et al, 2016; Hammers et al, 2017), our data, together with those of Lee and colleagues (Lee et al, 2022), indicate that GDF11 can rise to extremely high levels in vivo without apparent negative health consequences or premature death.
(A, B) Liquid chromatography–tandem mass spectrometry measurements of (A) GDF11 and (B) GDF8 in Gdf11+/+ (n = 7), Gdf11+/8aa (n = 8), and Gdf118aa/8aa (n = 10) mice (left plots); in Gdf8+/+ (n = 7), Gdf8+/11aa (n = 6), and Gdf811aa/11aa (n = 9) mice (middle plots); and in Gdf8+/+ (n = 10), Gdf8+/11MD (n = 8), and Gdf811MD/11MD (n = 10) mice (right plots). <LOD, below the level of detection. The same trends were observed in male and female mice separately. (C) Combined concentration of GDF11 and GDF8 in Gdf11Gdf8aa, Gdf8Gdf11aa, and Gdf8Gdf11MD mice as a measure of total ligand levels. Statistical analyses were performed by one-way ANOVA with Tukey’s correction for multiple comparisons. Amino acid residues of Gdf11 are represented in orange, and those of Gdf8, in blue. For Gdf11Gdf8aa and Gdf8Gdf11aa lines, one stripe denotes mono-allelic replacement, and two stripes denote bi-allelic replacement. For Gdf8Gdf11MD mice, half orange denotes mono-allelic replacement, and full orange denotes bi-allelic replacement.
GDF11 dampening in bi-allelic Gdf11Gdf8aa mutant embryos recapitulates developmental phenotype seen with Gdf11 loss of function
Germline deletion of Gdf11 results in perinatal lethality, and both homozygous (Gdf11−/−) and heterozygous (Gdf11+/−) disruption of Gdf11 cause developmental abnormalities of the skeleton (McPherron et al, 1999; Walker et al, 2016)—notably, the formation of extrathoracic vertebrae—and kidney agenesis in pups (Esquela & Lee, 2003; McPherron et al, 2009). We therefore assessed early-stage skeletal development in our chimeric embryos (Fig 3). Six sets of F4 mono-allelic mutant males were bred with mono-allelic mutant females within each Gdf11Gdf8aa, Gdf8Gdf11aa, and Gdf8Gdf11MD mouse line to generate F5 embryos that were harvested on embryonic day 18.5 (E18.5), eviscerated, and stained with Alizarin red and Alcian blue (Fig 3). Before evisceration, tissue samples from the posterior skin of prenatal pups were collected for genotyping by PCR validation and subcloning. Spleen and liver samples were also taken from each embryo for genotyping (data not shown) to confirm that maternal DNA from the fallopian tubes and gestational sacs would not obfuscate genotyping results by contaminating the collected embryonic tissue.
Skeletons from Gdf11Gdf8aa, Gdf8Gdf11aa, and Gdf8Gdf11MD embryos were harvested at E18.5 and stained with Alizarin red and Alcian blue. Gdf11+/+ (leftmost), Gdf118aa/8aa (second from left), Gdf8+/+ (middle), Gdf8+/11aa (second from right), and Gdf811MD/11MD (rightmost) vertebral columns, vertebrosternal ribs, full skeletons, and craniofacial bones are shown. (A) Gdf118aa/8aa skeletons exhibited abnormal axial vertebral patterning with T14 total axial vertebrae, compared with T13 total vertebrae in Gdf11+/+ skeletons. (B) All other mutant skeletons also had T13 total vertebrae. (C, D) Gdf118aa/8aa skeletons also had T8 vertebrosternal ribs, compared with T7 vertebrosternal ribs in Gdf11+/+ skeletons and (D) in other mutant skeletons. (E, F) Schematic of axial vertebral columns in (E) Gdf11Gdf8aa embryos and in (F) Gdf8Gdf11aa and Gdf8Gdf11MD embryos. Mono-allelic mutant mice Gdf11+/8aa, Gdf8+/11aa, and Gdf8+/11MD also had T13 total axial vertebrae and T7 vertebrosternal ribs. (G, H) In Gdf118aa/8aa and (H) Gdf811aa/11aa and Gdf811MD/11MD mutants, the full skeleton and craniofacial bones, in profile (middle row) and from above (bottom row), were indistinguishable from Gdf11+/+ and Gdf8+/+ mice, respectively. The limbs and digits were also similar to WT. Amino acid residues of Gdf11 are shown in orange, and those of Gdf8, in blue.
Upon imaging the skeletons, we discovered distinct skeletal transformations in the Gdf11Gdf8aa mouse line. Specifically, the bi-allelic Gdf118aa/8aa mutants exhibited one extrathoracic vertebra, with T14 axial vertebrae in total, compared with T13 vertebrae in WT mice (Fig 3A and E). We also observed an extra vertebrosternal rib in Gdf118aa/8aa mutants, resulting in a total of T8 ribs connected to the sternum, compared with T7 ribs in Gdf11+/+ and Gdf11+/8aa mice (Fig 3C and E). Phenotypically, this differential development directly compares to the axial skeletal phenotype observed in heterozygous Gdf11+/− mice, which also have 14 total thoracic vertebrae and 8 pairs of ribs fused to the sternum (McPherron et al, 1999). In contrast, homozygous Gdf11−/− knockout mice have 17–18 total thoracic vertebrae and 10–11 pairs of vertebrosternal ribs (McPherron et al, 1999). We further verified that the malformation documented occurred only in Gdf118aa/8aa mutants, and with 100% penetrance (Table 1). Moreover, given that circulating ligand levels in these chimeric mice remained unaltered, the observed differential skeletal phenotypes are specifically attributable to the dual amino acid changes made in the protein sequences and structures—not to alterations of circulating protein levels. In contrast, the Gdf8Gdf11aa mutant skeletons appeared indistinguishable from WT (Fig 3H), and none of the mono-allelic or bi-allelic mutants produced a measurable skeletal phenotype (Fig 3B, D, and F and Table 1). As stated previously, we generated the Gdf8Gdf11MD mutant line to investigate whether increasing GDF8 potency to the maximum level of GDF11, by replacing the entire mature domain of GDF8 with that of GDF11, would result in observable phenotypic outcomes. However, axial skeletal analysis of Gdf8Gdf11MD mice also revealed no measurable defects in the mutant skeletons, compared with WT (Fig 3B, D, and F and Table 1). Reported analyses of the related MstnGdf11/Gdf11 line generated by Lee and colleagues similarly found no abnormalities of axial skeletal patterning, though some decrement in bone density and alterations in trabeculae were noted, almost exclusively in males (Lee et al, 2022). These authors did not evaluate the impact of the dual amino acid substitutions reported here.
Skeletal analysis of Gdf11Gdf8aa, Gdf8Gdf11aa, and Gdf8Gdf11MD mouse embryos in C57BL/6J background.
We also investigated early-stage craniofacial bone development in Gdf11 and Gdf8 mutant mice (Fig 3G and H) because palatal defects have been reported in mice with Gdf11−/− deletion (McPherron et al, 1999; Cox et al, 2019) and in humans with Gdf11 loss-of-function alleles (Ravenscroft et al, 2021). In our inspection, the only skeletal differences were detected in the axial vertebral patterning and vertebrosternal rib count of Gdf118aa/8aa mutant mice (Fig 3A and C). No defects were noted in the limbs or cranium of Gdf11Gdf8aa, Gdf8Gdf11aa, or Gdf8Gdf11MD mutants, compared with Gdf11+/+ and Gdf8+/+ mice (Fig 3G and H). We also saw no palatal defects consistent with those previously reported for Gdf11-null mice (McPherron et al, 1999), nor did we observe a hole in the otic capsule of mutant mice, which has been reported in Gdf11 indel and gene-targeted mice (Goldstein et al, 2019) (Fig 3G and H). These results suggest that although full potency of GDF11 may not be necessary for craniofacial bone development, it is crucial for proper axial skeletal development. Therefore, dampening the potency of mature GDF11 with substitution of GDF8 residues is not compatible with maintaining fully normal developmental function, even with appropriate patterning of expression provided by the endogenous Gdf11 genomic locus. On the contrary, it appears that increasing the potency of GDF8, even to the maximum level of GDF11, does not elicit malformations detectable in early development. Together, these data underscore the notion that GDF11 and GDF8 are functionally distinct during development.
Gdf8Gdf11MD mutants exhibit decreased skeletal muscle mass, whereas the muscles of Gdf11Gdf8aa and Gdf8Gdf11aa mutants are not significantly altered
Next, we examined early-stage skeletal muscle and cardiac development in the Gdf11Gdf8aa, Gdf8Gdf11aa, and Gdf8Gdf11MD mouse lines. Prior studies indicate that genetic inactivation of Gdf8 dramatically increases muscle mass and alters fiber-type distribution across multiple animal species and in a dose-dependent manner (McPherron et al, 1997; McPherron & Lee, 1997), whereas boosting levels of GDF8 protein has been shown to drive muscle wasting (Zimmer et al, 2002; Artaza et al, 2007; Stolz et al, 2008). We therefore sought to determine whether enhancing the potency of mature GDF8 in the Gdf8Gdf11aa or Gdf8Gdf11MD mutants might reduce muscle mass compared with Gdf8+/+ mice, harvesting and analyzing the wet weight of the tibialis anterior (TA) (Fig 4C and D), quadriceps (Fig 4F and G), and triceps (Fig 4I and J) muscles postmortem across all three chimeric lines at 10–14 wk of age. Muscle weights were normalized to both total body weight (Fig 4A and B) and tibia bone length (Fig S3) in the Gdf11Gdf8aa, Gdf8Gdf11aa, and Gdf8Gdf11MD mutants (Fig 4E, H, and K) to assess possible changes in muscle mass. In addition, we harvested and weighed the kidneys from each mouse and normalized them to total body weight (Fig S4).
Tibialis anterior, quadriceps, and triceps muscles were collected postmortem from Gdf11Gdf8aa, Gdf8Gdf11aa, and Gdf8Gdf11MD mice at 10–14 wk of age. (A, B) Male and (B) female mice were weighed before dissection. (C, D, E) Weights of the tibialis anterior muscle in (C) male and (D) female mice were normalized to overall body weight (E) and showed a statistically significant decrease in both male and female Gdf811MD/11MD mutants, compared with Gdf8+/+. (F, G, H, I, J, K) Quadriceps weights in (F) male and (G) female mice normalized to body weight (H) showed a significant decrease in male Gdf811MD/11MD mice only, compared with Gdf8+/+, whereas the weight of the triceps in (I) male and (J) female mice normalized to body weight (K) showed a significant decrease in both male and female Gdf811MD/11MD mice, compared with Gdf8+/+. However, no significant differences were observed for normalized muscle weights in these mutant mice, compared with WT. Statistical analysis was performed by one-way ANOVA with Tukey’s correction for multiple comparisons. Amino acid residues of Gdf11 are represented in orange, and those of Gdf8, in blue. For Gdf11Gdf8aa and Gdf8Gdf11aa lines, one stripe and two stripes denote mono-allelic and bi-allelic replacement, respectively. For Gdf8Gdf11MD mice, half orange denotes mono-allelic replacement, and full orange denotes bi-allelic replacement. Also see Figs S3 and S4.
(A, B, C, D) Overall body weight (A), tibialis anterior muscle (B), quadriceps muscle (C), and triceps muscle (D) weights of Gdf11Gdf8aa, Gdf8Gdf11aa, and Gdf8Gdf11MD mice at 10–14 wk of age were normalized to tibia length. (B, D) In Gdf8Gdf11MD mice, there was a statistically significant decrease in the (B) tibialis anterior muscle normalized to tibia length between Gdf811MD/11MD (n = 18) versus Gdf8+/+ (n = 21, P < 0.01) mice, and in the (D) triceps muscle normalized to tibia length between Gdf811MD/11MD (n = 18) versus Gdf8+/11MD (n = 21, P < 0.05) mice. Statistical analysis was performed by one-way ANOVA with Tukey’s correction for multiple comparisons.
(A, B, C) Kidney weights (from two kidneys) of Gdf11Gdf8aa, Gdf8Gdf11aa, and Gdf8Gdf11MD (A) male and (B) female mice were taken at 10–14 wk of age and normalized to overall body weight (C). No mutant mouse of any chimeric line was represented with fewer than two kidneys. In Gdf11Gdf8aa, Gdf8Gdf11aa mice, no significant differences in kidney weights were found in mutants, compared with WT. (D, E, F) Similarly, the liver weight from (D) male and (E) female mice was normalized to body weight (F). (E) Significant decrease was found in the liver weight of (E) female Gdf811MD/11MD mutants (n = 9), compared with Gdf8+/+ mice (n = 10, P < 0.05). (D, F) However, the corresponding difference was not found in the liver weight of the males (D) or in the normalized weight of male and female Gdf8Gdf11MD mice (F). Statistical analysis was performed by one-way ANOVA with Tukey’s correction for multiple comparisons.
Across the chimeric amino acid Gdf8Gdf11aa and Gdf11Gdf8aa lines, we saw no significant differences in body weight (Fig 4A and B), TA (Fig 4C and D), quadriceps (Fig 4F and G), or triceps weights (Fig 4I and J). Skeletal muscle weights normalized to total body weight (Fig 4E, H, and K) and to tibia length (Fig S3B–D) were indistinguishable. Overall body weight normalized to tibia length was also insignificant in these two lines (Fig S3A). Across the Gdf8Gdf11MD mutants, however, the weights of these skeletal muscles were significantly decreased (Fig 4C, D, F, I, and J). Notably, TA (Fig 4C), quadriceps (Fig 4F), and triceps (Fig 4I) weights all were reduced to a statistically significant level in bi-allelic Gdf811MD/11MD mutant males (n = 9, P < 0.05), compared with sex-matched Gdf8+/+ mice (n = 10), with similarly significant decreases recorded in TA (Fig 4D) and triceps (Fig 4J) weights in Gdf811MD/11MD mutant females (n = 9, P < 0.05) as well. Although all three muscle groups in Gdf811MD/11MD mice normalized to overall body weight did not yield statistically significant differences (Fig 4E, H, and K), the raw weights of TA muscle normalized to tibia length (n = 18, P < 0.01) (Fig S3B) and triceps muscle normalized to tibia length (n = 18, P < 0.05) (Fig S3D) showed statistically significant decreases, compared with Gdf8+/+ mice (n = 21). Overall, we did not observe any distinct defects or malformations in muscle development or patterning at any point during development, and the chimeric mice appeared similar to WT into early adulthood. In all cases, separation by males and females resulted in shifts in the mean muscle mass between the sexes. Analysis of muscle mass in MstnGdf11/Gdf11 mice showed similar (∼10%) reductions in muscle mass in male, but not female, mice in which the mature domain of GDF8 was replaced by that of GDF11, with no differences in fiber composition (Lee et al, 2022). However, these studies did not include analysis of the dual amino acid substitutions reported here.
Although GDF11 loss of function frequently leads to kidney agenesis (Esquela & Lee, 2003; McPherron et al, 2009), we did not observe any chimeric mouse lacking a kidney. The combined weight of both kidneys (Fig S4A–C) also did not show any significant change in Gdf11Gdf8aa, Gdf8Gdf11aa, and Gdf8Gdf11MD mutants, compared with WT mice. Interestingly, we did record a statistically significant decrease in the liver weight of Gdf811MD/11MD females (n = 9), compared with Gdf8+/+ females (n = 10, P < 0.01) (Fig S4E), but a similarly significant decrease was not found in Gdf8Gdf11MD mutant males (Fig S4D), despite a modest, yet progressive, weight decline in the liver of Gdf8+/11MD mutants, followed by Gdf811MD/11MD mutants, compared with Gdf8+/+ mice (Fig S4D–F). This trend was not seen in Gdf8Gdf11aa mice (Fig S4D–F).
Taken together, these data indicate that the increased potency conferred to GDF8 by substitution of the GDF11 mature domain significantly impacted skeletal muscle mass in young Gdf8Gdf11MD mutant mice, in line with prior reports that endogenous GDF8 negatively regulates muscle development. The observed effects were distinct from those conferred to young Gdf8Gdf11aa mutant mice by substitution of only the Gdf11-like 89/91 residues—which did not yield any significant change in skeletal muscle mass. These data indicate a clear difference in potency between the mature domains of GDF11 and GDF8, and between the full GDF11 mature domain–replaced ligand and the double amino acid–substituted ligand generated here. In addition, though we did not observe signs of kidney agenesis in the Gdf8Gdf11MD mouse line, we recorded a trending decline in liver weight, most significant in Gdf8Gdf11MD mutant females. These data suggest a possible systemic effect on the liver—which does not produce GDF8—that could reflect a direct effect on liver hepatocytes or, more likely, be connected to the local GDF8 regulation of skeletal muscle mass. In such a scenario, skeletal muscle may be more sensitive to increased potency of GDF8, and as a result, a change in muscle mass indirectly affects liver development or homeostasis.
Baseline cardiac physiology and function remain unchanged in Gdf11Gdf8aa, Gdf8Gdf11aa, and Gdf8Gdf11MD mutants
Previous studies have shown that administration of exogenous recombinant GDF11 to aged mice reduces cardiac hypertrophy (Loffredo et al, 2013), and fetal cardiac GDF8 has also been implicated in early-stage heart development (Sharma et al, 1999). Since our in vitro experiments showed that GDF11 is a more potent signaling ligand than GDF8, there is potential for the genetically engineered mutants to induce changes in cardiac parameters. Therefore, we harvested and weighed the hearts of all chimeric mice at 10–14 wk of age (Fig 5A and B). In the Gdf11Gdf8aa line, we found a statistically significant increase in heart weight of Gdf118aa/8aa males (n = 13), compared with Gdf11+/+ males (n = 11, P < 0.05) and with Gdf11+/8aa males (n = 12, P < 0.05) (Fig 5A). However, this difference was not observed in Gdf11Gdf8aa female mice (Fig 5B), and the overall weight of Gdf11Gdf8aa mutant hearts normalized to body weight did not show a significant difference (Fig 5C). Analyses of Gdf8Gdf11aa and Gdf8Gdf11MD mouse lines showed no significant difference in the weight of mutant hearts, compared with that of Gdf8+/+ mice (Fig 5A and B). Further histopathological analysis of cardiac tissue also did not reveal significant differences in cardiomyocyte cross-sectional area (CSA) in any mutant lines compared with age- and sex-matched Gdf11+/+ and Gdf8+/+ mice (Fig 5D). We also investigated whether the mutations introduced to native Gdf11 and Gdf8 altered baseline cardiac physiology or function by performing blinded echocardiographic studies on all the chimeric mice at 10–14 wk of age (Fig 5E and F). In all three lines, echocardiographic imaging indicated equivalent baseline cardiac function. Fractional shortening (FS) (Fig 5E) and left ventricular heart dimensions (LVAW; LVPW; LVID) during systole and diastole (Fig 5F) were consistent across all genotypes, with no significant differences observed. Cardiac ejection fraction (ES) was also comparable across all three lines (Fig S5A–C).
(A, B, C) Hearts of Gdf11Gdf8aa, Gdf8Gdf11aa, and Gdf8Gdf11MD (A) male and (B) female mice were harvested and weighed at 10–14 wk of age and normalized to overall body weight (C). (D, E, F) Cardiomyocyte cross-sectional area (D), fractional shortening (E), and left ventricular heart dimensions (LVAW; LVPW; LVID) (F) during systole and diastole were measured across all three mouse lines. (A) Statistically significant increase in heart weight was found in (A) Gdf118aa/8aa males (n = 13), compared with Gdf11+/+ (n = 11, P < 0.05) and with Gdf11+/8aa males (n = 12, P < 0.05). (B, C) However, no similar difference was observed in (B) Gdf11Gdf8aa females, and the overall weight of Gdf11Gdf8aa mutant hearts normalized to body weight (C) did not yield a significant difference either. The cross-sectional area of cardiomyocytes was measured using FIJI software (scale bar = 100 µm). M-mode was used to measure left ventricular interventricular septal wall thickness (IVS/LVAW), left ventricular posterior wall thickness, and left ventricular internal diameter. Statistical analysis was performed by one-way ANOVA with Tukey’s correction for multiple comparisons. Amino acid residues of Gdf11 are represented in orange, and those of Gdf8, in blue. Also see Fig S5.
(A, B, C) Heart ejection fraction (ES) measurements were taken in (A) Gdf11Gdf8aa, (B) Gdf8Gdf11aa, and (C) Gdf8Gdf11MD mice and compared across all genotypes of each mouse line. Overall, no significant differences between any genotypes were found. Statistical analysis was performed by one-way ANOVA with Tukey’s correction for multiple comparisons.
These results indicate that the Gdf11Gdf8aa, Gdf8Gdf11aa, and Gdf8Gdf11MD mutant hearts have function and physiology similar to WT mice. Although no heart weight phenotype has been reported for Gdf11+/− mice, a difference was found in the heart weight of bi-allelic Gdf118aa/8aa mutant males (Fig 5A). However, this difference was not present in Gdf118aa/8aa female hearts (Fig 5B), and overall, we saw no significant differences across heart physiology and function of Gdf8Gdf11aa mutants. Furthermore, conferring the full potency of GDF11 to GDF8 did not significantly impact cardiac muscle CSA or ventricular dimensions in young Gdf8Gdf11MD mutant mice (Figs 5 and S5). If potency changes in either GDF11 or GDF8 indeed regulate cardiac muscle, then they do not appear to do so during development or young adulthood.
Gdf11Gdf8aa, Gdf8Gdf11aa, and Gdf8Gdf11MD mice exhibit normal regeneration of damaged muscle after cryoinjury
It has been reported that muscle repair after toxin-induced injury to skeletal myofibers is significantly enhanced in Mstn-null mice (McCroskery et al, 2005; Wagner et al, 2005), suggesting that endogenous GDF8 suppresses satellite cell proliferation. Debate continues as to whether GDF8 acts directly on muscle satellite cells and whether such action may account, at least in part, for the muscle hyperplasia or hypertrophy observed in Mstn-null mice (Thomas et al, 2000; Taylor et al, 2001; Amthor et al, 2009; Garikipati & Rodgers, 2012; Lee et al, 2012; George et al, 2013; Walker et al, 2016). To test what impact enhancing GDF8 potency has on muscle regenerative capacity, we subjected Gdf8Gdf11aa and Gdf8Gdf11MD mutants to cryoinjury (Fig 6). TA muscles of chimeric mice at 10–14 wk of age were cryoinjured on day 0, harvested at 7 and 14 d post-injury, and analyzed via H&E staining (Figs 6 and S6). Following cryoinjury, previously quiescent satellite cells in the basal lamina of myofibers are activated, giving rise to proliferating myoblasts (Dumont et al, 2015), which further differentiate and fuse together to form myotubes. Newly regenerated fibers can be distinguished by their central nuclei, with larger sized fibers at early time points in the repair process corresponding to more advanced muscle fiber regeneration (Mauro, 1961; Cosgrove et al, 2014; Sinha et al, 2014). We similarly challenged Gdf11Gdf8aa mutants in muscle regeneration assays, as conflicting reports have been published regarding the impact of changing levels of GDF11 on muscle repair (Sinha et al, 2014; Egerman et al, 2015).
The tibialis anterior muscles of male and female mice were harvested at 7 and 14 d post-injury and analyzed by H&E staining. (A, B, C) Cross-sectional area of the regenerating centrally nucleated fibers within the injured tibialis anterior muscle was measured using FIJI software (scale bar = 100 µm) and compared between (A) Gdf11Gdf8aa, (B) Gdf8Gdf11aa, and (C) Gdf8Gdf11MD mice at each time point. Overall, no significant differences were found in the frequency or size of regenerating myofibers in any genotype. However, at 14 d post-injury, bi-allelic Gdf8Gdf11MD/11MD mice exhibited modest increased cross-sectional area compared with Gdf8+/11MD, but not compared with Gdf8+/+ mice. Statistical analysis was performed by one-way ANOVA with Tukey’s correction for multiple comparisons. Also see Fig S6.
(A, B, C, D, E, F) Tibialis anterior muscles of male and female Gdf11Gdf8aa, Gdf8Gdf11aa, and Gdf8Gdf11MD mice were harvested at 7 d (A, B, C) and 14 d (D, E, F) post-injury and analyzed by H&E staining (respectively). The cross-sectional area of regenerating centrally nucleated fibers was measured using FIJI software (scale bar = 100 µm). No statistically significant difference in frequency or size of regenerating myofibers was found between genotypes in either male or female mice of any mutant line. Statistical analysis was performed by one-way ANOVA with Tukey’s correction for multiple comparisons.
Analysis of regenerating myofibers in cryodamaged muscles (7 or 14 d post-injury) showed no significant differences in average CSA or fiber size distribution in either Gdf8Gdf11aa or Gdf11Gdf8aa mice (Figs 6A–D and S6). These data indicate that neither increasing the potency of GDF8 nor dampening that of GDF11 is sufficient to alter the time course or outcome of muscle fiber regeneration in young adult mice. Taken together with the lack of muscle or body weight change in Gdf8Gdf11aa or Gdf11Gdf8aa mice, it appears that postnatal skeletal muscle is more tolerant of changes in GDF11 signaling potency than embryonic bone, which exhibited skeletal transformations when GDF11 potency was reduced. In the Gdf8Gdf11MD mutants, regenerating fiber size in Gdf8+/11MD and Gdf811MD/11MD mutants was indistinguishable from WT at both time points (Fig 6E and F). However, fiber size appeared slightly increased at 14 d post-injury in bi-allelic Gdf811MD/11MD mutants compared with mono-allelic Gdf8+/11MD mutants (Fig 6E). These data raise the possibility that the lack of mature GDF8, or the increased abundance of mature GDF11 protein, or the combination of these events might have a positive impact on the rate of muscle repair, consistent with prior reports of enhanced muscle regeneration after the loss of GDF8 (McCroskery et al, 2005; Wagner et al, 2005) or supplementation of GDF11 in vivo (Sinha et al, 2014).
Altogether, these data support the notion that the modest changes in ligand potency achieved through dual amino acid mutations confer a genetically linked phenotype in early-stage skeletal development within only Gdf11Gdf8aa mutant mice. In addition, the heightened ligand potency achieved through modifying the full mature domain of GDF8 in Gdf8Gdf11MD mutants confers a genetically linked phenotype impacting skeletal muscle mass in adult mice. Incredibly, in bi-allelic Gdf811MD/11MD mutants, circulating GDF11 concentration increased ∼50-fold more than normal, resulting in unexpectedly high levels of circulating GDF11 in vivo. Yet these chimeric mice were viable and survived normally into adulthood. The results indicate that gene regulation differences or differences in the prodomain regions may influence changes in ligand concentrations in vivo, and that ligand activity may not reflect alterations to ligand potency alone.
Discussion
Maintenance of GDF11 ligand potency and function is required for normal skeletal development
In this study, we investigated key phenotypes resulting from changes in the amino acid residues of GDF11 or GDF8, brought on by genetic modifications made to the Gdf11 or Gdf8 mature signaling domains in vivo. Results presented here indicate that changes in GDF11 and GDF8 potency elicit differential phenotypes (Table 2). Initial screening in early development revealed abnormal embryonic skeletal and vertebral transformations in the Gdf118aa/8aa mutant mice, similar to those of mice with heterozygous (Gdf11+/−) deletion of Gdf11 (McPherron et al, 1999). We followed these bi-allelic mutants of the Gdf11Gdf8aa line into adulthood to determine whether dampened GDF11 function in Gdf11Gdf8aa mutants results in additional phenotypic differences in organ growth and function postnatally. Assessments of overall body condition, skeletal and cardiac muscle mass, serum protein concentration, skeletal muscle repair, and baseline heart physiology and function were performed. Similar studies of the Gdf8Gdf11aa mutants were done in parallel to determine whether increased GDF8 potency in these mice resulted in differential phenotypes in adulthood. These efforts focused particularly on muscle size and function, given GDF8’s well-established negative regulation of muscle growth (McPherron et al, 1997; McPherron & Lee, 1997; Walker et al, 2016).
Developmental patterns and phenotypes of Gdf11Gdf8aa, Gdf8Gdf11aa, and Gdf8Gdf11MD mice.
Our findings show that reducing the potency of mature GDF11 toward that of mature GDF8 is insufficient to sustain normal developmental function. Specifically, exchanging two amino acid residues from the Gdf8 mature domain (G89 and E91) into the analogous location in the Gdf11 mature domain (D89 and Q91) decreased the potency of mature GDF11 to that of GDF8 (Walker et al, 2017) and resulted in skeletal transformations detectable during early development in Gdf11Gdf8aa mice. However, exchanging the same amino acids from the Gdf11 locus into the corresponding location in Gdf8 did not produce similar physiological defects, and Gdf8Gdf11aa mice carrying one or both chimeric alleles were indistinguishable from WT mice in all of the assays we performed.
Because both GDF8 and GDF11 have been shown to play critical roles in skeletal and cardiac muscle development and function, and to regulate other organ systems (McPherron et al, 1997; Lee & Lee, 2013; Walker et al, 2016), further experiments were conducted to investigate the phenotypes of Gdf11Gdf8aa and Gdf8Gdf11aa mutant mice in early adulthood. Despite the axial skeletal defects found in the Gdf11Gdf8aa line, reduction in GDF11 potency in these mutants did not impact postnatal skeletal muscle growth or regenerative activity, nor did it alter baseline cardiac physiology or function into adulthood. We observed no differences in body weight or skeletal muscle weight (TA, quadriceps, and triceps muscle; normalized to body weight and tibia length) postnatally in male or female mutants at 10–14 wk of age. Likewise, Gdf8Gdf11aa mice did not show significant differences in either sex at 10–14 wk of age. For both chimeric lines, we saw no differences in heart or kidney weight in either the Gdf11Gdf8aa or Gdf8Gdf11aa mutants. The only statistically significant finding occurred in the heart weight of Gdf118aa/8aa mutant males, compared with Gdf11+/+ and Gdf11+/8aa mice; however, this result was not observed in mutant females of the same line. The differences in the heart weight normalized to body weight also proved insignificant in the Gdf118aa/8aa mutants. Interestingly, Gdf8Gdf11aa mutants did not exhibit physiological changes to those found in Gdf11Gdf8aa mutants during embryonic development, nor did they produce measurable anatomic differences into adulthood.
Replacement of GDF8 mature domain with GDF11 decreases skeletal muscle mass and produces a 50-fold increase of circulating GDF11 levels in young adult mutants
Similar to Gdf8Gdf11aa mutants, chimeric Gdf8Gdf11MD mice also did not exhibit detectable transformations during embryonic skeletal development, either in mono-allelic Gdf8+/11MD or in bi-allelic Gdf811MD/11MD mutants. Furthermore, we detected neither cranial bone malformation nor cleft palates in the mutants. Therefore, raising the activity of GDF8 to the full potency of GDF11 did not produce measurable changes in osteogenesis. This result may reflect the different tissue-specific expression patterns of the Gdf11 and Gdf8 loci. We performed additional experiments to determine whether increased GDF8 function in Gdf8Gdf11MD mutants might result in phenotypic differences in muscle growth and function postnatally and into adulthood. In Gdf8Gdf11MD mice, we confirmed that the increase in the GDF8 potency produced by replacing the mature domain of Gdf8 with that of Gdf11 decreased skeletal muscle size/mass in several limb muscles, consistent with results from GDF8 supplementation studies (Zimmers et al, 2002; Stolz et al, 2008) and with the recently reported MstnGdf11/Gdf11 mice (Lee et al, 2022). Specifically, Gdf8Gdf11MD mutants exhibited decreased weight of the TA and triceps muscles in early postnatal life in both male and female mice. A significant decrease in quadriceps muscle mass was also noted in mutant males, with bi-allelic mutants exhibiting the greatest change.
Our Gdf8Gdf11MD mutants also presented a striking outcome in terms of circulating levels of GDF11, which were increased to ∼50-fold more than normal, similar to the 30–40-fold increase in circulating GDF11 protein reported for MstnGdf11/Gdf11 mice (Lee et al, 2022). Although some prior studies suggested that elevation of GDF11—even at moderate levels—could result in detrimental consequences in mice, including severe cachexia and premature death (Egerman et al, 2015; Glass, 2016; Schafer et al, 2016; Harper et al, 2018), these results demonstrate that substantial elevation of GDF11 is well tolerated, with only minor effects on skeletal muscle mass. Gdf8Gdf11MD chimeras were viable into adulthood and showed no signs of premature aging or other negative impacts on health or survival. The profound differences in circulating ligand levels in these gene-modified mice also suggest that gene regulatory differences encoded within the Gdf8 and Gdf11 genomic loci play an important role in determining systemic ligand abundance. Whether the substantial increase in GDF11 levels seen in Gdf8Gdf11MD mutants shifts the homeostasis of known antagonists or alters interactions with their respective N-terminal prodomains remains to be investigated. Finally, this study provides in vivo evidence supporting the molecular explanation for potency differences between GDF11 and GDF8 derived previously from structural and biochemical studies (Walker et al, 2017). In particular, while amino acid differences at residues 89 and 91 are responsible for much of the difference in potency observed between GDF11 and GDF8 (Walker et al, 2017), the other nine amino acids that distinguish the GDF11 and GDF8 mature domains clearly contribute as well.
We found no significant differences in heart weight or baseline cardiac physiology and function in Gdf8Gdf11MD mutants. However, whether similar results would be obtained in aged mutants or under conditions of transverse aortic constriction remains a question for future studies. Although supraphysiological elevation of GDF11 has been reported in several studies to impede recovery from muscle injury in young mice (Egerman et al, 2015; Hammers et al, 2017; Jones et al, 2018), and loss of GDF11 signaling in older animals has been suggested to underlie poorer regenerative outcomes with aging (Sinha et al, 2014), we also saw no difference in either the kinetics or ultimate outcome of muscle repair after injury in Gdf8Gdf11aa or Gdf11Gdf8aa mutants. With these data, we show that robust muscle repair activity is preserved in young mice both when GDF11 signaling is dampened and when GDF11 levels are raised, suggesting that muscle satellite cells and myofibers may be buffered to some extent against changes in GDF11 activity in young adulthood. The only other statistically significant finding occurred in the liver weight of Gdf811MD/11MD mutant females, compared with Gdf8+/+ mice and Gdf8+/11MD mutants, but not in that of Gdf811MD/11MD mutant males. At this time, we cannot differentiate whether increased GDF8 potency directly affected liver hepatocytes, leading to the observed change in organ size, or whether indirect effects, potentially linked to changes in skeletal muscle or other tissues, underlie this result. Gene expression analysis in the liver to examine whether regulation of metabolic genes is altered will be useful to verify the possible impact of the change in ligand potency.
Our chimeric mice, genetically modified to change unique amino acid residues between GDF11 and GDF8, demonstrate that sequence-determined structural differences in these ligands are critically important, and not simply accounted for by gene regulation differences alone. We have discovered that two specific amino acids in the fingertip region of the GDF11 mature domain are required for proper axial skeletal patterning during early-stage development, but do not appear to be crucial for regulating heart or skeletal muscle. In addition, substituting the full mature domain sequence of GDF11 into the Gdf8 locus, in place of mature GDF8, created a GDF8-null mouse with decreased skeletal muscle mass. Circulating GDF11 concentrations in this mutant were significantly higher than WT, showing that very high GDF11 blood levels can be tolerated and can overcome phenotypes typically associated with loss of GDF8 function. Our findings show that changing the ligand potency of GDF11 and GDF8 or altering their bioavailability within the mammalian system causes distinct measurable physiological effects, and that maintenance of functional GDF11 and GDF8 is necessary for proper development and adult tissue maintenance.
Materials and Methods
Mouse caretaking
Mouse handling and experimentation followed guidelines set forth by the Institutional Animal Care and Use Committee at Harvard University/Faculty of Arts and Sciences. Regular animal housing and care were carried out by the staff in Biological Laboratories in accordance with relevant Institutional Animal Care and Use Committee regulations and guidelines. Housing included density of two to four mice per cage, along with Enviro-Dri bedding, cotton nestlet, one red hut, automatic waterspout dispensing reverse osmosis/deionized water, and regular chow diet (Prolab IsoPro RMG3000 5P75/76).
Generation of mutant mouse lines
The CRISPR/Cas9 system was used to generate chimeric mice carrying amino acid mutations from the GDF8 and GDF11 interchanged in the mature domain. Two rounds of microinjections into C57BL/6J zygotes were performed for each line, delivering (1) purified S. pyogenes Cas9 mRNA, (2) in vitro–transcribed synthetic guide RNAs (sgRNAs), which targeted the native loci near the desired integration sites, and (3) the generated Gdf11 and Gdf8 ssDNA or Gdf11 dsDNA donor template. Zygote injections were performed by the Genome Modification Facility at Harvard University. Super-ovulated C57BL/6J female mice were mated to C57BL/6J males, and the fertilized zygotes were subsequently harvested from the oviducts. Zygotic pronuclei were microinjected with (1) purified Streptococcus pyogenes Cas9 mRNA (100 ng/µl; System Bioscience), (2) in vitro–transcribed synthetic guide RNAs, which targeted near the integration sites of the native loci, and (3) the generated Gdf11 and Gdf8 ssDNA donor construct. Microinjected zygotes were implanted into the oviducts of C57BL/6J surrogate females at 12 h post-coitum. To identify mosaic offspring that contained the desired chimeric Gdf11 and Gdf8 double amino acid–substituted sequences, progeny from the microinjected surrogate females were genotyped by Sanger sequencing and by PCR validation and subcloning. In the Gdf11 amino acid chimera, we produced 47 total pups, with 14 live pups and 33 dead pups. Of the 14 live pups, 4 exhibited a mosaic genotype, and 3 were ultimately selected as germline founders. In the Gdf8 amino acid chimera, we obtained 92 viable pups, with 0 dead pups. Of the 92 live pups, 49 showed a mosaic genotype upon screening for the targeted allele. Ultimately, four confirmed positive founders were used as breeders to establish the colony. In the Gdf8 full mature domain chimera, we obtained 80 viable pups, with two dead pups. Of the 80 live pups, 35 showed a mosaic genotype upon screening for the targeted allele. Ultimately, five confirmed positive founders were used as breeders to establish the colony. We bred these chosen mice with WT C57BL/6J mice and genotyped the resultant pups to confirm that our desired mutation was incorporated in the germline. We backcrossed the five-generation knock-in alleles in a C57BL/6J genetic background before characterization experiments. Male and female mice were selected based on gender and randomized before treatment for all proposed animal studies.
Genotyping
Initial mouse colony breeder genotypes were verified by sequencing. Subsequent progeny tissues were collected from the tail or the ear, and mixed in 1.5-ml Eppendorf tubes containing 300 µl of lysis solution (50 mM KCl, 10 mM Tris–HCl, pH 8.3, 2.5 mM MgCl2, 0.1 mg/ml gelatin, 0.45% NP-40, and 0.45% Tween-20 in ddH2O) with 10 µg/ml Proteinase K and incubated at 50°C for 10–12 h. Next, samples were placed in heating blocks at 98°C for 5 min to inactivate the Proteinase K. DNA was extracted using a standard phenol:chloroform/ethanol precipitation protocol, and genotyping was performed by PCR using Phusion High-Fidelity PCR Master Mix with HF Buffer. Primers used are as follows: for Gdf11Gdf8aa, FW: CCTGACCCTCAGCATCCTTTCA, RV: GGTCCTTACTTTGCCCCATCCT; and for Gdf8Gdf11aa and Gdf8Gdf11MD, FW: TGTGGTTGGTTTGTTTGTTTGT, RV: GCCTGTGGTGCTTGAATTCA. PCR products were digested with restriction enzyme AseI (R0526; New England Biolabs), with NEBuffer 3.1, at 37°C for 20–30 min, and analyzed on 1% agarose gel. Further genotyping was performed via Sanger sequencing to verify the nucleotide changes and the presence of the AseI site.
TLA sequencing
Bone marrow from Gdf11Gdf8aa, Gdf8Gdf11aa, and Gdf8Gdf11MD F5 mutant mice was harvested, homogenized, and subjected to ACK lysis. Harvested bone marrow (five vials, each containing 1 × 107 cells) was frozen and delivered to Cergentis B.V. for TLA sequencing analysis (de Vree et al, 2014). TLA sequencing used a locus-specific sequence for the targeted amplification and complete sequencing of Gdf11 and Gdf8 loci. The genomic DNA was cross-linked, digested, and re-ligated, before it was purified, and circular TLA fragments were then amplified with two independent sets of inverse primers, corresponding to the Gdf11 or Gdf8 locus-specific transgene, to identify the location of each targeting event across the whole genome. Primer sets used are as follows:
Gdf11Gdf8aa: Upstream, Fw: ACATTTGCTCCCATTACTGT, Rv: AGCAATAAGAACAAGGGAGC, Downstream, Fw: CAAGAGTCTTAAGAGGATGGG, Rv: GGGTAGTTTAGTAGCTCTCATAG.
Gdf8Gdf11aa: Upstream, Fw: GAATAGATGCAATGGTTGGC, Rv: AGAGTGTAGTGTTTAAGTAGCA, Downstream, Fw: CACAATTTGTTTATGCGGTTT, Rv: TCTCACTTCCTTGCCTAGAT.
Gdf8Gdf11MD: chr10 detection, Fw: CAAGTGGGTGTGTGGATAC, Rv: CTACCAAGATGTCCCCAATC, chr1 detection, Fw: GTAACTGCTCAGATTCCCAA, Rv: AGCTATTCCAAGGAACAACA. 5′ integration site: chr1:53,066,297 (tail) fused to Insert: 1, head (the same as chr10:128,885,435, tail) with four inserted bases ATCCCTTTTTAGAAGTCAAGGTGACAGACACACCCAAGAGGTCCCGAAGAAACCTAGGCCTGGACTGGATGAACACTCGAGTGAGTCCCGCTGCTGCCGATATCCTCTCACAGTGGACTTTGAGGCTTTTGGCTGGACTGGATCATC; 3′ integration site: Insert: 327, tail (the same as chr10:128,885,108, head) fused to chr1:53,066,629 (head) AACATGCTCTACTTCAATGACAAGCAGCAGATTATCTACGGCAAGATCCCTGGCATGGTGGTGGATCGATGTGGCTGCTCCTGAGCTTTGCATTAGGTTAGAAATTTTCCAAGTCATGGAAGGTCTTC. After amplification of the targeted locus, PCR amplicons of the complete Gdf11 and Gdf8 region of interest were purified and prepped for Illumina sequencing. Analysis was performed by aligning mutant sequences to the mouse mm10 reference genome sequence. In Gdf11Gdf8aa mice, two mutated nucleotides (G→C and T→C) were confirmed on chr10 in the mature domain of Gdf11, resulting in alteration of only the two targeted amino acids: Gdf11 D89 to G89 (Asp→Gly) and Q91 to E91 (Gln→Glu). In Gdf8Gdf11aa mice, two mutated nucleotides (G→A and G→C) were confirmed on chr1 in the mature domain of Gdf8, resulting in amino acid changes in Gdf8 G89 to D89 (Gly→Asp) and E91 to Q91 (Glu→Gln). In bi-allelic Gdf118aa/8aa and Gdf811aa/11aa samples, the mutated nucleotides were confirmed at 100% frequency, indicating that the mutations occurred on both alleles, whereas in mono-allelic Gdf11+/8aa and Gdf8+/11aa samples, the mutations were confirmed at ∼50% frequency, indicating occurrence of the desired mutations on only one allele. In the Gdf8Gdf11MD mice, correct integration of exon 3 of Gdf11 was confirmed on chr1 in place of the native exon 3 of Gdf8, indicating successful replacement of the GDF8 mature domain with that of GDF11. In bi-allelic Gdf811MD/11MD samples, no WT reads were present at the integration site and in the deleted region, confirming bi-allelic replacement of native GDF8, whereas in mono-allelic Gdf8+/11MD samples, WT reads were detected at the integration site and in the deleted region on one allele, confirming mono-allelic replacement of native GDF8.
Serum mass spectrometry of GDF11 and GDF8
For serum collection, after euthanasia, blood was collected via orbital bleeding into Microtainer tubes with a serum separator (BD) and incubated for 30 min at room temperature before centrifugation at 2,000g for 10 min at room temperature. The upper layer of serum was then transferred to a new tube and stored at −80°C before mass spectrometry analysis. Serum from chimeric mice (minimum 100 µl) was submitted to the Brigham and Women’s Hospital Brigham Research Assay Core (BRAC) for quantitative liquid chromatography–tandem mass spectrometry detection of GDF11 and GDF8 protein concentrations. The mouse serum was denatured and alkylated, followed by pH-based fractionation, using cation ion exchange SPE. After desalting and concentrating, the peptide mix was separated via liquid chromatography, followed by mass spectrometry analysis in a positive electrospray ionization mode. GDF11 and GDF8 concentrations were determined using unique proteotypic peptides from GDF11 and GDF8 as surrogate peptides, coupled with heavy-labeled unique peptides as internal standards. Included in the analysis were GDF8 concentrations and the mean GDF11 concentrations.
Skeletal preparation
Embryonic day 18.5 (E18.5) harvests (Lewandowski et al, 2019) were performed for skeletal and vertebral analyses of Gdf11Gdf8aa and Gdf8Gdf11aa mutants and compared with WT mice. For timed breeding, a vaginal plug observed in the female indicated embryonic day 0.5 (E0.5). Mouse embryos were harvested at E18.5, skinned, and eviscerated, and underwent washes of 100% ethanol and 100% acetone for 24 h at room temperature. The skeletons were then stained using a 0.3% Alcian blue (dyes bone) and 0.1% Alizarin red (dyes cartilage) solution at 37°C while oscillating for 72 h. Next, they were transferred to 1% KOH for 24 h on a rocker at room temperature, followed by a series of glycerol/KOH washes at (1) 20% glycerol/1% KOH, (2) 50% glycerol/1% KOH, and (3) 80% glycerol/1% KOH for 24 h at room temperature. The stained preparations were placed in 80% glycerol/1× PBS and imaged using a Nikon D750 camera attached to a Nikon SMZ1500 stereo microscope with an HR Plan APO 1× objective lens. Images were captured in NEF (Nikon Electronic Format) and processed and adjusted in the Adobe Camera Raw platform.
Tissue collection and analysis
Mice were ear-tagged and their genotype blinded at 10–14 wk of age. Adult mice were euthanized with CO2, and overall body weights were taken postmortem. For tissue collection, the heart was dissected, washed in PBS, and dried on paper towels before weighing. The TA muscle, quadriceps muscle, and triceps muscle were harvested and weighed. The tibia bone was excised, cleaned, and measured using electronic calipers. Statistical analysis was performed using Prism 8.4.2 for macOS. The muscle weights were divided by sex and genotype within each mouse line, and the results were represented as the mean within each group by genotype and sex. The combined muscle weights were normalized to body weight and tibia length. Groups more than two were compared by one-way ANOVA with Tukey’s correction for multiple comparisons.
Echocardiographic studies
Gdf11Gdf8aa mice (Gdf11+/+, n = 12; Gdf11+/8aa, n = 16; and Gdf118aa/8aa, n = 15) and Gdf8Gdf11aa mice (Gdf8+/+, n = 9; Gdf8+/11aa, n = 10; and Gdf811aa/11aa, n = 10), age 10–14 wk, were sedated with 0.1–0.5% inhaled isoflurane for echocardiography (Loffredo et al, 2013). Mice were placed on a heating pad, and echocardiograms were obtained at mid-papillary level with the Vevo 3100 (VisualSonics). The heart rate of every mouse studied was monitored and maintained at >400 bpm during echocardiographic procedure, so as to (1) avoid artificial myocardial depression brought on by exposure to isoflurane during imaging and (2) ensure consistent measurements across all study groups. Parasternal long-axis views, short-axis views, and two-dimensional M-mode were used to measure left ventricular interventricular septal wall thickness (IVS/LVAW), left ventricular posterior wall thickness (LVPW), and left ventricular internal diameter (LVID) during both systole and diastole in both mouse lines. Fractional shortening (FS%) was calculated with the VisualSonics software package. Statistical analysis was performed using Prism 8.4.2 for macOS. For echocardiographic and morphometric analyses, CSA, capillary density, and volume results were presented as the mean for each group by genotype and sex. At least three measurements were averaged and used for every data point from each mouse. Groups more than two were compared by one-way ANOVA with Tukey’s correction for multiple comparisons. All analyses were performed under blinded conditions.
Muscle cryoinjury
The muscle cryoinjury procedure (Oh et al, 2016) was chosen because of its ability to generate a reproducible injury area with a discrete border between injured and uninjured muscles. This border remains distinct during regeneration of injured muscle. Mice were anesthetized using isoflurane, and the skin over the left TA muscle was shaved and disinfected using Betadine, followed by wiping with 70% ethanol. The TA was then exposed by a small 3-mm incision. A metal probe with flat round bottom cooled down in dry ice was applied directly to exposed TA muscle for 5 s. The skin incision was then closed with synthetic absorbable suture (5-0 coated Vicryl) immediately after the injury. Buprenorphine (0.05–0.1 mg/kg, s.c.) was administered immediately after recovery from surgery, and subsequently every 8–12 h, for at least 48 h after surgery. Injured muscles were recovered for 7 or 14 d post-injury. The cryoinjury model employed has been widely used to assess muscle repair after damage and offers a number of advantages. In particular, cryoinjury can be performed such that the size of the lesion and severity of damage are highly similar across experimental animals, with preservation of regenerative muscle satellite cells that nucleate repair in the surrounding uninjured area (Gayraud-Morel, 2009; Dumont et al, 2015; Hardy et al, 2016).
Histology and CSA quantification
For cryosections, harvested TA muscles were flash-frozen in 2-methylbutane for 30 s followed by liquid nitrogen for 30 s. Samples were stored at −80°C before sectioning. Mouse hearts were rinsed with 1× PBS, embedded in compound at optimal cutting temperature, and frozen. TA samples and cardiomyocyte samples were sectioned at 10 µm. Hematoxylin and eosin (H&E) staining was used to visualize cardiomyocyte cross sections and regenerating myofibers in injured muscles. Sections were stained with hematoxylin for 3 min and washed with tap water for 2 min, followed by back-to-back washes in acid alcohol. Scott’s Bluing Reagent was used for 3 min for nuclear staining, followed by another tap water wash, and finally 2 min for eosin staining. Afterward, sections were dehydrated in ethanol and xylene, and subsequently mounted in Permount Mounting Medium (Cat. #17986-01; Electron Microscopy Sciences) and cured for 24 h before imaging. In regenerating myofibers after cryoinjury, a cross section representing the mid-belly of the TA where there was a clear representation of the injury was chosen. To quantify muscle fiber size, a series of images were taken spanning the entire regenerating area in the cross sections using a dual-head Olympus B×4 microscope with cellSens Standard software. Centrally nucleated myofibers were measured in each image using FIJI software (scale bar = 100 µm), resulting in ∼300–1,500 fibers collectively for each animal. Comparative analyses of more than two groups were performed by one-way ANOVA with Tukey’s correction for multiple comparisons. All analyses were performed under conditions in which the analyst was blinded to sample identity.
Statistical analyses
All data are presented as mean + SEM. For all data, n equals the number of biological replicates of animals used per experiment. The number of animals used for each group was determined based on total empirical data and anticipated completeness of datasets and was sufficient to detect differences in experimental outcomes, if present. Statistical analysis was performed using Prism 8.4.2 for macOS. Comparisons between two different experimental groups were assessed for statistical significance using a t test, and comparisons between more than two groups were performed by one-way ANOVA with Tukey’s correction for multiple comparisons. Statistical significance was accepted at P < 0.05. All experiments and analyses were performed under blinded conditions.
Acknowledgements
We thank Lin Wu and Laurie Chen from Harvard’s Genome Modification Facility for assistance in generating the Gdf11 and Gdf8 chimeric mice, Liming Peng and Shalender Bhasin for assistance with liquid chromatography–tandem mass spectrometry, Cergentis B.V. for TLA sequencing analysis, and Thompson and Wagers Lab members for helpful discussions. This work was supported by NIH grants R56AG062468 and R01AG047131 (to RT Lee), 4T32HL007208-39 (to RG Walker), R35GM134923 and R01AG072087 (to TB Thompson), and R01AG048917, R01AG057428, and an award from the Glenn Foundation (to AJ Wagers). The content is the responsibility of the authors and does not represent official views of Harvard University and its affiliated healthcare centers or the National Institutes of Health.
Author Contributions
J Lian: conceptualization, resources, data curation, software, formal analysis, supervision, validation, investigation, visualization, methodology, project administration, and writing—original draft, review, and editing.
RG Walker: conceptualization, resources, formal analysis, supervision, validation, investigation, visualization, methodology, project administration, and writing—review and editing.
A D’Amico: data curation, formal analysis, validation, investigation, and methodology.
A Vujic: data curation, software, formal analysis, validation, investigation, and methodology.
MJ Mills: data curation, validation, and investigation.
KA Messemer: resources, data curation, formal analysis, validation, and investigation.
KR Mendello: data curation, validation, and investigation.
JM Goldstein: validation and methodology.
KA Leacock: formal analysis.
S Epp: investigation.
EV Stimpfl: investigation.
TB Thompson: funding acquisition, methodology, and writing—review and editing.
AJ Wagers: funding acquisition, methodology, project administration, and writing—review and editing.
RT Lee: funding acquisition, methodology, project administration, and writing—review and editing.
Conflict of Interest Statement
RT Lee, RG Walker, and AJ Wagers are cofounders and members of the scientific advisory board for, and hold private equity in Elevian, Inc., a company that aims to develop medicines to restore regenerative capacity. Elevian also provides sponsored research support to the Lee Lab and Wagers Lab. Drs. RT Lee, RG Walker, AJ Wagers, and TB Thompson have filed patents related to GDF11 and GDF8 through their institutions.
Footnotes
- Received August 9, 2022.
- Revision received December 11, 2022.
- Accepted December 12, 2022.
- © 2023 Lian et al.
https://creativecommons.org/licenses/by/4.0/This article is available under a Creative Commons License (Attribution 4.0 International, as described at https://creativecommons.org/licenses/by/4.0/).
References
