Caught in the act – protein adaptation and the expanding roles of the PACS proteins in tissue homeostasis and disease

At the turn of the 20th century, Archibald Garrod, Walter Sutton and Thomas Hunt Morgan ushered in a new era of biological research by replacing observational studies with mechanistic analyses (Kelves and Hood, 1992). Their discoveries that chromosomes encode heritable traits set the foundation for understanding genetic inheritance. More than 30 years later, through inactivation and mutational studies of the common bread mold Neurospora, George Beadle and Edward Tatum ultimately determined that genes encode enzymes and laid a cornerstone of modern molecular biology through the ‘one gene – one protein’ hypothesis (Beadle and Tatum, 1941). This provocative model inevitably collapsed under the weight of subsequent discoveries, which revealed that genes frequently encode multiple proteins or collections of peptides that drive complex physiology and contribute to molecular promiscuity through mechanisms ranging from alternative RNA splicing to the proteolytic processing of polyproteins (‘one gene – many proteins’; Yang et al., 2016). The essence of these studies led to the present-day ‘one protein – multiple functions’ mantra, which recognizes that individual proteins often have multiple and seemingly disparate functions – a phenomenon known as ‘moonlighting’ (Jeffery, 1999; Henderson and Martin, 2014). The glycolytic enzymes glyceraldehyde-3-phosphate dehydrogenase (GAPDH) and enolase are widely cited examples of moonlighting proteins. Human GAPDH participates in as many as 20 different cellular functions ranging from the regulation of membrane traffic to the maintenance of genome integrity (Sirover, 2011). Human enolase can be present on the cell surface of activated monocytes where it serves as a plasminogen receptor that promotes recruitment of inflammatory cells to sites of tissue injury (Wygrecka et al., 2009).

Protein moonlighting, which by definition is limited to proteins that use a single primary sequence in the absence of posttranslational modifications to mediate multiple functions (Jeffery, 1999), is inherently limited for solving evolutionary challenges. Eventually, moonlighting gives way to gene duplication, which enables the resulting paralogs to acquire mutations that expand the molecular functions initiated by the ancestral protein. Inevitably, such a hit-and-miss approach to mutational diversification will hit a roadblock in the form of rigidly folded domains, such as SH2 or PDZ domains found in many proteins (Tompa et al., 2014). The dedicated structure of these ∼100-amino-acid protein domains, which are typically involved in high-affinity binding to discrete ligands, limits amino acid substitutions, thereby reducing their ability to increase the functional density of associated cellular proteins. Therefore, protein adaptation represents an additional and vital solution to increase the array of protein functions to an extent that is not possible by moonlighting and gene duplication alone (Tompa et al., 2014). Rather than coaxing rigid protein domains to accept new roles, such proteins – notably vertebrate proteins – have acquired intrinsically disordered regions (IDRs) to complement the limited utility of folded domains (Dunker et al., 2005; Pancsa and Tompa, 2016). IDRs do not fold into an autonomous structure, but contain flexible 5–15-amino-acid small linear motifs (SLiMs) that can assume an ensemble of ‘structures’, which in turn interact with a panoply of binding partners through low-affinity but high-specificity interactions (Davey et al., 2015). SLiMs are frequently sites of posttranslational modifications and undergo rapid evolution, greatly increasing the functional density of IDR-containing proteins (Tompa et al., 2014; Pancsa and Tompa, 2016). This plasticity enabled SLiMs from disparate proteins across a large taxonomic range to acquire similar amino acid mutations by convergent evolution so that they may bind to a common globular protein domain (Schlessinger et al., 2011; Davey et al., 2012, 2015; Van Roey et al., 2012; Jonas and Izaurralde, 2013). However, examples tracing how a single protein adapted to evolutionary pressure by acquiring specific SLiM mutations in order to fulfill new functions that are critical for vertebrate homeostasis are surprisingly limited. Studies on the phosphofurin acidic cluster sorting (PACS) proteins provide a powerful illustration of such evolutionary protein adaptations ‘caught in the act’ of connecting the increasingly complex organellar landscape that underpins vertebrate cell function.

Here, we discuss how PACS proteins first appeared in metazoans as membrane traffic regulators, and then expanded their functional repertoire in vertebrates by acquiring molecular switches, notably phosphorylation sites and nuclear localization signals (NLSs) within their SLiMs. These gain-of-function adaptations enabled the vertebrate PACS proteins to integrate cytoplasmic functions with nuclear gene expression. In particular, we focus on the acquisition of an Akt phosphorylation site and NLS in vertebrate PACS-2, which enable this protein to function as a metabolic switch that integrates cytoplasmic membrane traffic and interorganellar communication with nuclear gene expression in response to anabolic or catabolic cues.

The PACS proteins were discovered in a genetic screen for regulators of the secretory pathway proteinase furin (Wan et al., 1998; Thomas, 2002). Human PACS1 is located at 11q13.1-q13.2 and contains 24 exons and at least 12 alternatively spliced variants. Human PACS-2, which is located near the telomere at 14q32.33, also contains 24 exons and at least 11 alternatively spliced variants. Genome-wide association studies (GWAS) identified the human PACS1 locus as a susceptibility gene in severe early-onset obesity (Wheeler et al., 2013) and developmental delay (Deciphering Developmental Disorders Study, 2017), and mutations in PACS1 underlie ‘PACS-1 syndrome’ (Schuurs-Hoeijmakers et al., 2012, 2016; Stern et al., 2017), characterized by epileptic seizures, heart defects, hypotonia and craniofacial malformations. The PACS2 locus is highly susceptible to loss of heterozygosity in colorectal cancer, leading to a reduction or loss of PACS-2 protein expression (Anderson et al., 2001; Aslan et al., 2009). Similar to PACS1, mutations in PACS2 are associated with neonatal onset epilepsy, global developmental delay and intellectual disability (C. Thauvin and H. Olson, personal communication).

The canonical 963-amino-acid PACS-1 and 889-amino-acid PACS-2 proteins share 54% sequence identity. Bioinformatics analyses suggest both proteins lack folded globular domains and are instead nearly 50% disordered (Fig. 1). The ∼150-amino-acid cargo (furin)-binding regions (FBRs) in PACS-1 and PACS-2, which are predicted to be structured (Fig. 1), share nearly 80% sequence identity and bind client proteins, as well as the cytoplasmic membrane trafficking machinery as described below. PACS-1 and PACS-2 also share a disordered middle region (MR), which contains an autoregulatory domain as well as NLSs and, specifically in PACS-2, a critical Akt phosphorylation site that binds to 14-3-3 proteins. The function of the shared C-terminal region (CTR) is unknown. PACS-1 additionally contains an N-terminal extension called the atrophin-1-related region (ARR), which has homology to this nuclear transcriptional co-repressor (Zhang et al., 2006).

The PACS genes are a recent addition to the eukaryotic genome, appearing first in lower metazoans (Fig. 2). Invertebrates, including nematodes, arthropods and echinoderms, possess a single PACS locus that is apparently dedicated to membrane trafficking. Notably, Caernorhabditis elegans PACS (cePACS, T18H9.7a) localizes to early endosomes at the presynaptic terminus of the neuromuscular junction where it mediates synaptic transmission (Sieburth et al., 2005). Lower chordates, including Amphioxus and tunicates also possess only a single gene. The PACS gene was duplicated with the appearance of the vertebrates, resulting in the PACS-1 and PACS-2 genes.

The regulation of membrane traffic remains a role that is conserved among the vertebrate PACS proteins. The PACS-1 FBR binds to acidic clusters that can be phosphorylated by casein kinase 2 (CK2), as well as α-helices on a large number of client proteins (Wan et al., 1998; Youker et al., 2009; Dikeakos et al., 2012). Thus, PACS-1, which interacts with the clathrin adaptors AP-1 and AP-3, as well as the monomeric adaptor GGA3, mediates localization of furin and other client proteins to the trans-Golgi network (TGN) and also targets a subset of client proteins to the primary cilium, including the adaptor protein nephrocystin (also known as NPHP1) and the olfactory cyclic-nucleotide-gated ion channel (CNG), the latter by binding to the β1 subunit (CNGB1) (Fig. 3A) (Wan et al., 1998; Schermer et al., 2005; Jenkins et al., 2009; Youker et al., 2009). PACS-1 acquired a CK2-phosphorylated acidic cluster of its own, which is located in the disordered MR (see Fig. 1). CK2 phosphorylation of Ser²⁷⁸ in the PACS-1 autoregulatory domain controls intramolecular binding to the FBR, which regulates the interaction with client proteins (Scott et al., 2003). PACS-2, which interacts with the coatomer COPI, mediates the localization of cargo to the endoplasmic reticulum (ER) and, similar to cePACS, directs trafficking from early endosomes (Youker et al., 2009) (Fig. 3A, see also Fig. 4A). PACS-2 stabilizes a pool of the metalloproteinase ADAM17 (also known as TACE) on early endosomes, from where the enzyme is trafficked to the cell surface to shed ErbB ligands, including those that ligate the epidermal growth factor receptor (EGFR) (Dombernowsky et al., 2015). In the absence of PACS-2, ADAM17 is diverted to lysosomes, which reduces ErbB shedding. Correspondingly, EGFR signaling is reduced in the intestinal epithelium of PACS2^−/− mice (Dombernowsky et al., 2015). The role of the candidate autoregulatory domain in the PACS-2 MR has not been established.

Several pathogenic viruses acquired furin-like acidic clusters to exploit the endosomal sorting steps mediated by PACS-1 and PACS-2. This viral mimicry (see Davey et al., 2011) enables the viruses to assemble progeny, escape immune surveillance and prevent apoptosis. For example, PACS-1 binds to acidic clusters in the human cytomegalovirus (HCMV) envelope glycoprotein gB and the Epstein–Barr virus (EBV) tegument protein BBLF1 and localizes them to the TGN to support virus assembly (Crump et al., 2003; Chiu et al., 2012) (see Fig. 3A). Furthermore, PACS-2 interacts with a pair of small acidic clusters in the ubiquitin ligase K5 of Kaposi sarcoma herpesvirus (KSHV) at the ER. This interaction enables KSHV to downregulate the cell adhesion molecule CD31 (also known as PECAM1), which may contribute to KSHV-induced cancer (Mansouri et al., 2006). The HIV-1 accessory protein Nef uses a bipartite motif composed of a short acidic cluster and the αB helix to interact with both PACS-1 and PACS-2 (Piguet et al., 2000; Atkins et al., 2008; Dikeakos et al., 2012). This bipartite binding enables HIV-1 Nef to downregulate major histocompatibility complex class I (MHC-I) from the cell surface, which allows the virus to escape immune detection (Pawlak and Dikeakos, 2015) (Fig. 3B). To this end, Nef interacts with PACS-2 on early endosomes, which enables the HIV-1 protein to assemble a multi-kinase complex consisting of a Src family kinase (SFK), ζ-chain-associated protein kinase 70 (ZAP-70) and a class I phosphoinositide-3 kinase (PI3K) (Blagoveshchenskaya et al., 2002; Hung et al., 2007; Atkins et al., 2008). The activated multi-kinase complex increases the amount of phosphatidylinositol (3,4,5)-trisphosphate (PIP₃) underneath the plasma membrane, which recruits an ARF6 GEF to activate ARF6 and accelerate endocytosis of cell-surface MHC-I (Blagoveshchenskaya et al., 2002). Nef then connects the internalized MHC-I molecules to PACS-1 and AP-1 on an endosomal compartment to prevent their recycling to the cell surface and instead sequester them in the TGN, thereby protecting the virus from immune surveillance (Blagoveshchenskaya et al., 2002; Chaudhry et al., 2008; Noviello et al., 2008; Dikeakos et al., 2010; Dirk et al., 2016). Unlike KSHV and many other viruses, HIV-1 Nef does not induce degradation of downregulated MHC-I (Blagoveshchenskaya et al., 2002; Dikeakos et al., 2010; Dirk et al., 2016). This chink in the armor of HIV-1 provides an alternative approach to combat the virus by reversing the Nef-mediated immune evasion pathway. In support of this possibility, treatment of HIV-1-infected primary CD4⁺ T-cells with small-molecule inhibitors of the multi-kinase complex restores cell surface expression of MHC-I and sensitizes them to killing by CD8⁺ T-cells (Hung et al., 2007; Dikeakos et al., 2010; and M. Ostrowski, personal communication).

Surprisingly, while PACS-2 mediates trafficking of many pro-survival signaling molecules, it is also tasked with mediating death-ligand-induced apoptosis induced by TNF-related apoptosis-inducing ligand (TRAIL, also known as TNFSF10). This clinically important death ligand is an in vivo metastasis inhibitor that selectively kills diseased cells, including cancer cells and virally infected cells (Johnstone et al., 2008). TRAIL-induced cell death requires the coordinated permeabilization of multiple organelles, including mitochondria and lysosomes (Aslan and Thomas, 2009). TRAIL triggers mitochondrial outer membrane permeabilization (MOMP) by inducing dephosphorylation and proteolytic cleavage of the pro-apoptotic Bcl-2 protein Bid to form truncated Bid (tBid), which drives Bax-dependent MOMP and the consequent activation of executioner caspases (Aslan et al., 2009).

PACS-2 has essential roles in TRAIL-induced MOMP (Fig. 4B). In response to TRAIL, PACS-2 binds full-length dephosphorylated Bid and traffics it to mitochondria where it can be converted into tBid (Simmen et al., 2005). In parallel, TRAIL triggers assembly of a complex that contains PACS-2, Bim (also known as BCL2L11) and Bax, called the PIXosome, on lysosomal membranes (Werneburg et al., 2012). The PIXosome drives lysosomal membrane permeabilization (LMP), which releases cathepsin B and other hydrolases that are required for MOMP into the cytosol (Boya and Kroemer, 2008). Several mechanisms interfere with these proapoptotic functions of PACS-2. In hepatocellular carcinoma cells, inhibitors of apoptosis (IAPs) target PACS-2 for proteasomal degradation, thereby protecting the cancer cells from being killed by TRAIL (Guicciardi et al., 2014). In cardiomyocytes, miR499 prevents Bid-dependent apoptosis by downregulating PACS-2 (Wang et al., 2014). Not surprisingly, herpesviruses also exploit the ability of PACS-2 to traffic proteins to mitochondria. The HCMV protein viral inhibitor of apoptosis (vMIA) prevents cell death by trapping pro-apoptotic Bax on mitochondria (Arnoult et al., 2004). vMIA interacts with PACS-2 but not PACS-1 (our unpublished data), and the vMIA–PACS-2 interaction is required for efficient translocation of vMIA to mitochondria (Salka et al., 2017). Similar to the fate of Bax, vMIA traps PACS-2 on mitochondria. Notably, this sequestration of PACS-2 is coupled with a blunted translocation of Bid to mitochondria in response to death ligands, suggesting vMIA may trap both inactivated Bax and PACS-2 on mitochondria to inhibit apoptosis in HCMV-infected cells (Fig. 5A; L.T., T.S., J.A. and G.T., unpublished results). The PACS-2-dependent recruitment of Bid to mitochondria prior to formation of tBid suggests this translocation step may be a checkpoint for apoptotic regulation. In support of this possibility, PACS-1 also binds dephosphorylated Bid, but prevents translocation of Bid to mitochondria (T.S., L.T., J.A. and G.T., unpublished results).

Mitochondria-associated membranes (MAMs) are ER–mitochondria contacts, which were discovered as lipogeneic platforms in liver that are responsive to feeding and starvation (Bernhard et al., 1952; Vance, 1990). Subsequent studies further revealed that MAMs are a dynamic communication center that regulates Ca²⁺ signaling, lipid metabolism and autophagy (Raturi and Simmen, 2013; Sood et al., 2014; Theurey et al., 2016). Mammalian MAMs are formed by protein tethers, similar to the tetrameric tethering complexes in fungi known as ER–mitochondria encounter structures (ERMES) (Kornmann et al., 2009). However, of the more than 30 proteins involved in vertebrate MAM function, only two of them, GRAMD1A, which corresponds to yeast Lam6p, and MIRO (also known as RHOT1), which corresponds to yeast Gem1p, are conserved in ERMES (Kornmann et al., 2011; Elbaz-Alon et al., 2015). This increased complexity of animal MAMs suggests they acquired new roles beyond those controlled by ERMES (Herrera-Cruz and Simmen, 2017). This increased function correlates with the emerging realization that disturbances in MAM integrity are associated with diseases ranging from obesity to neurodegenerative disorders (Arruda and Hotamisligil, 2015; Paillusson et al., 2016).

Knockdown or knockout of PACS-2 detaches the ER from mitochondria and interferes with a number of key MAM functions (Simmen et al., 2005) (see Fig. 4C). For example, PACS-2 is required for starvation-induced autophagy because it promotes the recruitment of the early autophagy marker Atg14 to MAMs (Hamasaki et al., 2013). PACS-2 knockdown also blocks Ca²⁺-mediated apoptosis progression, which suggests that PACS-2 is required for efficient Ca²⁺ transfer between the ER and mitochondria (Simmen et al., 2005). This PACS-2 function may be coupled to its reported dynamic roles in the trafficking of calnexin and perhaps other ER Ca²⁺ regulatory proteins that modulate ER–mitochondria Ca²⁺ flux (Myhill et al., 2008). Mechanistically, PACS-2 may modulate MAM formation by preventing cleavage of the ER–mitochondria tethering protein BAP31 (also known as BCAP31) (Iwasawa et al., 2011). Disturbances in these PACS-2 functions could contribute to disease. For example, in obese mice challenged by a high-fat diet, PACS-2 is responsible for a chronic increase in MAM formation (Arruda et al., 2014), leading to toxic mitochondrial Ca²⁺ overload that consequently impairs mitochondrial oxidative capacity, exacerbates insulin resistance and disrupts glucose homeostasis (Arruda et al., 2014). Conversely, PACS-2 knockdown protects liver from overnutrition-induced steatosis and optimizes mitochondrial respiration and insulin sensitivity (Arruda et al., 2014). Aberrantly elevated PACS-2 and MAM formation are also found in neurons from Alzheimer's patients and Alzheimer's mouse models, suggesting that the deleterious consequences from abnormally strong MAM formation in this disease may also result from dysregulated PACS-2 expression (Area-Gomez et al., 2012; Hedskog et al., 2013).

The seemingly incongruous roles for PACS-2 in mediating pro-survival protein traffic and MAM function, as well as TRAIL-induced cell death, arose with the acquisition of an Akt phosphorylation site that switches the function of PACS-2 between these divergent roles. Akt phosphorylates PACS-2 on Ser⁴³⁷, which promotes high-affinity binding to 14-3-3 proteins (Aslan et al., 2009). In response to insulin, MAM-localized mTORC2 activates Akt, which phosphorylates PACS-2 to modulate MAM integrity (Betz et al., 2013) (Fig. 4C, top). Akt-mediated phosphorylation of Ser⁴³⁷ is also required for PACS-2-dependent membrane trafficking steps (Fig. 4A) (Aslan et al., 2009). By contrast, dephosphorylation of PACS-2 Ser⁴³⁷, which prevents 14-3-3 binding, blocks pro-survival membrane traffic and disrupts MAMs, but is required for apoptotic trafficking to mitochondria and lysosomes and may support autophagy induction (Fig. 4B,C, bottom) (Aslan et al., 2009; Werneburg et al., 2012; Betz et al., 2013; Hamasaki et al., 2013; and L.T. and G.T., unpublished data).

Akt and 14-3-3 proteins not only repress the apoptotic roles of PACS-2 but also regulate other proapoptotic proteins, including the Bcl-2 protein Bad and FOXO transcription factors (Datta et al., 1997, 2002; Brunet et al., 1999; Singh et al., 2010; Feehan and Shantz, 2016). The regulation of Bad and FOXO proteins by Akt and 14-3-3 is akin to a simple molecular ‘on or off’ switch (Van Roey et al., 2012). However, the regulation of the anabolic (pro-survival) roles of PACS-2 versus the catabolic (apoptotic) roles exerted by Akt and 14-3-3 appears to be more complex. We suggest that the Akt site in PACS-2 resembles a ‘bifurcation’ switch such that both the phosphorylated and dephosphorylated states of PACS-2 are active, but in opposing directions (anabolic versus catabolic pathways, respectively). This model is supported by the ability of 14-3-3 proteins to induce conformational changes in their partners that may enable PACS-2 to selectively bind to client proteins that are involved in anabolic (including pro-survival trafficking) pathways, which would be mediated by binding of 14-3-3 to phosphorylated (p)Ser⁴³⁷, versus catabolic (including autophagic and apoptotic trafficking) pathways, which would be mediated by PACS-2 that is dephosphorylated on Ser⁴³⁷ (Yaffe, 2002).

The extrinsic and intrinsic apoptotic pathways induced by TRAIL or DNA damage, respectively, use different molecular steps to trigger MOMP. Nonetheless, these pathways are intimately coupled, as p53 (also known as TP53) induces expression of the TRAIL receptor DR5 (also known as TNFRSF10B), and chemotherapeutics and TRAIL synergize to kill cancer cells (Sheikh et al., 1998; Ifeadi and Garnett-Benson, 2012). Surprisingly, PACS-2 has markedly different roles in its response to TRAIL compared with DNA damage; PACS2^−/− mice are impaired in TRAIL-induced apoptosis, but are sensitized to DNA-damage-induced apoptosis (Aslan et al., 2009; Barroso-Gonzalez et al., 2016). These opposing roles for PACS-2 are partly explained by the phosphorylation state of Ser⁴³⁷, which is reduced by TRAIL, but increased upon DNA damage (Barroso-Gonzalez et al., 2016). Whereas TRAIL uses dephosphorylated PACS-2 to induce MOMP and LMP, the DNA damage response uses phosphorylated PACS-2 to support cytostasis by coordinating NF-κB and Bcl-xL-dependent anti-apoptosis with p53- and p21 (CDKN1A)-dependent cell cycle arrest (Aslan et al., 2009; Barroso-Gonzalez et al., 2016).

Clues to understanding the roles of PACS-2 in mediating the p53–p21 and NF-κB–Bcl-xL pathways were provided by the DNA repair kinase ATM (Shiloh and Ziv, 2013). In response to DNA damage, nuclear ATM phosphorylates p53, which stabilizes the multi-functional transcription factor to induce its target genes, thus favoring cell cycle arrest and senescence over apoptosis (Xu and Baltimore, 1996; Sperka et al., 2012). Concurrently, a small pool of activated ATM translocates from the nucleus into the cytoplasm where it initiates a novel ‘nucleus-to-cytoplasm’ signaling pathway that promotes the NF-κB-dependent induction of anti-apoptotic Bcl-xL through an as-yet-unresolved mechanism (Miyamoto, 2011).

PACS-2 intersects both ATM pathways to coordinate induction of the NF-κB–Bcl-xL and p53–p21 pathways. Following DNA damage, cytoplasmic PACS-2 interacts with ATM, thereby sequestering the kinase in the cytoplasm where it induces TGF-β-activating kinase 1 (TAK1; also known as MAP3K7) to trigger activation of IκB kinase (IκK) (Wu et al., 2010; Barroso-Gonzalez et al., 2016) (Fig. 4D, top). ATM-activated IκK then stimulates the canonical IκBα–NF-κB induction of Bcl-xL. In the absence of PACS-2, ATM fails to accumulate in the cytoplasm, which reduces the induction of Bcl-xL. Consequently, induction of the apoptotic p53–Puma (Puma is also known as BBC3) pathway remains, thereby increasing MOMP and cell death (Barroso-Gonzalez et al., 2016).

To promote p53–p21-dependent cell cycle arrest over p53–Puma-dependent cell death, PACS-2 traffics to the nucleus where it increases the transactivation of p53 bound to the p21 promoter (Atkins et al., 2014) (Fig. 4D, bottom). Notably, in addition to the ATM-mediated phosphorylation of p53 within its N-terminal region, acetylation of critical lysine residues near the C-terminus of p53 is also required for it to promote maximal transcriptional activity of its target genes (Gu and Roeder, 1997). The class III histone deacetylase (HDAC) SIRT1, in turn, represses p53-mediated transcriptional activation by deacetylating p53 following DNA damage (Kruse and Gu, 2009). In response to DNA damage, Akt-phosphorylated PACS-2 enters the nucleus where it binds to and inhibits SIRT1, which protects acetylated p53 bound to the p21 promoter and, consequently, increases transcriptional output of p21 (Atkins et al., 2014; and L.T. and G.T., unpublished results). In the absence of PACS-2, the excessive SIRT1 activity reduces the p53-dependent induction of p21, which impedes cell cycle arrest and sensitizes cells to p53–Puma-dependent apoptosis. To access the nucleus, vertebrate PACS-2 acquired a polybasic nuclear localization signal (NLS) and a Crm1-dependent nuclear export signal (Atkins et al., 2014). Similar to the Ser⁴³⁷ Akt site of PACS-2, its NLS is located within a predicted IDR and is present only in vertebrate PACS proteins (Figs 1 and 2).

The acquisition of the Ser⁴³⁷ Akt site enabled cytoplasmic PACS-2 to switch between its homeostatic (pSer⁴³⁷) and apoptotic (dephosphorylated Ser⁴³⁷) roles (see Fig. 4). PACS-2 proteins in jawed vertebrates share an identical Akt motif nestled in a disordered region that apparently became fixed by positive selection (Fig. 2). Phylogenetic studies suggest this SLiM evolved rapidly, albeit along a circuitous path (Figs 1 and 2). An Akt consensus motif first appeared in lamprey PACS-1-related which coincided with an apparent gene duplication (Fig. 2). This consensus phosphorylation site was disrupted in fish PACS-1 by the acquisition of a proline residue at the +2 position. Surprisingly, in birds, the Akt motif reappeared in both PACS-1 and PACS-2. However, in mammals, the PACS-1 site again acquired a proline residue, this time in place of the phosphorylatable serine residue (Fig. 2). These findings suggest that PACS-2 acquired the Akt site to act as a vital ‘bifurcation’ switch between its anabolic versus catabolic pathways and that evolutionary pressure assigned this switch to mammalian PACS-2 by negatively selecting against the phosphorylation site in mammalian PACS-1 (Figs 2 and 4) (Davey et al., 2015).

Like the Akt site, the PACS-2 NLS, which is also located within a disordered region, is highly conserved across all jawed vertebrates (Figs 1 and 2). Phylogenetic studies suggest that a polybasic site with limited similarity to the vertebrate PACS NLS first appeared in amphioxus and highly homologous NLS sequences were first acquired in fish (Fig. 2). Acquisition of the NLS by jawed vertebrate PACS-2 parallels the evolutionary expansion of roles of p53 in directing apoptosis versus cell cycle arrest. In worms and flies, p53 is dedicated to driving apoptosis (Schumacher et al., 2001). However, vertebrate p53 additionally induces p21-dependent cell cycle arrest (Lu et al., 2009). Consistent with this finding, vertebrate PACS-2, but not C. elegans PACS, can localize in the nucleus (Atkins et al., 2014). Thus, the ability of PACS-2 to traffic to the nucleus, where it modulates SIRT1 to promote p53-dependent induction of p21, appears to have resulted from an evolutionary adaptation required to support the more complex demands of vertebrate p53 in responding to mild DNA damage by supporting p21-dependent cell cycle arrest. The predicted PACS-1 NLS is not fixed across all vertebrates. Mammals, birds and fish each express a characteristic NLS in their PACS-1 sequences. The significance of this variation, as well as validation of the PACS-1 polybasic site as a bona fide NLS, remains to be determined.

The PACS proteins first appeared in lower metazoans as membrane trafficking regulators and then adapted to evolutionary pressure by acquiring motifs to support the increasingly complex interorganellar communication pathways required for vertebrate cellular homeostasis (Figs 1 and 2). The association of mutations in the PACS1 and PACS2 genes with diseases ranging from obesity to epilepsy and cancer underscores the broad and important roles this gene family plays in vertebrate biology. As membrane trafficking regulators, the PACS proteins interact with sorting adaptors as well as a variety of trafficking motifs on client proteins, ranging from CK2-phosphorylatable acidic clusters to α-helices, in order to modulate endomembrane protein traffic (Figs 3 and 4). The PACS-1 and PACS-2 sorting pathways are also hijacked by pathogenic viruses to support production of progeny, immune evasion and protection from apoptosis.

The acquisition of the Ser⁴³⁷ Akt site enabled cytoplasmic PACS-2 to act as a bifurcation switch that separates its roles between endomembrane trafficking (phosphorylated PACS-2) and apoptotic signaling (unphosphorylated PACS-2) (see Fig. 4). The essential role for PACS-2 in mediating TRAIL-induced translocation of Bid to mitochondria was surprising as it expanded the prevailing model of Bid action. This conventional model posits that death receptors trigger dephosphorylation of Bid, which permits caspase-8-dependent proteolytic cleavage and myristoylation of human Bid at Asp^↓-Gly⁶¹ to form tBid (Li et al., 1998; Desagher et al., 2001; Kaufmann et al., 2012). However, and congruent with live-cell imaging analyses demonstrating that full-length Bid can translocate to mitochondria prior to the generation of tBid, we found that PACS-2 binds to and traffics full-length dephosphorylated Bid to mitochondria (Simmen et al., 2005; Ward et al., 2006). These findings suggest that myristoylation of tBid at Gly⁶¹ may not be absolutely required for Bid action. Indeed, granzyme B, which cleaves human Bid at Asp^↓-Ser⁷⁶ to generate a non-myristoylated Bid species, requires PACS-2 to recruit Bid to mitochondria (Li et al., 1998; Brasacchio et al., 2014). These findings suggest that following dephosphorylation, full-length Bid follows one of two distinct pathways leading to MOMP. One pathway is irreversible and relies on cleavage by caspase-8 to generate tBid, which obligates cleaved Bid to trigger MOMP (Fig. 5B). The other pathway is reversible and relies on the PACS-2-dependent translocation of full-length dephosphorylated Bid to mitochondria (Fig. 5C). Following translocation, Bid can be cleaved by multiple proteases, including caspase-8 and granzyme B, to trigger MOMP. Conversely, PACS-1 can bind de-phosphorylated Bid and block translocation to mitochondria, suggesting a mechanism to impede the apoptotic program. In HCMV-infected cells, vMIA traps PACS-2 and Bax on mitochondria, suppressing Bid recruitment and MOMP (Arnoult et al., 2004; Salka et al., 2017; Fig. 5A,C).

The PACS-2 Ser⁴³⁷ Akt site may also act as a bifurcation switch between anabolic and catabolic roles of the MAM. In response to insulin, MAM-localized mTORC2 activates Akt proteins to phosphorylate PACS-2 Ser⁴³⁷, which increases ER–mitochondria contacts (Betz et al., 2013). Accordingly, by increasing MAM integrity, Akt-phosphorylated PACS-2 may increase lipogenesis, modulate ER–mitochondria Ca²⁺ transfer and repress autophagy induction (Bernhard et al., 1952; Simmen et al., 2005; Betz et al., 2013). By contrast, starvation triggers PACS-2 Ser⁴³⁷ dephosphorylation, which would remodel MAMs to reduce lipogenesis but may increase ER–mitochondria Ca²⁺ transfer and autophagy (Fig. 4). Future work will investigate to what the extent the phosphorylation state of PACS-2 modulates MAM ‘thickness’ or changes in contacts between mitochondria with smooth or rough endoplasmic reticulum, which may in turn regulate the switch between anabolic versus catabolic functions (Giacomello and Pellegrini, 2016). It will also be interesting to determine whether PACS-2 (as well PACS-1) has roles at additional intermembrane contact sites in other contexts (Schrader et al., 2015). Finally, it will be important to determine to what extent PACS-2 Ser⁴³⁷ phosphorylation is regulated by mTORC2 and Akt localized to MAMs versus other compartments, and whether other signaling pathways might converge on Akt to phosphorylate PACS-2 (Hers et al., 2011; Betz and Hall, 2013). It will also be important to identify the PACS-2 Ser⁴³⁷ phosphatase.

Since autophagy is generally viewed as a survival mechanism, it may at first appear surprising that dephosphorylation of PACS-2 Ser⁴³⁷ is also required for TRAIL to induce Bid translocation to mitochondria and cell death (Simmen et al., 2005; Levine and Kroemer, 2008; Aslan et al., 2009; Rubinstein and Kimchi, 2012). Recent studies show, however, that autophagy is intimately coupled with TRAIL-induced cell death and can influence whether TRAIL kills cells by apoptosis (a death pathway in which cells are removed without membrane disruption) or necroptosis (an inflammatory necrotic-like death pathway that involves cell breakage) (Goodall et al., 2016). In response to TRAIL, induction of early stage autophagic machinery can promote assembly of the necrosome, which diverts the TRAIL-induced mode of killing from apoptosis to necroptosis. It will therefore be interesting to determine to what extent dephosphorylated PACS-2 modulates common or separate steps in the decision between autophagy and autophagy-modulated cell death.

The evolutionally recent acquisition of nuclear trafficking signals by PACS-1 and PACS-2 expanded their roles to also modulate nuclear gene expression (Atkins et al., 2014). The role of the PACS proteins in the nucleus is just beginning to be understood. Notably, PACS-2 is the most recent addition to a small collection of proteins that regulate SIRT1, which include deleted in breast cancer 1 (DBC1; also known as CCAR2), active regulator of SIRT1 (AROS; also known as RPS19BP1) and the moonlighting protein GAPDH (Kim et al., 2007, 2008; Atkins et al., 2014; Chang et al., 2015). It will be important to determine whether PACS-2 acts in the same or different pathways to the other SIRT1 regulators and whether PACS-2 or PACS-1 regulates HDACs in addition to SIRT1. Interestingly, the description of PACS-2 as an in vivo modulator of SIRT1 in response to DNA damage suggests that it may be involved in additional pathways controlled by SIRT1 (Brooks and Gu, 2009). For example, in response to fasting, liver SIRT1 increases the expression of genes that encode proteins involved in fatty acid oxidation and ketogenesis – pathways that protect from diet-induced obesity (Purushotham et al., 2009). Thus, the recent report that PACS-2 knockdown in liver protects mice from diet-induced obesity (Arruda et al., 2014), raises the possibility that the acquired NLS and Akt sites enable PACS-2 to coordinate SIRT1-dependent nuclear gene expression with MAM-dependent changes in mitochondrial respiration to modulate liver metabolism.

In summary, studies of the PACS proteins illustrate the vital role of protein adaptation in coordinating the seemingly autonomous actions of the nucleus, mitochondria and endomembrane systems in response to the complex challenges faced by vertebrate organ systems. Moreover, the findings that mutation or altered expression of PACS proteins are associated with pathologies ranging from cancer, obesity and viral pathogenesis to epilepsy and neurodegeneration points to a rich opportunity for insight into the molecular basis of disease through analysis of this adaptable gene family.

The authors thank R. Chaillet and J. Dacks for critically reading the manuscript, J. D. Dikeakos for helpful suggestions and C. H. Hung and C. Crump for contributions to studies described here. We also thank M. Ostrowski, C. Thauvin and H. Olson for communication of unpublished results.