CDK5RAP3, a UFL1 substrate adaptor, is crucial for liver development

The liver functions as a central regulator of synthesis, metabolism and detoxification (Gordillo et al., 2015). Liver development requires a complex regulation network, which has been studied extensively but remains largely undefined, especially at the post-translational level (Si-Tayeb et al., 2010). Conjugation and de-conjugation of ubiquitin and ubiquitin-like proteins (UBLs) to cellular proteins are among the most prevalent post-translational modifications (Hochstrasser, 2009; Schulman and Harper, 2009). This process plays a pivotal role in metazoan development by regulating cell cycle progression, apoptosis and cell fate determination (Bowerman and Kurz, 2006; Werner et al., 2015, 2017). The UFM1 conjugation system is a novel ubiquitin-like modification system that consists of UFM1, UBA5 (E1 activating enzyme), UFC1 (E2 conjugating enzyme), UFL1 (E3 ligase), UFSP1 and UFSP2 (UFM1-specific proteases) (Daniel and Liebau, 2014; Gavin et al., 2014; Kang et al., 2007; Komatsu et al., 2004; Padala et al., 2017; Tatsumi et al., 2010). Emerging evidence implicates the UFM1 system in the maintenance of endoplasmic reticulum (ER) homeostasis (Hu et al., 2014; Lemaire et al., 2011; Liu et al., 2017; Miller et al., 2017; Zhang et al., 2012). Recent genetic studies provide solid evidence for the essential role of the UFM1 system in animal development and hematopoiesis (Cai et al., 2015; Nahorski et al., 2018; Su et al., 2018; Tatsumi et al., 2011; Zhang et al., 2015). Nonetheless, the role of this system in liver development remains largely unclear. Clinically, the UFM1 system has been implicated in a range of human diseases. UFM1 modification is crucial for breast cancer development (Yoo et al., 2014). Disruption of the UFM1 cascade can result in early-onset encephalopathy (Colin et al., 2016; Mignon-Ravix et al., 2018; Muona et al., 2016). UBA5 mutation causes autosomal recessive cerebellar ataxia (Duan et al., 2016).

Recently, several cell-based studies have indicated the interactions between CDK5RAP3 and a few components of the UFM1 conjugation system, such as UFM1, UFL1 and UFBP1 (Kwon et al., 2010; Lemaire et al., 2011; Wu et al., 2010; Yoo et al., 2014). This suggests that CDK5RAP3 is a potential UFM1 substrate and in-depth investigation is necessary. CDK5RAP3, also known as LZAP and C53, was first identified as a binding protein of cyclin-dependent kinase 5 (CDK5) activator (Ching et al., 2000). Subsequent research showed that CDK5RAP3 interacts with a range of proteins, including RelA (Wang et al., 2007), ARF (Wang et al., 2006), Chk1 (Jiang et al., 2005), Chk2 (Jiang et al., 2005), PAK4 (Mak et al., 2011), Wip1 (Wamsley et al., 2017), γ-tubulin (Hořejší et al., 2012) and p38 MAPK (An et al., 2011). By interacting with these proteins, CDK5RAP3 exerts the role in regulating the cell cycle, cell survival, cell adherence/invasion, tumorigenesis and metastasis (Jiang et al., 2005; Jiang et al., 2009; Liu et al., 2011; Wang et al., 2007, 2006; Zhao et al., 2011). Moreover, a genetic study in zebrafish indicated that cdk5rap3 is essential for epiboly and gastrulation of early embryo (Liu et al., 2011). However, the role of Cdk5rap3 in mammalian development has not yet been explored.

In this study, we have generated complete and tissue-specific knockout mouse models, and reveal the crucial roles of CDK5RAP3 in liver development and hepatic functions. The Cdk5rap3-null mice died prenatally and showed severely hypoplastic livers. The hepatocyte-specific Cdk5rap3 knockout mice displayed liver hypoplasia and died after weaning. We identified the interaction between CDK5RAP3 and UFL1 in liver cells. Remarkably, the absence of CDK5RAP3 altered the profile of UFM1 substrates, suggesting CDK5RAP3 as a novel substrate adaptor for ufmylation. We further showed that Cdk5rap3 deficiency triggered ER stress and activated PERK and the IRE1α signaling pathway. Our results suggest that CDK5RAP3 serves as an important player in both the ufmylation system and mammalian liver development.

Cdk5rap3 gene targeting was carried out by the EUCOMM (European Conditional Mouse Mutagenesis) team (Skarnes et al., 2011). The lacZ-tagged conditional allele, Cdk5rap3^tm1a, is shown in Fig. 1A. Mutated Embryonic stem (ES) cell clones, as verified by PCR and Southern blot (Fig. S1B and Fig. 1B), were used to generate Cdk5rap3^tm1a/+ mice. Cdk5rap3^tm1a/+ mice were then crossed with EIIA-Cre mice to obtain Cdk5rap3^tm1b/+ mice, in which the floxed exons (#6-11) had been deleted ubiquitously (Fig. 1A). Cdk5rap3^tm1b/+ mice were phenotypically normal and were intercrossed to obtain homozygotes. Genotyping of the progeny failed to detect any Cdk5rap3^tm1b/tm1b (hereafter referred to as KO) pups, indicating that Cdk5rap3 knockout mice die in utero. To establish the timing of embryonic lethality, embryos were harvested at different developmental stages. Genotyping suggested that KO mice died from E16.5 onwards (Table 1). The expression of CDK5RAP3 was not detected in KO embryos via western blot (Fig. 1C).

To explore the reason for this lethality, embryos at different developmental stages were observed. At 14.5 onwards, KO embryos could be easily recognized by their significant growth retardation (Fig. 1E and Fig. S1C). The most striking defect was observed in the developing liver, occupying only a small fraction of the abdominal area (Fig. 1E; Fig. S1C). At E16.5, the mean weight of KO livers (LW) was 64.9% that of wild type, whereas mean total body weight (BW) of KO was 84.4% that of wild type (Fig. 1D). The LW/BW ratio is significantly smaller in KO embryos (Fig. 1D).

Profound anemia was also observed in Cdk5rap3 KO embryos (Fig. 1E; Fig. S1C). Histologic sections of E16.5 mutant embryos showed many nucleated erythrocytes in the peripheral blood, which were rarely seen in control embryos (Fig. S2A). This result revealed a defect in definitive erythropoiesis in KO embryos. Flow cytometry of E14.5 KO fetal livers showed decreased hematopoietic stem cells (HSCs) compared with WT livers (Fig. S2B-D), indicating impaired hematopoiesis. Analysis of mouse erythroid progenitors using the CD71/TER119 flow-cytometric assay indicated impaired erythropoiesis (data not shown). In addition, hemorrhage was also observed in KO embryos at E14.5 onward (data not shown).

To investigate Cdk5rap3 function in liver development, we analyzed the expression of CDK5RAP3 in E16.5 fetal livers by immunohistochemical staining and found that CDK5RAP3 was predominantly expressed in hepatocytes (Fig. 2A,A′). The colocalization of CDK5RAP3 and a hepatocyte marker, DLK-1, by immunofluorescence further indicated the hepatocyte-specific expression of CDK5RAP3 in fetal liver (Fig. 2B-D′). This specific expression suggested a possibly important role of CDK5RAP3 in hepatocytes. We observed that the expression of CDK5RAP3 was mostly restricted in the cytoplasm of hepatocytes (Fig. 2A-D′). Moreover, in adult mice, the expression of Cdk5rap3 is still broad but relatively higher in liver at both RNA and protein levels (Fig. 2E,F).

To investigate the origin of the liver hypoplasia phenotype, we examined the proliferation capacity of hepatocytes by examining expression of proliferating cell nuclear antigen (PCNA) and incorporation of bromodeoxyuridine (BrdU). Both PCNA and BrdU labeling were decreased in KO fetal livers compared with controls (Fig. 3A-B′; Fig. S3A-B′). We examined the level of phosphohistone H3 (pHH3) to determine cell cycle progression. Interestingly, more pHH3-positive cells were found in KO liver tissues, indicating G₂/M arrest (Fig. 3C,C′; Fig. S3C,C′). The influence of CDK5RAP3 on cell cycle was further investigated by western blot. Cyclin B1 expression was decreased in KO fetal livers, whereas Cyclin D1 expression remained unchanged after Cdk5rap3 deletion (Fig. S3I).

We asked whether cell death was also responsible for liver hypoplasia. At E14.5, expression of activated caspase 3, an apoptosis marker, was rarely detected in both wild-type and KO livers (Fig. S3D,D′). At E16.5, many apoptotic cells were observed in KO livers, whereas few apoptotic cells were found in wild-type livers (Fig. S3E,E′). To define the cell type of apoptotic cells, we carried out immunofluoresence with activated caspase 3 and DLK-1 on E16.5 wild-type and KO livers. Scarce hepatocytes displayed activated caspase 3 staining (Fig. S3F-H′). This result indicated that Cdk5rap3 is not crucial for hepatocyte survival. Instead, other cell types in KO livers are undergoing apoptosis.

At E16.5, most hepatocytes were polarized and differentiated, with clear cytoplasm in wild-type fetal livers, whereas many undifferentiated hepatocytes remained in the KO livers, which had high nuclear to cytoplasmic ratio and lacked polarity (Fig. 3D,D′). To identify the maturation status of the hepatocytes, the livers were examined for glycogen accumulation. Extensive glycogen accumulation indicated by periodic acid-Schiff (PAS) staining was evident in E16.5 wild-type livers but not in KO livers (Fig. 3E,E′). Next, we examined wild-type and KO livers for primitive bile ducts, which are positive for cytokeratin 19 (CK19) after the E16.5 stage. Several CK19-positive primitive ductular structures were visible in E16.5 wild-type livers, but KO livers displayed no detectable expression of CK19 (Fig. 3F,F′). The expression of HNF4ɑ, a master transcriptional regulator of hepatocyte differentiation (Si-Tayeb et al., 2010), was dramatically decreased in KO livers (Fig. 3G,G′). C/EBPɑ, which is restricted in fully differentiated hepatocytes (Birkenmeier et al., 1989), also showed obvious decrease in KO livers (Fig. 3H,H′). All these data suggest a role for Cdk5rap3 in hepatocyte maturation.

The cellular compositions in the fetal liver are complex. To verify the cell-autonomous role of Cdk5rap3 on hepatocyte maturation, we isolated hepatoblasts (E-cadherin⁺) from E14.5 fetal livers and then cultured in complete culture medium supplemented with appropriate factors promoting hepatic differentiation. At day 3, both wild-type and KO hepatoblasts proliferated rapidly, reached confluence and adopted a compact morphology (Fig. 3I,I′). At day 12, wild-type hepatoblasts showed drastic morphological changes: they adopted the typical morphology of mature hepatocytes with a small round nucleus and a dark cytoplasm, while the KO hepatoblasts displayed a disorganized structure (Fig. 3J,J′). Albumin, produced by differentiated hepatocytes, was detected only in wild-type hepatoblast cultures (Fig. S3J,J′). Altogether, these data suggest a role for Cdk5rap3 in hepatocyte proliferation and maturation, as well as in cholangiocyte differentiation.

Cdk5rap3 deletion resulted in both liver hypoplasia and severe anemia. Which phenotype(s) is responsible for the embryonic lethality needs to be clarified. We took advantage of different Cdk5rap3 conditional knockout mouse models to dissect the complexity (Fig. S1A). Specific deletion of Cdk5rap3 in embryonic hematopoietic cells using Vav-Cre mouse caused mild defects in hematopoiesis (data not shown), but the mutant mice survived to adulthood (Fig. S4A). We derived Cdk5rap3 conditional null mice in the liver. Alb-Cre mice were reported to express Cre in the liver from E15 (Krupczak-Hollis et al., 2004). Cdk5rap3^tm1d/tm1d; Alb-Cre mice were generated and surprisingly survived well into adulthood (Fig. S4B). The morphology and the histology of the livers in the mutant mice seemed normal (Fig. S4C,D). We presume this is due to insufficient deletion of Cdk5rap3 in hepatocytes (Fig. S4E,F).

We switched to use Foxa3-Cre transgenic mice, in which the Cre recombinase is active at E8.5 in the hepatic diverticulum (Lee et al., 2005). The Cdk5rap3^tm1d/tm1d; Foxa3-Cre mice (referred to as CKO hereafter) survived to birth, but started to die after weaning (Fig. 4A,B). Relatively high knockout efficiency was detected by both western blot and immunostaining (Fig. 4C-D′). The mutant mice were born without obvious defects, but weighed less starting from week 1 (Fig. 4E), suggesting abnormal energy consumption. We found that CKO livers were significantly smaller than control livers (Fig. 4A,F). When normalized to BW, CKO mice exhibited a significant reduction in LW/BW ratios compared with controls (Fig. 4F). Collectively, it seems that liver hypoplasia is the main reason for the lethality in Cdk5rap3-null mice.

In metabolism, 1-month-old CKO mice displayed severe hypoglycemia, as evidenced by the significantly lower serum glucose at both fasting and fed state (Fig. 4G). Decreased serum cholesterol (CHO) and triglyceride (TG), increased total bilirubin (TBIL), and elevated alanine aminotransferase (ALT) levels (Fig. S4G) all indicated abnormal liver function. However, no significant difference in serum aspartate transaminase (AST) levels was observed (Fig. S4G).

After weaning, energy intake is changed from mother's milk to solid food, which might represent a stressful adaption period for most mutant mice. Indeed, when we tested blood glucose at weaning, no significant difference was observed between CKO and control mice (Fig. S4H). However, after fasting for 6 h, blood glucose was significantly lower in CKO mice (Fig. S4H). Owing to glucose being a major source of energy, we wanted to know whether glucose feeding could rescue the CKO lethality. To test this, we fed CKO and control mice water supplemented with 10% glucose after weaning at week 4. The CKO mice did not die until week 7 (Fig. S4I), indicating that hypoglycemia is one of the causes for the lethality.

To determine whether Cdk5rap3 gene deletion in liver affected parenchymal organization, we prepared Hematoxylin and Eosin-stained sections from CKO and control mice at 0.5 and 1 month, respectively. Liver sections showed relatively organized parenchyma at 0.5 months (Fig. 5A,A′), whereas defective hepatic cord formation was observed in CKO mice at 1 month (Fig. 5B,B′). For the development of bile duct, all CKO livers at 1 month contained CK19-positive tubular and ductal plate-like structures, but their structural organization varied considerably from liver to liver (Fig. 5C,C′). Some CKO livers contained multiple, disorganized tubular structures that extended away from the portal tracts (Fig. 5C′).

To address the mechanism by which Cdk5rap3 deficiency may contribute to the abnormal liver development, we investigated the role of cell proliferation, apoptosis, as well as hepatocyte size. PCNA staining showed that a smaller number of PCNA-positive hepatocytes in CKO mice when compared with their controls at both 0.5 and 1 months (Fig. 5D-E′; Fig. S5A). Expression of cyclin B1 was unchanged, whereas that of cyclin D1 was significantly decreased in CKO livers (Fig. S5B). Consistent with our finding in embryos, no obvious apoptosis was detected in CKO livers at 1 month, as determined by the expression of activated caspase 3 (Fig. 5F,F′). This indicated that CDK5RAP3 is not important for hepatocyte survival postnatally. Interestingly, we observed that CKO mouse hepatocytes had a smaller cell area than controls at 1 month (Fig. 5B,B′). The hepatocyte size of control mice increased by 1.34-fold from 0.5 to 1 months, whereas there was only marginal increase for CKO mice (Fig. S5C). We deduced that the smaller liver size in CKO mice could be caused by both decreased hepatocyte size and impaired cell proliferation. More interestingly, shRNA-mediated Cdk5rap3 knockdown in human hepatocellular carcinoma HepG2 cells also caused a lower proliferation rate and reduced metabolic activity by MTT assay (Fig. S5D-F).

Hepatocyte maturation was detected using HNF4ɑ and C/EBPα expression. Decreased expression of both proteins in 1-month-old CKO livers suggests that Cdk5rap3-deficient hepatocytes are less mature (Fig. 5G-H′). Functionally, the glycogen accumulation (as determined by PAS staining) and lipid production (as determined by the Oil Red O staining) were both severely disrupted in CKO livers at 1 month (Fig. 5I-J′).

Next, we sought to determine how Cdk5rap3 mediates hepatocyte proliferation, differentiation and maturation. Considering CDK5RAP3 was reported as a binding protein in previous studies, we used co-immunoprecipitation (co-IP) to isolate its binding proteins in E16.5 wild-type liver cells, and analyzed them using mass spectrometry (Fig. S6A). Table 2 shows selective candidates of CDK5RAP3-interacting proteins. The top-ranked candidate, UFL1, is the E3 ligase in the UFM1 system. Previously, several studies demonstrated that CDK5RAP3 forms a complex with UFL1 in cultured cells (Kwon et al., 2010; Shiwaku et al., 2010; Wu et al., 2010; Yoo et al., 2014). Western blot further confirmed this interaction (Fig. 6A). However, no one has reported the function of CDK5RAP3 in the UFM1 conjugation system. We examined the expression of UFM1 system in KO, CKO and control livers. The expression of UBA5, UFC1, UFL1 and UFBP1 (a putative UFM1-conjugated substrate) remained unchanged (Fig. 6B; Fig. S6B). UFSP1 and UFSP2 expression was elevated after Cdk5rap3 loss in both KO and CKO mice, except that UFSP1was not detected in wild-type and KO fetal livers, possibly owing to low expression or complications with our antibody (Fig. 6B; Fig. S6B). We then explored whether the ufmylation profile was influenced by the loss of Cdk5rap3. Surprisingly, Cdk5rap3 deficiency altered the choice of UFM1-conjugated substrates in both KO and CKO livers (Fig. 6C,D; Fig. S6C). These results suggested that CDK5RAP3 acts as a ufmylation substrates adaptor. Importantly, the changed profiles of ufmylated substrates were the same at E16.5 and 1 month (when CKO mice had a high deletion rate) (Fig. 6C,D; Fig. S6C), suggesting that this change was not in a temporal manner, but rather a direct result of Cdk5rap3 deficiency. As both UFSP1 and UFSP2 were capable of releasing UFM1 from Ufm1-conjugated proteins, we deduce that UFSP1 and/or UFSP2 were upregulated to re-balance the UFM1 conjugation system.

Of note, the ufmylation profile in human HepG2 cells was also altered after Cdk5rap3 knockdown compared with the controls (Fig. S6D), suggesting a conservative role of CDK5RAP3 in mammals. Collectively, dysregulation of the ufmylation network might account for the liver hypoplasia observed in both complete and conditional knockout mice.

During postnatal development, the liver acquires an adult gene expression program, and initiates the expression of enzymes involved in major metabolic processes (Klaassen and Aleksunes, 2010). To comprehensively understand the hypoplastic livers observed in CDK5RAP3-deficient mice, we performed quantitative proteomics on liver tissues from both CKO and control mice. A total of 5125 proteins were profiled, among which 299 were identified to be differentially expressed between CKO livers and control livers (Fig. S7A). We subjected these differentially expressed proteins (DEPs) to gene ontology (GO) term analysis. Among cellular component (CC) ontologies, the DEPs were largely enriched in ER and ER-related terms, such as ER region, ER membrane and nuclear outer membrane-endoplasmic reticulum membrane network (NM-ER network) (Fig. 7A), and displayed a largely upregulated pattern in CKO livers compared with control livers (Fig. S7B). Conversely, among biological process (BP) ontologies, the DEPs were mainly enriched in metabolic processes, including several lipid metabolic processes (Fig. 7B). Using the DEP gene set, we performed gene set enrichment analysis (GSEA) under the GO term of endoplasmic reticulum (GO:0005783). Notably, the results showed that Cdk5rap3 loss led to the upregulation of 51 genes under the GO term (Fig. 7C). The heatmap of the upregulated enrichment genes under the ER ontology shows variations in protein expression among different livers (Fig. S7C). The expression changes of ER-related proteins can be partially explained by the abundant expression of CDK5RAP3 in the ER, as evidenced by the colocalization of CDK5RAP3 and protein disulfide isomerase (PDI): an ER marker (Fig. S7D).

The changed expression of ER-bound genes might disturb ER homeostasis, which will trigger ER stress and activate the unfolded protein response (UPR). ER stress-related proteins were analyzed in CKO and control livers. The PERK pathway leads to a transient global translational arrest through phosphorylation of eIF2α (Schröder and Kaufman, 2005). In our results, both p-PERK and p-eIF2α levels were found to be upregulated in CKO hepatocytes (Fig. 7D; Fig. S7G), which indicated that there was translational arrest in CKO mice. We detected that, after Cdk5rap3 deletion, the expression of both ATF4 and pro-apoptotic CHOP were upregulated (Fig. 7D; Fig. S7G). However, RT-qPCR analysis showed that Gadd34 expression, another target driven by ATF4, was not significantly changed (Fig. S7E). In addition, not all pro-apoptotic genes we analyzed [Bax, Bak1, Dr5 (Tnfrsf10b) and Bid] showed elevated expression after Cdk5rap3 loss, and the expression of the anti-apoptotic gene Bcl2 remains unchanged (Fig. S7E). This result supports our conclusion that Cdk5rap3 is not crucial for hepatocyte survival. For the IRE1α pathway, although IRE1α expression was not changed in CKO mice, the amount of both unspliced Xbp-1 and its active form [Xbp-1(s)] was elevated after Cdk5rap3 deletion, also indicating the activation of IRE1α pathway (Fig. 7D; Fig. S7G).

However, protein-level expression of BIP, the ER luminal chaperone, showed no change after Cdk5rap3 deletion in liver (Fig. S7F,G). ATF6, as a 90 kDa protein (p90ATF6), as well as its cleaved form (ATF6a), were both detected, but neither showed significant changes in CKO hepatocytes, indicating CDK5RAP3 has no effect on ATF6 signaling pathway (Fig. S7F,G). Taken together, these data indicated that Cdk5rap3 loss caused ER stress and activated IRE1α and PERK signaling pathways.

The developmental study of Cdk5rap3 during zebrafish morphogenesis indicated that Cdk5rap3 is indispensable for cell cycle progression, doming and zebrafish epiboly (Liu et al., 2011). Here, data from our mouse models have expanded the functional diversity of Cdk5rap3 to mammalian development, mainly in liver development and hematopoiesis. The most striking defect observed in Cdk5rap3 KO embryos was liver degeneration. Anemia and hemorrhage were also observed, suggesting impaired hematopoiesis and compromised vessel integrity. Identified as one of the candidate proteins for ufmylation, CDK5RAP3 seems to share similar biological functions with other components, such as UBA5, UFL1 and UFBP1, in hematopoiesis (Cai et al., 2015; Tatsumi et al., 2011; Zhang et al., 2015). Mice lacking Uba5 also died in utero, and exhibited severe anemia, owing to impaired erythropoiesis (Tatsumi et al., 2011). Of note, the rescued mice that expressed UBA5 in the erythroid lineage still died before E18.5, with severe liver hypoplasia, suggesting a possible role for Uba5 in liver development as well. Notwithstanding, to date, the underlying molecular mechanism for the UFM1 system remains poorly defined, and we still do not know whether the function of UFM1 system in liver development or hematopoiesis is ufmylation dependent or not.

Previous literature has proposed CDK5RAP3 as a UFM1 substrate, based on the interaction of CDK5RAP3 with UFM1, UFL1 and UFBP1 (Kwon et al., 2010; Lemaire et al., 2011; Wu et al., 2010; Yoo et al., 2014). However, this is not conclusive as direct evidence is lacking. In our study, we surprisingly found that the absence of Cdk5rap3 in hepatocytes altered the ufmylated substrates. We define CDK5RAP3 as the ufmylation adaptor, the first discovered so far. In ubiquitin-conjugation system, some substrate adaptors, such as the F-box proteins, have been well defined (Skaar et al., 2013). These substrate adaptors guarantee specific substrate recruitment and certain biological functions, and their dysregulation contributes to multiple pathologies (Skaar et al., 2013). Identifying and characterizing these ufmylated substrates, in the presence and absence of CDK5RAP3, is urgently needed and will greatly expand our knowledge of this UBL conjugation system and the precise role of CDK5RAP3 in it.

ER is an organelle responsible for the protein processing, folding, modification and trafficking, as well as lipid biosynthesis, which plays a pivotal role in integrating multiple metabolic signals required for cellular homeostasis (Hotamisligil, 2010; Schröder and Kaufman, 2005). At the subcellular level, ufmylation genes, such as Ufm1, Ufl1 and Cdk5rap3, are enriched in the ER (Shiwaku et al., 2010; Tatsumi et al., 2010; Wu et al., 2010). Previous studies have indicated that aberrant ufmylation triggers ER stress. Knockdown of ufmylation genes triggered UPR and amplification of the ER network (Zhang et al., 2012). Elevation of ER stress and activation of UPR were observed in Ufl1- and Ufbp1-deficient bone marrow cells and HSCs (Cai et al., 2015; Zhang et al., 2015). In this study, we also uncovered the biological function of Cdk5rap3 in maintaining ER homeostasis. In Cdk5rap3 knockout, ER-bound genes were affected. Moreover, Cdk5rap3 deficiency triggered ER stress and activated UPR, including the PERK and IRE1α signaling pathway. But how is ER stress triggered? Are ufmylated substrates involved to trigger ER stress? These are interesting questions to be addressed in future works.

ER homeostasis is of particular importance in hepatocytes. Hepatocytes are enriched in both smooth and rough ER. Numerous proteins and lipids are processed in the ER during liver development to direct and fulfill the proper functions in a temporal manner (Hetz, 2012). Perturbation of ER homeostasis will disrupt lipid metabolism and induce hepatic lipotoxicity (Zhou and Liu, 2014). Even worse, sustained or massive ER stress leads to hepatocyte apoptosis (Dara et al., 2011). Based on our data, a possible working mechanism for Cdk5rap3 in liver development is proposed (Fig. 7E). In wild-type hepatocyte, CDK5RAP3 binds to UFL1, which recruits specific substrates. These ufmylated substrates help maintain ER homeostasis. Hepatocytes go through normal proliferation, differentiation and maturation. In contrast, in Cdk5rap3-deficient hepatocyte, the ufmylated substrates were altered, which triggers ER stress. The unsolved ER stress impairs the hepatocyte proliferation, differentiation and maturation. It is plausible that restoration of ER homeostasis might at least partially rescue the defects observed in Cdk5rap3-deficient mice. Yet the possibility that CDK5RAP3 affects many proteins at the same time and simultaneously alters several cellular processes cannot be excluded. It is unclear whether the effect of Cdk5rap3 deletion is really dependent on its interaction with UFL1. In our future study, this mechanism needs to be strengthened by: (1) the identification of UFM1 substrates in both knockout and control livers; and (2) the examination of ufmylation patterns in Cdk5rap3-KO hepatocytes introduced with wild-type Cdk5rap3 and its mutant lacking UFL1-binding region.

Clinically, Cdk5rap3 is involved in gastric cancer (Wang et al., 2017; Zheng et al., 2018), colorectal cancer (Chen et al., 2011), hepatocellular carcinoma development and metastasis (Mak et al., 2011, 2012; Zhao et al., 2011). Our research has provided new insight in understanding the pathogenesis of these diseases, in which ufmylation and ER stress might be important factors. Indeed, ufmylation profile changes were observed in human hepatocellular carcinoma (R.Y., H.W., B.K., Y.J. and Y.H., unpublished). In addition, future identification and characterization of ufmylated substrates would provide novel therapeutic targets for diseases such as cancer, anemia and liver degeneration.

The Flp mice [B6.129S4-Gt(ROSA)26Sor^tm1(FLP1)Dym/J] (Farley et al., 2000), the EIIa-Cre mice [FVB/N-Tg(EIIa-cre)C5379Lmgd/J] (Lakso et al., 1996), the Vav-Cre mice [B6.Cg-Tg(Vav1-cre)A2Kio/J] (de Boer et al., 2003), the Alb-Cre mice [B6.Cg-Tg(Alb-cre)21Mgn/J] (Postic et al., 1999) and the Foxa3-Cre mice [Tg(Foxa3-cre)1Khk/Mmmh] (Lee et al., 2005) were used as described previously. All mouse experimental protocols were approved by the Institutional Animal Care and Use Committee at Peking Union Medical College and Chinese Academy of Medical Sciences. All animal care and experimental methods were carried out in accordance with the ARRIVE guidelines for animal experiments (www.nc3rs.org.uk/arrive-guidelines). All the mice used in this study were reared in specific pathogen free (SPF) facilities.

We used primary antibodies against activated caspase-3 (Cell Signaling Technology, 9664), ATF4 (Abcam, ab184909), ATF6 (Cell Signaling Technology, 65880), β-actin (Santa Cruz, sc-16), BiP (Cell Signaling Technology, 3177), BrdU (Abcam, ab6326), CDK5RAP3 (Abcam, ab157203; Santa Cruz, sc-134627), C/EBPα (Cell Signaling Technology, 8178), Chop (Abcam, ab11419), cyclin B1 (Santa Cruz, sc-245), cyclinD1 (Cell Signaling Technology, 2978), cytokeratin 19 (Abcam, ab52625), DLK (Abcam, ab119930), eIF2α (Santa Cruz, sc-11386), GAPDH (Santa Cruz, sc-25778), HNF4α (Santa Cruz, sc-6556), IREIα (Santa Cruz, sc-20790), PCNA (Santa Cruz, sc-56), PDI (Santa Cruz, sc-74551), p-eIF2α (Cell Signaling Technology, 9721), pHH3 (Abcam, ab14955), p-PERK (Cell Signaling Technology, 3179), UFBP1 (Santa Cruz, sc-85328), UFM1(Abcam, ab109305), UFSP1 (Santa Cruz, sc-398577), UFSP2 (Abcam, ab185965) and Xbp1(Abcam, ab37152). The antibodies for UBA5, UFC1 and UFL1 were kindly provided by Dr Masaaki Komatsu (Tokyo Metropolitan Institute of Medical Science). We purchased horseradish peroxidase (HRP)-linked secondary antibodies from Cell Signaling Technology. For more detailed information, see Table S2.

ES cell clone HEPD0516_2_A06 (cell type: JM8.N4) carrying the Cdk5rap3^tm1a allele was purchased from the EUCOMM team (RRID: SCR-003104). The ES cells were injected into C57BL/6 blastocysts to obtain chimeras. The resulting chimeric mice were crossed with B6(Cg)-Tyr^C-2J/J albino mice to obtain offspring carrying the Cdk5rap3^tm1a allele, namely Cdk5rap3^tm1a/+ mice. Cdk5rap3^tm1b/+ mice were produced by crossing Cdk5rap3^tm1a/+ mice with EIIA-Cre mice. The Cdk5rap3^tm1c/+ allele was produced by crossing Cdk5rap3^tm1a/+ mice with Flp mice. Cdk5rap3^tm1d/+ allele was generated by mating Cdk5rap3^tm1c/+ mice with different tissue-specific expressing Cre mice (Fig. 1A; Fig. S1A).

Southern blots were performed according to standard protocols (Church and Gilbert, 1984). Genomic DNA extracted from ES cells was digested with SacI and EcoRI separately, electrophoresed through a 0.9% agarose gel and transferred to Hybond-N+ (GE Healthcare). Hybridization was carried out using ³²P-labeled DNA probes made by Megaprime DNA Labelling System (GE Healthcare). The probes used to detect DNA fragments were synthesized by PCR. The corresponding primers for this PCR are listed in Table S1.

Hepatoblasts were isolated from E14.5 KO or wild-type mouse livers as described previously (Gailhouste, 2012). Briefly, fetal livers were collected and dissociated into single cells. Magnetic nanoparticles (StemCell Technologies) were used to isolate E-cadherin-positive hepatoblasts from the cell suspension. Cells were cultured in a complete culture medium supplemented with HGF, EGF and OSM, promoting hepatic differentiation.

Total RNA was extracted using Trizol reagent (Invitrogen). Reverse transcription was performed using the PrimeScript II 1st Strand cDNA Synthesis Kit (TaKaRa, D6210A). Real-time PCR was performed using the CFX96 Touch Real-Time PCR Detection System (Bio-Rad, 185-5196). Primer sequences for Cdk5rap3 and β-actin are listed in Table S1. The reactions were run in triplicate.

Whole-cell lysates were separated by SDS-polyacrylamide gel electrophoresis (SDS-PAGE), and electroblotted to polyvinylidene difluoride membranes (Bio-Rad). The membranes were blocked in 5% BSA dissolved in Tris-buffered saline containing 0.1% Tween-20 for 1 h at room temperature. The membranes were incubated with the primary antibody overnight at 4°C, followed by treatment with HRP-conjugated secondary antibody for 1 h at room temperature. Immuno-reactive bands were visualized using enhanced chemo-luminescence and the ChemiDoc XRS (Bio-Rad).

For the analysis of hematopoietic stem cells in fetal livers, fetal livers were dissected from E14.5 embryos. They were pipetted to dissociate them and cells were filtered through 40 µm nylon filters (BD Bioscience) to obtain a single-cell suspension. For HSCs analysis, cells were first incubated with lineage markers (Ter-119, CD3, B220 and Gr-1), CD34, Sca-1, Kit, FcrR and Flt3. Stained cells were analyzed by FACS Calibur (BD Bioscience). The resulting cell distribution profiles were analyzed using FlowJo software (Tree Star).

Tissues were collected and fixed in 10% phosphate-buffered formalin overnight at 4°C. Frozen or paraffin sections were prepared using standard procedures. For histology, slides were stained with Hematoxylin and Eosin. For immunohistochemical staining, sections were rehydrated and treated with 3% hydrogen peroxide to suppress the endogenous peroxidase activity. Antigen retrieval was achieved by boiling at 121°C for 15 min in 10 mM citrate buffer followed by gradual cooling to room temperature. Sections were incubated overnight with the primary antibodies at 4°C, followed by incubation of HRP-conjugated secondary antibodies. Immunohistochemical staining was developed using DAB (Sigma), and counterstained with Hematoxylin. For immunofluorescence staining, sections were incubated with the primary antibodies, followed by the appropriate Alexa Flour-coupled secondary antibodies (Invitrogen). Sections were photographed on a confocal fluorescence microscope (Olympus, FV1000).

For periodic acid-Schiff staining, sections were oxidized in 0.5% periodic acid (Sigma) and stained with Schiff's reagent (Sigma). Oil red O staining for neutral lipids was performed on fresh frozen tissue cut in 10 µm sections. Slides were air and fixed in 10% formalin. The slides were then rinsed in distilled water, placed in propylene glycol and stained with 0.5% oil red O solution at 60°C. After rinsing in 85% propylene glycol solution, slides were washed with distilled water and counterstained with Hematoxylin.

Serum was prepared from whole blood by centrifugation at 6000 g using the Microtainer serum separator tubes (BD Biosciences). The serum levels of CHO, TG, TBIL, ALT and ALP were measured using a HITACHI 7100 Automatic Analyzer (HITACHI, Japan). The serum level of glucose was measured by the ONETOUCH Ultra Family of Meters (BD Biosciences).

The knockdown experiments by shRNA were performed as previously described (Zhang et al., 2012). Lentiviral vector was constructed using pll3.7 vector. Lentiviruses were prepared using 293T packaging cell line according to the manufacturer's instruction (Invitrogen). The sequences of Cdk5rap3 and Ufm1 shRNA used are listed in Table S1. For knockdown assays, HepG2 cells were infected with lentiviruses expressing scrambled Cdk5rap3 or Ufm1 shRNA. The lentiviruses carry gfp element, and successfully transfected HepG2 cells will express GFP. After transfection for 24 h, the HepG2 cells were sorted by flow cytometry using GFP fluorescence. Knockdown efficiency was evaluated by immunoblotting.

HepG2 cells transfected with scrambled, Cdk5rap3 or Ufm1 shRNA were seeded in 96-well plates. Cell proliferation rate was evaluated at 24, 48, 72, 96 and 120 h. Briefly, 100 μl of 0.5 mg/ml of MTT solution was added to each well and incubated for 4 h at 37°C.The medium was then removed and 100 μl of dimethyl sulfoxide (DMSO, Sigma) was added to dissolve the content. Finally, absorbance was recorded at 570 nm with a reference filter at 630 nm. All tests were carried out in triplicate.

Immunoprecipitation was performed as previously described (Liu et al., 2016). Whole-cell lysates (WCLs) from E16.5 wild-type livers were prepared. WCLs were incubated with anti-CDK5RAP3 antibody and rabbit IgG at 4°C overnight. Dynabeads Protein A (Life Technologies) were added and the mixture was incubated for another 3 h. After being washed three times, the beads were boiled in SDS-PAGE loading buffer. Samples were separated by SDS-PAGE. After silver staining, bands of interest were cut for mass spectrometry analysis using nanoLC-LTQ-MS/MS (ThermoFinnigan), which was performed by the Core Facility Center, Institute of Biophysics Proteomics, Chinese Academy of Sciences. Tandem mass spectra were matched against the National Center for Biotechnology Information mouse database with SEQUEST BioWorks software (Thermo Fisher Scientific). For a peptide to be considered legitimately identified, it had to be the number one matched and had to achieve cross-correlation scores of 1.9 for [M+H]1+, 2.5 for [M+2H]2+ and 3.75 for [M+3H]3+. The scores of proteins are related to the number and the cross-correlation scores of enriched peptides, which calculated with SEQUEST BioWorks software.

Mouse liver proteins were extracted in lysis buffer [10 mM Tris-HCl (pH 8.0), 0.1 M NaH₂PO₄, 8 M urea, 0.02% SDS and 10 mM β-mercaptoethanol). The lysate was reduced with DTT and alkylated with iodoacetamide. The sample was then diluted in 2 M urea with buffer (5% AcN in 50 mM NH₄HCO₃) and digested with trypsin.

We labeled our samples with 10-channel tandem mass tag (TMT) reagents, as previously described (Zhou et al., 2012). Peptides were dissolved in 300 µl of 0.1% trifluoroacetic acid and vortexed for 5 min. The high pH spin columns (Pierce product 84868) were conditioned according to manufacturer's protocol and the sample was fractionated accordingly into 24 fractions and finally combined into 12, as described previously (Sun et al., 2018).

Peptides were dissolved in 10 µl of 0.1% formic acid (FA) and 5 µl was loaded onto a self-packed 60 cm long 75 μm internal diameter analytical column packed with 1.9 μm C18 beads (Dr Maisch). Elution was performed on a Dionex Ultimate 3000 RSLCnano (Thermo, Germany) at 225 nl per minute over a 160-min long gradient (buffer A was 0.1% FA in water and buffer B was 0.1% FA in acetonitrile). The elution scheme was as follows: 1% to 5% in 5 min, 5% to 30% in 140 min, 30% to 60% in 5 min, 60% to 99% in 5 min and 5 min at 99%. The column was then conditioned at 1% for 20 min. Ions were detected using a Fusion Orbitrap mass spectrometer (Thermo, Germany) set to acquire the top speed of a 3 s cycle for the full 180 min. The survey scan was collected in profile mode with the resolution set at 120,000, automatic gain control (AGC) was set to 400,000, maximum injection time was set to 50 ms, the scan range was 350 to 1500 with quadrupole isolation on and the RF lens was set at 60%. Tandem mass scans were collected in centroid mode after higher energy collision dissociation (HCD), the resolution was set to 30,000, AGC was set at 50,000, maximum injection time was set to 100 ms, the first mass was set at 110 m/z, isolation width was set at 0.7 m/z and collision energy was 40%. Dynamic exclusion was set to exclude ions for 20 s within a tolerance window of ±10 ppm.

All spectra were analyzed using Proteome Discoverer 2.1 with Sequest HT against Uniprot MOUSE protein sequence database (www.uniprot.org). Precursor mass tolerance was set to 20 ppm, while product ion tolerance was set to 0.05 Da. Dynamic modifications (max of 4 per peptide) included methionine oxidation (+15.9994), asparagine and glutamine deamidation (+0.98402), peptide n-terminal TMT (+229.16293) and protein n-terminal acetylation (+42.03670). Static modifications included cysteine carbamidomethylation (+57.02146) and lysine TMT (+229.16293). The percolator node was used to filter peptide spectral match (PSM) and peptides to an FDR of 1%.

The body of analysis was performed in R (www.r-project.org). For proteomic data, differentially expressed proteins were defined as being differentially expressed between CKO livers and control livers, with log₂FC>±0.5 and P<0.05. Gene symbols and Entrez IDs of such DEPs were retrieved by mapping the Uniprot accession numbers of DEPs in the R database org.Mm.eg.db. Over-representation analysis (ORA) was performed on the DEP genes to measure their enrichment in different GO terms acquired from the Gene Ontology Consortium (GOC, geneontology.org/). Both CC and BP GO terms were filtered to level 4 for the enrichment analysis. Enrichment results were filtered using a P value threshold of 0.05. Gene set enrichment analysis (GSEA) was performed using fold change as the gene-level stats. Plots were made using R packages ggplot2 (ggplot2.tidyverse.org/) and enrichplot (github.com/GuangchuangYu/enrichplot).

Data are presented as mean±s.d. unless stated otherwise. Differences between groups were assessed using a two-tailed t-test assuming unequal variance. Differences were considered to be significant at P<0.05. For over-representation analysis and gene set enrichment analysis, P values were corrected using the Benjamini-Hochberg procedure, with the resulting q value considered a measure of false discovery rate (FDR).

We thank Dr Chengran Xu (Peking University) for kindly giving us the Foxa3-Cre transgenic mice; Dr Masaaki Komatsu (Tokyo Metropolitan Institute of Medical Science) for providing the antibodies for UBA5, UFC1 and UFL1; and Dr Yongjun Yin and Andrew Hagan (Washington University in Saint Louis) for critical reading of the manuscript.