The up- and down-sides of organelle interconnectivity

Abstract

Interconnectivity and feedback control are hallmarks of biological systems. This includes communication between organelles, which allows them to function and adapt to changing cellular environments. While the specific mechanisms for all communications remain opaque, unraveling the wiring of organelle networks is critical to understand how biological systems are built and why they might collapse, as occurs in aging. A comprehensive understanding of all the routes involved in inter-organelle communication is still lacking but important themes are beginning to emerge, primarily in budding yeast. These routes are reviewed here in the context of sub-system proteostasis and Complex Adaptive Systems theory.

Introduction

“When we try to pick out anything by itself, we find it hitched to everything else in the Universe.”

John Muir (1911)

Most biologists of the 21st century will nod their head in agreement with Muir’s statement; for it is now well-appreciated that all living systems are based upon networks of interactions. This connectivity is observed at every level of biology, occurring between organisms in ecological communities, between tissues, cells, and within cells (Toju et al., 2017; Vidal et al., 2011). Progress has been made in identifying connections and mapping numerous networks, and this information has guided our appreciation that there are sub-networks that form discrete functional units within any network. The organelles (mitochondria, nuclei, lysosomes, endoplasmic reticulum, Golgi) are prime examples of such functional units that need to perform specific tasks but also to be fully integrated with, and responding and reacting to, the activity of other sub-systems (Butow and Avadhani, 2004; Eisenberg-Bord et al., 2016; Raffaello et al., 2016). Thus, while originally considered to be autonomous within the confines of the cell (Wilson, 1928), organelles, compartments and structures such as peroxisomes, lipid droplets, and the plasma membrane are now appreciated as sub-networks within cells whose activities need to be fully integrated (Costanzo et al., 2016). This is especially true for the protein quality control networks as the functionality of the sub-systems rely on a constant flow and exchange of “pristine” proteins between them.

Since the beginning of this century, there has been a great effort to identify and map biological networks and, taking advantage of sequenced genomes, many large experimentally generated data sets are laying a foundation for defining interaction network maps in different organisms (Vidal et al., 2011; Snider et al., 2015; Baryshnikova et al., 2013; Hughes and de Boer, 2013). However, given the relatively limited information on metazoan networks, we focus this review on organelle interaction and interdependency in the budding yeast Saccharomyces cerevisiae. Many of the benefits of this model organism have been presented before, but paramount is that, relative to other organisms, there is an amalgam of deep mechanistic understandings about many different subsystems, including organelles and protein quality control and a sophisticated toolset for developing new approaches and insights (Botstein and Fink, 2011). In addition, the most detailed, successful mapping efforts performed to date have been made with yeast: The first (nearly) complete genetic interaction network of S. cerevisiae was recently published (Costanzo et al., 2016).

Yet even this massive effort of determining interactions between all pairs of genes is but a first step in developing our understanding of biological complexity at the cellular level. These studies examined a single phenotype (growth) and were constrained to a single genetic background [thus not taking into account that many genes interact to create a phenotype (Mackay, 2014)] and to a single environmental condition (i.e. a single temperature and nutrient source). Furthermore, the large-scale efforts define interactions without necessarily explaining function at a similar scale.

Nevertheless, from data sets such as the genetic interaction network of S. cerevisiae (Costanzo et al., 2016), the genetic network of protein quality control (e.g. Hill et al., 2016), and functional inter-organelle interdependency (e.g. Hughes and Gottschling, 2012), interesting and unexpected hints are emerging with respect to how groups of genes/proteins known to be part of one organelle, buffer against defects in other organelles and/or compartments. By first identifying the determinants critical for such sub-system interactions and buffering, a deeper mechanistic understanding of how cells/organisms are put together and how they work may indeed be attainable.

Uncovering fundamental principles that define networks of interactions and how they assemble into functional sub-networks within a biological system requires network/systems theory. Such principles in biology have often been shaped by engineering, evolution and computational theories originally conceived decades ago and that continue to evolve and be refined (see Capra, 1996; Gleick, 1988; Johnson, 2012). There have been numerous thoughtful discussions that apply network/systems theory to various aspects of biology - e.g. (Barabási and Oltvai, 2004; Sun and Kim, 2011; Toju et al., 2017; Vidal et al., 2011; Kirkwood, 2011; Mangel, 2001). However, we are particularly influenced by considering biological interaction networks as “Complex Adaptive Systems” (Cilliers, 1998; Cohen, 2016; Mangel, 2001; Miller, 2015).

Relevant features of “Complex Adaptive Systems”

In its simplest terms, a complex system consists of a large number of elements (e.g. biological molecules) that interact and self-assemble in such a way that they have “emergent” properties that are not readily predicted by simply knowing about the individual elements. Here we simply outline a number of important concepts that are most relevant for our discussion [adapted from (Cilliers, 1998)].

• The complex system is dynamic and changes with time.

• Interactions do not need to be physical, they can also be considered as transfer of information.

• Any element can be influenced by more than one interaction, but most interactions are local. If information travels over longer distances, it has the potential to be modified en route.

• Elements often assemble into clusters (subsystems) which can cooperate or compete with one another. Subsystems can themselves have emergent properties that can be considered to understand the larger system.

• There can be loops within interactions, either directly or indirectly through multiple steps. These loops provide both positive and negative feedback.

• Complex systems are “open” - they interact with the environment. As a consequence, the border of a complex system is difficult to determine and the complex system is defined by its description.

• Complex systems operate under conditions far from equilibrium. Hence a constant flow of energy is required to maintain organization and the system’s survival. Equilibrium=death.

• Complex systems have a history. They not only change with time, but past experiences influence present behavior- i.e. initial conditions can matter!

• As elements adapt in complex systems, their adaptations are governed by probabilities tied to the system’s underlying fitness. There is always a chance that they will be in suboptimal circumstances and fail.

• This fail rate will be context dependent. For instance, if the element/system was optimized for fitness under a certain set of conditions, then when those conditions are no longer present, the response may no longer be as effective.

A cell provides some ready examples of these concepts. The biological process of duplicating a cell reminds us that life and its continued propagation is a time-dependent process and that the network of interactions needed to facilitate this are necessarily dynamic. Furthermore, the ability to adapt to situational change is also fundamental to all organisms. Even “simple” changes in metabolism require a “rewiring” of metabolic networks when an organism switches from, for example, using glucose as a carbon source to fatty acids. In fact, an emergent property of Complex Adaptive Systems is the “robustness” of the organism to perturbation (Félix and Barkoulas, 2015).

The rewiring that occurs during network adaptation may be considered on many time scales, from signaling events that occur in msec within cells, to rewiring of networks that occur over evolutionary time (i.e. when comparing networks between species or even different cells within a species). Here we will primarily consider time scales that occur within an individual organism’s lifetime. Furthermore, the adaptive changes in a network can be described not only with respect to time - i.e. the duration over which the re-wiring is maintained - but also the benefit they provide to the organism.

Organelles and large cellular complexes can be viewed as subsystems within a cell, and they provide an experimentally approachable level of phenotypic analysis. There is a rich history of knowledge amassed about processes occurring within organelles and large cellular complexes (Alberts et al., 2014), and importantly, quantifiable changes in their structure (size, shape, location, number) are readily followed by microscopy (Cohen and Schuldiner, 2011; Styles et al., 2016). Furthermore, detailed genetic, chemical and physiological screens have been carried out (especially in S. cerevisiae) that provide a basis for building a network of interactions that affect organelle structure (reviewed in (Giaever and Nislow, 2014)). Altogether, this makes organelles a superb experimental platform for exploring Complex Adaptive Systems within cells.

The upside of inter-organelle communication in protein quality control

Several examples have emerged that lend support to the notion that organelle communication is key to complex adaptive feedback control of subsystem proteostasis. Such interdependency between organelle functions and quality-control systems allows for adaptive, compensatory subsystem responses when one subsystem starts to fail. It should be noted that such adaptive, compensatory responses cannot achieve true cellular proteostasis per se, but rather allow the organism/cell to reach an alternative state of the proteome, which is compatible with function and survival in the face of irreversible damage to one or several subsystems (Fig. 1) as might occur during severe stress or aging. Interestingly, such compensatory feedback systems can actually extend lifespan in many organisms when activated by subsystem failures, such as, for example, mitochondrial dysfunction (e.g. Scheckhuber et al., 2007; Berendzen et al., 2016). As exemplified below, several principal ways have emerged by which organelles and cellular compartments can maintain protein quality-control adaptability through inter-organelle communication and exchange of biochemical information.

Figure 1. Examples of interconnections and compensatory responses of subsystem PQCs.

In all figures, a, b, are different types of organelles, “c” is the cytoplasm, and “d” the nucleus.

(A) Interconnection by shared PQC factors. State 1: A shared PQC factor (red circles) is utilized for quality control in several subsystems, a,b, and c. Perturbation: A transition from the normal state (state 1) to a chronic malfunctioning of one subsystem (a) during generates internal damage (orange thread) that requires a higher load of the shared PQC factor (red circles) to assuage. As a consequence, the PQC factor becomes limited at other subsystems (b & c) leading to elevated damage also in these systems. Response: The accumulated damage trigger a response (hatched arrow) in “d” to produce more of the PQC factor that can compensate for the titration of the shared factor by the chronically malfunction subsystem “a”. State 2: The new state (compensatory state) is different from the original state but might be compatible with function and survival in the face of a chronic functional decline of subsystem “a”.

(B) Interconnection and compensation by trafficking. State 1: The accumulation of a damaged and potentially toxic form of a molecule (orange thread) in subsystem “c” is in this example dealt with by a temporal PQC (purple diamond) destroying, or clearing, the damage and a spatial PQC (green and red circles) detoxifying the damaged molecule by spatial sequestration during a normal state. The latter PQC makes use of trafficking between subsystems “b” and “a” as detailed in the text. Perturbation: If the function of the temporal PQC in “c” is irreversibly diminished (crossed-over purple diamond), accumulation of damaged and toxic molecules increase. Response: A feedback signaling (hatched arrow) response can elevate the titer of a limiting spatial PQC factor (green circles) that boost the trafficking between “b” and “a” allowing the detoxification of damaged molecules. State2: The cell has reached a new functional state that compensate for the irreversible decline in temporal PQC.

(C) Interconnection by protons, metabolites and small molecules. State 1: The concentration of a small metabolite (blueness of subsystem indicate concentration of the metabolite) is regulated such that every subsystem displays a concentration compatible with function and damage control. Perturbation: In this example, subsystem “a” fails to retain the metabolite, which leaks into subsystem “c”, and later subsystem “b”. The change in the milieu of these subsystems is promoting damage (orange threads). Response: The accumulation of damage (and perhaps the concentration of the metabolite itself) triggers a response (hatched arrows) aimed at producing PQC factors (purple diamonds and green circles) that remove such damage from both subsystems. State 2: A new state is reached compensating for, but not fixing, the chronic failure of subsystem “a”.

Communication and surveillance by titration of quality-control factors

Proteostasis is maintained through a large number of proteins of the protein quality control (PQC) systems. The canonical proteins of the PQCs include molecular chaperones and their co-chaperones and nucleotide exchange factors, organelle-specific proteases, the proteasome machinery together with the ubiquitin tagging proteins (Balchin et al., 2016; Buchberger et al., 2010), and in some organisms, proteins such as the yeast Hsp104 disaggregase with a specific role in resolving protein aggregates (Glover and Lindquist 1998). When these factors fail to fully combat the accumulation of aberrant and misfolded proteins, such proteins tend to form oligomers, amorphous aggregates and/or amyloids, which may cause cytotoxicity (Dobson 2003; Douglas et al., 2008; Balchin et al., 2016; Buchberger et al., 2010). The cytosol and the organelles have their own PQC systems but some PQC proteins are shared between them, which sets the scene for trade-offs between organelle quality control. For example, the Hsp70s of the cytosol are not only required for maintaining a proper cytosolic proteome but are also required for 26S proteasome-dependent degradation of damaged/misfolded proteins extracted from the ER and mitochondria (Nakatsukasa et al., 2008). Thus, a breakdown of PQC in the ER can have repercussions on cytosolic PQC by titrating Hsp70s (titration principle outlined in Fig. 1A). Similarly, the titration of the Hsp40 chaperone, Sis1, in the cytosol by misfolded proteins, greatly diminished PQC in the nucleus; i.e. the removal of damaged proteins targeted for Sis1-dependent degradation by the 26S proteasome (Park et al., 2013). Such titration of PQC factors can be used as a cellular surveillance mechanism allowing feedback control adjusting subsystem quality control.

Another simple means of one organelle’s sensing, and responding to, the PQC of another type of organelle is for organelles to share one or several proteins with distinct functions dependent on their subsystem location, such as Cbs2; a protein shared between the mitochondria and nucleus. Cbs2 and Cbs1 are non-redundant mitochondrial translational activators of the mitochondrial-encoded cytochrome oxidase (COB) mRNA. In the absence of Cbs1 and the resulting breakdown of COB translation, Cbs2 is increasingly localized to the nucleus where it boosts Sir2-dependent genome maintenance and cytosolic PQC, which results in elevated stress resistance and longevity assurance in the face of a chronic collapse in mitochondrial function (Caballero et al., 2011). This serves as another example of how the demise of a process in one organelle, by adaptive compensatory responses, allows the cell to reach a new, but different, functional state compatible with survival.

Communication by inter-organelle trafficking

In parallel with temporal PQC aimed at fixing or degrading misfolded proteins, a spatial PQC is involved in sequestering aggregated proteins into specific sites within the cell (Fig. 2B). Upon proteostasis stress, such as heat shock, misfolded/damaged proteins first accumulate in small aggregates called stress foci (Spokoini et al., 2012), Q-bodies (Escusa-Toret et al., 2013), peripheral aggregates (Specht et al., 2011), and/or CytoQs (Miller et al., 2015). Some of these aggregates appear to be dispersed through the cytosol whereas others are associated with the ER, mitochondria, and vacuole (Escusa-Toret et al., 2013, Kaganovich et al., 2008, Miller et al., 2015, Specht et al., 2011, Zhou et al., 2014). Upon prolonged stress, these aggregates merge and are packed into inclusions, including the IPOD (Insoluble-Protein-Deposit), JUNQ (JUxta-Nuclear-Quality-control), and INQ (Intra-Nuclear-Quality-control)(Kaganovich et al., 2008; Miller et al., 2015; Spokoini et al., 2012). The formation of such inclusions is factor-dependent, requiring calmodulin and functional actin-cables along with chaperones and the Hsp104 disaggregase (Liu et al., 2010; 2011; Song et al., 2014; Specht et al., 2011; Yang et al., 2016; Zhao et al., 2016). In addition, Btn2 and Cur1, control the spatial deposition of misfolded proteins into discrete inclusions together with chaperones (e.g. Sis1) and small heat shock proteins in an as yet undefined manner (Malinovska et al., 2012; Park et al., 2013; Specht et al., 2011). Moreover, a recent genome-wide screen for genes involved in the formation and asymmetrical inheritance of protein inclusions pinpointed proteins involved in inter-organelle communication and endomembrane trafficking as unexpected auxiliary factors required for proper spatial PQC (Hill et al., 2016).

Figure 2. How inter-organelle interactions can mediate a cascade of decline that define a cellular aging process.

A schematic representation of the changes that occur during replicative aging of S. cerevisiae cells within, and between, the plasma membrane, vacuole, mitochondria and nucleus is presented in a series of five panels. Cellular age increase from top to bottom.

(A) Interactions in newborn mother cells. Representation of a newborn cell in which cytoplasmic protons (H⁺) are being pumped into the vacuole and across the plasma membrane (by the V-ATPase and Pma1, respectively). Consequently the vacuole is relatively acidic (represented by large “H⁺” and bright green vacuole color) and is able to store amino acids because of the activity of the antiporters (brown disc with dual arrows) that use the acidity of the vacuole to bring in amino acids. On the right side of the figure, mitochondria are shown providing the obligate cofactors, iron-sulfur clusters, to DNA repair and replication enzymes in the nucleus.

(B) Elevated Pma1 competes with V-ATPase for cytoplasmic protons. In this first step of the aging cascade, increased levels of Pma1 (3x more Pma1 trapezoids) pump more protons from the cytoplasm to the plasma membrane at the expense of the vacuole (smaller “H⁺” and less bright green)

(C) Reduced vacuole acidity leads to mitochondrial dysfunction. The amino acid antiporters require an acidic vacuole; amino acids are no longer imported into the vacuole and consequently buildup in the cytoplasm. The increased level of cytoplasmic amino acids leads to mitochondrial dysfunction (reduced ΔΨ; indicated by faded mitochondrial color).

(D) Reduced iron-sulfur cluster production follows mitochondrial dysfunction. Fewer iron-sulfur clusters make there way to the nucleus (dotted line).

(E) Genome instability increases with reduced DNA enzyme activities. Iron-sulfur cluster requiring DNA repair and replication enzymes are proposed to have reduced activity (orange to red) because of fewer available iron-sulfur clusters. Aberrant chromosomes are represented by dashed lines.

The endomembrane trafficking routes have been studied extensively in S. cerevisiae and many of the routes and factors discovered in this organisms have turned out to be evolutionary conserved (Gurkan et al., 2007). The exchange of material, including lipids, proteins and metabolites, is materialized by several interconnected trafficking routes; the exocytosis, or secretory (SEC), pathway provides a route for newly-synthesized proteins to the plasma membrane (or the external medium), whereas the VPS (Vacuolar Protein Sorting; through endosomes) and ALP (Alkaline Phosphatase) pathways direct proteins/endomembranes to the vacuole. Plasma membrane proteins, in turn, are internalized by the END pathway of endocytosis and directed to endosomes, which can be sorted to the vacuole for degradation or redirected to the Golgi through the recycling RCY pathway (Pelham 1999; Bröcker et al., 2010; Gurkan et al., 2007). A cascade of factors, including Rab-like GTPases, multi-subunit tethering factors, and SNARE proteins govern the specificity of trafficking to correct organelles and the docking of vesicles to specific target sites (Pelham 1999; Bröcker et al., 2010). In addition, lipid-signaling molecules, including phosphoinositides, are key regulators of endomembrane trafficking and its directionality (Odorizzi et al., 2000).

Some factors regulating the specificity of trafficking are, as mentioned, also required for the correct sorting of misfolded cytosolic proteins to specific aggregation sites and quality control compartments, including the IPOD. Specifically, defects in Golgi-to-vacuole trafficking, late endocytosis, and actin-Myo2-dependent vesicle tethering to the vacuole retard the formation of cytosolic Hsp104-associated IPOD inclusions (Hill et al., 2016; Eitzen et al., 2002), which are associated with the outer surface of the vacuole (Spokoini et al., 2012). Similarly, it was recently demonstrated that prion aggregates are sequestered into IPOD in a Myo2-based vesicular transport-dependent manner (Kumar et al., 2016). The fact that defects in such inter-organelle, Myo2-dependent, trafficking routes rendered misfolded reporter proteins toxic (Hill et al., 2016) suggests that inter-organelle trafficking is an important part of proper and protective PQC. Also, the data demonstrate that spatial cytosolic PQC is dependent on fully-functional ER, Golgi, and vacuolar organelles and the proper communication/trafficking between them.

The interconnection between organelle function/trafficking and spatial PQC may be accomplished be several mutually-inclusive mechanisms. First, it is possible that misfolded and aggregated proteins ‘hitchhike’ on endomembrane vesicles, as such proteins have been shown to interact with membrane lipids (Meriin et al., 2003; 2007). Second, members of the PQC that bind to misfolded and/or aggregated proteins may interact directly with membrane vesicles. A case in point is the physical interaction between the disaggregase Hsp104 and the yeast dynamin GTPase homolog, Vps1. Vps1 is required for pinching off vesicles from the Golgi and directing them, through late endosomes and MVB, to the vacuole (Smaczynska-de et al., 2010). This direct interaction between Vps1 and Hsp104 may carry along misfolded proteins on the surface of these vesicles en route to the vacuole. Prion aggregates may participate in an analogous shuttle, system as they have been reported to co-localize with components of the CVT vesicular transport machinery, including Myo2 and Atg8/9 (Kumar et al., 2016).

Spatial, endocytosis-dependent sequestration of aggregates into large IPOD inclusions also appears to provide the cell with the means of a compensatory adaptive response to failures in disaggregation: Cells lacking disaggregating activity display accelerated aging, which can be fully compensated for and reversed by boosting trafficking-dependent deposition of misfolded proteins into IPODs (principle outlined in Fig. 1B; Hill et al., 2016). Thus, boosting endomembrane-dependent spatial deposition of aggregates into large and discrete inclusions appears to bypass the need for temporal aggregate resolution (Hill et al.,).

Communication by direct inter-organelle contacts

Direct physical interaction between different organelles provides another means of inter-organelle communication, adaptability and the integration of compartmentalized processes that might impact on spatial PQC. Such interactions are directed by membrane contact sites (MCS), for example the vacuole-mitochondria patches (vCLAMPs), the ER-mitochondria encounter sites (ERMES), and the perinuclear ER-vacuolar junctions (NVJs), allowing close proximity of membranes of two organelles (Murley and Nunnari 2016: Lahiri et al., 2015; Elbaz and Schuldiner 2011). In addition, MCSs have been found between virtually all organelles, including the ER and plasma membrane, ER and mitochondria, ER and late endosomes/multivesicular bodies, ER/nucleus and vacuole, ER and Golgi, ER and Peroxisomes, ER and lipid droplets, mitochondria and vacuole, and mitochondria and peroxisomes. They provide a zone where signals and small molecules, including lipids and calcium and other ions, can be exchanged between intracellular compartments; all of which are important for coordinated control of membrane dynamics (Murley and Nunnari 2016: Lahiri et al., 2015; Elbaz and Schuldiner 2011).

The MCSs have received increased attention as they are linked to several human diseases and their functionality affects apoptosis, the immune response, organelle dynamics/performance, and ion and lipid homeostasis (Murley and Nunnari 2016: Lahiri et al., 2015; Elbaz and Schuldiner 2011). The role of MCSs in protein quality control is not understood but may be of particular importance as some of them facilitate the formation of lipid microenvironments enriched for specific lipids that may form a platform/domain for the spatial formation of protein quality control compartments. For example, the vacuolar protein Vac8, which we recently demonstrated to be required for the formation of protein inclusions (Hill et al., 2016), forms a nucleus/ER-vacuole MCS via interaction with the integral ER-proteins Nvj1 and Ltc1 (Murley and Nunnari 2016: Lahiri et al., 2015; Elbaz and Schuldiner 2011). Ltc1 is a sterol transporter required for the formation of sterol lipid-enriched vacuolar membrane domains during stress (Murley and Nunnari, 2016). The exact function of such domains is not known but is interesting in the context of PQC as sterols can promote the formation of lipid rafts, which have been shown to be associated with misfolded and aggregating disease proteins (Lewis and Hooper 2011; Agostini et al., 2013).

Communication by small molecules and ions

Due to organelle interconnectivity, alterations in the pool size of small molecules and ions at one organelle can cause repercussion and generate responses in distant, interconnected, locations. The pool size of protons is of special interest with respect to PQC since the pH of the environment affects the protonation state of its proteins, which is critical for non-covalent interactions and enzymatic activity. Fluctuations in pH can therefore destabilize the native form of a protein as well as protein-protein interactions (Matthew et al., 1985).

The vacuole/lysosome is the most acidic organelle in the cell and the loss of proper vacuolar pH control could affect interconnected subsystem PQCs in both direct and indirect ways (Fig. 1C): Apart from affecting proteins in the vacuole directly by such mechanisms, a collapse in vacuolar pH control is also expected to affect the protonation state of proteins in the cytosol and lead to protein misfolding/aggregation and a concomitant adaptive response elicited by the cytosolic PQC (Fig. 1C). Further, the reduced mitochondrial-membrane potential caused by the collapse in vacuolar pH control (Hughes & Gottschling, 2012), is expected to generate an increase in ROS production and mitochondrial protein damage (Machida & Tanaka, 1999, Pozniakovsky et al., 2005), which, in turn, elevate aggregation in the cytosol partly because protein import is impeded and unprocessed mitochondrial proteins accumulate in the cytosol (Erjavec et al., 2013). Interestingly, such defects generated by mitochondrial dysfunctions can be compensated for my a boost in cytosolic PQC and longevity assured despite chronic failures in mitochondrial function (Erjavec et al., 2013). Loss of vacuolar acidification will also cause a breakdown in vesicle trafficking and heterotypic and homotypic fusion to the vacuole (Coonrod et al., 2013; Baars et al., 2007), which, in turn, will cause aberrations in the spatial sequestration of aggregated proteins to the IPOD site (Hill et al., 2016). However, such effects on PQC control in the cytosol elicited by vacuolar dysfunction can be compensated for by boosting vacuole-directed endocytosis (Hill et al., 2016), suggesting that complex compensatory responses, including inte-organelle trafficking, can buffer against chronic failures in vacuolar pH control. The exact mechanism behind such compensatory responses remains to elucidated and may provide novel insights into cellular maintenance and longevity assurance in the face of organelle failures. A principle for a compensatory PQC response to defects in metabolite homeostasis in one organelle is depicted in figure 1C.

Communication between processes of organelle and damage inheritance

Yeast cytokinesis encompasses an asymmetrical inheritance of aggregated proteins and dysfunctional organelles and recent data suggest that these phenomena may be interrelated. Mother cell-biased segregation of aggregated proteins requires actin cable-dependent processes and the polarisome (Liu et al., 2010; Tessarz et al., 2009); a complex at the tip of the daughter cell required for actin cable nucleation (Fig. 2B). The role of actin cables in aggregate retention might be due to tethering of aggregate-associated vesicles/organelles to actin cables through the myosin protein Myo2 (Hill et al., 2016; Kumar et al., 2016). The filtering device involved in retaining dysfunctional mitochondria in mother cells exploits, like protein aggregates, actin cables and their retrograde flow of actin cables (Higuchi et al., 2013). Retrograde flow of actin cables is the result of actin nucleation taking place in the polarisome at the tip of the daughter cell. Thus, actin cables move ‘backwards’ towards the mother cell such that any cargo transported into the daughter cell by the myosin motor protein, Myo2, have to move (anterograde) against this actin flow. Reduced and healthy mitochondria move faster against this flow than oxidized (dysfunctional) mitochondria resulting in an enrichment of healthy mitochondria in the daughter cell during cytokinesis (Higuchi et al., 2013). Interestingly, the filtering process works in an even more stringent manner when the retrograde flow rate is increased by genetic means (Higuchi et al., 2013; Nyström 2013).

The coordinated retention of protein aggregates and dysfunctional mitochondria could, at least in part, be the result of smaller aggregates of the cytosol associating with the outer surface of mitochondria (Zhou et al., 2014) but other organelles, and the factors controlling their inheritance, have recently been demonstrated to be required for establishing asymmetrical segregation of aggregates (Hill et al., 2016). For example, the vacuole inheritance adaptor protein Vac17 is a limiting factor for both vacuole inheritance and the retention of aggregated proteins in mother cells (Hill et al., 2016). When overproduced, Vac17 boosts aggregate asymmetry and extends replicative lifespan in a Myo2-dependent manner. This is interesting since the transport and partitioning of organelles/vesicles, including the nucleus, vacuole, mitochondria, peroxisome, secretory vesicles, and lipid droplets, all rely on Myo2 in yeast (Hammer and Sellers, 2011). Different organelles use different protein adaptors to attach to Myo2 and since it has been suggested that Myo2 is limiting in the cell (Eves et al., 2012), these adaptors are expected to compete for Myo2 binding: As a result, organelles compete for their partitioning during cytokinesis. Such a competition might be key to rejuvenation, because filtering processes, such as those acting on dysfunctional mitochondria, appear to rely on restricted inheritance, i.e. functional and dysfunctional organelles compete for daughter inheritance (Higuchi et al., 2013).

Another key factor, besides Myo2, involved in the asymmetrical segregation of multiple factors and organelles is Bud6. Bud6 is member of the polarisome required for the asymmetrical segregation of protein aggregates and the segregation of all organelles dependent on actin cables for their inheritance. In addition, Bud6 is a key member of the ER diffusion barrier (Luedeke et al., 2005): Soluble proteins diffuse rapidly throughout the ER lumen whereas the diffusion of ER membrane proteins is restricted in a septin- and Bud6-dependent manner at the bud neck (Luedeke et al., 2005). In addition, misfolded proteins of the ER are retained in the mother cell by a lateral diffusion barrier consisting of septins, Bud6, the Bud1 GTPase, and sphingolipids (Clay et al., 2014).

While recent progress has started to uncovered how aggregates and specific dysfunctional organelles are prevented from being inherited by the progeny, we do not know how, and if, these individual filtering processes are coordinated - given the interconnected nature of organelle function and subsystem PQCs, it appears that they must be so to ensure a rejuvenated progeny but the mechanism behind such coordination remains to be uncovered.

The downside of interconnectivity: A cascade of decline

As noted above, interconnectivity of a network provides robustness and adaptability to a biological system. However if a subsystem becomes defective and does not re-establish normal function, the interconnectivity may lead to other subsystems becoming compromised too. Evidence for such a cascade of decline that spreads to other connected subsystems will be presented below by examining the replicative aging of S. cerevisiae cells. In a typical diploid lab strain, a single budding yeast (mother) cell divides asymmetrically, producing ~30 daughters, before arresting and ultimately lysing (Denoth Lippuner et al., 2014). This trait had made it a model system for cellular aging and has led to several discoveries about the aging process that are shared with aging in metazoan cells (Steinkraus et al., 2008).

We propose that Complex Adaptive Systems theory may not only help to explain how biological systems are put together, but also how their failures might drive the process of aging. We present a brief historical perspective that attempts to “connect-the-dots” in causal contributions during the yeast aging process, in order to provide the basis for the premise of a cascading decline, and to offer perspective for future discovery.

Mitochondria to Nuclear Genome

In diploid lab strains carrying heterozygous alleles, the frequency of loss-of-heterozygosity (LOH) increases 40-fold to 100-fold in the progeny of old mothers (>25 cell divisions) compared to that in the progeny from young mothers (McMurray and Gottschling, 2003). The increase in LOH correlates with the daughters of old mother cells displaying mitochondrial dysfunction, including a reduced inner mitochondrial membrane potential (ΔΨ) (Veatch et al., 2009). The ΔΨ is required for critical aspects of mitochondrial function (Kulawiak et al., 2013).

The connection between increased genomic instability and the dysfunctional mitochondria appeared to be through the mitochondrially-synthesized iron-sulfur clusters (ISCs) (Lill et al., 2012). They serve as cofactors in a number of proteins, including those critical to genome maintenance and integrity, such as enzymes involved in Base-Excision Repair, Nucleotide-Excision Repair, Double-Stranded Break repair and DNA Primase (Paul and Lill, 2015). This suggests that reduced iron-sulfur cluster production leads to compromised genome maintenance by reduced activity in one or more of these enzymes (Veatch et al., 2009).

Because daughters of old mother cells receive dysfunctional mitochondria (Veatch et al., 2009), this implies that the asymmetric preference of daughters to inherit functional mitochondria breaks down late in the mother’s life, and/or that the mitochondria in older mothers are dysfunctional (Higuchi et al., 2013).

Vacuole to Mitochondria

Mitochondria in aged mother cells are indeed dysfunctional: they are highly fragmented with dramatically reduced ΔΨ (Hughes and Gottschling, 2012). This dysfunction begins to appear in cells by the time they undergo ~10 cell divisions and becomes progressively worse with increasing age (i.e. more fragmentation). A genetic screen for genes that delayed the onset of the age-associated mitochondrial defect led to the discovery that the vacuole became less acidic early in the lifespan of the yeast mother cell and that several cell divisions later, this change led to the reduced mitochondrial ΔΨ and fragmentation (Hughes and Gottschling, 2012). The link between vacuolar pH increase and reduced mitochondrial function appears to be mediated by the vacuole’s reduced capacity to store neutral amino acids (import and storage require an acidic vacuole) (Russnak et al., 2001). How reduction in neutral amino acid storage leads to a decline in mitochondrial membrane potential remains to be determined.

In addition to the mitochondrial hyper-fragmentation and loss of membrane potential detected in aging mother cells, there is a large-scale remodeling of the mitochondrial proteome (Hughes et al., 2016). This too is initiated by the reduced vacuolar acidity and involves selective removal of a subset of membrane proteins from the mitochondrial inner and outer membranes, while leaving the remainder of the organelle intact. Selective removal of preexisting proteins is achieved by sorting into a mitochondrial-derived compartment, or MDC, followed by release through mitochondrial fission and elimination by autophagy. It appears that the MDC pathway provides protection to mitochondria in times of stress, because preventing MDC formation exacerbates mitochondrial dysfunction. Importantly, this degradation process is not wholesale elimination of mitochondria (mitophagy), but rather an apparent attempt at adaptation against full blown mitochondrial dysfunction (Sugiura et al., 2014). However, in very old yeast cells, MDC-mediated autophagy appears to collapse and MDC fragments become ubiquitous in the cytoplasm (Hughes et al., 2016).

Plasma Membrane to the Vacuole

What is the origin of the vacuole’s increase in pH, which begins shortly after a new mother goes through only a few cell divisions? It does not appear to be the result of vacuole damage, but rather a competition for free protons between the vacuole and the plasma membrane (Henderson et al., 2014). Pma1 is an ATP-dependent proton pump that generates the essential pH gradient across the plasma membrane, consuming as much as 40% of the cell’s ATP to pump protons from the cytoplasm to the external environment (Capieaux et al., 1989), which in turn facilitates selective transport of a variety of critical molecules into and out of the cell (Cyert and Philpott, 2013). Pma1 has an unusual property: it is a very long lived protein and continues to accumulate at the plasma membrane over the first several divisions of a mother cell’s life (Thayer et al., 2014). The vacuole pH increase appears to be tied to Pma1 by mass action. There are ~10⁶ Pma1 molecules/young mother cell and ~10⁵ vacuolar ATPase complexes/cell, which hydrolyzes ATP to pump cytoplasmic free protons to acidify the vacuole (Ghaemmaghami et al., 2003). For a cell the size of yeast, there are but 3000 free protons at pH 7. Thus there appears to be a competition between these two abundant ATP-dependent pumps for the same proton pool. It is worth noting that the reduction in vacuole acidification with age is not the same as a complete loss-of-function in vacuolar ATPase - the vacuole matrix is still more acidic than than the cytoplasm, but not as acidic as it is in younger cells (Hughes and Gottschling, 2012).

The examples presented here define a cascade of events in which different subsystems become altered/compromised because of their connectivity to one another. It is a cascade of events in the sense that once a new “state” is attained in one subsystem, it may take some time (several-to-many cell divisions) before its new state is manifested in an altered state of the interconnected subsystem. In the scenario described above and schematized in Figure 2, the accumulation of Pma1, which begins within the initial cell division, manifests a few divisions later in reduced vacuole acidity (Henderson et al., 2014). Once this new vacuole pH is achieved it takes 5–10 additional cell divisions before the mitochondrial alterations begin to appear (Hughes and Gottschling, 2012). And the mitochondrial defects in the mother do not appear to be passed on to the daughters until another ~10 cell divisions have occurred (Veatch et al., 2009). As the system (organism) adapts to intrinsic perturbations with age, the functional state changes over time (i.e. the network connections are different in young compared to old organisms).

The relative order of these age-associated events is preserved when the time of onset of an early age event is specifically delayed by lifespan extending mutations or growth conditions [e.g. tor1Δ, sch9Δ, low glucose - see (Hughes and Gottschling, 2012)]. The ability to monitor these cell biological changes through the various perturbations provides more evidence for the stated causal relationships. Yet it is worth noting again that in the context of a Complex Adaptive System, the phenotypes and inter-organelle relationships described here likely reflect a modicum of the cellular changes that are occurring. For instance, the change in vacuolar pH is likely eliciting other cellular events and responses in the cell that were not specifically examined (see above discussion on PQC). In addition, the examples presented here were in a particular set of strains and growth conditions. It is abundantly obvious that genetics and environment will “re-wire” the networks we have described. This presents challenges and opportunities to identify additional connections between subsystems that are altered during the aging process and how/whether they manifest as age-associated phenotypic cascades.

Concluding remarks

As we have described above, organelles serve as an experimentally tractable manifestation for observing, analyzing and understanding biological processes through networks guided by the principles of Complex Adaptive System theory. Goals to decipher the architecture required to build a complex adaptive system through coordinated and interactive organelle functions includes large experimentally generated data sets defining different interaction networks. Such approaches have been especially developed and successful using budding yeast as the model system. While some of these datasets are fairly “straightforward” in providing connections for a primary map - e.g. protein-protein interactions and environmentally-induced alterations in the proteome - others require additional computational modeling to infer connections - e.g. analysis of gene x gene phenotypic similarity – and the inferred connections need to be confirmed by meticulous in-depth analysis (Baryshnikova et al., 2013; Hughes and de Boer, 2013). In addition, the current state of the mapping efforts are limited, in part, due to technical adversities (e.g. inability to measure every type of interaction), historically defined interests (e.g. there is a disproportionate amount of data for cancer-related proteins/genes), the immense scale of commitment to carry out thorough genome-wide screens (the S. cerevisiae genetic interaction network was a nearly two decade-long effort) and the complexity of biology as we currently understand it (e.g. allelic variation between individuals in a population will continue to perplex efforts to understand phenotypes for the foreseeable future (Mackay 2014)). Furthermore, the large-scale efforts define interactions without necessarily explaining function at a similar scale. In fact, assigning mechanistic insights to the interactions is typically determined independently of the large-scale screening and/or as a “proof-of-principle” about the value of the screening effort. Thus, despite the herculean efforts in network analysis, most of the maps are first and incomplete drafts that offer less than perhaps anticipated with respect to understanding the complex adaptive system that embody a living cell/organism.

Nevertheless, we believe there are several ways forward in reaching a greater understanding of the architecture of living organisms and how the connections of their subsystems/organelles are wired to provide robustness and adaptability:

First, approaches aimed at combining available network data sets have the potential to identify how the integration of metabolic, genetic, protein, and organelle interaction networks might give rise to emergence, i.e. higher-order levels of complexity and adaptability. Analytically superimposing and relating data obtained from individual networks – i.e. go from a 2D analysis of networks to a 3D, 4D, etc. view of network interactions, constitute a challenge, though a successful outcome would have a great impact on biology.

Second, the newly finished genetic interaction map of yeast should be extremely useful as a blueprint, or Rosetta stone, for learning the language of organelle interconnectivity (Costanzo et al., 2016). The genetic interaction network reveals interesting and unexpected genetic connections between organelles, for example how the functions of some specific genes of one organelle are required for the fitness of cells upon the genetically-derived functional demise of another organelle. The specificity of such pairs of genes of different organelles buffering against each other’s loss of function could guide our efforts to elucidate how organelle buffering interactions are designed, i.e. through organelle contacts, inter-organelle trafficking, or small molecules.

Third, the analysis of how and why a complex adaptive system breaks down presents a rationale for understanding how it was put together in the first place. The analysis of the cellular etiology of aging offers such a unique opportunity: analyzing the sequential decline in organelle function and how this is causally connected during aging provides new insights into organelle wiring. In addition, thinking about aging in terms of subsystem compensations opens up new ways to reflect on how the system is built; for when a mutational defect in a certain organellar process fails to affect lifespan (or cause lifespan extension) it is often a result of the organism having evolved adaptive and interconnected feedback control that efficiently buffers against defects in one process. Analysis of the specificity of such buffering processes and how defects in one organelle is sensed by other sub-systems offers new insights into the wiring of the cell as a complex adaptive system.

Going beyond the yeast unicellular paradigm, it is clear that other species, and different cell types/tissues within a species, display distinct types of architecture and network interactions that manifest as differences in responses to change. Moreover, allelic differences between individuals may lead to slightly different network wirings, which will affect both compensatory responses and the deleterious domino-like effects of decline during disease and aging. This might suggest a bewildering level of complexity and species-/individual-dependent variation, and on top of this, we are still decades away from achieving a network that “defines” life even in simple unicellular organisms. Yet as more data accumulates, there is a strong indication that organelle interconnectivity may share common principles across kingdoms and phyla and that applying these principles to the complex adaptive systems of other cells/organisms is a path well worth exploring (Murley and Nunnari, 2016).