Versatile JMJD proteins: juggling histones and much more

Abstract

Jumonji C domain-containing (JMJD) proteins are found in bacteria, fungi, animals and plants. They belong to the 2-oxoglutarate-dependent oxygenase superfamily and are endowed with various enzymatic activities, including demethylation of histones and hydroxylation of non-histone proteins. Many JMJD proteins are involved in the epigenetic control of gene expression, yet they also modulate a myriad of other cellular processes. Here, we focus on the 33 human JMJD proteins and their established and controversial catalytic properties, survey their epigenetic and non-epigenetic functions, emphasize their contribution to sex-specific disease differences and highlight how they sense metabolic changes. All this underlines not only their key roles in development and homeostasis, but also that JMJD proteins are destined to become drug targets in multiple diseases.

JMJD proteins: modulators of histones and non-histone proteins

The molecular foundations of epigenetics are largely rooted in DNA methylation and histone posttranslational modifications, including arginine and lysine methylation [1]. In the 20^th century, histone methylation was widely considered to be an irreversible modification, but this view was unequivocally shattered in 2004 and 2006 with the discovery of two classes of lysine demethylases (KDMs), the KDM1A and KDM1B amine oxidases and the much more numerous JMJD proteins [2,3]. These enzymes counter histone lysine methyltransferases (KMTs), which together account for the dynamism of histone methylation affecting chromatin structure [4]. Yet, JMJD proteins also demethylate non-histone proteins, hydroxylate several amino acid side chains in proteins or RNA, cleave proteins or perform catalytic activity-independent tasks in and outside the cell nucleus [5,6]. Thus, the expansive JMJD protein family is implicated in a great variety of molecular processes beyond the modulation of chromatin and undeniably required for homeostasis and development.

On the flip side, abnormal JMJD activity is associated with many diseases. Prominently, several JMJD proteins were recognized as tumor suppressors or oncoproteins that, when displaying inactivating mutations or overexpression, respectively, facilitate tumorigenesis [2,7,8]. Some JMJD proteins even exert a dual function as a tumor suppressor and oncoprotein: for instance, KDM3B is an oncoprotein in colorectal cancer but a tumor suppressor in acute myeloid leukemia, while KDM5B fosters breast cancer but inhibits melanoma formation [9,10]. Notably, KDM5B may also promote melanomagenesis – likely after disease initiation – by enabling immune evasion [11], indicating a switch from tumor suppressor to promoter during disease progression.

Here, we provide a synopsis of the JMJD protein family and their epigenetic functions, with special emphasis on how their dysregulation can underlie sex-dependent differences in disease prevalence. In addition, we highlight other recent discoveries about non-epigenetic and even non-nuclear functions of JMJD proteins as well as their roles in sensing metabolic changes. This culminates in discussing how to leverage JMJD protein inactivation for disease treatment.

Broad catalytic activities

2-Oxoglutarate- and oxygen-dependent mechanisms of action

JMJD proteins belong to the 2-oxoglutarate (α-ketoglutarate)-dependent oxygenase superfamily, whose catalytic activity additionally requires Fe(II) and molecular oxygen while succinate and CO₂ are generated as byproducts [7,12]. The defining element of JMJD proteins is the Jumonji C (JmjC) domain encompassing a distorted double-stranded β-helix core fold with 8 anti-parallel β-strands. The JmjC domain, whose average length is ~160 amino acids but can vary by more than 90 amino acids between individual JMJD proteins, contains an HX(D/E)X_nH motif that is crucial for coordinating Fe(II). One mutation in this motif may be permissible (e.g., HXDX_nY in PHF2), but two as in JARID2 (SXDX_nV) seems to be detrimental for catalytic activity. JMJD proteins developed early during evolution, as they exist in yeasts, worms, flies, mammals, and plants, as well as bacteria [2,13]. In humans, 33 different JMJD proteins were identified (Figure 1, Table 1), making them one of the largest protein families. Confusingly, many JMJD proteins are known by several aliases, in part reflecting particular enzymatic activities (e.g., lysine demethylation for KDMs) or the presence of functional domains (e.g., PHD in PHF proteins), and their current official names that are employed throughout this review are not necessarily “user” friendly. This includes the naming of many JMJD proteins as KDMs similar to KDM1A/KDM1B, despite that the latter are not structurally homologous to JMJD proteins [2,3]. Further, KDM1 function does not require Fe(II) and 2-oxoglutarate, but rather KDM1 enzymatic activity involves the reduction of the cofactor flavin adenine dinucleotide (FAD) and its subsequent reoxidation leading to the production of H₂O₂ as a byproduct [2,14].

Figure 1. Phylogenetic tree of the 33 human JMJD proteins.

Localization of protein domains is according to UniProt, while phylogeny of the depicted isoforms (specified in Table 1) was assessed with SeaView Version 4. Closely related JMJD proteins are graphically aligned by their Jumonji C (JmjC) domains. Abbreviations for other domains are: ARID, AT-rich interaction domain; JmjN, Jumonji N domain; PHD, plant homeodomain; LRR, leucine-rich repeat; TPR, tetratricopeptide repeat. Proteins are named according to the Human Genome Organisation (HUGO) and respective abbreviations are: HIF1AN, hypoxia-inducible factor 1 subunit alpha inhibitor; HR, hairless; HSPBAP1, 27 kDa heat shock protein-associated protein 1; JARID2, Jumonji/ARID domain-containing protein 2; JMJD, Jumonji C domain-containing; KDM, lysine demethylase; PHF, PHD finger protein; RIOX, ribosomal hydroxylase; TYW5, tRNA wybutosine-synthesizing protein 5; UTY, ubiquitously transcribed TPR protein, Y-linked.

Table 1.

Reported catalytic activities of human JMJD proteins.

JMJD subfamilies (alternate names)	NCBI accession numbera	Reported catalytic activitiesb (histone demethylation sites, hydroxylation of non-histone proteins, and others)	References
KDM2A (FBXL11/JHDM1A) KDM2B (FBXL10/JHDM1B)	NP_036440.1 NP_115979.3	H3K4me3 (KDM2B) H3K36me2/1 H3K79me3/2 (KDM2B) Demethylation of lysine in non-histone protein	[2,3,57]
KDM3A (JMJD1A) KDM3B (JMJD1B) JMJD1C (KDM3C)	NP_060903.2 NP_057688.3 NP_116165.1	H3K9me2/1 H4R3me2/1 (KDM3B)? Demethylation of lysine in non-histone protein (KDM3A, KDM3C) Demethylation of arginine in unnatural peptides (KDM3A)	[9,18,58,77]
HR	NP_005135.2	H3K9me2/1	[5,9]
KDM4A (JMJD2A) KDM4B (JMJD2B) KDM4C (JMJD2C) KDM4D (JMJD2D) KDM4E (JMJD2E) KDM4F (JMJD2F)	NP_055478.2 NP_055830.1 NP_055876.2 NP_060509.2 NP_001155102.1 NP_001400684.1	H3K9me3/2 H1.4K26me3/2 H3K36me3/2 (not KDM4D-E) Demethylation of lysine in non-histone protein (KDM4C) H3R2me1 (KDM4E)? H3R2me2 (KDM4A, KDM4D, KDM4E)? H3R8me2 (KDM4E)? H3R26me2/1 (KDM4E)? H4R3me2 (KDM4D, KDM4E)? Demethylation of arginine in non-histone protein (KDM4E)?	[2,18,19,31,87]
KDM5A (JARID1A/RBP2) KDM5B (JARID1B/PLU-1) KDM5C (JARID1C/SMCX) KDM5D (JARID1D/SMCY)	NP_001036068.1 NP_001300971.1 NP_004178.2 NP_001140177.1	H3K4me3/2 H3R2me1 (KDM5C)? H3R2me2? H3R8me2/1 (KDM5C)? H4R3me2/1 (KDM5C)? Demethylation of arginine in non-histone protein (KDM5C, KDM5D)?	[10,18,19]
JARID2 (JMJ)	NP_004964.2	predicted to be catalytically inactive
KDM6A (UTX) UTY (KDM6C) KDM6B (JMJD3)	NP_001278344.1 NP_001245178.1 NP_001073893.1	H3K27me3/2 Demethylation of arginine in unnatural peptides (KDM6B)	[5,18]
KDM7A (KIAA1718) PHF2 (KDM7C) PHF8 (KDM7B)	NP_085150.1 NP_005383.3 NP_001171825.1	H3K9me2/1 H3K27me2 (KDM7A, PHF8) H4K20me1 (KDM7A, PHF8)	[5,48]
KDM8 (JMJD5) JMJD7	NP_001138820.1 NP_001108104.1	H3K36me2 (KDM8)? C-3 hydroxylation of arginine (KDM8) C-3 hydroxylation of lysine (JMJD7) Cleavage of arginine- or lysine-methylated histones	[6,12,21–23]
JMJD4 JMJD6 (PSR/PTDSR)	NP_075383.3 NP_001074930.1	H3R2me2/1 and H4R3me2/1 (JMJD6)? Demethylation of arginine in non-histone protein (JMJD6)? Demethylation of lysine in non-histone protein (JMJD4)? C-4 hydroxylation of lysine (JMJD4) C-5 hydroxylation of lysine (JMJD6) Cleavage of arginine-methylated MEPCE (JMJD6) Demethylation of 7SK RNA cap (JMJD6)? H2A.X tyrosine phosphorylation (JMJD6)?	[5,6,12,16,17,24–26,71,72]
JMJD8	NP_001005920.3	Demethylation of lysine in non-histone protein	[73]
RIOX1 (JMJD9/NO66) RIOX2 (JMJD10/NO52/MINA)	NP_078920.2 NP_694822.2	H3K4me3/2/1 (RIOX1)? H3K9me3 (RIOX2)? H3K36me3/2 (RIOX1)? C-3 hydroxylation of histidine	[5,6,12]
HIF1AN (FIH) TYW5	NP_060372.2 NP_001034782.1	C-3 hydroxylation of asparagine/aspartate/histidine (HIF1AN) tRNA hydroxylation (TYW5)	[5–7,15,20]
HSPBAP1 (PASS1)	NP_078886.2	unknown

^aRepresentative (mostly longest) isoforms are listed.

^bA bracket indicates that a catalytic activity has up to now not been reported for all subfamily members, but only for the specified protein(s) within the bracket. A question mark points out that this catalytic activity is considered to be uncertain and in need of corroboration.

Hypoxia-inducible factor (HIF) 1 subunit alpha inhibitor (HIF1AN) was the first JMJD protein whose enzymatic activity was uncovered in 2002. It hydroxylates the α subunits of the HIF transcription factor on an asparagine residue (Figure 2A). This hydroxylation precludes interaction with the CBP/p300 coactivators and thereby suppresses HIF activity [15]. However, it is still debated if decreased HIF1AN catalytic activity under hypoxia is seminal for the HIF-dependent cellular response to oxygen depletion [7]. HIF1AN was also identified to hydroxylate especially proteins with ankyrin repeats (see Glossary) not only on asparagine, but also on aspartate and histidine residues (Figure 2B, C), indicating promiscuity for HIF1AN catalytic activity [7].

Figure 2. Hydroxylation activities ascribed to JMJD proteins.

(A) Reaction mechanism exemplified by the C-3 hydroxylation of an asparagine residue by HIF1AN. Molecular oxygen and 2-oxoglutarate are required cofactors, while succinate and CO₂ are byproducts. (B) C-3 hydroxylation of aspartate. (C) C-3 hydroxylation of histidine. (D) Successive demethylation of N^ε-trimethylated lysine. The creation of a labile hemiaminal intermediate follows the same reaction mechanism as outlined in panel A, while the release of formaldehyde (CH₂O) does not require KDM enzymatic activity. (E) N^ω-methylation of arginine: monomethylation (left), asymmetrical (middle) or symmetrical (right) dimethylation. (F-H) C-5, C-4 and C-3 hydroxylation of a lysine residue. (I) C-3 hydroxylation of an arginine residue. (J) Conversion of guanosine into hydroxywybutosine in tRNA^Phe, including the step catalyzed by TYW5.

Heightened interest in JMJD proteins was elicited by the discovery that most of them demethylate histone lysine residues. The demethylation reaction is a two-step process: 2-oxoglutarate- and oxygen-dependent hydroxylation of an N^ε-methyl group, which creates a chemically labile hemiaminal, followed by the spontaneous release of formaldehyde (Figure 2D). This mechanism allows, in principle, JMJD proteins to demethylate mono-, di- and trimethylated lysine residues [2]. In addition, JMJD6 was reported to demethylate N^ω-methylarginine (Figure 2E) in histones via the same two-step mechanism as lysine demethylation [16], but this finding is highly controversial as is the purported arginine demethylation of non-histone proteins by JMJD6 [5]. Rather, JMJD6 performs C-5 lysyl hydroxylation (Figure 2F), including at multiple sites within unstructured lysine-rich domains [6,17]. In addition, when utilizing in part unnatural peptide substrates, some JMJD proteins were shown to demethylate mono- or dimethylated arginine residues in vitro [18,19] (Table 1). However, if arginine demethylation is performed by any JMJD protein in a physiological setting is still unresolved.

Like JMJD6, other JMJD family members may hydroxylate proteins (Table 1). This includes hydroxylation at the C-4 or C-3 position of lysine residues by JMJD4 and JMJD7, respectively, C-3 arginyl hydroxylation by JMJD5, and, like HIF1AN, C-3 histidyl hydroxylation by ribosomal hydroxylase (RIOX) 1 and 2 [12] (Figure 2C, G–I). Further, RIOX1 and RIOX2 were implicated in histone lysine demethylation, although this seems to be controversial [5,6]. Moreover, TYW5 is responsible for the hydroxylation of an unusual base in tRNA^Phe (Figure 2J) that – speculatively – may affect translational fidelity [20].

Catalytic action without 2-oxoglutarate and oxygen

In 2017, JMJD5 and JMJD7 were shown to clip lysine- or arginine-methylated histone tails at various residues, resulting into “tailless” nucleosomes that could facilitate transcription elongation [21,22]. The similarity of the JMJD5 JmjC domain with cathepsin L-type proteases rationalizes how JMJD5 can exert two fundamentally different catalytic activities. Notably, the JMJD5/7 peptidase function is unlikely to require 2-oxoglutarate and O₂ as cofactors, and Fe(II) in the catalytic center is replaceable by Zn²⁺ [21–23]. Similarly, JMJD6 may cleave the methylphosphate capping enzyme (MEPCE) at a methylated arginine residue [24]. A consequence of this cleavage was the activation of RNA polymerase II through releasing the stimulatory P-TEFb protein from the 7SK snRNP complex. Interestingly, the 7SK RNA is monomethylated by MEPCE at its 5’ γ-phosphate, implicating how JMJD6-mediated MEPCE destruction indirectly leads to reduced 7SK methyl capping rather than through a previously purported direct cap cleavage by JMJD6 [25]. Another unusual JMJD6 activity is tyrosine phosphorylation of histone H2A.X, which was dependent on both the JmjC domain and a serine-rich, 30 amino acid-long region [26]. However, this tyrosine kinase activity has not yet been independently corroborated and is hard to reconcile with the absence of structural similarities between JMJD6 and tyrosine kinases.

Epigenetic roles of JMJD proteins

Regulation of gene expression

Histones are tightly packed into nucleosomes, yet the N-termini of histones, especially H3 and H4, stick out from the nucleosome core and are comprised of many arginine and lysine residues that are amenable to posttranslational modifications. Both lysine and arginine methylation are part of the histone code that contributes to the epigenetic control of gene expression: e.g., H3K4me₃ at promoters and H3K4me₁ at enhancers stimulates transcription initiation, while H3K9me₃ and H3K27me₃ mark transcriptional repression [1,4,8]. But the role of any particular histone modification appears to be complex. For instance, H3K4me₃ is also present at inactive genes, yet together with H3K27me₃ puts such genes into a poised state that easily switches from a repressed to an active status [1,8]. By erasing the histone lysine code, many JMJD proteins activate or repress gene transcription depending on which lysine(s) they target, and this may also hold true for their potential function as erasers of arginine methylation [2,3]. A peculiarity is the catalytically inactive JARID2, which is actually needed to write the H3K27me₃ mark as a necessary component of a subset of Polycomb repressive complexes [27].

JMJD proteins do not sequence-specifically bind DNA, yet need to be recruited to chromatin to demethylate histones. For this, they harness different structural motifs (Figure 1). KDM5 proteins and JARID2 possess an ARID domain that non-specifically interacts with DNA [10]. KDM2 proteins contain leucine-rich repeats and an F-box that can be involved in protein-protein interactions [28,29], and the same holds true for the tetratricopeptide repeats in KDM6 proteins [30]. Hence, these motifs may facilitate the recruitment of KDMs to already chromatin-bound proteins. Similarly, several KDMs may utilize a Zn-finger for recruitment to chromatin. Notably, the Zn-fingers in KDM3 and KDM5 proteins are also required for catalytic activity, likely by inducing a necessary conformational change in the JmjC catalytic center, which is similar to the requirement of the JmjN domain for KDM4 and KDM5 catalytic activity [9,10,31]. Several JMJD proteins are endowed with PHD domains (Figure 1), which can bind to methylated or unmethylated histone sites [4,8]. In case of PHF8, its PHD domain latches onto H3K4me₃ and thereby guides PHF8 to its substrate, H3K9me₂, in close proximity on the same histone tail [32]. Tudor domains, which are present in KDM4A-C, also bind to various methylated histone lysine residues and could steer KDM4A-C to chromatin [4,31]. KDMs may additionally help each other to bind to chromatin through heteromer formation, as exemplified by KDM4A that can recruit KDM4C or KDM3A [31,33]. Lastly, several JMJD proteins contain intrinsically disordered domains that facilitate formation of phase-separated liquid condensates [34,35]. This may augment JMJD proximity to histone substrates if other chromatin-binding proteins or nucleosomes are included in the same condensates.

Culprits for sex bias

Genes encoded on X and Y chromosomes are poised to contribute to the sex bias of diseases. As the human X chromosome carries ~800 and the Y chromosome ~70 protein coding genes, most of the X-linked genes do not have a homolog on the Y chromosome and their double dosage in females could be toxic. To compensate for this, female somatic cells undergo random X chromosome inactivation, leading to one active and one inactive X chromosome. However, a few genes escape X chromosome inactivation, amongst which are KDM5C/SMCX and KDM6A/UTX that each have a homolog on the Y chromosome, namely KDM5D/SMCY and UTY/KDM6C [10,12] (Figure 3A). Hence, both females and males have two active alleles of each of those KDM pairs.

Figure 3. Consequences of mutational inactivation of sex chromosome-encoded JMJD proteins.

(A) Localization of KDM5C (SMCX: SMCY homolog, X-linked), KDM5D (SMCY: Selected mouse cDNA on Y), KDM6A (UTX: ubiquitously transcribed TPR protein, X-linked), UTY (KDM6C) and PHF8 on the human X or Y chromosome. (B) Inactivating germline mutation of one KDM5C allele. In females, this leads to the generation of two populations of somatic cells: one where the mutated KDM5C allele resides on the active X chromosome (Xa), and one where it resides on the inactivated X chromosome (Xi). Since KDM5C escapes X chromosome inactivation, both the Xa- and the Xi-encoded alleles are transcribed. (C) Consequences of pathogenic germline mutations in PHF8, which does not escape X chromosome inactivation (hence, the Xi-encoded PHF8 allele is not transcribed) and has no homolog on the Y chromosome.

If KDM5C and KDM5D were functionally equivalent, a mutation in one KDM5C allele should cause the same phenotype in females and males. This is not the case, since inactivating KDM5C mutations instigate the Claes-Jensen type of X-linked syndromic intellectual developmental disorder in males, while female carriers present at most with mild intellectual impairment (Figure 3B) [36,37]. The latter indicates that a single dose of KDM5C is sometimes somewhat insufficient for normal female development. On the other hand, a higher Kdm5C dosage stimulated adipose tissue expansion induced by high-fat diet in mice and KDM5C expression positively correlated with body mass index in humans, potentially explaining the differences in diet-induced adiposity between the sexes [38]. Another function that is specific for KDM5C and cannot be replaced by KDM5D pertains to the upregulation of Xist, a long non-coding RNA triggering X chromosome inactivation [39]. Finally, consistent with KDM5C being a tumor suppressor and KDM5D not being functionally equivalent [10], mutational inactivation of the sole KDM5C allele in males facilitates more often cancer development compared to loss of one allele in females, explaining for instance the ~10-fold higher number of KDM5C mutation carriers in male versus female clear cell renal cell carcinoma patients [40].

Similarly, UTY is incapable of replacing KDM6A, in part because UTY may have an intrinsically lower catalytic activity. Accordingly, pathogenic KDM6A mutations generally elicit in males a more severe manifestation of Kabuki syndrome type 2, a neurodevelopmental disorder characterized by intellectual disability, dwarfism and facial abnormalities [41]. Likewise, frequent pathogenic mutations of the tumor suppressor KDM6A contribute to the 3-fold higher preponderance of bladder cancer in males, and a mouse model demonstrated that the demethylase function of KDM6A was required to protect against bladder cancer development [40,42]. Furthermore, the higher KDM6A dosage in females seems to be responsible for a higher effector function of natural killer cells, which can incapacitate viruses, thus possibly accounting for the lower susceptibility of females to viral infections [43]. This may also be due to the fact that females will generally display higher expression of KDM6A, because estrogen induces KDM6A transcription [44].

The Y chromosomal KDM5D and UTY genes are not innocent bystanders. For instance, mutations in the KRAS oncogene, which are commonly found in sporadic colorectal cancer, induce the transcriptional upregulation of KDM5D, but not KDM5C, and higher KDM5D activity associates with increased metastasis and lethality in male colon cancer patients [45]. Y chromosome loss occurs in ~10–40% of bladder cancers, which would simultaneously delete KDM5D and UTY, and decreased KDM5D and UTY expression has been correlated with a poor prognosis. While individual ablation of KDM5D or UTY did not affect in vitro proliferation of bladder cancer cells, this stimulated in vivo tumor growth but only when the host was immunocompetent, because KDM5D or UTY loss created an immunosuppressive microenvironment that allowed tumor cells to evade immune surveillance [46]. Since the Y chromosome can be lost or encoded genes extremely downregulated with age, this may also partially account for the increased incidence of bladder cancer in elder men. Further, UTY repressed production of the proinflammatory CXCL9 and CXCL10 chemokines and thereby protected male mice from idiopathic pulmonary arterial hypertension, which is >4-fold more frequent in human females than males [47].

One more human JMJD protein is encoded on the X chromosome, but has no homolog on the Y chromosome and does not escape X inactivation: PHF8 [48]. Loss-of-function PHF8 mutations cause Siderius X-linked intellectual disability syndrome in males, while female carriers display no symptoms [49] despite the fact that half of somatic cells will have no functional PHF8 (Figure 3C). This suggests that PHF8 activity in a fraction of somatic cells is sufficient for normal development. Since PHF8 does not encode for a tumor suppressor, loss-of-function mutations have not been associated with increased cancer incidence in males in contrast to KDM5C and KDM6A.

Even when JMJD proteins are not encoded on sex chromosomes, they can be regulated in a sex-specific manner and thereby contribute to differences between genders. For instance, urinary levels of 2-oxoglutarate are about two-fold higher in males than females [50], suggesting that this differential cofactor concentration might generically account for sex-dependent divergent JMJD activities, yet with currently unknown consequences. Another example pertains to KDM4B and its different expression in males versus females, which sex-dependently affected the H3K9me₂ mark and thereby likely the speed of recovery – which may be reflective of post-stroke outcomes – from cerebral ischemia induced by internal carotid artery occlusion [51]. In conclusion, JMJD proteins, whether or not encoded on sex chromosomes, are likely important for the sex bias of many diseases, but could also be critical for sex-specific differences in development and homeostasis.

From epigenetics to genetics

The human genome is littered with retrotransposons. Normally, they are included within heterochromatin to preclude genomic instability that can lead to permanent genetic changes induced by excision and insertion of activated retrotransposons. H3K9me₃ is key to heterochromatin formation. Overactivity of KDM4A-C, which can remove this repressive mark, was accordingly shown to promote LINE-1 retrotransposition [52,53]. In contrast, KDM5B repressed retroelements independent of catalytic activity through recruitment of SETDB1, a H3K9 methyltransferase whose activity silenced retroelements [11]. Further, inactivation of KDM5C in clear cell renal cell carcinoma was associated with increased genomic rearrangements and disease aggressiveness [54], while biallelic germline mutations in KDM8 compromising catalytic activity led to developmental abnormalities, possibly as a consequence of increased DNA replication stress causing genomic instability [55]. Depletion of KDM3B with concurrent increase of H3K9me_2/1 marks led to transient site-specific extrachromosomal copy gains in mixed-lineage leukemia. When KDM3B activity was restored after a short time, these extrachromosomal copy gains vanished, but longer lasting KDM3B inactivation led to the insertion of those extrachromosomal DNA pieces into the genome [56]. Altogether, altering the activity of certain KDMs may cause heritable changes in the genome, which can underlie disease development but also provide genetic variants for continued evolution.

A potpourri of non-epigenetic functions

Although JMJD proteins have drawn their fame from their epigenetic activities, it has become clear that they also perform a host of non-epigenetic functions. For instance, in addition to histones, several JMJD proteins demethylate and thereby regulate transcription factors such as SRF and STAT3 [57,58]. Further, JMJD proteins can control transcription independent of catalytic activity [59]. This may include the interaction with and thereby modulation of transcription factors, such as inhibition of the tumor suppressor TP53 by KDM4D [60] or the promotion of FOS and TCF degradation by the F-box containing KDM2 proteins [61,62].

Another important means to control gene expression that is regulated by several JMJD proteins is RNA splicing, which occurs concurrent with gene transcription. Splice variants may drastically differ in their function and JMJD6 was shown to affect the alternative splicing of two glutaminase isoforms that have distinct activities and may contribute to the glutamine addiction of cancer cells [63]. Mechanistically, JMJD6 can affect splicing by modifying substrates, as exemplified by the lysyl hydroxylation of the U2AF65 splicing factor, and also in catalytic activity-independent ways [64]. Similarly, catalytic activity was not required for KDM3A to promote RNA binding of the splicing factor SRSF3 that impacted cell-cycle regulation [65]. Moreover, both KDM3A and JMJD6 promoted the generation of the V7 splice variant of the androgen receptor that can cause castration-resistant prostate cancer [66,67]. In the same vein, androgen deprivation led to the phosphorylation of KDM4B by protein kinase A, causing KDM4B binding to SRSF3 and thereby favoring the expression of the V7 splice variant [68]. This may in part explain why androgen deprivation, a mainstay of metastatic prostate cancer therapy, nearly always fails in the long term [69].

Many JMJD proteins are partially or predominantly localized to the cytoplasm where they shape various processes. While KDM4A is mostly residing within the cell nucleus, a substantial amount is also present in the cytoplasm [31] where KDM4A stimulates translation initiation, while JMJD4-mediated hydroxylation of the eukaryotic release factor 1 does so for translational termination [70,71]. The cytoplasmic JMJD4 protein may also demethylate and thereby inactivate RIG-I, a promoter of cholesterol synthesis, and in that way reduce diet-induced hepatic steatosis and eventually hepatocarcinogenesis [72]. Similarly, JMJD8 may demethylate and thereby suppress activity of the cytoplasmic AKT1 kinase [73]. However, this conflicts with the reported localization of JMJD8 to the endoplasmic reticulum, with a substantial portion of its JmjC domain buried within the membrane of this organelle. There, JMJD8 interacted with STING, which precluded formation of a STING-TBK1 complex, thus suppressing the STING immune response that normally contains breast tumor growth [74]. Another non-nuclear function is the association of KDM8 with microtubules during cell division. And lack of KDM8 caused flawed spindle assembly leading to mitotic defects [75]. In sum, the small selection of examples provided in this chapter shows that JMJD proteins influence cells in various ways other than through chromatin regulation. It may be that non-epigenetic JMJD functions, both inside and outside of the cell nucleus and either dependent or independent of catalytic activity, are far more prevalent than currently appreciated.

Sensing metabolism

The dependency of JMJD-mediated hydroxylation/demethylation on 2-oxoglutarate, oxygen, and Fe(II) ties JMJD activity to intracellular levels of these three agents, all of which are intimately regulated by the metabolic state of a cell. 2-Oxoglutarate is a metabolite of the Krebs cycle, which is fueled by acetyl coenzyme A production upon glycolysis or fatty acid degradation and upon glutaminolysis (Figure 4). Accordingly, increased catabolism may enhance 2-oxoglutarate levels and thereby enzymatic activities of JMJD proteins. On the other hand, accumulation of the Krebs cycle components fumarate and succinate, which can be caused by loss-of-function mutations in fumarate hydratase and succinate dehydrogenase in various cancers, should inhibit JMJD activity, since these two oncometabolites compete with 2-oxoglutarate for access to the catalytic center. And the same pertains to the oncometabolite 2-hydroxyglutarate, which is produced through side activities of lactate and malate dehydrogenase or upon neomorphic isocitrate dehydrogenase mutations [7,9,31]. One example of how oncometabolite-induced decrease of JMJD activity impacts cells is their inhibition of KDM4B that led to aberrant H3K9me₃ around DNA breaks suppressing their repair [76].

Figure 4. Regulation of JMJD catalytic activity through small molecules and metals.

The oncometabolites fumarate, succinate and R-/S-2-hydroxyglutarate all inhibit JMJD catalytic activity by competing with 2-oxoglutarate, whereas heavy metals may do so by competing with Fe²⁺. While LDH and MDH produce S-2-hydroxyglutarate, mutated IDH generates the respective R enantiomer. FH, fumarate hydratase; IDH, isocitrate dehydrogenase; LDH, lactate dehydrogenase; MDH, malate dehydrogenase; ROS, reactive oxygen species; SDH, succinate dehydrogenase.

Many JMJD proteins have a high affinity for oxygen and are therefore insensitive to physiological oxygen level changes such as hypoxia, which is particularly prevalent in solid tumors [7]. However, some JMJD proteins have a low affinity for oxygen and are thus sensitive to hypoxia. For instance, hypoxia inhibited KDM3A and thereby prevented lysine demethylation of the transcriptional coactivator PGC-1α leading to reduced mitochondrial biogenesis and tumor cell apoptosis [77]. Inflammation is another hypoxia-associated process that can trigger tissue injury. In T cells, hypoxia caused more H3K27me₃ via inhibition of KDM6A, which diminished T cell activity and thereby potentially the severity of inflammatory bowel disease [78]. And hypoxia also reduced KDM6A activity in murine myoblasts, which again enhanced H3K27 methylation and thereby blocked differentiation into myotubes [79].

Iron is an essential element and Fe(II) is needed for most catalytic activities of JMJD proteins. Thus, iron deficiency reduces JMJD activity, which can impair B cell-mediated immune responses (KDM2B, KDM3B, KDM4C), adipocyte differentiation (KDM3A), anabolic metabolism (KDM3B) or preeclampsia-related fibronectin homeostasis (JMJD6) [80–83]. Likewise, oxidation of Fe(II) by reactive oxygen species will impair JMJD function (Figure 4), which would explain why reducing agents like ascorbate (vitamin C), glutathione or cysteine stimulate JMJD activity [6,7,12,53]. Lastly, heavy metals such as Cd²⁺ or Ni²⁺ can compete with Fe²⁺ and thereby inhibit JMJD proteins, which may possibly underlie the manifestation of heavy metal-induced pathologies [9,84]. Altogether, JMJD proteins can sense metabolic changes in various ways and help to initiate appropriate cellular responses. Thus, JMJD proteins are crucial for cellular homeostasis and accordingly their dysregulation may have serious consequences, as highlighted by the development of obesity and metabolic syndrome upon knockout of Kdm3a or Kdm4b [85,86].

Concluding remarks

JMJD proteins are jack-of-all-trades that influence a multitude of cellular processes required for development and a healthy life (Figure 5). But when their activity becomes compromised, many diseases can arise, pointing out their utility as drug targets. Arguably most attention has been directed towards KDM4 proteins and their ability to promote multiple cancers, and thus various inhibitors have been designed to inhibit KDM4 catalytic activity or disable their PHD and Tudor domains that can read epigenetic marks [31,87]. In 2023, the orally available pan-KDM4 inhibitor TACH101 [88] has entered a Phase 1a/1b open-label study for patients with advanced or metastatic solid tumors (NCT05076552). Yet, no other JMJD inhibitor has hitherto entered a clinical trial. In part, this may be due to the non-specificity of current inhibitors that can lead to an abundance of side effects, redundancy within the JMJD family or their non-catalytic functions (see Outstanding questions). A solution to the latter could be the development of a PROTAC that will cause a particular targeted JMJD protein to become proteolytically degraded.

Figure 5. Biological processes affected by JMJD proteins.

A selection of known JMJD functions as described in this review; the center shows the catalytic core of KDM4A (amino acids 171–293) and was downloaded from the Protein Data Bank (accession number 2GP5). This figure was created with BioRender.com.

Outstanding Questions.

• What are the precise catalytic activities of each JMJD protein? Will any of them be really tyrosine kinases or (histone) arginine demethylases under physiological conditions? And which functions of JMJD proteins do not require catalytic activity?

• Is each member of a particular JMJD subfamily tasked with unique functions, or is there redundancy amongst subfamily members? Does this redundancy extend to other JMJD subfamilies with the same catalytic activity (e.g., both KDM3 and KDM7 proteins demethylate H3K9me_2/1)?

• How does JMJD enzymatic activity change the molecular properties of known and to-be-identified substrates, and what consequences would this have at the cellular and organismal level?

• How is JMJD gene expression controlled or JMJD protein activity modulated through posttranslational modifications, and what physiological relevance would that have?

• What are additional normal and pathological processes being critically influenced by JMJD proteins?

• Can specific inhibitors, or activators, of selected JMJD proteins be developed to ameliorate or even cure diseases?

Inhibition of KDM4 and/or KDM5 enzymatic activity may also be beneficial in the treatment of dilated and hypertrophic cardiomyopathy or cardiac fibrosis [87,89,90]. And likewise, inhibition of KDM3A and KDM3C may ameliorate heart hypertrophy and fibrosis [9]. JMJD proteins can also act as crucial host cell factors, as shown for KDM6B upon mycobacterial or salmonella infection or for KDM4D upon exposure to hepatitis B virus [91–93], suggesting that targeting human JMJD proteins can help fight infections that are major causes of morbidity and mortality. And since bacteria and fungi do possess JMJD proteins [2,13], inhibition of pathogen-encoded JMJD proteins could also be pursued to contain infections. As a proof-of-principle, JIB-04, a KDM4/5 inhibitor that has frequently been used to suppress cancer cell growth in the laboratory [87], was effective as an anti-Cryptococcus agent [94]. It is obvious that inhibition of JMJD proteins potentially could have a huge impact on the treatment of cancer, heart disease, infections and possibly many other illnesses. As such, there is an urgency to develop respective inhibitors and thoroughly assess them in clinical trials.

Another unsolved question pertains to the breadth of catalytic activities of JMJD proteins. Or what impact, if any, does a JMJD-mediated modification impose on a target protein? Similarly, JMJD proteins are themselves modified by e.g. phosphorylation, yet how this modulates their function is largely unclear [95]. More animal models are needed to decipher the exact physiological roles of each JMJD protein. Notably, JMJD catalytic activity likely decreases upon aging when 2-oxoglutarate levels decline by up to 10-fold with potentially serious consequences for health- and lifespan [96]; accordingly 2-oxoglutarate counteracted osteoporosis in aging mice by decreasing H3K9me₃/H3K27me₃ [97], two epigenetic marks that can be erased by several JMJD proteins. Activating certain JMJD proteins may thus be a fountain of youth, and consistently PHF8, KDM6B or UTX overexpression extended lifespan in Caenorhabditis elegans [98,99]. Albeit likely short of such miraculous exploits, JMJD proteins are nevertheless poised to surprise us with many more amazing feats.

Highlights.

• By demethylating or clipping histones, JMJD proteins induce chromatin changes that lie beneath the epigenetic control of gene expression.

• Through a wide repertoire of catalytic (hydroxylation, demethylation, proteolytic cleavage) and non-catalytic activities affecting non-histone proteins, JMJD proteins shape various cellular processes beyond epigenetics.

• Several oncometabolites, oxygen tension, reactive oxygen species and heavy metals all affect JMJD catalytic activity, which lets JMJD proteins sense metabolic changes.

• Pathogenic JMJD mutations or dysregulated JMJD activity underlie various diseases, including cancer and neurodevelopmental defects, and may also account for sex-dependent differences in disease prevalence and manifestation.

• Inhibiting or activating specific JMJD proteins may be efficacious in clinical therapy.

Glossary

7SK snRNP

a small nuclear ribonucleoprotein complex (snRNP) that encompasses the 7SK small nuclear RNA and several proteins, including P-TEFb. The latter is a protein kinase, which stimulates the elongation phase of RNA polymerase II-mediated transcription, but is inactive when sequestered within the 7SK snRNP.

Ankyrin repeat

a 33 amino acid-long motif that is repeated several times thereby creating solenoid structures. Establishes protein-protein interactions in many proteins, including Ankyrin that contains 24 ankyrin repeats.

ARID domain

AT-rich interaction domain that was originally found to bind to adenine/thymine-rich DNA in a way considered not to be sequence-specific. However, there are also ARID domains (as in KDM5 enzymes) that bind to stretches of DNA not being rich in adenine/thymine.

F-box

a structural motif consisting of ~50 amino acids that mediates interactions between proteins. Often found in proteins that confer specific substrate recognition on ubiquitin ligase complexes and thereby facilitate protein degradation.

Histone code

posttranslational modifications (e.g., methylation, acetylation, phosphorylation and ubiquitylation) of histone proteins that facilitate the recruitment of “readers” (e.g., proteins with PHD or Tudor domains) to chromatin in order to especially regulate gene transcription. “Writers” (e.g., methyl transferases) add these chemical marks, while “erasers” such as KDMs remove them.

LINE-1 (long interspersed nuclear element 1)

transposable elements that comprise ~15% of the human genome. They encode for factors required for transposition, including reverse transcriptase. Most LINE-1 elements are inactive and the activity of the others is normally repressed, but when activated these elements can cause genetic changes such as insertions or deletions in chromosomes.

Nucleosome

a structure consisting of two copies of each histones H2A, H2B, H3 and H4, around which ~147 bp of DNA are wrapped in ~1³/₄ turns. Nucleosomes are key to compacting DNA within the eukaryotic cell nucleus and can arrange into higher order structures. In particular N- and C-termini of histones stick out from the nucleosome core like tails.

Oncometabolite

a metabolite whose concentration is highly elevated in cancer cells. This can be due to gain-of-function or loss-of-function mutations in various metabolic enzymes.

PHD domain

originally found in and therefore named after plant homeodomain proteins. Through conserved cysteine and histidine residues, a PHD domain complexes Zn²⁺ and represents one subtype of Zn-fingers. It can function as an epigenetic reader by facilitating binding to (un)methylated histone lysine residues.

PROTAC

a proteolysis targeting chimera, which consists of two active domains chemically coupled to each other by a linker. One domain binds to a target protein, while the other one facilitates the recruitment of a ubiquitin ligase. In consequence, this leads to the degradation of the target protein through the 26S proteasome.

STING (stimulator of interferon genes)

endoplasmic reticulum protein, which is stimulated by 2’,3’-cyclic GMP-AMP that is produced in response to cytoplasmic DNA, as can occur upon viral infection or in autoimmune diseases. STING initiates an innate immune response that leads to upregulation of interferons and chemokines.

Tudor domain

characterized by a conserved structural fold first found in the Drosophila melanogaster TUDOR protein. It facilitates binding to methylated lysine and arginine residues especially in histones and is therefore considered to be an epigenetic reader.

X chromosome inactivation

random, heritable epigenetic event early in embryogenesis that silences either the maternal or paternal X chromosome in female somatic cells. A form of dosage compensation for X-linked genes in the female sex.